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 *dev_flow,
662 int tunnel __rte_unused, uint64_t layer_types,
663 uint64_t hash_fields)
665 struct rte_flow *flow = dev_flow->flow;
666 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
667 int rss_request_inner = flow->rss.level >= 2;
669 /* Check RSS hash level for tunnel. */
670 if (tunnel && rss_request_inner)
671 hash_fields |= IBV_RX_HASH_INNER;
672 else if (tunnel || rss_request_inner)
675 /* Check if requested layer matches RSS hash fields. */
676 if (!(flow->rss.types & layer_types))
682 * Lookup and set the ptype in the data Rx part. A single Ptype can be used,
683 * if several tunnel rules are used on this queue, the tunnel ptype will be
687 * Rx queue to update.
690 flow_rxq_tunnel_ptype_update(struct mlx5_rxq_ctrl *rxq_ctrl)
693 uint32_t tunnel_ptype = 0;
695 /* Look up for the ptype to use. */
696 for (i = 0; i != MLX5_FLOW_TUNNEL; ++i) {
697 if (!rxq_ctrl->flow_tunnels_n[i])
700 tunnel_ptype = tunnels_info[i].ptype;
706 rxq_ctrl->rxq.tunnel = tunnel_ptype;
710 * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) according to the devive
714 * Pointer to the Ethernet device structure.
716 * Pointer to flow structure.
717 * @param[in] dev_handle
718 * Pointer to device flow handle structure.
721 flow_drv_rxq_flags_set(struct rte_eth_dev *dev, struct rte_flow *flow,
722 struct mlx5_flow_handle *dev_handle)
724 struct mlx5_priv *priv = dev->data->dev_private;
725 const int mark = !!(dev_handle->act_flags &
726 (MLX5_FLOW_ACTION_FLAG | MLX5_FLOW_ACTION_MARK));
727 const int tunnel = !!(dev_handle->layers & MLX5_FLOW_LAYER_TUNNEL);
730 for (i = 0; i != flow->rss.queue_num; ++i) {
731 int idx = (*flow->rss.queue)[i];
732 struct mlx5_rxq_ctrl *rxq_ctrl =
733 container_of((*priv->rxqs)[idx],
734 struct mlx5_rxq_ctrl, rxq);
737 * To support metadata register copy on Tx loopback,
738 * this must be always enabled (metadata may arive
739 * from other port - not from local flows only.
741 if (priv->config.dv_flow_en &&
742 priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
743 mlx5_flow_ext_mreg_supported(dev)) {
744 rxq_ctrl->rxq.mark = 1;
745 rxq_ctrl->flow_mark_n = 1;
747 rxq_ctrl->rxq.mark = 1;
748 rxq_ctrl->flow_mark_n++;
753 /* Increase the counter matching the flow. */
754 for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
755 if ((tunnels_info[j].tunnel &
756 dev_handle->layers) ==
757 tunnels_info[j].tunnel) {
758 rxq_ctrl->flow_tunnels_n[j]++;
762 flow_rxq_tunnel_ptype_update(rxq_ctrl);
768 * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) for a flow
771 * Pointer to the Ethernet device structure.
773 * Pointer to flow structure.
776 flow_rxq_flags_set(struct rte_eth_dev *dev, struct rte_flow *flow)
778 struct mlx5_priv *priv = dev->data->dev_private;
780 struct mlx5_flow_handle *dev_handle;
782 SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
783 handle_idx, dev_handle, next)
784 flow_drv_rxq_flags_set(dev, flow, dev_handle);
788 * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
789 * device flow if no other flow uses it with the same kind of request.
792 * Pointer to Ethernet device.
794 * Pointer to flow structure.
795 * @param[in] dev_handle
796 * Pointer to the device flow handle structure.
799 flow_drv_rxq_flags_trim(struct rte_eth_dev *dev, struct rte_flow *flow,
800 struct mlx5_flow_handle *dev_handle)
802 struct mlx5_priv *priv = dev->data->dev_private;
803 const int mark = !!(dev_handle->act_flags &
804 (MLX5_FLOW_ACTION_FLAG | MLX5_FLOW_ACTION_MARK));
805 const int tunnel = !!(dev_handle->layers & MLX5_FLOW_LAYER_TUNNEL);
808 MLX5_ASSERT(dev->data->dev_started);
809 for (i = 0; i != flow->rss.queue_num; ++i) {
810 int idx = (*flow->rss.queue)[i];
811 struct mlx5_rxq_ctrl *rxq_ctrl =
812 container_of((*priv->rxqs)[idx],
813 struct mlx5_rxq_ctrl, rxq);
815 if (priv->config.dv_flow_en &&
816 priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
817 mlx5_flow_ext_mreg_supported(dev)) {
818 rxq_ctrl->rxq.mark = 1;
819 rxq_ctrl->flow_mark_n = 1;
821 rxq_ctrl->flow_mark_n--;
822 rxq_ctrl->rxq.mark = !!rxq_ctrl->flow_mark_n;
827 /* Decrease the counter matching the flow. */
828 for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
829 if ((tunnels_info[j].tunnel &
830 dev_handle->layers) ==
831 tunnels_info[j].tunnel) {
832 rxq_ctrl->flow_tunnels_n[j]--;
836 flow_rxq_tunnel_ptype_update(rxq_ctrl);
842 * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
843 * @p flow if no other flow uses it with the same kind of request.
846 * Pointer to Ethernet device.
848 * Pointer to the flow.
851 flow_rxq_flags_trim(struct rte_eth_dev *dev, struct rte_flow *flow)
853 struct mlx5_priv *priv = dev->data->dev_private;
855 struct mlx5_flow_handle *dev_handle;
857 SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
858 handle_idx, dev_handle, next)
859 flow_drv_rxq_flags_trim(dev, flow, dev_handle);
863 * Clear the Mark/Flag and Tunnel ptype information in all Rx queues.
866 * Pointer to Ethernet device.
869 flow_rxq_flags_clear(struct rte_eth_dev *dev)
871 struct mlx5_priv *priv = dev->data->dev_private;
874 for (i = 0; i != priv->rxqs_n; ++i) {
875 struct mlx5_rxq_ctrl *rxq_ctrl;
878 if (!(*priv->rxqs)[i])
880 rxq_ctrl = container_of((*priv->rxqs)[i],
881 struct mlx5_rxq_ctrl, rxq);
882 rxq_ctrl->flow_mark_n = 0;
883 rxq_ctrl->rxq.mark = 0;
884 for (j = 0; j != MLX5_FLOW_TUNNEL; ++j)
885 rxq_ctrl->flow_tunnels_n[j] = 0;
886 rxq_ctrl->rxq.tunnel = 0;
891 * return a pointer to the desired action in the list of actions.
894 * The list of actions to search the action in.
896 * The action to find.
899 * Pointer to the action in the list, if found. NULL otherwise.
901 const struct rte_flow_action *
902 mlx5_flow_find_action(const struct rte_flow_action *actions,
903 enum rte_flow_action_type action)
907 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++)
908 if (actions->type == action)
914 * Validate the flag action.
916 * @param[in] action_flags
917 * Bit-fields that holds the actions detected until now.
919 * Attributes of flow that includes this action.
921 * Pointer to error structure.
924 * 0 on success, a negative errno value otherwise and rte_errno is set.
927 mlx5_flow_validate_action_flag(uint64_t action_flags,
928 const struct rte_flow_attr *attr,
929 struct rte_flow_error *error)
931 if (action_flags & MLX5_FLOW_ACTION_MARK)
932 return rte_flow_error_set(error, EINVAL,
933 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
934 "can't mark and flag in same flow");
935 if (action_flags & MLX5_FLOW_ACTION_FLAG)
936 return rte_flow_error_set(error, EINVAL,
937 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
939 " actions in same flow");
941 return rte_flow_error_set(error, ENOTSUP,
942 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
943 "flag action not supported for "
949 * Validate the mark action.
952 * Pointer to the queue action.
953 * @param[in] action_flags
954 * Bit-fields that holds the actions detected until now.
956 * Attributes of flow that includes this action.
958 * Pointer to error structure.
961 * 0 on success, a negative errno value otherwise and rte_errno is set.
964 mlx5_flow_validate_action_mark(const struct rte_flow_action *action,
965 uint64_t action_flags,
966 const struct rte_flow_attr *attr,
967 struct rte_flow_error *error)
969 const struct rte_flow_action_mark *mark = action->conf;
972 return rte_flow_error_set(error, EINVAL,
973 RTE_FLOW_ERROR_TYPE_ACTION,
975 "configuration cannot be null");
976 if (mark->id >= MLX5_FLOW_MARK_MAX)
977 return rte_flow_error_set(error, EINVAL,
978 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
980 "mark id must in 0 <= id < "
981 RTE_STR(MLX5_FLOW_MARK_MAX));
982 if (action_flags & MLX5_FLOW_ACTION_FLAG)
983 return rte_flow_error_set(error, EINVAL,
984 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
985 "can't flag and mark in same flow");
986 if (action_flags & MLX5_FLOW_ACTION_MARK)
987 return rte_flow_error_set(error, EINVAL,
988 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
989 "can't have 2 mark actions in same"
992 return rte_flow_error_set(error, ENOTSUP,
993 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
994 "mark action not supported for "
1000 * Validate the drop action.
1002 * @param[in] action_flags
1003 * Bit-fields that holds the actions detected until now.
1005 * Attributes of flow that includes this action.
1007 * Pointer to error structure.
1010 * 0 on success, a negative errno value otherwise and rte_errno is set.
1013 mlx5_flow_validate_action_drop(uint64_t action_flags __rte_unused,
1014 const struct rte_flow_attr *attr,
1015 struct rte_flow_error *error)
1018 return rte_flow_error_set(error, ENOTSUP,
1019 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1020 "drop action not supported for "
1026 * Validate the queue action.
1029 * Pointer to the queue action.
1030 * @param[in] action_flags
1031 * Bit-fields that holds the actions detected until now.
1033 * Pointer to the Ethernet device structure.
1035 * Attributes of flow that includes this action.
1037 * Pointer to error structure.
1040 * 0 on success, a negative errno value otherwise and rte_errno is set.
1043 mlx5_flow_validate_action_queue(const struct rte_flow_action *action,
1044 uint64_t action_flags,
1045 struct rte_eth_dev *dev,
1046 const struct rte_flow_attr *attr,
1047 struct rte_flow_error *error)
1049 struct mlx5_priv *priv = dev->data->dev_private;
1050 const struct rte_flow_action_queue *queue = action->conf;
1052 if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1053 return rte_flow_error_set(error, EINVAL,
1054 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1055 "can't have 2 fate actions in"
1058 return rte_flow_error_set(error, EINVAL,
1059 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1060 NULL, "No Rx queues configured");
1061 if (queue->index >= priv->rxqs_n)
1062 return rte_flow_error_set(error, EINVAL,
1063 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1065 "queue index out of range");
1066 if (!(*priv->rxqs)[queue->index])
1067 return rte_flow_error_set(error, EINVAL,
1068 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1070 "queue is not configured");
1072 return rte_flow_error_set(error, ENOTSUP,
1073 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1074 "queue action not supported for "
1080 * Validate the rss action.
1083 * Pointer to the queue action.
1084 * @param[in] action_flags
1085 * Bit-fields that holds the actions detected until now.
1087 * Pointer to the Ethernet device structure.
1089 * Attributes of flow that includes this action.
1090 * @param[in] item_flags
1091 * Items that were detected.
1093 * Pointer to error structure.
1096 * 0 on success, a negative errno value otherwise and rte_errno is set.
1099 mlx5_flow_validate_action_rss(const struct rte_flow_action *action,
1100 uint64_t action_flags,
1101 struct rte_eth_dev *dev,
1102 const struct rte_flow_attr *attr,
1103 uint64_t item_flags,
1104 struct rte_flow_error *error)
1106 struct mlx5_priv *priv = dev->data->dev_private;
1107 const struct rte_flow_action_rss *rss = action->conf;
1108 int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1111 if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1112 return rte_flow_error_set(error, EINVAL,
1113 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1114 "can't have 2 fate actions"
1116 if (rss->func != RTE_ETH_HASH_FUNCTION_DEFAULT &&
1117 rss->func != RTE_ETH_HASH_FUNCTION_TOEPLITZ)
1118 return rte_flow_error_set(error, ENOTSUP,
1119 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1121 "RSS hash function not supported");
1122 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
1127 return rte_flow_error_set(error, ENOTSUP,
1128 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1130 "tunnel RSS is not supported");
1131 /* allow RSS key_len 0 in case of NULL (default) RSS key. */
1132 if (rss->key_len == 0 && rss->key != NULL)
1133 return rte_flow_error_set(error, ENOTSUP,
1134 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1136 "RSS hash key length 0");
1137 if (rss->key_len > 0 && rss->key_len < MLX5_RSS_HASH_KEY_LEN)
1138 return rte_flow_error_set(error, ENOTSUP,
1139 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1141 "RSS hash key too small");
1142 if (rss->key_len > MLX5_RSS_HASH_KEY_LEN)
1143 return rte_flow_error_set(error, ENOTSUP,
1144 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1146 "RSS hash key too large");
1147 if (rss->queue_num > priv->config.ind_table_max_size)
1148 return rte_flow_error_set(error, ENOTSUP,
1149 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1151 "number of queues too large");
1152 if (rss->types & MLX5_RSS_HF_MASK)
1153 return rte_flow_error_set(error, ENOTSUP,
1154 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1156 "some RSS protocols are not"
1158 if ((rss->types & (ETH_RSS_L3_SRC_ONLY | ETH_RSS_L3_DST_ONLY)) &&
1159 !(rss->types & ETH_RSS_IP))
1160 return rte_flow_error_set(error, EINVAL,
1161 RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1162 "L3 partial RSS requested but L3 RSS"
1163 " type not specified");
1164 if ((rss->types & (ETH_RSS_L4_SRC_ONLY | ETH_RSS_L4_DST_ONLY)) &&
1165 !(rss->types & (ETH_RSS_UDP | ETH_RSS_TCP)))
1166 return rte_flow_error_set(error, EINVAL,
1167 RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1168 "L4 partial RSS requested but L4 RSS"
1169 " type not specified");
1171 return rte_flow_error_set(error, EINVAL,
1172 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1173 NULL, "No Rx queues configured");
1174 if (!rss->queue_num)
1175 return rte_flow_error_set(error, EINVAL,
1176 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1177 NULL, "No queues configured");
1178 for (i = 0; i != rss->queue_num; ++i) {
1179 if (rss->queue[i] >= priv->rxqs_n)
1180 return rte_flow_error_set
1182 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1183 &rss->queue[i], "queue index out of range");
1184 if (!(*priv->rxqs)[rss->queue[i]])
1185 return rte_flow_error_set
1186 (error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1187 &rss->queue[i], "queue is not configured");
1190 return rte_flow_error_set(error, ENOTSUP,
1191 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1192 "rss action not supported for "
1194 if (rss->level > 1 && !tunnel)
1195 return rte_flow_error_set(error, EINVAL,
1196 RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1197 "inner RSS is not supported for "
1198 "non-tunnel flows");
1203 * Validate the count action.
1206 * Pointer to the Ethernet device structure.
1208 * Attributes of flow that includes this action.
1210 * Pointer to error structure.
1213 * 0 on success, a negative errno value otherwise and rte_errno is set.
1216 mlx5_flow_validate_action_count(struct rte_eth_dev *dev __rte_unused,
1217 const struct rte_flow_attr *attr,
1218 struct rte_flow_error *error)
1221 return rte_flow_error_set(error, ENOTSUP,
1222 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1223 "count action not supported for "
1229 * Verify the @p attributes will be correctly understood by the NIC and store
1230 * them in the @p flow if everything is correct.
1233 * Pointer to the Ethernet device structure.
1234 * @param[in] attributes
1235 * Pointer to flow attributes
1237 * Pointer to error structure.
1240 * 0 on success, a negative errno value otherwise and rte_errno is set.
1243 mlx5_flow_validate_attributes(struct rte_eth_dev *dev,
1244 const struct rte_flow_attr *attributes,
1245 struct rte_flow_error *error)
1247 struct mlx5_priv *priv = dev->data->dev_private;
1248 uint32_t priority_max = priv->config.flow_prio - 1;
1250 if (attributes->group)
1251 return rte_flow_error_set(error, ENOTSUP,
1252 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
1253 NULL, "groups is not supported");
1254 if (attributes->priority != MLX5_FLOW_PRIO_RSVD &&
1255 attributes->priority >= priority_max)
1256 return rte_flow_error_set(error, ENOTSUP,
1257 RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
1258 NULL, "priority out of range");
1259 if (attributes->egress)
1260 return rte_flow_error_set(error, ENOTSUP,
1261 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1262 "egress is not supported");
1263 if (attributes->transfer && !priv->config.dv_esw_en)
1264 return rte_flow_error_set(error, ENOTSUP,
1265 RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1266 NULL, "transfer is not supported");
1267 if (!attributes->ingress)
1268 return rte_flow_error_set(error, EINVAL,
1269 RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
1271 "ingress attribute is mandatory");
1276 * Validate ICMP6 item.
1279 * Item specification.
1280 * @param[in] item_flags
1281 * Bit-fields that holds the items detected until now.
1283 * Pointer to error structure.
1286 * 0 on success, a negative errno value otherwise and rte_errno is set.
1289 mlx5_flow_validate_item_icmp6(const struct rte_flow_item *item,
1290 uint64_t item_flags,
1291 uint8_t target_protocol,
1292 struct rte_flow_error *error)
1294 const struct rte_flow_item_icmp6 *mask = item->mask;
1295 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1296 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
1297 MLX5_FLOW_LAYER_OUTER_L3_IPV6;
1298 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1299 MLX5_FLOW_LAYER_OUTER_L4;
1302 if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMPV6)
1303 return rte_flow_error_set(error, EINVAL,
1304 RTE_FLOW_ERROR_TYPE_ITEM, item,
1305 "protocol filtering not compatible"
1306 " with ICMP6 layer");
1307 if (!(item_flags & l3m))
1308 return rte_flow_error_set(error, EINVAL,
1309 RTE_FLOW_ERROR_TYPE_ITEM, item,
1310 "IPv6 is mandatory to filter on"
1312 if (item_flags & l4m)
1313 return rte_flow_error_set(error, EINVAL,
1314 RTE_FLOW_ERROR_TYPE_ITEM, item,
1315 "multiple L4 layers not supported");
1317 mask = &rte_flow_item_icmp6_mask;
1318 ret = mlx5_flow_item_acceptable
1319 (item, (const uint8_t *)mask,
1320 (const uint8_t *)&rte_flow_item_icmp6_mask,
1321 sizeof(struct rte_flow_item_icmp6), error);
1328 * Validate ICMP item.
1331 * Item specification.
1332 * @param[in] item_flags
1333 * Bit-fields that holds the items detected until now.
1335 * Pointer to error structure.
1338 * 0 on success, a negative errno value otherwise and rte_errno is set.
1341 mlx5_flow_validate_item_icmp(const struct rte_flow_item *item,
1342 uint64_t item_flags,
1343 uint8_t target_protocol,
1344 struct rte_flow_error *error)
1346 const struct rte_flow_item_icmp *mask = item->mask;
1347 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1348 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
1349 MLX5_FLOW_LAYER_OUTER_L3_IPV4;
1350 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1351 MLX5_FLOW_LAYER_OUTER_L4;
1354 if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMP)
1355 return rte_flow_error_set(error, EINVAL,
1356 RTE_FLOW_ERROR_TYPE_ITEM, item,
1357 "protocol filtering not compatible"
1358 " with ICMP layer");
1359 if (!(item_flags & l3m))
1360 return rte_flow_error_set(error, EINVAL,
1361 RTE_FLOW_ERROR_TYPE_ITEM, item,
1362 "IPv4 is mandatory to filter"
1364 if (item_flags & l4m)
1365 return rte_flow_error_set(error, EINVAL,
1366 RTE_FLOW_ERROR_TYPE_ITEM, item,
1367 "multiple L4 layers not supported");
1369 mask = &rte_flow_item_icmp_mask;
1370 ret = mlx5_flow_item_acceptable
1371 (item, (const uint8_t *)mask,
1372 (const uint8_t *)&rte_flow_item_icmp_mask,
1373 sizeof(struct rte_flow_item_icmp), error);
1380 * Validate Ethernet item.
1383 * Item specification.
1384 * @param[in] item_flags
1385 * Bit-fields that holds the items detected until now.
1387 * Pointer to error structure.
1390 * 0 on success, a negative errno value otherwise and rte_errno is set.
1393 mlx5_flow_validate_item_eth(const struct rte_flow_item *item,
1394 uint64_t item_flags,
1395 struct rte_flow_error *error)
1397 const struct rte_flow_item_eth *mask = item->mask;
1398 const struct rte_flow_item_eth nic_mask = {
1399 .dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1400 .src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1401 .type = RTE_BE16(0xffff),
1404 int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1405 const uint64_t ethm = tunnel ? MLX5_FLOW_LAYER_INNER_L2 :
1406 MLX5_FLOW_LAYER_OUTER_L2;
1408 if (item_flags & ethm)
1409 return rte_flow_error_set(error, ENOTSUP,
1410 RTE_FLOW_ERROR_TYPE_ITEM, item,
1411 "multiple L2 layers not supported");
1412 if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_L3)) ||
1413 (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_L3)))
1414 return rte_flow_error_set(error, EINVAL,
1415 RTE_FLOW_ERROR_TYPE_ITEM, item,
1416 "L2 layer should not follow "
1418 if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_VLAN)) ||
1419 (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_VLAN)))
1420 return rte_flow_error_set(error, EINVAL,
1421 RTE_FLOW_ERROR_TYPE_ITEM, item,
1422 "L2 layer should not follow VLAN");
1424 mask = &rte_flow_item_eth_mask;
1425 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1426 (const uint8_t *)&nic_mask,
1427 sizeof(struct rte_flow_item_eth),
1433 * Validate VLAN item.
1436 * Item specification.
1437 * @param[in] item_flags
1438 * Bit-fields that holds the items detected until now.
1440 * Ethernet device flow is being created on.
1442 * Pointer to error structure.
1445 * 0 on success, a negative errno value otherwise and rte_errno is set.
1448 mlx5_flow_validate_item_vlan(const struct rte_flow_item *item,
1449 uint64_t item_flags,
1450 struct rte_eth_dev *dev,
1451 struct rte_flow_error *error)
1453 const struct rte_flow_item_vlan *spec = item->spec;
1454 const struct rte_flow_item_vlan *mask = item->mask;
1455 const struct rte_flow_item_vlan nic_mask = {
1456 .tci = RTE_BE16(UINT16_MAX),
1457 .inner_type = RTE_BE16(UINT16_MAX),
1459 uint16_t vlan_tag = 0;
1460 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1462 const uint64_t l34m = tunnel ? (MLX5_FLOW_LAYER_INNER_L3 |
1463 MLX5_FLOW_LAYER_INNER_L4) :
1464 (MLX5_FLOW_LAYER_OUTER_L3 |
1465 MLX5_FLOW_LAYER_OUTER_L4);
1466 const uint64_t vlanm = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
1467 MLX5_FLOW_LAYER_OUTER_VLAN;
1469 if (item_flags & vlanm)
1470 return rte_flow_error_set(error, EINVAL,
1471 RTE_FLOW_ERROR_TYPE_ITEM, item,
1472 "multiple VLAN layers not supported");
1473 else if ((item_flags & l34m) != 0)
1474 return rte_flow_error_set(error, EINVAL,
1475 RTE_FLOW_ERROR_TYPE_ITEM, item,
1476 "VLAN cannot follow L3/L4 layer");
1478 mask = &rte_flow_item_vlan_mask;
1479 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1480 (const uint8_t *)&nic_mask,
1481 sizeof(struct rte_flow_item_vlan),
1485 if (!tunnel && mask->tci != RTE_BE16(0x0fff)) {
1486 struct mlx5_priv *priv = dev->data->dev_private;
1488 if (priv->vmwa_context) {
1490 * Non-NULL context means we have a virtual machine
1491 * and SR-IOV enabled, we have to create VLAN interface
1492 * to make hypervisor to setup E-Switch vport
1493 * context correctly. We avoid creating the multiple
1494 * VLAN interfaces, so we cannot support VLAN tag mask.
1496 return rte_flow_error_set(error, EINVAL,
1497 RTE_FLOW_ERROR_TYPE_ITEM,
1499 "VLAN tag mask is not"
1500 " supported in virtual"
1505 vlan_tag = spec->tci;
1506 vlan_tag &= mask->tci;
1509 * From verbs perspective an empty VLAN is equivalent
1510 * to a packet without VLAN layer.
1513 return rte_flow_error_set(error, EINVAL,
1514 RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
1516 "VLAN cannot be empty");
1521 * Validate IPV4 item.
1524 * Item specification.
1525 * @param[in] item_flags
1526 * Bit-fields that holds the items detected until now.
1527 * @param[in] acc_mask
1528 * Acceptable mask, if NULL default internal default mask
1529 * will be used to check whether item fields are supported.
1531 * Pointer to error structure.
1534 * 0 on success, a negative errno value otherwise and rte_errno is set.
1537 mlx5_flow_validate_item_ipv4(const struct rte_flow_item *item,
1538 uint64_t item_flags,
1540 uint16_t ether_type,
1541 const struct rte_flow_item_ipv4 *acc_mask,
1542 struct rte_flow_error *error)
1544 const struct rte_flow_item_ipv4 *mask = item->mask;
1545 const struct rte_flow_item_ipv4 *spec = item->spec;
1546 const struct rte_flow_item_ipv4 nic_mask = {
1548 .src_addr = RTE_BE32(0xffffffff),
1549 .dst_addr = RTE_BE32(0xffffffff),
1550 .type_of_service = 0xff,
1551 .next_proto_id = 0xff,
1554 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1555 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1556 MLX5_FLOW_LAYER_OUTER_L3;
1557 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1558 MLX5_FLOW_LAYER_OUTER_L4;
1560 uint8_t next_proto = 0xFF;
1561 const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
1562 MLX5_FLOW_LAYER_OUTER_VLAN |
1563 MLX5_FLOW_LAYER_INNER_VLAN);
1565 if ((last_item & l2_vlan) && ether_type &&
1566 ether_type != RTE_ETHER_TYPE_IPV4)
1567 return rte_flow_error_set(error, EINVAL,
1568 RTE_FLOW_ERROR_TYPE_ITEM, item,
1569 "IPv4 cannot follow L2/VLAN layer "
1570 "which ether type is not IPv4");
1571 if (item_flags & MLX5_FLOW_LAYER_IPIP) {
1573 next_proto = mask->hdr.next_proto_id &
1574 spec->hdr.next_proto_id;
1575 if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
1576 return rte_flow_error_set(error, EINVAL,
1577 RTE_FLOW_ERROR_TYPE_ITEM,
1582 if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP)
1583 return rte_flow_error_set(error, EINVAL,
1584 RTE_FLOW_ERROR_TYPE_ITEM, item,
1585 "wrong tunnel type - IPv6 specified "
1586 "but IPv4 item provided");
1587 if (item_flags & l3m)
1588 return rte_flow_error_set(error, ENOTSUP,
1589 RTE_FLOW_ERROR_TYPE_ITEM, item,
1590 "multiple L3 layers not supported");
1591 else if (item_flags & l4m)
1592 return rte_flow_error_set(error, EINVAL,
1593 RTE_FLOW_ERROR_TYPE_ITEM, item,
1594 "L3 cannot follow an L4 layer.");
1595 else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
1596 !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
1597 return rte_flow_error_set(error, EINVAL,
1598 RTE_FLOW_ERROR_TYPE_ITEM, item,
1599 "L3 cannot follow an NVGRE layer.");
1601 mask = &rte_flow_item_ipv4_mask;
1602 else if (mask->hdr.next_proto_id != 0 &&
1603 mask->hdr.next_proto_id != 0xff)
1604 return rte_flow_error_set(error, EINVAL,
1605 RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
1606 "partial mask is not supported"
1608 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1609 acc_mask ? (const uint8_t *)acc_mask
1610 : (const uint8_t *)&nic_mask,
1611 sizeof(struct rte_flow_item_ipv4),
1619 * Validate IPV6 item.
1622 * Item specification.
1623 * @param[in] item_flags
1624 * Bit-fields that holds the items detected until now.
1625 * @param[in] acc_mask
1626 * Acceptable mask, if NULL default internal default mask
1627 * will be used to check whether item fields are supported.
1629 * Pointer to error structure.
1632 * 0 on success, a negative errno value otherwise and rte_errno is set.
1635 mlx5_flow_validate_item_ipv6(const struct rte_flow_item *item,
1636 uint64_t item_flags,
1638 uint16_t ether_type,
1639 const struct rte_flow_item_ipv6 *acc_mask,
1640 struct rte_flow_error *error)
1642 const struct rte_flow_item_ipv6 *mask = item->mask;
1643 const struct rte_flow_item_ipv6 *spec = item->spec;
1644 const struct rte_flow_item_ipv6 nic_mask = {
1647 "\xff\xff\xff\xff\xff\xff\xff\xff"
1648 "\xff\xff\xff\xff\xff\xff\xff\xff",
1650 "\xff\xff\xff\xff\xff\xff\xff\xff"
1651 "\xff\xff\xff\xff\xff\xff\xff\xff",
1652 .vtc_flow = RTE_BE32(0xffffffff),
1656 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1657 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1658 MLX5_FLOW_LAYER_OUTER_L3;
1659 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1660 MLX5_FLOW_LAYER_OUTER_L4;
1662 uint8_t next_proto = 0xFF;
1663 const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
1664 MLX5_FLOW_LAYER_OUTER_VLAN |
1665 MLX5_FLOW_LAYER_INNER_VLAN);
1667 if ((last_item & l2_vlan) && ether_type &&
1668 ether_type != RTE_ETHER_TYPE_IPV6)
1669 return rte_flow_error_set(error, EINVAL,
1670 RTE_FLOW_ERROR_TYPE_ITEM, item,
1671 "IPv6 cannot follow L2/VLAN layer "
1672 "which ether type is not IPv6");
1673 if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP) {
1675 next_proto = mask->hdr.proto & spec->hdr.proto;
1676 if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
1677 return rte_flow_error_set(error, EINVAL,
1678 RTE_FLOW_ERROR_TYPE_ITEM,
1683 if (item_flags & MLX5_FLOW_LAYER_IPIP)
1684 return rte_flow_error_set(error, EINVAL,
1685 RTE_FLOW_ERROR_TYPE_ITEM, item,
1686 "wrong tunnel type - IPv4 specified "
1687 "but IPv6 item provided");
1688 if (item_flags & l3m)
1689 return rte_flow_error_set(error, ENOTSUP,
1690 RTE_FLOW_ERROR_TYPE_ITEM, item,
1691 "multiple L3 layers not supported");
1692 else if (item_flags & l4m)
1693 return rte_flow_error_set(error, EINVAL,
1694 RTE_FLOW_ERROR_TYPE_ITEM, item,
1695 "L3 cannot follow an L4 layer.");
1696 else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
1697 !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
1698 return rte_flow_error_set(error, EINVAL,
1699 RTE_FLOW_ERROR_TYPE_ITEM, item,
1700 "L3 cannot follow an NVGRE layer.");
1702 mask = &rte_flow_item_ipv6_mask;
1703 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1704 acc_mask ? (const uint8_t *)acc_mask
1705 : (const uint8_t *)&nic_mask,
1706 sizeof(struct rte_flow_item_ipv6),
1714 * Validate UDP item.
1717 * Item specification.
1718 * @param[in] item_flags
1719 * Bit-fields that holds the items detected until now.
1720 * @param[in] target_protocol
1721 * The next protocol in the previous item.
1722 * @param[in] flow_mask
1723 * mlx5 flow-specific (DV, verbs, etc.) supported header fields mask.
1725 * Pointer to error structure.
1728 * 0 on success, a negative errno value otherwise and rte_errno is set.
1731 mlx5_flow_validate_item_udp(const struct rte_flow_item *item,
1732 uint64_t item_flags,
1733 uint8_t target_protocol,
1734 struct rte_flow_error *error)
1736 const struct rte_flow_item_udp *mask = item->mask;
1737 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1738 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1739 MLX5_FLOW_LAYER_OUTER_L3;
1740 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1741 MLX5_FLOW_LAYER_OUTER_L4;
1744 if (target_protocol != 0xff && target_protocol != IPPROTO_UDP)
1745 return rte_flow_error_set(error, EINVAL,
1746 RTE_FLOW_ERROR_TYPE_ITEM, item,
1747 "protocol filtering not compatible"
1749 if (!(item_flags & l3m))
1750 return rte_flow_error_set(error, EINVAL,
1751 RTE_FLOW_ERROR_TYPE_ITEM, item,
1752 "L3 is mandatory to filter on L4");
1753 if (item_flags & l4m)
1754 return rte_flow_error_set(error, EINVAL,
1755 RTE_FLOW_ERROR_TYPE_ITEM, item,
1756 "multiple L4 layers not supported");
1758 mask = &rte_flow_item_udp_mask;
1759 ret = mlx5_flow_item_acceptable
1760 (item, (const uint8_t *)mask,
1761 (const uint8_t *)&rte_flow_item_udp_mask,
1762 sizeof(struct rte_flow_item_udp), error);
1769 * Validate TCP item.
1772 * Item specification.
1773 * @param[in] item_flags
1774 * Bit-fields that holds the items detected until now.
1775 * @param[in] target_protocol
1776 * The next protocol in the previous item.
1778 * Pointer to error structure.
1781 * 0 on success, a negative errno value otherwise and rte_errno is set.
1784 mlx5_flow_validate_item_tcp(const struct rte_flow_item *item,
1785 uint64_t item_flags,
1786 uint8_t target_protocol,
1787 const struct rte_flow_item_tcp *flow_mask,
1788 struct rte_flow_error *error)
1790 const struct rte_flow_item_tcp *mask = item->mask;
1791 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1792 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1793 MLX5_FLOW_LAYER_OUTER_L3;
1794 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1795 MLX5_FLOW_LAYER_OUTER_L4;
1798 MLX5_ASSERT(flow_mask);
1799 if (target_protocol != 0xff && target_protocol != IPPROTO_TCP)
1800 return rte_flow_error_set(error, EINVAL,
1801 RTE_FLOW_ERROR_TYPE_ITEM, item,
1802 "protocol filtering not compatible"
1804 if (!(item_flags & l3m))
1805 return rte_flow_error_set(error, EINVAL,
1806 RTE_FLOW_ERROR_TYPE_ITEM, item,
1807 "L3 is mandatory to filter on L4");
1808 if (item_flags & l4m)
1809 return rte_flow_error_set(error, EINVAL,
1810 RTE_FLOW_ERROR_TYPE_ITEM, item,
1811 "multiple L4 layers not supported");
1813 mask = &rte_flow_item_tcp_mask;
1814 ret = mlx5_flow_item_acceptable
1815 (item, (const uint8_t *)mask,
1816 (const uint8_t *)flow_mask,
1817 sizeof(struct rte_flow_item_tcp), error);
1824 * Validate VXLAN item.
1827 * Item specification.
1828 * @param[in] item_flags
1829 * Bit-fields that holds the items detected until now.
1830 * @param[in] target_protocol
1831 * The next protocol in the previous item.
1833 * Pointer to error structure.
1836 * 0 on success, a negative errno value otherwise and rte_errno is set.
1839 mlx5_flow_validate_item_vxlan(const struct rte_flow_item *item,
1840 uint64_t item_flags,
1841 struct rte_flow_error *error)
1843 const struct rte_flow_item_vxlan *spec = item->spec;
1844 const struct rte_flow_item_vxlan *mask = item->mask;
1849 } id = { .vlan_id = 0, };
1852 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1853 return rte_flow_error_set(error, ENOTSUP,
1854 RTE_FLOW_ERROR_TYPE_ITEM, item,
1855 "multiple tunnel layers not"
1858 * Verify only UDPv4 is present as defined in
1859 * https://tools.ietf.org/html/rfc7348
1861 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
1862 return rte_flow_error_set(error, EINVAL,
1863 RTE_FLOW_ERROR_TYPE_ITEM, item,
1864 "no outer UDP layer found");
1866 mask = &rte_flow_item_vxlan_mask;
1867 ret = mlx5_flow_item_acceptable
1868 (item, (const uint8_t *)mask,
1869 (const uint8_t *)&rte_flow_item_vxlan_mask,
1870 sizeof(struct rte_flow_item_vxlan),
1875 memcpy(&id.vni[1], spec->vni, 3);
1876 memcpy(&id.vni[1], mask->vni, 3);
1878 if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
1879 return rte_flow_error_set(error, ENOTSUP,
1880 RTE_FLOW_ERROR_TYPE_ITEM, item,
1881 "VXLAN tunnel must be fully defined");
1886 * Validate VXLAN_GPE item.
1889 * Item specification.
1890 * @param[in] item_flags
1891 * Bit-fields that holds the items detected until now.
1893 * Pointer to the private data structure.
1894 * @param[in] target_protocol
1895 * The next protocol in the previous item.
1897 * Pointer to error structure.
1900 * 0 on success, a negative errno value otherwise and rte_errno is set.
1903 mlx5_flow_validate_item_vxlan_gpe(const struct rte_flow_item *item,
1904 uint64_t item_flags,
1905 struct rte_eth_dev *dev,
1906 struct rte_flow_error *error)
1908 struct mlx5_priv *priv = dev->data->dev_private;
1909 const struct rte_flow_item_vxlan_gpe *spec = item->spec;
1910 const struct rte_flow_item_vxlan_gpe *mask = item->mask;
1915 } id = { .vlan_id = 0, };
1917 if (!priv->config.l3_vxlan_en)
1918 return rte_flow_error_set(error, ENOTSUP,
1919 RTE_FLOW_ERROR_TYPE_ITEM, item,
1920 "L3 VXLAN is not enabled by device"
1921 " parameter and/or not configured in"
1923 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1924 return rte_flow_error_set(error, ENOTSUP,
1925 RTE_FLOW_ERROR_TYPE_ITEM, item,
1926 "multiple tunnel layers not"
1929 * Verify only UDPv4 is present as defined in
1930 * https://tools.ietf.org/html/rfc7348
1932 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
1933 return rte_flow_error_set(error, EINVAL,
1934 RTE_FLOW_ERROR_TYPE_ITEM, item,
1935 "no outer UDP layer found");
1937 mask = &rte_flow_item_vxlan_gpe_mask;
1938 ret = mlx5_flow_item_acceptable
1939 (item, (const uint8_t *)mask,
1940 (const uint8_t *)&rte_flow_item_vxlan_gpe_mask,
1941 sizeof(struct rte_flow_item_vxlan_gpe),
1947 return rte_flow_error_set(error, ENOTSUP,
1948 RTE_FLOW_ERROR_TYPE_ITEM,
1950 "VxLAN-GPE protocol"
1952 memcpy(&id.vni[1], spec->vni, 3);
1953 memcpy(&id.vni[1], mask->vni, 3);
1955 if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
1956 return rte_flow_error_set(error, ENOTSUP,
1957 RTE_FLOW_ERROR_TYPE_ITEM, item,
1958 "VXLAN-GPE tunnel must be fully"
1963 * Validate GRE Key item.
1966 * Item specification.
1967 * @param[in] item_flags
1968 * Bit flags to mark detected items.
1969 * @param[in] gre_item
1970 * Pointer to gre_item
1972 * Pointer to error structure.
1975 * 0 on success, a negative errno value otherwise and rte_errno is set.
1978 mlx5_flow_validate_item_gre_key(const struct rte_flow_item *item,
1979 uint64_t item_flags,
1980 const struct rte_flow_item *gre_item,
1981 struct rte_flow_error *error)
1983 const rte_be32_t *mask = item->mask;
1985 rte_be32_t gre_key_default_mask = RTE_BE32(UINT32_MAX);
1986 const struct rte_flow_item_gre *gre_spec;
1987 const struct rte_flow_item_gre *gre_mask;
1989 if (item_flags & MLX5_FLOW_LAYER_GRE_KEY)
1990 return rte_flow_error_set(error, ENOTSUP,
1991 RTE_FLOW_ERROR_TYPE_ITEM, item,
1992 "Multiple GRE key not support");
1993 if (!(item_flags & MLX5_FLOW_LAYER_GRE))
1994 return rte_flow_error_set(error, ENOTSUP,
1995 RTE_FLOW_ERROR_TYPE_ITEM, item,
1996 "No preceding GRE header");
1997 if (item_flags & MLX5_FLOW_LAYER_INNER)
1998 return rte_flow_error_set(error, ENOTSUP,
1999 RTE_FLOW_ERROR_TYPE_ITEM, item,
2000 "GRE key following a wrong item");
2001 gre_mask = gre_item->mask;
2003 gre_mask = &rte_flow_item_gre_mask;
2004 gre_spec = gre_item->spec;
2005 if (gre_spec && (gre_mask->c_rsvd0_ver & RTE_BE16(0x2000)) &&
2006 !(gre_spec->c_rsvd0_ver & RTE_BE16(0x2000)))
2007 return rte_flow_error_set(error, EINVAL,
2008 RTE_FLOW_ERROR_TYPE_ITEM, item,
2009 "Key bit must be on");
2012 mask = &gre_key_default_mask;
2013 ret = mlx5_flow_item_acceptable
2014 (item, (const uint8_t *)mask,
2015 (const uint8_t *)&gre_key_default_mask,
2016 sizeof(rte_be32_t), error);
2021 * Validate GRE item.
2024 * Item specification.
2025 * @param[in] item_flags
2026 * Bit flags to mark detected items.
2027 * @param[in] target_protocol
2028 * The next protocol in the previous item.
2030 * Pointer to error structure.
2033 * 0 on success, a negative errno value otherwise and rte_errno is set.
2036 mlx5_flow_validate_item_gre(const struct rte_flow_item *item,
2037 uint64_t item_flags,
2038 uint8_t target_protocol,
2039 struct rte_flow_error *error)
2041 const struct rte_flow_item_gre *spec __rte_unused = item->spec;
2042 const struct rte_flow_item_gre *mask = item->mask;
2044 const struct rte_flow_item_gre nic_mask = {
2045 .c_rsvd0_ver = RTE_BE16(0xB000),
2046 .protocol = RTE_BE16(UINT16_MAX),
2049 if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2050 return rte_flow_error_set(error, EINVAL,
2051 RTE_FLOW_ERROR_TYPE_ITEM, item,
2052 "protocol filtering not compatible"
2053 " with this GRE layer");
2054 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2055 return rte_flow_error_set(error, ENOTSUP,
2056 RTE_FLOW_ERROR_TYPE_ITEM, item,
2057 "multiple tunnel layers not"
2059 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2060 return rte_flow_error_set(error, ENOTSUP,
2061 RTE_FLOW_ERROR_TYPE_ITEM, item,
2062 "L3 Layer is missing");
2064 mask = &rte_flow_item_gre_mask;
2065 ret = mlx5_flow_item_acceptable
2066 (item, (const uint8_t *)mask,
2067 (const uint8_t *)&nic_mask,
2068 sizeof(struct rte_flow_item_gre), error);
2071 #ifndef HAVE_MLX5DV_DR
2072 #ifndef HAVE_IBV_DEVICE_MPLS_SUPPORT
2073 if (spec && (spec->protocol & mask->protocol))
2074 return rte_flow_error_set(error, ENOTSUP,
2075 RTE_FLOW_ERROR_TYPE_ITEM, item,
2076 "without MPLS support the"
2077 " specification cannot be used for"
2085 * Validate Geneve item.
2088 * Item specification.
2089 * @param[in] itemFlags
2090 * Bit-fields that holds the items detected until now.
2092 * Pointer to the private data structure.
2094 * Pointer to error structure.
2097 * 0 on success, a negative errno value otherwise and rte_errno is set.
2101 mlx5_flow_validate_item_geneve(const struct rte_flow_item *item,
2102 uint64_t item_flags,
2103 struct rte_eth_dev *dev,
2104 struct rte_flow_error *error)
2106 struct mlx5_priv *priv = dev->data->dev_private;
2107 const struct rte_flow_item_geneve *spec = item->spec;
2108 const struct rte_flow_item_geneve *mask = item->mask;
2111 uint8_t opt_len = priv->config.hca_attr.geneve_max_opt_len ?
2112 MLX5_GENEVE_OPT_LEN_1 : MLX5_GENEVE_OPT_LEN_0;
2113 const struct rte_flow_item_geneve nic_mask = {
2114 .ver_opt_len_o_c_rsvd0 = RTE_BE16(0x3f80),
2115 .vni = "\xff\xff\xff",
2116 .protocol = RTE_BE16(UINT16_MAX),
2119 if (!priv->config.hca_attr.tunnel_stateless_geneve_rx)
2120 return rte_flow_error_set(error, ENOTSUP,
2121 RTE_FLOW_ERROR_TYPE_ITEM, item,
2122 "L3 Geneve is not enabled by device"
2123 " parameter and/or not configured in"
2125 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2126 return rte_flow_error_set(error, ENOTSUP,
2127 RTE_FLOW_ERROR_TYPE_ITEM, item,
2128 "multiple tunnel layers not"
2131 * Verify only UDPv4 is present as defined in
2132 * https://tools.ietf.org/html/rfc7348
2134 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2135 return rte_flow_error_set(error, EINVAL,
2136 RTE_FLOW_ERROR_TYPE_ITEM, item,
2137 "no outer UDP layer found");
2139 mask = &rte_flow_item_geneve_mask;
2140 ret = mlx5_flow_item_acceptable
2141 (item, (const uint8_t *)mask,
2142 (const uint8_t *)&nic_mask,
2143 sizeof(struct rte_flow_item_geneve), error);
2147 gbhdr = rte_be_to_cpu_16(spec->ver_opt_len_o_c_rsvd0);
2148 if (MLX5_GENEVE_VER_VAL(gbhdr) ||
2149 MLX5_GENEVE_CRITO_VAL(gbhdr) ||
2150 MLX5_GENEVE_RSVD_VAL(gbhdr) || spec->rsvd1)
2151 return rte_flow_error_set(error, ENOTSUP,
2152 RTE_FLOW_ERROR_TYPE_ITEM,
2154 "Geneve protocol unsupported"
2155 " fields are being used");
2156 if (MLX5_GENEVE_OPTLEN_VAL(gbhdr) > opt_len)
2157 return rte_flow_error_set
2159 RTE_FLOW_ERROR_TYPE_ITEM,
2161 "Unsupported Geneve options length");
2163 if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2164 return rte_flow_error_set
2166 RTE_FLOW_ERROR_TYPE_ITEM, item,
2167 "Geneve tunnel must be fully defined");
2172 * Validate MPLS item.
2175 * Pointer to the rte_eth_dev structure.
2177 * Item specification.
2178 * @param[in] item_flags
2179 * Bit-fields that holds the items detected until now.
2180 * @param[in] prev_layer
2181 * The protocol layer indicated in previous item.
2183 * Pointer to error structure.
2186 * 0 on success, a negative errno value otherwise and rte_errno is set.
2189 mlx5_flow_validate_item_mpls(struct rte_eth_dev *dev __rte_unused,
2190 const struct rte_flow_item *item __rte_unused,
2191 uint64_t item_flags __rte_unused,
2192 uint64_t prev_layer __rte_unused,
2193 struct rte_flow_error *error)
2195 #ifdef HAVE_IBV_DEVICE_MPLS_SUPPORT
2196 const struct rte_flow_item_mpls *mask = item->mask;
2197 struct mlx5_priv *priv = dev->data->dev_private;
2200 if (!priv->config.mpls_en)
2201 return rte_flow_error_set(error, ENOTSUP,
2202 RTE_FLOW_ERROR_TYPE_ITEM, item,
2203 "MPLS not supported or"
2204 " disabled in firmware"
2206 /* MPLS over IP, UDP, GRE is allowed */
2207 if (!(prev_layer & (MLX5_FLOW_LAYER_OUTER_L3 |
2208 MLX5_FLOW_LAYER_OUTER_L4_UDP |
2209 MLX5_FLOW_LAYER_GRE)))
2210 return rte_flow_error_set(error, EINVAL,
2211 RTE_FLOW_ERROR_TYPE_ITEM, item,
2212 "protocol filtering not compatible"
2213 " with MPLS layer");
2214 /* Multi-tunnel isn't allowed but MPLS over GRE is an exception. */
2215 if ((item_flags & MLX5_FLOW_LAYER_TUNNEL) &&
2216 !(item_flags & MLX5_FLOW_LAYER_GRE))
2217 return rte_flow_error_set(error, ENOTSUP,
2218 RTE_FLOW_ERROR_TYPE_ITEM, item,
2219 "multiple tunnel layers not"
2222 mask = &rte_flow_item_mpls_mask;
2223 ret = mlx5_flow_item_acceptable
2224 (item, (const uint8_t *)mask,
2225 (const uint8_t *)&rte_flow_item_mpls_mask,
2226 sizeof(struct rte_flow_item_mpls), error);
2231 return rte_flow_error_set(error, ENOTSUP,
2232 RTE_FLOW_ERROR_TYPE_ITEM, item,
2233 "MPLS is not supported by Verbs, please"
2238 * Validate NVGRE item.
2241 * Item specification.
2242 * @param[in] item_flags
2243 * Bit flags to mark detected items.
2244 * @param[in] target_protocol
2245 * The next protocol in the previous item.
2247 * Pointer to error structure.
2250 * 0 on success, a negative errno value otherwise and rte_errno is set.
2253 mlx5_flow_validate_item_nvgre(const struct rte_flow_item *item,
2254 uint64_t item_flags,
2255 uint8_t target_protocol,
2256 struct rte_flow_error *error)
2258 const struct rte_flow_item_nvgre *mask = item->mask;
2261 if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2262 return rte_flow_error_set(error, EINVAL,
2263 RTE_FLOW_ERROR_TYPE_ITEM, item,
2264 "protocol filtering not compatible"
2265 " with this GRE layer");
2266 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2267 return rte_flow_error_set(error, ENOTSUP,
2268 RTE_FLOW_ERROR_TYPE_ITEM, item,
2269 "multiple tunnel layers not"
2271 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2272 return rte_flow_error_set(error, ENOTSUP,
2273 RTE_FLOW_ERROR_TYPE_ITEM, item,
2274 "L3 Layer is missing");
2276 mask = &rte_flow_item_nvgre_mask;
2277 ret = mlx5_flow_item_acceptable
2278 (item, (const uint8_t *)mask,
2279 (const uint8_t *)&rte_flow_item_nvgre_mask,
2280 sizeof(struct rte_flow_item_nvgre), error);
2286 /* Allocate unique ID for the split Q/RSS subflows. */
2288 flow_qrss_get_id(struct rte_eth_dev *dev)
2290 struct mlx5_priv *priv = dev->data->dev_private;
2291 uint32_t qrss_id, ret;
2293 ret = mlx5_flow_id_get(priv->qrss_id_pool, &qrss_id);
2296 MLX5_ASSERT(qrss_id);
2300 /* Free unique ID for the split Q/RSS subflows. */
2302 flow_qrss_free_id(struct rte_eth_dev *dev, uint32_t qrss_id)
2304 struct mlx5_priv *priv = dev->data->dev_private;
2307 mlx5_flow_id_release(priv->qrss_id_pool, qrss_id);
2311 * Release resource related QUEUE/RSS action split.
2314 * Pointer to Ethernet device.
2316 * Flow to release id's from.
2319 flow_mreg_split_qrss_release(struct rte_eth_dev *dev,
2320 struct rte_flow *flow)
2322 struct mlx5_priv *priv = dev->data->dev_private;
2323 uint32_t handle_idx;
2324 struct mlx5_flow_handle *dev_handle;
2326 SILIST_FOREACH(priv->sh->ipool[MLX5_IPOOL_MLX5_FLOW], flow->dev_handles,
2327 handle_idx, dev_handle, next)
2328 if (dev_handle->qrss_id)
2329 flow_qrss_free_id(dev, dev_handle->qrss_id);
2333 flow_null_validate(struct rte_eth_dev *dev __rte_unused,
2334 const struct rte_flow_attr *attr __rte_unused,
2335 const struct rte_flow_item items[] __rte_unused,
2336 const struct rte_flow_action actions[] __rte_unused,
2337 bool external __rte_unused,
2338 struct rte_flow_error *error)
2340 return rte_flow_error_set(error, ENOTSUP,
2341 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2344 static struct mlx5_flow *
2345 flow_null_prepare(struct rte_eth_dev *dev __rte_unused,
2346 const struct rte_flow_attr *attr __rte_unused,
2347 const struct rte_flow_item items[] __rte_unused,
2348 const struct rte_flow_action actions[] __rte_unused,
2349 struct rte_flow_error *error)
2351 rte_flow_error_set(error, ENOTSUP,
2352 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2357 flow_null_translate(struct rte_eth_dev *dev __rte_unused,
2358 struct mlx5_flow *dev_flow __rte_unused,
2359 const struct rte_flow_attr *attr __rte_unused,
2360 const struct rte_flow_item items[] __rte_unused,
2361 const struct rte_flow_action actions[] __rte_unused,
2362 struct rte_flow_error *error)
2364 return rte_flow_error_set(error, ENOTSUP,
2365 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2369 flow_null_apply(struct rte_eth_dev *dev __rte_unused,
2370 struct rte_flow *flow __rte_unused,
2371 struct rte_flow_error *error)
2373 return rte_flow_error_set(error, ENOTSUP,
2374 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2378 flow_null_remove(struct rte_eth_dev *dev __rte_unused,
2379 struct rte_flow *flow __rte_unused)
2384 flow_null_destroy(struct rte_eth_dev *dev __rte_unused,
2385 struct rte_flow *flow __rte_unused)
2390 flow_null_query(struct rte_eth_dev *dev __rte_unused,
2391 struct rte_flow *flow __rte_unused,
2392 const struct rte_flow_action *actions __rte_unused,
2393 void *data __rte_unused,
2394 struct rte_flow_error *error)
2396 return rte_flow_error_set(error, ENOTSUP,
2397 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2400 /* Void driver to protect from null pointer reference. */
2401 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops = {
2402 .validate = flow_null_validate,
2403 .prepare = flow_null_prepare,
2404 .translate = flow_null_translate,
2405 .apply = flow_null_apply,
2406 .remove = flow_null_remove,
2407 .destroy = flow_null_destroy,
2408 .query = flow_null_query,
2412 * Select flow driver type according to flow attributes and device
2416 * Pointer to the dev structure.
2418 * Pointer to the flow attributes.
2421 * flow driver type, MLX5_FLOW_TYPE_MAX otherwise.
2423 static enum mlx5_flow_drv_type
2424 flow_get_drv_type(struct rte_eth_dev *dev, const struct rte_flow_attr *attr)
2426 struct mlx5_priv *priv = dev->data->dev_private;
2427 enum mlx5_flow_drv_type type = MLX5_FLOW_TYPE_MAX;
2429 if (attr->transfer && priv->config.dv_esw_en)
2430 type = MLX5_FLOW_TYPE_DV;
2431 if (!attr->transfer)
2432 type = priv->config.dv_flow_en ? MLX5_FLOW_TYPE_DV :
2433 MLX5_FLOW_TYPE_VERBS;
2437 #define flow_get_drv_ops(type) flow_drv_ops[type]
2440 * Flow driver validation API. This abstracts calling driver specific functions.
2441 * The type of flow driver is determined according to flow attributes.
2444 * Pointer to the dev structure.
2446 * Pointer to the flow attributes.
2448 * Pointer to the list of items.
2449 * @param[in] actions
2450 * Pointer to the list of actions.
2451 * @param[in] external
2452 * This flow rule is created by request external to PMD.
2454 * Pointer to the error structure.
2457 * 0 on success, a negative errno value otherwise and rte_errno is set.
2460 flow_drv_validate(struct rte_eth_dev *dev,
2461 const struct rte_flow_attr *attr,
2462 const struct rte_flow_item items[],
2463 const struct rte_flow_action actions[],
2464 bool external, struct rte_flow_error *error)
2466 const struct mlx5_flow_driver_ops *fops;
2467 enum mlx5_flow_drv_type type = flow_get_drv_type(dev, attr);
2469 fops = flow_get_drv_ops(type);
2470 return fops->validate(dev, attr, items, actions, external, error);
2474 * Flow driver preparation API. This abstracts calling driver specific
2475 * functions. Parent flow (rte_flow) should have driver type (drv_type). It
2476 * calculates the size of memory required for device flow, allocates the memory,
2477 * initializes the device flow and returns the pointer.
2480 * This function initializes device flow structure such as dv or verbs in
2481 * struct mlx5_flow. However, it is caller's responsibility to initialize the
2482 * rest. For example, adding returning device flow to flow->dev_flow list and
2483 * setting backward reference to the flow should be done out of this function.
2484 * layers field is not filled either.
2487 * Pointer to the dev structure.
2489 * Pointer to the flow attributes.
2491 * Pointer to the list of items.
2492 * @param[in] actions
2493 * Pointer to the list of actions.
2495 * Pointer to the error structure.
2498 * Pointer to device flow on success, otherwise NULL and rte_errno is set.
2500 static inline struct mlx5_flow *
2501 flow_drv_prepare(struct rte_eth_dev *dev,
2502 const struct rte_flow *flow,
2503 const struct rte_flow_attr *attr,
2504 const struct rte_flow_item items[],
2505 const struct rte_flow_action actions[],
2506 struct rte_flow_error *error)
2508 const struct mlx5_flow_driver_ops *fops;
2509 enum mlx5_flow_drv_type type = flow->drv_type;
2511 MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2512 fops = flow_get_drv_ops(type);
2513 return fops->prepare(dev, attr, items, actions, error);
2517 * Flow driver translation API. This abstracts calling driver specific
2518 * functions. Parent flow (rte_flow) should have driver type (drv_type). It
2519 * translates a generic flow into a driver flow. flow_drv_prepare() must
2523 * dev_flow->layers could be filled as a result of parsing during translation
2524 * if needed by flow_drv_apply(). dev_flow->flow->actions can also be filled
2525 * if necessary. As a flow can have multiple dev_flows by RSS flow expansion,
2526 * flow->actions could be overwritten even though all the expanded dev_flows
2527 * have the same actions.
2530 * Pointer to the rte dev structure.
2531 * @param[in, out] dev_flow
2532 * Pointer to the mlx5 flow.
2534 * Pointer to the flow attributes.
2536 * Pointer to the list of items.
2537 * @param[in] actions
2538 * Pointer to the list of actions.
2540 * Pointer to the error structure.
2543 * 0 on success, a negative errno value otherwise and rte_errno is set.
2546 flow_drv_translate(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow,
2547 const struct rte_flow_attr *attr,
2548 const struct rte_flow_item items[],
2549 const struct rte_flow_action actions[],
2550 struct rte_flow_error *error)
2552 const struct mlx5_flow_driver_ops *fops;
2553 enum mlx5_flow_drv_type type = dev_flow->flow->drv_type;
2555 MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2556 fops = flow_get_drv_ops(type);
2557 return fops->translate(dev, dev_flow, attr, items, actions, error);
2561 * Flow driver apply API. This abstracts calling driver specific functions.
2562 * Parent flow (rte_flow) should have driver type (drv_type). It applies
2563 * translated driver flows on to device. flow_drv_translate() must precede.
2566 * Pointer to Ethernet device structure.
2567 * @param[in, out] flow
2568 * Pointer to flow structure.
2570 * Pointer to error structure.
2573 * 0 on success, a negative errno value otherwise and rte_errno is set.
2576 flow_drv_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
2577 struct rte_flow_error *error)
2579 const struct mlx5_flow_driver_ops *fops;
2580 enum mlx5_flow_drv_type type = flow->drv_type;
2582 MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2583 fops = flow_get_drv_ops(type);
2584 return fops->apply(dev, flow, error);
2588 * Flow driver remove API. This abstracts calling driver specific functions.
2589 * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
2590 * on device. All the resources of the flow should be freed by calling
2591 * flow_drv_destroy().
2594 * Pointer to Ethernet device.
2595 * @param[in, out] flow
2596 * Pointer to flow structure.
2599 flow_drv_remove(struct rte_eth_dev *dev, struct rte_flow *flow)
2601 const struct mlx5_flow_driver_ops *fops;
2602 enum mlx5_flow_drv_type type = flow->drv_type;
2604 MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2605 fops = flow_get_drv_ops(type);
2606 fops->remove(dev, flow);
2610 * Flow driver destroy API. This abstracts calling driver specific functions.
2611 * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
2612 * on device and releases resources of the flow.
2615 * Pointer to Ethernet device.
2616 * @param[in, out] flow
2617 * Pointer to flow structure.
2620 flow_drv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
2622 const struct mlx5_flow_driver_ops *fops;
2623 enum mlx5_flow_drv_type type = flow->drv_type;
2625 flow_mreg_split_qrss_release(dev, flow);
2626 MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2627 fops = flow_get_drv_ops(type);
2628 fops->destroy(dev, flow);
2632 * Validate a flow supported by the NIC.
2634 * @see rte_flow_validate()
2638 mlx5_flow_validate(struct rte_eth_dev *dev,
2639 const struct rte_flow_attr *attr,
2640 const struct rte_flow_item items[],
2641 const struct rte_flow_action actions[],
2642 struct rte_flow_error *error)
2646 ret = flow_drv_validate(dev, attr, items, actions, true, error);
2653 * Get RSS action from the action list.
2655 * @param[in] actions
2656 * Pointer to the list of actions.
2659 * Pointer to the RSS action if exist, else return NULL.
2661 static const struct rte_flow_action_rss*
2662 flow_get_rss_action(const struct rte_flow_action actions[])
2664 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2665 switch (actions->type) {
2666 case RTE_FLOW_ACTION_TYPE_RSS:
2667 return (const struct rte_flow_action_rss *)
2677 find_graph_root(const struct rte_flow_item pattern[], uint32_t rss_level)
2679 const struct rte_flow_item *item;
2680 unsigned int has_vlan = 0;
2682 for (item = pattern; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
2683 if (item->type == RTE_FLOW_ITEM_TYPE_VLAN) {
2689 return rss_level < 2 ? MLX5_EXPANSION_ROOT_ETH_VLAN :
2690 MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN;
2691 return rss_level < 2 ? MLX5_EXPANSION_ROOT :
2692 MLX5_EXPANSION_ROOT_OUTER;
2696 * Get layer flags from the prefix flow.
2698 * Some flows may be split to several subflows, the prefix subflow gets the
2699 * match items and the suffix sub flow gets the actions.
2700 * Some actions need the user defined match item flags to get the detail for
2702 * This function helps the suffix flow to get the item layer flags from prefix
2705 * @param[in] dev_flow
2706 * Pointer the created preifx subflow.
2709 * The layers get from prefix subflow.
2711 static inline uint64_t
2712 flow_get_prefix_layer_flags(struct mlx5_flow *dev_flow)
2714 uint64_t layers = 0;
2717 * Layers bits could be localization, but usually the compiler will
2718 * help to do the optimization work for source code.
2719 * If no decap actions, use the layers directly.
2721 if (!(dev_flow->handle->act_flags & MLX5_FLOW_ACTION_DECAP))
2722 return dev_flow->handle->layers;
2723 /* Convert L3 layers with decap action. */
2724 if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV4)
2725 layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV4;
2726 else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L3_IPV6)
2727 layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV6;
2728 /* Convert L4 layers with decap action. */
2729 if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_TCP)
2730 layers |= MLX5_FLOW_LAYER_OUTER_L4_TCP;
2731 else if (dev_flow->handle->layers & MLX5_FLOW_LAYER_INNER_L4_UDP)
2732 layers |= MLX5_FLOW_LAYER_OUTER_L4_UDP;
2737 * Get metadata split action information.
2739 * @param[in] actions
2740 * Pointer to the list of actions.
2742 * Pointer to the return pointer.
2743 * @param[out] qrss_type
2744 * Pointer to the action type to return. RTE_FLOW_ACTION_TYPE_END is returned
2745 * if no QUEUE/RSS is found.
2746 * @param[out] encap_idx
2747 * Pointer to the index of the encap action if exists, otherwise the last
2751 * Total number of actions.
2754 flow_parse_metadata_split_actions_info(const struct rte_flow_action actions[],
2755 const struct rte_flow_action **qrss,
2758 const struct rte_flow_action_raw_encap *raw_encap;
2760 int raw_decap_idx = -1;
2763 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2764 switch (actions->type) {
2765 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
2766 case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
2767 *encap_idx = actions_n;
2769 case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
2770 raw_decap_idx = actions_n;
2772 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
2773 raw_encap = actions->conf;
2774 if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
2775 *encap_idx = raw_decap_idx != -1 ?
2776 raw_decap_idx : actions_n;
2778 case RTE_FLOW_ACTION_TYPE_QUEUE:
2779 case RTE_FLOW_ACTION_TYPE_RSS:
2787 if (*encap_idx == -1)
2788 *encap_idx = actions_n;
2789 /* Count RTE_FLOW_ACTION_TYPE_END. */
2790 return actions_n + 1;
2794 * Check meter action from the action list.
2796 * @param[in] actions
2797 * Pointer to the list of actions.
2799 * Pointer to the meter exist flag.
2802 * Total number of actions.
2805 flow_check_meter_action(const struct rte_flow_action actions[], uint32_t *mtr)
2811 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2812 switch (actions->type) {
2813 case RTE_FLOW_ACTION_TYPE_METER:
2821 /* Count RTE_FLOW_ACTION_TYPE_END. */
2822 return actions_n + 1;
2826 * Check if the flow should be splited due to hairpin.
2827 * The reason for the split is that in current HW we can't
2828 * support encap on Rx, so if a flow have encap we move it
2832 * Pointer to Ethernet device.
2834 * Flow rule attributes.
2835 * @param[in] actions
2836 * Associated actions (list terminated by the END action).
2839 * > 0 the number of actions and the flow should be split,
2840 * 0 when no split required.
2843 flow_check_hairpin_split(struct rte_eth_dev *dev,
2844 const struct rte_flow_attr *attr,
2845 const struct rte_flow_action actions[])
2847 int queue_action = 0;
2850 const struct rte_flow_action_queue *queue;
2851 const struct rte_flow_action_rss *rss;
2852 const struct rte_flow_action_raw_encap *raw_encap;
2856 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2857 switch (actions->type) {
2858 case RTE_FLOW_ACTION_TYPE_QUEUE:
2859 queue = actions->conf;
2862 if (mlx5_rxq_get_type(dev, queue->index) !=
2863 MLX5_RXQ_TYPE_HAIRPIN)
2868 case RTE_FLOW_ACTION_TYPE_RSS:
2869 rss = actions->conf;
2870 if (rss == NULL || rss->queue_num == 0)
2872 if (mlx5_rxq_get_type(dev, rss->queue[0]) !=
2873 MLX5_RXQ_TYPE_HAIRPIN)
2878 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
2879 case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
2883 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
2884 raw_encap = actions->conf;
2885 if (raw_encap->size >
2886 (sizeof(struct rte_flow_item_eth) +
2887 sizeof(struct rte_flow_item_ipv4)))
2896 if (encap == 1 && queue_action)
2901 /* Declare flow create/destroy prototype in advance. */
2902 static struct rte_flow *
2903 flow_list_create(struct rte_eth_dev *dev, struct mlx5_flows *list,
2904 const struct rte_flow_attr *attr,
2905 const struct rte_flow_item items[],
2906 const struct rte_flow_action actions[],
2907 bool external, struct rte_flow_error *error);
2910 flow_list_destroy(struct rte_eth_dev *dev, struct mlx5_flows *list,
2911 struct rte_flow *flow);
2914 * Add a flow of copying flow metadata registers in RX_CP_TBL.
2916 * As mark_id is unique, if there's already a registered flow for the mark_id,
2917 * return by increasing the reference counter of the resource. Otherwise, create
2918 * the resource (mcp_res) and flow.
2921 * - If ingress port is ANY and reg_c[1] is mark_id,
2922 * flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
2924 * For default flow (zero mark_id), flow is like,
2925 * - If ingress port is ANY,
2926 * reg_b := reg_c[0] and jump to RX_ACT_TBL.
2929 * Pointer to Ethernet device.
2931 * ID of MARK action, zero means default flow for META.
2933 * Perform verbose error reporting if not NULL.
2936 * Associated resource on success, NULL otherwise and rte_errno is set.
2938 static struct mlx5_flow_mreg_copy_resource *
2939 flow_mreg_add_copy_action(struct rte_eth_dev *dev, uint32_t mark_id,
2940 struct rte_flow_error *error)
2942 struct mlx5_priv *priv = dev->data->dev_private;
2943 struct rte_flow_attr attr = {
2944 .group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
2947 struct mlx5_rte_flow_item_tag tag_spec = {
2950 struct rte_flow_item items[] = {
2951 [1] = { .type = RTE_FLOW_ITEM_TYPE_END, },
2953 struct rte_flow_action_mark ftag = {
2956 struct mlx5_flow_action_copy_mreg cp_mreg = {
2960 struct rte_flow_action_jump jump = {
2961 .group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
2963 struct rte_flow_action actions[] = {
2964 [3] = { .type = RTE_FLOW_ACTION_TYPE_END, },
2966 struct mlx5_flow_mreg_copy_resource *mcp_res;
2969 /* Fill the register fileds in the flow. */
2970 ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
2974 ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
2978 /* Check if already registered. */
2979 MLX5_ASSERT(priv->mreg_cp_tbl);
2980 mcp_res = (void *)mlx5_hlist_lookup(priv->mreg_cp_tbl, mark_id);
2982 /* For non-default rule. */
2983 if (mark_id != MLX5_DEFAULT_COPY_ID)
2985 MLX5_ASSERT(mark_id != MLX5_DEFAULT_COPY_ID ||
2986 mcp_res->refcnt == 1);
2989 /* Provide the full width of FLAG specific value. */
2990 if (mark_id == (priv->sh->dv_regc0_mask & MLX5_FLOW_MARK_DEFAULT))
2991 tag_spec.data = MLX5_FLOW_MARK_DEFAULT;
2992 /* Build a new flow. */
2993 if (mark_id != MLX5_DEFAULT_COPY_ID) {
2994 items[0] = (struct rte_flow_item){
2995 .type = MLX5_RTE_FLOW_ITEM_TYPE_TAG,
2998 items[1] = (struct rte_flow_item){
2999 .type = RTE_FLOW_ITEM_TYPE_END,
3001 actions[0] = (struct rte_flow_action){
3002 .type = MLX5_RTE_FLOW_ACTION_TYPE_MARK,
3005 actions[1] = (struct rte_flow_action){
3006 .type = MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3009 actions[2] = (struct rte_flow_action){
3010 .type = RTE_FLOW_ACTION_TYPE_JUMP,
3013 actions[3] = (struct rte_flow_action){
3014 .type = RTE_FLOW_ACTION_TYPE_END,
3017 /* Default rule, wildcard match. */
3018 attr.priority = MLX5_FLOW_PRIO_RSVD;
3019 items[0] = (struct rte_flow_item){
3020 .type = RTE_FLOW_ITEM_TYPE_END,
3022 actions[0] = (struct rte_flow_action){
3023 .type = MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3026 actions[1] = (struct rte_flow_action){
3027 .type = RTE_FLOW_ACTION_TYPE_JUMP,
3030 actions[2] = (struct rte_flow_action){
3031 .type = RTE_FLOW_ACTION_TYPE_END,
3034 /* Build a new entry. */
3035 mcp_res = rte_zmalloc(__func__, sizeof(*mcp_res), 0);
3041 * The copy Flows are not included in any list. There
3042 * ones are referenced from other Flows and can not
3043 * be applied, removed, deleted in ardbitrary order
3044 * by list traversing.
3046 mcp_res->flow = flow_list_create(dev, NULL, &attr, items,
3047 actions, false, error);
3051 mcp_res->hlist_ent.key = mark_id;
3052 ret = mlx5_hlist_insert(priv->mreg_cp_tbl,
3053 &mcp_res->hlist_ent);
3060 flow_list_destroy(dev, NULL, mcp_res->flow);
3066 * Release flow in RX_CP_TBL.
3069 * Pointer to Ethernet device.
3071 * Parent flow for wich copying is provided.
3074 flow_mreg_del_copy_action(struct rte_eth_dev *dev,
3075 struct rte_flow *flow)
3077 struct mlx5_flow_mreg_copy_resource *mcp_res = flow->mreg_copy;
3078 struct mlx5_priv *priv = dev->data->dev_private;
3080 if (!mcp_res || !priv->mreg_cp_tbl)
3082 if (flow->copy_applied) {
3083 MLX5_ASSERT(mcp_res->appcnt);
3084 flow->copy_applied = 0;
3086 if (!mcp_res->appcnt)
3087 flow_drv_remove(dev, mcp_res->flow);
3090 * We do not check availability of metadata registers here,
3091 * because copy resources are not allocated in this case.
3093 if (--mcp_res->refcnt)
3095 MLX5_ASSERT(mcp_res->flow);
3096 flow_list_destroy(dev, NULL, mcp_res->flow);
3097 mlx5_hlist_remove(priv->mreg_cp_tbl, &mcp_res->hlist_ent);
3099 flow->mreg_copy = NULL;
3103 * Start flow in RX_CP_TBL.
3106 * Pointer to Ethernet device.
3108 * Parent flow for wich copying is provided.
3111 * 0 on success, a negative errno value otherwise and rte_errno is set.
3114 flow_mreg_start_copy_action(struct rte_eth_dev *dev,
3115 struct rte_flow *flow)
3117 struct mlx5_flow_mreg_copy_resource *mcp_res = flow->mreg_copy;
3120 if (!mcp_res || flow->copy_applied)
3122 if (!mcp_res->appcnt) {
3123 ret = flow_drv_apply(dev, mcp_res->flow, NULL);
3128 flow->copy_applied = 1;
3133 * Stop flow in RX_CP_TBL.
3136 * Pointer to Ethernet device.
3138 * Parent flow for wich copying is provided.
3141 flow_mreg_stop_copy_action(struct rte_eth_dev *dev,
3142 struct rte_flow *flow)
3144 struct mlx5_flow_mreg_copy_resource *mcp_res = flow->mreg_copy;
3146 if (!mcp_res || !flow->copy_applied)
3148 MLX5_ASSERT(mcp_res->appcnt);
3150 flow->copy_applied = 0;
3151 if (!mcp_res->appcnt)
3152 flow_drv_remove(dev, mcp_res->flow);
3156 * Remove the default copy action from RX_CP_TBL.
3159 * Pointer to Ethernet device.
3162 flow_mreg_del_default_copy_action(struct rte_eth_dev *dev)
3164 struct mlx5_flow_mreg_copy_resource *mcp_res;
3165 struct mlx5_priv *priv = dev->data->dev_private;
3167 /* Check if default flow is registered. */
3168 if (!priv->mreg_cp_tbl)
3170 mcp_res = (void *)mlx5_hlist_lookup(priv->mreg_cp_tbl,
3171 MLX5_DEFAULT_COPY_ID);
3174 MLX5_ASSERT(mcp_res->flow);
3175 flow_list_destroy(dev, NULL, mcp_res->flow);
3176 mlx5_hlist_remove(priv->mreg_cp_tbl, &mcp_res->hlist_ent);
3181 * Add the default copy action in in RX_CP_TBL.
3184 * Pointer to Ethernet device.
3186 * Perform verbose error reporting if not NULL.
3189 * 0 for success, negative value otherwise and rte_errno is set.
3192 flow_mreg_add_default_copy_action(struct rte_eth_dev *dev,
3193 struct rte_flow_error *error)
3195 struct mlx5_priv *priv = dev->data->dev_private;
3196 struct mlx5_flow_mreg_copy_resource *mcp_res;
3198 /* Check whether extensive metadata feature is engaged. */
3199 if (!priv->config.dv_flow_en ||
3200 priv->config.dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
3201 !mlx5_flow_ext_mreg_supported(dev) ||
3202 !priv->sh->dv_regc0_mask)
3204 mcp_res = flow_mreg_add_copy_action(dev, MLX5_DEFAULT_COPY_ID, error);
3211 * Add a flow of copying flow metadata registers in RX_CP_TBL.
3213 * All the flow having Q/RSS action should be split by
3214 * flow_mreg_split_qrss_prep() to pass by RX_CP_TBL. A flow in the RX_CP_TBL
3215 * performs the following,
3216 * - CQE->flow_tag := reg_c[1] (MARK)
3217 * - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
3218 * As CQE's flow_tag is not a register, it can't be simply copied from reg_c[1]
3219 * but there should be a flow per each MARK ID set by MARK action.
3221 * For the aforementioned reason, if there's a MARK action in flow's action
3222 * list, a corresponding flow should be added to the RX_CP_TBL in order to copy
3223 * the MARK ID to CQE's flow_tag like,
3224 * - If reg_c[1] is mark_id,
3225 * flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
3227 * For SET_META action which stores value in reg_c[0], as the destination is
3228 * also a flow metadata register (reg_b), adding a default flow is enough. Zero
3229 * MARK ID means the default flow. The default flow looks like,
3230 * - For all flow, reg_b := reg_c[0] and jump to RX_ACT_TBL.
3233 * Pointer to Ethernet device.
3235 * Pointer to flow structure.
3236 * @param[in] actions
3237 * Pointer to the list of actions.
3239 * Perform verbose error reporting if not NULL.
3242 * 0 on success, negative value otherwise and rte_errno is set.
3245 flow_mreg_update_copy_table(struct rte_eth_dev *dev,
3246 struct rte_flow *flow,
3247 const struct rte_flow_action *actions,
3248 struct rte_flow_error *error)
3250 struct mlx5_priv *priv = dev->data->dev_private;
3251 struct mlx5_dev_config *config = &priv->config;
3252 struct mlx5_flow_mreg_copy_resource *mcp_res;
3253 const struct rte_flow_action_mark *mark;
3255 /* Check whether extensive metadata feature is engaged. */
3256 if (!config->dv_flow_en ||
3257 config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
3258 !mlx5_flow_ext_mreg_supported(dev) ||
3259 !priv->sh->dv_regc0_mask)
3261 /* Find MARK action. */
3262 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3263 switch (actions->type) {
3264 case RTE_FLOW_ACTION_TYPE_FLAG:
3265 mcp_res = flow_mreg_add_copy_action
3266 (dev, MLX5_FLOW_MARK_DEFAULT, error);
3269 flow->mreg_copy = mcp_res;
3270 if (dev->data->dev_started) {
3272 flow->copy_applied = 1;
3275 case RTE_FLOW_ACTION_TYPE_MARK:
3276 mark = (const struct rte_flow_action_mark *)
3279 flow_mreg_add_copy_action(dev, mark->id, error);
3282 flow->mreg_copy = mcp_res;
3283 if (dev->data->dev_started) {
3285 flow->copy_applied = 1;
3295 #define MLX5_MAX_SPLIT_ACTIONS 24
3296 #define MLX5_MAX_SPLIT_ITEMS 24
3299 * Split the hairpin flow.
3300 * Since HW can't support encap on Rx we move the encap to Tx.
3301 * If the count action is after the encap then we also
3302 * move the count action. in this case the count will also measure
3306 * Pointer to Ethernet device.
3307 * @param[in] actions
3308 * Associated actions (list terminated by the END action).
3309 * @param[out] actions_rx
3311 * @param[out] actions_tx
3313 * @param[out] pattern_tx
3314 * The pattern items for the Tx flow.
3315 * @param[out] flow_id
3316 * The flow ID connected to this flow.
3322 flow_hairpin_split(struct rte_eth_dev *dev,
3323 const struct rte_flow_action actions[],
3324 struct rte_flow_action actions_rx[],
3325 struct rte_flow_action actions_tx[],
3326 struct rte_flow_item pattern_tx[],
3329 struct mlx5_priv *priv = dev->data->dev_private;
3330 const struct rte_flow_action_raw_encap *raw_encap;
3331 const struct rte_flow_action_raw_decap *raw_decap;
3332 struct mlx5_rte_flow_action_set_tag *set_tag;
3333 struct rte_flow_action *tag_action;
3334 struct mlx5_rte_flow_item_tag *tag_item;
3335 struct rte_flow_item *item;
3339 mlx5_flow_id_get(priv->sh->flow_id_pool, flow_id);
3340 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3341 switch (actions->type) {
3342 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
3343 case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
3344 rte_memcpy(actions_tx, actions,
3345 sizeof(struct rte_flow_action));
3348 case RTE_FLOW_ACTION_TYPE_COUNT:
3350 rte_memcpy(actions_tx, actions,
3351 sizeof(struct rte_flow_action));
3354 rte_memcpy(actions_rx, actions,
3355 sizeof(struct rte_flow_action));
3359 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3360 raw_encap = actions->conf;
3361 if (raw_encap->size >
3362 (sizeof(struct rte_flow_item_eth) +
3363 sizeof(struct rte_flow_item_ipv4))) {
3364 memcpy(actions_tx, actions,
3365 sizeof(struct rte_flow_action));
3369 rte_memcpy(actions_rx, actions,
3370 sizeof(struct rte_flow_action));
3374 case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
3375 raw_decap = actions->conf;
3376 if (raw_decap->size <
3377 (sizeof(struct rte_flow_item_eth) +
3378 sizeof(struct rte_flow_item_ipv4))) {
3379 memcpy(actions_tx, actions,
3380 sizeof(struct rte_flow_action));
3383 rte_memcpy(actions_rx, actions,
3384 sizeof(struct rte_flow_action));
3389 rte_memcpy(actions_rx, actions,
3390 sizeof(struct rte_flow_action));
3395 /* Add set meta action and end action for the Rx flow. */
3396 tag_action = actions_rx;
3397 tag_action->type = MLX5_RTE_FLOW_ACTION_TYPE_TAG;
3399 rte_memcpy(actions_rx, actions, sizeof(struct rte_flow_action));
3401 set_tag = (void *)actions_rx;
3402 set_tag->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_RX, 0, NULL);
3403 MLX5_ASSERT(set_tag->id > REG_NONE);
3404 set_tag->data = *flow_id;
3405 tag_action->conf = set_tag;
3406 /* Create Tx item list. */
3407 rte_memcpy(actions_tx, actions, sizeof(struct rte_flow_action));
3408 addr = (void *)&pattern_tx[2];
3410 item->type = MLX5_RTE_FLOW_ITEM_TYPE_TAG;
3411 tag_item = (void *)addr;
3412 tag_item->data = *flow_id;
3413 tag_item->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_TX, 0, NULL);
3414 MLX5_ASSERT(set_tag->id > REG_NONE);
3415 item->spec = tag_item;
3416 addr += sizeof(struct mlx5_rte_flow_item_tag);
3417 tag_item = (void *)addr;
3418 tag_item->data = UINT32_MAX;
3419 tag_item->id = UINT16_MAX;
3420 item->mask = tag_item;
3421 addr += sizeof(struct mlx5_rte_flow_item_tag);
3424 item->type = RTE_FLOW_ITEM_TYPE_END;
3429 * The last stage of splitting chain, just creates the subflow
3430 * without any modification.
3433 * Pointer to Ethernet device.
3435 * Parent flow structure pointer.
3436 * @param[in, out] sub_flow
3437 * Pointer to return the created subflow, may be NULL.
3438 * @param[in] prefix_layers
3439 * Prefix subflow layers, may be 0.
3441 * Flow rule attributes.
3443 * Pattern specification (list terminated by the END pattern item).
3444 * @param[in] actions
3445 * Associated actions (list terminated by the END action).
3446 * @param[in] external
3447 * This flow rule is created by request external to PMD.
3449 * Perform verbose error reporting if not NULL.
3451 * 0 on success, negative value otherwise
3454 flow_create_split_inner(struct rte_eth_dev *dev,
3455 struct rte_flow *flow,
3456 struct mlx5_flow **sub_flow,
3457 uint64_t prefix_layers,
3458 const struct rte_flow_attr *attr,
3459 const struct rte_flow_item items[],
3460 const struct rte_flow_action actions[],
3461 bool external, struct rte_flow_error *error)
3463 struct mlx5_flow *dev_flow;
3465 dev_flow = flow_drv_prepare(dev, flow, attr, items, actions, error);
3468 dev_flow->flow = flow;
3469 dev_flow->external = external;
3470 /* Subflow object was created, we must include one in the list. */
3471 SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
3472 dev_flow->handle, next);
3474 * If dev_flow is as one of the suffix flow, some actions in suffix
3475 * flow may need some user defined item layer flags.
3478 dev_flow->handle->layers = prefix_layers;
3480 *sub_flow = dev_flow;
3481 return flow_drv_translate(dev, dev_flow, attr, items, actions, error);
3485 * Split the meter flow.
3487 * As meter flow will split to three sub flow, other than meter
3488 * action, the other actions make sense to only meter accepts
3489 * the packet. If it need to be dropped, no other additional
3490 * actions should be take.
3492 * One kind of special action which decapsulates the L3 tunnel
3493 * header will be in the prefix sub flow, as not to take the
3494 * L3 tunnel header into account.
3497 * Pointer to Ethernet device.
3499 * Pattern specification (list terminated by the END pattern item).
3500 * @param[out] sfx_items
3501 * Suffix flow match items (list terminated by the END pattern item).
3502 * @param[in] actions
3503 * Associated actions (list terminated by the END action).
3504 * @param[out] actions_sfx
3505 * Suffix flow actions.
3506 * @param[out] actions_pre
3507 * Prefix flow actions.
3508 * @param[out] pattern_sfx
3509 * The pattern items for the suffix flow.
3510 * @param[out] tag_sfx
3511 * Pointer to suffix flow tag.
3517 flow_meter_split_prep(struct rte_eth_dev *dev,
3518 const struct rte_flow_item items[],
3519 struct rte_flow_item sfx_items[],
3520 const struct rte_flow_action actions[],
3521 struct rte_flow_action actions_sfx[],
3522 struct rte_flow_action actions_pre[])
3524 struct rte_flow_action *tag_action = NULL;
3525 struct rte_flow_item *tag_item;
3526 struct mlx5_rte_flow_action_set_tag *set_tag;
3527 struct rte_flow_error error;
3528 const struct rte_flow_action_raw_encap *raw_encap;
3529 const struct rte_flow_action_raw_decap *raw_decap;
3530 struct mlx5_rte_flow_item_tag *tag_spec;
3531 struct mlx5_rte_flow_item_tag *tag_mask;
3533 bool copy_vlan = false;
3535 /* Prepare the actions for prefix and suffix flow. */
3536 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3537 struct rte_flow_action **action_cur = NULL;
3539 switch (actions->type) {
3540 case RTE_FLOW_ACTION_TYPE_METER:
3541 /* Add the extra tag action first. */
3542 tag_action = actions_pre;
3543 tag_action->type = MLX5_RTE_FLOW_ACTION_TYPE_TAG;
3545 action_cur = &actions_pre;
3547 case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
3548 case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
3549 action_cur = &actions_pre;
3551 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3552 raw_encap = actions->conf;
3553 if (raw_encap->size < MLX5_ENCAPSULATION_DECISION_SIZE)
3554 action_cur = &actions_pre;
3556 case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
3557 raw_decap = actions->conf;
3558 if (raw_decap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
3559 action_cur = &actions_pre;
3561 case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
3562 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
3569 action_cur = &actions_sfx;
3570 memcpy(*action_cur, actions, sizeof(struct rte_flow_action));
3573 /* Add end action to the actions. */
3574 actions_sfx->type = RTE_FLOW_ACTION_TYPE_END;
3575 actions_pre->type = RTE_FLOW_ACTION_TYPE_END;
3578 set_tag = (void *)actions_pre;
3579 set_tag->id = mlx5_flow_get_reg_id(dev, MLX5_MTR_SFX, 0, &error);
3581 * Get the id from the qrss_pool to make qrss share the id with meter.
3583 tag_id = flow_qrss_get_id(dev);
3584 set_tag->data = tag_id << MLX5_MTR_COLOR_BITS;
3586 tag_action->conf = set_tag;
3587 /* Prepare the suffix subflow items. */
3588 tag_item = sfx_items++;
3589 for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
3590 int item_type = items->type;
3592 switch (item_type) {
3593 case RTE_FLOW_ITEM_TYPE_PORT_ID:
3594 memcpy(sfx_items, items, sizeof(*sfx_items));
3597 case RTE_FLOW_ITEM_TYPE_VLAN:
3599 memcpy(sfx_items, items, sizeof(*sfx_items));
3601 * Convert to internal match item, it is used
3602 * for vlan push and set vid.
3604 sfx_items->type = MLX5_RTE_FLOW_ITEM_TYPE_VLAN;
3612 sfx_items->type = RTE_FLOW_ITEM_TYPE_END;
3614 tag_spec = (struct mlx5_rte_flow_item_tag *)sfx_items;
3615 tag_spec->data = tag_id << MLX5_MTR_COLOR_BITS;
3616 tag_spec->id = mlx5_flow_get_reg_id(dev, MLX5_MTR_SFX, 0, &error);
3617 tag_mask = tag_spec + 1;
3618 tag_mask->data = 0xffffff00;
3619 tag_item->type = MLX5_RTE_FLOW_ITEM_TYPE_TAG;
3620 tag_item->spec = tag_spec;
3621 tag_item->last = NULL;
3622 tag_item->mask = tag_mask;
3627 * Split action list having QUEUE/RSS for metadata register copy.
3629 * Once Q/RSS action is detected in user's action list, the flow action
3630 * should be split in order to copy metadata registers, which will happen in
3632 * - CQE->flow_tag := reg_c[1] (MARK)
3633 * - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
3634 * The Q/RSS action will be performed on RX_ACT_TBL after passing by RX_CP_TBL.
3635 * This is because the last action of each flow must be a terminal action
3636 * (QUEUE, RSS or DROP).
3638 * Flow ID must be allocated to identify actions in the RX_ACT_TBL and it is
3639 * stored and kept in the mlx5_flow structure per each sub_flow.
3641 * The Q/RSS action is replaced with,
3642 * - SET_TAG, setting the allocated flow ID to reg_c[2].
3643 * And the following JUMP action is added at the end,
3644 * - JUMP, to RX_CP_TBL.
3646 * A flow to perform remained Q/RSS action will be created in RX_ACT_TBL by
3647 * flow_create_split_metadata() routine. The flow will look like,
3648 * - If flow ID matches (reg_c[2]), perform Q/RSS.
3651 * Pointer to Ethernet device.
3652 * @param[out] split_actions
3653 * Pointer to store split actions to jump to CP_TBL.
3654 * @param[in] actions
3655 * Pointer to the list of original flow actions.
3657 * Pointer to the Q/RSS action.
3658 * @param[in] actions_n
3659 * Number of original actions.
3661 * Perform verbose error reporting if not NULL.
3664 * non-zero unique flow_id on success, otherwise 0 and
3665 * error/rte_error are set.
3668 flow_mreg_split_qrss_prep(struct rte_eth_dev *dev,
3669 struct rte_flow_action *split_actions,
3670 const struct rte_flow_action *actions,
3671 const struct rte_flow_action *qrss,
3672 int actions_n, struct rte_flow_error *error)
3674 struct mlx5_rte_flow_action_set_tag *set_tag;
3675 struct rte_flow_action_jump *jump;
3676 const int qrss_idx = qrss - actions;
3677 uint32_t flow_id = 0;
3681 * Given actions will be split
3682 * - Replace QUEUE/RSS action with SET_TAG to set flow ID.
3683 * - Add jump to mreg CP_TBL.
3684 * As a result, there will be one more action.
3687 memcpy(split_actions, actions, sizeof(*split_actions) * actions_n);
3688 set_tag = (void *)(split_actions + actions_n);
3690 * If tag action is not set to void(it means we are not the meter
3691 * suffix flow), add the tag action. Since meter suffix flow already
3692 * has the tag added.
3694 if (split_actions[qrss_idx].type != RTE_FLOW_ACTION_TYPE_VOID) {
3696 * Allocate the new subflow ID. This one is unique within
3697 * device and not shared with representors. Otherwise,
3698 * we would have to resolve multi-thread access synch
3699 * issue. Each flow on the shared device is appended
3700 * with source vport identifier, so the resulting
3701 * flows will be unique in the shared (by master and
3702 * representors) domain even if they have coinciding
3705 flow_id = flow_qrss_get_id(dev);
3707 return rte_flow_error_set(error, ENOMEM,
3708 RTE_FLOW_ERROR_TYPE_ACTION,
3709 NULL, "can't allocate id "
3710 "for split Q/RSS subflow");
3711 /* Internal SET_TAG action to set flow ID. */
3712 *set_tag = (struct mlx5_rte_flow_action_set_tag){
3715 ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0, error);
3719 /* Construct new actions array. */
3720 /* Replace QUEUE/RSS action. */
3721 split_actions[qrss_idx] = (struct rte_flow_action){
3722 .type = MLX5_RTE_FLOW_ACTION_TYPE_TAG,
3726 /* JUMP action to jump to mreg copy table (CP_TBL). */
3727 jump = (void *)(set_tag + 1);
3728 *jump = (struct rte_flow_action_jump){
3729 .group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
3731 split_actions[actions_n - 2] = (struct rte_flow_action){
3732 .type = RTE_FLOW_ACTION_TYPE_JUMP,
3735 split_actions[actions_n - 1] = (struct rte_flow_action){
3736 .type = RTE_FLOW_ACTION_TYPE_END,
3742 * Extend the given action list for Tx metadata copy.
3744 * Copy the given action list to the ext_actions and add flow metadata register
3745 * copy action in order to copy reg_a set by WQE to reg_c[0].
3747 * @param[out] ext_actions
3748 * Pointer to the extended action list.
3749 * @param[in] actions
3750 * Pointer to the list of actions.
3751 * @param[in] actions_n
3752 * Number of actions in the list.
3754 * Perform verbose error reporting if not NULL.
3755 * @param[in] encap_idx
3756 * The encap action inndex.
3759 * 0 on success, negative value otherwise
3762 flow_mreg_tx_copy_prep(struct rte_eth_dev *dev,
3763 struct rte_flow_action *ext_actions,
3764 const struct rte_flow_action *actions,
3765 int actions_n, struct rte_flow_error *error,
3768 struct mlx5_flow_action_copy_mreg *cp_mreg =
3769 (struct mlx5_flow_action_copy_mreg *)
3770 (ext_actions + actions_n + 1);
3773 ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
3777 ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_TX, 0, error);
3782 memcpy(ext_actions, actions, sizeof(*ext_actions) * encap_idx);
3783 if (encap_idx == actions_n - 1) {
3784 ext_actions[actions_n - 1] = (struct rte_flow_action){
3785 .type = MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3788 ext_actions[actions_n] = (struct rte_flow_action){
3789 .type = RTE_FLOW_ACTION_TYPE_END,
3792 ext_actions[encap_idx] = (struct rte_flow_action){
3793 .type = MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3796 memcpy(ext_actions + encap_idx + 1, actions + encap_idx,
3797 sizeof(*ext_actions) * (actions_n - encap_idx));
3803 * The splitting for metadata feature.
3805 * - Q/RSS action on NIC Rx should be split in order to pass by
3806 * the mreg copy table (RX_CP_TBL) and then it jumps to the
3807 * action table (RX_ACT_TBL) which has the split Q/RSS action.
3809 * - All the actions on NIC Tx should have a mreg copy action to
3810 * copy reg_a from WQE to reg_c[0].
3813 * Pointer to Ethernet device.
3815 * Parent flow structure pointer.
3816 * @param[in] prefix_layers
3817 * Prefix flow layer flags.
3819 * Flow rule attributes.
3821 * Pattern specification (list terminated by the END pattern item).
3822 * @param[in] actions
3823 * Associated actions (list terminated by the END action).
3824 * @param[in] external
3825 * This flow rule is created by request external to PMD.
3827 * Perform verbose error reporting if not NULL.
3829 * 0 on success, negative value otherwise
3832 flow_create_split_metadata(struct rte_eth_dev *dev,
3833 struct rte_flow *flow,
3834 uint64_t prefix_layers,
3835 const struct rte_flow_attr *attr,
3836 const struct rte_flow_item items[],
3837 const struct rte_flow_action actions[],
3838 bool external, struct rte_flow_error *error)
3840 struct mlx5_priv *priv = dev->data->dev_private;
3841 struct mlx5_dev_config *config = &priv->config;
3842 const struct rte_flow_action *qrss = NULL;
3843 struct rte_flow_action *ext_actions = NULL;
3844 struct mlx5_flow *dev_flow = NULL;
3845 uint32_t qrss_id = 0;
3852 /* Check whether extensive metadata feature is engaged. */
3853 if (!config->dv_flow_en ||
3854 config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
3855 !mlx5_flow_ext_mreg_supported(dev))
3856 return flow_create_split_inner(dev, flow, NULL, prefix_layers,
3857 attr, items, actions, external,
3859 actions_n = flow_parse_metadata_split_actions_info(actions, &qrss,
3862 /* Exclude hairpin flows from splitting. */
3863 if (qrss->type == RTE_FLOW_ACTION_TYPE_QUEUE) {
3864 const struct rte_flow_action_queue *queue;
3867 if (mlx5_rxq_get_type(dev, queue->index) ==
3868 MLX5_RXQ_TYPE_HAIRPIN)
3870 } else if (qrss->type == RTE_FLOW_ACTION_TYPE_RSS) {
3871 const struct rte_flow_action_rss *rss;
3874 if (mlx5_rxq_get_type(dev, rss->queue[0]) ==
3875 MLX5_RXQ_TYPE_HAIRPIN)
3880 /* Check if it is in meter suffix table. */
3881 mtr_sfx = attr->group == (attr->transfer ?
3882 (MLX5_FLOW_TABLE_LEVEL_SUFFIX - 1) :
3883 MLX5_FLOW_TABLE_LEVEL_SUFFIX);
3885 * Q/RSS action on NIC Rx should be split in order to pass by
3886 * the mreg copy table (RX_CP_TBL) and then it jumps to the
3887 * action table (RX_ACT_TBL) which has the split Q/RSS action.
3889 act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
3890 sizeof(struct rte_flow_action_set_tag) +
3891 sizeof(struct rte_flow_action_jump);
3892 ext_actions = rte_zmalloc(__func__, act_size, 0);
3894 return rte_flow_error_set(error, ENOMEM,
3895 RTE_FLOW_ERROR_TYPE_ACTION,
3896 NULL, "no memory to split "
3899 * If we are the suffix flow of meter, tag already exist.
3900 * Set the tag action to void.
3903 ext_actions[qrss - actions].type =
3904 RTE_FLOW_ACTION_TYPE_VOID;
3906 ext_actions[qrss - actions].type =
3907 MLX5_RTE_FLOW_ACTION_TYPE_TAG;
3909 * Create the new actions list with removed Q/RSS action
3910 * and appended set tag and jump to register copy table
3911 * (RX_CP_TBL). We should preallocate unique tag ID here
3912 * in advance, because it is needed for set tag action.
3914 qrss_id = flow_mreg_split_qrss_prep(dev, ext_actions, actions,
3915 qrss, actions_n, error);
3916 if (!mtr_sfx && !qrss_id) {
3920 } else if (attr->egress && !attr->transfer) {
3922 * All the actions on NIC Tx should have a metadata register
3923 * copy action to copy reg_a from WQE to reg_c[meta]
3925 act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
3926 sizeof(struct mlx5_flow_action_copy_mreg);
3927 ext_actions = rte_zmalloc(__func__, act_size, 0);
3929 return rte_flow_error_set(error, ENOMEM,
3930 RTE_FLOW_ERROR_TYPE_ACTION,
3931 NULL, "no memory to split "
3933 /* Create the action list appended with copy register. */
3934 ret = flow_mreg_tx_copy_prep(dev, ext_actions, actions,
3935 actions_n, error, encap_idx);
3939 /* Add the unmodified original or prefix subflow. */
3940 ret = flow_create_split_inner(dev, flow, &dev_flow, prefix_layers, attr,
3941 items, ext_actions ? ext_actions :
3942 actions, external, error);
3945 MLX5_ASSERT(dev_flow);
3947 const struct rte_flow_attr q_attr = {
3948 .group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
3951 /* Internal PMD action to set register. */
3952 struct mlx5_rte_flow_item_tag q_tag_spec = {
3956 struct rte_flow_item q_items[] = {
3958 .type = MLX5_RTE_FLOW_ITEM_TYPE_TAG,
3959 .spec = &q_tag_spec,
3964 .type = RTE_FLOW_ITEM_TYPE_END,
3967 struct rte_flow_action q_actions[] = {
3973 .type = RTE_FLOW_ACTION_TYPE_END,
3976 uint64_t layers = flow_get_prefix_layer_flags(dev_flow);
3979 * Configure the tag item only if there is no meter subflow.
3980 * Since tag is already marked in the meter suffix subflow
3981 * we can just use the meter suffix items as is.
3984 /* Not meter subflow. */
3985 MLX5_ASSERT(!mtr_sfx);
3987 * Put unique id in prefix flow due to it is destroyed
3988 * after suffix flow and id will be freed after there
3989 * is no actual flows with this id and identifier
3990 * reallocation becomes possible (for example, for
3991 * other flows in other threads).
3993 dev_flow->handle->qrss_id = qrss_id;
3994 ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0,
3998 q_tag_spec.id = ret;
4001 /* Add suffix subflow to execute Q/RSS. */
4002 ret = flow_create_split_inner(dev, flow, &dev_flow, layers,
4003 &q_attr, mtr_sfx ? items :
4008 /* qrss ID should be freed if failed. */
4010 MLX5_ASSERT(dev_flow);
4015 * We do not destroy the partially created sub_flows in case of error.
4016 * These ones are included into parent flow list and will be destroyed
4017 * by flow_drv_destroy.
4019 flow_qrss_free_id(dev, qrss_id);
4020 rte_free(ext_actions);
4025 * The splitting for meter feature.
4027 * - The meter flow will be split to two flows as prefix and
4028 * suffix flow. The packets make sense only it pass the prefix
4031 * - Reg_C_5 is used for the packet to match betweend prefix and
4035 * Pointer to Ethernet device.
4037 * Parent flow structure pointer.
4039 * Flow rule attributes.
4041 * Pattern specification (list terminated by the END pattern item).
4042 * @param[in] actions
4043 * Associated actions (list terminated by the END action).
4044 * @param[in] external
4045 * This flow rule is created by request external to PMD.
4047 * Perform verbose error reporting if not NULL.
4049 * 0 on success, negative value otherwise
4052 flow_create_split_meter(struct rte_eth_dev *dev,
4053 struct rte_flow *flow,
4054 const struct rte_flow_attr *attr,
4055 const struct rte_flow_item items[],
4056 const struct rte_flow_action actions[],
4057 bool external, struct rte_flow_error *error)
4059 struct mlx5_priv *priv = dev->data->dev_private;
4060 struct rte_flow_action *sfx_actions = NULL;
4061 struct rte_flow_action *pre_actions = NULL;
4062 struct rte_flow_item *sfx_items = NULL;
4063 struct mlx5_flow *dev_flow = NULL;
4064 struct rte_flow_attr sfx_attr = *attr;
4066 uint32_t mtr_tag_id = 0;
4073 actions_n = flow_check_meter_action(actions, &mtr);
4075 /* The five prefix actions: meter, decap, encap, tag, end. */
4076 act_size = sizeof(struct rte_flow_action) * (actions_n + 5) +
4077 sizeof(struct mlx5_rte_flow_action_set_tag);
4078 /* tag, vlan, port id, end. */
4079 #define METER_SUFFIX_ITEM 4
4080 item_size = sizeof(struct rte_flow_item) * METER_SUFFIX_ITEM +
4081 sizeof(struct mlx5_rte_flow_item_tag) * 2;
4082 sfx_actions = rte_zmalloc(__func__, (act_size + item_size), 0);
4084 return rte_flow_error_set(error, ENOMEM,
4085 RTE_FLOW_ERROR_TYPE_ACTION,
4086 NULL, "no memory to split "
4088 sfx_items = (struct rte_flow_item *)((char *)sfx_actions +
4090 pre_actions = sfx_actions + actions_n;
4091 mtr_tag_id = flow_meter_split_prep(dev, items, sfx_items,
4092 actions, sfx_actions,
4098 /* Add the prefix subflow. */
4099 ret = flow_create_split_inner(dev, flow, &dev_flow, 0, attr,
4100 items, pre_actions, external,
4106 dev_flow->handle->mtr_flow_id = mtr_tag_id;
4107 /* Setting the sfx group atrr. */
4108 sfx_attr.group = sfx_attr.transfer ?
4109 (MLX5_FLOW_TABLE_LEVEL_SUFFIX - 1) :
4110 MLX5_FLOW_TABLE_LEVEL_SUFFIX;
4112 /* Add the prefix subflow. */
4113 ret = flow_create_split_metadata(dev, flow, dev_flow ?
4114 flow_get_prefix_layer_flags(dev_flow) :
4116 sfx_items ? sfx_items : items,
4117 sfx_actions ? sfx_actions : actions,
4121 rte_free(sfx_actions);
4126 * Split the flow to subflow set. The splitters might be linked
4127 * in the chain, like this:
4128 * flow_create_split_outer() calls:
4129 * flow_create_split_meter() calls:
4130 * flow_create_split_metadata(meter_subflow_0) calls:
4131 * flow_create_split_inner(metadata_subflow_0)
4132 * flow_create_split_inner(metadata_subflow_1)
4133 * flow_create_split_inner(metadata_subflow_2)
4134 * flow_create_split_metadata(meter_subflow_1) calls:
4135 * flow_create_split_inner(metadata_subflow_0)
4136 * flow_create_split_inner(metadata_subflow_1)
4137 * flow_create_split_inner(metadata_subflow_2)
4139 * This provide flexible way to add new levels of flow splitting.
4140 * The all of successfully created subflows are included to the
4141 * parent flow dev_flow list.
4144 * Pointer to Ethernet device.
4146 * Parent flow structure pointer.
4148 * Flow rule attributes.
4150 * Pattern specification (list terminated by the END pattern item).
4151 * @param[in] actions
4152 * Associated actions (list terminated by the END action).
4153 * @param[in] external
4154 * This flow rule is created by request external to PMD.
4156 * Perform verbose error reporting if not NULL.
4158 * 0 on success, negative value otherwise
4161 flow_create_split_outer(struct rte_eth_dev *dev,
4162 struct rte_flow *flow,
4163 const struct rte_flow_attr *attr,
4164 const struct rte_flow_item items[],
4165 const struct rte_flow_action actions[],
4166 bool external, struct rte_flow_error *error)
4170 ret = flow_create_split_meter(dev, flow, attr, items,
4171 actions, external, error);
4172 MLX5_ASSERT(ret <= 0);
4177 * Create a flow and add it to @p list.
4180 * Pointer to Ethernet device.
4182 * Pointer to a TAILQ flow list. If this parameter NULL,
4183 * no list insertion occurred, flow is just created,
4184 * this is caller's responsibility to track the
4187 * Flow rule attributes.
4189 * Pattern specification (list terminated by the END pattern item).
4190 * @param[in] actions
4191 * Associated actions (list terminated by the END action).
4192 * @param[in] external
4193 * This flow rule is created by request external to PMD.
4195 * Perform verbose error reporting if not NULL.
4198 * A flow on success, NULL otherwise and rte_errno is set.
4200 static struct rte_flow *
4201 flow_list_create(struct rte_eth_dev *dev, struct mlx5_flows *list,
4202 const struct rte_flow_attr *attr,
4203 const struct rte_flow_item items[],
4204 const struct rte_flow_action actions[],
4205 bool external, struct rte_flow_error *error)
4207 struct mlx5_priv *priv = dev->data->dev_private;
4208 struct rte_flow *flow = NULL;
4209 struct mlx5_flow *dev_flow;
4210 const struct rte_flow_action_rss *rss;
4212 struct rte_flow_expand_rss buf;
4213 uint8_t buffer[2048];
4216 struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
4217 uint8_t buffer[2048];
4220 struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
4221 uint8_t buffer[2048];
4222 } actions_hairpin_tx;
4224 struct rte_flow_item items[MLX5_MAX_SPLIT_ITEMS];
4225 uint8_t buffer[2048];
4227 struct rte_flow_expand_rss *buf = &expand_buffer.buf;
4228 const struct rte_flow_action *p_actions_rx = actions;
4231 int hairpin_flow = 0;
4232 uint32_t hairpin_id = 0;
4233 struct rte_flow_attr attr_tx = { .priority = 0 };
4234 int ret = flow_drv_validate(dev, attr, items, p_actions_rx, external,
4239 hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
4240 if (hairpin_flow > 0) {
4241 if (hairpin_flow > MLX5_MAX_SPLIT_ACTIONS) {
4245 flow_hairpin_split(dev, actions, actions_rx.actions,
4246 actions_hairpin_tx.actions, items_tx.items,
4248 p_actions_rx = actions_rx.actions;
4250 flow_size = sizeof(struct rte_flow);
4251 rss = flow_get_rss_action(p_actions_rx);
4253 flow_size += RTE_ALIGN_CEIL(rss->queue_num * sizeof(uint16_t),
4256 flow_size += RTE_ALIGN_CEIL(sizeof(uint16_t), sizeof(void *));
4257 flow = rte_calloc(__func__, 1, flow_size, 0);
4260 goto error_before_flow;
4262 flow->drv_type = flow_get_drv_type(dev, attr);
4263 if (hairpin_id != 0)
4264 flow->hairpin_flow_id = hairpin_id;
4265 MLX5_ASSERT(flow->drv_type > MLX5_FLOW_TYPE_MIN &&
4266 flow->drv_type < MLX5_FLOW_TYPE_MAX);
4267 flow->rss.queue = (void *)(flow + 1);
4270 * The following information is required by
4271 * mlx5_flow_hashfields_adjust() in advance.
4273 flow->rss.level = rss->level;
4274 /* RSS type 0 indicates default RSS type (ETH_RSS_IP). */
4275 flow->rss.types = !rss->types ? ETH_RSS_IP : rss->types;
4277 flow->dev_handles = 0;
4278 if (rss && rss->types) {
4279 unsigned int graph_root;
4281 graph_root = find_graph_root(items, rss->level);
4282 ret = rte_flow_expand_rss(buf, sizeof(expand_buffer.buffer),
4284 mlx5_support_expansion,
4286 MLX5_ASSERT(ret > 0 &&
4287 (unsigned int)ret < sizeof(expand_buffer.buffer));
4290 buf->entry[0].pattern = (void *)(uintptr_t)items;
4293 * Record the start index when there is a nested call. All sub-flows
4294 * need to be translated before another calling.
4295 * No need to use ping-pong buffer to save memory here.
4297 if (priv->flow_idx) {
4298 MLX5_ASSERT(!priv->flow_nested_idx);
4299 priv->flow_nested_idx = priv->flow_idx;
4301 for (i = 0; i < buf->entries; ++i) {
4303 * The splitter may create multiple dev_flows,
4304 * depending on configuration. In the simplest
4305 * case it just creates unmodified original flow.
4307 ret = flow_create_split_outer(dev, flow, attr,
4308 buf->entry[i].pattern,
4309 p_actions_rx, external,
4314 /* Create the tx flow. */
4316 attr_tx.group = MLX5_HAIRPIN_TX_TABLE;
4317 attr_tx.ingress = 0;
4319 dev_flow = flow_drv_prepare(dev, flow, &attr_tx, items_tx.items,
4320 actions_hairpin_tx.actions, error);
4323 dev_flow->flow = flow;
4324 dev_flow->external = 0;
4325 SILIST_INSERT(&flow->dev_handles, dev_flow->handle_idx,
4326 dev_flow->handle, next);
4327 ret = flow_drv_translate(dev, dev_flow, &attr_tx,
4329 actions_hairpin_tx.actions, error);
4334 * Update the metadata register copy table. If extensive
4335 * metadata feature is enabled and registers are supported
4336 * we might create the extra rte_flow for each unique
4337 * MARK/FLAG action ID.
4339 * The table is updated for ingress Flows only, because
4340 * the egress Flows belong to the different device and
4341 * copy table should be updated in peer NIC Rx domain.
4343 if (attr->ingress &&
4344 (external || attr->group != MLX5_FLOW_MREG_CP_TABLE_GROUP)) {
4345 ret = flow_mreg_update_copy_table(dev, flow, actions, error);
4350 * If the flow is external (from application) OR device is started, then
4351 * the flow will be applied immediately.
4353 if (external || dev->data->dev_started) {
4354 ret = flow_drv_apply(dev, flow, error);
4359 TAILQ_INSERT_TAIL(list, flow, next);
4360 flow_rxq_flags_set(dev, flow);
4361 /* Nested flow creation index recovery. */
4362 priv->flow_idx = priv->flow_nested_idx;
4363 if (priv->flow_nested_idx)
4364 priv->flow_nested_idx = 0;
4368 ret = rte_errno; /* Save rte_errno before cleanup. */
4369 flow_mreg_del_copy_action(dev, flow);
4370 flow_drv_destroy(dev, flow);
4372 rte_errno = ret; /* Restore rte_errno. */
4376 mlx5_flow_id_release(priv->sh->flow_id_pool,
4379 priv->flow_idx = priv->flow_nested_idx;
4380 if (priv->flow_nested_idx)
4381 priv->flow_nested_idx = 0;
4386 * Create a dedicated flow rule on e-switch table 0 (root table), to direct all
4387 * incoming packets to table 1.
4389 * Other flow rules, requested for group n, will be created in
4390 * e-switch table n+1.
4391 * Jump action to e-switch group n will be created to group n+1.
4393 * Used when working in switchdev mode, to utilise advantages of table 1
4397 * Pointer to Ethernet device.
4400 * Pointer to flow on success, NULL otherwise and rte_errno is set.
4403 mlx5_flow_create_esw_table_zero_flow(struct rte_eth_dev *dev)
4405 const struct rte_flow_attr attr = {
4412 const struct rte_flow_item pattern = {
4413 .type = RTE_FLOW_ITEM_TYPE_END,
4415 struct rte_flow_action_jump jump = {
4418 const struct rte_flow_action actions[] = {
4420 .type = RTE_FLOW_ACTION_TYPE_JUMP,
4424 .type = RTE_FLOW_ACTION_TYPE_END,
4427 struct mlx5_priv *priv = dev->data->dev_private;
4428 struct rte_flow_error error;
4430 return flow_list_create(dev, &priv->ctrl_flows, &attr, &pattern,
4431 actions, false, &error);
4437 * @see rte_flow_create()
4441 mlx5_flow_create(struct rte_eth_dev *dev,
4442 const struct rte_flow_attr *attr,
4443 const struct rte_flow_item items[],
4444 const struct rte_flow_action actions[],
4445 struct rte_flow_error *error)
4447 struct mlx5_priv *priv = dev->data->dev_private;
4450 * If the device is not started yet, it is not allowed to created a
4451 * flow from application. PMD default flows and traffic control flows
4454 if (unlikely(!dev->data->dev_started)) {
4456 DRV_LOG(DEBUG, "port %u is not started when "
4457 "inserting a flow", dev->data->port_id);
4460 return flow_list_create(dev, &priv->flows,
4461 attr, items, actions, true, error);
4465 * Destroy a flow in a list.
4468 * Pointer to Ethernet device.
4470 * Pointer to a TAILQ flow list. If this parameter NULL,
4471 * there is no flow removal from the list.
4476 flow_list_destroy(struct rte_eth_dev *dev, struct mlx5_flows *list,
4477 struct rte_flow *flow)
4479 struct mlx5_priv *priv = dev->data->dev_private;
4482 * Update RX queue flags only if port is started, otherwise it is
4485 if (dev->data->dev_started)
4486 flow_rxq_flags_trim(dev, flow);
4487 if (flow->hairpin_flow_id)
4488 mlx5_flow_id_release(priv->sh->flow_id_pool,
4489 flow->hairpin_flow_id);
4490 flow_drv_destroy(dev, flow);
4492 TAILQ_REMOVE(list, flow, next);
4493 flow_mreg_del_copy_action(dev, flow);
4494 rte_free(flow->fdir);
4499 * Destroy all flows.
4502 * Pointer to Ethernet device.
4504 * Pointer to a TAILQ flow list.
4506 * If flushing is called avtively.
4509 mlx5_flow_list_flush(struct rte_eth_dev *dev, struct mlx5_flows *list,
4512 uint32_t num_flushed = 0;
4514 while (!TAILQ_EMPTY(list)) {
4515 struct rte_flow *flow;
4517 flow = TAILQ_FIRST(list);
4518 flow_list_destroy(dev, list, flow);
4522 DRV_LOG(INFO, "port %u: %u flows flushed before stopping",
4523 dev->data->port_id, num_flushed);
4531 * Pointer to Ethernet device.
4533 * Pointer to a TAILQ flow list.
4536 mlx5_flow_stop(struct rte_eth_dev *dev, struct mlx5_flows *list)
4538 struct rte_flow *flow;
4540 TAILQ_FOREACH_REVERSE(flow, list, mlx5_flows, next) {
4541 flow_drv_remove(dev, flow);
4542 flow_mreg_stop_copy_action(dev, flow);
4544 flow_mreg_del_default_copy_action(dev);
4545 flow_rxq_flags_clear(dev);
4552 * Pointer to Ethernet device.
4554 * Pointer to a TAILQ flow list.
4557 * 0 on success, a negative errno value otherwise and rte_errno is set.
4560 mlx5_flow_start(struct rte_eth_dev *dev, struct mlx5_flows *list)
4562 struct rte_flow *flow;
4563 struct rte_flow_error error;
4566 /* Make sure default copy action (reg_c[0] -> reg_b) is created. */
4567 ret = flow_mreg_add_default_copy_action(dev, &error);
4570 /* Apply Flows created by application. */
4571 TAILQ_FOREACH(flow, list, next) {
4572 ret = flow_mreg_start_copy_action(dev, flow);
4575 ret = flow_drv_apply(dev, flow, &error);
4578 flow_rxq_flags_set(dev, flow);
4582 ret = rte_errno; /* Save rte_errno before cleanup. */
4583 mlx5_flow_stop(dev, list);
4584 rte_errno = ret; /* Restore rte_errno. */
4589 * Stop all default actions for flows.
4592 * Pointer to Ethernet device.
4595 mlx5_flow_stop_default(struct rte_eth_dev *dev)
4597 flow_mreg_del_default_copy_action(dev);
4601 * Start all default actions for flows.
4604 * Pointer to Ethernet device.
4606 * 0 on success, a negative errno value otherwise and rte_errno is set.
4609 mlx5_flow_start_default(struct rte_eth_dev *dev)
4611 struct rte_flow_error error;
4613 /* Make sure default copy action (reg_c[0] -> reg_b) is created. */
4614 return flow_mreg_add_default_copy_action(dev, &error);
4618 * Allocate intermediate resources for flow creation.
4621 * Pointer to Ethernet device.
4624 mlx5_flow_alloc_intermediate(struct rte_eth_dev *dev)
4626 struct mlx5_priv *priv = dev->data->dev_private;
4628 if (!priv->inter_flows)
4629 priv->inter_flows = rte_calloc(__func__, MLX5_NUM_MAX_DEV_FLOWS,
4630 sizeof(struct mlx5_flow), 0);
4631 /* Reset the index. */
4633 priv->flow_nested_idx = 0;
4637 * Free intermediate resources for flows.
4640 * Pointer to Ethernet device.
4643 mlx5_flow_free_intermediate(struct rte_eth_dev *dev)
4645 struct mlx5_priv *priv = dev->data->dev_private;
4647 rte_free(priv->inter_flows);
4648 priv->inter_flows = NULL;
4652 * Verify the flow list is empty
4655 * Pointer to Ethernet device.
4657 * @return the number of flows not released.
4660 mlx5_flow_verify(struct rte_eth_dev *dev)
4662 struct mlx5_priv *priv = dev->data->dev_private;
4663 struct rte_flow *flow;
4666 TAILQ_FOREACH(flow, &priv->flows, next) {
4667 DRV_LOG(DEBUG, "port %u flow %p still referenced",
4668 dev->data->port_id, (void *)flow);
4675 * Enable default hairpin egress flow.
4678 * Pointer to Ethernet device.
4683 * 0 on success, a negative errno value otherwise and rte_errno is set.
4686 mlx5_ctrl_flow_source_queue(struct rte_eth_dev *dev,
4689 struct mlx5_priv *priv = dev->data->dev_private;
4690 const struct rte_flow_attr attr = {
4694 struct mlx5_rte_flow_item_tx_queue queue_spec = {
4697 struct mlx5_rte_flow_item_tx_queue queue_mask = {
4698 .queue = UINT32_MAX,
4700 struct rte_flow_item items[] = {
4702 .type = MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE,
4703 .spec = &queue_spec,
4705 .mask = &queue_mask,
4708 .type = RTE_FLOW_ITEM_TYPE_END,
4711 struct rte_flow_action_jump jump = {
4712 .group = MLX5_HAIRPIN_TX_TABLE,
4714 struct rte_flow_action actions[2];
4715 struct rte_flow *flow;
4716 struct rte_flow_error error;
4718 actions[0].type = RTE_FLOW_ACTION_TYPE_JUMP;
4719 actions[0].conf = &jump;
4720 actions[1].type = RTE_FLOW_ACTION_TYPE_END;
4721 flow = flow_list_create(dev, &priv->ctrl_flows,
4722 &attr, items, actions, false, &error);
4725 "Failed to create ctrl flow: rte_errno(%d),"
4726 " type(%d), message(%s)",
4727 rte_errno, error.type,
4728 error.message ? error.message : " (no stated reason)");
4735 * Enable a control flow configured from the control plane.
4738 * Pointer to Ethernet device.
4740 * An Ethernet flow spec to apply.
4742 * An Ethernet flow mask to apply.
4744 * A VLAN flow spec to apply.
4746 * A VLAN flow mask to apply.
4749 * 0 on success, a negative errno value otherwise and rte_errno is set.
4752 mlx5_ctrl_flow_vlan(struct rte_eth_dev *dev,
4753 struct rte_flow_item_eth *eth_spec,
4754 struct rte_flow_item_eth *eth_mask,
4755 struct rte_flow_item_vlan *vlan_spec,
4756 struct rte_flow_item_vlan *vlan_mask)
4758 struct mlx5_priv *priv = dev->data->dev_private;
4759 const struct rte_flow_attr attr = {
4761 .priority = MLX5_FLOW_PRIO_RSVD,
4763 struct rte_flow_item items[] = {
4765 .type = RTE_FLOW_ITEM_TYPE_ETH,
4771 .type = (vlan_spec) ? RTE_FLOW_ITEM_TYPE_VLAN :
4772 RTE_FLOW_ITEM_TYPE_END,
4778 .type = RTE_FLOW_ITEM_TYPE_END,
4781 uint16_t queue[priv->reta_idx_n];
4782 struct rte_flow_action_rss action_rss = {
4783 .func = RTE_ETH_HASH_FUNCTION_DEFAULT,
4785 .types = priv->rss_conf.rss_hf,
4786 .key_len = priv->rss_conf.rss_key_len,
4787 .queue_num = priv->reta_idx_n,
4788 .key = priv->rss_conf.rss_key,
4791 struct rte_flow_action actions[] = {
4793 .type = RTE_FLOW_ACTION_TYPE_RSS,
4794 .conf = &action_rss,
4797 .type = RTE_FLOW_ACTION_TYPE_END,
4800 struct rte_flow *flow;
4801 struct rte_flow_error error;
4804 if (!priv->reta_idx_n || !priv->rxqs_n) {
4807 for (i = 0; i != priv->reta_idx_n; ++i)
4808 queue[i] = (*priv->reta_idx)[i];
4809 flow = flow_list_create(dev, &priv->ctrl_flows,
4810 &attr, items, actions, false, &error);
4817 * Enable a flow control configured from the control plane.
4820 * Pointer to Ethernet device.
4822 * An Ethernet flow spec to apply.
4824 * An Ethernet flow mask to apply.
4827 * 0 on success, a negative errno value otherwise and rte_errno is set.
4830 mlx5_ctrl_flow(struct rte_eth_dev *dev,
4831 struct rte_flow_item_eth *eth_spec,
4832 struct rte_flow_item_eth *eth_mask)
4834 return mlx5_ctrl_flow_vlan(dev, eth_spec, eth_mask, NULL, NULL);
4840 * @see rte_flow_destroy()
4844 mlx5_flow_destroy(struct rte_eth_dev *dev,
4845 struct rte_flow *flow,
4846 struct rte_flow_error *error __rte_unused)
4848 struct mlx5_priv *priv = dev->data->dev_private;
4850 flow_list_destroy(dev, &priv->flows, flow);
4855 * Destroy all flows.
4857 * @see rte_flow_flush()
4861 mlx5_flow_flush(struct rte_eth_dev *dev,
4862 struct rte_flow_error *error __rte_unused)
4864 struct mlx5_priv *priv = dev->data->dev_private;
4866 mlx5_flow_list_flush(dev, &priv->flows, false);
4873 * @see rte_flow_isolate()
4877 mlx5_flow_isolate(struct rte_eth_dev *dev,
4879 struct rte_flow_error *error)
4881 struct mlx5_priv *priv = dev->data->dev_private;
4883 if (dev->data->dev_started) {
4884 rte_flow_error_set(error, EBUSY,
4885 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
4887 "port must be stopped first");
4890 priv->isolated = !!enable;
4892 dev->dev_ops = &mlx5_dev_ops_isolate;
4894 dev->dev_ops = &mlx5_dev_ops;
4901 * @see rte_flow_query()
4905 flow_drv_query(struct rte_eth_dev *dev,
4906 struct rte_flow *flow,
4907 const struct rte_flow_action *actions,
4909 struct rte_flow_error *error)
4911 const struct mlx5_flow_driver_ops *fops;
4912 enum mlx5_flow_drv_type ftype = flow->drv_type;
4914 MLX5_ASSERT(ftype > MLX5_FLOW_TYPE_MIN && ftype < MLX5_FLOW_TYPE_MAX);
4915 fops = flow_get_drv_ops(ftype);
4917 return fops->query(dev, flow, actions, data, error);
4923 * @see rte_flow_query()
4927 mlx5_flow_query(struct rte_eth_dev *dev,
4928 struct rte_flow *flow,
4929 const struct rte_flow_action *actions,
4931 struct rte_flow_error *error)
4935 ret = flow_drv_query(dev, flow, actions, data, error);
4942 * Convert a flow director filter to a generic flow.
4945 * Pointer to Ethernet device.
4946 * @param fdir_filter
4947 * Flow director filter to add.
4949 * Generic flow parameters structure.
4952 * 0 on success, a negative errno value otherwise and rte_errno is set.
4955 flow_fdir_filter_convert(struct rte_eth_dev *dev,
4956 const struct rte_eth_fdir_filter *fdir_filter,
4957 struct mlx5_fdir *attributes)
4959 struct mlx5_priv *priv = dev->data->dev_private;
4960 const struct rte_eth_fdir_input *input = &fdir_filter->input;
4961 const struct rte_eth_fdir_masks *mask =
4962 &dev->data->dev_conf.fdir_conf.mask;
4964 /* Validate queue number. */
4965 if (fdir_filter->action.rx_queue >= priv->rxqs_n) {
4966 DRV_LOG(ERR, "port %u invalid queue number %d",
4967 dev->data->port_id, fdir_filter->action.rx_queue);
4971 attributes->attr.ingress = 1;
4972 attributes->items[0] = (struct rte_flow_item) {
4973 .type = RTE_FLOW_ITEM_TYPE_ETH,
4974 .spec = &attributes->l2,
4975 .mask = &attributes->l2_mask,
4977 switch (fdir_filter->action.behavior) {
4978 case RTE_ETH_FDIR_ACCEPT:
4979 attributes->actions[0] = (struct rte_flow_action){
4980 .type = RTE_FLOW_ACTION_TYPE_QUEUE,
4981 .conf = &attributes->queue,
4984 case RTE_ETH_FDIR_REJECT:
4985 attributes->actions[0] = (struct rte_flow_action){
4986 .type = RTE_FLOW_ACTION_TYPE_DROP,
4990 DRV_LOG(ERR, "port %u invalid behavior %d",
4992 fdir_filter->action.behavior);
4993 rte_errno = ENOTSUP;
4996 attributes->queue.index = fdir_filter->action.rx_queue;
4998 switch (fdir_filter->input.flow_type) {
4999 case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
5000 case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
5001 case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
5002 attributes->l3.ipv4.hdr = (struct rte_ipv4_hdr){
5003 .src_addr = input->flow.ip4_flow.src_ip,
5004 .dst_addr = input->flow.ip4_flow.dst_ip,
5005 .time_to_live = input->flow.ip4_flow.ttl,
5006 .type_of_service = input->flow.ip4_flow.tos,
5008 attributes->l3_mask.ipv4.hdr = (struct rte_ipv4_hdr){
5009 .src_addr = mask->ipv4_mask.src_ip,
5010 .dst_addr = mask->ipv4_mask.dst_ip,
5011 .time_to_live = mask->ipv4_mask.ttl,
5012 .type_of_service = mask->ipv4_mask.tos,
5013 .next_proto_id = mask->ipv4_mask.proto,
5015 attributes->items[1] = (struct rte_flow_item){
5016 .type = RTE_FLOW_ITEM_TYPE_IPV4,
5017 .spec = &attributes->l3,
5018 .mask = &attributes->l3_mask,
5021 case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
5022 case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
5023 case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
5024 attributes->l3.ipv6.hdr = (struct rte_ipv6_hdr){
5025 .hop_limits = input->flow.ipv6_flow.hop_limits,
5026 .proto = input->flow.ipv6_flow.proto,
5029 memcpy(attributes->l3.ipv6.hdr.src_addr,
5030 input->flow.ipv6_flow.src_ip,
5031 RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
5032 memcpy(attributes->l3.ipv6.hdr.dst_addr,
5033 input->flow.ipv6_flow.dst_ip,
5034 RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
5035 memcpy(attributes->l3_mask.ipv6.hdr.src_addr,
5036 mask->ipv6_mask.src_ip,
5037 RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
5038 memcpy(attributes->l3_mask.ipv6.hdr.dst_addr,
5039 mask->ipv6_mask.dst_ip,
5040 RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
5041 attributes->items[1] = (struct rte_flow_item){
5042 .type = RTE_FLOW_ITEM_TYPE_IPV6,
5043 .spec = &attributes->l3,
5044 .mask = &attributes->l3_mask,
5048 DRV_LOG(ERR, "port %u invalid flow type%d",
5049 dev->data->port_id, fdir_filter->input.flow_type);
5050 rte_errno = ENOTSUP;
5054 switch (fdir_filter->input.flow_type) {
5055 case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
5056 attributes->l4.udp.hdr = (struct rte_udp_hdr){
5057 .src_port = input->flow.udp4_flow.src_port,
5058 .dst_port = input->flow.udp4_flow.dst_port,
5060 attributes->l4_mask.udp.hdr = (struct rte_udp_hdr){
5061 .src_port = mask->src_port_mask,
5062 .dst_port = mask->dst_port_mask,
5064 attributes->items[2] = (struct rte_flow_item){
5065 .type = RTE_FLOW_ITEM_TYPE_UDP,
5066 .spec = &attributes->l4,
5067 .mask = &attributes->l4_mask,
5070 case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
5071 attributes->l4.tcp.hdr = (struct rte_tcp_hdr){
5072 .src_port = input->flow.tcp4_flow.src_port,
5073 .dst_port = input->flow.tcp4_flow.dst_port,
5075 attributes->l4_mask.tcp.hdr = (struct rte_tcp_hdr){
5076 .src_port = mask->src_port_mask,
5077 .dst_port = mask->dst_port_mask,
5079 attributes->items[2] = (struct rte_flow_item){
5080 .type = RTE_FLOW_ITEM_TYPE_TCP,
5081 .spec = &attributes->l4,
5082 .mask = &attributes->l4_mask,
5085 case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
5086 attributes->l4.udp.hdr = (struct rte_udp_hdr){
5087 .src_port = input->flow.udp6_flow.src_port,
5088 .dst_port = input->flow.udp6_flow.dst_port,
5090 attributes->l4_mask.udp.hdr = (struct rte_udp_hdr){
5091 .src_port = mask->src_port_mask,
5092 .dst_port = mask->dst_port_mask,
5094 attributes->items[2] = (struct rte_flow_item){
5095 .type = RTE_FLOW_ITEM_TYPE_UDP,
5096 .spec = &attributes->l4,
5097 .mask = &attributes->l4_mask,
5100 case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
5101 attributes->l4.tcp.hdr = (struct rte_tcp_hdr){
5102 .src_port = input->flow.tcp6_flow.src_port,
5103 .dst_port = input->flow.tcp6_flow.dst_port,
5105 attributes->l4_mask.tcp.hdr = (struct rte_tcp_hdr){
5106 .src_port = mask->src_port_mask,
5107 .dst_port = mask->dst_port_mask,
5109 attributes->items[2] = (struct rte_flow_item){
5110 .type = RTE_FLOW_ITEM_TYPE_TCP,
5111 .spec = &attributes->l4,
5112 .mask = &attributes->l4_mask,
5115 case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
5116 case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
5119 DRV_LOG(ERR, "port %u invalid flow type%d",
5120 dev->data->port_id, fdir_filter->input.flow_type);
5121 rte_errno = ENOTSUP;
5127 #define FLOW_FDIR_CMP(f1, f2, fld) \
5128 memcmp(&(f1)->fld, &(f2)->fld, sizeof(f1->fld))
5131 * Compare two FDIR flows. If items and actions are identical, the two flows are
5135 * Pointer to Ethernet device.
5137 * FDIR flow to compare.
5139 * FDIR flow to compare.
5142 * Zero on match, 1 otherwise.
5145 flow_fdir_cmp(const struct mlx5_fdir *f1, const struct mlx5_fdir *f2)
5147 if (FLOW_FDIR_CMP(f1, f2, attr) ||
5148 FLOW_FDIR_CMP(f1, f2, l2) ||
5149 FLOW_FDIR_CMP(f1, f2, l2_mask) ||
5150 FLOW_FDIR_CMP(f1, f2, l3) ||
5151 FLOW_FDIR_CMP(f1, f2, l3_mask) ||
5152 FLOW_FDIR_CMP(f1, f2, l4) ||
5153 FLOW_FDIR_CMP(f1, f2, l4_mask) ||
5154 FLOW_FDIR_CMP(f1, f2, actions[0].type))
5156 if (f1->actions[0].type == RTE_FLOW_ACTION_TYPE_QUEUE &&
5157 FLOW_FDIR_CMP(f1, f2, queue))
5163 * Search device flow list to find out a matched FDIR flow.
5166 * Pointer to Ethernet device.
5168 * FDIR flow to lookup.
5171 * Pointer of flow if found, NULL otherwise.
5173 static struct rte_flow *
5174 flow_fdir_filter_lookup(struct rte_eth_dev *dev, struct mlx5_fdir *fdir_flow)
5176 struct mlx5_priv *priv = dev->data->dev_private;
5177 struct rte_flow *flow = NULL;
5179 MLX5_ASSERT(fdir_flow);
5180 TAILQ_FOREACH(flow, &priv->flows, next) {
5181 if (flow->fdir && !flow_fdir_cmp(flow->fdir, fdir_flow)) {
5182 DRV_LOG(DEBUG, "port %u found FDIR flow %p",
5183 dev->data->port_id, (void *)flow);
5191 * Add new flow director filter and store it in list.
5194 * Pointer to Ethernet device.
5195 * @param fdir_filter
5196 * Flow director filter to add.
5199 * 0 on success, a negative errno value otherwise and rte_errno is set.
5202 flow_fdir_filter_add(struct rte_eth_dev *dev,
5203 const struct rte_eth_fdir_filter *fdir_filter)
5205 struct mlx5_priv *priv = dev->data->dev_private;
5206 struct mlx5_fdir *fdir_flow;
5207 struct rte_flow *flow;
5210 fdir_flow = rte_zmalloc(__func__, sizeof(*fdir_flow), 0);
5215 ret = flow_fdir_filter_convert(dev, fdir_filter, fdir_flow);
5218 flow = flow_fdir_filter_lookup(dev, fdir_flow);
5223 flow = flow_list_create(dev, &priv->flows, &fdir_flow->attr,
5224 fdir_flow->items, fdir_flow->actions, true,
5228 MLX5_ASSERT(!flow->fdir);
5229 flow->fdir = fdir_flow;
5230 DRV_LOG(DEBUG, "port %u created FDIR flow %p",
5231 dev->data->port_id, (void *)flow);
5234 rte_free(fdir_flow);
5239 * Delete specific filter.
5242 * Pointer to Ethernet device.
5243 * @param fdir_filter
5244 * Filter to be deleted.
5247 * 0 on success, a negative errno value otherwise and rte_errno is set.
5250 flow_fdir_filter_delete(struct rte_eth_dev *dev,
5251 const struct rte_eth_fdir_filter *fdir_filter)
5253 struct mlx5_priv *priv = dev->data->dev_private;
5254 struct rte_flow *flow;
5255 struct mlx5_fdir fdir_flow = {
5260 ret = flow_fdir_filter_convert(dev, fdir_filter, &fdir_flow);
5263 flow = flow_fdir_filter_lookup(dev, &fdir_flow);
5268 flow_list_destroy(dev, &priv->flows, flow);
5269 DRV_LOG(DEBUG, "port %u deleted FDIR flow %p",
5270 dev->data->port_id, (void *)flow);
5275 * Update queue for specific filter.
5278 * Pointer to Ethernet device.
5279 * @param fdir_filter
5280 * Filter to be updated.
5283 * 0 on success, a negative errno value otherwise and rte_errno is set.
5286 flow_fdir_filter_update(struct rte_eth_dev *dev,
5287 const struct rte_eth_fdir_filter *fdir_filter)
5291 ret = flow_fdir_filter_delete(dev, fdir_filter);
5294 return flow_fdir_filter_add(dev, fdir_filter);
5298 * Flush all filters.
5301 * Pointer to Ethernet device.
5304 flow_fdir_filter_flush(struct rte_eth_dev *dev)
5306 struct mlx5_priv *priv = dev->data->dev_private;
5308 mlx5_flow_list_flush(dev, &priv->flows, false);
5312 * Get flow director information.
5315 * Pointer to Ethernet device.
5316 * @param[out] fdir_info
5317 * Resulting flow director information.
5320 flow_fdir_info_get(struct rte_eth_dev *dev, struct rte_eth_fdir_info *fdir_info)
5322 struct rte_eth_fdir_masks *mask =
5323 &dev->data->dev_conf.fdir_conf.mask;
5325 fdir_info->mode = dev->data->dev_conf.fdir_conf.mode;
5326 fdir_info->guarant_spc = 0;
5327 rte_memcpy(&fdir_info->mask, mask, sizeof(fdir_info->mask));
5328 fdir_info->max_flexpayload = 0;
5329 fdir_info->flow_types_mask[0] = 0;
5330 fdir_info->flex_payload_unit = 0;
5331 fdir_info->max_flex_payload_segment_num = 0;
5332 fdir_info->flex_payload_limit = 0;
5333 memset(&fdir_info->flex_conf, 0, sizeof(fdir_info->flex_conf));
5337 * Deal with flow director operations.
5340 * Pointer to Ethernet device.
5342 * Operation to perform.
5344 * Pointer to operation-specific structure.
5347 * 0 on success, a negative errno value otherwise and rte_errno is set.
5350 flow_fdir_ctrl_func(struct rte_eth_dev *dev, enum rte_filter_op filter_op,
5353 enum rte_fdir_mode fdir_mode =
5354 dev->data->dev_conf.fdir_conf.mode;
5356 if (filter_op == RTE_ETH_FILTER_NOP)
5358 if (fdir_mode != RTE_FDIR_MODE_PERFECT &&
5359 fdir_mode != RTE_FDIR_MODE_PERFECT_MAC_VLAN) {
5360 DRV_LOG(ERR, "port %u flow director mode %d not supported",
5361 dev->data->port_id, fdir_mode);
5365 switch (filter_op) {
5366 case RTE_ETH_FILTER_ADD:
5367 return flow_fdir_filter_add(dev, arg);
5368 case RTE_ETH_FILTER_UPDATE:
5369 return flow_fdir_filter_update(dev, arg);
5370 case RTE_ETH_FILTER_DELETE:
5371 return flow_fdir_filter_delete(dev, arg);
5372 case RTE_ETH_FILTER_FLUSH:
5373 flow_fdir_filter_flush(dev);
5375 case RTE_ETH_FILTER_INFO:
5376 flow_fdir_info_get(dev, arg);
5379 DRV_LOG(DEBUG, "port %u unknown operation %u",
5380 dev->data->port_id, filter_op);
5388 * Manage filter operations.
5391 * Pointer to Ethernet device structure.
5392 * @param filter_type
5395 * Operation to perform.
5397 * Pointer to operation-specific structure.
5400 * 0 on success, a negative errno value otherwise and rte_errno is set.
5403 mlx5_dev_filter_ctrl(struct rte_eth_dev *dev,
5404 enum rte_filter_type filter_type,
5405 enum rte_filter_op filter_op,
5408 switch (filter_type) {
5409 case RTE_ETH_FILTER_GENERIC:
5410 if (filter_op != RTE_ETH_FILTER_GET) {
5414 *(const void **)arg = &mlx5_flow_ops;
5416 case RTE_ETH_FILTER_FDIR:
5417 return flow_fdir_ctrl_func(dev, filter_op, arg);
5419 DRV_LOG(ERR, "port %u filter type (%d) not supported",
5420 dev->data->port_id, filter_type);
5421 rte_errno = ENOTSUP;
5428 * Create the needed meter and suffix tables.
5431 * Pointer to Ethernet device.
5433 * Pointer to the flow meter.
5436 * Pointer to table set on success, NULL otherwise.
5438 struct mlx5_meter_domains_infos *
5439 mlx5_flow_create_mtr_tbls(struct rte_eth_dev *dev,
5440 const struct mlx5_flow_meter *fm)
5442 const struct mlx5_flow_driver_ops *fops;
5444 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5445 return fops->create_mtr_tbls(dev, fm);
5449 * Destroy the meter table set.
5452 * Pointer to Ethernet device.
5454 * Pointer to the meter table set.
5460 mlx5_flow_destroy_mtr_tbls(struct rte_eth_dev *dev,
5461 struct mlx5_meter_domains_infos *tbls)
5463 const struct mlx5_flow_driver_ops *fops;
5465 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5466 return fops->destroy_mtr_tbls(dev, tbls);
5470 * Create policer rules.
5473 * Pointer to Ethernet device.
5475 * Pointer to flow meter structure.
5477 * Pointer to flow attributes.
5480 * 0 on success, -1 otherwise.
5483 mlx5_flow_create_policer_rules(struct rte_eth_dev *dev,
5484 struct mlx5_flow_meter *fm,
5485 const struct rte_flow_attr *attr)
5487 const struct mlx5_flow_driver_ops *fops;
5489 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5490 return fops->create_policer_rules(dev, fm, attr);
5494 * Destroy policer rules.
5497 * Pointer to flow meter structure.
5499 * Pointer to flow attributes.
5502 * 0 on success, -1 otherwise.
5505 mlx5_flow_destroy_policer_rules(struct rte_eth_dev *dev,
5506 struct mlx5_flow_meter *fm,
5507 const struct rte_flow_attr *attr)
5509 const struct mlx5_flow_driver_ops *fops;
5511 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5512 return fops->destroy_policer_rules(dev, fm, attr);
5516 * Allocate a counter.
5519 * Pointer to Ethernet device structure.
5522 * Index to allocated counter on success, 0 otherwise.
5525 mlx5_counter_alloc(struct rte_eth_dev *dev)
5527 const struct mlx5_flow_driver_ops *fops;
5528 struct rte_flow_attr attr = { .transfer = 0 };
5530 if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
5531 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5532 return fops->counter_alloc(dev);
5535 "port %u counter allocate is not supported.",
5536 dev->data->port_id);
5544 * Pointer to Ethernet device structure.
5546 * Index to counter to be free.
5549 mlx5_counter_free(struct rte_eth_dev *dev, uint32_t cnt)
5551 const struct mlx5_flow_driver_ops *fops;
5552 struct rte_flow_attr attr = { .transfer = 0 };
5554 if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
5555 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5556 fops->counter_free(dev, cnt);
5560 "port %u counter free is not supported.",
5561 dev->data->port_id);
5565 * Query counter statistics.
5568 * Pointer to Ethernet device structure.
5570 * Index to counter to query.
5572 * Set to clear counter statistics.
5574 * The counter hits packets number to save.
5576 * The counter hits bytes number to save.
5579 * 0 on success, a negative errno value otherwise.
5582 mlx5_counter_query(struct rte_eth_dev *dev, uint32_t cnt,
5583 bool clear, uint64_t *pkts, uint64_t *bytes)
5585 const struct mlx5_flow_driver_ops *fops;
5586 struct rte_flow_attr attr = { .transfer = 0 };
5588 if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
5589 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5590 return fops->counter_query(dev, cnt, clear, pkts, bytes);
5593 "port %u counter query is not supported.",
5594 dev->data->port_id);
5598 #define MLX5_POOL_QUERY_FREQ_US 1000000
5601 * Set the periodic procedure for triggering asynchronous batch queries for all
5602 * the counter pools.
5605 * Pointer to mlx5_ibv_shared object.
5608 mlx5_set_query_alarm(struct mlx5_ibv_shared *sh)
5610 struct mlx5_pools_container *cont = MLX5_CNT_CONTAINER(sh, 0, 0);
5611 uint32_t pools_n = rte_atomic16_read(&cont->n_valid);
5614 cont = MLX5_CNT_CONTAINER(sh, 1, 0);
5615 pools_n += rte_atomic16_read(&cont->n_valid);
5616 us = MLX5_POOL_QUERY_FREQ_US / pools_n;
5617 DRV_LOG(DEBUG, "Set alarm for %u pools each %u us", pools_n, us);
5618 if (rte_eal_alarm_set(us, mlx5_flow_query_alarm, sh)) {
5619 sh->cmng.query_thread_on = 0;
5620 DRV_LOG(ERR, "Cannot reinitialize query alarm");
5622 sh->cmng.query_thread_on = 1;
5627 * The periodic procedure for triggering asynchronous batch queries for all the
5628 * counter pools. This function is probably called by the host thread.
5631 * The parameter for the alarm process.
5634 mlx5_flow_query_alarm(void *arg)
5636 struct mlx5_ibv_shared *sh = arg;
5637 struct mlx5_devx_obj *dcs;
5640 uint8_t batch = sh->cmng.batch;
5641 uint16_t pool_index = sh->cmng.pool_index;
5642 struct mlx5_pools_container *cont;
5643 struct mlx5_pools_container *mcont;
5644 struct mlx5_flow_counter_pool *pool;
5646 if (sh->cmng.pending_queries >= MLX5_MAX_PENDING_QUERIES)
5649 cont = MLX5_CNT_CONTAINER(sh, batch, 1);
5650 mcont = MLX5_CNT_CONTAINER(sh, batch, 0);
5651 /* Check if resize was done and need to flip a container. */
5652 if (cont != mcont) {
5654 /* Clean the old container. */
5655 rte_free(cont->pools);
5656 memset(cont, 0, sizeof(*cont));
5659 /* Flip the host container. */
5660 sh->cmng.mhi[batch] ^= (uint8_t)2;
5664 /* 2 empty containers case is unexpected. */
5665 if (unlikely(batch != sh->cmng.batch))
5669 goto next_container;
5671 pool = cont->pools[pool_index];
5673 /* There is a pool query in progress. */
5676 LIST_FIRST(&sh->cmng.free_stat_raws);
5678 /* No free counter statistics raw memory. */
5680 dcs = (struct mlx5_devx_obj *)(uintptr_t)rte_atomic64_read
5682 offset = batch ? 0 : dcs->id % MLX5_COUNTERS_PER_POOL;
5684 * Identify the counters released between query trigger and query
5685 * handle more effiecntly. The counter released in this gap period
5686 * should wait for a new round of query as the new arrived packets
5687 * will not be taken into account.
5689 rte_atomic64_add(&pool->start_query_gen, 1);
5690 ret = mlx5_devx_cmd_flow_counter_query(dcs, 0, MLX5_COUNTERS_PER_POOL -
5692 pool->raw_hw->mem_mng->dm->id,
5694 (pool->raw_hw->data + offset),
5696 (uint64_t)(uintptr_t)pool);
5698 rte_atomic64_sub(&pool->start_query_gen, 1);
5699 DRV_LOG(ERR, "Failed to trigger asynchronous query for dcs ID"
5700 " %d", pool->min_dcs->id);
5701 pool->raw_hw = NULL;
5704 pool->raw_hw->min_dcs_id = dcs->id;
5705 LIST_REMOVE(pool->raw_hw, next);
5706 sh->cmng.pending_queries++;
5708 if (pool_index >= rte_atomic16_read(&cont->n_valid)) {
5713 sh->cmng.batch = batch;
5714 sh->cmng.pool_index = pool_index;
5715 mlx5_set_query_alarm(sh);
5719 * Handler for the HW respond about ready values from an asynchronous batch
5720 * query. This function is probably called by the host thread.
5723 * The pointer to the shared IB device context.
5724 * @param[in] async_id
5725 * The Devx async ID.
5727 * The status of the completion.
5730 mlx5_flow_async_pool_query_handle(struct mlx5_ibv_shared *sh,
5731 uint64_t async_id, int status)
5733 struct mlx5_flow_counter_pool *pool =
5734 (struct mlx5_flow_counter_pool *)(uintptr_t)async_id;
5735 struct mlx5_counter_stats_raw *raw_to_free;
5737 if (unlikely(status)) {
5738 rte_atomic64_sub(&pool->start_query_gen, 1);
5739 raw_to_free = pool->raw_hw;
5741 raw_to_free = pool->raw;
5742 rte_spinlock_lock(&pool->sl);
5743 pool->raw = pool->raw_hw;
5744 rte_spinlock_unlock(&pool->sl);
5745 MLX5_ASSERT(rte_atomic64_read(&pool->end_query_gen) + 1 ==
5746 rte_atomic64_read(&pool->start_query_gen));
5747 rte_atomic64_set(&pool->end_query_gen,
5748 rte_atomic64_read(&pool->start_query_gen));
5749 /* Be sure the new raw counters data is updated in memory. */
5752 LIST_INSERT_HEAD(&sh->cmng.free_stat_raws, raw_to_free, next);
5753 pool->raw_hw = NULL;
5754 sh->cmng.pending_queries--;
5758 * Translate the rte_flow group index to HW table value.
5760 * @param[in] attributes
5761 * Pointer to flow attributes
5762 * @param[in] external
5763 * Value is part of flow rule created by request external to PMD.
5765 * rte_flow group index value.
5766 * @param[out] fdb_def_rule
5767 * Whether fdb jump to table 1 is configured.
5771 * Pointer to error structure.
5774 * 0 on success, a negative errno value otherwise and rte_errno is set.
5777 mlx5_flow_group_to_table(const struct rte_flow_attr *attributes, bool external,
5778 uint32_t group, bool fdb_def_rule, uint32_t *table,
5779 struct rte_flow_error *error)
5781 if (attributes->transfer && external && fdb_def_rule) {
5782 if (group == UINT32_MAX)
5783 return rte_flow_error_set
5785 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
5787 "group index not supported");
5796 * Discover availability of metadata reg_c's.
5798 * Iteratively use test flows to check availability.
5801 * Pointer to the Ethernet device structure.
5804 * 0 on success, a negative errno value otherwise and rte_errno is set.
5807 mlx5_flow_discover_mreg_c(struct rte_eth_dev *dev)
5809 struct mlx5_priv *priv = dev->data->dev_private;
5810 struct mlx5_dev_config *config = &priv->config;
5811 enum modify_reg idx;
5814 /* reg_c[0] and reg_c[1] are reserved. */
5815 config->flow_mreg_c[n++] = REG_C_0;
5816 config->flow_mreg_c[n++] = REG_C_1;
5817 /* Discover availability of other reg_c's. */
5818 for (idx = REG_C_2; idx <= REG_C_7; ++idx) {
5819 struct rte_flow_attr attr = {
5820 .group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
5821 .priority = MLX5_FLOW_PRIO_RSVD,
5824 struct rte_flow_item items[] = {
5826 .type = RTE_FLOW_ITEM_TYPE_END,
5829 struct rte_flow_action actions[] = {
5831 .type = MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
5832 .conf = &(struct mlx5_flow_action_copy_mreg){
5838 .type = RTE_FLOW_ACTION_TYPE_JUMP,
5839 .conf = &(struct rte_flow_action_jump){
5840 .group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
5844 .type = RTE_FLOW_ACTION_TYPE_END,
5847 struct rte_flow *flow;
5848 struct rte_flow_error error;
5850 if (!config->dv_flow_en)
5852 /* Create internal flow, validation skips copy action. */
5853 flow = flow_list_create(dev, NULL, &attr, items,
5854 actions, false, &error);
5857 if (dev->data->dev_started || !flow_drv_apply(dev, flow, NULL))
5858 config->flow_mreg_c[n++] = idx;
5859 flow_list_destroy(dev, NULL, flow);
5861 for (; n < MLX5_MREG_C_NUM; ++n)
5862 config->flow_mreg_c[n] = REG_NONE;
5867 * Dump flow raw hw data to file
5870 * The pointer to Ethernet device.
5872 * A pointer to a file for output.
5874 * Perform verbose error reporting if not NULL. PMDs initialize this
5875 * structure in case of error only.
5877 * 0 on success, a nagative value otherwise.
5880 mlx5_flow_dev_dump(struct rte_eth_dev *dev,
5882 struct rte_flow_error *error __rte_unused)
5884 struct mlx5_priv *priv = dev->data->dev_private;
5885 struct mlx5_ibv_shared *sh = priv->sh;
5887 return mlx5_devx_cmd_flow_dump(sh->fdb_domain, sh->rx_domain,
5888 sh->tx_domain, file);