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)
318 struct mlx5_priv *priv = dev->data->dev_private;
320 struct ibv_flow_attr attr;
321 struct ibv_flow_spec_eth eth;
322 struct ibv_flow_spec_action_drop drop;
326 .port = (uint8_t)priv->ibv_port,
329 .type = IBV_FLOW_SPEC_ETH,
330 .size = sizeof(struct ibv_flow_spec_eth),
333 .size = sizeof(struct ibv_flow_spec_action_drop),
334 .type = IBV_FLOW_SPEC_ACTION_DROP,
337 struct ibv_flow *flow;
338 struct mlx5_hrxq *drop = mlx5_hrxq_drop_new(dev);
339 uint16_t vprio[] = { 8, 16 };
347 for (i = 0; i != RTE_DIM(vprio); i++) {
348 flow_attr.attr.priority = vprio[i] - 1;
349 flow = mlx5_glue->create_flow(drop->qp, &flow_attr.attr);
352 claim_zero(mlx5_glue->destroy_flow(flow));
355 mlx5_hrxq_drop_release(dev);
358 priority = RTE_DIM(priority_map_3);
361 priority = RTE_DIM(priority_map_5);
366 "port %u verbs maximum priority: %d expected 8/16",
367 dev->data->port_id, priority);
370 DRV_LOG(INFO, "port %u flow maximum priority: %d",
371 dev->data->port_id, priority);
376 * Adjust flow priority based on the highest layer and the request priority.
379 * Pointer to the Ethernet device structure.
380 * @param[in] priority
381 * The rule base priority.
382 * @param[in] subpriority
383 * The priority based on the items.
388 uint32_t mlx5_flow_adjust_priority(struct rte_eth_dev *dev, int32_t priority,
389 uint32_t subpriority)
392 struct mlx5_priv *priv = dev->data->dev_private;
394 switch (priv->config.flow_prio) {
395 case RTE_DIM(priority_map_3):
396 res = priority_map_3[priority][subpriority];
398 case RTE_DIM(priority_map_5):
399 res = priority_map_5[priority][subpriority];
406 * Verify the @p item specifications (spec, last, mask) are compatible with the
410 * Item specification.
412 * @p item->mask or flow default bit-masks.
413 * @param[in] nic_mask
414 * Bit-masks covering supported fields by the NIC to compare with user mask.
416 * Bit-masks size in bytes.
418 * Pointer to error structure.
421 * 0 on success, a negative errno value otherwise and rte_errno is set.
424 mlx5_flow_item_acceptable(const struct rte_flow_item *item,
426 const uint8_t *nic_mask,
428 struct rte_flow_error *error)
433 for (i = 0; i < size; ++i)
434 if ((nic_mask[i] | mask[i]) != nic_mask[i])
435 return rte_flow_error_set(error, ENOTSUP,
436 RTE_FLOW_ERROR_TYPE_ITEM,
438 "mask enables non supported"
440 if (!item->spec && (item->mask || item->last))
441 return rte_flow_error_set(error, EINVAL,
442 RTE_FLOW_ERROR_TYPE_ITEM, item,
443 "mask/last without a spec is not"
445 if (item->spec && item->last) {
451 for (i = 0; i < size; ++i) {
452 spec[i] = ((const uint8_t *)item->spec)[i] & mask[i];
453 last[i] = ((const uint8_t *)item->last)[i] & mask[i];
455 ret = memcmp(spec, last, size);
457 return rte_flow_error_set(error, EINVAL,
458 RTE_FLOW_ERROR_TYPE_ITEM,
460 "range is not valid");
466 * Adjust the hash fields according to the @p flow information.
468 * @param[in] dev_flow.
469 * Pointer to the mlx5_flow.
471 * 1 when the hash field is for a tunnel item.
472 * @param[in] layer_types
474 * @param[in] hash_fields
478 * The hash fileds that should be used.
481 mlx5_flow_hashfields_adjust(struct mlx5_flow *dev_flow,
482 int tunnel __rte_unused, uint64_t layer_types,
483 uint64_t hash_fields)
485 struct rte_flow *flow = dev_flow->flow;
486 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
487 int rss_request_inner = flow->rss.level >= 2;
489 /* Check RSS hash level for tunnel. */
490 if (tunnel && rss_request_inner)
491 hash_fields |= IBV_RX_HASH_INNER;
492 else if (tunnel || rss_request_inner)
495 /* Check if requested layer matches RSS hash fields. */
496 if (!(flow->rss.types & layer_types))
502 * Lookup and set the ptype in the data Rx part. A single Ptype can be used,
503 * if several tunnel rules are used on this queue, the tunnel ptype will be
507 * Rx queue to update.
510 flow_rxq_tunnel_ptype_update(struct mlx5_rxq_ctrl *rxq_ctrl)
513 uint32_t tunnel_ptype = 0;
515 /* Look up for the ptype to use. */
516 for (i = 0; i != MLX5_FLOW_TUNNEL; ++i) {
517 if (!rxq_ctrl->flow_tunnels_n[i])
520 tunnel_ptype = tunnels_info[i].ptype;
526 rxq_ctrl->rxq.tunnel = tunnel_ptype;
530 * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) according to the devive
534 * Pointer to the Ethernet device structure.
535 * @param[in] dev_flow
536 * Pointer to device flow structure.
539 flow_drv_rxq_flags_set(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow)
541 struct mlx5_priv *priv = dev->data->dev_private;
542 struct rte_flow *flow = dev_flow->flow;
543 const int mark = !!(flow->actions &
544 (MLX5_FLOW_ACTION_FLAG | MLX5_FLOW_ACTION_MARK));
545 const int tunnel = !!(dev_flow->layers & MLX5_FLOW_LAYER_TUNNEL);
548 for (i = 0; i != flow->rss.queue_num; ++i) {
549 int idx = (*flow->queue)[i];
550 struct mlx5_rxq_ctrl *rxq_ctrl =
551 container_of((*priv->rxqs)[idx],
552 struct mlx5_rxq_ctrl, rxq);
555 rxq_ctrl->rxq.mark = 1;
556 rxq_ctrl->flow_mark_n++;
561 /* Increase the counter matching the flow. */
562 for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
563 if ((tunnels_info[j].tunnel &
565 tunnels_info[j].tunnel) {
566 rxq_ctrl->flow_tunnels_n[j]++;
570 flow_rxq_tunnel_ptype_update(rxq_ctrl);
576 * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) for a flow
579 * Pointer to the Ethernet device structure.
581 * Pointer to flow structure.
584 flow_rxq_flags_set(struct rte_eth_dev *dev, struct rte_flow *flow)
586 struct mlx5_flow *dev_flow;
588 LIST_FOREACH(dev_flow, &flow->dev_flows, next)
589 flow_drv_rxq_flags_set(dev, dev_flow);
593 * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
594 * device flow if no other flow uses it with the same kind of request.
597 * Pointer to Ethernet device.
598 * @param[in] dev_flow
599 * Pointer to the device flow.
602 flow_drv_rxq_flags_trim(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow)
604 struct mlx5_priv *priv = dev->data->dev_private;
605 struct rte_flow *flow = dev_flow->flow;
606 const int mark = !!(flow->actions &
607 (MLX5_FLOW_ACTION_FLAG | MLX5_FLOW_ACTION_MARK));
608 const int tunnel = !!(dev_flow->layers & MLX5_FLOW_LAYER_TUNNEL);
611 assert(dev->data->dev_started);
612 for (i = 0; i != flow->rss.queue_num; ++i) {
613 int idx = (*flow->queue)[i];
614 struct mlx5_rxq_ctrl *rxq_ctrl =
615 container_of((*priv->rxqs)[idx],
616 struct mlx5_rxq_ctrl, rxq);
619 rxq_ctrl->flow_mark_n--;
620 rxq_ctrl->rxq.mark = !!rxq_ctrl->flow_mark_n;
625 /* Decrease the counter matching the flow. */
626 for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
627 if ((tunnels_info[j].tunnel &
629 tunnels_info[j].tunnel) {
630 rxq_ctrl->flow_tunnels_n[j]--;
634 flow_rxq_tunnel_ptype_update(rxq_ctrl);
640 * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
641 * @p flow if no other flow uses it with the same kind of request.
644 * Pointer to Ethernet device.
646 * Pointer to the flow.
649 flow_rxq_flags_trim(struct rte_eth_dev *dev, struct rte_flow *flow)
651 struct mlx5_flow *dev_flow;
653 LIST_FOREACH(dev_flow, &flow->dev_flows, next)
654 flow_drv_rxq_flags_trim(dev, dev_flow);
658 * Clear the Mark/Flag and Tunnel ptype information in all Rx queues.
661 * Pointer to Ethernet device.
664 flow_rxq_flags_clear(struct rte_eth_dev *dev)
666 struct mlx5_priv *priv = dev->data->dev_private;
669 for (i = 0; i != priv->rxqs_n; ++i) {
670 struct mlx5_rxq_ctrl *rxq_ctrl;
673 if (!(*priv->rxqs)[i])
675 rxq_ctrl = container_of((*priv->rxqs)[i],
676 struct mlx5_rxq_ctrl, rxq);
677 rxq_ctrl->flow_mark_n = 0;
678 rxq_ctrl->rxq.mark = 0;
679 for (j = 0; j != MLX5_FLOW_TUNNEL; ++j)
680 rxq_ctrl->flow_tunnels_n[j] = 0;
681 rxq_ctrl->rxq.tunnel = 0;
686 * Validate the flag action.
688 * @param[in] action_flags
689 * Bit-fields that holds the actions detected until now.
691 * Attributes of flow that includes this action.
693 * Pointer to error structure.
696 * 0 on success, a negative errno value otherwise and rte_errno is set.
699 mlx5_flow_validate_action_flag(uint64_t action_flags,
700 const struct rte_flow_attr *attr,
701 struct rte_flow_error *error)
704 if (action_flags & MLX5_FLOW_ACTION_DROP)
705 return rte_flow_error_set(error, EINVAL,
706 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
707 "can't drop and flag in same flow");
708 if (action_flags & MLX5_FLOW_ACTION_MARK)
709 return rte_flow_error_set(error, EINVAL,
710 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
711 "can't mark and flag in same flow");
712 if (action_flags & MLX5_FLOW_ACTION_FLAG)
713 return rte_flow_error_set(error, EINVAL,
714 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
716 " actions in same flow");
718 return rte_flow_error_set(error, ENOTSUP,
719 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
720 "flag action not supported for "
726 * Validate the mark action.
729 * Pointer to the queue action.
730 * @param[in] action_flags
731 * Bit-fields that holds the actions detected until now.
733 * Attributes of flow that includes this action.
735 * Pointer to error structure.
738 * 0 on success, a negative errno value otherwise and rte_errno is set.
741 mlx5_flow_validate_action_mark(const struct rte_flow_action *action,
742 uint64_t action_flags,
743 const struct rte_flow_attr *attr,
744 struct rte_flow_error *error)
746 const struct rte_flow_action_mark *mark = action->conf;
749 return rte_flow_error_set(error, EINVAL,
750 RTE_FLOW_ERROR_TYPE_ACTION,
752 "configuration cannot be null");
753 if (mark->id >= MLX5_FLOW_MARK_MAX)
754 return rte_flow_error_set(error, EINVAL,
755 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
757 "mark id must in 0 <= id < "
758 RTE_STR(MLX5_FLOW_MARK_MAX));
759 if (action_flags & MLX5_FLOW_ACTION_DROP)
760 return rte_flow_error_set(error, EINVAL,
761 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
762 "can't drop and mark in same flow");
763 if (action_flags & MLX5_FLOW_ACTION_FLAG)
764 return rte_flow_error_set(error, EINVAL,
765 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
766 "can't flag and mark in same flow");
767 if (action_flags & MLX5_FLOW_ACTION_MARK)
768 return rte_flow_error_set(error, EINVAL,
769 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
770 "can't have 2 mark actions in same"
773 return rte_flow_error_set(error, ENOTSUP,
774 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
775 "mark action not supported for "
781 * Validate the drop action.
783 * @param[in] action_flags
784 * Bit-fields that holds the actions detected until now.
786 * Attributes of flow that includes this action.
788 * Pointer to error structure.
791 * 0 on success, a negative errno value otherwise and rte_errno is set.
794 mlx5_flow_validate_action_drop(uint64_t action_flags,
795 const struct rte_flow_attr *attr,
796 struct rte_flow_error *error)
798 if (action_flags & MLX5_FLOW_ACTION_FLAG)
799 return rte_flow_error_set(error, EINVAL,
800 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
801 "can't drop and flag in same flow");
802 if (action_flags & MLX5_FLOW_ACTION_MARK)
803 return rte_flow_error_set(error, EINVAL,
804 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
805 "can't drop and mark in same flow");
806 if (action_flags & MLX5_FLOW_FATE_ACTIONS)
807 return rte_flow_error_set(error, EINVAL,
808 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
809 "can't have 2 fate actions in"
812 return rte_flow_error_set(error, ENOTSUP,
813 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
814 "drop action not supported for "
820 * Validate the queue action.
823 * Pointer to the queue action.
824 * @param[in] action_flags
825 * Bit-fields that holds the actions detected until now.
827 * Pointer to the Ethernet device structure.
829 * Attributes of flow that includes this action.
831 * Pointer to error structure.
834 * 0 on success, a negative errno value otherwise and rte_errno is set.
837 mlx5_flow_validate_action_queue(const struct rte_flow_action *action,
838 uint64_t action_flags,
839 struct rte_eth_dev *dev,
840 const struct rte_flow_attr *attr,
841 struct rte_flow_error *error)
843 struct mlx5_priv *priv = dev->data->dev_private;
844 const struct rte_flow_action_queue *queue = action->conf;
846 if (action_flags & MLX5_FLOW_FATE_ACTIONS)
847 return rte_flow_error_set(error, EINVAL,
848 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
849 "can't have 2 fate actions in"
852 return rte_flow_error_set(error, EINVAL,
853 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
854 NULL, "No Rx queues configured");
855 if (queue->index >= priv->rxqs_n)
856 return rte_flow_error_set(error, EINVAL,
857 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
859 "queue index out of range");
860 if (!(*priv->rxqs)[queue->index])
861 return rte_flow_error_set(error, EINVAL,
862 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
864 "queue is not configured");
866 return rte_flow_error_set(error, ENOTSUP,
867 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
868 "queue action not supported for "
874 * Validate the rss action.
877 * Pointer to the queue action.
878 * @param[in] action_flags
879 * Bit-fields that holds the actions detected until now.
881 * Pointer to the Ethernet device structure.
883 * Attributes of flow that includes this action.
885 * Pointer to error structure.
888 * 0 on success, a negative errno value otherwise and rte_errno is set.
891 mlx5_flow_validate_action_rss(const struct rte_flow_action *action,
892 uint64_t action_flags,
893 struct rte_eth_dev *dev,
894 const struct rte_flow_attr *attr,
895 struct rte_flow_error *error)
897 struct mlx5_priv *priv = dev->data->dev_private;
898 const struct rte_flow_action_rss *rss = action->conf;
901 if (action_flags & MLX5_FLOW_FATE_ACTIONS)
902 return rte_flow_error_set(error, EINVAL,
903 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
904 "can't have 2 fate actions"
906 if (rss->func != RTE_ETH_HASH_FUNCTION_DEFAULT &&
907 rss->func != RTE_ETH_HASH_FUNCTION_TOEPLITZ)
908 return rte_flow_error_set(error, ENOTSUP,
909 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
911 "RSS hash function not supported");
912 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
917 return rte_flow_error_set(error, ENOTSUP,
918 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
920 "tunnel RSS is not supported");
921 /* allow RSS key_len 0 in case of NULL (default) RSS key. */
922 if (rss->key_len == 0 && rss->key != NULL)
923 return rte_flow_error_set(error, ENOTSUP,
924 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
926 "RSS hash key length 0");
927 if (rss->key_len > 0 && rss->key_len < MLX5_RSS_HASH_KEY_LEN)
928 return rte_flow_error_set(error, ENOTSUP,
929 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
931 "RSS hash key too small");
932 if (rss->key_len > MLX5_RSS_HASH_KEY_LEN)
933 return rte_flow_error_set(error, ENOTSUP,
934 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
936 "RSS hash key too large");
937 if (rss->queue_num > priv->config.ind_table_max_size)
938 return rte_flow_error_set(error, ENOTSUP,
939 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
941 "number of queues too large");
942 if (rss->types & MLX5_RSS_HF_MASK)
943 return rte_flow_error_set(error, ENOTSUP,
944 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
946 "some RSS protocols are not"
949 return rte_flow_error_set(error, EINVAL,
950 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
951 NULL, "No Rx queues configured");
953 return rte_flow_error_set(error, EINVAL,
954 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
955 NULL, "No queues configured");
956 for (i = 0; i != rss->queue_num; ++i) {
957 if (!(*priv->rxqs)[rss->queue[i]])
958 return rte_flow_error_set
959 (error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
960 &rss->queue[i], "queue is not configured");
963 return rte_flow_error_set(error, ENOTSUP,
964 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
965 "rss action not supported for "
971 * Validate the count action.
974 * Pointer to the Ethernet device structure.
976 * Attributes of flow that includes this action.
978 * Pointer to error structure.
981 * 0 on success, a negative errno value otherwise and rte_errno is set.
984 mlx5_flow_validate_action_count(struct rte_eth_dev *dev __rte_unused,
985 const struct rte_flow_attr *attr,
986 struct rte_flow_error *error)
989 return rte_flow_error_set(error, ENOTSUP,
990 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
991 "count action not supported for "
997 * Verify the @p attributes will be correctly understood by the NIC and store
998 * them in the @p flow if everything is correct.
1001 * Pointer to the Ethernet device structure.
1002 * @param[in] attributes
1003 * Pointer to flow attributes
1005 * Pointer to error structure.
1008 * 0 on success, a negative errno value otherwise and rte_errno is set.
1011 mlx5_flow_validate_attributes(struct rte_eth_dev *dev,
1012 const struct rte_flow_attr *attributes,
1013 struct rte_flow_error *error)
1015 struct mlx5_priv *priv = dev->data->dev_private;
1016 uint32_t priority_max = priv->config.flow_prio - 1;
1018 if (attributes->group)
1019 return rte_flow_error_set(error, ENOTSUP,
1020 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
1021 NULL, "groups is not supported");
1022 if (attributes->priority != MLX5_FLOW_PRIO_RSVD &&
1023 attributes->priority >= priority_max)
1024 return rte_flow_error_set(error, ENOTSUP,
1025 RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
1026 NULL, "priority out of range");
1027 if (attributes->egress)
1028 return rte_flow_error_set(error, ENOTSUP,
1029 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1030 "egress is not supported");
1031 if (attributes->transfer)
1032 return rte_flow_error_set(error, ENOTSUP,
1033 RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1034 NULL, "transfer is not supported");
1035 if (!attributes->ingress)
1036 return rte_flow_error_set(error, EINVAL,
1037 RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
1039 "ingress attribute is mandatory");
1044 * Validate Ethernet item.
1047 * Item specification.
1048 * @param[in] item_flags
1049 * Bit-fields that holds the items detected until now.
1051 * Pointer to error structure.
1054 * 0 on success, a negative errno value otherwise and rte_errno is set.
1057 mlx5_flow_validate_item_eth(const struct rte_flow_item *item,
1058 uint64_t item_flags,
1059 struct rte_flow_error *error)
1061 const struct rte_flow_item_eth *mask = item->mask;
1062 const struct rte_flow_item_eth nic_mask = {
1063 .dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1064 .src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1065 .type = RTE_BE16(0xffff),
1068 int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1069 const uint64_t ethm = tunnel ? MLX5_FLOW_LAYER_INNER_L2 :
1070 MLX5_FLOW_LAYER_OUTER_L2;
1072 if (item_flags & ethm)
1073 return rte_flow_error_set(error, ENOTSUP,
1074 RTE_FLOW_ERROR_TYPE_ITEM, item,
1075 "multiple L2 layers not supported");
1077 mask = &rte_flow_item_eth_mask;
1078 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1079 (const uint8_t *)&nic_mask,
1080 sizeof(struct rte_flow_item_eth),
1086 * Validate VLAN item.
1089 * Item specification.
1090 * @param[in] item_flags
1091 * Bit-fields that holds the items detected until now.
1093 * Pointer to error structure.
1096 * 0 on success, a negative errno value otherwise and rte_errno is set.
1099 mlx5_flow_validate_item_vlan(const struct rte_flow_item *item,
1100 uint64_t item_flags,
1101 struct rte_flow_error *error)
1103 const struct rte_flow_item_vlan *spec = item->spec;
1104 const struct rte_flow_item_vlan *mask = item->mask;
1105 const struct rte_flow_item_vlan nic_mask = {
1106 .tci = RTE_BE16(0x0fff),
1107 .inner_type = RTE_BE16(0xffff),
1109 uint16_t vlan_tag = 0;
1110 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1112 const uint64_t l34m = tunnel ? (MLX5_FLOW_LAYER_INNER_L3 |
1113 MLX5_FLOW_LAYER_INNER_L4) :
1114 (MLX5_FLOW_LAYER_OUTER_L3 |
1115 MLX5_FLOW_LAYER_OUTER_L4);
1116 const uint64_t vlanm = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
1117 MLX5_FLOW_LAYER_OUTER_VLAN;
1119 if (item_flags & vlanm)
1120 return rte_flow_error_set(error, EINVAL,
1121 RTE_FLOW_ERROR_TYPE_ITEM, item,
1122 "multiple VLAN layers not supported");
1123 else if ((item_flags & l34m) != 0)
1124 return rte_flow_error_set(error, EINVAL,
1125 RTE_FLOW_ERROR_TYPE_ITEM, item,
1126 "L2 layer cannot follow L3/L4 layer");
1128 mask = &rte_flow_item_vlan_mask;
1129 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1130 (const uint8_t *)&nic_mask,
1131 sizeof(struct rte_flow_item_vlan),
1136 vlan_tag = spec->tci;
1137 vlan_tag &= mask->tci;
1140 * From verbs perspective an empty VLAN is equivalent
1141 * to a packet without VLAN layer.
1144 return rte_flow_error_set(error, EINVAL,
1145 RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
1147 "VLAN cannot be empty");
1152 * Validate IPV4 item.
1155 * Item specification.
1156 * @param[in] item_flags
1157 * Bit-fields that holds the items detected until now.
1158 * @param[in] acc_mask
1159 * Acceptable mask, if NULL default internal default mask
1160 * will be used to check whether item fields are supported.
1162 * Pointer to error structure.
1165 * 0 on success, a negative errno value otherwise and rte_errno is set.
1168 mlx5_flow_validate_item_ipv4(const struct rte_flow_item *item,
1169 uint64_t item_flags,
1170 const struct rte_flow_item_ipv4 *acc_mask,
1171 struct rte_flow_error *error)
1173 const struct rte_flow_item_ipv4 *mask = item->mask;
1174 const struct rte_flow_item_ipv4 nic_mask = {
1176 .src_addr = RTE_BE32(0xffffffff),
1177 .dst_addr = RTE_BE32(0xffffffff),
1178 .type_of_service = 0xff,
1179 .next_proto_id = 0xff,
1182 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1183 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1184 MLX5_FLOW_LAYER_OUTER_L3;
1185 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1186 MLX5_FLOW_LAYER_OUTER_L4;
1189 if (item_flags & l3m)
1190 return rte_flow_error_set(error, ENOTSUP,
1191 RTE_FLOW_ERROR_TYPE_ITEM, item,
1192 "multiple L3 layers not supported");
1193 else if (item_flags & l4m)
1194 return rte_flow_error_set(error, EINVAL,
1195 RTE_FLOW_ERROR_TYPE_ITEM, item,
1196 "L3 cannot follow an L4 layer.");
1198 mask = &rte_flow_item_ipv4_mask;
1199 else if (mask->hdr.next_proto_id != 0 &&
1200 mask->hdr.next_proto_id != 0xff)
1201 return rte_flow_error_set(error, EINVAL,
1202 RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
1203 "partial mask is not supported"
1205 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1206 acc_mask ? (const uint8_t *)acc_mask
1207 : (const uint8_t *)&nic_mask,
1208 sizeof(struct rte_flow_item_ipv4),
1216 * Validate IPV6 item.
1219 * Item specification.
1220 * @param[in] item_flags
1221 * Bit-fields that holds the items detected until now.
1222 * @param[in] acc_mask
1223 * Acceptable mask, if NULL default internal default mask
1224 * will be used to check whether item fields are supported.
1226 * Pointer to error structure.
1229 * 0 on success, a negative errno value otherwise and rte_errno is set.
1232 mlx5_flow_validate_item_ipv6(const struct rte_flow_item *item,
1233 uint64_t item_flags,
1234 const struct rte_flow_item_ipv6 *acc_mask,
1235 struct rte_flow_error *error)
1237 const struct rte_flow_item_ipv6 *mask = item->mask;
1238 const struct rte_flow_item_ipv6 nic_mask = {
1241 "\xff\xff\xff\xff\xff\xff\xff\xff"
1242 "\xff\xff\xff\xff\xff\xff\xff\xff",
1244 "\xff\xff\xff\xff\xff\xff\xff\xff"
1245 "\xff\xff\xff\xff\xff\xff\xff\xff",
1246 .vtc_flow = RTE_BE32(0xffffffff),
1251 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1252 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1253 MLX5_FLOW_LAYER_OUTER_L3;
1254 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1255 MLX5_FLOW_LAYER_OUTER_L4;
1258 if (item_flags & l3m)
1259 return rte_flow_error_set(error, ENOTSUP,
1260 RTE_FLOW_ERROR_TYPE_ITEM, item,
1261 "multiple L3 layers not supported");
1262 else if (item_flags & l4m)
1263 return rte_flow_error_set(error, EINVAL,
1264 RTE_FLOW_ERROR_TYPE_ITEM, item,
1265 "L3 cannot follow an L4 layer.");
1267 mask = &rte_flow_item_ipv6_mask;
1268 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1269 acc_mask ? (const uint8_t *)acc_mask
1270 : (const uint8_t *)&nic_mask,
1271 sizeof(struct rte_flow_item_ipv6),
1279 * Validate UDP item.
1282 * Item specification.
1283 * @param[in] item_flags
1284 * Bit-fields that holds the items detected until now.
1285 * @param[in] target_protocol
1286 * The next protocol in the previous item.
1287 * @param[in] flow_mask
1288 * mlx5 flow-specific (TCF, DV, verbs, etc.) supported header fields mask.
1290 * Pointer to error structure.
1293 * 0 on success, a negative errno value otherwise and rte_errno is set.
1296 mlx5_flow_validate_item_udp(const struct rte_flow_item *item,
1297 uint64_t item_flags,
1298 uint8_t target_protocol,
1299 struct rte_flow_error *error)
1301 const struct rte_flow_item_udp *mask = item->mask;
1302 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1303 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1304 MLX5_FLOW_LAYER_OUTER_L3;
1305 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1306 MLX5_FLOW_LAYER_OUTER_L4;
1309 if (target_protocol != 0xff && target_protocol != IPPROTO_UDP)
1310 return rte_flow_error_set(error, EINVAL,
1311 RTE_FLOW_ERROR_TYPE_ITEM, item,
1312 "protocol filtering not compatible"
1314 if (!(item_flags & l3m))
1315 return rte_flow_error_set(error, EINVAL,
1316 RTE_FLOW_ERROR_TYPE_ITEM, item,
1317 "L3 is mandatory to filter on L4");
1318 if (item_flags & l4m)
1319 return rte_flow_error_set(error, EINVAL,
1320 RTE_FLOW_ERROR_TYPE_ITEM, item,
1321 "multiple L4 layers not supported");
1323 mask = &rte_flow_item_udp_mask;
1324 ret = mlx5_flow_item_acceptable
1325 (item, (const uint8_t *)mask,
1326 (const uint8_t *)&rte_flow_item_udp_mask,
1327 sizeof(struct rte_flow_item_udp), error);
1334 * Validate TCP item.
1337 * Item specification.
1338 * @param[in] item_flags
1339 * Bit-fields that holds the items detected until now.
1340 * @param[in] target_protocol
1341 * The next protocol in the previous item.
1343 * Pointer to error structure.
1346 * 0 on success, a negative errno value otherwise and rte_errno is set.
1349 mlx5_flow_validate_item_tcp(const struct rte_flow_item *item,
1350 uint64_t item_flags,
1351 uint8_t target_protocol,
1352 const struct rte_flow_item_tcp *flow_mask,
1353 struct rte_flow_error *error)
1355 const struct rte_flow_item_tcp *mask = item->mask;
1356 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1357 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1358 MLX5_FLOW_LAYER_OUTER_L3;
1359 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1360 MLX5_FLOW_LAYER_OUTER_L4;
1364 if (target_protocol != 0xff && target_protocol != IPPROTO_TCP)
1365 return rte_flow_error_set(error, EINVAL,
1366 RTE_FLOW_ERROR_TYPE_ITEM, item,
1367 "protocol filtering not compatible"
1369 if (!(item_flags & l3m))
1370 return rte_flow_error_set(error, EINVAL,
1371 RTE_FLOW_ERROR_TYPE_ITEM, item,
1372 "L3 is mandatory to filter on L4");
1373 if (item_flags & l4m)
1374 return rte_flow_error_set(error, EINVAL,
1375 RTE_FLOW_ERROR_TYPE_ITEM, item,
1376 "multiple L4 layers not supported");
1378 mask = &rte_flow_item_tcp_mask;
1379 ret = mlx5_flow_item_acceptable
1380 (item, (const uint8_t *)mask,
1381 (const uint8_t *)flow_mask,
1382 sizeof(struct rte_flow_item_tcp), error);
1389 * Validate VXLAN item.
1392 * Item specification.
1393 * @param[in] item_flags
1394 * Bit-fields that holds the items detected until now.
1395 * @param[in] target_protocol
1396 * The next protocol in the previous item.
1398 * Pointer to error structure.
1401 * 0 on success, a negative errno value otherwise and rte_errno is set.
1404 mlx5_flow_validate_item_vxlan(const struct rte_flow_item *item,
1405 uint64_t item_flags,
1406 struct rte_flow_error *error)
1408 const struct rte_flow_item_vxlan *spec = item->spec;
1409 const struct rte_flow_item_vxlan *mask = item->mask;
1414 } id = { .vlan_id = 0, };
1415 uint32_t vlan_id = 0;
1418 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1419 return rte_flow_error_set(error, ENOTSUP,
1420 RTE_FLOW_ERROR_TYPE_ITEM, item,
1421 "multiple tunnel layers not"
1424 * Verify only UDPv4 is present as defined in
1425 * https://tools.ietf.org/html/rfc7348
1427 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
1428 return rte_flow_error_set(error, EINVAL,
1429 RTE_FLOW_ERROR_TYPE_ITEM, item,
1430 "no outer UDP layer found");
1432 mask = &rte_flow_item_vxlan_mask;
1433 ret = mlx5_flow_item_acceptable
1434 (item, (const uint8_t *)mask,
1435 (const uint8_t *)&rte_flow_item_vxlan_mask,
1436 sizeof(struct rte_flow_item_vxlan),
1441 memcpy(&id.vni[1], spec->vni, 3);
1442 vlan_id = id.vlan_id;
1443 memcpy(&id.vni[1], mask->vni, 3);
1444 vlan_id &= id.vlan_id;
1447 * Tunnel id 0 is equivalent as not adding a VXLAN layer, if
1448 * only this layer is defined in the Verbs specification it is
1449 * interpreted as wildcard and all packets will match this
1450 * rule, if it follows a full stack layer (ex: eth / ipv4 /
1451 * udp), all packets matching the layers before will also
1452 * match this rule. To avoid such situation, VNI 0 is
1453 * currently refused.
1456 return rte_flow_error_set(error, ENOTSUP,
1457 RTE_FLOW_ERROR_TYPE_ITEM, item,
1458 "VXLAN vni cannot be 0");
1459 if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
1460 return rte_flow_error_set(error, ENOTSUP,
1461 RTE_FLOW_ERROR_TYPE_ITEM, item,
1462 "VXLAN tunnel must be fully defined");
1467 * Validate VXLAN_GPE item.
1470 * Item specification.
1471 * @param[in] item_flags
1472 * Bit-fields that holds the items detected until now.
1474 * Pointer to the private data structure.
1475 * @param[in] target_protocol
1476 * The next protocol in the previous item.
1478 * Pointer to error structure.
1481 * 0 on success, a negative errno value otherwise and rte_errno is set.
1484 mlx5_flow_validate_item_vxlan_gpe(const struct rte_flow_item *item,
1485 uint64_t item_flags,
1486 struct rte_eth_dev *dev,
1487 struct rte_flow_error *error)
1489 struct mlx5_priv *priv = dev->data->dev_private;
1490 const struct rte_flow_item_vxlan_gpe *spec = item->spec;
1491 const struct rte_flow_item_vxlan_gpe *mask = item->mask;
1496 } id = { .vlan_id = 0, };
1497 uint32_t vlan_id = 0;
1499 if (!priv->config.l3_vxlan_en)
1500 return rte_flow_error_set(error, ENOTSUP,
1501 RTE_FLOW_ERROR_TYPE_ITEM, item,
1502 "L3 VXLAN is not enabled by device"
1503 " parameter and/or not configured in"
1505 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1506 return rte_flow_error_set(error, ENOTSUP,
1507 RTE_FLOW_ERROR_TYPE_ITEM, item,
1508 "multiple tunnel layers not"
1511 * Verify only UDPv4 is present as defined in
1512 * https://tools.ietf.org/html/rfc7348
1514 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
1515 return rte_flow_error_set(error, EINVAL,
1516 RTE_FLOW_ERROR_TYPE_ITEM, item,
1517 "no outer UDP layer found");
1519 mask = &rte_flow_item_vxlan_gpe_mask;
1520 ret = mlx5_flow_item_acceptable
1521 (item, (const uint8_t *)mask,
1522 (const uint8_t *)&rte_flow_item_vxlan_gpe_mask,
1523 sizeof(struct rte_flow_item_vxlan_gpe),
1529 return rte_flow_error_set(error, ENOTSUP,
1530 RTE_FLOW_ERROR_TYPE_ITEM,
1532 "VxLAN-GPE protocol"
1534 memcpy(&id.vni[1], spec->vni, 3);
1535 vlan_id = id.vlan_id;
1536 memcpy(&id.vni[1], mask->vni, 3);
1537 vlan_id &= id.vlan_id;
1540 * Tunnel id 0 is equivalent as not adding a VXLAN layer, if only this
1541 * layer is defined in the Verbs specification it is interpreted as
1542 * wildcard and all packets will match this rule, if it follows a full
1543 * stack layer (ex: eth / ipv4 / udp), all packets matching the layers
1544 * before will also match this rule. To avoid such situation, VNI 0
1545 * is currently refused.
1548 return rte_flow_error_set(error, ENOTSUP,
1549 RTE_FLOW_ERROR_TYPE_ITEM, item,
1550 "VXLAN-GPE vni cannot be 0");
1551 if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
1552 return rte_flow_error_set(error, ENOTSUP,
1553 RTE_FLOW_ERROR_TYPE_ITEM, item,
1554 "VXLAN-GPE tunnel must be fully"
1560 * Validate GRE item.
1563 * Item specification.
1564 * @param[in] item_flags
1565 * Bit flags to mark detected items.
1566 * @param[in] target_protocol
1567 * The next protocol in the previous item.
1569 * Pointer to error structure.
1572 * 0 on success, a negative errno value otherwise and rte_errno is set.
1575 mlx5_flow_validate_item_gre(const struct rte_flow_item *item,
1576 uint64_t item_flags,
1577 uint8_t target_protocol,
1578 struct rte_flow_error *error)
1580 const struct rte_flow_item_gre *spec __rte_unused = item->spec;
1581 const struct rte_flow_item_gre *mask = item->mask;
1584 if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
1585 return rte_flow_error_set(error, EINVAL,
1586 RTE_FLOW_ERROR_TYPE_ITEM, item,
1587 "protocol filtering not compatible"
1588 " with this GRE layer");
1589 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1590 return rte_flow_error_set(error, ENOTSUP,
1591 RTE_FLOW_ERROR_TYPE_ITEM, item,
1592 "multiple tunnel layers not"
1594 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
1595 return rte_flow_error_set(error, ENOTSUP,
1596 RTE_FLOW_ERROR_TYPE_ITEM, item,
1597 "L3 Layer is missing");
1599 mask = &rte_flow_item_gre_mask;
1600 ret = mlx5_flow_item_acceptable
1601 (item, (const uint8_t *)mask,
1602 (const uint8_t *)&rte_flow_item_gre_mask,
1603 sizeof(struct rte_flow_item_gre), error);
1606 #ifndef HAVE_IBV_DEVICE_MPLS_SUPPORT
1607 if (spec && (spec->protocol & mask->protocol))
1608 return rte_flow_error_set(error, ENOTSUP,
1609 RTE_FLOW_ERROR_TYPE_ITEM, item,
1610 "without MPLS support the"
1611 " specification cannot be used for"
1618 * Validate MPLS item.
1621 * Pointer to the rte_eth_dev structure.
1623 * Item specification.
1624 * @param[in] item_flags
1625 * Bit-fields that holds the items detected until now.
1626 * @param[in] prev_layer
1627 * The protocol layer indicated in previous item.
1629 * Pointer to error structure.
1632 * 0 on success, a negative errno value otherwise and rte_errno is set.
1635 mlx5_flow_validate_item_mpls(struct rte_eth_dev *dev __rte_unused,
1636 const struct rte_flow_item *item __rte_unused,
1637 uint64_t item_flags __rte_unused,
1638 uint64_t prev_layer __rte_unused,
1639 struct rte_flow_error *error)
1641 #ifdef HAVE_IBV_DEVICE_MPLS_SUPPORT
1642 const struct rte_flow_item_mpls *mask = item->mask;
1643 struct mlx5_priv *priv = dev->data->dev_private;
1646 if (!priv->config.mpls_en)
1647 return rte_flow_error_set(error, ENOTSUP,
1648 RTE_FLOW_ERROR_TYPE_ITEM, item,
1649 "MPLS not supported or"
1650 " disabled in firmware"
1652 /* MPLS over IP, UDP, GRE is allowed */
1653 if (!(prev_layer & (MLX5_FLOW_LAYER_OUTER_L3 |
1654 MLX5_FLOW_LAYER_OUTER_L4_UDP |
1655 MLX5_FLOW_LAYER_GRE)))
1656 return rte_flow_error_set(error, EINVAL,
1657 RTE_FLOW_ERROR_TYPE_ITEM, item,
1658 "protocol filtering not compatible"
1659 " with MPLS layer");
1660 /* Multi-tunnel isn't allowed but MPLS over GRE is an exception. */
1661 if ((item_flags & MLX5_FLOW_LAYER_TUNNEL) &&
1662 !(item_flags & MLX5_FLOW_LAYER_GRE))
1663 return rte_flow_error_set(error, ENOTSUP,
1664 RTE_FLOW_ERROR_TYPE_ITEM, item,
1665 "multiple tunnel layers not"
1668 mask = &rte_flow_item_mpls_mask;
1669 ret = mlx5_flow_item_acceptable
1670 (item, (const uint8_t *)mask,
1671 (const uint8_t *)&rte_flow_item_mpls_mask,
1672 sizeof(struct rte_flow_item_mpls), error);
1677 return rte_flow_error_set(error, ENOTSUP,
1678 RTE_FLOW_ERROR_TYPE_ITEM, item,
1679 "MPLS is not supported by Verbs, please"
1684 flow_null_validate(struct rte_eth_dev *dev __rte_unused,
1685 const struct rte_flow_attr *attr __rte_unused,
1686 const struct rte_flow_item items[] __rte_unused,
1687 const struct rte_flow_action actions[] __rte_unused,
1688 struct rte_flow_error *error __rte_unused)
1690 rte_errno = ENOTSUP;
1694 static struct mlx5_flow *
1695 flow_null_prepare(const struct rte_flow_attr *attr __rte_unused,
1696 const struct rte_flow_item items[] __rte_unused,
1697 const struct rte_flow_action actions[] __rte_unused,
1698 struct rte_flow_error *error __rte_unused)
1700 rte_errno = ENOTSUP;
1705 flow_null_translate(struct rte_eth_dev *dev __rte_unused,
1706 struct mlx5_flow *dev_flow __rte_unused,
1707 const struct rte_flow_attr *attr __rte_unused,
1708 const struct rte_flow_item items[] __rte_unused,
1709 const struct rte_flow_action actions[] __rte_unused,
1710 struct rte_flow_error *error __rte_unused)
1712 rte_errno = ENOTSUP;
1717 flow_null_apply(struct rte_eth_dev *dev __rte_unused,
1718 struct rte_flow *flow __rte_unused,
1719 struct rte_flow_error *error __rte_unused)
1721 rte_errno = ENOTSUP;
1726 flow_null_remove(struct rte_eth_dev *dev __rte_unused,
1727 struct rte_flow *flow __rte_unused)
1732 flow_null_destroy(struct rte_eth_dev *dev __rte_unused,
1733 struct rte_flow *flow __rte_unused)
1738 flow_null_query(struct rte_eth_dev *dev __rte_unused,
1739 struct rte_flow *flow __rte_unused,
1740 const struct rte_flow_action *actions __rte_unused,
1741 void *data __rte_unused,
1742 struct rte_flow_error *error __rte_unused)
1744 rte_errno = ENOTSUP;
1748 /* Void driver to protect from null pointer reference. */
1749 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops = {
1750 .validate = flow_null_validate,
1751 .prepare = flow_null_prepare,
1752 .translate = flow_null_translate,
1753 .apply = flow_null_apply,
1754 .remove = flow_null_remove,
1755 .destroy = flow_null_destroy,
1756 .query = flow_null_query,
1760 * Select flow driver type according to flow attributes and device
1764 * Pointer to the dev structure.
1766 * Pointer to the flow attributes.
1769 * flow driver type, MLX5_FLOW_TYPE_MAX otherwise.
1771 static enum mlx5_flow_drv_type
1772 flow_get_drv_type(struct rte_eth_dev *dev, const struct rte_flow_attr *attr)
1774 struct mlx5_priv *priv = dev->data->dev_private;
1775 enum mlx5_flow_drv_type type = MLX5_FLOW_TYPE_MAX;
1778 type = MLX5_FLOW_TYPE_TCF;
1780 type = priv->config.dv_flow_en ? MLX5_FLOW_TYPE_DV :
1781 MLX5_FLOW_TYPE_VERBS;
1785 #define flow_get_drv_ops(type) flow_drv_ops[type]
1788 * Flow driver validation API. This abstracts calling driver specific functions.
1789 * The type of flow driver is determined according to flow attributes.
1792 * Pointer to the dev structure.
1794 * Pointer to the flow attributes.
1796 * Pointer to the list of items.
1797 * @param[in] actions
1798 * Pointer to the list of actions.
1800 * Pointer to the error structure.
1803 * 0 on success, a negative errno value otherwise and rte_errno is set.
1806 flow_drv_validate(struct rte_eth_dev *dev,
1807 const struct rte_flow_attr *attr,
1808 const struct rte_flow_item items[],
1809 const struct rte_flow_action actions[],
1810 struct rte_flow_error *error)
1812 const struct mlx5_flow_driver_ops *fops;
1813 enum mlx5_flow_drv_type type = flow_get_drv_type(dev, attr);
1815 fops = flow_get_drv_ops(type);
1816 return fops->validate(dev, attr, items, actions, error);
1820 * Flow driver preparation API. This abstracts calling driver specific
1821 * functions. Parent flow (rte_flow) should have driver type (drv_type). It
1822 * calculates the size of memory required for device flow, allocates the memory,
1823 * initializes the device flow and returns the pointer.
1826 * This function initializes device flow structure such as dv, tcf or verbs in
1827 * struct mlx5_flow. However, it is caller's responsibility to initialize the
1828 * rest. For example, adding returning device flow to flow->dev_flow list and
1829 * setting backward reference to the flow should be done out of this function.
1830 * layers field is not filled either.
1833 * Pointer to the flow attributes.
1835 * Pointer to the list of items.
1836 * @param[in] actions
1837 * Pointer to the list of actions.
1839 * Pointer to the error structure.
1842 * Pointer to device flow on success, otherwise NULL and rte_errno is set.
1844 static inline struct mlx5_flow *
1845 flow_drv_prepare(const struct rte_flow *flow,
1846 const struct rte_flow_attr *attr,
1847 const struct rte_flow_item items[],
1848 const struct rte_flow_action actions[],
1849 struct rte_flow_error *error)
1851 const struct mlx5_flow_driver_ops *fops;
1852 enum mlx5_flow_drv_type type = flow->drv_type;
1854 assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
1855 fops = flow_get_drv_ops(type);
1856 return fops->prepare(attr, items, actions, error);
1860 * Flow driver translation API. This abstracts calling driver specific
1861 * functions. Parent flow (rte_flow) should have driver type (drv_type). It
1862 * translates a generic flow into a driver flow. flow_drv_prepare() must
1866 * dev_flow->layers could be filled as a result of parsing during translation
1867 * if needed by flow_drv_apply(). dev_flow->flow->actions can also be filled
1868 * if necessary. As a flow can have multiple dev_flows by RSS flow expansion,
1869 * flow->actions could be overwritten even though all the expanded dev_flows
1870 * have the same actions.
1873 * Pointer to the rte dev structure.
1874 * @param[in, out] dev_flow
1875 * Pointer to the mlx5 flow.
1877 * Pointer to the flow attributes.
1879 * Pointer to the list of items.
1880 * @param[in] actions
1881 * Pointer to the list of actions.
1883 * Pointer to the error structure.
1886 * 0 on success, a negative errno value otherwise and rte_errno is set.
1889 flow_drv_translate(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow,
1890 const struct rte_flow_attr *attr,
1891 const struct rte_flow_item items[],
1892 const struct rte_flow_action actions[],
1893 struct rte_flow_error *error)
1895 const struct mlx5_flow_driver_ops *fops;
1896 enum mlx5_flow_drv_type type = dev_flow->flow->drv_type;
1898 assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
1899 fops = flow_get_drv_ops(type);
1900 return fops->translate(dev, dev_flow, attr, items, actions, error);
1904 * Flow driver apply API. This abstracts calling driver specific functions.
1905 * Parent flow (rte_flow) should have driver type (drv_type). It applies
1906 * translated driver flows on to device. flow_drv_translate() must precede.
1909 * Pointer to Ethernet device structure.
1910 * @param[in, out] flow
1911 * Pointer to flow structure.
1913 * Pointer to error structure.
1916 * 0 on success, a negative errno value otherwise and rte_errno is set.
1919 flow_drv_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
1920 struct rte_flow_error *error)
1922 const struct mlx5_flow_driver_ops *fops;
1923 enum mlx5_flow_drv_type type = flow->drv_type;
1925 assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
1926 fops = flow_get_drv_ops(type);
1927 return fops->apply(dev, flow, error);
1931 * Flow driver remove API. This abstracts calling driver specific functions.
1932 * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
1933 * on device. All the resources of the flow should be freed by calling
1934 * flow_drv_destroy().
1937 * Pointer to Ethernet device.
1938 * @param[in, out] flow
1939 * Pointer to flow structure.
1942 flow_drv_remove(struct rte_eth_dev *dev, struct rte_flow *flow)
1944 const struct mlx5_flow_driver_ops *fops;
1945 enum mlx5_flow_drv_type type = flow->drv_type;
1947 assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
1948 fops = flow_get_drv_ops(type);
1949 fops->remove(dev, flow);
1953 * Flow driver destroy API. This abstracts calling driver specific functions.
1954 * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
1955 * on device and releases resources of the flow.
1958 * Pointer to Ethernet device.
1959 * @param[in, out] flow
1960 * Pointer to flow structure.
1963 flow_drv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
1965 const struct mlx5_flow_driver_ops *fops;
1966 enum mlx5_flow_drv_type type = flow->drv_type;
1968 assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
1969 fops = flow_get_drv_ops(type);
1970 fops->destroy(dev, flow);
1974 * Validate a flow supported by the NIC.
1976 * @see rte_flow_validate()
1980 mlx5_flow_validate(struct rte_eth_dev *dev,
1981 const struct rte_flow_attr *attr,
1982 const struct rte_flow_item items[],
1983 const struct rte_flow_action actions[],
1984 struct rte_flow_error *error)
1988 ret = flow_drv_validate(dev, attr, items, actions, error);
1995 * Get RSS action from the action list.
1997 * @param[in] actions
1998 * Pointer to the list of actions.
2001 * Pointer to the RSS action if exist, else return NULL.
2003 static const struct rte_flow_action_rss*
2004 flow_get_rss_action(const struct rte_flow_action actions[])
2006 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2007 switch (actions->type) {
2008 case RTE_FLOW_ACTION_TYPE_RSS:
2009 return (const struct rte_flow_action_rss *)
2019 find_graph_root(const struct rte_flow_item pattern[], uint32_t rss_level)
2021 const struct rte_flow_item *item;
2022 unsigned int has_vlan = 0;
2024 for (item = pattern; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
2025 if (item->type == RTE_FLOW_ITEM_TYPE_VLAN) {
2031 return rss_level < 2 ? MLX5_EXPANSION_ROOT_ETH_VLAN :
2032 MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN;
2033 return rss_level < 2 ? MLX5_EXPANSION_ROOT :
2034 MLX5_EXPANSION_ROOT_OUTER;
2038 * Create a flow and add it to @p list.
2041 * Pointer to Ethernet device.
2043 * Pointer to a TAILQ flow list.
2045 * Flow rule attributes.
2047 * Pattern specification (list terminated by the END pattern item).
2048 * @param[in] actions
2049 * Associated actions (list terminated by the END action).
2051 * Perform verbose error reporting if not NULL.
2054 * A flow on success, NULL otherwise and rte_errno is set.
2056 static struct rte_flow *
2057 flow_list_create(struct rte_eth_dev *dev, struct mlx5_flows *list,
2058 const struct rte_flow_attr *attr,
2059 const struct rte_flow_item items[],
2060 const struct rte_flow_action actions[],
2061 struct rte_flow_error *error)
2063 struct rte_flow *flow = NULL;
2064 struct mlx5_flow *dev_flow;
2065 const struct rte_flow_action_rss *rss;
2067 struct rte_flow_expand_rss buf;
2068 uint8_t buffer[2048];
2070 struct rte_flow_expand_rss *buf = &expand_buffer.buf;
2075 ret = flow_drv_validate(dev, attr, items, actions, error);
2078 flow_size = sizeof(struct rte_flow);
2079 rss = flow_get_rss_action(actions);
2081 flow_size += RTE_ALIGN_CEIL(rss->queue_num * sizeof(uint16_t),
2084 flow_size += RTE_ALIGN_CEIL(sizeof(uint16_t), sizeof(void *));
2085 flow = rte_calloc(__func__, 1, flow_size, 0);
2086 flow->drv_type = flow_get_drv_type(dev, attr);
2087 flow->ingress = attr->ingress;
2088 assert(flow->drv_type > MLX5_FLOW_TYPE_MIN &&
2089 flow->drv_type < MLX5_FLOW_TYPE_MAX);
2090 flow->queue = (void *)(flow + 1);
2091 LIST_INIT(&flow->dev_flows);
2092 if (rss && rss->types) {
2093 unsigned int graph_root;
2095 graph_root = find_graph_root(items, rss->level);
2096 ret = rte_flow_expand_rss(buf, sizeof(expand_buffer.buffer),
2098 mlx5_support_expansion,
2101 (unsigned int)ret < sizeof(expand_buffer.buffer));
2104 buf->entry[0].pattern = (void *)(uintptr_t)items;
2106 for (i = 0; i < buf->entries; ++i) {
2107 dev_flow = flow_drv_prepare(flow, attr, buf->entry[i].pattern,
2111 dev_flow->flow = flow;
2112 LIST_INSERT_HEAD(&flow->dev_flows, dev_flow, next);
2113 ret = flow_drv_translate(dev, dev_flow, attr,
2114 buf->entry[i].pattern,
2119 if (dev->data->dev_started) {
2120 ret = flow_drv_apply(dev, flow, error);
2124 TAILQ_INSERT_TAIL(list, flow, next);
2125 flow_rxq_flags_set(dev, flow);
2128 ret = rte_errno; /* Save rte_errno before cleanup. */
2130 flow_drv_destroy(dev, flow);
2132 rte_errno = ret; /* Restore rte_errno. */
2139 * @see rte_flow_create()
2143 mlx5_flow_create(struct rte_eth_dev *dev,
2144 const struct rte_flow_attr *attr,
2145 const struct rte_flow_item items[],
2146 const struct rte_flow_action actions[],
2147 struct rte_flow_error *error)
2149 struct mlx5_priv *priv = (struct mlx5_priv *)dev->data->dev_private;
2151 return flow_list_create(dev, &priv->flows,
2152 attr, items, actions, error);
2156 * Destroy a flow in a list.
2159 * Pointer to Ethernet device.
2161 * Pointer to a TAILQ flow list.
2166 flow_list_destroy(struct rte_eth_dev *dev, struct mlx5_flows *list,
2167 struct rte_flow *flow)
2170 * Update RX queue flags only if port is started, otherwise it is
2173 if (dev->data->dev_started)
2174 flow_rxq_flags_trim(dev, flow);
2175 flow_drv_destroy(dev, flow);
2176 TAILQ_REMOVE(list, flow, next);
2177 rte_free(flow->fdir);
2182 * Destroy all flows.
2185 * Pointer to Ethernet device.
2187 * Pointer to a TAILQ flow list.
2190 mlx5_flow_list_flush(struct rte_eth_dev *dev, struct mlx5_flows *list)
2192 while (!TAILQ_EMPTY(list)) {
2193 struct rte_flow *flow;
2195 flow = TAILQ_FIRST(list);
2196 flow_list_destroy(dev, list, flow);
2204 * Pointer to Ethernet device.
2206 * Pointer to a TAILQ flow list.
2209 mlx5_flow_stop(struct rte_eth_dev *dev, struct mlx5_flows *list)
2211 struct rte_flow *flow;
2213 TAILQ_FOREACH_REVERSE(flow, list, mlx5_flows, next)
2214 flow_drv_remove(dev, flow);
2215 flow_rxq_flags_clear(dev);
2222 * Pointer to Ethernet device.
2224 * Pointer to a TAILQ flow list.
2227 * 0 on success, a negative errno value otherwise and rte_errno is set.
2230 mlx5_flow_start(struct rte_eth_dev *dev, struct mlx5_flows *list)
2232 struct rte_flow *flow;
2233 struct rte_flow_error error;
2236 TAILQ_FOREACH(flow, list, next) {
2237 ret = flow_drv_apply(dev, flow, &error);
2240 flow_rxq_flags_set(dev, flow);
2244 ret = rte_errno; /* Save rte_errno before cleanup. */
2245 mlx5_flow_stop(dev, list);
2246 rte_errno = ret; /* Restore rte_errno. */
2251 * Verify the flow list is empty
2254 * Pointer to Ethernet device.
2256 * @return the number of flows not released.
2259 mlx5_flow_verify(struct rte_eth_dev *dev)
2261 struct mlx5_priv *priv = dev->data->dev_private;
2262 struct rte_flow *flow;
2265 TAILQ_FOREACH(flow, &priv->flows, next) {
2266 DRV_LOG(DEBUG, "port %u flow %p still referenced",
2267 dev->data->port_id, (void *)flow);
2274 * Enable a control flow configured from the control plane.
2277 * Pointer to Ethernet device.
2279 * An Ethernet flow spec to apply.
2281 * An Ethernet flow mask to apply.
2283 * A VLAN flow spec to apply.
2285 * A VLAN flow mask to apply.
2288 * 0 on success, a negative errno value otherwise and rte_errno is set.
2291 mlx5_ctrl_flow_vlan(struct rte_eth_dev *dev,
2292 struct rte_flow_item_eth *eth_spec,
2293 struct rte_flow_item_eth *eth_mask,
2294 struct rte_flow_item_vlan *vlan_spec,
2295 struct rte_flow_item_vlan *vlan_mask)
2297 struct mlx5_priv *priv = dev->data->dev_private;
2298 const struct rte_flow_attr attr = {
2300 .priority = MLX5_FLOW_PRIO_RSVD,
2302 struct rte_flow_item items[] = {
2304 .type = RTE_FLOW_ITEM_TYPE_ETH,
2310 .type = (vlan_spec) ? RTE_FLOW_ITEM_TYPE_VLAN :
2311 RTE_FLOW_ITEM_TYPE_END,
2317 .type = RTE_FLOW_ITEM_TYPE_END,
2320 uint16_t queue[priv->reta_idx_n];
2321 struct rte_flow_action_rss action_rss = {
2322 .func = RTE_ETH_HASH_FUNCTION_DEFAULT,
2324 .types = priv->rss_conf.rss_hf,
2325 .key_len = priv->rss_conf.rss_key_len,
2326 .queue_num = priv->reta_idx_n,
2327 .key = priv->rss_conf.rss_key,
2330 struct rte_flow_action actions[] = {
2332 .type = RTE_FLOW_ACTION_TYPE_RSS,
2333 .conf = &action_rss,
2336 .type = RTE_FLOW_ACTION_TYPE_END,
2339 struct rte_flow *flow;
2340 struct rte_flow_error error;
2343 if (!priv->reta_idx_n || !priv->rxqs_n) {
2346 for (i = 0; i != priv->reta_idx_n; ++i)
2347 queue[i] = (*priv->reta_idx)[i];
2348 flow = flow_list_create(dev, &priv->ctrl_flows,
2349 &attr, items, actions, &error);
2356 * Enable a flow control configured from the control plane.
2359 * Pointer to Ethernet device.
2361 * An Ethernet flow spec to apply.
2363 * An Ethernet flow mask to apply.
2366 * 0 on success, a negative errno value otherwise and rte_errno is set.
2369 mlx5_ctrl_flow(struct rte_eth_dev *dev,
2370 struct rte_flow_item_eth *eth_spec,
2371 struct rte_flow_item_eth *eth_mask)
2373 return mlx5_ctrl_flow_vlan(dev, eth_spec, eth_mask, NULL, NULL);
2379 * @see rte_flow_destroy()
2383 mlx5_flow_destroy(struct rte_eth_dev *dev,
2384 struct rte_flow *flow,
2385 struct rte_flow_error *error __rte_unused)
2387 struct mlx5_priv *priv = dev->data->dev_private;
2389 flow_list_destroy(dev, &priv->flows, flow);
2394 * Destroy all flows.
2396 * @see rte_flow_flush()
2400 mlx5_flow_flush(struct rte_eth_dev *dev,
2401 struct rte_flow_error *error __rte_unused)
2403 struct mlx5_priv *priv = dev->data->dev_private;
2405 mlx5_flow_list_flush(dev, &priv->flows);
2412 * @see rte_flow_isolate()
2416 mlx5_flow_isolate(struct rte_eth_dev *dev,
2418 struct rte_flow_error *error)
2420 struct mlx5_priv *priv = dev->data->dev_private;
2422 if (dev->data->dev_started) {
2423 rte_flow_error_set(error, EBUSY,
2424 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
2426 "port must be stopped first");
2429 priv->isolated = !!enable;
2431 dev->dev_ops = &mlx5_dev_ops_isolate;
2433 dev->dev_ops = &mlx5_dev_ops;
2440 * @see rte_flow_query()
2444 flow_drv_query(struct rte_eth_dev *dev,
2445 struct rte_flow *flow,
2446 const struct rte_flow_action *actions,
2448 struct rte_flow_error *error)
2450 const struct mlx5_flow_driver_ops *fops;
2451 enum mlx5_flow_drv_type ftype = flow->drv_type;
2453 assert(ftype > MLX5_FLOW_TYPE_MIN && ftype < MLX5_FLOW_TYPE_MAX);
2454 fops = flow_get_drv_ops(ftype);
2456 return fops->query(dev, flow, actions, data, error);
2462 * @see rte_flow_query()
2466 mlx5_flow_query(struct rte_eth_dev *dev,
2467 struct rte_flow *flow,
2468 const struct rte_flow_action *actions,
2470 struct rte_flow_error *error)
2474 ret = flow_drv_query(dev, flow, actions, data, error);
2481 * Convert a flow director filter to a generic flow.
2484 * Pointer to Ethernet device.
2485 * @param fdir_filter
2486 * Flow director filter to add.
2488 * Generic flow parameters structure.
2491 * 0 on success, a negative errno value otherwise and rte_errno is set.
2494 flow_fdir_filter_convert(struct rte_eth_dev *dev,
2495 const struct rte_eth_fdir_filter *fdir_filter,
2496 struct mlx5_fdir *attributes)
2498 struct mlx5_priv *priv = dev->data->dev_private;
2499 const struct rte_eth_fdir_input *input = &fdir_filter->input;
2500 const struct rte_eth_fdir_masks *mask =
2501 &dev->data->dev_conf.fdir_conf.mask;
2503 /* Validate queue number. */
2504 if (fdir_filter->action.rx_queue >= priv->rxqs_n) {
2505 DRV_LOG(ERR, "port %u invalid queue number %d",
2506 dev->data->port_id, fdir_filter->action.rx_queue);
2510 attributes->attr.ingress = 1;
2511 attributes->items[0] = (struct rte_flow_item) {
2512 .type = RTE_FLOW_ITEM_TYPE_ETH,
2513 .spec = &attributes->l2,
2514 .mask = &attributes->l2_mask,
2516 switch (fdir_filter->action.behavior) {
2517 case RTE_ETH_FDIR_ACCEPT:
2518 attributes->actions[0] = (struct rte_flow_action){
2519 .type = RTE_FLOW_ACTION_TYPE_QUEUE,
2520 .conf = &attributes->queue,
2523 case RTE_ETH_FDIR_REJECT:
2524 attributes->actions[0] = (struct rte_flow_action){
2525 .type = RTE_FLOW_ACTION_TYPE_DROP,
2529 DRV_LOG(ERR, "port %u invalid behavior %d",
2531 fdir_filter->action.behavior);
2532 rte_errno = ENOTSUP;
2535 attributes->queue.index = fdir_filter->action.rx_queue;
2537 switch (fdir_filter->input.flow_type) {
2538 case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
2539 case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
2540 case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
2541 attributes->l3.ipv4.hdr = (struct ipv4_hdr){
2542 .src_addr = input->flow.ip4_flow.src_ip,
2543 .dst_addr = input->flow.ip4_flow.dst_ip,
2544 .time_to_live = input->flow.ip4_flow.ttl,
2545 .type_of_service = input->flow.ip4_flow.tos,
2547 attributes->l3_mask.ipv4.hdr = (struct ipv4_hdr){
2548 .src_addr = mask->ipv4_mask.src_ip,
2549 .dst_addr = mask->ipv4_mask.dst_ip,
2550 .time_to_live = mask->ipv4_mask.ttl,
2551 .type_of_service = mask->ipv4_mask.tos,
2552 .next_proto_id = mask->ipv4_mask.proto,
2554 attributes->items[1] = (struct rte_flow_item){
2555 .type = RTE_FLOW_ITEM_TYPE_IPV4,
2556 .spec = &attributes->l3,
2557 .mask = &attributes->l3_mask,
2560 case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
2561 case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
2562 case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
2563 attributes->l3.ipv6.hdr = (struct ipv6_hdr){
2564 .hop_limits = input->flow.ipv6_flow.hop_limits,
2565 .proto = input->flow.ipv6_flow.proto,
2568 memcpy(attributes->l3.ipv6.hdr.src_addr,
2569 input->flow.ipv6_flow.src_ip,
2570 RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
2571 memcpy(attributes->l3.ipv6.hdr.dst_addr,
2572 input->flow.ipv6_flow.dst_ip,
2573 RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
2574 memcpy(attributes->l3_mask.ipv6.hdr.src_addr,
2575 mask->ipv6_mask.src_ip,
2576 RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
2577 memcpy(attributes->l3_mask.ipv6.hdr.dst_addr,
2578 mask->ipv6_mask.dst_ip,
2579 RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
2580 attributes->items[1] = (struct rte_flow_item){
2581 .type = RTE_FLOW_ITEM_TYPE_IPV6,
2582 .spec = &attributes->l3,
2583 .mask = &attributes->l3_mask,
2587 DRV_LOG(ERR, "port %u invalid flow type%d",
2588 dev->data->port_id, fdir_filter->input.flow_type);
2589 rte_errno = ENOTSUP;
2593 switch (fdir_filter->input.flow_type) {
2594 case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
2595 attributes->l4.udp.hdr = (struct udp_hdr){
2596 .src_port = input->flow.udp4_flow.src_port,
2597 .dst_port = input->flow.udp4_flow.dst_port,
2599 attributes->l4_mask.udp.hdr = (struct udp_hdr){
2600 .src_port = mask->src_port_mask,
2601 .dst_port = mask->dst_port_mask,
2603 attributes->items[2] = (struct rte_flow_item){
2604 .type = RTE_FLOW_ITEM_TYPE_UDP,
2605 .spec = &attributes->l4,
2606 .mask = &attributes->l4_mask,
2609 case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
2610 attributes->l4.tcp.hdr = (struct tcp_hdr){
2611 .src_port = input->flow.tcp4_flow.src_port,
2612 .dst_port = input->flow.tcp4_flow.dst_port,
2614 attributes->l4_mask.tcp.hdr = (struct tcp_hdr){
2615 .src_port = mask->src_port_mask,
2616 .dst_port = mask->dst_port_mask,
2618 attributes->items[2] = (struct rte_flow_item){
2619 .type = RTE_FLOW_ITEM_TYPE_TCP,
2620 .spec = &attributes->l4,
2621 .mask = &attributes->l4_mask,
2624 case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
2625 attributes->l4.udp.hdr = (struct udp_hdr){
2626 .src_port = input->flow.udp6_flow.src_port,
2627 .dst_port = input->flow.udp6_flow.dst_port,
2629 attributes->l4_mask.udp.hdr = (struct udp_hdr){
2630 .src_port = mask->src_port_mask,
2631 .dst_port = mask->dst_port_mask,
2633 attributes->items[2] = (struct rte_flow_item){
2634 .type = RTE_FLOW_ITEM_TYPE_UDP,
2635 .spec = &attributes->l4,
2636 .mask = &attributes->l4_mask,
2639 case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
2640 attributes->l4.tcp.hdr = (struct tcp_hdr){
2641 .src_port = input->flow.tcp6_flow.src_port,
2642 .dst_port = input->flow.tcp6_flow.dst_port,
2644 attributes->l4_mask.tcp.hdr = (struct tcp_hdr){
2645 .src_port = mask->src_port_mask,
2646 .dst_port = mask->dst_port_mask,
2648 attributes->items[2] = (struct rte_flow_item){
2649 .type = RTE_FLOW_ITEM_TYPE_TCP,
2650 .spec = &attributes->l4,
2651 .mask = &attributes->l4_mask,
2654 case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
2655 case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
2658 DRV_LOG(ERR, "port %u invalid flow type%d",
2659 dev->data->port_id, fdir_filter->input.flow_type);
2660 rte_errno = ENOTSUP;
2666 #define FLOW_FDIR_CMP(f1, f2, fld) \
2667 memcmp(&(f1)->fld, &(f2)->fld, sizeof(f1->fld))
2670 * Compare two FDIR flows. If items and actions are identical, the two flows are
2674 * Pointer to Ethernet device.
2676 * FDIR flow to compare.
2678 * FDIR flow to compare.
2681 * Zero on match, 1 otherwise.
2684 flow_fdir_cmp(const struct mlx5_fdir *f1, const struct mlx5_fdir *f2)
2686 if (FLOW_FDIR_CMP(f1, f2, attr) ||
2687 FLOW_FDIR_CMP(f1, f2, l2) ||
2688 FLOW_FDIR_CMP(f1, f2, l2_mask) ||
2689 FLOW_FDIR_CMP(f1, f2, l3) ||
2690 FLOW_FDIR_CMP(f1, f2, l3_mask) ||
2691 FLOW_FDIR_CMP(f1, f2, l4) ||
2692 FLOW_FDIR_CMP(f1, f2, l4_mask) ||
2693 FLOW_FDIR_CMP(f1, f2, actions[0].type))
2695 if (f1->actions[0].type == RTE_FLOW_ACTION_TYPE_QUEUE &&
2696 FLOW_FDIR_CMP(f1, f2, queue))
2702 * Search device flow list to find out a matched FDIR flow.
2705 * Pointer to Ethernet device.
2707 * FDIR flow to lookup.
2710 * Pointer of flow if found, NULL otherwise.
2712 static struct rte_flow *
2713 flow_fdir_filter_lookup(struct rte_eth_dev *dev, struct mlx5_fdir *fdir_flow)
2715 struct mlx5_priv *priv = dev->data->dev_private;
2716 struct rte_flow *flow = NULL;
2719 TAILQ_FOREACH(flow, &priv->flows, next) {
2720 if (flow->fdir && !flow_fdir_cmp(flow->fdir, fdir_flow)) {
2721 DRV_LOG(DEBUG, "port %u found FDIR flow %p",
2722 dev->data->port_id, (void *)flow);
2730 * Add new flow director filter and store it in list.
2733 * Pointer to Ethernet device.
2734 * @param fdir_filter
2735 * Flow director filter to add.
2738 * 0 on success, a negative errno value otherwise and rte_errno is set.
2741 flow_fdir_filter_add(struct rte_eth_dev *dev,
2742 const struct rte_eth_fdir_filter *fdir_filter)
2744 struct mlx5_priv *priv = dev->data->dev_private;
2745 struct mlx5_fdir *fdir_flow;
2746 struct rte_flow *flow;
2749 fdir_flow = rte_zmalloc(__func__, sizeof(*fdir_flow), 0);
2754 ret = flow_fdir_filter_convert(dev, fdir_filter, fdir_flow);
2757 flow = flow_fdir_filter_lookup(dev, fdir_flow);
2762 flow = flow_list_create(dev, &priv->flows, &fdir_flow->attr,
2763 fdir_flow->items, fdir_flow->actions, NULL);
2766 assert(!flow->fdir);
2767 flow->fdir = fdir_flow;
2768 DRV_LOG(DEBUG, "port %u created FDIR flow %p",
2769 dev->data->port_id, (void *)flow);
2772 rte_free(fdir_flow);
2777 * Delete specific filter.
2780 * Pointer to Ethernet device.
2781 * @param fdir_filter
2782 * Filter to be deleted.
2785 * 0 on success, a negative errno value otherwise and rte_errno is set.
2788 flow_fdir_filter_delete(struct rte_eth_dev *dev,
2789 const struct rte_eth_fdir_filter *fdir_filter)
2791 struct mlx5_priv *priv = dev->data->dev_private;
2792 struct rte_flow *flow;
2793 struct mlx5_fdir fdir_flow = {
2798 ret = flow_fdir_filter_convert(dev, fdir_filter, &fdir_flow);
2801 flow = flow_fdir_filter_lookup(dev, &fdir_flow);
2806 flow_list_destroy(dev, &priv->flows, flow);
2807 DRV_LOG(DEBUG, "port %u deleted FDIR flow %p",
2808 dev->data->port_id, (void *)flow);
2813 * Update queue for specific filter.
2816 * Pointer to Ethernet device.
2817 * @param fdir_filter
2818 * Filter to be updated.
2821 * 0 on success, a negative errno value otherwise and rte_errno is set.
2824 flow_fdir_filter_update(struct rte_eth_dev *dev,
2825 const struct rte_eth_fdir_filter *fdir_filter)
2829 ret = flow_fdir_filter_delete(dev, fdir_filter);
2832 return flow_fdir_filter_add(dev, fdir_filter);
2836 * Flush all filters.
2839 * Pointer to Ethernet device.
2842 flow_fdir_filter_flush(struct rte_eth_dev *dev)
2844 struct mlx5_priv *priv = dev->data->dev_private;
2846 mlx5_flow_list_flush(dev, &priv->flows);
2850 * Get flow director information.
2853 * Pointer to Ethernet device.
2854 * @param[out] fdir_info
2855 * Resulting flow director information.
2858 flow_fdir_info_get(struct rte_eth_dev *dev, struct rte_eth_fdir_info *fdir_info)
2860 struct rte_eth_fdir_masks *mask =
2861 &dev->data->dev_conf.fdir_conf.mask;
2863 fdir_info->mode = dev->data->dev_conf.fdir_conf.mode;
2864 fdir_info->guarant_spc = 0;
2865 rte_memcpy(&fdir_info->mask, mask, sizeof(fdir_info->mask));
2866 fdir_info->max_flexpayload = 0;
2867 fdir_info->flow_types_mask[0] = 0;
2868 fdir_info->flex_payload_unit = 0;
2869 fdir_info->max_flex_payload_segment_num = 0;
2870 fdir_info->flex_payload_limit = 0;
2871 memset(&fdir_info->flex_conf, 0, sizeof(fdir_info->flex_conf));
2875 * Deal with flow director operations.
2878 * Pointer to Ethernet device.
2880 * Operation to perform.
2882 * Pointer to operation-specific structure.
2885 * 0 on success, a negative errno value otherwise and rte_errno is set.
2888 flow_fdir_ctrl_func(struct rte_eth_dev *dev, enum rte_filter_op filter_op,
2891 enum rte_fdir_mode fdir_mode =
2892 dev->data->dev_conf.fdir_conf.mode;
2894 if (filter_op == RTE_ETH_FILTER_NOP)
2896 if (fdir_mode != RTE_FDIR_MODE_PERFECT &&
2897 fdir_mode != RTE_FDIR_MODE_PERFECT_MAC_VLAN) {
2898 DRV_LOG(ERR, "port %u flow director mode %d not supported",
2899 dev->data->port_id, fdir_mode);
2903 switch (filter_op) {
2904 case RTE_ETH_FILTER_ADD:
2905 return flow_fdir_filter_add(dev, arg);
2906 case RTE_ETH_FILTER_UPDATE:
2907 return flow_fdir_filter_update(dev, arg);
2908 case RTE_ETH_FILTER_DELETE:
2909 return flow_fdir_filter_delete(dev, arg);
2910 case RTE_ETH_FILTER_FLUSH:
2911 flow_fdir_filter_flush(dev);
2913 case RTE_ETH_FILTER_INFO:
2914 flow_fdir_info_get(dev, arg);
2917 DRV_LOG(DEBUG, "port %u unknown operation %u",
2918 dev->data->port_id, filter_op);
2926 * Manage filter operations.
2929 * Pointer to Ethernet device structure.
2930 * @param filter_type
2933 * Operation to perform.
2935 * Pointer to operation-specific structure.
2938 * 0 on success, a negative errno value otherwise and rte_errno is set.
2941 mlx5_dev_filter_ctrl(struct rte_eth_dev *dev,
2942 enum rte_filter_type filter_type,
2943 enum rte_filter_op filter_op,
2946 switch (filter_type) {
2947 case RTE_ETH_FILTER_GENERIC:
2948 if (filter_op != RTE_ETH_FILTER_GET) {
2952 *(const void **)arg = &mlx5_flow_ops;
2954 case RTE_ETH_FILTER_FDIR:
2955 return flow_fdir_ctrl_func(dev, filter_op, arg);
2957 DRV_LOG(ERR, "port %u filter type (%d) not supported",
2958 dev->data->port_id, filter_type);
2959 rte_errno = ENOTSUP;