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),
171 .type = RTE_FLOW_ITEM_TYPE_VXLAN,
173 [MLX5_EXPANSION_VXLAN_GPE] = {
174 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_ETH,
176 MLX5_EXPANSION_IPV6),
177 .type = RTE_FLOW_ITEM_TYPE_VXLAN_GPE,
179 [MLX5_EXPANSION_GRE] = {
180 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4),
181 .type = RTE_FLOW_ITEM_TYPE_GRE,
183 [MLX5_EXPANSION_MPLS] = {
184 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
185 MLX5_EXPANSION_IPV6),
186 .type = RTE_FLOW_ITEM_TYPE_MPLS,
188 [MLX5_EXPANSION_ETH] = {
189 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
190 MLX5_EXPANSION_IPV6),
191 .type = RTE_FLOW_ITEM_TYPE_ETH,
193 [MLX5_EXPANSION_ETH_VLAN] = {
194 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_VLAN),
195 .type = RTE_FLOW_ITEM_TYPE_ETH,
197 [MLX5_EXPANSION_VLAN] = {
198 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4,
199 MLX5_EXPANSION_IPV6),
200 .type = RTE_FLOW_ITEM_TYPE_VLAN,
202 [MLX5_EXPANSION_IPV4] = {
203 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV4_UDP,
204 MLX5_EXPANSION_IPV4_TCP),
205 .type = RTE_FLOW_ITEM_TYPE_IPV4,
206 .rss_types = ETH_RSS_IPV4 | ETH_RSS_FRAG_IPV4 |
207 ETH_RSS_NONFRAG_IPV4_OTHER,
209 [MLX5_EXPANSION_IPV4_UDP] = {
210 .type = RTE_FLOW_ITEM_TYPE_UDP,
211 .rss_types = ETH_RSS_NONFRAG_IPV4_UDP,
213 [MLX5_EXPANSION_IPV4_TCP] = {
214 .type = RTE_FLOW_ITEM_TYPE_TCP,
215 .rss_types = ETH_RSS_NONFRAG_IPV4_TCP,
217 [MLX5_EXPANSION_IPV6] = {
218 .next = RTE_FLOW_EXPAND_RSS_NEXT(MLX5_EXPANSION_IPV6_UDP,
219 MLX5_EXPANSION_IPV6_TCP),
220 .type = RTE_FLOW_ITEM_TYPE_IPV6,
221 .rss_types = ETH_RSS_IPV6 | ETH_RSS_FRAG_IPV6 |
222 ETH_RSS_NONFRAG_IPV6_OTHER,
224 [MLX5_EXPANSION_IPV6_UDP] = {
225 .type = RTE_FLOW_ITEM_TYPE_UDP,
226 .rss_types = ETH_RSS_NONFRAG_IPV6_UDP,
228 [MLX5_EXPANSION_IPV6_TCP] = {
229 .type = RTE_FLOW_ITEM_TYPE_TCP,
230 .rss_types = ETH_RSS_NONFRAG_IPV6_TCP,
234 static const struct rte_flow_ops mlx5_flow_ops = {
235 .validate = mlx5_flow_validate,
236 .create = mlx5_flow_create,
237 .destroy = mlx5_flow_destroy,
238 .flush = mlx5_flow_flush,
239 .isolate = mlx5_flow_isolate,
240 .query = mlx5_flow_query,
241 .dev_dump = mlx5_flow_dev_dump,
244 /* Convert FDIR request to Generic flow. */
246 struct rte_flow_attr attr;
247 struct rte_flow_item items[4];
248 struct rte_flow_item_eth l2;
249 struct rte_flow_item_eth l2_mask;
251 struct rte_flow_item_ipv4 ipv4;
252 struct rte_flow_item_ipv6 ipv6;
255 struct rte_flow_item_ipv4 ipv4;
256 struct rte_flow_item_ipv6 ipv6;
259 struct rte_flow_item_udp udp;
260 struct rte_flow_item_tcp tcp;
263 struct rte_flow_item_udp udp;
264 struct rte_flow_item_tcp tcp;
266 struct rte_flow_action actions[2];
267 struct rte_flow_action_queue queue;
270 /* Map of Verbs to Flow priority with 8 Verbs priorities. */
271 static const uint32_t priority_map_3[][MLX5_PRIORITY_MAP_MAX] = {
272 { 0, 1, 2 }, { 2, 3, 4 }, { 5, 6, 7 },
275 /* Map of Verbs to Flow priority with 16 Verbs priorities. */
276 static const uint32_t priority_map_5[][MLX5_PRIORITY_MAP_MAX] = {
277 { 0, 1, 2 }, { 3, 4, 5 }, { 6, 7, 8 },
278 { 9, 10, 11 }, { 12, 13, 14 },
281 /* Tunnel information. */
282 struct mlx5_flow_tunnel_info {
283 uint64_t tunnel; /**< Tunnel bit (see MLX5_FLOW_*). */
284 uint32_t ptype; /**< Tunnel Ptype (see RTE_PTYPE_*). */
287 static struct mlx5_flow_tunnel_info tunnels_info[] = {
289 .tunnel = MLX5_FLOW_LAYER_VXLAN,
290 .ptype = RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L4_UDP,
293 .tunnel = MLX5_FLOW_LAYER_GENEVE,
294 .ptype = RTE_PTYPE_TUNNEL_GENEVE | RTE_PTYPE_L4_UDP,
297 .tunnel = MLX5_FLOW_LAYER_VXLAN_GPE,
298 .ptype = RTE_PTYPE_TUNNEL_VXLAN_GPE | RTE_PTYPE_L4_UDP,
301 .tunnel = MLX5_FLOW_LAYER_GRE,
302 .ptype = RTE_PTYPE_TUNNEL_GRE,
305 .tunnel = MLX5_FLOW_LAYER_MPLS | MLX5_FLOW_LAYER_OUTER_L4_UDP,
306 .ptype = RTE_PTYPE_TUNNEL_MPLS_IN_UDP | RTE_PTYPE_L4_UDP,
309 .tunnel = MLX5_FLOW_LAYER_MPLS,
310 .ptype = RTE_PTYPE_TUNNEL_MPLS_IN_GRE,
313 .tunnel = MLX5_FLOW_LAYER_NVGRE,
314 .ptype = RTE_PTYPE_TUNNEL_NVGRE,
317 .tunnel = MLX5_FLOW_LAYER_IPIP,
318 .ptype = RTE_PTYPE_TUNNEL_IP,
321 .tunnel = MLX5_FLOW_LAYER_IPV6_ENCAP,
322 .ptype = RTE_PTYPE_TUNNEL_IP,
325 .tunnel = MLX5_FLOW_LAYER_GTP,
326 .ptype = RTE_PTYPE_TUNNEL_GTPU,
331 * Translate tag ID to register.
334 * Pointer to the Ethernet device structure.
336 * The feature that request the register.
338 * The request register ID.
340 * Error description in case of any.
343 * The request register on success, a negative errno
344 * value otherwise and rte_errno is set.
347 mlx5_flow_get_reg_id(struct rte_eth_dev *dev,
348 enum mlx5_feature_name feature,
350 struct rte_flow_error *error)
352 struct mlx5_priv *priv = dev->data->dev_private;
353 struct mlx5_dev_config *config = &priv->config;
354 enum modify_reg start_reg;
355 bool skip_mtr_reg = false;
358 case MLX5_HAIRPIN_RX:
360 case MLX5_HAIRPIN_TX:
362 case MLX5_METADATA_RX:
363 switch (config->dv_xmeta_en) {
364 case MLX5_XMETA_MODE_LEGACY:
366 case MLX5_XMETA_MODE_META16:
368 case MLX5_XMETA_MODE_META32:
372 case MLX5_METADATA_TX:
374 case MLX5_METADATA_FDB:
375 switch (config->dv_xmeta_en) {
376 case MLX5_XMETA_MODE_LEGACY:
378 case MLX5_XMETA_MODE_META16:
380 case MLX5_XMETA_MODE_META32:
385 switch (config->dv_xmeta_en) {
386 case MLX5_XMETA_MODE_LEGACY:
388 case MLX5_XMETA_MODE_META16:
390 case MLX5_XMETA_MODE_META32:
396 * If meter color and flow match share one register, flow match
397 * should use the meter color register for match.
399 if (priv->mtr_reg_share)
400 return priv->mtr_color_reg;
402 return priv->mtr_color_reg != REG_C_2 ? REG_C_2 :
405 RTE_ASSERT(priv->mtr_color_reg != REG_NONE);
406 return priv->mtr_color_reg;
409 * Metadata COPY_MARK register using is in meter suffix sub
410 * flow while with meter. It's safe to share the same register.
412 return priv->mtr_color_reg != REG_C_2 ? REG_C_2 : REG_C_3;
415 * If meter is enable, it will engage the register for color
416 * match and flow match. If meter color match is not using the
417 * REG_C_2, need to skip the REG_C_x be used by meter color
419 * If meter is disable, free to use all available registers.
421 start_reg = priv->mtr_color_reg != REG_C_2 ? REG_C_2 :
422 (priv->mtr_reg_share ? REG_C_3 : REG_C_4);
423 skip_mtr_reg = !!(priv->mtr_en && start_reg == REG_C_2);
424 if (id > (REG_C_7 - start_reg))
425 return rte_flow_error_set(error, EINVAL,
426 RTE_FLOW_ERROR_TYPE_ITEM,
427 NULL, "invalid tag id");
428 if (config->flow_mreg_c[id + start_reg - REG_C_0] == REG_NONE)
429 return rte_flow_error_set(error, ENOTSUP,
430 RTE_FLOW_ERROR_TYPE_ITEM,
431 NULL, "unsupported tag id");
433 * This case means meter is using the REG_C_x great than 2.
434 * Take care not to conflict with meter color REG_C_x.
435 * If the available index REG_C_y >= REG_C_x, skip the
438 if (skip_mtr_reg && config->flow_mreg_c
439 [id + start_reg - REG_C_0] >= priv->mtr_color_reg) {
440 if (config->flow_mreg_c
441 [id + 1 + start_reg - REG_C_0] != REG_NONE)
442 return config->flow_mreg_c
443 [id + 1 + start_reg - REG_C_0];
444 return rte_flow_error_set(error, ENOTSUP,
445 RTE_FLOW_ERROR_TYPE_ITEM,
446 NULL, "unsupported tag id");
448 return config->flow_mreg_c[id + start_reg - REG_C_0];
451 return rte_flow_error_set(error, EINVAL,
452 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
453 NULL, "invalid feature name");
457 * Check extensive flow metadata register support.
460 * Pointer to rte_eth_dev structure.
463 * True if device supports extensive flow metadata register, otherwise false.
466 mlx5_flow_ext_mreg_supported(struct rte_eth_dev *dev)
468 struct mlx5_priv *priv = dev->data->dev_private;
469 struct mlx5_dev_config *config = &priv->config;
472 * Having available reg_c can be regarded inclusively as supporting
473 * extensive flow metadata register, which could mean,
474 * - metadata register copy action by modify header.
475 * - 16 modify header actions is supported.
476 * - reg_c's are preserved across different domain (FDB and NIC) on
477 * packet loopback by flow lookup miss.
479 return config->flow_mreg_c[2] != REG_NONE;
483 * Discover the maximum number of priority available.
486 * Pointer to the Ethernet device structure.
489 * number of supported flow priority on success, a negative errno
490 * value otherwise and rte_errno is set.
493 mlx5_flow_discover_priorities(struct rte_eth_dev *dev)
495 struct mlx5_priv *priv = dev->data->dev_private;
497 struct ibv_flow_attr attr;
498 struct ibv_flow_spec_eth eth;
499 struct ibv_flow_spec_action_drop drop;
503 .port = (uint8_t)priv->ibv_port,
506 .type = IBV_FLOW_SPEC_ETH,
507 .size = sizeof(struct ibv_flow_spec_eth),
510 .size = sizeof(struct ibv_flow_spec_action_drop),
511 .type = IBV_FLOW_SPEC_ACTION_DROP,
514 struct ibv_flow *flow;
515 struct mlx5_hrxq *drop = mlx5_hrxq_drop_new(dev);
516 uint16_t vprio[] = { 8, 16 };
524 for (i = 0; i != RTE_DIM(vprio); i++) {
525 flow_attr.attr.priority = vprio[i] - 1;
526 flow = mlx5_glue->create_flow(drop->qp, &flow_attr.attr);
529 claim_zero(mlx5_glue->destroy_flow(flow));
532 mlx5_hrxq_drop_release(dev);
535 priority = RTE_DIM(priority_map_3);
538 priority = RTE_DIM(priority_map_5);
543 "port %u verbs maximum priority: %d expected 8/16",
544 dev->data->port_id, priority);
547 DRV_LOG(INFO, "port %u flow maximum priority: %d",
548 dev->data->port_id, priority);
553 * Adjust flow priority based on the highest layer and the request priority.
556 * Pointer to the Ethernet device structure.
557 * @param[in] priority
558 * The rule base priority.
559 * @param[in] subpriority
560 * The priority based on the items.
565 uint32_t mlx5_flow_adjust_priority(struct rte_eth_dev *dev, int32_t priority,
566 uint32_t subpriority)
569 struct mlx5_priv *priv = dev->data->dev_private;
571 switch (priv->config.flow_prio) {
572 case RTE_DIM(priority_map_3):
573 res = priority_map_3[priority][subpriority];
575 case RTE_DIM(priority_map_5):
576 res = priority_map_5[priority][subpriority];
583 * Verify the @p item specifications (spec, last, mask) are compatible with the
587 * Item specification.
589 * @p item->mask or flow default bit-masks.
590 * @param[in] nic_mask
591 * Bit-masks covering supported fields by the NIC to compare with user mask.
593 * Bit-masks size in bytes.
595 * Pointer to error structure.
598 * 0 on success, a negative errno value otherwise and rte_errno is set.
601 mlx5_flow_item_acceptable(const struct rte_flow_item *item,
603 const uint8_t *nic_mask,
605 struct rte_flow_error *error)
610 for (i = 0; i < size; ++i)
611 if ((nic_mask[i] | mask[i]) != nic_mask[i])
612 return rte_flow_error_set(error, ENOTSUP,
613 RTE_FLOW_ERROR_TYPE_ITEM,
615 "mask enables non supported"
617 if (!item->spec && (item->mask || item->last))
618 return rte_flow_error_set(error, EINVAL,
619 RTE_FLOW_ERROR_TYPE_ITEM, item,
620 "mask/last without a spec is not"
622 if (item->spec && item->last) {
628 for (i = 0; i < size; ++i) {
629 spec[i] = ((const uint8_t *)item->spec)[i] & mask[i];
630 last[i] = ((const uint8_t *)item->last)[i] & mask[i];
632 ret = memcmp(spec, last, size);
634 return rte_flow_error_set(error, EINVAL,
635 RTE_FLOW_ERROR_TYPE_ITEM,
637 "range is not valid");
643 * Adjust the hash fields according to the @p flow information.
645 * @param[in] dev_flow.
646 * Pointer to the mlx5_flow.
648 * 1 when the hash field is for a tunnel item.
649 * @param[in] layer_types
651 * @param[in] hash_fields
655 * The hash fields that should be used.
658 mlx5_flow_hashfields_adjust(struct mlx5_flow *dev_flow,
659 int tunnel __rte_unused, uint64_t layer_types,
660 uint64_t hash_fields)
662 struct rte_flow *flow = dev_flow->flow;
663 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
664 int rss_request_inner = flow->rss.level >= 2;
666 /* Check RSS hash level for tunnel. */
667 if (tunnel && rss_request_inner)
668 hash_fields |= IBV_RX_HASH_INNER;
669 else if (tunnel || rss_request_inner)
672 /* Check if requested layer matches RSS hash fields. */
673 if (!(flow->rss.types & layer_types))
679 * Lookup and set the ptype in the data Rx part. A single Ptype can be used,
680 * if several tunnel rules are used on this queue, the tunnel ptype will be
684 * Rx queue to update.
687 flow_rxq_tunnel_ptype_update(struct mlx5_rxq_ctrl *rxq_ctrl)
690 uint32_t tunnel_ptype = 0;
692 /* Look up for the ptype to use. */
693 for (i = 0; i != MLX5_FLOW_TUNNEL; ++i) {
694 if (!rxq_ctrl->flow_tunnels_n[i])
697 tunnel_ptype = tunnels_info[i].ptype;
703 rxq_ctrl->rxq.tunnel = tunnel_ptype;
707 * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) according to the devive
711 * Pointer to the Ethernet device structure.
712 * @param[in] dev_flow
713 * Pointer to device flow structure.
716 flow_drv_rxq_flags_set(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow)
718 struct mlx5_priv *priv = dev->data->dev_private;
719 struct rte_flow *flow = dev_flow->flow;
720 const int mark = !!(dev_flow->actions &
721 (MLX5_FLOW_ACTION_FLAG | MLX5_FLOW_ACTION_MARK));
722 const int tunnel = !!(dev_flow->layers & MLX5_FLOW_LAYER_TUNNEL);
725 for (i = 0; i != flow->rss.queue_num; ++i) {
726 int idx = (*flow->rss.queue)[i];
727 struct mlx5_rxq_ctrl *rxq_ctrl =
728 container_of((*priv->rxqs)[idx],
729 struct mlx5_rxq_ctrl, rxq);
732 * To support metadata register copy on Tx loopback,
733 * this must be always enabled (metadata may arive
734 * from other port - not from local flows only.
736 if (priv->config.dv_flow_en &&
737 priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
738 mlx5_flow_ext_mreg_supported(dev)) {
739 rxq_ctrl->rxq.mark = 1;
740 rxq_ctrl->flow_mark_n = 1;
742 rxq_ctrl->rxq.mark = 1;
743 rxq_ctrl->flow_mark_n++;
748 /* Increase the counter matching the flow. */
749 for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
750 if ((tunnels_info[j].tunnel &
752 tunnels_info[j].tunnel) {
753 rxq_ctrl->flow_tunnels_n[j]++;
757 flow_rxq_tunnel_ptype_update(rxq_ctrl);
763 * Set the Rx queue flags (Mark/Flag and Tunnel Ptypes) for a flow
766 * Pointer to the Ethernet device structure.
768 * Pointer to flow structure.
771 flow_rxq_flags_set(struct rte_eth_dev *dev, struct rte_flow *flow)
773 struct mlx5_flow *dev_flow;
775 LIST_FOREACH(dev_flow, &flow->dev_flows, next)
776 flow_drv_rxq_flags_set(dev, dev_flow);
780 * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
781 * device flow if no other flow uses it with the same kind of request.
784 * Pointer to Ethernet device.
785 * @param[in] dev_flow
786 * Pointer to the device flow.
789 flow_drv_rxq_flags_trim(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow)
791 struct mlx5_priv *priv = dev->data->dev_private;
792 struct rte_flow *flow = dev_flow->flow;
793 const int mark = !!(dev_flow->actions &
794 (MLX5_FLOW_ACTION_FLAG | MLX5_FLOW_ACTION_MARK));
795 const int tunnel = !!(dev_flow->layers & MLX5_FLOW_LAYER_TUNNEL);
798 assert(dev->data->dev_started);
799 for (i = 0; i != flow->rss.queue_num; ++i) {
800 int idx = (*flow->rss.queue)[i];
801 struct mlx5_rxq_ctrl *rxq_ctrl =
802 container_of((*priv->rxqs)[idx],
803 struct mlx5_rxq_ctrl, rxq);
805 if (priv->config.dv_flow_en &&
806 priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
807 mlx5_flow_ext_mreg_supported(dev)) {
808 rxq_ctrl->rxq.mark = 1;
809 rxq_ctrl->flow_mark_n = 1;
811 rxq_ctrl->flow_mark_n--;
812 rxq_ctrl->rxq.mark = !!rxq_ctrl->flow_mark_n;
817 /* Decrease the counter matching the flow. */
818 for (j = 0; j != MLX5_FLOW_TUNNEL; ++j) {
819 if ((tunnels_info[j].tunnel &
821 tunnels_info[j].tunnel) {
822 rxq_ctrl->flow_tunnels_n[j]--;
826 flow_rxq_tunnel_ptype_update(rxq_ctrl);
832 * Clear the Rx queue flags (Mark/Flag and Tunnel Ptype) associated with the
833 * @p flow if no other flow uses it with the same kind of request.
836 * Pointer to Ethernet device.
838 * Pointer to the flow.
841 flow_rxq_flags_trim(struct rte_eth_dev *dev, struct rte_flow *flow)
843 struct mlx5_flow *dev_flow;
845 LIST_FOREACH(dev_flow, &flow->dev_flows, next)
846 flow_drv_rxq_flags_trim(dev, dev_flow);
850 * Clear the Mark/Flag and Tunnel ptype information in all Rx queues.
853 * Pointer to Ethernet device.
856 flow_rxq_flags_clear(struct rte_eth_dev *dev)
858 struct mlx5_priv *priv = dev->data->dev_private;
861 for (i = 0; i != priv->rxqs_n; ++i) {
862 struct mlx5_rxq_ctrl *rxq_ctrl;
865 if (!(*priv->rxqs)[i])
867 rxq_ctrl = container_of((*priv->rxqs)[i],
868 struct mlx5_rxq_ctrl, rxq);
869 rxq_ctrl->flow_mark_n = 0;
870 rxq_ctrl->rxq.mark = 0;
871 for (j = 0; j != MLX5_FLOW_TUNNEL; ++j)
872 rxq_ctrl->flow_tunnels_n[j] = 0;
873 rxq_ctrl->rxq.tunnel = 0;
878 * return a pointer to the desired action in the list of actions.
881 * The list of actions to search the action in.
883 * The action to find.
886 * Pointer to the action in the list, if found. NULL otherwise.
888 const struct rte_flow_action *
889 mlx5_flow_find_action(const struct rte_flow_action *actions,
890 enum rte_flow_action_type action)
894 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++)
895 if (actions->type == action)
901 * Validate the flag action.
903 * @param[in] action_flags
904 * Bit-fields that holds the actions detected until now.
906 * Attributes of flow that includes this action.
908 * Pointer to error structure.
911 * 0 on success, a negative errno value otherwise and rte_errno is set.
914 mlx5_flow_validate_action_flag(uint64_t action_flags,
915 const struct rte_flow_attr *attr,
916 struct rte_flow_error *error)
918 if (action_flags & MLX5_FLOW_ACTION_MARK)
919 return rte_flow_error_set(error, EINVAL,
920 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
921 "can't mark and flag in same flow");
922 if (action_flags & MLX5_FLOW_ACTION_FLAG)
923 return rte_flow_error_set(error, EINVAL,
924 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
926 " actions in same flow");
928 return rte_flow_error_set(error, ENOTSUP,
929 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
930 "flag action not supported for "
936 * Validate the mark action.
939 * Pointer to the queue action.
940 * @param[in] action_flags
941 * Bit-fields that holds the actions detected until now.
943 * Attributes of flow that includes this action.
945 * Pointer to error structure.
948 * 0 on success, a negative errno value otherwise and rte_errno is set.
951 mlx5_flow_validate_action_mark(const struct rte_flow_action *action,
952 uint64_t action_flags,
953 const struct rte_flow_attr *attr,
954 struct rte_flow_error *error)
956 const struct rte_flow_action_mark *mark = action->conf;
959 return rte_flow_error_set(error, EINVAL,
960 RTE_FLOW_ERROR_TYPE_ACTION,
962 "configuration cannot be null");
963 if (mark->id >= MLX5_FLOW_MARK_MAX)
964 return rte_flow_error_set(error, EINVAL,
965 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
967 "mark id must in 0 <= id < "
968 RTE_STR(MLX5_FLOW_MARK_MAX));
969 if (action_flags & MLX5_FLOW_ACTION_FLAG)
970 return rte_flow_error_set(error, EINVAL,
971 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
972 "can't flag and mark in same flow");
973 if (action_flags & MLX5_FLOW_ACTION_MARK)
974 return rte_flow_error_set(error, EINVAL,
975 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
976 "can't have 2 mark actions in same"
979 return rte_flow_error_set(error, ENOTSUP,
980 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
981 "mark action not supported for "
987 * Validate the drop action.
989 * @param[in] action_flags
990 * Bit-fields that holds the actions detected until now.
992 * Attributes of flow that includes this action.
994 * Pointer to error structure.
997 * 0 on success, a negative errno value otherwise and rte_errno is set.
1000 mlx5_flow_validate_action_drop(uint64_t action_flags __rte_unused,
1001 const struct rte_flow_attr *attr,
1002 struct rte_flow_error *error)
1005 return rte_flow_error_set(error, ENOTSUP,
1006 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1007 "drop action not supported for "
1013 * Validate the queue action.
1016 * Pointer to the queue action.
1017 * @param[in] action_flags
1018 * Bit-fields that holds the actions detected until now.
1020 * Pointer to the Ethernet device structure.
1022 * Attributes of flow that includes this action.
1024 * Pointer to error structure.
1027 * 0 on success, a negative errno value otherwise and rte_errno is set.
1030 mlx5_flow_validate_action_queue(const struct rte_flow_action *action,
1031 uint64_t action_flags,
1032 struct rte_eth_dev *dev,
1033 const struct rte_flow_attr *attr,
1034 struct rte_flow_error *error)
1036 struct mlx5_priv *priv = dev->data->dev_private;
1037 const struct rte_flow_action_queue *queue = action->conf;
1039 if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1040 return rte_flow_error_set(error, EINVAL,
1041 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1042 "can't have 2 fate actions in"
1045 return rte_flow_error_set(error, EINVAL,
1046 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1047 NULL, "No Rx queues configured");
1048 if (queue->index >= priv->rxqs_n)
1049 return rte_flow_error_set(error, EINVAL,
1050 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1052 "queue index out of range");
1053 if (!(*priv->rxqs)[queue->index])
1054 return rte_flow_error_set(error, EINVAL,
1055 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1057 "queue is not configured");
1059 return rte_flow_error_set(error, ENOTSUP,
1060 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1061 "queue action not supported for "
1067 * Validate the rss action.
1070 * Pointer to the queue action.
1071 * @param[in] action_flags
1072 * Bit-fields that holds the actions detected until now.
1074 * Pointer to the Ethernet device structure.
1076 * Attributes of flow that includes this action.
1077 * @param[in] item_flags
1078 * Items that were detected.
1080 * Pointer to error structure.
1083 * 0 on success, a negative errno value otherwise and rte_errno is set.
1086 mlx5_flow_validate_action_rss(const struct rte_flow_action *action,
1087 uint64_t action_flags,
1088 struct rte_eth_dev *dev,
1089 const struct rte_flow_attr *attr,
1090 uint64_t item_flags,
1091 struct rte_flow_error *error)
1093 struct mlx5_priv *priv = dev->data->dev_private;
1094 const struct rte_flow_action_rss *rss = action->conf;
1095 int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1098 if (action_flags & MLX5_FLOW_FATE_ACTIONS)
1099 return rte_flow_error_set(error, EINVAL,
1100 RTE_FLOW_ERROR_TYPE_ACTION, NULL,
1101 "can't have 2 fate actions"
1103 if (rss->func != RTE_ETH_HASH_FUNCTION_DEFAULT &&
1104 rss->func != RTE_ETH_HASH_FUNCTION_TOEPLITZ)
1105 return rte_flow_error_set(error, ENOTSUP,
1106 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1108 "RSS hash function not supported");
1109 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
1114 return rte_flow_error_set(error, ENOTSUP,
1115 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1117 "tunnel RSS is not supported");
1118 /* allow RSS key_len 0 in case of NULL (default) RSS key. */
1119 if (rss->key_len == 0 && rss->key != NULL)
1120 return rte_flow_error_set(error, ENOTSUP,
1121 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1123 "RSS hash key length 0");
1124 if (rss->key_len > 0 && rss->key_len < MLX5_RSS_HASH_KEY_LEN)
1125 return rte_flow_error_set(error, ENOTSUP,
1126 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1128 "RSS hash key too small");
1129 if (rss->key_len > MLX5_RSS_HASH_KEY_LEN)
1130 return rte_flow_error_set(error, ENOTSUP,
1131 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1133 "RSS hash key too large");
1134 if (rss->queue_num > priv->config.ind_table_max_size)
1135 return rte_flow_error_set(error, ENOTSUP,
1136 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1138 "number of queues too large");
1139 if (rss->types & MLX5_RSS_HF_MASK)
1140 return rte_flow_error_set(error, ENOTSUP,
1141 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1143 "some RSS protocols are not"
1145 if ((rss->types & (ETH_RSS_L3_SRC_ONLY | ETH_RSS_L3_DST_ONLY)) &&
1146 !(rss->types & ETH_RSS_IP))
1147 return rte_flow_error_set(error, EINVAL,
1148 RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1149 "L3 partial RSS requested but L3 RSS"
1150 " type not specified");
1151 if ((rss->types & (ETH_RSS_L4_SRC_ONLY | ETH_RSS_L4_DST_ONLY)) &&
1152 !(rss->types & (ETH_RSS_UDP | ETH_RSS_TCP)))
1153 return rte_flow_error_set(error, EINVAL,
1154 RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1155 "L4 partial RSS requested but L4 RSS"
1156 " type not specified");
1158 return rte_flow_error_set(error, EINVAL,
1159 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1160 NULL, "No Rx queues configured");
1161 if (!rss->queue_num)
1162 return rte_flow_error_set(error, EINVAL,
1163 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1164 NULL, "No queues configured");
1165 for (i = 0; i != rss->queue_num; ++i) {
1166 if (rss->queue[i] >= priv->rxqs_n)
1167 return rte_flow_error_set
1169 RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1170 &rss->queue[i], "queue index out of range");
1171 if (!(*priv->rxqs)[rss->queue[i]])
1172 return rte_flow_error_set
1173 (error, EINVAL, RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1174 &rss->queue[i], "queue is not configured");
1177 return rte_flow_error_set(error, ENOTSUP,
1178 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1179 "rss action not supported for "
1181 if (rss->level > 1 && !tunnel)
1182 return rte_flow_error_set(error, EINVAL,
1183 RTE_FLOW_ERROR_TYPE_ACTION_CONF, NULL,
1184 "inner RSS is not supported for "
1185 "non-tunnel flows");
1190 * Validate the count action.
1193 * Pointer to the Ethernet device structure.
1195 * Attributes of flow that includes this action.
1197 * Pointer to error structure.
1200 * 0 on success, a negative errno value otherwise and rte_errno is set.
1203 mlx5_flow_validate_action_count(struct rte_eth_dev *dev __rte_unused,
1204 const struct rte_flow_attr *attr,
1205 struct rte_flow_error *error)
1208 return rte_flow_error_set(error, ENOTSUP,
1209 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1210 "count action not supported for "
1216 * Verify the @p attributes will be correctly understood by the NIC and store
1217 * them in the @p flow if everything is correct.
1220 * Pointer to the Ethernet device structure.
1221 * @param[in] attributes
1222 * Pointer to flow attributes
1224 * Pointer to error structure.
1227 * 0 on success, a negative errno value otherwise and rte_errno is set.
1230 mlx5_flow_validate_attributes(struct rte_eth_dev *dev,
1231 const struct rte_flow_attr *attributes,
1232 struct rte_flow_error *error)
1234 struct mlx5_priv *priv = dev->data->dev_private;
1235 uint32_t priority_max = priv->config.flow_prio - 1;
1237 if (attributes->group)
1238 return rte_flow_error_set(error, ENOTSUP,
1239 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
1240 NULL, "groups is not supported");
1241 if (attributes->priority != MLX5_FLOW_PRIO_RSVD &&
1242 attributes->priority >= priority_max)
1243 return rte_flow_error_set(error, ENOTSUP,
1244 RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
1245 NULL, "priority out of range");
1246 if (attributes->egress)
1247 return rte_flow_error_set(error, ENOTSUP,
1248 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, NULL,
1249 "egress is not supported");
1250 if (attributes->transfer && !priv->config.dv_esw_en)
1251 return rte_flow_error_set(error, ENOTSUP,
1252 RTE_FLOW_ERROR_TYPE_ATTR_TRANSFER,
1253 NULL, "transfer is not supported");
1254 if (!attributes->ingress)
1255 return rte_flow_error_set(error, EINVAL,
1256 RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
1258 "ingress attribute is mandatory");
1263 * Validate ICMP6 item.
1266 * Item specification.
1267 * @param[in] item_flags
1268 * Bit-fields that holds the items detected until now.
1270 * Pointer to error structure.
1273 * 0 on success, a negative errno value otherwise and rte_errno is set.
1276 mlx5_flow_validate_item_icmp6(const struct rte_flow_item *item,
1277 uint64_t item_flags,
1278 uint8_t target_protocol,
1279 struct rte_flow_error *error)
1281 const struct rte_flow_item_icmp6 *mask = item->mask;
1282 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1283 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
1284 MLX5_FLOW_LAYER_OUTER_L3_IPV6;
1285 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1286 MLX5_FLOW_LAYER_OUTER_L4;
1289 if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMPV6)
1290 return rte_flow_error_set(error, EINVAL,
1291 RTE_FLOW_ERROR_TYPE_ITEM, item,
1292 "protocol filtering not compatible"
1293 " with ICMP6 layer");
1294 if (!(item_flags & l3m))
1295 return rte_flow_error_set(error, EINVAL,
1296 RTE_FLOW_ERROR_TYPE_ITEM, item,
1297 "IPv6 is mandatory to filter on"
1299 if (item_flags & l4m)
1300 return rte_flow_error_set(error, EINVAL,
1301 RTE_FLOW_ERROR_TYPE_ITEM, item,
1302 "multiple L4 layers not supported");
1304 mask = &rte_flow_item_icmp6_mask;
1305 ret = mlx5_flow_item_acceptable
1306 (item, (const uint8_t *)mask,
1307 (const uint8_t *)&rte_flow_item_icmp6_mask,
1308 sizeof(struct rte_flow_item_icmp6), error);
1315 * Validate ICMP item.
1318 * Item specification.
1319 * @param[in] item_flags
1320 * Bit-fields that holds the items detected until now.
1322 * Pointer to error structure.
1325 * 0 on success, a negative errno value otherwise and rte_errno is set.
1328 mlx5_flow_validate_item_icmp(const struct rte_flow_item *item,
1329 uint64_t item_flags,
1330 uint8_t target_protocol,
1331 struct rte_flow_error *error)
1333 const struct rte_flow_item_icmp *mask = item->mask;
1334 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1335 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
1336 MLX5_FLOW_LAYER_OUTER_L3_IPV4;
1337 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1338 MLX5_FLOW_LAYER_OUTER_L4;
1341 if (target_protocol != 0xFF && target_protocol != IPPROTO_ICMP)
1342 return rte_flow_error_set(error, EINVAL,
1343 RTE_FLOW_ERROR_TYPE_ITEM, item,
1344 "protocol filtering not compatible"
1345 " with ICMP layer");
1346 if (!(item_flags & l3m))
1347 return rte_flow_error_set(error, EINVAL,
1348 RTE_FLOW_ERROR_TYPE_ITEM, item,
1349 "IPv4 is mandatory to filter"
1351 if (item_flags & l4m)
1352 return rte_flow_error_set(error, EINVAL,
1353 RTE_FLOW_ERROR_TYPE_ITEM, item,
1354 "multiple L4 layers not supported");
1356 mask = &rte_flow_item_icmp_mask;
1357 ret = mlx5_flow_item_acceptable
1358 (item, (const uint8_t *)mask,
1359 (const uint8_t *)&rte_flow_item_icmp_mask,
1360 sizeof(struct rte_flow_item_icmp), error);
1367 * Validate Ethernet item.
1370 * Item specification.
1371 * @param[in] item_flags
1372 * Bit-fields that holds the items detected until now.
1374 * Pointer to error structure.
1377 * 0 on success, a negative errno value otherwise and rte_errno is set.
1380 mlx5_flow_validate_item_eth(const struct rte_flow_item *item,
1381 uint64_t item_flags,
1382 struct rte_flow_error *error)
1384 const struct rte_flow_item_eth *mask = item->mask;
1385 const struct rte_flow_item_eth nic_mask = {
1386 .dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1387 .src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
1388 .type = RTE_BE16(0xffff),
1391 int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1392 const uint64_t ethm = tunnel ? MLX5_FLOW_LAYER_INNER_L2 :
1393 MLX5_FLOW_LAYER_OUTER_L2;
1395 if (item_flags & ethm)
1396 return rte_flow_error_set(error, ENOTSUP,
1397 RTE_FLOW_ERROR_TYPE_ITEM, item,
1398 "multiple L2 layers not supported");
1399 if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_L3)) ||
1400 (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_L3)))
1401 return rte_flow_error_set(error, EINVAL,
1402 RTE_FLOW_ERROR_TYPE_ITEM, item,
1403 "L2 layer should not follow "
1405 if ((!tunnel && (item_flags & MLX5_FLOW_LAYER_OUTER_VLAN)) ||
1406 (tunnel && (item_flags & MLX5_FLOW_LAYER_INNER_VLAN)))
1407 return rte_flow_error_set(error, EINVAL,
1408 RTE_FLOW_ERROR_TYPE_ITEM, item,
1409 "L2 layer should not follow VLAN");
1411 mask = &rte_flow_item_eth_mask;
1412 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1413 (const uint8_t *)&nic_mask,
1414 sizeof(struct rte_flow_item_eth),
1420 * Validate VLAN item.
1423 * Item specification.
1424 * @param[in] item_flags
1425 * Bit-fields that holds the items detected until now.
1427 * Ethernet device flow is being created on.
1429 * Pointer to error structure.
1432 * 0 on success, a negative errno value otherwise and rte_errno is set.
1435 mlx5_flow_validate_item_vlan(const struct rte_flow_item *item,
1436 uint64_t item_flags,
1437 struct rte_eth_dev *dev,
1438 struct rte_flow_error *error)
1440 const struct rte_flow_item_vlan *spec = item->spec;
1441 const struct rte_flow_item_vlan *mask = item->mask;
1442 const struct rte_flow_item_vlan nic_mask = {
1443 .tci = RTE_BE16(UINT16_MAX),
1444 .inner_type = RTE_BE16(UINT16_MAX),
1446 uint16_t vlan_tag = 0;
1447 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1449 const uint64_t l34m = tunnel ? (MLX5_FLOW_LAYER_INNER_L3 |
1450 MLX5_FLOW_LAYER_INNER_L4) :
1451 (MLX5_FLOW_LAYER_OUTER_L3 |
1452 MLX5_FLOW_LAYER_OUTER_L4);
1453 const uint64_t vlanm = tunnel ? MLX5_FLOW_LAYER_INNER_VLAN :
1454 MLX5_FLOW_LAYER_OUTER_VLAN;
1456 if (item_flags & vlanm)
1457 return rte_flow_error_set(error, EINVAL,
1458 RTE_FLOW_ERROR_TYPE_ITEM, item,
1459 "multiple VLAN layers not supported");
1460 else if ((item_flags & l34m) != 0)
1461 return rte_flow_error_set(error, EINVAL,
1462 RTE_FLOW_ERROR_TYPE_ITEM, item,
1463 "VLAN cannot follow L3/L4 layer");
1465 mask = &rte_flow_item_vlan_mask;
1466 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1467 (const uint8_t *)&nic_mask,
1468 sizeof(struct rte_flow_item_vlan),
1472 if (!tunnel && mask->tci != RTE_BE16(0x0fff)) {
1473 struct mlx5_priv *priv = dev->data->dev_private;
1475 if (priv->vmwa_context) {
1477 * Non-NULL context means we have a virtual machine
1478 * and SR-IOV enabled, we have to create VLAN interface
1479 * to make hypervisor to setup E-Switch vport
1480 * context correctly. We avoid creating the multiple
1481 * VLAN interfaces, so we cannot support VLAN tag mask.
1483 return rte_flow_error_set(error, EINVAL,
1484 RTE_FLOW_ERROR_TYPE_ITEM,
1486 "VLAN tag mask is not"
1487 " supported in virtual"
1492 vlan_tag = spec->tci;
1493 vlan_tag &= mask->tci;
1496 * From verbs perspective an empty VLAN is equivalent
1497 * to a packet without VLAN layer.
1500 return rte_flow_error_set(error, EINVAL,
1501 RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
1503 "VLAN cannot be empty");
1508 * Validate IPV4 item.
1511 * Item specification.
1512 * @param[in] item_flags
1513 * Bit-fields that holds the items detected until now.
1514 * @param[in] acc_mask
1515 * Acceptable mask, if NULL default internal default mask
1516 * will be used to check whether item fields are supported.
1518 * Pointer to error structure.
1521 * 0 on success, a negative errno value otherwise and rte_errno is set.
1524 mlx5_flow_validate_item_ipv4(const struct rte_flow_item *item,
1525 uint64_t item_flags,
1527 uint16_t ether_type,
1528 const struct rte_flow_item_ipv4 *acc_mask,
1529 struct rte_flow_error *error)
1531 const struct rte_flow_item_ipv4 *mask = item->mask;
1532 const struct rte_flow_item_ipv4 *spec = item->spec;
1533 const struct rte_flow_item_ipv4 nic_mask = {
1535 .src_addr = RTE_BE32(0xffffffff),
1536 .dst_addr = RTE_BE32(0xffffffff),
1537 .type_of_service = 0xff,
1538 .next_proto_id = 0xff,
1541 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1542 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1543 MLX5_FLOW_LAYER_OUTER_L3;
1544 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1545 MLX5_FLOW_LAYER_OUTER_L4;
1547 uint8_t next_proto = 0xFF;
1548 const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
1549 MLX5_FLOW_LAYER_OUTER_VLAN |
1550 MLX5_FLOW_LAYER_INNER_VLAN);
1552 if ((last_item & l2_vlan) && ether_type &&
1553 ether_type != RTE_ETHER_TYPE_IPV4)
1554 return rte_flow_error_set(error, EINVAL,
1555 RTE_FLOW_ERROR_TYPE_ITEM, item,
1556 "IPv4 cannot follow L2/VLAN layer "
1557 "which ether type is not IPv4");
1558 if (item_flags & MLX5_FLOW_LAYER_IPIP) {
1560 next_proto = mask->hdr.next_proto_id &
1561 spec->hdr.next_proto_id;
1562 if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
1563 return rte_flow_error_set(error, EINVAL,
1564 RTE_FLOW_ERROR_TYPE_ITEM,
1569 if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP)
1570 return rte_flow_error_set(error, EINVAL,
1571 RTE_FLOW_ERROR_TYPE_ITEM, item,
1572 "wrong tunnel type - IPv6 specified "
1573 "but IPv4 item provided");
1574 if (item_flags & l3m)
1575 return rte_flow_error_set(error, ENOTSUP,
1576 RTE_FLOW_ERROR_TYPE_ITEM, item,
1577 "multiple L3 layers not supported");
1578 else if (item_flags & l4m)
1579 return rte_flow_error_set(error, EINVAL,
1580 RTE_FLOW_ERROR_TYPE_ITEM, item,
1581 "L3 cannot follow an L4 layer.");
1582 else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
1583 !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
1584 return rte_flow_error_set(error, EINVAL,
1585 RTE_FLOW_ERROR_TYPE_ITEM, item,
1586 "L3 cannot follow an NVGRE layer.");
1588 mask = &rte_flow_item_ipv4_mask;
1589 else if (mask->hdr.next_proto_id != 0 &&
1590 mask->hdr.next_proto_id != 0xff)
1591 return rte_flow_error_set(error, EINVAL,
1592 RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
1593 "partial mask is not supported"
1595 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1596 acc_mask ? (const uint8_t *)acc_mask
1597 : (const uint8_t *)&nic_mask,
1598 sizeof(struct rte_flow_item_ipv4),
1606 * Validate IPV6 item.
1609 * Item specification.
1610 * @param[in] item_flags
1611 * Bit-fields that holds the items detected until now.
1612 * @param[in] acc_mask
1613 * Acceptable mask, if NULL default internal default mask
1614 * will be used to check whether item fields are supported.
1616 * Pointer to error structure.
1619 * 0 on success, a negative errno value otherwise and rte_errno is set.
1622 mlx5_flow_validate_item_ipv6(const struct rte_flow_item *item,
1623 uint64_t item_flags,
1625 uint16_t ether_type,
1626 const struct rte_flow_item_ipv6 *acc_mask,
1627 struct rte_flow_error *error)
1629 const struct rte_flow_item_ipv6 *mask = item->mask;
1630 const struct rte_flow_item_ipv6 *spec = item->spec;
1631 const struct rte_flow_item_ipv6 nic_mask = {
1634 "\xff\xff\xff\xff\xff\xff\xff\xff"
1635 "\xff\xff\xff\xff\xff\xff\xff\xff",
1637 "\xff\xff\xff\xff\xff\xff\xff\xff"
1638 "\xff\xff\xff\xff\xff\xff\xff\xff",
1639 .vtc_flow = RTE_BE32(0xffffffff),
1644 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1645 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1646 MLX5_FLOW_LAYER_OUTER_L3;
1647 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1648 MLX5_FLOW_LAYER_OUTER_L4;
1650 uint8_t next_proto = 0xFF;
1651 const uint64_t l2_vlan = (MLX5_FLOW_LAYER_L2 |
1652 MLX5_FLOW_LAYER_OUTER_VLAN |
1653 MLX5_FLOW_LAYER_INNER_VLAN);
1655 if ((last_item & l2_vlan) && ether_type &&
1656 ether_type != RTE_ETHER_TYPE_IPV6)
1657 return rte_flow_error_set(error, EINVAL,
1658 RTE_FLOW_ERROR_TYPE_ITEM, item,
1659 "IPv6 cannot follow L2/VLAN layer "
1660 "which ether type is not IPv6");
1661 if (item_flags & MLX5_FLOW_LAYER_IPV6_ENCAP) {
1663 next_proto = mask->hdr.proto & spec->hdr.proto;
1664 if (next_proto == IPPROTO_IPIP || next_proto == IPPROTO_IPV6)
1665 return rte_flow_error_set(error, EINVAL,
1666 RTE_FLOW_ERROR_TYPE_ITEM,
1671 if (item_flags & MLX5_FLOW_LAYER_IPIP)
1672 return rte_flow_error_set(error, EINVAL,
1673 RTE_FLOW_ERROR_TYPE_ITEM, item,
1674 "wrong tunnel type - IPv4 specified "
1675 "but IPv6 item provided");
1676 if (item_flags & l3m)
1677 return rte_flow_error_set(error, ENOTSUP,
1678 RTE_FLOW_ERROR_TYPE_ITEM, item,
1679 "multiple L3 layers not supported");
1680 else if (item_flags & l4m)
1681 return rte_flow_error_set(error, EINVAL,
1682 RTE_FLOW_ERROR_TYPE_ITEM, item,
1683 "L3 cannot follow an L4 layer.");
1684 else if ((item_flags & MLX5_FLOW_LAYER_NVGRE) &&
1685 !(item_flags & MLX5_FLOW_LAYER_INNER_L2))
1686 return rte_flow_error_set(error, EINVAL,
1687 RTE_FLOW_ERROR_TYPE_ITEM, item,
1688 "L3 cannot follow an NVGRE layer.");
1690 mask = &rte_flow_item_ipv6_mask;
1691 ret = mlx5_flow_item_acceptable(item, (const uint8_t *)mask,
1692 acc_mask ? (const uint8_t *)acc_mask
1693 : (const uint8_t *)&nic_mask,
1694 sizeof(struct rte_flow_item_ipv6),
1702 * Validate UDP item.
1705 * Item specification.
1706 * @param[in] item_flags
1707 * Bit-fields that holds the items detected until now.
1708 * @param[in] target_protocol
1709 * The next protocol in the previous item.
1710 * @param[in] flow_mask
1711 * mlx5 flow-specific (DV, verbs, etc.) supported header fields mask.
1713 * Pointer to error structure.
1716 * 0 on success, a negative errno value otherwise and rte_errno is set.
1719 mlx5_flow_validate_item_udp(const struct rte_flow_item *item,
1720 uint64_t item_flags,
1721 uint8_t target_protocol,
1722 struct rte_flow_error *error)
1724 const struct rte_flow_item_udp *mask = item->mask;
1725 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1726 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1727 MLX5_FLOW_LAYER_OUTER_L3;
1728 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1729 MLX5_FLOW_LAYER_OUTER_L4;
1732 if (target_protocol != 0xff && target_protocol != IPPROTO_UDP)
1733 return rte_flow_error_set(error, EINVAL,
1734 RTE_FLOW_ERROR_TYPE_ITEM, item,
1735 "protocol filtering not compatible"
1737 if (!(item_flags & l3m))
1738 return rte_flow_error_set(error, EINVAL,
1739 RTE_FLOW_ERROR_TYPE_ITEM, item,
1740 "L3 is mandatory to filter on L4");
1741 if (item_flags & l4m)
1742 return rte_flow_error_set(error, EINVAL,
1743 RTE_FLOW_ERROR_TYPE_ITEM, item,
1744 "multiple L4 layers not supported");
1746 mask = &rte_flow_item_udp_mask;
1747 ret = mlx5_flow_item_acceptable
1748 (item, (const uint8_t *)mask,
1749 (const uint8_t *)&rte_flow_item_udp_mask,
1750 sizeof(struct rte_flow_item_udp), error);
1757 * Validate TCP item.
1760 * Item specification.
1761 * @param[in] item_flags
1762 * Bit-fields that holds the items detected until now.
1763 * @param[in] target_protocol
1764 * The next protocol in the previous item.
1766 * Pointer to error structure.
1769 * 0 on success, a negative errno value otherwise and rte_errno is set.
1772 mlx5_flow_validate_item_tcp(const struct rte_flow_item *item,
1773 uint64_t item_flags,
1774 uint8_t target_protocol,
1775 const struct rte_flow_item_tcp *flow_mask,
1776 struct rte_flow_error *error)
1778 const struct rte_flow_item_tcp *mask = item->mask;
1779 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1780 const uint64_t l3m = tunnel ? MLX5_FLOW_LAYER_INNER_L3 :
1781 MLX5_FLOW_LAYER_OUTER_L3;
1782 const uint64_t l4m = tunnel ? MLX5_FLOW_LAYER_INNER_L4 :
1783 MLX5_FLOW_LAYER_OUTER_L4;
1787 if (target_protocol != 0xff && target_protocol != IPPROTO_TCP)
1788 return rte_flow_error_set(error, EINVAL,
1789 RTE_FLOW_ERROR_TYPE_ITEM, item,
1790 "protocol filtering not compatible"
1792 if (!(item_flags & l3m))
1793 return rte_flow_error_set(error, EINVAL,
1794 RTE_FLOW_ERROR_TYPE_ITEM, item,
1795 "L3 is mandatory to filter on L4");
1796 if (item_flags & l4m)
1797 return rte_flow_error_set(error, EINVAL,
1798 RTE_FLOW_ERROR_TYPE_ITEM, item,
1799 "multiple L4 layers not supported");
1801 mask = &rte_flow_item_tcp_mask;
1802 ret = mlx5_flow_item_acceptable
1803 (item, (const uint8_t *)mask,
1804 (const uint8_t *)flow_mask,
1805 sizeof(struct rte_flow_item_tcp), error);
1812 * Validate VXLAN item.
1815 * Item specification.
1816 * @param[in] item_flags
1817 * Bit-fields that holds the items detected until now.
1818 * @param[in] target_protocol
1819 * The next protocol in the previous item.
1821 * Pointer to error structure.
1824 * 0 on success, a negative errno value otherwise and rte_errno is set.
1827 mlx5_flow_validate_item_vxlan(const struct rte_flow_item *item,
1828 uint64_t item_flags,
1829 struct rte_flow_error *error)
1831 const struct rte_flow_item_vxlan *spec = item->spec;
1832 const struct rte_flow_item_vxlan *mask = item->mask;
1837 } id = { .vlan_id = 0, };
1838 uint32_t vlan_id = 0;
1841 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1842 return rte_flow_error_set(error, ENOTSUP,
1843 RTE_FLOW_ERROR_TYPE_ITEM, item,
1844 "multiple tunnel layers not"
1847 * Verify only UDPv4 is present as defined in
1848 * https://tools.ietf.org/html/rfc7348
1850 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
1851 return rte_flow_error_set(error, EINVAL,
1852 RTE_FLOW_ERROR_TYPE_ITEM, item,
1853 "no outer UDP layer found");
1855 mask = &rte_flow_item_vxlan_mask;
1856 ret = mlx5_flow_item_acceptable
1857 (item, (const uint8_t *)mask,
1858 (const uint8_t *)&rte_flow_item_vxlan_mask,
1859 sizeof(struct rte_flow_item_vxlan),
1864 memcpy(&id.vni[1], spec->vni, 3);
1865 vlan_id = id.vlan_id;
1866 memcpy(&id.vni[1], mask->vni, 3);
1867 vlan_id &= id.vlan_id;
1870 * Tunnel id 0 is equivalent as not adding a VXLAN layer, if
1871 * only this layer is defined in the Verbs specification it is
1872 * interpreted as wildcard and all packets will match this
1873 * rule, if it follows a full stack layer (ex: eth / ipv4 /
1874 * udp), all packets matching the layers before will also
1875 * match this rule. To avoid such situation, VNI 0 is
1876 * currently refused.
1879 return rte_flow_error_set(error, ENOTSUP,
1880 RTE_FLOW_ERROR_TYPE_ITEM, item,
1881 "VXLAN vni cannot be 0");
1882 if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
1883 return rte_flow_error_set(error, ENOTSUP,
1884 RTE_FLOW_ERROR_TYPE_ITEM, item,
1885 "VXLAN tunnel must be fully defined");
1890 * Validate VXLAN_GPE item.
1893 * Item specification.
1894 * @param[in] item_flags
1895 * Bit-fields that holds the items detected until now.
1897 * Pointer to the private data structure.
1898 * @param[in] target_protocol
1899 * The next protocol in the previous item.
1901 * Pointer to error structure.
1904 * 0 on success, a negative errno value otherwise and rte_errno is set.
1907 mlx5_flow_validate_item_vxlan_gpe(const struct rte_flow_item *item,
1908 uint64_t item_flags,
1909 struct rte_eth_dev *dev,
1910 struct rte_flow_error *error)
1912 struct mlx5_priv *priv = dev->data->dev_private;
1913 const struct rte_flow_item_vxlan_gpe *spec = item->spec;
1914 const struct rte_flow_item_vxlan_gpe *mask = item->mask;
1919 } id = { .vlan_id = 0, };
1920 uint32_t vlan_id = 0;
1922 if (!priv->config.l3_vxlan_en)
1923 return rte_flow_error_set(error, ENOTSUP,
1924 RTE_FLOW_ERROR_TYPE_ITEM, item,
1925 "L3 VXLAN is not enabled by device"
1926 " parameter and/or not configured in"
1928 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
1929 return rte_flow_error_set(error, ENOTSUP,
1930 RTE_FLOW_ERROR_TYPE_ITEM, item,
1931 "multiple tunnel layers not"
1934 * Verify only UDPv4 is present as defined in
1935 * https://tools.ietf.org/html/rfc7348
1937 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
1938 return rte_flow_error_set(error, EINVAL,
1939 RTE_FLOW_ERROR_TYPE_ITEM, item,
1940 "no outer UDP layer found");
1942 mask = &rte_flow_item_vxlan_gpe_mask;
1943 ret = mlx5_flow_item_acceptable
1944 (item, (const uint8_t *)mask,
1945 (const uint8_t *)&rte_flow_item_vxlan_gpe_mask,
1946 sizeof(struct rte_flow_item_vxlan_gpe),
1952 return rte_flow_error_set(error, ENOTSUP,
1953 RTE_FLOW_ERROR_TYPE_ITEM,
1955 "VxLAN-GPE protocol"
1957 memcpy(&id.vni[1], spec->vni, 3);
1958 vlan_id = id.vlan_id;
1959 memcpy(&id.vni[1], mask->vni, 3);
1960 vlan_id &= id.vlan_id;
1963 * Tunnel id 0 is equivalent as not adding a VXLAN layer, if only this
1964 * layer is defined in the Verbs specification it is interpreted as
1965 * wildcard and all packets will match this rule, if it follows a full
1966 * stack layer (ex: eth / ipv4 / udp), all packets matching the layers
1967 * before will also match this rule. To avoid such situation, VNI 0
1968 * is currently refused.
1971 return rte_flow_error_set(error, ENOTSUP,
1972 RTE_FLOW_ERROR_TYPE_ITEM, item,
1973 "VXLAN-GPE vni cannot be 0");
1974 if (!(item_flags & MLX5_FLOW_LAYER_OUTER))
1975 return rte_flow_error_set(error, ENOTSUP,
1976 RTE_FLOW_ERROR_TYPE_ITEM, item,
1977 "VXLAN-GPE tunnel must be fully"
1982 * Validate GRE Key item.
1985 * Item specification.
1986 * @param[in] item_flags
1987 * Bit flags to mark detected items.
1988 * @param[in] gre_item
1989 * Pointer to gre_item
1991 * Pointer to error structure.
1994 * 0 on success, a negative errno value otherwise and rte_errno is set.
1997 mlx5_flow_validate_item_gre_key(const struct rte_flow_item *item,
1998 uint64_t item_flags,
1999 const struct rte_flow_item *gre_item,
2000 struct rte_flow_error *error)
2002 const rte_be32_t *mask = item->mask;
2004 rte_be32_t gre_key_default_mask = RTE_BE32(UINT32_MAX);
2005 const struct rte_flow_item_gre *gre_spec;
2006 const struct rte_flow_item_gre *gre_mask;
2008 if (item_flags & MLX5_FLOW_LAYER_GRE_KEY)
2009 return rte_flow_error_set(error, ENOTSUP,
2010 RTE_FLOW_ERROR_TYPE_ITEM, item,
2011 "Multiple GRE key not support");
2012 if (!(item_flags & MLX5_FLOW_LAYER_GRE))
2013 return rte_flow_error_set(error, ENOTSUP,
2014 RTE_FLOW_ERROR_TYPE_ITEM, item,
2015 "No preceding GRE header");
2016 if (item_flags & MLX5_FLOW_LAYER_INNER)
2017 return rte_flow_error_set(error, ENOTSUP,
2018 RTE_FLOW_ERROR_TYPE_ITEM, item,
2019 "GRE key following a wrong item");
2020 gre_mask = gre_item->mask;
2022 gre_mask = &rte_flow_item_gre_mask;
2023 gre_spec = gre_item->spec;
2024 if (gre_spec && (gre_mask->c_rsvd0_ver & RTE_BE16(0x2000)) &&
2025 !(gre_spec->c_rsvd0_ver & RTE_BE16(0x2000)))
2026 return rte_flow_error_set(error, EINVAL,
2027 RTE_FLOW_ERROR_TYPE_ITEM, item,
2028 "Key bit must be on");
2031 mask = &gre_key_default_mask;
2032 ret = mlx5_flow_item_acceptable
2033 (item, (const uint8_t *)mask,
2034 (const uint8_t *)&gre_key_default_mask,
2035 sizeof(rte_be32_t), error);
2040 * Validate GRE item.
2043 * Item specification.
2044 * @param[in] item_flags
2045 * Bit flags to mark detected items.
2046 * @param[in] target_protocol
2047 * The next protocol in the previous item.
2049 * Pointer to error structure.
2052 * 0 on success, a negative errno value otherwise and rte_errno is set.
2055 mlx5_flow_validate_item_gre(const struct rte_flow_item *item,
2056 uint64_t item_flags,
2057 uint8_t target_protocol,
2058 struct rte_flow_error *error)
2060 const struct rte_flow_item_gre *spec __rte_unused = item->spec;
2061 const struct rte_flow_item_gre *mask = item->mask;
2063 const struct rte_flow_item_gre nic_mask = {
2064 .c_rsvd0_ver = RTE_BE16(0xB000),
2065 .protocol = RTE_BE16(UINT16_MAX),
2068 if (target_protocol != 0xff && target_protocol != IPPROTO_GRE)
2069 return rte_flow_error_set(error, EINVAL,
2070 RTE_FLOW_ERROR_TYPE_ITEM, item,
2071 "protocol filtering not compatible"
2072 " with this GRE layer");
2073 if (item_flags & MLX5_FLOW_LAYER_TUNNEL)
2074 return rte_flow_error_set(error, ENOTSUP,
2075 RTE_FLOW_ERROR_TYPE_ITEM, item,
2076 "multiple tunnel layers not"
2078 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
2079 return rte_flow_error_set(error, ENOTSUP,
2080 RTE_FLOW_ERROR_TYPE_ITEM, item,
2081 "L3 Layer is missing");
2083 mask = &rte_flow_item_gre_mask;
2084 ret = mlx5_flow_item_acceptable
2085 (item, (const uint8_t *)mask,
2086 (const uint8_t *)&nic_mask,
2087 sizeof(struct rte_flow_item_gre), error);
2090 #ifndef HAVE_MLX5DV_DR
2091 #ifndef HAVE_IBV_DEVICE_MPLS_SUPPORT
2092 if (spec && (spec->protocol & mask->protocol))
2093 return rte_flow_error_set(error, ENOTSUP,
2094 RTE_FLOW_ERROR_TYPE_ITEM, item,
2095 "without MPLS support the"
2096 " specification cannot be used for"
2104 * Validate Geneve item.
2107 * Item specification.
2108 * @param[in] itemFlags
2109 * Bit-fields that holds the items detected until now.
2111 * Pointer to the private data structure.
2113 * Pointer to error structure.
2116 * 0 on success, a negative errno value otherwise and rte_errno is set.
2120 mlx5_flow_validate_item_geneve(const struct rte_flow_item *item,
2121 uint64_t item_flags,
2122 struct rte_eth_dev *dev,
2123 struct rte_flow_error *error)
2125 struct mlx5_priv *priv = dev->data->dev_private;
2126 const struct rte_flow_item_geneve *spec = item->spec;
2127 const struct rte_flow_item_geneve *mask = item->mask;
2130 uint8_t opt_len = priv->config.hca_attr.geneve_max_opt_len ?
2131 MLX5_GENEVE_OPT_LEN_1 : MLX5_GENEVE_OPT_LEN_0;
2132 const struct rte_flow_item_geneve nic_mask = {
2133 .ver_opt_len_o_c_rsvd0 = RTE_BE16(0x3f80),
2134 .vni = "\xff\xff\xff",
2135 .protocol = RTE_BE16(UINT16_MAX),
2138 if (!(priv->config.hca_attr.flex_parser_protocols &
2139 MLX5_HCA_FLEX_GENEVE_ENABLED) ||
2140 !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);
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 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 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 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 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 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 port id item from the item list.
2670 * Pointer to the list of items.
2673 * Pointer to the port id item if exist, else return NULL.
2675 static const struct rte_flow_item *
2676 find_port_id_item(const struct rte_flow_item *item)
2679 for (; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
2680 if (item->type == RTE_FLOW_ITEM_TYPE_PORT_ID)
2687 * Get RSS action from the action list.
2689 * @param[in] actions
2690 * Pointer to the list of actions.
2693 * Pointer to the RSS action if exist, else return NULL.
2695 static const struct rte_flow_action_rss*
2696 flow_get_rss_action(const struct rte_flow_action actions[])
2698 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2699 switch (actions->type) {
2700 case RTE_FLOW_ACTION_TYPE_RSS:
2701 return (const struct rte_flow_action_rss *)
2711 find_graph_root(const struct rte_flow_item pattern[], uint32_t rss_level)
2713 const struct rte_flow_item *item;
2714 unsigned int has_vlan = 0;
2716 for (item = pattern; item->type != RTE_FLOW_ITEM_TYPE_END; item++) {
2717 if (item->type == RTE_FLOW_ITEM_TYPE_VLAN) {
2723 return rss_level < 2 ? MLX5_EXPANSION_ROOT_ETH_VLAN :
2724 MLX5_EXPANSION_ROOT_OUTER_ETH_VLAN;
2725 return rss_level < 2 ? MLX5_EXPANSION_ROOT :
2726 MLX5_EXPANSION_ROOT_OUTER;
2730 * Get QUEUE/RSS action from the action list.
2732 * @param[in] actions
2733 * Pointer to the list of actions.
2735 * Pointer to the return pointer.
2736 * @param[out] qrss_type
2737 * Pointer to the action type to return. RTE_FLOW_ACTION_TYPE_END is returned
2738 * if no QUEUE/RSS is found.
2741 * Total number of actions.
2744 flow_parse_qrss_action(const struct rte_flow_action actions[],
2745 const struct rte_flow_action **qrss)
2749 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2750 switch (actions->type) {
2751 case RTE_FLOW_ACTION_TYPE_QUEUE:
2752 case RTE_FLOW_ACTION_TYPE_RSS:
2760 /* Count RTE_FLOW_ACTION_TYPE_END. */
2761 return actions_n + 1;
2765 * Check meter action from the action list.
2767 * @param[in] actions
2768 * Pointer to the list of actions.
2770 * Pointer to the meter exist flag.
2773 * Total number of actions.
2776 flow_check_meter_action(const struct rte_flow_action actions[], uint32_t *mtr)
2782 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2783 switch (actions->type) {
2784 case RTE_FLOW_ACTION_TYPE_METER:
2792 /* Count RTE_FLOW_ACTION_TYPE_END. */
2793 return actions_n + 1;
2797 * Check if the flow should be splited due to hairpin.
2798 * The reason for the split is that in current HW we can't
2799 * support encap on Rx, so if a flow have encap we move it
2803 * Pointer to Ethernet device.
2805 * Flow rule attributes.
2806 * @param[in] actions
2807 * Associated actions (list terminated by the END action).
2810 * > 0 the number of actions and the flow should be split,
2811 * 0 when no split required.
2814 flow_check_hairpin_split(struct rte_eth_dev *dev,
2815 const struct rte_flow_attr *attr,
2816 const struct rte_flow_action actions[])
2818 int queue_action = 0;
2821 const struct rte_flow_action_queue *queue;
2822 const struct rte_flow_action_rss *rss;
2823 const struct rte_flow_action_raw_encap *raw_encap;
2827 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2828 switch (actions->type) {
2829 case RTE_FLOW_ACTION_TYPE_QUEUE:
2830 queue = actions->conf;
2833 if (mlx5_rxq_get_type(dev, queue->index) !=
2834 MLX5_RXQ_TYPE_HAIRPIN)
2839 case RTE_FLOW_ACTION_TYPE_RSS:
2840 rss = actions->conf;
2841 if (rss == NULL || rss->queue_num == 0)
2843 if (mlx5_rxq_get_type(dev, rss->queue[0]) !=
2844 MLX5_RXQ_TYPE_HAIRPIN)
2849 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
2850 case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
2854 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
2855 raw_encap = actions->conf;
2856 if (raw_encap->size >
2857 (sizeof(struct rte_flow_item_eth) +
2858 sizeof(struct rte_flow_item_ipv4)))
2867 if (encap == 1 && queue_action)
2872 /* Declare flow create/destroy prototype in advance. */
2873 static struct rte_flow *
2874 flow_list_create(struct rte_eth_dev *dev, struct mlx5_flows *list,
2875 const struct rte_flow_attr *attr,
2876 const struct rte_flow_item items[],
2877 const struct rte_flow_action actions[],
2878 bool external, struct rte_flow_error *error);
2881 flow_list_destroy(struct rte_eth_dev *dev, struct mlx5_flows *list,
2882 struct rte_flow *flow);
2885 * Add a flow of copying flow metadata registers in RX_CP_TBL.
2887 * As mark_id is unique, if there's already a registered flow for the mark_id,
2888 * return by increasing the reference counter of the resource. Otherwise, create
2889 * the resource (mcp_res) and flow.
2892 * - If ingress port is ANY and reg_c[1] is mark_id,
2893 * flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
2895 * For default flow (zero mark_id), flow is like,
2896 * - If ingress port is ANY,
2897 * reg_b := reg_c[0] and jump to RX_ACT_TBL.
2900 * Pointer to Ethernet device.
2902 * ID of MARK action, zero means default flow for META.
2904 * Perform verbose error reporting if not NULL.
2907 * Associated resource on success, NULL otherwise and rte_errno is set.
2909 static struct mlx5_flow_mreg_copy_resource *
2910 flow_mreg_add_copy_action(struct rte_eth_dev *dev, uint32_t mark_id,
2911 struct rte_flow_error *error)
2913 struct mlx5_priv *priv = dev->data->dev_private;
2914 struct rte_flow_attr attr = {
2915 .group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
2918 struct mlx5_rte_flow_item_tag tag_spec = {
2921 struct rte_flow_item items[] = {
2922 [1] = { .type = RTE_FLOW_ITEM_TYPE_END, },
2924 struct rte_flow_action_mark ftag = {
2927 struct mlx5_flow_action_copy_mreg cp_mreg = {
2931 struct rte_flow_action_jump jump = {
2932 .group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
2934 struct rte_flow_action actions[] = {
2935 [3] = { .type = RTE_FLOW_ACTION_TYPE_END, },
2937 struct mlx5_flow_mreg_copy_resource *mcp_res;
2940 /* Fill the register fileds in the flow. */
2941 ret = mlx5_flow_get_reg_id(dev, MLX5_FLOW_MARK, 0, error);
2945 ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
2949 /* Check if already registered. */
2950 assert(priv->mreg_cp_tbl);
2951 mcp_res = (void *)mlx5_hlist_lookup(priv->mreg_cp_tbl, mark_id);
2953 /* For non-default rule. */
2954 if (mark_id != MLX5_DEFAULT_COPY_ID)
2956 assert(mark_id != MLX5_DEFAULT_COPY_ID || mcp_res->refcnt == 1);
2959 /* Provide the full width of FLAG specific value. */
2960 if (mark_id == (priv->sh->dv_regc0_mask & MLX5_FLOW_MARK_DEFAULT))
2961 tag_spec.data = MLX5_FLOW_MARK_DEFAULT;
2962 /* Build a new flow. */
2963 if (mark_id != MLX5_DEFAULT_COPY_ID) {
2964 items[0] = (struct rte_flow_item){
2965 .type = MLX5_RTE_FLOW_ITEM_TYPE_TAG,
2968 items[1] = (struct rte_flow_item){
2969 .type = RTE_FLOW_ITEM_TYPE_END,
2971 actions[0] = (struct rte_flow_action){
2972 .type = MLX5_RTE_FLOW_ACTION_TYPE_MARK,
2975 actions[1] = (struct rte_flow_action){
2976 .type = MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
2979 actions[2] = (struct rte_flow_action){
2980 .type = RTE_FLOW_ACTION_TYPE_JUMP,
2983 actions[3] = (struct rte_flow_action){
2984 .type = RTE_FLOW_ACTION_TYPE_END,
2987 /* Default rule, wildcard match. */
2988 attr.priority = MLX5_FLOW_PRIO_RSVD;
2989 items[0] = (struct rte_flow_item){
2990 .type = RTE_FLOW_ITEM_TYPE_END,
2992 actions[0] = (struct rte_flow_action){
2993 .type = MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
2996 actions[1] = (struct rte_flow_action){
2997 .type = RTE_FLOW_ACTION_TYPE_JUMP,
3000 actions[2] = (struct rte_flow_action){
3001 .type = RTE_FLOW_ACTION_TYPE_END,
3004 /* Build a new entry. */
3005 mcp_res = rte_zmalloc(__func__, sizeof(*mcp_res), 0);
3011 * The copy Flows are not included in any list. There
3012 * ones are referenced from other Flows and can not
3013 * be applied, removed, deleted in ardbitrary order
3014 * by list traversing.
3016 mcp_res->flow = flow_list_create(dev, NULL, &attr, items,
3017 actions, false, error);
3021 mcp_res->hlist_ent.key = mark_id;
3022 ret = mlx5_hlist_insert(priv->mreg_cp_tbl,
3023 &mcp_res->hlist_ent);
3030 flow_list_destroy(dev, NULL, mcp_res->flow);
3036 * Release flow in RX_CP_TBL.
3039 * Pointer to Ethernet device.
3041 * Parent flow for wich copying is provided.
3044 flow_mreg_del_copy_action(struct rte_eth_dev *dev,
3045 struct rte_flow *flow)
3047 struct mlx5_flow_mreg_copy_resource *mcp_res = flow->mreg_copy;
3048 struct mlx5_priv *priv = dev->data->dev_private;
3050 if (!mcp_res || !priv->mreg_cp_tbl)
3052 if (flow->copy_applied) {
3053 assert(mcp_res->appcnt);
3054 flow->copy_applied = 0;
3056 if (!mcp_res->appcnt)
3057 flow_drv_remove(dev, mcp_res->flow);
3060 * We do not check availability of metadata registers here,
3061 * because copy resources are not allocated in this case.
3063 if (--mcp_res->refcnt)
3065 assert(mcp_res->flow);
3066 flow_list_destroy(dev, NULL, mcp_res->flow);
3067 mlx5_hlist_remove(priv->mreg_cp_tbl, &mcp_res->hlist_ent);
3069 flow->mreg_copy = NULL;
3073 * Start flow in RX_CP_TBL.
3076 * Pointer to Ethernet device.
3078 * Parent flow for wich copying is provided.
3081 * 0 on success, a negative errno value otherwise and rte_errno is set.
3084 flow_mreg_start_copy_action(struct rte_eth_dev *dev,
3085 struct rte_flow *flow)
3087 struct mlx5_flow_mreg_copy_resource *mcp_res = flow->mreg_copy;
3090 if (!mcp_res || flow->copy_applied)
3092 if (!mcp_res->appcnt) {
3093 ret = flow_drv_apply(dev, mcp_res->flow, NULL);
3098 flow->copy_applied = 1;
3103 * Stop flow in RX_CP_TBL.
3106 * Pointer to Ethernet device.
3108 * Parent flow for wich copying is provided.
3111 flow_mreg_stop_copy_action(struct rte_eth_dev *dev,
3112 struct rte_flow *flow)
3114 struct mlx5_flow_mreg_copy_resource *mcp_res = flow->mreg_copy;
3116 if (!mcp_res || !flow->copy_applied)
3118 assert(mcp_res->appcnt);
3120 flow->copy_applied = 0;
3121 if (!mcp_res->appcnt)
3122 flow_drv_remove(dev, mcp_res->flow);
3126 * Remove the default copy action from RX_CP_TBL.
3129 * Pointer to Ethernet device.
3132 flow_mreg_del_default_copy_action(struct rte_eth_dev *dev)
3134 struct mlx5_flow_mreg_copy_resource *mcp_res;
3135 struct mlx5_priv *priv = dev->data->dev_private;
3137 /* Check if default flow is registered. */
3138 if (!priv->mreg_cp_tbl)
3140 mcp_res = (void *)mlx5_hlist_lookup(priv->mreg_cp_tbl,
3141 MLX5_DEFAULT_COPY_ID);
3144 assert(mcp_res->flow);
3145 flow_list_destroy(dev, NULL, mcp_res->flow);
3146 mlx5_hlist_remove(priv->mreg_cp_tbl, &mcp_res->hlist_ent);
3151 * Add the default copy action in in RX_CP_TBL.
3154 * Pointer to Ethernet device.
3156 * Perform verbose error reporting if not NULL.
3159 * 0 for success, negative value otherwise and rte_errno is set.
3162 flow_mreg_add_default_copy_action(struct rte_eth_dev *dev,
3163 struct rte_flow_error *error)
3165 struct mlx5_priv *priv = dev->data->dev_private;
3166 struct mlx5_flow_mreg_copy_resource *mcp_res;
3168 /* Check whether extensive metadata feature is engaged. */
3169 if (!priv->config.dv_flow_en ||
3170 priv->config.dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
3171 !mlx5_flow_ext_mreg_supported(dev) ||
3172 !priv->sh->dv_regc0_mask)
3174 mcp_res = flow_mreg_add_copy_action(dev, MLX5_DEFAULT_COPY_ID, error);
3181 * Add a flow of copying flow metadata registers in RX_CP_TBL.
3183 * All the flow having Q/RSS action should be split by
3184 * flow_mreg_split_qrss_prep() to pass by RX_CP_TBL. A flow in the RX_CP_TBL
3185 * performs the following,
3186 * - CQE->flow_tag := reg_c[1] (MARK)
3187 * - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
3188 * As CQE's flow_tag is not a register, it can't be simply copied from reg_c[1]
3189 * but there should be a flow per each MARK ID set by MARK action.
3191 * For the aforementioned reason, if there's a MARK action in flow's action
3192 * list, a corresponding flow should be added to the RX_CP_TBL in order to copy
3193 * the MARK ID to CQE's flow_tag like,
3194 * - If reg_c[1] is mark_id,
3195 * flow_tag := mark_id, reg_b := reg_c[0] and jump to RX_ACT_TBL.
3197 * For SET_META action which stores value in reg_c[0], as the destination is
3198 * also a flow metadata register (reg_b), adding a default flow is enough. Zero
3199 * MARK ID means the default flow. The default flow looks like,
3200 * - For all flow, reg_b := reg_c[0] and jump to RX_ACT_TBL.
3203 * Pointer to Ethernet device.
3205 * Pointer to flow structure.
3206 * @param[in] actions
3207 * Pointer to the list of actions.
3209 * Perform verbose error reporting if not NULL.
3212 * 0 on success, negative value otherwise and rte_errno is set.
3215 flow_mreg_update_copy_table(struct rte_eth_dev *dev,
3216 struct rte_flow *flow,
3217 const struct rte_flow_action *actions,
3218 struct rte_flow_error *error)
3220 struct mlx5_priv *priv = dev->data->dev_private;
3221 struct mlx5_dev_config *config = &priv->config;
3222 struct mlx5_flow_mreg_copy_resource *mcp_res;
3223 const struct rte_flow_action_mark *mark;
3225 /* Check whether extensive metadata feature is engaged. */
3226 if (!config->dv_flow_en ||
3227 config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
3228 !mlx5_flow_ext_mreg_supported(dev) ||
3229 !priv->sh->dv_regc0_mask)
3231 /* Find MARK action. */
3232 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3233 switch (actions->type) {
3234 case RTE_FLOW_ACTION_TYPE_FLAG:
3235 mcp_res = flow_mreg_add_copy_action
3236 (dev, MLX5_FLOW_MARK_DEFAULT, error);
3239 flow->mreg_copy = mcp_res;
3240 if (dev->data->dev_started) {
3242 flow->copy_applied = 1;
3245 case RTE_FLOW_ACTION_TYPE_MARK:
3246 mark = (const struct rte_flow_action_mark *)
3249 flow_mreg_add_copy_action(dev, mark->id, error);
3252 flow->mreg_copy = mcp_res;
3253 if (dev->data->dev_started) {
3255 flow->copy_applied = 1;
3265 #define MLX5_MAX_SPLIT_ACTIONS 24
3266 #define MLX5_MAX_SPLIT_ITEMS 24
3269 * Split the hairpin flow.
3270 * Since HW can't support encap on Rx we move the encap to Tx.
3271 * If the count action is after the encap then we also
3272 * move the count action. in this case the count will also measure
3276 * Pointer to Ethernet device.
3277 * @param[in] actions
3278 * Associated actions (list terminated by the END action).
3279 * @param[out] actions_rx
3281 * @param[out] actions_tx
3283 * @param[out] pattern_tx
3284 * The pattern items for the Tx flow.
3285 * @param[out] flow_id
3286 * The flow ID connected to this flow.
3292 flow_hairpin_split(struct rte_eth_dev *dev,
3293 const struct rte_flow_action actions[],
3294 struct rte_flow_action actions_rx[],
3295 struct rte_flow_action actions_tx[],
3296 struct rte_flow_item pattern_tx[],
3299 struct mlx5_priv *priv = dev->data->dev_private;
3300 const struct rte_flow_action_raw_encap *raw_encap;
3301 const struct rte_flow_action_raw_decap *raw_decap;
3302 struct mlx5_rte_flow_action_set_tag *set_tag;
3303 struct rte_flow_action *tag_action;
3304 struct mlx5_rte_flow_item_tag *tag_item;
3305 struct rte_flow_item *item;
3309 mlx5_flow_id_get(priv->sh->flow_id_pool, flow_id);
3310 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3311 switch (actions->type) {
3312 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
3313 case RTE_FLOW_ACTION_TYPE_NVGRE_ENCAP:
3314 rte_memcpy(actions_tx, actions,
3315 sizeof(struct rte_flow_action));
3318 case RTE_FLOW_ACTION_TYPE_COUNT:
3320 rte_memcpy(actions_tx, actions,
3321 sizeof(struct rte_flow_action));
3324 rte_memcpy(actions_rx, actions,
3325 sizeof(struct rte_flow_action));
3329 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3330 raw_encap = actions->conf;
3331 if (raw_encap->size >
3332 (sizeof(struct rte_flow_item_eth) +
3333 sizeof(struct rte_flow_item_ipv4))) {
3334 memcpy(actions_tx, actions,
3335 sizeof(struct rte_flow_action));
3339 rte_memcpy(actions_rx, actions,
3340 sizeof(struct rte_flow_action));
3344 case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
3345 raw_decap = actions->conf;
3346 if (raw_decap->size <
3347 (sizeof(struct rte_flow_item_eth) +
3348 sizeof(struct rte_flow_item_ipv4))) {
3349 memcpy(actions_tx, actions,
3350 sizeof(struct rte_flow_action));
3353 rte_memcpy(actions_rx, actions,
3354 sizeof(struct rte_flow_action));
3359 rte_memcpy(actions_rx, actions,
3360 sizeof(struct rte_flow_action));
3365 /* Add set meta action and end action for the Rx flow. */
3366 tag_action = actions_rx;
3367 tag_action->type = MLX5_RTE_FLOW_ACTION_TYPE_TAG;
3369 rte_memcpy(actions_rx, actions, sizeof(struct rte_flow_action));
3371 set_tag = (void *)actions_rx;
3372 set_tag->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_RX, 0, NULL);
3373 assert(set_tag->id > REG_NONE);
3374 set_tag->data = *flow_id;
3375 tag_action->conf = set_tag;
3376 /* Create Tx item list. */
3377 rte_memcpy(actions_tx, actions, sizeof(struct rte_flow_action));
3378 addr = (void *)&pattern_tx[2];
3380 item->type = MLX5_RTE_FLOW_ITEM_TYPE_TAG;
3381 tag_item = (void *)addr;
3382 tag_item->data = *flow_id;
3383 tag_item->id = mlx5_flow_get_reg_id(dev, MLX5_HAIRPIN_TX, 0, NULL);
3384 assert(set_tag->id > REG_NONE);
3385 item->spec = tag_item;
3386 addr += sizeof(struct mlx5_rte_flow_item_tag);
3387 tag_item = (void *)addr;
3388 tag_item->data = UINT32_MAX;
3389 tag_item->id = UINT16_MAX;
3390 item->mask = tag_item;
3391 addr += sizeof(struct mlx5_rte_flow_item_tag);
3394 item->type = RTE_FLOW_ITEM_TYPE_END;
3399 * The last stage of splitting chain, just creates the subflow
3400 * without any modification.
3403 * Pointer to Ethernet device.
3405 * Parent flow structure pointer.
3406 * @param[in, out] sub_flow
3407 * Pointer to return the created subflow, may be NULL.
3409 * Flow rule attributes.
3411 * Pattern specification (list terminated by the END pattern item).
3412 * @param[in] actions
3413 * Associated actions (list terminated by the END action).
3414 * @param[in] external
3415 * This flow rule is created by request external to PMD.
3417 * Perform verbose error reporting if not NULL.
3419 * 0 on success, negative value otherwise
3422 flow_create_split_inner(struct rte_eth_dev *dev,
3423 struct rte_flow *flow,
3424 struct mlx5_flow **sub_flow,
3425 const struct rte_flow_attr *attr,
3426 const struct rte_flow_item items[],
3427 const struct rte_flow_action actions[],
3428 bool external, struct rte_flow_error *error)
3430 struct mlx5_flow *dev_flow;
3432 dev_flow = flow_drv_prepare(flow, attr, items, actions, error);
3435 dev_flow->flow = flow;
3436 dev_flow->external = external;
3437 /* Subflow object was created, we must include one in the list. */
3438 LIST_INSERT_HEAD(&flow->dev_flows, dev_flow, next);
3440 *sub_flow = dev_flow;
3441 return flow_drv_translate(dev, dev_flow, attr, items, actions, error);
3445 * Split the meter flow.
3447 * As meter flow will split to three sub flow, other than meter
3448 * action, the other actions make sense to only meter accepts
3449 * the packet. If it need to be dropped, no other additional
3450 * actions should be take.
3452 * One kind of special action which decapsulates the L3 tunnel
3453 * header will be in the prefix sub flow, as not to take the
3454 * L3 tunnel header into account.
3457 * Pointer to Ethernet device.
3458 * @param[in] actions
3459 * Associated actions (list terminated by the END action).
3460 * @param[out] actions_sfx
3461 * Suffix flow actions.
3462 * @param[out] actions_pre
3463 * Prefix flow actions.
3464 * @param[out] pattern_sfx
3465 * The pattern items for the suffix flow.
3466 * @param[out] tag_sfx
3467 * Pointer to suffix flow tag.
3473 flow_meter_split_prep(struct rte_eth_dev *dev,
3474 const struct rte_flow_action actions[],
3475 struct rte_flow_action actions_sfx[],
3476 struct rte_flow_action actions_pre[])
3478 struct rte_flow_action *tag_action;
3479 struct mlx5_rte_flow_action_set_tag *set_tag;
3480 struct rte_flow_error error;
3481 const struct rte_flow_action_raw_encap *raw_encap;
3482 const struct rte_flow_action_raw_decap *raw_decap;
3485 /* Add the extra tag action first. */
3486 tag_action = actions_pre;
3487 tag_action->type = MLX5_RTE_FLOW_ACTION_TYPE_TAG;
3489 /* Prepare the actions for prefix and suffix flow. */
3490 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3491 switch (actions->type) {
3492 case RTE_FLOW_ACTION_TYPE_METER:
3493 case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
3494 case RTE_FLOW_ACTION_TYPE_NVGRE_DECAP:
3495 memcpy(actions_pre, actions,
3496 sizeof(struct rte_flow_action));
3499 case RTE_FLOW_ACTION_TYPE_RAW_ENCAP:
3500 raw_encap = actions->conf;
3501 if (raw_encap->size >
3502 (sizeof(struct rte_flow_item_eth) +
3503 sizeof(struct rte_flow_item_ipv4))) {
3504 memcpy(actions_sfx, actions,
3505 sizeof(struct rte_flow_action));
3508 rte_memcpy(actions_pre, actions,
3509 sizeof(struct rte_flow_action));
3513 case RTE_FLOW_ACTION_TYPE_RAW_DECAP:
3514 raw_decap = actions->conf;
3515 /* Size 0 decap means 50 bytes as vxlan decap. */
3516 if (raw_decap->size && (raw_decap->size <
3517 (sizeof(struct rte_flow_item_eth) +
3518 sizeof(struct rte_flow_item_ipv4)))) {
3519 memcpy(actions_sfx, actions,
3520 sizeof(struct rte_flow_action));
3523 rte_memcpy(actions_pre, actions,
3524 sizeof(struct rte_flow_action));
3529 memcpy(actions_sfx, actions,
3530 sizeof(struct rte_flow_action));
3535 /* Add end action to the actions. */
3536 actions_sfx->type = RTE_FLOW_ACTION_TYPE_END;
3537 actions_pre->type = RTE_FLOW_ACTION_TYPE_END;
3540 set_tag = (void *)actions_pre;
3541 set_tag->id = mlx5_flow_get_reg_id(dev, MLX5_MTR_SFX, 0, &error);
3543 * Get the id from the qrss_pool to make qrss share the id with meter.
3545 tag_id = flow_qrss_get_id(dev);
3546 set_tag->data = tag_id << MLX5_MTR_COLOR_BITS;
3547 tag_action->conf = set_tag;
3552 * Split action list having QUEUE/RSS for metadata register copy.
3554 * Once Q/RSS action is detected in user's action list, the flow action
3555 * should be split in order to copy metadata registers, which will happen in
3557 * - CQE->flow_tag := reg_c[1] (MARK)
3558 * - CQE->flow_table_metadata (reg_b) := reg_c[0] (META)
3559 * The Q/RSS action will be performed on RX_ACT_TBL after passing by RX_CP_TBL.
3560 * This is because the last action of each flow must be a terminal action
3561 * (QUEUE, RSS or DROP).
3563 * Flow ID must be allocated to identify actions in the RX_ACT_TBL and it is
3564 * stored and kept in the mlx5_flow structure per each sub_flow.
3566 * The Q/RSS action is replaced with,
3567 * - SET_TAG, setting the allocated flow ID to reg_c[2].
3568 * And the following JUMP action is added at the end,
3569 * - JUMP, to RX_CP_TBL.
3571 * A flow to perform remained Q/RSS action will be created in RX_ACT_TBL by
3572 * flow_create_split_metadata() routine. The flow will look like,
3573 * - If flow ID matches (reg_c[2]), perform Q/RSS.
3576 * Pointer to Ethernet device.
3577 * @param[out] split_actions
3578 * Pointer to store split actions to jump to CP_TBL.
3579 * @param[in] actions
3580 * Pointer to the list of original flow actions.
3582 * Pointer to the Q/RSS action.
3583 * @param[in] actions_n
3584 * Number of original actions.
3586 * Perform verbose error reporting if not NULL.
3589 * non-zero unique flow_id on success, otherwise 0 and
3590 * error/rte_error are set.
3593 flow_mreg_split_qrss_prep(struct rte_eth_dev *dev,
3594 struct rte_flow_action *split_actions,
3595 const struct rte_flow_action *actions,
3596 const struct rte_flow_action *qrss,
3597 int actions_n, struct rte_flow_error *error)
3599 struct mlx5_rte_flow_action_set_tag *set_tag;
3600 struct rte_flow_action_jump *jump;
3601 const int qrss_idx = qrss - actions;
3602 uint32_t flow_id = 0;
3606 * Given actions will be split
3607 * - Replace QUEUE/RSS action with SET_TAG to set flow ID.
3608 * - Add jump to mreg CP_TBL.
3609 * As a result, there will be one more action.
3612 memcpy(split_actions, actions, sizeof(*split_actions) * actions_n);
3613 set_tag = (void *)(split_actions + actions_n);
3615 * If tag action is not set to void(it means we are not the meter
3616 * suffix flow), add the tag action. Since meter suffix flow already
3617 * has the tag added.
3619 if (split_actions[qrss_idx].type != RTE_FLOW_ACTION_TYPE_VOID) {
3621 * Allocate the new subflow ID. This one is unique within
3622 * device and not shared with representors. Otherwise,
3623 * we would have to resolve multi-thread access synch
3624 * issue. Each flow on the shared device is appended
3625 * with source vport identifier, so the resulting
3626 * flows will be unique in the shared (by master and
3627 * representors) domain even if they have coinciding
3630 flow_id = flow_qrss_get_id(dev);
3632 return rte_flow_error_set(error, ENOMEM,
3633 RTE_FLOW_ERROR_TYPE_ACTION,
3634 NULL, "can't allocate id "
3635 "for split Q/RSS subflow");
3636 /* Internal SET_TAG action to set flow ID. */
3637 *set_tag = (struct mlx5_rte_flow_action_set_tag){
3640 ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0, error);
3644 /* Construct new actions array. */
3645 /* Replace QUEUE/RSS action. */
3646 split_actions[qrss_idx] = (struct rte_flow_action){
3647 .type = MLX5_RTE_FLOW_ACTION_TYPE_TAG,
3651 /* JUMP action to jump to mreg copy table (CP_TBL). */
3652 jump = (void *)(set_tag + 1);
3653 *jump = (struct rte_flow_action_jump){
3654 .group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
3656 split_actions[actions_n - 2] = (struct rte_flow_action){
3657 .type = RTE_FLOW_ACTION_TYPE_JUMP,
3660 split_actions[actions_n - 1] = (struct rte_flow_action){
3661 .type = RTE_FLOW_ACTION_TYPE_END,
3667 * Extend the given action list for Tx metadata copy.
3669 * Copy the given action list to the ext_actions and add flow metadata register
3670 * copy action in order to copy reg_a set by WQE to reg_c[0].
3672 * @param[out] ext_actions
3673 * Pointer to the extended action list.
3674 * @param[in] actions
3675 * Pointer to the list of actions.
3676 * @param[in] actions_n
3677 * Number of actions in the list.
3679 * Perform verbose error reporting if not NULL.
3682 * 0 on success, negative value otherwise
3685 flow_mreg_tx_copy_prep(struct rte_eth_dev *dev,
3686 struct rte_flow_action *ext_actions,
3687 const struct rte_flow_action *actions,
3688 int actions_n, struct rte_flow_error *error)
3690 struct mlx5_flow_action_copy_mreg *cp_mreg =
3691 (struct mlx5_flow_action_copy_mreg *)
3692 (ext_actions + actions_n + 1);
3695 ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_RX, 0, error);
3699 ret = mlx5_flow_get_reg_id(dev, MLX5_METADATA_TX, 0, error);
3703 memcpy(ext_actions, actions,
3704 sizeof(*ext_actions) * actions_n);
3705 ext_actions[actions_n - 1] = (struct rte_flow_action){
3706 .type = MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
3709 ext_actions[actions_n] = (struct rte_flow_action){
3710 .type = RTE_FLOW_ACTION_TYPE_END,
3716 * The splitting for metadata feature.
3718 * - Q/RSS action on NIC Rx should be split in order to pass by
3719 * the mreg copy table (RX_CP_TBL) and then it jumps to the
3720 * action table (RX_ACT_TBL) which has the split Q/RSS action.
3722 * - All the actions on NIC Tx should have a mreg copy action to
3723 * copy reg_a from WQE to reg_c[0].
3726 * Pointer to Ethernet device.
3728 * Parent flow structure pointer.
3730 * Flow rule attributes.
3732 * Pattern specification (list terminated by the END pattern item).
3733 * @param[in] actions
3734 * Associated actions (list terminated by the END action).
3735 * @param[in] external
3736 * This flow rule is created by request external to PMD.
3738 * Perform verbose error reporting if not NULL.
3740 * 0 on success, negative value otherwise
3743 flow_create_split_metadata(struct rte_eth_dev *dev,
3744 struct rte_flow *flow,
3745 const struct rte_flow_attr *attr,
3746 const struct rte_flow_item items[],
3747 const struct rte_flow_action actions[],
3748 bool external, struct rte_flow_error *error)
3750 struct mlx5_priv *priv = dev->data->dev_private;
3751 struct mlx5_dev_config *config = &priv->config;
3752 const struct rte_flow_action *qrss = NULL;
3753 struct rte_flow_action *ext_actions = NULL;
3754 struct mlx5_flow *dev_flow = NULL;
3755 uint32_t qrss_id = 0;
3761 /* Check whether extensive metadata feature is engaged. */
3762 if (!config->dv_flow_en ||
3763 config->dv_xmeta_en == MLX5_XMETA_MODE_LEGACY ||
3764 !mlx5_flow_ext_mreg_supported(dev))
3765 return flow_create_split_inner(dev, flow, NULL, attr, items,
3766 actions, external, error);
3767 actions_n = flow_parse_qrss_action(actions, &qrss);
3769 /* Exclude hairpin flows from splitting. */
3770 if (qrss->type == RTE_FLOW_ACTION_TYPE_QUEUE) {
3771 const struct rte_flow_action_queue *queue;
3774 if (mlx5_rxq_get_type(dev, queue->index) ==
3775 MLX5_RXQ_TYPE_HAIRPIN)
3777 } else if (qrss->type == RTE_FLOW_ACTION_TYPE_RSS) {
3778 const struct rte_flow_action_rss *rss;
3781 if (mlx5_rxq_get_type(dev, rss->queue[0]) ==
3782 MLX5_RXQ_TYPE_HAIRPIN)
3787 /* Check if it is in meter suffix table. */
3788 mtr_sfx = attr->group == (attr->transfer ?
3789 (MLX5_FLOW_TABLE_LEVEL_SUFFIX - 1) :
3790 MLX5_FLOW_TABLE_LEVEL_SUFFIX);
3792 * Q/RSS action on NIC Rx should be split in order to pass by
3793 * the mreg copy table (RX_CP_TBL) and then it jumps to the
3794 * action table (RX_ACT_TBL) which has the split Q/RSS action.
3796 act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
3797 sizeof(struct rte_flow_action_set_tag) +
3798 sizeof(struct rte_flow_action_jump);
3799 ext_actions = rte_zmalloc(__func__, act_size, 0);
3801 return rte_flow_error_set(error, ENOMEM,
3802 RTE_FLOW_ERROR_TYPE_ACTION,
3803 NULL, "no memory to split "
3806 * If we are the suffix flow of meter, tag already exist.
3807 * Set the tag action to void.
3810 ext_actions[qrss - actions].type =
3811 RTE_FLOW_ACTION_TYPE_VOID;
3813 ext_actions[qrss - actions].type =
3814 MLX5_RTE_FLOW_ACTION_TYPE_TAG;
3816 * Create the new actions list with removed Q/RSS action
3817 * and appended set tag and jump to register copy table
3818 * (RX_CP_TBL). We should preallocate unique tag ID here
3819 * in advance, because it is needed for set tag action.
3821 qrss_id = flow_mreg_split_qrss_prep(dev, ext_actions, actions,
3822 qrss, actions_n, error);
3823 if (!mtr_sfx && !qrss_id) {
3827 } else if (attr->egress && !attr->transfer) {
3829 * All the actions on NIC Tx should have a metadata register
3830 * copy action to copy reg_a from WQE to reg_c[meta]
3832 act_size = sizeof(struct rte_flow_action) * (actions_n + 1) +
3833 sizeof(struct mlx5_flow_action_copy_mreg);
3834 ext_actions = rte_zmalloc(__func__, act_size, 0);
3836 return rte_flow_error_set(error, ENOMEM,
3837 RTE_FLOW_ERROR_TYPE_ACTION,
3838 NULL, "no memory to split "
3840 /* Create the action list appended with copy register. */
3841 ret = flow_mreg_tx_copy_prep(dev, ext_actions, actions,
3846 /* Add the unmodified original or prefix subflow. */
3847 ret = flow_create_split_inner(dev, flow, &dev_flow, attr, items,
3848 ext_actions ? ext_actions : actions,
3854 const struct rte_flow_attr q_attr = {
3855 .group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
3858 /* Internal PMD action to set register. */
3859 struct mlx5_rte_flow_item_tag q_tag_spec = {
3863 struct rte_flow_item q_items[] = {
3865 .type = MLX5_RTE_FLOW_ITEM_TYPE_TAG,
3866 .spec = &q_tag_spec,
3871 .type = RTE_FLOW_ITEM_TYPE_END,
3874 struct rte_flow_action q_actions[] = {
3880 .type = RTE_FLOW_ACTION_TYPE_END,
3883 uint64_t hash_fields = dev_flow->hash_fields;
3886 * Configure the tag item only if there is no meter subflow.
3887 * Since tag is already marked in the meter suffix subflow
3888 * we can just use the meter suffix items as is.
3891 /* Not meter subflow. */
3894 * Put unique id in prefix flow due to it is destroyed
3895 * after suffix flow and id will be freed after there
3896 * is no actual flows with this id and identifier
3897 * reallocation becomes possible (for example, for
3898 * other flows in other threads).
3900 dev_flow->qrss_id = qrss_id;
3902 ret = mlx5_flow_get_reg_id(dev, MLX5_COPY_MARK, 0,
3906 q_tag_spec.id = ret;
3909 /* Add suffix subflow to execute Q/RSS. */
3910 ret = flow_create_split_inner(dev, flow, &dev_flow,
3911 &q_attr, mtr_sfx ? items :
3917 dev_flow->hash_fields = hash_fields;
3922 * We do not destroy the partially created sub_flows in case of error.
3923 * These ones are included into parent flow list and will be destroyed
3924 * by flow_drv_destroy.
3926 flow_qrss_free_id(dev, qrss_id);
3927 rte_free(ext_actions);
3932 * The splitting for meter feature.
3934 * - The meter flow will be split to two flows as prefix and
3935 * suffix flow. The packets make sense only it pass the prefix
3938 * - Reg_C_5 is used for the packet to match betweend prefix and
3942 * Pointer to Ethernet device.
3944 * Parent flow structure pointer.
3946 * Flow rule attributes.
3948 * Pattern specification (list terminated by the END pattern item).
3949 * @param[in] actions
3950 * Associated actions (list terminated by the END action).
3951 * @param[in] external
3952 * This flow rule is created by request external to PMD.
3954 * Perform verbose error reporting if not NULL.
3956 * 0 on success, negative value otherwise
3959 flow_create_split_meter(struct rte_eth_dev *dev,
3960 struct rte_flow *flow,
3961 const struct rte_flow_attr *attr,
3962 const struct rte_flow_item items[],
3963 const struct rte_flow_action actions[],
3964 bool external, struct rte_flow_error *error)
3966 struct mlx5_priv *priv = dev->data->dev_private;
3967 struct rte_flow_action *sfx_actions = NULL;
3968 struct rte_flow_action *pre_actions = NULL;
3969 struct rte_flow_item *sfx_items = NULL;
3970 const struct rte_flow_item *sfx_port_id_item;
3971 struct mlx5_flow *dev_flow = NULL;
3972 struct rte_flow_attr sfx_attr = *attr;
3974 uint32_t mtr_tag_id = 0;
3981 actions_n = flow_check_meter_action(actions, &mtr);
3983 struct mlx5_rte_flow_item_tag *tag_spec;
3984 struct mlx5_rte_flow_item_tag *tag_mask;
3985 /* The five prefix actions: meter, decap, encap, tag, end. */
3986 act_size = sizeof(struct rte_flow_action) * (actions_n + 5) +
3987 sizeof(struct rte_flow_action_set_tag);
3989 #define METER_SUFFIX_ITEM 3
3990 item_size = sizeof(struct rte_flow_item) * METER_SUFFIX_ITEM +
3991 sizeof(struct mlx5_rte_flow_item_tag) * 2;
3992 sfx_actions = rte_zmalloc(__func__, (act_size + item_size), 0);
3994 return rte_flow_error_set(error, ENOMEM,
3995 RTE_FLOW_ERROR_TYPE_ACTION,
3996 NULL, "no memory to split "
3998 pre_actions = sfx_actions + actions_n;
3999 mtr_tag_id = flow_meter_split_prep(dev, actions, sfx_actions,
4005 /* Add the prefix subflow. */
4006 ret = flow_create_split_inner(dev, flow, &dev_flow, attr, items,
4007 pre_actions, external, error);
4012 dev_flow->mtr_flow_id = mtr_tag_id;
4013 /* Prepare the suffix flow match pattern. */
4014 sfx_items = (struct rte_flow_item *)((char *)sfx_actions +
4016 tag_spec = (struct mlx5_rte_flow_item_tag *)(sfx_items +
4018 tag_spec->data = dev_flow->mtr_flow_id << MLX5_MTR_COLOR_BITS;
4019 tag_spec->id = mlx5_flow_get_reg_id(dev, MLX5_MTR_SFX, 0,
4021 tag_mask = tag_spec + 1;
4022 tag_mask->data = 0xffffff00;
4023 sfx_items->type = MLX5_RTE_FLOW_ITEM_TYPE_TAG;
4024 sfx_items->spec = tag_spec;
4025 sfx_items->last = NULL;
4026 sfx_items->mask = tag_mask;
4028 sfx_port_id_item = find_port_id_item(items);
4029 if (sfx_port_id_item) {
4030 memcpy(sfx_items, sfx_port_id_item,
4031 sizeof(*sfx_items));
4034 sfx_items->type = RTE_FLOW_ITEM_TYPE_END;
4035 sfx_items -= sfx_port_id_item ? 2 : 1;
4036 /* Setting the sfx group atrr. */
4037 sfx_attr.group = sfx_attr.transfer ?
4038 (MLX5_FLOW_TABLE_LEVEL_SUFFIX - 1) :
4039 MLX5_FLOW_TABLE_LEVEL_SUFFIX;
4041 /* Add the prefix subflow. */
4042 ret = flow_create_split_metadata(dev, flow, &sfx_attr,
4043 sfx_items ? sfx_items : items,
4044 sfx_actions ? sfx_actions : actions,
4048 rte_free(sfx_actions);
4053 * Split the flow to subflow set. The splitters might be linked
4054 * in the chain, like this:
4055 * flow_create_split_outer() calls:
4056 * flow_create_split_meter() calls:
4057 * flow_create_split_metadata(meter_subflow_0) calls:
4058 * flow_create_split_inner(metadata_subflow_0)
4059 * flow_create_split_inner(metadata_subflow_1)
4060 * flow_create_split_inner(metadata_subflow_2)
4061 * flow_create_split_metadata(meter_subflow_1) calls:
4062 * flow_create_split_inner(metadata_subflow_0)
4063 * flow_create_split_inner(metadata_subflow_1)
4064 * flow_create_split_inner(metadata_subflow_2)
4066 * This provide flexible way to add new levels of flow splitting.
4067 * The all of successfully created subflows are included to the
4068 * parent flow dev_flow list.
4071 * Pointer to Ethernet device.
4073 * Parent flow structure pointer.
4075 * Flow rule attributes.
4077 * Pattern specification (list terminated by the END pattern item).
4078 * @param[in] actions
4079 * Associated actions (list terminated by the END action).
4080 * @param[in] external
4081 * This flow rule is created by request external to PMD.
4083 * Perform verbose error reporting if not NULL.
4085 * 0 on success, negative value otherwise
4088 flow_create_split_outer(struct rte_eth_dev *dev,
4089 struct rte_flow *flow,
4090 const struct rte_flow_attr *attr,
4091 const struct rte_flow_item items[],
4092 const struct rte_flow_action actions[],
4093 bool external, struct rte_flow_error *error)
4097 ret = flow_create_split_meter(dev, flow, attr, items,
4098 actions, external, error);
4104 * Create a flow and add it to @p list.
4107 * Pointer to Ethernet device.
4109 * Pointer to a TAILQ flow list. If this parameter NULL,
4110 * no list insertion occurred, flow is just created,
4111 * this is caller's responsibility to track the
4114 * Flow rule attributes.
4116 * Pattern specification (list terminated by the END pattern item).
4117 * @param[in] actions
4118 * Associated actions (list terminated by the END action).
4119 * @param[in] external
4120 * This flow rule is created by request external to PMD.
4122 * Perform verbose error reporting if not NULL.
4125 * A flow on success, NULL otherwise and rte_errno is set.
4127 static struct rte_flow *
4128 flow_list_create(struct rte_eth_dev *dev, struct mlx5_flows *list,
4129 const struct rte_flow_attr *attr,
4130 const struct rte_flow_item items[],
4131 const struct rte_flow_action actions[],
4132 bool external, struct rte_flow_error *error)
4134 struct mlx5_priv *priv = dev->data->dev_private;
4135 struct rte_flow *flow = NULL;
4136 struct mlx5_flow *dev_flow;
4137 const struct rte_flow_action_rss *rss;
4139 struct rte_flow_expand_rss buf;
4140 uint8_t buffer[2048];
4143 struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
4144 uint8_t buffer[2048];
4147 struct rte_flow_action actions[MLX5_MAX_SPLIT_ACTIONS];
4148 uint8_t buffer[2048];
4149 } actions_hairpin_tx;
4151 struct rte_flow_item items[MLX5_MAX_SPLIT_ITEMS];
4152 uint8_t buffer[2048];
4154 struct rte_flow_expand_rss *buf = &expand_buffer.buf;
4155 const struct rte_flow_action *p_actions_rx = actions;
4159 int hairpin_flow = 0;
4160 uint32_t hairpin_id = 0;
4161 struct rte_flow_attr attr_tx = { .priority = 0 };
4163 hairpin_flow = flow_check_hairpin_split(dev, attr, actions);
4164 if (hairpin_flow > 0) {
4165 if (hairpin_flow > MLX5_MAX_SPLIT_ACTIONS) {
4169 flow_hairpin_split(dev, actions, actions_rx.actions,
4170 actions_hairpin_tx.actions, items_tx.items,
4172 p_actions_rx = actions_rx.actions;
4174 ret = flow_drv_validate(dev, attr, items, p_actions_rx, external,
4177 goto error_before_flow;
4178 flow_size = sizeof(struct rte_flow);
4179 rss = flow_get_rss_action(p_actions_rx);
4181 flow_size += RTE_ALIGN_CEIL(rss->queue_num * sizeof(uint16_t),
4184 flow_size += RTE_ALIGN_CEIL(sizeof(uint16_t), sizeof(void *));
4185 flow = rte_calloc(__func__, 1, flow_size, 0);
4188 goto error_before_flow;
4190 flow->drv_type = flow_get_drv_type(dev, attr);
4191 if (hairpin_id != 0)
4192 flow->hairpin_flow_id = hairpin_id;
4193 assert(flow->drv_type > MLX5_FLOW_TYPE_MIN &&
4194 flow->drv_type < MLX5_FLOW_TYPE_MAX);
4195 flow->rss.queue = (void *)(flow + 1);
4198 * The following information is required by
4199 * mlx5_flow_hashfields_adjust() in advance.
4201 flow->rss.level = rss->level;
4202 /* RSS type 0 indicates default RSS type (ETH_RSS_IP). */
4203 flow->rss.types = !rss->types ? ETH_RSS_IP : rss->types;
4205 LIST_INIT(&flow->dev_flows);
4206 if (rss && rss->types) {
4207 unsigned int graph_root;
4209 graph_root = find_graph_root(items, rss->level);
4210 ret = rte_flow_expand_rss(buf, sizeof(expand_buffer.buffer),
4212 mlx5_support_expansion,
4215 (unsigned int)ret < sizeof(expand_buffer.buffer));
4218 buf->entry[0].pattern = (void *)(uintptr_t)items;
4220 for (i = 0; i < buf->entries; ++i) {
4222 * The splitter may create multiple dev_flows,
4223 * depending on configuration. In the simplest
4224 * case it just creates unmodified original flow.
4226 ret = flow_create_split_outer(dev, flow, attr,
4227 buf->entry[i].pattern,
4228 p_actions_rx, external,
4233 /* Create the tx flow. */
4235 attr_tx.group = MLX5_HAIRPIN_TX_TABLE;
4236 attr_tx.ingress = 0;
4238 dev_flow = flow_drv_prepare(flow, &attr_tx, items_tx.items,
4239 actions_hairpin_tx.actions, error);
4242 dev_flow->flow = flow;
4243 dev_flow->external = 0;
4244 LIST_INSERT_HEAD(&flow->dev_flows, dev_flow, next);
4245 ret = flow_drv_translate(dev, dev_flow, &attr_tx,
4247 actions_hairpin_tx.actions, error);
4252 * Update the metadata register copy table. If extensive
4253 * metadata feature is enabled and registers are supported
4254 * we might create the extra rte_flow for each unique
4255 * MARK/FLAG action ID.
4257 * The table is updated for ingress Flows only, because
4258 * the egress Flows belong to the different device and
4259 * copy table should be updated in peer NIC Rx domain.
4261 if (attr->ingress &&
4262 (external || attr->group != MLX5_FLOW_MREG_CP_TABLE_GROUP)) {
4263 ret = flow_mreg_update_copy_table(dev, flow, actions, error);
4267 if (dev->data->dev_started) {
4268 ret = flow_drv_apply(dev, flow, error);
4273 TAILQ_INSERT_TAIL(list, flow, next);
4274 flow_rxq_flags_set(dev, flow);
4278 mlx5_flow_id_release(priv->sh->flow_id_pool,
4283 flow_mreg_del_copy_action(dev, flow);
4284 ret = rte_errno; /* Save rte_errno before cleanup. */
4285 if (flow->hairpin_flow_id)
4286 mlx5_flow_id_release(priv->sh->flow_id_pool,
4287 flow->hairpin_flow_id);
4289 flow_drv_destroy(dev, flow);
4291 rte_errno = ret; /* Restore rte_errno. */
4296 * Create a dedicated flow rule on e-switch table 0 (root table), to direct all
4297 * incoming packets to table 1.
4299 * Other flow rules, requested for group n, will be created in
4300 * e-switch table n+1.
4301 * Jump action to e-switch group n will be created to group n+1.
4303 * Used when working in switchdev mode, to utilise advantages of table 1
4307 * Pointer to Ethernet device.
4310 * Pointer to flow on success, NULL otherwise and rte_errno is set.
4313 mlx5_flow_create_esw_table_zero_flow(struct rte_eth_dev *dev)
4315 const struct rte_flow_attr attr = {
4322 const struct rte_flow_item pattern = {
4323 .type = RTE_FLOW_ITEM_TYPE_END,
4325 struct rte_flow_action_jump jump = {
4328 const struct rte_flow_action actions[] = {
4330 .type = RTE_FLOW_ACTION_TYPE_JUMP,
4334 .type = RTE_FLOW_ACTION_TYPE_END,
4337 struct mlx5_priv *priv = dev->data->dev_private;
4338 struct rte_flow_error error;
4340 return flow_list_create(dev, &priv->ctrl_flows, &attr, &pattern,
4341 actions, false, &error);
4347 * @see rte_flow_create()
4351 mlx5_flow_create(struct rte_eth_dev *dev,
4352 const struct rte_flow_attr *attr,
4353 const struct rte_flow_item items[],
4354 const struct rte_flow_action actions[],
4355 struct rte_flow_error *error)
4357 struct mlx5_priv *priv = dev->data->dev_private;
4359 return flow_list_create(dev, &priv->flows,
4360 attr, items, actions, true, error);
4364 * Destroy a flow in a list.
4367 * Pointer to Ethernet device.
4369 * Pointer to a TAILQ flow list. If this parameter NULL,
4370 * there is no flow removal from the list.
4375 flow_list_destroy(struct rte_eth_dev *dev, struct mlx5_flows *list,
4376 struct rte_flow *flow)
4378 struct mlx5_priv *priv = dev->data->dev_private;
4381 * Update RX queue flags only if port is started, otherwise it is
4384 if (dev->data->dev_started)
4385 flow_rxq_flags_trim(dev, flow);
4386 if (flow->hairpin_flow_id)
4387 mlx5_flow_id_release(priv->sh->flow_id_pool,
4388 flow->hairpin_flow_id);
4389 flow_drv_destroy(dev, flow);
4391 TAILQ_REMOVE(list, flow, next);
4392 flow_mreg_del_copy_action(dev, flow);
4393 rte_free(flow->fdir);
4398 * Destroy all flows.
4401 * Pointer to Ethernet device.
4403 * Pointer to a TAILQ flow list.
4406 mlx5_flow_list_flush(struct rte_eth_dev *dev, struct mlx5_flows *list)
4408 while (!TAILQ_EMPTY(list)) {
4409 struct rte_flow *flow;
4411 flow = TAILQ_FIRST(list);
4412 flow_list_destroy(dev, list, flow);
4420 * Pointer to Ethernet device.
4422 * Pointer to a TAILQ flow list.
4425 mlx5_flow_stop(struct rte_eth_dev *dev, struct mlx5_flows *list)
4427 struct rte_flow *flow;
4429 TAILQ_FOREACH_REVERSE(flow, list, mlx5_flows, next) {
4430 flow_drv_remove(dev, flow);
4431 flow_mreg_stop_copy_action(dev, flow);
4433 flow_mreg_del_default_copy_action(dev);
4434 flow_rxq_flags_clear(dev);
4441 * Pointer to Ethernet device.
4443 * Pointer to a TAILQ flow list.
4446 * 0 on success, a negative errno value otherwise and rte_errno is set.
4449 mlx5_flow_start(struct rte_eth_dev *dev, struct mlx5_flows *list)
4451 struct rte_flow *flow;
4452 struct rte_flow_error error;
4455 /* Make sure default copy action (reg_c[0] -> reg_b) is created. */
4456 ret = flow_mreg_add_default_copy_action(dev, &error);
4459 /* Apply Flows created by application. */
4460 TAILQ_FOREACH(flow, list, next) {
4461 ret = flow_mreg_start_copy_action(dev, flow);
4464 ret = flow_drv_apply(dev, flow, &error);
4467 flow_rxq_flags_set(dev, flow);
4471 ret = rte_errno; /* Save rte_errno before cleanup. */
4472 mlx5_flow_stop(dev, list);
4473 rte_errno = ret; /* Restore rte_errno. */
4478 * Verify the flow list is empty
4481 * Pointer to Ethernet device.
4483 * @return the number of flows not released.
4486 mlx5_flow_verify(struct rte_eth_dev *dev)
4488 struct mlx5_priv *priv = dev->data->dev_private;
4489 struct rte_flow *flow;
4492 TAILQ_FOREACH(flow, &priv->flows, next) {
4493 DRV_LOG(DEBUG, "port %u flow %p still referenced",
4494 dev->data->port_id, (void *)flow);
4501 * Enable default hairpin egress flow.
4504 * Pointer to Ethernet device.
4509 * 0 on success, a negative errno value otherwise and rte_errno is set.
4512 mlx5_ctrl_flow_source_queue(struct rte_eth_dev *dev,
4515 struct mlx5_priv *priv = dev->data->dev_private;
4516 const struct rte_flow_attr attr = {
4520 struct mlx5_rte_flow_item_tx_queue queue_spec = {
4523 struct mlx5_rte_flow_item_tx_queue queue_mask = {
4524 .queue = UINT32_MAX,
4526 struct rte_flow_item items[] = {
4528 .type = MLX5_RTE_FLOW_ITEM_TYPE_TX_QUEUE,
4529 .spec = &queue_spec,
4531 .mask = &queue_mask,
4534 .type = RTE_FLOW_ITEM_TYPE_END,
4537 struct rte_flow_action_jump jump = {
4538 .group = MLX5_HAIRPIN_TX_TABLE,
4540 struct rte_flow_action actions[2];
4541 struct rte_flow *flow;
4542 struct rte_flow_error error;
4544 actions[0].type = RTE_FLOW_ACTION_TYPE_JUMP;
4545 actions[0].conf = &jump;
4546 actions[1].type = RTE_FLOW_ACTION_TYPE_END;
4547 flow = flow_list_create(dev, &priv->ctrl_flows,
4548 &attr, items, actions, false, &error);
4551 "Failed to create ctrl flow: rte_errno(%d),"
4552 " type(%d), message(%s)",
4553 rte_errno, error.type,
4554 error.message ? error.message : " (no stated reason)");
4561 * Enable a control flow configured from the control plane.
4564 * Pointer to Ethernet device.
4566 * An Ethernet flow spec to apply.
4568 * An Ethernet flow mask to apply.
4570 * A VLAN flow spec to apply.
4572 * A VLAN flow mask to apply.
4575 * 0 on success, a negative errno value otherwise and rte_errno is set.
4578 mlx5_ctrl_flow_vlan(struct rte_eth_dev *dev,
4579 struct rte_flow_item_eth *eth_spec,
4580 struct rte_flow_item_eth *eth_mask,
4581 struct rte_flow_item_vlan *vlan_spec,
4582 struct rte_flow_item_vlan *vlan_mask)
4584 struct mlx5_priv *priv = dev->data->dev_private;
4585 const struct rte_flow_attr attr = {
4587 .priority = MLX5_FLOW_PRIO_RSVD,
4589 struct rte_flow_item items[] = {
4591 .type = RTE_FLOW_ITEM_TYPE_ETH,
4597 .type = (vlan_spec) ? RTE_FLOW_ITEM_TYPE_VLAN :
4598 RTE_FLOW_ITEM_TYPE_END,
4604 .type = RTE_FLOW_ITEM_TYPE_END,
4607 uint16_t queue[priv->reta_idx_n];
4608 struct rte_flow_action_rss action_rss = {
4609 .func = RTE_ETH_HASH_FUNCTION_DEFAULT,
4611 .types = priv->rss_conf.rss_hf,
4612 .key_len = priv->rss_conf.rss_key_len,
4613 .queue_num = priv->reta_idx_n,
4614 .key = priv->rss_conf.rss_key,
4617 struct rte_flow_action actions[] = {
4619 .type = RTE_FLOW_ACTION_TYPE_RSS,
4620 .conf = &action_rss,
4623 .type = RTE_FLOW_ACTION_TYPE_END,
4626 struct rte_flow *flow;
4627 struct rte_flow_error error;
4630 if (!priv->reta_idx_n || !priv->rxqs_n) {
4633 for (i = 0; i != priv->reta_idx_n; ++i)
4634 queue[i] = (*priv->reta_idx)[i];
4635 flow = flow_list_create(dev, &priv->ctrl_flows,
4636 &attr, items, actions, false, &error);
4643 * Enable a flow control configured from the control plane.
4646 * Pointer to Ethernet device.
4648 * An Ethernet flow spec to apply.
4650 * An Ethernet flow mask to apply.
4653 * 0 on success, a negative errno value otherwise and rte_errno is set.
4656 mlx5_ctrl_flow(struct rte_eth_dev *dev,
4657 struct rte_flow_item_eth *eth_spec,
4658 struct rte_flow_item_eth *eth_mask)
4660 return mlx5_ctrl_flow_vlan(dev, eth_spec, eth_mask, NULL, NULL);
4666 * @see rte_flow_destroy()
4670 mlx5_flow_destroy(struct rte_eth_dev *dev,
4671 struct rte_flow *flow,
4672 struct rte_flow_error *error __rte_unused)
4674 struct mlx5_priv *priv = dev->data->dev_private;
4676 flow_list_destroy(dev, &priv->flows, flow);
4681 * Destroy all flows.
4683 * @see rte_flow_flush()
4687 mlx5_flow_flush(struct rte_eth_dev *dev,
4688 struct rte_flow_error *error __rte_unused)
4690 struct mlx5_priv *priv = dev->data->dev_private;
4692 mlx5_flow_list_flush(dev, &priv->flows);
4699 * @see rte_flow_isolate()
4703 mlx5_flow_isolate(struct rte_eth_dev *dev,
4705 struct rte_flow_error *error)
4707 struct mlx5_priv *priv = dev->data->dev_private;
4709 if (dev->data->dev_started) {
4710 rte_flow_error_set(error, EBUSY,
4711 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
4713 "port must be stopped first");
4716 priv->isolated = !!enable;
4718 dev->dev_ops = &mlx5_dev_ops_isolate;
4720 dev->dev_ops = &mlx5_dev_ops;
4727 * @see rte_flow_query()
4731 flow_drv_query(struct rte_eth_dev *dev,
4732 struct rte_flow *flow,
4733 const struct rte_flow_action *actions,
4735 struct rte_flow_error *error)
4737 const struct mlx5_flow_driver_ops *fops;
4738 enum mlx5_flow_drv_type ftype = flow->drv_type;
4740 assert(ftype > MLX5_FLOW_TYPE_MIN && ftype < MLX5_FLOW_TYPE_MAX);
4741 fops = flow_get_drv_ops(ftype);
4743 return fops->query(dev, flow, actions, data, error);
4749 * @see rte_flow_query()
4753 mlx5_flow_query(struct rte_eth_dev *dev,
4754 struct rte_flow *flow,
4755 const struct rte_flow_action *actions,
4757 struct rte_flow_error *error)
4761 ret = flow_drv_query(dev, flow, actions, data, error);
4768 * Convert a flow director filter to a generic flow.
4771 * Pointer to Ethernet device.
4772 * @param fdir_filter
4773 * Flow director filter to add.
4775 * Generic flow parameters structure.
4778 * 0 on success, a negative errno value otherwise and rte_errno is set.
4781 flow_fdir_filter_convert(struct rte_eth_dev *dev,
4782 const struct rte_eth_fdir_filter *fdir_filter,
4783 struct mlx5_fdir *attributes)
4785 struct mlx5_priv *priv = dev->data->dev_private;
4786 const struct rte_eth_fdir_input *input = &fdir_filter->input;
4787 const struct rte_eth_fdir_masks *mask =
4788 &dev->data->dev_conf.fdir_conf.mask;
4790 /* Validate queue number. */
4791 if (fdir_filter->action.rx_queue >= priv->rxqs_n) {
4792 DRV_LOG(ERR, "port %u invalid queue number %d",
4793 dev->data->port_id, fdir_filter->action.rx_queue);
4797 attributes->attr.ingress = 1;
4798 attributes->items[0] = (struct rte_flow_item) {
4799 .type = RTE_FLOW_ITEM_TYPE_ETH,
4800 .spec = &attributes->l2,
4801 .mask = &attributes->l2_mask,
4803 switch (fdir_filter->action.behavior) {
4804 case RTE_ETH_FDIR_ACCEPT:
4805 attributes->actions[0] = (struct rte_flow_action){
4806 .type = RTE_FLOW_ACTION_TYPE_QUEUE,
4807 .conf = &attributes->queue,
4810 case RTE_ETH_FDIR_REJECT:
4811 attributes->actions[0] = (struct rte_flow_action){
4812 .type = RTE_FLOW_ACTION_TYPE_DROP,
4816 DRV_LOG(ERR, "port %u invalid behavior %d",
4818 fdir_filter->action.behavior);
4819 rte_errno = ENOTSUP;
4822 attributes->queue.index = fdir_filter->action.rx_queue;
4824 switch (fdir_filter->input.flow_type) {
4825 case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
4826 case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
4827 case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
4828 attributes->l3.ipv4.hdr = (struct rte_ipv4_hdr){
4829 .src_addr = input->flow.ip4_flow.src_ip,
4830 .dst_addr = input->flow.ip4_flow.dst_ip,
4831 .time_to_live = input->flow.ip4_flow.ttl,
4832 .type_of_service = input->flow.ip4_flow.tos,
4834 attributes->l3_mask.ipv4.hdr = (struct rte_ipv4_hdr){
4835 .src_addr = mask->ipv4_mask.src_ip,
4836 .dst_addr = mask->ipv4_mask.dst_ip,
4837 .time_to_live = mask->ipv4_mask.ttl,
4838 .type_of_service = mask->ipv4_mask.tos,
4839 .next_proto_id = mask->ipv4_mask.proto,
4841 attributes->items[1] = (struct rte_flow_item){
4842 .type = RTE_FLOW_ITEM_TYPE_IPV4,
4843 .spec = &attributes->l3,
4844 .mask = &attributes->l3_mask,
4847 case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
4848 case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
4849 case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
4850 attributes->l3.ipv6.hdr = (struct rte_ipv6_hdr){
4851 .hop_limits = input->flow.ipv6_flow.hop_limits,
4852 .proto = input->flow.ipv6_flow.proto,
4855 memcpy(attributes->l3.ipv6.hdr.src_addr,
4856 input->flow.ipv6_flow.src_ip,
4857 RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
4858 memcpy(attributes->l3.ipv6.hdr.dst_addr,
4859 input->flow.ipv6_flow.dst_ip,
4860 RTE_DIM(attributes->l3.ipv6.hdr.src_addr));
4861 memcpy(attributes->l3_mask.ipv6.hdr.src_addr,
4862 mask->ipv6_mask.src_ip,
4863 RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
4864 memcpy(attributes->l3_mask.ipv6.hdr.dst_addr,
4865 mask->ipv6_mask.dst_ip,
4866 RTE_DIM(attributes->l3_mask.ipv6.hdr.src_addr));
4867 attributes->items[1] = (struct rte_flow_item){
4868 .type = RTE_FLOW_ITEM_TYPE_IPV6,
4869 .spec = &attributes->l3,
4870 .mask = &attributes->l3_mask,
4874 DRV_LOG(ERR, "port %u invalid flow type%d",
4875 dev->data->port_id, fdir_filter->input.flow_type);
4876 rte_errno = ENOTSUP;
4880 switch (fdir_filter->input.flow_type) {
4881 case RTE_ETH_FLOW_NONFRAG_IPV4_UDP:
4882 attributes->l4.udp.hdr = (struct rte_udp_hdr){
4883 .src_port = input->flow.udp4_flow.src_port,
4884 .dst_port = input->flow.udp4_flow.dst_port,
4886 attributes->l4_mask.udp.hdr = (struct rte_udp_hdr){
4887 .src_port = mask->src_port_mask,
4888 .dst_port = mask->dst_port_mask,
4890 attributes->items[2] = (struct rte_flow_item){
4891 .type = RTE_FLOW_ITEM_TYPE_UDP,
4892 .spec = &attributes->l4,
4893 .mask = &attributes->l4_mask,
4896 case RTE_ETH_FLOW_NONFRAG_IPV4_TCP:
4897 attributes->l4.tcp.hdr = (struct rte_tcp_hdr){
4898 .src_port = input->flow.tcp4_flow.src_port,
4899 .dst_port = input->flow.tcp4_flow.dst_port,
4901 attributes->l4_mask.tcp.hdr = (struct rte_tcp_hdr){
4902 .src_port = mask->src_port_mask,
4903 .dst_port = mask->dst_port_mask,
4905 attributes->items[2] = (struct rte_flow_item){
4906 .type = RTE_FLOW_ITEM_TYPE_TCP,
4907 .spec = &attributes->l4,
4908 .mask = &attributes->l4_mask,
4911 case RTE_ETH_FLOW_NONFRAG_IPV6_UDP:
4912 attributes->l4.udp.hdr = (struct rte_udp_hdr){
4913 .src_port = input->flow.udp6_flow.src_port,
4914 .dst_port = input->flow.udp6_flow.dst_port,
4916 attributes->l4_mask.udp.hdr = (struct rte_udp_hdr){
4917 .src_port = mask->src_port_mask,
4918 .dst_port = mask->dst_port_mask,
4920 attributes->items[2] = (struct rte_flow_item){
4921 .type = RTE_FLOW_ITEM_TYPE_UDP,
4922 .spec = &attributes->l4,
4923 .mask = &attributes->l4_mask,
4926 case RTE_ETH_FLOW_NONFRAG_IPV6_TCP:
4927 attributes->l4.tcp.hdr = (struct rte_tcp_hdr){
4928 .src_port = input->flow.tcp6_flow.src_port,
4929 .dst_port = input->flow.tcp6_flow.dst_port,
4931 attributes->l4_mask.tcp.hdr = (struct rte_tcp_hdr){
4932 .src_port = mask->src_port_mask,
4933 .dst_port = mask->dst_port_mask,
4935 attributes->items[2] = (struct rte_flow_item){
4936 .type = RTE_FLOW_ITEM_TYPE_TCP,
4937 .spec = &attributes->l4,
4938 .mask = &attributes->l4_mask,
4941 case RTE_ETH_FLOW_NONFRAG_IPV4_OTHER:
4942 case RTE_ETH_FLOW_NONFRAG_IPV6_OTHER:
4945 DRV_LOG(ERR, "port %u invalid flow type%d",
4946 dev->data->port_id, fdir_filter->input.flow_type);
4947 rte_errno = ENOTSUP;
4953 #define FLOW_FDIR_CMP(f1, f2, fld) \
4954 memcmp(&(f1)->fld, &(f2)->fld, sizeof(f1->fld))
4957 * Compare two FDIR flows. If items and actions are identical, the two flows are
4961 * Pointer to Ethernet device.
4963 * FDIR flow to compare.
4965 * FDIR flow to compare.
4968 * Zero on match, 1 otherwise.
4971 flow_fdir_cmp(const struct mlx5_fdir *f1, const struct mlx5_fdir *f2)
4973 if (FLOW_FDIR_CMP(f1, f2, attr) ||
4974 FLOW_FDIR_CMP(f1, f2, l2) ||
4975 FLOW_FDIR_CMP(f1, f2, l2_mask) ||
4976 FLOW_FDIR_CMP(f1, f2, l3) ||
4977 FLOW_FDIR_CMP(f1, f2, l3_mask) ||
4978 FLOW_FDIR_CMP(f1, f2, l4) ||
4979 FLOW_FDIR_CMP(f1, f2, l4_mask) ||
4980 FLOW_FDIR_CMP(f1, f2, actions[0].type))
4982 if (f1->actions[0].type == RTE_FLOW_ACTION_TYPE_QUEUE &&
4983 FLOW_FDIR_CMP(f1, f2, queue))
4989 * Search device flow list to find out a matched FDIR flow.
4992 * Pointer to Ethernet device.
4994 * FDIR flow to lookup.
4997 * Pointer of flow if found, NULL otherwise.
4999 static struct rte_flow *
5000 flow_fdir_filter_lookup(struct rte_eth_dev *dev, struct mlx5_fdir *fdir_flow)
5002 struct mlx5_priv *priv = dev->data->dev_private;
5003 struct rte_flow *flow = NULL;
5006 TAILQ_FOREACH(flow, &priv->flows, next) {
5007 if (flow->fdir && !flow_fdir_cmp(flow->fdir, fdir_flow)) {
5008 DRV_LOG(DEBUG, "port %u found FDIR flow %p",
5009 dev->data->port_id, (void *)flow);
5017 * Add new flow director filter and store it in list.
5020 * Pointer to Ethernet device.
5021 * @param fdir_filter
5022 * Flow director filter to add.
5025 * 0 on success, a negative errno value otherwise and rte_errno is set.
5028 flow_fdir_filter_add(struct rte_eth_dev *dev,
5029 const struct rte_eth_fdir_filter *fdir_filter)
5031 struct mlx5_priv *priv = dev->data->dev_private;
5032 struct mlx5_fdir *fdir_flow;
5033 struct rte_flow *flow;
5036 fdir_flow = rte_zmalloc(__func__, sizeof(*fdir_flow), 0);
5041 ret = flow_fdir_filter_convert(dev, fdir_filter, fdir_flow);
5044 flow = flow_fdir_filter_lookup(dev, fdir_flow);
5049 flow = flow_list_create(dev, &priv->flows, &fdir_flow->attr,
5050 fdir_flow->items, fdir_flow->actions, true,
5054 assert(!flow->fdir);
5055 flow->fdir = fdir_flow;
5056 DRV_LOG(DEBUG, "port %u created FDIR flow %p",
5057 dev->data->port_id, (void *)flow);
5060 rte_free(fdir_flow);
5065 * Delete specific filter.
5068 * Pointer to Ethernet device.
5069 * @param fdir_filter
5070 * Filter to be deleted.
5073 * 0 on success, a negative errno value otherwise and rte_errno is set.
5076 flow_fdir_filter_delete(struct rte_eth_dev *dev,
5077 const struct rte_eth_fdir_filter *fdir_filter)
5079 struct mlx5_priv *priv = dev->data->dev_private;
5080 struct rte_flow *flow;
5081 struct mlx5_fdir fdir_flow = {
5086 ret = flow_fdir_filter_convert(dev, fdir_filter, &fdir_flow);
5089 flow = flow_fdir_filter_lookup(dev, &fdir_flow);
5094 flow_list_destroy(dev, &priv->flows, flow);
5095 DRV_LOG(DEBUG, "port %u deleted FDIR flow %p",
5096 dev->data->port_id, (void *)flow);
5101 * Update queue for specific filter.
5104 * Pointer to Ethernet device.
5105 * @param fdir_filter
5106 * Filter to be updated.
5109 * 0 on success, a negative errno value otherwise and rte_errno is set.
5112 flow_fdir_filter_update(struct rte_eth_dev *dev,
5113 const struct rte_eth_fdir_filter *fdir_filter)
5117 ret = flow_fdir_filter_delete(dev, fdir_filter);
5120 return flow_fdir_filter_add(dev, fdir_filter);
5124 * Flush all filters.
5127 * Pointer to Ethernet device.
5130 flow_fdir_filter_flush(struct rte_eth_dev *dev)
5132 struct mlx5_priv *priv = dev->data->dev_private;
5134 mlx5_flow_list_flush(dev, &priv->flows);
5138 * Get flow director information.
5141 * Pointer to Ethernet device.
5142 * @param[out] fdir_info
5143 * Resulting flow director information.
5146 flow_fdir_info_get(struct rte_eth_dev *dev, struct rte_eth_fdir_info *fdir_info)
5148 struct rte_eth_fdir_masks *mask =
5149 &dev->data->dev_conf.fdir_conf.mask;
5151 fdir_info->mode = dev->data->dev_conf.fdir_conf.mode;
5152 fdir_info->guarant_spc = 0;
5153 rte_memcpy(&fdir_info->mask, mask, sizeof(fdir_info->mask));
5154 fdir_info->max_flexpayload = 0;
5155 fdir_info->flow_types_mask[0] = 0;
5156 fdir_info->flex_payload_unit = 0;
5157 fdir_info->max_flex_payload_segment_num = 0;
5158 fdir_info->flex_payload_limit = 0;
5159 memset(&fdir_info->flex_conf, 0, sizeof(fdir_info->flex_conf));
5163 * Deal with flow director operations.
5166 * Pointer to Ethernet device.
5168 * Operation to perform.
5170 * Pointer to operation-specific structure.
5173 * 0 on success, a negative errno value otherwise and rte_errno is set.
5176 flow_fdir_ctrl_func(struct rte_eth_dev *dev, enum rte_filter_op filter_op,
5179 enum rte_fdir_mode fdir_mode =
5180 dev->data->dev_conf.fdir_conf.mode;
5182 if (filter_op == RTE_ETH_FILTER_NOP)
5184 if (fdir_mode != RTE_FDIR_MODE_PERFECT &&
5185 fdir_mode != RTE_FDIR_MODE_PERFECT_MAC_VLAN) {
5186 DRV_LOG(ERR, "port %u flow director mode %d not supported",
5187 dev->data->port_id, fdir_mode);
5191 switch (filter_op) {
5192 case RTE_ETH_FILTER_ADD:
5193 return flow_fdir_filter_add(dev, arg);
5194 case RTE_ETH_FILTER_UPDATE:
5195 return flow_fdir_filter_update(dev, arg);
5196 case RTE_ETH_FILTER_DELETE:
5197 return flow_fdir_filter_delete(dev, arg);
5198 case RTE_ETH_FILTER_FLUSH:
5199 flow_fdir_filter_flush(dev);
5201 case RTE_ETH_FILTER_INFO:
5202 flow_fdir_info_get(dev, arg);
5205 DRV_LOG(DEBUG, "port %u unknown operation %u",
5206 dev->data->port_id, filter_op);
5214 * Manage filter operations.
5217 * Pointer to Ethernet device structure.
5218 * @param filter_type
5221 * Operation to perform.
5223 * Pointer to operation-specific structure.
5226 * 0 on success, a negative errno value otherwise and rte_errno is set.
5229 mlx5_dev_filter_ctrl(struct rte_eth_dev *dev,
5230 enum rte_filter_type filter_type,
5231 enum rte_filter_op filter_op,
5234 switch (filter_type) {
5235 case RTE_ETH_FILTER_GENERIC:
5236 if (filter_op != RTE_ETH_FILTER_GET) {
5240 *(const void **)arg = &mlx5_flow_ops;
5242 case RTE_ETH_FILTER_FDIR:
5243 return flow_fdir_ctrl_func(dev, filter_op, arg);
5245 DRV_LOG(ERR, "port %u filter type (%d) not supported",
5246 dev->data->port_id, filter_type);
5247 rte_errno = ENOTSUP;
5254 * Create the needed meter and suffix tables.
5257 * Pointer to Ethernet device.
5259 * Pointer to the flow meter.
5262 * Pointer to table set on success, NULL otherwise.
5264 struct mlx5_meter_domains_infos *
5265 mlx5_flow_create_mtr_tbls(struct rte_eth_dev *dev,
5266 const struct mlx5_flow_meter *fm)
5268 const struct mlx5_flow_driver_ops *fops;
5270 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5271 return fops->create_mtr_tbls(dev, fm);
5275 * Destroy the meter table set.
5278 * Pointer to Ethernet device.
5280 * Pointer to the meter table set.
5286 mlx5_flow_destroy_mtr_tbls(struct rte_eth_dev *dev,
5287 struct mlx5_meter_domains_infos *tbls)
5289 const struct mlx5_flow_driver_ops *fops;
5291 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5292 return fops->destroy_mtr_tbls(dev, tbls);
5296 * Create policer rules.
5299 * Pointer to Ethernet device.
5301 * Pointer to flow meter structure.
5303 * Pointer to flow attributes.
5306 * 0 on success, -1 otherwise.
5309 mlx5_flow_create_policer_rules(struct rte_eth_dev *dev,
5310 struct mlx5_flow_meter *fm,
5311 const struct rte_flow_attr *attr)
5313 const struct mlx5_flow_driver_ops *fops;
5315 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5316 return fops->create_policer_rules(dev, fm, attr);
5320 * Destroy policer rules.
5323 * Pointer to flow meter structure.
5325 * Pointer to flow attributes.
5328 * 0 on success, -1 otherwise.
5331 mlx5_flow_destroy_policer_rules(struct rte_eth_dev *dev,
5332 struct mlx5_flow_meter *fm,
5333 const struct rte_flow_attr *attr)
5335 const struct mlx5_flow_driver_ops *fops;
5337 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5338 return fops->destroy_policer_rules(dev, fm, attr);
5342 * Allocate a counter.
5345 * Pointer to Ethernet device structure.
5348 * Pointer to allocated counter on success, NULL otherwise.
5350 struct mlx5_flow_counter *
5351 mlx5_counter_alloc(struct rte_eth_dev *dev)
5353 const struct mlx5_flow_driver_ops *fops;
5354 struct rte_flow_attr attr = { .transfer = 0 };
5356 if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
5357 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5358 return fops->counter_alloc(dev);
5361 "port %u counter allocate is not supported.",
5362 dev->data->port_id);
5370 * Pointer to Ethernet device structure.
5372 * Pointer to counter to be free.
5375 mlx5_counter_free(struct rte_eth_dev *dev, struct mlx5_flow_counter *cnt)
5377 const struct mlx5_flow_driver_ops *fops;
5378 struct rte_flow_attr attr = { .transfer = 0 };
5380 if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
5381 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5382 fops->counter_free(dev, cnt);
5386 "port %u counter free is not supported.",
5387 dev->data->port_id);
5391 * Query counter statistics.
5394 * Pointer to Ethernet device structure.
5396 * Pointer to counter to query.
5398 * Set to clear counter statistics.
5400 * The counter hits packets number to save.
5402 * The counter hits bytes number to save.
5405 * 0 on success, a negative errno value otherwise.
5408 mlx5_counter_query(struct rte_eth_dev *dev, struct mlx5_flow_counter *cnt,
5409 bool clear, uint64_t *pkts, uint64_t *bytes)
5411 const struct mlx5_flow_driver_ops *fops;
5412 struct rte_flow_attr attr = { .transfer = 0 };
5414 if (flow_get_drv_type(dev, &attr) == MLX5_FLOW_TYPE_DV) {
5415 fops = flow_get_drv_ops(MLX5_FLOW_TYPE_DV);
5416 return fops->counter_query(dev, cnt, clear, pkts, bytes);
5419 "port %u counter query is not supported.",
5420 dev->data->port_id);
5424 #define MLX5_POOL_QUERY_FREQ_US 1000000
5427 * Set the periodic procedure for triggering asynchronous batch queries for all
5428 * the counter pools.
5431 * Pointer to mlx5_ibv_shared object.
5434 mlx5_set_query_alarm(struct mlx5_ibv_shared *sh)
5436 struct mlx5_pools_container *cont = MLX5_CNT_CONTAINER(sh, 0, 0);
5437 uint32_t pools_n = rte_atomic16_read(&cont->n_valid);
5440 cont = MLX5_CNT_CONTAINER(sh, 1, 0);
5441 pools_n += rte_atomic16_read(&cont->n_valid);
5442 us = MLX5_POOL_QUERY_FREQ_US / pools_n;
5443 DRV_LOG(DEBUG, "Set alarm for %u pools each %u us", pools_n, us);
5444 if (rte_eal_alarm_set(us, mlx5_flow_query_alarm, sh)) {
5445 sh->cmng.query_thread_on = 0;
5446 DRV_LOG(ERR, "Cannot reinitialize query alarm");
5448 sh->cmng.query_thread_on = 1;
5453 * The periodic procedure for triggering asynchronous batch queries for all the
5454 * counter pools. This function is probably called by the host thread.
5457 * The parameter for the alarm process.
5460 mlx5_flow_query_alarm(void *arg)
5462 struct mlx5_ibv_shared *sh = arg;
5463 struct mlx5_devx_obj *dcs;
5466 uint8_t batch = sh->cmng.batch;
5467 uint16_t pool_index = sh->cmng.pool_index;
5468 struct mlx5_pools_container *cont;
5469 struct mlx5_pools_container *mcont;
5470 struct mlx5_flow_counter_pool *pool;
5472 if (sh->cmng.pending_queries >= MLX5_MAX_PENDING_QUERIES)
5475 cont = MLX5_CNT_CONTAINER(sh, batch, 1);
5476 mcont = MLX5_CNT_CONTAINER(sh, batch, 0);
5477 /* Check if resize was done and need to flip a container. */
5478 if (cont != mcont) {
5480 /* Clean the old container. */
5481 rte_free(cont->pools);
5482 memset(cont, 0, sizeof(*cont));
5485 /* Flip the host container. */
5486 sh->cmng.mhi[batch] ^= (uint8_t)2;
5490 /* 2 empty containers case is unexpected. */
5491 if (unlikely(batch != sh->cmng.batch))
5495 goto next_container;
5497 pool = cont->pools[pool_index];
5499 /* There is a pool query in progress. */
5502 LIST_FIRST(&sh->cmng.free_stat_raws);
5504 /* No free counter statistics raw memory. */
5506 dcs = (struct mlx5_devx_obj *)(uintptr_t)rte_atomic64_read
5508 offset = batch ? 0 : dcs->id % MLX5_COUNTERS_PER_POOL;
5509 ret = mlx5_devx_cmd_flow_counter_query(dcs, 0, MLX5_COUNTERS_PER_POOL -
5511 pool->raw_hw->mem_mng->dm->id,
5513 (pool->raw_hw->data + offset),
5515 (uint64_t)(uintptr_t)pool);
5517 DRV_LOG(ERR, "Failed to trigger asynchronous query for dcs ID"
5518 " %d", pool->min_dcs->id);
5519 pool->raw_hw = NULL;
5522 pool->raw_hw->min_dcs_id = dcs->id;
5523 LIST_REMOVE(pool->raw_hw, next);
5524 sh->cmng.pending_queries++;
5526 if (pool_index >= rte_atomic16_read(&cont->n_valid)) {
5531 sh->cmng.batch = batch;
5532 sh->cmng.pool_index = pool_index;
5533 mlx5_set_query_alarm(sh);
5537 * Handler for the HW respond about ready values from an asynchronous batch
5538 * query. This function is probably called by the host thread.
5541 * The pointer to the shared IB device context.
5542 * @param[in] async_id
5543 * The Devx async ID.
5545 * The status of the completion.
5548 mlx5_flow_async_pool_query_handle(struct mlx5_ibv_shared *sh,
5549 uint64_t async_id, int status)
5551 struct mlx5_flow_counter_pool *pool =
5552 (struct mlx5_flow_counter_pool *)(uintptr_t)async_id;
5553 struct mlx5_counter_stats_raw *raw_to_free;
5555 if (unlikely(status)) {
5556 raw_to_free = pool->raw_hw;
5558 raw_to_free = pool->raw;
5559 rte_spinlock_lock(&pool->sl);
5560 pool->raw = pool->raw_hw;
5561 rte_spinlock_unlock(&pool->sl);
5562 rte_atomic64_add(&pool->query_gen, 1);
5563 /* Be sure the new raw counters data is updated in memory. */
5566 LIST_INSERT_HEAD(&sh->cmng.free_stat_raws, raw_to_free, next);
5567 pool->raw_hw = NULL;
5568 sh->cmng.pending_queries--;
5572 * Translate the rte_flow group index to HW table value.
5574 * @param[in] attributes
5575 * Pointer to flow attributes
5576 * @param[in] external
5577 * Value is part of flow rule created by request external to PMD.
5579 * rte_flow group index value.
5583 * Pointer to error structure.
5586 * 0 on success, a negative errno value otherwise and rte_errno is set.
5589 mlx5_flow_group_to_table(const struct rte_flow_attr *attributes, bool external,
5590 uint32_t group, uint32_t *table,
5591 struct rte_flow_error *error)
5593 if (attributes->transfer && external) {
5594 if (group == UINT32_MAX)
5595 return rte_flow_error_set
5597 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
5599 "group index not supported");
5608 * Discover availability of metadata reg_c's.
5610 * Iteratively use test flows to check availability.
5613 * Pointer to the Ethernet device structure.
5616 * 0 on success, a negative errno value otherwise and rte_errno is set.
5619 mlx5_flow_discover_mreg_c(struct rte_eth_dev *dev)
5621 struct mlx5_priv *priv = dev->data->dev_private;
5622 struct mlx5_dev_config *config = &priv->config;
5623 enum modify_reg idx;
5626 /* reg_c[0] and reg_c[1] are reserved. */
5627 config->flow_mreg_c[n++] = REG_C_0;
5628 config->flow_mreg_c[n++] = REG_C_1;
5629 /* Discover availability of other reg_c's. */
5630 for (idx = REG_C_2; idx <= REG_C_7; ++idx) {
5631 struct rte_flow_attr attr = {
5632 .group = MLX5_FLOW_MREG_CP_TABLE_GROUP,
5633 .priority = MLX5_FLOW_PRIO_RSVD,
5636 struct rte_flow_item items[] = {
5638 .type = RTE_FLOW_ITEM_TYPE_END,
5641 struct rte_flow_action actions[] = {
5643 .type = MLX5_RTE_FLOW_ACTION_TYPE_COPY_MREG,
5644 .conf = &(struct mlx5_flow_action_copy_mreg){
5650 .type = RTE_FLOW_ACTION_TYPE_JUMP,
5651 .conf = &(struct rte_flow_action_jump){
5652 .group = MLX5_FLOW_MREG_ACT_TABLE_GROUP,
5656 .type = RTE_FLOW_ACTION_TYPE_END,
5659 struct rte_flow *flow;
5660 struct rte_flow_error error;
5662 if (!config->dv_flow_en)
5664 /* Create internal flow, validation skips copy action. */
5665 flow = flow_list_create(dev, NULL, &attr, items,
5666 actions, false, &error);
5669 if (dev->data->dev_started || !flow_drv_apply(dev, flow, NULL))
5670 config->flow_mreg_c[n++] = idx;
5671 flow_list_destroy(dev, NULL, flow);
5673 for (; n < MLX5_MREG_C_NUM; ++n)
5674 config->flow_mreg_c[n] = REG_NONE;
5679 * Dump flow raw hw data to file
5682 * The pointer to Ethernet device.
5684 * A pointer to a file for output.
5686 * Perform verbose error reporting if not NULL. PMDs initialize this
5687 * structure in case of error only.
5689 * 0 on success, a nagative value otherwise.
5692 mlx5_flow_dev_dump(struct rte_eth_dev *dev,
5694 struct rte_flow_error *error __rte_unused)
5696 struct mlx5_priv *priv = dev->data->dev_private;
5697 struct mlx5_ibv_shared *sh = priv->sh;
5699 return mlx5_devx_cmd_flow_dump(sh->fdb_domain, sh->rx_domain,
5700 sh->tx_domain, file);