1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright 2016 6WIND S.A.
3 * Copyright 2016 Mellanox Technologies, Ltd
6 #include <netinet/in.h>
14 /* ISO C doesn't support unnamed structs/unions, disabling -pedantic. */
16 #pragma GCC diagnostic ignored "-Wpedantic"
18 #include <infiniband/verbs.h>
20 #pragma GCC diagnostic error "-Wpedantic"
23 #include <rte_common.h>
24 #include <rte_ether.h>
25 #include <rte_ethdev_driver.h>
27 #include <rte_flow_driver.h>
28 #include <rte_malloc.h>
31 #include <mlx5_glue.h>
32 #include <mlx5_devx_cmds.h>
35 #include "mlx5_defs.h"
37 #include "mlx5_flow.h"
38 #include "mlx5_rxtx.h"
40 /* Dev ops structure defined in mlx5.c */
41 extern const struct eth_dev_ops mlx5_dev_ops;
42 extern const struct eth_dev_ops mlx5_dev_ops_isolate;
44 /** Device flow drivers. */
45 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
46 extern const struct mlx5_flow_driver_ops mlx5_flow_dv_drv_ops;
48 extern const struct mlx5_flow_driver_ops mlx5_flow_verbs_drv_ops;
50 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops;
52 const struct mlx5_flow_driver_ops *flow_drv_ops[] = {
53 [MLX5_FLOW_TYPE_MIN] = &mlx5_flow_null_drv_ops,
54 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
55 [MLX5_FLOW_TYPE_DV] = &mlx5_flow_dv_drv_ops,
57 [MLX5_FLOW_TYPE_VERBS] = &mlx5_flow_verbs_drv_ops,
58 [MLX5_FLOW_TYPE_MAX] = &mlx5_flow_null_drv_ops
63 MLX5_EXPANSION_ROOT_OUTER,
64 MLX5_EXPANSION_ROOT_ETH_VLAN,
65 MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN,
66 MLX5_EXPANSION_OUTER_ETH,
67 MLX5_EXPANSION_OUTER_ETH_VLAN,
68 MLX5_EXPANSION_OUTER_VLAN,
69 MLX5_EXPANSION_OUTER_IPV4,
70 MLX5_EXPANSION_OUTER_IPV4_UDP,
71 MLX5_EXPANSION_OUTER_IPV4_TCP,
72 MLX5_EXPANSION_OUTER_IPV6,
73 MLX5_EXPANSION_OUTER_IPV6_UDP,
74 MLX5_EXPANSION_OUTER_IPV6_TCP,
76 MLX5_EXPANSION_VXLAN_GPE,
80 MLX5_EXPANSION_ETH_VLAN,
83 MLX5_EXPANSION_IPV4_UDP,
84 MLX5_EXPANSION_IPV4_TCP,
86 MLX5_EXPANSION_IPV6_UDP,
87 MLX5_EXPANSION_IPV6_TCP,
90 /** Supported expansion of items. */
91 static const struct rte_flow_expand_node mlx5_support_expansion[] = {
92 [MLX5_EXPANSION_ROOT] = {
93 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
96 .type = RTE_FLOW_ITEM_TYPE_END,
98 [MLX5_EXPANSION_ROOT_OUTER] = {
99 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_ETH,
100 MLX5_EXPANSION_OUTER_IPV4,
101 MLX5_EXPANSION_OUTER_IPV6),
102 .type = RTE_FLOW_ITEM_TYPE_END,
104 [MLX5_EXPANSION_ROOT_ETH_VLAN] = {
105 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH_VLAN),
106 .type = RTE_FLOW_ITEM_TYPE_END,
108 [MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN] = {
109 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_ETH_VLAN),
110 .type = RTE_FLOW_ITEM_TYPE_END,
112 [MLX5_EXPANSION_OUTER_ETH] = {
113 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
114 MLX5_EXPANSION_OUTER_IPV6,
115 MLX5_EXPANSION_MPLS),
116 .type = RTE_FLOW_ITEM_TYPE_ETH,
119 [MLX5_EXPANSION_OUTER_ETH_VLAN] = {
120 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_VLAN),
121 .type = RTE_FLOW_ITEM_TYPE_ETH,
124 [MLX5_EXPANSION_OUTER_VLAN] = {
125 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
126 MLX5_EXPANSION_OUTER_IPV6),
127 .type = RTE_FLOW_ITEM_TYPE_VLAN,
129 [MLX5_EXPANSION_OUTER_IPV4] = {
130 .next = RTE_FLOW_EXPAND_RSS_NEXT
131 (MLX5_EXPANSION_OUTER_IPV4_UDP,
132 MLX5_EXPANSION_OUTER_IPV4_TCP,
135 MLX5_EXPANSION_IPV6),
136 .type = RTE_FLOW_ITEM_TYPE_IPV4,
137 .rss_types = ETH_RSS_IPV4 | ETH_RSS_FRAG_IPV4 |
138 ETH_RSS_NONFRAG_IPV4_OTHER,
140 [MLX5_EXPANSION_OUTER_IPV4_UDP] = {
141 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
142 MLX5_EXPANSION_VXLAN_GPE),
143 .type = RTE_FLOW_ITEM_TYPE_UDP,
144 .rss_types = ETH_RSS_NONFRAG_IPV4_UDP,
146 [MLX5_EXPANSION_OUTER_IPV4_TCP] = {
147 .type = RTE_FLOW_ITEM_TYPE_TCP,
148 .rss_types = ETH_RSS_NONFRAG_IPV4_TCP,
150 [MLX5_EXPANSION_OUTER_IPV6] = {
151 .next = RTE_FLOW_EXPAND_RSS_NEXT
152 (MLX5_EXPANSION_OUTER_IPV6_UDP,
153 MLX5_EXPANSION_OUTER_IPV6_TCP,
155 MLX5_EXPANSION_IPV6),
156 .type = RTE_FLOW_ITEM_TYPE_IPV6,
157 .rss_types = ETH_RSS_IPV6 | ETH_RSS_FRAG_IPV6 |
158 ETH_RSS_NONFRAG_IPV6_OTHER,
160 [MLX5_EXPANSION_OUTER_IPV6_UDP] = {
161 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
162 MLX5_EXPANSION_VXLAN_GPE),
163 .type = RTE_FLOW_ITEM_TYPE_UDP,
164 .rss_types = ETH_RSS_NONFRAG_IPV6_UDP,
166 [MLX5_EXPANSION_OUTER_IPV6_TCP] = {
167 .type = RTE_FLOW_ITEM_TYPE_TCP,
168 .rss_types = ETH_RSS_NONFRAG_IPV6_TCP,
170 [MLX5_EXPANSION_VXLAN] = {
171 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
173 MLX5_EXPANSION_IPV6),
174 .type = RTE_FLOW_ITEM_TYPE_VXLAN,
176 [MLX5_EXPANSION_VXLAN_GPE] = {
177 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
179 MLX5_EXPANSION_IPV6),
180 .type = RTE_FLOW_ITEM_TYPE_VXLAN_GPE,
182 [MLX5_EXPANSION_GRE] = {
183 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4),
184 .type = RTE_FLOW_ITEM_TYPE_GRE,
186 [MLX5_EXPANSION_MPLS] = {
187 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
188 MLX5_EXPANSION_IPV6),
189 .type = RTE_FLOW_ITEM_TYPE_MPLS,
191 [MLX5_EXPANSION_ETH] = {
192 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
193 MLX5_EXPANSION_IPV6),
194 .type = RTE_FLOW_ITEM_TYPE_ETH,
196 [MLX5_EXPANSION_ETH_VLAN] = {
197 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VLAN),
198 .type = RTE_FLOW_ITEM_TYPE_ETH,
200 [MLX5_EXPANSION_VLAN] = {
201 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
202 MLX5_EXPANSION_IPV6),
203 .type = RTE_FLOW_ITEM_TYPE_VLAN,
205 [MLX5_EXPANSION_IPV4] = {
206 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4_UDP,
207 MLX5_EXPANSION_IPV4_TCP),
208 .type = RTE_FLOW_ITEM_TYPE_IPV4,
209 .rss_types = ETH_RSS_IPV4 | ETH_RSS_FRAG_IPV4 |
210 ETH_RSS_NONFRAG_IPV4_OTHER,
212 [MLX5_EXPANSION_IPV4_UDP] = {
213 .type = RTE_FLOW_ITEM_TYPE_UDP,
214 .rss_types = ETH_RSS_NONFRAG_IPV4_UDP,
216 [MLX5_EXPANSION_IPV4_TCP] = {
217 .type = RTE_FLOW_ITEM_TYPE_TCP,
218 .rss_types = ETH_RSS_NONFRAG_IPV4_TCP,
220 [MLX5_EXPANSION_IPV6] = {
221 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV6_UDP,
222 MLX5_EXPANSION_IPV6_TCP),
223 .type = RTE_FLOW_ITEM_TYPE_IPV6,
224 .rss_types = ETH_RSS_IPV6 | ETH_RSS_FRAG_IPV6 |
225 ETH_RSS_NONFRAG_IPV6_OTHER,
227 [MLX5_EXPANSION_IPV6_UDP] = {
228 .type = RTE_FLOW_ITEM_TYPE_UDP,
229 .rss_types = ETH_RSS_NONFRAG_IPV6_UDP,
231 [MLX5_EXPANSION_IPV6_TCP] = {
232 .type = RTE_FLOW_ITEM_TYPE_TCP,
233 .rss_types = ETH_RSS_NONFRAG_IPV6_TCP,
237 static const struct rte_flow_ops mlx5_flow_ops = {
238 .validate = mlx5_flow_validate,
239 .create = mlx5_flow_create,
240 .destroy = mlx5_flow_destroy,
241 .flush = mlx5_flow_flush,
242 .isolate = mlx5_flow_isolate,
243 .query = mlx5_flow_query,
244 .dev_dump = mlx5_flow_dev_dump,
247 /* Convert FDIR request to Generic flow. */
249 struct rte_flow_attr attr;
250 struct rte_flow_item items[4];
251 struct rte_flow_item_eth l2;
252 struct rte_flow_item_eth l2_mask;
254 struct rte_flow_item_ipv4 ipv4;
255 struct rte_flow_item_ipv6 ipv6;
258 struct rte_flow_item_ipv4 ipv4;
259 struct rte_flow_item_ipv6 ipv6;
262 struct rte_flow_item_udp udp;
263 struct rte_flow_item_tcp tcp;
266 struct rte_flow_item_udp udp;
267 struct rte_flow_item_tcp tcp;
269 struct rte_flow_action actions[2];
270 struct rte_flow_action_queue queue;
273 /* Map of Verbs to Flow priority with 8 Verbs priorities. */
274 static const uint32_t priority_map_3[][MLX5_PRIORITY_MAP_MAX] = {
275 { 0, 1, 2 }, { 2, 3, 4 }, { 5, 6, 7 },
278 /* Map of Verbs to Flow priority with 16 Verbs priorities. */
279 static const uint32_t priority_map_5[][MLX5_PRIORITY_MAP_MAX] = {
280 { 0, 1, 2 }, { 3, 4, 5 }, { 6, 7, 8 },
281 { 9, 10, 11 }, { 12, 13, 14 },
284 /* Tunnel information. */
285 struct mlx5_flow_tunnel_info {
286 uint64_t tunnel; /**< Tunnel bit (see MLX5_FLOW_*). */
287 uint32_t ptype; /**< Tunnel Ptype (see RTE_PTYPE_*). */
290 static struct mlx5_flow_tunnel_info tunnels_info[] = {
292 .tunnel = MLX5_FLOW_LAYER_VXLAN,
293 .ptype = RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L4_UDP,
296 .tunnel = MLX5_FLOW_LAYER_GENEVE,
297 .ptype = RTE_PTYPE_TUNNEL_GENEVE | RTE_PTYPE_L4_UDP,
300 .tunnel = MLX5_FLOW_LAYER_VXLAN_GPE,
301 .ptype = RTE_PTYPE_TUNNEL_VXLAN_GPE | RTE_PTYPE_L4_UDP,
304 .tunnel = MLX5_FLOW_LAYER_GRE,
305 .ptype = RTE_PTYPE_TUNNEL_GRE,
308 .tunnel = MLX5_FLOW_LAYER_MPLS | MLX5_FLOW_LAYER_OUTER_L4_UDP,
309 .ptype = RTE_PTYPE_TUNNEL_MPLS_IN_UDP | RTE_PTYPE_L4_UDP,
312 .tunnel = MLX5_FLOW_LAYER_MPLS,
313 .ptype = RTE_PTYPE_TUNNEL_MPLS_IN_GRE,
316 .tunnel = MLX5_FLOW_LAYER_NVGRE,
317 .ptype = RTE_PTYPE_TUNNEL_NVGRE,
320 .tunnel = MLX5_FLOW_LAYER_IPIP,
321 .ptype = RTE_PTYPE_TUNNEL_IP,
324 .tunnel = MLX5_FLOW_LAYER_IPV6_ENCAP,
325 .ptype = RTE_PTYPE_TUNNEL_IP,
328 .tunnel = MLX5_FLOW_LAYER_GTP,
329 .ptype = RTE_PTYPE_TUNNEL_GTPU,
334 * Translate tag ID to register.
337 * Pointer to the Ethernet device structure.
339 * The feature that request the register.
341 * The request register ID.
343 * Error description in case of any.
346 * The request register on success, a negative errno
347 * value otherwise and rte_errno is set.
350 mlx5_flow_get_reg_id(struct rte_eth_dev *dev,
351 enum mlx5_feature_name feature,
353 struct rte_flow_error *error)
355 struct mlx5_priv *priv = dev->data->dev_private;
356 struct mlx5_dev_config *config = &priv->config;
357 enum modify_reg start_reg;
358 bool skip_mtr_reg = false;
361 case MLX5_HAIRPIN_RX:
363 case MLX5_HAIRPIN_TX:
365 case MLX5_METADATA_RX:
366 switch (config->dv_xmeta_en) {
367 case MLX5_XMETA_MODE_LEGACY:
369 case MLX5_XMETA_MODE_META16:
371 case MLX5_XMETA_MODE_META32:
375 case MLX5_METADATA_TX:
377 case MLX5_METADATA_FDB:
378 switch (config->dv_xmeta_en) {
379 case MLX5_XMETA_MODE_LEGACY:
381 case MLX5_XMETA_MODE_META16:
383 case MLX5_XMETA_MODE_META32:
388 switch (config->dv_xmeta_en) {
389 case MLX5_XMETA_MODE_LEGACY:
391 case MLX5_XMETA_MODE_META16:
393 case MLX5_XMETA_MODE_META32:
399 * If meter color and flow match share one register, flow match
400 * should use the meter color register for match.
402 if (priv->mtr_reg_share)
403 return priv->mtr_color_reg;
405 return priv->mtr_color_reg != REG_C_2 ? REG_C_2 :
408 MLX5_ASSERT(priv->mtr_color_reg != REG_NONE);
409 return priv->mtr_color_reg;
412 * Metadata COPY_MARK register using is in meter suffix sub
413 * flow while with meter. It's safe to share the same register.
415 return priv->mtr_color_reg != REG_C_2 ? REG_C_2 : REG_C_3;
418 * If meter is enable, it will engage the register for color
419 * match and flow match. If meter color match is not using the
420 * REG_C_2, need to skip the REG_C_x be used by meter color
422 * If meter is disable, free to use all available registers.
424 start_reg = priv->mtr_color_reg != REG_C_2 ? REG_C_2 :
425 (priv->mtr_reg_share ? REG_C_3 : REG_C_4);
426 skip_mtr_reg = !!(priv->mtr_en && start_reg == REG_C_2);
427 if (id > (REG_C_7 - start_reg))
428 return rte_flow_error_set(error, EINVAL,
429 RTE_FLOW_ERROR_TYPE_ITEM,
430 NULL, "invalid tag id");
431 if (config->flow_mreg_c[id + start_reg - REG_C_0] == REG_NONE)
432 return rte_flow_error_set(error, ENOTSUP,
433 RTE_FLOW_ERROR_TYPE_ITEM,
434 NULL, "unsupported tag id");
436 * This case means meter is using the REG_C_x great than 2.
437 * Take care not to conflict with meter color REG_C_x.
438 * If the available index REG_C_y >= REG_C_x, skip the
441 if (skip_mtr_reg && config->flow_mreg_c
442 [id + start_reg - REG_C_0] >= priv->mtr_color_reg) {
443 if (config->flow_mreg_c
444 [id + 1 + start_reg - REG_C_0] != REG_NONE)
445 return config->flow_mreg_c
446 [id + 1 + start_reg - REG_C_0];
447 return rte_flow_error_set(error, ENOTSUP,
448 RTE_FLOW_ERROR_TYPE_ITEM,
449 NULL, "unsupported tag id");
451 return config->flow_mreg_c[id + start_reg - REG_C_0];
454 return rte_flow_error_set(error, EINVAL,
455 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
456 NULL, "invalid feature name");
460 * Check extensive flow metadata register support.
463 * Pointer to rte_eth_dev structure.
466 * True if device supports extensive flow metadata register, otherwise false.
469 mlx5_flow_ext_mreg_supported(struct rte_eth_dev *dev)
471 struct mlx5_priv *priv = dev->data->dev_private;
472 struct mlx5_dev_config *config = &priv->config;
475 * Having available reg_c can be regarded inclusively as supporting
476 * extensive flow metadata register, which could mean,
477 * - metadata register copy action by modify header.
478 * - 16 modify header actions is supported.
479 * - reg_c's are preserved across different domain (FDB and NIC) on
480 * packet loopback by flow lookup miss.
482 return config->flow_mreg_c[2] != REG_NONE;
486 * Discover the maximum number of priority available.
489 * Pointer to the Ethernet device structure.
492 * number of supported flow priority on success, a negative errno
493 * value otherwise and rte_errno is set.
496 mlx5_flow_discover_priorities(struct rte_eth_dev *dev)
498 struct mlx5_priv *priv = dev->data->dev_private;
500 struct ibv_flow_attr attr;
501 struct ibv_flow_spec_eth eth;
502 struct ibv_flow_spec_action_drop drop;
506 .port = (uint8_t)priv->ibv_port,
509 .type = IBV_FLOW_SPEC_ETH,
510 .size = sizeof(struct ibv_flow_spec_eth),
513 .size = sizeof(struct ibv_flow_spec_action_drop),
514 .type = IBV_FLOW_SPEC_ACTION_DROP,
517 struct ibv_flow *flow;
518 struct mlx5_hrxq *drop = mlx5_hrxq_drop_new(dev);
519 uint16_t vprio[] = { 8, 16 };
527 for (i = 0; i != RTE_DIM(vprio); i++) {
528 flow_attr.attr.priority = vprio[i] - 1;
529 flow = mlx5_glue->create_flow(drop->qp, &flow_attr.attr);
532 claim_zero(mlx5_glue->destroy_flow(flow));
535 mlx5_hrxq_drop_release(dev);
538 priority = RTE_DIM(priority_map_3);
541 priority = RTE_DIM(priority_map_5);
546 "port %u verbs maximum priority: %d expected 8/16",
547 dev->data->port_id, priority);
550 DRV_LOG(INFO, "port %u flow maximum priority: %d",
551 dev->data->port_id, priority);
556 * Adjust flow priority based on the highest layer and the request priority.
559 * Pointer to the Ethernet device structure.
560 * @param[in] priority
561 * The rule base priority.
562 * @param[in] subpriority
563 * The priority based on the items.
568 uint32_t mlx5_flow_adjust_priority(struct rte_eth_dev *dev, int32_t priority,
569 uint32_t subpriority)
572 struct mlx5_priv *priv = dev->data->dev_private;
574 switch (priv->config.flow_prio) {
575 case RTE_DIM(priority_map_3):
576 res = priority_map_3[priority][subpriority];
578 case RTE_DIM(priority_map_5):
579 res = priority_map_5[priority][subpriority];
586 * Verify the @p item specifications (spec, last, mask) are compatible with the
590 * Item specification.
592 * @p item->mask or flow default bit-masks.
593 * @param[in] nic_mask
594 * Bit-masks covering supported fields by the NIC to compare with user mask.
596 * Bit-masks size in bytes.
598 * Pointer to error structure.
601 * 0 on success, a negative errno value otherwise and rte_errno is set.
604 mlx5_flow_item_acceptable(const struct rte_flow_item *item,
606 const uint8_t *nic_mask,
608 struct rte_flow_error *error)
612 MLX5_ASSERT(nic_mask);
613 for (i = 0; i < size; ++i)
614 if ((nic_mask[i] | mask[i]) != nic_mask[i])
615 return rte_flow_error_set(error, ENOTSUP,
616 RTE_FLOW_ERROR_TYPE_ITEM,
618 "mask enables non supported"
620 if (!item->spec && (item->mask || item->last))
621 return rte_flow_error_set(error, EINVAL,
622 RTE_FLOW_ERROR_TYPE_ITEM, item,
623 "mask/last without a spec is not"
625 if (item->spec && item->last) {
631 for (i = 0; i < size; ++i) {
632 spec[i] = ((const uint8_t *)item->spec)[i] & mask[i];
633 last[i] = ((const uint8_t *)item->last)[i] & mask[i];
635 ret = memcmp(spec, last, size);
637 return rte_flow_error_set(error, EINVAL,
638 RTE_FLOW_ERROR_TYPE_ITEM,
640 "range is not valid");
646 * Adjust the hash fields according to the @p flow information.
648 * @param[in] dev_flow.
649 * Pointer to the mlx5_flow.
651 * 1 when the hash field is for a tunnel item.
652 * @param[in] layer_types
654 * @param[in] hash_fields
658 * The hash fields that should be used.
661 mlx5_flow_hashfields_adjust(struct mlx5_flow_rss_desc *rss_desc,
662 int tunnel __rte_unused, uint64_t layer_types,
663 uint64_t hash_fields)
665 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
666 int rss_request_inner = rss_desc->level >= 2;
668 /* Check RSS hash level for tunnel. */
669 if (tunnel && rss_request_inner)
670 hash_fields |= IBV_RX_HASH_INNER;
671 else if (tunnel || rss_request_inner)
674 /* Check if requested layer matches RSS hash fields. */
675 if (!(rss_desc->types & layer_types))
681 * Lookup and set the ptype in the data Rx part. A single Ptype can be used,
682 * if several tunnel rules are used on this queue, the tunnel ptype will be
686 * Rx queue to update.
689 flow_rxq_tunnel_ptype_update(struct mlx5_rxq_ctrl *rxq_ctrl)
692 uint32_t tunnel_ptype = 0;
694 /* Look up for the ptype to use. */
695 for (i = 0; i != MLX5_FLOW_TUNNEL; ++i) {
696 if (!rxq_ctrl->flow_tunnels_n[i])
699 tunnel_ptype = tunnels_info[i].ptype;
705 rxq_ctrl->rxq.tunnel = tunnel_ptype;
709 * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) according to the devive
713 * Pointer to the Ethernet device structure.
714 * @param[in] dev_handle
715 * Pointer to device flow handle structure.
718 flow_drv_rxq_flags_set(struct rte_eth_dev *dev,
719 struct mlx5_flow_handle *dev_handle)
721 struct mlx5_priv *priv = dev->data->dev_private;
722 const int mark = dev_handle->mark;
723 const int tunnel = !!(dev_handle->layers & MLX5_FLOW_LAYER_TUNNEL);
724 struct mlx5_hrxq *hrxq;
727 if (dev_handle->fate_action != MLX5_FLOW_FATE_QUEUE)
729 hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
730 dev_handle->rix_hrxq);
733 for (i = 0; i != hrxq->ind_table->queues_n; ++i) {
734 int idx = hrxq->ind_table->queues[i];
735 struct mlx5_rxq_ctrl *rxq_ctrl =
736 container_of((*priv->rxqs)[idx],
737 struct mlx5_rxq_ctrl, rxq);
740 * To support metadata register copy on Tx loopback,
741 * this must be always enabled (metadata may arive
742 * from other port - not from local flows only.
744 if (priv->config.dv_flow_en &&
745 priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
746 mlx5_flow_ext_mreg_supported(dev)) {
747 rxq_ctrl->rxq.mark = 1;
748 rxq_ctrl->flow_mark_n = 1;
750 rxq_ctrl->rxq.mark = 1;
751 rxq_ctrl->flow_mark_n++;
756 /* Increase the counter matching the flow. */
757 for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
758 if ((tunnels_info[j].tunnel &
759 dev_handle->layers) ==
760 tunnels_info[j].tunnel) {
761 rxq_ctrl->flow_tunnels_n[j]++;
765 flow_rxq_tunnel_ptype_update(rxq_ctrl);
771 * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) for a flow
774 * Pointer to the Ethernet device structure.
776 * Pointer to flow structure.
779 flow_rxq_flags_set(struct rte_eth_dev *dev, struct rte_flow *flow)
781 struct mlx5_priv *priv = dev->data->dev_private;
783 struct mlx5_flow_handle *dev_handle;
785 SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
786 handle_idx, dev_handle, next)
787 flow_drv_rxq_flags_set(dev, dev_handle);
791 * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
792 * device flow if no other flow uses it with the same kind of request.
795 * Pointer to Ethernet device.
796 * @param[in] dev_handle
797 * Pointer to the device flow handle structure.
800 flow_drv_rxq_flags_trim(struct rte_eth_dev *dev,
801 struct mlx5_flow_handle *dev_handle)
803 struct mlx5_priv *priv = dev->data->dev_private;
804 const int mark = dev_handle->mark;
805 const int tunnel = !!(dev_handle->layers & MLX5_FLOW_LAYER_TUNNEL);
806 struct mlx5_hrxq *hrxq;
809 if (dev_handle->fate_action != MLX5_FLOW_FATE_QUEUE)
811 hrxq = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_HRXQ],
812 dev_handle->rix_hrxq);
815 MLX5_ASSERT(dev->data->dev_started);
816 for (i = 0; i != hrxq->ind_table->queues_n; ++i) {
817 int idx = hrxq->ind_table->queues[i];
818 struct mlx5_rxq_ctrl *rxq_ctrl =
819 container_of((*priv->rxqs)[idx],
820 struct mlx5_rxq_ctrl, rxq);
822 if (priv->config.dv_flow_en &&
823 priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
824 mlx5_flow_ext_mreg_supported(dev)) {
825 rxq_ctrl->rxq.mark = 1;
826 rxq_ctrl->flow_mark_n = 1;
828 rxq_ctrl->flow_mark_n--;
829 rxq_ctrl->rxq.mark = !!rxq_ctrl->flow_mark_n;
834 /* Decrease the counter matching the flow. */
835 for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
836 if ((tunnels_info[j].tunnel &
837 dev_handle->layers) ==
838 tunnels_info[j].tunnel) {
839 rxq_ctrl->flow_tunnels_n[j]--;
843 flow_rxq_tunnel_ptype_update(rxq_ctrl);
849 * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
850 * @p flow if no other flow uses it with the same kind of request.
853 * Pointer to Ethernet device.
855 * Pointer to the flow.
858 flow_rxq_flags_trim(struct rte_eth_dev *dev, struct rte_flow *flow)
860 struct mlx5_priv *priv = dev->data->dev_private;
862 struct mlx5_flow_handle *dev_handle;
864 SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
865 handle_idx, dev_handle, next)
866 flow_drv_rxq_flags_trim(dev, dev_handle);
870 * Clear the Mark/Flag and Tunnel ptype information in all Rx queues.
873 * Pointer to Ethernet device.
876 flow_rxq_flags_clear(struct rte_eth_dev *dev)
878 struct mlx5_priv *priv = dev->data->dev_private;
881 for (i = 0; i != priv->rxqs_n; ++i) {
882 struct mlx5_rxq_ctrl *rxq_ctrl;
885 if (!(*priv->rxqs)[i])
887 rxq_ctrl = container_of((*priv->rxqs)[i],
888 struct mlx5_rxq_ctrl, rxq);
889 rxq_ctrl->flow_mark_n = 0;
890 rxq_ctrl->rxq.mark = 0;
891 for (j = 0; j != MLX5_FLOW_TUNNEL; ++j)
892 rxq_ctrl->flow_tunnels_n[j] = 0;
893 rxq_ctrl->rxq.tunnel = 0;
898 * return a pointer to the desired action in the list of actions.
901 * The list of actions to search the action in.
903 * The action to find.
906 * Pointer to the action in the list, if found. NULL otherwise.
908 const struct rte_flow_action *
909 mlx5_flow_find_action(const struct rte_flow_action *actions,
910 enum rte_flow_action_type action)
914 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++)
915 if (actions->type == action)
921 * Validate the flag action.
923 * @param[in] action_flags
924 * Bit-fields that holds the actions detected until now.
926 * Attributes of flow that includes this action.
928 * Pointer to error structure.
931 * 0 on success, a negative errno value otherwise and rte_errno is set.
934 mlx5_flow_validate_action_flag(uint64_t action_flags,
935 const struct rte_flow_attr *attr,
936 struct rte_flow_error *error)
938 if (action_flags & MLX5_FLOW_ACTION_MARK)
939 return rte_flow_error_set(error, EINVAL,
940 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
941 "can't mark and flag in same flow");
942 if (action_flags & MLX5_FLOW_ACTION_FLAG)
943 return rte_flow_error_set(error, EINVAL,
944 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
946 " actions in same flow");
948 return rte_flow_error_set(error, ENOTSUP,
949 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
950 "flag action not supported for "
956 * Validate the mark action.
959 * Pointer to the queue action.
960 * @param[in] action_flags
961 * Bit-fields that holds the actions detected until now.
963 * Attributes of flow that includes this action.
965 * Pointer to error structure.
968 * 0 on success, a negative errno value otherwise and rte_errno is set.
971 mlx5_flow_validate_action_mark(const struct rte_flow_action *action,
972 uint64_t action_flags,
973 const struct rte_flow_attr *attr,
974 struct rte_flow_error *error)
976 const struct rte_flow_action_mark *mark = action->conf;
979 return rte_flow_error_set(error, EINVAL,
980 RTE_FLOW_ERROR_TYPE_ACTION,
982 "configuration cannot be null");
983 if (mark->id >= MLX5_FLOW_MARK_MAX)
984 return rte_flow_error_set(error, EINVAL,
985 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
987 "mark id must in 0 <= id < "
988 RTE_STR(MLX5_FLOW_MARK_MAX));
989 if (action_flags & MLX5_FLOW_ACTION_FLAG)
990 return rte_flow_error_set(error, EINVAL,
991 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
992 "can't flag and mark in same flow");
993 if (action_flags & MLX5_FLOW_ACTION_MARK)
994 return rte_flow_error_set(error, EINVAL,
995 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
996 "can't have 2 mark actions in same"
999 return rte_flow_error_set(error, ENOTSUP,
1000 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1001 "mark action not supported for "
1007 * Validate the drop action.
1009 * @param[in] action_flags
1010 * Bit-fields that holds the actions detected until now.
1012 * Attributes of flow that includes this action.
1014 * Pointer to error structure.
1017 * 0 on success, a negative errno value otherwise and rte_errno is set.
1020 mlx5_flow_validate_action_drop(uint64_t action_flags __rte_unused,
1021 const struct rte_flow_attr *attr,
1022 struct rte_flow_error *error)
1025 return rte_flow_error_set(error, ENOTSUP,
1026 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1027 "drop action not supported for "
1033 * Validate the queue action.
1036 * Pointer to the queue action.
1037 * @param[in] action_flags
1038 * Bit-fields that holds the actions detected until now.
1040 * Pointer to the Ethernet device structure.
1042 * Attributes of flow that includes this action.
1044 * Pointer to error structure.
1047 * 0 on success, a negative errno value otherwise and rte_errno is set.
1050 mlx5_flow_validate_action_queue(const struct rte_flow_action *action,
1051 uint64_t action_flags,
1052 struct rte_eth_dev *dev,
1053 const struct rte_flow_attr *attr,
1054 struct rte_flow_error *error)
1056 struct mlx5_priv *priv = dev->data->dev_private;
1057 const struct rte_flow_action_queue *queue = action->conf;
1059 if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1060 return rte_flow_error_set(error, EINVAL,
1061 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1062 "can't have 2 fate actions in"
1065 return rte_flow_error_set(error, EINVAL,
1066 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1067 NULL, "No Rx queues configured");
1068 if (queue->index >= priv->rxqs_n)
1069 return rte_flow_error_set(error, EINVAL,
1070 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1072 "queue index out of range");
1073 if (!(*priv->rxqs)[queue->index])
1074 return rte_flow_error_set(error, EINVAL,
1075 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1077 "queue is not configured");
1079 return rte_flow_error_set(error, ENOTSUP,
1080 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1081 "queue action not supported for "
1087 * Validate the rss action.
1090 * Pointer to the queue action.
1091 * @param[in] action_flags
1092 * Bit-fields that holds the actions detected until now.
1094 * Pointer to the Ethernet device structure.
1096 * Attributes of flow that includes this action.
1097 * @param[in] item_flags
1098 * Items that were detected.
1100 * Pointer to error structure.
1103 * 0 on success, a negative errno value otherwise and rte_errno is set.
1106 mlx5_flow_validate_action_rss(const struct rte_flow_action *action,
1107 uint64_t action_flags,
1108 struct rte_eth_dev *dev,
1109 const struct rte_flow_attr *attr,
1110 uint64_t item_flags,
1111 struct rte_flow_error *error)
1113 struct mlx5_priv *priv = dev->data->dev_private;
1114 const struct rte_flow_action_rss *rss = action->conf;
1115 int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1118 if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1119 return rte_flow_error_set(error, EINVAL,
1120 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1121 "can't have 2 fate actions"
1123 if (rss->func != RTE_ETH_HASH_FUNCTION_DEFAULT &&
1124 rss->func != RTE_ETH_HASH_FUNCTION_TOEPLITZ)
1125 return rte_flow_error_set(error, ENOTSUP,
1126 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1128 "RSS hash function not supported");
1129 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
1134 return rte_flow_error_set(error, ENOTSUP,
1135 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1137 "tunnel RSS is not supported");
1138 /* allow RSS key_len 0 in case of NULL (default) RSS key. */
1139 if (rss->key_len == 0 && rss->key != NULL)
1140 return rte_flow_error_set(error, ENOTSUP,
1141 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1143 "RSS hash key length 0");
1144 if (rss->key_len > 0 && rss->key_len < MLX5_RSS_HASH_KEY_LEN)
1145 return rte_flow_error_set(error, ENOTSUP,
1146 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1148 "RSS hash key too small");
1149 if (rss->key_len > MLX5_RSS_HASH_KEY_LEN)
1150 return rte_flow_error_set(error, ENOTSUP,
1151 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1153 "RSS hash key too large");
1154 if (rss->queue_num > priv->config.ind_table_max_size)
1155 return rte_flow_error_set(error, ENOTSUP,
1156 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1158 "number of queues too large");
1159 if (rss->types & MLX5_RSS_HF_MASK)
1160 return rte_flow_error_set(error, ENOTSUP,
1161 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1163 "some RSS protocols are not"
1165 if ((rss->types & (ETH_RSS_L3_SRC_ONLY | ETH_RSS_L3_DST_ONLY)) &&
1166 !(rss->types & ETH_RSS_IP))
1167 return rte_flow_error_set(error, EINVAL,
1168 RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1169 "L3 partial RSS requested but L3 RSS"
1170 " type not specified");
1171 if ((rss->types & (ETH_RSS_L4_SRC_ONLY | ETH_RSS_L4_DST_ONLY)) &&
1172 !(rss->types & (ETH_RSS_UDP | ETH_RSS_TCP)))
1173 return rte_flow_error_set(error, EINVAL,
1174 RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1175 "L4 partial RSS requested but L4 RSS"
1176 " type not specified");
1178 return rte_flow_error_set(error, EINVAL,
1179 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1180 NULL, "No Rx queues configured");
1181 if (!rss->queue_num)
1182 return rte_flow_error_set(error, EINVAL,
1183 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1184 NULL, "No queues configured");
1185 for (i = 0; i != rss->queue_num; ++i) {
1186 if (rss->queue[i] >= priv->rxqs_n)
1187 return rte_flow_error_set
1189 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1190 &rss->queue[i], "queue index out of range");
1191 if (!(*priv->rxqs)[rss->queue[i]])
1192 return rte_flow_error_set
1193 (error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1194 &rss->queue[i], "queue is not configured");
1197 return rte_flow_error_set(error, ENOTSUP,
1198 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1199 "rss action not supported for "
1201 if (rss->level > 1 && !tunnel)
1202 return rte_flow_error_set(error, EINVAL,
1203 RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1204 "inner RSS is not supported for "
1205 "non-tunnel flows");
1210 * Validate the count action.
1213 * Pointer to the Ethernet device structure.
1215 * Attributes of flow that includes this action.
1217 * Pointer to error structure.
1220 * 0 on success, a negative errno value otherwise and rte_errno is set.
1223 mlx5_flow_validate_action_count(struct rte_eth_dev *dev __rte_unused,
1224 const struct rte_flow_attr *attr,
1225 struct rte_flow_error *error)
1228 return rte_flow_error_set(error, ENOTSUP,
1229 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1230 "count action not supported for "
1236 * Verify the @p attributes will be correctly understood by the NIC and store
1237 * them in the @p flow if everything is correct.
1240 * Pointer to the Ethernet device structure.
1241 * @param[in] attributes
1242 * Pointer to flow attributes
1244 * Pointer to error structure.
1247 * 0 on success, a negative errno value otherwise and rte_errno is set.
1250 mlx5_flow_validate_attributes(struct rte_eth_dev *dev,
1251 const struct rte_flow_attr *attributes,
1252 struct rte_flow_error *error)
1254 struct mlx5_priv *priv = dev->data->dev_private;
1255 uint32_t priority_max = priv->config.flow_prio - 1;
1257 if (attributes->group)
1258 return rte_flow_error_set(error, ENOTSUP,
1259 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
1260 NULL, "groups is not supported");
1261 if (attributes->priority != MLX5_FLOW_PRIO_RSVD &&
1262 attributes->priority >= priority_max)
1263 return rte_flow_error_set(error, ENOTSUP,
1264 RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
1265 NULL, "priority out of range");
1266 if (attributes->egress)
1267 return rte_flow_error_set(error, ENOTSUP,
1268 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1269 "egress is not supported");
1270 if (attributes->transfer && !priv->config.dv_esw_en)
1271 return rte_flow_error_set(error, ENOTSUP,
1272 RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1273 NULL, "transfer is not supported");
1274 if (!attributes->ingress)
1275 return rte_flow_error_set(error, EINVAL,
1276 RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
1278 "ingress attribute is mandatory");
1283 * Validate ICMP6 item.
1286 * Item specification.
1287 * @param[in] item_flags
1288 * Bit-fields that holds the items detected until now.
1290 * Pointer to error structure.
1293 * 0 on success, a negative errno value otherwise and rte_errno is set.
1296 mlx5_flow_validate_item_icmp6(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_icmp6 *mask = item->mask;
1302 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1303 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
1304 MLX5_FLOW_LAYER_OUTER_L3_IPV6;
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_ICMPV6)
1310 return rte_flow_error_set(error, EINVAL,
1311 RTE_FLOW_ERROR_TYPE_ITEM, item,
1312 "protocol filtering not compatible"
1313 " with ICMP6 layer");
1314 if (!(item_flags & l3m))
1315 return rte_flow_error_set(error, EINVAL,
1316 RTE_FLOW_ERROR_TYPE_ITEM, item,
1317 "IPv6 is mandatory to filter on"
1319 if (item_flags & l4m)
1320 return rte_flow_error_set(error, EINVAL,
1321 RTE_FLOW_ERROR_TYPE_ITEM, item,
1322 "multiple L4 layers not supported");
1324 mask = &rte_flow_item_icmp6_mask;
1325 ret = mlx5_flow_item_acceptable
1326 (item, (const uint8_t *)mask,
1327 (const uint8_t *)&rte_flow_item_icmp6_mask,
1328 sizeof(struct rte_flow_item_icmp6), error);
1335 * Validate ICMP item.
1338 * Item specification.
1339 * @param[in] item_flags
1340 * Bit-fields that holds the items detected until now.
1342 * Pointer to error structure.
1345 * 0 on success, a negative errno value otherwise and rte_errno is set.
1348 mlx5_flow_validate_item_icmp(const struct rte_flow_item *item,
1349 uint64_t item_flags,
1350 uint8_t target_protocol,
1351 struct rte_flow_error *error)
1353 const struct rte_flow_item_icmp *mask = item->mask;
1354 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1355 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
1356 MLX5_FLOW_LAYER_OUTER_L3_IPV4;
1357 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1358 MLX5_FLOW_LAYER_OUTER_L4;
1361 if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMP)
1362 return rte_flow_error_set(error, EINVAL,
1363 RTE_FLOW_ERROR_TYPE_ITEM, item,
1364 "protocol filtering not compatible"
1365 " with ICMP layer");
1366 if (!(item_flags & l3m))
1367 return rte_flow_error_set(error, EINVAL,
1368 RTE_FLOW_ERROR_TYPE_ITEM, item,
1369 "IPv4 is mandatory to filter"
1371 if (item_flags & l4m)
1372 return rte_flow_error_set(error, EINVAL,
1373 RTE_FLOW_ERROR_TYPE_ITEM, item,
1374 "multiple L4 layers not supported");
1376 mask = &rte_flow_item_icmp_mask;
1377 ret = mlx5_flow_item_acceptable
1378 (item, (const uint8_t *)mask,
1379 (const uint8_t *)&rte_flow_item_icmp_mask,
1380 sizeof(struct rte_flow_item_icmp), error);
1387 * Validate Ethernet item.
1390 * Item specification.
1391 * @param[in] item_flags
1392 * Bit-fields that holds the items detected until now.
1394 * Pointer to error structure.
1397 * 0 on success, a negative errno value otherwise and rte_errno is set.
1400 mlx5_flow_validate_item_eth(const struct rte_flow_item *item,
1401 uint64_t item_flags,
1402 struct rte_flow_error *error)
1404 const struct rte_flow_item_eth *mask = item->mask;
1405 const struct rte_flow_item_eth nic_mask = {
1406 .dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1407 .src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1408 .type = RTE_BE16(0xffff),
1411 int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1412 const uint64_t ethm = tunnel ? MLX5_FLOW_LAYER_INNER_L2 :
1413 MLX5_FLOW_LAYER_OUTER_L2;
1415 if (item_flags & ethm)
1416 return rte_flow_error_set(error, ENOTSUP,
1417 RTE_FLOW_ERROR_TYPE_ITEM, item,
1418 "multiple L2 layers not supported");
1419 if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_L3)) ||
1420 (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_L3)))
1421 return rte_flow_error_set(error, EINVAL,
1422 RTE_FLOW_ERROR_TYPE_ITEM, item,
1423 "L2 layer should not follow "
1425 if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_VLAN)) ||
1426 (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_VLAN)))
1427 return rte_flow_error_set(error, EINVAL,
1428 RTE_FLOW_ERROR_TYPE_ITEM, item,
1429 "L2 layer should not follow VLAN");
1431 mask = &rte_flow_item_eth_mask;
1432 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1433 (const uint8_t *)&nic_mask,
1434 sizeof(struct rte_flow_item_eth),
1440 * Validate VLAN item.
1443 * Item specification.
1444 * @param[in] item_flags
1445 * Bit-fields that holds the items detected until now.
1447 * Ethernet device flow is being created on.
1449 * Pointer to error structure.
1452 * 0 on success, a negative errno value otherwise and rte_errno is set.
1455 mlx5_flow_validate_item_vlan(const struct rte_flow_item *item,
1456 uint64_t item_flags,
1457 struct rte_eth_dev *dev,
1458 struct rte_flow_error *error)
1460 const struct rte_flow_item_vlan *spec = item->spec;
1461 const struct rte_flow_item_vlan *mask = item->mask;
1462 const struct rte_flow_item_vlan nic_mask = {
1463 .tci = RTE_BE16(UINT16_MAX),
1464 .inner_type = RTE_BE16(UINT16_MAX),
1466 uint16_t vlan_tag = 0;
1467 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1469 const uint64_t l34m = tunnel ? (MLX5_FLOW_LAYER_INNER_L3 |
1470 MLX5_FLOW_LAYER_INNER_L4) :
1471 (MLX5_FLOW_LAYER_OUTER_L3 |
1472 MLX5_FLOW_LAYER_OUTER_L4);
1473 const uint64_t vlanm = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
1474 MLX5_FLOW_LAYER_OUTER_VLAN;
1476 if (item_flags & vlanm)
1477 return rte_flow_error_set(error, EINVAL,
1478 RTE_FLOW_ERROR_TYPE_ITEM, item,
1479 "multiple VLAN layers not supported");
1480 else if ((item_flags & l34m) != 0)
1481 return rte_flow_error_set(error, EINVAL,
1482 RTE_FLOW_ERROR_TYPE_ITEM, item,
1483 "VLAN cannot follow L3/L4 layer");
1485 mask = &rte_flow_item_vlan_mask;
1486 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1487 (const uint8_t *)&nic_mask,
1488 sizeof(struct rte_flow_item_vlan),
1492 if (!tunnel && mask->tci != RTE_BE16(0x0fff)) {
1493 struct mlx5_priv *priv = dev->data->dev_private;
1495 if (priv->vmwa_context) {
1497 * Non-NULL context means we have a virtual machine
1498 * and SR-IOV enabled, we have to create VLAN interface
1499 * to make hypervisor to setup E-Switch vport
1500 * context correctly. We avoid creating the multiple
1501 * VLAN interfaces, so we cannot support VLAN tag mask.
1503 return rte_flow_error_set(error, EINVAL,
1504 RTE_FLOW_ERROR_TYPE_ITEM,
1506 "VLAN tag mask is not"
1507 " supported in virtual"
1512 vlan_tag = spec->tci;
1513 vlan_tag &= mask->tci;
1516 * From verbs perspective an empty VLAN is equivalent
1517 * to a packet without VLAN layer.
1520 return rte_flow_error_set(error, EINVAL,
1521 RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
1523 "VLAN cannot be empty");
1528 * Validate IPV4 item.
1531 * Item specification.
1532 * @param[in] item_flags
1533 * Bit-fields that holds the items detected until now.
1534 * @param[in] acc_mask
1535 * Acceptable mask, if NULL default internal default mask
1536 * will be used to check whether item fields are supported.
1538 * Pointer to error structure.
1541 * 0 on success, a negative errno value otherwise and rte_errno is set.
1544 mlx5_flow_validate_item_ipv4(const struct rte_flow_item *item,
1545 uint64_t item_flags,
1547 uint16_t ether_type,
1548 const struct rte_flow_item_ipv4 *acc_mask,
1549 struct rte_flow_error *error)
1551 const struct rte_flow_item_ipv4 *mask = item->mask;
1552 const struct rte_flow_item_ipv4 *spec = item->spec;
1553 const struct rte_flow_item_ipv4 nic_mask = {
1555 .src_addr = RTE_BE32(0xffffffff),
1556 .dst_addr = RTE_BE32(0xffffffff),
1557 .type_of_service = 0xff,
1558 .next_proto_id = 0xff,
1561 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1562 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1563 MLX5_FLOW_LAYER_OUTER_L3;
1564 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1565 MLX5_FLOW_LAYER_OUTER_L4;
1567 uint8_t next_proto = 0xFF;
1568 const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
1569 MLX5_FLOW_LAYER_OUTER_VLAN |
1570 MLX5_FLOW_LAYER_INNER_VLAN);
1572 if ((last_item & l2_vlan) && ether_type &&
1573 ether_type != RTE_ETHER_TYPE_IPV4)
1574 return rte_flow_error_set(error, EINVAL,
1575 RTE_FLOW_ERROR_TYPE_ITEM, item,
1576 "IPv4 cannot follow L2/VLAN layer "
1577 "which ether type is not IPv4");
1578 if (item_flags & MLX5_FLOW_LAYER_IPIP) {
1580 next_proto = mask->hdr.next_proto_id &
1581 spec->hdr.next_proto_id;
1582 if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
1583 return rte_flow_error_set(error, EINVAL,
1584 RTE_FLOW_ERROR_TYPE_ITEM,
1589 if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP)
1590 return rte_flow_error_set(error, EINVAL,
1591 RTE_FLOW_ERROR_TYPE_ITEM, item,
1592 "wrong tunnel type - IPv6 specified "
1593 "but IPv4 item provided");
1594 if (item_flags & l3m)
1595 return rte_flow_error_set(error, ENOTSUP,
1596 RTE_FLOW_ERROR_TYPE_ITEM, item,
1597 "multiple L3 layers not supported");
1598 else if (item_flags & l4m)
1599 return rte_flow_error_set(error, EINVAL,
1600 RTE_FLOW_ERROR_TYPE_ITEM, item,
1601 "L3 cannot follow an L4 layer.");
1602 else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
1603 !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
1604 return rte_flow_error_set(error, EINVAL,
1605 RTE_FLOW_ERROR_TYPE_ITEM, item,
1606 "L3 cannot follow an NVGRE layer.");
1608 mask = &rte_flow_item_ipv4_mask;
1609 else if (mask->hdr.next_proto_id != 0 &&
1610 mask->hdr.next_proto_id != 0xff)
1611 return rte_flow_error_set(error, EINVAL,
1612 RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
1613 "partial mask is not supported"
1615 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1616 acc_mask ? (const uint8_t *)acc_mask
1617 : (const uint8_t *)&nic_mask,
1618 sizeof(struct rte_flow_item_ipv4),
1626 * Validate IPV6 item.
1629 * Item specification.
1630 * @param[in] item_flags
1631 * Bit-fields that holds the items detected until now.
1632 * @param[in] acc_mask
1633 * Acceptable mask, if NULL default internal default mask
1634 * will be used to check whether item fields are supported.
1636 * Pointer to error structure.
1639 * 0 on success, a negative errno value otherwise and rte_errno is set.
1642 mlx5_flow_validate_item_ipv6(const struct rte_flow_item *item,
1643 uint64_t item_flags,
1645 uint16_t ether_type,
1646 const struct rte_flow_item_ipv6 *acc_mask,
1647 struct rte_flow_error *error)
1649 const struct rte_flow_item_ipv6 *mask = item->mask;
1650 const struct rte_flow_item_ipv6 *spec = item->spec;
1651 const struct rte_flow_item_ipv6 nic_mask = {
1654 "\xff\xff\xff\xff\xff\xff\xff\xff"
1655 "\xff\xff\xff\xff\xff\xff\xff\xff",
1657 "\xff\xff\xff\xff\xff\xff\xff\xff"
1658 "\xff\xff\xff\xff\xff\xff\xff\xff",
1659 .vtc_flow = RTE_BE32(0xffffffff),
1663 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1664 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1665 MLX5_FLOW_LAYER_OUTER_L3;
1666 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1667 MLX5_FLOW_LAYER_OUTER_L4;
1669 uint8_t next_proto = 0xFF;
1670 const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
1671 MLX5_FLOW_LAYER_OUTER_VLAN |
1672 MLX5_FLOW_LAYER_INNER_VLAN);
1674 if ((last_item & l2_vlan) && ether_type &&
1675 ether_type != RTE_ETHER_TYPE_IPV6)
1676 return rte_flow_error_set(error, EINVAL,
1677 RTE_FLOW_ERROR_TYPE_ITEM, item,
1678 "IPv6 cannot follow L2/VLAN layer "
1679 "which ether type is not IPv6");
1680 if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP) {
1682 next_proto = mask->hdr.proto & spec->hdr.proto;
1683 if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
1684 return rte_flow_error_set(error, EINVAL,
1685 RTE_FLOW_ERROR_TYPE_ITEM,
1690 if (item_flags & MLX5_FLOW_LAYER_IPIP)
1691 return rte_flow_error_set(error, EINVAL,
1692 RTE_FLOW_ERROR_TYPE_ITEM, item,
1693 "wrong tunnel type - IPv4 specified "
1694 "but IPv6 item provided");
1695 if (item_flags & l3m)
1696 return rte_flow_error_set(error, ENOTSUP,
1697 RTE_FLOW_ERROR_TYPE_ITEM, item,
1698 "multiple L3 layers not supported");
1699 else if (item_flags & l4m)
1700 return rte_flow_error_set(error, EINVAL,
1701 RTE_FLOW_ERROR_TYPE_ITEM, item,
1702 "L3 cannot follow an L4 layer.");
1703 else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
1704 !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
1705 return rte_flow_error_set(error, EINVAL,
1706 RTE_FLOW_ERROR_TYPE_ITEM, item,
1707 "L3 cannot follow an NVGRE layer.");
1709 mask = &rte_flow_item_ipv6_mask;
1710 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1711 acc_mask ? (const uint8_t *)acc_mask
1712 : (const uint8_t *)&nic_mask,
1713 sizeof(struct rte_flow_item_ipv6),
1721 * Validate UDP item.
1724 * Item specification.
1725 * @param[in] item_flags
1726 * Bit-fields that holds the items detected until now.
1727 * @param[in] target_protocol
1728 * The next protocol in the previous item.
1729 * @param[in] flow_mask
1730 * mlx5 flow-specific (DV, verbs, etc.) supported header fields mask.
1732 * Pointer to error structure.
1735 * 0 on success, a negative errno value otherwise and rte_errno is set.
1738 mlx5_flow_validate_item_udp(const struct rte_flow_item *item,
1739 uint64_t item_flags,
1740 uint8_t target_protocol,
1741 struct rte_flow_error *error)
1743 const struct rte_flow_item_udp *mask = item->mask;
1744 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1745 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1746 MLX5_FLOW_LAYER_OUTER_L3;
1747 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1748 MLX5_FLOW_LAYER_OUTER_L4;
1751 if (target_protocol != 0xff && target_protocol != IPPROTO_UDP)
1752 return rte_flow_error_set(error, EINVAL,
1753 RTE_FLOW_ERROR_TYPE_ITEM, item,
1754 "protocol filtering not compatible"
1756 if (!(item_flags & l3m))
1757 return rte_flow_error_set(error, EINVAL,
1758 RTE_FLOW_ERROR_TYPE_ITEM, item,
1759 "L3 is mandatory to filter on L4");
1760 if (item_flags & l4m)
1761 return rte_flow_error_set(error, EINVAL,
1762 RTE_FLOW_ERROR_TYPE_ITEM, item,
1763 "multiple L4 layers not supported");
1765 mask = &rte_flow_item_udp_mask;
1766 ret = mlx5_flow_item_acceptable
1767 (item, (const uint8_t *)mask,
1768 (const uint8_t *)&rte_flow_item_udp_mask,
1769 sizeof(struct rte_flow_item_udp), error);
1776 * Validate TCP item.
1779 * Item specification.
1780 * @param[in] item_flags
1781 * Bit-fields that holds the items detected until now.
1782 * @param[in] target_protocol
1783 * The next protocol in the previous item.
1785 * Pointer to error structure.
1788 * 0 on success, a negative errno value otherwise and rte_errno is set.
1791 mlx5_flow_validate_item_tcp(const struct rte_flow_item *item,
1792 uint64_t item_flags,
1793 uint8_t target_protocol,
1794 const struct rte_flow_item_tcp *flow_mask,
1795 struct rte_flow_error *error)
1797 const struct rte_flow_item_tcp *mask = item->mask;
1798 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1799 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1800 MLX5_FLOW_LAYER_OUTER_L3;
1801 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1802 MLX5_FLOW_LAYER_OUTER_L4;
1805 MLX5_ASSERT(flow_mask);
1806 if (target_protocol != 0xff && target_protocol != IPPROTO_TCP)
1807 return rte_flow_error_set(error, EINVAL,
1808 RTE_FLOW_ERROR_TYPE_ITEM, item,
1809 "protocol filtering not compatible"
1811 if (!(item_flags & l3m))
1812 return rte_flow_error_set(error, EINVAL,
1813 RTE_FLOW_ERROR_TYPE_ITEM, item,
1814 "L3 is mandatory to filter on L4");
1815 if (item_flags & l4m)
1816 return rte_flow_error_set(error, EINVAL,
1817 RTE_FLOW_ERROR_TYPE_ITEM, item,
1818 "multiple L4 layers not supported");
1820 mask = &rte_flow_item_tcp_mask;
1821 ret = mlx5_flow_item_acceptable
1822 (item, (const uint8_t *)mask,
1823 (const uint8_t *)flow_mask,
1824 sizeof(struct rte_flow_item_tcp), error);
1831 * Validate VXLAN item.
1834 * Item specification.
1835 * @param[in] item_flags
1836 * Bit-fields that holds the items detected until now.
1837 * @param[in] target_protocol
1838 * The next protocol in the previous item.
1840 * Pointer to error structure.
1843 * 0 on success, a negative errno value otherwise and rte_errno is set.
1846 mlx5_flow_validate_item_vxlan(const struct rte_flow_item *item,
1847 uint64_t item_flags,
1848 struct rte_flow_error *error)
1850 const struct rte_flow_item_vxlan *spec = item->spec;
1851 const struct rte_flow_item_vxlan *mask = item->mask;
1856 } id = { .vlan_id = 0, };
1859 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1860 return rte_flow_error_set(error, ENOTSUP,
1861 RTE_FLOW_ERROR_TYPE_ITEM, item,
1862 "multiple tunnel layers not"
1865 * Verify only UDPv4 is present as defined in
1866 * https://tools.ietf.org/html/rfc7348
1868 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
1869 return rte_flow_error_set(error, EINVAL,
1870 RTE_FLOW_ERROR_TYPE_ITEM, item,
1871 "no outer UDP layer found");
1873 mask = &rte_flow_item_vxlan_mask;
1874 ret = mlx5_flow_item_acceptable
1875 (item, (const uint8_t *)mask,
1876 (const uint8_t *)&rte_flow_item_vxlan_mask,
1877 sizeof(struct rte_flow_item_vxlan),
1882 memcpy(&id.vni[1], spec->vni, 3);
1883 memcpy(&id.vni[1], mask->vni, 3);
1885 if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
1886 return rte_flow_error_set(error, ENOTSUP,
1887 RTE_FLOW_ERROR_TYPE_ITEM, item,
1888 "VXLAN tunnel must be fully defined");
1893 * Validate VXLAN_GPE item.
1896 * Item specification.
1897 * @param[in] item_flags
1898 * Bit-fields that holds the items detected until now.
1900 * Pointer to the private data structure.
1901 * @param[in] target_protocol
1902 * The next protocol in the previous item.
1904 * Pointer to error structure.
1907 * 0 on success, a negative errno value otherwise and rte_errno is set.
1910 mlx5_flow_validate_item_vxlan_gpe(const struct rte_flow_item *item,
1911 uint64_t item_flags,
1912 struct rte_eth_dev *dev,
1913 struct rte_flow_error *error)
1915 struct mlx5_priv *priv = dev->data->dev_private;
1916 const struct rte_flow_item_vxlan_gpe *spec = item->spec;
1917 const struct rte_flow_item_vxlan_gpe *mask = item->mask;
1922 } id = { .vlan_id = 0, };
1924 if (!priv->config.l3_vxlan_en)
1925 return rte_flow_error_set(error, ENOTSUP,
1926 RTE_FLOW_ERROR_TYPE_ITEM, item,
1927 "L3 VXLAN is not enabled by device"
1928 " parameter and/or not configured in"
1930 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1931 return rte_flow_error_set(error, ENOTSUP,
1932 RTE_FLOW_ERROR_TYPE_ITEM, item,
1933 "multiple tunnel layers not"
1936 * Verify only UDPv4 is present as defined in
1937 * https://tools.ietf.org/html/rfc7348
1939 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
1940 return rte_flow_error_set(error, EINVAL,
1941 RTE_FLOW_ERROR_TYPE_ITEM, item,
1942 "no outer UDP layer found");
1944 mask = &rte_flow_item_vxlan_gpe_mask;
1945 ret = mlx5_flow_item_acceptable
1946 (item, (const uint8_t *)mask,
1947 (const uint8_t *)&rte_flow_item_vxlan_gpe_mask,
1948 sizeof(struct rte_flow_item_vxlan_gpe),
1954 return rte_flow_error_set(error, ENOTSUP,
1955 RTE_FLOW_ERROR_TYPE_ITEM,
1957 "VxLAN-GPE protocol"
1959 memcpy(&id.vni[1], spec->vni, 3);
1960 memcpy(&id.vni[1], mask->vni, 3);
1962 if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
1963 return rte_flow_error_set(error, ENOTSUP,
1964 RTE_FLOW_ERROR_TYPE_ITEM, item,
1965 "VXLAN-GPE tunnel must be fully"
1970 * Validate GRE Key item.
1973 * Item specification.
1974 * @param[in] item_flags
1975 * Bit flags to mark detected items.
1976 * @param[in] gre_item
1977 * Pointer to gre_item
1979 * Pointer to error structure.
1982 * 0 on success, a negative errno value otherwise and rte_errno is set.
1985 mlx5_flow_validate_item_gre_key(const struct rte_flow_item *item,
1986 uint64_t item_flags,
1987 const struct rte_flow_item *gre_item,
1988 struct rte_flow_error *error)
1990 const rte_be32_t *mask = item->mask;
1992 rte_be32_t gre_key_default_mask = RTE_BE32(UINT32_MAX);
1993 const struct rte_flow_item_gre *gre_spec;
1994 const struct rte_flow_item_gre *gre_mask;
1996 if (item_flags & MLX5_FLOW_LAYER_GRE_KEY)
1997 return rte_flow_error_set(error, ENOTSUP,
1998 RTE_FLOW_ERROR_TYPE_ITEM, item,
1999 "Multiple GRE key not support");
2000 if (!(item_flags & MLX5_FLOW_LAYER_GRE))
2001 return rte_flow_error_set(error, ENOTSUP,
2002 RTE_FLOW_ERROR_TYPE_ITEM, item,
2003 "No preceding GRE header");
2004 if (item_flags & MLX5_FLOW_LAYER_INNER)
2005 return rte_flow_error_set(error, ENOTSUP,
2006 RTE_FLOW_ERROR_TYPE_ITEM, item,
2007 "GRE key following a wrong item");
2008 gre_mask = gre_item->mask;
2010 gre_mask = &rte_flow_item_gre_mask;
2011 gre_spec = gre_item->spec;
2012 if (gre_spec && (gre_mask->c_rsvd0_ver & RTE_BE16(0x2000)) &&
2013 !(gre_spec->c_rsvd0_ver & RTE_BE16(0x2000)))
2014 return rte_flow_error_set(error, EINVAL,
2015 RTE_FLOW_ERROR_TYPE_ITEM, item,
2016 "Key bit must be on");
2019 mask = &gre_key_default_mask;
2020 ret = mlx5_flow_item_acceptable
2021 (item, (const uint8_t *)mask,
2022 (const uint8_t *)&gre_key_default_mask,
2023 sizeof(rte_be32_t), error);
2028 * Validate GRE item.
2031 * Item specification.
2032 * @param[in] item_flags
2033 * Bit flags to mark detected items.
2034 * @param[in] target_protocol
2035 * The next protocol in the previous item.
2037 * Pointer to error structure.
2040 * 0 on success, a negative errno value otherwise and rte_errno is set.
2043 mlx5_flow_validate_item_gre(const struct rte_flow_item *item,
2044 uint64_t item_flags,
2045 uint8_t target_protocol,
2046 struct rte_flow_error *error)
2048 const struct rte_flow_item_gre *spec __rte_unused = item->spec;
2049 const struct rte_flow_item_gre *mask = item->mask;
2051 const struct rte_flow_item_gre nic_mask = {
2052 .c_rsvd0_ver = RTE_BE16(0xB000),
2053 .protocol = RTE_BE16(UINT16_MAX),
2056 if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2057 return rte_flow_error_set(error, EINVAL,
2058 RTE_FLOW_ERROR_TYPE_ITEM, item,
2059 "protocol filtering not compatible"
2060 " with this GRE layer");
2061 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2062 return rte_flow_error_set(error, ENOTSUP,
2063 RTE_FLOW_ERROR_TYPE_ITEM, item,
2064 "multiple tunnel layers not"
2066 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2067 return rte_flow_error_set(error, ENOTSUP,
2068 RTE_FLOW_ERROR_TYPE_ITEM, item,
2069 "L3 Layer is missing");
2071 mask = &rte_flow_item_gre_mask;
2072 ret = mlx5_flow_item_acceptable
2073 (item, (const uint8_t *)mask,
2074 (const uint8_t *)&nic_mask,
2075 sizeof(struct rte_flow_item_gre), error);
2078 #ifndef HAVE_MLX5DV_DR
2079 #ifndef HAVE_IBV_DEVICE_MPLS_SUPPORT
2080 if (spec && (spec->protocol & mask->protocol))
2081 return rte_flow_error_set(error, ENOTSUP,
2082 RTE_FLOW_ERROR_TYPE_ITEM, item,
2083 "without MPLS support the"
2084 " specification cannot be used for"
2092 * Validate Geneve item.
2095 * Item specification.
2096 * @param[in] itemFlags
2097 * Bit-fields that holds the items detected until now.
2099 * Pointer to the private data structure.
2101 * Pointer to error structure.
2104 * 0 on success, a negative errno value otherwise and rte_errno is set.
2108 mlx5_flow_validate_item_geneve(const struct rte_flow_item *item,
2109 uint64_t item_flags,
2110 struct rte_eth_dev *dev,
2111 struct rte_flow_error *error)
2113 struct mlx5_priv *priv = dev->data->dev_private;
2114 const struct rte_flow_item_geneve *spec = item->spec;
2115 const struct rte_flow_item_geneve *mask = item->mask;
2118 uint8_t opt_len = priv->config.hca_attr.geneve_max_opt_len ?
2119 MLX5_GENEVE_OPT_LEN_1 : MLX5_GENEVE_OPT_LEN_0;
2120 const struct rte_flow_item_geneve nic_mask = {
2121 .ver_opt_len_o_c_rsvd0 = RTE_BE16(0x3f80),
2122 .vni = "\xff\xff\xff",
2123 .protocol = RTE_BE16(UINT16_MAX),
2126 if (!priv->config.hca_attr.tunnel_stateless_geneve_rx)
2127 return rte_flow_error_set(error, ENOTSUP,
2128 RTE_FLOW_ERROR_TYPE_ITEM, item,
2129 "L3 Geneve is not enabled by device"
2130 " parameter and/or not configured in"
2132 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2133 return rte_flow_error_set(error, ENOTSUP,
2134 RTE_FLOW_ERROR_TYPE_ITEM, item,
2135 "multiple tunnel layers not"
2138 * Verify only UDPv4 is present as defined in
2139 * https://tools.ietf.org/html/rfc7348
2141 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2142 return rte_flow_error_set(error, EINVAL,
2143 RTE_FLOW_ERROR_TYPE_ITEM, item,
2144 "no outer UDP layer found");
2146 mask = &rte_flow_item_geneve_mask;
2147 ret = mlx5_flow_item_acceptable
2148 (item, (const uint8_t *)mask,
2149 (const uint8_t *)&nic_mask,
2150 sizeof(struct rte_flow_item_geneve), error);
2154 gbhdr = rte_be_to_cpu_16(spec->ver_opt_len_o_c_rsvd0);
2155 if (MLX5_GENEVE_VER_VAL(gbhdr) ||
2156 MLX5_GENEVE_CRITO_VAL(gbhdr) ||
2157 MLX5_GENEVE_RSVD_VAL(gbhdr) || spec->rsvd1)
2158 return rte_flow_error_set(error, ENOTSUP,
2159 RTE_FLOW_ERROR_TYPE_ITEM,
2161 "Geneve protocol unsupported"
2162 " fields are being used");
2163 if (MLX5_GENEVE_OPTLEN_VAL(gbhdr) > opt_len)
2164 return rte_flow_error_set
2166 RTE_FLOW_ERROR_TYPE_ITEM,
2168 "Unsupported Geneve options length");
2170 if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2171 return rte_flow_error_set
2173 RTE_FLOW_ERROR_TYPE_ITEM, item,
2174 "Geneve tunnel must be fully defined");
2179 * Validate MPLS item.
2182 * Pointer to the rte_eth_dev structure.
2184 * Item specification.
2185 * @param[in] item_flags
2186 * Bit-fields that holds the items detected until now.
2187 * @param[in] prev_layer
2188 * The protocol layer indicated in previous item.
2190 * Pointer to error structure.
2193 * 0 on success, a negative errno value otherwise and rte_errno is set.
2196 mlx5_flow_validate_item_mpls(struct rte_eth_dev *dev __rte_unused,
2197 const struct rte_flow_item *item __rte_unused,
2198 uint64_t item_flags __rte_unused,
2199 uint64_t prev_layer __rte_unused,
2200 struct rte_flow_error *error)
2202 #ifdef HAVE_IBV_DEVICE_MPLS_SUPPORT
2203 const struct rte_flow_item_mpls *mask = item->mask;
2204 struct mlx5_priv *priv = dev->data->dev_private;
2207 if (!priv->config.mpls_en)
2208 return rte_flow_error_set(error, ENOTSUP,
2209 RTE_FLOW_ERROR_TYPE_ITEM, item,
2210 "MPLS not supported or"
2211 " disabled in firmware"
2213 /* MPLS over IP, UDP, GRE is allowed */
2214 if (!(prev_layer & (MLX5_FLOW_LAYER_OUTER_L3 |
2215 MLX5_FLOW_LAYER_OUTER_L4_UDP |
2216 MLX5_FLOW_LAYER_GRE)))
2217 return rte_flow_error_set(error, EINVAL,
2218 RTE_FLOW_ERROR_TYPE_ITEM, item,
2219 "protocol filtering not compatible"
2220 " with MPLS layer");
2221 /* Multi-tunnel isn't allowed but MPLS over GRE is an exception. */
2222 if ((item_flags & MLX5_FLOW_LAYER_TUNNEL) &&
2223 !(item_flags & MLX5_FLOW_LAYER_GRE))
2224 return rte_flow_error_set(error, ENOTSUP,
2225 RTE_FLOW_ERROR_TYPE_ITEM, item,
2226 "multiple tunnel layers not"
2229 mask = &rte_flow_item_mpls_mask;
2230 ret = mlx5_flow_item_acceptable
2231 (item, (const uint8_t *)mask,
2232 (const uint8_t *)&rte_flow_item_mpls_mask,
2233 sizeof(struct rte_flow_item_mpls), error);
2238 return rte_flow_error_set(error, ENOTSUP,
2239 RTE_FLOW_ERROR_TYPE_ITEM, item,
2240 "MPLS is not supported by Verbs, please"
2245 * Validate NVGRE item.
2248 * Item specification.
2249 * @param[in] item_flags
2250 * Bit flags to mark detected items.
2251 * @param[in] target_protocol
2252 * The next protocol in the previous item.
2254 * Pointer to error structure.
2257 * 0 on success, a negative errno value otherwise and rte_errno is set.
2260 mlx5_flow_validate_item_nvgre(const struct rte_flow_item *item,
2261 uint64_t item_flags,
2262 uint8_t target_protocol,
2263 struct rte_flow_error *error)
2265 const struct rte_flow_item_nvgre *mask = item->mask;
2268 if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2269 return rte_flow_error_set(error, EINVAL,
2270 RTE_FLOW_ERROR_TYPE_ITEM, item,
2271 "protocol filtering not compatible"
2272 " with this GRE layer");
2273 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2274 return rte_flow_error_set(error, ENOTSUP,
2275 RTE_FLOW_ERROR_TYPE_ITEM, item,
2276 "multiple tunnel layers not"
2278 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2279 return rte_flow_error_set(error, ENOTSUP,
2280 RTE_FLOW_ERROR_TYPE_ITEM, item,
2281 "L3 Layer is missing");
2283 mask = &rte_flow_item_nvgre_mask;
2284 ret = mlx5_flow_item_acceptable
2285 (item, (const uint8_t *)mask,
2286 (const uint8_t *)&rte_flow_item_nvgre_mask,
2287 sizeof(struct rte_flow_item_nvgre), error);
2293 /* Allocate unique ID for the split Q/RSS subflows. */
2295 flow_qrss_get_id(struct rte_eth_dev *dev)
2297 struct mlx5_priv *priv = dev->data->dev_private;
2298 uint32_t qrss_id, ret;
2300 ret = mlx5_flow_id_get(priv->qrss_id_pool, &qrss_id);
2303 MLX5_ASSERT(qrss_id);
2307 /* Free unique ID for the split Q/RSS subflows. */
2309 flow_qrss_free_id(struct rte_eth_dev *dev, uint32_t qrss_id)
2311 struct mlx5_priv *priv = dev->data->dev_private;
2314 mlx5_flow_id_release(priv->qrss_id_pool, qrss_id);
2318 * Release resource related QUEUE/RSS action split.
2321 * Pointer to Ethernet device.
2323 * Flow to release id's from.
2326 flow_mreg_split_qrss_release(struct rte_eth_dev *dev,
2327 struct rte_flow *flow)
2329 struct mlx5_priv *priv = dev->data->dev_private;
2330 uint32_t handle_idx;
2331 struct mlx5_flow_handle *dev_handle;
2333 SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
2334 handle_idx, dev_handle, next)
2335 if (dev_handle->split_flow_id)
2336 flow_qrss_free_id(dev, dev_handle->split_flow_id);
2340 flow_null_validate(struct rte_eth_dev *dev __rte_unused,
2341 const struct rte_flow_attr *attr __rte_unused,
2342 const struct rte_flow_item items[] __rte_unused,
2343 const struct rte_flow_action actions[] __rte_unused,
2344 bool external __rte_unused,
2345 struct rte_flow_error *error)
2347 return rte_flow_error_set(error, ENOTSUP,
2348 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2351 static struct mlx5_flow *
2352 flow_null_prepare(struct rte_eth_dev *dev __rte_unused,
2353 const struct rte_flow_attr *attr __rte_unused,
2354 const struct rte_flow_item items[] __rte_unused,
2355 const struct rte_flow_action actions[] __rte_unused,
2356 struct rte_flow_error *error)
2358 rte_flow_error_set(error, ENOTSUP,
2359 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2364 flow_null_translate(struct rte_eth_dev *dev __rte_unused,
2365 struct mlx5_flow *dev_flow __rte_unused,
2366 const struct rte_flow_attr *attr __rte_unused,
2367 const struct rte_flow_item items[] __rte_unused,
2368 const struct rte_flow_action actions[] __rte_unused,
2369 struct rte_flow_error *error)
2371 return rte_flow_error_set(error, ENOTSUP,
2372 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2376 flow_null_apply(struct rte_eth_dev *dev __rte_unused,
2377 struct rte_flow *flow __rte_unused,
2378 struct rte_flow_error *error)
2380 return rte_flow_error_set(error, ENOTSUP,
2381 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2385 flow_null_remove(struct rte_eth_dev *dev __rte_unused,
2386 struct rte_flow *flow __rte_unused)
2391 flow_null_destroy(struct rte_eth_dev *dev __rte_unused,
2392 struct rte_flow *flow __rte_unused)
2397 flow_null_query(struct rte_eth_dev *dev __rte_unused,
2398 struct rte_flow *flow __rte_unused,
2399 const struct rte_flow_action *actions __rte_unused,
2400 void *data __rte_unused,
2401 struct rte_flow_error *error)
2403 return rte_flow_error_set(error, ENOTSUP,
2404 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2407 /* Void driver to protect from null pointer reference. */
2408 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops = {
2409 .validate = flow_null_validate,
2410 .prepare = flow_null_prepare,
2411 .translate = flow_null_translate,
2412 .apply = flow_null_apply,
2413 .remove = flow_null_remove,
2414 .destroy = flow_null_destroy,
2415 .query = flow_null_query,
2419 * Select flow driver type according to flow attributes and device
2423 * Pointer to the dev structure.
2425 * Pointer to the flow attributes.
2428 * flow driver type, MLX5_FLOW_TYPE_MAX otherwise.
2430 static enum mlx5_flow_drv_type
2431 flow_get_drv_type(struct rte_eth_dev *dev, const struct rte_flow_attr *attr)
2433 struct mlx5_priv *priv = dev->data->dev_private;
2434 enum mlx5_flow_drv_type type = MLX5_FLOW_TYPE_MAX;
2436 if (attr->transfer && priv->config.dv_esw_en)
2437 type = MLX5_FLOW_TYPE_DV;
2438 if (!attr->transfer)
2439 type = priv->config.dv_flow_en ? MLX5_FLOW_TYPE_DV :
2440 MLX5_FLOW_TYPE_VERBS;
2444 #define flow_get_drv_ops(type) flow_drv_ops[type]
2447 * Flow driver validation API. This abstracts calling driver specific functions.
2448 * The type of flow driver is determined according to flow attributes.
2451 * Pointer to the dev structure.
2453 * Pointer to the flow attributes.
2455 * Pointer to the list of items.
2456 * @param[in] actions
2457 * Pointer to the list of actions.
2458 * @param[in] external
2459 * This flow rule is created by request external to PMD.
2461 * Pointer to the error structure.
2464 * 0 on success, a negative errno value otherwise and rte_errno is set.
2467 flow_drv_validate(struct rte_eth_dev *dev,
2468 const struct rte_flow_attr *attr,
2469 const struct rte_flow_item items[],
2470 const struct rte_flow_action actions[],
2471 bool external, struct rte_flow_error *error)
2473 const struct mlx5_flow_driver_ops *fops;
2474 enum mlx5_flow_drv_type type = flow_get_drv_type(dev, attr);
2476 fops = flow_get_drv_ops(type);
2477 return fops->validate(dev, attr, items, actions, external, error);
2481 * Flow driver preparation API. This abstracts calling driver specific
2482 * functions. Parent flow (rte_flow) should have driver type (drv_type). It
2483 * calculates the size of memory required for device flow, allocates the memory,
2484 * initializes the device flow and returns the pointer.
2487 * This function initializes device flow structure such as dv or verbs in
2488 * struct mlx5_flow. However, it is caller's responsibility to initialize the
2489 * rest. For example, adding returning device flow to flow->dev_flow list and
2490 * setting backward reference to the flow should be done out of this function.
2491 * layers field is not filled either.
2494 * Pointer to the dev structure.
2496 * Pointer to the flow attributes.
2498 * Pointer to the list of items.
2499 * @param[in] actions
2500 * Pointer to the list of actions.
2502 * Pointer to the error structure.
2505 * Pointer to device flow on success, otherwise NULL and rte_errno is set.
2507 static inline struct mlx5_flow *
2508 flow_drv_prepare(struct rte_eth_dev *dev,
2509 const struct rte_flow *flow,
2510 const struct rte_flow_attr *attr,
2511 const struct rte_flow_item items[],
2512 const struct rte_flow_action actions[],
2513 struct rte_flow_error *error)
2515 const struct mlx5_flow_driver_ops *fops;
2516 enum mlx5_flow_drv_type type = flow->drv_type;
2518 MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2519 fops = flow_get_drv_ops(type);
2520 return fops->prepare(dev, attr, items, actions, error);
2524 * Flow driver translation API. This abstracts calling driver specific
2525 * functions. Parent flow (rte_flow) should have driver type (drv_type). It
2526 * translates a generic flow into a driver flow. flow_drv_prepare() must
2530 * dev_flow->layers could be filled as a result of parsing during translation
2531 * if needed by flow_drv_apply(). dev_flow->flow->actions can also be filled
2532 * if necessary. As a flow can have multiple dev_flows by RSS flow expansion,
2533 * flow->actions could be overwritten even though all the expanded dev_flows
2534 * have the same actions.
2537 * Pointer to the rte dev structure.
2538 * @param[in, out] dev_flow
2539 * Pointer to the mlx5 flow.
2541 * Pointer to the flow attributes.
2543 * Pointer to the list of items.
2544 * @param[in] actions
2545 * Pointer to the list of actions.
2547 * Pointer to the error structure.
2550 * 0 on success, a negative errno value otherwise and rte_errno is set.
2553 flow_drv_translate(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow,
2554 const struct rte_flow_attr *attr,
2555 const struct rte_flow_item items[],
2556 const struct rte_flow_action actions[],
2557 struct rte_flow_error *error)
2559 const struct mlx5_flow_driver_ops *fops;
2560 enum mlx5_flow_drv_type type = dev_flow->flow->drv_type;
2562 MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2563 fops = flow_get_drv_ops(type);
2564 return fops->translate(dev, dev_flow, attr, items, actions, error);
2568 * Flow driver apply API. This abstracts calling driver specific functions.
2569 * Parent flow (rte_flow) should have driver type (drv_type). It applies
2570 * translated driver flows on to device. flow_drv_translate() must precede.
2573 * Pointer to Ethernet device structure.
2574 * @param[in, out] flow
2575 * Pointer to flow structure.
2577 * Pointer to error structure.
2580 * 0 on success, a negative errno value otherwise and rte_errno is set.
2583 flow_drv_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
2584 struct rte_flow_error *error)
2586 const struct mlx5_flow_driver_ops *fops;
2587 enum mlx5_flow_drv_type type = flow->drv_type;
2589 MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2590 fops = flow_get_drv_ops(type);
2591 return fops->apply(dev, flow, error);
2595 * Flow driver remove API. This abstracts calling driver specific functions.
2596 * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
2597 * on device. All the resources of the flow should be freed by calling
2598 * flow_drv_destroy().
2601 * Pointer to Ethernet device.
2602 * @param[in, out] flow
2603 * Pointer to flow structure.
2606 flow_drv_remove(struct rte_eth_dev *dev, struct rte_flow *flow)
2608 const struct mlx5_flow_driver_ops *fops;
2609 enum mlx5_flow_drv_type type = flow->drv_type;
2611 MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2612 fops = flow_get_drv_ops(type);
2613 fops->remove(dev, flow);
2617 * Flow driver destroy API. This abstracts calling driver specific functions.
2618 * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
2619 * on device and releases resources of the flow.
2622 * Pointer to Ethernet device.
2623 * @param[in, out] flow
2624 * Pointer to flow structure.
2627 flow_drv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
2629 const struct mlx5_flow_driver_ops *fops;
2630 enum mlx5_flow_drv_type type = flow->drv_type;
2632 flow_mreg_split_qrss_release(dev, flow);
2633 MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2634 fops = flow_get_drv_ops(type);
2635 fops->destroy(dev, flow);
2639 * Validate a flow supported by the NIC.
2641 * @see rte_flow_validate()
2645 mlx5_flow_validate(struct rte_eth_dev *dev,
2646 const struct rte_flow_attr *attr,
2647 const struct rte_flow_item items[],
2648 const struct rte_flow_action actions[],
2649 struct rte_flow_error *error)
2653 ret = flow_drv_validate(dev, attr, items, actions, true, error);
2660 * Get RSS action from the action list.
2662 * @param[in] actions
2663 * Pointer to the list of actions.
2666 * Pointer to the RSS action if exist, else return NULL.
2668 static const struct rte_flow_action_rss*
2669 flow_get_rss_action(const struct rte_flow_action actions[])
2671 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2672 switch (actions->type) {
2673 case RTE_FLOW_ACTION_TYPE_RSS:
2674 return (const struct rte_flow_action_rss *)
2684 find_graph_root(const struct rte_flow_item pattern[], uint32_t rss_level)
2686 const struct rte_flow_item *item;
2687 unsigned int has_vlan = 0;
2689 for (item = pattern; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
2690 if (item->type == RTE_FLOW_ITEM_TYPE_VLAN) {
2696 return rss_level < 2 ? MLX5_EXPANSION_ROOT_ETH_VLAN :
2697 MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN;
2698 return rss_level < 2 ? MLX5_EXPANSION_ROOT :
2699 MLX5_EXPANSION_ROOT_OUTER;
2703 * Get layer flags from the prefix flow.
2705 * Some flows may be split to several subflows, the prefix subflow gets the
2706 * match items and the suffix sub flow gets the actions.
2707 * Some actions need the user defined match item flags to get the detail for
2709 * This function helps the suffix flow to get the item layer flags from prefix
2712 * @param[in] dev_flow
2713 * Pointer the created preifx subflow.
2716 * The layers get from prefix subflow.
2718 static inline uint64_t
2719 flow_get_prefix_layer_flags(struct mlx5_flow *dev_flow)
2721 uint64_t layers = 0;
2724 * Layers bits could be localization, but usually the compiler will
2725 * help to do the optimization work for source code.
2726 * If no decap actions, use the layers directly.
2728 if (!(dev_flow->act_flags & MLX5_FLOW_ACTION_DECAP))
2729 return dev_flow->handle->layers;
2730 /* Convert L3 layers with decap action. */
2731 if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV4)
2732 layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV4;
2733 else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV6)
2734 layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV6;
2735 /* Convert L4 layers with decap action. */
2736 if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_TCP)
2737 layers |= MLX5_FLOW_LAYER_OUTER_L4_TCP;
2738 else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_UDP)
2739 layers |= MLX5_FLOW_LAYER_OUTER_L4_UDP;
2744 * Get metadata split action information.
2746 * @param[in] actions
2747 * Pointer to the list of actions.
2749 * Pointer to the return pointer.
2750 * @param[out] qrss_type
2751 * Pointer to the action type to return. RTE_FLOW_ACTION_TYPE_END is returned
2752 * if no QUEUE/RSS is found.
2753 * @param[out] encap_idx
2754 * Pointer to the index of the encap action if exists, otherwise the last
2758 * Total number of actions.
2761 flow_parse_metadata_split_actions_info(const struct rte_flow_action actions[],
2762 const struct rte_flow_action **qrss,
2765 const struct rte_flow_action_raw_encap *raw_encap;
2767 int raw_decap_idx = -1;
2770 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2771 switch (actions->type) {
2772 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
2773 case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
2774 *encap_idx = actions_n;
2776 case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
2777 raw_decap_idx = actions_n;
2779 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
2780 raw_encap = actions->conf;
2781 if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
2782 *encap_idx = raw_decap_idx != -1 ?
2783 raw_decap_idx : actions_n;
2785 case RTE_FLOW_ACTION_TYPE_QUEUE:
2786 case RTE_FLOW_ACTION_TYPE_RSS:
2794 if (*encap_idx == -1)
2795 *encap_idx = actions_n;
2796 /* Count RTE_FLOW_ACTION_TYPE_END. */
2797 return actions_n + 1;
2801 * Check meter action from the action list.
2803 * @param[in] actions
2804 * Pointer to the list of actions.
2806 * Pointer to the meter exist flag.
2809 * Total number of actions.
2812 flow_check_meter_action(const struct rte_flow_action actions[], uint32_t *mtr)
2818 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2819 switch (actions->type) {
2820 case RTE_FLOW_ACTION_TYPE_METER:
2828 /* Count RTE_FLOW_ACTION_TYPE_END. */
2829 return actions_n + 1;
2833 * Check if the flow should be splited due to hairpin.
2834 * The reason for the split is that in current HW we can't
2835 * support encap on Rx, so if a flow have encap we move it
2839 * Pointer to Ethernet device.
2841 * Flow rule attributes.
2842 * @param[in] actions
2843 * Associated actions (list terminated by the END action).
2846 * > 0 the number of actions and the flow should be split,
2847 * 0 when no split required.
2850 flow_check_hairpin_split(struct rte_eth_dev *dev,
2851 const struct rte_flow_attr *attr,
2852 const struct rte_flow_action actions[])
2854 int queue_action = 0;
2857 const struct rte_flow_action_queue *queue;
2858 const struct rte_flow_action_rss *rss;
2859 const struct rte_flow_action_raw_encap *raw_encap;
2863 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2864 switch (actions->type) {
2865 case RTE_FLOW_ACTION_TYPE_QUEUE:
2866 queue = actions->conf;
2869 if (mlx5_rxq_get_type(dev, queue->index) !=
2870 MLX5_RXQ_TYPE_HAIRPIN)
2875 case RTE_FLOW_ACTION_TYPE_RSS:
2876 rss = actions->conf;
2877 if (rss == NULL || rss->queue_num == 0)
2879 if (mlx5_rxq_get_type(dev, rss->queue[0]) !=
2880 MLX5_RXQ_TYPE_HAIRPIN)
2885 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
2886 case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
2890 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
2891 raw_encap = actions->conf;
2892 if (raw_encap->size >
2893 (sizeof(struct rte_flow_item_eth) +
2894 sizeof(struct rte_flow_item_ipv4)))
2903 if (encap == 1 && queue_action)
2908 /* Declare flow create/destroy prototype in advance. */
2909 static struct rte_flow *
2910 flow_list_create(struct rte_eth_dev *dev, struct mlx5_flows *list,
2911 const struct rte_flow_attr *attr,
2912 const struct rte_flow_item items[],
2913 const struct rte_flow_action actions[],
2914 bool external, struct rte_flow_error *error);
2917 flow_list_destroy(struct rte_eth_dev *dev, struct mlx5_flows *list,
2918 struct rte_flow *flow);
2921 * Add a flow of copying flow metadata registers in RX_CP_TBL.
2923 * As mark_id is unique, if there's already a registered flow for the mark_id,
2924 * return by increasing the reference counter of the resource. Otherwise, create
2925 * the resource (mcp_res) and flow.
2928 * - If ingress port is ANY and reg_c[1] is mark_id,
2929 * flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
2931 * For default flow (zero mark_id), flow is like,
2932 * - If ingress port is ANY,
2933 * reg_b := reg_c[0] and jump to RX_ACT_TBL.
2936 * Pointer to Ethernet device.
2938 * ID of MARK action, zero means default flow for META.
2940 * Perform verbose error reporting if not NULL.
2943 * Associated resource on success, NULL otherwise and rte_errno is set.
2945 static struct mlx5_flow_mreg_copy_resource *
2946 flow_mreg_add_copy_action(struct rte_eth_dev *dev, uint32_t mark_id,
2947 struct rte_flow_error *error)
2949 struct mlx5_priv *priv = dev->data->dev_private;
2950 struct rte_flow_attr attr = {
2951 .group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
2954 struct mlx5_rte_flow_item_tag tag_spec = {
2957 struct rte_flow_item items[] = {
2958 [1] = { .type = RTE_FLOW_ITEM_TYPE_END, },
2960 struct rte_flow_action_mark ftag = {
2963 struct mlx5_flow_action_copy_mreg cp_mreg = {
2967 struct rte_flow_action_jump jump = {
2968 .group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
2970 struct rte_flow_action actions[] = {
2971 [3] = { .type = RTE_FLOW_ACTION_TYPE_END, },
2973 struct mlx5_flow_mreg_copy_resource *mcp_res;
2977 /* Fill the register fileds in the flow. */
2978 ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
2982 ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
2986 /* Check if already registered. */
2987 MLX5_ASSERT(priv->mreg_cp_tbl);
2988 mcp_res = (void *)mlx5_hlist_lookup(priv->mreg_cp_tbl, mark_id);
2990 /* For non-default rule. */
2991 if (mark_id != MLX5_DEFAULT_COPY_ID)
2993 MLX5_ASSERT(mark_id != MLX5_DEFAULT_COPY_ID ||
2994 mcp_res->refcnt == 1);
2997 /* Provide the full width of FLAG specific value. */
2998 if (mark_id == (priv->sh->dv_regc0_mask & MLX5_FLOW_MARK_DEFAULT))
2999 tag_spec.data = MLX5_FLOW_MARK_DEFAULT;
3000 /* Build a new flow. */
3001 if (mark_id != MLX5_DEFAULT_COPY_ID) {
3002 items[0] = (struct rte_flow_item){
3003 .type = MLX5_RTE_FLOW_ITEM_TYPE_TAG,
3006 items[1] = (struct rte_flow_item){
3007 .type = RTE_FLOW_ITEM_TYPE_END,
3009 actions[0] = (struct rte_flow_action){
3010 .type = MLX5_RTE_FLOW_ACTION_TYPE_MARK,
3013 actions[1] = (struct rte_flow_action){
3014 .type = MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3017 actions[2] = (struct rte_flow_action){
3018 .type = RTE_FLOW_ACTION_TYPE_JUMP,
3021 actions[3] = (struct rte_flow_action){
3022 .type = RTE_FLOW_ACTION_TYPE_END,
3025 /* Default rule, wildcard match. */
3026 attr.priority = MLX5_FLOW_PRIO_RSVD;
3027 items[0] = (struct rte_flow_item){
3028 .type = RTE_FLOW_ITEM_TYPE_END,
3030 actions[0] = (struct rte_flow_action){
3031 .type = MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3034 actions[1] = (struct rte_flow_action){
3035 .type = RTE_FLOW_ACTION_TYPE_JUMP,
3038 actions[2] = (struct rte_flow_action){
3039 .type = RTE_FLOW_ACTION_TYPE_END,
3042 /* Build a new entry. */
3043 mcp_res = mlx5_ipool_zmalloc(priv->sh->ipool[MLX5_IPOOL_MCP], &idx);
3050 * The copy Flows are not included in any list. There
3051 * ones are referenced from other Flows and can not
3052 * be applied, removed, deleted in ardbitrary order
3053 * by list traversing.
3055 mcp_res->flow = flow_list_create(dev, NULL, &attr, items,
3056 actions, false, error);
3060 mcp_res->hlist_ent.key = mark_id;
3061 ret = mlx5_hlist_insert(priv->mreg_cp_tbl,
3062 &mcp_res->hlist_ent);
3069 flow_list_destroy(dev, NULL, mcp_res->flow);
3070 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], mcp_res->idx);
3075 * Release flow in RX_CP_TBL.
3078 * Pointer to Ethernet device.
3080 * Parent flow for wich copying is provided.
3083 flow_mreg_del_copy_action(struct rte_eth_dev *dev,
3084 struct rte_flow *flow)
3086 struct mlx5_flow_mreg_copy_resource *mcp_res;
3087 struct mlx5_priv *priv = dev->data->dev_private;
3089 if (!flow->rix_mreg_copy)
3091 mcp_res = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MCP],
3092 flow->rix_mreg_copy);
3093 if (!mcp_res || !priv->mreg_cp_tbl)
3095 if (flow->copy_applied) {
3096 MLX5_ASSERT(mcp_res->appcnt);
3097 flow->copy_applied = 0;
3099 if (!mcp_res->appcnt)
3100 flow_drv_remove(dev, mcp_res->flow);
3103 * We do not check availability of metadata registers here,
3104 * because copy resources are not allocated in this case.
3106 if (--mcp_res->refcnt)
3108 MLX5_ASSERT(mcp_res->flow);
3109 flow_list_destroy(dev, NULL, mcp_res->flow);
3110 mlx5_hlist_remove(priv->mreg_cp_tbl, &mcp_res->hlist_ent);
3111 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], mcp_res->idx);
3112 flow->rix_mreg_copy = 0;
3116 * Start flow in RX_CP_TBL.
3119 * Pointer to Ethernet device.
3121 * Parent flow for wich copying is provided.
3124 * 0 on success, a negative errno value otherwise and rte_errno is set.
3127 flow_mreg_start_copy_action(struct rte_eth_dev *dev,
3128 struct rte_flow *flow)
3130 struct mlx5_flow_mreg_copy_resource *mcp_res;
3131 struct mlx5_priv *priv = dev->data->dev_private;
3134 if (!flow->rix_mreg_copy || flow->copy_applied)
3136 mcp_res = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MCP],
3137 flow->rix_mreg_copy);
3140 if (!mcp_res->appcnt) {
3141 ret = flow_drv_apply(dev, mcp_res->flow, NULL);
3146 flow->copy_applied = 1;
3151 * Stop flow in RX_CP_TBL.
3154 * Pointer to Ethernet device.
3156 * Parent flow for wich copying is provided.
3159 flow_mreg_stop_copy_action(struct rte_eth_dev *dev,
3160 struct rte_flow *flow)
3162 struct mlx5_flow_mreg_copy_resource *mcp_res;
3163 struct mlx5_priv *priv = dev->data->dev_private;
3165 if (!flow->rix_mreg_copy || !flow->copy_applied)
3167 mcp_res = mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_MCP],
3168 flow->rix_mreg_copy);
3171 MLX5_ASSERT(mcp_res->appcnt);
3173 flow->copy_applied = 0;
3174 if (!mcp_res->appcnt)
3175 flow_drv_remove(dev, mcp_res->flow);
3179 * Remove the default copy action from RX_CP_TBL.
3182 * Pointer to Ethernet device.
3185 flow_mreg_del_default_copy_action(struct rte_eth_dev *dev)
3187 struct mlx5_flow_mreg_copy_resource *mcp_res;
3188 struct mlx5_priv *priv = dev->data->dev_private;
3190 /* Check if default flow is registered. */
3191 if (!priv->mreg_cp_tbl)
3193 mcp_res = (void *)mlx5_hlist_lookup(priv->mreg_cp_tbl,
3194 MLX5_DEFAULT_COPY_ID);
3197 MLX5_ASSERT(mcp_res->flow);
3198 flow_list_destroy(dev, NULL, mcp_res->flow);
3199 mlx5_hlist_remove(priv->mreg_cp_tbl, &mcp_res->hlist_ent);
3200 mlx5_ipool_free(priv->sh->ipool[MLX5_IPOOL_MCP], mcp_res->idx);
3204 * Add the default copy action in in RX_CP_TBL.
3207 * Pointer to Ethernet device.
3209 * Perform verbose error reporting if not NULL.
3212 * 0 for success, negative value otherwise and rte_errno is set.
3215 flow_mreg_add_default_copy_action(struct rte_eth_dev *dev,
3216 struct rte_flow_error *error)
3218 struct mlx5_priv *priv = dev->data->dev_private;
3219 struct mlx5_flow_mreg_copy_resource *mcp_res;
3221 /* Check whether extensive metadata feature is engaged. */
3222 if (!priv->config.dv_flow_en ||
3223 priv->config.dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
3224 !mlx5_flow_ext_mreg_supported(dev) ||
3225 !priv->sh->dv_regc0_mask)
3227 mcp_res = flow_mreg_add_copy_action(dev, MLX5_DEFAULT_COPY_ID, error);
3234 * Add a flow of copying flow metadata registers in RX_CP_TBL.
3236 * All the flow having Q/RSS action should be split by
3237 * flow_mreg_split_qrss_prep() to pass by RX_CP_TBL. A flow in the RX_CP_TBL
3238 * performs the following,
3239 * - CQE->flow_tag := reg_c[1] (MARK)
3240 * - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
3241 * As CQE's flow_tag is not a register, it can't be simply copied from reg_c[1]
3242 * but there should be a flow per each MARK ID set by MARK action.
3244 * For the aforementioned reason, if there's a MARK action in flow's action
3245 * list, a corresponding flow should be added to the RX_CP_TBL in order to copy
3246 * the MARK ID to CQE's flow_tag like,
3247 * - If reg_c[1] is mark_id,
3248 * flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
3250 * For SET_META action which stores value in reg_c[0], as the destination is
3251 * also a flow metadata register (reg_b), adding a default flow is enough. Zero
3252 * MARK ID means the default flow. The default flow looks like,
3253 * - For all flow, reg_b := reg_c[0] and jump to RX_ACT_TBL.
3256 * Pointer to Ethernet device.
3258 * Pointer to flow structure.
3259 * @param[in] actions
3260 * Pointer to the list of actions.
3262 * Perform verbose error reporting if not NULL.
3265 * 0 on success, negative value otherwise and rte_errno is set.
3268 flow_mreg_update_copy_table(struct rte_eth_dev *dev,
3269 struct rte_flow *flow,
3270 const struct rte_flow_action *actions,
3271 struct rte_flow_error *error)
3273 struct mlx5_priv *priv = dev->data->dev_private;
3274 struct mlx5_dev_config *config = &priv->config;
3275 struct mlx5_flow_mreg_copy_resource *mcp_res;
3276 const struct rte_flow_action_mark *mark;
3278 /* Check whether extensive metadata feature is engaged. */
3279 if (!config->dv_flow_en ||
3280 config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
3281 !mlx5_flow_ext_mreg_supported(dev) ||
3282 !priv->sh->dv_regc0_mask)
3284 /* Find MARK action. */
3285 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3286 switch (actions->type) {
3287 case RTE_FLOW_ACTION_TYPE_FLAG:
3288 mcp_res = flow_mreg_add_copy_action
3289 (dev, MLX5_FLOW_MARK_DEFAULT, error);
3292 flow->rix_mreg_copy = mcp_res->idx;
3293 if (dev->data->dev_started) {
3295 flow->copy_applied = 1;
3298 case RTE_FLOW_ACTION_TYPE_MARK:
3299 mark = (const struct rte_flow_action_mark *)
3302 flow_mreg_add_copy_action(dev, mark->id, error);
3305 flow->rix_mreg_copy = mcp_res->idx;
3306 if (dev->data->dev_started) {
3308 flow->copy_applied = 1;
3318 #define MLX5_MAX_SPLIT_ACTIONS 24
3319 #define MLX5_MAX_SPLIT_ITEMS 24
3322 * Split the hairpin flow.
3323 * Since HW can't support encap on Rx we move the encap to Tx.
3324 * If the count action is after the encap then we also
3325 * move the count action. in this case the count will also measure
3329 * Pointer to Ethernet device.
3330 * @param[in] actions
3331 * Associated actions (list terminated by the END action).
3332 * @param[out] actions_rx
3334 * @param[out] actions_tx
3336 * @param[out] pattern_tx
3337 * The pattern items for the Tx flow.
3338 * @param[out] flow_id
3339 * The flow ID connected to this flow.
3345 flow_hairpin_split(struct rte_eth_dev *dev,
3346 const struct rte_flow_action actions[],
3347 struct rte_flow_action actions_rx[],
3348 struct rte_flow_action actions_tx[],
3349 struct rte_flow_item pattern_tx[],
3352 struct mlx5_priv *priv = dev->data->dev_private;
3353 const struct rte_flow_action_raw_encap *raw_encap;
3354 const struct rte_flow_action_raw_decap *raw_decap;
3355 struct mlx5_rte_flow_action_set_tag *set_tag;
3356 struct rte_flow_action *tag_action;
3357 struct mlx5_rte_flow_item_tag *tag_item;
3358 struct rte_flow_item *item;
3362 mlx5_flow_id_get(priv->sh->flow_id_pool, flow_id);
3363 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3364 switch (actions->type) {
3365 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
3366 case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
3367 rte_memcpy(actions_tx, actions,
3368 sizeof(struct rte_flow_action));
3371 case RTE_FLOW_ACTION_TYPE_COUNT:
3373 rte_memcpy(actions_tx, actions,
3374 sizeof(struct rte_flow_action));
3377 rte_memcpy(actions_rx, actions,
3378 sizeof(struct rte_flow_action));
3382 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3383 raw_encap = actions->conf;
3384 if (raw_encap->size >
3385 (sizeof(struct rte_flow_item_eth) +
3386 sizeof(struct rte_flow_item_ipv4))) {
3387 memcpy(actions_tx, actions,
3388 sizeof(struct rte_flow_action));
3392 rte_memcpy(actions_rx, actions,
3393 sizeof(struct rte_flow_action));
3397 case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
3398 raw_decap = actions->conf;
3399 if (raw_decap->size <
3400 (sizeof(struct rte_flow_item_eth) +
3401 sizeof(struct rte_flow_item_ipv4))) {
3402 memcpy(actions_tx, actions,
3403 sizeof(struct rte_flow_action));
3406 rte_memcpy(actions_rx, actions,
3407 sizeof(struct rte_flow_action));
3412 rte_memcpy(actions_rx, actions,
3413 sizeof(struct rte_flow_action));
3418 /* Add set meta action and end action for the Rx flow. */
3419 tag_action = actions_rx;
3420 tag_action->type = MLX5_RTE_FLOW_ACTION_TYPE_TAG;
3422 rte_memcpy(actions_rx, actions, sizeof(struct rte_flow_action));
3424 set_tag = (void *)actions_rx;
3425 set_tag->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_RX, 0, NULL);
3426 MLX5_ASSERT(set_tag->id > REG_NONE);
3427 set_tag->data = *flow_id;
3428 tag_action->conf = set_tag;
3429 /* Create Tx item list. */
3430 rte_memcpy(actions_tx, actions, sizeof(struct rte_flow_action));
3431 addr = (void *)&pattern_tx[2];
3433 item->type = MLX5_RTE_FLOW_ITEM_TYPE_TAG;
3434 tag_item = (void *)addr;
3435 tag_item->data = *flow_id;
3436 tag_item->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_TX, 0, NULL);
3437 MLX5_ASSERT(set_tag->id > REG_NONE);
3438 item->spec = tag_item;
3439 addr += sizeof(struct mlx5_rte_flow_item_tag);
3440 tag_item = (void *)addr;
3441 tag_item->data = UINT32_MAX;
3442 tag_item->id = UINT16_MAX;
3443 item->mask = tag_item;
3444 addr += sizeof(struct mlx5_rte_flow_item_tag);
3447 item->type = RTE_FLOW_ITEM_TYPE_END;
3452 * The last stage of splitting chain, just creates the subflow
3453 * without any modification.
3456 * Pointer to Ethernet device.
3458 * Parent flow structure pointer.
3459 * @param[in, out] sub_flow
3460 * Pointer to return the created subflow, may be NULL.
3461 * @param[in] prefix_layers
3462 * Prefix subflow layers, may be 0.
3464 * Flow rule attributes.
3466 * Pattern specification (list terminated by the END pattern item).
3467 * @param[in] actions
3468 * Associated actions (list terminated by the END action).
3469 * @param[in] external
3470 * This flow rule is created by request external to PMD.
3472 * Perform verbose error reporting if not NULL.
3474 * 0 on success, negative value otherwise
3477 flow_create_split_inner(struct rte_eth_dev *dev,
3478 struct rte_flow *flow,
3479 struct mlx5_flow **sub_flow,
3480 uint64_t prefix_layers,
3481 const struct rte_flow_attr *attr,
3482 const struct rte_flow_item items[],
3483 const struct rte_flow_action actions[],
3484 bool external, struct rte_flow_error *error)
3486 struct mlx5_flow *dev_flow;
3488 dev_flow = flow_drv_prepare(dev, flow, attr, items, actions, error);
3491 dev_flow->flow = flow;
3492 dev_flow->external = external;
3493 /* Subflow object was created, we must include one in the list. */
3494 SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
3495 dev_flow->handle, next);
3497 * If dev_flow is as one of the suffix flow, some actions in suffix
3498 * flow may need some user defined item layer flags.
3501 dev_flow->handle->layers = prefix_layers;
3503 *sub_flow = dev_flow;
3504 return flow_drv_translate(dev, dev_flow, attr, items, actions, error);
3508 * Split the meter flow.
3510 * As meter flow will split to three sub flow, other than meter
3511 * action, the other actions make sense to only meter accepts
3512 * the packet. If it need to be dropped, no other additional
3513 * actions should be take.
3515 * One kind of special action which decapsulates the L3 tunnel
3516 * header will be in the prefix sub flow, as not to take the
3517 * L3 tunnel header into account.
3520 * Pointer to Ethernet device.
3522 * Pattern specification (list terminated by the END pattern item).
3523 * @param[out] sfx_items
3524 * Suffix flow match items (list terminated by the END pattern item).
3525 * @param[in] actions
3526 * Associated actions (list terminated by the END action).
3527 * @param[out] actions_sfx
3528 * Suffix flow actions.
3529 * @param[out] actions_pre
3530 * Prefix flow actions.
3531 * @param[out] pattern_sfx
3532 * The pattern items for the suffix flow.
3533 * @param[out] tag_sfx
3534 * Pointer to suffix flow tag.
3540 flow_meter_split_prep(struct rte_eth_dev *dev,
3541 const struct rte_flow_item items[],
3542 struct rte_flow_item sfx_items[],
3543 const struct rte_flow_action actions[],
3544 struct rte_flow_action actions_sfx[],
3545 struct rte_flow_action actions_pre[])
3547 struct rte_flow_action *tag_action = NULL;
3548 struct rte_flow_item *tag_item;
3549 struct mlx5_rte_flow_action_set_tag *set_tag;
3550 struct rte_flow_error error;
3551 const struct rte_flow_action_raw_encap *raw_encap;
3552 const struct rte_flow_action_raw_decap *raw_decap;
3553 struct mlx5_rte_flow_item_tag *tag_spec;
3554 struct mlx5_rte_flow_item_tag *tag_mask;
3556 bool copy_vlan = false;
3558 /* Prepare the actions for prefix and suffix flow. */
3559 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3560 struct rte_flow_action **action_cur = NULL;
3562 switch (actions->type) {
3563 case RTE_FLOW_ACTION_TYPE_METER:
3564 /* Add the extra tag action first. */
3565 tag_action = actions_pre;
3566 tag_action->type = MLX5_RTE_FLOW_ACTION_TYPE_TAG;
3568 action_cur = &actions_pre;
3570 case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
3571 case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
3572 action_cur = &actions_pre;
3574 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3575 raw_encap = actions->conf;
3576 if (raw_encap->size < MLX5_ENCAPSULATION_DECISION_SIZE)
3577 action_cur = &actions_pre;
3579 case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
3580 raw_decap = actions->conf;
3581 if (raw_decap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
3582 action_cur = &actions_pre;
3584 case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
3585 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
3592 action_cur = &actions_sfx;
3593 memcpy(*action_cur, actions, sizeof(struct rte_flow_action));
3596 /* Add end action to the actions. */
3597 actions_sfx->type = RTE_FLOW_ACTION_TYPE_END;
3598 actions_pre->type = RTE_FLOW_ACTION_TYPE_END;
3601 set_tag = (void *)actions_pre;
3602 set_tag->id = mlx5_flow_get_reg_id(dev, MLX5_MTR_SFX, 0, &error);
3604 * Get the id from the qrss_pool to make qrss share the id with meter.
3606 tag_id = flow_qrss_get_id(dev);
3607 set_tag->data = tag_id << MLX5_MTR_COLOR_BITS;
3609 tag_action->conf = set_tag;
3610 /* Prepare the suffix subflow items. */
3611 tag_item = sfx_items++;
3612 for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
3613 int item_type = items->type;
3615 switch (item_type) {
3616 case RTE_FLOW_ITEM_TYPE_PORT_ID:
3617 memcpy(sfx_items, items, sizeof(*sfx_items));
3620 case RTE_FLOW_ITEM_TYPE_VLAN:
3622 memcpy(sfx_items, items, sizeof(*sfx_items));
3624 * Convert to internal match item, it is used
3625 * for vlan push and set vid.
3627 sfx_items->type = MLX5_RTE_FLOW_ITEM_TYPE_VLAN;
3635 sfx_items->type = RTE_FLOW_ITEM_TYPE_END;
3637 tag_spec = (struct mlx5_rte_flow_item_tag *)sfx_items;
3638 tag_spec->data = tag_id << MLX5_MTR_COLOR_BITS;
3639 tag_spec->id = mlx5_flow_get_reg_id(dev, MLX5_MTR_SFX, 0, &error);
3640 tag_mask = tag_spec + 1;
3641 tag_mask->data = 0xffffff00;
3642 tag_item->type = MLX5_RTE_FLOW_ITEM_TYPE_TAG;
3643 tag_item->spec = tag_spec;
3644 tag_item->last = NULL;
3645 tag_item->mask = tag_mask;
3650 * Split action list having QUEUE/RSS for metadata register copy.
3652 * Once Q/RSS action is detected in user's action list, the flow action
3653 * should be split in order to copy metadata registers, which will happen in
3655 * - CQE->flow_tag := reg_c[1] (MARK)
3656 * - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
3657 * The Q/RSS action will be performed on RX_ACT_TBL after passing by RX_CP_TBL.
3658 * This is because the last action of each flow must be a terminal action
3659 * (QUEUE, RSS or DROP).
3661 * Flow ID must be allocated to identify actions in the RX_ACT_TBL and it is
3662 * stored and kept in the mlx5_flow structure per each sub_flow.
3664 * The Q/RSS action is replaced with,
3665 * - SET_TAG, setting the allocated flow ID to reg_c[2].
3666 * And the following JUMP action is added at the end,
3667 * - JUMP, to RX_CP_TBL.
3669 * A flow to perform remained Q/RSS action will be created in RX_ACT_TBL by
3670 * flow_create_split_metadata() routine. The flow will look like,
3671 * - If flow ID matches (reg_c[2]), perform Q/RSS.
3674 * Pointer to Ethernet device.
3675 * @param[out] split_actions
3676 * Pointer to store split actions to jump to CP_TBL.
3677 * @param[in] actions
3678 * Pointer to the list of original flow actions.
3680 * Pointer to the Q/RSS action.
3681 * @param[in] actions_n
3682 * Number of original actions.
3684 * Perform verbose error reporting if not NULL.
3687 * non-zero unique flow_id on success, otherwise 0 and
3688 * error/rte_error are set.
3691 flow_mreg_split_qrss_prep(struct rte_eth_dev *dev,
3692 struct rte_flow_action *split_actions,
3693 const struct rte_flow_action *actions,
3694 const struct rte_flow_action *qrss,
3695 int actions_n, struct rte_flow_error *error)
3697 struct mlx5_rte_flow_action_set_tag *set_tag;
3698 struct rte_flow_action_jump *jump;
3699 const int qrss_idx = qrss - actions;
3700 uint32_t flow_id = 0;
3704 * Given actions will be split
3705 * - Replace QUEUE/RSS action with SET_TAG to set flow ID.
3706 * - Add jump to mreg CP_TBL.
3707 * As a result, there will be one more action.
3710 memcpy(split_actions, actions, sizeof(*split_actions) * actions_n);
3711 set_tag = (void *)(split_actions + actions_n);
3713 * If tag action is not set to void(it means we are not the meter
3714 * suffix flow), add the tag action. Since meter suffix flow already
3715 * has the tag added.
3717 if (split_actions[qrss_idx].type != RTE_FLOW_ACTION_TYPE_VOID) {
3719 * Allocate the new subflow ID. This one is unique within
3720 * device and not shared with representors. Otherwise,
3721 * we would have to resolve multi-thread access synch
3722 * issue. Each flow on the shared device is appended
3723 * with source vport identifier, so the resulting
3724 * flows will be unique in the shared (by master and
3725 * representors) domain even if they have coinciding
3728 flow_id = flow_qrss_get_id(dev);
3730 return rte_flow_error_set(error, ENOMEM,
3731 RTE_FLOW_ERROR_TYPE_ACTION,
3732 NULL, "can't allocate id "
3733 "for split Q/RSS subflow");
3734 /* Internal SET_TAG action to set flow ID. */
3735 *set_tag = (struct mlx5_rte_flow_action_set_tag){
3738 ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0, error);
3742 /* Construct new actions array. */
3743 /* Replace QUEUE/RSS action. */
3744 split_actions[qrss_idx] = (struct rte_flow_action){
3745 .type = MLX5_RTE_FLOW_ACTION_TYPE_TAG,
3749 /* JUMP action to jump to mreg copy table (CP_TBL). */
3750 jump = (void *)(set_tag + 1);
3751 *jump = (struct rte_flow_action_jump){
3752 .group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
3754 split_actions[actions_n - 2] = (struct rte_flow_action){
3755 .type = RTE_FLOW_ACTION_TYPE_JUMP,
3758 split_actions[actions_n - 1] = (struct rte_flow_action){
3759 .type = RTE_FLOW_ACTION_TYPE_END,
3765 * Extend the given action list for Tx metadata copy.
3767 * Copy the given action list to the ext_actions and add flow metadata register
3768 * copy action in order to copy reg_a set by WQE to reg_c[0].
3770 * @param[out] ext_actions
3771 * Pointer to the extended action list.
3772 * @param[in] actions
3773 * Pointer to the list of actions.
3774 * @param[in] actions_n
3775 * Number of actions in the list.
3777 * Perform verbose error reporting if not NULL.
3778 * @param[in] encap_idx
3779 * The encap action inndex.
3782 * 0 on success, negative value otherwise
3785 flow_mreg_tx_copy_prep(struct rte_eth_dev *dev,
3786 struct rte_flow_action *ext_actions,
3787 const struct rte_flow_action *actions,
3788 int actions_n, struct rte_flow_error *error,
3791 struct mlx5_flow_action_copy_mreg *cp_mreg =
3792 (struct mlx5_flow_action_copy_mreg *)
3793 (ext_actions + actions_n + 1);
3796 ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
3800 ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_TX, 0, error);
3805 memcpy(ext_actions, actions, sizeof(*ext_actions) * encap_idx);
3806 if (encap_idx == actions_n - 1) {
3807 ext_actions[actions_n - 1] = (struct rte_flow_action){
3808 .type = MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3811 ext_actions[actions_n] = (struct rte_flow_action){
3812 .type = RTE_FLOW_ACTION_TYPE_END,
3815 ext_actions[encap_idx] = (struct rte_flow_action){
3816 .type = MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3819 memcpy(ext_actions + encap_idx + 1, actions + encap_idx,
3820 sizeof(*ext_actions) * (actions_n - encap_idx));
3826 * The splitting for metadata feature.
3828 * - Q/RSS action on NIC Rx should be split in order to pass by
3829 * the mreg copy table (RX_CP_TBL) and then it jumps to the
3830 * action table (RX_ACT_TBL) which has the split Q/RSS action.
3832 * - All the actions on NIC Tx should have a mreg copy action to
3833 * copy reg_a from WQE to reg_c[0].
3836 * Pointer to Ethernet device.
3838 * Parent flow structure pointer.
3839 * @param[in] prefix_layers
3840 * Prefix flow layer flags.
3842 * Flow rule attributes.
3844 * Pattern specification (list terminated by the END pattern item).
3845 * @param[in] actions
3846 * Associated actions (list terminated by the END action).
3847 * @param[in] external
3848 * This flow rule is created by request external to PMD.
3850 * Perform verbose error reporting if not NULL.
3852 * 0 on success, negative value otherwise
3855 flow_create_split_metadata(struct rte_eth_dev *dev,
3856 struct rte_flow *flow,
3857 uint64_t prefix_layers,
3858 const struct rte_flow_attr *attr,
3859 const struct rte_flow_item items[],
3860 const struct rte_flow_action actions[],
3861 bool external, struct rte_flow_error *error)
3863 struct mlx5_priv *priv = dev->data->dev_private;
3864 struct mlx5_dev_config *config = &priv->config;
3865 const struct rte_flow_action *qrss = NULL;
3866 struct rte_flow_action *ext_actions = NULL;
3867 struct mlx5_flow *dev_flow = NULL;
3868 uint32_t qrss_id = 0;
3875 /* Check whether extensive metadata feature is engaged. */
3876 if (!config->dv_flow_en ||
3877 config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
3878 !mlx5_flow_ext_mreg_supported(dev))
3879 return flow_create_split_inner(dev, flow, NULL, prefix_layers,
3880 attr, items, actions, external,
3882 actions_n = flow_parse_metadata_split_actions_info(actions, &qrss,
3885 /* Exclude hairpin flows from splitting. */
3886 if (qrss->type == RTE_FLOW_ACTION_TYPE_QUEUE) {
3887 const struct rte_flow_action_queue *queue;
3890 if (mlx5_rxq_get_type(dev, queue->index) ==
3891 MLX5_RXQ_TYPE_HAIRPIN)
3893 } else if (qrss->type == RTE_FLOW_ACTION_TYPE_RSS) {
3894 const struct rte_flow_action_rss *rss;
3897 if (mlx5_rxq_get_type(dev, rss->queue[0]) ==
3898 MLX5_RXQ_TYPE_HAIRPIN)
3903 /* Check if it is in meter suffix table. */
3904 mtr_sfx = attr->group == (attr->transfer ?
3905 (MLX5_FLOW_TABLE_LEVEL_SUFFIX - 1) :
3906 MLX5_FLOW_TABLE_LEVEL_SUFFIX);
3908 * Q/RSS action on NIC Rx should be split in order to pass by
3909 * the mreg copy table (RX_CP_TBL) and then it jumps to the
3910 * action table (RX_ACT_TBL) which has the split Q/RSS action.
3912 act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
3913 sizeof(struct rte_flow_action_set_tag) +
3914 sizeof(struct rte_flow_action_jump);
3915 ext_actions = rte_zmalloc(__func__, act_size, 0);
3917 return rte_flow_error_set(error, ENOMEM,
3918 RTE_FLOW_ERROR_TYPE_ACTION,
3919 NULL, "no memory to split "
3922 * If we are the suffix flow of meter, tag already exist.
3923 * Set the tag action to void.
3926 ext_actions[qrss - actions].type =
3927 RTE_FLOW_ACTION_TYPE_VOID;
3929 ext_actions[qrss - actions].type =
3930 MLX5_RTE_FLOW_ACTION_TYPE_TAG;
3932 * Create the new actions list with removed Q/RSS action
3933 * and appended set tag and jump to register copy table
3934 * (RX_CP_TBL). We should preallocate unique tag ID here
3935 * in advance, because it is needed for set tag action.
3937 qrss_id = flow_mreg_split_qrss_prep(dev, ext_actions, actions,
3938 qrss, actions_n, error);
3939 if (!mtr_sfx && !qrss_id) {
3943 } else if (attr->egress && !attr->transfer) {
3945 * All the actions on NIC Tx should have a metadata register
3946 * copy action to copy reg_a from WQE to reg_c[meta]
3948 act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
3949 sizeof(struct mlx5_flow_action_copy_mreg);
3950 ext_actions = rte_zmalloc(__func__, act_size, 0);
3952 return rte_flow_error_set(error, ENOMEM,
3953 RTE_FLOW_ERROR_TYPE_ACTION,
3954 NULL, "no memory to split "
3956 /* Create the action list appended with copy register. */
3957 ret = flow_mreg_tx_copy_prep(dev, ext_actions, actions,
3958 actions_n, error, encap_idx);
3962 /* Add the unmodified original or prefix subflow. */
3963 ret = flow_create_split_inner(dev, flow, &dev_flow, prefix_layers, attr,
3964 items, ext_actions ? ext_actions :
3965 actions, external, error);
3968 MLX5_ASSERT(dev_flow);
3970 const struct rte_flow_attr q_attr = {
3971 .group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
3974 /* Internal PMD action to set register. */
3975 struct mlx5_rte_flow_item_tag q_tag_spec = {
3979 struct rte_flow_item q_items[] = {
3981 .type = MLX5_RTE_FLOW_ITEM_TYPE_TAG,
3982 .spec = &q_tag_spec,
3987 .type = RTE_FLOW_ITEM_TYPE_END,
3990 struct rte_flow_action q_actions[] = {
3996 .type = RTE_FLOW_ACTION_TYPE_END,
3999 uint64_t layers = flow_get_prefix_layer_flags(dev_flow);
4002 * Configure the tag item only if there is no meter subflow.
4003 * Since tag is already marked in the meter suffix subflow
4004 * we can just use the meter suffix items as is.
4007 /* Not meter subflow. */
4008 MLX5_ASSERT(!mtr_sfx);
4010 * Put unique id in prefix flow due to it is destroyed
4011 * after suffix flow and id will be freed after there
4012 * is no actual flows with this id and identifier
4013 * reallocation becomes possible (for example, for
4014 * other flows in other threads).
4016 dev_flow->handle->split_flow_id = qrss_id;
4017 ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0,
4021 q_tag_spec.id = ret;
4024 /* Add suffix subflow to execute Q/RSS. */
4025 ret = flow_create_split_inner(dev, flow, &dev_flow, layers,
4026 &q_attr, mtr_sfx ? items :
4031 /* qrss ID should be freed if failed. */
4033 MLX5_ASSERT(dev_flow);
4038 * We do not destroy the partially created sub_flows in case of error.
4039 * These ones are included into parent flow list and will be destroyed
4040 * by flow_drv_destroy.
4042 flow_qrss_free_id(dev, qrss_id);
4043 rte_free(ext_actions);
4048 * The splitting for meter feature.
4050 * - The meter flow will be split to two flows as prefix and
4051 * suffix flow. The packets make sense only it pass the prefix
4054 * - Reg_C_5 is used for the packet to match betweend prefix and
4058 * Pointer to Ethernet device.
4060 * Parent flow structure pointer.
4062 * Flow rule attributes.
4064 * Pattern specification (list terminated by the END pattern item).
4065 * @param[in] actions
4066 * Associated actions (list terminated by the END action).
4067 * @param[in] external
4068 * This flow rule is created by request external to PMD.
4070 * Perform verbose error reporting if not NULL.
4072 * 0 on success, negative value otherwise
4075 flow_create_split_meter(struct rte_eth_dev *dev,
4076 struct rte_flow *flow,
4077 const struct rte_flow_attr *attr,
4078 const struct rte_flow_item items[],
4079 const struct rte_flow_action actions[],
4080 bool external, struct rte_flow_error *error)
4082 struct mlx5_priv *priv = dev->data->dev_private;
4083 struct rte_flow_action *sfx_actions = NULL;
4084 struct rte_flow_action *pre_actions = NULL;
4085 struct rte_flow_item *sfx_items = NULL;
4086 struct mlx5_flow *dev_flow = NULL;
4087 struct rte_flow_attr sfx_attr = *attr;
4089 uint32_t mtr_tag_id = 0;
4096 actions_n = flow_check_meter_action(actions, &mtr);
4098 /* The five prefix actions: meter, decap, encap, tag, end. */
4099 act_size = sizeof(struct rte_flow_action) * (actions_n + 5) +
4100 sizeof(struct mlx5_rte_flow_action_set_tag);
4101 /* tag, vlan, port id, end. */
4102 #define METER_SUFFIX_ITEM 4
4103 item_size = sizeof(struct rte_flow_item) * METER_SUFFIX_ITEM +
4104 sizeof(struct mlx5_rte_flow_item_tag) * 2;
4105 sfx_actions = rte_zmalloc(__func__, (act_size + item_size), 0);
4107 return rte_flow_error_set(error, ENOMEM,
4108 RTE_FLOW_ERROR_TYPE_ACTION,
4109 NULL, "no memory to split "
4111 sfx_items = (struct rte_flow_item *)((char *)sfx_actions +
4113 pre_actions = sfx_actions + actions_n;
4114 mtr_tag_id = flow_meter_split_prep(dev, items, sfx_items,
4115 actions, sfx_actions,
4121 /* Add the prefix subflow. */
4122 ret = flow_create_split_inner(dev, flow, &dev_flow, 0, attr,
4123 items, pre_actions, external,
4129 dev_flow->handle->split_flow_id = mtr_tag_id;
4130 /* Setting the sfx group atrr. */
4131 sfx_attr.group = sfx_attr.transfer ?
4132 (MLX5_FLOW_TABLE_LEVEL_SUFFIX - 1) :
4133 MLX5_FLOW_TABLE_LEVEL_SUFFIX;
4135 /* Add the prefix subflow. */
4136 ret = flow_create_split_metadata(dev, flow, dev_flow ?
4137 flow_get_prefix_layer_flags(dev_flow) :
4139 sfx_items ? sfx_items : items,
4140 sfx_actions ? sfx_actions : actions,
4144 rte_free(sfx_actions);
4149 * Split the flow to subflow set. The splitters might be linked
4150 * in the chain, like this:
4151 * flow_create_split_outer() calls:
4152 * flow_create_split_meter() calls:
4153 * flow_create_split_metadata(meter_subflow_0) calls:
4154 * flow_create_split_inner(metadata_subflow_0)
4155 * flow_create_split_inner(metadata_subflow_1)
4156 * flow_create_split_inner(metadata_subflow_2)
4157 * flow_create_split_metadata(meter_subflow_1) calls:
4158 * flow_create_split_inner(metadata_subflow_0)
4159 * flow_create_split_inner(metadata_subflow_1)
4160 * flow_create_split_inner(metadata_subflow_2)
4162 * This provide flexible way to add new levels of flow splitting.
4163 * The all of successfully created subflows are included to the
4164 * parent flow dev_flow list.
4167 * Pointer to Ethernet device.
4169 * Parent flow structure pointer.
4171 * Flow rule attributes.
4173 * Pattern specification (list terminated by the END pattern item).
4174 * @param[in] actions
4175 * Associated actions (list terminated by the END action).
4176 * @param[in] external
4177 * This flow rule is created by request external to PMD.
4179 * Perform verbose error reporting if not NULL.
4181 * 0 on success, negative value otherwise
4184 flow_create_split_outer(struct rte_eth_dev *dev,
4185 struct rte_flow *flow,
4186 const struct rte_flow_attr *attr,
4187 const struct rte_flow_item items[],
4188 const struct rte_flow_action actions[],
4189 bool external, struct rte_flow_error *error)
4193 ret = flow_create_split_meter(dev, flow, attr, items,
4194 actions, external, error);
4195 MLX5_ASSERT(ret <= 0);
4200 * Create a flow and add it to @p list.
4203 * Pointer to Ethernet device.
4205 * Pointer to a TAILQ flow list. If this parameter NULL,
4206 * no list insertion occurred, flow is just created,
4207 * this is caller's responsibility to track the
4210 * Flow rule attributes.
4212 * Pattern specification (list terminated by the END pattern item).
4213 * @param[in] actions
4214 * Associated actions (list terminated by the END action).
4215 * @param[in] external
4216 * This flow rule is created by request external to PMD.
4218 * Perform verbose error reporting if not NULL.
4221 * A flow on success, NULL otherwise and rte_errno is set.
4223 static struct rte_flow *
4224 flow_list_create(struct rte_eth_dev *dev, struct mlx5_flows *list,
4225 const struct rte_flow_attr *attr,
4226 const struct rte_flow_item items[],
4227 const struct rte_flow_action actions[],
4228 bool external, struct rte_flow_error *error)
4230 struct mlx5_priv *priv = dev->data->dev_private;
4231 struct rte_flow *flow = NULL;
4232 struct mlx5_flow *dev_flow;
4233 const struct rte_flow_action_rss *rss;
4235 struct rte_flow_expand_rss buf;
4236 uint8_t buffer[2048];
4239 struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
4240 uint8_t buffer[2048];
4243 struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
4244 uint8_t buffer[2048];
4245 } actions_hairpin_tx;
4247 struct rte_flow_item items[MLX5_MAX_SPLIT_ITEMS];
4248 uint8_t buffer[2048];
4250 struct rte_flow_expand_rss *buf = &expand_buffer.buf;
4251 struct mlx5_flow_rss_desc *rss_desc = &((struct mlx5_flow_rss_desc *)
4252 priv->rss_desc)[!!priv->flow_idx];
4253 const struct rte_flow_action *p_actions_rx = actions;
4255 int hairpin_flow = 0;
4256 uint32_t hairpin_id = 0;
4257 struct rte_flow_attr attr_tx = { .priority = 0 };
4258 int ret = flow_drv_validate(dev, attr, items, p_actions_rx, external,
4263 hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
4264 if (hairpin_flow > 0) {
4265 if (hairpin_flow > MLX5_MAX_SPLIT_ACTIONS) {
4269 flow_hairpin_split(dev, actions, actions_rx.actions,
4270 actions_hairpin_tx.actions, items_tx.items,
4272 p_actions_rx = actions_rx.actions;
4274 flow = rte_calloc(__func__, 1, sizeof(struct rte_flow), 0);
4277 goto error_before_flow;
4279 flow->drv_type = flow_get_drv_type(dev, attr);
4280 if (hairpin_id != 0)
4281 flow->hairpin_flow_id = hairpin_id;
4282 MLX5_ASSERT(flow->drv_type > MLX5_FLOW_TYPE_MIN &&
4283 flow->drv_type < MLX5_FLOW_TYPE_MAX);
4284 memset(rss_desc, 0, sizeof(*rss_desc));
4285 rss = flow_get_rss_action(p_actions_rx);
4288 * The following information is required by
4289 * mlx5_flow_hashfields_adjust() in advance.
4291 rss_desc->level = rss->level;
4292 /* RSS type 0 indicates default RSS type (ETH_RSS_IP). */
4293 rss_desc->types = !rss->types ? ETH_RSS_IP : rss->types;
4295 flow->dev_handles = 0;
4296 if (rss && rss->types) {
4297 unsigned int graph_root;
4299 graph_root = find_graph_root(items, rss->level);
4300 ret = rte_flow_expand_rss(buf, sizeof(expand_buffer.buffer),
4302 mlx5_support_expansion,
4304 MLX5_ASSERT(ret > 0 &&
4305 (unsigned int)ret < sizeof(expand_buffer.buffer));
4308 buf->entry[0].pattern = (void *)(uintptr_t)items;
4311 * Record the start index when there is a nested call. All sub-flows
4312 * need to be translated before another calling.
4313 * No need to use ping-pong buffer to save memory here.
4315 if (priv->flow_idx) {
4316 MLX5_ASSERT(!priv->flow_nested_idx);
4317 priv->flow_nested_idx = priv->flow_idx;
4319 for (i = 0; i < buf->entries; ++i) {
4321 * The splitter may create multiple dev_flows,
4322 * depending on configuration. In the simplest
4323 * case it just creates unmodified original flow.
4325 ret = flow_create_split_outer(dev, flow, attr,
4326 buf->entry[i].pattern,
4327 p_actions_rx, external,
4332 /* Create the tx flow. */
4334 attr_tx.group = MLX5_HAIRPIN_TX_TABLE;
4335 attr_tx.ingress = 0;
4337 dev_flow = flow_drv_prepare(dev, flow, &attr_tx, items_tx.items,
4338 actions_hairpin_tx.actions, error);
4341 dev_flow->flow = flow;
4342 dev_flow->external = 0;
4343 SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
4344 dev_flow->handle, next);
4345 ret = flow_drv_translate(dev, dev_flow, &attr_tx,
4347 actions_hairpin_tx.actions, error);
4352 * Update the metadata register copy table. If extensive
4353 * metadata feature is enabled and registers are supported
4354 * we might create the extra rte_flow for each unique
4355 * MARK/FLAG action ID.
4357 * The table is updated for ingress Flows only, because
4358 * the egress Flows belong to the different device and
4359 * copy table should be updated in peer NIC Rx domain.
4361 if (attr->ingress &&
4362 (external || attr->group != MLX5_FLOW_MREG_CP_TABLE_GROUP)) {
4363 ret = flow_mreg_update_copy_table(dev, flow, actions, error);
4368 * If the flow is external (from application) OR device is started, then
4369 * the flow will be applied immediately.
4371 if (external || dev->data->dev_started) {
4372 ret = flow_drv_apply(dev, flow, error);
4377 TAILQ_INSERT_TAIL(list, flow, next);
4378 flow_rxq_flags_set(dev, flow);
4379 /* Nested flow creation index recovery. */
4380 priv->flow_idx = priv->flow_nested_idx;
4381 if (priv->flow_nested_idx)
4382 priv->flow_nested_idx = 0;
4386 ret = rte_errno; /* Save rte_errno before cleanup. */
4387 flow_mreg_del_copy_action(dev, flow);
4388 flow_drv_destroy(dev, flow);
4390 rte_errno = ret; /* Restore rte_errno. */
4394 mlx5_flow_id_release(priv->sh->flow_id_pool,
4397 priv->flow_idx = priv->flow_nested_idx;
4398 if (priv->flow_nested_idx)
4399 priv->flow_nested_idx = 0;
4404 * Create a dedicated flow rule on e-switch table 0 (root table), to direct all
4405 * incoming packets to table 1.
4407 * Other flow rules, requested for group n, will be created in
4408 * e-switch table n+1.
4409 * Jump action to e-switch group n will be created to group n+1.
4411 * Used when working in switchdev mode, to utilise advantages of table 1
4415 * Pointer to Ethernet device.
4418 * Pointer to flow on success, NULL otherwise and rte_errno is set.
4421 mlx5_flow_create_esw_table_zero_flow(struct rte_eth_dev *dev)
4423 const struct rte_flow_attr attr = {
4430 const struct rte_flow_item pattern = {
4431 .type = RTE_FLOW_ITEM_TYPE_END,
4433 struct rte_flow_action_jump jump = {
4436 const struct rte_flow_action actions[] = {
4438 .type = RTE_FLOW_ACTION_TYPE_JUMP,
4442 .type = RTE_FLOW_ACTION_TYPE_END,
4445 struct mlx5_priv *priv = dev->data->dev_private;
4446 struct rte_flow_error error;
4448 return flow_list_create(dev, &priv->ctrl_flows, &attr, &pattern,
4449 actions, false, &error);
4455 * @see rte_flow_create()
4459 mlx5_flow_create(struct rte_eth_dev *dev,
4460 const struct rte_flow_attr *attr,
4461 const struct rte_flow_item items[],
4462 const struct rte_flow_action actions[],
4463 struct rte_flow_error *error)
4465 struct mlx5_priv *priv = dev->data->dev_private;
4468 * If the device is not started yet, it is not allowed to created a
4469 * flow from application. PMD default flows and traffic control flows
4472 if (unlikely(!dev->data->dev_started)) {
4474 DRV_LOG(DEBUG, "port %u is not started when "
4475 "inserting a flow", dev->data->port_id);
4478 return flow_list_create(dev, &priv->flows,
4479 attr, items, actions, true, error);
4483 * Destroy a flow in a list.
4486 * Pointer to Ethernet device.
4488 * Pointer to a TAILQ flow list. If this parameter NULL,
4489 * there is no flow removal from the list.
4494 flow_list_destroy(struct rte_eth_dev *dev, struct mlx5_flows *list,
4495 struct rte_flow *flow)
4497 struct mlx5_priv *priv = dev->data->dev_private;
4498 struct mlx5_fdir_flow *priv_fdir_flow = NULL;
4501 * Update RX queue flags only if port is started, otherwise it is
4504 if (dev->data->dev_started)
4505 flow_rxq_flags_trim(dev, flow);
4506 if (flow->hairpin_flow_id)
4507 mlx5_flow_id_release(priv->sh->flow_id_pool,
4508 flow->hairpin_flow_id);
4509 flow_drv_destroy(dev, flow);
4511 TAILQ_REMOVE(list, flow, next);
4512 flow_mreg_del_copy_action(dev, flow);
4514 LIST_FOREACH(priv_fdir_flow, &priv->fdir_flows, next) {
4515 if (priv_fdir_flow->flow == flow)
4518 if (priv_fdir_flow) {
4519 LIST_REMOVE(priv_fdir_flow, next);
4520 rte_free(priv_fdir_flow->fdir);
4521 rte_free(priv_fdir_flow);
4528 * Destroy all flows.
4531 * Pointer to Ethernet device.
4533 * Pointer to a TAILQ flow list.
4535 * If flushing is called avtively.
4538 mlx5_flow_list_flush(struct rte_eth_dev *dev, struct mlx5_flows *list,
4541 uint32_t num_flushed = 0;
4543 while (!TAILQ_EMPTY(list)) {
4544 struct rte_flow *flow;
4546 flow = TAILQ_FIRST(list);
4547 flow_list_destroy(dev, list, flow);
4551 DRV_LOG(INFO, "port %u: %u flows flushed before stopping",
4552 dev->data->port_id, num_flushed);
4560 * Pointer to Ethernet device.
4562 * Pointer to a TAILQ flow list.
4565 mlx5_flow_stop(struct rte_eth_dev *dev, struct mlx5_flows *list)
4567 struct rte_flow *flow;
4569 TAILQ_FOREACH_REVERSE(flow, list, mlx5_flows, next) {
4570 flow_drv_remove(dev, flow);
4571 flow_mreg_stop_copy_action(dev, flow);
4573 flow_mreg_del_default_copy_action(dev);
4574 flow_rxq_flags_clear(dev);
4581 * Pointer to Ethernet device.
4583 * Pointer to a TAILQ flow list.
4586 * 0 on success, a negative errno value otherwise and rte_errno is set.
4589 mlx5_flow_start(struct rte_eth_dev *dev, struct mlx5_flows *list)
4591 struct rte_flow *flow;
4592 struct rte_flow_error error;
4595 /* Make sure default copy action (reg_c[0] -> reg_b) is created. */
4596 ret = flow_mreg_add_default_copy_action(dev, &error);
4599 /* Apply Flows created by application. */
4600 TAILQ_FOREACH(flow, list, next) {
4601 ret = flow_mreg_start_copy_action(dev, flow);
4604 ret = flow_drv_apply(dev, flow, &error);
4607 flow_rxq_flags_set(dev, flow);
4611 ret = rte_errno; /* Save rte_errno before cleanup. */
4612 mlx5_flow_stop(dev, list);
4613 rte_errno = ret; /* Restore rte_errno. */
4618 * Stop all default actions for flows.
4621 * Pointer to Ethernet device.
4624 mlx5_flow_stop_default(struct rte_eth_dev *dev)
4626 flow_mreg_del_default_copy_action(dev);
4630 * Start all default actions for flows.
4633 * Pointer to Ethernet device.
4635 * 0 on success, a negative errno value otherwise and rte_errno is set.
4638 mlx5_flow_start_default(struct rte_eth_dev *dev)
4640 struct rte_flow_error error;
4642 /* Make sure default copy action (reg_c[0] -> reg_b) is created. */
4643 return flow_mreg_add_default_copy_action(dev, &error);
4647 * Allocate intermediate resources for flow creation.
4650 * Pointer to Ethernet device.
4653 mlx5_flow_alloc_intermediate(struct rte_eth_dev *dev)
4655 struct mlx5_priv *priv = dev->data->dev_private;
4657 if (!priv->inter_flows) {
4658 priv->inter_flows = rte_calloc(__func__, 1,
4659 MLX5_NUM_MAX_DEV_FLOWS *
4660 sizeof(struct mlx5_flow) +
4661 (sizeof(struct mlx5_flow_rss_desc) +
4662 sizeof(uint16_t) * UINT16_MAX) * 2, 0);
4663 if (!priv->inter_flows) {
4664 DRV_LOG(ERR, "can't allocate intermediate memory.");
4668 priv->rss_desc = &((struct mlx5_flow *)priv->inter_flows)
4669 [MLX5_NUM_MAX_DEV_FLOWS];
4670 /* Reset the index. */
4672 priv->flow_nested_idx = 0;
4676 * Free intermediate resources for flows.
4679 * Pointer to Ethernet device.
4682 mlx5_flow_free_intermediate(struct rte_eth_dev *dev)
4684 struct mlx5_priv *priv = dev->data->dev_private;
4686 rte_free(priv->inter_flows);
4687 priv->inter_flows = NULL;
4691 * Verify the flow list is empty
4694 * Pointer to Ethernet device.
4696 * @return the number of flows not released.
4699 mlx5_flow_verify(struct rte_eth_dev *dev)
4701 struct mlx5_priv *priv = dev->data->dev_private;
4702 struct rte_flow *flow;
4705 TAILQ_FOREACH(flow, &priv->flows, next) {
4706 DRV_LOG(DEBUG, "port %u flow %p still referenced",
4707 dev->data->port_id, (void *)flow);
4714 * Enable default hairpin egress flow.
4717 * Pointer to Ethernet device.
4722 * 0 on success, a negative errno value otherwise and rte_errno is set.
4725 mlx5_ctrl_flow_source_queue(struct rte_eth_dev *dev,
4728 struct mlx5_priv *priv = dev->data->dev_private;
4729 const struct rte_flow_attr attr = {
4733 struct mlx5_rte_flow_item_tx_queue queue_spec = {
4736 struct mlx5_rte_flow_item_tx_queue queue_mask = {
4737 .queue = UINT32_MAX,
4739 struct rte_flow_item items[] = {
4741 .type = MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE,
4742 .spec = &queue_spec,
4744 .mask = &queue_mask,
4747 .type = RTE_FLOW_ITEM_TYPE_END,
4750 struct rte_flow_action_jump jump = {
4751 .group = MLX5_HAIRPIN_TX_TABLE,
4753 struct rte_flow_action actions[2];
4754 struct rte_flow *flow;
4755 struct rte_flow_error error;
4757 actions[0].type = RTE_FLOW_ACTION_TYPE_JUMP;
4758 actions[0].conf = &jump;
4759 actions[1].type = RTE_FLOW_ACTION_TYPE_END;
4760 flow = flow_list_create(dev, &priv->ctrl_flows,
4761 &attr, items, actions, false, &error);
4764 "Failed to create ctrl flow: rte_errno(%d),"
4765 " type(%d), message(%s)",
4766 rte_errno, error.type,
4767 error.message ? error.message : " (no stated reason)");
4774 * Enable a control flow configured from the control plane.
4777 * Pointer to Ethernet device.
4779 * An Ethernet flow spec to apply.
4781 * An Ethernet flow mask to apply.
4783 * A VLAN flow spec to apply.
4785 * A VLAN flow mask to apply.
4788 * 0 on success, a negative errno value otherwise and rte_errno is set.
4791 mlx5_ctrl_flow_vlan(struct rte_eth_dev *dev,
4792 struct rte_flow_item_eth *eth_spec,
4793 struct rte_flow_item_eth *eth_mask,
4794 struct rte_flow_item_vlan *vlan_spec,
4795 struct rte_flow_item_vlan *vlan_mask)
4797 struct mlx5_priv *priv = dev->data->dev_private;
4798 const struct rte_flow_attr attr = {
4800 .priority = MLX5_FLOW_PRIO_RSVD,
4802 struct rte_flow_item items[] = {
4804 .type = RTE_FLOW_ITEM_TYPE_ETH,
4810 .type = (vlan_spec) ? RTE_FLOW_ITEM_TYPE_VLAN :
4811 RTE_FLOW_ITEM_TYPE_END,
4817 .type = RTE_FLOW_ITEM_TYPE_END,
4820 uint16_t queue[priv->reta_idx_n];
4821 struct rte_flow_action_rss action_rss = {
4822 .func = RTE_ETH_HASH_FUNCTION_DEFAULT,
4824 .types = priv->rss_conf.rss_hf,
4825 .key_len = priv->rss_conf.rss_key_len,
4826 .queue_num = priv->reta_idx_n,
4827 .key = priv->rss_conf.rss_key,
4830 struct rte_flow_action actions[] = {
4832 .type = RTE_FLOW_ACTION_TYPE_RSS,
4833 .conf = &action_rss,
4836 .type = RTE_FLOW_ACTION_TYPE_END,
4839 struct rte_flow *flow;
4840 struct rte_flow_error error;
4843 if (!priv->reta_idx_n || !priv->rxqs_n) {
4846 for (i = 0; i != priv->reta_idx_n; ++i)
4847 queue[i] = (*priv->reta_idx)[i];
4848 flow = flow_list_create(dev, &priv->ctrl_flows,
4849 &attr, items, actions, false, &error);
4856 * Enable a flow control configured from the control plane.
4859 * Pointer to Ethernet device.
4861 * An Ethernet flow spec to apply.
4863 * An Ethernet flow mask to apply.
4866 * 0 on success, a negative errno value otherwise and rte_errno is set.
4869 mlx5_ctrl_flow(struct rte_eth_dev *dev,
4870 struct rte_flow_item_eth *eth_spec,
4871 struct rte_flow_item_eth *eth_mask)
4873 return mlx5_ctrl_flow_vlan(dev, eth_spec, eth_mask, NULL, NULL);
4879 * @see rte_flow_destroy()
4883 mlx5_flow_destroy(struct rte_eth_dev *dev,
4884 struct rte_flow *flow,
4885 struct rte_flow_error *error __rte_unused)
4887 struct mlx5_priv *priv = dev->data->dev_private;
4889 flow_list_destroy(dev, &priv->flows, flow);
4894 * Destroy all flows.
4896 * @see rte_flow_flush()
4900 mlx5_flow_flush(struct rte_eth_dev *dev,
4901 struct rte_flow_error *error __rte_unused)
4903 struct mlx5_priv *priv = dev->data->dev_private;
4905 mlx5_flow_list_flush(dev, &priv->flows, false);
4912 * @see rte_flow_isolate()
4916 mlx5_flow_isolate(struct rte_eth_dev *dev,
4918 struct rte_flow_error *error)
4920 struct mlx5_priv *priv = dev->data->dev_private;
4922 if (dev->data->dev_started) {
4923 rte_flow_error_set(error, EBUSY,
4924 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
4926 "port must be stopped first");
4929 priv->isolated = !!enable;
4931 dev->dev_ops = &mlx5_dev_ops_isolate;
4933 dev->dev_ops = &mlx5_dev_ops;
4940 * @see rte_flow_query()
4944 flow_drv_query(struct rte_eth_dev *dev,
4945 struct rte_flow *flow,
4946 const struct rte_flow_action *actions,
4948 struct rte_flow_error *error)
4950 const struct mlx5_flow_driver_ops *fops;
4951 enum mlx5_flow_drv_type ftype = flow->drv_type;
4953 MLX5_ASSERT(ftype > MLX5_FLOW_TYPE_MIN && ftype < MLX5_FLOW_TYPE_MAX);
4954 fops = flow_get_drv_ops(ftype);
4956 return fops->query(dev, flow, actions, data, error);
4962 * @see rte_flow_query()
4966 mlx5_flow_query(struct rte_eth_dev *dev,
4967 struct rte_flow *flow,
4968 const struct rte_flow_action *actions,
4970 struct rte_flow_error *error)
4974 ret = flow_drv_query(dev, flow, actions, data, error);
4981 * Convert a flow director filter to a generic flow.
4984 * Pointer to Ethernet device.
4985 * @param fdir_filter
4986 * Flow director filter to add.
4988 * Generic flow parameters structure.
4991 * 0 on success, a negative errno value otherwise and rte_errno is set.
4994 flow_fdir_filter_convert(struct rte_eth_dev *dev,
4995 const struct rte_eth_fdir_filter *fdir_filter,
4996 struct mlx5_fdir *attributes)
4998 struct mlx5_priv *priv = dev->data->dev_private;
4999 const struct rte_eth_fdir_input *input = &fdir_filter->input;
5000 const struct rte_eth_fdir_masks *mask =
5001 &dev->data->dev_conf.fdir_conf.mask;
5003 /* Validate queue number. */
5004 if (fdir_filter->action.rx_queue >= priv->rxqs_n) {
5005 DRV_LOG(ERR, "port %u invalid queue number %d",
5006 dev->data->port_id, fdir_filter->action.rx_queue);
5010 attributes->attr.ingress = 1;
5011 attributes->items[0] = (struct rte_flow_item) {
5012 .type = RTE_FLOW_ITEM_TYPE_ETH,
5013 .spec = &attributes->l2,
5014 .mask = &attributes->l2_mask,
5016 switch (fdir_filter->action.behavior) {
5017 case RTE_ETH_FDIR_ACCEPT:
5018 attributes->actions[0] = (struct rte_flow_action){
5019 .type = RTE_FLOW_ACTION_TYPE_QUEUE,
5020 .conf = &attributes->queue,
5023 case RTE_ETH_FDIR_REJECT:
5024 attributes->actions[0] = (struct rte_flow_action){
5025 .type = RTE_FLOW_ACTION_TYPE_DROP,
5029 DRV_LOG(ERR, "port %u invalid behavior %d",
5031 fdir_filter->action.behavior);
5032 rte_errno = ENOTSUP;
5035 attributes->queue.index = fdir_filter->action.rx_queue;
5037 switch (fdir_filter->input.flow_type) {
5038 case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
5039 case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
5040 case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
5041 attributes->l3.ipv4.hdr = (struct rte_ipv4_hdr){
5042 .src_addr = input->flow.ip4_flow.src_ip,
5043 .dst_addr = input->flow.ip4_flow.dst_ip,
5044 .time_to_live = input->flow.ip4_flow.ttl,
5045 .type_of_service = input->flow.ip4_flow.tos,
5047 attributes->l3_mask.ipv4.hdr = (struct rte_ipv4_hdr){
5048 .src_addr = mask->ipv4_mask.src_ip,
5049 .dst_addr = mask->ipv4_mask.dst_ip,
5050 .time_to_live = mask->ipv4_mask.ttl,
5051 .type_of_service = mask->ipv4_mask.tos,
5052 .next_proto_id = mask->ipv4_mask.proto,
5054 attributes->items[1] = (struct rte_flow_item){
5055 .type = RTE_FLOW_ITEM_TYPE_IPV4,
5056 .spec = &attributes->l3,
5057 .mask = &attributes->l3_mask,
5060 case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
5061 case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
5062 case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
5063 attributes->l3.ipv6.hdr = (struct rte_ipv6_hdr){
5064 .hop_limits = input->flow.ipv6_flow.hop_limits,
5065 .proto = input->flow.ipv6_flow.proto,
5068 memcpy(attributes->l3.ipv6.hdr.src_addr,
5069 input->flow.ipv6_flow.src_ip,
5070 RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
5071 memcpy(attributes->l3.ipv6.hdr.dst_addr,
5072 input->flow.ipv6_flow.dst_ip,
5073 RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
5074 memcpy(attributes->l3_mask.ipv6.hdr.src_addr,
5075 mask->ipv6_mask.src_ip,
5076 RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
5077 memcpy(attributes->l3_mask.ipv6.hdr.dst_addr,
5078 mask->ipv6_mask.dst_ip,
5079 RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
5080 attributes->items[1] = (struct rte_flow_item){
5081 .type = RTE_FLOW_ITEM_TYPE_IPV6,
5082 .spec = &attributes->l3,
5083 .mask = &attributes->l3_mask,
5087 DRV_LOG(ERR, "port %u invalid flow type%d",
5088 dev->data->port_id, fdir_filter->input.flow_type);
5089 rte_errno = ENOTSUP;
5093 switch (fdir_filter->input.flow_type) {
5094 case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
5095 attributes->l4.udp.hdr = (struct rte_udp_hdr){
5096 .src_port = input->flow.udp4_flow.src_port,
5097 .dst_port = input->flow.udp4_flow.dst_port,
5099 attributes->l4_mask.udp.hdr = (struct rte_udp_hdr){
5100 .src_port = mask->src_port_mask,
5101 .dst_port = mask->dst_port_mask,
5103 attributes->items[2] = (struct rte_flow_item){
5104 .type = RTE_FLOW_ITEM_TYPE_UDP,
5105 .spec = &attributes->l4,
5106 .mask = &attributes->l4_mask,
5109 case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
5110 attributes->l4.tcp.hdr = (struct rte_tcp_hdr){
5111 .src_port = input->flow.tcp4_flow.src_port,
5112 .dst_port = input->flow.tcp4_flow.dst_port,
5114 attributes->l4_mask.tcp.hdr = (struct rte_tcp_hdr){
5115 .src_port = mask->src_port_mask,
5116 .dst_port = mask->dst_port_mask,
5118 attributes->items[2] = (struct rte_flow_item){
5119 .type = RTE_FLOW_ITEM_TYPE_TCP,
5120 .spec = &attributes->l4,
5121 .mask = &attributes->l4_mask,
5124 case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
5125 attributes->l4.udp.hdr = (struct rte_udp_hdr){
5126 .src_port = input->flow.udp6_flow.src_port,
5127 .dst_port = input->flow.udp6_flow.dst_port,
5129 attributes->l4_mask.udp.hdr = (struct rte_udp_hdr){
5130 .src_port = mask->src_port_mask,
5131 .dst_port = mask->dst_port_mask,
5133 attributes->items[2] = (struct rte_flow_item){
5134 .type = RTE_FLOW_ITEM_TYPE_UDP,
5135 .spec = &attributes->l4,
5136 .mask = &attributes->l4_mask,
5139 case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
5140 attributes->l4.tcp.hdr = (struct rte_tcp_hdr){
5141 .src_port = input->flow.tcp6_flow.src_port,
5142 .dst_port = input->flow.tcp6_flow.dst_port,
5144 attributes->l4_mask.tcp.hdr = (struct rte_tcp_hdr){
5145 .src_port = mask->src_port_mask,
5146 .dst_port = mask->dst_port_mask,
5148 attributes->items[2] = (struct rte_flow_item){
5149 .type = RTE_FLOW_ITEM_TYPE_TCP,
5150 .spec = &attributes->l4,
5151 .mask = &attributes->l4_mask,
5154 case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
5155 case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
5158 DRV_LOG(ERR, "port %u invalid flow type%d",
5159 dev->data->port_id, fdir_filter->input.flow_type);
5160 rte_errno = ENOTSUP;
5166 #define FLOW_FDIR_CMP(f1, f2, fld) \
5167 memcmp(&(f1)->fld, &(f2)->fld, sizeof(f1->fld))
5170 * Compare two FDIR flows. If items and actions are identical, the two flows are
5174 * Pointer to Ethernet device.
5176 * FDIR flow to compare.
5178 * FDIR flow to compare.
5181 * Zero on match, 1 otherwise.
5184 flow_fdir_cmp(const struct mlx5_fdir *f1, const struct mlx5_fdir *f2)
5186 if (FLOW_FDIR_CMP(f1, f2, attr) ||
5187 FLOW_FDIR_CMP(f1, f2, l2) ||
5188 FLOW_FDIR_CMP(f1, f2, l2_mask) ||
5189 FLOW_FDIR_CMP(f1, f2, l3) ||
5190 FLOW_FDIR_CMP(f1, f2, l3_mask) ||
5191 FLOW_FDIR_CMP(f1, f2, l4) ||
5192 FLOW_FDIR_CMP(f1, f2, l4_mask) ||
5193 FLOW_FDIR_CMP(f1, f2, actions[0].type))
5195 if (f1->actions[0].type == RTE_FLOW_ACTION_TYPE_QUEUE &&
5196 FLOW_FDIR_CMP(f1, f2, queue))
5202 * Search device flow list to find out a matched FDIR flow.
5205 * Pointer to Ethernet device.
5207 * FDIR flow to lookup.
5210 * Pointer of flow if found, NULL otherwise.
5212 static struct rte_flow *
5213 flow_fdir_filter_lookup(struct rte_eth_dev *dev, struct mlx5_fdir *fdir_flow)
5215 struct mlx5_priv *priv = dev->data->dev_private;
5216 struct rte_flow *flow = NULL;
5217 struct mlx5_fdir_flow *priv_fdir_flow = NULL;
5219 MLX5_ASSERT(fdir_flow);
5220 LIST_FOREACH(priv_fdir_flow, &priv->fdir_flows, next) {
5221 if (!flow_fdir_cmp(priv_fdir_flow->fdir, fdir_flow)) {
5222 DRV_LOG(DEBUG, "port %u found FDIR flow %p",
5223 dev->data->port_id, (void *)flow);
5224 flow = priv_fdir_flow->flow;
5232 * Add new flow director filter and store it in list.
5235 * Pointer to Ethernet device.
5236 * @param fdir_filter
5237 * Flow director filter to add.
5240 * 0 on success, a negative errno value otherwise and rte_errno is set.
5243 flow_fdir_filter_add(struct rte_eth_dev *dev,
5244 const struct rte_eth_fdir_filter *fdir_filter)
5246 struct mlx5_priv *priv = dev->data->dev_private;
5247 struct mlx5_fdir *fdir_flow;
5248 struct rte_flow *flow;
5249 struct mlx5_fdir_flow *priv_fdir_flow = NULL;
5252 fdir_flow = rte_zmalloc(__func__, sizeof(*fdir_flow), 0);
5257 ret = flow_fdir_filter_convert(dev, fdir_filter, fdir_flow);
5260 flow = flow_fdir_filter_lookup(dev, fdir_flow);
5265 priv_fdir_flow = rte_zmalloc(__func__, sizeof(struct mlx5_fdir_flow),
5267 if (!priv_fdir_flow) {
5271 flow = flow_list_create(dev, &priv->flows, &fdir_flow->attr,
5272 fdir_flow->items, fdir_flow->actions, true,
5277 priv_fdir_flow->fdir = fdir_flow;
5278 priv_fdir_flow->flow = flow;
5279 LIST_INSERT_HEAD(&priv->fdir_flows, priv_fdir_flow, next);
5280 DRV_LOG(DEBUG, "port %u created FDIR flow %p",
5281 dev->data->port_id, (void *)flow);
5284 rte_free(priv_fdir_flow);
5285 rte_free(fdir_flow);
5290 * Delete specific filter.
5293 * Pointer to Ethernet device.
5294 * @param fdir_filter
5295 * Filter to be deleted.
5298 * 0 on success, a negative errno value otherwise and rte_errno is set.
5301 flow_fdir_filter_delete(struct rte_eth_dev *dev,
5302 const struct rte_eth_fdir_filter *fdir_filter)
5304 struct mlx5_priv *priv = dev->data->dev_private;
5305 struct rte_flow *flow;
5306 struct mlx5_fdir fdir_flow = {
5309 struct mlx5_fdir_flow *priv_fdir_flow = NULL;
5312 ret = flow_fdir_filter_convert(dev, fdir_filter, &fdir_flow);
5315 LIST_FOREACH(priv_fdir_flow, &priv->fdir_flows, next) {
5316 /* Find the fdir in priv list */
5317 if (!flow_fdir_cmp(priv_fdir_flow->fdir, &fdir_flow))
5320 if (!priv_fdir_flow)
5322 LIST_REMOVE(priv_fdir_flow, next);
5323 flow = priv_fdir_flow->flow;
5324 /* Fdir resource will be releasd after flow destroy. */
5326 flow_list_destroy(dev, &priv->flows, flow);
5327 rte_free(priv_fdir_flow->fdir);
5328 rte_free(priv_fdir_flow);
5329 DRV_LOG(DEBUG, "port %u deleted FDIR flow %p",
5330 dev->data->port_id, (void *)flow);
5335 * Update queue for specific filter.
5338 * Pointer to Ethernet device.
5339 * @param fdir_filter
5340 * Filter to be updated.
5343 * 0 on success, a negative errno value otherwise and rte_errno is set.
5346 flow_fdir_filter_update(struct rte_eth_dev *dev,
5347 const struct rte_eth_fdir_filter *fdir_filter)
5351 ret = flow_fdir_filter_delete(dev, fdir_filter);
5354 return flow_fdir_filter_add(dev, fdir_filter);
5358 * Flush all filters.
5361 * Pointer to Ethernet device.
5364 flow_fdir_filter_flush(struct rte_eth_dev *dev)
5366 struct mlx5_priv *priv = dev->data->dev_private;
5367 struct mlx5_fdir_flow *priv_fdir_flow = NULL;
5369 while (!LIST_EMPTY(&priv->fdir_flows)) {
5370 priv_fdir_flow = LIST_FIRST(&priv->fdir_flows);
5371 LIST_REMOVE(priv_fdir_flow, next);
5372 priv_fdir_flow->flow->fdir = 0;
5373 flow_list_destroy(dev, &priv->flows, priv_fdir_flow->flow);
5374 rte_free(priv_fdir_flow->fdir);
5375 rte_free(priv_fdir_flow);
5380 * Get flow director information.
5383 * Pointer to Ethernet device.
5384 * @param[out] fdir_info
5385 * Resulting flow director information.
5388 flow_fdir_info_get(struct rte_eth_dev *dev, struct rte_eth_fdir_info *fdir_info)
5390 struct rte_eth_fdir_masks *mask =
5391 &dev->data->dev_conf.fdir_conf.mask;
5393 fdir_info->mode = dev->data->dev_conf.fdir_conf.mode;
5394 fdir_info->guarant_spc = 0;
5395 rte_memcpy(&fdir_info->mask, mask, sizeof(fdir_info->mask));
5396 fdir_info->max_flexpayload = 0;
5397 fdir_info->flow_types_mask[0] = 0;
5398 fdir_info->flex_payload_unit = 0;
5399 fdir_info->max_flex_payload_segment_num = 0;
5400 fdir_info->flex_payload_limit = 0;
5401 memset(&fdir_info->flex_conf, 0, sizeof(fdir_info->flex_conf));
5405 * Deal with flow director operations.
5408 * Pointer to Ethernet device.
5410 * Operation to perform.
5412 * Pointer to operation-specific structure.
5415 * 0 on success, a negative errno value otherwise and rte_errno is set.
5418 flow_fdir_ctrl_func(struct rte_eth_dev *dev, enum rte_filter_op filter_op,
5421 enum rte_fdir_mode fdir_mode =
5422 dev->data->dev_conf.fdir_conf.mode;
5424 if (filter_op == RTE_ETH_FILTER_NOP)
5426 if (fdir_mode != RTE_FDIR_MODE_PERFECT &&
5427 fdir_mode != RTE_FDIR_MODE_PERFECT_MAC_VLAN) {
5428 DRV_LOG(ERR, "port %u flow director mode %d not supported",
5429 dev->data->port_id, fdir_mode);
5433 switch (filter_op) {
5434 case RTE_ETH_FILTER_ADD:
5435 return flow_fdir_filter_add(dev, arg);
5436 case RTE_ETH_FILTER_UPDATE:
5437 return flow_fdir_filter_update(dev, arg);
5438 case RTE_ETH_FILTER_DELETE:
5439 return flow_fdir_filter_delete(dev, arg);
5440 case RTE_ETH_FILTER_FLUSH:
5441 flow_fdir_filter_flush(dev);
5443 case RTE_ETH_FILTER_INFO:
5444 flow_fdir_info_get(dev, arg);
5447 DRV_LOG(DEBUG, "port %u unknown operation %u",
5448 dev->data->port_id, filter_op);
5456 * Manage filter operations.
5459 * Pointer to Ethernet device structure.
5460 * @param filter_type
5463 * Operation to perform.
5465 * Pointer to operation-specific structure.
5468 * 0 on success, a negative errno value otherwise and rte_errno is set.
5471 mlx5_dev_filter_ctrl(struct rte_eth_dev *dev,
5472 enum rte_filter_type filter_type,
5473 enum rte_filter_op filter_op,
5476 switch (filter_type) {
5477 case RTE_ETH_FILTER_GENERIC:
5478 if (filter_op != RTE_ETH_FILTER_GET) {
5482 *(const void **)arg = &mlx5_flow_ops;
5484 case RTE_ETH_FILTER_FDIR:
5485 return flow_fdir_ctrl_func(dev, filter_op, arg);
5487 DRV_LOG(ERR, "port %u filter type (%d) not supported",
5488 dev->data->port_id, filter_type);
5489 rte_errno = ENOTSUP;
5496 * Create the needed meter and suffix tables.
5499 * Pointer to Ethernet device.
5501 * Pointer to the flow meter.
5504 * Pointer to table set on success, NULL otherwise.
5506 struct mlx5_meter_domains_infos *
5507 mlx5_flow_create_mtr_tbls(struct rte_eth_dev *dev,
5508 const struct mlx5_flow_meter *fm)
5510 const struct mlx5_flow_driver_ops *fops;
5512 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5513 return fops->create_mtr_tbls(dev, fm);
5517 * Destroy the meter table set.
5520 * Pointer to Ethernet device.
5522 * Pointer to the meter table set.
5528 mlx5_flow_destroy_mtr_tbls(struct rte_eth_dev *dev,
5529 struct mlx5_meter_domains_infos *tbls)
5531 const struct mlx5_flow_driver_ops *fops;
5533 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5534 return fops->destroy_mtr_tbls(dev, tbls);
5538 * Create policer rules.
5541 * Pointer to Ethernet device.
5543 * Pointer to flow meter structure.
5545 * Pointer to flow attributes.
5548 * 0 on success, -1 otherwise.
5551 mlx5_flow_create_policer_rules(struct rte_eth_dev *dev,
5552 struct mlx5_flow_meter *fm,
5553 const struct rte_flow_attr *attr)
5555 const struct mlx5_flow_driver_ops *fops;
5557 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5558 return fops->create_policer_rules(dev, fm, attr);
5562 * Destroy policer rules.
5565 * Pointer to flow meter structure.
5567 * Pointer to flow attributes.
5570 * 0 on success, -1 otherwise.
5573 mlx5_flow_destroy_policer_rules(struct rte_eth_dev *dev,
5574 struct mlx5_flow_meter *fm,
5575 const struct rte_flow_attr *attr)
5577 const struct mlx5_flow_driver_ops *fops;
5579 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5580 return fops->destroy_policer_rules(dev, fm, attr);
5584 * Allocate a counter.
5587 * Pointer to Ethernet device structure.
5590 * Index to allocated counter on success, 0 otherwise.
5593 mlx5_counter_alloc(struct rte_eth_dev *dev)
5595 const struct mlx5_flow_driver_ops *fops;
5596 struct rte_flow_attr attr = { .transfer = 0 };
5598 if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
5599 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5600 return fops->counter_alloc(dev);
5603 "port %u counter allocate is not supported.",
5604 dev->data->port_id);
5612 * Pointer to Ethernet device structure.
5614 * Index to counter to be free.
5617 mlx5_counter_free(struct rte_eth_dev *dev, uint32_t cnt)
5619 const struct mlx5_flow_driver_ops *fops;
5620 struct rte_flow_attr attr = { .transfer = 0 };
5622 if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
5623 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5624 fops->counter_free(dev, cnt);
5628 "port %u counter free is not supported.",
5629 dev->data->port_id);
5633 * Query counter statistics.
5636 * Pointer to Ethernet device structure.
5638 * Index to counter to query.
5640 * Set to clear counter statistics.
5642 * The counter hits packets number to save.
5644 * The counter hits bytes number to save.
5647 * 0 on success, a negative errno value otherwise.
5650 mlx5_counter_query(struct rte_eth_dev *dev, uint32_t cnt,
5651 bool clear, uint64_t *pkts, uint64_t *bytes)
5653 const struct mlx5_flow_driver_ops *fops;
5654 struct rte_flow_attr attr = { .transfer = 0 };
5656 if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
5657 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5658 return fops->counter_query(dev, cnt, clear, pkts, bytes);
5661 "port %u counter query is not supported.",
5662 dev->data->port_id);
5666 #define MLX5_POOL_QUERY_FREQ_US 1000000
5669 * Set the periodic procedure for triggering asynchronous batch queries for all
5670 * the counter pools.
5673 * Pointer to mlx5_ibv_shared object.
5676 mlx5_set_query_alarm(struct mlx5_ibv_shared *sh)
5678 struct mlx5_pools_container *cont = MLX5_CNT_CONTAINER(sh, 0, 0);
5679 uint32_t pools_n = rte_atomic16_read(&cont->n_valid);
5682 cont = MLX5_CNT_CONTAINER(sh, 1, 0);
5683 pools_n += rte_atomic16_read(&cont->n_valid);
5684 us = MLX5_POOL_QUERY_FREQ_US / pools_n;
5685 DRV_LOG(DEBUG, "Set alarm for %u pools each %u us", pools_n, us);
5686 if (rte_eal_alarm_set(us, mlx5_flow_query_alarm, sh)) {
5687 sh->cmng.query_thread_on = 0;
5688 DRV_LOG(ERR, "Cannot reinitialize query alarm");
5690 sh->cmng.query_thread_on = 1;
5695 * The periodic procedure for triggering asynchronous batch queries for all the
5696 * counter pools. This function is probably called by the host thread.
5699 * The parameter for the alarm process.
5702 mlx5_flow_query_alarm(void *arg)
5704 struct mlx5_ibv_shared *sh = arg;
5705 struct mlx5_devx_obj *dcs;
5708 uint8_t batch = sh->cmng.batch;
5709 uint16_t pool_index = sh->cmng.pool_index;
5710 struct mlx5_pools_container *cont;
5711 struct mlx5_pools_container *mcont;
5712 struct mlx5_flow_counter_pool *pool;
5714 if (sh->cmng.pending_queries >= MLX5_MAX_PENDING_QUERIES)
5717 cont = MLX5_CNT_CONTAINER(sh, batch, 1);
5718 mcont = MLX5_CNT_CONTAINER(sh, batch, 0);
5719 /* Check if resize was done and need to flip a container. */
5720 if (cont != mcont) {
5722 /* Clean the old container. */
5723 rte_free(cont->pools);
5724 memset(cont, 0, sizeof(*cont));
5727 /* Flip the host container. */
5728 sh->cmng.mhi[batch] ^= (uint8_t)2;
5732 /* 2 empty containers case is unexpected. */
5733 if (unlikely(batch != sh->cmng.batch))
5737 goto next_container;
5739 pool = cont->pools[pool_index];
5741 /* There is a pool query in progress. */
5744 LIST_FIRST(&sh->cmng.free_stat_raws);
5746 /* No free counter statistics raw memory. */
5748 dcs = (struct mlx5_devx_obj *)(uintptr_t)rte_atomic64_read
5750 offset = batch ? 0 : dcs->id % MLX5_COUNTERS_PER_POOL;
5752 * Identify the counters released between query trigger and query
5753 * handle more effiecntly. The counter released in this gap period
5754 * should wait for a new round of query as the new arrived packets
5755 * will not be taken into account.
5757 rte_atomic64_add(&pool->start_query_gen, 1);
5758 ret = mlx5_devx_cmd_flow_counter_query(dcs, 0, MLX5_COUNTERS_PER_POOL -
5760 pool->raw_hw->mem_mng->dm->id,
5762 (pool->raw_hw->data + offset),
5764 (uint64_t)(uintptr_t)pool);
5766 rte_atomic64_sub(&pool->start_query_gen, 1);
5767 DRV_LOG(ERR, "Failed to trigger asynchronous query for dcs ID"
5768 " %d", pool->min_dcs->id);
5769 pool->raw_hw = NULL;
5772 pool->raw_hw->min_dcs_id = dcs->id;
5773 LIST_REMOVE(pool->raw_hw, next);
5774 sh->cmng.pending_queries++;
5776 if (pool_index >= rte_atomic16_read(&cont->n_valid)) {
5781 sh->cmng.batch = batch;
5782 sh->cmng.pool_index = pool_index;
5783 mlx5_set_query_alarm(sh);
5787 * Handler for the HW respond about ready values from an asynchronous batch
5788 * query. This function is probably called by the host thread.
5791 * The pointer to the shared IB device context.
5792 * @param[in] async_id
5793 * The Devx async ID.
5795 * The status of the completion.
5798 mlx5_flow_async_pool_query_handle(struct mlx5_ibv_shared *sh,
5799 uint64_t async_id, int status)
5801 struct mlx5_flow_counter_pool *pool =
5802 (struct mlx5_flow_counter_pool *)(uintptr_t)async_id;
5803 struct mlx5_counter_stats_raw *raw_to_free;
5805 if (unlikely(status)) {
5806 rte_atomic64_sub(&pool->start_query_gen, 1);
5807 raw_to_free = pool->raw_hw;
5809 raw_to_free = pool->raw;
5810 rte_spinlock_lock(&pool->sl);
5811 pool->raw = pool->raw_hw;
5812 rte_spinlock_unlock(&pool->sl);
5813 MLX5_ASSERT(rte_atomic64_read(&pool->end_query_gen) + 1 ==
5814 rte_atomic64_read(&pool->start_query_gen));
5815 rte_atomic64_set(&pool->end_query_gen,
5816 rte_atomic64_read(&pool->start_query_gen));
5817 /* Be sure the new raw counters data is updated in memory. */
5820 LIST_INSERT_HEAD(&sh->cmng.free_stat_raws, raw_to_free, next);
5821 pool->raw_hw = NULL;
5822 sh->cmng.pending_queries--;
5826 * Translate the rte_flow group index to HW table value.
5828 * @param[in] attributes
5829 * Pointer to flow attributes
5830 * @param[in] external
5831 * Value is part of flow rule created by request external to PMD.
5833 * rte_flow group index value.
5834 * @param[out] fdb_def_rule
5835 * Whether fdb jump to table 1 is configured.
5839 * Pointer to error structure.
5842 * 0 on success, a negative errno value otherwise and rte_errno is set.
5845 mlx5_flow_group_to_table(const struct rte_flow_attr *attributes, bool external,
5846 uint32_t group, bool fdb_def_rule, uint32_t *table,
5847 struct rte_flow_error *error)
5849 if (attributes->transfer && external && fdb_def_rule) {
5850 if (group == UINT32_MAX)
5851 return rte_flow_error_set
5853 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
5855 "group index not supported");
5864 * Discover availability of metadata reg_c's.
5866 * Iteratively use test flows to check availability.
5869 * Pointer to the Ethernet device structure.
5872 * 0 on success, a negative errno value otherwise and rte_errno is set.
5875 mlx5_flow_discover_mreg_c(struct rte_eth_dev *dev)
5877 struct mlx5_priv *priv = dev->data->dev_private;
5878 struct mlx5_dev_config *config = &priv->config;
5879 enum modify_reg idx;
5882 /* reg_c[0] and reg_c[1] are reserved. */
5883 config->flow_mreg_c[n++] = REG_C_0;
5884 config->flow_mreg_c[n++] = REG_C_1;
5885 /* Discover availability of other reg_c's. */
5886 for (idx = REG_C_2; idx <= REG_C_7; ++idx) {
5887 struct rte_flow_attr attr = {
5888 .group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
5889 .priority = MLX5_FLOW_PRIO_RSVD,
5892 struct rte_flow_item items[] = {
5894 .type = RTE_FLOW_ITEM_TYPE_END,
5897 struct rte_flow_action actions[] = {
5899 .type = MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
5900 .conf = &(struct mlx5_flow_action_copy_mreg){
5906 .type = RTE_FLOW_ACTION_TYPE_JUMP,
5907 .conf = &(struct rte_flow_action_jump){
5908 .group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
5912 .type = RTE_FLOW_ACTION_TYPE_END,
5915 struct rte_flow *flow;
5916 struct rte_flow_error error;
5918 if (!config->dv_flow_en)
5920 /* Create internal flow, validation skips copy action. */
5921 flow = flow_list_create(dev, NULL, &attr, items,
5922 actions, false, &error);
5925 if (dev->data->dev_started || !flow_drv_apply(dev, flow, NULL))
5926 config->flow_mreg_c[n++] = idx;
5927 flow_list_destroy(dev, NULL, flow);
5929 for (; n < MLX5_MREG_C_NUM; ++n)
5930 config->flow_mreg_c[n] = REG_NONE;
5935 * Dump flow raw hw data to file
5938 * The pointer to Ethernet device.
5940 * A pointer to a file for output.
5942 * Perform verbose error reporting if not NULL. PMDs initialize this
5943 * structure in case of error only.
5945 * 0 on success, a nagative value otherwise.
5948 mlx5_flow_dev_dump(struct rte_eth_dev *dev,
5950 struct rte_flow_error *error __rte_unused)
5952 struct mlx5_priv *priv = dev->data->dev_private;
5953 struct mlx5_ibv_shared *sh = priv->sh;
5955 return mlx5_devx_cmd_flow_dump(sh->fdb_domain, sh->rx_domain,
5956 sh->tx_domain, file);