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_item items[4];
243 struct rte_flow_item_eth l2;
244 struct rte_flow_item_eth l2_mask;
246 struct rte_flow_item_ipv4 ipv4;
247 struct rte_flow_item_ipv6 ipv6;
250 struct rte_flow_item_ipv4 ipv4;
251 struct rte_flow_item_ipv6 ipv6;
254 struct rte_flow_item_udp udp;
255 struct rte_flow_item_tcp tcp;
258 struct rte_flow_item_udp udp;
259 struct rte_flow_item_tcp tcp;
261 struct rte_flow_action actions[2];
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 uint64_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_UDP | 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, uint64_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"
850 return rte_flow_error_set(error, EINVAL,
851 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
852 NULL, "No Rx queues configured");
853 if (queue->index >= priv->rxqs_n)
854 return rte_flow_error_set(error, EINVAL,
855 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
857 "queue index out of range");
858 if (!(*priv->rxqs)[queue->index])
859 return rte_flow_error_set(error, EINVAL,
860 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
862 "queue is not configured");
864 return rte_flow_error_set(error, ENOTSUP,
865 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
866 "queue action not supported for "
872 * Validate the rss action.
875 * Pointer to the queue action.
876 * @param[in] action_flags
877 * Bit-fields that holds the actions detected until now.
879 * Pointer to the Ethernet device structure.
881 * Attributes of flow that includes this action.
883 * Pointer to error structure.
886 * 0 on success, a negative errno value otherwise and rte_ernno is set.
889 mlx5_flow_validate_action_rss(const struct rte_flow_action *action,
890 uint64_t action_flags,
891 struct rte_eth_dev *dev,
892 const struct rte_flow_attr *attr,
893 struct rte_flow_error *error)
895 struct priv *priv = dev->data->dev_private;
896 const struct rte_flow_action_rss *rss = action->conf;
899 if (action_flags & MLX5_FLOW_FATE_ACTIONS)
900 return rte_flow_error_set(error, EINVAL,
901 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
902 "can't have 2 fate actions"
904 if (rss->func != RTE_ETH_HASH_FUNCTION_DEFAULT &&
905 rss->func != RTE_ETH_HASH_FUNCTION_TOEPLITZ)
906 return rte_flow_error_set(error, ENOTSUP,
907 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
909 "RSS hash function not supported");
910 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
915 return rte_flow_error_set(error, ENOTSUP,
916 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
918 "tunnel RSS is not supported");
919 /* allow RSS key_len 0 in case of NULL (default) RSS key. */
920 if (rss->key_len == 0 && rss->key != NULL)
921 return rte_flow_error_set(error, ENOTSUP,
922 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
924 "RSS hash key length 0");
925 if (rss->key_len > 0 && rss->key_len < MLX5_RSS_HASH_KEY_LEN)
926 return rte_flow_error_set(error, ENOTSUP,
927 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
929 "RSS hash key too small");
930 if (rss->key_len > MLX5_RSS_HASH_KEY_LEN)
931 return rte_flow_error_set(error, ENOTSUP,
932 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
934 "RSS hash key too large");
935 if (rss->queue_num > priv->config.ind_table_max_size)
936 return rte_flow_error_set(error, ENOTSUP,
937 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
939 "number of queues too large");
940 if (rss->types & MLX5_RSS_HF_MASK)
941 return rte_flow_error_set(error, ENOTSUP,
942 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
944 "some RSS protocols are not"
947 return rte_flow_error_set(error, EINVAL,
948 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
949 NULL, "No Rx queues configured");
950 for (i = 0; i != rss->queue_num; ++i) {
951 if (!(*priv->rxqs)[rss->queue[i]])
952 return rte_flow_error_set
953 (error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
954 &rss->queue[i], "queue is not configured");
957 return rte_flow_error_set(error, ENOTSUP,
958 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
959 "rss action not supported for "
965 * Validate the count action.
968 * Pointer to the Ethernet device structure.
970 * Attributes of flow that includes this action.
972 * Pointer to error structure.
975 * 0 on success, a negative errno value otherwise and rte_ernno is set.
978 mlx5_flow_validate_action_count(struct rte_eth_dev *dev __rte_unused,
979 const struct rte_flow_attr *attr,
980 struct rte_flow_error *error)
983 return rte_flow_error_set(error, ENOTSUP,
984 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
985 "count action not supported for "
991 * Verify the @p attributes will be correctly understood by the NIC and store
992 * them in the @p flow if everything is correct.
995 * Pointer to the Ethernet device structure.
996 * @param[in] attributes
997 * Pointer to flow attributes
999 * Pointer to error structure.
1002 * 0 on success, a negative errno value otherwise and rte_errno is set.
1005 mlx5_flow_validate_attributes(struct rte_eth_dev *dev,
1006 const struct rte_flow_attr *attributes,
1007 struct rte_flow_error *error)
1009 struct priv *priv = dev->data->dev_private;
1010 uint32_t priority_max = priv->config.flow_prio - 1;
1012 if (attributes->group)
1013 return rte_flow_error_set(error, ENOTSUP,
1014 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
1015 NULL, "groups is not supported");
1016 if (attributes->priority != MLX5_FLOW_PRIO_RSVD &&
1017 attributes->priority >= priority_max)
1018 return rte_flow_error_set(error, ENOTSUP,
1019 RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
1020 NULL, "priority out of range");
1021 if (attributes->egress)
1022 return rte_flow_error_set(error, ENOTSUP,
1023 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1024 "egress is not supported");
1025 if (attributes->transfer)
1026 return rte_flow_error_set(error, ENOTSUP,
1027 RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1028 NULL, "transfer is not supported");
1029 if (!attributes->ingress)
1030 return rte_flow_error_set(error, EINVAL,
1031 RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
1033 "ingress attribute is mandatory");
1038 * Validate Ethernet item.
1041 * Item specification.
1042 * @param[in] item_flags
1043 * Bit-fields that holds the items detected until now.
1045 * Pointer to error structure.
1048 * 0 on success, a negative errno value otherwise and rte_errno is set.
1051 mlx5_flow_validate_item_eth(const struct rte_flow_item *item,
1052 uint64_t item_flags,
1053 struct rte_flow_error *error)
1055 const struct rte_flow_item_eth *mask = item->mask;
1056 const struct rte_flow_item_eth nic_mask = {
1057 .dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1058 .src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1059 .type = RTE_BE16(0xffff),
1062 int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1063 const uint64_t ethm = tunnel ? MLX5_FLOW_LAYER_INNER_L2 :
1064 MLX5_FLOW_LAYER_OUTER_L2;
1066 if (item_flags & ethm)
1067 return rte_flow_error_set(error, ENOTSUP,
1068 RTE_FLOW_ERROR_TYPE_ITEM, item,
1069 "multiple L2 layers not supported");
1071 mask = &rte_flow_item_eth_mask;
1072 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1073 (const uint8_t *)&nic_mask,
1074 sizeof(struct rte_flow_item_eth),
1080 * Validate VLAN item.
1083 * Item specification.
1084 * @param[in] item_flags
1085 * Bit-fields that holds the items detected until now.
1087 * Pointer to error structure.
1090 * 0 on success, a negative errno value otherwise and rte_errno is set.
1093 mlx5_flow_validate_item_vlan(const struct rte_flow_item *item,
1094 uint64_t item_flags,
1095 struct rte_flow_error *error)
1097 const struct rte_flow_item_vlan *spec = item->spec;
1098 const struct rte_flow_item_vlan *mask = item->mask;
1099 const struct rte_flow_item_vlan nic_mask = {
1100 .tci = RTE_BE16(0x0fff),
1101 .inner_type = RTE_BE16(0xffff),
1103 uint16_t vlan_tag = 0;
1104 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1106 const uint64_t l34m = tunnel ? (MLX5_FLOW_LAYER_INNER_L3 |
1107 MLX5_FLOW_LAYER_INNER_L4) :
1108 (MLX5_FLOW_LAYER_OUTER_L3 |
1109 MLX5_FLOW_LAYER_OUTER_L4);
1110 const uint64_t vlanm = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
1111 MLX5_FLOW_LAYER_OUTER_VLAN;
1113 if (item_flags & vlanm)
1114 return rte_flow_error_set(error, EINVAL,
1115 RTE_FLOW_ERROR_TYPE_ITEM, item,
1116 "multiple VLAN layers not supported");
1117 else if ((item_flags & l34m) != 0)
1118 return rte_flow_error_set(error, EINVAL,
1119 RTE_FLOW_ERROR_TYPE_ITEM, item,
1120 "L2 layer cannot follow L3/L4 layer");
1122 mask = &rte_flow_item_vlan_mask;
1123 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1124 (const uint8_t *)&nic_mask,
1125 sizeof(struct rte_flow_item_vlan),
1130 vlan_tag = spec->tci;
1131 vlan_tag &= mask->tci;
1134 * From verbs perspective an empty VLAN is equivalent
1135 * to a packet without VLAN layer.
1138 return rte_flow_error_set(error, EINVAL,
1139 RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
1141 "VLAN cannot be empty");
1146 * Validate IPV4 item.
1149 * Item specification.
1150 * @param[in] item_flags
1151 * Bit-fields that holds the items detected until now.
1152 * @param[in] acc_mask
1153 * Acceptable mask, if NULL default internal default mask
1154 * will be used to check whether item fields are supported.
1156 * Pointer to error structure.
1159 * 0 on success, a negative errno value otherwise and rte_errno is set.
1162 mlx5_flow_validate_item_ipv4(const struct rte_flow_item *item,
1163 uint64_t item_flags,
1164 const struct rte_flow_item_ipv4 *acc_mask,
1165 struct rte_flow_error *error)
1167 const struct rte_flow_item_ipv4 *mask = item->mask;
1168 const struct rte_flow_item_ipv4 nic_mask = {
1170 .src_addr = RTE_BE32(0xffffffff),
1171 .dst_addr = RTE_BE32(0xffffffff),
1172 .type_of_service = 0xff,
1173 .next_proto_id = 0xff,
1176 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1177 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1178 MLX5_FLOW_LAYER_OUTER_L3;
1179 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1180 MLX5_FLOW_LAYER_OUTER_L4;
1183 if (item_flags & l3m)
1184 return rte_flow_error_set(error, ENOTSUP,
1185 RTE_FLOW_ERROR_TYPE_ITEM, item,
1186 "multiple L3 layers not supported");
1187 else if (item_flags & l4m)
1188 return rte_flow_error_set(error, EINVAL,
1189 RTE_FLOW_ERROR_TYPE_ITEM, item,
1190 "L3 cannot follow an L4 layer.");
1192 mask = &rte_flow_item_ipv4_mask;
1193 else if (mask->hdr.next_proto_id != 0 &&
1194 mask->hdr.next_proto_id != 0xff)
1195 return rte_flow_error_set(error, EINVAL,
1196 RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
1197 "partial mask is not supported"
1199 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1200 acc_mask ? (const uint8_t *)acc_mask
1201 : (const uint8_t *)&nic_mask,
1202 sizeof(struct rte_flow_item_ipv4),
1210 * Validate IPV6 item.
1213 * Item specification.
1214 * @param[in] item_flags
1215 * Bit-fields that holds the items detected until now.
1216 * @param[in] acc_mask
1217 * Acceptable mask, if NULL default internal default mask
1218 * will be used to check whether item fields are supported.
1220 * Pointer to error structure.
1223 * 0 on success, a negative errno value otherwise and rte_errno is set.
1226 mlx5_flow_validate_item_ipv6(const struct rte_flow_item *item,
1227 uint64_t item_flags,
1228 const struct rte_flow_item_ipv6 *acc_mask,
1229 struct rte_flow_error *error)
1231 const struct rte_flow_item_ipv6 *mask = item->mask;
1232 const struct rte_flow_item_ipv6 nic_mask = {
1235 "\xff\xff\xff\xff\xff\xff\xff\xff"
1236 "\xff\xff\xff\xff\xff\xff\xff\xff",
1238 "\xff\xff\xff\xff\xff\xff\xff\xff"
1239 "\xff\xff\xff\xff\xff\xff\xff\xff",
1240 .vtc_flow = RTE_BE32(0xffffffff),
1245 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1246 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1247 MLX5_FLOW_LAYER_OUTER_L3;
1248 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1249 MLX5_FLOW_LAYER_OUTER_L4;
1252 if (item_flags & l3m)
1253 return rte_flow_error_set(error, ENOTSUP,
1254 RTE_FLOW_ERROR_TYPE_ITEM, item,
1255 "multiple L3 layers not supported");
1256 else if (item_flags & l4m)
1257 return rte_flow_error_set(error, EINVAL,
1258 RTE_FLOW_ERROR_TYPE_ITEM, item,
1259 "L3 cannot follow an L4 layer.");
1261 mask = &rte_flow_item_ipv6_mask;
1262 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1263 acc_mask ? (const uint8_t *)acc_mask
1264 : (const uint8_t *)&nic_mask,
1265 sizeof(struct rte_flow_item_ipv6),
1273 * Validate UDP item.
1276 * Item specification.
1277 * @param[in] item_flags
1278 * Bit-fields that holds the items detected until now.
1279 * @param[in] target_protocol
1280 * The next protocol in the previous item.
1281 * @param[in] flow_mask
1282 * mlx5 flow-specific (TCF, DV, verbs, etc.) supported header fields mask.
1284 * Pointer to error structure.
1287 * 0 on success, a negative errno value otherwise and rte_errno is set.
1290 mlx5_flow_validate_item_udp(const struct rte_flow_item *item,
1291 uint64_t item_flags,
1292 uint8_t target_protocol,
1293 struct rte_flow_error *error)
1295 const struct rte_flow_item_udp *mask = item->mask;
1296 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1297 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1298 MLX5_FLOW_LAYER_OUTER_L3;
1299 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1300 MLX5_FLOW_LAYER_OUTER_L4;
1303 if (target_protocol != 0xff && target_protocol != IPPROTO_UDP)
1304 return rte_flow_error_set(error, EINVAL,
1305 RTE_FLOW_ERROR_TYPE_ITEM, item,
1306 "protocol filtering not compatible"
1308 if (!(item_flags & l3m))
1309 return rte_flow_error_set(error, EINVAL,
1310 RTE_FLOW_ERROR_TYPE_ITEM, item,
1311 "L3 is mandatory to filter on L4");
1312 if (item_flags & l4m)
1313 return rte_flow_error_set(error, EINVAL,
1314 RTE_FLOW_ERROR_TYPE_ITEM, item,
1315 "multiple L4 layers not supported");
1317 mask = &rte_flow_item_udp_mask;
1318 ret = mlx5_flow_item_acceptable
1319 (item, (const uint8_t *)mask,
1320 (const uint8_t *)&rte_flow_item_udp_mask,
1321 sizeof(struct rte_flow_item_udp), error);
1328 * Validate TCP item.
1331 * Item specification.
1332 * @param[in] item_flags
1333 * Bit-fields that holds the items detected until now.
1334 * @param[in] target_protocol
1335 * The next protocol in the previous item.
1337 * Pointer to error structure.
1340 * 0 on success, a negative errno value otherwise and rte_errno is set.
1343 mlx5_flow_validate_item_tcp(const struct rte_flow_item *item,
1344 uint64_t item_flags,
1345 uint8_t target_protocol,
1346 const struct rte_flow_item_tcp *flow_mask,
1347 struct rte_flow_error *error)
1349 const struct rte_flow_item_tcp *mask = item->mask;
1350 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1351 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1352 MLX5_FLOW_LAYER_OUTER_L3;
1353 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1354 MLX5_FLOW_LAYER_OUTER_L4;
1358 if (target_protocol != 0xff && target_protocol != IPPROTO_TCP)
1359 return rte_flow_error_set(error, EINVAL,
1360 RTE_FLOW_ERROR_TYPE_ITEM, item,
1361 "protocol filtering not compatible"
1363 if (!(item_flags & l3m))
1364 return rte_flow_error_set(error, EINVAL,
1365 RTE_FLOW_ERROR_TYPE_ITEM, item,
1366 "L3 is mandatory to filter on L4");
1367 if (item_flags & l4m)
1368 return rte_flow_error_set(error, EINVAL,
1369 RTE_FLOW_ERROR_TYPE_ITEM, item,
1370 "multiple L4 layers not supported");
1372 mask = &rte_flow_item_tcp_mask;
1373 ret = mlx5_flow_item_acceptable
1374 (item, (const uint8_t *)mask,
1375 (const uint8_t *)flow_mask,
1376 sizeof(struct rte_flow_item_tcp), error);
1383 * Validate VXLAN item.
1386 * Item specification.
1387 * @param[in] item_flags
1388 * Bit-fields that holds the items detected until now.
1389 * @param[in] target_protocol
1390 * The next protocol in the previous item.
1392 * Pointer to error structure.
1395 * 0 on success, a negative errno value otherwise and rte_errno is set.
1398 mlx5_flow_validate_item_vxlan(const struct rte_flow_item *item,
1399 uint64_t item_flags,
1400 struct rte_flow_error *error)
1402 const struct rte_flow_item_vxlan *spec = item->spec;
1403 const struct rte_flow_item_vxlan *mask = item->mask;
1408 } id = { .vlan_id = 0, };
1409 uint32_t vlan_id = 0;
1412 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1413 return rte_flow_error_set(error, ENOTSUP,
1414 RTE_FLOW_ERROR_TYPE_ITEM, item,
1415 "multiple tunnel layers not"
1418 * Verify only UDPv4 is present as defined in
1419 * https://tools.ietf.org/html/rfc7348
1421 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
1422 return rte_flow_error_set(error, EINVAL,
1423 RTE_FLOW_ERROR_TYPE_ITEM, item,
1424 "no outer UDP layer found");
1426 mask = &rte_flow_item_vxlan_mask;
1427 ret = mlx5_flow_item_acceptable
1428 (item, (const uint8_t *)mask,
1429 (const uint8_t *)&rte_flow_item_vxlan_mask,
1430 sizeof(struct rte_flow_item_vxlan),
1435 memcpy(&id.vni[1], spec->vni, 3);
1436 vlan_id = id.vlan_id;
1437 memcpy(&id.vni[1], mask->vni, 3);
1438 vlan_id &= id.vlan_id;
1441 * Tunnel id 0 is equivalent as not adding a VXLAN layer, if
1442 * only this layer is defined in the Verbs specification it is
1443 * interpreted as wildcard and all packets will match this
1444 * rule, if it follows a full stack layer (ex: eth / ipv4 /
1445 * udp), all packets matching the layers before will also
1446 * match this rule. To avoid such situation, VNI 0 is
1447 * currently refused.
1450 return rte_flow_error_set(error, ENOTSUP,
1451 RTE_FLOW_ERROR_TYPE_ITEM, item,
1452 "VXLAN vni cannot be 0");
1453 if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
1454 return rte_flow_error_set(error, ENOTSUP,
1455 RTE_FLOW_ERROR_TYPE_ITEM, item,
1456 "VXLAN tunnel must be fully defined");
1461 * Validate VXLAN_GPE item.
1464 * Item specification.
1465 * @param[in] item_flags
1466 * Bit-fields that holds the items detected until now.
1468 * Pointer to the private data structure.
1469 * @param[in] target_protocol
1470 * The next protocol in the previous item.
1472 * Pointer to error structure.
1475 * 0 on success, a negative errno value otherwise and rte_errno is set.
1478 mlx5_flow_validate_item_vxlan_gpe(const struct rte_flow_item *item,
1479 uint64_t item_flags,
1480 struct rte_eth_dev *dev,
1481 struct rte_flow_error *error)
1483 struct priv *priv = dev->data->dev_private;
1484 const struct rte_flow_item_vxlan_gpe *spec = item->spec;
1485 const struct rte_flow_item_vxlan_gpe *mask = item->mask;
1490 } id = { .vlan_id = 0, };
1491 uint32_t vlan_id = 0;
1493 if (!priv->config.l3_vxlan_en)
1494 return rte_flow_error_set(error, ENOTSUP,
1495 RTE_FLOW_ERROR_TYPE_ITEM, item,
1496 "L3 VXLAN is not enabled by device"
1497 " parameter and/or not configured in"
1499 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1500 return rte_flow_error_set(error, ENOTSUP,
1501 RTE_FLOW_ERROR_TYPE_ITEM, item,
1502 "multiple tunnel layers not"
1505 * Verify only UDPv4 is present as defined in
1506 * https://tools.ietf.org/html/rfc7348
1508 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
1509 return rte_flow_error_set(error, EINVAL,
1510 RTE_FLOW_ERROR_TYPE_ITEM, item,
1511 "no outer UDP layer found");
1513 mask = &rte_flow_item_vxlan_gpe_mask;
1514 ret = mlx5_flow_item_acceptable
1515 (item, (const uint8_t *)mask,
1516 (const uint8_t *)&rte_flow_item_vxlan_gpe_mask,
1517 sizeof(struct rte_flow_item_vxlan_gpe),
1523 return rte_flow_error_set(error, ENOTSUP,
1524 RTE_FLOW_ERROR_TYPE_ITEM,
1526 "VxLAN-GPE protocol"
1528 memcpy(&id.vni[1], spec->vni, 3);
1529 vlan_id = id.vlan_id;
1530 memcpy(&id.vni[1], mask->vni, 3);
1531 vlan_id &= id.vlan_id;
1534 * Tunnel id 0 is equivalent as not adding a VXLAN layer, if only this
1535 * layer is defined in the Verbs specification it is interpreted as
1536 * wildcard and all packets will match this rule, if it follows a full
1537 * stack layer (ex: eth / ipv4 / udp), all packets matching the layers
1538 * before will also match this rule. To avoid such situation, VNI 0
1539 * is currently refused.
1542 return rte_flow_error_set(error, ENOTSUP,
1543 RTE_FLOW_ERROR_TYPE_ITEM, item,
1544 "VXLAN-GPE vni cannot be 0");
1545 if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
1546 return rte_flow_error_set(error, ENOTSUP,
1547 RTE_FLOW_ERROR_TYPE_ITEM, item,
1548 "VXLAN-GPE tunnel must be fully"
1554 * Validate GRE item.
1557 * Item specification.
1558 * @param[in] item_flags
1559 * Bit flags to mark detected items.
1560 * @param[in] target_protocol
1561 * The next protocol in the previous item.
1563 * Pointer to error structure.
1566 * 0 on success, a negative errno value otherwise and rte_errno is set.
1569 mlx5_flow_validate_item_gre(const struct rte_flow_item *item,
1570 uint64_t item_flags,
1571 uint8_t target_protocol,
1572 struct rte_flow_error *error)
1574 const struct rte_flow_item_gre *spec __rte_unused = item->spec;
1575 const struct rte_flow_item_gre *mask = item->mask;
1578 if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
1579 return rte_flow_error_set(error, EINVAL,
1580 RTE_FLOW_ERROR_TYPE_ITEM, item,
1581 "protocol filtering not compatible"
1582 " with this GRE layer");
1583 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1584 return rte_flow_error_set(error, ENOTSUP,
1585 RTE_FLOW_ERROR_TYPE_ITEM, item,
1586 "multiple tunnel layers not"
1588 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
1589 return rte_flow_error_set(error, ENOTSUP,
1590 RTE_FLOW_ERROR_TYPE_ITEM, item,
1591 "L3 Layer is missing");
1593 mask = &rte_flow_item_gre_mask;
1594 ret = mlx5_flow_item_acceptable
1595 (item, (const uint8_t *)mask,
1596 (const uint8_t *)&rte_flow_item_gre_mask,
1597 sizeof(struct rte_flow_item_gre), error);
1600 #ifndef HAVE_IBV_DEVICE_MPLS_SUPPORT
1601 if (spec && (spec->protocol & mask->protocol))
1602 return rte_flow_error_set(error, ENOTSUP,
1603 RTE_FLOW_ERROR_TYPE_ITEM, item,
1604 "without MPLS support the"
1605 " specification cannot be used for"
1612 * Validate MPLS item.
1615 * Pointer to the rte_eth_dev structure.
1617 * Item specification.
1618 * @param[in] item_flags
1619 * Bit-fields that holds the items detected until now.
1620 * @param[in] prev_layer
1621 * The protocol layer indicated in previous item.
1623 * Pointer to error structure.
1626 * 0 on success, a negative errno value otherwise and rte_errno is set.
1629 mlx5_flow_validate_item_mpls(struct rte_eth_dev *dev __rte_unused,
1630 const struct rte_flow_item *item __rte_unused,
1631 uint64_t item_flags __rte_unused,
1632 uint64_t prev_layer __rte_unused,
1633 struct rte_flow_error *error)
1635 #ifdef HAVE_IBV_DEVICE_MPLS_SUPPORT
1636 const struct rte_flow_item_mpls *mask = item->mask;
1637 struct priv *priv = dev->data->dev_private;
1640 if (!priv->config.mpls_en)
1641 return rte_flow_error_set(error, ENOTSUP,
1642 RTE_FLOW_ERROR_TYPE_ITEM, item,
1643 "MPLS not supported or"
1644 " disabled in firmware"
1646 /* MPLS over IP, UDP, GRE is allowed */
1647 if (!(prev_layer & (MLX5_FLOW_LAYER_OUTER_L3 |
1648 MLX5_FLOW_LAYER_OUTER_L4_UDP |
1649 MLX5_FLOW_LAYER_GRE)))
1650 return rte_flow_error_set(error, EINVAL,
1651 RTE_FLOW_ERROR_TYPE_ITEM, item,
1652 "protocol filtering not compatible"
1653 " with MPLS layer");
1654 /* Multi-tunnel isn't allowed but MPLS over GRE is an exception. */
1655 if ((item_flags & MLX5_FLOW_LAYER_TUNNEL) &&
1656 !(item_flags & MLX5_FLOW_LAYER_GRE))
1657 return rte_flow_error_set(error, ENOTSUP,
1658 RTE_FLOW_ERROR_TYPE_ITEM, item,
1659 "multiple tunnel layers not"
1662 mask = &rte_flow_item_mpls_mask;
1663 ret = mlx5_flow_item_acceptable
1664 (item, (const uint8_t *)mask,
1665 (const uint8_t *)&rte_flow_item_mpls_mask,
1666 sizeof(struct rte_flow_item_mpls), error);
1671 return rte_flow_error_set(error, ENOTSUP,
1672 RTE_FLOW_ERROR_TYPE_ITEM, item,
1673 "MPLS is not supported by Verbs, please"
1678 flow_null_validate(struct rte_eth_dev *dev __rte_unused,
1679 const struct rte_flow_attr *attr __rte_unused,
1680 const struct rte_flow_item items[] __rte_unused,
1681 const struct rte_flow_action actions[] __rte_unused,
1682 struct rte_flow_error *error __rte_unused)
1684 rte_errno = ENOTSUP;
1688 static struct mlx5_flow *
1689 flow_null_prepare(const struct rte_flow_attr *attr __rte_unused,
1690 const struct rte_flow_item items[] __rte_unused,
1691 const struct rte_flow_action actions[] __rte_unused,
1692 struct rte_flow_error *error __rte_unused)
1694 rte_errno = ENOTSUP;
1699 flow_null_translate(struct rte_eth_dev *dev __rte_unused,
1700 struct mlx5_flow *dev_flow __rte_unused,
1701 const struct rte_flow_attr *attr __rte_unused,
1702 const struct rte_flow_item items[] __rte_unused,
1703 const struct rte_flow_action actions[] __rte_unused,
1704 struct rte_flow_error *error __rte_unused)
1706 rte_errno = ENOTSUP;
1711 flow_null_apply(struct rte_eth_dev *dev __rte_unused,
1712 struct rte_flow *flow __rte_unused,
1713 struct rte_flow_error *error __rte_unused)
1715 rte_errno = ENOTSUP;
1720 flow_null_remove(struct rte_eth_dev *dev __rte_unused,
1721 struct rte_flow *flow __rte_unused)
1726 flow_null_destroy(struct rte_eth_dev *dev __rte_unused,
1727 struct rte_flow *flow __rte_unused)
1732 flow_null_query(struct rte_eth_dev *dev __rte_unused,
1733 struct rte_flow *flow __rte_unused,
1734 const struct rte_flow_action *actions __rte_unused,
1735 void *data __rte_unused,
1736 struct rte_flow_error *error __rte_unused)
1738 rte_errno = ENOTSUP;
1742 /* Void driver to protect from null pointer reference. */
1743 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops = {
1744 .validate = flow_null_validate,
1745 .prepare = flow_null_prepare,
1746 .translate = flow_null_translate,
1747 .apply = flow_null_apply,
1748 .remove = flow_null_remove,
1749 .destroy = flow_null_destroy,
1750 .query = flow_null_query,
1754 * Select flow driver type according to flow attributes and device
1758 * Pointer to the dev structure.
1760 * Pointer to the flow attributes.
1763 * flow driver type, MLX5_FLOW_TYPE_MAX otherwise.
1765 static enum mlx5_flow_drv_type
1766 flow_get_drv_type(struct rte_eth_dev *dev, const struct rte_flow_attr *attr)
1768 struct priv *priv = dev->data->dev_private;
1769 enum mlx5_flow_drv_type type = MLX5_FLOW_TYPE_MAX;
1772 type = MLX5_FLOW_TYPE_TCF;
1774 type = priv->config.dv_flow_en ? MLX5_FLOW_TYPE_DV :
1775 MLX5_FLOW_TYPE_VERBS;
1779 #define flow_get_drv_ops(type) flow_drv_ops[type]
1782 * Flow driver validation API. This abstracts calling driver specific functions.
1783 * The type of flow driver is determined according to flow attributes.
1786 * Pointer to the dev structure.
1788 * Pointer to the flow attributes.
1790 * Pointer to the list of items.
1791 * @param[in] actions
1792 * Pointer to the list of actions.
1794 * Pointer to the error structure.
1797 * 0 on success, a negative errno value otherwise and rte_ernno is set.
1800 flow_drv_validate(struct rte_eth_dev *dev,
1801 const struct rte_flow_attr *attr,
1802 const struct rte_flow_item items[],
1803 const struct rte_flow_action actions[],
1804 struct rte_flow_error *error)
1806 const struct mlx5_flow_driver_ops *fops;
1807 enum mlx5_flow_drv_type type = flow_get_drv_type(dev, attr);
1809 fops = flow_get_drv_ops(type);
1810 return fops->validate(dev, attr, items, actions, error);
1814 * Flow driver preparation API. This abstracts calling driver specific
1815 * functions. Parent flow (rte_flow) should have driver type (drv_type). It
1816 * calculates the size of memory required for device flow, allocates the memory,
1817 * initializes the device flow and returns the pointer.
1820 * This function initializes device flow structure such as dv, tcf or verbs in
1821 * struct mlx5_flow. However, it is caller's responsibility to initialize the
1822 * rest. For example, adding returning device flow to flow->dev_flow list and
1823 * setting backward reference to the flow should be done out of this function.
1824 * layers field is not filled either.
1827 * Pointer to the flow attributes.
1829 * Pointer to the list of items.
1830 * @param[in] actions
1831 * Pointer to the list of actions.
1833 * Pointer to the error structure.
1836 * Pointer to device flow on success, otherwise NULL and rte_ernno is set.
1838 static inline struct mlx5_flow *
1839 flow_drv_prepare(const struct rte_flow *flow,
1840 const struct rte_flow_attr *attr,
1841 const struct rte_flow_item items[],
1842 const struct rte_flow_action actions[],
1843 struct rte_flow_error *error)
1845 const struct mlx5_flow_driver_ops *fops;
1846 enum mlx5_flow_drv_type type = flow->drv_type;
1848 assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
1849 fops = flow_get_drv_ops(type);
1850 return fops->prepare(attr, items, actions, error);
1854 * Flow driver translation API. This abstracts calling driver specific
1855 * functions. Parent flow (rte_flow) should have driver type (drv_type). It
1856 * translates a generic flow into a driver flow. flow_drv_prepare() must
1860 * dev_flow->layers could be filled as a result of parsing during translation
1861 * if needed by flow_drv_apply(). dev_flow->flow->actions can also be filled
1862 * if necessary. As a flow can have multiple dev_flows by RSS flow expansion,
1863 * flow->actions could be overwritten even though all the expanded dev_flows
1864 * have the same actions.
1867 * Pointer to the rte dev structure.
1868 * @param[in, out] dev_flow
1869 * Pointer to the mlx5 flow.
1871 * Pointer to the flow attributes.
1873 * Pointer to the list of items.
1874 * @param[in] actions
1875 * Pointer to the list of actions.
1877 * Pointer to the error structure.
1880 * 0 on success, a negative errno value otherwise and rte_ernno is set.
1883 flow_drv_translate(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow,
1884 const struct rte_flow_attr *attr,
1885 const struct rte_flow_item items[],
1886 const struct rte_flow_action actions[],
1887 struct rte_flow_error *error)
1889 const struct mlx5_flow_driver_ops *fops;
1890 enum mlx5_flow_drv_type type = dev_flow->flow->drv_type;
1892 assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
1893 fops = flow_get_drv_ops(type);
1894 return fops->translate(dev, dev_flow, attr, items, actions, error);
1898 * Flow driver apply API. This abstracts calling driver specific functions.
1899 * Parent flow (rte_flow) should have driver type (drv_type). It applies
1900 * translated driver flows on to device. flow_drv_translate() must precede.
1903 * Pointer to Ethernet device structure.
1904 * @param[in, out] flow
1905 * Pointer to flow structure.
1907 * Pointer to error structure.
1910 * 0 on success, a negative errno value otherwise and rte_errno is set.
1913 flow_drv_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
1914 struct rte_flow_error *error)
1916 const struct mlx5_flow_driver_ops *fops;
1917 enum mlx5_flow_drv_type type = flow->drv_type;
1919 assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
1920 fops = flow_get_drv_ops(type);
1921 return fops->apply(dev, flow, error);
1925 * Flow driver remove API. This abstracts calling driver specific functions.
1926 * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
1927 * on device. All the resources of the flow should be freed by calling
1928 * flow_drv_destroy().
1931 * Pointer to Ethernet device.
1932 * @param[in, out] flow
1933 * Pointer to flow structure.
1936 flow_drv_remove(struct rte_eth_dev *dev, struct rte_flow *flow)
1938 const struct mlx5_flow_driver_ops *fops;
1939 enum mlx5_flow_drv_type type = flow->drv_type;
1941 assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
1942 fops = flow_get_drv_ops(type);
1943 fops->remove(dev, flow);
1947 * Flow driver destroy API. This abstracts calling driver specific functions.
1948 * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
1949 * on device and releases resources of the flow.
1952 * Pointer to Ethernet device.
1953 * @param[in, out] flow
1954 * Pointer to flow structure.
1957 flow_drv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
1959 const struct mlx5_flow_driver_ops *fops;
1960 enum mlx5_flow_drv_type type = flow->drv_type;
1962 assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
1963 fops = flow_get_drv_ops(type);
1964 fops->destroy(dev, flow);
1968 * Validate a flow supported by the NIC.
1970 * @see rte_flow_validate()
1974 mlx5_flow_validate(struct rte_eth_dev *dev,
1975 const struct rte_flow_attr *attr,
1976 const struct rte_flow_item items[],
1977 const struct rte_flow_action actions[],
1978 struct rte_flow_error *error)
1982 ret = flow_drv_validate(dev, attr, items, actions, error);
1989 * Get RSS action from the action list.
1991 * @param[in] actions
1992 * Pointer to the list of actions.
1995 * Pointer to the RSS action if exist, else return NULL.
1997 static const struct rte_flow_action_rss*
1998 flow_get_rss_action(const struct rte_flow_action actions[])
2000 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2001 switch (actions->type) {
2002 case RTE_FLOW_ACTION_TYPE_RSS:
2003 return (const struct rte_flow_action_rss *)
2013 find_graph_root(const struct rte_flow_item pattern[], uint32_t rss_level)
2015 const struct rte_flow_item *item;
2016 unsigned int has_vlan = 0;
2018 for (item = pattern; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
2019 if (item->type == RTE_FLOW_ITEM_TYPE_VLAN) {
2025 return rss_level < 2 ? MLX5_EXPANSION_ROOT_ETH_VLAN :
2026 MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN;
2027 return rss_level < 2 ? MLX5_EXPANSION_ROOT :
2028 MLX5_EXPANSION_ROOT_OUTER;
2032 * Create a flow and add it to @p list.
2035 * Pointer to Ethernet device.
2037 * Pointer to a TAILQ flow list.
2039 * Flow rule attributes.
2041 * Pattern specification (list terminated by the END pattern item).
2042 * @param[in] actions
2043 * Associated actions (list terminated by the END action).
2045 * Perform verbose error reporting if not NULL.
2048 * A flow on success, NULL otherwise and rte_errno is set.
2050 static struct rte_flow *
2051 flow_list_create(struct rte_eth_dev *dev, struct mlx5_flows *list,
2052 const struct rte_flow_attr *attr,
2053 const struct rte_flow_item items[],
2054 const struct rte_flow_action actions[],
2055 struct rte_flow_error *error)
2057 struct rte_flow *flow = NULL;
2058 struct mlx5_flow *dev_flow;
2059 const struct rte_flow_action_rss *rss;
2061 struct rte_flow_expand_rss buf;
2062 uint8_t buffer[2048];
2064 struct rte_flow_expand_rss *buf = &expand_buffer.buf;
2069 ret = flow_drv_validate(dev, attr, items, actions, error);
2072 flow_size = sizeof(struct rte_flow);
2073 rss = flow_get_rss_action(actions);
2075 flow_size += RTE_ALIGN_CEIL(rss->queue_num * sizeof(uint16_t),
2078 flow_size += RTE_ALIGN_CEIL(sizeof(uint16_t), sizeof(void *));
2079 flow = rte_calloc(__func__, 1, flow_size, 0);
2080 flow->drv_type = flow_get_drv_type(dev, attr);
2081 assert(flow->drv_type > MLX5_FLOW_TYPE_MIN &&
2082 flow->drv_type < MLX5_FLOW_TYPE_MAX);
2083 flow->queue = (void *)(flow + 1);
2084 LIST_INIT(&flow->dev_flows);
2085 if (rss && rss->types) {
2086 unsigned int graph_root;
2088 graph_root = find_graph_root(items, rss->level);
2089 ret = rte_flow_expand_rss(buf, sizeof(expand_buffer.buffer),
2091 mlx5_support_expansion,
2094 (unsigned int)ret < sizeof(expand_buffer.buffer));
2097 buf->entry[0].pattern = (void *)(uintptr_t)items;
2099 for (i = 0; i < buf->entries; ++i) {
2100 dev_flow = flow_drv_prepare(flow, attr, buf->entry[i].pattern,
2104 dev_flow->flow = flow;
2105 LIST_INSERT_HEAD(&flow->dev_flows, dev_flow, next);
2106 ret = flow_drv_translate(dev, dev_flow, attr,
2107 buf->entry[i].pattern,
2112 if (dev->data->dev_started) {
2113 ret = flow_drv_apply(dev, flow, error);
2117 TAILQ_INSERT_TAIL(list, flow, next);
2118 flow_rxq_flags_set(dev, flow);
2121 ret = rte_errno; /* Save rte_errno before cleanup. */
2123 flow_drv_destroy(dev, flow);
2125 rte_errno = ret; /* Restore rte_errno. */
2132 * @see rte_flow_create()
2136 mlx5_flow_create(struct rte_eth_dev *dev,
2137 const struct rte_flow_attr *attr,
2138 const struct rte_flow_item items[],
2139 const struct rte_flow_action actions[],
2140 struct rte_flow_error *error)
2142 return flow_list_create(dev,
2143 &((struct priv *)dev->data->dev_private)->flows,
2144 attr, items, actions, error);
2148 * Destroy a flow in a list.
2151 * Pointer to Ethernet device.
2153 * Pointer to a TAILQ flow list.
2158 flow_list_destroy(struct rte_eth_dev *dev, struct mlx5_flows *list,
2159 struct rte_flow *flow)
2162 * Update RX queue flags only if port is started, otherwise it is
2165 if (dev->data->dev_started)
2166 flow_rxq_flags_trim(dev, flow);
2167 flow_drv_destroy(dev, flow);
2168 TAILQ_REMOVE(list, flow, next);
2169 rte_free(flow->fdir);
2174 * Destroy all flows.
2177 * Pointer to Ethernet device.
2179 * Pointer to a TAILQ flow list.
2182 mlx5_flow_list_flush(struct rte_eth_dev *dev, struct mlx5_flows *list)
2184 while (!TAILQ_EMPTY(list)) {
2185 struct rte_flow *flow;
2187 flow = TAILQ_FIRST(list);
2188 flow_list_destroy(dev, list, flow);
2196 * Pointer to Ethernet device.
2198 * Pointer to a TAILQ flow list.
2201 mlx5_flow_stop(struct rte_eth_dev *dev, struct mlx5_flows *list)
2203 struct rte_flow *flow;
2205 TAILQ_FOREACH_REVERSE(flow, list, mlx5_flows, next)
2206 flow_drv_remove(dev, flow);
2207 flow_rxq_flags_clear(dev);
2214 * Pointer to Ethernet device.
2216 * Pointer to a TAILQ flow list.
2219 * 0 on success, a negative errno value otherwise and rte_errno is set.
2222 mlx5_flow_start(struct rte_eth_dev *dev, struct mlx5_flows *list)
2224 struct rte_flow *flow;
2225 struct rte_flow_error error;
2228 TAILQ_FOREACH(flow, list, next) {
2229 ret = flow_drv_apply(dev, flow, &error);
2232 flow_rxq_flags_set(dev, flow);
2236 ret = rte_errno; /* Save rte_errno before cleanup. */
2237 mlx5_flow_stop(dev, list);
2238 rte_errno = ret; /* Restore rte_errno. */
2243 * Verify the flow list is empty
2246 * Pointer to Ethernet device.
2248 * @return the number of flows not released.
2251 mlx5_flow_verify(struct rte_eth_dev *dev)
2253 struct priv *priv = dev->data->dev_private;
2254 struct rte_flow *flow;
2257 TAILQ_FOREACH(flow, &priv->flows, next) {
2258 DRV_LOG(DEBUG, "port %u flow %p still referenced",
2259 dev->data->port_id, (void *)flow);
2266 * Enable a control flow configured from the control plane.
2269 * Pointer to Ethernet device.
2271 * An Ethernet flow spec to apply.
2273 * An Ethernet flow mask to apply.
2275 * A VLAN flow spec to apply.
2277 * A VLAN flow mask to apply.
2280 * 0 on success, a negative errno value otherwise and rte_errno is set.
2283 mlx5_ctrl_flow_vlan(struct rte_eth_dev *dev,
2284 struct rte_flow_item_eth *eth_spec,
2285 struct rte_flow_item_eth *eth_mask,
2286 struct rte_flow_item_vlan *vlan_spec,
2287 struct rte_flow_item_vlan *vlan_mask)
2289 struct priv *priv = dev->data->dev_private;
2290 const struct rte_flow_attr attr = {
2292 .priority = MLX5_FLOW_PRIO_RSVD,
2294 struct rte_flow_item items[] = {
2296 .type = RTE_FLOW_ITEM_TYPE_ETH,
2302 .type = (vlan_spec) ? RTE_FLOW_ITEM_TYPE_VLAN :
2303 RTE_FLOW_ITEM_TYPE_END,
2309 .type = RTE_FLOW_ITEM_TYPE_END,
2312 uint16_t queue[priv->reta_idx_n];
2313 struct rte_flow_action_rss action_rss = {
2314 .func = RTE_ETH_HASH_FUNCTION_DEFAULT,
2316 .types = priv->rss_conf.rss_hf,
2317 .key_len = priv->rss_conf.rss_key_len,
2318 .queue_num = priv->reta_idx_n,
2319 .key = priv->rss_conf.rss_key,
2322 struct rte_flow_action actions[] = {
2324 .type = RTE_FLOW_ACTION_TYPE_RSS,
2325 .conf = &action_rss,
2328 .type = RTE_FLOW_ACTION_TYPE_END,
2331 struct rte_flow *flow;
2332 struct rte_flow_error error;
2335 if (!priv->reta_idx_n || !priv->rxqs_n) {
2338 for (i = 0; i != priv->reta_idx_n; ++i)
2339 queue[i] = (*priv->reta_idx)[i];
2340 flow = flow_list_create(dev, &priv->ctrl_flows,
2341 &attr, items, actions, &error);
2348 * Enable a flow control configured from the control plane.
2351 * Pointer to Ethernet device.
2353 * An Ethernet flow spec to apply.
2355 * An Ethernet flow mask to apply.
2358 * 0 on success, a negative errno value otherwise and rte_errno is set.
2361 mlx5_ctrl_flow(struct rte_eth_dev *dev,
2362 struct rte_flow_item_eth *eth_spec,
2363 struct rte_flow_item_eth *eth_mask)
2365 return mlx5_ctrl_flow_vlan(dev, eth_spec, eth_mask, NULL, NULL);
2371 * @see rte_flow_destroy()
2375 mlx5_flow_destroy(struct rte_eth_dev *dev,
2376 struct rte_flow *flow,
2377 struct rte_flow_error *error __rte_unused)
2379 struct priv *priv = dev->data->dev_private;
2381 flow_list_destroy(dev, &priv->flows, flow);
2386 * Destroy all flows.
2388 * @see rte_flow_flush()
2392 mlx5_flow_flush(struct rte_eth_dev *dev,
2393 struct rte_flow_error *error __rte_unused)
2395 struct priv *priv = dev->data->dev_private;
2397 mlx5_flow_list_flush(dev, &priv->flows);
2404 * @see rte_flow_isolate()
2408 mlx5_flow_isolate(struct rte_eth_dev *dev,
2410 struct rte_flow_error *error)
2412 struct priv *priv = dev->data->dev_private;
2414 if (dev->data->dev_started) {
2415 rte_flow_error_set(error, EBUSY,
2416 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
2418 "port must be stopped first");
2421 priv->isolated = !!enable;
2423 dev->dev_ops = &mlx5_dev_ops_isolate;
2425 dev->dev_ops = &mlx5_dev_ops;
2432 * @see rte_flow_query()
2436 flow_drv_query(struct rte_eth_dev *dev,
2437 struct rte_flow *flow,
2438 const struct rte_flow_action *actions,
2440 struct rte_flow_error *error)
2442 const struct mlx5_flow_driver_ops *fops;
2443 enum mlx5_flow_drv_type ftype = flow->drv_type;
2445 assert(ftype > MLX5_FLOW_TYPE_MIN && ftype < MLX5_FLOW_TYPE_MAX);
2446 fops = flow_get_drv_ops(ftype);
2448 return fops->query(dev, flow, actions, data, error);
2454 * @see rte_flow_query()
2458 mlx5_flow_query(struct rte_eth_dev *dev,
2459 struct rte_flow *flow,
2460 const struct rte_flow_action *actions,
2462 struct rte_flow_error *error)
2466 ret = flow_drv_query(dev, flow, actions, data, error);
2473 * Convert a flow director filter to a generic flow.
2476 * Pointer to Ethernet device.
2477 * @param fdir_filter
2478 * Flow director filter to add.
2480 * Generic flow parameters structure.
2483 * 0 on success, a negative errno value otherwise and rte_errno is set.
2486 flow_fdir_filter_convert(struct rte_eth_dev *dev,
2487 const struct rte_eth_fdir_filter *fdir_filter,
2488 struct mlx5_fdir *attributes)
2490 struct priv *priv = dev->data->dev_private;
2491 const struct rte_eth_fdir_input *input = &fdir_filter->input;
2492 const struct rte_eth_fdir_masks *mask =
2493 &dev->data->dev_conf.fdir_conf.mask;
2495 /* Validate queue number. */
2496 if (fdir_filter->action.rx_queue >= priv->rxqs_n) {
2497 DRV_LOG(ERR, "port %u invalid queue number %d",
2498 dev->data->port_id, fdir_filter->action.rx_queue);
2502 attributes->attr.ingress = 1;
2503 attributes->items[0] = (struct rte_flow_item) {
2504 .type = RTE_FLOW_ITEM_TYPE_ETH,
2505 .spec = &attributes->l2,
2506 .mask = &attributes->l2_mask,
2508 switch (fdir_filter->action.behavior) {
2509 case RTE_ETH_FDIR_ACCEPT:
2510 attributes->actions[0] = (struct rte_flow_action){
2511 .type = RTE_FLOW_ACTION_TYPE_QUEUE,
2512 .conf = &attributes->queue,
2515 case RTE_ETH_FDIR_REJECT:
2516 attributes->actions[0] = (struct rte_flow_action){
2517 .type = RTE_FLOW_ACTION_TYPE_DROP,
2521 DRV_LOG(ERR, "port %u invalid behavior %d",
2523 fdir_filter->action.behavior);
2524 rte_errno = ENOTSUP;
2527 attributes->queue.index = fdir_filter->action.rx_queue;
2529 switch (fdir_filter->input.flow_type) {
2530 case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
2531 case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
2532 case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
2533 attributes->l3.ipv4.hdr = (struct ipv4_hdr){
2534 .src_addr = input->flow.ip4_flow.src_ip,
2535 .dst_addr = input->flow.ip4_flow.dst_ip,
2536 .time_to_live = input->flow.ip4_flow.ttl,
2537 .type_of_service = input->flow.ip4_flow.tos,
2539 attributes->l3_mask.ipv4.hdr = (struct ipv4_hdr){
2540 .src_addr = mask->ipv4_mask.src_ip,
2541 .dst_addr = mask->ipv4_mask.dst_ip,
2542 .time_to_live = mask->ipv4_mask.ttl,
2543 .type_of_service = mask->ipv4_mask.tos,
2544 .next_proto_id = mask->ipv4_mask.proto,
2546 attributes->items[1] = (struct rte_flow_item){
2547 .type = RTE_FLOW_ITEM_TYPE_IPV4,
2548 .spec = &attributes->l3,
2549 .mask = &attributes->l3_mask,
2552 case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
2553 case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
2554 case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
2555 attributes->l3.ipv6.hdr = (struct ipv6_hdr){
2556 .hop_limits = input->flow.ipv6_flow.hop_limits,
2557 .proto = input->flow.ipv6_flow.proto,
2560 memcpy(attributes->l3.ipv6.hdr.src_addr,
2561 input->flow.ipv6_flow.src_ip,
2562 RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
2563 memcpy(attributes->l3.ipv6.hdr.dst_addr,
2564 input->flow.ipv6_flow.dst_ip,
2565 RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
2566 memcpy(attributes->l3_mask.ipv6.hdr.src_addr,
2567 mask->ipv6_mask.src_ip,
2568 RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
2569 memcpy(attributes->l3_mask.ipv6.hdr.dst_addr,
2570 mask->ipv6_mask.dst_ip,
2571 RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
2572 attributes->items[1] = (struct rte_flow_item){
2573 .type = RTE_FLOW_ITEM_TYPE_IPV6,
2574 .spec = &attributes->l3,
2575 .mask = &attributes->l3_mask,
2579 DRV_LOG(ERR, "port %u invalid flow type%d",
2580 dev->data->port_id, fdir_filter->input.flow_type);
2581 rte_errno = ENOTSUP;
2585 switch (fdir_filter->input.flow_type) {
2586 case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
2587 attributes->l4.udp.hdr = (struct udp_hdr){
2588 .src_port = input->flow.udp4_flow.src_port,
2589 .dst_port = input->flow.udp4_flow.dst_port,
2591 attributes->l4_mask.udp.hdr = (struct udp_hdr){
2592 .src_port = mask->src_port_mask,
2593 .dst_port = mask->dst_port_mask,
2595 attributes->items[2] = (struct rte_flow_item){
2596 .type = RTE_FLOW_ITEM_TYPE_UDP,
2597 .spec = &attributes->l4,
2598 .mask = &attributes->l4_mask,
2601 case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
2602 attributes->l4.tcp.hdr = (struct tcp_hdr){
2603 .src_port = input->flow.tcp4_flow.src_port,
2604 .dst_port = input->flow.tcp4_flow.dst_port,
2606 attributes->l4_mask.tcp.hdr = (struct tcp_hdr){
2607 .src_port = mask->src_port_mask,
2608 .dst_port = mask->dst_port_mask,
2610 attributes->items[2] = (struct rte_flow_item){
2611 .type = RTE_FLOW_ITEM_TYPE_TCP,
2612 .spec = &attributes->l4,
2613 .mask = &attributes->l4_mask,
2616 case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
2617 attributes->l4.udp.hdr = (struct udp_hdr){
2618 .src_port = input->flow.udp6_flow.src_port,
2619 .dst_port = input->flow.udp6_flow.dst_port,
2621 attributes->l4_mask.udp.hdr = (struct udp_hdr){
2622 .src_port = mask->src_port_mask,
2623 .dst_port = mask->dst_port_mask,
2625 attributes->items[2] = (struct rte_flow_item){
2626 .type = RTE_FLOW_ITEM_TYPE_UDP,
2627 .spec = &attributes->l4,
2628 .mask = &attributes->l4_mask,
2631 case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
2632 attributes->l4.tcp.hdr = (struct tcp_hdr){
2633 .src_port = input->flow.tcp6_flow.src_port,
2634 .dst_port = input->flow.tcp6_flow.dst_port,
2636 attributes->l4_mask.tcp.hdr = (struct tcp_hdr){
2637 .src_port = mask->src_port_mask,
2638 .dst_port = mask->dst_port_mask,
2640 attributes->items[2] = (struct rte_flow_item){
2641 .type = RTE_FLOW_ITEM_TYPE_TCP,
2642 .spec = &attributes->l4,
2643 .mask = &attributes->l4_mask,
2646 case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
2647 case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
2650 DRV_LOG(ERR, "port %u invalid flow type%d",
2651 dev->data->port_id, fdir_filter->input.flow_type);
2652 rte_errno = ENOTSUP;
2658 #define FLOW_FDIR_CMP(f1, f2, fld) \
2659 memcmp(&(f1)->fld, &(f2)->fld, sizeof(f1->fld))
2662 * Compare two FDIR flows. If items and actions are identical, the two flows are
2666 * Pointer to Ethernet device.
2668 * FDIR flow to compare.
2670 * FDIR flow to compare.
2673 * Zero on match, 1 otherwise.
2676 flow_fdir_cmp(const struct mlx5_fdir *f1, const struct mlx5_fdir *f2)
2678 if (FLOW_FDIR_CMP(f1, f2, attr) ||
2679 FLOW_FDIR_CMP(f1, f2, l2) ||
2680 FLOW_FDIR_CMP(f1, f2, l2_mask) ||
2681 FLOW_FDIR_CMP(f1, f2, l3) ||
2682 FLOW_FDIR_CMP(f1, f2, l3_mask) ||
2683 FLOW_FDIR_CMP(f1, f2, l4) ||
2684 FLOW_FDIR_CMP(f1, f2, l4_mask) ||
2685 FLOW_FDIR_CMP(f1, f2, actions[0].type))
2687 if (f1->actions[0].type == RTE_FLOW_ACTION_TYPE_QUEUE &&
2688 FLOW_FDIR_CMP(f1, f2, queue))
2694 * Search device flow list to find out a matched FDIR flow.
2697 * Pointer to Ethernet device.
2699 * FDIR flow to lookup.
2702 * Pointer of flow if found, NULL otherwise.
2704 static struct rte_flow *
2705 flow_fdir_filter_lookup(struct rte_eth_dev *dev, struct mlx5_fdir *fdir_flow)
2707 struct priv *priv = dev->data->dev_private;
2708 struct rte_flow *flow = NULL;
2711 TAILQ_FOREACH(flow, &priv->flows, next) {
2712 if (flow->fdir && !flow_fdir_cmp(flow->fdir, fdir_flow)) {
2713 DRV_LOG(DEBUG, "port %u found FDIR flow %p",
2714 dev->data->port_id, (void *)flow);
2722 * Add new flow director filter and store it in list.
2725 * Pointer to Ethernet device.
2726 * @param fdir_filter
2727 * Flow director filter to add.
2730 * 0 on success, a negative errno value otherwise and rte_errno is set.
2733 flow_fdir_filter_add(struct rte_eth_dev *dev,
2734 const struct rte_eth_fdir_filter *fdir_filter)
2736 struct priv *priv = dev->data->dev_private;
2737 struct mlx5_fdir *fdir_flow;
2738 struct rte_flow *flow;
2741 fdir_flow = rte_zmalloc(__func__, sizeof(*fdir_flow), 0);
2746 ret = flow_fdir_filter_convert(dev, fdir_filter, fdir_flow);
2749 flow = flow_fdir_filter_lookup(dev, fdir_flow);
2754 flow = flow_list_create(dev, &priv->flows, &fdir_flow->attr,
2755 fdir_flow->items, fdir_flow->actions, NULL);
2758 assert(!flow->fdir);
2759 flow->fdir = fdir_flow;
2760 DRV_LOG(DEBUG, "port %u created FDIR flow %p",
2761 dev->data->port_id, (void *)flow);
2764 rte_free(fdir_flow);
2769 * Delete specific filter.
2772 * Pointer to Ethernet device.
2773 * @param fdir_filter
2774 * Filter to be deleted.
2777 * 0 on success, a negative errno value otherwise and rte_errno is set.
2780 flow_fdir_filter_delete(struct rte_eth_dev *dev,
2781 const struct rte_eth_fdir_filter *fdir_filter)
2783 struct priv *priv = dev->data->dev_private;
2784 struct rte_flow *flow;
2785 struct mlx5_fdir fdir_flow = {
2790 ret = flow_fdir_filter_convert(dev, fdir_filter, &fdir_flow);
2793 flow = flow_fdir_filter_lookup(dev, &fdir_flow);
2798 flow_list_destroy(dev, &priv->flows, flow);
2799 DRV_LOG(DEBUG, "port %u deleted FDIR flow %p",
2800 dev->data->port_id, (void *)flow);
2805 * Update queue for specific filter.
2808 * Pointer to Ethernet device.
2809 * @param fdir_filter
2810 * Filter to be updated.
2813 * 0 on success, a negative errno value otherwise and rte_errno is set.
2816 flow_fdir_filter_update(struct rte_eth_dev *dev,
2817 const struct rte_eth_fdir_filter *fdir_filter)
2821 ret = flow_fdir_filter_delete(dev, fdir_filter);
2824 return flow_fdir_filter_add(dev, fdir_filter);
2828 * Flush all filters.
2831 * Pointer to Ethernet device.
2834 flow_fdir_filter_flush(struct rte_eth_dev *dev)
2836 struct priv *priv = dev->data->dev_private;
2838 mlx5_flow_list_flush(dev, &priv->flows);
2842 * Get flow director information.
2845 * Pointer to Ethernet device.
2846 * @param[out] fdir_info
2847 * Resulting flow director information.
2850 flow_fdir_info_get(struct rte_eth_dev *dev, struct rte_eth_fdir_info *fdir_info)
2852 struct rte_eth_fdir_masks *mask =
2853 &dev->data->dev_conf.fdir_conf.mask;
2855 fdir_info->mode = dev->data->dev_conf.fdir_conf.mode;
2856 fdir_info->guarant_spc = 0;
2857 rte_memcpy(&fdir_info->mask, mask, sizeof(fdir_info->mask));
2858 fdir_info->max_flexpayload = 0;
2859 fdir_info->flow_types_mask[0] = 0;
2860 fdir_info->flex_payload_unit = 0;
2861 fdir_info->max_flex_payload_segment_num = 0;
2862 fdir_info->flex_payload_limit = 0;
2863 memset(&fdir_info->flex_conf, 0, sizeof(fdir_info->flex_conf));
2867 * Deal with flow director operations.
2870 * Pointer to Ethernet device.
2872 * Operation to perform.
2874 * Pointer to operation-specific structure.
2877 * 0 on success, a negative errno value otherwise and rte_errno is set.
2880 flow_fdir_ctrl_func(struct rte_eth_dev *dev, enum rte_filter_op filter_op,
2883 enum rte_fdir_mode fdir_mode =
2884 dev->data->dev_conf.fdir_conf.mode;
2886 if (filter_op == RTE_ETH_FILTER_NOP)
2888 if (fdir_mode != RTE_FDIR_MODE_PERFECT &&
2889 fdir_mode != RTE_FDIR_MODE_PERFECT_MAC_VLAN) {
2890 DRV_LOG(ERR, "port %u flow director mode %d not supported",
2891 dev->data->port_id, fdir_mode);
2895 switch (filter_op) {
2896 case RTE_ETH_FILTER_ADD:
2897 return flow_fdir_filter_add(dev, arg);
2898 case RTE_ETH_FILTER_UPDATE:
2899 return flow_fdir_filter_update(dev, arg);
2900 case RTE_ETH_FILTER_DELETE:
2901 return flow_fdir_filter_delete(dev, arg);
2902 case RTE_ETH_FILTER_FLUSH:
2903 flow_fdir_filter_flush(dev);
2905 case RTE_ETH_FILTER_INFO:
2906 flow_fdir_info_get(dev, arg);
2909 DRV_LOG(DEBUG, "port %u unknown operation %u",
2910 dev->data->port_id, filter_op);
2918 * Manage filter operations.
2921 * Pointer to Ethernet device structure.
2922 * @param filter_type
2925 * Operation to perform.
2927 * Pointer to operation-specific structure.
2930 * 0 on success, a negative errno value otherwise and rte_errno is set.
2933 mlx5_dev_filter_ctrl(struct rte_eth_dev *dev,
2934 enum rte_filter_type filter_type,
2935 enum rte_filter_op filter_op,
2938 switch (filter_type) {
2939 case RTE_ETH_FILTER_GENERIC:
2940 if (filter_op != RTE_ETH_FILTER_GET) {
2944 *(const void **)arg = &mlx5_flow_ops;
2946 case RTE_ETH_FILTER_FDIR:
2947 return flow_fdir_ctrl_func(dev, filter_op, arg);
2949 DRV_LOG(ERR, "port %u filter type (%d) not supported",
2950 dev->data->port_id, filter_type);
2951 rte_errno = ENOTSUP;