4 * Copyright 2016 6WIND S.A.
5 * Copyright 2016 Mellanox.
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
11 * * Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * * Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in
15 * the documentation and/or other materials provided with the
17 * * Neither the name of 6WIND S.A. nor the names of its
18 * contributors may be used to endorse or promote products derived
19 * from this software without specific prior written permission.
21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
24 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
25 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
26 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
27 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
31 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
34 #include <sys/queue.h>
38 /* ISO C doesn't support unnamed structs/unions, disabling -pedantic. */
40 #pragma GCC diagnostic ignored "-Wpedantic"
42 #include <infiniband/verbs.h>
44 #pragma GCC diagnostic error "-Wpedantic"
47 #include <rte_ethdev.h>
49 #include <rte_flow_driver.h>
50 #include <rte_malloc.h>
56 mlx5_flow_create_eth(const struct rte_flow_item *item,
57 const void *default_mask,
61 mlx5_flow_create_vlan(const struct rte_flow_item *item,
62 const void *default_mask,
66 mlx5_flow_create_ipv4(const struct rte_flow_item *item,
67 const void *default_mask,
71 mlx5_flow_create_ipv6(const struct rte_flow_item *item,
72 const void *default_mask,
76 mlx5_flow_create_udp(const struct rte_flow_item *item,
77 const void *default_mask,
81 mlx5_flow_create_tcp(const struct rte_flow_item *item,
82 const void *default_mask,
86 mlx5_flow_create_vxlan(const struct rte_flow_item *item,
87 const void *default_mask,
91 LIST_ENTRY(rte_flow) next; /**< Pointer to the next flow structure. */
92 struct ibv_exp_flow_attr *ibv_attr; /**< Pointer to Verbs attributes. */
93 struct ibv_exp_rwq_ind_table *ind_table; /**< Indirection table. */
94 struct ibv_qp *qp; /**< Verbs queue pair. */
95 struct ibv_exp_flow *ibv_flow; /**< Verbs flow. */
96 struct ibv_exp_wq *wq; /**< Verbs work queue. */
97 struct ibv_cq *cq; /**< Verbs completion queue. */
98 struct rxq *rxq; /**< Pointer to the queue, NULL if drop queue. */
99 uint32_t mark:1; /**< Set if the flow is marked. */
102 /** Static initializer for items. */
104 (const enum rte_flow_item_type []){ \
105 __VA_ARGS__, RTE_FLOW_ITEM_TYPE_END, \
108 /** Structure to generate a simple graph of layers supported by the NIC. */
109 struct mlx5_flow_items {
110 /** List of possible actions for these items. */
111 const enum rte_flow_action_type *const actions;
112 /** Bit-masks corresponding to the possibilities for the item. */
115 * Default bit-masks to use when item->mask is not provided. When
116 * \default_mask is also NULL, the full supported bit-mask (\mask) is
119 const void *default_mask;
120 /** Bit-masks size in bytes. */
121 const unsigned int mask_sz;
123 * Conversion function from rte_flow to NIC specific flow.
126 * rte_flow item to convert.
127 * @param default_mask
128 * Default bit-masks to use when item->mask is not provided.
130 * Internal structure to store the conversion.
133 * 0 on success, negative value otherwise.
135 int (*convert)(const struct rte_flow_item *item,
136 const void *default_mask,
138 /** Size in bytes of the destination structure. */
139 const unsigned int dst_sz;
140 /** List of possible following items. */
141 const enum rte_flow_item_type *const items;
144 /** Valid action for this PMD. */
145 static const enum rte_flow_action_type valid_actions[] = {
146 RTE_FLOW_ACTION_TYPE_DROP,
147 RTE_FLOW_ACTION_TYPE_QUEUE,
148 RTE_FLOW_ACTION_TYPE_MARK,
149 RTE_FLOW_ACTION_TYPE_END,
152 /** Graph of supported items and associated actions. */
153 static const struct mlx5_flow_items mlx5_flow_items[] = {
154 [RTE_FLOW_ITEM_TYPE_END] = {
155 .items = ITEMS(RTE_FLOW_ITEM_TYPE_ETH,
156 RTE_FLOW_ITEM_TYPE_VXLAN),
158 [RTE_FLOW_ITEM_TYPE_ETH] = {
159 .items = ITEMS(RTE_FLOW_ITEM_TYPE_VLAN,
160 RTE_FLOW_ITEM_TYPE_IPV4,
161 RTE_FLOW_ITEM_TYPE_IPV6),
162 .actions = valid_actions,
163 .mask = &(const struct rte_flow_item_eth){
164 .dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
165 .src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
167 .default_mask = &rte_flow_item_eth_mask,
168 .mask_sz = sizeof(struct rte_flow_item_eth),
169 .convert = mlx5_flow_create_eth,
170 .dst_sz = sizeof(struct ibv_exp_flow_spec_eth),
172 [RTE_FLOW_ITEM_TYPE_VLAN] = {
173 .items = ITEMS(RTE_FLOW_ITEM_TYPE_IPV4,
174 RTE_FLOW_ITEM_TYPE_IPV6),
175 .actions = valid_actions,
176 .mask = &(const struct rte_flow_item_vlan){
179 .default_mask = &rte_flow_item_vlan_mask,
180 .mask_sz = sizeof(struct rte_flow_item_vlan),
181 .convert = mlx5_flow_create_vlan,
184 [RTE_FLOW_ITEM_TYPE_IPV4] = {
185 .items = ITEMS(RTE_FLOW_ITEM_TYPE_UDP,
186 RTE_FLOW_ITEM_TYPE_TCP),
187 .actions = valid_actions,
188 .mask = &(const struct rte_flow_item_ipv4){
192 .type_of_service = -1,
196 .default_mask = &rte_flow_item_ipv4_mask,
197 .mask_sz = sizeof(struct rte_flow_item_ipv4),
198 .convert = mlx5_flow_create_ipv4,
199 .dst_sz = sizeof(struct ibv_exp_flow_spec_ipv4_ext),
201 [RTE_FLOW_ITEM_TYPE_IPV6] = {
202 .items = ITEMS(RTE_FLOW_ITEM_TYPE_UDP,
203 RTE_FLOW_ITEM_TYPE_TCP),
204 .actions = valid_actions,
205 .mask = &(const struct rte_flow_item_ipv6){
208 0xff, 0xff, 0xff, 0xff,
209 0xff, 0xff, 0xff, 0xff,
210 0xff, 0xff, 0xff, 0xff,
211 0xff, 0xff, 0xff, 0xff,
214 0xff, 0xff, 0xff, 0xff,
215 0xff, 0xff, 0xff, 0xff,
216 0xff, 0xff, 0xff, 0xff,
217 0xff, 0xff, 0xff, 0xff,
221 .default_mask = &rte_flow_item_ipv6_mask,
222 .mask_sz = sizeof(struct rte_flow_item_ipv6),
223 .convert = mlx5_flow_create_ipv6,
224 .dst_sz = sizeof(struct ibv_exp_flow_spec_ipv6),
226 [RTE_FLOW_ITEM_TYPE_UDP] = {
227 .items = ITEMS(RTE_FLOW_ITEM_TYPE_VXLAN),
228 .actions = valid_actions,
229 .mask = &(const struct rte_flow_item_udp){
235 .default_mask = &rte_flow_item_udp_mask,
236 .mask_sz = sizeof(struct rte_flow_item_udp),
237 .convert = mlx5_flow_create_udp,
238 .dst_sz = sizeof(struct ibv_exp_flow_spec_tcp_udp),
240 [RTE_FLOW_ITEM_TYPE_TCP] = {
241 .actions = valid_actions,
242 .mask = &(const struct rte_flow_item_tcp){
248 .default_mask = &rte_flow_item_tcp_mask,
249 .mask_sz = sizeof(struct rte_flow_item_tcp),
250 .convert = mlx5_flow_create_tcp,
251 .dst_sz = sizeof(struct ibv_exp_flow_spec_tcp_udp),
253 [RTE_FLOW_ITEM_TYPE_VXLAN] = {
254 .items = ITEMS(RTE_FLOW_ITEM_TYPE_ETH),
255 .actions = valid_actions,
256 .mask = &(const struct rte_flow_item_vxlan){
257 .vni = "\xff\xff\xff",
259 .default_mask = &rte_flow_item_vxlan_mask,
260 .mask_sz = sizeof(struct rte_flow_item_vxlan),
261 .convert = mlx5_flow_create_vxlan,
262 .dst_sz = sizeof(struct ibv_exp_flow_spec_tunnel),
266 /** Structure to pass to the conversion function. */
268 struct ibv_exp_flow_attr *ibv_attr; /**< Verbs attribute. */
269 unsigned int offset; /**< Offset in bytes in the ibv_attr buffer. */
270 uint32_t inner; /**< Set once VXLAN is encountered. */
273 struct mlx5_flow_action {
274 uint32_t queue:1; /**< Target is a receive queue. */
275 uint32_t drop:1; /**< Target is a drop queue. */
276 uint32_t mark:1; /**< Mark is present in the flow. */
277 uint32_t queue_id; /**< Identifier of the queue. */
278 uint32_t mark_id; /**< Mark identifier. */
282 * Check support for a given item.
285 * Item specification.
287 * Bit-masks covering supported fields to compare with spec, last and mask in
290 * Bit-Mask size in bytes.
296 mlx5_flow_item_validate(const struct rte_flow_item *item,
297 const uint8_t *mask, unsigned int size)
301 if (!item->spec && (item->mask || item->last))
303 if (item->spec && !item->mask) {
305 const uint8_t *spec = item->spec;
307 for (i = 0; i < size; ++i)
308 if ((spec[i] | mask[i]) != mask[i])
311 if (item->last && !item->mask) {
313 const uint8_t *spec = item->last;
315 for (i = 0; i < size; ++i)
316 if ((spec[i] | mask[i]) != mask[i])
321 const uint8_t *spec = item->mask;
323 for (i = 0; i < size; ++i)
324 if ((spec[i] | mask[i]) != mask[i])
327 if (item->spec && item->last) {
330 const uint8_t *apply = mask;
335 for (i = 0; i < size; ++i) {
336 spec[i] = ((const uint8_t *)item->spec)[i] & apply[i];
337 last[i] = ((const uint8_t *)item->last)[i] & apply[i];
339 ret = memcmp(spec, last, size);
345 * Validate a flow supported by the NIC.
348 * Pointer to private structure.
350 * Flow rule attributes.
352 * Pattern specification (list terminated by the END pattern item).
354 * Associated actions (list terminated by the END action).
356 * Perform verbose error reporting if not NULL.
357 * @param[in, out] flow
358 * Flow structure to update.
361 * 0 on success, a negative errno value otherwise and rte_errno is set.
364 priv_flow_validate(struct priv *priv,
365 const struct rte_flow_attr *attr,
366 const struct rte_flow_item items[],
367 const struct rte_flow_action actions[],
368 struct rte_flow_error *error,
369 struct mlx5_flow *flow)
371 const struct mlx5_flow_items *cur_item = mlx5_flow_items;
372 struct mlx5_flow_action action = {
380 rte_flow_error_set(error, ENOTSUP,
381 RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
383 "groups are not supported");
386 if (attr->priority) {
387 rte_flow_error_set(error, ENOTSUP,
388 RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
390 "priorities are not supported");
394 rte_flow_error_set(error, ENOTSUP,
395 RTE_FLOW_ERROR_TYPE_ATTR_EGRESS,
397 "egress is not supported");
400 if (!attr->ingress) {
401 rte_flow_error_set(error, ENOTSUP,
402 RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
404 "only ingress is supported");
407 for (; items->type != RTE_FLOW_ITEM_TYPE_END; ++items) {
408 const struct mlx5_flow_items *token = NULL;
412 if (items->type == RTE_FLOW_ITEM_TYPE_VOID)
416 cur_item->items[i] != RTE_FLOW_ITEM_TYPE_END;
418 if (cur_item->items[i] == items->type) {
419 token = &mlx5_flow_items[items->type];
424 goto exit_item_not_supported;
426 err = mlx5_flow_item_validate(items,
427 (const uint8_t *)cur_item->mask,
430 goto exit_item_not_supported;
431 if (flow->ibv_attr && cur_item->convert) {
432 err = cur_item->convert(items,
433 (cur_item->default_mask ?
434 cur_item->default_mask :
438 goto exit_item_not_supported;
440 flow->offset += cur_item->dst_sz;
442 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; ++actions) {
443 if (actions->type == RTE_FLOW_ACTION_TYPE_VOID) {
445 } else if (actions->type == RTE_FLOW_ACTION_TYPE_DROP) {
447 } else if (actions->type == RTE_FLOW_ACTION_TYPE_QUEUE) {
448 const struct rte_flow_action_queue *queue =
449 (const struct rte_flow_action_queue *)
452 if (!queue || (queue->index > (priv->rxqs_n - 1)))
453 goto exit_action_not_supported;
455 } else if (actions->type == RTE_FLOW_ACTION_TYPE_MARK) {
456 const struct rte_flow_action_mark *mark =
457 (const struct rte_flow_action_mark *)
460 if (mark && (mark->id >= MLX5_FLOW_MARK_MAX)) {
461 rte_flow_error_set(error, ENOTSUP,
462 RTE_FLOW_ERROR_TYPE_ACTION,
464 "mark must be between 0"
470 goto exit_action_not_supported;
473 if (action.mark && !flow->ibv_attr && !action.drop)
474 flow->offset += sizeof(struct ibv_exp_flow_spec_action_tag);
475 if (!action.queue && !action.drop) {
476 rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_HANDLE,
477 NULL, "no valid action");
481 exit_item_not_supported:
482 rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ITEM,
483 items, "item not supported");
485 exit_action_not_supported:
486 rte_flow_error_set(error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
487 actions, "action not supported");
492 * Validate a flow supported by the NIC.
494 * @see rte_flow_validate()
498 mlx5_flow_validate(struct rte_eth_dev *dev,
499 const struct rte_flow_attr *attr,
500 const struct rte_flow_item items[],
501 const struct rte_flow_action actions[],
502 struct rte_flow_error *error)
504 struct priv *priv = dev->data->dev_private;
506 struct mlx5_flow flow = { .offset = sizeof(struct ibv_exp_flow_attr) };
509 ret = priv_flow_validate(priv, attr, items, actions, error, &flow);
515 * Convert Ethernet item to Verbs specification.
518 * Item specification.
519 * @param default_mask[in]
520 * Default bit-masks to use when item->mask is not provided.
521 * @param data[in, out]
525 mlx5_flow_create_eth(const struct rte_flow_item *item,
526 const void *default_mask,
529 const struct rte_flow_item_eth *spec = item->spec;
530 const struct rte_flow_item_eth *mask = item->mask;
531 struct mlx5_flow *flow = (struct mlx5_flow *)data;
532 struct ibv_exp_flow_spec_eth *eth;
533 const unsigned int eth_size = sizeof(struct ibv_exp_flow_spec_eth);
536 ++flow->ibv_attr->num_of_specs;
537 flow->ibv_attr->priority = 2;
538 eth = (void *)((uintptr_t)flow->ibv_attr + flow->offset);
539 *eth = (struct ibv_exp_flow_spec_eth) {
540 .type = flow->inner | IBV_EXP_FLOW_SPEC_ETH,
547 memcpy(eth->val.dst_mac, spec->dst.addr_bytes, ETHER_ADDR_LEN);
548 memcpy(eth->val.src_mac, spec->src.addr_bytes, ETHER_ADDR_LEN);
549 memcpy(eth->mask.dst_mac, mask->dst.addr_bytes, ETHER_ADDR_LEN);
550 memcpy(eth->mask.src_mac, mask->src.addr_bytes, ETHER_ADDR_LEN);
551 /* Remove unwanted bits from values. */
552 for (i = 0; i < ETHER_ADDR_LEN; ++i) {
553 eth->val.dst_mac[i] &= eth->mask.dst_mac[i];
554 eth->val.src_mac[i] &= eth->mask.src_mac[i];
560 * Convert VLAN item to Verbs specification.
563 * Item specification.
564 * @param default_mask[in]
565 * Default bit-masks to use when item->mask is not provided.
566 * @param data[in, out]
570 mlx5_flow_create_vlan(const struct rte_flow_item *item,
571 const void *default_mask,
574 const struct rte_flow_item_vlan *spec = item->spec;
575 const struct rte_flow_item_vlan *mask = item->mask;
576 struct mlx5_flow *flow = (struct mlx5_flow *)data;
577 struct ibv_exp_flow_spec_eth *eth;
578 const unsigned int eth_size = sizeof(struct ibv_exp_flow_spec_eth);
580 eth = (void *)((uintptr_t)flow->ibv_attr + flow->offset - eth_size);
585 eth->val.vlan_tag = spec->tci;
586 eth->mask.vlan_tag = mask->tci;
587 eth->val.vlan_tag &= eth->mask.vlan_tag;
592 * Convert IPv4 item to Verbs specification.
595 * Item specification.
596 * @param default_mask[in]
597 * Default bit-masks to use when item->mask is not provided.
598 * @param data[in, out]
602 mlx5_flow_create_ipv4(const struct rte_flow_item *item,
603 const void *default_mask,
606 const struct rte_flow_item_ipv4 *spec = item->spec;
607 const struct rte_flow_item_ipv4 *mask = item->mask;
608 struct mlx5_flow *flow = (struct mlx5_flow *)data;
609 struct ibv_exp_flow_spec_ipv4_ext *ipv4;
610 unsigned int ipv4_size = sizeof(struct ibv_exp_flow_spec_ipv4_ext);
612 ++flow->ibv_attr->num_of_specs;
613 flow->ibv_attr->priority = 1;
614 ipv4 = (void *)((uintptr_t)flow->ibv_attr + flow->offset);
615 *ipv4 = (struct ibv_exp_flow_spec_ipv4_ext) {
616 .type = flow->inner | IBV_EXP_FLOW_SPEC_IPV4_EXT,
623 ipv4->val = (struct ibv_exp_flow_ipv4_ext_filter){
624 .src_ip = spec->hdr.src_addr,
625 .dst_ip = spec->hdr.dst_addr,
626 .proto = spec->hdr.next_proto_id,
627 .tos = spec->hdr.type_of_service,
629 ipv4->mask = (struct ibv_exp_flow_ipv4_ext_filter){
630 .src_ip = mask->hdr.src_addr,
631 .dst_ip = mask->hdr.dst_addr,
632 .proto = mask->hdr.next_proto_id,
633 .tos = mask->hdr.type_of_service,
635 /* Remove unwanted bits from values. */
636 ipv4->val.src_ip &= ipv4->mask.src_ip;
637 ipv4->val.dst_ip &= ipv4->mask.dst_ip;
638 ipv4->val.proto &= ipv4->mask.proto;
639 ipv4->val.tos &= ipv4->mask.tos;
644 * Convert IPv6 item to Verbs specification.
647 * Item specification.
648 * @param default_mask[in]
649 * Default bit-masks to use when item->mask is not provided.
650 * @param data[in, out]
654 mlx5_flow_create_ipv6(const struct rte_flow_item *item,
655 const void *default_mask,
658 const struct rte_flow_item_ipv6 *spec = item->spec;
659 const struct rte_flow_item_ipv6 *mask = item->mask;
660 struct mlx5_flow *flow = (struct mlx5_flow *)data;
661 struct ibv_exp_flow_spec_ipv6 *ipv6;
662 unsigned int ipv6_size = sizeof(struct ibv_exp_flow_spec_ipv6);
665 ++flow->ibv_attr->num_of_specs;
666 flow->ibv_attr->priority = 1;
667 ipv6 = (void *)((uintptr_t)flow->ibv_attr + flow->offset);
668 *ipv6 = (struct ibv_exp_flow_spec_ipv6) {
669 .type = flow->inner | IBV_EXP_FLOW_SPEC_IPV6,
676 memcpy(ipv6->val.src_ip, spec->hdr.src_addr,
677 RTE_DIM(ipv6->val.src_ip));
678 memcpy(ipv6->val.dst_ip, spec->hdr.dst_addr,
679 RTE_DIM(ipv6->val.dst_ip));
680 memcpy(ipv6->mask.src_ip, mask->hdr.src_addr,
681 RTE_DIM(ipv6->mask.src_ip));
682 memcpy(ipv6->mask.dst_ip, mask->hdr.dst_addr,
683 RTE_DIM(ipv6->mask.dst_ip));
684 /* Remove unwanted bits from values. */
685 for (i = 0; i < RTE_DIM(ipv6->val.src_ip); ++i) {
686 ipv6->val.src_ip[i] &= ipv6->mask.src_ip[i];
687 ipv6->val.dst_ip[i] &= ipv6->mask.dst_ip[i];
693 * Convert UDP item to Verbs specification.
696 * Item specification.
697 * @param default_mask[in]
698 * Default bit-masks to use when item->mask is not provided.
699 * @param data[in, out]
703 mlx5_flow_create_udp(const struct rte_flow_item *item,
704 const void *default_mask,
707 const struct rte_flow_item_udp *spec = item->spec;
708 const struct rte_flow_item_udp *mask = item->mask;
709 struct mlx5_flow *flow = (struct mlx5_flow *)data;
710 struct ibv_exp_flow_spec_tcp_udp *udp;
711 unsigned int udp_size = sizeof(struct ibv_exp_flow_spec_tcp_udp);
713 ++flow->ibv_attr->num_of_specs;
714 flow->ibv_attr->priority = 0;
715 udp = (void *)((uintptr_t)flow->ibv_attr + flow->offset);
716 *udp = (struct ibv_exp_flow_spec_tcp_udp) {
717 .type = flow->inner | IBV_EXP_FLOW_SPEC_UDP,
724 udp->val.dst_port = spec->hdr.dst_port;
725 udp->val.src_port = spec->hdr.src_port;
726 udp->mask.dst_port = mask->hdr.dst_port;
727 udp->mask.src_port = mask->hdr.src_port;
728 /* Remove unwanted bits from values. */
729 udp->val.src_port &= udp->mask.src_port;
730 udp->val.dst_port &= udp->mask.dst_port;
735 * Convert TCP item to Verbs specification.
738 * Item specification.
739 * @param default_mask[in]
740 * Default bit-masks to use when item->mask is not provided.
741 * @param data[in, out]
745 mlx5_flow_create_tcp(const struct rte_flow_item *item,
746 const void *default_mask,
749 const struct rte_flow_item_tcp *spec = item->spec;
750 const struct rte_flow_item_tcp *mask = item->mask;
751 struct mlx5_flow *flow = (struct mlx5_flow *)data;
752 struct ibv_exp_flow_spec_tcp_udp *tcp;
753 unsigned int tcp_size = sizeof(struct ibv_exp_flow_spec_tcp_udp);
755 ++flow->ibv_attr->num_of_specs;
756 flow->ibv_attr->priority = 0;
757 tcp = (void *)((uintptr_t)flow->ibv_attr + flow->offset);
758 *tcp = (struct ibv_exp_flow_spec_tcp_udp) {
759 .type = flow->inner | IBV_EXP_FLOW_SPEC_TCP,
766 tcp->val.dst_port = spec->hdr.dst_port;
767 tcp->val.src_port = spec->hdr.src_port;
768 tcp->mask.dst_port = mask->hdr.dst_port;
769 tcp->mask.src_port = mask->hdr.src_port;
770 /* Remove unwanted bits from values. */
771 tcp->val.src_port &= tcp->mask.src_port;
772 tcp->val.dst_port &= tcp->mask.dst_port;
777 * Convert VXLAN item to Verbs specification.
780 * Item specification.
781 * @param default_mask[in]
782 * Default bit-masks to use when item->mask is not provided.
783 * @param data[in, out]
787 mlx5_flow_create_vxlan(const struct rte_flow_item *item,
788 const void *default_mask,
791 const struct rte_flow_item_vxlan *spec = item->spec;
792 const struct rte_flow_item_vxlan *mask = item->mask;
793 struct mlx5_flow *flow = (struct mlx5_flow *)data;
794 struct ibv_exp_flow_spec_tunnel *vxlan;
795 unsigned int size = sizeof(struct ibv_exp_flow_spec_tunnel);
801 ++flow->ibv_attr->num_of_specs;
802 flow->ibv_attr->priority = 0;
804 vxlan = (void *)((uintptr_t)flow->ibv_attr + flow->offset);
805 *vxlan = (struct ibv_exp_flow_spec_tunnel) {
806 .type = flow->inner | IBV_EXP_FLOW_SPEC_VXLAN_TUNNEL,
809 flow->inner = IBV_EXP_FLOW_SPEC_INNER;
814 memcpy(&id.vni[1], spec->vni, 3);
815 vxlan->val.tunnel_id = id.vlan_id;
816 memcpy(&id.vni[1], mask->vni, 3);
817 vxlan->mask.tunnel_id = id.vlan_id;
818 /* Remove unwanted bits from values. */
819 vxlan->val.tunnel_id &= vxlan->mask.tunnel_id;
824 * Convert mark/flag action to Verbs specification.
827 * Pointer to MLX5 flow structure.
832 mlx5_flow_create_flag_mark(struct mlx5_flow *flow, uint32_t mark_id)
834 struct ibv_exp_flow_spec_action_tag *tag;
835 unsigned int size = sizeof(struct ibv_exp_flow_spec_action_tag);
837 tag = (void *)((uintptr_t)flow->ibv_attr + flow->offset);
838 *tag = (struct ibv_exp_flow_spec_action_tag){
839 .type = IBV_EXP_FLOW_SPEC_ACTION_TAG,
841 .tag_id = mlx5_flow_mark_set(mark_id),
843 ++flow->ibv_attr->num_of_specs;
848 * Complete flow rule creation.
851 * Pointer to private structure.
853 * Verbs flow attributes.
855 * Target action structure.
857 * Perform verbose error reporting if not NULL.
860 * A flow if the rule could be created.
862 static struct rte_flow *
863 priv_flow_create_action_queue(struct priv *priv,
864 struct ibv_exp_flow_attr *ibv_attr,
865 struct mlx5_flow_action *action,
866 struct rte_flow_error *error)
868 struct rxq_ctrl *rxq;
869 struct rte_flow *rte_flow;
873 rte_flow = rte_calloc(__func__, 1, sizeof(*rte_flow), 0);
875 rte_flow_error_set(error, ENOMEM, RTE_FLOW_ERROR_TYPE_HANDLE,
876 NULL, "cannot allocate flow memory");
881 ibv_exp_create_cq(priv->ctx, 1, NULL, NULL, 0,
882 &(struct ibv_exp_cq_init_attr){
886 rte_flow_error_set(error, ENOMEM,
887 RTE_FLOW_ERROR_TYPE_HANDLE,
888 NULL, "cannot allocate CQ");
891 rte_flow->wq = ibv_exp_create_wq(priv->ctx,
892 &(struct ibv_exp_wq_init_attr){
893 .wq_type = IBV_EXP_WQT_RQ,
900 rxq = container_of((*priv->rxqs)[action->queue_id],
901 struct rxq_ctrl, rxq);
902 rte_flow->rxq = &rxq->rxq;
903 rxq->rxq.mark |= action->mark;
904 rte_flow->wq = rxq->wq;
906 rte_flow->mark = action->mark;
907 rte_flow->ibv_attr = ibv_attr;
908 rte_flow->ind_table = ibv_exp_create_rwq_ind_table(
910 &(struct ibv_exp_rwq_ind_table_init_attr){
912 .log_ind_tbl_size = 0,
913 .ind_tbl = &rte_flow->wq,
916 if (!rte_flow->ind_table) {
917 rte_flow_error_set(error, ENOMEM, RTE_FLOW_ERROR_TYPE_HANDLE,
918 NULL, "cannot allocate indirection table");
921 rte_flow->qp = ibv_exp_create_qp(
923 &(struct ibv_exp_qp_init_attr){
924 .qp_type = IBV_QPT_RAW_PACKET,
926 IBV_EXP_QP_INIT_ATTR_PD |
927 IBV_EXP_QP_INIT_ATTR_PORT |
928 IBV_EXP_QP_INIT_ATTR_RX_HASH,
930 .rx_hash_conf = &(struct ibv_exp_rx_hash_conf){
932 IBV_EXP_RX_HASH_FUNC_TOEPLITZ,
933 .rx_hash_key_len = rss_hash_default_key_len,
934 .rx_hash_key = rss_hash_default_key,
935 .rx_hash_fields_mask = 0,
936 .rwq_ind_tbl = rte_flow->ind_table,
938 .port_num = priv->port,
941 rte_flow_error_set(error, ENOMEM, RTE_FLOW_ERROR_TYPE_HANDLE,
942 NULL, "cannot allocate QP");
945 rte_flow->ibv_flow = ibv_exp_create_flow(rte_flow->qp,
947 if (!rte_flow->ibv_flow) {
948 rte_flow_error_set(error, ENOMEM, RTE_FLOW_ERROR_TYPE_HANDLE,
949 NULL, "flow rule creation failure");
956 ibv_destroy_qp(rte_flow->qp);
957 if (rte_flow->ind_table)
958 ibv_exp_destroy_rwq_ind_table(rte_flow->ind_table);
959 if (!rte_flow->rxq && rte_flow->wq)
960 ibv_exp_destroy_wq(rte_flow->wq);
961 if (!rte_flow->rxq && rte_flow->cq)
962 ibv_destroy_cq(rte_flow->cq);
963 rte_free(rte_flow->ibv_attr);
972 * Pointer to private structure.
974 * Flow rule attributes.
976 * Pattern specification (list terminated by the END pattern item).
978 * Associated actions (list terminated by the END action).
980 * Perform verbose error reporting if not NULL.
983 * A flow on success, NULL otherwise.
985 static struct rte_flow *
986 priv_flow_create(struct priv *priv,
987 const struct rte_flow_attr *attr,
988 const struct rte_flow_item items[],
989 const struct rte_flow_action actions[],
990 struct rte_flow_error *error)
992 struct rte_flow *rte_flow;
993 struct mlx5_flow_action action;
994 struct mlx5_flow flow = { .offset = sizeof(struct ibv_exp_flow_attr), };
997 err = priv_flow_validate(priv, attr, items, actions, error, &flow);
1000 flow.ibv_attr = rte_malloc(__func__, flow.offset, 0);
1001 flow.offset = sizeof(struct ibv_exp_flow_attr);
1002 if (!flow.ibv_attr) {
1003 rte_flow_error_set(error, ENOMEM, RTE_FLOW_ERROR_TYPE_HANDLE,
1004 NULL, "cannot allocate ibv_attr memory");
1007 *flow.ibv_attr = (struct ibv_exp_flow_attr){
1008 .type = IBV_EXP_FLOW_ATTR_NORMAL,
1009 .size = sizeof(struct ibv_exp_flow_attr),
1010 .priority = attr->priority,
1017 claim_zero(priv_flow_validate(priv, attr, items, actions,
1019 action = (struct mlx5_flow_action){
1023 .mark_id = MLX5_FLOW_MARK_DEFAULT,
1025 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; ++actions) {
1026 if (actions->type == RTE_FLOW_ACTION_TYPE_VOID) {
1028 } else if (actions->type == RTE_FLOW_ACTION_TYPE_QUEUE) {
1031 ((const struct rte_flow_action_queue *)
1032 actions->conf)->index;
1033 } else if (actions->type == RTE_FLOW_ACTION_TYPE_DROP) {
1036 } else if (actions->type == RTE_FLOW_ACTION_TYPE_MARK) {
1037 const struct rte_flow_action_mark *mark =
1038 (const struct rte_flow_action_mark *)
1042 action.mark_id = mark->id;
1043 action.mark = !action.drop;
1045 rte_flow_error_set(error, ENOTSUP,
1046 RTE_FLOW_ERROR_TYPE_ACTION,
1047 actions, "unsupported action");
1052 mlx5_flow_create_flag_mark(&flow, action.mark_id);
1053 flow.offset += sizeof(struct ibv_exp_flow_spec_action_tag);
1055 rte_flow = priv_flow_create_action_queue(priv, flow.ibv_attr,
1059 rte_free(flow.ibv_attr);
1066 * @see rte_flow_create()
1070 mlx5_flow_create(struct rte_eth_dev *dev,
1071 const struct rte_flow_attr *attr,
1072 const struct rte_flow_item items[],
1073 const struct rte_flow_action actions[],
1074 struct rte_flow_error *error)
1076 struct priv *priv = dev->data->dev_private;
1077 struct rte_flow *flow;
1080 flow = priv_flow_create(priv, attr, items, actions, error);
1082 LIST_INSERT_HEAD(&priv->flows, flow, next);
1083 DEBUG("Flow created %p", (void *)flow);
1093 * Pointer to private structure.
1098 priv_flow_destroy(struct priv *priv,
1099 struct rte_flow *flow)
1102 LIST_REMOVE(flow, next);
1104 claim_zero(ibv_exp_destroy_flow(flow->ibv_flow));
1106 claim_zero(ibv_destroy_qp(flow->qp));
1107 if (flow->ind_table)
1108 claim_zero(ibv_exp_destroy_rwq_ind_table(flow->ind_table));
1109 if (!flow->rxq && flow->wq)
1110 claim_zero(ibv_exp_destroy_wq(flow->wq));
1111 if (!flow->rxq && flow->cq)
1112 claim_zero(ibv_destroy_cq(flow->cq));
1114 struct rte_flow *tmp;
1115 uint32_t mark_n = 0;
1117 for (tmp = LIST_FIRST(&priv->flows);
1119 tmp = LIST_NEXT(tmp, next)) {
1120 if ((flow->rxq == tmp->rxq) && tmp->mark)
1123 flow->rxq->mark = !!mark_n;
1125 rte_free(flow->ibv_attr);
1126 DEBUG("Flow destroyed %p", (void *)flow);
1133 * @see rte_flow_destroy()
1137 mlx5_flow_destroy(struct rte_eth_dev *dev,
1138 struct rte_flow *flow,
1139 struct rte_flow_error *error)
1141 struct priv *priv = dev->data->dev_private;
1145 priv_flow_destroy(priv, flow);
1151 * Destroy all flows.
1154 * Pointer to private structure.
1157 priv_flow_flush(struct priv *priv)
1159 while (!LIST_EMPTY(&priv->flows)) {
1160 struct rte_flow *flow;
1162 flow = LIST_FIRST(&priv->flows);
1163 priv_flow_destroy(priv, flow);
1168 * Destroy all flows.
1170 * @see rte_flow_flush()
1174 mlx5_flow_flush(struct rte_eth_dev *dev,
1175 struct rte_flow_error *error)
1177 struct priv *priv = dev->data->dev_private;
1181 priv_flow_flush(priv);
1189 * Called by dev_stop() to remove all flows.
1192 * Pointer to private structure.
1195 priv_flow_stop(struct priv *priv)
1197 struct rte_flow *flow;
1199 for (flow = LIST_FIRST(&priv->flows);
1201 flow = LIST_NEXT(flow, next)) {
1202 claim_zero(ibv_exp_destroy_flow(flow->ibv_flow));
1203 flow->ibv_flow = NULL;
1205 flow->rxq->mark = 0;
1206 DEBUG("Flow %p removed", (void *)flow);
1214 * Pointer to private structure.
1217 * 0 on success, a errno value otherwise and rte_errno is set.
1220 priv_flow_start(struct priv *priv)
1222 struct rte_flow *flow;
1224 for (flow = LIST_FIRST(&priv->flows);
1226 flow = LIST_NEXT(flow, next)) {
1227 flow->ibv_flow = ibv_exp_create_flow(flow->qp,
1229 if (!flow->ibv_flow) {
1230 DEBUG("Flow %p cannot be applied", (void *)flow);
1234 DEBUG("Flow %p applied", (void *)flow);
1236 flow->rxq->mark |= flow->mark;