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>
30 #include <mlx5_glue.h>
31 #include <mlx5_devx_cmds.h>
34 #include "mlx5_defs.h"
36 #include "mlx5_flow.h"
37 #include "mlx5_rxtx.h"
39 /* Dev ops structure defined in mlx5.c */
40 extern const struct eth_dev_ops mlx5_dev_ops;
41 extern const struct eth_dev_ops mlx5_dev_ops_isolate;
43 /** Device flow drivers. */
44 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
45 extern const struct mlx5_flow_driver_ops mlx5_flow_dv_drv_ops;
47 extern const struct mlx5_flow_driver_ops mlx5_flow_verbs_drv_ops;
49 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops;
51 const struct mlx5_flow_driver_ops *flow_drv_ops[] = {
52 [MLX5_FLOW_TYPE_MIN] = &mlx5_flow_null_drv_ops,
53 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
54 [MLX5_FLOW_TYPE_DV] = &mlx5_flow_dv_drv_ops,
56 [MLX5_FLOW_TYPE_VERBS] = &mlx5_flow_verbs_drv_ops,
57 [MLX5_FLOW_TYPE_MAX] = &mlx5_flow_null_drv_ops
62 MLX5_EXPANSION_ROOT_OUTER,
63 MLX5_EXPANSION_ROOT_ETH_VLAN,
64 MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN,
65 MLX5_EXPANSION_OUTER_ETH,
66 MLX5_EXPANSION_OUTER_ETH_VLAN,
67 MLX5_EXPANSION_OUTER_VLAN,
68 MLX5_EXPANSION_OUTER_IPV4,
69 MLX5_EXPANSION_OUTER_IPV4_UDP,
70 MLX5_EXPANSION_OUTER_IPV4_TCP,
71 MLX5_EXPANSION_OUTER_IPV6,
72 MLX5_EXPANSION_OUTER_IPV6_UDP,
73 MLX5_EXPANSION_OUTER_IPV6_TCP,
75 MLX5_EXPANSION_VXLAN_GPE,
79 MLX5_EXPANSION_ETH_VLAN,
82 MLX5_EXPANSION_IPV4_UDP,
83 MLX5_EXPANSION_IPV4_TCP,
85 MLX5_EXPANSION_IPV6_UDP,
86 MLX5_EXPANSION_IPV6_TCP,
89 /** Supported expansion of items. */
90 static const struct rte_flow_expand_node mlx5_support_expansion[] = {
91 [MLX5_EXPANSION_ROOT] = {
92 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
95 .type = RTE_FLOW_ITEM_TYPE_END,
97 [MLX5_EXPANSION_ROOT_OUTER] = {
98 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_ETH,
99 MLX5_EXPANSION_OUTER_IPV4,
100 MLX5_EXPANSION_OUTER_IPV6),
101 .type = RTE_FLOW_ITEM_TYPE_END,
103 [MLX5_EXPANSION_ROOT_ETH_VLAN] = {
104 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH_VLAN),
105 .type = RTE_FLOW_ITEM_TYPE_END,
107 [MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN] = {
108 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_ETH_VLAN),
109 .type = RTE_FLOW_ITEM_TYPE_END,
111 [MLX5_EXPANSION_OUTER_ETH] = {
112 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
113 MLX5_EXPANSION_OUTER_IPV6,
114 MLX5_EXPANSION_MPLS),
115 .type = RTE_FLOW_ITEM_TYPE_ETH,
118 [MLX5_EXPANSION_OUTER_ETH_VLAN] = {
119 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_VLAN),
120 .type = RTE_FLOW_ITEM_TYPE_ETH,
123 [MLX5_EXPANSION_OUTER_VLAN] = {
124 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_OUTER_IPV4,
125 MLX5_EXPANSION_OUTER_IPV6),
126 .type = RTE_FLOW_ITEM_TYPE_VLAN,
128 [MLX5_EXPANSION_OUTER_IPV4] = {
129 .next = RTE_FLOW_EXPAND_RSS_NEXT
130 (MLX5_EXPANSION_OUTER_IPV4_UDP,
131 MLX5_EXPANSION_OUTER_IPV4_TCP,
134 MLX5_EXPANSION_IPV6),
135 .type = RTE_FLOW_ITEM_TYPE_IPV4,
136 .rss_types = ETH_RSS_IPV4 | ETH_RSS_FRAG_IPV4 |
137 ETH_RSS_NONFRAG_IPV4_OTHER,
139 [MLX5_EXPANSION_OUTER_IPV4_UDP] = {
140 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
141 MLX5_EXPANSION_VXLAN_GPE),
142 .type = RTE_FLOW_ITEM_TYPE_UDP,
143 .rss_types = ETH_RSS_NONFRAG_IPV4_UDP,
145 [MLX5_EXPANSION_OUTER_IPV4_TCP] = {
146 .type = RTE_FLOW_ITEM_TYPE_TCP,
147 .rss_types = ETH_RSS_NONFRAG_IPV4_TCP,
149 [MLX5_EXPANSION_OUTER_IPV6] = {
150 .next = RTE_FLOW_EXPAND_RSS_NEXT
151 (MLX5_EXPANSION_OUTER_IPV6_UDP,
152 MLX5_EXPANSION_OUTER_IPV6_TCP,
154 MLX5_EXPANSION_IPV6),
155 .type = RTE_FLOW_ITEM_TYPE_IPV6,
156 .rss_types = ETH_RSS_IPV6 | ETH_RSS_FRAG_IPV6 |
157 ETH_RSS_NONFRAG_IPV6_OTHER,
159 [MLX5_EXPANSION_OUTER_IPV6_UDP] = {
160 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VXLAN,
161 MLX5_EXPANSION_VXLAN_GPE),
162 .type = RTE_FLOW_ITEM_TYPE_UDP,
163 .rss_types = ETH_RSS_NONFRAG_IPV6_UDP,
165 [MLX5_EXPANSION_OUTER_IPV6_TCP] = {
166 .type = RTE_FLOW_ITEM_TYPE_TCP,
167 .rss_types = ETH_RSS_NONFRAG_IPV6_TCP,
169 [MLX5_EXPANSION_VXLAN] = {
170 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
172 MLX5_EXPANSION_IPV6),
173 .type = RTE_FLOW_ITEM_TYPE_VXLAN,
175 [MLX5_EXPANSION_VXLAN_GPE] = {
176 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
178 MLX5_EXPANSION_IPV6),
179 .type = RTE_FLOW_ITEM_TYPE_VXLAN_GPE,
181 [MLX5_EXPANSION_GRE] = {
182 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4),
183 .type = RTE_FLOW_ITEM_TYPE_GRE,
185 [MLX5_EXPANSION_MPLS] = {
186 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
187 MLX5_EXPANSION_IPV6),
188 .type = RTE_FLOW_ITEM_TYPE_MPLS,
190 [MLX5_EXPANSION_ETH] = {
191 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
192 MLX5_EXPANSION_IPV6),
193 .type = RTE_FLOW_ITEM_TYPE_ETH,
195 [MLX5_EXPANSION_ETH_VLAN] = {
196 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VLAN),
197 .type = RTE_FLOW_ITEM_TYPE_ETH,
199 [MLX5_EXPANSION_VLAN] = {
200 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
201 MLX5_EXPANSION_IPV6),
202 .type = RTE_FLOW_ITEM_TYPE_VLAN,
204 [MLX5_EXPANSION_IPV4] = {
205 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4_UDP,
206 MLX5_EXPANSION_IPV4_TCP),
207 .type = RTE_FLOW_ITEM_TYPE_IPV4,
208 .rss_types = ETH_RSS_IPV4 | ETH_RSS_FRAG_IPV4 |
209 ETH_RSS_NONFRAG_IPV4_OTHER,
211 [MLX5_EXPANSION_IPV4_UDP] = {
212 .type = RTE_FLOW_ITEM_TYPE_UDP,
213 .rss_types = ETH_RSS_NONFRAG_IPV4_UDP,
215 [MLX5_EXPANSION_IPV4_TCP] = {
216 .type = RTE_FLOW_ITEM_TYPE_TCP,
217 .rss_types = ETH_RSS_NONFRAG_IPV4_TCP,
219 [MLX5_EXPANSION_IPV6] = {
220 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV6_UDP,
221 MLX5_EXPANSION_IPV6_TCP),
222 .type = RTE_FLOW_ITEM_TYPE_IPV6,
223 .rss_types = ETH_RSS_IPV6 | ETH_RSS_FRAG_IPV6 |
224 ETH_RSS_NONFRAG_IPV6_OTHER,
226 [MLX5_EXPANSION_IPV6_UDP] = {
227 .type = RTE_FLOW_ITEM_TYPE_UDP,
228 .rss_types = ETH_RSS_NONFRAG_IPV6_UDP,
230 [MLX5_EXPANSION_IPV6_TCP] = {
231 .type = RTE_FLOW_ITEM_TYPE_TCP,
232 .rss_types = ETH_RSS_NONFRAG_IPV6_TCP,
236 static const struct rte_flow_ops mlx5_flow_ops = {
237 .validate = mlx5_flow_validate,
238 .create = mlx5_flow_create,
239 .destroy = mlx5_flow_destroy,
240 .flush = mlx5_flow_flush,
241 .isolate = mlx5_flow_isolate,
242 .query = mlx5_flow_query,
243 .dev_dump = mlx5_flow_dev_dump,
246 /* Convert FDIR request to Generic flow. */
248 struct rte_flow_attr attr;
249 struct rte_flow_item items[4];
250 struct rte_flow_item_eth l2;
251 struct rte_flow_item_eth l2_mask;
253 struct rte_flow_item_ipv4 ipv4;
254 struct rte_flow_item_ipv6 ipv6;
257 struct rte_flow_item_ipv4 ipv4;
258 struct rte_flow_item_ipv6 ipv6;
261 struct rte_flow_item_udp udp;
262 struct rte_flow_item_tcp tcp;
265 struct rte_flow_item_udp udp;
266 struct rte_flow_item_tcp tcp;
268 struct rte_flow_action actions[2];
269 struct rte_flow_action_queue queue;
272 /* Map of Verbs to Flow priority with 8 Verbs priorities. */
273 static const uint32_t priority_map_3[][MLX5_PRIORITY_MAP_MAX] = {
274 { 0, 1, 2 }, { 2, 3, 4 }, { 5, 6, 7 },
277 /* Map of Verbs to Flow priority with 16 Verbs priorities. */
278 static const uint32_t priority_map_5[][MLX5_PRIORITY_MAP_MAX] = {
279 { 0, 1, 2 }, { 3, 4, 5 }, { 6, 7, 8 },
280 { 9, 10, 11 }, { 12, 13, 14 },
283 /* Tunnel information. */
284 struct mlx5_flow_tunnel_info {
285 uint64_t tunnel; /**< Tunnel bit (see MLX5_FLOW_*). */
286 uint32_t ptype; /**< Tunnel Ptype (see RTE_PTYPE_*). */
289 static struct mlx5_flow_tunnel_info tunnels_info[] = {
291 .tunnel = MLX5_FLOW_LAYER_VXLAN,
292 .ptype = RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L4_UDP,
295 .tunnel = MLX5_FLOW_LAYER_GENEVE,
296 .ptype = RTE_PTYPE_TUNNEL_GENEVE | RTE_PTYPE_L4_UDP,
299 .tunnel = MLX5_FLOW_LAYER_VXLAN_GPE,
300 .ptype = RTE_PTYPE_TUNNEL_VXLAN_GPE | RTE_PTYPE_L4_UDP,
303 .tunnel = MLX5_FLOW_LAYER_GRE,
304 .ptype = RTE_PTYPE_TUNNEL_GRE,
307 .tunnel = MLX5_FLOW_LAYER_MPLS | MLX5_FLOW_LAYER_OUTER_L4_UDP,
308 .ptype = RTE_PTYPE_TUNNEL_MPLS_IN_UDP | RTE_PTYPE_L4_UDP,
311 .tunnel = MLX5_FLOW_LAYER_MPLS,
312 .ptype = RTE_PTYPE_TUNNEL_MPLS_IN_GRE,
315 .tunnel = MLX5_FLOW_LAYER_NVGRE,
316 .ptype = RTE_PTYPE_TUNNEL_NVGRE,
319 .tunnel = MLX5_FLOW_LAYER_IPIP,
320 .ptype = RTE_PTYPE_TUNNEL_IP,
323 .tunnel = MLX5_FLOW_LAYER_IPV6_ENCAP,
324 .ptype = RTE_PTYPE_TUNNEL_IP,
327 .tunnel = MLX5_FLOW_LAYER_GTP,
328 .ptype = RTE_PTYPE_TUNNEL_GTPU,
333 * Translate tag ID to register.
336 * Pointer to the Ethernet device structure.
338 * The feature that request the register.
340 * The request register ID.
342 * Error description in case of any.
345 * The request register on success, a negative errno
346 * value otherwise and rte_errno is set.
349 mlx5_flow_get_reg_id(struct rte_eth_dev *dev,
350 enum mlx5_feature_name feature,
352 struct rte_flow_error *error)
354 struct mlx5_priv *priv = dev->data->dev_private;
355 struct mlx5_dev_config *config = &priv->config;
356 enum modify_reg start_reg;
357 bool skip_mtr_reg = false;
360 case MLX5_HAIRPIN_RX:
362 case MLX5_HAIRPIN_TX:
364 case MLX5_METADATA_RX:
365 switch (config->dv_xmeta_en) {
366 case MLX5_XMETA_MODE_LEGACY:
368 case MLX5_XMETA_MODE_META16:
370 case MLX5_XMETA_MODE_META32:
374 case MLX5_METADATA_TX:
376 case MLX5_METADATA_FDB:
377 switch (config->dv_xmeta_en) {
378 case MLX5_XMETA_MODE_LEGACY:
380 case MLX5_XMETA_MODE_META16:
382 case MLX5_XMETA_MODE_META32:
387 switch (config->dv_xmeta_en) {
388 case MLX5_XMETA_MODE_LEGACY:
390 case MLX5_XMETA_MODE_META16:
392 case MLX5_XMETA_MODE_META32:
398 * If meter color and flow match share one register, flow match
399 * should use the meter color register for match.
401 if (priv->mtr_reg_share)
402 return priv->mtr_color_reg;
404 return priv->mtr_color_reg != REG_C_2 ? REG_C_2 :
407 MLX5_ASSERT(priv->mtr_color_reg != REG_NONE);
408 return priv->mtr_color_reg;
411 * Metadata COPY_MARK register using is in meter suffix sub
412 * flow while with meter. It's safe to share the same register.
414 return priv->mtr_color_reg != REG_C_2 ? REG_C_2 : REG_C_3;
417 * If meter is enable, it will engage the register for color
418 * match and flow match. If meter color match is not using the
419 * REG_C_2, need to skip the REG_C_x be used by meter color
421 * If meter is disable, free to use all available registers.
423 start_reg = priv->mtr_color_reg != REG_C_2 ? REG_C_2 :
424 (priv->mtr_reg_share ? REG_C_3 : REG_C_4);
425 skip_mtr_reg = !!(priv->mtr_en && start_reg == REG_C_2);
426 if (id > (REG_C_7 - start_reg))
427 return rte_flow_error_set(error, EINVAL,
428 RTE_FLOW_ERROR_TYPE_ITEM,
429 NULL, "invalid tag id");
430 if (config->flow_mreg_c[id + start_reg - REG_C_0] == REG_NONE)
431 return rte_flow_error_set(error, ENOTSUP,
432 RTE_FLOW_ERROR_TYPE_ITEM,
433 NULL, "unsupported tag id");
435 * This case means meter is using the REG_C_x great than 2.
436 * Take care not to conflict with meter color REG_C_x.
437 * If the available index REG_C_y >= REG_C_x, skip the
440 if (skip_mtr_reg && config->flow_mreg_c
441 [id + start_reg - REG_C_0] >= priv->mtr_color_reg) {
442 if (config->flow_mreg_c
443 [id + 1 + start_reg - REG_C_0] != REG_NONE)
444 return config->flow_mreg_c
445 [id + 1 + start_reg - REG_C_0];
446 return rte_flow_error_set(error, ENOTSUP,
447 RTE_FLOW_ERROR_TYPE_ITEM,
448 NULL, "unsupported tag id");
450 return config->flow_mreg_c[id + start_reg - REG_C_0];
453 return rte_flow_error_set(error, EINVAL,
454 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
455 NULL, "invalid feature name");
459 * Check extensive flow metadata register support.
462 * Pointer to rte_eth_dev structure.
465 * True if device supports extensive flow metadata register, otherwise false.
468 mlx5_flow_ext_mreg_supported(struct rte_eth_dev *dev)
470 struct mlx5_priv *priv = dev->data->dev_private;
471 struct mlx5_dev_config *config = &priv->config;
474 * Having available reg_c can be regarded inclusively as supporting
475 * extensive flow metadata register, which could mean,
476 * - metadata register copy action by modify header.
477 * - 16 modify header actions is supported.
478 * - reg_c's are preserved across different domain (FDB and NIC) on
479 * packet loopback by flow lookup miss.
481 return config->flow_mreg_c[2] != REG_NONE;
485 * Discover the maximum number of priority available.
488 * Pointer to the Ethernet device structure.
491 * number of supported flow priority on success, a negative errno
492 * value otherwise and rte_errno is set.
495 mlx5_flow_discover_priorities(struct rte_eth_dev *dev)
497 struct mlx5_priv *priv = dev->data->dev_private;
499 struct ibv_flow_attr attr;
500 struct ibv_flow_spec_eth eth;
501 struct ibv_flow_spec_action_drop drop;
505 .port = (uint8_t)priv->ibv_port,
508 .type = IBV_FLOW_SPEC_ETH,
509 .size = sizeof(struct ibv_flow_spec_eth),
512 .size = sizeof(struct ibv_flow_spec_action_drop),
513 .type = IBV_FLOW_SPEC_ACTION_DROP,
516 struct ibv_flow *flow;
517 struct mlx5_hrxq *drop = mlx5_hrxq_drop_new(dev);
518 uint16_t vprio[] = { 8, 16 };
526 for (i = 0; i != RTE_DIM(vprio); i++) {
527 flow_attr.attr.priority = vprio[i] - 1;
528 flow = mlx5_glue->create_flow(drop->qp, &flow_attr.attr);
531 claim_zero(mlx5_glue->destroy_flow(flow));
534 mlx5_hrxq_drop_release(dev);
537 priority = RTE_DIM(priority_map_3);
540 priority = RTE_DIM(priority_map_5);
545 "port %u verbs maximum priority: %d expected 8/16",
546 dev->data->port_id, priority);
549 DRV_LOG(INFO, "port %u flow maximum priority: %d",
550 dev->data->port_id, priority);
555 * Adjust flow priority based on the highest layer and the request priority.
558 * Pointer to the Ethernet device structure.
559 * @param[in] priority
560 * The rule base priority.
561 * @param[in] subpriority
562 * The priority based on the items.
567 uint32_t mlx5_flow_adjust_priority(struct rte_eth_dev *dev, int32_t priority,
568 uint32_t subpriority)
571 struct mlx5_priv *priv = dev->data->dev_private;
573 switch (priv->config.flow_prio) {
574 case RTE_DIM(priority_map_3):
575 res = priority_map_3[priority][subpriority];
577 case RTE_DIM(priority_map_5):
578 res = priority_map_5[priority][subpriority];
585 * Verify the @p item specifications (spec, last, mask) are compatible with the
589 * Item specification.
591 * @p item->mask or flow default bit-masks.
592 * @param[in] nic_mask
593 * Bit-masks covering supported fields by the NIC to compare with user mask.
595 * Bit-masks size in bytes.
597 * Pointer to error structure.
600 * 0 on success, a negative errno value otherwise and rte_errno is set.
603 mlx5_flow_item_acceptable(const struct rte_flow_item *item,
605 const uint8_t *nic_mask,
607 struct rte_flow_error *error)
611 MLX5_ASSERT(nic_mask);
612 for (i = 0; i < size; ++i)
613 if ((nic_mask[i] | mask[i]) != nic_mask[i])
614 return rte_flow_error_set(error, ENOTSUP,
615 RTE_FLOW_ERROR_TYPE_ITEM,
617 "mask enables non supported"
619 if (!item->spec && (item->mask || item->last))
620 return rte_flow_error_set(error, EINVAL,
621 RTE_FLOW_ERROR_TYPE_ITEM, item,
622 "mask/last without a spec is not"
624 if (item->spec && item->last) {
630 for (i = 0; i < size; ++i) {
631 spec[i] = ((const uint8_t *)item->spec)[i] & mask[i];
632 last[i] = ((const uint8_t *)item->last)[i] & mask[i];
634 ret = memcmp(spec, last, size);
636 return rte_flow_error_set(error, EINVAL,
637 RTE_FLOW_ERROR_TYPE_ITEM,
639 "range is not valid");
645 * Adjust the hash fields according to the @p flow information.
647 * @param[in] dev_flow.
648 * Pointer to the mlx5_flow.
650 * 1 when the hash field is for a tunnel item.
651 * @param[in] layer_types
653 * @param[in] hash_fields
657 * The hash fields that should be used.
660 mlx5_flow_hashfields_adjust(struct mlx5_flow *dev_flow,
661 int tunnel __rte_unused, uint64_t layer_types,
662 uint64_t hash_fields)
664 struct rte_flow *flow = dev_flow->flow;
665 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
666 int rss_request_inner = flow->rss.level >= 2;
668 /* Check RSS hash level for tunnel. */
669 if (tunnel && rss_request_inner)
670 hash_fields |= IBV_RX_HASH_INNER;
671 else if (tunnel || rss_request_inner)
674 /* Check if requested layer matches RSS hash fields. */
675 if (!(flow->rss.types & layer_types))
681 * Lookup and set the ptype in the data Rx part. A single Ptype can be used,
682 * if several tunnel rules are used on this queue, the tunnel ptype will be
686 * Rx queue to update.
689 flow_rxq_tunnel_ptype_update(struct mlx5_rxq_ctrl *rxq_ctrl)
692 uint32_t tunnel_ptype = 0;
694 /* Look up for the ptype to use. */
695 for (i = 0; i != MLX5_FLOW_TUNNEL; ++i) {
696 if (!rxq_ctrl->flow_tunnels_n[i])
699 tunnel_ptype = tunnels_info[i].ptype;
705 rxq_ctrl->rxq.tunnel = tunnel_ptype;
709 * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) according to the devive
713 * Pointer to the Ethernet device structure.
714 * @param[in] dev_flow
715 * Pointer to device flow structure.
718 flow_drv_rxq_flags_set(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow)
720 struct mlx5_priv *priv = dev->data->dev_private;
721 struct rte_flow *flow = dev_flow->flow;
722 const int mark = !!(dev_flow->actions &
723 (MLX5_FLOW_ACTION_FLAG | MLX5_FLOW_ACTION_MARK));
724 const int tunnel = !!(dev_flow->layers & MLX5_FLOW_LAYER_TUNNEL);
727 for (i = 0; i != flow->rss.queue_num; ++i) {
728 int idx = (*flow->rss.queue)[i];
729 struct mlx5_rxq_ctrl *rxq_ctrl =
730 container_of((*priv->rxqs)[idx],
731 struct mlx5_rxq_ctrl, rxq);
734 * To support metadata register copy on Tx loopback,
735 * this must be always enabled (metadata may arive
736 * from other port - not from local flows only.
738 if (priv->config.dv_flow_en &&
739 priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
740 mlx5_flow_ext_mreg_supported(dev)) {
741 rxq_ctrl->rxq.mark = 1;
742 rxq_ctrl->flow_mark_n = 1;
744 rxq_ctrl->rxq.mark = 1;
745 rxq_ctrl->flow_mark_n++;
750 /* Increase the counter matching the flow. */
751 for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
752 if ((tunnels_info[j].tunnel &
754 tunnels_info[j].tunnel) {
755 rxq_ctrl->flow_tunnels_n[j]++;
759 flow_rxq_tunnel_ptype_update(rxq_ctrl);
765 * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) for a flow
768 * Pointer to the Ethernet device structure.
770 * Pointer to flow structure.
773 flow_rxq_flags_set(struct rte_eth_dev *dev, struct rte_flow *flow)
775 struct mlx5_flow *dev_flow;
777 LIST_FOREACH(dev_flow, &flow->dev_flows, next)
778 flow_drv_rxq_flags_set(dev, dev_flow);
782 * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
783 * device flow if no other flow uses it with the same kind of request.
786 * Pointer to Ethernet device.
787 * @param[in] dev_flow
788 * Pointer to the device flow.
791 flow_drv_rxq_flags_trim(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow)
793 struct mlx5_priv *priv = dev->data->dev_private;
794 struct rte_flow *flow = dev_flow->flow;
795 const int mark = !!(dev_flow->actions &
796 (MLX5_FLOW_ACTION_FLAG | MLX5_FLOW_ACTION_MARK));
797 const int tunnel = !!(dev_flow->layers & MLX5_FLOW_LAYER_TUNNEL);
800 MLX5_ASSERT(dev->data->dev_started);
801 for (i = 0; i != flow->rss.queue_num; ++i) {
802 int idx = (*flow->rss.queue)[i];
803 struct mlx5_rxq_ctrl *rxq_ctrl =
804 container_of((*priv->rxqs)[idx],
805 struct mlx5_rxq_ctrl, rxq);
807 if (priv->config.dv_flow_en &&
808 priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
809 mlx5_flow_ext_mreg_supported(dev)) {
810 rxq_ctrl->rxq.mark = 1;
811 rxq_ctrl->flow_mark_n = 1;
813 rxq_ctrl->flow_mark_n--;
814 rxq_ctrl->rxq.mark = !!rxq_ctrl->flow_mark_n;
819 /* Decrease the counter matching the flow. */
820 for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
821 if ((tunnels_info[j].tunnel &
823 tunnels_info[j].tunnel) {
824 rxq_ctrl->flow_tunnels_n[j]--;
828 flow_rxq_tunnel_ptype_update(rxq_ctrl);
834 * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
835 * @p flow if no other flow uses it with the same kind of request.
838 * Pointer to Ethernet device.
840 * Pointer to the flow.
843 flow_rxq_flags_trim(struct rte_eth_dev *dev, struct rte_flow *flow)
845 struct mlx5_flow *dev_flow;
847 LIST_FOREACH(dev_flow, &flow->dev_flows, next)
848 flow_drv_rxq_flags_trim(dev, dev_flow);
852 * Clear the Mark/Flag and Tunnel ptype information in all Rx queues.
855 * Pointer to Ethernet device.
858 flow_rxq_flags_clear(struct rte_eth_dev *dev)
860 struct mlx5_priv *priv = dev->data->dev_private;
863 for (i = 0; i != priv->rxqs_n; ++i) {
864 struct mlx5_rxq_ctrl *rxq_ctrl;
867 if (!(*priv->rxqs)[i])
869 rxq_ctrl = container_of((*priv->rxqs)[i],
870 struct mlx5_rxq_ctrl, rxq);
871 rxq_ctrl->flow_mark_n = 0;
872 rxq_ctrl->rxq.mark = 0;
873 for (j = 0; j != MLX5_FLOW_TUNNEL; ++j)
874 rxq_ctrl->flow_tunnels_n[j] = 0;
875 rxq_ctrl->rxq.tunnel = 0;
880 * return a pointer to the desired action in the list of actions.
883 * The list of actions to search the action in.
885 * The action to find.
888 * Pointer to the action in the list, if found. NULL otherwise.
890 const struct rte_flow_action *
891 mlx5_flow_find_action(const struct rte_flow_action *actions,
892 enum rte_flow_action_type action)
896 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++)
897 if (actions->type == action)
903 * Validate the flag action.
905 * @param[in] action_flags
906 * Bit-fields that holds the actions detected until now.
908 * Attributes of flow that includes this action.
910 * Pointer to error structure.
913 * 0 on success, a negative errno value otherwise and rte_errno is set.
916 mlx5_flow_validate_action_flag(uint64_t action_flags,
917 const struct rte_flow_attr *attr,
918 struct rte_flow_error *error)
920 if (action_flags & MLX5_FLOW_ACTION_MARK)
921 return rte_flow_error_set(error, EINVAL,
922 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
923 "can't mark and flag in same flow");
924 if (action_flags & MLX5_FLOW_ACTION_FLAG)
925 return rte_flow_error_set(error, EINVAL,
926 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
928 " actions in same flow");
930 return rte_flow_error_set(error, ENOTSUP,
931 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
932 "flag action not supported for "
938 * Validate the mark action.
941 * Pointer to the queue action.
942 * @param[in] action_flags
943 * Bit-fields that holds the actions detected until now.
945 * Attributes of flow that includes this action.
947 * Pointer to error structure.
950 * 0 on success, a negative errno value otherwise and rte_errno is set.
953 mlx5_flow_validate_action_mark(const struct rte_flow_action *action,
954 uint64_t action_flags,
955 const struct rte_flow_attr *attr,
956 struct rte_flow_error *error)
958 const struct rte_flow_action_mark *mark = action->conf;
961 return rte_flow_error_set(error, EINVAL,
962 RTE_FLOW_ERROR_TYPE_ACTION,
964 "configuration cannot be null");
965 if (mark->id >= MLX5_FLOW_MARK_MAX)
966 return rte_flow_error_set(error, EINVAL,
967 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
969 "mark id must in 0 <= id < "
970 RTE_STR(MLX5_FLOW_MARK_MAX));
971 if (action_flags & MLX5_FLOW_ACTION_FLAG)
972 return rte_flow_error_set(error, EINVAL,
973 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
974 "can't flag and mark in same flow");
975 if (action_flags & MLX5_FLOW_ACTION_MARK)
976 return rte_flow_error_set(error, EINVAL,
977 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
978 "can't have 2 mark actions in same"
981 return rte_flow_error_set(error, ENOTSUP,
982 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
983 "mark action not supported for "
989 * Validate the drop action.
991 * @param[in] action_flags
992 * Bit-fields that holds the actions detected until now.
994 * Attributes of flow that includes this action.
996 * Pointer to error structure.
999 * 0 on success, a negative errno value otherwise and rte_errno is set.
1002 mlx5_flow_validate_action_drop(uint64_t action_flags __rte_unused,
1003 const struct rte_flow_attr *attr,
1004 struct rte_flow_error *error)
1007 return rte_flow_error_set(error, ENOTSUP,
1008 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1009 "drop action not supported for "
1015 * Validate the queue action.
1018 * Pointer to the queue action.
1019 * @param[in] action_flags
1020 * Bit-fields that holds the actions detected until now.
1022 * Pointer to the Ethernet device structure.
1024 * Attributes of flow that includes this action.
1026 * Pointer to error structure.
1029 * 0 on success, a negative errno value otherwise and rte_errno is set.
1032 mlx5_flow_validate_action_queue(const struct rte_flow_action *action,
1033 uint64_t action_flags,
1034 struct rte_eth_dev *dev,
1035 const struct rte_flow_attr *attr,
1036 struct rte_flow_error *error)
1038 struct mlx5_priv *priv = dev->data->dev_private;
1039 const struct rte_flow_action_queue *queue = action->conf;
1041 if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1042 return rte_flow_error_set(error, EINVAL,
1043 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1044 "can't have 2 fate actions in"
1047 return rte_flow_error_set(error, EINVAL,
1048 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1049 NULL, "No Rx queues configured");
1050 if (queue->index >= priv->rxqs_n)
1051 return rte_flow_error_set(error, EINVAL,
1052 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1054 "queue index out of range");
1055 if (!(*priv->rxqs)[queue->index])
1056 return rte_flow_error_set(error, EINVAL,
1057 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1059 "queue is not configured");
1061 return rte_flow_error_set(error, ENOTSUP,
1062 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1063 "queue action not supported for "
1069 * Validate the rss action.
1072 * Pointer to the queue action.
1073 * @param[in] action_flags
1074 * Bit-fields that holds the actions detected until now.
1076 * Pointer to the Ethernet device structure.
1078 * Attributes of flow that includes this action.
1079 * @param[in] item_flags
1080 * Items that were detected.
1082 * Pointer to error structure.
1085 * 0 on success, a negative errno value otherwise and rte_errno is set.
1088 mlx5_flow_validate_action_rss(const struct rte_flow_action *action,
1089 uint64_t action_flags,
1090 struct rte_eth_dev *dev,
1091 const struct rte_flow_attr *attr,
1092 uint64_t item_flags,
1093 struct rte_flow_error *error)
1095 struct mlx5_priv *priv = dev->data->dev_private;
1096 const struct rte_flow_action_rss *rss = action->conf;
1097 int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1100 if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1101 return rte_flow_error_set(error, EINVAL,
1102 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1103 "can't have 2 fate actions"
1105 if (rss->func != RTE_ETH_HASH_FUNCTION_DEFAULT &&
1106 rss->func != RTE_ETH_HASH_FUNCTION_TOEPLITZ)
1107 return rte_flow_error_set(error, ENOTSUP,
1108 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1110 "RSS hash function not supported");
1111 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
1116 return rte_flow_error_set(error, ENOTSUP,
1117 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1119 "tunnel RSS is not supported");
1120 /* allow RSS key_len 0 in case of NULL (default) RSS key. */
1121 if (rss->key_len == 0 && rss->key != NULL)
1122 return rte_flow_error_set(error, ENOTSUP,
1123 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1125 "RSS hash key length 0");
1126 if (rss->key_len > 0 && rss->key_len < MLX5_RSS_HASH_KEY_LEN)
1127 return rte_flow_error_set(error, ENOTSUP,
1128 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1130 "RSS hash key too small");
1131 if (rss->key_len > MLX5_RSS_HASH_KEY_LEN)
1132 return rte_flow_error_set(error, ENOTSUP,
1133 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1135 "RSS hash key too large");
1136 if (rss->queue_num > priv->config.ind_table_max_size)
1137 return rte_flow_error_set(error, ENOTSUP,
1138 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1140 "number of queues too large");
1141 if (rss->types & MLX5_RSS_HF_MASK)
1142 return rte_flow_error_set(error, ENOTSUP,
1143 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1145 "some RSS protocols are not"
1147 if ((rss->types & (ETH_RSS_L3_SRC_ONLY | ETH_RSS_L3_DST_ONLY)) &&
1148 !(rss->types & ETH_RSS_IP))
1149 return rte_flow_error_set(error, EINVAL,
1150 RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1151 "L3 partial RSS requested but L3 RSS"
1152 " type not specified");
1153 if ((rss->types & (ETH_RSS_L4_SRC_ONLY | ETH_RSS_L4_DST_ONLY)) &&
1154 !(rss->types & (ETH_RSS_UDP | ETH_RSS_TCP)))
1155 return rte_flow_error_set(error, EINVAL,
1156 RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1157 "L4 partial RSS requested but L4 RSS"
1158 " type not specified");
1160 return rte_flow_error_set(error, EINVAL,
1161 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1162 NULL, "No Rx queues configured");
1163 if (!rss->queue_num)
1164 return rte_flow_error_set(error, EINVAL,
1165 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1166 NULL, "No queues configured");
1167 for (i = 0; i != rss->queue_num; ++i) {
1168 if (rss->queue[i] >= priv->rxqs_n)
1169 return rte_flow_error_set
1171 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1172 &rss->queue[i], "queue index out of range");
1173 if (!(*priv->rxqs)[rss->queue[i]])
1174 return rte_flow_error_set
1175 (error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1176 &rss->queue[i], "queue is not configured");
1179 return rte_flow_error_set(error, ENOTSUP,
1180 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1181 "rss action not supported for "
1183 if (rss->level > 1 && !tunnel)
1184 return rte_flow_error_set(error, EINVAL,
1185 RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1186 "inner RSS is not supported for "
1187 "non-tunnel flows");
1192 * Validate the count action.
1195 * Pointer to the Ethernet device structure.
1197 * Attributes of flow that includes this action.
1199 * Pointer to error structure.
1202 * 0 on success, a negative errno value otherwise and rte_errno is set.
1205 mlx5_flow_validate_action_count(struct rte_eth_dev *dev __rte_unused,
1206 const struct rte_flow_attr *attr,
1207 struct rte_flow_error *error)
1210 return rte_flow_error_set(error, ENOTSUP,
1211 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1212 "count action not supported for "
1218 * Verify the @p attributes will be correctly understood by the NIC and store
1219 * them in the @p flow if everything is correct.
1222 * Pointer to the Ethernet device structure.
1223 * @param[in] attributes
1224 * Pointer to flow attributes
1226 * Pointer to error structure.
1229 * 0 on success, a negative errno value otherwise and rte_errno is set.
1232 mlx5_flow_validate_attributes(struct rte_eth_dev *dev,
1233 const struct rte_flow_attr *attributes,
1234 struct rte_flow_error *error)
1236 struct mlx5_priv *priv = dev->data->dev_private;
1237 uint32_t priority_max = priv->config.flow_prio - 1;
1239 if (attributes->group)
1240 return rte_flow_error_set(error, ENOTSUP,
1241 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
1242 NULL, "groups is not supported");
1243 if (attributes->priority != MLX5_FLOW_PRIO_RSVD &&
1244 attributes->priority >= priority_max)
1245 return rte_flow_error_set(error, ENOTSUP,
1246 RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
1247 NULL, "priority out of range");
1248 if (attributes->egress)
1249 return rte_flow_error_set(error, ENOTSUP,
1250 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1251 "egress is not supported");
1252 if (attributes->transfer && !priv->config.dv_esw_en)
1253 return rte_flow_error_set(error, ENOTSUP,
1254 RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1255 NULL, "transfer is not supported");
1256 if (!attributes->ingress)
1257 return rte_flow_error_set(error, EINVAL,
1258 RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
1260 "ingress attribute is mandatory");
1265 * Validate ICMP6 item.
1268 * Item specification.
1269 * @param[in] item_flags
1270 * Bit-fields that holds the items detected until now.
1272 * Pointer to error structure.
1275 * 0 on success, a negative errno value otherwise and rte_errno is set.
1278 mlx5_flow_validate_item_icmp6(const struct rte_flow_item *item,
1279 uint64_t item_flags,
1280 uint8_t target_protocol,
1281 struct rte_flow_error *error)
1283 const struct rte_flow_item_icmp6 *mask = item->mask;
1284 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1285 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
1286 MLX5_FLOW_LAYER_OUTER_L3_IPV6;
1287 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1288 MLX5_FLOW_LAYER_OUTER_L4;
1291 if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMPV6)
1292 return rte_flow_error_set(error, EINVAL,
1293 RTE_FLOW_ERROR_TYPE_ITEM, item,
1294 "protocol filtering not compatible"
1295 " with ICMP6 layer");
1296 if (!(item_flags & l3m))
1297 return rte_flow_error_set(error, EINVAL,
1298 RTE_FLOW_ERROR_TYPE_ITEM, item,
1299 "IPv6 is mandatory to filter on"
1301 if (item_flags & l4m)
1302 return rte_flow_error_set(error, EINVAL,
1303 RTE_FLOW_ERROR_TYPE_ITEM, item,
1304 "multiple L4 layers not supported");
1306 mask = &rte_flow_item_icmp6_mask;
1307 ret = mlx5_flow_item_acceptable
1308 (item, (const uint8_t *)mask,
1309 (const uint8_t *)&rte_flow_item_icmp6_mask,
1310 sizeof(struct rte_flow_item_icmp6), error);
1317 * Validate ICMP item.
1320 * Item specification.
1321 * @param[in] item_flags
1322 * Bit-fields that holds the items detected until now.
1324 * Pointer to error structure.
1327 * 0 on success, a negative errno value otherwise and rte_errno is set.
1330 mlx5_flow_validate_item_icmp(const struct rte_flow_item *item,
1331 uint64_t item_flags,
1332 uint8_t target_protocol,
1333 struct rte_flow_error *error)
1335 const struct rte_flow_item_icmp *mask = item->mask;
1336 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1337 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
1338 MLX5_FLOW_LAYER_OUTER_L3_IPV4;
1339 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1340 MLX5_FLOW_LAYER_OUTER_L4;
1343 if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMP)
1344 return rte_flow_error_set(error, EINVAL,
1345 RTE_FLOW_ERROR_TYPE_ITEM, item,
1346 "protocol filtering not compatible"
1347 " with ICMP layer");
1348 if (!(item_flags & l3m))
1349 return rte_flow_error_set(error, EINVAL,
1350 RTE_FLOW_ERROR_TYPE_ITEM, item,
1351 "IPv4 is mandatory to filter"
1353 if (item_flags & l4m)
1354 return rte_flow_error_set(error, EINVAL,
1355 RTE_FLOW_ERROR_TYPE_ITEM, item,
1356 "multiple L4 layers not supported");
1358 mask = &rte_flow_item_icmp_mask;
1359 ret = mlx5_flow_item_acceptable
1360 (item, (const uint8_t *)mask,
1361 (const uint8_t *)&rte_flow_item_icmp_mask,
1362 sizeof(struct rte_flow_item_icmp), error);
1369 * Validate Ethernet item.
1372 * Item specification.
1373 * @param[in] item_flags
1374 * Bit-fields that holds the items detected until now.
1376 * Pointer to error structure.
1379 * 0 on success, a negative errno value otherwise and rte_errno is set.
1382 mlx5_flow_validate_item_eth(const struct rte_flow_item *item,
1383 uint64_t item_flags,
1384 struct rte_flow_error *error)
1386 const struct rte_flow_item_eth *mask = item->mask;
1387 const struct rte_flow_item_eth nic_mask = {
1388 .dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1389 .src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1390 .type = RTE_BE16(0xffff),
1393 int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1394 const uint64_t ethm = tunnel ? MLX5_FLOW_LAYER_INNER_L2 :
1395 MLX5_FLOW_LAYER_OUTER_L2;
1397 if (item_flags & ethm)
1398 return rte_flow_error_set(error, ENOTSUP,
1399 RTE_FLOW_ERROR_TYPE_ITEM, item,
1400 "multiple L2 layers not supported");
1401 if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_L3)) ||
1402 (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_L3)))
1403 return rte_flow_error_set(error, EINVAL,
1404 RTE_FLOW_ERROR_TYPE_ITEM, item,
1405 "L2 layer should not follow "
1407 if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_VLAN)) ||
1408 (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_VLAN)))
1409 return rte_flow_error_set(error, EINVAL,
1410 RTE_FLOW_ERROR_TYPE_ITEM, item,
1411 "L2 layer should not follow VLAN");
1413 mask = &rte_flow_item_eth_mask;
1414 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1415 (const uint8_t *)&nic_mask,
1416 sizeof(struct rte_flow_item_eth),
1422 * Validate VLAN item.
1425 * Item specification.
1426 * @param[in] item_flags
1427 * Bit-fields that holds the items detected until now.
1429 * Ethernet device flow is being created on.
1431 * Pointer to error structure.
1434 * 0 on success, a negative errno value otherwise and rte_errno is set.
1437 mlx5_flow_validate_item_vlan(const struct rte_flow_item *item,
1438 uint64_t item_flags,
1439 struct rte_eth_dev *dev,
1440 struct rte_flow_error *error)
1442 const struct rte_flow_item_vlan *spec = item->spec;
1443 const struct rte_flow_item_vlan *mask = item->mask;
1444 const struct rte_flow_item_vlan nic_mask = {
1445 .tci = RTE_BE16(UINT16_MAX),
1446 .inner_type = RTE_BE16(UINT16_MAX),
1448 uint16_t vlan_tag = 0;
1449 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1451 const uint64_t l34m = tunnel ? (MLX5_FLOW_LAYER_INNER_L3 |
1452 MLX5_FLOW_LAYER_INNER_L4) :
1453 (MLX5_FLOW_LAYER_OUTER_L3 |
1454 MLX5_FLOW_LAYER_OUTER_L4);
1455 const uint64_t vlanm = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
1456 MLX5_FLOW_LAYER_OUTER_VLAN;
1458 if (item_flags & vlanm)
1459 return rte_flow_error_set(error, EINVAL,
1460 RTE_FLOW_ERROR_TYPE_ITEM, item,
1461 "multiple VLAN layers not supported");
1462 else if ((item_flags & l34m) != 0)
1463 return rte_flow_error_set(error, EINVAL,
1464 RTE_FLOW_ERROR_TYPE_ITEM, item,
1465 "VLAN cannot follow L3/L4 layer");
1467 mask = &rte_flow_item_vlan_mask;
1468 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1469 (const uint8_t *)&nic_mask,
1470 sizeof(struct rte_flow_item_vlan),
1474 if (!tunnel && mask->tci != RTE_BE16(0x0fff)) {
1475 struct mlx5_priv *priv = dev->data->dev_private;
1477 if (priv->vmwa_context) {
1479 * Non-NULL context means we have a virtual machine
1480 * and SR-IOV enabled, we have to create VLAN interface
1481 * to make hypervisor to setup E-Switch vport
1482 * context correctly. We avoid creating the multiple
1483 * VLAN interfaces, so we cannot support VLAN tag mask.
1485 return rte_flow_error_set(error, EINVAL,
1486 RTE_FLOW_ERROR_TYPE_ITEM,
1488 "VLAN tag mask is not"
1489 " supported in virtual"
1494 vlan_tag = spec->tci;
1495 vlan_tag &= mask->tci;
1498 * From verbs perspective an empty VLAN is equivalent
1499 * to a packet without VLAN layer.
1502 return rte_flow_error_set(error, EINVAL,
1503 RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
1505 "VLAN cannot be empty");
1510 * Validate IPV4 item.
1513 * Item specification.
1514 * @param[in] item_flags
1515 * Bit-fields that holds the items detected until now.
1516 * @param[in] acc_mask
1517 * Acceptable mask, if NULL default internal default mask
1518 * will be used to check whether item fields are supported.
1520 * Pointer to error structure.
1523 * 0 on success, a negative errno value otherwise and rte_errno is set.
1526 mlx5_flow_validate_item_ipv4(const struct rte_flow_item *item,
1527 uint64_t item_flags,
1529 uint16_t ether_type,
1530 const struct rte_flow_item_ipv4 *acc_mask,
1531 struct rte_flow_error *error)
1533 const struct rte_flow_item_ipv4 *mask = item->mask;
1534 const struct rte_flow_item_ipv4 *spec = item->spec;
1535 const struct rte_flow_item_ipv4 nic_mask = {
1537 .src_addr = RTE_BE32(0xffffffff),
1538 .dst_addr = RTE_BE32(0xffffffff),
1539 .type_of_service = 0xff,
1540 .next_proto_id = 0xff,
1543 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1544 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1545 MLX5_FLOW_LAYER_OUTER_L3;
1546 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1547 MLX5_FLOW_LAYER_OUTER_L4;
1549 uint8_t next_proto = 0xFF;
1550 const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
1551 MLX5_FLOW_LAYER_OUTER_VLAN |
1552 MLX5_FLOW_LAYER_INNER_VLAN);
1554 if ((last_item & l2_vlan) && ether_type &&
1555 ether_type != RTE_ETHER_TYPE_IPV4)
1556 return rte_flow_error_set(error, EINVAL,
1557 RTE_FLOW_ERROR_TYPE_ITEM, item,
1558 "IPv4 cannot follow L2/VLAN layer "
1559 "which ether type is not IPv4");
1560 if (item_flags & MLX5_FLOW_LAYER_IPIP) {
1562 next_proto = mask->hdr.next_proto_id &
1563 spec->hdr.next_proto_id;
1564 if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
1565 return rte_flow_error_set(error, EINVAL,
1566 RTE_FLOW_ERROR_TYPE_ITEM,
1571 if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP)
1572 return rte_flow_error_set(error, EINVAL,
1573 RTE_FLOW_ERROR_TYPE_ITEM, item,
1574 "wrong tunnel type - IPv6 specified "
1575 "but IPv4 item provided");
1576 if (item_flags & l3m)
1577 return rte_flow_error_set(error, ENOTSUP,
1578 RTE_FLOW_ERROR_TYPE_ITEM, item,
1579 "multiple L3 layers not supported");
1580 else if (item_flags & l4m)
1581 return rte_flow_error_set(error, EINVAL,
1582 RTE_FLOW_ERROR_TYPE_ITEM, item,
1583 "L3 cannot follow an L4 layer.");
1584 else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
1585 !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
1586 return rte_flow_error_set(error, EINVAL,
1587 RTE_FLOW_ERROR_TYPE_ITEM, item,
1588 "L3 cannot follow an NVGRE layer.");
1590 mask = &rte_flow_item_ipv4_mask;
1591 else if (mask->hdr.next_proto_id != 0 &&
1592 mask->hdr.next_proto_id != 0xff)
1593 return rte_flow_error_set(error, EINVAL,
1594 RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
1595 "partial mask is not supported"
1597 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1598 acc_mask ? (const uint8_t *)acc_mask
1599 : (const uint8_t *)&nic_mask,
1600 sizeof(struct rte_flow_item_ipv4),
1608 * Validate IPV6 item.
1611 * Item specification.
1612 * @param[in] item_flags
1613 * Bit-fields that holds the items detected until now.
1614 * @param[in] acc_mask
1615 * Acceptable mask, if NULL default internal default mask
1616 * will be used to check whether item fields are supported.
1618 * Pointer to error structure.
1621 * 0 on success, a negative errno value otherwise and rte_errno is set.
1624 mlx5_flow_validate_item_ipv6(const struct rte_flow_item *item,
1625 uint64_t item_flags,
1627 uint16_t ether_type,
1628 const struct rte_flow_item_ipv6 *acc_mask,
1629 struct rte_flow_error *error)
1631 const struct rte_flow_item_ipv6 *mask = item->mask;
1632 const struct rte_flow_item_ipv6 *spec = item->spec;
1633 const struct rte_flow_item_ipv6 nic_mask = {
1636 "\xff\xff\xff\xff\xff\xff\xff\xff"
1637 "\xff\xff\xff\xff\xff\xff\xff\xff",
1639 "\xff\xff\xff\xff\xff\xff\xff\xff"
1640 "\xff\xff\xff\xff\xff\xff\xff\xff",
1641 .vtc_flow = RTE_BE32(0xffffffff),
1646 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1647 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1648 MLX5_FLOW_LAYER_OUTER_L3;
1649 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1650 MLX5_FLOW_LAYER_OUTER_L4;
1652 uint8_t next_proto = 0xFF;
1653 const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
1654 MLX5_FLOW_LAYER_OUTER_VLAN |
1655 MLX5_FLOW_LAYER_INNER_VLAN);
1657 if ((last_item & l2_vlan) && ether_type &&
1658 ether_type != RTE_ETHER_TYPE_IPV6)
1659 return rte_flow_error_set(error, EINVAL,
1660 RTE_FLOW_ERROR_TYPE_ITEM, item,
1661 "IPv6 cannot follow L2/VLAN layer "
1662 "which ether type is not IPv6");
1663 if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP) {
1665 next_proto = mask->hdr.proto & spec->hdr.proto;
1666 if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
1667 return rte_flow_error_set(error, EINVAL,
1668 RTE_FLOW_ERROR_TYPE_ITEM,
1673 if (item_flags & MLX5_FLOW_LAYER_IPIP)
1674 return rte_flow_error_set(error, EINVAL,
1675 RTE_FLOW_ERROR_TYPE_ITEM, item,
1676 "wrong tunnel type - IPv4 specified "
1677 "but IPv6 item provided");
1678 if (item_flags & l3m)
1679 return rte_flow_error_set(error, ENOTSUP,
1680 RTE_FLOW_ERROR_TYPE_ITEM, item,
1681 "multiple L3 layers not supported");
1682 else if (item_flags & l4m)
1683 return rte_flow_error_set(error, EINVAL,
1684 RTE_FLOW_ERROR_TYPE_ITEM, item,
1685 "L3 cannot follow an L4 layer.");
1686 else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
1687 !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
1688 return rte_flow_error_set(error, EINVAL,
1689 RTE_FLOW_ERROR_TYPE_ITEM, item,
1690 "L3 cannot follow an NVGRE layer.");
1692 mask = &rte_flow_item_ipv6_mask;
1693 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1694 acc_mask ? (const uint8_t *)acc_mask
1695 : (const uint8_t *)&nic_mask,
1696 sizeof(struct rte_flow_item_ipv6),
1704 * Validate UDP item.
1707 * Item specification.
1708 * @param[in] item_flags
1709 * Bit-fields that holds the items detected until now.
1710 * @param[in] target_protocol
1711 * The next protocol in the previous item.
1712 * @param[in] flow_mask
1713 * mlx5 flow-specific (DV, verbs, etc.) supported header fields mask.
1715 * Pointer to error structure.
1718 * 0 on success, a negative errno value otherwise and rte_errno is set.
1721 mlx5_flow_validate_item_udp(const struct rte_flow_item *item,
1722 uint64_t item_flags,
1723 uint8_t target_protocol,
1724 struct rte_flow_error *error)
1726 const struct rte_flow_item_udp *mask = item->mask;
1727 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1728 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1729 MLX5_FLOW_LAYER_OUTER_L3;
1730 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1731 MLX5_FLOW_LAYER_OUTER_L4;
1734 if (target_protocol != 0xff && target_protocol != IPPROTO_UDP)
1735 return rte_flow_error_set(error, EINVAL,
1736 RTE_FLOW_ERROR_TYPE_ITEM, item,
1737 "protocol filtering not compatible"
1739 if (!(item_flags & l3m))
1740 return rte_flow_error_set(error, EINVAL,
1741 RTE_FLOW_ERROR_TYPE_ITEM, item,
1742 "L3 is mandatory to filter on L4");
1743 if (item_flags & l4m)
1744 return rte_flow_error_set(error, EINVAL,
1745 RTE_FLOW_ERROR_TYPE_ITEM, item,
1746 "multiple L4 layers not supported");
1748 mask = &rte_flow_item_udp_mask;
1749 ret = mlx5_flow_item_acceptable
1750 (item, (const uint8_t *)mask,
1751 (const uint8_t *)&rte_flow_item_udp_mask,
1752 sizeof(struct rte_flow_item_udp), error);
1759 * Validate TCP item.
1762 * Item specification.
1763 * @param[in] item_flags
1764 * Bit-fields that holds the items detected until now.
1765 * @param[in] target_protocol
1766 * The next protocol in the previous item.
1768 * Pointer to error structure.
1771 * 0 on success, a negative errno value otherwise and rte_errno is set.
1774 mlx5_flow_validate_item_tcp(const struct rte_flow_item *item,
1775 uint64_t item_flags,
1776 uint8_t target_protocol,
1777 const struct rte_flow_item_tcp *flow_mask,
1778 struct rte_flow_error *error)
1780 const struct rte_flow_item_tcp *mask = item->mask;
1781 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1782 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1783 MLX5_FLOW_LAYER_OUTER_L3;
1784 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1785 MLX5_FLOW_LAYER_OUTER_L4;
1788 MLX5_ASSERT(flow_mask);
1789 if (target_protocol != 0xff && target_protocol != IPPROTO_TCP)
1790 return rte_flow_error_set(error, EINVAL,
1791 RTE_FLOW_ERROR_TYPE_ITEM, item,
1792 "protocol filtering not compatible"
1794 if (!(item_flags & l3m))
1795 return rte_flow_error_set(error, EINVAL,
1796 RTE_FLOW_ERROR_TYPE_ITEM, item,
1797 "L3 is mandatory to filter on L4");
1798 if (item_flags & l4m)
1799 return rte_flow_error_set(error, EINVAL,
1800 RTE_FLOW_ERROR_TYPE_ITEM, item,
1801 "multiple L4 layers not supported");
1803 mask = &rte_flow_item_tcp_mask;
1804 ret = mlx5_flow_item_acceptable
1805 (item, (const uint8_t *)mask,
1806 (const uint8_t *)flow_mask,
1807 sizeof(struct rte_flow_item_tcp), error);
1814 * Validate VXLAN item.
1817 * Item specification.
1818 * @param[in] item_flags
1819 * Bit-fields that holds the items detected until now.
1820 * @param[in] target_protocol
1821 * The next protocol in the previous item.
1823 * Pointer to error structure.
1826 * 0 on success, a negative errno value otherwise and rte_errno is set.
1829 mlx5_flow_validate_item_vxlan(const struct rte_flow_item *item,
1830 uint64_t item_flags,
1831 struct rte_flow_error *error)
1833 const struct rte_flow_item_vxlan *spec = item->spec;
1834 const struct rte_flow_item_vxlan *mask = item->mask;
1839 } id = { .vlan_id = 0, };
1840 uint32_t vlan_id = 0;
1843 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1844 return rte_flow_error_set(error, ENOTSUP,
1845 RTE_FLOW_ERROR_TYPE_ITEM, item,
1846 "multiple tunnel layers not"
1849 * Verify only UDPv4 is present as defined in
1850 * https://tools.ietf.org/html/rfc7348
1852 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
1853 return rte_flow_error_set(error, EINVAL,
1854 RTE_FLOW_ERROR_TYPE_ITEM, item,
1855 "no outer UDP layer found");
1857 mask = &rte_flow_item_vxlan_mask;
1858 ret = mlx5_flow_item_acceptable
1859 (item, (const uint8_t *)mask,
1860 (const uint8_t *)&rte_flow_item_vxlan_mask,
1861 sizeof(struct rte_flow_item_vxlan),
1866 memcpy(&id.vni[1], spec->vni, 3);
1867 vlan_id = id.vlan_id;
1868 memcpy(&id.vni[1], mask->vni, 3);
1869 vlan_id &= id.vlan_id;
1872 * Tunnel id 0 is equivalent as not adding a VXLAN layer, if
1873 * only this layer is defined in the Verbs specification it is
1874 * interpreted as wildcard and all packets will match this
1875 * rule, if it follows a full stack layer (ex: eth / ipv4 /
1876 * udp), all packets matching the layers before will also
1877 * match this rule. To avoid such situation, VNI 0 is
1878 * currently refused.
1881 return rte_flow_error_set(error, ENOTSUP,
1882 RTE_FLOW_ERROR_TYPE_ITEM, item,
1883 "VXLAN vni cannot be 0");
1884 if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
1885 return rte_flow_error_set(error, ENOTSUP,
1886 RTE_FLOW_ERROR_TYPE_ITEM, item,
1887 "VXLAN tunnel must be fully defined");
1892 * Validate VXLAN_GPE item.
1895 * Item specification.
1896 * @param[in] item_flags
1897 * Bit-fields that holds the items detected until now.
1899 * Pointer to the private data structure.
1900 * @param[in] target_protocol
1901 * The next protocol in the previous item.
1903 * Pointer to error structure.
1906 * 0 on success, a negative errno value otherwise and rte_errno is set.
1909 mlx5_flow_validate_item_vxlan_gpe(const struct rte_flow_item *item,
1910 uint64_t item_flags,
1911 struct rte_eth_dev *dev,
1912 struct rte_flow_error *error)
1914 struct mlx5_priv *priv = dev->data->dev_private;
1915 const struct rte_flow_item_vxlan_gpe *spec = item->spec;
1916 const struct rte_flow_item_vxlan_gpe *mask = item->mask;
1921 } id = { .vlan_id = 0, };
1922 uint32_t vlan_id = 0;
1924 if (!priv->config.l3_vxlan_en)
1925 return rte_flow_error_set(error, ENOTSUP,
1926 RTE_FLOW_ERROR_TYPE_ITEM, item,
1927 "L3 VXLAN is not enabled by device"
1928 " parameter and/or not configured in"
1930 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1931 return rte_flow_error_set(error, ENOTSUP,
1932 RTE_FLOW_ERROR_TYPE_ITEM, item,
1933 "multiple tunnel layers not"
1936 * Verify only UDPv4 is present as defined in
1937 * https://tools.ietf.org/html/rfc7348
1939 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
1940 return rte_flow_error_set(error, EINVAL,
1941 RTE_FLOW_ERROR_TYPE_ITEM, item,
1942 "no outer UDP layer found");
1944 mask = &rte_flow_item_vxlan_gpe_mask;
1945 ret = mlx5_flow_item_acceptable
1946 (item, (const uint8_t *)mask,
1947 (const uint8_t *)&rte_flow_item_vxlan_gpe_mask,
1948 sizeof(struct rte_flow_item_vxlan_gpe),
1954 return rte_flow_error_set(error, ENOTSUP,
1955 RTE_FLOW_ERROR_TYPE_ITEM,
1957 "VxLAN-GPE protocol"
1959 memcpy(&id.vni[1], spec->vni, 3);
1960 vlan_id = id.vlan_id;
1961 memcpy(&id.vni[1], mask->vni, 3);
1962 vlan_id &= id.vlan_id;
1965 * Tunnel id 0 is equivalent as not adding a VXLAN layer, if only this
1966 * layer is defined in the Verbs specification it is interpreted as
1967 * wildcard and all packets will match this rule, if it follows a full
1968 * stack layer (ex: eth / ipv4 / udp), all packets matching the layers
1969 * before will also match this rule. To avoid such situation, VNI 0
1970 * is currently refused.
1973 return rte_flow_error_set(error, ENOTSUP,
1974 RTE_FLOW_ERROR_TYPE_ITEM, item,
1975 "VXLAN-GPE vni cannot be 0");
1976 if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
1977 return rte_flow_error_set(error, ENOTSUP,
1978 RTE_FLOW_ERROR_TYPE_ITEM, item,
1979 "VXLAN-GPE tunnel must be fully"
1984 * Validate GRE Key item.
1987 * Item specification.
1988 * @param[in] item_flags
1989 * Bit flags to mark detected items.
1990 * @param[in] gre_item
1991 * Pointer to gre_item
1993 * Pointer to error structure.
1996 * 0 on success, a negative errno value otherwise and rte_errno is set.
1999 mlx5_flow_validate_item_gre_key(const struct rte_flow_item *item,
2000 uint64_t item_flags,
2001 const struct rte_flow_item *gre_item,
2002 struct rte_flow_error *error)
2004 const rte_be32_t *mask = item->mask;
2006 rte_be32_t gre_key_default_mask = RTE_BE32(UINT32_MAX);
2007 const struct rte_flow_item_gre *gre_spec;
2008 const struct rte_flow_item_gre *gre_mask;
2010 if (item_flags & MLX5_FLOW_LAYER_GRE_KEY)
2011 return rte_flow_error_set(error, ENOTSUP,
2012 RTE_FLOW_ERROR_TYPE_ITEM, item,
2013 "Multiple GRE key not support");
2014 if (!(item_flags & MLX5_FLOW_LAYER_GRE))
2015 return rte_flow_error_set(error, ENOTSUP,
2016 RTE_FLOW_ERROR_TYPE_ITEM, item,
2017 "No preceding GRE header");
2018 if (item_flags & MLX5_FLOW_LAYER_INNER)
2019 return rte_flow_error_set(error, ENOTSUP,
2020 RTE_FLOW_ERROR_TYPE_ITEM, item,
2021 "GRE key following a wrong item");
2022 gre_mask = gre_item->mask;
2024 gre_mask = &rte_flow_item_gre_mask;
2025 gre_spec = gre_item->spec;
2026 if (gre_spec && (gre_mask->c_rsvd0_ver & RTE_BE16(0x2000)) &&
2027 !(gre_spec->c_rsvd0_ver & RTE_BE16(0x2000)))
2028 return rte_flow_error_set(error, EINVAL,
2029 RTE_FLOW_ERROR_TYPE_ITEM, item,
2030 "Key bit must be on");
2033 mask = &gre_key_default_mask;
2034 ret = mlx5_flow_item_acceptable
2035 (item, (const uint8_t *)mask,
2036 (const uint8_t *)&gre_key_default_mask,
2037 sizeof(rte_be32_t), error);
2042 * Validate GRE item.
2045 * Item specification.
2046 * @param[in] item_flags
2047 * Bit flags to mark detected items.
2048 * @param[in] target_protocol
2049 * The next protocol in the previous item.
2051 * Pointer to error structure.
2054 * 0 on success, a negative errno value otherwise and rte_errno is set.
2057 mlx5_flow_validate_item_gre(const struct rte_flow_item *item,
2058 uint64_t item_flags,
2059 uint8_t target_protocol,
2060 struct rte_flow_error *error)
2062 const struct rte_flow_item_gre *spec __rte_unused = item->spec;
2063 const struct rte_flow_item_gre *mask = item->mask;
2065 const struct rte_flow_item_gre nic_mask = {
2066 .c_rsvd0_ver = RTE_BE16(0xB000),
2067 .protocol = RTE_BE16(UINT16_MAX),
2070 if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2071 return rte_flow_error_set(error, EINVAL,
2072 RTE_FLOW_ERROR_TYPE_ITEM, item,
2073 "protocol filtering not compatible"
2074 " with this GRE layer");
2075 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2076 return rte_flow_error_set(error, ENOTSUP,
2077 RTE_FLOW_ERROR_TYPE_ITEM, item,
2078 "multiple tunnel layers not"
2080 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2081 return rte_flow_error_set(error, ENOTSUP,
2082 RTE_FLOW_ERROR_TYPE_ITEM, item,
2083 "L3 Layer is missing");
2085 mask = &rte_flow_item_gre_mask;
2086 ret = mlx5_flow_item_acceptable
2087 (item, (const uint8_t *)mask,
2088 (const uint8_t *)&nic_mask,
2089 sizeof(struct rte_flow_item_gre), error);
2092 #ifndef HAVE_MLX5DV_DR
2093 #ifndef HAVE_IBV_DEVICE_MPLS_SUPPORT
2094 if (spec && (spec->protocol & mask->protocol))
2095 return rte_flow_error_set(error, ENOTSUP,
2096 RTE_FLOW_ERROR_TYPE_ITEM, item,
2097 "without MPLS support the"
2098 " specification cannot be used for"
2106 * Validate Geneve item.
2109 * Item specification.
2110 * @param[in] itemFlags
2111 * Bit-fields that holds the items detected until now.
2113 * Pointer to the private data structure.
2115 * Pointer to error structure.
2118 * 0 on success, a negative errno value otherwise and rte_errno is set.
2122 mlx5_flow_validate_item_geneve(const struct rte_flow_item *item,
2123 uint64_t item_flags,
2124 struct rte_eth_dev *dev,
2125 struct rte_flow_error *error)
2127 struct mlx5_priv *priv = dev->data->dev_private;
2128 const struct rte_flow_item_geneve *spec = item->spec;
2129 const struct rte_flow_item_geneve *mask = item->mask;
2132 uint8_t opt_len = priv->config.hca_attr.geneve_max_opt_len ?
2133 MLX5_GENEVE_OPT_LEN_1 : MLX5_GENEVE_OPT_LEN_0;
2134 const struct rte_flow_item_geneve nic_mask = {
2135 .ver_opt_len_o_c_rsvd0 = RTE_BE16(0x3f80),
2136 .vni = "\xff\xff\xff",
2137 .protocol = RTE_BE16(UINT16_MAX),
2140 if (!priv->config.hca_attr.tunnel_stateless_geneve_rx)
2141 return rte_flow_error_set(error, ENOTSUP,
2142 RTE_FLOW_ERROR_TYPE_ITEM, item,
2143 "L3 Geneve is not enabled by device"
2144 " parameter and/or not configured in"
2146 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2147 return rte_flow_error_set(error, ENOTSUP,
2148 RTE_FLOW_ERROR_TYPE_ITEM, item,
2149 "multiple tunnel layers not"
2152 * Verify only UDPv4 is present as defined in
2153 * https://tools.ietf.org/html/rfc7348
2155 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2156 return rte_flow_error_set(error, EINVAL,
2157 RTE_FLOW_ERROR_TYPE_ITEM, item,
2158 "no outer UDP layer found");
2160 mask = &rte_flow_item_geneve_mask;
2161 ret = mlx5_flow_item_acceptable
2162 (item, (const uint8_t *)mask,
2163 (const uint8_t *)&nic_mask,
2164 sizeof(struct rte_flow_item_geneve), error);
2168 gbhdr = rte_be_to_cpu_16(spec->ver_opt_len_o_c_rsvd0);
2169 if (MLX5_GENEVE_VER_VAL(gbhdr) ||
2170 MLX5_GENEVE_CRITO_VAL(gbhdr) ||
2171 MLX5_GENEVE_RSVD_VAL(gbhdr) || spec->rsvd1)
2172 return rte_flow_error_set(error, ENOTSUP,
2173 RTE_FLOW_ERROR_TYPE_ITEM,
2175 "Geneve protocol unsupported"
2176 " fields are being used");
2177 if (MLX5_GENEVE_OPTLEN_VAL(gbhdr) > opt_len)
2178 return rte_flow_error_set
2180 RTE_FLOW_ERROR_TYPE_ITEM,
2182 "Unsupported Geneve options length");
2184 if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
2185 return rte_flow_error_set
2187 RTE_FLOW_ERROR_TYPE_ITEM, item,
2188 "Geneve tunnel must be fully defined");
2193 * Validate MPLS item.
2196 * Pointer to the rte_eth_dev structure.
2198 * Item specification.
2199 * @param[in] item_flags
2200 * Bit-fields that holds the items detected until now.
2201 * @param[in] prev_layer
2202 * The protocol layer indicated in previous item.
2204 * Pointer to error structure.
2207 * 0 on success, a negative errno value otherwise and rte_errno is set.
2210 mlx5_flow_validate_item_mpls(struct rte_eth_dev *dev __rte_unused,
2211 const struct rte_flow_item *item __rte_unused,
2212 uint64_t item_flags __rte_unused,
2213 uint64_t prev_layer __rte_unused,
2214 struct rte_flow_error *error)
2216 #ifdef HAVE_IBV_DEVICE_MPLS_SUPPORT
2217 const struct rte_flow_item_mpls *mask = item->mask;
2218 struct mlx5_priv *priv = dev->data->dev_private;
2221 if (!priv->config.mpls_en)
2222 return rte_flow_error_set(error, ENOTSUP,
2223 RTE_FLOW_ERROR_TYPE_ITEM, item,
2224 "MPLS not supported or"
2225 " disabled in firmware"
2227 /* MPLS over IP, UDP, GRE is allowed */
2228 if (!(prev_layer & (MLX5_FLOW_LAYER_OUTER_L3 |
2229 MLX5_FLOW_LAYER_OUTER_L4_UDP |
2230 MLX5_FLOW_LAYER_GRE)))
2231 return rte_flow_error_set(error, EINVAL,
2232 RTE_FLOW_ERROR_TYPE_ITEM, item,
2233 "protocol filtering not compatible"
2234 " with MPLS layer");
2235 /* Multi-tunnel isn't allowed but MPLS over GRE is an exception. */
2236 if ((item_flags & MLX5_FLOW_LAYER_TUNNEL) &&
2237 !(item_flags & MLX5_FLOW_LAYER_GRE))
2238 return rte_flow_error_set(error, ENOTSUP,
2239 RTE_FLOW_ERROR_TYPE_ITEM, item,
2240 "multiple tunnel layers not"
2243 mask = &rte_flow_item_mpls_mask;
2244 ret = mlx5_flow_item_acceptable
2245 (item, (const uint8_t *)mask,
2246 (const uint8_t *)&rte_flow_item_mpls_mask,
2247 sizeof(struct rte_flow_item_mpls), error);
2252 return rte_flow_error_set(error, ENOTSUP,
2253 RTE_FLOW_ERROR_TYPE_ITEM, item,
2254 "MPLS is not supported by Verbs, please"
2259 * Validate NVGRE item.
2262 * Item specification.
2263 * @param[in] item_flags
2264 * Bit flags to mark detected items.
2265 * @param[in] target_protocol
2266 * The next protocol in the previous item.
2268 * Pointer to error structure.
2271 * 0 on success, a negative errno value otherwise and rte_errno is set.
2274 mlx5_flow_validate_item_nvgre(const struct rte_flow_item *item,
2275 uint64_t item_flags,
2276 uint8_t target_protocol,
2277 struct rte_flow_error *error)
2279 const struct rte_flow_item_nvgre *mask = item->mask;
2282 if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2283 return rte_flow_error_set(error, EINVAL,
2284 RTE_FLOW_ERROR_TYPE_ITEM, item,
2285 "protocol filtering not compatible"
2286 " with this GRE layer");
2287 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2288 return rte_flow_error_set(error, ENOTSUP,
2289 RTE_FLOW_ERROR_TYPE_ITEM, item,
2290 "multiple tunnel layers not"
2292 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2293 return rte_flow_error_set(error, ENOTSUP,
2294 RTE_FLOW_ERROR_TYPE_ITEM, item,
2295 "L3 Layer is missing");
2297 mask = &rte_flow_item_nvgre_mask;
2298 ret = mlx5_flow_item_acceptable
2299 (item, (const uint8_t *)mask,
2300 (const uint8_t *)&rte_flow_item_nvgre_mask,
2301 sizeof(struct rte_flow_item_nvgre), error);
2307 /* Allocate unique ID for the split Q/RSS subflows. */
2309 flow_qrss_get_id(struct rte_eth_dev *dev)
2311 struct mlx5_priv *priv = dev->data->dev_private;
2312 uint32_t qrss_id, ret;
2314 ret = mlx5_flow_id_get(priv->qrss_id_pool, &qrss_id);
2317 MLX5_ASSERT(qrss_id);
2321 /* Free unique ID for the split Q/RSS subflows. */
2323 flow_qrss_free_id(struct rte_eth_dev *dev, uint32_t qrss_id)
2325 struct mlx5_priv *priv = dev->data->dev_private;
2328 mlx5_flow_id_release(priv->qrss_id_pool, qrss_id);
2332 * Release resource related QUEUE/RSS action split.
2335 * Pointer to Ethernet device.
2337 * Flow to release id's from.
2340 flow_mreg_split_qrss_release(struct rte_eth_dev *dev,
2341 struct rte_flow *flow)
2343 struct mlx5_flow *dev_flow;
2345 LIST_FOREACH(dev_flow, &flow->dev_flows, next)
2346 if (dev_flow->qrss_id)
2347 flow_qrss_free_id(dev, dev_flow->qrss_id);
2351 flow_null_validate(struct rte_eth_dev *dev __rte_unused,
2352 const struct rte_flow_attr *attr __rte_unused,
2353 const struct rte_flow_item items[] __rte_unused,
2354 const struct rte_flow_action actions[] __rte_unused,
2355 bool external __rte_unused,
2356 struct rte_flow_error *error)
2358 return rte_flow_error_set(error, ENOTSUP,
2359 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2362 static struct mlx5_flow *
2363 flow_null_prepare(const struct rte_flow_attr *attr __rte_unused,
2364 const struct rte_flow_item items[] __rte_unused,
2365 const struct rte_flow_action actions[] __rte_unused,
2366 struct rte_flow_error *error)
2368 rte_flow_error_set(error, ENOTSUP,
2369 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2374 flow_null_translate(struct rte_eth_dev *dev __rte_unused,
2375 struct mlx5_flow *dev_flow __rte_unused,
2376 const struct rte_flow_attr *attr __rte_unused,
2377 const struct rte_flow_item items[] __rte_unused,
2378 const struct rte_flow_action actions[] __rte_unused,
2379 struct rte_flow_error *error)
2381 return rte_flow_error_set(error, ENOTSUP,
2382 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2386 flow_null_apply(struct rte_eth_dev *dev __rte_unused,
2387 struct rte_flow *flow __rte_unused,
2388 struct rte_flow_error *error)
2390 return rte_flow_error_set(error, ENOTSUP,
2391 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2395 flow_null_remove(struct rte_eth_dev *dev __rte_unused,
2396 struct rte_flow *flow __rte_unused)
2401 flow_null_destroy(struct rte_eth_dev *dev __rte_unused,
2402 struct rte_flow *flow __rte_unused)
2407 flow_null_query(struct rte_eth_dev *dev __rte_unused,
2408 struct rte_flow *flow __rte_unused,
2409 const struct rte_flow_action *actions __rte_unused,
2410 void *data __rte_unused,
2411 struct rte_flow_error *error)
2413 return rte_flow_error_set(error, ENOTSUP,
2414 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL, NULL);
2417 /* Void driver to protect from null pointer reference. */
2418 const struct mlx5_flow_driver_ops mlx5_flow_null_drv_ops = {
2419 .validate = flow_null_validate,
2420 .prepare = flow_null_prepare,
2421 .translate = flow_null_translate,
2422 .apply = flow_null_apply,
2423 .remove = flow_null_remove,
2424 .destroy = flow_null_destroy,
2425 .query = flow_null_query,
2429 * Select flow driver type according to flow attributes and device
2433 * Pointer to the dev structure.
2435 * Pointer to the flow attributes.
2438 * flow driver type, MLX5_FLOW_TYPE_MAX otherwise.
2440 static enum mlx5_flow_drv_type
2441 flow_get_drv_type(struct rte_eth_dev *dev, const struct rte_flow_attr *attr)
2443 struct mlx5_priv *priv = dev->data->dev_private;
2444 enum mlx5_flow_drv_type type = MLX5_FLOW_TYPE_MAX;
2446 if (attr->transfer && priv->config.dv_esw_en)
2447 type = MLX5_FLOW_TYPE_DV;
2448 if (!attr->transfer)
2449 type = priv->config.dv_flow_en ? MLX5_FLOW_TYPE_DV :
2450 MLX5_FLOW_TYPE_VERBS;
2454 #define flow_get_drv_ops(type) flow_drv_ops[type]
2457 * Flow driver validation API. This abstracts calling driver specific functions.
2458 * The type of flow driver is determined according to flow attributes.
2461 * Pointer to the dev structure.
2463 * Pointer to the flow attributes.
2465 * Pointer to the list of items.
2466 * @param[in] actions
2467 * Pointer to the list of actions.
2468 * @param[in] external
2469 * This flow rule is created by request external to PMD.
2471 * Pointer to the error structure.
2474 * 0 on success, a negative errno value otherwise and rte_errno is set.
2477 flow_drv_validate(struct rte_eth_dev *dev,
2478 const struct rte_flow_attr *attr,
2479 const struct rte_flow_item items[],
2480 const struct rte_flow_action actions[],
2481 bool external, struct rte_flow_error *error)
2483 const struct mlx5_flow_driver_ops *fops;
2484 enum mlx5_flow_drv_type type = flow_get_drv_type(dev, attr);
2486 fops = flow_get_drv_ops(type);
2487 return fops->validate(dev, attr, items, actions, external, error);
2491 * Flow driver preparation API. This abstracts calling driver specific
2492 * functions. Parent flow (rte_flow) should have driver type (drv_type). It
2493 * calculates the size of memory required for device flow, allocates the memory,
2494 * initializes the device flow and returns the pointer.
2497 * This function initializes device flow structure such as dv or verbs in
2498 * struct mlx5_flow. However, it is caller's responsibility to initialize the
2499 * rest. For example, adding returning device flow to flow->dev_flow list and
2500 * setting backward reference to the flow should be done out of this function.
2501 * layers field is not filled either.
2504 * Pointer to the flow attributes.
2506 * Pointer to the list of items.
2507 * @param[in] actions
2508 * Pointer to the list of actions.
2510 * Pointer to the error structure.
2513 * Pointer to device flow on success, otherwise NULL and rte_errno is set.
2515 static inline struct mlx5_flow *
2516 flow_drv_prepare(const struct rte_flow *flow,
2517 const struct rte_flow_attr *attr,
2518 const struct rte_flow_item items[],
2519 const struct rte_flow_action actions[],
2520 struct rte_flow_error *error)
2522 const struct mlx5_flow_driver_ops *fops;
2523 enum mlx5_flow_drv_type type = flow->drv_type;
2525 MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2526 fops = flow_get_drv_ops(type);
2527 return fops->prepare(attr, items, actions, error);
2531 * Flow driver translation API. This abstracts calling driver specific
2532 * functions. Parent flow (rte_flow) should have driver type (drv_type). It
2533 * translates a generic flow into a driver flow. flow_drv_prepare() must
2537 * dev_flow->layers could be filled as a result of parsing during translation
2538 * if needed by flow_drv_apply(). dev_flow->flow->actions can also be filled
2539 * if necessary. As a flow can have multiple dev_flows by RSS flow expansion,
2540 * flow->actions could be overwritten even though all the expanded dev_flows
2541 * have the same actions.
2544 * Pointer to the rte dev structure.
2545 * @param[in, out] dev_flow
2546 * Pointer to the mlx5 flow.
2548 * Pointer to the flow attributes.
2550 * Pointer to the list of items.
2551 * @param[in] actions
2552 * Pointer to the list of actions.
2554 * Pointer to the error structure.
2557 * 0 on success, a negative errno value otherwise and rte_errno is set.
2560 flow_drv_translate(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow,
2561 const struct rte_flow_attr *attr,
2562 const struct rte_flow_item items[],
2563 const struct rte_flow_action actions[],
2564 struct rte_flow_error *error)
2566 const struct mlx5_flow_driver_ops *fops;
2567 enum mlx5_flow_drv_type type = dev_flow->flow->drv_type;
2569 MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2570 fops = flow_get_drv_ops(type);
2571 return fops->translate(dev, dev_flow, attr, items, actions, error);
2575 * Flow driver apply API. This abstracts calling driver specific functions.
2576 * Parent flow (rte_flow) should have driver type (drv_type). It applies
2577 * translated driver flows on to device. flow_drv_translate() must precede.
2580 * Pointer to Ethernet device structure.
2581 * @param[in, out] flow
2582 * Pointer to flow structure.
2584 * Pointer to error structure.
2587 * 0 on success, a negative errno value otherwise and rte_errno is set.
2590 flow_drv_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
2591 struct rte_flow_error *error)
2593 const struct mlx5_flow_driver_ops *fops;
2594 enum mlx5_flow_drv_type type = flow->drv_type;
2596 MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2597 fops = flow_get_drv_ops(type);
2598 return fops->apply(dev, flow, error);
2602 * Flow driver remove API. This abstracts calling driver specific functions.
2603 * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
2604 * on device. All the resources of the flow should be freed by calling
2605 * flow_drv_destroy().
2608 * Pointer to Ethernet device.
2609 * @param[in, out] flow
2610 * Pointer to flow structure.
2613 flow_drv_remove(struct rte_eth_dev *dev, struct rte_flow *flow)
2615 const struct mlx5_flow_driver_ops *fops;
2616 enum mlx5_flow_drv_type type = flow->drv_type;
2618 MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2619 fops = flow_get_drv_ops(type);
2620 fops->remove(dev, flow);
2624 * Flow driver destroy API. This abstracts calling driver specific functions.
2625 * Parent flow (rte_flow) should have driver type (drv_type). It removes a flow
2626 * on device and releases resources of the flow.
2629 * Pointer to Ethernet device.
2630 * @param[in, out] flow
2631 * Pointer to flow structure.
2634 flow_drv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
2636 const struct mlx5_flow_driver_ops *fops;
2637 enum mlx5_flow_drv_type type = flow->drv_type;
2639 flow_mreg_split_qrss_release(dev, flow);
2640 MLX5_ASSERT(type > MLX5_FLOW_TYPE_MIN && type < MLX5_FLOW_TYPE_MAX);
2641 fops = flow_get_drv_ops(type);
2642 fops->destroy(dev, flow);
2646 * Validate a flow supported by the NIC.
2648 * @see rte_flow_validate()
2652 mlx5_flow_validate(struct rte_eth_dev *dev,
2653 const struct rte_flow_attr *attr,
2654 const struct rte_flow_item items[],
2655 const struct rte_flow_action actions[],
2656 struct rte_flow_error *error)
2660 ret = flow_drv_validate(dev, attr, items, actions, true, error);
2667 * Get RSS action from the action list.
2669 * @param[in] actions
2670 * Pointer to the list of actions.
2673 * Pointer to the RSS action if exist, else return NULL.
2675 static const struct rte_flow_action_rss*
2676 flow_get_rss_action(const struct rte_flow_action actions[])
2678 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2679 switch (actions->type) {
2680 case RTE_FLOW_ACTION_TYPE_RSS:
2681 return (const struct rte_flow_action_rss *)
2691 find_graph_root(const struct rte_flow_item pattern[], uint32_t rss_level)
2693 const struct rte_flow_item *item;
2694 unsigned int has_vlan = 0;
2696 for (item = pattern; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
2697 if (item->type == RTE_FLOW_ITEM_TYPE_VLAN) {
2703 return rss_level < 2 ? MLX5_EXPANSION_ROOT_ETH_VLAN :
2704 MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN;
2705 return rss_level < 2 ? MLX5_EXPANSION_ROOT :
2706 MLX5_EXPANSION_ROOT_OUTER;
2710 * Get layer flags from the prefix flow.
2712 * Some flows may be split to several subflows, the prefix subflow gets the
2713 * match items and the suffix sub flow gets the actions.
2714 * Some actions need the user defined match item flags to get the detail for
2716 * This function helps the suffix flow to get the item layer flags from prefix
2719 * @param[in] dev_flow
2720 * Pointer the created preifx subflow.
2723 * The layers get from prefix subflow.
2725 static inline uint64_t
2726 flow_get_prefix_layer_flags(struct mlx5_flow *dev_flow)
2728 uint64_t layers = 0;
2730 /* If no decap actions, use the layers directly. */
2731 if (!(dev_flow->actions & MLX5_FLOW_ACTION_DECAP))
2732 return dev_flow->layers;
2733 /* Convert L3 layers with decap action. */
2734 if (dev_flow->layers & MLX5_FLOW_LAYER_INNER_L3_IPV4)
2735 layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV4;
2736 else if (dev_flow->layers & MLX5_FLOW_LAYER_INNER_L3_IPV6)
2737 layers |= MLX5_FLOW_LAYER_OUTER_L3_IPV6;
2738 /* Convert L4 layers with decap action. */
2739 if (dev_flow->layers & MLX5_FLOW_LAYER_INNER_L4_TCP)
2740 layers |= MLX5_FLOW_LAYER_OUTER_L4_TCP;
2741 else if (dev_flow->layers & MLX5_FLOW_LAYER_INNER_L4_UDP)
2742 layers |= MLX5_FLOW_LAYER_OUTER_L4_UDP;
2747 * Get metadata split action information.
2749 * @param[in] actions
2750 * Pointer to the list of actions.
2752 * Pointer to the return pointer.
2753 * @param[out] qrss_type
2754 * Pointer to the action type to return. RTE_FLOW_ACTION_TYPE_END is returned
2755 * if no QUEUE/RSS is found.
2756 * @param[out] encap_idx
2757 * Pointer to the index of the encap action if exists, otherwise the last
2761 * Total number of actions.
2764 flow_parse_metadata_split_actions_info(const struct rte_flow_action actions[],
2765 const struct rte_flow_action **qrss,
2768 const struct rte_flow_action_raw_encap *raw_encap;
2770 int raw_decap_idx = -1;
2773 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2774 switch (actions->type) {
2775 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
2776 case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
2777 *encap_idx = actions_n;
2779 case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
2780 raw_decap_idx = actions_n;
2782 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
2783 raw_encap = actions->conf;
2784 if (raw_encap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
2785 *encap_idx = raw_decap_idx != -1 ?
2786 raw_decap_idx : actions_n;
2788 case RTE_FLOW_ACTION_TYPE_QUEUE:
2789 case RTE_FLOW_ACTION_TYPE_RSS:
2797 if (*encap_idx == -1)
2798 *encap_idx = actions_n;
2799 /* Count RTE_FLOW_ACTION_TYPE_END. */
2800 return actions_n + 1;
2804 * Check meter action from the action list.
2806 * @param[in] actions
2807 * Pointer to the list of actions.
2809 * Pointer to the meter exist flag.
2812 * Total number of actions.
2815 flow_check_meter_action(const struct rte_flow_action actions[], uint32_t *mtr)
2821 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2822 switch (actions->type) {
2823 case RTE_FLOW_ACTION_TYPE_METER:
2831 /* Count RTE_FLOW_ACTION_TYPE_END. */
2832 return actions_n + 1;
2836 * Check if the flow should be splited due to hairpin.
2837 * The reason for the split is that in current HW we can't
2838 * support encap on Rx, so if a flow have encap we move it
2842 * Pointer to Ethernet device.
2844 * Flow rule attributes.
2845 * @param[in] actions
2846 * Associated actions (list terminated by the END action).
2849 * > 0 the number of actions and the flow should be split,
2850 * 0 when no split required.
2853 flow_check_hairpin_split(struct rte_eth_dev *dev,
2854 const struct rte_flow_attr *attr,
2855 const struct rte_flow_action actions[])
2857 int queue_action = 0;
2860 const struct rte_flow_action_queue *queue;
2861 const struct rte_flow_action_rss *rss;
2862 const struct rte_flow_action_raw_encap *raw_encap;
2866 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2867 switch (actions->type) {
2868 case RTE_FLOW_ACTION_TYPE_QUEUE:
2869 queue = actions->conf;
2872 if (mlx5_rxq_get_type(dev, queue->index) !=
2873 MLX5_RXQ_TYPE_HAIRPIN)
2878 case RTE_FLOW_ACTION_TYPE_RSS:
2879 rss = actions->conf;
2880 if (rss == NULL || rss->queue_num == 0)
2882 if (mlx5_rxq_get_type(dev, rss->queue[0]) !=
2883 MLX5_RXQ_TYPE_HAIRPIN)
2888 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
2889 case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
2893 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
2894 raw_encap = actions->conf;
2895 if (raw_encap->size >
2896 (sizeof(struct rte_flow_item_eth) +
2897 sizeof(struct rte_flow_item_ipv4)))
2906 if (encap == 1 && queue_action)
2911 /* Declare flow create/destroy prototype in advance. */
2912 static struct rte_flow *
2913 flow_list_create(struct rte_eth_dev *dev, struct mlx5_flows *list,
2914 const struct rte_flow_attr *attr,
2915 const struct rte_flow_item items[],
2916 const struct rte_flow_action actions[],
2917 bool external, struct rte_flow_error *error);
2920 flow_list_destroy(struct rte_eth_dev *dev, struct mlx5_flows *list,
2921 struct rte_flow *flow);
2924 * Add a flow of copying flow metadata registers in RX_CP_TBL.
2926 * As mark_id is unique, if there's already a registered flow for the mark_id,
2927 * return by increasing the reference counter of the resource. Otherwise, create
2928 * the resource (mcp_res) and flow.
2931 * - If ingress port is ANY and reg_c[1] is mark_id,
2932 * flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
2934 * For default flow (zero mark_id), flow is like,
2935 * - If ingress port is ANY,
2936 * reg_b := reg_c[0] and jump to RX_ACT_TBL.
2939 * Pointer to Ethernet device.
2941 * ID of MARK action, zero means default flow for META.
2943 * Perform verbose error reporting if not NULL.
2946 * Associated resource on success, NULL otherwise and rte_errno is set.
2948 static struct mlx5_flow_mreg_copy_resource *
2949 flow_mreg_add_copy_action(struct rte_eth_dev *dev, uint32_t mark_id,
2950 struct rte_flow_error *error)
2952 struct mlx5_priv *priv = dev->data->dev_private;
2953 struct rte_flow_attr attr = {
2954 .group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
2957 struct mlx5_rte_flow_item_tag tag_spec = {
2960 struct rte_flow_item items[] = {
2961 [1] = { .type = RTE_FLOW_ITEM_TYPE_END, },
2963 struct rte_flow_action_mark ftag = {
2966 struct mlx5_flow_action_copy_mreg cp_mreg = {
2970 struct rte_flow_action_jump jump = {
2971 .group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
2973 struct rte_flow_action actions[] = {
2974 [3] = { .type = RTE_FLOW_ACTION_TYPE_END, },
2976 struct mlx5_flow_mreg_copy_resource *mcp_res;
2979 /* Fill the register fileds in the flow. */
2980 ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
2984 ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
2988 /* Check if already registered. */
2989 MLX5_ASSERT(priv->mreg_cp_tbl);
2990 mcp_res = (void *)mlx5_hlist_lookup(priv->mreg_cp_tbl, mark_id);
2992 /* For non-default rule. */
2993 if (mark_id != MLX5_DEFAULT_COPY_ID)
2995 MLX5_ASSERT(mark_id != MLX5_DEFAULT_COPY_ID ||
2996 mcp_res->refcnt == 1);
2999 /* Provide the full width of FLAG specific value. */
3000 if (mark_id == (priv->sh->dv_regc0_mask & MLX5_FLOW_MARK_DEFAULT))
3001 tag_spec.data = MLX5_FLOW_MARK_DEFAULT;
3002 /* Build a new flow. */
3003 if (mark_id != MLX5_DEFAULT_COPY_ID) {
3004 items[0] = (struct rte_flow_item){
3005 .type = MLX5_RTE_FLOW_ITEM_TYPE_TAG,
3008 items[1] = (struct rte_flow_item){
3009 .type = RTE_FLOW_ITEM_TYPE_END,
3011 actions[0] = (struct rte_flow_action){
3012 .type = MLX5_RTE_FLOW_ACTION_TYPE_MARK,
3015 actions[1] = (struct rte_flow_action){
3016 .type = MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3019 actions[2] = (struct rte_flow_action){
3020 .type = RTE_FLOW_ACTION_TYPE_JUMP,
3023 actions[3] = (struct rte_flow_action){
3024 .type = RTE_FLOW_ACTION_TYPE_END,
3027 /* Default rule, wildcard match. */
3028 attr.priority = MLX5_FLOW_PRIO_RSVD;
3029 items[0] = (struct rte_flow_item){
3030 .type = RTE_FLOW_ITEM_TYPE_END,
3032 actions[0] = (struct rte_flow_action){
3033 .type = MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3036 actions[1] = (struct rte_flow_action){
3037 .type = RTE_FLOW_ACTION_TYPE_JUMP,
3040 actions[2] = (struct rte_flow_action){
3041 .type = RTE_FLOW_ACTION_TYPE_END,
3044 /* Build a new entry. */
3045 mcp_res = rte_zmalloc(__func__, sizeof(*mcp_res), 0);
3051 * The copy Flows are not included in any list. There
3052 * ones are referenced from other Flows and can not
3053 * be applied, removed, deleted in ardbitrary order
3054 * by list traversing.
3056 mcp_res->flow = flow_list_create(dev, NULL, &attr, items,
3057 actions, false, error);
3061 mcp_res->hlist_ent.key = mark_id;
3062 ret = mlx5_hlist_insert(priv->mreg_cp_tbl,
3063 &mcp_res->hlist_ent);
3070 flow_list_destroy(dev, NULL, mcp_res->flow);
3076 * Release flow in RX_CP_TBL.
3079 * Pointer to Ethernet device.
3081 * Parent flow for wich copying is provided.
3084 flow_mreg_del_copy_action(struct rte_eth_dev *dev,
3085 struct rte_flow *flow)
3087 struct mlx5_flow_mreg_copy_resource *mcp_res = flow->mreg_copy;
3088 struct mlx5_priv *priv = dev->data->dev_private;
3090 if (!mcp_res || !priv->mreg_cp_tbl)
3092 if (flow->copy_applied) {
3093 MLX5_ASSERT(mcp_res->appcnt);
3094 flow->copy_applied = 0;
3096 if (!mcp_res->appcnt)
3097 flow_drv_remove(dev, mcp_res->flow);
3100 * We do not check availability of metadata registers here,
3101 * because copy resources are not allocated in this case.
3103 if (--mcp_res->refcnt)
3105 MLX5_ASSERT(mcp_res->flow);
3106 flow_list_destroy(dev, NULL, mcp_res->flow);
3107 mlx5_hlist_remove(priv->mreg_cp_tbl, &mcp_res->hlist_ent);
3109 flow->mreg_copy = NULL;
3113 * Start flow in RX_CP_TBL.
3116 * Pointer to Ethernet device.
3118 * Parent flow for wich copying is provided.
3121 * 0 on success, a negative errno value otherwise and rte_errno is set.
3124 flow_mreg_start_copy_action(struct rte_eth_dev *dev,
3125 struct rte_flow *flow)
3127 struct mlx5_flow_mreg_copy_resource *mcp_res = flow->mreg_copy;
3130 if (!mcp_res || flow->copy_applied)
3132 if (!mcp_res->appcnt) {
3133 ret = flow_drv_apply(dev, mcp_res->flow, NULL);
3138 flow->copy_applied = 1;
3143 * Stop flow in RX_CP_TBL.
3146 * Pointer to Ethernet device.
3148 * Parent flow for wich copying is provided.
3151 flow_mreg_stop_copy_action(struct rte_eth_dev *dev,
3152 struct rte_flow *flow)
3154 struct mlx5_flow_mreg_copy_resource *mcp_res = flow->mreg_copy;
3156 if (!mcp_res || !flow->copy_applied)
3158 MLX5_ASSERT(mcp_res->appcnt);
3160 flow->copy_applied = 0;
3161 if (!mcp_res->appcnt)
3162 flow_drv_remove(dev, mcp_res->flow);
3166 * Remove the default copy action from RX_CP_TBL.
3169 * Pointer to Ethernet device.
3172 flow_mreg_del_default_copy_action(struct rte_eth_dev *dev)
3174 struct mlx5_flow_mreg_copy_resource *mcp_res;
3175 struct mlx5_priv *priv = dev->data->dev_private;
3177 /* Check if default flow is registered. */
3178 if (!priv->mreg_cp_tbl)
3180 mcp_res = (void *)mlx5_hlist_lookup(priv->mreg_cp_tbl,
3181 MLX5_DEFAULT_COPY_ID);
3184 MLX5_ASSERT(mcp_res->flow);
3185 flow_list_destroy(dev, NULL, mcp_res->flow);
3186 mlx5_hlist_remove(priv->mreg_cp_tbl, &mcp_res->hlist_ent);
3191 * Add the default copy action in in RX_CP_TBL.
3194 * Pointer to Ethernet device.
3196 * Perform verbose error reporting if not NULL.
3199 * 0 for success, negative value otherwise and rte_errno is set.
3202 flow_mreg_add_default_copy_action(struct rte_eth_dev *dev,
3203 struct rte_flow_error *error)
3205 struct mlx5_priv *priv = dev->data->dev_private;
3206 struct mlx5_flow_mreg_copy_resource *mcp_res;
3208 /* Check whether extensive metadata feature is engaged. */
3209 if (!priv->config.dv_flow_en ||
3210 priv->config.dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
3211 !mlx5_flow_ext_mreg_supported(dev) ||
3212 !priv->sh->dv_regc0_mask)
3214 mcp_res = flow_mreg_add_copy_action(dev, MLX5_DEFAULT_COPY_ID, error);
3221 * Add a flow of copying flow metadata registers in RX_CP_TBL.
3223 * All the flow having Q/RSS action should be split by
3224 * flow_mreg_split_qrss_prep() to pass by RX_CP_TBL. A flow in the RX_CP_TBL
3225 * performs the following,
3226 * - CQE->flow_tag := reg_c[1] (MARK)
3227 * - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
3228 * As CQE's flow_tag is not a register, it can't be simply copied from reg_c[1]
3229 * but there should be a flow per each MARK ID set by MARK action.
3231 * For the aforementioned reason, if there's a MARK action in flow's action
3232 * list, a corresponding flow should be added to the RX_CP_TBL in order to copy
3233 * the MARK ID to CQE's flow_tag like,
3234 * - If reg_c[1] is mark_id,
3235 * flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
3237 * For SET_META action which stores value in reg_c[0], as the destination is
3238 * also a flow metadata register (reg_b), adding a default flow is enough. Zero
3239 * MARK ID means the default flow. The default flow looks like,
3240 * - For all flow, reg_b := reg_c[0] and jump to RX_ACT_TBL.
3243 * Pointer to Ethernet device.
3245 * Pointer to flow structure.
3246 * @param[in] actions
3247 * Pointer to the list of actions.
3249 * Perform verbose error reporting if not NULL.
3252 * 0 on success, negative value otherwise and rte_errno is set.
3255 flow_mreg_update_copy_table(struct rte_eth_dev *dev,
3256 struct rte_flow *flow,
3257 const struct rte_flow_action *actions,
3258 struct rte_flow_error *error)
3260 struct mlx5_priv *priv = dev->data->dev_private;
3261 struct mlx5_dev_config *config = &priv->config;
3262 struct mlx5_flow_mreg_copy_resource *mcp_res;
3263 const struct rte_flow_action_mark *mark;
3265 /* Check whether extensive metadata feature is engaged. */
3266 if (!config->dv_flow_en ||
3267 config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
3268 !mlx5_flow_ext_mreg_supported(dev) ||
3269 !priv->sh->dv_regc0_mask)
3271 /* Find MARK action. */
3272 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3273 switch (actions->type) {
3274 case RTE_FLOW_ACTION_TYPE_FLAG:
3275 mcp_res = flow_mreg_add_copy_action
3276 (dev, MLX5_FLOW_MARK_DEFAULT, error);
3279 flow->mreg_copy = mcp_res;
3280 if (dev->data->dev_started) {
3282 flow->copy_applied = 1;
3285 case RTE_FLOW_ACTION_TYPE_MARK:
3286 mark = (const struct rte_flow_action_mark *)
3289 flow_mreg_add_copy_action(dev, mark->id, error);
3292 flow->mreg_copy = mcp_res;
3293 if (dev->data->dev_started) {
3295 flow->copy_applied = 1;
3305 #define MLX5_MAX_SPLIT_ACTIONS 24
3306 #define MLX5_MAX_SPLIT_ITEMS 24
3309 * Split the hairpin flow.
3310 * Since HW can't support encap on Rx we move the encap to Tx.
3311 * If the count action is after the encap then we also
3312 * move the count action. in this case the count will also measure
3316 * Pointer to Ethernet device.
3317 * @param[in] actions
3318 * Associated actions (list terminated by the END action).
3319 * @param[out] actions_rx
3321 * @param[out] actions_tx
3323 * @param[out] pattern_tx
3324 * The pattern items for the Tx flow.
3325 * @param[out] flow_id
3326 * The flow ID connected to this flow.
3332 flow_hairpin_split(struct rte_eth_dev *dev,
3333 const struct rte_flow_action actions[],
3334 struct rte_flow_action actions_rx[],
3335 struct rte_flow_action actions_tx[],
3336 struct rte_flow_item pattern_tx[],
3339 struct mlx5_priv *priv = dev->data->dev_private;
3340 const struct rte_flow_action_raw_encap *raw_encap;
3341 const struct rte_flow_action_raw_decap *raw_decap;
3342 struct mlx5_rte_flow_action_set_tag *set_tag;
3343 struct rte_flow_action *tag_action;
3344 struct mlx5_rte_flow_item_tag *tag_item;
3345 struct rte_flow_item *item;
3349 mlx5_flow_id_get(priv->sh->flow_id_pool, flow_id);
3350 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3351 switch (actions->type) {
3352 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
3353 case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
3354 rte_memcpy(actions_tx, actions,
3355 sizeof(struct rte_flow_action));
3358 case RTE_FLOW_ACTION_TYPE_COUNT:
3360 rte_memcpy(actions_tx, actions,
3361 sizeof(struct rte_flow_action));
3364 rte_memcpy(actions_rx, actions,
3365 sizeof(struct rte_flow_action));
3369 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3370 raw_encap = actions->conf;
3371 if (raw_encap->size >
3372 (sizeof(struct rte_flow_item_eth) +
3373 sizeof(struct rte_flow_item_ipv4))) {
3374 memcpy(actions_tx, actions,
3375 sizeof(struct rte_flow_action));
3379 rte_memcpy(actions_rx, actions,
3380 sizeof(struct rte_flow_action));
3384 case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
3385 raw_decap = actions->conf;
3386 if (raw_decap->size <
3387 (sizeof(struct rte_flow_item_eth) +
3388 sizeof(struct rte_flow_item_ipv4))) {
3389 memcpy(actions_tx, actions,
3390 sizeof(struct rte_flow_action));
3393 rte_memcpy(actions_rx, actions,
3394 sizeof(struct rte_flow_action));
3399 rte_memcpy(actions_rx, actions,
3400 sizeof(struct rte_flow_action));
3405 /* Add set meta action and end action for the Rx flow. */
3406 tag_action = actions_rx;
3407 tag_action->type = MLX5_RTE_FLOW_ACTION_TYPE_TAG;
3409 rte_memcpy(actions_rx, actions, sizeof(struct rte_flow_action));
3411 set_tag = (void *)actions_rx;
3412 set_tag->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_RX, 0, NULL);
3413 MLX5_ASSERT(set_tag->id > REG_NONE);
3414 set_tag->data = *flow_id;
3415 tag_action->conf = set_tag;
3416 /* Create Tx item list. */
3417 rte_memcpy(actions_tx, actions, sizeof(struct rte_flow_action));
3418 addr = (void *)&pattern_tx[2];
3420 item->type = MLX5_RTE_FLOW_ITEM_TYPE_TAG;
3421 tag_item = (void *)addr;
3422 tag_item->data = *flow_id;
3423 tag_item->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_TX, 0, NULL);
3424 MLX5_ASSERT(set_tag->id > REG_NONE);
3425 item->spec = tag_item;
3426 addr += sizeof(struct mlx5_rte_flow_item_tag);
3427 tag_item = (void *)addr;
3428 tag_item->data = UINT32_MAX;
3429 tag_item->id = UINT16_MAX;
3430 item->mask = tag_item;
3431 addr += sizeof(struct mlx5_rte_flow_item_tag);
3434 item->type = RTE_FLOW_ITEM_TYPE_END;
3439 * The last stage of splitting chain, just creates the subflow
3440 * without any modification.
3443 * Pointer to Ethernet device.
3445 * Parent flow structure pointer.
3446 * @param[in, out] sub_flow
3447 * Pointer to return the created subflow, may be NULL.
3448 * @param[in] prefix_layers
3449 * Prefix subflow layers, may be 0.
3451 * Flow rule attributes.
3453 * Pattern specification (list terminated by the END pattern item).
3454 * @param[in] actions
3455 * Associated actions (list terminated by the END action).
3456 * @param[in] external
3457 * This flow rule is created by request external to PMD.
3459 * Perform verbose error reporting if not NULL.
3461 * 0 on success, negative value otherwise
3464 flow_create_split_inner(struct rte_eth_dev *dev,
3465 struct rte_flow *flow,
3466 struct mlx5_flow **sub_flow,
3467 uint64_t prefix_layers,
3468 const struct rte_flow_attr *attr,
3469 const struct rte_flow_item items[],
3470 const struct rte_flow_action actions[],
3471 bool external, struct rte_flow_error *error)
3473 struct mlx5_flow *dev_flow;
3475 dev_flow = flow_drv_prepare(flow, attr, items, actions, error);
3478 dev_flow->flow = flow;
3479 dev_flow->external = external;
3480 /* Subflow object was created, we must include one in the list. */
3481 LIST_INSERT_HEAD(&flow->dev_flows, dev_flow, next);
3483 * If dev_flow is as one of the suffix flow, some actions in suffix
3484 * flow may need some user defined item layer flags.
3487 dev_flow->layers = prefix_layers;
3489 *sub_flow = dev_flow;
3490 return flow_drv_translate(dev, dev_flow, attr, items, actions, error);
3494 * Split the meter flow.
3496 * As meter flow will split to three sub flow, other than meter
3497 * action, the other actions make sense to only meter accepts
3498 * the packet. If it need to be dropped, no other additional
3499 * actions should be take.
3501 * One kind of special action which decapsulates the L3 tunnel
3502 * header will be in the prefix sub flow, as not to take the
3503 * L3 tunnel header into account.
3506 * Pointer to Ethernet device.
3508 * Pattern specification (list terminated by the END pattern item).
3509 * @param[out] sfx_items
3510 * Suffix flow match items (list terminated by the END pattern item).
3511 * @param[in] actions
3512 * Associated actions (list terminated by the END action).
3513 * @param[out] actions_sfx
3514 * Suffix flow actions.
3515 * @param[out] actions_pre
3516 * Prefix flow actions.
3517 * @param[out] pattern_sfx
3518 * The pattern items for the suffix flow.
3519 * @param[out] tag_sfx
3520 * Pointer to suffix flow tag.
3526 flow_meter_split_prep(struct rte_eth_dev *dev,
3527 const struct rte_flow_item items[],
3528 struct rte_flow_item sfx_items[],
3529 const struct rte_flow_action actions[],
3530 struct rte_flow_action actions_sfx[],
3531 struct rte_flow_action actions_pre[])
3533 struct rte_flow_action *tag_action = NULL;
3534 struct rte_flow_item *tag_item;
3535 struct mlx5_rte_flow_action_set_tag *set_tag;
3536 struct rte_flow_error error;
3537 const struct rte_flow_action_raw_encap *raw_encap;
3538 const struct rte_flow_action_raw_decap *raw_decap;
3539 struct mlx5_rte_flow_item_tag *tag_spec;
3540 struct mlx5_rte_flow_item_tag *tag_mask;
3542 bool copy_vlan = false;
3544 /* Prepare the actions for prefix and suffix flow. */
3545 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3546 struct rte_flow_action **action_cur = NULL;
3548 switch (actions->type) {
3549 case RTE_FLOW_ACTION_TYPE_METER:
3550 /* Add the extra tag action first. */
3551 tag_action = actions_pre;
3552 tag_action->type = MLX5_RTE_FLOW_ACTION_TYPE_TAG;
3554 action_cur = &actions_pre;
3556 case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
3557 case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
3558 action_cur = &actions_pre;
3560 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3561 raw_encap = actions->conf;
3562 if (raw_encap->size < MLX5_ENCAPSULATION_DECISION_SIZE)
3563 action_cur = &actions_pre;
3565 case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
3566 raw_decap = actions->conf;
3567 if (raw_decap->size > MLX5_ENCAPSULATION_DECISION_SIZE)
3568 action_cur = &actions_pre;
3570 case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
3571 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
3578 action_cur = &actions_sfx;
3579 memcpy(*action_cur, actions, sizeof(struct rte_flow_action));
3582 /* Add end action to the actions. */
3583 actions_sfx->type = RTE_FLOW_ACTION_TYPE_END;
3584 actions_pre->type = RTE_FLOW_ACTION_TYPE_END;
3587 set_tag = (void *)actions_pre;
3588 set_tag->id = mlx5_flow_get_reg_id(dev, MLX5_MTR_SFX, 0, &error);
3590 * Get the id from the qrss_pool to make qrss share the id with meter.
3592 tag_id = flow_qrss_get_id(dev);
3593 set_tag->data = tag_id << MLX5_MTR_COLOR_BITS;
3595 tag_action->conf = set_tag;
3596 /* Prepare the suffix subflow items. */
3597 tag_item = sfx_items++;
3598 for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
3599 int item_type = items->type;
3601 switch (item_type) {
3602 case RTE_FLOW_ITEM_TYPE_PORT_ID:
3603 memcpy(sfx_items, items, sizeof(*sfx_items));
3606 case RTE_FLOW_ITEM_TYPE_VLAN:
3608 memcpy(sfx_items, items, sizeof(*sfx_items));
3610 * Convert to internal match item, it is used
3611 * for vlan push and set vid.
3613 sfx_items->type = MLX5_RTE_FLOW_ITEM_TYPE_VLAN;
3621 sfx_items->type = RTE_FLOW_ITEM_TYPE_END;
3623 tag_spec = (struct mlx5_rte_flow_item_tag *)sfx_items;
3624 tag_spec->data = tag_id << MLX5_MTR_COLOR_BITS;
3625 tag_spec->id = mlx5_flow_get_reg_id(dev, MLX5_MTR_SFX, 0, &error);
3626 tag_mask = tag_spec + 1;
3627 tag_mask->data = 0xffffff00;
3628 tag_item->type = MLX5_RTE_FLOW_ITEM_TYPE_TAG;
3629 tag_item->spec = tag_spec;
3630 tag_item->last = NULL;
3631 tag_item->mask = tag_mask;
3636 * Split action list having QUEUE/RSS for metadata register copy.
3638 * Once Q/RSS action is detected in user's action list, the flow action
3639 * should be split in order to copy metadata registers, which will happen in
3641 * - CQE->flow_tag := reg_c[1] (MARK)
3642 * - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
3643 * The Q/RSS action will be performed on RX_ACT_TBL after passing by RX_CP_TBL.
3644 * This is because the last action of each flow must be a terminal action
3645 * (QUEUE, RSS or DROP).
3647 * Flow ID must be allocated to identify actions in the RX_ACT_TBL and it is
3648 * stored and kept in the mlx5_flow structure per each sub_flow.
3650 * The Q/RSS action is replaced with,
3651 * - SET_TAG, setting the allocated flow ID to reg_c[2].
3652 * And the following JUMP action is added at the end,
3653 * - JUMP, to RX_CP_TBL.
3655 * A flow to perform remained Q/RSS action will be created in RX_ACT_TBL by
3656 * flow_create_split_metadata() routine. The flow will look like,
3657 * - If flow ID matches (reg_c[2]), perform Q/RSS.
3660 * Pointer to Ethernet device.
3661 * @param[out] split_actions
3662 * Pointer to store split actions to jump to CP_TBL.
3663 * @param[in] actions
3664 * Pointer to the list of original flow actions.
3666 * Pointer to the Q/RSS action.
3667 * @param[in] actions_n
3668 * Number of original actions.
3670 * Perform verbose error reporting if not NULL.
3673 * non-zero unique flow_id on success, otherwise 0 and
3674 * error/rte_error are set.
3677 flow_mreg_split_qrss_prep(struct rte_eth_dev *dev,
3678 struct rte_flow_action *split_actions,
3679 const struct rte_flow_action *actions,
3680 const struct rte_flow_action *qrss,
3681 int actions_n, struct rte_flow_error *error)
3683 struct mlx5_rte_flow_action_set_tag *set_tag;
3684 struct rte_flow_action_jump *jump;
3685 const int qrss_idx = qrss - actions;
3686 uint32_t flow_id = 0;
3690 * Given actions will be split
3691 * - Replace QUEUE/RSS action with SET_TAG to set flow ID.
3692 * - Add jump to mreg CP_TBL.
3693 * As a result, there will be one more action.
3696 memcpy(split_actions, actions, sizeof(*split_actions) * actions_n);
3697 set_tag = (void *)(split_actions + actions_n);
3699 * If tag action is not set to void(it means we are not the meter
3700 * suffix flow), add the tag action. Since meter suffix flow already
3701 * has the tag added.
3703 if (split_actions[qrss_idx].type != RTE_FLOW_ACTION_TYPE_VOID) {
3705 * Allocate the new subflow ID. This one is unique within
3706 * device and not shared with representors. Otherwise,
3707 * we would have to resolve multi-thread access synch
3708 * issue. Each flow on the shared device is appended
3709 * with source vport identifier, so the resulting
3710 * flows will be unique in the shared (by master and
3711 * representors) domain even if they have coinciding
3714 flow_id = flow_qrss_get_id(dev);
3716 return rte_flow_error_set(error, ENOMEM,
3717 RTE_FLOW_ERROR_TYPE_ACTION,
3718 NULL, "can't allocate id "
3719 "for split Q/RSS subflow");
3720 /* Internal SET_TAG action to set flow ID. */
3721 *set_tag = (struct mlx5_rte_flow_action_set_tag){
3724 ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0, error);
3728 /* Construct new actions array. */
3729 /* Replace QUEUE/RSS action. */
3730 split_actions[qrss_idx] = (struct rte_flow_action){
3731 .type = MLX5_RTE_FLOW_ACTION_TYPE_TAG,
3735 /* JUMP action to jump to mreg copy table (CP_TBL). */
3736 jump = (void *)(set_tag + 1);
3737 *jump = (struct rte_flow_action_jump){
3738 .group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
3740 split_actions[actions_n - 2] = (struct rte_flow_action){
3741 .type = RTE_FLOW_ACTION_TYPE_JUMP,
3744 split_actions[actions_n - 1] = (struct rte_flow_action){
3745 .type = RTE_FLOW_ACTION_TYPE_END,
3751 * Extend the given action list for Tx metadata copy.
3753 * Copy the given action list to the ext_actions and add flow metadata register
3754 * copy action in order to copy reg_a set by WQE to reg_c[0].
3756 * @param[out] ext_actions
3757 * Pointer to the extended action list.
3758 * @param[in] actions
3759 * Pointer to the list of actions.
3760 * @param[in] actions_n
3761 * Number of actions in the list.
3763 * Perform verbose error reporting if not NULL.
3764 * @param[in] encap_idx
3765 * The encap action inndex.
3768 * 0 on success, negative value otherwise
3771 flow_mreg_tx_copy_prep(struct rte_eth_dev *dev,
3772 struct rte_flow_action *ext_actions,
3773 const struct rte_flow_action *actions,
3774 int actions_n, struct rte_flow_error *error,
3777 struct mlx5_flow_action_copy_mreg *cp_mreg =
3778 (struct mlx5_flow_action_copy_mreg *)
3779 (ext_actions + actions_n + 1);
3782 ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
3786 ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_TX, 0, error);
3791 memcpy(ext_actions, actions, sizeof(*ext_actions) * encap_idx);
3792 if (encap_idx == actions_n - 1) {
3793 ext_actions[actions_n - 1] = (struct rte_flow_action){
3794 .type = MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3797 ext_actions[actions_n] = (struct rte_flow_action){
3798 .type = RTE_FLOW_ACTION_TYPE_END,
3801 ext_actions[encap_idx] = (struct rte_flow_action){
3802 .type = MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3805 memcpy(ext_actions + encap_idx + 1, actions + encap_idx,
3806 sizeof(*ext_actions) * (actions_n - encap_idx));
3812 * The splitting for metadata feature.
3814 * - Q/RSS action on NIC Rx should be split in order to pass by
3815 * the mreg copy table (RX_CP_TBL) and then it jumps to the
3816 * action table (RX_ACT_TBL) which has the split Q/RSS action.
3818 * - All the actions on NIC Tx should have a mreg copy action to
3819 * copy reg_a from WQE to reg_c[0].
3822 * Pointer to Ethernet device.
3824 * Parent flow structure pointer.
3825 * @param[in] prefix_layers
3826 * Prefix flow layer flags.
3828 * Flow rule attributes.
3830 * Pattern specification (list terminated by the END pattern item).
3831 * @param[in] actions
3832 * Associated actions (list terminated by the END action).
3833 * @param[in] external
3834 * This flow rule is created by request external to PMD.
3836 * Perform verbose error reporting if not NULL.
3838 * 0 on success, negative value otherwise
3841 flow_create_split_metadata(struct rte_eth_dev *dev,
3842 struct rte_flow *flow,
3843 uint64_t prefix_layers,
3844 const struct rte_flow_attr *attr,
3845 const struct rte_flow_item items[],
3846 const struct rte_flow_action actions[],
3847 bool external, struct rte_flow_error *error)
3849 struct mlx5_priv *priv = dev->data->dev_private;
3850 struct mlx5_dev_config *config = &priv->config;
3851 const struct rte_flow_action *qrss = NULL;
3852 struct rte_flow_action *ext_actions = NULL;
3853 struct mlx5_flow *dev_flow = NULL;
3854 uint32_t qrss_id = 0;
3861 /* Check whether extensive metadata feature is engaged. */
3862 if (!config->dv_flow_en ||
3863 config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
3864 !mlx5_flow_ext_mreg_supported(dev))
3865 return flow_create_split_inner(dev, flow, NULL, prefix_layers,
3866 attr, items, actions, external,
3868 actions_n = flow_parse_metadata_split_actions_info(actions, &qrss,
3871 /* Exclude hairpin flows from splitting. */
3872 if (qrss->type == RTE_FLOW_ACTION_TYPE_QUEUE) {
3873 const struct rte_flow_action_queue *queue;
3876 if (mlx5_rxq_get_type(dev, queue->index) ==
3877 MLX5_RXQ_TYPE_HAIRPIN)
3879 } else if (qrss->type == RTE_FLOW_ACTION_TYPE_RSS) {
3880 const struct rte_flow_action_rss *rss;
3883 if (mlx5_rxq_get_type(dev, rss->queue[0]) ==
3884 MLX5_RXQ_TYPE_HAIRPIN)
3889 /* Check if it is in meter suffix table. */
3890 mtr_sfx = attr->group == (attr->transfer ?
3891 (MLX5_FLOW_TABLE_LEVEL_SUFFIX - 1) :
3892 MLX5_FLOW_TABLE_LEVEL_SUFFIX);
3894 * Q/RSS action on NIC Rx should be split in order to pass by
3895 * the mreg copy table (RX_CP_TBL) and then it jumps to the
3896 * action table (RX_ACT_TBL) which has the split Q/RSS action.
3898 act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
3899 sizeof(struct rte_flow_action_set_tag) +
3900 sizeof(struct rte_flow_action_jump);
3901 ext_actions = rte_zmalloc(__func__, act_size, 0);
3903 return rte_flow_error_set(error, ENOMEM,
3904 RTE_FLOW_ERROR_TYPE_ACTION,
3905 NULL, "no memory to split "
3908 * If we are the suffix flow of meter, tag already exist.
3909 * Set the tag action to void.
3912 ext_actions[qrss - actions].type =
3913 RTE_FLOW_ACTION_TYPE_VOID;
3915 ext_actions[qrss - actions].type =
3916 MLX5_RTE_FLOW_ACTION_TYPE_TAG;
3918 * Create the new actions list with removed Q/RSS action
3919 * and appended set tag and jump to register copy table
3920 * (RX_CP_TBL). We should preallocate unique tag ID here
3921 * in advance, because it is needed for set tag action.
3923 qrss_id = flow_mreg_split_qrss_prep(dev, ext_actions, actions,
3924 qrss, actions_n, error);
3925 if (!mtr_sfx && !qrss_id) {
3929 } else if (attr->egress && !attr->transfer) {
3931 * All the actions on NIC Tx should have a metadata register
3932 * copy action to copy reg_a from WQE to reg_c[meta]
3934 act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
3935 sizeof(struct mlx5_flow_action_copy_mreg);
3936 ext_actions = rte_zmalloc(__func__, act_size, 0);
3938 return rte_flow_error_set(error, ENOMEM,
3939 RTE_FLOW_ERROR_TYPE_ACTION,
3940 NULL, "no memory to split "
3942 /* Create the action list appended with copy register. */
3943 ret = flow_mreg_tx_copy_prep(dev, ext_actions, actions,
3944 actions_n, error, encap_idx);
3948 /* Add the unmodified original or prefix subflow. */
3949 ret = flow_create_split_inner(dev, flow, &dev_flow, prefix_layers, attr,
3950 items, ext_actions ? ext_actions :
3951 actions, external, error);
3954 MLX5_ASSERT(dev_flow);
3956 const struct rte_flow_attr q_attr = {
3957 .group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
3960 /* Internal PMD action to set register. */
3961 struct mlx5_rte_flow_item_tag q_tag_spec = {
3965 struct rte_flow_item q_items[] = {
3967 .type = MLX5_RTE_FLOW_ITEM_TYPE_TAG,
3968 .spec = &q_tag_spec,
3973 .type = RTE_FLOW_ITEM_TYPE_END,
3976 struct rte_flow_action q_actions[] = {
3982 .type = RTE_FLOW_ACTION_TYPE_END,
3985 uint64_t layers = flow_get_prefix_layer_flags(dev_flow);
3988 * Configure the tag item only if there is no meter subflow.
3989 * Since tag is already marked in the meter suffix subflow
3990 * we can just use the meter suffix items as is.
3993 /* Not meter subflow. */
3994 MLX5_ASSERT(!mtr_sfx);
3996 * Put unique id in prefix flow due to it is destroyed
3997 * after suffix flow and id will be freed after there
3998 * is no actual flows with this id and identifier
3999 * reallocation becomes possible (for example, for
4000 * other flows in other threads).
4002 dev_flow->qrss_id = qrss_id;
4004 ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0,
4008 q_tag_spec.id = ret;
4011 /* Add suffix subflow to execute Q/RSS. */
4012 ret = flow_create_split_inner(dev, flow, &dev_flow, layers,
4013 &q_attr, mtr_sfx ? items :
4018 MLX5_ASSERT(dev_flow);
4023 * We do not destroy the partially created sub_flows in case of error.
4024 * These ones are included into parent flow list and will be destroyed
4025 * by flow_drv_destroy.
4027 flow_qrss_free_id(dev, qrss_id);
4028 rte_free(ext_actions);
4033 * The splitting for meter feature.
4035 * - The meter flow will be split to two flows as prefix and
4036 * suffix flow. The packets make sense only it pass the prefix
4039 * - Reg_C_5 is used for the packet to match betweend prefix and
4043 * Pointer to Ethernet device.
4045 * Parent flow structure pointer.
4047 * Flow rule attributes.
4049 * Pattern specification (list terminated by the END pattern item).
4050 * @param[in] actions
4051 * Associated actions (list terminated by the END action).
4052 * @param[in] external
4053 * This flow rule is created by request external to PMD.
4055 * Perform verbose error reporting if not NULL.
4057 * 0 on success, negative value otherwise
4060 flow_create_split_meter(struct rte_eth_dev *dev,
4061 struct rte_flow *flow,
4062 const struct rte_flow_attr *attr,
4063 const struct rte_flow_item items[],
4064 const struct rte_flow_action actions[],
4065 bool external, struct rte_flow_error *error)
4067 struct mlx5_priv *priv = dev->data->dev_private;
4068 struct rte_flow_action *sfx_actions = NULL;
4069 struct rte_flow_action *pre_actions = NULL;
4070 struct rte_flow_item *sfx_items = NULL;
4071 struct mlx5_flow *dev_flow = NULL;
4072 struct rte_flow_attr sfx_attr = *attr;
4074 uint32_t mtr_tag_id = 0;
4081 actions_n = flow_check_meter_action(actions, &mtr);
4083 /* The five prefix actions: meter, decap, encap, tag, end. */
4084 act_size = sizeof(struct rte_flow_action) * (actions_n + 5) +
4085 sizeof(struct mlx5_rte_flow_action_set_tag);
4086 /* tag, vlan, port id, end. */
4087 #define METER_SUFFIX_ITEM 4
4088 item_size = sizeof(struct rte_flow_item) * METER_SUFFIX_ITEM +
4089 sizeof(struct mlx5_rte_flow_item_tag) * 2;
4090 sfx_actions = rte_zmalloc(__func__, (act_size + item_size), 0);
4092 return rte_flow_error_set(error, ENOMEM,
4093 RTE_FLOW_ERROR_TYPE_ACTION,
4094 NULL, "no memory to split "
4096 sfx_items = (struct rte_flow_item *)((char *)sfx_actions +
4098 pre_actions = sfx_actions + actions_n;
4099 mtr_tag_id = flow_meter_split_prep(dev, items, sfx_items,
4100 actions, sfx_actions,
4106 /* Add the prefix subflow. */
4107 ret = flow_create_split_inner(dev, flow, &dev_flow, 0, attr,
4108 items, pre_actions, external,
4114 dev_flow->mtr_flow_id = mtr_tag_id;
4115 /* Setting the sfx group atrr. */
4116 sfx_attr.group = sfx_attr.transfer ?
4117 (MLX5_FLOW_TABLE_LEVEL_SUFFIX - 1) :
4118 MLX5_FLOW_TABLE_LEVEL_SUFFIX;
4120 /* Add the prefix subflow. */
4121 ret = flow_create_split_metadata(dev, flow, dev_flow ?
4122 flow_get_prefix_layer_flags(dev_flow) :
4124 sfx_items ? sfx_items : items,
4125 sfx_actions ? sfx_actions : actions,
4129 rte_free(sfx_actions);
4134 * Split the flow to subflow set. The splitters might be linked
4135 * in the chain, like this:
4136 * flow_create_split_outer() calls:
4137 * flow_create_split_meter() calls:
4138 * flow_create_split_metadata(meter_subflow_0) calls:
4139 * flow_create_split_inner(metadata_subflow_0)
4140 * flow_create_split_inner(metadata_subflow_1)
4141 * flow_create_split_inner(metadata_subflow_2)
4142 * flow_create_split_metadata(meter_subflow_1) calls:
4143 * flow_create_split_inner(metadata_subflow_0)
4144 * flow_create_split_inner(metadata_subflow_1)
4145 * flow_create_split_inner(metadata_subflow_2)
4147 * This provide flexible way to add new levels of flow splitting.
4148 * The all of successfully created subflows are included to the
4149 * parent flow dev_flow list.
4152 * Pointer to Ethernet device.
4154 * Parent flow structure pointer.
4156 * Flow rule attributes.
4158 * Pattern specification (list terminated by the END pattern item).
4159 * @param[in] actions
4160 * Associated actions (list terminated by the END action).
4161 * @param[in] external
4162 * This flow rule is created by request external to PMD.
4164 * Perform verbose error reporting if not NULL.
4166 * 0 on success, negative value otherwise
4169 flow_create_split_outer(struct rte_eth_dev *dev,
4170 struct rte_flow *flow,
4171 const struct rte_flow_attr *attr,
4172 const struct rte_flow_item items[],
4173 const struct rte_flow_action actions[],
4174 bool external, struct rte_flow_error *error)
4178 ret = flow_create_split_meter(dev, flow, attr, items,
4179 actions, external, error);
4180 MLX5_ASSERT(ret <= 0);
4185 * Create a flow and add it to @p list.
4188 * Pointer to Ethernet device.
4190 * Pointer to a TAILQ flow list. If this parameter NULL,
4191 * no list insertion occurred, flow is just created,
4192 * this is caller's responsibility to track the
4195 * Flow rule attributes.
4197 * Pattern specification (list terminated by the END pattern item).
4198 * @param[in] actions
4199 * Associated actions (list terminated by the END action).
4200 * @param[in] external
4201 * This flow rule is created by request external to PMD.
4203 * Perform verbose error reporting if not NULL.
4206 * A flow on success, NULL otherwise and rte_errno is set.
4208 static struct rte_flow *
4209 flow_list_create(struct rte_eth_dev *dev, struct mlx5_flows *list,
4210 const struct rte_flow_attr *attr,
4211 const struct rte_flow_item items[],
4212 const struct rte_flow_action actions[],
4213 bool external, struct rte_flow_error *error)
4215 struct mlx5_priv *priv = dev->data->dev_private;
4216 struct rte_flow *flow = NULL;
4217 struct mlx5_flow *dev_flow;
4218 const struct rte_flow_action_rss *rss;
4220 struct rte_flow_expand_rss buf;
4221 uint8_t buffer[2048];
4224 struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
4225 uint8_t buffer[2048];
4228 struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
4229 uint8_t buffer[2048];
4230 } actions_hairpin_tx;
4232 struct rte_flow_item items[MLX5_MAX_SPLIT_ITEMS];
4233 uint8_t buffer[2048];
4235 struct rte_flow_expand_rss *buf = &expand_buffer.buf;
4236 const struct rte_flow_action *p_actions_rx = actions;
4239 int hairpin_flow = 0;
4240 uint32_t hairpin_id = 0;
4241 struct rte_flow_attr attr_tx = { .priority = 0 };
4242 int ret = flow_drv_validate(dev, attr, items, p_actions_rx, external,
4247 hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
4248 if (hairpin_flow > 0) {
4249 if (hairpin_flow > MLX5_MAX_SPLIT_ACTIONS) {
4253 flow_hairpin_split(dev, actions, actions_rx.actions,
4254 actions_hairpin_tx.actions, items_tx.items,
4256 p_actions_rx = actions_rx.actions;
4258 flow_size = sizeof(struct rte_flow);
4259 rss = flow_get_rss_action(p_actions_rx);
4261 flow_size += RTE_ALIGN_CEIL(rss->queue_num * sizeof(uint16_t),
4264 flow_size += RTE_ALIGN_CEIL(sizeof(uint16_t), sizeof(void *));
4265 flow = rte_calloc(__func__, 1, flow_size, 0);
4268 goto error_before_flow;
4270 flow->drv_type = flow_get_drv_type(dev, attr);
4271 if (hairpin_id != 0)
4272 flow->hairpin_flow_id = hairpin_id;
4273 MLX5_ASSERT(flow->drv_type > MLX5_FLOW_TYPE_MIN &&
4274 flow->drv_type < MLX5_FLOW_TYPE_MAX);
4275 flow->rss.queue = (void *)(flow + 1);
4278 * The following information is required by
4279 * mlx5_flow_hashfields_adjust() in advance.
4281 flow->rss.level = rss->level;
4282 /* RSS type 0 indicates default RSS type (ETH_RSS_IP). */
4283 flow->rss.types = !rss->types ? ETH_RSS_IP : rss->types;
4285 LIST_INIT(&flow->dev_flows);
4286 if (rss && rss->types) {
4287 unsigned int graph_root;
4289 graph_root = find_graph_root(items, rss->level);
4290 ret = rte_flow_expand_rss(buf, sizeof(expand_buffer.buffer),
4292 mlx5_support_expansion,
4294 MLX5_ASSERT(ret > 0 &&
4295 (unsigned int)ret < sizeof(expand_buffer.buffer));
4298 buf->entry[0].pattern = (void *)(uintptr_t)items;
4300 for (i = 0; i < buf->entries; ++i) {
4302 * The splitter may create multiple dev_flows,
4303 * depending on configuration. In the simplest
4304 * case it just creates unmodified original flow.
4306 ret = flow_create_split_outer(dev, flow, attr,
4307 buf->entry[i].pattern,
4308 p_actions_rx, external,
4313 /* Create the tx flow. */
4315 attr_tx.group = MLX5_HAIRPIN_TX_TABLE;
4316 attr_tx.ingress = 0;
4318 dev_flow = flow_drv_prepare(flow, &attr_tx, items_tx.items,
4319 actions_hairpin_tx.actions, error);
4322 dev_flow->flow = flow;
4323 dev_flow->external = 0;
4324 LIST_INSERT_HEAD(&flow->dev_flows, dev_flow, next);
4325 ret = flow_drv_translate(dev, dev_flow, &attr_tx,
4327 actions_hairpin_tx.actions, error);
4332 * Update the metadata register copy table. If extensive
4333 * metadata feature is enabled and registers are supported
4334 * we might create the extra rte_flow for each unique
4335 * MARK/FLAG action ID.
4337 * The table is updated for ingress Flows only, because
4338 * the egress Flows belong to the different device and
4339 * copy table should be updated in peer NIC Rx domain.
4341 if (attr->ingress &&
4342 (external || attr->group != MLX5_FLOW_MREG_CP_TABLE_GROUP)) {
4343 ret = flow_mreg_update_copy_table(dev, flow, actions, error);
4347 if (dev->data->dev_started) {
4348 ret = flow_drv_apply(dev, flow, error);
4353 TAILQ_INSERT_TAIL(list, flow, next);
4354 flow_rxq_flags_set(dev, flow);
4358 mlx5_flow_id_release(priv->sh->flow_id_pool,
4363 flow_mreg_del_copy_action(dev, flow);
4364 ret = rte_errno; /* Save rte_errno before cleanup. */
4365 if (flow->hairpin_flow_id)
4366 mlx5_flow_id_release(priv->sh->flow_id_pool,
4367 flow->hairpin_flow_id);
4369 flow_drv_destroy(dev, flow);
4371 rte_errno = ret; /* Restore rte_errno. */
4376 * Create a dedicated flow rule on e-switch table 0 (root table), to direct all
4377 * incoming packets to table 1.
4379 * Other flow rules, requested for group n, will be created in
4380 * e-switch table n+1.
4381 * Jump action to e-switch group n will be created to group n+1.
4383 * Used when working in switchdev mode, to utilise advantages of table 1
4387 * Pointer to Ethernet device.
4390 * Pointer to flow on success, NULL otherwise and rte_errno is set.
4393 mlx5_flow_create_esw_table_zero_flow(struct rte_eth_dev *dev)
4395 const struct rte_flow_attr attr = {
4402 const struct rte_flow_item pattern = {
4403 .type = RTE_FLOW_ITEM_TYPE_END,
4405 struct rte_flow_action_jump jump = {
4408 const struct rte_flow_action actions[] = {
4410 .type = RTE_FLOW_ACTION_TYPE_JUMP,
4414 .type = RTE_FLOW_ACTION_TYPE_END,
4417 struct mlx5_priv *priv = dev->data->dev_private;
4418 struct rte_flow_error error;
4420 return flow_list_create(dev, &priv->ctrl_flows, &attr, &pattern,
4421 actions, false, &error);
4427 * @see rte_flow_create()
4431 mlx5_flow_create(struct rte_eth_dev *dev,
4432 const struct rte_flow_attr *attr,
4433 const struct rte_flow_item items[],
4434 const struct rte_flow_action actions[],
4435 struct rte_flow_error *error)
4437 struct mlx5_priv *priv = dev->data->dev_private;
4439 return flow_list_create(dev, &priv->flows,
4440 attr, items, actions, true, error);
4444 * Destroy a flow in a list.
4447 * Pointer to Ethernet device.
4449 * Pointer to a TAILQ flow list. If this parameter NULL,
4450 * there is no flow removal from the list.
4455 flow_list_destroy(struct rte_eth_dev *dev, struct mlx5_flows *list,
4456 struct rte_flow *flow)
4458 struct mlx5_priv *priv = dev->data->dev_private;
4461 * Update RX queue flags only if port is started, otherwise it is
4464 if (dev->data->dev_started)
4465 flow_rxq_flags_trim(dev, flow);
4466 if (flow->hairpin_flow_id)
4467 mlx5_flow_id_release(priv->sh->flow_id_pool,
4468 flow->hairpin_flow_id);
4469 flow_drv_destroy(dev, flow);
4471 TAILQ_REMOVE(list, flow, next);
4472 flow_mreg_del_copy_action(dev, flow);
4473 rte_free(flow->fdir);
4478 * Destroy all flows.
4481 * Pointer to Ethernet device.
4483 * Pointer to a TAILQ flow list.
4486 mlx5_flow_list_flush(struct rte_eth_dev *dev, struct mlx5_flows *list)
4488 while (!TAILQ_EMPTY(list)) {
4489 struct rte_flow *flow;
4491 flow = TAILQ_FIRST(list);
4492 flow_list_destroy(dev, list, flow);
4500 * Pointer to Ethernet device.
4502 * Pointer to a TAILQ flow list.
4505 mlx5_flow_stop(struct rte_eth_dev *dev, struct mlx5_flows *list)
4507 struct rte_flow *flow;
4509 TAILQ_FOREACH_REVERSE(flow, list, mlx5_flows, next) {
4510 flow_drv_remove(dev, flow);
4511 flow_mreg_stop_copy_action(dev, flow);
4513 flow_mreg_del_default_copy_action(dev);
4514 flow_rxq_flags_clear(dev);
4521 * Pointer to Ethernet device.
4523 * Pointer to a TAILQ flow list.
4526 * 0 on success, a negative errno value otherwise and rte_errno is set.
4529 mlx5_flow_start(struct rte_eth_dev *dev, struct mlx5_flows *list)
4531 struct rte_flow *flow;
4532 struct rte_flow_error error;
4535 /* Make sure default copy action (reg_c[0] -> reg_b) is created. */
4536 ret = flow_mreg_add_default_copy_action(dev, &error);
4539 /* Apply Flows created by application. */
4540 TAILQ_FOREACH(flow, list, next) {
4541 ret = flow_mreg_start_copy_action(dev, flow);
4544 ret = flow_drv_apply(dev, flow, &error);
4547 flow_rxq_flags_set(dev, flow);
4551 ret = rte_errno; /* Save rte_errno before cleanup. */
4552 mlx5_flow_stop(dev, list);
4553 rte_errno = ret; /* Restore rte_errno. */
4558 * Verify the flow list is empty
4561 * Pointer to Ethernet device.
4563 * @return the number of flows not released.
4566 mlx5_flow_verify(struct rte_eth_dev *dev)
4568 struct mlx5_priv *priv = dev->data->dev_private;
4569 struct rte_flow *flow;
4572 TAILQ_FOREACH(flow, &priv->flows, next) {
4573 DRV_LOG(DEBUG, "port %u flow %p still referenced",
4574 dev->data->port_id, (void *)flow);
4581 * Enable default hairpin egress flow.
4584 * Pointer to Ethernet device.
4589 * 0 on success, a negative errno value otherwise and rte_errno is set.
4592 mlx5_ctrl_flow_source_queue(struct rte_eth_dev *dev,
4595 struct mlx5_priv *priv = dev->data->dev_private;
4596 const struct rte_flow_attr attr = {
4600 struct mlx5_rte_flow_item_tx_queue queue_spec = {
4603 struct mlx5_rte_flow_item_tx_queue queue_mask = {
4604 .queue = UINT32_MAX,
4606 struct rte_flow_item items[] = {
4608 .type = MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE,
4609 .spec = &queue_spec,
4611 .mask = &queue_mask,
4614 .type = RTE_FLOW_ITEM_TYPE_END,
4617 struct rte_flow_action_jump jump = {
4618 .group = MLX5_HAIRPIN_TX_TABLE,
4620 struct rte_flow_action actions[2];
4621 struct rte_flow *flow;
4622 struct rte_flow_error error;
4624 actions[0].type = RTE_FLOW_ACTION_TYPE_JUMP;
4625 actions[0].conf = &jump;
4626 actions[1].type = RTE_FLOW_ACTION_TYPE_END;
4627 flow = flow_list_create(dev, &priv->ctrl_flows,
4628 &attr, items, actions, false, &error);
4631 "Failed to create ctrl flow: rte_errno(%d),"
4632 " type(%d), message(%s)",
4633 rte_errno, error.type,
4634 error.message ? error.message : " (no stated reason)");
4641 * Enable a control flow configured from the control plane.
4644 * Pointer to Ethernet device.
4646 * An Ethernet flow spec to apply.
4648 * An Ethernet flow mask to apply.
4650 * A VLAN flow spec to apply.
4652 * A VLAN flow mask to apply.
4655 * 0 on success, a negative errno value otherwise and rte_errno is set.
4658 mlx5_ctrl_flow_vlan(struct rte_eth_dev *dev,
4659 struct rte_flow_item_eth *eth_spec,
4660 struct rte_flow_item_eth *eth_mask,
4661 struct rte_flow_item_vlan *vlan_spec,
4662 struct rte_flow_item_vlan *vlan_mask)
4664 struct mlx5_priv *priv = dev->data->dev_private;
4665 const struct rte_flow_attr attr = {
4667 .priority = MLX5_FLOW_PRIO_RSVD,
4669 struct rte_flow_item items[] = {
4671 .type = RTE_FLOW_ITEM_TYPE_ETH,
4677 .type = (vlan_spec) ? RTE_FLOW_ITEM_TYPE_VLAN :
4678 RTE_FLOW_ITEM_TYPE_END,
4684 .type = RTE_FLOW_ITEM_TYPE_END,
4687 uint16_t queue[priv->reta_idx_n];
4688 struct rte_flow_action_rss action_rss = {
4689 .func = RTE_ETH_HASH_FUNCTION_DEFAULT,
4691 .types = priv->rss_conf.rss_hf,
4692 .key_len = priv->rss_conf.rss_key_len,
4693 .queue_num = priv->reta_idx_n,
4694 .key = priv->rss_conf.rss_key,
4697 struct rte_flow_action actions[] = {
4699 .type = RTE_FLOW_ACTION_TYPE_RSS,
4700 .conf = &action_rss,
4703 .type = RTE_FLOW_ACTION_TYPE_END,
4706 struct rte_flow *flow;
4707 struct rte_flow_error error;
4710 if (!priv->reta_idx_n || !priv->rxqs_n) {
4713 for (i = 0; i != priv->reta_idx_n; ++i)
4714 queue[i] = (*priv->reta_idx)[i];
4715 flow = flow_list_create(dev, &priv->ctrl_flows,
4716 &attr, items, actions, false, &error);
4723 * Enable a flow control configured from the control plane.
4726 * Pointer to Ethernet device.
4728 * An Ethernet flow spec to apply.
4730 * An Ethernet flow mask to apply.
4733 * 0 on success, a negative errno value otherwise and rte_errno is set.
4736 mlx5_ctrl_flow(struct rte_eth_dev *dev,
4737 struct rte_flow_item_eth *eth_spec,
4738 struct rte_flow_item_eth *eth_mask)
4740 return mlx5_ctrl_flow_vlan(dev, eth_spec, eth_mask, NULL, NULL);
4746 * @see rte_flow_destroy()
4750 mlx5_flow_destroy(struct rte_eth_dev *dev,
4751 struct rte_flow *flow,
4752 struct rte_flow_error *error __rte_unused)
4754 struct mlx5_priv *priv = dev->data->dev_private;
4756 flow_list_destroy(dev, &priv->flows, flow);
4761 * Destroy all flows.
4763 * @see rte_flow_flush()
4767 mlx5_flow_flush(struct rte_eth_dev *dev,
4768 struct rte_flow_error *error __rte_unused)
4770 struct mlx5_priv *priv = dev->data->dev_private;
4772 mlx5_flow_list_flush(dev, &priv->flows);
4779 * @see rte_flow_isolate()
4783 mlx5_flow_isolate(struct rte_eth_dev *dev,
4785 struct rte_flow_error *error)
4787 struct mlx5_priv *priv = dev->data->dev_private;
4789 if (dev->data->dev_started) {
4790 rte_flow_error_set(error, EBUSY,
4791 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
4793 "port must be stopped first");
4796 priv->isolated = !!enable;
4798 dev->dev_ops = &mlx5_dev_ops_isolate;
4800 dev->dev_ops = &mlx5_dev_ops;
4807 * @see rte_flow_query()
4811 flow_drv_query(struct rte_eth_dev *dev,
4812 struct rte_flow *flow,
4813 const struct rte_flow_action *actions,
4815 struct rte_flow_error *error)
4817 const struct mlx5_flow_driver_ops *fops;
4818 enum mlx5_flow_drv_type ftype = flow->drv_type;
4820 MLX5_ASSERT(ftype > MLX5_FLOW_TYPE_MIN && ftype < MLX5_FLOW_TYPE_MAX);
4821 fops = flow_get_drv_ops(ftype);
4823 return fops->query(dev, flow, actions, data, error);
4829 * @see rte_flow_query()
4833 mlx5_flow_query(struct rte_eth_dev *dev,
4834 struct rte_flow *flow,
4835 const struct rte_flow_action *actions,
4837 struct rte_flow_error *error)
4841 ret = flow_drv_query(dev, flow, actions, data, error);
4848 * Convert a flow director filter to a generic flow.
4851 * Pointer to Ethernet device.
4852 * @param fdir_filter
4853 * Flow director filter to add.
4855 * Generic flow parameters structure.
4858 * 0 on success, a negative errno value otherwise and rte_errno is set.
4861 flow_fdir_filter_convert(struct rte_eth_dev *dev,
4862 const struct rte_eth_fdir_filter *fdir_filter,
4863 struct mlx5_fdir *attributes)
4865 struct mlx5_priv *priv = dev->data->dev_private;
4866 const struct rte_eth_fdir_input *input = &fdir_filter->input;
4867 const struct rte_eth_fdir_masks *mask =
4868 &dev->data->dev_conf.fdir_conf.mask;
4870 /* Validate queue number. */
4871 if (fdir_filter->action.rx_queue >= priv->rxqs_n) {
4872 DRV_LOG(ERR, "port %u invalid queue number %d",
4873 dev->data->port_id, fdir_filter->action.rx_queue);
4877 attributes->attr.ingress = 1;
4878 attributes->items[0] = (struct rte_flow_item) {
4879 .type = RTE_FLOW_ITEM_TYPE_ETH,
4880 .spec = &attributes->l2,
4881 .mask = &attributes->l2_mask,
4883 switch (fdir_filter->action.behavior) {
4884 case RTE_ETH_FDIR_ACCEPT:
4885 attributes->actions[0] = (struct rte_flow_action){
4886 .type = RTE_FLOW_ACTION_TYPE_QUEUE,
4887 .conf = &attributes->queue,
4890 case RTE_ETH_FDIR_REJECT:
4891 attributes->actions[0] = (struct rte_flow_action){
4892 .type = RTE_FLOW_ACTION_TYPE_DROP,
4896 DRV_LOG(ERR, "port %u invalid behavior %d",
4898 fdir_filter->action.behavior);
4899 rte_errno = ENOTSUP;
4902 attributes->queue.index = fdir_filter->action.rx_queue;
4904 switch (fdir_filter->input.flow_type) {
4905 case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
4906 case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
4907 case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
4908 attributes->l3.ipv4.hdr = (struct rte_ipv4_hdr){
4909 .src_addr = input->flow.ip4_flow.src_ip,
4910 .dst_addr = input->flow.ip4_flow.dst_ip,
4911 .time_to_live = input->flow.ip4_flow.ttl,
4912 .type_of_service = input->flow.ip4_flow.tos,
4914 attributes->l3_mask.ipv4.hdr = (struct rte_ipv4_hdr){
4915 .src_addr = mask->ipv4_mask.src_ip,
4916 .dst_addr = mask->ipv4_mask.dst_ip,
4917 .time_to_live = mask->ipv4_mask.ttl,
4918 .type_of_service = mask->ipv4_mask.tos,
4919 .next_proto_id = mask->ipv4_mask.proto,
4921 attributes->items[1] = (struct rte_flow_item){
4922 .type = RTE_FLOW_ITEM_TYPE_IPV4,
4923 .spec = &attributes->l3,
4924 .mask = &attributes->l3_mask,
4927 case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
4928 case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
4929 case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
4930 attributes->l3.ipv6.hdr = (struct rte_ipv6_hdr){
4931 .hop_limits = input->flow.ipv6_flow.hop_limits,
4932 .proto = input->flow.ipv6_flow.proto,
4935 memcpy(attributes->l3.ipv6.hdr.src_addr,
4936 input->flow.ipv6_flow.src_ip,
4937 RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
4938 memcpy(attributes->l3.ipv6.hdr.dst_addr,
4939 input->flow.ipv6_flow.dst_ip,
4940 RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
4941 memcpy(attributes->l3_mask.ipv6.hdr.src_addr,
4942 mask->ipv6_mask.src_ip,
4943 RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
4944 memcpy(attributes->l3_mask.ipv6.hdr.dst_addr,
4945 mask->ipv6_mask.dst_ip,
4946 RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
4947 attributes->items[1] = (struct rte_flow_item){
4948 .type = RTE_FLOW_ITEM_TYPE_IPV6,
4949 .spec = &attributes->l3,
4950 .mask = &attributes->l3_mask,
4954 DRV_LOG(ERR, "port %u invalid flow type%d",
4955 dev->data->port_id, fdir_filter->input.flow_type);
4956 rte_errno = ENOTSUP;
4960 switch (fdir_filter->input.flow_type) {
4961 case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
4962 attributes->l4.udp.hdr = (struct rte_udp_hdr){
4963 .src_port = input->flow.udp4_flow.src_port,
4964 .dst_port = input->flow.udp4_flow.dst_port,
4966 attributes->l4_mask.udp.hdr = (struct rte_udp_hdr){
4967 .src_port = mask->src_port_mask,
4968 .dst_port = mask->dst_port_mask,
4970 attributes->items[2] = (struct rte_flow_item){
4971 .type = RTE_FLOW_ITEM_TYPE_UDP,
4972 .spec = &attributes->l4,
4973 .mask = &attributes->l4_mask,
4976 case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
4977 attributes->l4.tcp.hdr = (struct rte_tcp_hdr){
4978 .src_port = input->flow.tcp4_flow.src_port,
4979 .dst_port = input->flow.tcp4_flow.dst_port,
4981 attributes->l4_mask.tcp.hdr = (struct rte_tcp_hdr){
4982 .src_port = mask->src_port_mask,
4983 .dst_port = mask->dst_port_mask,
4985 attributes->items[2] = (struct rte_flow_item){
4986 .type = RTE_FLOW_ITEM_TYPE_TCP,
4987 .spec = &attributes->l4,
4988 .mask = &attributes->l4_mask,
4991 case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
4992 attributes->l4.udp.hdr = (struct rte_udp_hdr){
4993 .src_port = input->flow.udp6_flow.src_port,
4994 .dst_port = input->flow.udp6_flow.dst_port,
4996 attributes->l4_mask.udp.hdr = (struct rte_udp_hdr){
4997 .src_port = mask->src_port_mask,
4998 .dst_port = mask->dst_port_mask,
5000 attributes->items[2] = (struct rte_flow_item){
5001 .type = RTE_FLOW_ITEM_TYPE_UDP,
5002 .spec = &attributes->l4,
5003 .mask = &attributes->l4_mask,
5006 case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
5007 attributes->l4.tcp.hdr = (struct rte_tcp_hdr){
5008 .src_port = input->flow.tcp6_flow.src_port,
5009 .dst_port = input->flow.tcp6_flow.dst_port,
5011 attributes->l4_mask.tcp.hdr = (struct rte_tcp_hdr){
5012 .src_port = mask->src_port_mask,
5013 .dst_port = mask->dst_port_mask,
5015 attributes->items[2] = (struct rte_flow_item){
5016 .type = RTE_FLOW_ITEM_TYPE_TCP,
5017 .spec = &attributes->l4,
5018 .mask = &attributes->l4_mask,
5021 case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
5022 case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
5025 DRV_LOG(ERR, "port %u invalid flow type%d",
5026 dev->data->port_id, fdir_filter->input.flow_type);
5027 rte_errno = ENOTSUP;
5033 #define FLOW_FDIR_CMP(f1, f2, fld) \
5034 memcmp(&(f1)->fld, &(f2)->fld, sizeof(f1->fld))
5037 * Compare two FDIR flows. If items and actions are identical, the two flows are
5041 * Pointer to Ethernet device.
5043 * FDIR flow to compare.
5045 * FDIR flow to compare.
5048 * Zero on match, 1 otherwise.
5051 flow_fdir_cmp(const struct mlx5_fdir *f1, const struct mlx5_fdir *f2)
5053 if (FLOW_FDIR_CMP(f1, f2, attr) ||
5054 FLOW_FDIR_CMP(f1, f2, l2) ||
5055 FLOW_FDIR_CMP(f1, f2, l2_mask) ||
5056 FLOW_FDIR_CMP(f1, f2, l3) ||
5057 FLOW_FDIR_CMP(f1, f2, l3_mask) ||
5058 FLOW_FDIR_CMP(f1, f2, l4) ||
5059 FLOW_FDIR_CMP(f1, f2, l4_mask) ||
5060 FLOW_FDIR_CMP(f1, f2, actions[0].type))
5062 if (f1->actions[0].type == RTE_FLOW_ACTION_TYPE_QUEUE &&
5063 FLOW_FDIR_CMP(f1, f2, queue))
5069 * Search device flow list to find out a matched FDIR flow.
5072 * Pointer to Ethernet device.
5074 * FDIR flow to lookup.
5077 * Pointer of flow if found, NULL otherwise.
5079 static struct rte_flow *
5080 flow_fdir_filter_lookup(struct rte_eth_dev *dev, struct mlx5_fdir *fdir_flow)
5082 struct mlx5_priv *priv = dev->data->dev_private;
5083 struct rte_flow *flow = NULL;
5085 MLX5_ASSERT(fdir_flow);
5086 TAILQ_FOREACH(flow, &priv->flows, next) {
5087 if (flow->fdir && !flow_fdir_cmp(flow->fdir, fdir_flow)) {
5088 DRV_LOG(DEBUG, "port %u found FDIR flow %p",
5089 dev->data->port_id, (void *)flow);
5097 * Add new flow director filter and store it in list.
5100 * Pointer to Ethernet device.
5101 * @param fdir_filter
5102 * Flow director filter to add.
5105 * 0 on success, a negative errno value otherwise and rte_errno is set.
5108 flow_fdir_filter_add(struct rte_eth_dev *dev,
5109 const struct rte_eth_fdir_filter *fdir_filter)
5111 struct mlx5_priv *priv = dev->data->dev_private;
5112 struct mlx5_fdir *fdir_flow;
5113 struct rte_flow *flow;
5116 fdir_flow = rte_zmalloc(__func__, sizeof(*fdir_flow), 0);
5121 ret = flow_fdir_filter_convert(dev, fdir_filter, fdir_flow);
5124 flow = flow_fdir_filter_lookup(dev, fdir_flow);
5129 flow = flow_list_create(dev, &priv->flows, &fdir_flow->attr,
5130 fdir_flow->items, fdir_flow->actions, true,
5134 MLX5_ASSERT(!flow->fdir);
5135 flow->fdir = fdir_flow;
5136 DRV_LOG(DEBUG, "port %u created FDIR flow %p",
5137 dev->data->port_id, (void *)flow);
5140 rte_free(fdir_flow);
5145 * Delete specific filter.
5148 * Pointer to Ethernet device.
5149 * @param fdir_filter
5150 * Filter to be deleted.
5153 * 0 on success, a negative errno value otherwise and rte_errno is set.
5156 flow_fdir_filter_delete(struct rte_eth_dev *dev,
5157 const struct rte_eth_fdir_filter *fdir_filter)
5159 struct mlx5_priv *priv = dev->data->dev_private;
5160 struct rte_flow *flow;
5161 struct mlx5_fdir fdir_flow = {
5166 ret = flow_fdir_filter_convert(dev, fdir_filter, &fdir_flow);
5169 flow = flow_fdir_filter_lookup(dev, &fdir_flow);
5174 flow_list_destroy(dev, &priv->flows, flow);
5175 DRV_LOG(DEBUG, "port %u deleted FDIR flow %p",
5176 dev->data->port_id, (void *)flow);
5181 * Update queue for specific filter.
5184 * Pointer to Ethernet device.
5185 * @param fdir_filter
5186 * Filter to be updated.
5189 * 0 on success, a negative errno value otherwise and rte_errno is set.
5192 flow_fdir_filter_update(struct rte_eth_dev *dev,
5193 const struct rte_eth_fdir_filter *fdir_filter)
5197 ret = flow_fdir_filter_delete(dev, fdir_filter);
5200 return flow_fdir_filter_add(dev, fdir_filter);
5204 * Flush all filters.
5207 * Pointer to Ethernet device.
5210 flow_fdir_filter_flush(struct rte_eth_dev *dev)
5212 struct mlx5_priv *priv = dev->data->dev_private;
5214 mlx5_flow_list_flush(dev, &priv->flows);
5218 * Get flow director information.
5221 * Pointer to Ethernet device.
5222 * @param[out] fdir_info
5223 * Resulting flow director information.
5226 flow_fdir_info_get(struct rte_eth_dev *dev, struct rte_eth_fdir_info *fdir_info)
5228 struct rte_eth_fdir_masks *mask =
5229 &dev->data->dev_conf.fdir_conf.mask;
5231 fdir_info->mode = dev->data->dev_conf.fdir_conf.mode;
5232 fdir_info->guarant_spc = 0;
5233 rte_memcpy(&fdir_info->mask, mask, sizeof(fdir_info->mask));
5234 fdir_info->max_flexpayload = 0;
5235 fdir_info->flow_types_mask[0] = 0;
5236 fdir_info->flex_payload_unit = 0;
5237 fdir_info->max_flex_payload_segment_num = 0;
5238 fdir_info->flex_payload_limit = 0;
5239 memset(&fdir_info->flex_conf, 0, sizeof(fdir_info->flex_conf));
5243 * Deal with flow director operations.
5246 * Pointer to Ethernet device.
5248 * Operation to perform.
5250 * Pointer to operation-specific structure.
5253 * 0 on success, a negative errno value otherwise and rte_errno is set.
5256 flow_fdir_ctrl_func(struct rte_eth_dev *dev, enum rte_filter_op filter_op,
5259 enum rte_fdir_mode fdir_mode =
5260 dev->data->dev_conf.fdir_conf.mode;
5262 if (filter_op == RTE_ETH_FILTER_NOP)
5264 if (fdir_mode != RTE_FDIR_MODE_PERFECT &&
5265 fdir_mode != RTE_FDIR_MODE_PERFECT_MAC_VLAN) {
5266 DRV_LOG(ERR, "port %u flow director mode %d not supported",
5267 dev->data->port_id, fdir_mode);
5271 switch (filter_op) {
5272 case RTE_ETH_FILTER_ADD:
5273 return flow_fdir_filter_add(dev, arg);
5274 case RTE_ETH_FILTER_UPDATE:
5275 return flow_fdir_filter_update(dev, arg);
5276 case RTE_ETH_FILTER_DELETE:
5277 return flow_fdir_filter_delete(dev, arg);
5278 case RTE_ETH_FILTER_FLUSH:
5279 flow_fdir_filter_flush(dev);
5281 case RTE_ETH_FILTER_INFO:
5282 flow_fdir_info_get(dev, arg);
5285 DRV_LOG(DEBUG, "port %u unknown operation %u",
5286 dev->data->port_id, filter_op);
5294 * Manage filter operations.
5297 * Pointer to Ethernet device structure.
5298 * @param filter_type
5301 * Operation to perform.
5303 * Pointer to operation-specific structure.
5306 * 0 on success, a negative errno value otherwise and rte_errno is set.
5309 mlx5_dev_filter_ctrl(struct rte_eth_dev *dev,
5310 enum rte_filter_type filter_type,
5311 enum rte_filter_op filter_op,
5314 switch (filter_type) {
5315 case RTE_ETH_FILTER_GENERIC:
5316 if (filter_op != RTE_ETH_FILTER_GET) {
5320 *(const void **)arg = &mlx5_flow_ops;
5322 case RTE_ETH_FILTER_FDIR:
5323 return flow_fdir_ctrl_func(dev, filter_op, arg);
5325 DRV_LOG(ERR, "port %u filter type (%d) not supported",
5326 dev->data->port_id, filter_type);
5327 rte_errno = ENOTSUP;
5334 * Create the needed meter and suffix tables.
5337 * Pointer to Ethernet device.
5339 * Pointer to the flow meter.
5342 * Pointer to table set on success, NULL otherwise.
5344 struct mlx5_meter_domains_infos *
5345 mlx5_flow_create_mtr_tbls(struct rte_eth_dev *dev,
5346 const struct mlx5_flow_meter *fm)
5348 const struct mlx5_flow_driver_ops *fops;
5350 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5351 return fops->create_mtr_tbls(dev, fm);
5355 * Destroy the meter table set.
5358 * Pointer to Ethernet device.
5360 * Pointer to the meter table set.
5366 mlx5_flow_destroy_mtr_tbls(struct rte_eth_dev *dev,
5367 struct mlx5_meter_domains_infos *tbls)
5369 const struct mlx5_flow_driver_ops *fops;
5371 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5372 return fops->destroy_mtr_tbls(dev, tbls);
5376 * Create policer rules.
5379 * Pointer to Ethernet device.
5381 * Pointer to flow meter structure.
5383 * Pointer to flow attributes.
5386 * 0 on success, -1 otherwise.
5389 mlx5_flow_create_policer_rules(struct rte_eth_dev *dev,
5390 struct mlx5_flow_meter *fm,
5391 const struct rte_flow_attr *attr)
5393 const struct mlx5_flow_driver_ops *fops;
5395 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5396 return fops->create_policer_rules(dev, fm, attr);
5400 * Destroy policer rules.
5403 * Pointer to flow meter structure.
5405 * Pointer to flow attributes.
5408 * 0 on success, -1 otherwise.
5411 mlx5_flow_destroy_policer_rules(struct rte_eth_dev *dev,
5412 struct mlx5_flow_meter *fm,
5413 const struct rte_flow_attr *attr)
5415 const struct mlx5_flow_driver_ops *fops;
5417 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5418 return fops->destroy_policer_rules(dev, fm, attr);
5422 * Allocate a counter.
5425 * Pointer to Ethernet device structure.
5428 * Pointer to allocated counter on success, NULL otherwise.
5430 struct mlx5_flow_counter *
5431 mlx5_counter_alloc(struct rte_eth_dev *dev)
5433 const struct mlx5_flow_driver_ops *fops;
5434 struct rte_flow_attr attr = { .transfer = 0 };
5436 if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
5437 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5438 return fops->counter_alloc(dev);
5441 "port %u counter allocate is not supported.",
5442 dev->data->port_id);
5450 * Pointer to Ethernet device structure.
5452 * Pointer to counter to be free.
5455 mlx5_counter_free(struct rte_eth_dev *dev, struct mlx5_flow_counter *cnt)
5457 const struct mlx5_flow_driver_ops *fops;
5458 struct rte_flow_attr attr = { .transfer = 0 };
5460 if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
5461 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5462 fops->counter_free(dev, cnt);
5466 "port %u counter free is not supported.",
5467 dev->data->port_id);
5471 * Query counter statistics.
5474 * Pointer to Ethernet device structure.
5476 * Pointer to counter to query.
5478 * Set to clear counter statistics.
5480 * The counter hits packets number to save.
5482 * The counter hits bytes number to save.
5485 * 0 on success, a negative errno value otherwise.
5488 mlx5_counter_query(struct rte_eth_dev *dev, struct mlx5_flow_counter *cnt,
5489 bool clear, uint64_t *pkts, uint64_t *bytes)
5491 const struct mlx5_flow_driver_ops *fops;
5492 struct rte_flow_attr attr = { .transfer = 0 };
5494 if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
5495 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5496 return fops->counter_query(dev, cnt, clear, pkts, bytes);
5499 "port %u counter query is not supported.",
5500 dev->data->port_id);
5504 #define MLX5_POOL_QUERY_FREQ_US 1000000
5507 * Set the periodic procedure for triggering asynchronous batch queries for all
5508 * the counter pools.
5511 * Pointer to mlx5_ibv_shared object.
5514 mlx5_set_query_alarm(struct mlx5_ibv_shared *sh)
5516 struct mlx5_pools_container *cont = MLX5_CNT_CONTAINER(sh, 0, 0);
5517 uint32_t pools_n = rte_atomic16_read(&cont->n_valid);
5520 cont = MLX5_CNT_CONTAINER(sh, 1, 0);
5521 pools_n += rte_atomic16_read(&cont->n_valid);
5522 us = MLX5_POOL_QUERY_FREQ_US / pools_n;
5523 DRV_LOG(DEBUG, "Set alarm for %u pools each %u us", pools_n, us);
5524 if (rte_eal_alarm_set(us, mlx5_flow_query_alarm, sh)) {
5525 sh->cmng.query_thread_on = 0;
5526 DRV_LOG(ERR, "Cannot reinitialize query alarm");
5528 sh->cmng.query_thread_on = 1;
5533 * The periodic procedure for triggering asynchronous batch queries for all the
5534 * counter pools. This function is probably called by the host thread.
5537 * The parameter for the alarm process.
5540 mlx5_flow_query_alarm(void *arg)
5542 struct mlx5_ibv_shared *sh = arg;
5543 struct mlx5_devx_obj *dcs;
5546 uint8_t batch = sh->cmng.batch;
5547 uint16_t pool_index = sh->cmng.pool_index;
5548 struct mlx5_pools_container *cont;
5549 struct mlx5_pools_container *mcont;
5550 struct mlx5_flow_counter_pool *pool;
5552 if (sh->cmng.pending_queries >= MLX5_MAX_PENDING_QUERIES)
5555 cont = MLX5_CNT_CONTAINER(sh, batch, 1);
5556 mcont = MLX5_CNT_CONTAINER(sh, batch, 0);
5557 /* Check if resize was done and need to flip a container. */
5558 if (cont != mcont) {
5560 /* Clean the old container. */
5561 rte_free(cont->pools);
5562 memset(cont, 0, sizeof(*cont));
5565 /* Flip the host container. */
5566 sh->cmng.mhi[batch] ^= (uint8_t)2;
5570 /* 2 empty containers case is unexpected. */
5571 if (unlikely(batch != sh->cmng.batch))
5575 goto next_container;
5577 pool = cont->pools[pool_index];
5579 /* There is a pool query in progress. */
5582 LIST_FIRST(&sh->cmng.free_stat_raws);
5584 /* No free counter statistics raw memory. */
5586 dcs = (struct mlx5_devx_obj *)(uintptr_t)rte_atomic64_read
5588 offset = batch ? 0 : dcs->id % MLX5_COUNTERS_PER_POOL;
5589 ret = mlx5_devx_cmd_flow_counter_query(dcs, 0, MLX5_COUNTERS_PER_POOL -
5591 pool->raw_hw->mem_mng->dm->id,
5593 (pool->raw_hw->data + offset),
5595 (uint64_t)(uintptr_t)pool);
5597 DRV_LOG(ERR, "Failed to trigger asynchronous query for dcs ID"
5598 " %d", pool->min_dcs->id);
5599 pool->raw_hw = NULL;
5602 pool->raw_hw->min_dcs_id = dcs->id;
5603 LIST_REMOVE(pool->raw_hw, next);
5604 sh->cmng.pending_queries++;
5606 if (pool_index >= rte_atomic16_read(&cont->n_valid)) {
5611 sh->cmng.batch = batch;
5612 sh->cmng.pool_index = pool_index;
5613 mlx5_set_query_alarm(sh);
5617 * Handler for the HW respond about ready values from an asynchronous batch
5618 * query. This function is probably called by the host thread.
5621 * The pointer to the shared IB device context.
5622 * @param[in] async_id
5623 * The Devx async ID.
5625 * The status of the completion.
5628 mlx5_flow_async_pool_query_handle(struct mlx5_ibv_shared *sh,
5629 uint64_t async_id, int status)
5631 struct mlx5_flow_counter_pool *pool =
5632 (struct mlx5_flow_counter_pool *)(uintptr_t)async_id;
5633 struct mlx5_counter_stats_raw *raw_to_free;
5635 if (unlikely(status)) {
5636 raw_to_free = pool->raw_hw;
5638 raw_to_free = pool->raw;
5639 rte_spinlock_lock(&pool->sl);
5640 pool->raw = pool->raw_hw;
5641 rte_spinlock_unlock(&pool->sl);
5642 rte_atomic64_add(&pool->query_gen, 1);
5643 /* Be sure the new raw counters data is updated in memory. */
5646 LIST_INSERT_HEAD(&sh->cmng.free_stat_raws, raw_to_free, next);
5647 pool->raw_hw = NULL;
5648 sh->cmng.pending_queries--;
5652 * Translate the rte_flow group index to HW table value.
5654 * @param[in] attributes
5655 * Pointer to flow attributes
5656 * @param[in] external
5657 * Value is part of flow rule created by request external to PMD.
5659 * rte_flow group index value.
5660 * @param[out] fdb_def_rule
5661 * Whether fdb jump to table 1 is configured.
5665 * Pointer to error structure.
5668 * 0 on success, a negative errno value otherwise and rte_errno is set.
5671 mlx5_flow_group_to_table(const struct rte_flow_attr *attributes, bool external,
5672 uint32_t group, bool fdb_def_rule, uint32_t *table,
5673 struct rte_flow_error *error)
5675 if (attributes->transfer && external && fdb_def_rule) {
5676 if (group == UINT32_MAX)
5677 return rte_flow_error_set
5679 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
5681 "group index not supported");
5690 * Discover availability of metadata reg_c's.
5692 * Iteratively use test flows to check availability.
5695 * Pointer to the Ethernet device structure.
5698 * 0 on success, a negative errno value otherwise and rte_errno is set.
5701 mlx5_flow_discover_mreg_c(struct rte_eth_dev *dev)
5703 struct mlx5_priv *priv = dev->data->dev_private;
5704 struct mlx5_dev_config *config = &priv->config;
5705 enum modify_reg idx;
5708 /* reg_c[0] and reg_c[1] are reserved. */
5709 config->flow_mreg_c[n++] = REG_C_0;
5710 config->flow_mreg_c[n++] = REG_C_1;
5711 /* Discover availability of other reg_c's. */
5712 for (idx = REG_C_2; idx <= REG_C_7; ++idx) {
5713 struct rte_flow_attr attr = {
5714 .group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
5715 .priority = MLX5_FLOW_PRIO_RSVD,
5718 struct rte_flow_item items[] = {
5720 .type = RTE_FLOW_ITEM_TYPE_END,
5723 struct rte_flow_action actions[] = {
5725 .type = MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
5726 .conf = &(struct mlx5_flow_action_copy_mreg){
5732 .type = RTE_FLOW_ACTION_TYPE_JUMP,
5733 .conf = &(struct rte_flow_action_jump){
5734 .group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
5738 .type = RTE_FLOW_ACTION_TYPE_END,
5741 struct rte_flow *flow;
5742 struct rte_flow_error error;
5744 if (!config->dv_flow_en)
5746 /* Create internal flow, validation skips copy action. */
5747 flow = flow_list_create(dev, NULL, &attr, items,
5748 actions, false, &error);
5751 if (dev->data->dev_started || !flow_drv_apply(dev, flow, NULL))
5752 config->flow_mreg_c[n++] = idx;
5753 flow_list_destroy(dev, NULL, flow);
5755 for (; n < MLX5_MREG_C_NUM; ++n)
5756 config->flow_mreg_c[n] = REG_NONE;
5761 * Dump flow raw hw data to file
5764 * The pointer to Ethernet device.
5766 * A pointer to a file for output.
5768 * Perform verbose error reporting if not NULL. PMDs initialize this
5769 * structure in case of error only.
5771 * 0 on success, a nagative value otherwise.
5774 mlx5_flow_dev_dump(struct rte_eth_dev *dev,
5776 struct rte_flow_error *error __rte_unused)
5778 struct mlx5_priv *priv = dev->data->dev_private;
5779 struct mlx5_ibv_shared *sh = priv->sh;
5781 return mlx5_devx_cmd_flow_dump(sh->fdb_domain, sh->rx_domain,
5782 sh->tx_domain, file);