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_GRE | RTE_PTYPE_L4_UDP,
300 .tunnel = MLX5_FLOW_LAYER_MPLS,
301 .ptype = RTE_PTYPE_TUNNEL_MPLS_IN_GRE,
306 * Discover the maximum number of priority available.
309 * Pointer to the Ethernet device structure.
312 * number of supported flow priority on success, a negative errno
313 * value otherwise and rte_errno is set.
316 mlx5_flow_discover_priorities(struct rte_eth_dev *dev)
319 struct ibv_flow_attr attr;
320 struct ibv_flow_spec_eth eth;
321 struct ibv_flow_spec_action_drop drop;
327 .type = IBV_FLOW_SPEC_ETH,
328 .size = sizeof(struct ibv_flow_spec_eth),
331 .size = sizeof(struct ibv_flow_spec_action_drop),
332 .type = IBV_FLOW_SPEC_ACTION_DROP,
335 struct ibv_flow *flow;
336 struct mlx5_hrxq *drop = mlx5_hrxq_drop_new(dev);
337 uint16_t vprio[] = { 8, 16 };
345 for (i = 0; i != RTE_DIM(vprio); i++) {
346 flow_attr.attr.priority = vprio[i] - 1;
347 flow = mlx5_glue->create_flow(drop->qp, &flow_attr.attr);
350 claim_zero(mlx5_glue->destroy_flow(flow));
355 priority = RTE_DIM(priority_map_3);
358 priority = RTE_DIM(priority_map_5);
363 "port %u verbs maximum priority: %d expected 8/16",
364 dev->data->port_id, vprio[i]);
367 mlx5_hrxq_drop_release(dev);
368 DRV_LOG(INFO, "port %u flow maximum priority: %d",
369 dev->data->port_id, priority);
374 * Adjust flow priority based on the highest layer and the request priority.
377 * Pointer to the Ethernet device structure.
378 * @param[in] priority
379 * The rule base priority.
380 * @param[in] subpriority
381 * The priority based on the items.
386 uint32_t mlx5_flow_adjust_priority(struct rte_eth_dev *dev, int32_t priority,
387 uint32_t subpriority)
390 struct priv *priv = dev->data->dev_private;
392 switch (priv->config.flow_prio) {
393 case RTE_DIM(priority_map_3):
394 res = priority_map_3[priority][subpriority];
396 case RTE_DIM(priority_map_5):
397 res = priority_map_5[priority][subpriority];
404 * Verify the @p item specifications (spec, last, mask) are compatible with the
408 * Item specification.
410 * @p item->mask or flow default bit-masks.
411 * @param[in] nic_mask
412 * Bit-masks covering supported fields by the NIC to compare with user mask.
414 * Bit-masks size in bytes.
416 * Pointer to error structure.
419 * 0 on success, a negative errno value otherwise and rte_errno is set.
422 mlx5_flow_item_acceptable(const struct rte_flow_item *item,
424 const uint8_t *nic_mask,
426 struct rte_flow_error *error)
431 for (i = 0; i < size; ++i)
432 if ((nic_mask[i] | mask[i]) != nic_mask[i])
433 return rte_flow_error_set(error, ENOTSUP,
434 RTE_FLOW_ERROR_TYPE_ITEM,
436 "mask enables non supported"
438 if (!item->spec && (item->mask || item->last))
439 return rte_flow_error_set(error, EINVAL,
440 RTE_FLOW_ERROR_TYPE_ITEM, item,
441 "mask/last without a spec is not"
443 if (item->spec && item->last) {
449 for (i = 0; i < size; ++i) {
450 spec[i] = ((const uint8_t *)item->spec)[i] & mask[i];
451 last[i] = ((const uint8_t *)item->last)[i] & mask[i];
453 ret = memcmp(spec, last, size);
455 return rte_flow_error_set(error, EINVAL,
456 RTE_FLOW_ERROR_TYPE_ITEM,
458 "range is not valid");
464 * Adjust the hash fields according to the @p flow information.
466 * @param[in] dev_flow.
467 * Pointer to the mlx5_flow.
469 * 1 when the hash field is for a tunnel item.
470 * @param[in] layer_types
472 * @param[in] hash_fields
476 * The hash fileds that should be used.
479 mlx5_flow_hashfields_adjust(struct mlx5_flow *dev_flow,
480 int tunnel __rte_unused, 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"
849 if (queue->index >= priv->rxqs_n)
850 return rte_flow_error_set(error, EINVAL,
851 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
853 "queue index out of range");
854 if (!(*priv->rxqs)[queue->index])
855 return rte_flow_error_set(error, EINVAL,
856 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
858 "queue is not configured");
860 return rte_flow_error_set(error, ENOTSUP,
861 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
862 "queue action not supported for "
868 * Validate the rss action.
871 * Pointer to the queue action.
872 * @param[in] action_flags
873 * Bit-fields that holds the actions detected until now.
875 * Pointer to the Ethernet device structure.
877 * Attributes of flow that includes this action.
879 * Pointer to error structure.
882 * 0 on success, a negative errno value otherwise and rte_ernno is set.
885 mlx5_flow_validate_action_rss(const struct rte_flow_action *action,
886 uint64_t action_flags,
887 struct rte_eth_dev *dev,
888 const struct rte_flow_attr *attr,
889 struct rte_flow_error *error)
891 struct priv *priv = dev->data->dev_private;
892 const struct rte_flow_action_rss *rss = action->conf;
895 if (action_flags & MLX5_FLOW_FATE_ACTIONS)
896 return rte_flow_error_set(error, EINVAL,
897 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
898 "can't have 2 fate actions"
900 if (rss->func != RTE_ETH_HASH_FUNCTION_DEFAULT &&
901 rss->func != RTE_ETH_HASH_FUNCTION_TOEPLITZ)
902 return rte_flow_error_set(error, ENOTSUP,
903 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
905 "RSS hash function not supported");
906 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
911 return rte_flow_error_set(error, ENOTSUP,
912 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
914 "tunnel RSS is not supported");
915 if (rss->key_len < MLX5_RSS_HASH_KEY_LEN)
916 return rte_flow_error_set(error, ENOTSUP,
917 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
919 "RSS hash key too small");
920 if (rss->key_len > MLX5_RSS_HASH_KEY_LEN)
921 return rte_flow_error_set(error, ENOTSUP,
922 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
924 "RSS hash key too large");
925 if (rss->queue_num > priv->config.ind_table_max_size)
926 return rte_flow_error_set(error, ENOTSUP,
927 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
929 "number of queues too large");
930 if (rss->types & MLX5_RSS_HF_MASK)
931 return rte_flow_error_set(error, ENOTSUP,
932 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
934 "some RSS protocols are not"
936 for (i = 0; i != rss->queue_num; ++i) {
937 if (!(*priv->rxqs)[rss->queue[i]])
938 return rte_flow_error_set
939 (error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
940 &rss->queue[i], "queue is not configured");
943 return rte_flow_error_set(error, ENOTSUP,
944 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
945 "rss action not supported for "
951 * Validate the count action.
954 * Pointer to the Ethernet device structure.
956 * Attributes of flow that includes this action.
958 * Pointer to error structure.
961 * 0 on success, a negative errno value otherwise and rte_ernno is set.
964 mlx5_flow_validate_action_count(struct rte_eth_dev *dev __rte_unused,
965 const struct rte_flow_attr *attr,
966 struct rte_flow_error *error)
969 return rte_flow_error_set(error, ENOTSUP,
970 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
971 "count action not supported for "
977 * Verify the @p attributes will be correctly understood by the NIC and store
978 * them in the @p flow if everything is correct.
981 * Pointer to the Ethernet device structure.
982 * @param[in] attributes
983 * Pointer to flow attributes
985 * Pointer to error structure.
988 * 0 on success, a negative errno value otherwise and rte_errno is set.
991 mlx5_flow_validate_attributes(struct rte_eth_dev *dev,
992 const struct rte_flow_attr *attributes,
993 struct rte_flow_error *error)
995 struct priv *priv = dev->data->dev_private;
996 uint32_t priority_max = priv->config.flow_prio - 1;
998 if (attributes->group)
999 return rte_flow_error_set(error, ENOTSUP,
1000 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
1001 NULL, "groups is not supported");
1002 if (attributes->priority != MLX5_FLOW_PRIO_RSVD &&
1003 attributes->priority >= priority_max)
1004 return rte_flow_error_set(error, ENOTSUP,
1005 RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
1006 NULL, "priority out of range");
1007 if (attributes->egress)
1008 return rte_flow_error_set(error, ENOTSUP,
1009 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1010 "egress is not supported");
1011 if (attributes->transfer)
1012 return rte_flow_error_set(error, ENOTSUP,
1013 RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1014 NULL, "transfer is not supported");
1015 if (!attributes->ingress)
1016 return rte_flow_error_set(error, EINVAL,
1017 RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
1019 "ingress attribute is mandatory");
1024 * Validate Ethernet item.
1027 * Item specification.
1028 * @param[in] item_flags
1029 * Bit-fields that holds the items detected until now.
1031 * Pointer to error structure.
1034 * 0 on success, a negative errno value otherwise and rte_errno is set.
1037 mlx5_flow_validate_item_eth(const struct rte_flow_item *item,
1038 uint64_t item_flags,
1039 struct rte_flow_error *error)
1041 const struct rte_flow_item_eth *mask = item->mask;
1042 const struct rte_flow_item_eth nic_mask = {
1043 .dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1044 .src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1045 .type = RTE_BE16(0xffff),
1048 int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1049 const uint64_t ethm = tunnel ? MLX5_FLOW_LAYER_INNER_L2 :
1050 MLX5_FLOW_LAYER_OUTER_L2;
1052 if (item_flags & ethm)
1053 return rte_flow_error_set(error, ENOTSUP,
1054 RTE_FLOW_ERROR_TYPE_ITEM, item,
1055 "multiple L2 layers not supported");
1057 mask = &rte_flow_item_eth_mask;
1058 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1059 (const uint8_t *)&nic_mask,
1060 sizeof(struct rte_flow_item_eth),
1066 * Validate VLAN item.
1069 * Item specification.
1070 * @param[in] item_flags
1071 * Bit-fields that holds the items detected until now.
1073 * Pointer to error structure.
1076 * 0 on success, a negative errno value otherwise and rte_errno is set.
1079 mlx5_flow_validate_item_vlan(const struct rte_flow_item *item,
1080 uint64_t item_flags,
1081 struct rte_flow_error *error)
1083 const struct rte_flow_item_vlan *spec = item->spec;
1084 const struct rte_flow_item_vlan *mask = item->mask;
1085 const struct rte_flow_item_vlan nic_mask = {
1086 .tci = RTE_BE16(0x0fff),
1087 .inner_type = RTE_BE16(0xffff),
1089 uint16_t vlan_tag = 0;
1090 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1092 const uint64_t l34m = tunnel ? (MLX5_FLOW_LAYER_INNER_L3 |
1093 MLX5_FLOW_LAYER_INNER_L4) :
1094 (MLX5_FLOW_LAYER_OUTER_L3 |
1095 MLX5_FLOW_LAYER_OUTER_L4);
1096 const uint64_t vlanm = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
1097 MLX5_FLOW_LAYER_OUTER_VLAN;
1099 if (item_flags & vlanm)
1100 return rte_flow_error_set(error, EINVAL,
1101 RTE_FLOW_ERROR_TYPE_ITEM, item,
1102 "multiple VLAN layers not supported");
1103 else if ((item_flags & l34m) != 0)
1104 return rte_flow_error_set(error, EINVAL,
1105 RTE_FLOW_ERROR_TYPE_ITEM, item,
1106 "L2 layer cannot follow L3/L4 layer");
1108 mask = &rte_flow_item_vlan_mask;
1109 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1110 (const uint8_t *)&nic_mask,
1111 sizeof(struct rte_flow_item_vlan),
1116 vlan_tag = spec->tci;
1117 vlan_tag &= mask->tci;
1120 * From verbs perspective an empty VLAN is equivalent
1121 * to a packet without VLAN layer.
1124 return rte_flow_error_set(error, EINVAL,
1125 RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
1127 "VLAN cannot be empty");
1132 * Validate IPV4 item.
1135 * Item specification.
1136 * @param[in] item_flags
1137 * Bit-fields that holds the items detected until now.
1139 * Pointer to error structure.
1142 * 0 on success, a negative errno value otherwise and rte_errno is set.
1145 mlx5_flow_validate_item_ipv4(const struct rte_flow_item *item,
1146 uint64_t item_flags,
1147 struct rte_flow_error *error)
1149 const struct rte_flow_item_ipv4 *mask = item->mask;
1150 const struct rte_flow_item_ipv4 nic_mask = {
1152 .src_addr = RTE_BE32(0xffffffff),
1153 .dst_addr = RTE_BE32(0xffffffff),
1154 .type_of_service = 0xff,
1155 .next_proto_id = 0xff,
1158 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1159 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1160 MLX5_FLOW_LAYER_OUTER_L3;
1161 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1162 MLX5_FLOW_LAYER_OUTER_L4;
1165 if (item_flags & l3m)
1166 return rte_flow_error_set(error, ENOTSUP,
1167 RTE_FLOW_ERROR_TYPE_ITEM, item,
1168 "multiple L3 layers not supported");
1169 else if (item_flags & l4m)
1170 return rte_flow_error_set(error, EINVAL,
1171 RTE_FLOW_ERROR_TYPE_ITEM, item,
1172 "L3 cannot follow an L4 layer.");
1174 mask = &rte_flow_item_ipv4_mask;
1175 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1176 (const uint8_t *)&nic_mask,
1177 sizeof(struct rte_flow_item_ipv4),
1185 * Validate IPV6 item.
1188 * Item specification.
1189 * @param[in] item_flags
1190 * Bit-fields that holds the items detected until now.
1192 * Pointer to error structure.
1195 * 0 on success, a negative errno value otherwise and rte_errno is set.
1198 mlx5_flow_validate_item_ipv6(const struct rte_flow_item *item,
1199 uint64_t item_flags,
1200 struct rte_flow_error *error)
1202 const struct rte_flow_item_ipv6 *mask = item->mask;
1203 const struct rte_flow_item_ipv6 nic_mask = {
1206 "\xff\xff\xff\xff\xff\xff\xff\xff"
1207 "\xff\xff\xff\xff\xff\xff\xff\xff",
1209 "\xff\xff\xff\xff\xff\xff\xff\xff"
1210 "\xff\xff\xff\xff\xff\xff\xff\xff",
1211 .vtc_flow = RTE_BE32(0xffffffff),
1216 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1217 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1218 MLX5_FLOW_LAYER_OUTER_L3;
1219 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1220 MLX5_FLOW_LAYER_OUTER_L4;
1223 if (item_flags & l3m)
1224 return rte_flow_error_set(error, ENOTSUP,
1225 RTE_FLOW_ERROR_TYPE_ITEM, item,
1226 "multiple L3 layers not supported");
1227 else if (item_flags & l4m)
1228 return rte_flow_error_set(error, EINVAL,
1229 RTE_FLOW_ERROR_TYPE_ITEM, item,
1230 "L3 cannot follow an L4 layer.");
1232 * IPv6 is not recognised by the NIC inside a GRE tunnel.
1233 * Such support has to be disabled as the rule will be
1234 * accepted. Issue reproduced with Mellanox OFED 4.3-3.0.2.1 and
1235 * Mellanox OFED 4.4-1.0.0.0.
1237 if (tunnel && item_flags & MLX5_FLOW_LAYER_GRE)
1238 return rte_flow_error_set(error, ENOTSUP,
1239 RTE_FLOW_ERROR_TYPE_ITEM, item,
1240 "IPv6 inside a GRE tunnel is"
1241 " not recognised.");
1243 mask = &rte_flow_item_ipv6_mask;
1244 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1245 (const uint8_t *)&nic_mask,
1246 sizeof(struct rte_flow_item_ipv6),
1254 * Validate UDP item.
1257 * Item specification.
1258 * @param[in] item_flags
1259 * Bit-fields that holds the items detected until now.
1260 * @param[in] target_protocol
1261 * The next protocol in the previous item.
1262 * @param[in] flow_mask
1263 * mlx5 flow-specific (TCF, DV, verbs, etc.) supported header fields mask.
1265 * Pointer to error structure.
1268 * 0 on success, a negative errno value otherwise and rte_errno is set.
1271 mlx5_flow_validate_item_udp(const struct rte_flow_item *item,
1272 uint64_t item_flags,
1273 uint8_t target_protocol,
1274 struct rte_flow_error *error)
1276 const struct rte_flow_item_udp *mask = item->mask;
1277 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1278 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1279 MLX5_FLOW_LAYER_OUTER_L3;
1280 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1281 MLX5_FLOW_LAYER_OUTER_L4;
1284 if (target_protocol != 0xff && target_protocol != IPPROTO_UDP)
1285 return rte_flow_error_set(error, EINVAL,
1286 RTE_FLOW_ERROR_TYPE_ITEM, item,
1287 "protocol filtering not compatible"
1289 if (!(item_flags & l3m))
1290 return rte_flow_error_set(error, EINVAL,
1291 RTE_FLOW_ERROR_TYPE_ITEM, item,
1292 "L3 is mandatory to filter on L4");
1293 if (item_flags & l4m)
1294 return rte_flow_error_set(error, EINVAL,
1295 RTE_FLOW_ERROR_TYPE_ITEM, item,
1296 "multiple L4 layers not supported");
1298 mask = &rte_flow_item_udp_mask;
1299 ret = mlx5_flow_item_acceptable
1300 (item, (const uint8_t *)mask,
1301 (const uint8_t *)&rte_flow_item_udp_mask,
1302 sizeof(struct rte_flow_item_udp), error);
1309 * Validate TCP item.
1312 * Item specification.
1313 * @param[in] item_flags
1314 * Bit-fields that holds the items detected until now.
1315 * @param[in] target_protocol
1316 * The next protocol in the previous item.
1318 * Pointer to error structure.
1321 * 0 on success, a negative errno value otherwise and rte_errno is set.
1324 mlx5_flow_validate_item_tcp(const struct rte_flow_item *item,
1325 uint64_t item_flags,
1326 uint8_t target_protocol,
1327 const struct rte_flow_item_tcp *flow_mask,
1328 struct rte_flow_error *error)
1330 const struct rte_flow_item_tcp *mask = item->mask;
1331 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1332 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1333 MLX5_FLOW_LAYER_OUTER_L3;
1334 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1335 MLX5_FLOW_LAYER_OUTER_L4;
1339 if (target_protocol != 0xff && target_protocol != IPPROTO_TCP)
1340 return rte_flow_error_set(error, EINVAL,
1341 RTE_FLOW_ERROR_TYPE_ITEM, item,
1342 "protocol filtering not compatible"
1344 if (!(item_flags & l3m))
1345 return rte_flow_error_set(error, EINVAL,
1346 RTE_FLOW_ERROR_TYPE_ITEM, item,
1347 "L3 is mandatory to filter on L4");
1348 if (item_flags & l4m)
1349 return rte_flow_error_set(error, EINVAL,
1350 RTE_FLOW_ERROR_TYPE_ITEM, item,
1351 "multiple L4 layers not supported");
1353 mask = &rte_flow_item_tcp_mask;
1354 ret = mlx5_flow_item_acceptable
1355 (item, (const uint8_t *)mask,
1356 (const uint8_t *)flow_mask,
1357 sizeof(struct rte_flow_item_tcp), error);
1364 * Validate VXLAN item.
1367 * Item specification.
1368 * @param[in] item_flags
1369 * Bit-fields that holds the items detected until now.
1370 * @param[in] target_protocol
1371 * The next protocol in the previous item.
1373 * Pointer to error structure.
1376 * 0 on success, a negative errno value otherwise and rte_errno is set.
1379 mlx5_flow_validate_item_vxlan(const struct rte_flow_item *item,
1380 uint64_t item_flags,
1381 struct rte_flow_error *error)
1383 const struct rte_flow_item_vxlan *spec = item->spec;
1384 const struct rte_flow_item_vxlan *mask = item->mask;
1389 } id = { .vlan_id = 0, };
1390 uint32_t vlan_id = 0;
1393 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1394 return rte_flow_error_set(error, ENOTSUP,
1395 RTE_FLOW_ERROR_TYPE_ITEM, item,
1396 "multiple tunnel layers not"
1399 * Verify only UDPv4 is present as defined in
1400 * https://tools.ietf.org/html/rfc7348
1402 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
1403 return rte_flow_error_set(error, EINVAL,
1404 RTE_FLOW_ERROR_TYPE_ITEM, item,
1405 "no outer UDP layer found");
1407 mask = &rte_flow_item_vxlan_mask;
1408 ret = mlx5_flow_item_acceptable
1409 (item, (const uint8_t *)mask,
1410 (const uint8_t *)&rte_flow_item_vxlan_mask,
1411 sizeof(struct rte_flow_item_vxlan),
1416 memcpy(&id.vni[1], spec->vni, 3);
1417 vlan_id = id.vlan_id;
1418 memcpy(&id.vni[1], mask->vni, 3);
1419 vlan_id &= id.vlan_id;
1422 * Tunnel id 0 is equivalent as not adding a VXLAN layer, if
1423 * only this layer is defined in the Verbs specification it is
1424 * interpreted as wildcard and all packets will match this
1425 * rule, if it follows a full stack layer (ex: eth / ipv4 /
1426 * udp), all packets matching the layers before will also
1427 * match this rule. To avoid such situation, VNI 0 is
1428 * currently refused.
1431 return rte_flow_error_set(error, ENOTSUP,
1432 RTE_FLOW_ERROR_TYPE_ITEM, item,
1433 "VXLAN vni cannot be 0");
1434 if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
1435 return rte_flow_error_set(error, ENOTSUP,
1436 RTE_FLOW_ERROR_TYPE_ITEM, item,
1437 "VXLAN tunnel must be fully defined");
1442 * Validate VXLAN_GPE item.
1445 * Item specification.
1446 * @param[in] item_flags
1447 * Bit-fields that holds the items detected until now.
1449 * Pointer to the private data structure.
1450 * @param[in] target_protocol
1451 * The next protocol in the previous item.
1453 * Pointer to error structure.
1456 * 0 on success, a negative errno value otherwise and rte_errno is set.
1459 mlx5_flow_validate_item_vxlan_gpe(const struct rte_flow_item *item,
1460 uint64_t item_flags,
1461 struct rte_eth_dev *dev,
1462 struct rte_flow_error *error)
1464 struct priv *priv = dev->data->dev_private;
1465 const struct rte_flow_item_vxlan_gpe *spec = item->spec;
1466 const struct rte_flow_item_vxlan_gpe *mask = item->mask;
1471 } id = { .vlan_id = 0, };
1472 uint32_t vlan_id = 0;
1474 if (!priv->config.l3_vxlan_en)
1475 return rte_flow_error_set(error, ENOTSUP,
1476 RTE_FLOW_ERROR_TYPE_ITEM, item,
1477 "L3 VXLAN is not enabled by device"
1478 " parameter and/or not configured in"
1480 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1481 return rte_flow_error_set(error, ENOTSUP,
1482 RTE_FLOW_ERROR_TYPE_ITEM, item,
1483 "multiple tunnel layers not"
1486 * Verify only UDPv4 is present as defined in
1487 * https://tools.ietf.org/html/rfc7348
1489 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
1490 return rte_flow_error_set(error, EINVAL,
1491 RTE_FLOW_ERROR_TYPE_ITEM, item,
1492 "no outer UDP layer found");
1494 mask = &rte_flow_item_vxlan_gpe_mask;
1495 ret = mlx5_flow_item_acceptable
1496 (item, (const uint8_t *)mask,
1497 (const uint8_t *)&rte_flow_item_vxlan_gpe_mask,
1498 sizeof(struct rte_flow_item_vxlan_gpe),
1504 return rte_flow_error_set(error, ENOTSUP,
1505 RTE_FLOW_ERROR_TYPE_ITEM,
1507 "VxLAN-GPE protocol"
1509 memcpy(&id.vni[1], spec->vni, 3);
1510 vlan_id = id.vlan_id;
1511 memcpy(&id.vni[1], mask->vni, 3);
1512 vlan_id &= id.vlan_id;
1515 * Tunnel id 0 is equivalent as not adding a VXLAN layer, if only this
1516 * layer is defined in the Verbs specification it is interpreted as
1517 * wildcard and all packets will match this rule, if it follows a full
1518 * stack layer (ex: eth / ipv4 / udp), all packets matching the layers
1519 * before will also match this rule. To avoid such situation, VNI 0
1520 * is currently refused.
1523 return rte_flow_error_set(error, ENOTSUP,
1524 RTE_FLOW_ERROR_TYPE_ITEM, item,
1525 "VXLAN-GPE vni cannot be 0");
1526 if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
1527 return rte_flow_error_set(error, ENOTSUP,
1528 RTE_FLOW_ERROR_TYPE_ITEM, item,
1529 "VXLAN-GPE tunnel must be fully"
1535 * Validate GRE item.
1538 * Item specification.
1539 * @param[in] item_flags
1540 * Bit flags to mark detected items.
1541 * @param[in] target_protocol
1542 * The next protocol in the previous item.
1544 * Pointer to error structure.
1547 * 0 on success, a negative errno value otherwise and rte_errno is set.
1550 mlx5_flow_validate_item_gre(const struct rte_flow_item *item,
1551 uint64_t item_flags,
1552 uint8_t target_protocol,
1553 struct rte_flow_error *error)
1555 const struct rte_flow_item_gre *spec __rte_unused = item->spec;
1556 const struct rte_flow_item_gre *mask = item->mask;
1559 if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
1560 return rte_flow_error_set(error, EINVAL,
1561 RTE_FLOW_ERROR_TYPE_ITEM, item,
1562 "protocol filtering not compatible"
1563 " with this GRE layer");
1564 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1565 return rte_flow_error_set(error, ENOTSUP,
1566 RTE_FLOW_ERROR_TYPE_ITEM, item,
1567 "multiple tunnel layers not"
1569 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
1570 return rte_flow_error_set(error, ENOTSUP,
1571 RTE_FLOW_ERROR_TYPE_ITEM, item,
1572 "L3 Layer is missing");
1574 mask = &rte_flow_item_gre_mask;
1575 ret = mlx5_flow_item_acceptable
1576 (item, (const uint8_t *)mask,
1577 (const uint8_t *)&rte_flow_item_gre_mask,
1578 sizeof(struct rte_flow_item_gre), error);
1581 #ifndef HAVE_IBV_DEVICE_MPLS_SUPPORT
1582 if (spec && (spec->protocol & mask->protocol))
1583 return rte_flow_error_set(error, ENOTSUP,
1584 RTE_FLOW_ERROR_TYPE_ITEM, item,
1585 "without MPLS support the"
1586 " specification cannot be used for"
1593 * Validate MPLS item.
1596 * Item specification.
1597 * @param[in] item_flags
1598 * Bit-fields that holds the items detected until now.
1599 * @param[in] target_protocol
1600 * The next protocol in the previous item.
1602 * Pointer to error structure.
1605 * 0 on success, a negative errno value otherwise and rte_errno is set.
1608 mlx5_flow_validate_item_mpls(const struct rte_flow_item *item __rte_unused,
1609 uint64_t item_flags __rte_unused,
1610 uint8_t target_protocol __rte_unused,
1611 struct rte_flow_error *error)
1613 #ifdef HAVE_IBV_DEVICE_MPLS_SUPPORT
1614 const struct rte_flow_item_mpls *mask = item->mask;
1617 if (target_protocol != 0xff && target_protocol != IPPROTO_MPLS)
1618 return rte_flow_error_set(error, EINVAL,
1619 RTE_FLOW_ERROR_TYPE_ITEM, item,
1620 "protocol filtering not compatible"
1621 " with MPLS layer");
1622 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1623 return rte_flow_error_set(error, ENOTSUP,
1624 RTE_FLOW_ERROR_TYPE_ITEM, item,
1625 "multiple tunnel layers not"
1628 mask = &rte_flow_item_mpls_mask;
1629 ret = mlx5_flow_item_acceptable
1630 (item, (const uint8_t *)mask,
1631 (const uint8_t *)&rte_flow_item_mpls_mask,
1632 sizeof(struct rte_flow_item_mpls), error);
1637 return rte_flow_error_set(error, ENOTSUP,
1638 RTE_FLOW_ERROR_TYPE_ITEM, item,
1639 "MPLS is not supported by Verbs, please"
1644 flow_null_validate(struct rte_eth_dev *dev __rte_unused,
1645 const struct rte_flow_attr *attr __rte_unused,
1646 const struct rte_flow_item items[] __rte_unused,
1647 const struct rte_flow_action actions[] __rte_unused,
1648 struct rte_flow_error *error __rte_unused)
1650 rte_errno = ENOTSUP;
1654 static struct mlx5_flow *
1655 flow_null_prepare(const struct rte_flow_attr *attr __rte_unused,
1656 const struct rte_flow_item items[] __rte_unused,
1657 const struct rte_flow_action actions[] __rte_unused,
1658 uint64_t *item_flags __rte_unused,
1659 uint64_t *action_flags __rte_unused,
1660 struct rte_flow_error *error __rte_unused)
1662 rte_errno = ENOTSUP;
1667 flow_null_translate(struct rte_eth_dev *dev __rte_unused,
1668 struct mlx5_flow *dev_flow __rte_unused,
1669 const struct rte_flow_attr *attr __rte_unused,
1670 const struct rte_flow_item items[] __rte_unused,
1671 const struct rte_flow_action actions[] __rte_unused,
1672 struct rte_flow_error *error __rte_unused)
1674 rte_errno = ENOTSUP;
1679 flow_null_apply(struct rte_eth_dev *dev __rte_unused,
1680 struct rte_flow *flow __rte_unused,
1681 struct rte_flow_error *error __rte_unused)
1683 rte_errno = ENOTSUP;
1688 flow_null_remove(struct rte_eth_dev *dev __rte_unused,
1689 struct rte_flow *flow __rte_unused)
1694 flow_null_destroy(struct rte_eth_dev *dev __rte_unused,
1695 struct rte_flow *flow __rte_unused)
1700 flow_null_query(struct rte_eth_dev *dev __rte_unused,
1701 struct rte_flow *flow __rte_unused,
1702 const struct rte_flow_action *actions __rte_unused,
1703 void *data __rte_unused,
1704 struct rte_flow_error *error __rte_unused)
1706 rte_errno = ENOTSUP;
1710 /* Void driver to protect from null pointer reference. */
1711 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops = {
1712 .validate = flow_null_validate,
1713 .prepare = flow_null_prepare,
1714 .translate = flow_null_translate,
1715 .apply = flow_null_apply,
1716 .remove = flow_null_remove,
1717 .destroy = flow_null_destroy,
1718 .query = flow_null_query,
1722 * Select flow driver type according to flow attributes and device
1726 * Pointer to the dev structure.
1728 * Pointer to the flow attributes.
1731 * flow driver type, MLX5_FLOW_TYPE_MAX otherwise.
1733 static enum mlx5_flow_drv_type
1734 flow_get_drv_type(struct rte_eth_dev *dev, const struct rte_flow_attr *attr)
1736 struct priv *priv = dev->data->dev_private;
1737 enum mlx5_flow_drv_type type = MLX5_FLOW_TYPE_MAX;
1740 type = MLX5_FLOW_TYPE_TCF;
1742 type = priv->config.dv_flow_en ? MLX5_FLOW_TYPE_DV :
1743 MLX5_FLOW_TYPE_VERBS;
1747 #define flow_get_drv_ops(type) flow_drv_ops[type]
1750 * Flow driver validation API. This abstracts calling driver specific functions.
1751 * The type of flow driver is determined according to flow attributes.
1754 * Pointer to the dev structure.
1756 * Pointer to the flow attributes.
1758 * Pointer to the list of items.
1759 * @param[in] actions
1760 * Pointer to the list of actions.
1762 * Pointer to the error structure.
1765 * 0 on success, a negative errno value otherwise and rte_ernno is set.
1768 flow_drv_validate(struct rte_eth_dev *dev,
1769 const struct rte_flow_attr *attr,
1770 const struct rte_flow_item items[],
1771 const struct rte_flow_action actions[],
1772 struct rte_flow_error *error)
1774 const struct mlx5_flow_driver_ops *fops;
1775 enum mlx5_flow_drv_type type = flow_get_drv_type(dev, attr);
1777 fops = flow_get_drv_ops(type);
1778 return fops->validate(dev, attr, items, actions, error);
1782 * Flow driver preparation API. This abstracts calling driver specific
1783 * functions. Parent flow (rte_flow) should have driver type (drv_type). It
1784 * calculates the size of memory required for device flow, allocates the memory,
1785 * initializes the device flow and returns the pointer.
1788 * Pointer to the flow attributes.
1790 * Pointer to the list of items.
1791 * @param[in] actions
1792 * Pointer to the list of actions.
1793 * @param[out] item_flags
1794 * Pointer to bit mask of all items detected.
1795 * @param[out] action_flags
1796 * Pointer to bit mask of all actions detected.
1798 * Pointer to the error structure.
1801 * Pointer to device flow on success, otherwise NULL and rte_ernno is set.
1803 static inline struct mlx5_flow *
1804 flow_drv_prepare(struct rte_flow *flow,
1805 const struct rte_flow_attr *attr,
1806 const struct rte_flow_item items[],
1807 const struct rte_flow_action actions[],
1808 uint64_t *item_flags,
1809 uint64_t *action_flags,
1810 struct rte_flow_error *error)
1812 const struct mlx5_flow_driver_ops *fops;
1813 enum mlx5_flow_drv_type type = flow->drv_type;
1815 assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
1816 fops = flow_get_drv_ops(type);
1817 return fops->prepare(attr, items, actions, item_flags, action_flags,
1822 * Flow driver translation API. This abstracts calling driver specific
1823 * functions. Parent flow (rte_flow) should have driver type (drv_type). It
1824 * translates a generic flow into a driver flow. flow_drv_prepare() must
1829 * Pointer to the rte dev structure.
1830 * @param[in, out] dev_flow
1831 * Pointer to the mlx5 flow.
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 * 0 on success, a negative errno value otherwise and rte_ernno is set.
1845 flow_drv_translate(struct rte_eth_dev *dev, struct mlx5_flow *dev_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 = dev_flow->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->translate(dev, dev_flow, attr, items, actions, error);
1860 * Flow driver apply API. This abstracts calling driver specific functions.
1861 * Parent flow (rte_flow) should have driver type (drv_type). It applies
1862 * translated driver flows on to device. flow_drv_translate() must precede.
1865 * Pointer to Ethernet device structure.
1866 * @param[in, out] flow
1867 * Pointer to flow structure.
1869 * Pointer to error structure.
1872 * 0 on success, a negative errno value otherwise and rte_errno is set.
1875 flow_drv_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
1876 struct rte_flow_error *error)
1878 const struct mlx5_flow_driver_ops *fops;
1879 enum mlx5_flow_drv_type type = flow->drv_type;
1881 assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
1882 fops = flow_get_drv_ops(type);
1883 return fops->apply(dev, flow, error);
1887 * Flow driver remove API. This abstracts calling driver specific functions.
1888 * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
1889 * on device. All the resources of the flow should be freed by calling
1890 * flow_dv_destroy().
1893 * Pointer to Ethernet device.
1894 * @param[in, out] flow
1895 * Pointer to flow structure.
1898 flow_drv_remove(struct rte_eth_dev *dev, struct rte_flow *flow)
1900 const struct mlx5_flow_driver_ops *fops;
1901 enum mlx5_flow_drv_type type = flow->drv_type;
1903 assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
1904 fops = flow_get_drv_ops(type);
1905 fops->remove(dev, flow);
1909 * Flow driver destroy API. This abstracts calling driver specific functions.
1910 * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
1911 * on device and releases resources of the flow.
1914 * Pointer to Ethernet device.
1915 * @param[in, out] flow
1916 * Pointer to flow structure.
1919 flow_drv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
1921 const struct mlx5_flow_driver_ops *fops;
1922 enum mlx5_flow_drv_type type = flow->drv_type;
1924 assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
1925 fops = flow_get_drv_ops(type);
1926 fops->destroy(dev, flow);
1930 * Validate a flow supported by the NIC.
1932 * @see rte_flow_validate()
1936 mlx5_flow_validate(struct rte_eth_dev *dev,
1937 const struct rte_flow_attr *attr,
1938 const struct rte_flow_item items[],
1939 const struct rte_flow_action actions[],
1940 struct rte_flow_error *error)
1944 ret = flow_drv_validate(dev, attr, items, actions, error);
1951 * Get RSS action from the action list.
1953 * @param[in] actions
1954 * Pointer to the list of actions.
1957 * Pointer to the RSS action if exist, else return NULL.
1959 static const struct rte_flow_action_rss*
1960 flow_get_rss_action(const struct rte_flow_action actions[])
1962 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
1963 switch (actions->type) {
1964 case RTE_FLOW_ACTION_TYPE_RSS:
1965 return (const struct rte_flow_action_rss *)
1975 find_graph_root(const struct rte_flow_item pattern[], uint32_t rss_level)
1977 const struct rte_flow_item *item;
1978 unsigned int has_vlan = 0;
1980 for (item = pattern; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
1981 if (item->type == RTE_FLOW_ITEM_TYPE_VLAN) {
1987 return rss_level < 2 ? MLX5_EXPANSION_ROOT_ETH_VLAN :
1988 MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN;
1989 return rss_level < 2 ? MLX5_EXPANSION_ROOT :
1990 MLX5_EXPANSION_ROOT_OUTER;
1994 * Create a flow and add it to @p list.
1997 * Pointer to Ethernet device.
1999 * Pointer to a TAILQ flow list.
2001 * Flow rule attributes.
2003 * Pattern specification (list terminated by the END pattern item).
2004 * @param[in] actions
2005 * Associated actions (list terminated by the END action).
2007 * Perform verbose error reporting if not NULL.
2010 * A flow on success, NULL otherwise and rte_errno is set.
2012 static struct rte_flow *
2013 flow_list_create(struct rte_eth_dev *dev, struct mlx5_flows *list,
2014 const struct rte_flow_attr *attr,
2015 const struct rte_flow_item items[],
2016 const struct rte_flow_action actions[],
2017 struct rte_flow_error *error)
2019 struct rte_flow *flow = NULL;
2020 struct mlx5_flow *dev_flow;
2021 uint64_t action_flags = 0;
2022 uint64_t item_flags = 0;
2023 const struct rte_flow_action_rss *rss;
2025 struct rte_flow_expand_rss buf;
2026 uint8_t buffer[2048];
2028 struct rte_flow_expand_rss *buf = &expand_buffer.buf;
2033 ret = flow_drv_validate(dev, attr, items, actions, error);
2036 flow_size = sizeof(struct rte_flow);
2037 rss = flow_get_rss_action(actions);
2039 flow_size += RTE_ALIGN_CEIL(rss->queue_num * sizeof(uint16_t),
2042 flow_size += RTE_ALIGN_CEIL(sizeof(uint16_t), sizeof(void *));
2043 flow = rte_calloc(__func__, 1, flow_size, 0);
2044 flow->drv_type = flow_get_drv_type(dev, attr);
2045 assert(flow->drv_type > MLX5_FLOW_TYPE_MIN &&
2046 flow->drv_type < MLX5_FLOW_TYPE_MAX);
2047 flow->queue = (void *)(flow + 1);
2048 LIST_INIT(&flow->dev_flows);
2049 if (rss && rss->types) {
2050 unsigned int graph_root;
2052 graph_root = find_graph_root(items, rss->level);
2053 ret = rte_flow_expand_rss(buf, sizeof(expand_buffer.buffer),
2055 mlx5_support_expansion,
2058 (unsigned int)ret < sizeof(expand_buffer.buffer));
2061 buf->entry[0].pattern = (void *)(uintptr_t)items;
2063 for (i = 0; i < buf->entries; ++i) {
2064 dev_flow = flow_drv_prepare(flow, attr, buf->entry[i].pattern,
2065 actions, &item_flags, &action_flags,
2069 dev_flow->flow = flow;
2070 dev_flow->layers = item_flags;
2071 /* Store actions once as expanded flows have same actions. */
2073 flow->actions = action_flags;
2074 assert(flow->actions == action_flags);
2075 LIST_INSERT_HEAD(&flow->dev_flows, dev_flow, next);
2076 ret = flow_drv_translate(dev, dev_flow, attr,
2077 buf->entry[i].pattern,
2082 if (dev->data->dev_started) {
2083 ret = flow_drv_apply(dev, flow, error);
2087 TAILQ_INSERT_TAIL(list, flow, next);
2088 flow_rxq_flags_set(dev, flow);
2091 ret = rte_errno; /* Save rte_errno before cleanup. */
2093 flow_drv_destroy(dev, flow);
2095 rte_errno = ret; /* Restore rte_errno. */
2102 * @see rte_flow_create()
2106 mlx5_flow_create(struct rte_eth_dev *dev,
2107 const struct rte_flow_attr *attr,
2108 const struct rte_flow_item items[],
2109 const struct rte_flow_action actions[],
2110 struct rte_flow_error *error)
2112 return flow_list_create(dev,
2113 &((struct priv *)dev->data->dev_private)->flows,
2114 attr, items, actions, error);
2118 * Destroy a flow in a list.
2121 * Pointer to Ethernet device.
2123 * Pointer to a TAILQ flow list.
2128 flow_list_destroy(struct rte_eth_dev *dev, struct mlx5_flows *list,
2129 struct rte_flow *flow)
2131 flow_drv_destroy(dev, flow);
2132 TAILQ_REMOVE(list, flow, next);
2134 * Update RX queue flags only if port is started, otherwise it is
2137 if (dev->data->dev_started)
2138 flow_rxq_flags_trim(dev, flow);
2139 rte_free(flow->fdir);
2144 * Destroy all flows.
2147 * Pointer to Ethernet device.
2149 * Pointer to a TAILQ flow list.
2152 mlx5_flow_list_flush(struct rte_eth_dev *dev, struct mlx5_flows *list)
2154 while (!TAILQ_EMPTY(list)) {
2155 struct rte_flow *flow;
2157 flow = TAILQ_FIRST(list);
2158 flow_list_destroy(dev, list, flow);
2166 * Pointer to Ethernet device.
2168 * Pointer to a TAILQ flow list.
2171 mlx5_flow_stop(struct rte_eth_dev *dev, struct mlx5_flows *list)
2173 struct rte_flow *flow;
2175 TAILQ_FOREACH_REVERSE(flow, list, mlx5_flows, next)
2176 flow_drv_remove(dev, flow);
2177 flow_rxq_flags_clear(dev);
2184 * Pointer to Ethernet device.
2186 * Pointer to a TAILQ flow list.
2189 * 0 on success, a negative errno value otherwise and rte_errno is set.
2192 mlx5_flow_start(struct rte_eth_dev *dev, struct mlx5_flows *list)
2194 struct rte_flow *flow;
2195 struct rte_flow_error error;
2198 TAILQ_FOREACH(flow, list, next) {
2199 ret = flow_drv_apply(dev, flow, &error);
2202 flow_rxq_flags_set(dev, flow);
2206 ret = rte_errno; /* Save rte_errno before cleanup. */
2207 mlx5_flow_stop(dev, list);
2208 rte_errno = ret; /* Restore rte_errno. */
2213 * Verify the flow list is empty
2216 * Pointer to Ethernet device.
2218 * @return the number of flows not released.
2221 mlx5_flow_verify(struct rte_eth_dev *dev)
2223 struct priv *priv = dev->data->dev_private;
2224 struct rte_flow *flow;
2227 TAILQ_FOREACH(flow, &priv->flows, next) {
2228 DRV_LOG(DEBUG, "port %u flow %p still referenced",
2229 dev->data->port_id, (void *)flow);
2236 * Enable a control flow configured from the control plane.
2239 * Pointer to Ethernet device.
2241 * An Ethernet flow spec to apply.
2243 * An Ethernet flow mask to apply.
2245 * A VLAN flow spec to apply.
2247 * A VLAN flow mask to apply.
2250 * 0 on success, a negative errno value otherwise and rte_errno is set.
2253 mlx5_ctrl_flow_vlan(struct rte_eth_dev *dev,
2254 struct rte_flow_item_eth *eth_spec,
2255 struct rte_flow_item_eth *eth_mask,
2256 struct rte_flow_item_vlan *vlan_spec,
2257 struct rte_flow_item_vlan *vlan_mask)
2259 struct priv *priv = dev->data->dev_private;
2260 const struct rte_flow_attr attr = {
2262 .priority = MLX5_FLOW_PRIO_RSVD,
2264 struct rte_flow_item items[] = {
2266 .type = RTE_FLOW_ITEM_TYPE_ETH,
2272 .type = (vlan_spec) ? RTE_FLOW_ITEM_TYPE_VLAN :
2273 RTE_FLOW_ITEM_TYPE_END,
2279 .type = RTE_FLOW_ITEM_TYPE_END,
2282 uint16_t queue[priv->reta_idx_n];
2283 struct rte_flow_action_rss action_rss = {
2284 .func = RTE_ETH_HASH_FUNCTION_DEFAULT,
2286 .types = priv->rss_conf.rss_hf,
2287 .key_len = priv->rss_conf.rss_key_len,
2288 .queue_num = priv->reta_idx_n,
2289 .key = priv->rss_conf.rss_key,
2292 struct rte_flow_action actions[] = {
2294 .type = RTE_FLOW_ACTION_TYPE_RSS,
2295 .conf = &action_rss,
2298 .type = RTE_FLOW_ACTION_TYPE_END,
2301 struct rte_flow *flow;
2302 struct rte_flow_error error;
2305 if (!priv->reta_idx_n) {
2309 for (i = 0; i != priv->reta_idx_n; ++i)
2310 queue[i] = (*priv->reta_idx)[i];
2311 flow = flow_list_create(dev, &priv->ctrl_flows,
2312 &attr, items, actions, &error);
2319 * Enable a flow control configured from the control plane.
2322 * Pointer to Ethernet device.
2324 * An Ethernet flow spec to apply.
2326 * An Ethernet flow mask to apply.
2329 * 0 on success, a negative errno value otherwise and rte_errno is set.
2332 mlx5_ctrl_flow(struct rte_eth_dev *dev,
2333 struct rte_flow_item_eth *eth_spec,
2334 struct rte_flow_item_eth *eth_mask)
2336 return mlx5_ctrl_flow_vlan(dev, eth_spec, eth_mask, NULL, NULL);
2342 * @see rte_flow_destroy()
2346 mlx5_flow_destroy(struct rte_eth_dev *dev,
2347 struct rte_flow *flow,
2348 struct rte_flow_error *error __rte_unused)
2350 struct priv *priv = dev->data->dev_private;
2352 flow_list_destroy(dev, &priv->flows, flow);
2357 * Destroy all flows.
2359 * @see rte_flow_flush()
2363 mlx5_flow_flush(struct rte_eth_dev *dev,
2364 struct rte_flow_error *error __rte_unused)
2366 struct priv *priv = dev->data->dev_private;
2368 mlx5_flow_list_flush(dev, &priv->flows);
2375 * @see rte_flow_isolate()
2379 mlx5_flow_isolate(struct rte_eth_dev *dev,
2381 struct rte_flow_error *error)
2383 struct priv *priv = dev->data->dev_private;
2385 if (dev->data->dev_started) {
2386 rte_flow_error_set(error, EBUSY,
2387 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
2389 "port must be stopped first");
2392 priv->isolated = !!enable;
2394 dev->dev_ops = &mlx5_dev_ops_isolate;
2396 dev->dev_ops = &mlx5_dev_ops;
2403 * @see rte_flow_query()
2407 flow_drv_query(struct rte_eth_dev *dev,
2408 struct rte_flow *flow,
2409 const struct rte_flow_action *actions,
2411 struct rte_flow_error *error)
2413 const struct mlx5_flow_driver_ops *fops;
2414 enum mlx5_flow_drv_type ftype = flow->drv_type;
2416 assert(ftype > MLX5_FLOW_TYPE_MIN && ftype < MLX5_FLOW_TYPE_MAX);
2417 fops = flow_get_drv_ops(ftype);
2419 return fops->query(dev, flow, actions, data, error);
2425 * @see rte_flow_query()
2429 mlx5_flow_query(struct rte_eth_dev *dev,
2430 struct rte_flow *flow,
2431 const struct rte_flow_action *actions,
2433 struct rte_flow_error *error)
2437 ret = flow_drv_query(dev, flow, actions, data, error);
2444 * Convert a flow director filter to a generic flow.
2447 * Pointer to Ethernet device.
2448 * @param fdir_filter
2449 * Flow director filter to add.
2451 * Generic flow parameters structure.
2454 * 0 on success, a negative errno value otherwise and rte_errno is set.
2457 flow_fdir_filter_convert(struct rte_eth_dev *dev,
2458 const struct rte_eth_fdir_filter *fdir_filter,
2459 struct mlx5_fdir *attributes)
2461 struct priv *priv = dev->data->dev_private;
2462 const struct rte_eth_fdir_input *input = &fdir_filter->input;
2463 const struct rte_eth_fdir_masks *mask =
2464 &dev->data->dev_conf.fdir_conf.mask;
2466 /* Validate queue number. */
2467 if (fdir_filter->action.rx_queue >= priv->rxqs_n) {
2468 DRV_LOG(ERR, "port %u invalid queue number %d",
2469 dev->data->port_id, fdir_filter->action.rx_queue);
2473 attributes->attr.ingress = 1;
2474 attributes->items[0] = (struct rte_flow_item) {
2475 .type = RTE_FLOW_ITEM_TYPE_ETH,
2476 .spec = &attributes->l2,
2477 .mask = &attributes->l2_mask,
2479 switch (fdir_filter->action.behavior) {
2480 case RTE_ETH_FDIR_ACCEPT:
2481 attributes->actions[0] = (struct rte_flow_action){
2482 .type = RTE_FLOW_ACTION_TYPE_QUEUE,
2483 .conf = &attributes->queue,
2486 case RTE_ETH_FDIR_REJECT:
2487 attributes->actions[0] = (struct rte_flow_action){
2488 .type = RTE_FLOW_ACTION_TYPE_DROP,
2492 DRV_LOG(ERR, "port %u invalid behavior %d",
2494 fdir_filter->action.behavior);
2495 rte_errno = ENOTSUP;
2498 attributes->queue.index = fdir_filter->action.rx_queue;
2500 switch (fdir_filter->input.flow_type) {
2501 case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
2502 case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
2503 case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
2504 attributes->l3.ipv4.hdr = (struct ipv4_hdr){
2505 .src_addr = input->flow.ip4_flow.src_ip,
2506 .dst_addr = input->flow.ip4_flow.dst_ip,
2507 .time_to_live = input->flow.ip4_flow.ttl,
2508 .type_of_service = input->flow.ip4_flow.tos,
2510 attributes->l3_mask.ipv4.hdr = (struct ipv4_hdr){
2511 .src_addr = mask->ipv4_mask.src_ip,
2512 .dst_addr = mask->ipv4_mask.dst_ip,
2513 .time_to_live = mask->ipv4_mask.ttl,
2514 .type_of_service = mask->ipv4_mask.tos,
2515 .next_proto_id = mask->ipv4_mask.proto,
2517 attributes->items[1] = (struct rte_flow_item){
2518 .type = RTE_FLOW_ITEM_TYPE_IPV4,
2519 .spec = &attributes->l3,
2520 .mask = &attributes->l3_mask,
2523 case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
2524 case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
2525 case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
2526 attributes->l3.ipv6.hdr = (struct ipv6_hdr){
2527 .hop_limits = input->flow.ipv6_flow.hop_limits,
2528 .proto = input->flow.ipv6_flow.proto,
2531 memcpy(attributes->l3.ipv6.hdr.src_addr,
2532 input->flow.ipv6_flow.src_ip,
2533 RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
2534 memcpy(attributes->l3.ipv6.hdr.dst_addr,
2535 input->flow.ipv6_flow.dst_ip,
2536 RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
2537 memcpy(attributes->l3_mask.ipv6.hdr.src_addr,
2538 mask->ipv6_mask.src_ip,
2539 RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
2540 memcpy(attributes->l3_mask.ipv6.hdr.dst_addr,
2541 mask->ipv6_mask.dst_ip,
2542 RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
2543 attributes->items[1] = (struct rte_flow_item){
2544 .type = RTE_FLOW_ITEM_TYPE_IPV6,
2545 .spec = &attributes->l3,
2546 .mask = &attributes->l3_mask,
2550 DRV_LOG(ERR, "port %u invalid flow type%d",
2551 dev->data->port_id, fdir_filter->input.flow_type);
2552 rte_errno = ENOTSUP;
2556 switch (fdir_filter->input.flow_type) {
2557 case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
2558 attributes->l4.udp.hdr = (struct udp_hdr){
2559 .src_port = input->flow.udp4_flow.src_port,
2560 .dst_port = input->flow.udp4_flow.dst_port,
2562 attributes->l4_mask.udp.hdr = (struct udp_hdr){
2563 .src_port = mask->src_port_mask,
2564 .dst_port = mask->dst_port_mask,
2566 attributes->items[2] = (struct rte_flow_item){
2567 .type = RTE_FLOW_ITEM_TYPE_UDP,
2568 .spec = &attributes->l4,
2569 .mask = &attributes->l4_mask,
2572 case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
2573 attributes->l4.tcp.hdr = (struct tcp_hdr){
2574 .src_port = input->flow.tcp4_flow.src_port,
2575 .dst_port = input->flow.tcp4_flow.dst_port,
2577 attributes->l4_mask.tcp.hdr = (struct tcp_hdr){
2578 .src_port = mask->src_port_mask,
2579 .dst_port = mask->dst_port_mask,
2581 attributes->items[2] = (struct rte_flow_item){
2582 .type = RTE_FLOW_ITEM_TYPE_TCP,
2583 .spec = &attributes->l4,
2584 .mask = &attributes->l4_mask,
2587 case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
2588 attributes->l4.udp.hdr = (struct udp_hdr){
2589 .src_port = input->flow.udp6_flow.src_port,
2590 .dst_port = input->flow.udp6_flow.dst_port,
2592 attributes->l4_mask.udp.hdr = (struct udp_hdr){
2593 .src_port = mask->src_port_mask,
2594 .dst_port = mask->dst_port_mask,
2596 attributes->items[2] = (struct rte_flow_item){
2597 .type = RTE_FLOW_ITEM_TYPE_UDP,
2598 .spec = &attributes->l4,
2599 .mask = &attributes->l4_mask,
2602 case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
2603 attributes->l4.tcp.hdr = (struct tcp_hdr){
2604 .src_port = input->flow.tcp6_flow.src_port,
2605 .dst_port = input->flow.tcp6_flow.dst_port,
2607 attributes->l4_mask.tcp.hdr = (struct tcp_hdr){
2608 .src_port = mask->src_port_mask,
2609 .dst_port = mask->dst_port_mask,
2611 attributes->items[2] = (struct rte_flow_item){
2612 .type = RTE_FLOW_ITEM_TYPE_TCP,
2613 .spec = &attributes->l4,
2614 .mask = &attributes->l4_mask,
2617 case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
2618 case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
2621 DRV_LOG(ERR, "port %u invalid flow type%d",
2622 dev->data->port_id, fdir_filter->input.flow_type);
2623 rte_errno = ENOTSUP;
2629 #define FLOW_FDIR_CMP(f1, f2, fld) \
2630 memcmp(&(f1)->fld, &(f2)->fld, sizeof(f1->fld))
2633 * Compare two FDIR flows. If items and actions are identical, the two flows are
2637 * Pointer to Ethernet device.
2639 * FDIR flow to compare.
2641 * FDIR flow to compare.
2644 * Zero on match, 1 otherwise.
2647 flow_fdir_cmp(const struct mlx5_fdir *f1, const struct mlx5_fdir *f2)
2649 if (FLOW_FDIR_CMP(f1, f2, attr) ||
2650 FLOW_FDIR_CMP(f1, f2, l2) ||
2651 FLOW_FDIR_CMP(f1, f2, l2_mask) ||
2652 FLOW_FDIR_CMP(f1, f2, l3) ||
2653 FLOW_FDIR_CMP(f1, f2, l3_mask) ||
2654 FLOW_FDIR_CMP(f1, f2, l4) ||
2655 FLOW_FDIR_CMP(f1, f2, l4_mask) ||
2656 FLOW_FDIR_CMP(f1, f2, actions[0]))
2658 if (f1->actions[0].type == RTE_FLOW_ACTION_TYPE_QUEUE &&
2659 FLOW_FDIR_CMP(f1, f2, queue))
2665 * Search device flow list to find out a matched FDIR flow.
2668 * Pointer to Ethernet device.
2670 * FDIR flow to lookup.
2673 * Pointer of flow if found, NULL otherwise.
2675 static struct rte_flow *
2676 flow_fdir_filter_lookup(struct rte_eth_dev *dev, struct mlx5_fdir *fdir_flow)
2678 struct priv *priv = dev->data->dev_private;
2679 struct rte_flow *flow = NULL;
2682 TAILQ_FOREACH(flow, &priv->flows, next) {
2683 if (flow->fdir && !flow_fdir_cmp(flow->fdir, fdir_flow)) {
2684 DRV_LOG(DEBUG, "port %u found FDIR flow %p",
2685 dev->data->port_id, (void *)flow);
2693 * Add new flow director filter and store it in list.
2696 * Pointer to Ethernet device.
2697 * @param fdir_filter
2698 * Flow director filter to add.
2701 * 0 on success, a negative errno value otherwise and rte_errno is set.
2704 flow_fdir_filter_add(struct rte_eth_dev *dev,
2705 const struct rte_eth_fdir_filter *fdir_filter)
2707 struct priv *priv = dev->data->dev_private;
2708 struct mlx5_fdir *fdir_flow;
2709 struct rte_flow *flow;
2712 fdir_flow = rte_zmalloc(__func__, sizeof(*fdir_flow), 0);
2717 ret = flow_fdir_filter_convert(dev, fdir_filter, fdir_flow);
2720 flow = flow_fdir_filter_lookup(dev, fdir_flow);
2725 flow = flow_list_create(dev, &priv->flows, &fdir_flow->attr,
2726 fdir_flow->items, fdir_flow->actions, NULL);
2729 assert(!flow->fdir);
2730 flow->fdir = fdir_flow;
2731 DRV_LOG(DEBUG, "port %u created FDIR flow %p",
2732 dev->data->port_id, (void *)flow);
2735 rte_free(fdir_flow);
2740 * Delete specific filter.
2743 * Pointer to Ethernet device.
2744 * @param fdir_filter
2745 * Filter to be deleted.
2748 * 0 on success, a negative errno value otherwise and rte_errno is set.
2751 flow_fdir_filter_delete(struct rte_eth_dev *dev,
2752 const struct rte_eth_fdir_filter *fdir_filter)
2754 struct priv *priv = dev->data->dev_private;
2755 struct rte_flow *flow;
2756 struct mlx5_fdir fdir_flow = {
2761 ret = flow_fdir_filter_convert(dev, fdir_filter, &fdir_flow);
2764 flow = flow_fdir_filter_lookup(dev, &fdir_flow);
2769 flow_list_destroy(dev, &priv->flows, flow);
2770 DRV_LOG(DEBUG, "port %u deleted FDIR flow %p",
2771 dev->data->port_id, (void *)flow);
2776 * Update queue for specific filter.
2779 * Pointer to Ethernet device.
2780 * @param fdir_filter
2781 * Filter to be updated.
2784 * 0 on success, a negative errno value otherwise and rte_errno is set.
2787 flow_fdir_filter_update(struct rte_eth_dev *dev,
2788 const struct rte_eth_fdir_filter *fdir_filter)
2792 ret = flow_fdir_filter_delete(dev, fdir_filter);
2795 return flow_fdir_filter_add(dev, fdir_filter);
2799 * Flush all filters.
2802 * Pointer to Ethernet device.
2805 flow_fdir_filter_flush(struct rte_eth_dev *dev)
2807 struct priv *priv = dev->data->dev_private;
2809 mlx5_flow_list_flush(dev, &priv->flows);
2813 * Get flow director information.
2816 * Pointer to Ethernet device.
2817 * @param[out] fdir_info
2818 * Resulting flow director information.
2821 flow_fdir_info_get(struct rte_eth_dev *dev, struct rte_eth_fdir_info *fdir_info)
2823 struct rte_eth_fdir_masks *mask =
2824 &dev->data->dev_conf.fdir_conf.mask;
2826 fdir_info->mode = dev->data->dev_conf.fdir_conf.mode;
2827 fdir_info->guarant_spc = 0;
2828 rte_memcpy(&fdir_info->mask, mask, sizeof(fdir_info->mask));
2829 fdir_info->max_flexpayload = 0;
2830 fdir_info->flow_types_mask[0] = 0;
2831 fdir_info->flex_payload_unit = 0;
2832 fdir_info->max_flex_payload_segment_num = 0;
2833 fdir_info->flex_payload_limit = 0;
2834 memset(&fdir_info->flex_conf, 0, sizeof(fdir_info->flex_conf));
2838 * Deal with flow director operations.
2841 * Pointer to Ethernet device.
2843 * Operation to perform.
2845 * Pointer to operation-specific structure.
2848 * 0 on success, a negative errno value otherwise and rte_errno is set.
2851 flow_fdir_ctrl_func(struct rte_eth_dev *dev, enum rte_filter_op filter_op,
2854 enum rte_fdir_mode fdir_mode =
2855 dev->data->dev_conf.fdir_conf.mode;
2857 if (filter_op == RTE_ETH_FILTER_NOP)
2859 if (fdir_mode != RTE_FDIR_MODE_PERFECT &&
2860 fdir_mode != RTE_FDIR_MODE_PERFECT_MAC_VLAN) {
2861 DRV_LOG(ERR, "port %u flow director mode %d not supported",
2862 dev->data->port_id, fdir_mode);
2866 switch (filter_op) {
2867 case RTE_ETH_FILTER_ADD:
2868 return flow_fdir_filter_add(dev, arg);
2869 case RTE_ETH_FILTER_UPDATE:
2870 return flow_fdir_filter_update(dev, arg);
2871 case RTE_ETH_FILTER_DELETE:
2872 return flow_fdir_filter_delete(dev, arg);
2873 case RTE_ETH_FILTER_FLUSH:
2874 flow_fdir_filter_flush(dev);
2876 case RTE_ETH_FILTER_INFO:
2877 flow_fdir_info_get(dev, arg);
2880 DRV_LOG(DEBUG, "port %u unknown operation %u",
2881 dev->data->port_id, filter_op);
2889 * Manage filter operations.
2892 * Pointer to Ethernet device structure.
2893 * @param filter_type
2896 * Operation to perform.
2898 * Pointer to operation-specific structure.
2901 * 0 on success, a negative errno value otherwise and rte_errno is set.
2904 mlx5_dev_filter_ctrl(struct rte_eth_dev *dev,
2905 enum rte_filter_type filter_type,
2906 enum rte_filter_op filter_op,
2909 switch (filter_type) {
2910 case RTE_ETH_FILTER_GENERIC:
2911 if (filter_op != RTE_ETH_FILTER_GET) {
2915 *(const void **)arg = &mlx5_flow_ops;
2917 case RTE_ETH_FILTER_FDIR:
2918 return flow_fdir_ctrl_func(dev, filter_op, arg);
2920 DRV_LOG(ERR, "port %u filter type (%d) not supported",
2921 dev->data->port_id, filter_type);
2922 rte_errno = ENOTSUP;