1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright 2016 6WIND S.A.
3 * Copyright 2016 Mellanox Technologies, Ltd
6 #include <netinet/in.h>
13 /* ISO C doesn't support unnamed structs/unions, disabling -pedantic. */
15 #pragma GCC diagnostic ignored "-Wpedantic"
17 #include <infiniband/verbs.h>
19 #pragma GCC diagnostic error "-Wpedantic"
22 #include <rte_common.h>
23 #include <rte_ether.h>
24 #include <rte_ethdev_driver.h>
26 #include <rte_flow_driver.h>
27 #include <rte_malloc.h>
31 #include "mlx5_defs.h"
32 #include "mlx5_flow.h"
33 #include "mlx5_glue.h"
35 #include "mlx5_rxtx.h"
37 /* Dev ops structure defined in mlx5.c */
38 extern const struct eth_dev_ops mlx5_dev_ops;
39 extern const struct eth_dev_ops mlx5_dev_ops_isolate;
41 /** Device flow drivers. */
42 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
43 extern const struct mlx5_flow_driver_ops mlx5_flow_dv_drv_ops;
45 extern const struct mlx5_flow_driver_ops mlx5_flow_verbs_drv_ops;
47 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops;
49 const struct mlx5_flow_driver_ops *flow_drv_ops[] = {
50 [MLX5_FLOW_TYPE_MIN] = &mlx5_flow_null_drv_ops,
51 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
52 [MLX5_FLOW_TYPE_DV] = &mlx5_flow_dv_drv_ops,
54 [MLX5_FLOW_TYPE_VERBS] = &mlx5_flow_verbs_drv_ops,
55 [MLX5_FLOW_TYPE_MAX] = &mlx5_flow_null_drv_ops
60 MLX5_EXPANSION_ROOT_OUTER,
61 MLX5_EXPANSION_ROOT_ETH_VLAN,
62 MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN,
63 MLX5_EXPANSION_OUTER_ETH,
64 MLX5_EXPANSION_OUTER_ETH_VLAN,
65 MLX5_EXPANSION_OUTER_VLAN,
66 MLX5_EXPANSION_OUTER_IPV4,
67 MLX5_EXPANSION_OUTER_IPV4_UDP,
68 MLX5_EXPANSION_OUTER_IPV4_TCP,
69 MLX5_EXPANSION_OUTER_IPV6,
70 MLX5_EXPANSION_OUTER_IPV6_UDP,
71 MLX5_EXPANSION_OUTER_IPV6_TCP,
73 MLX5_EXPANSION_VXLAN_GPE,
77 MLX5_EXPANSION_ETH_VLAN,
80 MLX5_EXPANSION_IPV4_UDP,
81 MLX5_EXPANSION_IPV4_TCP,
83 MLX5_EXPANSION_IPV6_UDP,
84 MLX5_EXPANSION_IPV6_TCP,
87 /** Supported expansion of items. */
88 static const struct rte_flow_expand_node mlx5_support_expansion[] = {
89 [MLX5_EXPANSION_ROOT] = {
90 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
93 .type = RTE_FLOW_ITEM_TYPE_END,
95 [MLX5_EXPANSION_ROOT_OUTER] = {
96 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_ETH,
97 MLX5_EXPANSION_OUTER_IPV4,
98 MLX5_EXPANSION_OUTER_IPV6),
99 .type = RTE_FLOW_ITEM_TYPE_END,
101 [MLX5_EXPANSION_ROOT_ETH_VLAN] = {
102 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH_VLAN),
103 .type = RTE_FLOW_ITEM_TYPE_END,
105 [MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN] = {
106 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_ETH_VLAN),
107 .type = RTE_FLOW_ITEM_TYPE_END,
109 [MLX5_EXPANSION_OUTER_ETH] = {
110 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
111 MLX5_EXPANSION_OUTER_IPV6,
112 MLX5_EXPANSION_MPLS),
113 .type = RTE_FLOW_ITEM_TYPE_ETH,
116 [MLX5_EXPANSION_OUTER_ETH_VLAN] = {
117 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_VLAN),
118 .type = RTE_FLOW_ITEM_TYPE_ETH,
121 [MLX5_EXPANSION_OUTER_VLAN] = {
122 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
123 MLX5_EXPANSION_OUTER_IPV6),
124 .type = RTE_FLOW_ITEM_TYPE_VLAN,
126 [MLX5_EXPANSION_OUTER_IPV4] = {
127 .next = RTE_FLOW_EXPAND_RSS_NEXT
128 (MLX5_EXPANSION_OUTER_IPV4_UDP,
129 MLX5_EXPANSION_OUTER_IPV4_TCP,
132 MLX5_EXPANSION_IPV6),
133 .type = RTE_FLOW_ITEM_TYPE_IPV4,
134 .rss_types = ETH_RSS_IPV4 | ETH_RSS_FRAG_IPV4 |
135 ETH_RSS_NONFRAG_IPV4_OTHER,
137 [MLX5_EXPANSION_OUTER_IPV4_UDP] = {
138 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
139 MLX5_EXPANSION_VXLAN_GPE),
140 .type = RTE_FLOW_ITEM_TYPE_UDP,
141 .rss_types = ETH_RSS_NONFRAG_IPV4_UDP,
143 [MLX5_EXPANSION_OUTER_IPV4_TCP] = {
144 .type = RTE_FLOW_ITEM_TYPE_TCP,
145 .rss_types = ETH_RSS_NONFRAG_IPV4_TCP,
147 [MLX5_EXPANSION_OUTER_IPV6] = {
148 .next = RTE_FLOW_EXPAND_RSS_NEXT
149 (MLX5_EXPANSION_OUTER_IPV6_UDP,
150 MLX5_EXPANSION_OUTER_IPV6_TCP,
152 MLX5_EXPANSION_IPV6),
153 .type = RTE_FLOW_ITEM_TYPE_IPV6,
154 .rss_types = ETH_RSS_IPV6 | ETH_RSS_FRAG_IPV6 |
155 ETH_RSS_NONFRAG_IPV6_OTHER,
157 [MLX5_EXPANSION_OUTER_IPV6_UDP] = {
158 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
159 MLX5_EXPANSION_VXLAN_GPE),
160 .type = RTE_FLOW_ITEM_TYPE_UDP,
161 .rss_types = ETH_RSS_NONFRAG_IPV6_UDP,
163 [MLX5_EXPANSION_OUTER_IPV6_TCP] = {
164 .type = RTE_FLOW_ITEM_TYPE_TCP,
165 .rss_types = ETH_RSS_NONFRAG_IPV6_TCP,
167 [MLX5_EXPANSION_VXLAN] = {
168 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH),
169 .type = RTE_FLOW_ITEM_TYPE_VXLAN,
171 [MLX5_EXPANSION_VXLAN_GPE] = {
172 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
174 MLX5_EXPANSION_IPV6),
175 .type = RTE_FLOW_ITEM_TYPE_VXLAN_GPE,
177 [MLX5_EXPANSION_GRE] = {
178 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4),
179 .type = RTE_FLOW_ITEM_TYPE_GRE,
181 [MLX5_EXPANSION_MPLS] = {
182 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
183 MLX5_EXPANSION_IPV6),
184 .type = RTE_FLOW_ITEM_TYPE_MPLS,
186 [MLX5_EXPANSION_ETH] = {
187 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
188 MLX5_EXPANSION_IPV6),
189 .type = RTE_FLOW_ITEM_TYPE_ETH,
191 [MLX5_EXPANSION_ETH_VLAN] = {
192 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VLAN),
193 .type = RTE_FLOW_ITEM_TYPE_ETH,
195 [MLX5_EXPANSION_VLAN] = {
196 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
197 MLX5_EXPANSION_IPV6),
198 .type = RTE_FLOW_ITEM_TYPE_VLAN,
200 [MLX5_EXPANSION_IPV4] = {
201 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4_UDP,
202 MLX5_EXPANSION_IPV4_TCP),
203 .type = RTE_FLOW_ITEM_TYPE_IPV4,
204 .rss_types = ETH_RSS_IPV4 | ETH_RSS_FRAG_IPV4 |
205 ETH_RSS_NONFRAG_IPV4_OTHER,
207 [MLX5_EXPANSION_IPV4_UDP] = {
208 .type = RTE_FLOW_ITEM_TYPE_UDP,
209 .rss_types = ETH_RSS_NONFRAG_IPV4_UDP,
211 [MLX5_EXPANSION_IPV4_TCP] = {
212 .type = RTE_FLOW_ITEM_TYPE_TCP,
213 .rss_types = ETH_RSS_NONFRAG_IPV4_TCP,
215 [MLX5_EXPANSION_IPV6] = {
216 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV6_UDP,
217 MLX5_EXPANSION_IPV6_TCP),
218 .type = RTE_FLOW_ITEM_TYPE_IPV6,
219 .rss_types = ETH_RSS_IPV6 | ETH_RSS_FRAG_IPV6 |
220 ETH_RSS_NONFRAG_IPV6_OTHER,
222 [MLX5_EXPANSION_IPV6_UDP] = {
223 .type = RTE_FLOW_ITEM_TYPE_UDP,
224 .rss_types = ETH_RSS_NONFRAG_IPV6_UDP,
226 [MLX5_EXPANSION_IPV6_TCP] = {
227 .type = RTE_FLOW_ITEM_TYPE_TCP,
228 .rss_types = ETH_RSS_NONFRAG_IPV6_TCP,
232 static const struct rte_flow_ops mlx5_flow_ops = {
233 .validate = mlx5_flow_validate,
234 .create = mlx5_flow_create,
235 .destroy = mlx5_flow_destroy,
236 .flush = mlx5_flow_flush,
237 .isolate = mlx5_flow_isolate,
238 .query = mlx5_flow_query,
241 /* Convert FDIR request to Generic flow. */
243 struct rte_flow_attr attr;
244 struct rte_flow_item items[4];
245 struct rte_flow_item_eth l2;
246 struct rte_flow_item_eth l2_mask;
248 struct rte_flow_item_ipv4 ipv4;
249 struct rte_flow_item_ipv6 ipv6;
252 struct rte_flow_item_ipv4 ipv4;
253 struct rte_flow_item_ipv6 ipv6;
256 struct rte_flow_item_udp udp;
257 struct rte_flow_item_tcp tcp;
260 struct rte_flow_item_udp udp;
261 struct rte_flow_item_tcp tcp;
263 struct rte_flow_action actions[2];
264 struct rte_flow_action_queue queue;
267 /* Map of Verbs to Flow priority with 8 Verbs priorities. */
268 static const uint32_t priority_map_3[][MLX5_PRIORITY_MAP_MAX] = {
269 { 0, 1, 2 }, { 2, 3, 4 }, { 5, 6, 7 },
272 /* Map of Verbs to Flow priority with 16 Verbs priorities. */
273 static const uint32_t priority_map_5[][MLX5_PRIORITY_MAP_MAX] = {
274 { 0, 1, 2 }, { 3, 4, 5 }, { 6, 7, 8 },
275 { 9, 10, 11 }, { 12, 13, 14 },
278 /* Tunnel information. */
279 struct mlx5_flow_tunnel_info {
280 uint64_t tunnel; /**< Tunnel bit (see MLX5_FLOW_*). */
281 uint32_t ptype; /**< Tunnel Ptype (see RTE_PTYPE_*). */
284 static struct mlx5_flow_tunnel_info tunnels_info[] = {
286 .tunnel = MLX5_FLOW_LAYER_VXLAN,
287 .ptype = RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L4_UDP,
290 .tunnel = MLX5_FLOW_LAYER_GENEVE,
291 .ptype = RTE_PTYPE_TUNNEL_GENEVE | RTE_PTYPE_L4_UDP,
294 .tunnel = MLX5_FLOW_LAYER_VXLAN_GPE,
295 .ptype = RTE_PTYPE_TUNNEL_VXLAN_GPE | RTE_PTYPE_L4_UDP,
298 .tunnel = MLX5_FLOW_LAYER_GRE,
299 .ptype = RTE_PTYPE_TUNNEL_GRE,
302 .tunnel = MLX5_FLOW_LAYER_MPLS | MLX5_FLOW_LAYER_OUTER_L4_UDP,
303 .ptype = RTE_PTYPE_TUNNEL_MPLS_IN_UDP | RTE_PTYPE_L4_UDP,
306 .tunnel = MLX5_FLOW_LAYER_MPLS,
307 .ptype = RTE_PTYPE_TUNNEL_MPLS_IN_GRE,
310 .tunnel = MLX5_FLOW_LAYER_NVGRE,
311 .ptype = RTE_PTYPE_TUNNEL_NVGRE,
314 .tunnel = MLX5_FLOW_LAYER_IPIP,
315 .ptype = RTE_PTYPE_TUNNEL_IP,
318 .tunnel = MLX5_FLOW_LAYER_IPV6_ENCAP,
319 .ptype = RTE_PTYPE_TUNNEL_IP,
324 * Translate tag ID to register.
327 * Pointer to the Ethernet device structure.
329 * The feature that request the register.
331 * The request register ID.
333 * Error description in case of any.
336 * The request register on success, a negative errno
337 * value otherwise and rte_errno is set.
340 mlx5_flow_get_reg_id(struct rte_eth_dev *dev,
341 enum mlx5_feature_name feature,
343 struct rte_flow_error *error)
345 struct mlx5_priv *priv = dev->data->dev_private;
346 struct mlx5_dev_config *config = &priv->config;
347 enum modify_reg start_reg;
350 case MLX5_HAIRPIN_RX:
352 case MLX5_HAIRPIN_TX:
354 case MLX5_METADATA_RX:
355 switch (config->dv_xmeta_en) {
356 case MLX5_XMETA_MODE_LEGACY:
358 case MLX5_XMETA_MODE_META16:
360 case MLX5_XMETA_MODE_META32:
364 case MLX5_METADATA_TX:
366 case MLX5_METADATA_FDB:
367 switch (config->dv_xmeta_en) {
368 case MLX5_XMETA_MODE_LEGACY:
370 case MLX5_XMETA_MODE_META16:
372 case MLX5_XMETA_MODE_META32:
377 switch (config->dv_xmeta_en) {
378 case MLX5_XMETA_MODE_LEGACY:
380 case MLX5_XMETA_MODE_META16:
382 case MLX5_XMETA_MODE_META32:
389 * Metadata COPY_MARK register using is in meter suffix sub
390 * flow while with meter. It's safe to share the same register.
392 return priv->mtr_color_reg != REG_C_2 ? REG_C_2 : REG_C_3;
394 RTE_ASSERT(priv->mtr_color_reg != REG_NONE);
395 return priv->mtr_color_reg;
398 * If meter is enable, it will engage two registers for color
399 * match and flow match. If meter color match is not using the
400 * REG_C_2, need to skip the REG_C_x be used by meter color
402 * If meter is disable, free to use all available registers.
404 if (priv->mtr_color_reg != REG_NONE)
405 start_reg = priv->mtr_color_reg != REG_C_2 ? REG_C_3 :
409 if (id > (REG_C_7 - start_reg))
410 return rte_flow_error_set(error, EINVAL,
411 RTE_FLOW_ERROR_TYPE_ITEM,
412 NULL, "invalid tag id");
413 if (config->flow_mreg_c[id + start_reg - REG_C_0] == REG_NONE)
414 return rte_flow_error_set(error, ENOTSUP,
415 RTE_FLOW_ERROR_TYPE_ITEM,
416 NULL, "unsupported tag id");
418 * This case means meter is using the REG_C_x great than 2.
419 * Take care not to conflict with meter color REG_C_x.
420 * If the available index REG_C_y >= REG_C_x, skip the
423 if (start_reg == REG_C_3 && config->flow_mreg_c
424 [id + REG_C_3 - REG_C_0] >= priv->mtr_color_reg) {
425 if (config->flow_mreg_c[id + 1 + REG_C_3 - REG_C_0] !=
427 return config->flow_mreg_c
428 [id + 1 + REG_C_3 - REG_C_0];
429 return rte_flow_error_set(error, ENOTSUP,
430 RTE_FLOW_ERROR_TYPE_ITEM,
431 NULL, "unsupported tag id");
433 return config->flow_mreg_c[id + start_reg - REG_C_0];
436 return rte_flow_error_set(error, EINVAL,
437 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
438 NULL, "invalid feature name");
442 * Check extensive flow metadata register support.
445 * Pointer to rte_eth_dev structure.
448 * True if device supports extensive flow metadata register, otherwise false.
451 mlx5_flow_ext_mreg_supported(struct rte_eth_dev *dev)
453 struct mlx5_priv *priv = dev->data->dev_private;
454 struct mlx5_dev_config *config = &priv->config;
457 * Having available reg_c can be regarded inclusively as supporting
458 * extensive flow metadata register, which could mean,
459 * - metadata register copy action by modify header.
460 * - 16 modify header actions is supported.
461 * - reg_c's are preserved across different domain (FDB and NIC) on
462 * packet loopback by flow lookup miss.
464 return config->flow_mreg_c[2] != REG_NONE;
468 * Discover the maximum number of priority available.
471 * Pointer to the Ethernet device structure.
474 * number of supported flow priority on success, a negative errno
475 * value otherwise and rte_errno is set.
478 mlx5_flow_discover_priorities(struct rte_eth_dev *dev)
480 struct mlx5_priv *priv = dev->data->dev_private;
482 struct ibv_flow_attr attr;
483 struct ibv_flow_spec_eth eth;
484 struct ibv_flow_spec_action_drop drop;
488 .port = (uint8_t)priv->ibv_port,
491 .type = IBV_FLOW_SPEC_ETH,
492 .size = sizeof(struct ibv_flow_spec_eth),
495 .size = sizeof(struct ibv_flow_spec_action_drop),
496 .type = IBV_FLOW_SPEC_ACTION_DROP,
499 struct ibv_flow *flow;
500 struct mlx5_hrxq *drop = mlx5_hrxq_drop_new(dev);
501 uint16_t vprio[] = { 8, 16 };
509 for (i = 0; i != RTE_DIM(vprio); i++) {
510 flow_attr.attr.priority = vprio[i] - 1;
511 flow = mlx5_glue->create_flow(drop->qp, &flow_attr.attr);
514 claim_zero(mlx5_glue->destroy_flow(flow));
517 mlx5_hrxq_drop_release(dev);
520 priority = RTE_DIM(priority_map_3);
523 priority = RTE_DIM(priority_map_5);
528 "port %u verbs maximum priority: %d expected 8/16",
529 dev->data->port_id, priority);
532 DRV_LOG(INFO, "port %u flow maximum priority: %d",
533 dev->data->port_id, priority);
538 * Adjust flow priority based on the highest layer and the request priority.
541 * Pointer to the Ethernet device structure.
542 * @param[in] priority
543 * The rule base priority.
544 * @param[in] subpriority
545 * The priority based on the items.
550 uint32_t mlx5_flow_adjust_priority(struct rte_eth_dev *dev, int32_t priority,
551 uint32_t subpriority)
554 struct mlx5_priv *priv = dev->data->dev_private;
556 switch (priv->config.flow_prio) {
557 case RTE_DIM(priority_map_3):
558 res = priority_map_3[priority][subpriority];
560 case RTE_DIM(priority_map_5):
561 res = priority_map_5[priority][subpriority];
568 * Verify the @p item specifications (spec, last, mask) are compatible with the
572 * Item specification.
574 * @p item->mask or flow default bit-masks.
575 * @param[in] nic_mask
576 * Bit-masks covering supported fields by the NIC to compare with user mask.
578 * Bit-masks size in bytes.
580 * Pointer to error structure.
583 * 0 on success, a negative errno value otherwise and rte_errno is set.
586 mlx5_flow_item_acceptable(const struct rte_flow_item *item,
588 const uint8_t *nic_mask,
590 struct rte_flow_error *error)
595 for (i = 0; i < size; ++i)
596 if ((nic_mask[i] | mask[i]) != nic_mask[i])
597 return rte_flow_error_set(error, ENOTSUP,
598 RTE_FLOW_ERROR_TYPE_ITEM,
600 "mask enables non supported"
602 if (!item->spec && (item->mask || item->last))
603 return rte_flow_error_set(error, EINVAL,
604 RTE_FLOW_ERROR_TYPE_ITEM, item,
605 "mask/last without a spec is not"
607 if (item->spec && item->last) {
613 for (i = 0; i < size; ++i) {
614 spec[i] = ((const uint8_t *)item->spec)[i] & mask[i];
615 last[i] = ((const uint8_t *)item->last)[i] & mask[i];
617 ret = memcmp(spec, last, size);
619 return rte_flow_error_set(error, EINVAL,
620 RTE_FLOW_ERROR_TYPE_ITEM,
622 "range is not valid");
628 * Adjust the hash fields according to the @p flow information.
630 * @param[in] dev_flow.
631 * Pointer to the mlx5_flow.
633 * 1 when the hash field is for a tunnel item.
634 * @param[in] layer_types
636 * @param[in] hash_fields
640 * The hash fields that should be used.
643 mlx5_flow_hashfields_adjust(struct mlx5_flow *dev_flow,
644 int tunnel __rte_unused, uint64_t layer_types,
645 uint64_t hash_fields)
647 struct rte_flow *flow = dev_flow->flow;
648 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
649 int rss_request_inner = flow->rss.level >= 2;
651 /* Check RSS hash level for tunnel. */
652 if (tunnel && rss_request_inner)
653 hash_fields |= IBV_RX_HASH_INNER;
654 else if (tunnel || rss_request_inner)
657 /* Check if requested layer matches RSS hash fields. */
658 if (!(flow->rss.types & layer_types))
664 * Lookup and set the ptype in the data Rx part. A single Ptype can be used,
665 * if several tunnel rules are used on this queue, the tunnel ptype will be
669 * Rx queue to update.
672 flow_rxq_tunnel_ptype_update(struct mlx5_rxq_ctrl *rxq_ctrl)
675 uint32_t tunnel_ptype = 0;
677 /* Look up for the ptype to use. */
678 for (i = 0; i != MLX5_FLOW_TUNNEL; ++i) {
679 if (!rxq_ctrl->flow_tunnels_n[i])
682 tunnel_ptype = tunnels_info[i].ptype;
688 rxq_ctrl->rxq.tunnel = tunnel_ptype;
692 * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) according to the devive
696 * Pointer to the Ethernet device structure.
697 * @param[in] dev_flow
698 * Pointer to device flow structure.
701 flow_drv_rxq_flags_set(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow)
703 struct mlx5_priv *priv = dev->data->dev_private;
704 struct rte_flow *flow = dev_flow->flow;
705 const int mark = !!(dev_flow->actions &
706 (MLX5_FLOW_ACTION_FLAG | MLX5_FLOW_ACTION_MARK));
707 const int tunnel = !!(dev_flow->layers & MLX5_FLOW_LAYER_TUNNEL);
710 for (i = 0; i != flow->rss.queue_num; ++i) {
711 int idx = (*flow->rss.queue)[i];
712 struct mlx5_rxq_ctrl *rxq_ctrl =
713 container_of((*priv->rxqs)[idx],
714 struct mlx5_rxq_ctrl, rxq);
717 * To support metadata register copy on Tx loopback,
718 * this must be always enabled (metadata may arive
719 * from other port - not from local flows only.
721 if (priv->config.dv_flow_en &&
722 priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
723 mlx5_flow_ext_mreg_supported(dev)) {
724 rxq_ctrl->rxq.mark = 1;
725 rxq_ctrl->flow_mark_n = 1;
727 rxq_ctrl->rxq.mark = 1;
728 rxq_ctrl->flow_mark_n++;
733 /* Increase the counter matching the flow. */
734 for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
735 if ((tunnels_info[j].tunnel &
737 tunnels_info[j].tunnel) {
738 rxq_ctrl->flow_tunnels_n[j]++;
742 flow_rxq_tunnel_ptype_update(rxq_ctrl);
748 * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) for a flow
751 * Pointer to the Ethernet device structure.
753 * Pointer to flow structure.
756 flow_rxq_flags_set(struct rte_eth_dev *dev, struct rte_flow *flow)
758 struct mlx5_flow *dev_flow;
760 LIST_FOREACH(dev_flow, &flow->dev_flows, next)
761 flow_drv_rxq_flags_set(dev, dev_flow);
765 * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
766 * device flow if no other flow uses it with the same kind of request.
769 * Pointer to Ethernet device.
770 * @param[in] dev_flow
771 * Pointer to the device flow.
774 flow_drv_rxq_flags_trim(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow)
776 struct mlx5_priv *priv = dev->data->dev_private;
777 struct rte_flow *flow = dev_flow->flow;
778 const int mark = !!(dev_flow->actions &
779 (MLX5_FLOW_ACTION_FLAG | MLX5_FLOW_ACTION_MARK));
780 const int tunnel = !!(dev_flow->layers & MLX5_FLOW_LAYER_TUNNEL);
783 assert(dev->data->dev_started);
784 for (i = 0; i != flow->rss.queue_num; ++i) {
785 int idx = (*flow->rss.queue)[i];
786 struct mlx5_rxq_ctrl *rxq_ctrl =
787 container_of((*priv->rxqs)[idx],
788 struct mlx5_rxq_ctrl, rxq);
790 if (priv->config.dv_flow_en &&
791 priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
792 mlx5_flow_ext_mreg_supported(dev)) {
793 rxq_ctrl->rxq.mark = 1;
794 rxq_ctrl->flow_mark_n = 1;
796 rxq_ctrl->flow_mark_n--;
797 rxq_ctrl->rxq.mark = !!rxq_ctrl->flow_mark_n;
802 /* Decrease the counter matching the flow. */
803 for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
804 if ((tunnels_info[j].tunnel &
806 tunnels_info[j].tunnel) {
807 rxq_ctrl->flow_tunnels_n[j]--;
811 flow_rxq_tunnel_ptype_update(rxq_ctrl);
817 * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
818 * @p flow if no other flow uses it with the same kind of request.
821 * Pointer to Ethernet device.
823 * Pointer to the flow.
826 flow_rxq_flags_trim(struct rte_eth_dev *dev, struct rte_flow *flow)
828 struct mlx5_flow *dev_flow;
830 LIST_FOREACH(dev_flow, &flow->dev_flows, next)
831 flow_drv_rxq_flags_trim(dev, dev_flow);
835 * Clear the Mark/Flag and Tunnel ptype information in all Rx queues.
838 * Pointer to Ethernet device.
841 flow_rxq_flags_clear(struct rte_eth_dev *dev)
843 struct mlx5_priv *priv = dev->data->dev_private;
846 for (i = 0; i != priv->rxqs_n; ++i) {
847 struct mlx5_rxq_ctrl *rxq_ctrl;
850 if (!(*priv->rxqs)[i])
852 rxq_ctrl = container_of((*priv->rxqs)[i],
853 struct mlx5_rxq_ctrl, rxq);
854 rxq_ctrl->flow_mark_n = 0;
855 rxq_ctrl->rxq.mark = 0;
856 for (j = 0; j != MLX5_FLOW_TUNNEL; ++j)
857 rxq_ctrl->flow_tunnels_n[j] = 0;
858 rxq_ctrl->rxq.tunnel = 0;
863 * return a pointer to the desired action in the list of actions.
866 * The list of actions to search the action in.
868 * The action to find.
871 * Pointer to the action in the list, if found. NULL otherwise.
873 const struct rte_flow_action *
874 mlx5_flow_find_action(const struct rte_flow_action *actions,
875 enum rte_flow_action_type action)
879 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++)
880 if (actions->type == action)
886 * Validate the flag action.
888 * @param[in] action_flags
889 * Bit-fields that holds the actions detected until now.
891 * Attributes of flow that includes this action.
893 * Pointer to error structure.
896 * 0 on success, a negative errno value otherwise and rte_errno is set.
899 mlx5_flow_validate_action_flag(uint64_t action_flags,
900 const struct rte_flow_attr *attr,
901 struct rte_flow_error *error)
904 if (action_flags & MLX5_FLOW_ACTION_DROP)
905 return rte_flow_error_set(error, EINVAL,
906 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
907 "can't drop and flag in same flow");
908 if (action_flags & MLX5_FLOW_ACTION_MARK)
909 return rte_flow_error_set(error, EINVAL,
910 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
911 "can't mark and flag in same flow");
912 if (action_flags & MLX5_FLOW_ACTION_FLAG)
913 return rte_flow_error_set(error, EINVAL,
914 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
916 " actions in same flow");
918 return rte_flow_error_set(error, ENOTSUP,
919 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
920 "flag action not supported for "
926 * Validate the mark action.
929 * Pointer to the queue action.
930 * @param[in] action_flags
931 * Bit-fields that holds the actions detected until now.
933 * Attributes of flow that includes this action.
935 * Pointer to error structure.
938 * 0 on success, a negative errno value otherwise and rte_errno is set.
941 mlx5_flow_validate_action_mark(const struct rte_flow_action *action,
942 uint64_t action_flags,
943 const struct rte_flow_attr *attr,
944 struct rte_flow_error *error)
946 const struct rte_flow_action_mark *mark = action->conf;
949 return rte_flow_error_set(error, EINVAL,
950 RTE_FLOW_ERROR_TYPE_ACTION,
952 "configuration cannot be null");
953 if (mark->id >= MLX5_FLOW_MARK_MAX)
954 return rte_flow_error_set(error, EINVAL,
955 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
957 "mark id must in 0 <= id < "
958 RTE_STR(MLX5_FLOW_MARK_MAX));
959 if (action_flags & MLX5_FLOW_ACTION_DROP)
960 return rte_flow_error_set(error, EINVAL,
961 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
962 "can't drop and mark in same flow");
963 if (action_flags & MLX5_FLOW_ACTION_FLAG)
964 return rte_flow_error_set(error, EINVAL,
965 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
966 "can't flag and mark in same flow");
967 if (action_flags & MLX5_FLOW_ACTION_MARK)
968 return rte_flow_error_set(error, EINVAL,
969 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
970 "can't have 2 mark actions in same"
973 return rte_flow_error_set(error, ENOTSUP,
974 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
975 "mark action not supported for "
981 * Validate the drop action.
983 * @param[in] action_flags
984 * Bit-fields that holds the actions detected until now.
986 * Attributes of flow that includes this action.
988 * Pointer to error structure.
991 * 0 on success, a negative errno value otherwise and rte_errno is set.
994 mlx5_flow_validate_action_drop(uint64_t action_flags,
995 const struct rte_flow_attr *attr,
996 struct rte_flow_error *error)
998 if (action_flags & MLX5_FLOW_ACTION_FLAG)
999 return rte_flow_error_set(error, EINVAL,
1000 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1001 "can't drop and flag in same flow");
1002 if (action_flags & MLX5_FLOW_ACTION_MARK)
1003 return rte_flow_error_set(error, EINVAL,
1004 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1005 "can't drop and mark in same flow");
1006 if (action_flags & (MLX5_FLOW_FATE_ACTIONS |
1007 MLX5_FLOW_FATE_ESWITCH_ACTIONS))
1008 return rte_flow_error_set(error, EINVAL,
1009 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1010 "can't have 2 fate actions in"
1013 return rte_flow_error_set(error, ENOTSUP,
1014 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1015 "drop action not supported for "
1021 * Validate the queue action.
1024 * Pointer to the queue action.
1025 * @param[in] action_flags
1026 * Bit-fields that holds the actions detected until now.
1028 * Pointer to the Ethernet device structure.
1030 * Attributes of flow that includes this action.
1032 * Pointer to error structure.
1035 * 0 on success, a negative errno value otherwise and rte_errno is set.
1038 mlx5_flow_validate_action_queue(const struct rte_flow_action *action,
1039 uint64_t action_flags,
1040 struct rte_eth_dev *dev,
1041 const struct rte_flow_attr *attr,
1042 struct rte_flow_error *error)
1044 struct mlx5_priv *priv = dev->data->dev_private;
1045 const struct rte_flow_action_queue *queue = action->conf;
1047 if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1048 return rte_flow_error_set(error, EINVAL,
1049 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1050 "can't have 2 fate actions in"
1053 return rte_flow_error_set(error, EINVAL,
1054 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1055 NULL, "No Rx queues configured");
1056 if (queue->index >= priv->rxqs_n)
1057 return rte_flow_error_set(error, EINVAL,
1058 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1060 "queue index out of range");
1061 if (!(*priv->rxqs)[queue->index])
1062 return rte_flow_error_set(error, EINVAL,
1063 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1065 "queue is not configured");
1067 return rte_flow_error_set(error, ENOTSUP,
1068 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1069 "queue action not supported for "
1075 * Validate the rss action.
1078 * Pointer to the queue action.
1079 * @param[in] action_flags
1080 * Bit-fields that holds the actions detected until now.
1082 * Pointer to the Ethernet device structure.
1084 * Attributes of flow that includes this action.
1085 * @param[in] item_flags
1086 * Items that were detected.
1088 * Pointer to error structure.
1091 * 0 on success, a negative errno value otherwise and rte_errno is set.
1094 mlx5_flow_validate_action_rss(const struct rte_flow_action *action,
1095 uint64_t action_flags,
1096 struct rte_eth_dev *dev,
1097 const struct rte_flow_attr *attr,
1098 uint64_t item_flags,
1099 struct rte_flow_error *error)
1101 struct mlx5_priv *priv = dev->data->dev_private;
1102 const struct rte_flow_action_rss *rss = action->conf;
1103 int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1106 if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1107 return rte_flow_error_set(error, EINVAL,
1108 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1109 "can't have 2 fate actions"
1111 if (rss->func != RTE_ETH_HASH_FUNCTION_DEFAULT &&
1112 rss->func != RTE_ETH_HASH_FUNCTION_TOEPLITZ)
1113 return rte_flow_error_set(error, ENOTSUP,
1114 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1116 "RSS hash function not supported");
1117 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
1122 return rte_flow_error_set(error, ENOTSUP,
1123 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1125 "tunnel RSS is not supported");
1126 /* allow RSS key_len 0 in case of NULL (default) RSS key. */
1127 if (rss->key_len == 0 && rss->key != NULL)
1128 return rte_flow_error_set(error, ENOTSUP,
1129 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1131 "RSS hash key length 0");
1132 if (rss->key_len > 0 && rss->key_len < MLX5_RSS_HASH_KEY_LEN)
1133 return rte_flow_error_set(error, ENOTSUP,
1134 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1136 "RSS hash key too small");
1137 if (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 large");
1142 if (rss->queue_num > priv->config.ind_table_max_size)
1143 return rte_flow_error_set(error, ENOTSUP,
1144 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1146 "number of queues too large");
1147 if (rss->types & MLX5_RSS_HF_MASK)
1148 return rte_flow_error_set(error, ENOTSUP,
1149 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1151 "some RSS protocols are not"
1153 if ((rss->types & (ETH_RSS_L3_SRC_ONLY | ETH_RSS_L3_DST_ONLY)) &&
1154 !(rss->types & ETH_RSS_IP))
1155 return rte_flow_error_set(error, EINVAL,
1156 RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1157 "L3 partial RSS requested but L3 RSS"
1158 " type not specified");
1159 if ((rss->types & (ETH_RSS_L4_SRC_ONLY | ETH_RSS_L4_DST_ONLY)) &&
1160 !(rss->types & (ETH_RSS_UDP | ETH_RSS_TCP)))
1161 return rte_flow_error_set(error, EINVAL,
1162 RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1163 "L4 partial RSS requested but L4 RSS"
1164 " type not specified");
1166 return rte_flow_error_set(error, EINVAL,
1167 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1168 NULL, "No Rx queues configured");
1169 if (!rss->queue_num)
1170 return rte_flow_error_set(error, EINVAL,
1171 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1172 NULL, "No queues configured");
1173 for (i = 0; i != rss->queue_num; ++i) {
1174 if (rss->queue[i] >= priv->rxqs_n)
1175 return rte_flow_error_set
1177 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1178 &rss->queue[i], "queue index out of range");
1179 if (!(*priv->rxqs)[rss->queue[i]])
1180 return rte_flow_error_set
1181 (error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1182 &rss->queue[i], "queue is not configured");
1185 return rte_flow_error_set(error, ENOTSUP,
1186 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1187 "rss action not supported for "
1189 if (rss->level > 1 && !tunnel)
1190 return rte_flow_error_set(error, EINVAL,
1191 RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1192 "inner RSS is not supported for "
1193 "non-tunnel flows");
1198 * Validate the count action.
1201 * Pointer to the Ethernet device structure.
1203 * Attributes of flow that includes this action.
1205 * Pointer to error structure.
1208 * 0 on success, a negative errno value otherwise and rte_errno is set.
1211 mlx5_flow_validate_action_count(struct rte_eth_dev *dev __rte_unused,
1212 const struct rte_flow_attr *attr,
1213 struct rte_flow_error *error)
1216 return rte_flow_error_set(error, ENOTSUP,
1217 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1218 "count action not supported for "
1224 * Verify the @p attributes will be correctly understood by the NIC and store
1225 * them in the @p flow if everything is correct.
1228 * Pointer to the Ethernet device structure.
1229 * @param[in] attributes
1230 * Pointer to flow attributes
1232 * Pointer to error structure.
1235 * 0 on success, a negative errno value otherwise and rte_errno is set.
1238 mlx5_flow_validate_attributes(struct rte_eth_dev *dev,
1239 const struct rte_flow_attr *attributes,
1240 struct rte_flow_error *error)
1242 struct mlx5_priv *priv = dev->data->dev_private;
1243 uint32_t priority_max = priv->config.flow_prio - 1;
1245 if (attributes->group)
1246 return rte_flow_error_set(error, ENOTSUP,
1247 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
1248 NULL, "groups is not supported");
1249 if (attributes->priority != MLX5_FLOW_PRIO_RSVD &&
1250 attributes->priority >= priority_max)
1251 return rte_flow_error_set(error, ENOTSUP,
1252 RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
1253 NULL, "priority out of range");
1254 if (attributes->egress)
1255 return rte_flow_error_set(error, ENOTSUP,
1256 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1257 "egress is not supported");
1258 if (attributes->transfer && !priv->config.dv_esw_en)
1259 return rte_flow_error_set(error, ENOTSUP,
1260 RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1261 NULL, "transfer is not supported");
1262 if (!attributes->ingress)
1263 return rte_flow_error_set(error, EINVAL,
1264 RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
1266 "ingress attribute is mandatory");
1271 * Validate ICMP6 item.
1274 * Item specification.
1275 * @param[in] item_flags
1276 * Bit-fields that holds the items detected until now.
1278 * Pointer to error structure.
1281 * 0 on success, a negative errno value otherwise and rte_errno is set.
1284 mlx5_flow_validate_item_icmp6(const struct rte_flow_item *item,
1285 uint64_t item_flags,
1286 uint8_t target_protocol,
1287 struct rte_flow_error *error)
1289 const struct rte_flow_item_icmp6 *mask = item->mask;
1290 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1291 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
1292 MLX5_FLOW_LAYER_OUTER_L3_IPV6;
1293 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1294 MLX5_FLOW_LAYER_OUTER_L4;
1297 if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMPV6)
1298 return rte_flow_error_set(error, EINVAL,
1299 RTE_FLOW_ERROR_TYPE_ITEM, item,
1300 "protocol filtering not compatible"
1301 " with ICMP6 layer");
1302 if (!(item_flags & l3m))
1303 return rte_flow_error_set(error, EINVAL,
1304 RTE_FLOW_ERROR_TYPE_ITEM, item,
1305 "IPv6 is mandatory to filter on"
1307 if (item_flags & l4m)
1308 return rte_flow_error_set(error, EINVAL,
1309 RTE_FLOW_ERROR_TYPE_ITEM, item,
1310 "multiple L4 layers not supported");
1312 mask = &rte_flow_item_icmp6_mask;
1313 ret = mlx5_flow_item_acceptable
1314 (item, (const uint8_t *)mask,
1315 (const uint8_t *)&rte_flow_item_icmp6_mask,
1316 sizeof(struct rte_flow_item_icmp6), error);
1323 * Validate ICMP item.
1326 * Item specification.
1327 * @param[in] item_flags
1328 * Bit-fields that holds the items detected until now.
1330 * Pointer to error structure.
1333 * 0 on success, a negative errno value otherwise and rte_errno is set.
1336 mlx5_flow_validate_item_icmp(const struct rte_flow_item *item,
1337 uint64_t item_flags,
1338 uint8_t target_protocol,
1339 struct rte_flow_error *error)
1341 const struct rte_flow_item_icmp *mask = item->mask;
1342 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1343 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
1344 MLX5_FLOW_LAYER_OUTER_L3_IPV4;
1345 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1346 MLX5_FLOW_LAYER_OUTER_L4;
1349 if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMP)
1350 return rte_flow_error_set(error, EINVAL,
1351 RTE_FLOW_ERROR_TYPE_ITEM, item,
1352 "protocol filtering not compatible"
1353 " with ICMP layer");
1354 if (!(item_flags & l3m))
1355 return rte_flow_error_set(error, EINVAL,
1356 RTE_FLOW_ERROR_TYPE_ITEM, item,
1357 "IPv4 is mandatory to filter"
1359 if (item_flags & l4m)
1360 return rte_flow_error_set(error, EINVAL,
1361 RTE_FLOW_ERROR_TYPE_ITEM, item,
1362 "multiple L4 layers not supported");
1364 mask = &rte_flow_item_icmp_mask;
1365 ret = mlx5_flow_item_acceptable
1366 (item, (const uint8_t *)mask,
1367 (const uint8_t *)&rte_flow_item_icmp_mask,
1368 sizeof(struct rte_flow_item_icmp), error);
1375 * Validate Ethernet item.
1378 * Item specification.
1379 * @param[in] item_flags
1380 * Bit-fields that holds the items detected until now.
1382 * Pointer to error structure.
1385 * 0 on success, a negative errno value otherwise and rte_errno is set.
1388 mlx5_flow_validate_item_eth(const struct rte_flow_item *item,
1389 uint64_t item_flags,
1390 struct rte_flow_error *error)
1392 const struct rte_flow_item_eth *mask = item->mask;
1393 const struct rte_flow_item_eth nic_mask = {
1394 .dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1395 .src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1396 .type = RTE_BE16(0xffff),
1399 int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1400 const uint64_t ethm = tunnel ? MLX5_FLOW_LAYER_INNER_L2 :
1401 MLX5_FLOW_LAYER_OUTER_L2;
1403 if (item_flags & ethm)
1404 return rte_flow_error_set(error, ENOTSUP,
1405 RTE_FLOW_ERROR_TYPE_ITEM, item,
1406 "multiple L2 layers not supported");
1407 if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_L3)) ||
1408 (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_L3)))
1409 return rte_flow_error_set(error, EINVAL,
1410 RTE_FLOW_ERROR_TYPE_ITEM, item,
1411 "L2 layer should not follow "
1413 if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_VLAN)) ||
1414 (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_VLAN)))
1415 return rte_flow_error_set(error, EINVAL,
1416 RTE_FLOW_ERROR_TYPE_ITEM, item,
1417 "L2 layer should not follow VLAN");
1419 mask = &rte_flow_item_eth_mask;
1420 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1421 (const uint8_t *)&nic_mask,
1422 sizeof(struct rte_flow_item_eth),
1428 * Validate VLAN item.
1431 * Item specification.
1432 * @param[in] item_flags
1433 * Bit-fields that holds the items detected until now.
1435 * Ethernet device flow is being created on.
1437 * Pointer to error structure.
1440 * 0 on success, a negative errno value otherwise and rte_errno is set.
1443 mlx5_flow_validate_item_vlan(const struct rte_flow_item *item,
1444 uint64_t item_flags,
1445 struct rte_eth_dev *dev,
1446 struct rte_flow_error *error)
1448 const struct rte_flow_item_vlan *spec = item->spec;
1449 const struct rte_flow_item_vlan *mask = item->mask;
1450 const struct rte_flow_item_vlan nic_mask = {
1451 .tci = RTE_BE16(UINT16_MAX),
1452 .inner_type = RTE_BE16(UINT16_MAX),
1454 uint16_t vlan_tag = 0;
1455 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1457 const uint64_t l34m = tunnel ? (MLX5_FLOW_LAYER_INNER_L3 |
1458 MLX5_FLOW_LAYER_INNER_L4) :
1459 (MLX5_FLOW_LAYER_OUTER_L3 |
1460 MLX5_FLOW_LAYER_OUTER_L4);
1461 const uint64_t vlanm = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
1462 MLX5_FLOW_LAYER_OUTER_VLAN;
1464 if (item_flags & vlanm)
1465 return rte_flow_error_set(error, EINVAL,
1466 RTE_FLOW_ERROR_TYPE_ITEM, item,
1467 "multiple VLAN layers not supported");
1468 else if ((item_flags & l34m) != 0)
1469 return rte_flow_error_set(error, EINVAL,
1470 RTE_FLOW_ERROR_TYPE_ITEM, item,
1471 "VLAN cannot follow L3/L4 layer");
1473 mask = &rte_flow_item_vlan_mask;
1474 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1475 (const uint8_t *)&nic_mask,
1476 sizeof(struct rte_flow_item_vlan),
1480 if (!tunnel && mask->tci != RTE_BE16(0x0fff)) {
1481 struct mlx5_priv *priv = dev->data->dev_private;
1483 if (priv->vmwa_context) {
1485 * Non-NULL context means we have a virtual machine
1486 * and SR-IOV enabled, we have to create VLAN interface
1487 * to make hypervisor to setup E-Switch vport
1488 * context correctly. We avoid creating the multiple
1489 * VLAN interfaces, so we cannot support VLAN tag mask.
1491 return rte_flow_error_set(error, EINVAL,
1492 RTE_FLOW_ERROR_TYPE_ITEM,
1494 "VLAN tag mask is not"
1495 " supported in virtual"
1500 vlan_tag = spec->tci;
1501 vlan_tag &= mask->tci;
1504 * From verbs perspective an empty VLAN is equivalent
1505 * to a packet without VLAN layer.
1508 return rte_flow_error_set(error, EINVAL,
1509 RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
1511 "VLAN cannot be empty");
1516 * Validate IPV4 item.
1519 * Item specification.
1520 * @param[in] item_flags
1521 * Bit-fields that holds the items detected until now.
1522 * @param[in] acc_mask
1523 * Acceptable mask, if NULL default internal default mask
1524 * will be used to check whether item fields are supported.
1526 * Pointer to error structure.
1529 * 0 on success, a negative errno value otherwise and rte_errno is set.
1532 mlx5_flow_validate_item_ipv4(const struct rte_flow_item *item,
1533 uint64_t item_flags,
1535 uint16_t ether_type,
1536 const struct rte_flow_item_ipv4 *acc_mask,
1537 struct rte_flow_error *error)
1539 const struct rte_flow_item_ipv4 *mask = item->mask;
1540 const struct rte_flow_item_ipv4 *spec = item->spec;
1541 const struct rte_flow_item_ipv4 nic_mask = {
1543 .src_addr = RTE_BE32(0xffffffff),
1544 .dst_addr = RTE_BE32(0xffffffff),
1545 .type_of_service = 0xff,
1546 .next_proto_id = 0xff,
1549 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1550 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1551 MLX5_FLOW_LAYER_OUTER_L3;
1552 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1553 MLX5_FLOW_LAYER_OUTER_L4;
1555 uint8_t next_proto = 0xFF;
1556 const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
1557 MLX5_FLOW_LAYER_OUTER_VLAN |
1558 MLX5_FLOW_LAYER_INNER_VLAN);
1560 if ((last_item & l2_vlan) && ether_type &&
1561 ether_type != RTE_ETHER_TYPE_IPV4)
1562 return rte_flow_error_set(error, EINVAL,
1563 RTE_FLOW_ERROR_TYPE_ITEM, item,
1564 "IPv4 cannot follow L2/VLAN layer "
1565 "which ether type is not IPv4");
1566 if (item_flags & MLX5_FLOW_LAYER_IPIP) {
1568 next_proto = mask->hdr.next_proto_id &
1569 spec->hdr.next_proto_id;
1570 if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
1571 return rte_flow_error_set(error, EINVAL,
1572 RTE_FLOW_ERROR_TYPE_ITEM,
1577 if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP)
1578 return rte_flow_error_set(error, EINVAL,
1579 RTE_FLOW_ERROR_TYPE_ITEM, item,
1580 "wrong tunnel type - IPv6 specified "
1581 "but IPv4 item provided");
1582 if (item_flags & l3m)
1583 return rte_flow_error_set(error, ENOTSUP,
1584 RTE_FLOW_ERROR_TYPE_ITEM, item,
1585 "multiple L3 layers not supported");
1586 else if (item_flags & l4m)
1587 return rte_flow_error_set(error, EINVAL,
1588 RTE_FLOW_ERROR_TYPE_ITEM, item,
1589 "L3 cannot follow an L4 layer.");
1590 else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
1591 !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
1592 return rte_flow_error_set(error, EINVAL,
1593 RTE_FLOW_ERROR_TYPE_ITEM, item,
1594 "L3 cannot follow an NVGRE layer.");
1596 mask = &rte_flow_item_ipv4_mask;
1597 else if (mask->hdr.next_proto_id != 0 &&
1598 mask->hdr.next_proto_id != 0xff)
1599 return rte_flow_error_set(error, EINVAL,
1600 RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
1601 "partial mask is not supported"
1603 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1604 acc_mask ? (const uint8_t *)acc_mask
1605 : (const uint8_t *)&nic_mask,
1606 sizeof(struct rte_flow_item_ipv4),
1614 * Validate IPV6 item.
1617 * Item specification.
1618 * @param[in] item_flags
1619 * Bit-fields that holds the items detected until now.
1620 * @param[in] acc_mask
1621 * Acceptable mask, if NULL default internal default mask
1622 * will be used to check whether item fields are supported.
1624 * Pointer to error structure.
1627 * 0 on success, a negative errno value otherwise and rte_errno is set.
1630 mlx5_flow_validate_item_ipv6(const struct rte_flow_item *item,
1631 uint64_t item_flags,
1633 uint16_t ether_type,
1634 const struct rte_flow_item_ipv6 *acc_mask,
1635 struct rte_flow_error *error)
1637 const struct rte_flow_item_ipv6 *mask = item->mask;
1638 const struct rte_flow_item_ipv6 *spec = item->spec;
1639 const struct rte_flow_item_ipv6 nic_mask = {
1642 "\xff\xff\xff\xff\xff\xff\xff\xff"
1643 "\xff\xff\xff\xff\xff\xff\xff\xff",
1645 "\xff\xff\xff\xff\xff\xff\xff\xff"
1646 "\xff\xff\xff\xff\xff\xff\xff\xff",
1647 .vtc_flow = RTE_BE32(0xffffffff),
1652 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1653 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1654 MLX5_FLOW_LAYER_OUTER_L3;
1655 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1656 MLX5_FLOW_LAYER_OUTER_L4;
1658 uint8_t next_proto = 0xFF;
1659 const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
1660 MLX5_FLOW_LAYER_OUTER_VLAN |
1661 MLX5_FLOW_LAYER_INNER_VLAN);
1663 if ((last_item & l2_vlan) && ether_type &&
1664 ether_type != RTE_ETHER_TYPE_IPV6)
1665 return rte_flow_error_set(error, EINVAL,
1666 RTE_FLOW_ERROR_TYPE_ITEM, item,
1667 "IPv6 cannot follow L2/VLAN layer "
1668 "which ether type is not IPv6");
1669 if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP) {
1671 next_proto = mask->hdr.proto & spec->hdr.proto;
1672 if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
1673 return rte_flow_error_set(error, EINVAL,
1674 RTE_FLOW_ERROR_TYPE_ITEM,
1679 if (item_flags & MLX5_FLOW_LAYER_IPIP)
1680 return rte_flow_error_set(error, EINVAL,
1681 RTE_FLOW_ERROR_TYPE_ITEM, item,
1682 "wrong tunnel type - IPv4 specified "
1683 "but IPv6 item provided");
1684 if (item_flags & l3m)
1685 return rte_flow_error_set(error, ENOTSUP,
1686 RTE_FLOW_ERROR_TYPE_ITEM, item,
1687 "multiple L3 layers not supported");
1688 else if (item_flags & l4m)
1689 return rte_flow_error_set(error, EINVAL,
1690 RTE_FLOW_ERROR_TYPE_ITEM, item,
1691 "L3 cannot follow an L4 layer.");
1692 else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
1693 !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
1694 return rte_flow_error_set(error, EINVAL,
1695 RTE_FLOW_ERROR_TYPE_ITEM, item,
1696 "L3 cannot follow an NVGRE layer.");
1698 mask = &rte_flow_item_ipv6_mask;
1699 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1700 acc_mask ? (const uint8_t *)acc_mask
1701 : (const uint8_t *)&nic_mask,
1702 sizeof(struct rte_flow_item_ipv6),
1710 * Validate UDP item.
1713 * Item specification.
1714 * @param[in] item_flags
1715 * Bit-fields that holds the items detected until now.
1716 * @param[in] target_protocol
1717 * The next protocol in the previous item.
1718 * @param[in] flow_mask
1719 * mlx5 flow-specific (DV, verbs, etc.) supported header fields mask.
1721 * Pointer to error structure.
1724 * 0 on success, a negative errno value otherwise and rte_errno is set.
1727 mlx5_flow_validate_item_udp(const struct rte_flow_item *item,
1728 uint64_t item_flags,
1729 uint8_t target_protocol,
1730 struct rte_flow_error *error)
1732 const struct rte_flow_item_udp *mask = item->mask;
1733 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1734 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1735 MLX5_FLOW_LAYER_OUTER_L3;
1736 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1737 MLX5_FLOW_LAYER_OUTER_L4;
1740 if (target_protocol != 0xff && target_protocol != IPPROTO_UDP)
1741 return rte_flow_error_set(error, EINVAL,
1742 RTE_FLOW_ERROR_TYPE_ITEM, item,
1743 "protocol filtering not compatible"
1745 if (!(item_flags & l3m))
1746 return rte_flow_error_set(error, EINVAL,
1747 RTE_FLOW_ERROR_TYPE_ITEM, item,
1748 "L3 is mandatory to filter on L4");
1749 if (item_flags & l4m)
1750 return rte_flow_error_set(error, EINVAL,
1751 RTE_FLOW_ERROR_TYPE_ITEM, item,
1752 "multiple L4 layers not supported");
1754 mask = &rte_flow_item_udp_mask;
1755 ret = mlx5_flow_item_acceptable
1756 (item, (const uint8_t *)mask,
1757 (const uint8_t *)&rte_flow_item_udp_mask,
1758 sizeof(struct rte_flow_item_udp), error);
1765 * Validate TCP item.
1768 * Item specification.
1769 * @param[in] item_flags
1770 * Bit-fields that holds the items detected until now.
1771 * @param[in] target_protocol
1772 * The next protocol in the previous item.
1774 * Pointer to error structure.
1777 * 0 on success, a negative errno value otherwise and rte_errno is set.
1780 mlx5_flow_validate_item_tcp(const struct rte_flow_item *item,
1781 uint64_t item_flags,
1782 uint8_t target_protocol,
1783 const struct rte_flow_item_tcp *flow_mask,
1784 struct rte_flow_error *error)
1786 const struct rte_flow_item_tcp *mask = item->mask;
1787 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1788 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1789 MLX5_FLOW_LAYER_OUTER_L3;
1790 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1791 MLX5_FLOW_LAYER_OUTER_L4;
1795 if (target_protocol != 0xff && target_protocol != IPPROTO_TCP)
1796 return rte_flow_error_set(error, EINVAL,
1797 RTE_FLOW_ERROR_TYPE_ITEM, item,
1798 "protocol filtering not compatible"
1800 if (!(item_flags & l3m))
1801 return rte_flow_error_set(error, EINVAL,
1802 RTE_FLOW_ERROR_TYPE_ITEM, item,
1803 "L3 is mandatory to filter on L4");
1804 if (item_flags & l4m)
1805 return rte_flow_error_set(error, EINVAL,
1806 RTE_FLOW_ERROR_TYPE_ITEM, item,
1807 "multiple L4 layers not supported");
1809 mask = &rte_flow_item_tcp_mask;
1810 ret = mlx5_flow_item_acceptable
1811 (item, (const uint8_t *)mask,
1812 (const uint8_t *)flow_mask,
1813 sizeof(struct rte_flow_item_tcp), error);
1820 * Validate VXLAN item.
1823 * Item specification.
1824 * @param[in] item_flags
1825 * Bit-fields that holds the items detected until now.
1826 * @param[in] target_protocol
1827 * The next protocol in the previous item.
1829 * Pointer to error structure.
1832 * 0 on success, a negative errno value otherwise and rte_errno is set.
1835 mlx5_flow_validate_item_vxlan(const struct rte_flow_item *item,
1836 uint64_t item_flags,
1837 struct rte_flow_error *error)
1839 const struct rte_flow_item_vxlan *spec = item->spec;
1840 const struct rte_flow_item_vxlan *mask = item->mask;
1845 } id = { .vlan_id = 0, };
1846 uint32_t vlan_id = 0;
1849 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1850 return rte_flow_error_set(error, ENOTSUP,
1851 RTE_FLOW_ERROR_TYPE_ITEM, item,
1852 "multiple tunnel layers not"
1855 * Verify only UDPv4 is present as defined in
1856 * https://tools.ietf.org/html/rfc7348
1858 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
1859 return rte_flow_error_set(error, EINVAL,
1860 RTE_FLOW_ERROR_TYPE_ITEM, item,
1861 "no outer UDP layer found");
1863 mask = &rte_flow_item_vxlan_mask;
1864 ret = mlx5_flow_item_acceptable
1865 (item, (const uint8_t *)mask,
1866 (const uint8_t *)&rte_flow_item_vxlan_mask,
1867 sizeof(struct rte_flow_item_vxlan),
1872 memcpy(&id.vni[1], spec->vni, 3);
1873 vlan_id = id.vlan_id;
1874 memcpy(&id.vni[1], mask->vni, 3);
1875 vlan_id &= id.vlan_id;
1878 * Tunnel id 0 is equivalent as not adding a VXLAN layer, if
1879 * only this layer is defined in the Verbs specification it is
1880 * interpreted as wildcard and all packets will match this
1881 * rule, if it follows a full stack layer (ex: eth / ipv4 /
1882 * udp), all packets matching the layers before will also
1883 * match this rule. To avoid such situation, VNI 0 is
1884 * currently refused.
1887 return rte_flow_error_set(error, ENOTSUP,
1888 RTE_FLOW_ERROR_TYPE_ITEM, item,
1889 "VXLAN vni cannot be 0");
1890 if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
1891 return rte_flow_error_set(error, ENOTSUP,
1892 RTE_FLOW_ERROR_TYPE_ITEM, item,
1893 "VXLAN tunnel must be fully defined");
1898 * Validate VXLAN_GPE item.
1901 * Item specification.
1902 * @param[in] item_flags
1903 * Bit-fields that holds the items detected until now.
1905 * Pointer to the private data structure.
1906 * @param[in] target_protocol
1907 * The next protocol in the previous item.
1909 * Pointer to error structure.
1912 * 0 on success, a negative errno value otherwise and rte_errno is set.
1915 mlx5_flow_validate_item_vxlan_gpe(const struct rte_flow_item *item,
1916 uint64_t item_flags,
1917 struct rte_eth_dev *dev,
1918 struct rte_flow_error *error)
1920 struct mlx5_priv *priv = dev->data->dev_private;
1921 const struct rte_flow_item_vxlan_gpe *spec = item->spec;
1922 const struct rte_flow_item_vxlan_gpe *mask = item->mask;
1927 } id = { .vlan_id = 0, };
1928 uint32_t vlan_id = 0;
1930 if (!priv->config.l3_vxlan_en)
1931 return rte_flow_error_set(error, ENOTSUP,
1932 RTE_FLOW_ERROR_TYPE_ITEM, item,
1933 "L3 VXLAN is not enabled by device"
1934 " parameter and/or not configured in"
1936 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1937 return rte_flow_error_set(error, ENOTSUP,
1938 RTE_FLOW_ERROR_TYPE_ITEM, item,
1939 "multiple tunnel layers not"
1942 * Verify only UDPv4 is present as defined in
1943 * https://tools.ietf.org/html/rfc7348
1945 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
1946 return rte_flow_error_set(error, EINVAL,
1947 RTE_FLOW_ERROR_TYPE_ITEM, item,
1948 "no outer UDP layer found");
1950 mask = &rte_flow_item_vxlan_gpe_mask;
1951 ret = mlx5_flow_item_acceptable
1952 (item, (const uint8_t *)mask,
1953 (const uint8_t *)&rte_flow_item_vxlan_gpe_mask,
1954 sizeof(struct rte_flow_item_vxlan_gpe),
1960 return rte_flow_error_set(error, ENOTSUP,
1961 RTE_FLOW_ERROR_TYPE_ITEM,
1963 "VxLAN-GPE protocol"
1965 memcpy(&id.vni[1], spec->vni, 3);
1966 vlan_id = id.vlan_id;
1967 memcpy(&id.vni[1], mask->vni, 3);
1968 vlan_id &= id.vlan_id;
1971 * Tunnel id 0 is equivalent as not adding a VXLAN layer, if only this
1972 * layer is defined in the Verbs specification it is interpreted as
1973 * wildcard and all packets will match this rule, if it follows a full
1974 * stack layer (ex: eth / ipv4 / udp), all packets matching the layers
1975 * before will also match this rule. To avoid such situation, VNI 0
1976 * is currently refused.
1979 return rte_flow_error_set(error, ENOTSUP,
1980 RTE_FLOW_ERROR_TYPE_ITEM, item,
1981 "VXLAN-GPE vni cannot be 0");
1982 if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
1983 return rte_flow_error_set(error, ENOTSUP,
1984 RTE_FLOW_ERROR_TYPE_ITEM, item,
1985 "VXLAN-GPE tunnel must be fully"
1990 * Validate GRE Key item.
1993 * Item specification.
1994 * @param[in] item_flags
1995 * Bit flags to mark detected items.
1996 * @param[in] gre_item
1997 * Pointer to gre_item
1999 * Pointer to error structure.
2002 * 0 on success, a negative errno value otherwise and rte_errno is set.
2005 mlx5_flow_validate_item_gre_key(const struct rte_flow_item *item,
2006 uint64_t item_flags,
2007 const struct rte_flow_item *gre_item,
2008 struct rte_flow_error *error)
2010 const rte_be32_t *mask = item->mask;
2012 rte_be32_t gre_key_default_mask = RTE_BE32(UINT32_MAX);
2013 const struct rte_flow_item_gre *gre_spec;
2014 const struct rte_flow_item_gre *gre_mask;
2016 if (item_flags & MLX5_FLOW_LAYER_GRE_KEY)
2017 return rte_flow_error_set(error, ENOTSUP,
2018 RTE_FLOW_ERROR_TYPE_ITEM, item,
2019 "Multiple GRE key not support");
2020 if (!(item_flags & MLX5_FLOW_LAYER_GRE))
2021 return rte_flow_error_set(error, ENOTSUP,
2022 RTE_FLOW_ERROR_TYPE_ITEM, item,
2023 "No preceding GRE header");
2024 if (item_flags & MLX5_FLOW_LAYER_INNER)
2025 return rte_flow_error_set(error, ENOTSUP,
2026 RTE_FLOW_ERROR_TYPE_ITEM, item,
2027 "GRE key following a wrong item");
2028 gre_mask = gre_item->mask;
2030 gre_mask = &rte_flow_item_gre_mask;
2031 gre_spec = gre_item->spec;
2032 if (gre_spec && (gre_mask->c_rsvd0_ver & RTE_BE16(0x2000)) &&
2033 !(gre_spec->c_rsvd0_ver & RTE_BE16(0x2000)))
2034 return rte_flow_error_set(error, EINVAL,
2035 RTE_FLOW_ERROR_TYPE_ITEM, item,
2036 "Key bit must be on");
2039 mask = &gre_key_default_mask;
2040 ret = mlx5_flow_item_acceptable
2041 (item, (const uint8_t *)mask,
2042 (const uint8_t *)&gre_key_default_mask,
2043 sizeof(rte_be32_t), error);
2048 * Validate GRE item.
2051 * Item specification.
2052 * @param[in] item_flags
2053 * Bit flags to mark detected items.
2054 * @param[in] target_protocol
2055 * The next protocol in the previous item.
2057 * Pointer to error structure.
2060 * 0 on success, a negative errno value otherwise and rte_errno is set.
2063 mlx5_flow_validate_item_gre(const struct rte_flow_item *item,
2064 uint64_t item_flags,
2065 uint8_t target_protocol,
2066 struct rte_flow_error *error)
2068 const struct rte_flow_item_gre *spec __rte_unused = item->spec;
2069 const struct rte_flow_item_gre *mask = item->mask;
2071 const struct rte_flow_item_gre nic_mask = {
2072 .c_rsvd0_ver = RTE_BE16(0xB000),
2073 .protocol = RTE_BE16(UINT16_MAX),
2076 if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2077 return rte_flow_error_set(error, EINVAL,
2078 RTE_FLOW_ERROR_TYPE_ITEM, item,
2079 "protocol filtering not compatible"
2080 " with this GRE layer");
2081 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2082 return rte_flow_error_set(error, ENOTSUP,
2083 RTE_FLOW_ERROR_TYPE_ITEM, item,
2084 "multiple tunnel layers not"
2086 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2087 return rte_flow_error_set(error, ENOTSUP,
2088 RTE_FLOW_ERROR_TYPE_ITEM, item,
2089 "L3 Layer is missing");
2091 mask = &rte_flow_item_gre_mask;
2092 ret = mlx5_flow_item_acceptable
2093 (item, (const uint8_t *)mask,
2094 (const uint8_t *)&nic_mask,
2095 sizeof(struct rte_flow_item_gre), error);
2098 #ifndef HAVE_MLX5DV_DR
2099 #ifndef HAVE_IBV_DEVICE_MPLS_SUPPORT
2100 if (spec && (spec->protocol & mask->protocol))
2101 return rte_flow_error_set(error, ENOTSUP,
2102 RTE_FLOW_ERROR_TYPE_ITEM, item,
2103 "without MPLS support the"
2104 " specification cannot be used for"
2112 * Validate Geneve item.
2115 * Item specification.
2116 * @param[in] itemFlags
2117 * Bit-fields that holds the items detected until now.
2119 * Pointer to the private data structure.
2121 * Pointer to error structure.
2124 * 0 on success, a negative errno value otherwise and rte_errno is set.
2128 mlx5_flow_validate_item_geneve(const struct rte_flow_item *item,
2129 uint64_t item_flags,
2130 struct rte_eth_dev *dev,
2131 struct rte_flow_error *error)
2133 struct mlx5_priv *priv = dev->data->dev_private;
2134 const struct rte_flow_item_geneve *spec = item->spec;
2135 const struct rte_flow_item_geneve *mask = item->mask;
2138 uint8_t opt_len = priv->config.hca_attr.geneve_max_opt_len ?
2139 MLX5_GENEVE_OPT_LEN_1 : MLX5_GENEVE_OPT_LEN_0;
2140 const struct rte_flow_item_geneve nic_mask = {
2141 .ver_opt_len_o_c_rsvd0 = RTE_BE16(0x3f80),
2142 .vni = "\xff\xff\xff",
2143 .protocol = RTE_BE16(UINT16_MAX),
2146 if (!(priv->config.hca_attr.flex_parser_protocols &
2147 MLX5_HCA_FLEX_GENEVE_ENABLED) ||
2148 !priv->config.hca_attr.tunnel_stateless_geneve_rx)
2149 return rte_flow_error_set(error, ENOTSUP,
2150 RTE_FLOW_ERROR_TYPE_ITEM, item,
2151 "L3 Geneve is not enabled by device"
2152 " parameter and/or not configured in"
2154 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2155 return rte_flow_error_set(error, ENOTSUP,
2156 RTE_FLOW_ERROR_TYPE_ITEM, item,
2157 "multiple tunnel layers not"
2160 * Verify only UDPv4 is present as defined in
2161 * https://tools.ietf.org/html/rfc7348
2163 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2164 return rte_flow_error_set(error, EINVAL,
2165 RTE_FLOW_ERROR_TYPE_ITEM, item,
2166 "no outer UDP layer found");
2168 mask = &rte_flow_item_geneve_mask;
2169 ret = mlx5_flow_item_acceptable
2170 (item, (const uint8_t *)mask,
2171 (const uint8_t *)&nic_mask,
2172 sizeof(struct rte_flow_item_geneve), error);
2176 gbhdr = rte_be_to_cpu_16(spec->ver_opt_len_o_c_rsvd0);
2177 if (MLX5_GENEVE_VER_VAL(gbhdr) ||
2178 MLX5_GENEVE_CRITO_VAL(gbhdr) ||
2179 MLX5_GENEVE_RSVD_VAL(gbhdr) || spec->rsvd1)
2180 return rte_flow_error_set(error, ENOTSUP,
2181 RTE_FLOW_ERROR_TYPE_ITEM,
2183 "Geneve protocol unsupported"
2184 " fields are being used");
2185 if (MLX5_GENEVE_OPTLEN_VAL(gbhdr) > opt_len)
2186 return rte_flow_error_set
2188 RTE_FLOW_ERROR_TYPE_ITEM,
2190 "Unsupported Geneve options length");
2192 if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2193 return rte_flow_error_set
2195 RTE_FLOW_ERROR_TYPE_ITEM, item,
2196 "Geneve tunnel must be fully defined");
2201 * Validate MPLS item.
2204 * Pointer to the rte_eth_dev structure.
2206 * Item specification.
2207 * @param[in] item_flags
2208 * Bit-fields that holds the items detected until now.
2209 * @param[in] prev_layer
2210 * The protocol layer indicated in previous item.
2212 * Pointer to error structure.
2215 * 0 on success, a negative errno value otherwise and rte_errno is set.
2218 mlx5_flow_validate_item_mpls(struct rte_eth_dev *dev __rte_unused,
2219 const struct rte_flow_item *item __rte_unused,
2220 uint64_t item_flags __rte_unused,
2221 uint64_t prev_layer __rte_unused,
2222 struct rte_flow_error *error)
2224 #ifdef HAVE_IBV_DEVICE_MPLS_SUPPORT
2225 const struct rte_flow_item_mpls *mask = item->mask;
2226 struct mlx5_priv *priv = dev->data->dev_private;
2229 if (!priv->config.mpls_en)
2230 return rte_flow_error_set(error, ENOTSUP,
2231 RTE_FLOW_ERROR_TYPE_ITEM, item,
2232 "MPLS not supported or"
2233 " disabled in firmware"
2235 /* MPLS over IP, UDP, GRE is allowed */
2236 if (!(prev_layer & (MLX5_FLOW_LAYER_OUTER_L3 |
2237 MLX5_FLOW_LAYER_OUTER_L4_UDP |
2238 MLX5_FLOW_LAYER_GRE)))
2239 return rte_flow_error_set(error, EINVAL,
2240 RTE_FLOW_ERROR_TYPE_ITEM, item,
2241 "protocol filtering not compatible"
2242 " with MPLS layer");
2243 /* Multi-tunnel isn't allowed but MPLS over GRE is an exception. */
2244 if ((item_flags & MLX5_FLOW_LAYER_TUNNEL) &&
2245 !(item_flags & MLX5_FLOW_LAYER_GRE))
2246 return rte_flow_error_set(error, ENOTSUP,
2247 RTE_FLOW_ERROR_TYPE_ITEM, item,
2248 "multiple tunnel layers not"
2251 mask = &rte_flow_item_mpls_mask;
2252 ret = mlx5_flow_item_acceptable
2253 (item, (const uint8_t *)mask,
2254 (const uint8_t *)&rte_flow_item_mpls_mask,
2255 sizeof(struct rte_flow_item_mpls), error);
2260 return rte_flow_error_set(error, ENOTSUP,
2261 RTE_FLOW_ERROR_TYPE_ITEM, item,
2262 "MPLS is not supported by Verbs, please"
2267 * Validate NVGRE item.
2270 * Item specification.
2271 * @param[in] item_flags
2272 * Bit flags to mark detected items.
2273 * @param[in] target_protocol
2274 * The next protocol in the previous item.
2276 * Pointer to error structure.
2279 * 0 on success, a negative errno value otherwise and rte_errno is set.
2282 mlx5_flow_validate_item_nvgre(const struct rte_flow_item *item,
2283 uint64_t item_flags,
2284 uint8_t target_protocol,
2285 struct rte_flow_error *error)
2287 const struct rte_flow_item_nvgre *mask = item->mask;
2290 if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2291 return rte_flow_error_set(error, EINVAL,
2292 RTE_FLOW_ERROR_TYPE_ITEM, item,
2293 "protocol filtering not compatible"
2294 " with this GRE layer");
2295 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2296 return rte_flow_error_set(error, ENOTSUP,
2297 RTE_FLOW_ERROR_TYPE_ITEM, item,
2298 "multiple tunnel layers not"
2300 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2301 return rte_flow_error_set(error, ENOTSUP,
2302 RTE_FLOW_ERROR_TYPE_ITEM, item,
2303 "L3 Layer is missing");
2305 mask = &rte_flow_item_nvgre_mask;
2306 ret = mlx5_flow_item_acceptable
2307 (item, (const uint8_t *)mask,
2308 (const uint8_t *)&rte_flow_item_nvgre_mask,
2309 sizeof(struct rte_flow_item_nvgre), error);
2315 /* Allocate unique ID for the split Q/RSS subflows. */
2317 flow_qrss_get_id(struct rte_eth_dev *dev)
2319 struct mlx5_priv *priv = dev->data->dev_private;
2320 uint32_t qrss_id, ret;
2322 ret = mlx5_flow_id_get(priv->qrss_id_pool, &qrss_id);
2329 /* Free unique ID for the split Q/RSS subflows. */
2331 flow_qrss_free_id(struct rte_eth_dev *dev, uint32_t qrss_id)
2333 struct mlx5_priv *priv = dev->data->dev_private;
2336 mlx5_flow_id_release(priv->qrss_id_pool, qrss_id);
2340 * Release resource related QUEUE/RSS action split.
2343 * Pointer to Ethernet device.
2345 * Flow to release id's from.
2348 flow_mreg_split_qrss_release(struct rte_eth_dev *dev,
2349 struct rte_flow *flow)
2351 struct mlx5_flow *dev_flow;
2353 LIST_FOREACH(dev_flow, &flow->dev_flows, next)
2354 if (dev_flow->qrss_id)
2355 flow_qrss_free_id(dev, dev_flow->qrss_id);
2359 flow_null_validate(struct rte_eth_dev *dev __rte_unused,
2360 const struct rte_flow_attr *attr __rte_unused,
2361 const struct rte_flow_item items[] __rte_unused,
2362 const struct rte_flow_action actions[] __rte_unused,
2363 bool external __rte_unused,
2364 struct rte_flow_error *error)
2366 return rte_flow_error_set(error, ENOTSUP,
2367 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2370 static struct mlx5_flow *
2371 flow_null_prepare(const struct rte_flow_attr *attr __rte_unused,
2372 const struct rte_flow_item items[] __rte_unused,
2373 const struct rte_flow_action actions[] __rte_unused,
2374 struct rte_flow_error *error)
2376 rte_flow_error_set(error, ENOTSUP,
2377 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2382 flow_null_translate(struct rte_eth_dev *dev __rte_unused,
2383 struct mlx5_flow *dev_flow __rte_unused,
2384 const struct rte_flow_attr *attr __rte_unused,
2385 const struct rte_flow_item items[] __rte_unused,
2386 const struct rte_flow_action actions[] __rte_unused,
2387 struct rte_flow_error *error)
2389 return rte_flow_error_set(error, ENOTSUP,
2390 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2394 flow_null_apply(struct rte_eth_dev *dev __rte_unused,
2395 struct rte_flow *flow __rte_unused,
2396 struct rte_flow_error *error)
2398 return rte_flow_error_set(error, ENOTSUP,
2399 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2403 flow_null_remove(struct rte_eth_dev *dev __rte_unused,
2404 struct rte_flow *flow __rte_unused)
2409 flow_null_destroy(struct rte_eth_dev *dev __rte_unused,
2410 struct rte_flow *flow __rte_unused)
2415 flow_null_query(struct rte_eth_dev *dev __rte_unused,
2416 struct rte_flow *flow __rte_unused,
2417 const struct rte_flow_action *actions __rte_unused,
2418 void *data __rte_unused,
2419 struct rte_flow_error *error)
2421 return rte_flow_error_set(error, ENOTSUP,
2422 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2425 /* Void driver to protect from null pointer reference. */
2426 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops = {
2427 .validate = flow_null_validate,
2428 .prepare = flow_null_prepare,
2429 .translate = flow_null_translate,
2430 .apply = flow_null_apply,
2431 .remove = flow_null_remove,
2432 .destroy = flow_null_destroy,
2433 .query = flow_null_query,
2437 * Select flow driver type according to flow attributes and device
2441 * Pointer to the dev structure.
2443 * Pointer to the flow attributes.
2446 * flow driver type, MLX5_FLOW_TYPE_MAX otherwise.
2448 static enum mlx5_flow_drv_type
2449 flow_get_drv_type(struct rte_eth_dev *dev, const struct rte_flow_attr *attr)
2451 struct mlx5_priv *priv = dev->data->dev_private;
2452 enum mlx5_flow_drv_type type = MLX5_FLOW_TYPE_MAX;
2454 if (attr->transfer && priv->config.dv_esw_en)
2455 type = MLX5_FLOW_TYPE_DV;
2456 if (!attr->transfer)
2457 type = priv->config.dv_flow_en ? MLX5_FLOW_TYPE_DV :
2458 MLX5_FLOW_TYPE_VERBS;
2462 #define flow_get_drv_ops(type) flow_drv_ops[type]
2465 * Flow driver validation API. This abstracts calling driver specific functions.
2466 * The type of flow driver is determined according to flow attributes.
2469 * Pointer to the dev structure.
2471 * Pointer to the flow attributes.
2473 * Pointer to the list of items.
2474 * @param[in] actions
2475 * Pointer to the list of actions.
2476 * @param[in] external
2477 * This flow rule is created by request external to PMD.
2479 * Pointer to the error structure.
2482 * 0 on success, a negative errno value otherwise and rte_errno is set.
2485 flow_drv_validate(struct rte_eth_dev *dev,
2486 const struct rte_flow_attr *attr,
2487 const struct rte_flow_item items[],
2488 const struct rte_flow_action actions[],
2489 bool external, struct rte_flow_error *error)
2491 const struct mlx5_flow_driver_ops *fops;
2492 enum mlx5_flow_drv_type type = flow_get_drv_type(dev, attr);
2494 fops = flow_get_drv_ops(type);
2495 return fops->validate(dev, attr, items, actions, external, error);
2499 * Flow driver preparation API. This abstracts calling driver specific
2500 * functions. Parent flow (rte_flow) should have driver type (drv_type). It
2501 * calculates the size of memory required for device flow, allocates the memory,
2502 * initializes the device flow and returns the pointer.
2505 * This function initializes device flow structure such as dv or verbs in
2506 * struct mlx5_flow. However, it is caller's responsibility to initialize the
2507 * rest. For example, adding returning device flow to flow->dev_flow list and
2508 * setting backward reference to the flow should be done out of this function.
2509 * layers field is not filled either.
2512 * Pointer to the flow attributes.
2514 * Pointer to the list of items.
2515 * @param[in] actions
2516 * Pointer to the list of actions.
2518 * Pointer to the error structure.
2521 * Pointer to device flow on success, otherwise NULL and rte_errno is set.
2523 static inline struct mlx5_flow *
2524 flow_drv_prepare(const struct rte_flow *flow,
2525 const struct rte_flow_attr *attr,
2526 const struct rte_flow_item items[],
2527 const struct rte_flow_action actions[],
2528 struct rte_flow_error *error)
2530 const struct mlx5_flow_driver_ops *fops;
2531 enum mlx5_flow_drv_type type = flow->drv_type;
2533 assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2534 fops = flow_get_drv_ops(type);
2535 return fops->prepare(attr, items, actions, error);
2539 * Flow driver translation API. This abstracts calling driver specific
2540 * functions. Parent flow (rte_flow) should have driver type (drv_type). It
2541 * translates a generic flow into a driver flow. flow_drv_prepare() must
2545 * dev_flow->layers could be filled as a result of parsing during translation
2546 * if needed by flow_drv_apply(). dev_flow->flow->actions can also be filled
2547 * if necessary. As a flow can have multiple dev_flows by RSS flow expansion,
2548 * flow->actions could be overwritten even though all the expanded dev_flows
2549 * have the same actions.
2552 * Pointer to the rte dev structure.
2553 * @param[in, out] dev_flow
2554 * Pointer to the mlx5 flow.
2556 * Pointer to the flow attributes.
2558 * Pointer to the list of items.
2559 * @param[in] actions
2560 * Pointer to the list of actions.
2562 * Pointer to the error structure.
2565 * 0 on success, a negative errno value otherwise and rte_errno is set.
2568 flow_drv_translate(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow,
2569 const struct rte_flow_attr *attr,
2570 const struct rte_flow_item items[],
2571 const struct rte_flow_action actions[],
2572 struct rte_flow_error *error)
2574 const struct mlx5_flow_driver_ops *fops;
2575 enum mlx5_flow_drv_type type = dev_flow->flow->drv_type;
2577 assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2578 fops = flow_get_drv_ops(type);
2579 return fops->translate(dev, dev_flow, attr, items, actions, error);
2583 * Flow driver apply API. This abstracts calling driver specific functions.
2584 * Parent flow (rte_flow) should have driver type (drv_type). It applies
2585 * translated driver flows on to device. flow_drv_translate() must precede.
2588 * Pointer to Ethernet device structure.
2589 * @param[in, out] flow
2590 * Pointer to flow structure.
2592 * Pointer to error structure.
2595 * 0 on success, a negative errno value otherwise and rte_errno is set.
2598 flow_drv_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
2599 struct rte_flow_error *error)
2601 const struct mlx5_flow_driver_ops *fops;
2602 enum mlx5_flow_drv_type type = flow->drv_type;
2604 assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2605 fops = flow_get_drv_ops(type);
2606 return fops->apply(dev, flow, error);
2610 * Flow driver remove 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. All the resources of the flow should be freed by calling
2613 * flow_drv_destroy().
2616 * Pointer to Ethernet device.
2617 * @param[in, out] flow
2618 * Pointer to flow structure.
2621 flow_drv_remove(struct rte_eth_dev *dev, struct rte_flow *flow)
2623 const struct mlx5_flow_driver_ops *fops;
2624 enum mlx5_flow_drv_type type = flow->drv_type;
2626 assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2627 fops = flow_get_drv_ops(type);
2628 fops->remove(dev, flow);
2632 * Flow driver destroy API. This abstracts calling driver specific functions.
2633 * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
2634 * on device and releases resources of the flow.
2637 * Pointer to Ethernet device.
2638 * @param[in, out] flow
2639 * Pointer to flow structure.
2642 flow_drv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
2644 const struct mlx5_flow_driver_ops *fops;
2645 enum mlx5_flow_drv_type type = flow->drv_type;
2647 flow_mreg_split_qrss_release(dev, flow);
2648 assert(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2649 fops = flow_get_drv_ops(type);
2650 fops->destroy(dev, flow);
2654 * Validate a flow supported by the NIC.
2656 * @see rte_flow_validate()
2660 mlx5_flow_validate(struct rte_eth_dev *dev,
2661 const struct rte_flow_attr *attr,
2662 const struct rte_flow_item items[],
2663 const struct rte_flow_action actions[],
2664 struct rte_flow_error *error)
2668 ret = flow_drv_validate(dev, attr, items, actions, true, error);
2675 * Get port id item from the item list.
2678 * Pointer to the list of items.
2681 * Pointer to the port id item if exist, else return NULL.
2683 static const struct rte_flow_item *
2684 find_port_id_item(const struct rte_flow_item *item)
2687 for (; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
2688 if (item->type == RTE_FLOW_ITEM_TYPE_PORT_ID)
2695 * Get RSS action from the action list.
2697 * @param[in] actions
2698 * Pointer to the list of actions.
2701 * Pointer to the RSS action if exist, else return NULL.
2703 static const struct rte_flow_action_rss*
2704 flow_get_rss_action(const struct rte_flow_action actions[])
2706 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2707 switch (actions->type) {
2708 case RTE_FLOW_ACTION_TYPE_RSS:
2709 return (const struct rte_flow_action_rss *)
2719 find_graph_root(const struct rte_flow_item pattern[], uint32_t rss_level)
2721 const struct rte_flow_item *item;
2722 unsigned int has_vlan = 0;
2724 for (item = pattern; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
2725 if (item->type == RTE_FLOW_ITEM_TYPE_VLAN) {
2731 return rss_level < 2 ? MLX5_EXPANSION_ROOT_ETH_VLAN :
2732 MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN;
2733 return rss_level < 2 ? MLX5_EXPANSION_ROOT :
2734 MLX5_EXPANSION_ROOT_OUTER;
2738 * Get QUEUE/RSS action from the action list.
2740 * @param[in] actions
2741 * Pointer to the list of actions.
2743 * Pointer to the return pointer.
2744 * @param[out] qrss_type
2745 * Pointer to the action type to return. RTE_FLOW_ACTION_TYPE_END is returned
2746 * if no QUEUE/RSS is found.
2749 * Total number of actions.
2752 flow_parse_qrss_action(const struct rte_flow_action actions[],
2753 const struct rte_flow_action **qrss)
2757 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2758 switch (actions->type) {
2759 case RTE_FLOW_ACTION_TYPE_QUEUE:
2760 case RTE_FLOW_ACTION_TYPE_RSS:
2768 /* Count RTE_FLOW_ACTION_TYPE_END. */
2769 return actions_n + 1;
2773 * Check meter action from the action list.
2775 * @param[in] actions
2776 * Pointer to the list of actions.
2778 * Pointer to the meter exist flag.
2781 * Total number of actions.
2784 flow_check_meter_action(const struct rte_flow_action actions[], uint32_t *mtr)
2790 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2791 switch (actions->type) {
2792 case RTE_FLOW_ACTION_TYPE_METER:
2800 /* Count RTE_FLOW_ACTION_TYPE_END. */
2801 return actions_n + 1;
2805 * Check if the flow should be splited due to hairpin.
2806 * The reason for the split is that in current HW we can't
2807 * support encap on Rx, so if a flow have encap we move it
2811 * Pointer to Ethernet device.
2813 * Flow rule attributes.
2814 * @param[in] actions
2815 * Associated actions (list terminated by the END action).
2818 * > 0 the number of actions and the flow should be split,
2819 * 0 when no split required.
2822 flow_check_hairpin_split(struct rte_eth_dev *dev,
2823 const struct rte_flow_attr *attr,
2824 const struct rte_flow_action actions[])
2826 int queue_action = 0;
2829 const struct rte_flow_action_queue *queue;
2830 const struct rte_flow_action_rss *rss;
2831 const struct rte_flow_action_raw_encap *raw_encap;
2835 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2836 switch (actions->type) {
2837 case RTE_FLOW_ACTION_TYPE_QUEUE:
2838 queue = actions->conf;
2841 if (mlx5_rxq_get_type(dev, queue->index) !=
2842 MLX5_RXQ_TYPE_HAIRPIN)
2847 case RTE_FLOW_ACTION_TYPE_RSS:
2848 rss = actions->conf;
2849 if (rss == NULL || rss->queue_num == 0)
2851 if (mlx5_rxq_get_type(dev, rss->queue[0]) !=
2852 MLX5_RXQ_TYPE_HAIRPIN)
2857 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
2858 case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
2862 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
2863 raw_encap = actions->conf;
2864 if (raw_encap->size >
2865 (sizeof(struct rte_flow_item_eth) +
2866 sizeof(struct rte_flow_item_ipv4)))
2875 if (encap == 1 && queue_action)
2880 /* Declare flow create/destroy prototype in advance. */
2881 static struct rte_flow *
2882 flow_list_create(struct rte_eth_dev *dev, struct mlx5_flows *list,
2883 const struct rte_flow_attr *attr,
2884 const struct rte_flow_item items[],
2885 const struct rte_flow_action actions[],
2886 bool external, struct rte_flow_error *error);
2889 flow_list_destroy(struct rte_eth_dev *dev, struct mlx5_flows *list,
2890 struct rte_flow *flow);
2893 * Add a flow of copying flow metadata registers in RX_CP_TBL.
2895 * As mark_id is unique, if there's already a registered flow for the mark_id,
2896 * return by increasing the reference counter of the resource. Otherwise, create
2897 * the resource (mcp_res) and flow.
2900 * - If ingress port is ANY and reg_c[1] is mark_id,
2901 * flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
2903 * For default flow (zero mark_id), flow is like,
2904 * - If ingress port is ANY,
2905 * reg_b := reg_c[0] and jump to RX_ACT_TBL.
2908 * Pointer to Ethernet device.
2910 * ID of MARK action, zero means default flow for META.
2912 * Perform verbose error reporting if not NULL.
2915 * Associated resource on success, NULL otherwise and rte_errno is set.
2917 static struct mlx5_flow_mreg_copy_resource *
2918 flow_mreg_add_copy_action(struct rte_eth_dev *dev, uint32_t mark_id,
2919 struct rte_flow_error *error)
2921 struct mlx5_priv *priv = dev->data->dev_private;
2922 struct rte_flow_attr attr = {
2923 .group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
2926 struct mlx5_rte_flow_item_tag tag_spec = {
2929 struct rte_flow_item items[] = {
2930 [1] = { .type = RTE_FLOW_ITEM_TYPE_END, },
2932 struct rte_flow_action_mark ftag = {
2935 struct mlx5_flow_action_copy_mreg cp_mreg = {
2939 struct rte_flow_action_jump jump = {
2940 .group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
2942 struct rte_flow_action actions[] = {
2943 [3] = { .type = RTE_FLOW_ACTION_TYPE_END, },
2945 struct mlx5_flow_mreg_copy_resource *mcp_res;
2948 /* Fill the register fileds in the flow. */
2949 ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
2953 ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
2957 /* Check if already registered. */
2958 assert(priv->mreg_cp_tbl);
2959 mcp_res = (void *)mlx5_hlist_lookup(priv->mreg_cp_tbl, mark_id);
2961 /* For non-default rule. */
2962 if (mark_id != MLX5_DEFAULT_COPY_ID)
2964 assert(mark_id != MLX5_DEFAULT_COPY_ID || mcp_res->refcnt == 1);
2967 /* Provide the full width of FLAG specific value. */
2968 if (mark_id == (priv->sh->dv_regc0_mask & MLX5_FLOW_MARK_DEFAULT))
2969 tag_spec.data = MLX5_FLOW_MARK_DEFAULT;
2970 /* Build a new flow. */
2971 if (mark_id != MLX5_DEFAULT_COPY_ID) {
2972 items[0] = (struct rte_flow_item){
2973 .type = MLX5_RTE_FLOW_ITEM_TYPE_TAG,
2976 items[1] = (struct rte_flow_item){
2977 .type = RTE_FLOW_ITEM_TYPE_END,
2979 actions[0] = (struct rte_flow_action){
2980 .type = MLX5_RTE_FLOW_ACTION_TYPE_MARK,
2983 actions[1] = (struct rte_flow_action){
2984 .type = MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
2987 actions[2] = (struct rte_flow_action){
2988 .type = RTE_FLOW_ACTION_TYPE_JUMP,
2991 actions[3] = (struct rte_flow_action){
2992 .type = RTE_FLOW_ACTION_TYPE_END,
2995 /* Default rule, wildcard match. */
2996 attr.priority = MLX5_FLOW_PRIO_RSVD;
2997 items[0] = (struct rte_flow_item){
2998 .type = RTE_FLOW_ITEM_TYPE_END,
3000 actions[0] = (struct rte_flow_action){
3001 .type = MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3004 actions[1] = (struct rte_flow_action){
3005 .type = RTE_FLOW_ACTION_TYPE_JUMP,
3008 actions[2] = (struct rte_flow_action){
3009 .type = RTE_FLOW_ACTION_TYPE_END,
3012 /* Build a new entry. */
3013 mcp_res = rte_zmalloc(__func__, sizeof(*mcp_res), 0);
3019 * The copy Flows are not included in any list. There
3020 * ones are referenced from other Flows and can not
3021 * be applied, removed, deleted in ardbitrary order
3022 * by list traversing.
3024 mcp_res->flow = flow_list_create(dev, NULL, &attr, items,
3025 actions, false, error);
3029 mcp_res->hlist_ent.key = mark_id;
3030 ret = mlx5_hlist_insert(priv->mreg_cp_tbl,
3031 &mcp_res->hlist_ent);
3038 flow_list_destroy(dev, NULL, mcp_res->flow);
3044 * Release flow in RX_CP_TBL.
3047 * Pointer to Ethernet device.
3049 * Parent flow for wich copying is provided.
3052 flow_mreg_del_copy_action(struct rte_eth_dev *dev,
3053 struct rte_flow *flow)
3055 struct mlx5_flow_mreg_copy_resource *mcp_res = flow->mreg_copy;
3056 struct mlx5_priv *priv = dev->data->dev_private;
3058 if (!mcp_res || !priv->mreg_cp_tbl)
3060 if (flow->copy_applied) {
3061 assert(mcp_res->appcnt);
3062 flow->copy_applied = 0;
3064 if (!mcp_res->appcnt)
3065 flow_drv_remove(dev, mcp_res->flow);
3068 * We do not check availability of metadata registers here,
3069 * because copy resources are not allocated in this case.
3071 if (--mcp_res->refcnt)
3073 assert(mcp_res->flow);
3074 flow_list_destroy(dev, NULL, mcp_res->flow);
3075 mlx5_hlist_remove(priv->mreg_cp_tbl, &mcp_res->hlist_ent);
3077 flow->mreg_copy = NULL;
3081 * Start flow in RX_CP_TBL.
3084 * Pointer to Ethernet device.
3086 * Parent flow for wich copying is provided.
3089 * 0 on success, a negative errno value otherwise and rte_errno is set.
3092 flow_mreg_start_copy_action(struct rte_eth_dev *dev,
3093 struct rte_flow *flow)
3095 struct mlx5_flow_mreg_copy_resource *mcp_res = flow->mreg_copy;
3098 if (!mcp_res || flow->copy_applied)
3100 if (!mcp_res->appcnt) {
3101 ret = flow_drv_apply(dev, mcp_res->flow, NULL);
3106 flow->copy_applied = 1;
3111 * Stop flow in RX_CP_TBL.
3114 * Pointer to Ethernet device.
3116 * Parent flow for wich copying is provided.
3119 flow_mreg_stop_copy_action(struct rte_eth_dev *dev,
3120 struct rte_flow *flow)
3122 struct mlx5_flow_mreg_copy_resource *mcp_res = flow->mreg_copy;
3124 if (!mcp_res || !flow->copy_applied)
3126 assert(mcp_res->appcnt);
3128 flow->copy_applied = 0;
3129 if (!mcp_res->appcnt)
3130 flow_drv_remove(dev, mcp_res->flow);
3134 * Remove the default copy action from RX_CP_TBL.
3137 * Pointer to Ethernet device.
3140 flow_mreg_del_default_copy_action(struct rte_eth_dev *dev)
3142 struct mlx5_flow_mreg_copy_resource *mcp_res;
3143 struct mlx5_priv *priv = dev->data->dev_private;
3145 /* Check if default flow is registered. */
3146 if (!priv->mreg_cp_tbl)
3148 mcp_res = (void *)mlx5_hlist_lookup(priv->mreg_cp_tbl,
3149 MLX5_DEFAULT_COPY_ID);
3152 assert(mcp_res->flow);
3153 flow_list_destroy(dev, NULL, mcp_res->flow);
3154 mlx5_hlist_remove(priv->mreg_cp_tbl, &mcp_res->hlist_ent);
3159 * Add the default copy action in in RX_CP_TBL.
3162 * Pointer to Ethernet device.
3164 * Perform verbose error reporting if not NULL.
3167 * 0 for success, negative value otherwise and rte_errno is set.
3170 flow_mreg_add_default_copy_action(struct rte_eth_dev *dev,
3171 struct rte_flow_error *error)
3173 struct mlx5_priv *priv = dev->data->dev_private;
3174 struct mlx5_flow_mreg_copy_resource *mcp_res;
3176 /* Check whether extensive metadata feature is engaged. */
3177 if (!priv->config.dv_flow_en ||
3178 priv->config.dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
3179 !mlx5_flow_ext_mreg_supported(dev) ||
3180 !priv->sh->dv_regc0_mask)
3182 mcp_res = flow_mreg_add_copy_action(dev, MLX5_DEFAULT_COPY_ID, error);
3189 * Add a flow of copying flow metadata registers in RX_CP_TBL.
3191 * All the flow having Q/RSS action should be split by
3192 * flow_mreg_split_qrss_prep() to pass by RX_CP_TBL. A flow in the RX_CP_TBL
3193 * performs the following,
3194 * - CQE->flow_tag := reg_c[1] (MARK)
3195 * - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
3196 * As CQE's flow_tag is not a register, it can't be simply copied from reg_c[1]
3197 * but there should be a flow per each MARK ID set by MARK action.
3199 * For the aforementioned reason, if there's a MARK action in flow's action
3200 * list, a corresponding flow should be added to the RX_CP_TBL in order to copy
3201 * the MARK ID to CQE's flow_tag like,
3202 * - If reg_c[1] is mark_id,
3203 * flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
3205 * For SET_META action which stores value in reg_c[0], as the destination is
3206 * also a flow metadata register (reg_b), adding a default flow is enough. Zero
3207 * MARK ID means the default flow. The default flow looks like,
3208 * - For all flow, reg_b := reg_c[0] and jump to RX_ACT_TBL.
3211 * Pointer to Ethernet device.
3213 * Pointer to flow structure.
3214 * @param[in] actions
3215 * Pointer to the list of actions.
3217 * Perform verbose error reporting if not NULL.
3220 * 0 on success, negative value otherwise and rte_errno is set.
3223 flow_mreg_update_copy_table(struct rte_eth_dev *dev,
3224 struct rte_flow *flow,
3225 const struct rte_flow_action *actions,
3226 struct rte_flow_error *error)
3228 struct mlx5_priv *priv = dev->data->dev_private;
3229 struct mlx5_dev_config *config = &priv->config;
3230 struct mlx5_flow_mreg_copy_resource *mcp_res;
3231 const struct rte_flow_action_mark *mark;
3233 /* Check whether extensive metadata feature is engaged. */
3234 if (!config->dv_flow_en ||
3235 config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
3236 !mlx5_flow_ext_mreg_supported(dev) ||
3237 !priv->sh->dv_regc0_mask)
3239 /* Find MARK action. */
3240 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3241 switch (actions->type) {
3242 case RTE_FLOW_ACTION_TYPE_FLAG:
3243 mcp_res = flow_mreg_add_copy_action
3244 (dev, MLX5_FLOW_MARK_DEFAULT, error);
3247 flow->mreg_copy = mcp_res;
3248 if (dev->data->dev_started) {
3250 flow->copy_applied = 1;
3253 case RTE_FLOW_ACTION_TYPE_MARK:
3254 mark = (const struct rte_flow_action_mark *)
3257 flow_mreg_add_copy_action(dev, mark->id, error);
3260 flow->mreg_copy = mcp_res;
3261 if (dev->data->dev_started) {
3263 flow->copy_applied = 1;
3273 #define MLX5_MAX_SPLIT_ACTIONS 24
3274 #define MLX5_MAX_SPLIT_ITEMS 24
3277 * Split the hairpin flow.
3278 * Since HW can't support encap on Rx we move the encap to Tx.
3279 * If the count action is after the encap then we also
3280 * move the count action. in this case the count will also measure
3284 * Pointer to Ethernet device.
3285 * @param[in] actions
3286 * Associated actions (list terminated by the END action).
3287 * @param[out] actions_rx
3289 * @param[out] actions_tx
3291 * @param[out] pattern_tx
3292 * The pattern items for the Tx flow.
3293 * @param[out] flow_id
3294 * The flow ID connected to this flow.
3300 flow_hairpin_split(struct rte_eth_dev *dev,
3301 const struct rte_flow_action actions[],
3302 struct rte_flow_action actions_rx[],
3303 struct rte_flow_action actions_tx[],
3304 struct rte_flow_item pattern_tx[],
3307 struct mlx5_priv *priv = dev->data->dev_private;
3308 const struct rte_flow_action_raw_encap *raw_encap;
3309 const struct rte_flow_action_raw_decap *raw_decap;
3310 struct mlx5_rte_flow_action_set_tag *set_tag;
3311 struct rte_flow_action *tag_action;
3312 struct mlx5_rte_flow_item_tag *tag_item;
3313 struct rte_flow_item *item;
3317 mlx5_flow_id_get(priv->sh->flow_id_pool, flow_id);
3318 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3319 switch (actions->type) {
3320 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
3321 case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
3322 rte_memcpy(actions_tx, actions,
3323 sizeof(struct rte_flow_action));
3326 case RTE_FLOW_ACTION_TYPE_COUNT:
3328 rte_memcpy(actions_tx, actions,
3329 sizeof(struct rte_flow_action));
3332 rte_memcpy(actions_rx, actions,
3333 sizeof(struct rte_flow_action));
3337 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3338 raw_encap = actions->conf;
3339 if (raw_encap->size >
3340 (sizeof(struct rte_flow_item_eth) +
3341 sizeof(struct rte_flow_item_ipv4))) {
3342 memcpy(actions_tx, actions,
3343 sizeof(struct rte_flow_action));
3347 rte_memcpy(actions_rx, actions,
3348 sizeof(struct rte_flow_action));
3352 case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
3353 raw_decap = actions->conf;
3354 if (raw_decap->size <
3355 (sizeof(struct rte_flow_item_eth) +
3356 sizeof(struct rte_flow_item_ipv4))) {
3357 memcpy(actions_tx, actions,
3358 sizeof(struct rte_flow_action));
3361 rte_memcpy(actions_rx, actions,
3362 sizeof(struct rte_flow_action));
3367 rte_memcpy(actions_rx, actions,
3368 sizeof(struct rte_flow_action));
3373 /* Add set meta action and end action for the Rx flow. */
3374 tag_action = actions_rx;
3375 tag_action->type = MLX5_RTE_FLOW_ACTION_TYPE_TAG;
3377 rte_memcpy(actions_rx, actions, sizeof(struct rte_flow_action));
3379 set_tag = (void *)actions_rx;
3380 set_tag->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_RX, 0, NULL);
3381 assert(set_tag->id > REG_NONE);
3382 set_tag->data = *flow_id;
3383 tag_action->conf = set_tag;
3384 /* Create Tx item list. */
3385 rte_memcpy(actions_tx, actions, sizeof(struct rte_flow_action));
3386 addr = (void *)&pattern_tx[2];
3388 item->type = MLX5_RTE_FLOW_ITEM_TYPE_TAG;
3389 tag_item = (void *)addr;
3390 tag_item->data = *flow_id;
3391 tag_item->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_TX, 0, NULL);
3392 assert(set_tag->id > REG_NONE);
3393 item->spec = tag_item;
3394 addr += sizeof(struct mlx5_rte_flow_item_tag);
3395 tag_item = (void *)addr;
3396 tag_item->data = UINT32_MAX;
3397 tag_item->id = UINT16_MAX;
3398 item->mask = tag_item;
3399 addr += sizeof(struct mlx5_rte_flow_item_tag);
3402 item->type = RTE_FLOW_ITEM_TYPE_END;
3407 * The last stage of splitting chain, just creates the subflow
3408 * without any modification.
3411 * Pointer to Ethernet device.
3413 * Parent flow structure pointer.
3414 * @param[in, out] sub_flow
3415 * Pointer to return the created subflow, may be NULL.
3417 * Flow rule attributes.
3419 * Pattern specification (list terminated by the END pattern item).
3420 * @param[in] actions
3421 * Associated actions (list terminated by the END action).
3422 * @param[in] external
3423 * This flow rule is created by request external to PMD.
3425 * Perform verbose error reporting if not NULL.
3427 * 0 on success, negative value otherwise
3430 flow_create_split_inner(struct rte_eth_dev *dev,
3431 struct rte_flow *flow,
3432 struct mlx5_flow **sub_flow,
3433 const struct rte_flow_attr *attr,
3434 const struct rte_flow_item items[],
3435 const struct rte_flow_action actions[],
3436 bool external, struct rte_flow_error *error)
3438 struct mlx5_flow *dev_flow;
3440 dev_flow = flow_drv_prepare(flow, attr, items, actions, error);
3443 dev_flow->flow = flow;
3444 dev_flow->external = external;
3445 /* Subflow object was created, we must include one in the list. */
3446 LIST_INSERT_HEAD(&flow->dev_flows, dev_flow, next);
3448 *sub_flow = dev_flow;
3449 return flow_drv_translate(dev, dev_flow, attr, items, actions, error);
3453 * Split the meter flow.
3455 * As meter flow will split to three sub flow, other than meter
3456 * action, the other actions make sense to only meter accepts
3457 * the packet. If it need to be dropped, no other additional
3458 * actions should be take.
3460 * One kind of special action which decapsulates the L3 tunnel
3461 * header will be in the prefix sub flow, as not to take the
3462 * L3 tunnel header into account.
3465 * Pointer to Ethernet device.
3466 * @param[in] actions
3467 * Associated actions (list terminated by the END action).
3468 * @param[out] actions_sfx
3469 * Suffix flow actions.
3470 * @param[out] actions_pre
3471 * Prefix flow actions.
3472 * @param[out] pattern_sfx
3473 * The pattern items for the suffix flow.
3474 * @param[out] tag_sfx
3475 * Pointer to suffix flow tag.
3481 flow_meter_split_prep(struct rte_eth_dev *dev,
3482 const struct rte_flow_action actions[],
3483 struct rte_flow_action actions_sfx[],
3484 struct rte_flow_action actions_pre[])
3486 struct rte_flow_action *tag_action;
3487 struct mlx5_rte_flow_action_set_tag *set_tag;
3488 struct rte_flow_error error;
3489 const struct rte_flow_action_raw_encap *raw_encap;
3490 const struct rte_flow_action_raw_decap *raw_decap;
3493 /* Add the extra tag action first. */
3494 tag_action = actions_pre;
3495 tag_action->type = MLX5_RTE_FLOW_ACTION_TYPE_TAG;
3497 /* Prepare the actions for prefix and suffix flow. */
3498 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3499 switch (actions->type) {
3500 case RTE_FLOW_ACTION_TYPE_METER:
3501 case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
3502 case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
3503 memcpy(actions_pre, actions,
3504 sizeof(struct rte_flow_action));
3507 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3508 raw_encap = actions->conf;
3509 if (raw_encap->size >
3510 (sizeof(struct rte_flow_item_eth) +
3511 sizeof(struct rte_flow_item_ipv4))) {
3512 memcpy(actions_sfx, actions,
3513 sizeof(struct rte_flow_action));
3516 rte_memcpy(actions_pre, actions,
3517 sizeof(struct rte_flow_action));
3521 case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
3522 raw_decap = actions->conf;
3523 /* Size 0 decap means 50 bytes as vxlan decap. */
3524 if (raw_decap->size && (raw_decap->size <
3525 (sizeof(struct rte_flow_item_eth) +
3526 sizeof(struct rte_flow_item_ipv4)))) {
3527 memcpy(actions_sfx, actions,
3528 sizeof(struct rte_flow_action));
3531 rte_memcpy(actions_pre, actions,
3532 sizeof(struct rte_flow_action));
3537 memcpy(actions_sfx, actions,
3538 sizeof(struct rte_flow_action));
3543 /* Add end action to the actions. */
3544 actions_sfx->type = RTE_FLOW_ACTION_TYPE_END;
3545 actions_pre->type = RTE_FLOW_ACTION_TYPE_END;
3548 set_tag = (void *)actions_pre;
3549 set_tag->id = mlx5_flow_get_reg_id(dev, MLX5_MTR_SFX, 0, &error);
3551 * Get the id from the qrss_pool to make qrss share the id with meter.
3553 tag_id = flow_qrss_get_id(dev);
3554 set_tag->data = rte_cpu_to_be_32(tag_id);
3555 tag_action->conf = set_tag;
3560 * Split action list having QUEUE/RSS for metadata register copy.
3562 * Once Q/RSS action is detected in user's action list, the flow action
3563 * should be split in order to copy metadata registers, which will happen in
3565 * - CQE->flow_tag := reg_c[1] (MARK)
3566 * - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
3567 * The Q/RSS action will be performed on RX_ACT_TBL after passing by RX_CP_TBL.
3568 * This is because the last action of each flow must be a terminal action
3569 * (QUEUE, RSS or DROP).
3571 * Flow ID must be allocated to identify actions in the RX_ACT_TBL and it is
3572 * stored and kept in the mlx5_flow structure per each sub_flow.
3574 * The Q/RSS action is replaced with,
3575 * - SET_TAG, setting the allocated flow ID to reg_c[2].
3576 * And the following JUMP action is added at the end,
3577 * - JUMP, to RX_CP_TBL.
3579 * A flow to perform remained Q/RSS action will be created in RX_ACT_TBL by
3580 * flow_create_split_metadata() routine. The flow will look like,
3581 * - If flow ID matches (reg_c[2]), perform Q/RSS.
3584 * Pointer to Ethernet device.
3585 * @param[out] split_actions
3586 * Pointer to store split actions to jump to CP_TBL.
3587 * @param[in] actions
3588 * Pointer to the list of original flow actions.
3590 * Pointer to the Q/RSS action.
3591 * @param[in] actions_n
3592 * Number of original actions.
3594 * Perform verbose error reporting if not NULL.
3597 * non-zero unique flow_id on success, otherwise 0 and
3598 * error/rte_error are set.
3601 flow_mreg_split_qrss_prep(struct rte_eth_dev *dev,
3602 struct rte_flow_action *split_actions,
3603 const struct rte_flow_action *actions,
3604 const struct rte_flow_action *qrss,
3605 int actions_n, struct rte_flow_error *error)
3607 struct mlx5_rte_flow_action_set_tag *set_tag;
3608 struct rte_flow_action_jump *jump;
3609 const int qrss_idx = qrss - actions;
3610 uint32_t flow_id = 0;
3614 * Given actions will be split
3615 * - Replace QUEUE/RSS action with SET_TAG to set flow ID.
3616 * - Add jump to mreg CP_TBL.
3617 * As a result, there will be one more action.
3620 memcpy(split_actions, actions, sizeof(*split_actions) * actions_n);
3621 set_tag = (void *)(split_actions + actions_n);
3623 * If tag action is not set to void(it means we are not the meter
3624 * suffix flow), add the tag action. Since meter suffix flow already
3625 * has the tag added.
3627 if (split_actions[qrss_idx].type != RTE_FLOW_ACTION_TYPE_VOID) {
3629 * Allocate the new subflow ID. This one is unique within
3630 * device and not shared with representors. Otherwise,
3631 * we would have to resolve multi-thread access synch
3632 * issue. Each flow on the shared device is appended
3633 * with source vport identifier, so the resulting
3634 * flows will be unique in the shared (by master and
3635 * representors) domain even if they have coinciding
3638 flow_id = flow_qrss_get_id(dev);
3640 return rte_flow_error_set(error, ENOMEM,
3641 RTE_FLOW_ERROR_TYPE_ACTION,
3642 NULL, "can't allocate id "
3643 "for split Q/RSS subflow");
3644 /* Internal SET_TAG action to set flow ID. */
3645 *set_tag = (struct mlx5_rte_flow_action_set_tag){
3648 ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0, error);
3652 /* Construct new actions array. */
3653 /* Replace QUEUE/RSS action. */
3654 split_actions[qrss_idx] = (struct rte_flow_action){
3655 .type = MLX5_RTE_FLOW_ACTION_TYPE_TAG,
3659 /* JUMP action to jump to mreg copy table (CP_TBL). */
3660 jump = (void *)(set_tag + 1);
3661 *jump = (struct rte_flow_action_jump){
3662 .group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
3664 split_actions[actions_n - 2] = (struct rte_flow_action){
3665 .type = RTE_FLOW_ACTION_TYPE_JUMP,
3668 split_actions[actions_n - 1] = (struct rte_flow_action){
3669 .type = RTE_FLOW_ACTION_TYPE_END,
3675 * Extend the given action list for Tx metadata copy.
3677 * Copy the given action list to the ext_actions and add flow metadata register
3678 * copy action in order to copy reg_a set by WQE to reg_c[0].
3680 * @param[out] ext_actions
3681 * Pointer to the extended action list.
3682 * @param[in] actions
3683 * Pointer to the list of actions.
3684 * @param[in] actions_n
3685 * Number of actions in the list.
3687 * Perform verbose error reporting if not NULL.
3690 * 0 on success, negative value otherwise
3693 flow_mreg_tx_copy_prep(struct rte_eth_dev *dev,
3694 struct rte_flow_action *ext_actions,
3695 const struct rte_flow_action *actions,
3696 int actions_n, struct rte_flow_error *error)
3698 struct mlx5_flow_action_copy_mreg *cp_mreg =
3699 (struct mlx5_flow_action_copy_mreg *)
3700 (ext_actions + actions_n + 1);
3703 ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
3707 ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_TX, 0, error);
3711 memcpy(ext_actions, actions,
3712 sizeof(*ext_actions) * actions_n);
3713 ext_actions[actions_n - 1] = (struct rte_flow_action){
3714 .type = MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3717 ext_actions[actions_n] = (struct rte_flow_action){
3718 .type = RTE_FLOW_ACTION_TYPE_END,
3724 * The splitting for metadata feature.
3726 * - Q/RSS action on NIC Rx should be split in order to pass by
3727 * the mreg copy table (RX_CP_TBL) and then it jumps to the
3728 * action table (RX_ACT_TBL) which has the split Q/RSS action.
3730 * - All the actions on NIC Tx should have a mreg copy action to
3731 * copy reg_a from WQE to reg_c[0].
3734 * Pointer to Ethernet device.
3736 * Parent flow structure pointer.
3738 * Flow rule attributes.
3740 * Pattern specification (list terminated by the END pattern item).
3741 * @param[in] actions
3742 * Associated actions (list terminated by the END action).
3743 * @param[in] external
3744 * This flow rule is created by request external to PMD.
3746 * Perform verbose error reporting if not NULL.
3748 * 0 on success, negative value otherwise
3751 flow_create_split_metadata(struct rte_eth_dev *dev,
3752 struct rte_flow *flow,
3753 const struct rte_flow_attr *attr,
3754 const struct rte_flow_item items[],
3755 const struct rte_flow_action actions[],
3756 bool external, struct rte_flow_error *error)
3758 struct mlx5_priv *priv = dev->data->dev_private;
3759 struct mlx5_dev_config *config = &priv->config;
3760 const struct rte_flow_action *qrss = NULL;
3761 struct rte_flow_action *ext_actions = NULL;
3762 struct mlx5_flow *dev_flow = NULL;
3763 uint32_t qrss_id = 0;
3769 /* Check whether extensive metadata feature is engaged. */
3770 if (!config->dv_flow_en ||
3771 config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
3772 !mlx5_flow_ext_mreg_supported(dev))
3773 return flow_create_split_inner(dev, flow, NULL, attr, items,
3774 actions, external, error);
3775 actions_n = flow_parse_qrss_action(actions, &qrss);
3777 /* Exclude hairpin flows from splitting. */
3778 if (qrss->type == RTE_FLOW_ACTION_TYPE_QUEUE) {
3779 const struct rte_flow_action_queue *queue;
3782 if (mlx5_rxq_get_type(dev, queue->index) ==
3783 MLX5_RXQ_TYPE_HAIRPIN)
3785 } else if (qrss->type == RTE_FLOW_ACTION_TYPE_RSS) {
3786 const struct rte_flow_action_rss *rss;
3789 if (mlx5_rxq_get_type(dev, rss->queue[0]) ==
3790 MLX5_RXQ_TYPE_HAIRPIN)
3795 /* Check if it is in meter suffix table. */
3796 mtr_sfx = attr->group == (attr->transfer ?
3797 (MLX5_FLOW_TABLE_LEVEL_SUFFIX - 1) :
3798 MLX5_FLOW_TABLE_LEVEL_SUFFIX);
3800 * Q/RSS action on NIC Rx should be split in order to pass by
3801 * the mreg copy table (RX_CP_TBL) and then it jumps to the
3802 * action table (RX_ACT_TBL) which has the split Q/RSS action.
3804 act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
3805 sizeof(struct rte_flow_action_set_tag) +
3806 sizeof(struct rte_flow_action_jump);
3807 ext_actions = rte_zmalloc(__func__, act_size, 0);
3809 return rte_flow_error_set(error, ENOMEM,
3810 RTE_FLOW_ERROR_TYPE_ACTION,
3811 NULL, "no memory to split "
3814 * If we are the suffix flow of meter, tag already exist.
3815 * Set the tag action to void.
3818 ext_actions[qrss - actions].type =
3819 RTE_FLOW_ACTION_TYPE_VOID;
3821 ext_actions[qrss - actions].type =
3822 MLX5_RTE_FLOW_ACTION_TYPE_TAG;
3824 * Create the new actions list with removed Q/RSS action
3825 * and appended set tag and jump to register copy table
3826 * (RX_CP_TBL). We should preallocate unique tag ID here
3827 * in advance, because it is needed for set tag action.
3829 qrss_id = flow_mreg_split_qrss_prep(dev, ext_actions, actions,
3830 qrss, actions_n, error);
3831 if (!mtr_sfx && !qrss_id) {
3835 } else if (attr->egress && !attr->transfer) {
3837 * All the actions on NIC Tx should have a metadata register
3838 * copy action to copy reg_a from WQE to reg_c[meta]
3840 act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
3841 sizeof(struct mlx5_flow_action_copy_mreg);
3842 ext_actions = rte_zmalloc(__func__, act_size, 0);
3844 return rte_flow_error_set(error, ENOMEM,
3845 RTE_FLOW_ERROR_TYPE_ACTION,
3846 NULL, "no memory to split "
3848 /* Create the action list appended with copy register. */
3849 ret = flow_mreg_tx_copy_prep(dev, ext_actions, actions,
3854 /* Add the unmodified original or prefix subflow. */
3855 ret = flow_create_split_inner(dev, flow, &dev_flow, attr, items,
3856 ext_actions ? ext_actions : actions,
3862 const struct rte_flow_attr q_attr = {
3863 .group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
3866 /* Internal PMD action to set register. */
3867 struct mlx5_rte_flow_item_tag q_tag_spec = {
3871 struct rte_flow_item q_items[] = {
3873 .type = MLX5_RTE_FLOW_ITEM_TYPE_TAG,
3874 .spec = &q_tag_spec,
3879 .type = RTE_FLOW_ITEM_TYPE_END,
3882 struct rte_flow_action q_actions[] = {
3888 .type = RTE_FLOW_ACTION_TYPE_END,
3891 uint64_t hash_fields = dev_flow->hash_fields;
3894 * Configure the tag item only if there is no meter subflow.
3895 * Since tag is already marked in the meter suffix subflow
3896 * we can just use the meter suffix items as is.
3899 /* Not meter subflow. */
3902 * Put unique id in prefix flow due to it is destroyed
3903 * after suffix flow and id will be freed after there
3904 * is no actual flows with this id and identifier
3905 * reallocation becomes possible (for example, for
3906 * other flows in other threads).
3908 dev_flow->qrss_id = qrss_id;
3910 ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0,
3914 q_tag_spec.id = ret;
3917 /* Add suffix subflow to execute Q/RSS. */
3918 ret = flow_create_split_inner(dev, flow, &dev_flow,
3919 &q_attr, mtr_sfx ? items :
3925 dev_flow->hash_fields = hash_fields;
3930 * We do not destroy the partially created sub_flows in case of error.
3931 * These ones are included into parent flow list and will be destroyed
3932 * by flow_drv_destroy.
3934 flow_qrss_free_id(dev, qrss_id);
3935 rte_free(ext_actions);
3940 * The splitting for meter feature.
3942 * - The meter flow will be split to two flows as prefix and
3943 * suffix flow. The packets make sense only it pass the prefix
3946 * - Reg_C_5 is used for the packet to match betweend prefix and
3950 * Pointer to Ethernet device.
3952 * Parent flow structure pointer.
3954 * Flow rule attributes.
3956 * Pattern specification (list terminated by the END pattern item).
3957 * @param[in] actions
3958 * Associated actions (list terminated by the END action).
3959 * @param[in] external
3960 * This flow rule is created by request external to PMD.
3962 * Perform verbose error reporting if not NULL.
3964 * 0 on success, negative value otherwise
3967 flow_create_split_meter(struct rte_eth_dev *dev,
3968 struct rte_flow *flow,
3969 const struct rte_flow_attr *attr,
3970 const struct rte_flow_item items[],
3971 const struct rte_flow_action actions[],
3972 bool external, struct rte_flow_error *error)
3974 struct mlx5_priv *priv = dev->data->dev_private;
3975 struct rte_flow_action *sfx_actions = NULL;
3976 struct rte_flow_action *pre_actions = NULL;
3977 struct rte_flow_item *sfx_items = NULL;
3978 const struct rte_flow_item *sfx_port_id_item;
3979 struct mlx5_flow *dev_flow = NULL;
3980 struct rte_flow_attr sfx_attr = *attr;
3982 uint32_t mtr_tag_id = 0;
3989 actions_n = flow_check_meter_action(actions, &mtr);
3991 struct mlx5_rte_flow_item_tag *tag_spec;
3992 /* The five prefix actions: meter, decap, encap, tag, end. */
3993 act_size = sizeof(struct rte_flow_action) * (actions_n + 5) +
3994 sizeof(struct rte_flow_action_set_tag);
3996 #define METER_SUFFIX_ITEM 3
3997 item_size = sizeof(struct rte_flow_item) * METER_SUFFIX_ITEM +
3998 sizeof(struct mlx5_rte_flow_item_tag);
3999 sfx_actions = rte_zmalloc(__func__, (act_size + item_size), 0);
4001 return rte_flow_error_set(error, ENOMEM,
4002 RTE_FLOW_ERROR_TYPE_ACTION,
4003 NULL, "no memory to split "
4005 pre_actions = sfx_actions + actions_n;
4006 mtr_tag_id = flow_meter_split_prep(dev, actions, sfx_actions,
4012 /* Add the prefix subflow. */
4013 ret = flow_create_split_inner(dev, flow, &dev_flow, attr, items,
4014 pre_actions, external, error);
4019 dev_flow->mtr_flow_id = mtr_tag_id;
4020 /* Prepare the suffix flow match pattern. */
4021 sfx_items = (struct rte_flow_item *)((char *)sfx_actions +
4023 tag_spec = (struct mlx5_rte_flow_item_tag *)(sfx_items +
4025 tag_spec->data = rte_cpu_to_be_32(dev_flow->mtr_flow_id);
4026 tag_spec->id = mlx5_flow_get_reg_id(dev, MLX5_MTR_SFX, 0,
4028 sfx_items->type = MLX5_RTE_FLOW_ITEM_TYPE_TAG;
4029 sfx_items->spec = tag_spec;
4030 sfx_items->last = NULL;
4031 sfx_items->mask = NULL;
4033 sfx_port_id_item = find_port_id_item(items);
4034 if (sfx_port_id_item) {
4035 memcpy(sfx_items, sfx_port_id_item,
4036 sizeof(*sfx_items));
4039 sfx_items->type = RTE_FLOW_ITEM_TYPE_END;
4040 sfx_items -= METER_SUFFIX_ITEM;
4041 /* Setting the sfx group atrr. */
4042 sfx_attr.group = sfx_attr.transfer ?
4043 (MLX5_FLOW_TABLE_LEVEL_SUFFIX - 1) :
4044 MLX5_FLOW_TABLE_LEVEL_SUFFIX;
4046 /* Add the prefix subflow. */
4047 ret = flow_create_split_metadata(dev, flow, &sfx_attr,
4048 sfx_items ? sfx_items : items,
4049 sfx_actions ? sfx_actions : actions,
4053 rte_free(sfx_actions);
4058 * Split the flow to subflow set. The splitters might be linked
4059 * in the chain, like this:
4060 * flow_create_split_outer() calls:
4061 * flow_create_split_meter() calls:
4062 * flow_create_split_metadata(meter_subflow_0) calls:
4063 * flow_create_split_inner(metadata_subflow_0)
4064 * flow_create_split_inner(metadata_subflow_1)
4065 * flow_create_split_inner(metadata_subflow_2)
4066 * flow_create_split_metadata(meter_subflow_1) calls:
4067 * flow_create_split_inner(metadata_subflow_0)
4068 * flow_create_split_inner(metadata_subflow_1)
4069 * flow_create_split_inner(metadata_subflow_2)
4071 * This provide flexible way to add new levels of flow splitting.
4072 * The all of successfully created subflows are included to the
4073 * parent flow dev_flow list.
4076 * Pointer to Ethernet device.
4078 * Parent flow structure pointer.
4080 * Flow rule attributes.
4082 * Pattern specification (list terminated by the END pattern item).
4083 * @param[in] actions
4084 * Associated actions (list terminated by the END action).
4085 * @param[in] external
4086 * This flow rule is created by request external to PMD.
4088 * Perform verbose error reporting if not NULL.
4090 * 0 on success, negative value otherwise
4093 flow_create_split_outer(struct rte_eth_dev *dev,
4094 struct rte_flow *flow,
4095 const struct rte_flow_attr *attr,
4096 const struct rte_flow_item items[],
4097 const struct rte_flow_action actions[],
4098 bool external, struct rte_flow_error *error)
4102 ret = flow_create_split_meter(dev, flow, attr, items,
4103 actions, external, error);
4109 * Create a flow and add it to @p list.
4112 * Pointer to Ethernet device.
4114 * Pointer to a TAILQ flow list. If this parameter NULL,
4115 * no list insertion occurred, flow is just created,
4116 * this is caller's responsibility to track the
4119 * Flow rule attributes.
4121 * Pattern specification (list terminated by the END pattern item).
4122 * @param[in] actions
4123 * Associated actions (list terminated by the END action).
4124 * @param[in] external
4125 * This flow rule is created by request external to PMD.
4127 * Perform verbose error reporting if not NULL.
4130 * A flow on success, NULL otherwise and rte_errno is set.
4132 static struct rte_flow *
4133 flow_list_create(struct rte_eth_dev *dev, struct mlx5_flows *list,
4134 const struct rte_flow_attr *attr,
4135 const struct rte_flow_item items[],
4136 const struct rte_flow_action actions[],
4137 bool external, struct rte_flow_error *error)
4139 struct mlx5_priv *priv = dev->data->dev_private;
4140 struct rte_flow *flow = NULL;
4141 struct mlx5_flow *dev_flow;
4142 const struct rte_flow_action_rss *rss;
4144 struct rte_flow_expand_rss buf;
4145 uint8_t buffer[2048];
4148 struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
4149 uint8_t buffer[2048];
4152 struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
4153 uint8_t buffer[2048];
4154 } actions_hairpin_tx;
4156 struct rte_flow_item items[MLX5_MAX_SPLIT_ITEMS];
4157 uint8_t buffer[2048];
4159 struct rte_flow_expand_rss *buf = &expand_buffer.buf;
4160 const struct rte_flow_action *p_actions_rx = actions;
4164 int hairpin_flow = 0;
4165 uint32_t hairpin_id = 0;
4166 struct rte_flow_attr attr_tx = { .priority = 0 };
4168 hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
4169 if (hairpin_flow > 0) {
4170 if (hairpin_flow > MLX5_MAX_SPLIT_ACTIONS) {
4174 flow_hairpin_split(dev, actions, actions_rx.actions,
4175 actions_hairpin_tx.actions, items_tx.items,
4177 p_actions_rx = actions_rx.actions;
4179 ret = flow_drv_validate(dev, attr, items, p_actions_rx, external,
4182 goto error_before_flow;
4183 flow_size = sizeof(struct rte_flow);
4184 rss = flow_get_rss_action(p_actions_rx);
4186 flow_size += RTE_ALIGN_CEIL(rss->queue_num * sizeof(uint16_t),
4189 flow_size += RTE_ALIGN_CEIL(sizeof(uint16_t), sizeof(void *));
4190 flow = rte_calloc(__func__, 1, flow_size, 0);
4193 goto error_before_flow;
4195 flow->drv_type = flow_get_drv_type(dev, attr);
4196 if (hairpin_id != 0)
4197 flow->hairpin_flow_id = hairpin_id;
4198 assert(flow->drv_type > MLX5_FLOW_TYPE_MIN &&
4199 flow->drv_type < MLX5_FLOW_TYPE_MAX);
4200 flow->rss.queue = (void *)(flow + 1);
4203 * The following information is required by
4204 * mlx5_flow_hashfields_adjust() in advance.
4206 flow->rss.level = rss->level;
4207 /* RSS type 0 indicates default RSS type (ETH_RSS_IP). */
4208 flow->rss.types = !rss->types ? ETH_RSS_IP : rss->types;
4210 LIST_INIT(&flow->dev_flows);
4211 if (rss && rss->types) {
4212 unsigned int graph_root;
4214 graph_root = find_graph_root(items, rss->level);
4215 ret = rte_flow_expand_rss(buf, sizeof(expand_buffer.buffer),
4217 mlx5_support_expansion,
4220 (unsigned int)ret < sizeof(expand_buffer.buffer));
4223 buf->entry[0].pattern = (void *)(uintptr_t)items;
4225 for (i = 0; i < buf->entries; ++i) {
4227 * The splitter may create multiple dev_flows,
4228 * depending on configuration. In the simplest
4229 * case it just creates unmodified original flow.
4231 ret = flow_create_split_outer(dev, flow, attr,
4232 buf->entry[i].pattern,
4233 p_actions_rx, external,
4238 /* Create the tx flow. */
4240 attr_tx.group = MLX5_HAIRPIN_TX_TABLE;
4241 attr_tx.ingress = 0;
4243 dev_flow = flow_drv_prepare(flow, &attr_tx, items_tx.items,
4244 actions_hairpin_tx.actions, error);
4247 dev_flow->flow = flow;
4248 dev_flow->external = 0;
4249 LIST_INSERT_HEAD(&flow->dev_flows, dev_flow, next);
4250 ret = flow_drv_translate(dev, dev_flow, &attr_tx,
4252 actions_hairpin_tx.actions, error);
4257 * Update the metadata register copy table. If extensive
4258 * metadata feature is enabled and registers are supported
4259 * we might create the extra rte_flow for each unique
4260 * MARK/FLAG action ID.
4262 * The table is updated for ingress Flows only, because
4263 * the egress Flows belong to the different device and
4264 * copy table should be updated in peer NIC Rx domain.
4266 if (attr->ingress &&
4267 (external || attr->group != MLX5_FLOW_MREG_CP_TABLE_GROUP)) {
4268 ret = flow_mreg_update_copy_table(dev, flow, actions, error);
4272 if (dev->data->dev_started) {
4273 ret = flow_drv_apply(dev, flow, error);
4278 TAILQ_INSERT_TAIL(list, flow, next);
4279 flow_rxq_flags_set(dev, flow);
4283 mlx5_flow_id_release(priv->sh->flow_id_pool,
4288 flow_mreg_del_copy_action(dev, flow);
4289 ret = rte_errno; /* Save rte_errno before cleanup. */
4290 if (flow->hairpin_flow_id)
4291 mlx5_flow_id_release(priv->sh->flow_id_pool,
4292 flow->hairpin_flow_id);
4294 flow_drv_destroy(dev, flow);
4296 rte_errno = ret; /* Restore rte_errno. */
4301 * Create a dedicated flow rule on e-switch table 0 (root table), to direct all
4302 * incoming packets to table 1.
4304 * Other flow rules, requested for group n, will be created in
4305 * e-switch table n+1.
4306 * Jump action to e-switch group n will be created to group n+1.
4308 * Used when working in switchdev mode, to utilise advantages of table 1
4312 * Pointer to Ethernet device.
4315 * Pointer to flow on success, NULL otherwise and rte_errno is set.
4318 mlx5_flow_create_esw_table_zero_flow(struct rte_eth_dev *dev)
4320 const struct rte_flow_attr attr = {
4327 const struct rte_flow_item pattern = {
4328 .type = RTE_FLOW_ITEM_TYPE_END,
4330 struct rte_flow_action_jump jump = {
4333 const struct rte_flow_action actions[] = {
4335 .type = RTE_FLOW_ACTION_TYPE_JUMP,
4339 .type = RTE_FLOW_ACTION_TYPE_END,
4342 struct mlx5_priv *priv = dev->data->dev_private;
4343 struct rte_flow_error error;
4345 return flow_list_create(dev, &priv->ctrl_flows, &attr, &pattern,
4346 actions, false, &error);
4352 * @see rte_flow_create()
4356 mlx5_flow_create(struct rte_eth_dev *dev,
4357 const struct rte_flow_attr *attr,
4358 const struct rte_flow_item items[],
4359 const struct rte_flow_action actions[],
4360 struct rte_flow_error *error)
4362 struct mlx5_priv *priv = dev->data->dev_private;
4364 return flow_list_create(dev, &priv->flows,
4365 attr, items, actions, true, error);
4369 * Destroy a flow in a list.
4372 * Pointer to Ethernet device.
4374 * Pointer to a TAILQ flow list. If this parameter NULL,
4375 * there is no flow removal from the list.
4380 flow_list_destroy(struct rte_eth_dev *dev, struct mlx5_flows *list,
4381 struct rte_flow *flow)
4383 struct mlx5_priv *priv = dev->data->dev_private;
4386 * Update RX queue flags only if port is started, otherwise it is
4389 if (dev->data->dev_started)
4390 flow_rxq_flags_trim(dev, flow);
4391 if (flow->hairpin_flow_id)
4392 mlx5_flow_id_release(priv->sh->flow_id_pool,
4393 flow->hairpin_flow_id);
4394 flow_drv_destroy(dev, flow);
4396 TAILQ_REMOVE(list, flow, next);
4397 flow_mreg_del_copy_action(dev, flow);
4398 rte_free(flow->fdir);
4403 * Destroy all flows.
4406 * Pointer to Ethernet device.
4408 * Pointer to a TAILQ flow list.
4411 mlx5_flow_list_flush(struct rte_eth_dev *dev, struct mlx5_flows *list)
4413 while (!TAILQ_EMPTY(list)) {
4414 struct rte_flow *flow;
4416 flow = TAILQ_FIRST(list);
4417 flow_list_destroy(dev, list, flow);
4425 * Pointer to Ethernet device.
4427 * Pointer to a TAILQ flow list.
4430 mlx5_flow_stop(struct rte_eth_dev *dev, struct mlx5_flows *list)
4432 struct rte_flow *flow;
4434 TAILQ_FOREACH_REVERSE(flow, list, mlx5_flows, next) {
4435 flow_drv_remove(dev, flow);
4436 flow_mreg_stop_copy_action(dev, flow);
4438 flow_mreg_del_default_copy_action(dev);
4439 flow_rxq_flags_clear(dev);
4446 * Pointer to Ethernet device.
4448 * Pointer to a TAILQ flow list.
4451 * 0 on success, a negative errno value otherwise and rte_errno is set.
4454 mlx5_flow_start(struct rte_eth_dev *dev, struct mlx5_flows *list)
4456 struct rte_flow *flow;
4457 struct rte_flow_error error;
4460 /* Make sure default copy action (reg_c[0] -> reg_b) is created. */
4461 ret = flow_mreg_add_default_copy_action(dev, &error);
4464 /* Apply Flows created by application. */
4465 TAILQ_FOREACH(flow, list, next) {
4466 ret = flow_mreg_start_copy_action(dev, flow);
4469 ret = flow_drv_apply(dev, flow, &error);
4472 flow_rxq_flags_set(dev, flow);
4476 ret = rte_errno; /* Save rte_errno before cleanup. */
4477 mlx5_flow_stop(dev, list);
4478 rte_errno = ret; /* Restore rte_errno. */
4483 * Verify the flow list is empty
4486 * Pointer to Ethernet device.
4488 * @return the number of flows not released.
4491 mlx5_flow_verify(struct rte_eth_dev *dev)
4493 struct mlx5_priv *priv = dev->data->dev_private;
4494 struct rte_flow *flow;
4497 TAILQ_FOREACH(flow, &priv->flows, next) {
4498 DRV_LOG(DEBUG, "port %u flow %p still referenced",
4499 dev->data->port_id, (void *)flow);
4506 * Enable default hairpin egress flow.
4509 * Pointer to Ethernet device.
4514 * 0 on success, a negative errno value otherwise and rte_errno is set.
4517 mlx5_ctrl_flow_source_queue(struct rte_eth_dev *dev,
4520 struct mlx5_priv *priv = dev->data->dev_private;
4521 const struct rte_flow_attr attr = {
4525 struct mlx5_rte_flow_item_tx_queue queue_spec = {
4528 struct mlx5_rte_flow_item_tx_queue queue_mask = {
4529 .queue = UINT32_MAX,
4531 struct rte_flow_item items[] = {
4533 .type = MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE,
4534 .spec = &queue_spec,
4536 .mask = &queue_mask,
4539 .type = RTE_FLOW_ITEM_TYPE_END,
4542 struct rte_flow_action_jump jump = {
4543 .group = MLX5_HAIRPIN_TX_TABLE,
4545 struct rte_flow_action actions[2];
4546 struct rte_flow *flow;
4547 struct rte_flow_error error;
4549 actions[0].type = RTE_FLOW_ACTION_TYPE_JUMP;
4550 actions[0].conf = &jump;
4551 actions[1].type = RTE_FLOW_ACTION_TYPE_END;
4552 flow = flow_list_create(dev, &priv->ctrl_flows,
4553 &attr, items, actions, false, &error);
4556 "Failed to create ctrl flow: rte_errno(%d),"
4557 " type(%d), message(%s)",
4558 rte_errno, error.type,
4559 error.message ? error.message : " (no stated reason)");
4566 * Enable a control flow configured from the control plane.
4569 * Pointer to Ethernet device.
4571 * An Ethernet flow spec to apply.
4573 * An Ethernet flow mask to apply.
4575 * A VLAN flow spec to apply.
4577 * A VLAN flow mask to apply.
4580 * 0 on success, a negative errno value otherwise and rte_errno is set.
4583 mlx5_ctrl_flow_vlan(struct rte_eth_dev *dev,
4584 struct rte_flow_item_eth *eth_spec,
4585 struct rte_flow_item_eth *eth_mask,
4586 struct rte_flow_item_vlan *vlan_spec,
4587 struct rte_flow_item_vlan *vlan_mask)
4589 struct mlx5_priv *priv = dev->data->dev_private;
4590 const struct rte_flow_attr attr = {
4592 .priority = MLX5_FLOW_PRIO_RSVD,
4594 struct rte_flow_item items[] = {
4596 .type = RTE_FLOW_ITEM_TYPE_ETH,
4602 .type = (vlan_spec) ? RTE_FLOW_ITEM_TYPE_VLAN :
4603 RTE_FLOW_ITEM_TYPE_END,
4609 .type = RTE_FLOW_ITEM_TYPE_END,
4612 uint16_t queue[priv->reta_idx_n];
4613 struct rte_flow_action_rss action_rss = {
4614 .func = RTE_ETH_HASH_FUNCTION_DEFAULT,
4616 .types = priv->rss_conf.rss_hf,
4617 .key_len = priv->rss_conf.rss_key_len,
4618 .queue_num = priv->reta_idx_n,
4619 .key = priv->rss_conf.rss_key,
4622 struct rte_flow_action actions[] = {
4624 .type = RTE_FLOW_ACTION_TYPE_RSS,
4625 .conf = &action_rss,
4628 .type = RTE_FLOW_ACTION_TYPE_END,
4631 struct rte_flow *flow;
4632 struct rte_flow_error error;
4635 if (!priv->reta_idx_n || !priv->rxqs_n) {
4638 for (i = 0; i != priv->reta_idx_n; ++i)
4639 queue[i] = (*priv->reta_idx)[i];
4640 flow = flow_list_create(dev, &priv->ctrl_flows,
4641 &attr, items, actions, false, &error);
4648 * Enable a flow control configured from the control plane.
4651 * Pointer to Ethernet device.
4653 * An Ethernet flow spec to apply.
4655 * An Ethernet flow mask to apply.
4658 * 0 on success, a negative errno value otherwise and rte_errno is set.
4661 mlx5_ctrl_flow(struct rte_eth_dev *dev,
4662 struct rte_flow_item_eth *eth_spec,
4663 struct rte_flow_item_eth *eth_mask)
4665 return mlx5_ctrl_flow_vlan(dev, eth_spec, eth_mask, NULL, NULL);
4671 * @see rte_flow_destroy()
4675 mlx5_flow_destroy(struct rte_eth_dev *dev,
4676 struct rte_flow *flow,
4677 struct rte_flow_error *error __rte_unused)
4679 struct mlx5_priv *priv = dev->data->dev_private;
4681 flow_list_destroy(dev, &priv->flows, flow);
4686 * Destroy all flows.
4688 * @see rte_flow_flush()
4692 mlx5_flow_flush(struct rte_eth_dev *dev,
4693 struct rte_flow_error *error __rte_unused)
4695 struct mlx5_priv *priv = dev->data->dev_private;
4697 mlx5_flow_list_flush(dev, &priv->flows);
4704 * @see rte_flow_isolate()
4708 mlx5_flow_isolate(struct rte_eth_dev *dev,
4710 struct rte_flow_error *error)
4712 struct mlx5_priv *priv = dev->data->dev_private;
4714 if (dev->data->dev_started) {
4715 rte_flow_error_set(error, EBUSY,
4716 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
4718 "port must be stopped first");
4721 priv->isolated = !!enable;
4723 dev->dev_ops = &mlx5_dev_ops_isolate;
4725 dev->dev_ops = &mlx5_dev_ops;
4732 * @see rte_flow_query()
4736 flow_drv_query(struct rte_eth_dev *dev,
4737 struct rte_flow *flow,
4738 const struct rte_flow_action *actions,
4740 struct rte_flow_error *error)
4742 const struct mlx5_flow_driver_ops *fops;
4743 enum mlx5_flow_drv_type ftype = flow->drv_type;
4745 assert(ftype > MLX5_FLOW_TYPE_MIN && ftype < MLX5_FLOW_TYPE_MAX);
4746 fops = flow_get_drv_ops(ftype);
4748 return fops->query(dev, flow, actions, data, error);
4754 * @see rte_flow_query()
4758 mlx5_flow_query(struct rte_eth_dev *dev,
4759 struct rte_flow *flow,
4760 const struct rte_flow_action *actions,
4762 struct rte_flow_error *error)
4766 ret = flow_drv_query(dev, flow, actions, data, error);
4773 * Convert a flow director filter to a generic flow.
4776 * Pointer to Ethernet device.
4777 * @param fdir_filter
4778 * Flow director filter to add.
4780 * Generic flow parameters structure.
4783 * 0 on success, a negative errno value otherwise and rte_errno is set.
4786 flow_fdir_filter_convert(struct rte_eth_dev *dev,
4787 const struct rte_eth_fdir_filter *fdir_filter,
4788 struct mlx5_fdir *attributes)
4790 struct mlx5_priv *priv = dev->data->dev_private;
4791 const struct rte_eth_fdir_input *input = &fdir_filter->input;
4792 const struct rte_eth_fdir_masks *mask =
4793 &dev->data->dev_conf.fdir_conf.mask;
4795 /* Validate queue number. */
4796 if (fdir_filter->action.rx_queue >= priv->rxqs_n) {
4797 DRV_LOG(ERR, "port %u invalid queue number %d",
4798 dev->data->port_id, fdir_filter->action.rx_queue);
4802 attributes->attr.ingress = 1;
4803 attributes->items[0] = (struct rte_flow_item) {
4804 .type = RTE_FLOW_ITEM_TYPE_ETH,
4805 .spec = &attributes->l2,
4806 .mask = &attributes->l2_mask,
4808 switch (fdir_filter->action.behavior) {
4809 case RTE_ETH_FDIR_ACCEPT:
4810 attributes->actions[0] = (struct rte_flow_action){
4811 .type = RTE_FLOW_ACTION_TYPE_QUEUE,
4812 .conf = &attributes->queue,
4815 case RTE_ETH_FDIR_REJECT:
4816 attributes->actions[0] = (struct rte_flow_action){
4817 .type = RTE_FLOW_ACTION_TYPE_DROP,
4821 DRV_LOG(ERR, "port %u invalid behavior %d",
4823 fdir_filter->action.behavior);
4824 rte_errno = ENOTSUP;
4827 attributes->queue.index = fdir_filter->action.rx_queue;
4829 switch (fdir_filter->input.flow_type) {
4830 case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
4831 case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
4832 case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
4833 attributes->l3.ipv4.hdr = (struct rte_ipv4_hdr){
4834 .src_addr = input->flow.ip4_flow.src_ip,
4835 .dst_addr = input->flow.ip4_flow.dst_ip,
4836 .time_to_live = input->flow.ip4_flow.ttl,
4837 .type_of_service = input->flow.ip4_flow.tos,
4839 attributes->l3_mask.ipv4.hdr = (struct rte_ipv4_hdr){
4840 .src_addr = mask->ipv4_mask.src_ip,
4841 .dst_addr = mask->ipv4_mask.dst_ip,
4842 .time_to_live = mask->ipv4_mask.ttl,
4843 .type_of_service = mask->ipv4_mask.tos,
4844 .next_proto_id = mask->ipv4_mask.proto,
4846 attributes->items[1] = (struct rte_flow_item){
4847 .type = RTE_FLOW_ITEM_TYPE_IPV4,
4848 .spec = &attributes->l3,
4849 .mask = &attributes->l3_mask,
4852 case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
4853 case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
4854 case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
4855 attributes->l3.ipv6.hdr = (struct rte_ipv6_hdr){
4856 .hop_limits = input->flow.ipv6_flow.hop_limits,
4857 .proto = input->flow.ipv6_flow.proto,
4860 memcpy(attributes->l3.ipv6.hdr.src_addr,
4861 input->flow.ipv6_flow.src_ip,
4862 RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
4863 memcpy(attributes->l3.ipv6.hdr.dst_addr,
4864 input->flow.ipv6_flow.dst_ip,
4865 RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
4866 memcpy(attributes->l3_mask.ipv6.hdr.src_addr,
4867 mask->ipv6_mask.src_ip,
4868 RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
4869 memcpy(attributes->l3_mask.ipv6.hdr.dst_addr,
4870 mask->ipv6_mask.dst_ip,
4871 RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
4872 attributes->items[1] = (struct rte_flow_item){
4873 .type = RTE_FLOW_ITEM_TYPE_IPV6,
4874 .spec = &attributes->l3,
4875 .mask = &attributes->l3_mask,
4879 DRV_LOG(ERR, "port %u invalid flow type%d",
4880 dev->data->port_id, fdir_filter->input.flow_type);
4881 rte_errno = ENOTSUP;
4885 switch (fdir_filter->input.flow_type) {
4886 case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
4887 attributes->l4.udp.hdr = (struct rte_udp_hdr){
4888 .src_port = input->flow.udp4_flow.src_port,
4889 .dst_port = input->flow.udp4_flow.dst_port,
4891 attributes->l4_mask.udp.hdr = (struct rte_udp_hdr){
4892 .src_port = mask->src_port_mask,
4893 .dst_port = mask->dst_port_mask,
4895 attributes->items[2] = (struct rte_flow_item){
4896 .type = RTE_FLOW_ITEM_TYPE_UDP,
4897 .spec = &attributes->l4,
4898 .mask = &attributes->l4_mask,
4901 case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
4902 attributes->l4.tcp.hdr = (struct rte_tcp_hdr){
4903 .src_port = input->flow.tcp4_flow.src_port,
4904 .dst_port = input->flow.tcp4_flow.dst_port,
4906 attributes->l4_mask.tcp.hdr = (struct rte_tcp_hdr){
4907 .src_port = mask->src_port_mask,
4908 .dst_port = mask->dst_port_mask,
4910 attributes->items[2] = (struct rte_flow_item){
4911 .type = RTE_FLOW_ITEM_TYPE_TCP,
4912 .spec = &attributes->l4,
4913 .mask = &attributes->l4_mask,
4916 case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
4917 attributes->l4.udp.hdr = (struct rte_udp_hdr){
4918 .src_port = input->flow.udp6_flow.src_port,
4919 .dst_port = input->flow.udp6_flow.dst_port,
4921 attributes->l4_mask.udp.hdr = (struct rte_udp_hdr){
4922 .src_port = mask->src_port_mask,
4923 .dst_port = mask->dst_port_mask,
4925 attributes->items[2] = (struct rte_flow_item){
4926 .type = RTE_FLOW_ITEM_TYPE_UDP,
4927 .spec = &attributes->l4,
4928 .mask = &attributes->l4_mask,
4931 case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
4932 attributes->l4.tcp.hdr = (struct rte_tcp_hdr){
4933 .src_port = input->flow.tcp6_flow.src_port,
4934 .dst_port = input->flow.tcp6_flow.dst_port,
4936 attributes->l4_mask.tcp.hdr = (struct rte_tcp_hdr){
4937 .src_port = mask->src_port_mask,
4938 .dst_port = mask->dst_port_mask,
4940 attributes->items[2] = (struct rte_flow_item){
4941 .type = RTE_FLOW_ITEM_TYPE_TCP,
4942 .spec = &attributes->l4,
4943 .mask = &attributes->l4_mask,
4946 case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
4947 case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
4950 DRV_LOG(ERR, "port %u invalid flow type%d",
4951 dev->data->port_id, fdir_filter->input.flow_type);
4952 rte_errno = ENOTSUP;
4958 #define FLOW_FDIR_CMP(f1, f2, fld) \
4959 memcmp(&(f1)->fld, &(f2)->fld, sizeof(f1->fld))
4962 * Compare two FDIR flows. If items and actions are identical, the two flows are
4966 * Pointer to Ethernet device.
4968 * FDIR flow to compare.
4970 * FDIR flow to compare.
4973 * Zero on match, 1 otherwise.
4976 flow_fdir_cmp(const struct mlx5_fdir *f1, const struct mlx5_fdir *f2)
4978 if (FLOW_FDIR_CMP(f1, f2, attr) ||
4979 FLOW_FDIR_CMP(f1, f2, l2) ||
4980 FLOW_FDIR_CMP(f1, f2, l2_mask) ||
4981 FLOW_FDIR_CMP(f1, f2, l3) ||
4982 FLOW_FDIR_CMP(f1, f2, l3_mask) ||
4983 FLOW_FDIR_CMP(f1, f2, l4) ||
4984 FLOW_FDIR_CMP(f1, f2, l4_mask) ||
4985 FLOW_FDIR_CMP(f1, f2, actions[0].type))
4987 if (f1->actions[0].type == RTE_FLOW_ACTION_TYPE_QUEUE &&
4988 FLOW_FDIR_CMP(f1, f2, queue))
4994 * Search device flow list to find out a matched FDIR flow.
4997 * Pointer to Ethernet device.
4999 * FDIR flow to lookup.
5002 * Pointer of flow if found, NULL otherwise.
5004 static struct rte_flow *
5005 flow_fdir_filter_lookup(struct rte_eth_dev *dev, struct mlx5_fdir *fdir_flow)
5007 struct mlx5_priv *priv = dev->data->dev_private;
5008 struct rte_flow *flow = NULL;
5011 TAILQ_FOREACH(flow, &priv->flows, next) {
5012 if (flow->fdir && !flow_fdir_cmp(flow->fdir, fdir_flow)) {
5013 DRV_LOG(DEBUG, "port %u found FDIR flow %p",
5014 dev->data->port_id, (void *)flow);
5022 * Add new flow director filter and store it in list.
5025 * Pointer to Ethernet device.
5026 * @param fdir_filter
5027 * Flow director filter to add.
5030 * 0 on success, a negative errno value otherwise and rte_errno is set.
5033 flow_fdir_filter_add(struct rte_eth_dev *dev,
5034 const struct rte_eth_fdir_filter *fdir_filter)
5036 struct mlx5_priv *priv = dev->data->dev_private;
5037 struct mlx5_fdir *fdir_flow;
5038 struct rte_flow *flow;
5041 fdir_flow = rte_zmalloc(__func__, sizeof(*fdir_flow), 0);
5046 ret = flow_fdir_filter_convert(dev, fdir_filter, fdir_flow);
5049 flow = flow_fdir_filter_lookup(dev, fdir_flow);
5054 flow = flow_list_create(dev, &priv->flows, &fdir_flow->attr,
5055 fdir_flow->items, fdir_flow->actions, true,
5059 assert(!flow->fdir);
5060 flow->fdir = fdir_flow;
5061 DRV_LOG(DEBUG, "port %u created FDIR flow %p",
5062 dev->data->port_id, (void *)flow);
5065 rte_free(fdir_flow);
5070 * Delete specific filter.
5073 * Pointer to Ethernet device.
5074 * @param fdir_filter
5075 * Filter to be deleted.
5078 * 0 on success, a negative errno value otherwise and rte_errno is set.
5081 flow_fdir_filter_delete(struct rte_eth_dev *dev,
5082 const struct rte_eth_fdir_filter *fdir_filter)
5084 struct mlx5_priv *priv = dev->data->dev_private;
5085 struct rte_flow *flow;
5086 struct mlx5_fdir fdir_flow = {
5091 ret = flow_fdir_filter_convert(dev, fdir_filter, &fdir_flow);
5094 flow = flow_fdir_filter_lookup(dev, &fdir_flow);
5099 flow_list_destroy(dev, &priv->flows, flow);
5100 DRV_LOG(DEBUG, "port %u deleted FDIR flow %p",
5101 dev->data->port_id, (void *)flow);
5106 * Update queue for specific filter.
5109 * Pointer to Ethernet device.
5110 * @param fdir_filter
5111 * Filter to be updated.
5114 * 0 on success, a negative errno value otherwise and rte_errno is set.
5117 flow_fdir_filter_update(struct rte_eth_dev *dev,
5118 const struct rte_eth_fdir_filter *fdir_filter)
5122 ret = flow_fdir_filter_delete(dev, fdir_filter);
5125 return flow_fdir_filter_add(dev, fdir_filter);
5129 * Flush all filters.
5132 * Pointer to Ethernet device.
5135 flow_fdir_filter_flush(struct rte_eth_dev *dev)
5137 struct mlx5_priv *priv = dev->data->dev_private;
5139 mlx5_flow_list_flush(dev, &priv->flows);
5143 * Get flow director information.
5146 * Pointer to Ethernet device.
5147 * @param[out] fdir_info
5148 * Resulting flow director information.
5151 flow_fdir_info_get(struct rte_eth_dev *dev, struct rte_eth_fdir_info *fdir_info)
5153 struct rte_eth_fdir_masks *mask =
5154 &dev->data->dev_conf.fdir_conf.mask;
5156 fdir_info->mode = dev->data->dev_conf.fdir_conf.mode;
5157 fdir_info->guarant_spc = 0;
5158 rte_memcpy(&fdir_info->mask, mask, sizeof(fdir_info->mask));
5159 fdir_info->max_flexpayload = 0;
5160 fdir_info->flow_types_mask[0] = 0;
5161 fdir_info->flex_payload_unit = 0;
5162 fdir_info->max_flex_payload_segment_num = 0;
5163 fdir_info->flex_payload_limit = 0;
5164 memset(&fdir_info->flex_conf, 0, sizeof(fdir_info->flex_conf));
5168 * Deal with flow director operations.
5171 * Pointer to Ethernet device.
5173 * Operation to perform.
5175 * Pointer to operation-specific structure.
5178 * 0 on success, a negative errno value otherwise and rte_errno is set.
5181 flow_fdir_ctrl_func(struct rte_eth_dev *dev, enum rte_filter_op filter_op,
5184 enum rte_fdir_mode fdir_mode =
5185 dev->data->dev_conf.fdir_conf.mode;
5187 if (filter_op == RTE_ETH_FILTER_NOP)
5189 if (fdir_mode != RTE_FDIR_MODE_PERFECT &&
5190 fdir_mode != RTE_FDIR_MODE_PERFECT_MAC_VLAN) {
5191 DRV_LOG(ERR, "port %u flow director mode %d not supported",
5192 dev->data->port_id, fdir_mode);
5196 switch (filter_op) {
5197 case RTE_ETH_FILTER_ADD:
5198 return flow_fdir_filter_add(dev, arg);
5199 case RTE_ETH_FILTER_UPDATE:
5200 return flow_fdir_filter_update(dev, arg);
5201 case RTE_ETH_FILTER_DELETE:
5202 return flow_fdir_filter_delete(dev, arg);
5203 case RTE_ETH_FILTER_FLUSH:
5204 flow_fdir_filter_flush(dev);
5206 case RTE_ETH_FILTER_INFO:
5207 flow_fdir_info_get(dev, arg);
5210 DRV_LOG(DEBUG, "port %u unknown operation %u",
5211 dev->data->port_id, filter_op);
5219 * Manage filter operations.
5222 * Pointer to Ethernet device structure.
5223 * @param filter_type
5226 * Operation to perform.
5228 * Pointer to operation-specific structure.
5231 * 0 on success, a negative errno value otherwise and rte_errno is set.
5234 mlx5_dev_filter_ctrl(struct rte_eth_dev *dev,
5235 enum rte_filter_type filter_type,
5236 enum rte_filter_op filter_op,
5239 switch (filter_type) {
5240 case RTE_ETH_FILTER_GENERIC:
5241 if (filter_op != RTE_ETH_FILTER_GET) {
5245 *(const void **)arg = &mlx5_flow_ops;
5247 case RTE_ETH_FILTER_FDIR:
5248 return flow_fdir_ctrl_func(dev, filter_op, arg);
5250 DRV_LOG(ERR, "port %u filter type (%d) not supported",
5251 dev->data->port_id, filter_type);
5252 rte_errno = ENOTSUP;
5259 * Create the needed meter and suffix tables.
5262 * Pointer to Ethernet device.
5264 * Pointer to the flow meter.
5267 * Pointer to table set on success, NULL otherwise.
5269 struct mlx5_meter_domains_infos *
5270 mlx5_flow_create_mtr_tbls(struct rte_eth_dev *dev,
5271 const struct mlx5_flow_meter *fm)
5273 const struct mlx5_flow_driver_ops *fops;
5275 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5276 return fops->create_mtr_tbls(dev, fm);
5280 * Destroy the meter table set.
5283 * Pointer to Ethernet device.
5285 * Pointer to the meter table set.
5291 mlx5_flow_destroy_mtr_tbls(struct rte_eth_dev *dev,
5292 struct mlx5_meter_domains_infos *tbls)
5294 const struct mlx5_flow_driver_ops *fops;
5296 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5297 return fops->destroy_mtr_tbls(dev, tbls);
5301 * Create policer rules.
5304 * Pointer to Ethernet device.
5306 * Pointer to flow meter structure.
5308 * Pointer to flow attributes.
5311 * 0 on success, -1 otherwise.
5314 mlx5_flow_create_policer_rules(struct rte_eth_dev *dev,
5315 struct mlx5_flow_meter *fm,
5316 const struct rte_flow_attr *attr)
5318 const struct mlx5_flow_driver_ops *fops;
5320 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5321 return fops->create_policer_rules(dev, fm, attr);
5325 * Destroy policer rules.
5328 * Pointer to flow meter structure.
5330 * Pointer to flow attributes.
5333 * 0 on success, -1 otherwise.
5336 mlx5_flow_destroy_policer_rules(struct rte_eth_dev *dev,
5337 struct mlx5_flow_meter *fm,
5338 const struct rte_flow_attr *attr)
5340 const struct mlx5_flow_driver_ops *fops;
5342 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5343 return fops->destroy_policer_rules(dev, fm, attr);
5347 * Allocate a counter.
5350 * Pointer to Ethernet device structure.
5353 * Pointer to allocated counter on success, NULL otherwise.
5355 struct mlx5_flow_counter *
5356 mlx5_counter_alloc(struct rte_eth_dev *dev)
5358 const struct mlx5_flow_driver_ops *fops;
5359 struct rte_flow_attr attr = { .transfer = 0 };
5361 if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
5362 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5363 return fops->counter_alloc(dev);
5366 "port %u counter allocate is not supported.",
5367 dev->data->port_id);
5375 * Pointer to Ethernet device structure.
5377 * Pointer to counter to be free.
5380 mlx5_counter_free(struct rte_eth_dev *dev, struct mlx5_flow_counter *cnt)
5382 const struct mlx5_flow_driver_ops *fops;
5383 struct rte_flow_attr attr = { .transfer = 0 };
5385 if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
5386 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5387 fops->counter_free(dev, cnt);
5391 "port %u counter free is not supported.",
5392 dev->data->port_id);
5396 * Query counter statistics.
5399 * Pointer to Ethernet device structure.
5401 * Pointer to counter to query.
5403 * Set to clear counter statistics.
5405 * The counter hits packets number to save.
5407 * The counter hits bytes number to save.
5410 * 0 on success, a negative errno value otherwise.
5413 mlx5_counter_query(struct rte_eth_dev *dev, struct mlx5_flow_counter *cnt,
5414 bool clear, uint64_t *pkts, uint64_t *bytes)
5416 const struct mlx5_flow_driver_ops *fops;
5417 struct rte_flow_attr attr = { .transfer = 0 };
5419 if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
5420 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5421 return fops->counter_query(dev, cnt, clear, pkts, bytes);
5424 "port %u counter query is not supported.",
5425 dev->data->port_id);
5429 #define MLX5_POOL_QUERY_FREQ_US 1000000
5432 * Set the periodic procedure for triggering asynchronous batch queries for all
5433 * the counter pools.
5436 * Pointer to mlx5_ibv_shared object.
5439 mlx5_set_query_alarm(struct mlx5_ibv_shared *sh)
5441 struct mlx5_pools_container *cont = MLX5_CNT_CONTAINER(sh, 0, 0);
5442 uint32_t pools_n = rte_atomic16_read(&cont->n_valid);
5445 cont = MLX5_CNT_CONTAINER(sh, 1, 0);
5446 pools_n += rte_atomic16_read(&cont->n_valid);
5447 us = MLX5_POOL_QUERY_FREQ_US / pools_n;
5448 DRV_LOG(DEBUG, "Set alarm for %u pools each %u us", pools_n, us);
5449 if (rte_eal_alarm_set(us, mlx5_flow_query_alarm, sh)) {
5450 sh->cmng.query_thread_on = 0;
5451 DRV_LOG(ERR, "Cannot reinitialize query alarm");
5453 sh->cmng.query_thread_on = 1;
5458 * The periodic procedure for triggering asynchronous batch queries for all the
5459 * counter pools. This function is probably called by the host thread.
5462 * The parameter for the alarm process.
5465 mlx5_flow_query_alarm(void *arg)
5467 struct mlx5_ibv_shared *sh = arg;
5468 struct mlx5_devx_obj *dcs;
5471 uint8_t batch = sh->cmng.batch;
5472 uint16_t pool_index = sh->cmng.pool_index;
5473 struct mlx5_pools_container *cont;
5474 struct mlx5_pools_container *mcont;
5475 struct mlx5_flow_counter_pool *pool;
5477 if (sh->cmng.pending_queries >= MLX5_MAX_PENDING_QUERIES)
5480 cont = MLX5_CNT_CONTAINER(sh, batch, 1);
5481 mcont = MLX5_CNT_CONTAINER(sh, batch, 0);
5482 /* Check if resize was done and need to flip a container. */
5483 if (cont != mcont) {
5485 /* Clean the old container. */
5486 rte_free(cont->pools);
5487 memset(cont, 0, sizeof(*cont));
5490 /* Flip the host container. */
5491 sh->cmng.mhi[batch] ^= (uint8_t)2;
5495 /* 2 empty containers case is unexpected. */
5496 if (unlikely(batch != sh->cmng.batch))
5500 goto next_container;
5502 pool = cont->pools[pool_index];
5504 /* There is a pool query in progress. */
5507 LIST_FIRST(&sh->cmng.free_stat_raws);
5509 /* No free counter statistics raw memory. */
5511 dcs = (struct mlx5_devx_obj *)(uintptr_t)rte_atomic64_read
5513 offset = batch ? 0 : dcs->id % MLX5_COUNTERS_PER_POOL;
5514 ret = mlx5_devx_cmd_flow_counter_query(dcs, 0, MLX5_COUNTERS_PER_POOL -
5516 pool->raw_hw->mem_mng->dm->id,
5518 (pool->raw_hw->data + offset),
5520 (uint64_t)(uintptr_t)pool);
5522 DRV_LOG(ERR, "Failed to trigger asynchronous query for dcs ID"
5523 " %d", pool->min_dcs->id);
5524 pool->raw_hw = NULL;
5527 pool->raw_hw->min_dcs_id = dcs->id;
5528 LIST_REMOVE(pool->raw_hw, next);
5529 sh->cmng.pending_queries++;
5531 if (pool_index >= rte_atomic16_read(&cont->n_valid)) {
5536 sh->cmng.batch = batch;
5537 sh->cmng.pool_index = pool_index;
5538 mlx5_set_query_alarm(sh);
5542 * Handler for the HW respond about ready values from an asynchronous batch
5543 * query. This function is probably called by the host thread.
5546 * The pointer to the shared IB device context.
5547 * @param[in] async_id
5548 * The Devx async ID.
5550 * The status of the completion.
5553 mlx5_flow_async_pool_query_handle(struct mlx5_ibv_shared *sh,
5554 uint64_t async_id, int status)
5556 struct mlx5_flow_counter_pool *pool =
5557 (struct mlx5_flow_counter_pool *)(uintptr_t)async_id;
5558 struct mlx5_counter_stats_raw *raw_to_free;
5560 if (unlikely(status)) {
5561 raw_to_free = pool->raw_hw;
5563 raw_to_free = pool->raw;
5564 rte_spinlock_lock(&pool->sl);
5565 pool->raw = pool->raw_hw;
5566 rte_spinlock_unlock(&pool->sl);
5567 rte_atomic64_add(&pool->query_gen, 1);
5568 /* Be sure the new raw counters data is updated in memory. */
5571 LIST_INSERT_HEAD(&sh->cmng.free_stat_raws, raw_to_free, next);
5572 pool->raw_hw = NULL;
5573 sh->cmng.pending_queries--;
5577 * Translate the rte_flow group index to HW table value.
5579 * @param[in] attributes
5580 * Pointer to flow attributes
5581 * @param[in] external
5582 * Value is part of flow rule created by request external to PMD.
5584 * rte_flow group index value.
5588 * Pointer to error structure.
5591 * 0 on success, a negative errno value otherwise and rte_errno is set.
5594 mlx5_flow_group_to_table(const struct rte_flow_attr *attributes, bool external,
5595 uint32_t group, uint32_t *table,
5596 struct rte_flow_error *error)
5598 if (attributes->transfer && external) {
5599 if (group == UINT32_MAX)
5600 return rte_flow_error_set
5602 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
5604 "group index not supported");
5613 * Discover availability of metadata reg_c's.
5615 * Iteratively use test flows to check availability.
5618 * Pointer to the Ethernet device structure.
5621 * 0 on success, a negative errno value otherwise and rte_errno is set.
5624 mlx5_flow_discover_mreg_c(struct rte_eth_dev *dev)
5626 struct mlx5_priv *priv = dev->data->dev_private;
5627 struct mlx5_dev_config *config = &priv->config;
5628 enum modify_reg idx;
5631 /* reg_c[0] and reg_c[1] are reserved. */
5632 config->flow_mreg_c[n++] = REG_C_0;
5633 config->flow_mreg_c[n++] = REG_C_1;
5634 /* Discover availability of other reg_c's. */
5635 for (idx = REG_C_2; idx <= REG_C_7; ++idx) {
5636 struct rte_flow_attr attr = {
5637 .group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
5638 .priority = MLX5_FLOW_PRIO_RSVD,
5641 struct rte_flow_item items[] = {
5643 .type = RTE_FLOW_ITEM_TYPE_END,
5646 struct rte_flow_action actions[] = {
5648 .type = MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
5649 .conf = &(struct mlx5_flow_action_copy_mreg){
5655 .type = RTE_FLOW_ACTION_TYPE_JUMP,
5656 .conf = &(struct rte_flow_action_jump){
5657 .group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
5661 .type = RTE_FLOW_ACTION_TYPE_END,
5664 struct rte_flow *flow;
5665 struct rte_flow_error error;
5667 if (!config->dv_flow_en)
5669 /* Create internal flow, validation skips copy action. */
5670 flow = flow_list_create(dev, NULL, &attr, items,
5671 actions, false, &error);
5674 if (dev->data->dev_started || !flow_drv_apply(dev, flow, NULL))
5675 config->flow_mreg_c[n++] = idx;
5676 flow_list_destroy(dev, NULL, flow);
5678 for (; n < MLX5_MREG_C_NUM; ++n)
5679 config->flow_mreg_c[n] = REG_NONE;