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_eth_ctrl.h>
25 #include <rte_ethdev_driver.h>
27 #include <rte_flow_driver.h>
28 #include <rte_malloc.h>
32 #include "mlx5_defs.h"
34 #include "mlx5_glue.h"
35 #include "mlx5_flow.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_tcf_drv_ops;
46 extern const struct mlx5_flow_driver_ops mlx5_flow_verbs_drv_ops;
48 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops;
50 const struct mlx5_flow_driver_ops *flow_drv_ops[] = {
51 [MLX5_FLOW_TYPE_MIN] = &mlx5_flow_null_drv_ops,
52 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
53 [MLX5_FLOW_TYPE_DV] = &mlx5_flow_dv_drv_ops,
55 [MLX5_FLOW_TYPE_TCF] = &mlx5_flow_tcf_drv_ops,
56 [MLX5_FLOW_TYPE_VERBS] = &mlx5_flow_verbs_drv_ops,
57 [MLX5_FLOW_TYPE_MAX] = &mlx5_flow_null_drv_ops
62 MLX5_EXPANSION_ROOT_OUTER,
63 MLX5_EXPANSION_ROOT_ETH_VLAN,
64 MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN,
65 MLX5_EXPANSION_OUTER_ETH,
66 MLX5_EXPANSION_OUTER_ETH_VLAN,
67 MLX5_EXPANSION_OUTER_VLAN,
68 MLX5_EXPANSION_OUTER_IPV4,
69 MLX5_EXPANSION_OUTER_IPV4_UDP,
70 MLX5_EXPANSION_OUTER_IPV4_TCP,
71 MLX5_EXPANSION_OUTER_IPV6,
72 MLX5_EXPANSION_OUTER_IPV6_UDP,
73 MLX5_EXPANSION_OUTER_IPV6_TCP,
75 MLX5_EXPANSION_VXLAN_GPE,
79 MLX5_EXPANSION_ETH_VLAN,
82 MLX5_EXPANSION_IPV4_UDP,
83 MLX5_EXPANSION_IPV4_TCP,
85 MLX5_EXPANSION_IPV6_UDP,
86 MLX5_EXPANSION_IPV6_TCP,
89 /** Supported expansion of items. */
90 static const struct rte_flow_expand_node mlx5_support_expansion[] = {
91 [MLX5_EXPANSION_ROOT] = {
92 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
95 .type = RTE_FLOW_ITEM_TYPE_END,
97 [MLX5_EXPANSION_ROOT_OUTER] = {
98 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_ETH,
99 MLX5_EXPANSION_OUTER_IPV4,
100 MLX5_EXPANSION_OUTER_IPV6),
101 .type = RTE_FLOW_ITEM_TYPE_END,
103 [MLX5_EXPANSION_ROOT_ETH_VLAN] = {
104 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH_VLAN),
105 .type = RTE_FLOW_ITEM_TYPE_END,
107 [MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN] = {
108 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_ETH_VLAN),
109 .type = RTE_FLOW_ITEM_TYPE_END,
111 [MLX5_EXPANSION_OUTER_ETH] = {
112 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
113 MLX5_EXPANSION_OUTER_IPV6,
114 MLX5_EXPANSION_MPLS),
115 .type = RTE_FLOW_ITEM_TYPE_ETH,
118 [MLX5_EXPANSION_OUTER_ETH_VLAN] = {
119 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_VLAN),
120 .type = RTE_FLOW_ITEM_TYPE_ETH,
123 [MLX5_EXPANSION_OUTER_VLAN] = {
124 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
125 MLX5_EXPANSION_OUTER_IPV6),
126 .type = RTE_FLOW_ITEM_TYPE_VLAN,
128 [MLX5_EXPANSION_OUTER_IPV4] = {
129 .next = RTE_FLOW_EXPAND_RSS_NEXT
130 (MLX5_EXPANSION_OUTER_IPV4_UDP,
131 MLX5_EXPANSION_OUTER_IPV4_TCP,
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),
151 .type = RTE_FLOW_ITEM_TYPE_IPV6,
152 .rss_types = ETH_RSS_IPV6 | ETH_RSS_FRAG_IPV6 |
153 ETH_RSS_NONFRAG_IPV6_OTHER,
155 [MLX5_EXPANSION_OUTER_IPV6_UDP] = {
156 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
157 MLX5_EXPANSION_VXLAN_GPE),
158 .type = RTE_FLOW_ITEM_TYPE_UDP,
159 .rss_types = ETH_RSS_NONFRAG_IPV6_UDP,
161 [MLX5_EXPANSION_OUTER_IPV6_TCP] = {
162 .type = RTE_FLOW_ITEM_TYPE_TCP,
163 .rss_types = ETH_RSS_NONFRAG_IPV6_TCP,
165 [MLX5_EXPANSION_VXLAN] = {
166 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH),
167 .type = RTE_FLOW_ITEM_TYPE_VXLAN,
169 [MLX5_EXPANSION_VXLAN_GPE] = {
170 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
172 MLX5_EXPANSION_IPV6),
173 .type = RTE_FLOW_ITEM_TYPE_VXLAN_GPE,
175 [MLX5_EXPANSION_GRE] = {
176 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4),
177 .type = RTE_FLOW_ITEM_TYPE_GRE,
179 [MLX5_EXPANSION_MPLS] = {
180 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
181 MLX5_EXPANSION_IPV6),
182 .type = RTE_FLOW_ITEM_TYPE_MPLS,
184 [MLX5_EXPANSION_ETH] = {
185 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
186 MLX5_EXPANSION_IPV6),
187 .type = RTE_FLOW_ITEM_TYPE_ETH,
189 [MLX5_EXPANSION_ETH_VLAN] = {
190 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VLAN),
191 .type = RTE_FLOW_ITEM_TYPE_ETH,
193 [MLX5_EXPANSION_VLAN] = {
194 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
195 MLX5_EXPANSION_IPV6),
196 .type = RTE_FLOW_ITEM_TYPE_VLAN,
198 [MLX5_EXPANSION_IPV4] = {
199 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4_UDP,
200 MLX5_EXPANSION_IPV4_TCP),
201 .type = RTE_FLOW_ITEM_TYPE_IPV4,
202 .rss_types = ETH_RSS_IPV4 | ETH_RSS_FRAG_IPV4 |
203 ETH_RSS_NONFRAG_IPV4_OTHER,
205 [MLX5_EXPANSION_IPV4_UDP] = {
206 .type = RTE_FLOW_ITEM_TYPE_UDP,
207 .rss_types = ETH_RSS_NONFRAG_IPV4_UDP,
209 [MLX5_EXPANSION_IPV4_TCP] = {
210 .type = RTE_FLOW_ITEM_TYPE_TCP,
211 .rss_types = ETH_RSS_NONFRAG_IPV4_TCP,
213 [MLX5_EXPANSION_IPV6] = {
214 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV6_UDP,
215 MLX5_EXPANSION_IPV6_TCP),
216 .type = RTE_FLOW_ITEM_TYPE_IPV6,
217 .rss_types = ETH_RSS_IPV6 | ETH_RSS_FRAG_IPV6 |
218 ETH_RSS_NONFRAG_IPV6_OTHER,
220 [MLX5_EXPANSION_IPV6_UDP] = {
221 .type = RTE_FLOW_ITEM_TYPE_UDP,
222 .rss_types = ETH_RSS_NONFRAG_IPV6_UDP,
224 [MLX5_EXPANSION_IPV6_TCP] = {
225 .type = RTE_FLOW_ITEM_TYPE_TCP,
226 .rss_types = ETH_RSS_NONFRAG_IPV6_TCP,
230 static const struct rte_flow_ops mlx5_flow_ops = {
231 .validate = mlx5_flow_validate,
232 .create = mlx5_flow_create,
233 .destroy = mlx5_flow_destroy,
234 .flush = mlx5_flow_flush,
235 .isolate = mlx5_flow_isolate,
236 .query = mlx5_flow_query,
239 /* Convert FDIR request to Generic flow. */
241 struct rte_flow_attr attr;
242 struct rte_flow_action actions[2];
243 struct rte_flow_item items[4];
244 struct rte_flow_item_eth l2;
245 struct rte_flow_item_eth l2_mask;
247 struct rte_flow_item_ipv4 ipv4;
248 struct rte_flow_item_ipv6 ipv6;
251 struct rte_flow_item_ipv4 ipv4;
252 struct rte_flow_item_ipv6 ipv6;
255 struct rte_flow_item_udp udp;
256 struct rte_flow_item_tcp tcp;
259 struct rte_flow_item_udp udp;
260 struct rte_flow_item_tcp tcp;
262 struct rte_flow_action_queue queue;
265 /* Map of Verbs to Flow priority with 8 Verbs priorities. */
266 static const uint32_t priority_map_3[][MLX5_PRIORITY_MAP_MAX] = {
267 { 0, 1, 2 }, { 2, 3, 4 }, { 5, 6, 7 },
270 /* Map of Verbs to Flow priority with 16 Verbs priorities. */
271 static const uint32_t priority_map_5[][MLX5_PRIORITY_MAP_MAX] = {
272 { 0, 1, 2 }, { 3, 4, 5 }, { 6, 7, 8 },
273 { 9, 10, 11 }, { 12, 13, 14 },
276 /* Tunnel information. */
277 struct mlx5_flow_tunnel_info {
278 uint32_t tunnel; /**< Tunnel bit (see MLX5_FLOW_*). */
279 uint32_t ptype; /**< Tunnel Ptype (see RTE_PTYPE_*). */
282 static struct mlx5_flow_tunnel_info tunnels_info[] = {
284 .tunnel = MLX5_FLOW_LAYER_VXLAN,
285 .ptype = RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L4_UDP,
288 .tunnel = MLX5_FLOW_LAYER_VXLAN_GPE,
289 .ptype = RTE_PTYPE_TUNNEL_VXLAN_GPE | RTE_PTYPE_L4_UDP,
292 .tunnel = MLX5_FLOW_LAYER_GRE,
293 .ptype = RTE_PTYPE_TUNNEL_GRE,
296 .tunnel = MLX5_FLOW_LAYER_MPLS | MLX5_FLOW_LAYER_OUTER_L4_UDP,
297 .ptype = RTE_PTYPE_TUNNEL_MPLS_IN_GRE | RTE_PTYPE_L4_UDP,
300 .tunnel = MLX5_FLOW_LAYER_MPLS,
301 .ptype = RTE_PTYPE_TUNNEL_MPLS_IN_GRE,
306 * Discover the maximum number of priority available.
309 * Pointer to the Ethernet device structure.
312 * number of supported flow priority on success, a negative errno
313 * value otherwise and rte_errno is set.
316 mlx5_flow_discover_priorities(struct rte_eth_dev *dev)
319 struct ibv_flow_attr attr;
320 struct ibv_flow_spec_eth eth;
321 struct ibv_flow_spec_action_drop drop;
327 .type = IBV_FLOW_SPEC_ETH,
328 .size = sizeof(struct ibv_flow_spec_eth),
331 .size = sizeof(struct ibv_flow_spec_action_drop),
332 .type = IBV_FLOW_SPEC_ACTION_DROP,
335 struct ibv_flow *flow;
336 struct mlx5_hrxq *drop = mlx5_hrxq_drop_new(dev);
337 uint16_t vprio[] = { 8, 16 };
345 for (i = 0; i != RTE_DIM(vprio); i++) {
346 flow_attr.attr.priority = vprio[i] - 1;
347 flow = mlx5_glue->create_flow(drop->qp, &flow_attr.attr);
350 claim_zero(mlx5_glue->destroy_flow(flow));
355 priority = RTE_DIM(priority_map_3);
358 priority = RTE_DIM(priority_map_5);
363 "port %u verbs maximum priority: %d expected 8/16",
364 dev->data->port_id, vprio[i]);
367 mlx5_hrxq_drop_release(dev);
368 DRV_LOG(INFO, "port %u flow maximum priority: %d",
369 dev->data->port_id, priority);
374 * Adjust flow priority based on the highest layer and the request priority.
377 * Pointer to the Ethernet device structure.
378 * @param[in] priority
379 * The rule base priority.
380 * @param[in] subpriority
381 * The priority based on the items.
386 uint32_t mlx5_flow_adjust_priority(struct rte_eth_dev *dev, int32_t priority,
387 uint32_t subpriority)
390 struct priv *priv = dev->data->dev_private;
392 switch (priv->config.flow_prio) {
393 case RTE_DIM(priority_map_3):
394 res = priority_map_3[priority][subpriority];
396 case RTE_DIM(priority_map_5):
397 res = priority_map_5[priority][subpriority];
404 * Verify the @p item specifications (spec, last, mask) are compatible with the
408 * Item specification.
410 * @p item->mask or flow default bit-masks.
411 * @param[in] nic_mask
412 * Bit-masks covering supported fields by the NIC to compare with user mask.
414 * Bit-masks size in bytes.
416 * Pointer to error structure.
419 * 0 on success, a negative errno value otherwise and rte_errno is set.
422 mlx5_flow_item_acceptable(const struct rte_flow_item *item,
424 const uint8_t *nic_mask,
426 struct rte_flow_error *error)
431 for (i = 0; i < size; ++i)
432 if ((nic_mask[i] | mask[i]) != nic_mask[i])
433 return rte_flow_error_set(error, ENOTSUP,
434 RTE_FLOW_ERROR_TYPE_ITEM,
436 "mask enables non supported"
438 if (!item->spec && (item->mask || item->last))
439 return rte_flow_error_set(error, EINVAL,
440 RTE_FLOW_ERROR_TYPE_ITEM, item,
441 "mask/last without a spec is not"
443 if (item->spec && item->last) {
449 for (i = 0; i < size; ++i) {
450 spec[i] = ((const uint8_t *)item->spec)[i] & mask[i];
451 last[i] = ((const uint8_t *)item->last)[i] & mask[i];
453 ret = memcmp(spec, last, size);
455 return rte_flow_error_set(error, EINVAL,
456 RTE_FLOW_ERROR_TYPE_ITEM,
458 "range is not valid");
464 * Adjust the hash fields according to the @p flow information.
466 * @param[in] dev_flow.
467 * Pointer to the mlx5_flow.
469 * 1 when the hash field is for a tunnel item.
470 * @param[in] layer_types
472 * @param[in] hash_fields
476 * The hash fileds that should be used.
479 mlx5_flow_hashfields_adjust(struct mlx5_flow *dev_flow,
480 int tunnel __rte_unused, uint32_t layer_types,
481 uint64_t hash_fields)
483 struct rte_flow *flow = dev_flow->flow;
484 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
485 int rss_request_inner = flow->rss.level >= 2;
487 /* Check RSS hash level for tunnel. */
488 if (tunnel && rss_request_inner)
489 hash_fields |= IBV_RX_HASH_INNER;
490 else if (tunnel || rss_request_inner)
493 /* Check if requested layer matches RSS hash fields. */
494 if (!(flow->rss.types & layer_types))
500 * Lookup and set the ptype in the data Rx part. A single Ptype can be used,
501 * if several tunnel rules are used on this queue, the tunnel ptype will be
505 * Rx queue to update.
508 flow_rxq_tunnel_ptype_update(struct mlx5_rxq_ctrl *rxq_ctrl)
511 uint32_t tunnel_ptype = 0;
513 /* Look up for the ptype to use. */
514 for (i = 0; i != MLX5_FLOW_TUNNEL; ++i) {
515 if (!rxq_ctrl->flow_tunnels_n[i])
518 tunnel_ptype = tunnels_info[i].ptype;
524 rxq_ctrl->rxq.tunnel = tunnel_ptype;
528 * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) according to the devive
532 * Pointer to the Ethernet device structure.
533 * @param[in] dev_flow
534 * Pointer to device flow structure.
537 flow_drv_rxq_flags_set(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow)
539 struct priv *priv = dev->data->dev_private;
540 struct rte_flow *flow = dev_flow->flow;
541 const int mark = !!(flow->actions &
542 (MLX5_FLOW_ACTION_FLAG | MLX5_FLOW_ACTION_MARK));
543 const int tunnel = !!(dev_flow->layers & MLX5_FLOW_LAYER_TUNNEL);
546 for (i = 0; i != flow->rss.queue_num; ++i) {
547 int idx = (*flow->queue)[i];
548 struct mlx5_rxq_ctrl *rxq_ctrl =
549 container_of((*priv->rxqs)[idx],
550 struct mlx5_rxq_ctrl, rxq);
553 rxq_ctrl->rxq.mark = 1;
554 rxq_ctrl->flow_mark_n++;
559 /* Increase the counter matching the flow. */
560 for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
561 if ((tunnels_info[j].tunnel &
563 tunnels_info[j].tunnel) {
564 rxq_ctrl->flow_tunnels_n[j]++;
568 flow_rxq_tunnel_ptype_update(rxq_ctrl);
574 * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) for a flow
577 * Pointer to the Ethernet device structure.
579 * Pointer to flow structure.
582 flow_rxq_flags_set(struct rte_eth_dev *dev, struct rte_flow *flow)
584 struct mlx5_flow *dev_flow;
586 LIST_FOREACH(dev_flow, &flow->dev_flows, next)
587 flow_drv_rxq_flags_set(dev, dev_flow);
591 * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
592 * device flow if no other flow uses it with the same kind of request.
595 * Pointer to Ethernet device.
596 * @param[in] dev_flow
597 * Pointer to the device flow.
600 flow_drv_rxq_flags_trim(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow)
602 struct priv *priv = dev->data->dev_private;
603 struct rte_flow *flow = dev_flow->flow;
604 const int mark = !!(flow->actions &
605 (MLX5_FLOW_ACTION_FLAG | MLX5_FLOW_ACTION_MARK));
606 const int tunnel = !!(dev_flow->layers & MLX5_FLOW_LAYER_TUNNEL);
609 assert(dev->data->dev_started);
610 for (i = 0; i != flow->rss.queue_num; ++i) {
611 int idx = (*flow->queue)[i];
612 struct mlx5_rxq_ctrl *rxq_ctrl =
613 container_of((*priv->rxqs)[idx],
614 struct mlx5_rxq_ctrl, rxq);
617 rxq_ctrl->flow_mark_n--;
618 rxq_ctrl->rxq.mark = !!rxq_ctrl->flow_mark_n;
623 /* Decrease the counter matching the flow. */
624 for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
625 if ((tunnels_info[j].tunnel &
627 tunnels_info[j].tunnel) {
628 rxq_ctrl->flow_tunnels_n[j]--;
632 flow_rxq_tunnel_ptype_update(rxq_ctrl);
638 * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
639 * @p flow if no other flow uses it with the same kind of request.
642 * Pointer to Ethernet device.
644 * Pointer to the flow.
647 flow_rxq_flags_trim(struct rte_eth_dev *dev, struct rte_flow *flow)
649 struct mlx5_flow *dev_flow;
651 LIST_FOREACH(dev_flow, &flow->dev_flows, next)
652 flow_drv_rxq_flags_trim(dev, dev_flow);
656 * Clear the Mark/Flag and Tunnel ptype information in all Rx queues.
659 * Pointer to Ethernet device.
662 flow_rxq_flags_clear(struct rte_eth_dev *dev)
664 struct priv *priv = dev->data->dev_private;
667 for (i = 0; i != priv->rxqs_n; ++i) {
668 struct mlx5_rxq_ctrl *rxq_ctrl;
671 if (!(*priv->rxqs)[i])
673 rxq_ctrl = container_of((*priv->rxqs)[i],
674 struct mlx5_rxq_ctrl, rxq);
675 rxq_ctrl->flow_mark_n = 0;
676 rxq_ctrl->rxq.mark = 0;
677 for (j = 0; j != MLX5_FLOW_TUNNEL; ++j)
678 rxq_ctrl->flow_tunnels_n[j] = 0;
679 rxq_ctrl->rxq.tunnel = 0;
684 * Validate the flag action.
686 * @param[in] action_flags
687 * Bit-fields that holds the actions detected until now.
689 * Attributes of flow that includes this action.
691 * Pointer to error structure.
694 * 0 on success, a negative errno value otherwise and rte_errno is set.
697 mlx5_flow_validate_action_flag(uint64_t action_flags,
698 const struct rte_flow_attr *attr,
699 struct rte_flow_error *error)
702 if (action_flags & MLX5_FLOW_ACTION_DROP)
703 return rte_flow_error_set(error, EINVAL,
704 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
705 "can't drop and flag in same flow");
706 if (action_flags & MLX5_FLOW_ACTION_MARK)
707 return rte_flow_error_set(error, EINVAL,
708 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
709 "can't mark and flag in same flow");
710 if (action_flags & MLX5_FLOW_ACTION_FLAG)
711 return rte_flow_error_set(error, EINVAL,
712 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
714 " actions in same flow");
716 return rte_flow_error_set(error, ENOTSUP,
717 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
718 "flag action not supported for "
724 * Validate the mark action.
727 * Pointer to the queue action.
728 * @param[in] action_flags
729 * Bit-fields that holds the actions detected until now.
731 * Attributes of flow that includes this action.
733 * Pointer to error structure.
736 * 0 on success, a negative errno value otherwise and rte_errno is set.
739 mlx5_flow_validate_action_mark(const struct rte_flow_action *action,
740 uint64_t action_flags,
741 const struct rte_flow_attr *attr,
742 struct rte_flow_error *error)
744 const struct rte_flow_action_mark *mark = action->conf;
747 return rte_flow_error_set(error, EINVAL,
748 RTE_FLOW_ERROR_TYPE_ACTION,
750 "configuration cannot be null");
751 if (mark->id >= MLX5_FLOW_MARK_MAX)
752 return rte_flow_error_set(error, EINVAL,
753 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
755 "mark id must in 0 <= id < "
756 RTE_STR(MLX5_FLOW_MARK_MAX));
757 if (action_flags & MLX5_FLOW_ACTION_DROP)
758 return rte_flow_error_set(error, EINVAL,
759 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
760 "can't drop and mark in same flow");
761 if (action_flags & MLX5_FLOW_ACTION_FLAG)
762 return rte_flow_error_set(error, EINVAL,
763 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
764 "can't flag and mark in same flow");
765 if (action_flags & MLX5_FLOW_ACTION_MARK)
766 return rte_flow_error_set(error, EINVAL,
767 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
768 "can't have 2 mark actions in same"
771 return rte_flow_error_set(error, ENOTSUP,
772 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
773 "mark action not supported for "
779 * Validate the drop 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_ernno is set.
792 mlx5_flow_validate_action_drop(uint64_t action_flags,
793 const struct rte_flow_attr *attr,
794 struct rte_flow_error *error)
796 if (action_flags & MLX5_FLOW_ACTION_FLAG)
797 return rte_flow_error_set(error, EINVAL,
798 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
799 "can't drop and flag in same flow");
800 if (action_flags & MLX5_FLOW_ACTION_MARK)
801 return rte_flow_error_set(error, EINVAL,
802 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
803 "can't drop and mark in same flow");
804 if (action_flags & MLX5_FLOW_FATE_ACTIONS)
805 return rte_flow_error_set(error, EINVAL,
806 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
807 "can't have 2 fate actions in"
810 return rte_flow_error_set(error, ENOTSUP,
811 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
812 "drop action not supported for "
818 * Validate the queue action.
821 * Pointer to the queue action.
822 * @param[in] action_flags
823 * Bit-fields that holds the actions detected until now.
825 * Pointer to the Ethernet device structure.
827 * Attributes of flow that includes this action.
829 * Pointer to error structure.
832 * 0 on success, a negative errno value otherwise and rte_ernno is set.
835 mlx5_flow_validate_action_queue(const struct rte_flow_action *action,
836 uint64_t action_flags,
837 struct rte_eth_dev *dev,
838 const struct rte_flow_attr *attr,
839 struct rte_flow_error *error)
841 struct priv *priv = dev->data->dev_private;
842 const struct rte_flow_action_queue *queue = action->conf;
844 if (action_flags & MLX5_FLOW_FATE_ACTIONS)
845 return rte_flow_error_set(error, EINVAL,
846 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
847 "can't have 2 fate actions in"
849 if (queue->index >= priv->rxqs_n)
850 return rte_flow_error_set(error, EINVAL,
851 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
853 "queue index out of range");
854 if (!(*priv->rxqs)[queue->index])
855 return rte_flow_error_set(error, EINVAL,
856 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
858 "queue is not configured");
860 return rte_flow_error_set(error, ENOTSUP,
861 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
862 "queue action not supported for "
868 * Validate the rss action.
871 * Pointer to the queue action.
872 * @param[in] action_flags
873 * Bit-fields that holds the actions detected until now.
875 * Pointer to the Ethernet device structure.
877 * Attributes of flow that includes this action.
879 * Pointer to error structure.
882 * 0 on success, a negative errno value otherwise and rte_ernno is set.
885 mlx5_flow_validate_action_rss(const struct rte_flow_action *action,
886 uint64_t action_flags,
887 struct rte_eth_dev *dev,
888 const struct rte_flow_attr *attr,
889 struct rte_flow_error *error)
891 struct priv *priv = dev->data->dev_private;
892 const struct rte_flow_action_rss *rss = action->conf;
895 if (action_flags & MLX5_FLOW_FATE_ACTIONS)
896 return rte_flow_error_set(error, EINVAL,
897 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
898 "can't have 2 fate actions"
900 if (rss->func != RTE_ETH_HASH_FUNCTION_DEFAULT &&
901 rss->func != RTE_ETH_HASH_FUNCTION_TOEPLITZ)
902 return rte_flow_error_set(error, ENOTSUP,
903 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
905 "RSS hash function not supported");
906 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
911 return rte_flow_error_set(error, ENOTSUP,
912 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
914 "tunnel RSS is not supported");
915 if (rss->key_len < MLX5_RSS_HASH_KEY_LEN)
916 return rte_flow_error_set(error, ENOTSUP,
917 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
919 "RSS hash key too small");
920 if (rss->key_len > MLX5_RSS_HASH_KEY_LEN)
921 return rte_flow_error_set(error, ENOTSUP,
922 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
924 "RSS hash key too large");
925 if (rss->queue_num > priv->config.ind_table_max_size)
926 return rte_flow_error_set(error, ENOTSUP,
927 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
929 "number of queues too large");
930 if (rss->types & MLX5_RSS_HF_MASK)
931 return rte_flow_error_set(error, ENOTSUP,
932 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
934 "some RSS protocols are not"
936 for (i = 0; i != rss->queue_num; ++i) {
937 if (!(*priv->rxqs)[rss->queue[i]])
938 return rte_flow_error_set
939 (error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
940 &rss->queue[i], "queue is not configured");
943 return rte_flow_error_set(error, ENOTSUP,
944 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
945 "rss action not supported for "
951 * Validate the count action.
954 * Pointer to the Ethernet device structure.
956 * Attributes of flow that includes this action.
958 * Pointer to error structure.
961 * 0 on success, a negative errno value otherwise and rte_ernno is set.
964 mlx5_flow_validate_action_count(struct rte_eth_dev *dev __rte_unused,
965 const struct rte_flow_attr *attr,
966 struct rte_flow_error *error)
969 return rte_flow_error_set(error, ENOTSUP,
970 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
971 "count action not supported for "
977 * Verify the @p attributes will be correctly understood by the NIC and store
978 * them in the @p flow if everything is correct.
981 * Pointer to the Ethernet device structure.
982 * @param[in] attributes
983 * Pointer to flow attributes
985 * Pointer to error structure.
988 * 0 on success, a negative errno value otherwise and rte_errno is set.
991 mlx5_flow_validate_attributes(struct rte_eth_dev *dev,
992 const struct rte_flow_attr *attributes,
993 struct rte_flow_error *error)
995 struct priv *priv = dev->data->dev_private;
996 uint32_t priority_max = priv->config.flow_prio - 1;
998 if (attributes->group)
999 return rte_flow_error_set(error, ENOTSUP,
1000 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
1001 NULL, "groups is not supported");
1002 if (attributes->priority != MLX5_FLOW_PRIO_RSVD &&
1003 attributes->priority >= priority_max)
1004 return rte_flow_error_set(error, ENOTSUP,
1005 RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
1006 NULL, "priority out of range");
1007 if (attributes->egress)
1008 return rte_flow_error_set(error, ENOTSUP,
1009 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1010 "egress is not supported");
1011 if (attributes->transfer)
1012 return rte_flow_error_set(error, ENOTSUP,
1013 RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1014 NULL, "transfer is not supported");
1015 if (!attributes->ingress)
1016 return rte_flow_error_set(error, EINVAL,
1017 RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
1019 "ingress attribute is mandatory");
1024 * Validate Ethernet item.
1027 * Item specification.
1028 * @param[in] item_flags
1029 * Bit-fields that holds the items detected until now.
1031 * Pointer to error structure.
1034 * 0 on success, a negative errno value otherwise and rte_errno is set.
1037 mlx5_flow_validate_item_eth(const struct rte_flow_item *item,
1038 uint64_t item_flags,
1039 struct rte_flow_error *error)
1041 const struct rte_flow_item_eth *mask = item->mask;
1042 const struct rte_flow_item_eth nic_mask = {
1043 .dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1044 .src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1045 .type = RTE_BE16(0xffff),
1048 int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1050 if (item_flags & MLX5_FLOW_LAYER_OUTER_L2)
1051 return rte_flow_error_set(error, ENOTSUP,
1052 RTE_FLOW_ERROR_TYPE_ITEM, item,
1053 "3 levels of l2 are not supported");
1054 if ((item_flags & MLX5_FLOW_LAYER_INNER_L2) && !tunnel)
1055 return rte_flow_error_set(error, ENOTSUP,
1056 RTE_FLOW_ERROR_TYPE_ITEM, item,
1057 "2 L2 without tunnel are not supported");
1059 mask = &rte_flow_item_eth_mask;
1060 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1061 (const uint8_t *)&nic_mask,
1062 sizeof(struct rte_flow_item_eth),
1068 * Validate VLAN item.
1071 * Item specification.
1072 * @param[in] item_flags
1073 * Bit-fields that holds the items detected until now.
1075 * Pointer to error structure.
1078 * 0 on success, a negative errno value otherwise and rte_errno is set.
1081 mlx5_flow_validate_item_vlan(const struct rte_flow_item *item,
1083 struct rte_flow_error *error)
1085 const struct rte_flow_item_vlan *spec = item->spec;
1086 const struct rte_flow_item_vlan *mask = item->mask;
1087 const struct rte_flow_item_vlan nic_mask = {
1088 .tci = RTE_BE16(0x0fff),
1089 .inner_type = RTE_BE16(0xffff),
1091 uint16_t vlan_tag = 0;
1092 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1094 const uint32_t l34m = tunnel ? (MLX5_FLOW_LAYER_INNER_L3 |
1095 MLX5_FLOW_LAYER_INNER_L4) :
1096 (MLX5_FLOW_LAYER_OUTER_L3 |
1097 MLX5_FLOW_LAYER_OUTER_L4);
1098 const uint32_t vlanm = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
1099 MLX5_FLOW_LAYER_OUTER_VLAN;
1101 if (item_flags & vlanm)
1102 return rte_flow_error_set(error, EINVAL,
1103 RTE_FLOW_ERROR_TYPE_ITEM, item,
1104 "VLAN layer already configured");
1105 else if ((item_flags & l34m) != 0)
1106 return rte_flow_error_set(error, EINVAL,
1107 RTE_FLOW_ERROR_TYPE_ITEM, item,
1108 "L2 layer cannot follow L3/L4 layer");
1110 mask = &rte_flow_item_vlan_mask;
1111 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1112 (const uint8_t *)&nic_mask,
1113 sizeof(struct rte_flow_item_vlan),
1118 vlan_tag = spec->tci;
1119 vlan_tag &= mask->tci;
1122 * From verbs perspective an empty VLAN is equivalent
1123 * to a packet without VLAN layer.
1126 return rte_flow_error_set(error, EINVAL,
1127 RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
1129 "VLAN cannot be empty");
1134 * Validate IPV4 item.
1137 * Item specification.
1138 * @param[in] item_flags
1139 * Bit-fields that holds the items detected until now.
1141 * Pointer to error structure.
1144 * 0 on success, a negative errno value otherwise and rte_errno is set.
1147 mlx5_flow_validate_item_ipv4(const struct rte_flow_item *item,
1149 struct rte_flow_error *error)
1151 const struct rte_flow_item_ipv4 *mask = item->mask;
1152 const struct rte_flow_item_ipv4 nic_mask = {
1154 .src_addr = RTE_BE32(0xffffffff),
1155 .dst_addr = RTE_BE32(0xffffffff),
1156 .type_of_service = 0xff,
1157 .next_proto_id = 0xff,
1160 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1163 if (item_flags & (tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1164 MLX5_FLOW_LAYER_OUTER_L3))
1165 return rte_flow_error_set(error, ENOTSUP,
1166 RTE_FLOW_ERROR_TYPE_ITEM, item,
1167 "multiple L3 layers not supported");
1168 else if (item_flags & (tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1169 MLX5_FLOW_LAYER_OUTER_L4))
1170 return rte_flow_error_set(error, EINVAL,
1171 RTE_FLOW_ERROR_TYPE_ITEM, item,
1172 "L3 cannot follow an L4 layer.");
1174 mask = &rte_flow_item_ipv4_mask;
1175 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1176 (const uint8_t *)&nic_mask,
1177 sizeof(struct rte_flow_item_ipv4),
1185 * Validate IPV6 item.
1188 * Item specification.
1189 * @param[in] item_flags
1190 * Bit-fields that holds the items detected until now.
1192 * Pointer to error structure.
1195 * 0 on success, a negative errno value otherwise and rte_errno is set.
1198 mlx5_flow_validate_item_ipv6(const struct rte_flow_item *item,
1199 uint64_t item_flags,
1200 struct rte_flow_error *error)
1202 const struct rte_flow_item_ipv6 *mask = item->mask;
1203 const struct rte_flow_item_ipv6 nic_mask = {
1206 "\xff\xff\xff\xff\xff\xff\xff\xff"
1207 "\xff\xff\xff\xff\xff\xff\xff\xff",
1209 "\xff\xff\xff\xff\xff\xff\xff\xff"
1210 "\xff\xff\xff\xff\xff\xff\xff\xff",
1211 .vtc_flow = RTE_BE32(0xffffffff),
1216 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1219 if (item_flags & (tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1220 MLX5_FLOW_LAYER_OUTER_L3))
1221 return rte_flow_error_set(error, ENOTSUP,
1222 RTE_FLOW_ERROR_TYPE_ITEM, item,
1223 "multiple L3 layers not supported");
1224 else if (item_flags & (tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1225 MLX5_FLOW_LAYER_OUTER_L4))
1226 return rte_flow_error_set(error, EINVAL,
1227 RTE_FLOW_ERROR_TYPE_ITEM, item,
1228 "L3 cannot follow an L4 layer.");
1230 * IPv6 is not recognised by the NIC inside a GRE tunnel.
1231 * Such support has to be disabled as the rule will be
1232 * accepted. Issue reproduced with Mellanox OFED 4.3-3.0.2.1 and
1233 * Mellanox OFED 4.4-1.0.0.0.
1235 if (tunnel && item_flags & MLX5_FLOW_LAYER_GRE)
1236 return rte_flow_error_set(error, ENOTSUP,
1237 RTE_FLOW_ERROR_TYPE_ITEM, item,
1238 "IPv6 inside a GRE tunnel is"
1239 " not recognised.");
1241 mask = &rte_flow_item_ipv6_mask;
1242 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1243 (const uint8_t *)&nic_mask,
1244 sizeof(struct rte_flow_item_ipv6),
1252 * Validate UDP item.
1255 * Item specification.
1256 * @param[in] item_flags
1257 * Bit-fields that holds the items detected until now.
1258 * @param[in] target_protocol
1259 * The next protocol in the previous item.
1260 * @param[in] flow_mask
1261 * mlx5 flow-specific (TCF, DV, verbs, etc.) supported header fields mask.
1263 * Pointer to error structure.
1266 * 0 on success, a negative errno value otherwise and rte_errno is set.
1269 mlx5_flow_validate_item_udp(const struct rte_flow_item *item,
1270 uint64_t item_flags,
1271 uint8_t target_protocol,
1272 struct rte_flow_error *error)
1274 const struct rte_flow_item_udp *mask = item->mask;
1275 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1278 if (target_protocol != 0xff && target_protocol != IPPROTO_UDP)
1279 return rte_flow_error_set(error, EINVAL,
1280 RTE_FLOW_ERROR_TYPE_ITEM, item,
1281 "protocol filtering not compatible"
1283 if (!(item_flags & (tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1284 MLX5_FLOW_LAYER_OUTER_L3)))
1285 return rte_flow_error_set(error, EINVAL,
1286 RTE_FLOW_ERROR_TYPE_ITEM, item,
1287 "L3 is mandatory to filter on L4");
1288 if (item_flags & (tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1289 MLX5_FLOW_LAYER_OUTER_L4))
1290 return rte_flow_error_set(error, EINVAL,
1291 RTE_FLOW_ERROR_TYPE_ITEM, item,
1292 "L4 layer is already present");
1294 mask = &rte_flow_item_udp_mask;
1295 ret = mlx5_flow_item_acceptable
1296 (item, (const uint8_t *)mask,
1297 (const uint8_t *)&rte_flow_item_udp_mask,
1298 sizeof(struct rte_flow_item_udp), error);
1305 * Validate TCP item.
1308 * Item specification.
1309 * @param[in] item_flags
1310 * Bit-fields that holds the items detected until now.
1311 * @param[in] target_protocol
1312 * The next protocol in the previous item.
1314 * Pointer to error structure.
1317 * 0 on success, a negative errno value otherwise and rte_errno is set.
1320 mlx5_flow_validate_item_tcp(const struct rte_flow_item *item,
1321 uint64_t item_flags,
1322 uint8_t target_protocol,
1323 const struct rte_flow_item_tcp *flow_mask,
1324 struct rte_flow_error *error)
1326 const struct rte_flow_item_tcp *mask = item->mask;
1327 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1331 if (target_protocol != 0xff && target_protocol != IPPROTO_TCP)
1332 return rte_flow_error_set(error, EINVAL,
1333 RTE_FLOW_ERROR_TYPE_ITEM, item,
1334 "protocol filtering not compatible"
1336 if (!(item_flags & (tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1337 MLX5_FLOW_LAYER_OUTER_L3)))
1338 return rte_flow_error_set(error, EINVAL,
1339 RTE_FLOW_ERROR_TYPE_ITEM, item,
1340 "L3 is mandatory to filter on L4");
1341 if (item_flags & (tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1342 MLX5_FLOW_LAYER_OUTER_L4))
1343 return rte_flow_error_set(error, EINVAL,
1344 RTE_FLOW_ERROR_TYPE_ITEM, item,
1345 "L4 layer is already present");
1347 mask = &rte_flow_item_tcp_mask;
1348 ret = mlx5_flow_item_acceptable
1349 (item, (const uint8_t *)mask,
1350 (const uint8_t *)flow_mask,
1351 sizeof(struct rte_flow_item_tcp), error);
1358 * Validate VXLAN item.
1361 * Item specification.
1362 * @param[in] item_flags
1363 * Bit-fields that holds the items detected until now.
1364 * @param[in] target_protocol
1365 * The next protocol in the previous item.
1367 * Pointer to error structure.
1370 * 0 on success, a negative errno value otherwise and rte_errno is set.
1373 mlx5_flow_validate_item_vxlan(const struct rte_flow_item *item,
1374 uint64_t item_flags,
1375 struct rte_flow_error *error)
1377 const struct rte_flow_item_vxlan *spec = item->spec;
1378 const struct rte_flow_item_vxlan *mask = item->mask;
1383 } id = { .vlan_id = 0, };
1384 uint32_t vlan_id = 0;
1387 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1388 return rte_flow_error_set(error, ENOTSUP,
1389 RTE_FLOW_ERROR_TYPE_ITEM, item,
1390 "a tunnel is already present");
1392 * Verify only UDPv4 is present as defined in
1393 * https://tools.ietf.org/html/rfc7348
1395 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
1396 return rte_flow_error_set(error, EINVAL,
1397 RTE_FLOW_ERROR_TYPE_ITEM, item,
1398 "no outer UDP layer found");
1400 mask = &rte_flow_item_vxlan_mask;
1401 ret = mlx5_flow_item_acceptable
1402 (item, (const uint8_t *)mask,
1403 (const uint8_t *)&rte_flow_item_vxlan_mask,
1404 sizeof(struct rte_flow_item_vxlan),
1409 memcpy(&id.vni[1], spec->vni, 3);
1410 vlan_id = id.vlan_id;
1411 memcpy(&id.vni[1], mask->vni, 3);
1412 vlan_id &= id.vlan_id;
1415 * Tunnel id 0 is equivalent as not adding a VXLAN layer, if
1416 * only this layer is defined in the Verbs specification it is
1417 * interpreted as wildcard and all packets will match this
1418 * rule, if it follows a full stack layer (ex: eth / ipv4 /
1419 * udp), all packets matching the layers before will also
1420 * match this rule. To avoid such situation, VNI 0 is
1421 * currently refused.
1424 return rte_flow_error_set(error, ENOTSUP,
1425 RTE_FLOW_ERROR_TYPE_ITEM, item,
1426 "VXLAN vni cannot be 0");
1427 if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
1428 return rte_flow_error_set(error, ENOTSUP,
1429 RTE_FLOW_ERROR_TYPE_ITEM, item,
1430 "VXLAN tunnel must be fully defined");
1435 * Validate VXLAN_GPE item.
1438 * Item specification.
1439 * @param[in] item_flags
1440 * Bit-fields that holds the items detected until now.
1442 * Pointer to the private data structure.
1443 * @param[in] target_protocol
1444 * The next protocol in the previous item.
1446 * Pointer to error structure.
1449 * 0 on success, a negative errno value otherwise and rte_errno is set.
1452 mlx5_flow_validate_item_vxlan_gpe(const struct rte_flow_item *item,
1453 uint64_t item_flags,
1454 struct rte_eth_dev *dev,
1455 struct rte_flow_error *error)
1457 struct priv *priv = dev->data->dev_private;
1458 const struct rte_flow_item_vxlan_gpe *spec = item->spec;
1459 const struct rte_flow_item_vxlan_gpe *mask = item->mask;
1464 } id = { .vlan_id = 0, };
1465 uint32_t vlan_id = 0;
1467 if (!priv->config.l3_vxlan_en)
1468 return rte_flow_error_set(error, ENOTSUP,
1469 RTE_FLOW_ERROR_TYPE_ITEM, item,
1470 "L3 VXLAN is not enabled by device"
1471 " parameter and/or not configured in"
1473 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1474 return rte_flow_error_set(error, ENOTSUP,
1475 RTE_FLOW_ERROR_TYPE_ITEM, item,
1476 "a tunnel is already present");
1478 * Verify only UDPv4 is present as defined in
1479 * https://tools.ietf.org/html/rfc7348
1481 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
1482 return rte_flow_error_set(error, EINVAL,
1483 RTE_FLOW_ERROR_TYPE_ITEM, item,
1484 "no outer UDP layer found");
1486 mask = &rte_flow_item_vxlan_gpe_mask;
1487 ret = mlx5_flow_item_acceptable
1488 (item, (const uint8_t *)mask,
1489 (const uint8_t *)&rte_flow_item_vxlan_gpe_mask,
1490 sizeof(struct rte_flow_item_vxlan_gpe),
1496 return rte_flow_error_set(error, ENOTSUP,
1497 RTE_FLOW_ERROR_TYPE_ITEM,
1499 "VxLAN-GPE protocol"
1501 memcpy(&id.vni[1], spec->vni, 3);
1502 vlan_id = id.vlan_id;
1503 memcpy(&id.vni[1], mask->vni, 3);
1504 vlan_id &= id.vlan_id;
1507 * Tunnel id 0 is equivalent as not adding a VXLAN layer, if only this
1508 * layer is defined in the Verbs specification it is interpreted as
1509 * wildcard and all packets will match this rule, if it follows a full
1510 * stack layer (ex: eth / ipv4 / udp), all packets matching the layers
1511 * before will also match this rule. To avoid such situation, VNI 0
1512 * is currently refused.
1515 return rte_flow_error_set(error, ENOTSUP,
1516 RTE_FLOW_ERROR_TYPE_ITEM, item,
1517 "VXLAN-GPE vni cannot be 0");
1518 if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
1519 return rte_flow_error_set(error, ENOTSUP,
1520 RTE_FLOW_ERROR_TYPE_ITEM, item,
1521 "VXLAN-GPE tunnel must be fully"
1527 * Validate GRE item.
1530 * Item specification.
1531 * @param[in] item_flags
1532 * Bit flags to mark detected items.
1533 * @param[in] target_protocol
1534 * The next protocol in the previous item.
1536 * Pointer to error structure.
1539 * 0 on success, a negative errno value otherwise and rte_errno is set.
1542 mlx5_flow_validate_item_gre(const struct rte_flow_item *item,
1543 uint64_t item_flags,
1544 uint8_t target_protocol,
1545 struct rte_flow_error *error)
1547 const struct rte_flow_item_gre *spec __rte_unused = item->spec;
1548 const struct rte_flow_item_gre *mask = item->mask;
1551 if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
1552 return rte_flow_error_set(error, EINVAL,
1553 RTE_FLOW_ERROR_TYPE_ITEM, item,
1554 "protocol filtering not compatible"
1555 " with this GRE layer");
1556 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1557 return rte_flow_error_set(error, ENOTSUP,
1558 RTE_FLOW_ERROR_TYPE_ITEM, item,
1559 "a tunnel is already present");
1560 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
1561 return rte_flow_error_set(error, ENOTSUP,
1562 RTE_FLOW_ERROR_TYPE_ITEM, item,
1563 "L3 Layer is missing");
1565 mask = &rte_flow_item_gre_mask;
1566 ret = mlx5_flow_item_acceptable
1567 (item, (const uint8_t *)mask,
1568 (const uint8_t *)&rte_flow_item_gre_mask,
1569 sizeof(struct rte_flow_item_gre), error);
1572 #ifndef HAVE_IBV_DEVICE_MPLS_SUPPORT
1573 if (spec && (spec->protocol & mask->protocol))
1574 return rte_flow_error_set(error, ENOTSUP,
1575 RTE_FLOW_ERROR_TYPE_ITEM, item,
1576 "without MPLS support the"
1577 " specification cannot be used for"
1584 * Validate MPLS item.
1587 * Item specification.
1588 * @param[in] item_flags
1589 * Bit-fields that holds the items detected until now.
1590 * @param[in] target_protocol
1591 * The next protocol in the previous item.
1593 * Pointer to error structure.
1596 * 0 on success, a negative errno value otherwise and rte_errno is set.
1599 mlx5_flow_validate_item_mpls(const struct rte_flow_item *item __rte_unused,
1600 uint64_t item_flags __rte_unused,
1601 uint8_t target_protocol __rte_unused,
1602 struct rte_flow_error *error)
1604 #ifdef HAVE_IBV_DEVICE_MPLS_SUPPORT
1605 const struct rte_flow_item_mpls *mask = item->mask;
1608 if (target_protocol != 0xff && target_protocol != IPPROTO_MPLS)
1609 return rte_flow_error_set(error, EINVAL,
1610 RTE_FLOW_ERROR_TYPE_ITEM, item,
1611 "protocol filtering not compatible"
1612 " with MPLS layer");
1613 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1614 return rte_flow_error_set(error, ENOTSUP,
1615 RTE_FLOW_ERROR_TYPE_ITEM, item,
1616 "a tunnel is already"
1619 mask = &rte_flow_item_mpls_mask;
1620 ret = mlx5_flow_item_acceptable
1621 (item, (const uint8_t *)mask,
1622 (const uint8_t *)&rte_flow_item_mpls_mask,
1623 sizeof(struct rte_flow_item_mpls), error);
1628 return rte_flow_error_set(error, ENOTSUP,
1629 RTE_FLOW_ERROR_TYPE_ITEM, item,
1630 "MPLS is not supported by Verbs, please"
1635 flow_null_validate(struct rte_eth_dev *dev __rte_unused,
1636 const struct rte_flow_attr *attr __rte_unused,
1637 const struct rte_flow_item items[] __rte_unused,
1638 const struct rte_flow_action actions[] __rte_unused,
1639 struct rte_flow_error *error __rte_unused)
1641 rte_errno = ENOTSUP;
1645 static struct mlx5_flow *
1646 flow_null_prepare(const struct rte_flow_attr *attr __rte_unused,
1647 const struct rte_flow_item items[] __rte_unused,
1648 const struct rte_flow_action actions[] __rte_unused,
1649 uint64_t *item_flags __rte_unused,
1650 uint64_t *action_flags __rte_unused,
1651 struct rte_flow_error *error __rte_unused)
1653 rte_errno = ENOTSUP;
1658 flow_null_translate(struct rte_eth_dev *dev __rte_unused,
1659 struct mlx5_flow *dev_flow __rte_unused,
1660 const struct rte_flow_attr *attr __rte_unused,
1661 const struct rte_flow_item items[] __rte_unused,
1662 const struct rte_flow_action actions[] __rte_unused,
1663 struct rte_flow_error *error __rte_unused)
1665 rte_errno = ENOTSUP;
1670 flow_null_apply(struct rte_eth_dev *dev __rte_unused,
1671 struct rte_flow *flow __rte_unused,
1672 struct rte_flow_error *error __rte_unused)
1674 rte_errno = ENOTSUP;
1679 flow_null_remove(struct rte_eth_dev *dev __rte_unused,
1680 struct rte_flow *flow __rte_unused)
1685 flow_null_destroy(struct rte_eth_dev *dev __rte_unused,
1686 struct rte_flow *flow __rte_unused)
1691 flow_null_query(struct rte_eth_dev *dev __rte_unused,
1692 struct rte_flow *flow __rte_unused,
1693 const struct rte_flow_action *actions __rte_unused,
1694 void *data __rte_unused,
1695 struct rte_flow_error *error __rte_unused)
1697 rte_errno = ENOTSUP;
1701 /* Void driver to protect from null pointer reference. */
1702 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops = {
1703 .validate = flow_null_validate,
1704 .prepare = flow_null_prepare,
1705 .translate = flow_null_translate,
1706 .apply = flow_null_apply,
1707 .remove = flow_null_remove,
1708 .destroy = flow_null_destroy,
1709 .query = flow_null_query,
1713 * Select flow driver type according to flow attributes and device
1717 * Pointer to the dev structure.
1719 * Pointer to the flow attributes.
1722 * flow driver type, MLX5_FLOW_TYPE_MAX otherwise.
1724 static enum mlx5_flow_drv_type
1725 flow_get_drv_type(struct rte_eth_dev *dev, const struct rte_flow_attr *attr)
1727 struct priv *priv = dev->data->dev_private;
1728 enum mlx5_flow_drv_type type = MLX5_FLOW_TYPE_MAX;
1731 type = MLX5_FLOW_TYPE_TCF;
1733 type = priv->config.dv_flow_en ? MLX5_FLOW_TYPE_DV :
1734 MLX5_FLOW_TYPE_VERBS;
1738 #define flow_get_drv_ops(type) flow_drv_ops[type]
1741 * Flow driver validation API. This abstracts calling driver specific functions.
1742 * The type of flow driver is determined according to flow attributes.
1745 * Pointer to the dev structure.
1747 * Pointer to the flow attributes.
1749 * Pointer to the list of items.
1750 * @param[in] actions
1751 * Pointer to the list of actions.
1753 * Pointer to the error structure.
1756 * 0 on success, a negative errno value otherwise and rte_ernno is set.
1759 flow_drv_validate(struct rte_eth_dev *dev,
1760 const struct rte_flow_attr *attr,
1761 const struct rte_flow_item items[],
1762 const struct rte_flow_action actions[],
1763 struct rte_flow_error *error)
1765 const struct mlx5_flow_driver_ops *fops;
1766 enum mlx5_flow_drv_type type = flow_get_drv_type(dev, attr);
1768 fops = flow_get_drv_ops(type);
1769 return fops->validate(dev, attr, items, actions, error);
1773 * Flow driver preparation API. This abstracts calling driver specific
1774 * functions. Parent flow (rte_flow) should have driver type (drv_type). It
1775 * calculates the size of memory required for device flow, allocates the memory,
1776 * initializes the device flow and returns the pointer.
1779 * Pointer to the flow attributes.
1781 * Pointer to the list of items.
1782 * @param[in] actions
1783 * Pointer to the list of actions.
1784 * @param[out] item_flags
1785 * Pointer to bit mask of all items detected.
1786 * @param[out] action_flags
1787 * Pointer to bit mask of all actions detected.
1789 * Pointer to the error structure.
1792 * Pointer to device flow on success, otherwise NULL and rte_ernno is set.
1794 static inline struct mlx5_flow *
1795 flow_drv_prepare(struct rte_flow *flow,
1796 const struct rte_flow_attr *attr,
1797 const struct rte_flow_item items[],
1798 const struct rte_flow_action actions[],
1799 uint64_t *item_flags,
1800 uint64_t *action_flags,
1801 struct rte_flow_error *error)
1803 const struct mlx5_flow_driver_ops *fops;
1804 enum mlx5_flow_drv_type type = flow->drv_type;
1806 assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
1807 fops = flow_get_drv_ops(type);
1808 return fops->prepare(attr, items, actions, item_flags, action_flags,
1813 * Flow driver translation API. This abstracts calling driver specific
1814 * functions. Parent flow (rte_flow) should have driver type (drv_type). It
1815 * translates a generic flow into a driver flow. flow_drv_prepare() must
1820 * Pointer to the rte dev structure.
1821 * @param[in, out] dev_flow
1822 * Pointer to the mlx5 flow.
1824 * Pointer to the flow attributes.
1826 * Pointer to the list of items.
1827 * @param[in] actions
1828 * Pointer to the list of actions.
1830 * Pointer to the error structure.
1833 * 0 on success, a negative errno value otherwise and rte_ernno is set.
1836 flow_drv_translate(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow,
1837 const struct rte_flow_attr *attr,
1838 const struct rte_flow_item items[],
1839 const struct rte_flow_action actions[],
1840 struct rte_flow_error *error)
1842 const struct mlx5_flow_driver_ops *fops;
1843 enum mlx5_flow_drv_type type = dev_flow->flow->drv_type;
1845 assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
1846 fops = flow_get_drv_ops(type);
1847 return fops->translate(dev, dev_flow, attr, items, actions, error);
1851 * Flow driver apply API. This abstracts calling driver specific functions.
1852 * Parent flow (rte_flow) should have driver type (drv_type). It applies
1853 * translated driver flows on to device. flow_drv_translate() must precede.
1856 * Pointer to Ethernet device structure.
1857 * @param[in, out] flow
1858 * Pointer to flow structure.
1860 * Pointer to error structure.
1863 * 0 on success, a negative errno value otherwise and rte_errno is set.
1866 flow_drv_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
1867 struct rte_flow_error *error)
1869 const struct mlx5_flow_driver_ops *fops;
1870 enum mlx5_flow_drv_type type = flow->drv_type;
1872 assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
1873 fops = flow_get_drv_ops(type);
1874 return fops->apply(dev, flow, error);
1878 * Flow driver remove API. This abstracts calling driver specific functions.
1879 * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
1880 * on device. All the resources of the flow should be freed by calling
1881 * flow_dv_destroy().
1884 * Pointer to Ethernet device.
1885 * @param[in, out] flow
1886 * Pointer to flow structure.
1889 flow_drv_remove(struct rte_eth_dev *dev, struct rte_flow *flow)
1891 const struct mlx5_flow_driver_ops *fops;
1892 enum mlx5_flow_drv_type type = flow->drv_type;
1894 assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
1895 fops = flow_get_drv_ops(type);
1896 fops->remove(dev, flow);
1900 * Flow driver destroy API. This abstracts calling driver specific functions.
1901 * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
1902 * on device and releases resources of the flow.
1905 * Pointer to Ethernet device.
1906 * @param[in, out] flow
1907 * Pointer to flow structure.
1910 flow_drv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
1912 const struct mlx5_flow_driver_ops *fops;
1913 enum mlx5_flow_drv_type type = flow->drv_type;
1915 assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
1916 fops = flow_get_drv_ops(type);
1917 fops->destroy(dev, flow);
1921 * Validate a flow supported by the NIC.
1923 * @see rte_flow_validate()
1927 mlx5_flow_validate(struct rte_eth_dev *dev,
1928 const struct rte_flow_attr *attr,
1929 const struct rte_flow_item items[],
1930 const struct rte_flow_action actions[],
1931 struct rte_flow_error *error)
1935 ret = flow_drv_validate(dev, attr, items, actions, error);
1942 * Get RSS action from the action list.
1944 * @param[in] actions
1945 * Pointer to the list of actions.
1948 * Pointer to the RSS action if exist, else return NULL.
1950 static const struct rte_flow_action_rss*
1951 flow_get_rss_action(const struct rte_flow_action actions[])
1953 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
1954 switch (actions->type) {
1955 case RTE_FLOW_ACTION_TYPE_RSS:
1956 return (const struct rte_flow_action_rss *)
1966 find_graph_root(const struct rte_flow_item pattern[], uint32_t rss_level)
1968 const struct rte_flow_item *item;
1969 unsigned int has_vlan = 0;
1971 for (item = pattern; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
1972 if (item->type == RTE_FLOW_ITEM_TYPE_VLAN) {
1978 return rss_level < 2 ? MLX5_EXPANSION_ROOT_ETH_VLAN :
1979 MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN;
1980 return rss_level < 2 ? MLX5_EXPANSION_ROOT :
1981 MLX5_EXPANSION_ROOT_OUTER;
1985 * Create a flow and add it to @p list.
1988 * Pointer to Ethernet device.
1990 * Pointer to a TAILQ flow list.
1992 * Flow rule attributes.
1994 * Pattern specification (list terminated by the END pattern item).
1995 * @param[in] actions
1996 * Associated actions (list terminated by the END action).
1998 * Perform verbose error reporting if not NULL.
2001 * A flow on success, NULL otherwise and rte_errno is set.
2003 static struct rte_flow *
2004 flow_list_create(struct rte_eth_dev *dev, struct mlx5_flows *list,
2005 const struct rte_flow_attr *attr,
2006 const struct rte_flow_item items[],
2007 const struct rte_flow_action actions[],
2008 struct rte_flow_error *error)
2010 struct rte_flow *flow = NULL;
2011 struct mlx5_flow *dev_flow;
2012 uint64_t action_flags = 0;
2013 uint64_t item_flags = 0;
2014 const struct rte_flow_action_rss *rss;
2016 struct rte_flow_expand_rss buf;
2017 uint8_t buffer[2048];
2019 struct rte_flow_expand_rss *buf = &expand_buffer.buf;
2024 ret = flow_drv_validate(dev, attr, items, actions, error);
2027 flow_size = sizeof(struct rte_flow);
2028 rss = flow_get_rss_action(actions);
2030 flow_size += RTE_ALIGN_CEIL(rss->queue_num * sizeof(uint16_t),
2033 flow_size += RTE_ALIGN_CEIL(sizeof(uint16_t), sizeof(void *));
2034 flow = rte_calloc(__func__, 1, flow_size, 0);
2035 flow->drv_type = flow_get_drv_type(dev, attr);
2036 assert(flow->drv_type > MLX5_FLOW_TYPE_MIN &&
2037 flow->drv_type < MLX5_FLOW_TYPE_MAX);
2038 flow->queue = (void *)(flow + 1);
2039 LIST_INIT(&flow->dev_flows);
2040 if (rss && rss->types) {
2041 unsigned int graph_root;
2043 graph_root = find_graph_root(items, rss->level);
2044 ret = rte_flow_expand_rss(buf, sizeof(expand_buffer.buffer),
2046 mlx5_support_expansion,
2049 (unsigned int)ret < sizeof(expand_buffer.buffer));
2052 buf->entry[0].pattern = (void *)(uintptr_t)items;
2054 for (i = 0; i < buf->entries; ++i) {
2055 dev_flow = flow_drv_prepare(flow, attr, buf->entry[i].pattern,
2056 actions, &item_flags, &action_flags,
2060 dev_flow->flow = flow;
2061 dev_flow->layers = item_flags;
2062 /* Store actions once as expanded flows have same actions. */
2064 flow->actions = action_flags;
2065 assert(flow->actions == action_flags);
2066 LIST_INSERT_HEAD(&flow->dev_flows, dev_flow, next);
2067 ret = flow_drv_translate(dev, dev_flow, attr,
2068 buf->entry[i].pattern,
2073 if (dev->data->dev_started) {
2074 ret = flow_drv_apply(dev, flow, error);
2078 TAILQ_INSERT_TAIL(list, flow, next);
2079 flow_rxq_flags_set(dev, flow);
2082 ret = rte_errno; /* Save rte_errno before cleanup. */
2084 flow_drv_destroy(dev, flow);
2086 rte_errno = ret; /* Restore rte_errno. */
2093 * @see rte_flow_create()
2097 mlx5_flow_create(struct rte_eth_dev *dev,
2098 const struct rte_flow_attr *attr,
2099 const struct rte_flow_item items[],
2100 const struct rte_flow_action actions[],
2101 struct rte_flow_error *error)
2103 return flow_list_create(dev,
2104 &((struct priv *)dev->data->dev_private)->flows,
2105 attr, items, actions, error);
2109 * Destroy a flow in a list.
2112 * Pointer to Ethernet device.
2114 * Pointer to a TAILQ flow list.
2119 flow_list_destroy(struct rte_eth_dev *dev, struct mlx5_flows *list,
2120 struct rte_flow *flow)
2122 flow_drv_destroy(dev, flow);
2123 TAILQ_REMOVE(list, flow, next);
2125 * Update RX queue flags only if port is started, otherwise it is
2128 if (dev->data->dev_started)
2129 flow_rxq_flags_trim(dev, flow);
2134 * Destroy all flows.
2137 * Pointer to Ethernet device.
2139 * Pointer to a TAILQ flow list.
2142 mlx5_flow_list_flush(struct rte_eth_dev *dev, struct mlx5_flows *list)
2144 while (!TAILQ_EMPTY(list)) {
2145 struct rte_flow *flow;
2147 flow = TAILQ_FIRST(list);
2148 flow_list_destroy(dev, list, flow);
2156 * Pointer to Ethernet device.
2158 * Pointer to a TAILQ flow list.
2161 mlx5_flow_stop(struct rte_eth_dev *dev, struct mlx5_flows *list)
2163 struct rte_flow *flow;
2165 TAILQ_FOREACH_REVERSE(flow, list, mlx5_flows, next)
2166 flow_drv_remove(dev, flow);
2167 flow_rxq_flags_clear(dev);
2174 * Pointer to Ethernet device.
2176 * Pointer to a TAILQ flow list.
2179 * 0 on success, a negative errno value otherwise and rte_errno is set.
2182 mlx5_flow_start(struct rte_eth_dev *dev, struct mlx5_flows *list)
2184 struct rte_flow *flow;
2185 struct rte_flow_error error;
2188 TAILQ_FOREACH(flow, list, next) {
2189 ret = flow_drv_apply(dev, flow, &error);
2192 flow_rxq_flags_set(dev, flow);
2196 ret = rte_errno; /* Save rte_errno before cleanup. */
2197 mlx5_flow_stop(dev, list);
2198 rte_errno = ret; /* Restore rte_errno. */
2203 * Verify the flow list is empty
2206 * Pointer to Ethernet device.
2208 * @return the number of flows not released.
2211 mlx5_flow_verify(struct rte_eth_dev *dev)
2213 struct priv *priv = dev->data->dev_private;
2214 struct rte_flow *flow;
2217 TAILQ_FOREACH(flow, &priv->flows, next) {
2218 DRV_LOG(DEBUG, "port %u flow %p still referenced",
2219 dev->data->port_id, (void *)flow);
2226 * Enable a control flow configured from the control plane.
2229 * Pointer to Ethernet device.
2231 * An Ethernet flow spec to apply.
2233 * An Ethernet flow mask to apply.
2235 * A VLAN flow spec to apply.
2237 * A VLAN flow mask to apply.
2240 * 0 on success, a negative errno value otherwise and rte_errno is set.
2243 mlx5_ctrl_flow_vlan(struct rte_eth_dev *dev,
2244 struct rte_flow_item_eth *eth_spec,
2245 struct rte_flow_item_eth *eth_mask,
2246 struct rte_flow_item_vlan *vlan_spec,
2247 struct rte_flow_item_vlan *vlan_mask)
2249 struct priv *priv = dev->data->dev_private;
2250 const struct rte_flow_attr attr = {
2252 .priority = MLX5_FLOW_PRIO_RSVD,
2254 struct rte_flow_item items[] = {
2256 .type = RTE_FLOW_ITEM_TYPE_ETH,
2262 .type = (vlan_spec) ? RTE_FLOW_ITEM_TYPE_VLAN :
2263 RTE_FLOW_ITEM_TYPE_END,
2269 .type = RTE_FLOW_ITEM_TYPE_END,
2272 uint16_t queue[priv->reta_idx_n];
2273 struct rte_flow_action_rss action_rss = {
2274 .func = RTE_ETH_HASH_FUNCTION_DEFAULT,
2276 .types = priv->rss_conf.rss_hf,
2277 .key_len = priv->rss_conf.rss_key_len,
2278 .queue_num = priv->reta_idx_n,
2279 .key = priv->rss_conf.rss_key,
2282 struct rte_flow_action actions[] = {
2284 .type = RTE_FLOW_ACTION_TYPE_RSS,
2285 .conf = &action_rss,
2288 .type = RTE_FLOW_ACTION_TYPE_END,
2291 struct rte_flow *flow;
2292 struct rte_flow_error error;
2295 if (!priv->reta_idx_n) {
2299 for (i = 0; i != priv->reta_idx_n; ++i)
2300 queue[i] = (*priv->reta_idx)[i];
2301 flow = flow_list_create(dev, &priv->ctrl_flows,
2302 &attr, items, actions, &error);
2309 * Enable a flow control configured from the control plane.
2312 * Pointer to Ethernet device.
2314 * An Ethernet flow spec to apply.
2316 * An Ethernet flow mask to apply.
2319 * 0 on success, a negative errno value otherwise and rte_errno is set.
2322 mlx5_ctrl_flow(struct rte_eth_dev *dev,
2323 struct rte_flow_item_eth *eth_spec,
2324 struct rte_flow_item_eth *eth_mask)
2326 return mlx5_ctrl_flow_vlan(dev, eth_spec, eth_mask, NULL, NULL);
2332 * @see rte_flow_destroy()
2336 mlx5_flow_destroy(struct rte_eth_dev *dev,
2337 struct rte_flow *flow,
2338 struct rte_flow_error *error __rte_unused)
2340 struct priv *priv = dev->data->dev_private;
2342 flow_list_destroy(dev, &priv->flows, flow);
2347 * Destroy all flows.
2349 * @see rte_flow_flush()
2353 mlx5_flow_flush(struct rte_eth_dev *dev,
2354 struct rte_flow_error *error __rte_unused)
2356 struct priv *priv = dev->data->dev_private;
2358 mlx5_flow_list_flush(dev, &priv->flows);
2365 * @see rte_flow_isolate()
2369 mlx5_flow_isolate(struct rte_eth_dev *dev,
2371 struct rte_flow_error *error)
2373 struct priv *priv = dev->data->dev_private;
2375 if (dev->data->dev_started) {
2376 rte_flow_error_set(error, EBUSY,
2377 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
2379 "port must be stopped first");
2382 priv->isolated = !!enable;
2384 dev->dev_ops = &mlx5_dev_ops_isolate;
2386 dev->dev_ops = &mlx5_dev_ops;
2393 * @see rte_flow_query()
2397 flow_drv_query(struct rte_eth_dev *dev,
2398 struct rte_flow *flow,
2399 const struct rte_flow_action *actions,
2401 struct rte_flow_error *error)
2403 const struct mlx5_flow_driver_ops *fops;
2404 enum mlx5_flow_drv_type ftype = flow->drv_type;
2406 assert(ftype > MLX5_FLOW_TYPE_MIN && ftype < MLX5_FLOW_TYPE_MAX);
2407 fops = flow_get_drv_ops(ftype);
2409 return fops->query(dev, flow, actions, data, error);
2415 * @see rte_flow_query()
2419 mlx5_flow_query(struct rte_eth_dev *dev,
2420 struct rte_flow *flow,
2421 const struct rte_flow_action *actions,
2423 struct rte_flow_error *error)
2427 ret = flow_drv_query(dev, flow, actions, data, error);
2434 * Convert a flow director filter to a generic flow.
2437 * Pointer to Ethernet device.
2438 * @param fdir_filter
2439 * Flow director filter to add.
2441 * Generic flow parameters structure.
2444 * 0 on success, a negative errno value otherwise and rte_errno is set.
2447 mlx5_fdir_filter_convert(struct rte_eth_dev *dev,
2448 const struct rte_eth_fdir_filter *fdir_filter,
2449 struct mlx5_fdir *attributes)
2451 struct priv *priv = dev->data->dev_private;
2452 const struct rte_eth_fdir_input *input = &fdir_filter->input;
2453 const struct rte_eth_fdir_masks *mask =
2454 &dev->data->dev_conf.fdir_conf.mask;
2456 /* Validate queue number. */
2457 if (fdir_filter->action.rx_queue >= priv->rxqs_n) {
2458 DRV_LOG(ERR, "port %u invalid queue number %d",
2459 dev->data->port_id, fdir_filter->action.rx_queue);
2463 attributes->attr.ingress = 1;
2464 attributes->items[0] = (struct rte_flow_item) {
2465 .type = RTE_FLOW_ITEM_TYPE_ETH,
2466 .spec = &attributes->l2,
2467 .mask = &attributes->l2_mask,
2469 switch (fdir_filter->action.behavior) {
2470 case RTE_ETH_FDIR_ACCEPT:
2471 attributes->actions[0] = (struct rte_flow_action){
2472 .type = RTE_FLOW_ACTION_TYPE_QUEUE,
2473 .conf = &attributes->queue,
2476 case RTE_ETH_FDIR_REJECT:
2477 attributes->actions[0] = (struct rte_flow_action){
2478 .type = RTE_FLOW_ACTION_TYPE_DROP,
2482 DRV_LOG(ERR, "port %u invalid behavior %d",
2484 fdir_filter->action.behavior);
2485 rte_errno = ENOTSUP;
2488 attributes->queue.index = fdir_filter->action.rx_queue;
2490 switch (fdir_filter->input.flow_type) {
2491 case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
2492 case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
2493 case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
2494 attributes->l3.ipv4.hdr = (struct ipv4_hdr){
2495 .src_addr = input->flow.ip4_flow.src_ip,
2496 .dst_addr = input->flow.ip4_flow.dst_ip,
2497 .time_to_live = input->flow.ip4_flow.ttl,
2498 .type_of_service = input->flow.ip4_flow.tos,
2500 attributes->l3_mask.ipv4.hdr = (struct ipv4_hdr){
2501 .src_addr = mask->ipv4_mask.src_ip,
2502 .dst_addr = mask->ipv4_mask.dst_ip,
2503 .time_to_live = mask->ipv4_mask.ttl,
2504 .type_of_service = mask->ipv4_mask.tos,
2505 .next_proto_id = mask->ipv4_mask.proto,
2507 attributes->items[1] = (struct rte_flow_item){
2508 .type = RTE_FLOW_ITEM_TYPE_IPV4,
2509 .spec = &attributes->l3,
2510 .mask = &attributes->l3_mask,
2513 case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
2514 case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
2515 case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
2516 attributes->l3.ipv6.hdr = (struct ipv6_hdr){
2517 .hop_limits = input->flow.ipv6_flow.hop_limits,
2518 .proto = input->flow.ipv6_flow.proto,
2521 memcpy(attributes->l3.ipv6.hdr.src_addr,
2522 input->flow.ipv6_flow.src_ip,
2523 RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
2524 memcpy(attributes->l3.ipv6.hdr.dst_addr,
2525 input->flow.ipv6_flow.dst_ip,
2526 RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
2527 memcpy(attributes->l3_mask.ipv6.hdr.src_addr,
2528 mask->ipv6_mask.src_ip,
2529 RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
2530 memcpy(attributes->l3_mask.ipv6.hdr.dst_addr,
2531 mask->ipv6_mask.dst_ip,
2532 RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
2533 attributes->items[1] = (struct rte_flow_item){
2534 .type = RTE_FLOW_ITEM_TYPE_IPV6,
2535 .spec = &attributes->l3,
2536 .mask = &attributes->l3_mask,
2540 DRV_LOG(ERR, "port %u invalid flow type%d",
2541 dev->data->port_id, fdir_filter->input.flow_type);
2542 rte_errno = ENOTSUP;
2546 switch (fdir_filter->input.flow_type) {
2547 case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
2548 attributes->l4.udp.hdr = (struct udp_hdr){
2549 .src_port = input->flow.udp4_flow.src_port,
2550 .dst_port = input->flow.udp4_flow.dst_port,
2552 attributes->l4_mask.udp.hdr = (struct udp_hdr){
2553 .src_port = mask->src_port_mask,
2554 .dst_port = mask->dst_port_mask,
2556 attributes->items[2] = (struct rte_flow_item){
2557 .type = RTE_FLOW_ITEM_TYPE_UDP,
2558 .spec = &attributes->l4,
2559 .mask = &attributes->l4_mask,
2562 case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
2563 attributes->l4.tcp.hdr = (struct tcp_hdr){
2564 .src_port = input->flow.tcp4_flow.src_port,
2565 .dst_port = input->flow.tcp4_flow.dst_port,
2567 attributes->l4_mask.tcp.hdr = (struct tcp_hdr){
2568 .src_port = mask->src_port_mask,
2569 .dst_port = mask->dst_port_mask,
2571 attributes->items[2] = (struct rte_flow_item){
2572 .type = RTE_FLOW_ITEM_TYPE_TCP,
2573 .spec = &attributes->l4,
2574 .mask = &attributes->l4_mask,
2577 case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
2578 attributes->l4.udp.hdr = (struct udp_hdr){
2579 .src_port = input->flow.udp6_flow.src_port,
2580 .dst_port = input->flow.udp6_flow.dst_port,
2582 attributes->l4_mask.udp.hdr = (struct udp_hdr){
2583 .src_port = mask->src_port_mask,
2584 .dst_port = mask->dst_port_mask,
2586 attributes->items[2] = (struct rte_flow_item){
2587 .type = RTE_FLOW_ITEM_TYPE_UDP,
2588 .spec = &attributes->l4,
2589 .mask = &attributes->l4_mask,
2592 case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
2593 attributes->l4.tcp.hdr = (struct tcp_hdr){
2594 .src_port = input->flow.tcp6_flow.src_port,
2595 .dst_port = input->flow.tcp6_flow.dst_port,
2597 attributes->l4_mask.tcp.hdr = (struct tcp_hdr){
2598 .src_port = mask->src_port_mask,
2599 .dst_port = mask->dst_port_mask,
2601 attributes->items[2] = (struct rte_flow_item){
2602 .type = RTE_FLOW_ITEM_TYPE_TCP,
2603 .spec = &attributes->l4,
2604 .mask = &attributes->l4_mask,
2607 case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
2608 case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
2611 DRV_LOG(ERR, "port %u invalid flow type%d",
2612 dev->data->port_id, fdir_filter->input.flow_type);
2613 rte_errno = ENOTSUP;
2620 * Add new flow director filter and store it in list.
2623 * Pointer to Ethernet device.
2624 * @param fdir_filter
2625 * Flow director filter to add.
2628 * 0 on success, a negative errno value otherwise and rte_errno is set.
2631 mlx5_fdir_filter_add(struct rte_eth_dev *dev,
2632 const struct rte_eth_fdir_filter *fdir_filter)
2634 struct priv *priv = dev->data->dev_private;
2635 struct mlx5_fdir attributes = {
2638 .dst.addr_bytes = "\x00\x00\x00\x00\x00\x00",
2639 .src.addr_bytes = "\x00\x00\x00\x00\x00\x00",
2643 struct rte_flow_error error;
2644 struct rte_flow *flow;
2647 ret = mlx5_fdir_filter_convert(dev, fdir_filter, &attributes);
2650 flow = flow_list_create(dev, &priv->flows, &attributes.attr,
2651 attributes.items, attributes.actions, &error);
2653 DRV_LOG(DEBUG, "port %u FDIR created %p", dev->data->port_id,
2661 * Delete specific filter.
2664 * Pointer to Ethernet device.
2665 * @param fdir_filter
2666 * Filter to be deleted.
2669 * 0 on success, a negative errno value otherwise and rte_errno is set.
2672 mlx5_fdir_filter_delete(struct rte_eth_dev *dev __rte_unused,
2673 const struct rte_eth_fdir_filter *fdir_filter
2676 rte_errno = ENOTSUP;
2681 * Update queue for specific filter.
2684 * Pointer to Ethernet device.
2685 * @param fdir_filter
2686 * Filter to be updated.
2689 * 0 on success, a negative errno value otherwise and rte_errno is set.
2692 mlx5_fdir_filter_update(struct rte_eth_dev *dev,
2693 const struct rte_eth_fdir_filter *fdir_filter)
2697 ret = mlx5_fdir_filter_delete(dev, fdir_filter);
2700 return mlx5_fdir_filter_add(dev, fdir_filter);
2704 * Flush all filters.
2707 * Pointer to Ethernet device.
2710 mlx5_fdir_filter_flush(struct rte_eth_dev *dev)
2712 struct priv *priv = dev->data->dev_private;
2714 mlx5_flow_list_flush(dev, &priv->flows);
2718 * Get flow director information.
2721 * Pointer to Ethernet device.
2722 * @param[out] fdir_info
2723 * Resulting flow director information.
2726 mlx5_fdir_info_get(struct rte_eth_dev *dev, struct rte_eth_fdir_info *fdir_info)
2728 struct rte_eth_fdir_masks *mask =
2729 &dev->data->dev_conf.fdir_conf.mask;
2731 fdir_info->mode = dev->data->dev_conf.fdir_conf.mode;
2732 fdir_info->guarant_spc = 0;
2733 rte_memcpy(&fdir_info->mask, mask, sizeof(fdir_info->mask));
2734 fdir_info->max_flexpayload = 0;
2735 fdir_info->flow_types_mask[0] = 0;
2736 fdir_info->flex_payload_unit = 0;
2737 fdir_info->max_flex_payload_segment_num = 0;
2738 fdir_info->flex_payload_limit = 0;
2739 memset(&fdir_info->flex_conf, 0, sizeof(fdir_info->flex_conf));
2743 * Deal with flow director operations.
2746 * Pointer to Ethernet device.
2748 * Operation to perform.
2750 * Pointer to operation-specific structure.
2753 * 0 on success, a negative errno value otherwise and rte_errno is set.
2756 mlx5_fdir_ctrl_func(struct rte_eth_dev *dev, enum rte_filter_op filter_op,
2759 enum rte_fdir_mode fdir_mode =
2760 dev->data->dev_conf.fdir_conf.mode;
2762 if (filter_op == RTE_ETH_FILTER_NOP)
2764 if (fdir_mode != RTE_FDIR_MODE_PERFECT &&
2765 fdir_mode != RTE_FDIR_MODE_PERFECT_MAC_VLAN) {
2766 DRV_LOG(ERR, "port %u flow director mode %d not supported",
2767 dev->data->port_id, fdir_mode);
2771 switch (filter_op) {
2772 case RTE_ETH_FILTER_ADD:
2773 return mlx5_fdir_filter_add(dev, arg);
2774 case RTE_ETH_FILTER_UPDATE:
2775 return mlx5_fdir_filter_update(dev, arg);
2776 case RTE_ETH_FILTER_DELETE:
2777 return mlx5_fdir_filter_delete(dev, arg);
2778 case RTE_ETH_FILTER_FLUSH:
2779 mlx5_fdir_filter_flush(dev);
2781 case RTE_ETH_FILTER_INFO:
2782 mlx5_fdir_info_get(dev, arg);
2785 DRV_LOG(DEBUG, "port %u unknown operation %u",
2786 dev->data->port_id, filter_op);
2794 * Manage filter operations.
2797 * Pointer to Ethernet device structure.
2798 * @param filter_type
2801 * Operation to perform.
2803 * Pointer to operation-specific structure.
2806 * 0 on success, a negative errno value otherwise and rte_errno is set.
2809 mlx5_dev_filter_ctrl(struct rte_eth_dev *dev,
2810 enum rte_filter_type filter_type,
2811 enum rte_filter_op filter_op,
2814 switch (filter_type) {
2815 case RTE_ETH_FILTER_GENERIC:
2816 if (filter_op != RTE_ETH_FILTER_GET) {
2820 *(const void **)arg = &mlx5_flow_ops;
2822 case RTE_ETH_FILTER_FDIR:
2823 return mlx5_fdir_ctrl_func(dev, filter_op, arg);
2825 DRV_LOG(ERR, "port %u filter type (%d) not supported",
2826 dev->data->port_id, filter_type);
2827 rte_errno = ENOTSUP;