1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright 2018 Mellanox Technologies, Ltd
5 #include <netinet/in.h>
12 /* ISO C doesn't support unnamed structs/unions, disabling -pedantic. */
14 #pragma GCC diagnostic ignored "-Wpedantic"
16 #include <infiniband/verbs.h>
18 #pragma GCC diagnostic error "-Wpedantic"
21 #include <rte_common.h>
22 #include <rte_ether.h>
23 #include <rte_ethdev_driver.h>
25 #include <rte_flow_driver.h>
26 #include <rte_malloc.h>
30 #include "mlx5_defs.h"
31 #include "mlx5_flow.h"
32 #include "mlx5_glue.h"
34 #include "mlx5_rxtx.h"
36 #define VERBS_SPEC_INNER(item_flags) \
37 (!!((item_flags) & MLX5_FLOW_LAYER_TUNNEL) ? IBV_FLOW_SPEC_INNER : 0)
40 * Create Verbs flow counter with Verbs library.
43 * Pointer to the Ethernet device structure.
44 * @param[in, out] counter
45 * mlx5 flow counter object, contains the counter id,
46 * handle of created Verbs flow counter is returned
47 * in cs field (if counters are supported).
50 * 0 On success else a negative errno value is returned
51 * and rte_errno is set.
54 flow_verbs_counter_create(struct rte_eth_dev *dev,
55 struct mlx5_flow_counter *counter)
57 #if defined(HAVE_IBV_DEVICE_COUNTERS_SET_V42)
58 struct mlx5_priv *priv = dev->data->dev_private;
59 struct ibv_context *ctx = priv->sh->ctx;
60 struct ibv_counter_set_init_attr init = {
61 .counter_set_id = counter->id};
63 counter->cs = mlx5_glue->create_counter_set(ctx, &init);
69 #elif defined(HAVE_IBV_DEVICE_COUNTERS_SET_V45)
70 struct mlx5_priv *priv = dev->data->dev_private;
71 struct ibv_context *ctx = priv->sh->ctx;
72 struct ibv_counters_init_attr init = {0};
73 struct ibv_counter_attach_attr attach;
76 memset(&attach, 0, sizeof(attach));
77 counter->cs = mlx5_glue->create_counters(ctx, &init);
82 attach.counter_desc = IBV_COUNTER_PACKETS;
84 ret = mlx5_glue->attach_counters(counter->cs, &attach, NULL);
86 attach.counter_desc = IBV_COUNTER_BYTES;
88 ret = mlx5_glue->attach_counters
89 (counter->cs, &attach, NULL);
92 claim_zero(mlx5_glue->destroy_counters(counter->cs));
107 * Get a flow counter.
110 * Pointer to the Ethernet device structure.
112 * Indicate if this counter is shared with other flows.
114 * Counter identifier.
117 * A pointer to the counter, NULL otherwise and rte_errno is set.
119 static struct mlx5_flow_counter *
120 flow_verbs_counter_new(struct rte_eth_dev *dev, uint32_t shared, uint32_t id)
122 struct mlx5_priv *priv = dev->data->dev_private;
123 struct mlx5_flow_counter *cnt;
127 TAILQ_FOREACH(cnt, &priv->sh->cmng.flow_counters, next) {
128 if (cnt->shared && cnt->id == id) {
134 cnt = rte_calloc(__func__, 1, sizeof(*cnt), 0);
140 cnt->shared = shared;
144 /* Create counter with Verbs. */
145 ret = flow_verbs_counter_create(dev, cnt);
147 TAILQ_INSERT_HEAD(&priv->sh->cmng.flow_counters, cnt, next);
150 /* Some error occurred in Verbs library. */
157 * Release a flow counter.
160 * Pointer to the Ethernet device structure.
162 * Pointer to the counter handler.
165 flow_verbs_counter_release(struct rte_eth_dev *dev,
166 struct mlx5_flow_counter *counter)
168 struct mlx5_priv *priv = dev->data->dev_private;
170 if (--counter->ref_cnt == 0) {
171 #if defined(HAVE_IBV_DEVICE_COUNTERS_SET_V42)
172 claim_zero(mlx5_glue->destroy_counter_set(counter->cs));
173 #elif defined(HAVE_IBV_DEVICE_COUNTERS_SET_V45)
174 claim_zero(mlx5_glue->destroy_counters(counter->cs));
176 TAILQ_REMOVE(&priv->sh->cmng.flow_counters, counter, next);
182 * Query a flow counter via Verbs library call.
184 * @see rte_flow_query()
188 flow_verbs_counter_query(struct rte_eth_dev *dev __rte_unused,
189 struct rte_flow *flow, void *data,
190 struct rte_flow_error *error)
192 #if defined(HAVE_IBV_DEVICE_COUNTERS_SET_V42) || \
193 defined(HAVE_IBV_DEVICE_COUNTERS_SET_V45)
194 if (flow->actions & MLX5_FLOW_ACTION_COUNT) {
195 struct rte_flow_query_count *qc = data;
196 uint64_t counters[2] = {0, 0};
197 #if defined(HAVE_IBV_DEVICE_COUNTERS_SET_V42)
198 struct ibv_query_counter_set_attr query_cs_attr = {
199 .cs = flow->counter->cs,
200 .query_flags = IBV_COUNTER_SET_FORCE_UPDATE,
202 struct ibv_counter_set_data query_out = {
204 .outlen = 2 * sizeof(uint64_t),
206 int err = mlx5_glue->query_counter_set(&query_cs_attr,
208 #elif defined(HAVE_IBV_DEVICE_COUNTERS_SET_V45)
209 int err = mlx5_glue->query_counters
210 (flow->counter->cs, counters,
212 IBV_READ_COUNTERS_ATTR_PREFER_CACHED);
215 return rte_flow_error_set
217 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
219 "cannot read counter");
222 qc->hits = counters[0] - flow->counter->hits;
223 qc->bytes = counters[1] - flow->counter->bytes;
225 flow->counter->hits = counters[0];
226 flow->counter->bytes = counters[1];
230 return rte_flow_error_set(error, EINVAL,
231 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
233 "flow does not have counter");
237 return rte_flow_error_set(error, ENOTSUP,
238 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
240 "counters are not available");
245 * Add a verbs item specification into @p verbs.
248 * Pointer to verbs structure.
250 * Create specification.
252 * Size in bytes of the specification to copy.
255 flow_verbs_spec_add(struct mlx5_flow_verbs *verbs, void *src, unsigned int size)
261 assert(verbs->specs);
262 dst = (void *)(verbs->specs + verbs->size);
263 memcpy(dst, src, size);
264 ++verbs->attr->num_of_specs;
269 * Convert the @p item into a Verbs specification. This function assumes that
270 * the input is valid and that there is space to insert the requested item
273 * @param[in, out] dev_flow
274 * Pointer to dev_flow structure.
276 * Item specification.
277 * @param[in] item_flags
281 flow_verbs_translate_item_eth(struct mlx5_flow *dev_flow,
282 const struct rte_flow_item *item,
285 const struct rte_flow_item_eth *spec = item->spec;
286 const struct rte_flow_item_eth *mask = item->mask;
287 const unsigned int size = sizeof(struct ibv_flow_spec_eth);
288 struct ibv_flow_spec_eth eth = {
289 .type = IBV_FLOW_SPEC_ETH | VERBS_SPEC_INNER(item_flags),
294 mask = &rte_flow_item_eth_mask;
298 memcpy(ð.val.dst_mac, spec->dst.addr_bytes,
300 memcpy(ð.val.src_mac, spec->src.addr_bytes,
302 eth.val.ether_type = spec->type;
303 memcpy(ð.mask.dst_mac, mask->dst.addr_bytes,
305 memcpy(ð.mask.src_mac, mask->src.addr_bytes,
307 eth.mask.ether_type = mask->type;
308 /* Remove unwanted bits from values. */
309 for (i = 0; i < RTE_ETHER_ADDR_LEN; ++i) {
310 eth.val.dst_mac[i] &= eth.mask.dst_mac[i];
311 eth.val.src_mac[i] &= eth.mask.src_mac[i];
313 eth.val.ether_type &= eth.mask.ether_type;
315 flow_verbs_spec_add(&dev_flow->verbs, ð, size);
319 * Update the VLAN tag in the Verbs Ethernet specification.
320 * This function assumes that the input is valid and there is space to add
321 * the requested item.
323 * @param[in, out] attr
324 * Pointer to Verbs attributes structure.
326 * Verbs structure containing the VLAN information to copy.
329 flow_verbs_item_vlan_update(struct ibv_flow_attr *attr,
330 struct ibv_flow_spec_eth *eth)
333 const enum ibv_flow_spec_type search = eth->type;
334 struct ibv_spec_header *hdr = (struct ibv_spec_header *)
335 ((uint8_t *)attr + sizeof(struct ibv_flow_attr));
337 for (i = 0; i != attr->num_of_specs; ++i) {
338 if (hdr->type == search) {
339 struct ibv_flow_spec_eth *e =
340 (struct ibv_flow_spec_eth *)hdr;
342 e->val.vlan_tag = eth->val.vlan_tag;
343 e->mask.vlan_tag = eth->mask.vlan_tag;
344 e->val.ether_type = eth->val.ether_type;
345 e->mask.ether_type = eth->mask.ether_type;
348 hdr = (struct ibv_spec_header *)((uint8_t *)hdr + hdr->size);
353 * Convert the @p item into a Verbs specification. This function assumes that
354 * the input is valid and that there is space to insert the requested item
357 * @param[in, out] dev_flow
358 * Pointer to dev_flow structure.
360 * Item specification.
361 * @param[in] item_flags
365 flow_verbs_translate_item_vlan(struct mlx5_flow *dev_flow,
366 const struct rte_flow_item *item,
369 const struct rte_flow_item_vlan *spec = item->spec;
370 const struct rte_flow_item_vlan *mask = item->mask;
371 unsigned int size = sizeof(struct ibv_flow_spec_eth);
372 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
373 struct ibv_flow_spec_eth eth = {
374 .type = IBV_FLOW_SPEC_ETH | VERBS_SPEC_INNER(item_flags),
377 const uint32_t l2m = tunnel ? MLX5_FLOW_LAYER_INNER_L2 :
378 MLX5_FLOW_LAYER_OUTER_L2;
381 mask = &rte_flow_item_vlan_mask;
383 eth.val.vlan_tag = spec->tci;
384 eth.mask.vlan_tag = mask->tci;
385 eth.val.vlan_tag &= eth.mask.vlan_tag;
386 eth.val.ether_type = spec->inner_type;
387 eth.mask.ether_type = mask->inner_type;
388 eth.val.ether_type &= eth.mask.ether_type;
390 if (!(item_flags & l2m))
391 flow_verbs_spec_add(&dev_flow->verbs, ð, size);
393 flow_verbs_item_vlan_update(dev_flow->verbs.attr, ð);
395 dev_flow->verbs.vf_vlan.tag =
396 rte_be_to_cpu_16(spec->tci) & 0x0fff;
400 * Convert the @p item into a Verbs specification. This function assumes that
401 * the input is valid and that there is space to insert the requested item
404 * @param[in, out] dev_flow
405 * Pointer to dev_flow structure.
407 * Item specification.
408 * @param[in] item_flags
412 flow_verbs_translate_item_ipv4(struct mlx5_flow *dev_flow,
413 const struct rte_flow_item *item,
416 const struct rte_flow_item_ipv4 *spec = item->spec;
417 const struct rte_flow_item_ipv4 *mask = item->mask;
418 unsigned int size = sizeof(struct ibv_flow_spec_ipv4_ext);
419 struct ibv_flow_spec_ipv4_ext ipv4 = {
420 .type = IBV_FLOW_SPEC_IPV4_EXT | VERBS_SPEC_INNER(item_flags),
425 mask = &rte_flow_item_ipv4_mask;
427 ipv4.val = (struct ibv_flow_ipv4_ext_filter){
428 .src_ip = spec->hdr.src_addr,
429 .dst_ip = spec->hdr.dst_addr,
430 .proto = spec->hdr.next_proto_id,
431 .tos = spec->hdr.type_of_service,
433 ipv4.mask = (struct ibv_flow_ipv4_ext_filter){
434 .src_ip = mask->hdr.src_addr,
435 .dst_ip = mask->hdr.dst_addr,
436 .proto = mask->hdr.next_proto_id,
437 .tos = mask->hdr.type_of_service,
439 /* Remove unwanted bits from values. */
440 ipv4.val.src_ip &= ipv4.mask.src_ip;
441 ipv4.val.dst_ip &= ipv4.mask.dst_ip;
442 ipv4.val.proto &= ipv4.mask.proto;
443 ipv4.val.tos &= ipv4.mask.tos;
445 flow_verbs_spec_add(&dev_flow->verbs, &ipv4, size);
449 * Convert the @p item into a Verbs specification. This function assumes that
450 * the input is valid and that there is space to insert the requested item
453 * @param[in, out] dev_flow
454 * Pointer to dev_flow structure.
456 * Item specification.
457 * @param[in] item_flags
461 flow_verbs_translate_item_ipv6(struct mlx5_flow *dev_flow,
462 const struct rte_flow_item *item,
465 const struct rte_flow_item_ipv6 *spec = item->spec;
466 const struct rte_flow_item_ipv6 *mask = item->mask;
467 unsigned int size = sizeof(struct ibv_flow_spec_ipv6);
468 struct ibv_flow_spec_ipv6 ipv6 = {
469 .type = IBV_FLOW_SPEC_IPV6 | VERBS_SPEC_INNER(item_flags),
474 mask = &rte_flow_item_ipv6_mask;
477 uint32_t vtc_flow_val;
478 uint32_t vtc_flow_mask;
480 memcpy(&ipv6.val.src_ip, spec->hdr.src_addr,
481 RTE_DIM(ipv6.val.src_ip));
482 memcpy(&ipv6.val.dst_ip, spec->hdr.dst_addr,
483 RTE_DIM(ipv6.val.dst_ip));
484 memcpy(&ipv6.mask.src_ip, mask->hdr.src_addr,
485 RTE_DIM(ipv6.mask.src_ip));
486 memcpy(&ipv6.mask.dst_ip, mask->hdr.dst_addr,
487 RTE_DIM(ipv6.mask.dst_ip));
488 vtc_flow_val = rte_be_to_cpu_32(spec->hdr.vtc_flow);
489 vtc_flow_mask = rte_be_to_cpu_32(mask->hdr.vtc_flow);
490 ipv6.val.flow_label =
491 rte_cpu_to_be_32((vtc_flow_val & RTE_IPV6_HDR_FL_MASK) >>
492 RTE_IPV6_HDR_FL_SHIFT);
493 ipv6.val.traffic_class = (vtc_flow_val & RTE_IPV6_HDR_TC_MASK) >>
494 RTE_IPV6_HDR_TC_SHIFT;
495 ipv6.val.next_hdr = spec->hdr.proto;
496 ipv6.val.hop_limit = spec->hdr.hop_limits;
497 ipv6.mask.flow_label =
498 rte_cpu_to_be_32((vtc_flow_mask & RTE_IPV6_HDR_FL_MASK) >>
499 RTE_IPV6_HDR_FL_SHIFT);
500 ipv6.mask.traffic_class = (vtc_flow_mask & RTE_IPV6_HDR_TC_MASK) >>
501 RTE_IPV6_HDR_TC_SHIFT;
502 ipv6.mask.next_hdr = mask->hdr.proto;
503 ipv6.mask.hop_limit = mask->hdr.hop_limits;
504 /* Remove unwanted bits from values. */
505 for (i = 0; i < RTE_DIM(ipv6.val.src_ip); ++i) {
506 ipv6.val.src_ip[i] &= ipv6.mask.src_ip[i];
507 ipv6.val.dst_ip[i] &= ipv6.mask.dst_ip[i];
509 ipv6.val.flow_label &= ipv6.mask.flow_label;
510 ipv6.val.traffic_class &= ipv6.mask.traffic_class;
511 ipv6.val.next_hdr &= ipv6.mask.next_hdr;
512 ipv6.val.hop_limit &= ipv6.mask.hop_limit;
514 flow_verbs_spec_add(&dev_flow->verbs, &ipv6, size);
518 * Convert the @p item into a Verbs specification. This function assumes that
519 * the input is valid and that there is space to insert the requested item
522 * @param[in, out] dev_flow
523 * Pointer to dev_flow structure.
525 * Item specification.
526 * @param[in] item_flags
530 flow_verbs_translate_item_tcp(struct mlx5_flow *dev_flow,
531 const struct rte_flow_item *item,
532 uint64_t item_flags __rte_unused)
534 const struct rte_flow_item_tcp *spec = item->spec;
535 const struct rte_flow_item_tcp *mask = item->mask;
536 unsigned int size = sizeof(struct ibv_flow_spec_tcp_udp);
537 struct ibv_flow_spec_tcp_udp tcp = {
538 .type = IBV_FLOW_SPEC_TCP | VERBS_SPEC_INNER(item_flags),
543 mask = &rte_flow_item_tcp_mask;
545 tcp.val.dst_port = spec->hdr.dst_port;
546 tcp.val.src_port = spec->hdr.src_port;
547 tcp.mask.dst_port = mask->hdr.dst_port;
548 tcp.mask.src_port = mask->hdr.src_port;
549 /* Remove unwanted bits from values. */
550 tcp.val.src_port &= tcp.mask.src_port;
551 tcp.val.dst_port &= tcp.mask.dst_port;
553 flow_verbs_spec_add(&dev_flow->verbs, &tcp, size);
557 * Convert the @p item into a Verbs specification. This function assumes that
558 * the input is valid and that there is space to insert the requested item
561 * @param[in, out] dev_flow
562 * Pointer to dev_flow structure.
564 * Item specification.
565 * @param[in] item_flags
569 flow_verbs_translate_item_udp(struct mlx5_flow *dev_flow,
570 const struct rte_flow_item *item,
571 uint64_t item_flags __rte_unused)
573 const struct rte_flow_item_udp *spec = item->spec;
574 const struct rte_flow_item_udp *mask = item->mask;
575 unsigned int size = sizeof(struct ibv_flow_spec_tcp_udp);
576 struct ibv_flow_spec_tcp_udp udp = {
577 .type = IBV_FLOW_SPEC_UDP | VERBS_SPEC_INNER(item_flags),
582 mask = &rte_flow_item_udp_mask;
584 udp.val.dst_port = spec->hdr.dst_port;
585 udp.val.src_port = spec->hdr.src_port;
586 udp.mask.dst_port = mask->hdr.dst_port;
587 udp.mask.src_port = mask->hdr.src_port;
588 /* Remove unwanted bits from values. */
589 udp.val.src_port &= udp.mask.src_port;
590 udp.val.dst_port &= udp.mask.dst_port;
592 flow_verbs_spec_add(&dev_flow->verbs, &udp, size);
596 * Convert the @p item into a Verbs specification. This function assumes that
597 * the input is valid and that there is space to insert the requested item
600 * @param[in, out] dev_flow
601 * Pointer to dev_flow structure.
603 * Item specification.
604 * @param[in] item_flags
608 flow_verbs_translate_item_vxlan(struct mlx5_flow *dev_flow,
609 const struct rte_flow_item *item,
610 uint64_t item_flags __rte_unused)
612 const struct rte_flow_item_vxlan *spec = item->spec;
613 const struct rte_flow_item_vxlan *mask = item->mask;
614 unsigned int size = sizeof(struct ibv_flow_spec_tunnel);
615 struct ibv_flow_spec_tunnel vxlan = {
616 .type = IBV_FLOW_SPEC_VXLAN_TUNNEL,
622 } id = { .vlan_id = 0, };
625 mask = &rte_flow_item_vxlan_mask;
627 memcpy(&id.vni[1], spec->vni, 3);
628 vxlan.val.tunnel_id = id.vlan_id;
629 memcpy(&id.vni[1], mask->vni, 3);
630 vxlan.mask.tunnel_id = id.vlan_id;
631 /* Remove unwanted bits from values. */
632 vxlan.val.tunnel_id &= vxlan.mask.tunnel_id;
634 flow_verbs_spec_add(&dev_flow->verbs, &vxlan, size);
638 * Convert the @p item into a Verbs specification. This function assumes that
639 * the input is valid and that there is space to insert the requested item
642 * @param[in, out] dev_flow
643 * Pointer to dev_flow structure.
645 * Item specification.
646 * @param[in] item_flags
650 flow_verbs_translate_item_vxlan_gpe(struct mlx5_flow *dev_flow,
651 const struct rte_flow_item *item,
652 uint64_t item_flags __rte_unused)
654 const struct rte_flow_item_vxlan_gpe *spec = item->spec;
655 const struct rte_flow_item_vxlan_gpe *mask = item->mask;
656 unsigned int size = sizeof(struct ibv_flow_spec_tunnel);
657 struct ibv_flow_spec_tunnel vxlan_gpe = {
658 .type = IBV_FLOW_SPEC_VXLAN_TUNNEL,
664 } id = { .vlan_id = 0, };
667 mask = &rte_flow_item_vxlan_gpe_mask;
669 memcpy(&id.vni[1], spec->vni, 3);
670 vxlan_gpe.val.tunnel_id = id.vlan_id;
671 memcpy(&id.vni[1], mask->vni, 3);
672 vxlan_gpe.mask.tunnel_id = id.vlan_id;
673 /* Remove unwanted bits from values. */
674 vxlan_gpe.val.tunnel_id &= vxlan_gpe.mask.tunnel_id;
676 flow_verbs_spec_add(&dev_flow->verbs, &vxlan_gpe, size);
680 * Update the protocol in Verbs IPv4/IPv6 spec.
682 * @param[in, out] attr
683 * Pointer to Verbs attributes structure.
685 * Specification type to search in order to update the IP protocol.
686 * @param[in] protocol
687 * Protocol value to set if none is present in the specification.
690 flow_verbs_item_gre_ip_protocol_update(struct ibv_flow_attr *attr,
691 enum ibv_flow_spec_type search,
695 struct ibv_spec_header *hdr = (struct ibv_spec_header *)
696 ((uint8_t *)attr + sizeof(struct ibv_flow_attr));
700 for (i = 0; i != attr->num_of_specs; ++i) {
701 if (hdr->type == search) {
703 struct ibv_flow_spec_ipv4_ext *ipv4;
704 struct ibv_flow_spec_ipv6 *ipv6;
708 case IBV_FLOW_SPEC_IPV4_EXT:
709 ip.ipv4 = (struct ibv_flow_spec_ipv4_ext *)hdr;
710 if (!ip.ipv4->val.proto) {
711 ip.ipv4->val.proto = protocol;
712 ip.ipv4->mask.proto = 0xff;
715 case IBV_FLOW_SPEC_IPV6:
716 ip.ipv6 = (struct ibv_flow_spec_ipv6 *)hdr;
717 if (!ip.ipv6->val.next_hdr) {
718 ip.ipv6->val.next_hdr = protocol;
719 ip.ipv6->mask.next_hdr = 0xff;
727 hdr = (struct ibv_spec_header *)((uint8_t *)hdr + hdr->size);
732 * Convert the @p item into a Verbs specification. This function assumes that
733 * the input is valid and that there is space to insert the requested item
736 * @param[in, out] dev_flow
737 * Pointer to dev_flow structure.
739 * Item specification.
740 * @param[in] item_flags
744 flow_verbs_translate_item_gre(struct mlx5_flow *dev_flow,
745 const struct rte_flow_item *item __rte_unused,
748 struct mlx5_flow_verbs *verbs = &dev_flow->verbs;
749 #ifndef HAVE_IBV_DEVICE_MPLS_SUPPORT
750 unsigned int size = sizeof(struct ibv_flow_spec_tunnel);
751 struct ibv_flow_spec_tunnel tunnel = {
752 .type = IBV_FLOW_SPEC_VXLAN_TUNNEL,
756 const struct rte_flow_item_gre *spec = item->spec;
757 const struct rte_flow_item_gre *mask = item->mask;
758 unsigned int size = sizeof(struct ibv_flow_spec_gre);
759 struct ibv_flow_spec_gre tunnel = {
760 .type = IBV_FLOW_SPEC_GRE,
765 mask = &rte_flow_item_gre_mask;
767 tunnel.val.c_ks_res0_ver = spec->c_rsvd0_ver;
768 tunnel.val.protocol = spec->protocol;
769 tunnel.mask.c_ks_res0_ver = mask->c_rsvd0_ver;
770 tunnel.mask.protocol = mask->protocol;
771 /* Remove unwanted bits from values. */
772 tunnel.val.c_ks_res0_ver &= tunnel.mask.c_ks_res0_ver;
773 tunnel.val.protocol &= tunnel.mask.protocol;
774 tunnel.val.key &= tunnel.mask.key;
777 if (item_flags & MLX5_FLOW_LAYER_OUTER_L3_IPV4)
778 flow_verbs_item_gre_ip_protocol_update(verbs->attr,
779 IBV_FLOW_SPEC_IPV4_EXT,
782 flow_verbs_item_gre_ip_protocol_update(verbs->attr,
785 flow_verbs_spec_add(verbs, &tunnel, size);
789 * Convert the @p action into a Verbs specification. This function assumes that
790 * the input is valid and that there is space to insert the requested action
791 * into the flow. This function also return the action that was added.
793 * @param[in, out] dev_flow
794 * Pointer to dev_flow structure.
796 * Item specification.
797 * @param[in] item_flags
801 flow_verbs_translate_item_mpls(struct mlx5_flow *dev_flow __rte_unused,
802 const struct rte_flow_item *item __rte_unused,
803 uint64_t item_flags __rte_unused)
805 #ifdef HAVE_IBV_DEVICE_MPLS_SUPPORT
806 const struct rte_flow_item_mpls *spec = item->spec;
807 const struct rte_flow_item_mpls *mask = item->mask;
808 unsigned int size = sizeof(struct ibv_flow_spec_mpls);
809 struct ibv_flow_spec_mpls mpls = {
810 .type = IBV_FLOW_SPEC_MPLS,
815 mask = &rte_flow_item_mpls_mask;
817 memcpy(&mpls.val.label, spec, sizeof(mpls.val.label));
818 memcpy(&mpls.mask.label, mask, sizeof(mpls.mask.label));
819 /* Remove unwanted bits from values. */
820 mpls.val.label &= mpls.mask.label;
822 flow_verbs_spec_add(&dev_flow->verbs, &mpls, size);
827 * Convert the @p action into a Verbs specification. This function assumes that
828 * the input is valid and that there is space to insert the requested action
831 * @param[in] dev_flow
832 * Pointer to mlx5_flow.
834 * Action configuration.
837 flow_verbs_translate_action_drop
838 (struct mlx5_flow *dev_flow,
839 const struct rte_flow_action *action __rte_unused)
841 unsigned int size = sizeof(struct ibv_flow_spec_action_drop);
842 struct ibv_flow_spec_action_drop drop = {
843 .type = IBV_FLOW_SPEC_ACTION_DROP,
847 flow_verbs_spec_add(&dev_flow->verbs, &drop, size);
851 * Convert the @p action into a Verbs specification. This function assumes that
852 * the input is valid and that there is space to insert the requested action
855 * @param[in] dev_flow
856 * Pointer to mlx5_flow.
858 * Action configuration.
861 flow_verbs_translate_action_queue(struct mlx5_flow *dev_flow,
862 const struct rte_flow_action *action)
864 const struct rte_flow_action_queue *queue = action->conf;
865 struct rte_flow *flow = dev_flow->flow;
868 (*flow->queue)[0] = queue->index;
869 flow->rss.queue_num = 1;
873 * Convert the @p action into a Verbs specification. This function assumes that
874 * the input is valid and that there is space to insert the requested action
878 * Action configuration.
879 * @param[in, out] action_flags
880 * Pointer to the detected actions.
881 * @param[in] dev_flow
882 * Pointer to mlx5_flow.
885 flow_verbs_translate_action_rss(struct mlx5_flow *dev_flow,
886 const struct rte_flow_action *action)
888 const struct rte_flow_action_rss *rss = action->conf;
889 const uint8_t *rss_key;
890 struct rte_flow *flow = dev_flow->flow;
893 memcpy((*flow->queue), rss->queue,
894 rss->queue_num * sizeof(uint16_t));
895 flow->rss.queue_num = rss->queue_num;
896 /* NULL RSS key indicates default RSS key. */
897 rss_key = !rss->key ? rss_hash_default_key : rss->key;
898 memcpy(flow->key, rss_key, MLX5_RSS_HASH_KEY_LEN);
899 /* RSS type 0 indicates default RSS type (ETH_RSS_IP). */
900 flow->rss.types = !rss->types ? ETH_RSS_IP : rss->types;
901 flow->rss.level = rss->level;
905 * Convert the @p action into a Verbs specification. This function assumes that
906 * the input is valid and that there is space to insert the requested action
909 * @param[in] dev_flow
910 * Pointer to mlx5_flow.
912 * Action configuration.
915 flow_verbs_translate_action_flag
916 (struct mlx5_flow *dev_flow,
917 const struct rte_flow_action *action __rte_unused)
919 unsigned int size = sizeof(struct ibv_flow_spec_action_tag);
920 struct ibv_flow_spec_action_tag tag = {
921 .type = IBV_FLOW_SPEC_ACTION_TAG,
923 .tag_id = mlx5_flow_mark_set(MLX5_FLOW_MARK_DEFAULT),
926 flow_verbs_spec_add(&dev_flow->verbs, &tag, size);
930 * Convert the @p action into a Verbs specification. This function assumes that
931 * the input is valid and that there is space to insert the requested action
934 * @param[in] dev_flow
935 * Pointer to mlx5_flow.
937 * Action configuration.
940 flow_verbs_translate_action_mark(struct mlx5_flow *dev_flow,
941 const struct rte_flow_action *action)
943 const struct rte_flow_action_mark *mark = action->conf;
944 unsigned int size = sizeof(struct ibv_flow_spec_action_tag);
945 struct ibv_flow_spec_action_tag tag = {
946 .type = IBV_FLOW_SPEC_ACTION_TAG,
948 .tag_id = mlx5_flow_mark_set(mark->id),
951 flow_verbs_spec_add(&dev_flow->verbs, &tag, size);
955 * Convert the @p action into a Verbs specification. This function assumes that
956 * the input is valid and that there is space to insert the requested action
960 * Pointer to the Ethernet device structure.
962 * Action configuration.
963 * @param[in] dev_flow
964 * Pointer to mlx5_flow.
966 * Pointer to error structure.
969 * 0 On success else a negative errno value is returned and rte_errno is set.
972 flow_verbs_translate_action_count(struct mlx5_flow *dev_flow,
973 const struct rte_flow_action *action,
974 struct rte_eth_dev *dev,
975 struct rte_flow_error *error)
977 const struct rte_flow_action_count *count = action->conf;
978 struct rte_flow *flow = dev_flow->flow;
979 #if defined(HAVE_IBV_DEVICE_COUNTERS_SET_V42) || \
980 defined(HAVE_IBV_DEVICE_COUNTERS_SET_V45)
981 unsigned int size = sizeof(struct ibv_flow_spec_counter_action);
982 struct ibv_flow_spec_counter_action counter = {
983 .type = IBV_FLOW_SPEC_ACTION_COUNT,
988 if (!flow->counter) {
989 flow->counter = flow_verbs_counter_new(dev, count->shared,
992 return rte_flow_error_set(error, rte_errno,
993 RTE_FLOW_ERROR_TYPE_ACTION,
998 #if defined(HAVE_IBV_DEVICE_COUNTERS_SET_V42)
999 counter.counter_set_handle = flow->counter->cs->handle;
1000 flow_verbs_spec_add(&dev_flow->verbs, &counter, size);
1001 #elif defined(HAVE_IBV_DEVICE_COUNTERS_SET_V45)
1002 counter.counters = flow->counter->cs;
1003 flow_verbs_spec_add(&dev_flow->verbs, &counter, size);
1009 * Internal validation function. For validating both actions and items.
1012 * Pointer to the Ethernet device structure.
1014 * Pointer to the flow attributes.
1016 * Pointer to the list of items.
1017 * @param[in] actions
1018 * Pointer to the list of actions.
1020 * Pointer to the error structure.
1023 * 0 on success, a negative errno value otherwise and rte_errno is set.
1026 flow_verbs_validate(struct rte_eth_dev *dev,
1027 const struct rte_flow_attr *attr,
1028 const struct rte_flow_item items[],
1029 const struct rte_flow_action actions[],
1030 struct rte_flow_error *error)
1033 uint64_t action_flags = 0;
1034 uint64_t item_flags = 0;
1035 uint64_t last_item = 0;
1036 uint8_t next_protocol = 0xff;
1040 ret = mlx5_flow_validate_attributes(dev, attr, error);
1043 for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
1044 int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1047 switch (items->type) {
1048 case RTE_FLOW_ITEM_TYPE_VOID:
1050 case RTE_FLOW_ITEM_TYPE_ETH:
1051 ret = mlx5_flow_validate_item_eth(items, item_flags,
1055 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L2 :
1056 MLX5_FLOW_LAYER_OUTER_L2;
1058 case RTE_FLOW_ITEM_TYPE_VLAN:
1059 ret = mlx5_flow_validate_item_vlan(items, item_flags,
1063 last_item = tunnel ? (MLX5_FLOW_LAYER_INNER_L2 |
1064 MLX5_FLOW_LAYER_INNER_VLAN) :
1065 (MLX5_FLOW_LAYER_OUTER_L2 |
1066 MLX5_FLOW_LAYER_OUTER_VLAN);
1068 case RTE_FLOW_ITEM_TYPE_IPV4:
1069 ret = mlx5_flow_validate_item_ipv4(items, item_flags,
1073 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
1074 MLX5_FLOW_LAYER_OUTER_L3_IPV4;
1075 if (items->mask != NULL &&
1076 ((const struct rte_flow_item_ipv4 *)
1077 items->mask)->hdr.next_proto_id) {
1079 ((const struct rte_flow_item_ipv4 *)
1080 (items->spec))->hdr.next_proto_id;
1082 ((const struct rte_flow_item_ipv4 *)
1083 (items->mask))->hdr.next_proto_id;
1085 /* Reset for inner layer. */
1086 next_protocol = 0xff;
1089 case RTE_FLOW_ITEM_TYPE_IPV6:
1090 ret = mlx5_flow_validate_item_ipv6(items, item_flags,
1094 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
1095 MLX5_FLOW_LAYER_OUTER_L3_IPV6;
1096 if (items->mask != NULL &&
1097 ((const struct rte_flow_item_ipv6 *)
1098 items->mask)->hdr.proto) {
1100 ((const struct rte_flow_item_ipv6 *)
1101 items->spec)->hdr.proto;
1103 ((const struct rte_flow_item_ipv6 *)
1104 items->mask)->hdr.proto;
1106 /* Reset for inner layer. */
1107 next_protocol = 0xff;
1110 case RTE_FLOW_ITEM_TYPE_UDP:
1111 ret = mlx5_flow_validate_item_udp(items, item_flags,
1116 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L4_UDP :
1117 MLX5_FLOW_LAYER_OUTER_L4_UDP;
1119 case RTE_FLOW_ITEM_TYPE_TCP:
1120 ret = mlx5_flow_validate_item_tcp
1123 &rte_flow_item_tcp_mask,
1127 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L4_TCP :
1128 MLX5_FLOW_LAYER_OUTER_L4_TCP;
1130 case RTE_FLOW_ITEM_TYPE_VXLAN:
1131 ret = mlx5_flow_validate_item_vxlan(items, item_flags,
1135 last_item = MLX5_FLOW_LAYER_VXLAN;
1137 case RTE_FLOW_ITEM_TYPE_VXLAN_GPE:
1138 ret = mlx5_flow_validate_item_vxlan_gpe(items,
1143 last_item = MLX5_FLOW_LAYER_VXLAN_GPE;
1145 case RTE_FLOW_ITEM_TYPE_GRE:
1146 ret = mlx5_flow_validate_item_gre(items, item_flags,
1147 next_protocol, error);
1150 last_item = MLX5_FLOW_LAYER_GRE;
1152 case RTE_FLOW_ITEM_TYPE_MPLS:
1153 ret = mlx5_flow_validate_item_mpls(dev, items,
1158 last_item = MLX5_FLOW_LAYER_MPLS;
1161 return rte_flow_error_set(error, ENOTSUP,
1162 RTE_FLOW_ERROR_TYPE_ITEM,
1163 NULL, "item not supported");
1165 item_flags |= last_item;
1167 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
1168 switch (actions->type) {
1169 case RTE_FLOW_ACTION_TYPE_VOID:
1171 case RTE_FLOW_ACTION_TYPE_FLAG:
1172 ret = mlx5_flow_validate_action_flag(action_flags,
1177 action_flags |= MLX5_FLOW_ACTION_FLAG;
1179 case RTE_FLOW_ACTION_TYPE_MARK:
1180 ret = mlx5_flow_validate_action_mark(actions,
1186 action_flags |= MLX5_FLOW_ACTION_MARK;
1188 case RTE_FLOW_ACTION_TYPE_DROP:
1189 ret = mlx5_flow_validate_action_drop(action_flags,
1194 action_flags |= MLX5_FLOW_ACTION_DROP;
1196 case RTE_FLOW_ACTION_TYPE_QUEUE:
1197 ret = mlx5_flow_validate_action_queue(actions,
1203 action_flags |= MLX5_FLOW_ACTION_QUEUE;
1205 case RTE_FLOW_ACTION_TYPE_RSS:
1206 ret = mlx5_flow_validate_action_rss(actions,
1212 action_flags |= MLX5_FLOW_ACTION_RSS;
1214 case RTE_FLOW_ACTION_TYPE_COUNT:
1215 ret = mlx5_flow_validate_action_count(dev, attr, error);
1218 action_flags |= MLX5_FLOW_ACTION_COUNT;
1221 return rte_flow_error_set(error, ENOTSUP,
1222 RTE_FLOW_ERROR_TYPE_ACTION,
1224 "action not supported");
1227 if (!(action_flags & MLX5_FLOW_FATE_ACTIONS))
1228 return rte_flow_error_set(error, EINVAL,
1229 RTE_FLOW_ERROR_TYPE_ACTION, actions,
1230 "no fate action is found");
1235 * Calculate the required bytes that are needed for the action part of the verbs
1238 * @param[in] actions
1239 * Pointer to the list of actions.
1242 * The size of the memory needed for all actions.
1245 flow_verbs_get_actions_size(const struct rte_flow_action actions[])
1249 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
1250 switch (actions->type) {
1251 case RTE_FLOW_ACTION_TYPE_VOID:
1253 case RTE_FLOW_ACTION_TYPE_FLAG:
1254 size += sizeof(struct ibv_flow_spec_action_tag);
1256 case RTE_FLOW_ACTION_TYPE_MARK:
1257 size += sizeof(struct ibv_flow_spec_action_tag);
1259 case RTE_FLOW_ACTION_TYPE_DROP:
1260 size += sizeof(struct ibv_flow_spec_action_drop);
1262 case RTE_FLOW_ACTION_TYPE_QUEUE:
1264 case RTE_FLOW_ACTION_TYPE_RSS:
1266 case RTE_FLOW_ACTION_TYPE_COUNT:
1267 #if defined(HAVE_IBV_DEVICE_COUNTERS_SET_V42) || \
1268 defined(HAVE_IBV_DEVICE_COUNTERS_SET_V45)
1269 size += sizeof(struct ibv_flow_spec_counter_action);
1280 * Calculate the required bytes that are needed for the item part of the verbs
1284 * Pointer to the list of items.
1287 * The size of the memory needed for all items.
1290 flow_verbs_get_items_size(const struct rte_flow_item items[])
1294 for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
1295 switch (items->type) {
1296 case RTE_FLOW_ITEM_TYPE_VOID:
1298 case RTE_FLOW_ITEM_TYPE_ETH:
1299 size += sizeof(struct ibv_flow_spec_eth);
1301 case RTE_FLOW_ITEM_TYPE_VLAN:
1302 size += sizeof(struct ibv_flow_spec_eth);
1304 case RTE_FLOW_ITEM_TYPE_IPV4:
1305 size += sizeof(struct ibv_flow_spec_ipv4_ext);
1307 case RTE_FLOW_ITEM_TYPE_IPV6:
1308 size += sizeof(struct ibv_flow_spec_ipv6);
1310 case RTE_FLOW_ITEM_TYPE_UDP:
1311 size += sizeof(struct ibv_flow_spec_tcp_udp);
1313 case RTE_FLOW_ITEM_TYPE_TCP:
1314 size += sizeof(struct ibv_flow_spec_tcp_udp);
1316 case RTE_FLOW_ITEM_TYPE_VXLAN:
1317 size += sizeof(struct ibv_flow_spec_tunnel);
1319 case RTE_FLOW_ITEM_TYPE_VXLAN_GPE:
1320 size += sizeof(struct ibv_flow_spec_tunnel);
1322 #ifdef HAVE_IBV_DEVICE_MPLS_SUPPORT
1323 case RTE_FLOW_ITEM_TYPE_GRE:
1324 size += sizeof(struct ibv_flow_spec_gre);
1326 case RTE_FLOW_ITEM_TYPE_MPLS:
1327 size += sizeof(struct ibv_flow_spec_mpls);
1330 case RTE_FLOW_ITEM_TYPE_GRE:
1331 size += sizeof(struct ibv_flow_spec_tunnel);
1342 * Internal preparation function. Allocate mlx5_flow with the required size.
1343 * The required size is calculate based on the actions and items. This function
1344 * also returns the detected actions and items for later use.
1347 * Pointer to the flow attributes.
1349 * Pointer to the list of items.
1350 * @param[in] actions
1351 * Pointer to the list of actions.
1353 * Pointer to the error structure.
1356 * Pointer to mlx5_flow object on success, otherwise NULL and rte_errno
1359 static struct mlx5_flow *
1360 flow_verbs_prepare(const struct rte_flow_attr *attr __rte_unused,
1361 const struct rte_flow_item items[],
1362 const struct rte_flow_action actions[],
1363 struct rte_flow_error *error)
1365 uint32_t size = sizeof(struct mlx5_flow) + sizeof(struct ibv_flow_attr);
1366 struct mlx5_flow *flow;
1368 size += flow_verbs_get_actions_size(actions);
1369 size += flow_verbs_get_items_size(items);
1370 flow = rte_calloc(__func__, 1, size, 0);
1372 rte_flow_error_set(error, ENOMEM,
1373 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
1374 "not enough memory to create flow");
1377 flow->verbs.attr = (void *)(flow + 1);
1379 (uint8_t *)(flow + 1) + sizeof(struct ibv_flow_attr);
1384 * Fill the flow with verb spec.
1387 * Pointer to Ethernet device.
1388 * @param[in, out] dev_flow
1389 * Pointer to the mlx5 flow.
1391 * Pointer to the flow attributes.
1393 * Pointer to the list of items.
1394 * @param[in] actions
1395 * Pointer to the list of actions.
1397 * Pointer to the error structure.
1400 * 0 on success, else a negative errno value otherwise and rte_errno is set.
1403 flow_verbs_translate(struct rte_eth_dev *dev,
1404 struct mlx5_flow *dev_flow,
1405 const struct rte_flow_attr *attr,
1406 const struct rte_flow_item items[],
1407 const struct rte_flow_action actions[],
1408 struct rte_flow_error *error)
1410 struct rte_flow *flow = dev_flow->flow;
1411 uint64_t item_flags = 0;
1412 uint64_t action_flags = 0;
1413 uint64_t priority = attr->priority;
1414 uint32_t subpriority = 0;
1415 struct mlx5_priv *priv = dev->data->dev_private;
1417 if (priority == MLX5_FLOW_PRIO_RSVD)
1418 priority = priv->config.flow_prio - 1;
1419 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
1422 switch (actions->type) {
1423 case RTE_FLOW_ACTION_TYPE_VOID:
1425 case RTE_FLOW_ACTION_TYPE_FLAG:
1426 flow_verbs_translate_action_flag(dev_flow, actions);
1427 action_flags |= MLX5_FLOW_ACTION_FLAG;
1429 case RTE_FLOW_ACTION_TYPE_MARK:
1430 flow_verbs_translate_action_mark(dev_flow, actions);
1431 action_flags |= MLX5_FLOW_ACTION_MARK;
1433 case RTE_FLOW_ACTION_TYPE_DROP:
1434 flow_verbs_translate_action_drop(dev_flow, actions);
1435 action_flags |= MLX5_FLOW_ACTION_DROP;
1437 case RTE_FLOW_ACTION_TYPE_QUEUE:
1438 flow_verbs_translate_action_queue(dev_flow, actions);
1439 action_flags |= MLX5_FLOW_ACTION_QUEUE;
1441 case RTE_FLOW_ACTION_TYPE_RSS:
1442 flow_verbs_translate_action_rss(dev_flow, actions);
1443 action_flags |= MLX5_FLOW_ACTION_RSS;
1445 case RTE_FLOW_ACTION_TYPE_COUNT:
1446 ret = flow_verbs_translate_action_count(dev_flow,
1451 action_flags |= MLX5_FLOW_ACTION_COUNT;
1454 return rte_flow_error_set(error, ENOTSUP,
1455 RTE_FLOW_ERROR_TYPE_ACTION,
1457 "action not supported");
1460 flow->actions = action_flags;
1461 for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
1462 int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1464 switch (items->type) {
1465 case RTE_FLOW_ITEM_TYPE_VOID:
1467 case RTE_FLOW_ITEM_TYPE_ETH:
1468 flow_verbs_translate_item_eth(dev_flow, items,
1470 subpriority = MLX5_PRIORITY_MAP_L2;
1471 item_flags |= tunnel ? MLX5_FLOW_LAYER_INNER_L2 :
1472 MLX5_FLOW_LAYER_OUTER_L2;
1474 case RTE_FLOW_ITEM_TYPE_VLAN:
1475 flow_verbs_translate_item_vlan(dev_flow, items,
1477 subpriority = MLX5_PRIORITY_MAP_L2;
1478 item_flags |= tunnel ? (MLX5_FLOW_LAYER_INNER_L2 |
1479 MLX5_FLOW_LAYER_INNER_VLAN) :
1480 (MLX5_FLOW_LAYER_OUTER_L2 |
1481 MLX5_FLOW_LAYER_OUTER_VLAN);
1483 case RTE_FLOW_ITEM_TYPE_IPV4:
1484 flow_verbs_translate_item_ipv4(dev_flow, items,
1486 subpriority = MLX5_PRIORITY_MAP_L3;
1487 dev_flow->verbs.hash_fields |=
1488 mlx5_flow_hashfields_adjust
1490 MLX5_IPV4_LAYER_TYPES,
1491 MLX5_IPV4_IBV_RX_HASH);
1492 item_flags |= tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
1493 MLX5_FLOW_LAYER_OUTER_L3_IPV4;
1495 case RTE_FLOW_ITEM_TYPE_IPV6:
1496 flow_verbs_translate_item_ipv6(dev_flow, items,
1498 subpriority = MLX5_PRIORITY_MAP_L3;
1499 dev_flow->verbs.hash_fields |=
1500 mlx5_flow_hashfields_adjust
1502 MLX5_IPV6_LAYER_TYPES,
1503 MLX5_IPV6_IBV_RX_HASH);
1504 item_flags |= tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
1505 MLX5_FLOW_LAYER_OUTER_L3_IPV6;
1507 case RTE_FLOW_ITEM_TYPE_TCP:
1508 flow_verbs_translate_item_tcp(dev_flow, items,
1510 subpriority = MLX5_PRIORITY_MAP_L4;
1511 dev_flow->verbs.hash_fields |=
1512 mlx5_flow_hashfields_adjust
1513 (dev_flow, tunnel, ETH_RSS_TCP,
1514 (IBV_RX_HASH_SRC_PORT_TCP |
1515 IBV_RX_HASH_DST_PORT_TCP));
1516 item_flags |= tunnel ? MLX5_FLOW_LAYER_INNER_L4_TCP :
1517 MLX5_FLOW_LAYER_OUTER_L4_TCP;
1519 case RTE_FLOW_ITEM_TYPE_UDP:
1520 flow_verbs_translate_item_udp(dev_flow, items,
1522 subpriority = MLX5_PRIORITY_MAP_L4;
1523 dev_flow->verbs.hash_fields |=
1524 mlx5_flow_hashfields_adjust
1525 (dev_flow, tunnel, ETH_RSS_UDP,
1526 (IBV_RX_HASH_SRC_PORT_UDP |
1527 IBV_RX_HASH_DST_PORT_UDP));
1528 item_flags |= tunnel ? MLX5_FLOW_LAYER_INNER_L4_UDP :
1529 MLX5_FLOW_LAYER_OUTER_L4_UDP;
1531 case RTE_FLOW_ITEM_TYPE_VXLAN:
1532 flow_verbs_translate_item_vxlan(dev_flow, items,
1534 subpriority = MLX5_PRIORITY_MAP_L2;
1535 item_flags |= MLX5_FLOW_LAYER_VXLAN;
1537 case RTE_FLOW_ITEM_TYPE_VXLAN_GPE:
1538 flow_verbs_translate_item_vxlan_gpe(dev_flow, items,
1540 subpriority = MLX5_PRIORITY_MAP_L2;
1541 item_flags |= MLX5_FLOW_LAYER_VXLAN_GPE;
1543 case RTE_FLOW_ITEM_TYPE_GRE:
1544 flow_verbs_translate_item_gre(dev_flow, items,
1546 subpriority = MLX5_PRIORITY_MAP_L2;
1547 item_flags |= MLX5_FLOW_LAYER_GRE;
1549 case RTE_FLOW_ITEM_TYPE_MPLS:
1550 flow_verbs_translate_item_mpls(dev_flow, items,
1552 subpriority = MLX5_PRIORITY_MAP_L2;
1553 item_flags |= MLX5_FLOW_LAYER_MPLS;
1556 return rte_flow_error_set(error, ENOTSUP,
1557 RTE_FLOW_ERROR_TYPE_ITEM,
1559 "item not supported");
1562 dev_flow->layers = item_flags;
1563 dev_flow->verbs.attr->priority =
1564 mlx5_flow_adjust_priority(dev, priority, subpriority);
1565 dev_flow->verbs.attr->port = (uint8_t)priv->ibv_port;
1570 * Remove the flow from the NIC but keeps it in memory.
1573 * Pointer to the Ethernet device structure.
1574 * @param[in, out] flow
1575 * Pointer to flow structure.
1578 flow_verbs_remove(struct rte_eth_dev *dev, struct rte_flow *flow)
1580 struct mlx5_flow_verbs *verbs;
1581 struct mlx5_flow *dev_flow;
1585 LIST_FOREACH(dev_flow, &flow->dev_flows, next) {
1586 verbs = &dev_flow->verbs;
1588 claim_zero(mlx5_glue->destroy_flow(verbs->flow));
1592 if (flow->actions & MLX5_FLOW_ACTION_DROP)
1593 mlx5_hrxq_drop_release(dev);
1595 mlx5_hrxq_release(dev, verbs->hrxq);
1598 if (dev_flow->verbs.vf_vlan.tag &&
1599 dev_flow->verbs.vf_vlan.created) {
1600 mlx5_vlan_vmwa_release(dev, &dev_flow->verbs.vf_vlan);
1606 * Remove the flow from the NIC and the memory.
1609 * Pointer to the Ethernet device structure.
1610 * @param[in, out] flow
1611 * Pointer to flow structure.
1614 flow_verbs_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
1616 struct mlx5_flow *dev_flow;
1620 flow_verbs_remove(dev, flow);
1621 while (!LIST_EMPTY(&flow->dev_flows)) {
1622 dev_flow = LIST_FIRST(&flow->dev_flows);
1623 LIST_REMOVE(dev_flow, next);
1626 if (flow->counter) {
1627 flow_verbs_counter_release(dev, flow->counter);
1628 flow->counter = NULL;
1633 * Apply the flow to the NIC.
1636 * Pointer to the Ethernet device structure.
1637 * @param[in, out] flow
1638 * Pointer to flow structure.
1640 * Pointer to error structure.
1643 * 0 on success, a negative errno value otherwise and rte_errno is set.
1646 flow_verbs_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
1647 struct rte_flow_error *error)
1649 struct mlx5_priv *priv = dev->data->dev_private;
1650 struct mlx5_flow_verbs *verbs;
1651 struct mlx5_flow *dev_flow;
1654 LIST_FOREACH(dev_flow, &flow->dev_flows, next) {
1655 verbs = &dev_flow->verbs;
1656 if (flow->actions & MLX5_FLOW_ACTION_DROP) {
1657 verbs->hrxq = mlx5_hrxq_drop_new(dev);
1661 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
1662 "cannot get drop hash queue");
1666 struct mlx5_hrxq *hrxq;
1668 hrxq = mlx5_hrxq_get(dev, flow->key,
1669 MLX5_RSS_HASH_KEY_LEN,
1672 flow->rss.queue_num);
1674 hrxq = mlx5_hrxq_new(dev, flow->key,
1675 MLX5_RSS_HASH_KEY_LEN,
1678 flow->rss.queue_num,
1679 !!(dev_flow->layers &
1680 MLX5_FLOW_LAYER_TUNNEL));
1684 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
1685 "cannot get hash queue");
1690 verbs->flow = mlx5_glue->create_flow(verbs->hrxq->qp,
1693 rte_flow_error_set(error, errno,
1694 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
1696 "hardware refuses to create flow");
1699 if (priv->vmwa_context &&
1700 dev_flow->verbs.vf_vlan.tag &&
1701 !dev_flow->verbs.vf_vlan.created) {
1703 * The rule contains the VLAN pattern.
1704 * For VF we are going to create VLAN
1705 * interface to make hypervisor set correct
1706 * e-Switch vport context.
1708 mlx5_vlan_vmwa_acquire(dev, &dev_flow->verbs.vf_vlan);
1713 err = rte_errno; /* Save rte_errno before cleanup. */
1714 LIST_FOREACH(dev_flow, &flow->dev_flows, next) {
1715 verbs = &dev_flow->verbs;
1717 if (flow->actions & MLX5_FLOW_ACTION_DROP)
1718 mlx5_hrxq_drop_release(dev);
1720 mlx5_hrxq_release(dev, verbs->hrxq);
1723 if (dev_flow->verbs.vf_vlan.tag &&
1724 dev_flow->verbs.vf_vlan.created) {
1725 mlx5_vlan_vmwa_release(dev, &dev_flow->verbs.vf_vlan);
1728 rte_errno = err; /* Restore rte_errno. */
1735 * @see rte_flow_query()
1739 flow_verbs_query(struct rte_eth_dev *dev,
1740 struct rte_flow *flow,
1741 const struct rte_flow_action *actions,
1743 struct rte_flow_error *error)
1747 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
1748 switch (actions->type) {
1749 case RTE_FLOW_ACTION_TYPE_VOID:
1751 case RTE_FLOW_ACTION_TYPE_COUNT:
1752 ret = flow_verbs_counter_query(dev, flow, data, error);
1755 return rte_flow_error_set(error, ENOTSUP,
1756 RTE_FLOW_ERROR_TYPE_ACTION,
1758 "action not supported");
1764 const struct mlx5_flow_driver_ops mlx5_flow_verbs_drv_ops = {
1765 .validate = flow_verbs_validate,
1766 .prepare = flow_verbs_prepare,
1767 .translate = flow_verbs_translate,
1768 .apply = flow_verbs_apply,
1769 .remove = flow_verbs_remove,
1770 .destroy = flow_verbs_destroy,
1771 .query = flow_verbs_query,