]> git.proxmox.com Git - ceph.git/blob - ceph/src/spdk/dpdk/drivers/net/mlx5/mlx5_flow_verbs.c
projects / ceph.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
import 15.2.0 Octopus source
[ceph.git] / ceph / src / spdk / dpdk / drivers / net / mlx5 / mlx5_flow_verbs.c
1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright 2018 Mellanox Technologies, Ltd
3 */
4
5 #include <netinet/in.h>
6 #include <sys/queue.h>
7 #include <stdalign.h>
8 #include <stdint.h>
9 #include <string.h>
10
11 /* Verbs header. */
12 /* ISO C doesn't support unnamed structs/unions, disabling -pedantic. */
13 #ifdef PEDANTIC
14 #pragma GCC diagnostic ignored "-Wpedantic"
15 #endif
16 #include <infiniband/verbs.h>
17 #ifdef PEDANTIC
18 #pragma GCC diagnostic error "-Wpedantic"
19 #endif
20
21 #include <rte_common.h>
22 #include <rte_ether.h>
23 #include <rte_ethdev_driver.h>
24 #include <rte_flow.h>
25 #include <rte_flow_driver.h>
26 #include <rte_malloc.h>
27 #include <rte_ip.h>
28
29 #include "mlx5.h"
30 #include "mlx5_defs.h"
31 #include "mlx5_flow.h"
32 #include "mlx5_glue.h"
33 #include "mlx5_prm.h"
34 #include "mlx5_rxtx.h"
35
36 #define VERBS_SPEC_INNER(item_flags) \
37 (!!((item_flags) & MLX5_FLOW_LAYER_TUNNEL) ? IBV_FLOW_SPEC_INNER : 0)
38
39 /**
40 * Create Verbs flow counter with Verbs library.
41 *
42 * @param[in] dev
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).
48 *
49 * @return
50 * 0 On success else a negative errno value is returned
51 * and rte_errno is set.
52 */
53 static int
54 flow_verbs_counter_create(struct rte_eth_dev *dev,
55 struct mlx5_flow_counter *counter)
56 {
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};
62
63 counter->cs = mlx5_glue->create_counter_set(ctx, &init);
64 if (!counter->cs) {
65 rte_errno = ENOTSUP;
66 return -ENOTSUP;
67 }
68 return 0;
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;
74 int ret;
75
76 memset(&attach, 0, sizeof(attach));
77 counter->cs = mlx5_glue->create_counters(ctx, &init);
78 if (!counter->cs) {
79 rte_errno = ENOTSUP;
80 return -ENOTSUP;
81 }
82 attach.counter_desc = IBV_COUNTER_PACKETS;
83 attach.index = 0;
84 ret = mlx5_glue->attach_counters(counter->cs, &attach, NULL);
85 if (!ret) {
86 attach.counter_desc = IBV_COUNTER_BYTES;
87 attach.index = 1;
88 ret = mlx5_glue->attach_counters
89 (counter->cs, &attach, NULL);
90 }
91 if (ret) {
92 claim_zero(mlx5_glue->destroy_counters(counter->cs));
93 counter->cs = NULL;
94 rte_errno = ret;
95 return -ret;
96 }
97 return 0;
98 #else
99 (void)dev;
100 (void)counter;
101 rte_errno = ENOTSUP;
102 return -ENOTSUP;
103 #endif
104 }
105
106 /**
107 * Get a flow counter.
108 *
109 * @param[in] dev
110 * Pointer to the Ethernet device structure.
111 * @param[in] shared
112 * Indicate if this counter is shared with other flows.
113 * @param[in] id
114 * Counter identifier.
115 *
116 * @return
117 * A pointer to the counter, NULL otherwise and rte_errno is set.
118 */
119 static struct mlx5_flow_counter *
120 flow_verbs_counter_new(struct rte_eth_dev *dev, uint32_t shared, uint32_t id)
121 {
122 struct mlx5_priv *priv = dev->data->dev_private;
123 struct mlx5_flow_counter *cnt;
124 int ret;
125
126 if (shared) {
127 LIST_FOREACH(cnt, &priv->flow_counters, next) {
128 if (cnt->shared && cnt->id == id) {
129 cnt->ref_cnt++;
130 return cnt;
131 }
132 }
133 }
134 cnt = rte_calloc(__func__, 1, sizeof(*cnt), 0);
135 if (!cnt) {
136 rte_errno = ENOMEM;
137 return NULL;
138 }
139 cnt->id = id;
140 cnt->shared = shared;
141 cnt->ref_cnt = 1;
142 cnt->hits = 0;
143 cnt->bytes = 0;
144 /* Create counter with Verbs. */
145 ret = flow_verbs_counter_create(dev, cnt);
146 if (!ret) {
147 LIST_INSERT_HEAD(&priv->flow_counters, cnt, next);
148 return cnt;
149 }
150 /* Some error occurred in Verbs library. */
151 rte_free(cnt);
152 rte_errno = -ret;
153 return NULL;
154 }
155
156 /**
157 * Release a flow counter.
158 *
159 * @param[in] counter
160 * Pointer to the counter handler.
161 */
162 static void
163 flow_verbs_counter_release(struct mlx5_flow_counter *counter)
164 {
165 if (--counter->ref_cnt == 0) {
166 #if defined(HAVE_IBV_DEVICE_COUNTERS_SET_V42)
167 claim_zero(mlx5_glue->destroy_counter_set(counter->cs));
168 #elif defined(HAVE_IBV_DEVICE_COUNTERS_SET_V45)
169 claim_zero(mlx5_glue->destroy_counters(counter->cs));
170 #endif
171 LIST_REMOVE(counter, next);
172 rte_free(counter);
173 }
174 }
175
176 /**
177 * Query a flow counter via Verbs library call.
178 *
179 * @see rte_flow_query()
180 * @see rte_flow_ops
181 */
182 static int
183 flow_verbs_counter_query(struct rte_eth_dev *dev __rte_unused,
184 struct rte_flow *flow, void *data,
185 struct rte_flow_error *error)
186 {
187 #if defined(HAVE_IBV_DEVICE_COUNTERS_SET_V42) || \
188 defined(HAVE_IBV_DEVICE_COUNTERS_SET_V45)
189 if (flow->actions & MLX5_FLOW_ACTION_COUNT) {
190 struct rte_flow_query_count *qc = data;
191 uint64_t counters[2] = {0, 0};
192 #if defined(HAVE_IBV_DEVICE_COUNTERS_SET_V42)
193 struct ibv_query_counter_set_attr query_cs_attr = {
194 .cs = flow->counter->cs,
195 .query_flags = IBV_COUNTER_SET_FORCE_UPDATE,
196 };
197 struct ibv_counter_set_data query_out = {
198 .out = counters,
199 .outlen = 2 * sizeof(uint64_t),
200 };
201 int err = mlx5_glue->query_counter_set(&query_cs_attr,
202 &query_out);
203 #elif defined(HAVE_IBV_DEVICE_COUNTERS_SET_V45)
204 int err = mlx5_glue->query_counters
205 (flow->counter->cs, counters,
206 RTE_DIM(counters),
207 IBV_READ_COUNTERS_ATTR_PREFER_CACHED);
208 #endif
209 if (err)
210 return rte_flow_error_set
211 (error, err,
212 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
213 NULL,
214 "cannot read counter");
215 qc->hits_set = 1;
216 qc->bytes_set = 1;
217 qc->hits = counters[0] - flow->counter->hits;
218 qc->bytes = counters[1] - flow->counter->bytes;
219 if (qc->reset) {
220 flow->counter->hits = counters[0];
221 flow->counter->bytes = counters[1];
222 }
223 return 0;
224 }
225 return rte_flow_error_set(error, EINVAL,
226 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
227 NULL,
228 "flow does not have counter");
229 #else
230 (void)flow;
231 (void)data;
232 return rte_flow_error_set(error, ENOTSUP,
233 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
234 NULL,
235 "counters are not available");
236 #endif
237 }
238
239 /**
240 * Add a verbs item specification into @p verbs.
241 *
242 * @param[out] verbs
243 * Pointer to verbs structure.
244 * @param[in] src
245 * Create specification.
246 * @param[in] size
247 * Size in bytes of the specification to copy.
248 */
249 static void
250 flow_verbs_spec_add(struct mlx5_flow_verbs *verbs, void *src, unsigned int size)
251 {
252 void *dst;
253
254 if (!verbs)
255 return;
256 assert(verbs->specs);
257 dst = (void *)(verbs->specs + verbs->size);
258 memcpy(dst, src, size);
259 ++verbs->attr->num_of_specs;
260 verbs->size += size;
261 }
262
263 /**
264 * Convert the @p item into a Verbs specification. This function assumes that
265 * the input is valid and that there is space to insert the requested item
266 * into the flow.
267 *
268 * @param[in, out] dev_flow
269 * Pointer to dev_flow structure.
270 * @param[in] item
271 * Item specification.
272 * @param[in] item_flags
273 * Parsed item flags.
274 */
275 static void
276 flow_verbs_translate_item_eth(struct mlx5_flow *dev_flow,
277 const struct rte_flow_item *item,
278 uint64_t item_flags)
279 {
280 const struct rte_flow_item_eth *spec = item->spec;
281 const struct rte_flow_item_eth *mask = item->mask;
282 const unsigned int size = sizeof(struct ibv_flow_spec_eth);
283 struct ibv_flow_spec_eth eth = {
284 .type = IBV_FLOW_SPEC_ETH | VERBS_SPEC_INNER(item_flags),
285 .size = size,
286 };
287
288 if (!mask)
289 mask = &rte_flow_item_eth_mask;
290 if (spec) {
291 unsigned int i;
292
293 memcpy(&eth.val.dst_mac, spec->dst.addr_bytes, ETHER_ADDR_LEN);
294 memcpy(&eth.val.src_mac, spec->src.addr_bytes, ETHER_ADDR_LEN);
295 eth.val.ether_type = spec->type;
296 memcpy(&eth.mask.dst_mac, mask->dst.addr_bytes, ETHER_ADDR_LEN);
297 memcpy(&eth.mask.src_mac, mask->src.addr_bytes, ETHER_ADDR_LEN);
298 eth.mask.ether_type = mask->type;
299 /* Remove unwanted bits from values. */
300 for (i = 0; i < ETHER_ADDR_LEN; ++i) {
301 eth.val.dst_mac[i] &= eth.mask.dst_mac[i];
302 eth.val.src_mac[i] &= eth.mask.src_mac[i];
303 }
304 eth.val.ether_type &= eth.mask.ether_type;
305 }
306 flow_verbs_spec_add(&dev_flow->verbs, &eth, size);
307 }
308
309 /**
310 * Update the VLAN tag in the Verbs Ethernet specification.
311 * This function assumes that the input is valid and there is space to add
312 * the requested item.
313 *
314 * @param[in, out] attr
315 * Pointer to Verbs attributes structure.
316 * @param[in] eth
317 * Verbs structure containing the VLAN information to copy.
318 */
319 static void
320 flow_verbs_item_vlan_update(struct ibv_flow_attr *attr,
321 struct ibv_flow_spec_eth *eth)
322 {
323 unsigned int i;
324 const enum ibv_flow_spec_type search = eth->type;
325 struct ibv_spec_header *hdr = (struct ibv_spec_header *)
326 ((uint8_t *)attr + sizeof(struct ibv_flow_attr));
327
328 for (i = 0; i != attr->num_of_specs; ++i) {
329 if (hdr->type == search) {
330 struct ibv_flow_spec_eth *e =
331 (struct ibv_flow_spec_eth *)hdr;
332
333 e->val.vlan_tag = eth->val.vlan_tag;
334 e->mask.vlan_tag = eth->mask.vlan_tag;
335 e->val.ether_type = eth->val.ether_type;
336 e->mask.ether_type = eth->mask.ether_type;
337 break;
338 }
339 hdr = (struct ibv_spec_header *)((uint8_t *)hdr + hdr->size);
340 }
341 }
342
343 /**
344 * Convert the @p item into a Verbs specification. This function assumes that
345 * the input is valid and that there is space to insert the requested item
346 * into the flow.
347 *
348 * @param[in, out] dev_flow
349 * Pointer to dev_flow structure.
350 * @param[in] item
351 * Item specification.
352 * @param[in] item_flags
353 * Parsed item flags.
354 */
355 static void
356 flow_verbs_translate_item_vlan(struct mlx5_flow *dev_flow,
357 const struct rte_flow_item *item,
358 uint64_t item_flags)
359 {
360 const struct rte_flow_item_vlan *spec = item->spec;
361 const struct rte_flow_item_vlan *mask = item->mask;
362 unsigned int size = sizeof(struct ibv_flow_spec_eth);
363 const int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
364 struct ibv_flow_spec_eth eth = {
365 .type = IBV_FLOW_SPEC_ETH | VERBS_SPEC_INNER(item_flags),
366 .size = size,
367 };
368 const uint32_t l2m = tunnel ? MLX5_FLOW_LAYER_INNER_L2 :
369 MLX5_FLOW_LAYER_OUTER_L2;
370
371 if (!mask)
372 mask = &rte_flow_item_vlan_mask;
373 if (spec) {
374 eth.val.vlan_tag = spec->tci;
375 eth.mask.vlan_tag = mask->tci;
376 eth.val.vlan_tag &= eth.mask.vlan_tag;
377 eth.val.ether_type = spec->inner_type;
378 eth.mask.ether_type = mask->inner_type;
379 eth.val.ether_type &= eth.mask.ether_type;
380 }
381 if (!(item_flags & l2m))
382 flow_verbs_spec_add(&dev_flow->verbs, &eth, size);
383 else
384 flow_verbs_item_vlan_update(dev_flow->verbs.attr, &eth);
385 }
386
387 /**
388 * Convert the @p item into a Verbs specification. This function assumes that
389 * the input is valid and that there is space to insert the requested item
390 * into the flow.
391 *
392 * @param[in, out] dev_flow
393 * Pointer to dev_flow structure.
394 * @param[in] item
395 * Item specification.
396 * @param[in] item_flags
397 * Parsed item flags.
398 */
399 static void
400 flow_verbs_translate_item_ipv4(struct mlx5_flow *dev_flow,
401 const struct rte_flow_item *item,
402 uint64_t item_flags)
403 {
404 const struct rte_flow_item_ipv4 *spec = item->spec;
405 const struct rte_flow_item_ipv4 *mask = item->mask;
406 unsigned int size = sizeof(struct ibv_flow_spec_ipv4_ext);
407 struct ibv_flow_spec_ipv4_ext ipv4 = {
408 .type = IBV_FLOW_SPEC_IPV4_EXT | VERBS_SPEC_INNER(item_flags),
409 .size = size,
410 };
411
412 if (!mask)
413 mask = &rte_flow_item_ipv4_mask;
414 if (spec) {
415 ipv4.val = (struct ibv_flow_ipv4_ext_filter){
416 .src_ip = spec->hdr.src_addr,
417 .dst_ip = spec->hdr.dst_addr,
418 .proto = spec->hdr.next_proto_id,
419 .tos = spec->hdr.type_of_service,
420 };
421 ipv4.mask = (struct ibv_flow_ipv4_ext_filter){
422 .src_ip = mask->hdr.src_addr,
423 .dst_ip = mask->hdr.dst_addr,
424 .proto = mask->hdr.next_proto_id,
425 .tos = mask->hdr.type_of_service,
426 };
427 /* Remove unwanted bits from values. */
428 ipv4.val.src_ip &= ipv4.mask.src_ip;
429 ipv4.val.dst_ip &= ipv4.mask.dst_ip;
430 ipv4.val.proto &= ipv4.mask.proto;
431 ipv4.val.tos &= ipv4.mask.tos;
432 }
433 flow_verbs_spec_add(&dev_flow->verbs, &ipv4, size);
434 }
435
436 /**
437 * Convert the @p item into a Verbs specification. This function assumes that
438 * the input is valid and that there is space to insert the requested item
439 * into the flow.
440 *
441 * @param[in, out] dev_flow
442 * Pointer to dev_flow structure.
443 * @param[in] item
444 * Item specification.
445 * @param[in] item_flags
446 * Parsed item flags.
447 */
448 static void
449 flow_verbs_translate_item_ipv6(struct mlx5_flow *dev_flow,
450 const struct rte_flow_item *item,
451 uint64_t item_flags)
452 {
453 const struct rte_flow_item_ipv6 *spec = item->spec;
454 const struct rte_flow_item_ipv6 *mask = item->mask;
455 unsigned int size = sizeof(struct ibv_flow_spec_ipv6);
456 struct ibv_flow_spec_ipv6 ipv6 = {
457 .type = IBV_FLOW_SPEC_IPV6 | VERBS_SPEC_INNER(item_flags),
458 .size = size,
459 };
460
461 if (!mask)
462 mask = &rte_flow_item_ipv6_mask;
463 if (spec) {
464 unsigned int i;
465 uint32_t vtc_flow_val;
466 uint32_t vtc_flow_mask;
467
468 memcpy(&ipv6.val.src_ip, spec->hdr.src_addr,
469 RTE_DIM(ipv6.val.src_ip));
470 memcpy(&ipv6.val.dst_ip, spec->hdr.dst_addr,
471 RTE_DIM(ipv6.val.dst_ip));
472 memcpy(&ipv6.mask.src_ip, mask->hdr.src_addr,
473 RTE_DIM(ipv6.mask.src_ip));
474 memcpy(&ipv6.mask.dst_ip, mask->hdr.dst_addr,
475 RTE_DIM(ipv6.mask.dst_ip));
476 vtc_flow_val = rte_be_to_cpu_32(spec->hdr.vtc_flow);
477 vtc_flow_mask = rte_be_to_cpu_32(mask->hdr.vtc_flow);
478 ipv6.val.flow_label =
479 rte_cpu_to_be_32((vtc_flow_val & IPV6_HDR_FL_MASK) >>
480 IPV6_HDR_FL_SHIFT);
481 ipv6.val.traffic_class = (vtc_flow_val & IPV6_HDR_TC_MASK) >>
482 IPV6_HDR_TC_SHIFT;
483 ipv6.val.next_hdr = spec->hdr.proto;
484 ipv6.val.hop_limit = spec->hdr.hop_limits;
485 ipv6.mask.flow_label =
486 rte_cpu_to_be_32((vtc_flow_mask & IPV6_HDR_FL_MASK) >>
487 IPV6_HDR_FL_SHIFT);
488 ipv6.mask.traffic_class = (vtc_flow_mask & IPV6_HDR_TC_MASK) >>
489 IPV6_HDR_TC_SHIFT;
490 ipv6.mask.next_hdr = mask->hdr.proto;
491 ipv6.mask.hop_limit = mask->hdr.hop_limits;
492 /* Remove unwanted bits from values. */
493 for (i = 0; i < RTE_DIM(ipv6.val.src_ip); ++i) {
494 ipv6.val.src_ip[i] &= ipv6.mask.src_ip[i];
495 ipv6.val.dst_ip[i] &= ipv6.mask.dst_ip[i];
496 }
497 ipv6.val.flow_label &= ipv6.mask.flow_label;
498 ipv6.val.traffic_class &= ipv6.mask.traffic_class;
499 ipv6.val.next_hdr &= ipv6.mask.next_hdr;
500 ipv6.val.hop_limit &= ipv6.mask.hop_limit;
501 }
502 flow_verbs_spec_add(&dev_flow->verbs, &ipv6, size);
503 }
504
505 /**
506 * Convert the @p item into a Verbs specification. This function assumes that
507 * the input is valid and that there is space to insert the requested item
508 * into the flow.
509 *
510 * @param[in, out] dev_flow
511 * Pointer to dev_flow structure.
512 * @param[in] item
513 * Item specification.
514 * @param[in] item_flags
515 * Parsed item flags.
516 */
517 static void
518 flow_verbs_translate_item_tcp(struct mlx5_flow *dev_flow,
519 const struct rte_flow_item *item,
520 uint64_t item_flags __rte_unused)
521 {
522 const struct rte_flow_item_tcp *spec = item->spec;
523 const struct rte_flow_item_tcp *mask = item->mask;
524 unsigned int size = sizeof(struct ibv_flow_spec_tcp_udp);
525 struct ibv_flow_spec_tcp_udp tcp = {
526 .type = IBV_FLOW_SPEC_TCP | VERBS_SPEC_INNER(item_flags),
527 .size = size,
528 };
529
530 if (!mask)
531 mask = &rte_flow_item_tcp_mask;
532 if (spec) {
533 tcp.val.dst_port = spec->hdr.dst_port;
534 tcp.val.src_port = spec->hdr.src_port;
535 tcp.mask.dst_port = mask->hdr.dst_port;
536 tcp.mask.src_port = mask->hdr.src_port;
537 /* Remove unwanted bits from values. */
538 tcp.val.src_port &= tcp.mask.src_port;
539 tcp.val.dst_port &= tcp.mask.dst_port;
540 }
541 flow_verbs_spec_add(&dev_flow->verbs, &tcp, size);
542 }
543
544 /**
545 * Convert the @p item into a Verbs specification. This function assumes that
546 * the input is valid and that there is space to insert the requested item
547 * into the flow.
548 *
549 * @param[in, out] dev_flow
550 * Pointer to dev_flow structure.
551 * @param[in] item
552 * Item specification.
553 * @param[in] item_flags
554 * Parsed item flags.
555 */
556 static void
557 flow_verbs_translate_item_udp(struct mlx5_flow *dev_flow,
558 const struct rte_flow_item *item,
559 uint64_t item_flags __rte_unused)
560 {
561 const struct rte_flow_item_udp *spec = item->spec;
562 const struct rte_flow_item_udp *mask = item->mask;
563 unsigned int size = sizeof(struct ibv_flow_spec_tcp_udp);
564 struct ibv_flow_spec_tcp_udp udp = {
565 .type = IBV_FLOW_SPEC_UDP | VERBS_SPEC_INNER(item_flags),
566 .size = size,
567 };
568
569 if (!mask)
570 mask = &rte_flow_item_udp_mask;
571 if (spec) {
572 udp.val.dst_port = spec->hdr.dst_port;
573 udp.val.src_port = spec->hdr.src_port;
574 udp.mask.dst_port = mask->hdr.dst_port;
575 udp.mask.src_port = mask->hdr.src_port;
576 /* Remove unwanted bits from values. */
577 udp.val.src_port &= udp.mask.src_port;
578 udp.val.dst_port &= udp.mask.dst_port;
579 }
580 flow_verbs_spec_add(&dev_flow->verbs, &udp, size);
581 }
582
583 /**
584 * Convert the @p item into a Verbs specification. This function assumes that
585 * the input is valid and that there is space to insert the requested item
586 * into the flow.
587 *
588 * @param[in, out] dev_flow
589 * Pointer to dev_flow structure.
590 * @param[in] item
591 * Item specification.
592 * @param[in] item_flags
593 * Parsed item flags.
594 */
595 static void
596 flow_verbs_translate_item_vxlan(struct mlx5_flow *dev_flow,
597 const struct rte_flow_item *item,
598 uint64_t item_flags __rte_unused)
599 {
600 const struct rte_flow_item_vxlan *spec = item->spec;
601 const struct rte_flow_item_vxlan *mask = item->mask;
602 unsigned int size = sizeof(struct ibv_flow_spec_tunnel);
603 struct ibv_flow_spec_tunnel vxlan = {
604 .type = IBV_FLOW_SPEC_VXLAN_TUNNEL,
605 .size = size,
606 };
607 union vni {
608 uint32_t vlan_id;
609 uint8_t vni[4];
610 } id = { .vlan_id = 0, };
611
612 if (!mask)
613 mask = &rte_flow_item_vxlan_mask;
614 if (spec) {
615 memcpy(&id.vni[1], spec->vni, 3);
616 vxlan.val.tunnel_id = id.vlan_id;
617 memcpy(&id.vni[1], mask->vni, 3);
618 vxlan.mask.tunnel_id = id.vlan_id;
619 /* Remove unwanted bits from values. */
620 vxlan.val.tunnel_id &= vxlan.mask.tunnel_id;
621 }
622 flow_verbs_spec_add(&dev_flow->verbs, &vxlan, size);
623 }
624
625 /**
626 * Convert the @p item into a Verbs specification. This function assumes that
627 * the input is valid and that there is space to insert the requested item
628 * into the flow.
629 *
630 * @param[in, out] dev_flow
631 * Pointer to dev_flow structure.
632 * @param[in] item
633 * Item specification.
634 * @param[in] item_flags
635 * Parsed item flags.
636 */
637 static void
638 flow_verbs_translate_item_vxlan_gpe(struct mlx5_flow *dev_flow,
639 const struct rte_flow_item *item,
640 uint64_t item_flags __rte_unused)
641 {
642 const struct rte_flow_item_vxlan_gpe *spec = item->spec;
643 const struct rte_flow_item_vxlan_gpe *mask = item->mask;
644 unsigned int size = sizeof(struct ibv_flow_spec_tunnel);
645 struct ibv_flow_spec_tunnel vxlan_gpe = {
646 .type = IBV_FLOW_SPEC_VXLAN_TUNNEL,
647 .size = size,
648 };
649 union vni {
650 uint32_t vlan_id;
651 uint8_t vni[4];
652 } id = { .vlan_id = 0, };
653
654 if (!mask)
655 mask = &rte_flow_item_vxlan_gpe_mask;
656 if (spec) {
657 memcpy(&id.vni[1], spec->vni, 3);
658 vxlan_gpe.val.tunnel_id = id.vlan_id;
659 memcpy(&id.vni[1], mask->vni, 3);
660 vxlan_gpe.mask.tunnel_id = id.vlan_id;
661 /* Remove unwanted bits from values. */
662 vxlan_gpe.val.tunnel_id &= vxlan_gpe.mask.tunnel_id;
663 }
664 flow_verbs_spec_add(&dev_flow->verbs, &vxlan_gpe, size);
665 }
666
667 /**
668 * Update the protocol in Verbs IPv4/IPv6 spec.
669 *
670 * @param[in, out] attr
671 * Pointer to Verbs attributes structure.
672 * @param[in] search
673 * Specification type to search in order to update the IP protocol.
674 * @param[in] protocol
675 * Protocol value to set if none is present in the specification.
676 */
677 static void
678 flow_verbs_item_gre_ip_protocol_update(struct ibv_flow_attr *attr,
679 enum ibv_flow_spec_type search,
680 uint8_t protocol)
681 {
682 unsigned int i;
683 struct ibv_spec_header *hdr = (struct ibv_spec_header *)
684 ((uint8_t *)attr + sizeof(struct ibv_flow_attr));
685
686 if (!attr)
687 return;
688 for (i = 0; i != attr->num_of_specs; ++i) {
689 if (hdr->type == search) {
690 union {
691 struct ibv_flow_spec_ipv4_ext *ipv4;
692 struct ibv_flow_spec_ipv6 *ipv6;
693 } ip;
694
695 switch (search) {
696 case IBV_FLOW_SPEC_IPV4_EXT:
697 ip.ipv4 = (struct ibv_flow_spec_ipv4_ext *)hdr;
698 if (!ip.ipv4->val.proto) {
699 ip.ipv4->val.proto = protocol;
700 ip.ipv4->mask.proto = 0xff;
701 }
702 break;
703 case IBV_FLOW_SPEC_IPV6:
704 ip.ipv6 = (struct ibv_flow_spec_ipv6 *)hdr;
705 if (!ip.ipv6->val.next_hdr) {
706 ip.ipv6->val.next_hdr = protocol;
707 ip.ipv6->mask.next_hdr = 0xff;
708 }
709 break;
710 default:
711 break;
712 }
713 break;
714 }
715 hdr = (struct ibv_spec_header *)((uint8_t *)hdr + hdr->size);
716 }
717 }
718
719 /**
720 * Convert the @p item into a Verbs specification. This function assumes that
721 * the input is valid and that there is space to insert the requested item
722 * into the flow.
723 *
724 * @param[in, out] dev_flow
725 * Pointer to dev_flow structure.
726 * @param[in] item
727 * Item specification.
728 * @param[in] item_flags
729 * Parsed item flags.
730 */
731 static void
732 flow_verbs_translate_item_gre(struct mlx5_flow *dev_flow,
733 const struct rte_flow_item *item __rte_unused,
734 uint64_t item_flags)
735 {
736 struct mlx5_flow_verbs *verbs = &dev_flow->verbs;
737 #ifndef HAVE_IBV_DEVICE_MPLS_SUPPORT
738 unsigned int size = sizeof(struct ibv_flow_spec_tunnel);
739 struct ibv_flow_spec_tunnel tunnel = {
740 .type = IBV_FLOW_SPEC_VXLAN_TUNNEL,
741 .size = size,
742 };
743 #else
744 const struct rte_flow_item_gre *spec = item->spec;
745 const struct rte_flow_item_gre *mask = item->mask;
746 unsigned int size = sizeof(struct ibv_flow_spec_gre);
747 struct ibv_flow_spec_gre tunnel = {
748 .type = IBV_FLOW_SPEC_GRE,
749 .size = size,
750 };
751
752 if (!mask)
753 mask = &rte_flow_item_gre_mask;
754 if (spec) {
755 tunnel.val.c_ks_res0_ver = spec->c_rsvd0_ver;
756 tunnel.val.protocol = spec->protocol;
757 tunnel.mask.c_ks_res0_ver = mask->c_rsvd0_ver;
758 tunnel.mask.protocol = mask->protocol;
759 /* Remove unwanted bits from values. */
760 tunnel.val.c_ks_res0_ver &= tunnel.mask.c_ks_res0_ver;
761 tunnel.val.protocol &= tunnel.mask.protocol;
762 tunnel.val.key &= tunnel.mask.key;
763 }
764 #endif
765 if (item_flags & MLX5_FLOW_LAYER_OUTER_L3_IPV4)
766 flow_verbs_item_gre_ip_protocol_update(verbs->attr,
767 IBV_FLOW_SPEC_IPV4_EXT,
768 IPPROTO_GRE);
769 else
770 flow_verbs_item_gre_ip_protocol_update(verbs->attr,
771 IBV_FLOW_SPEC_IPV6,
772 IPPROTO_GRE);
773 flow_verbs_spec_add(verbs, &tunnel, size);
774 }
775
776 /**
777 * Convert the @p action into a Verbs specification. This function assumes that
778 * the input is valid and that there is space to insert the requested action
779 * into the flow. This function also return the action that was added.
780 *
781 * @param[in, out] dev_flow
782 * Pointer to dev_flow structure.
783 * @param[in] item
784 * Item specification.
785 * @param[in] item_flags
786 * Parsed item flags.
787 */
788 static void
789 flow_verbs_translate_item_mpls(struct mlx5_flow *dev_flow __rte_unused,
790 const struct rte_flow_item *item __rte_unused,
791 uint64_t item_flags __rte_unused)
792 {
793 #ifdef HAVE_IBV_DEVICE_MPLS_SUPPORT
794 const struct rte_flow_item_mpls *spec = item->spec;
795 const struct rte_flow_item_mpls *mask = item->mask;
796 unsigned int size = sizeof(struct ibv_flow_spec_mpls);
797 struct ibv_flow_spec_mpls mpls = {
798 .type = IBV_FLOW_SPEC_MPLS,
799 .size = size,
800 };
801
802 if (!mask)
803 mask = &rte_flow_item_mpls_mask;
804 if (spec) {
805 memcpy(&mpls.val.label, spec, sizeof(mpls.val.label));
806 memcpy(&mpls.mask.label, mask, sizeof(mpls.mask.label));
807 /* Remove unwanted bits from values. */
808 mpls.val.label &= mpls.mask.label;
809 }
810 flow_verbs_spec_add(&dev_flow->verbs, &mpls, size);
811 #endif
812 }
813
814 /**
815 * Convert the @p action into a Verbs specification. This function assumes that
816 * the input is valid and that there is space to insert the requested action
817 * into the flow.
818 *
819 * @param[in] dev_flow
820 * Pointer to mlx5_flow.
821 * @param[in] action
822 * Action configuration.
823 */
824 static void
825 flow_verbs_translate_action_drop
826 (struct mlx5_flow *dev_flow,
827 const struct rte_flow_action *action __rte_unused)
828 {
829 unsigned int size = sizeof(struct ibv_flow_spec_action_drop);
830 struct ibv_flow_spec_action_drop drop = {
831 .type = IBV_FLOW_SPEC_ACTION_DROP,
832 .size = size,
833 };
834
835 flow_verbs_spec_add(&dev_flow->verbs, &drop, size);
836 }
837
838 /**
839 * Convert the @p action into a Verbs specification. This function assumes that
840 * the input is valid and that there is space to insert the requested action
841 * into the flow.
842 *
843 * @param[in] dev_flow
844 * Pointer to mlx5_flow.
845 * @param[in] action
846 * Action configuration.
847 */
848 static void
849 flow_verbs_translate_action_queue(struct mlx5_flow *dev_flow,
850 const struct rte_flow_action *action)
851 {
852 const struct rte_flow_action_queue *queue = action->conf;
853 struct rte_flow *flow = dev_flow->flow;
854
855 if (flow->queue)
856 (*flow->queue)[0] = queue->index;
857 flow->rss.queue_num = 1;
858 }
859
860 /**
861 * Convert the @p action into a Verbs specification. This function assumes that
862 * the input is valid and that there is space to insert the requested action
863 * into the flow.
864 *
865 * @param[in] action
866 * Action configuration.
867 * @param[in, out] action_flags
868 * Pointer to the detected actions.
869 * @param[in] dev_flow
870 * Pointer to mlx5_flow.
871 */
872 static void
873 flow_verbs_translate_action_rss(struct mlx5_flow *dev_flow,
874 const struct rte_flow_action *action)
875 {
876 const struct rte_flow_action_rss *rss = action->conf;
877 const uint8_t *rss_key;
878 struct rte_flow *flow = dev_flow->flow;
879
880 if (flow->queue)
881 memcpy((*flow->queue), rss->queue,
882 rss->queue_num * sizeof(uint16_t));
883 flow->rss.queue_num = rss->queue_num;
884 /* NULL RSS key indicates default RSS key. */
885 rss_key = !rss->key ? rss_hash_default_key : rss->key;
886 memcpy(flow->key, rss_key, MLX5_RSS_HASH_KEY_LEN);
887 /* RSS type 0 indicates default RSS type (ETH_RSS_IP). */
888 flow->rss.types = !rss->types ? ETH_RSS_IP : rss->types;
889 flow->rss.level = rss->level;
890 }
891
892 /**
893 * Convert the @p action into a Verbs specification. This function assumes that
894 * the input is valid and that there is space to insert the requested action
895 * into the flow.
896 *
897 * @param[in] dev_flow
898 * Pointer to mlx5_flow.
899 * @param[in] action
900 * Action configuration.
901 */
902 static void
903 flow_verbs_translate_action_flag
904 (struct mlx5_flow *dev_flow,
905 const struct rte_flow_action *action __rte_unused)
906 {
907 unsigned int size = sizeof(struct ibv_flow_spec_action_tag);
908 struct ibv_flow_spec_action_tag tag = {
909 .type = IBV_FLOW_SPEC_ACTION_TAG,
910 .size = size,
911 .tag_id = mlx5_flow_mark_set(MLX5_FLOW_MARK_DEFAULT),
912 };
913
914 flow_verbs_spec_add(&dev_flow->verbs, &tag, size);
915 }
916
917 /**
918 * Convert the @p action into a Verbs specification. This function assumes that
919 * the input is valid and that there is space to insert the requested action
920 * into the flow.
921 *
922 * @param[in] dev_flow
923 * Pointer to mlx5_flow.
924 * @param[in] action
925 * Action configuration.
926 */
927 static void
928 flow_verbs_translate_action_mark(struct mlx5_flow *dev_flow,
929 const struct rte_flow_action *action)
930 {
931 const struct rte_flow_action_mark *mark = action->conf;
932 unsigned int size = sizeof(struct ibv_flow_spec_action_tag);
933 struct ibv_flow_spec_action_tag tag = {
934 .type = IBV_FLOW_SPEC_ACTION_TAG,
935 .size = size,
936 .tag_id = mlx5_flow_mark_set(mark->id),
937 };
938
939 flow_verbs_spec_add(&dev_flow->verbs, &tag, size);
940 }
941
942 /**
943 * Convert the @p action into a Verbs specification. This function assumes that
944 * the input is valid and that there is space to insert the requested action
945 * into the flow.
946 *
947 * @param[in] dev
948 * Pointer to the Ethernet device structure.
949 * @param[in] action
950 * Action configuration.
951 * @param[in] dev_flow
952 * Pointer to mlx5_flow.
953 * @param[out] error
954 * Pointer to error structure.
955 *
956 * @return
957 * 0 On success else a negative errno value is returned and rte_errno is set.
958 */
959 static int
960 flow_verbs_translate_action_count(struct mlx5_flow *dev_flow,
961 const struct rte_flow_action *action,
962 struct rte_eth_dev *dev,
963 struct rte_flow_error *error)
964 {
965 const struct rte_flow_action_count *count = action->conf;
966 struct rte_flow *flow = dev_flow->flow;
967 #if defined(HAVE_IBV_DEVICE_COUNTERS_SET_V42) || \
968 defined(HAVE_IBV_DEVICE_COUNTERS_SET_V45)
969 unsigned int size = sizeof(struct ibv_flow_spec_counter_action);
970 struct ibv_flow_spec_counter_action counter = {
971 .type = IBV_FLOW_SPEC_ACTION_COUNT,
972 .size = size,
973 };
974 #endif
975
976 if (!flow->counter) {
977 flow->counter = flow_verbs_counter_new(dev, count->shared,
978 count->id);
979 if (!flow->counter)
980 return rte_flow_error_set(error, rte_errno,
981 RTE_FLOW_ERROR_TYPE_ACTION,
982 action,
983 "cannot get counter"
984 " context.");
985 }
986 #if defined(HAVE_IBV_DEVICE_COUNTERS_SET_V42)
987 counter.counter_set_handle = flow->counter->cs->handle;
988 flow_verbs_spec_add(&dev_flow->verbs, &counter, size);
989 #elif defined(HAVE_IBV_DEVICE_COUNTERS_SET_V45)
990 counter.counters = flow->counter->cs;
991 flow_verbs_spec_add(&dev_flow->verbs, &counter, size);
992 #endif
993 return 0;
994 }
995
996 /**
997 * Internal validation function. For validating both actions and items.
998 *
999 * @param[in] dev
1000 * Pointer to the Ethernet device structure.
1001 * @param[in] attr
1002 * Pointer to the flow attributes.
1003 * @param[in] items
1004 * Pointer to the list of items.
1005 * @param[in] actions
1006 * Pointer to the list of actions.
1007 * @param[out] error
1008 * Pointer to the error structure.
1009 *
1010 * @return
1011 * 0 on success, a negative errno value otherwise and rte_errno is set.
1012 */
1013 static int
1014 flow_verbs_validate(struct rte_eth_dev *dev,
1015 const struct rte_flow_attr *attr,
1016 const struct rte_flow_item items[],
1017 const struct rte_flow_action actions[],
1018 struct rte_flow_error *error)
1019 {
1020 int ret;
1021 uint64_t action_flags = 0;
1022 uint64_t item_flags = 0;
1023 uint64_t last_item = 0;
1024 uint8_t next_protocol = 0xff;
1025
1026 if (items == NULL)
1027 return -1;
1028 ret = mlx5_flow_validate_attributes(dev, attr, error);
1029 if (ret < 0)
1030 return ret;
1031 for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
1032 int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1033 int ret = 0;
1034
1035 switch (items->type) {
1036 case RTE_FLOW_ITEM_TYPE_VOID:
1037 break;
1038 case RTE_FLOW_ITEM_TYPE_ETH:
1039 ret = mlx5_flow_validate_item_eth(items, item_flags,
1040 error);
1041 if (ret < 0)
1042 return ret;
1043 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L2 :
1044 MLX5_FLOW_LAYER_OUTER_L2;
1045 break;
1046 case RTE_FLOW_ITEM_TYPE_VLAN:
1047 ret = mlx5_flow_validate_item_vlan(items, item_flags,
1048 error);
1049 if (ret < 0)
1050 return ret;
1051 last_item = tunnel ? (MLX5_FLOW_LAYER_INNER_L2 |
1052 MLX5_FLOW_LAYER_INNER_VLAN) :
1053 (MLX5_FLOW_LAYER_OUTER_L2 |
1054 MLX5_FLOW_LAYER_OUTER_VLAN);
1055 break;
1056 case RTE_FLOW_ITEM_TYPE_IPV4:
1057 ret = mlx5_flow_validate_item_ipv4(items, item_flags,
1058 NULL, error);
1059 if (ret < 0)
1060 return ret;
1061 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
1062 MLX5_FLOW_LAYER_OUTER_L3_IPV4;
1063 if (items->mask != NULL &&
1064 ((const struct rte_flow_item_ipv4 *)
1065 items->mask)->hdr.next_proto_id) {
1066 next_protocol =
1067 ((const struct rte_flow_item_ipv4 *)
1068 (items->spec))->hdr.next_proto_id;
1069 next_protocol &=
1070 ((const struct rte_flow_item_ipv4 *)
1071 (items->mask))->hdr.next_proto_id;
1072 } else {
1073 /* Reset for inner layer. */
1074 next_protocol = 0xff;
1075 }
1076 break;
1077 case RTE_FLOW_ITEM_TYPE_IPV6:
1078 ret = mlx5_flow_validate_item_ipv6(items, item_flags,
1079 NULL, error);
1080 if (ret < 0)
1081 return ret;
1082 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
1083 MLX5_FLOW_LAYER_OUTER_L3_IPV6;
1084 if (items->mask != NULL &&
1085 ((const struct rte_flow_item_ipv6 *)
1086 items->mask)->hdr.proto) {
1087 next_protocol =
1088 ((const struct rte_flow_item_ipv6 *)
1089 items->spec)->hdr.proto;
1090 next_protocol &=
1091 ((const struct rte_flow_item_ipv6 *)
1092 items->mask)->hdr.proto;
1093 } else {
1094 /* Reset for inner layer. */
1095 next_protocol = 0xff;
1096 }
1097 break;
1098 case RTE_FLOW_ITEM_TYPE_UDP:
1099 ret = mlx5_flow_validate_item_udp(items, item_flags,
1100 next_protocol,
1101 error);
1102 if (ret < 0)
1103 return ret;
1104 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L4_UDP :
1105 MLX5_FLOW_LAYER_OUTER_L4_UDP;
1106 break;
1107 case RTE_FLOW_ITEM_TYPE_TCP:
1108 ret = mlx5_flow_validate_item_tcp
1109 (items, item_flags,
1110 next_protocol,
1111 &rte_flow_item_tcp_mask,
1112 error);
1113 if (ret < 0)
1114 return ret;
1115 last_item = tunnel ? MLX5_FLOW_LAYER_INNER_L4_TCP :
1116 MLX5_FLOW_LAYER_OUTER_L4_TCP;
1117 break;
1118 case RTE_FLOW_ITEM_TYPE_VXLAN:
1119 ret = mlx5_flow_validate_item_vxlan(items, item_flags,
1120 error);
1121 if (ret < 0)
1122 return ret;
1123 last_item = MLX5_FLOW_LAYER_VXLAN;
1124 break;
1125 case RTE_FLOW_ITEM_TYPE_VXLAN_GPE:
1126 ret = mlx5_flow_validate_item_vxlan_gpe(items,
1127 item_flags,
1128 dev, error);
1129 if (ret < 0)
1130 return ret;
1131 last_item = MLX5_FLOW_LAYER_VXLAN_GPE;
1132 break;
1133 case RTE_FLOW_ITEM_TYPE_GRE:
1134 ret = mlx5_flow_validate_item_gre(items, item_flags,
1135 next_protocol, error);
1136 if (ret < 0)
1137 return ret;
1138 last_item = MLX5_FLOW_LAYER_GRE;
1139 break;
1140 case RTE_FLOW_ITEM_TYPE_MPLS:
1141 ret = mlx5_flow_validate_item_mpls(dev, items,
1142 item_flags,
1143 last_item, error);
1144 if (ret < 0)
1145 return ret;
1146 last_item = MLX5_FLOW_LAYER_MPLS;
1147 break;
1148 default:
1149 return rte_flow_error_set(error, ENOTSUP,
1150 RTE_FLOW_ERROR_TYPE_ITEM,
1151 NULL, "item not supported");
1152 }
1153 item_flags |= last_item;
1154 }
1155 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
1156 switch (actions->type) {
1157 case RTE_FLOW_ACTION_TYPE_VOID:
1158 break;
1159 case RTE_FLOW_ACTION_TYPE_FLAG:
1160 ret = mlx5_flow_validate_action_flag(action_flags,
1161 attr,
1162 error);
1163 if (ret < 0)
1164 return ret;
1165 action_flags |= MLX5_FLOW_ACTION_FLAG;
1166 break;
1167 case RTE_FLOW_ACTION_TYPE_MARK:
1168 ret = mlx5_flow_validate_action_mark(actions,
1169 action_flags,
1170 attr,
1171 error);
1172 if (ret < 0)
1173 return ret;
1174 action_flags |= MLX5_FLOW_ACTION_MARK;
1175 break;
1176 case RTE_FLOW_ACTION_TYPE_DROP:
1177 ret = mlx5_flow_validate_action_drop(action_flags,
1178 attr,
1179 error);
1180 if (ret < 0)
1181 return ret;
1182 action_flags |= MLX5_FLOW_ACTION_DROP;
1183 break;
1184 case RTE_FLOW_ACTION_TYPE_QUEUE:
1185 ret = mlx5_flow_validate_action_queue(actions,
1186 action_flags, dev,
1187 attr,
1188 error);
1189 if (ret < 0)
1190 return ret;
1191 action_flags |= MLX5_FLOW_ACTION_QUEUE;
1192 break;
1193 case RTE_FLOW_ACTION_TYPE_RSS:
1194 ret = mlx5_flow_validate_action_rss(actions,
1195 action_flags, dev,
1196 attr, item_flags,
1197 error);
1198 if (ret < 0)
1199 return ret;
1200 action_flags |= MLX5_FLOW_ACTION_RSS;
1201 break;
1202 case RTE_FLOW_ACTION_TYPE_COUNT:
1203 ret = mlx5_flow_validate_action_count(dev, attr, error);
1204 if (ret < 0)
1205 return ret;
1206 action_flags |= MLX5_FLOW_ACTION_COUNT;
1207 break;
1208 default:
1209 return rte_flow_error_set(error, ENOTSUP,
1210 RTE_FLOW_ERROR_TYPE_ACTION,
1211 actions,
1212 "action not supported");
1213 }
1214 }
1215 if (!(action_flags & MLX5_FLOW_FATE_ACTIONS))
1216 return rte_flow_error_set(error, EINVAL,
1217 RTE_FLOW_ERROR_TYPE_ACTION, actions,
1218 "no fate action is found");
1219 return 0;
1220 }
1221
1222 /**
1223 * Calculate the required bytes that are needed for the action part of the verbs
1224 * flow.
1225 *
1226 * @param[in] actions
1227 * Pointer to the list of actions.
1228 *
1229 * @return
1230 * The size of the memory needed for all actions.
1231 */
1232 static int
1233 flow_verbs_get_actions_size(const struct rte_flow_action actions[])
1234 {
1235 int size = 0;
1236
1237 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
1238 switch (actions->type) {
1239 case RTE_FLOW_ACTION_TYPE_VOID:
1240 break;
1241 case RTE_FLOW_ACTION_TYPE_FLAG:
1242 size += sizeof(struct ibv_flow_spec_action_tag);
1243 break;
1244 case RTE_FLOW_ACTION_TYPE_MARK:
1245 size += sizeof(struct ibv_flow_spec_action_tag);
1246 break;
1247 case RTE_FLOW_ACTION_TYPE_DROP:
1248 size += sizeof(struct ibv_flow_spec_action_drop);
1249 break;
1250 case RTE_FLOW_ACTION_TYPE_QUEUE:
1251 break;
1252 case RTE_FLOW_ACTION_TYPE_RSS:
1253 break;
1254 case RTE_FLOW_ACTION_TYPE_COUNT:
1255 #if defined(HAVE_IBV_DEVICE_COUNTERS_SET_V42) || \
1256 defined(HAVE_IBV_DEVICE_COUNTERS_SET_V45)
1257 size += sizeof(struct ibv_flow_spec_counter_action);
1258 #endif
1259 break;
1260 default:
1261 break;
1262 }
1263 }
1264 return size;
1265 }
1266
1267 /**
1268 * Calculate the required bytes that are needed for the item part of the verbs
1269 * flow.
1270 *
1271 * @param[in] items
1272 * Pointer to the list of items.
1273 *
1274 * @return
1275 * The size of the memory needed for all items.
1276 */
1277 static int
1278 flow_verbs_get_items_size(const struct rte_flow_item items[])
1279 {
1280 int size = 0;
1281
1282 for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
1283 switch (items->type) {
1284 case RTE_FLOW_ITEM_TYPE_VOID:
1285 break;
1286 case RTE_FLOW_ITEM_TYPE_ETH:
1287 size += sizeof(struct ibv_flow_spec_eth);
1288 break;
1289 case RTE_FLOW_ITEM_TYPE_VLAN:
1290 size += sizeof(struct ibv_flow_spec_eth);
1291 break;
1292 case RTE_FLOW_ITEM_TYPE_IPV4:
1293 size += sizeof(struct ibv_flow_spec_ipv4_ext);
1294 break;
1295 case RTE_FLOW_ITEM_TYPE_IPV6:
1296 size += sizeof(struct ibv_flow_spec_ipv6);
1297 break;
1298 case RTE_FLOW_ITEM_TYPE_UDP:
1299 size += sizeof(struct ibv_flow_spec_tcp_udp);
1300 break;
1301 case RTE_FLOW_ITEM_TYPE_TCP:
1302 size += sizeof(struct ibv_flow_spec_tcp_udp);
1303 break;
1304 case RTE_FLOW_ITEM_TYPE_VXLAN:
1305 size += sizeof(struct ibv_flow_spec_tunnel);
1306 break;
1307 case RTE_FLOW_ITEM_TYPE_VXLAN_GPE:
1308 size += sizeof(struct ibv_flow_spec_tunnel);
1309 break;
1310 #ifdef HAVE_IBV_DEVICE_MPLS_SUPPORT
1311 case RTE_FLOW_ITEM_TYPE_GRE:
1312 size += sizeof(struct ibv_flow_spec_gre);
1313 break;
1314 case RTE_FLOW_ITEM_TYPE_MPLS:
1315 size += sizeof(struct ibv_flow_spec_mpls);
1316 break;
1317 #else
1318 case RTE_FLOW_ITEM_TYPE_GRE:
1319 size += sizeof(struct ibv_flow_spec_tunnel);
1320 break;
1321 #endif
1322 default:
1323 break;
1324 }
1325 }
1326 return size;
1327 }
1328
1329 /**
1330 * Internal preparation function. Allocate mlx5_flow with the required size.
1331 * The required size is calculate based on the actions and items. This function
1332 * also returns the detected actions and items for later use.
1333 *
1334 * @param[in] attr
1335 * Pointer to the flow attributes.
1336 * @param[in] items
1337 * Pointer to the list of items.
1338 * @param[in] actions
1339 * Pointer to the list of actions.
1340 * @param[out] error
1341 * Pointer to the error structure.
1342 *
1343 * @return
1344 * Pointer to mlx5_flow object on success, otherwise NULL and rte_errno
1345 * is set.
1346 */
1347 static struct mlx5_flow *
1348 flow_verbs_prepare(const struct rte_flow_attr *attr __rte_unused,
1349 const struct rte_flow_item items[],
1350 const struct rte_flow_action actions[],
1351 struct rte_flow_error *error)
1352 {
1353 uint32_t size = sizeof(struct mlx5_flow) + sizeof(struct ibv_flow_attr);
1354 struct mlx5_flow *flow;
1355
1356 size += flow_verbs_get_actions_size(actions);
1357 size += flow_verbs_get_items_size(items);
1358 flow = rte_calloc(__func__, 1, size, 0);
1359 if (!flow) {
1360 rte_flow_error_set(error, ENOMEM,
1361 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
1362 "not enough memory to create flow");
1363 return NULL;
1364 }
1365 flow->verbs.attr = (void *)(flow + 1);
1366 flow->verbs.specs =
1367 (uint8_t *)(flow + 1) + sizeof(struct ibv_flow_attr);
1368 return flow;
1369 }
1370
1371 /**
1372 * Fill the flow with verb spec.
1373 *
1374 * @param[in] dev
1375 * Pointer to Ethernet device.
1376 * @param[in, out] dev_flow
1377 * Pointer to the mlx5 flow.
1378 * @param[in] attr
1379 * Pointer to the flow attributes.
1380 * @param[in] items
1381 * Pointer to the list of items.
1382 * @param[in] actions
1383 * Pointer to the list of actions.
1384 * @param[out] error
1385 * Pointer to the error structure.
1386 *
1387 * @return
1388 * 0 on success, else a negative errno value otherwise and rte_errno is set.
1389 */
1390 static int
1391 flow_verbs_translate(struct rte_eth_dev *dev,
1392 struct mlx5_flow *dev_flow,
1393 const struct rte_flow_attr *attr,
1394 const struct rte_flow_item items[],
1395 const struct rte_flow_action actions[],
1396 struct rte_flow_error *error)
1397 {
1398 struct rte_flow *flow = dev_flow->flow;
1399 uint64_t item_flags = 0;
1400 uint64_t action_flags = 0;
1401 uint64_t priority = attr->priority;
1402 uint32_t subpriority = 0;
1403 struct mlx5_priv *priv = dev->data->dev_private;
1404
1405 if (priority == MLX5_FLOW_PRIO_RSVD)
1406 priority = priv->config.flow_prio - 1;
1407 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
1408 int ret;
1409
1410 switch (actions->type) {
1411 case RTE_FLOW_ACTION_TYPE_VOID:
1412 break;
1413 case RTE_FLOW_ACTION_TYPE_FLAG:
1414 flow_verbs_translate_action_flag(dev_flow, actions);
1415 action_flags |= MLX5_FLOW_ACTION_FLAG;
1416 break;
1417 case RTE_FLOW_ACTION_TYPE_MARK:
1418 flow_verbs_translate_action_mark(dev_flow, actions);
1419 action_flags |= MLX5_FLOW_ACTION_MARK;
1420 break;
1421 case RTE_FLOW_ACTION_TYPE_DROP:
1422 flow_verbs_translate_action_drop(dev_flow, actions);
1423 action_flags |= MLX5_FLOW_ACTION_DROP;
1424 break;
1425 case RTE_FLOW_ACTION_TYPE_QUEUE:
1426 flow_verbs_translate_action_queue(dev_flow, actions);
1427 action_flags |= MLX5_FLOW_ACTION_QUEUE;
1428 break;
1429 case RTE_FLOW_ACTION_TYPE_RSS:
1430 flow_verbs_translate_action_rss(dev_flow, actions);
1431 action_flags |= MLX5_FLOW_ACTION_RSS;
1432 break;
1433 case RTE_FLOW_ACTION_TYPE_COUNT:
1434 ret = flow_verbs_translate_action_count(dev_flow,
1435 actions,
1436 dev, error);
1437 if (ret < 0)
1438 return ret;
1439 action_flags |= MLX5_FLOW_ACTION_COUNT;
1440 break;
1441 default:
1442 return rte_flow_error_set(error, ENOTSUP,
1443 RTE_FLOW_ERROR_TYPE_ACTION,
1444 actions,
1445 "action not supported");
1446 }
1447 }
1448 flow->actions = action_flags;
1449 for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
1450 int tunnel = !!(item_flags & MLX5_FLOW_LAYER_TUNNEL);
1451
1452 switch (items->type) {
1453 case RTE_FLOW_ITEM_TYPE_VOID:
1454 break;
1455 case RTE_FLOW_ITEM_TYPE_ETH:
1456 flow_verbs_translate_item_eth(dev_flow, items,
1457 item_flags);
1458 subpriority = MLX5_PRIORITY_MAP_L2;
1459 item_flags |= tunnel ? MLX5_FLOW_LAYER_INNER_L2 :
1460 MLX5_FLOW_LAYER_OUTER_L2;
1461 break;
1462 case RTE_FLOW_ITEM_TYPE_VLAN:
1463 flow_verbs_translate_item_vlan(dev_flow, items,
1464 item_flags);
1465 subpriority = MLX5_PRIORITY_MAP_L2;
1466 item_flags |= tunnel ? (MLX5_FLOW_LAYER_INNER_L2 |
1467 MLX5_FLOW_LAYER_INNER_VLAN) :
1468 (MLX5_FLOW_LAYER_OUTER_L2 |
1469 MLX5_FLOW_LAYER_OUTER_VLAN);
1470 break;
1471 case RTE_FLOW_ITEM_TYPE_IPV4:
1472 flow_verbs_translate_item_ipv4(dev_flow, items,
1473 item_flags);
1474 subpriority = MLX5_PRIORITY_MAP_L3;
1475 dev_flow->verbs.hash_fields |=
1476 mlx5_flow_hashfields_adjust
1477 (dev_flow, tunnel,
1478 MLX5_IPV4_LAYER_TYPES,
1479 MLX5_IPV4_IBV_RX_HASH);
1480 item_flags |= tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV4 :
1481 MLX5_FLOW_LAYER_OUTER_L3_IPV4;
1482 break;
1483 case RTE_FLOW_ITEM_TYPE_IPV6:
1484 flow_verbs_translate_item_ipv6(dev_flow, items,
1485 item_flags);
1486 subpriority = MLX5_PRIORITY_MAP_L3;
1487 dev_flow->verbs.hash_fields |=
1488 mlx5_flow_hashfields_adjust
1489 (dev_flow, tunnel,
1490 MLX5_IPV6_LAYER_TYPES,
1491 MLX5_IPV6_IBV_RX_HASH);
1492 item_flags |= tunnel ? MLX5_FLOW_LAYER_INNER_L3_IPV6 :
1493 MLX5_FLOW_LAYER_OUTER_L3_IPV6;
1494 break;
1495 case RTE_FLOW_ITEM_TYPE_TCP:
1496 flow_verbs_translate_item_tcp(dev_flow, items,
1497 item_flags);
1498 subpriority = MLX5_PRIORITY_MAP_L4;
1499 dev_flow->verbs.hash_fields |=
1500 mlx5_flow_hashfields_adjust
1501 (dev_flow, tunnel, ETH_RSS_TCP,
1502 (IBV_RX_HASH_SRC_PORT_TCP |
1503 IBV_RX_HASH_DST_PORT_TCP));
1504 item_flags |= tunnel ? MLX5_FLOW_LAYER_INNER_L4_TCP :
1505 MLX5_FLOW_LAYER_OUTER_L4_TCP;
1506 break;
1507 case RTE_FLOW_ITEM_TYPE_UDP:
1508 flow_verbs_translate_item_udp(dev_flow, items,
1509 item_flags);
1510 subpriority = MLX5_PRIORITY_MAP_L4;
1511 dev_flow->verbs.hash_fields |=
1512 mlx5_flow_hashfields_adjust
1513 (dev_flow, tunnel, ETH_RSS_UDP,
1514 (IBV_RX_HASH_SRC_PORT_UDP |
1515 IBV_RX_HASH_DST_PORT_UDP));
1516 item_flags |= tunnel ? MLX5_FLOW_LAYER_INNER_L4_UDP :
1517 MLX5_FLOW_LAYER_OUTER_L4_UDP;
1518 break;
1519 case RTE_FLOW_ITEM_TYPE_VXLAN:
1520 flow_verbs_translate_item_vxlan(dev_flow, items,
1521 item_flags);
1522 subpriority = MLX5_PRIORITY_MAP_L2;
1523 item_flags |= MLX5_FLOW_LAYER_VXLAN;
1524 break;
1525 case RTE_FLOW_ITEM_TYPE_VXLAN_GPE:
1526 flow_verbs_translate_item_vxlan_gpe(dev_flow, items,
1527 item_flags);
1528 subpriority = MLX5_PRIORITY_MAP_L2;
1529 item_flags |= MLX5_FLOW_LAYER_VXLAN_GPE;
1530 break;
1531 case RTE_FLOW_ITEM_TYPE_GRE:
1532 flow_verbs_translate_item_gre(dev_flow, items,
1533 item_flags);
1534 subpriority = MLX5_PRIORITY_MAP_L2;
1535 item_flags |= MLX5_FLOW_LAYER_GRE;
1536 break;
1537 case RTE_FLOW_ITEM_TYPE_MPLS:
1538 flow_verbs_translate_item_mpls(dev_flow, items,
1539 item_flags);
1540 subpriority = MLX5_PRIORITY_MAP_L2;
1541 item_flags |= MLX5_FLOW_LAYER_MPLS;
1542 break;
1543 default:
1544 return rte_flow_error_set(error, ENOTSUP,
1545 RTE_FLOW_ERROR_TYPE_ITEM,
1546 NULL,
1547 "item not supported");
1548 }
1549 }
1550 dev_flow->layers = item_flags;
1551 dev_flow->verbs.attr->priority =
1552 mlx5_flow_adjust_priority(dev, priority, subpriority);
1553 dev_flow->verbs.attr->port = (uint8_t)priv->ibv_port;
1554 return 0;
1555 }
1556
1557 /**
1558 * Remove the flow from the NIC but keeps it in memory.
1559 *
1560 * @param[in] dev
1561 * Pointer to the Ethernet device structure.
1562 * @param[in, out] flow
1563 * Pointer to flow structure.
1564 */
1565 static void
1566 flow_verbs_remove(struct rte_eth_dev *dev, struct rte_flow *flow)
1567 {
1568 struct mlx5_flow_verbs *verbs;
1569 struct mlx5_flow *dev_flow;
1570
1571 if (!flow)
1572 return;
1573 LIST_FOREACH(dev_flow, &flow->dev_flows, next) {
1574 verbs = &dev_flow->verbs;
1575 if (verbs->flow) {
1576 claim_zero(mlx5_glue->destroy_flow(verbs->flow));
1577 verbs->flow = NULL;
1578 }
1579 if (verbs->hrxq) {
1580 if (flow->actions & MLX5_FLOW_ACTION_DROP)
1581 mlx5_hrxq_drop_release(dev);
1582 else
1583 mlx5_hrxq_release(dev, verbs->hrxq);
1584 verbs->hrxq = NULL;
1585 }
1586 }
1587 }
1588
1589 /**
1590 * Remove the flow from the NIC and the memory.
1591 *
1592 * @param[in] dev
1593 * Pointer to the Ethernet device structure.
1594 * @param[in, out] flow
1595 * Pointer to flow structure.
1596 */
1597 static void
1598 flow_verbs_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
1599 {
1600 struct mlx5_flow *dev_flow;
1601
1602 if (!flow)
1603 return;
1604 flow_verbs_remove(dev, flow);
1605 while (!LIST_EMPTY(&flow->dev_flows)) {
1606 dev_flow = LIST_FIRST(&flow->dev_flows);
1607 LIST_REMOVE(dev_flow, next);
1608 rte_free(dev_flow);
1609 }
1610 if (flow->counter) {
1611 flow_verbs_counter_release(flow->counter);
1612 flow->counter = NULL;
1613 }
1614 }
1615
1616 /**
1617 * Apply the flow to the NIC.
1618 *
1619 * @param[in] dev
1620 * Pointer to the Ethernet device structure.
1621 * @param[in, out] flow
1622 * Pointer to flow structure.
1623 * @param[out] error
1624 * Pointer to error structure.
1625 *
1626 * @return
1627 * 0 on success, a negative errno value otherwise and rte_errno is set.
1628 */
1629 static int
1630 flow_verbs_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
1631 struct rte_flow_error *error)
1632 {
1633 struct mlx5_flow_verbs *verbs;
1634 struct mlx5_flow *dev_flow;
1635 int err;
1636
1637 LIST_FOREACH(dev_flow, &flow->dev_flows, next) {
1638 verbs = &dev_flow->verbs;
1639 if (flow->actions & MLX5_FLOW_ACTION_DROP) {
1640 verbs->hrxq = mlx5_hrxq_drop_new(dev);
1641 if (!verbs->hrxq) {
1642 rte_flow_error_set
1643 (error, errno,
1644 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
1645 "cannot get drop hash queue");
1646 goto error;
1647 }
1648 } else {
1649 struct mlx5_hrxq *hrxq;
1650
1651 hrxq = mlx5_hrxq_get(dev, flow->key,
1652 MLX5_RSS_HASH_KEY_LEN,
1653 verbs->hash_fields,
1654 (*flow->queue),
1655 flow->rss.queue_num);
1656 if (!hrxq)
1657 hrxq = mlx5_hrxq_new(dev, flow->key,
1658 MLX5_RSS_HASH_KEY_LEN,
1659 verbs->hash_fields,
1660 (*flow->queue),
1661 flow->rss.queue_num,
1662 !!(dev_flow->layers &
1663 MLX5_FLOW_LAYER_TUNNEL));
1664 if (!hrxq) {
1665 rte_flow_error_set
1666 (error, rte_errno,
1667 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
1668 "cannot get hash queue");
1669 goto error;
1670 }
1671 verbs->hrxq = hrxq;
1672 }
1673 verbs->flow = mlx5_glue->create_flow(verbs->hrxq->qp,
1674 verbs->attr);
1675 if (!verbs->flow) {
1676 rte_flow_error_set(error, errno,
1677 RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
1678 NULL,
1679 "hardware refuses to create flow");
1680 goto error;
1681 }
1682 }
1683 return 0;
1684 error:
1685 err = rte_errno; /* Save rte_errno before cleanup. */
1686 LIST_FOREACH(dev_flow, &flow->dev_flows, next) {
1687 verbs = &dev_flow->verbs;
1688 if (verbs->hrxq) {
1689 if (flow->actions & MLX5_FLOW_ACTION_DROP)
1690 mlx5_hrxq_drop_release(dev);
1691 else
1692 mlx5_hrxq_release(dev, verbs->hrxq);
1693 verbs->hrxq = NULL;
1694 }
1695 }
1696 rte_errno = err; /* Restore rte_errno. */
1697 return -rte_errno;
1698 }
1699
1700 /**
1701 * Query a flow.
1702 *
1703 * @see rte_flow_query()
1704 * @see rte_flow_ops
1705 */
1706 static int
1707 flow_verbs_query(struct rte_eth_dev *dev,
1708 struct rte_flow *flow,
1709 const struct rte_flow_action *actions,
1710 void *data,
1711 struct rte_flow_error *error)
1712 {
1713 int ret = -EINVAL;
1714
1715 for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
1716 switch (actions->type) {
1717 case RTE_FLOW_ACTION_TYPE_VOID:
1718 break;
1719 case RTE_FLOW_ACTION_TYPE_COUNT:
1720 ret = flow_verbs_counter_query(dev, flow, data, error);
1721 break;
1722 default:
1723 return rte_flow_error_set(error, ENOTSUP,
1724 RTE_FLOW_ERROR_TYPE_ACTION,
1725 actions,
1726 "action not supported");
1727 }
1728 }
1729 return ret;
1730 }
1731
1732 const struct mlx5_flow_driver_ops mlx5_flow_verbs_drv_ops = {
1733 .validate = flow_verbs_validate,
1734 .prepare = flow_verbs_prepare,
1735 .translate = flow_verbs_translate,
1736 .apply = flow_verbs_apply,
1737 .remove = flow_verbs_remove,
1738 .destroy = flow_verbs_destroy,
1739 .query = flow_verbs_query,
1740 };
Ceph