xref: /dpdk/drivers/net/bnxt/bnxt_flow.c (revision 68a03efeed657e6e05f281479b33b51102797e15)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright(c) 2014-2021 Broadcom
3  * All rights reserved.
4  */
5 
6 #include <sys/queue.h>
7 
8 #include <rte_log.h>
9 #include <rte_malloc.h>
10 #include <rte_flow.h>
11 #include <rte_flow_driver.h>
12 #include <rte_tailq.h>
13 #include <rte_alarm.h>
14 #include <rte_cycles.h>
15 
16 #include "bnxt.h"
17 #include "bnxt_filter.h"
18 #include "bnxt_hwrm.h"
19 #include "bnxt_ring.h"
20 #include "bnxt_rxq.h"
21 #include "bnxt_rxr.h"
22 #include "bnxt_vnic.h"
23 #include "hsi_struct_def_dpdk.h"
24 
25 static int
26 bnxt_flow_args_validate(const struct rte_flow_attr *attr,
27 			const struct rte_flow_item pattern[],
28 			const struct rte_flow_action actions[],
29 			struct rte_flow_error *error)
30 {
31 	if (!pattern) {
32 		rte_flow_error_set(error,
33 				   EINVAL,
34 				   RTE_FLOW_ERROR_TYPE_ITEM_NUM,
35 				   NULL,
36 				   "NULL pattern.");
37 		return -rte_errno;
38 	}
39 
40 	if (!actions) {
41 		rte_flow_error_set(error,
42 				   EINVAL,
43 				   RTE_FLOW_ERROR_TYPE_ACTION_NUM,
44 				   NULL,
45 				   "NULL action.");
46 		return -rte_errno;
47 	}
48 
49 	if (!attr) {
50 		rte_flow_error_set(error,
51 				   EINVAL,
52 				   RTE_FLOW_ERROR_TYPE_ATTR,
53 				   NULL,
54 				   "NULL attribute.");
55 		return -rte_errno;
56 	}
57 
58 	return 0;
59 }
60 
61 static const struct rte_flow_item *
62 bnxt_flow_non_void_item(const struct rte_flow_item *cur)
63 {
64 	while (1) {
65 		if (cur->type != RTE_FLOW_ITEM_TYPE_VOID)
66 			return cur;
67 		cur++;
68 	}
69 }
70 
71 static const struct rte_flow_action *
72 bnxt_flow_non_void_action(const struct rte_flow_action *cur)
73 {
74 	while (1) {
75 		if (cur->type != RTE_FLOW_ACTION_TYPE_VOID)
76 			return cur;
77 		cur++;
78 	}
79 }
80 
81 static int
82 bnxt_filter_type_check(const struct rte_flow_item pattern[],
83 		       struct rte_flow_error *error)
84 {
85 	const struct rte_flow_item *item =
86 		bnxt_flow_non_void_item(pattern);
87 	int use_ntuple = 1;
88 	bool has_vlan = 0;
89 
90 	while (item->type != RTE_FLOW_ITEM_TYPE_END) {
91 		switch (item->type) {
92 		case RTE_FLOW_ITEM_TYPE_ANY:
93 		case RTE_FLOW_ITEM_TYPE_ETH:
94 			use_ntuple = 0;
95 			break;
96 		case RTE_FLOW_ITEM_TYPE_VLAN:
97 			use_ntuple = 0;
98 			has_vlan = 1;
99 			break;
100 		case RTE_FLOW_ITEM_TYPE_IPV4:
101 		case RTE_FLOW_ITEM_TYPE_IPV6:
102 		case RTE_FLOW_ITEM_TYPE_TCP:
103 		case RTE_FLOW_ITEM_TYPE_UDP:
104 			/* FALLTHROUGH */
105 			/* need ntuple match, reset exact match */
106 			use_ntuple |= 1;
107 			break;
108 		default:
109 			PMD_DRV_LOG(DEBUG, "Unknown Flow type\n");
110 			use_ntuple |= 0;
111 		}
112 		item++;
113 	}
114 
115 	if (has_vlan && use_ntuple) {
116 		PMD_DRV_LOG(ERR,
117 			    "VLAN flow cannot use NTUPLE filter\n");
118 		rte_flow_error_set(error, EINVAL,
119 				   RTE_FLOW_ERROR_TYPE_ITEM,
120 				   item,
121 				   "Cannot use VLAN with NTUPLE");
122 		return -rte_errno;
123 	}
124 
125 	return use_ntuple;
126 }
127 
128 static int
129 bnxt_validate_and_parse_flow_type(struct bnxt *bp,
130 				  const struct rte_flow_attr *attr,
131 				  const struct rte_flow_item pattern[],
132 				  struct rte_flow_error *error,
133 				  struct bnxt_filter_info *filter)
134 {
135 	const struct rte_flow_item *item = bnxt_flow_non_void_item(pattern);
136 	const struct rte_flow_item_vlan *vlan_spec, *vlan_mask;
137 	const struct rte_flow_item_ipv4 *ipv4_spec, *ipv4_mask;
138 	const struct rte_flow_item_ipv6 *ipv6_spec, *ipv6_mask;
139 	const struct rte_flow_item_tcp *tcp_spec, *tcp_mask;
140 	const struct rte_flow_item_udp *udp_spec, *udp_mask;
141 	const struct rte_flow_item_eth *eth_spec, *eth_mask;
142 	const struct rte_ether_addr *dst, *src;
143 	const struct rte_flow_item_nvgre *nvgre_spec;
144 	const struct rte_flow_item_nvgre *nvgre_mask;
145 	const struct rte_flow_item_gre *gre_spec;
146 	const struct rte_flow_item_gre *gre_mask;
147 	const struct rte_flow_item_vxlan *vxlan_spec;
148 	const struct rte_flow_item_vxlan *vxlan_mask;
149 	uint8_t vni_mask[] = {0xFF, 0xFF, 0xFF};
150 	uint8_t tni_mask[] = {0xFF, 0xFF, 0xFF};
151 	const struct rte_flow_item_vf *vf_spec;
152 	uint32_t tenant_id_be = 0, valid_flags = 0;
153 	bool vni_masked = 0;
154 	bool tni_masked = 0;
155 	uint32_t en_ethertype;
156 	uint8_t inner = 0;
157 	uint32_t vf = 0;
158 	uint32_t en = 0;
159 	int use_ntuple;
160 	int dflt_vnic;
161 
162 	use_ntuple = bnxt_filter_type_check(pattern, error);
163 	if (use_ntuple < 0)
164 		return use_ntuple;
165 	PMD_DRV_LOG(DEBUG, "Use NTUPLE %d\n", use_ntuple);
166 
167 	filter->filter_type = use_ntuple ?
168 		HWRM_CFA_NTUPLE_FILTER : HWRM_CFA_L2_FILTER;
169 	en_ethertype = use_ntuple ?
170 		NTUPLE_FLTR_ALLOC_INPUT_EN_ETHERTYPE :
171 		EM_FLOW_ALLOC_INPUT_EN_ETHERTYPE;
172 
173 	while (item->type != RTE_FLOW_ITEM_TYPE_END) {
174 		if (item->last) {
175 			/* last or range is NOT supported as match criteria */
176 			rte_flow_error_set(error, EINVAL,
177 					   RTE_FLOW_ERROR_TYPE_ITEM,
178 					   item,
179 					   "No support for range");
180 			return -rte_errno;
181 		}
182 
183 		switch (item->type) {
184 		case RTE_FLOW_ITEM_TYPE_ANY:
185 			inner =
186 			((const struct rte_flow_item_any *)item->spec)->num > 3;
187 			if (inner)
188 				PMD_DRV_LOG(DEBUG, "Parse inner header\n");
189 			break;
190 		case RTE_FLOW_ITEM_TYPE_ETH:
191 			if (!item->spec)
192 				break;
193 
194 			eth_spec = item->spec;
195 
196 			if (item->mask)
197 				eth_mask = item->mask;
198 			else
199 				eth_mask = &rte_flow_item_eth_mask;
200 
201 			/* Source MAC address mask cannot be partially set.
202 			 * Should be All 0's or all 1's.
203 			 * Destination MAC address mask must not be partially
204 			 * set. Should be all 1's or all 0's.
205 			 */
206 			if ((!rte_is_zero_ether_addr(&eth_mask->src) &&
207 			     !rte_is_broadcast_ether_addr(&eth_mask->src)) ||
208 			    (!rte_is_zero_ether_addr(&eth_mask->dst) &&
209 			     !rte_is_broadcast_ether_addr(&eth_mask->dst))) {
210 				rte_flow_error_set(error,
211 						   EINVAL,
212 						   RTE_FLOW_ERROR_TYPE_ITEM,
213 						   item,
214 						   "MAC_addr mask not valid");
215 				return -rte_errno;
216 			}
217 
218 			/* Mask is not allowed. Only exact matches are */
219 			if (eth_mask->type &&
220 			    eth_mask->type != RTE_BE16(0xffff)) {
221 				rte_flow_error_set(error, EINVAL,
222 						   RTE_FLOW_ERROR_TYPE_ITEM,
223 						   item,
224 						   "ethertype mask not valid");
225 				return -rte_errno;
226 			}
227 
228 			if (rte_is_broadcast_ether_addr(&eth_mask->dst)) {
229 				dst = &eth_spec->dst;
230 				if (!rte_is_valid_assigned_ether_addr(dst)) {
231 					rte_flow_error_set(error,
232 							   EINVAL,
233 							   RTE_FLOW_ERROR_TYPE_ITEM,
234 							   item,
235 							   "DMAC is invalid");
236 					PMD_DRV_LOG(ERR,
237 						    "DMAC is invalid!\n");
238 					return -rte_errno;
239 				}
240 				rte_memcpy(filter->dst_macaddr,
241 					   &eth_spec->dst, RTE_ETHER_ADDR_LEN);
242 				en |= use_ntuple ?
243 					NTUPLE_FLTR_ALLOC_INPUT_EN_DST_MACADDR :
244 					EM_FLOW_ALLOC_INPUT_EN_DST_MACADDR;
245 				valid_flags |= inner ?
246 					BNXT_FLOW_L2_INNER_DST_VALID_FLAG :
247 					BNXT_FLOW_L2_DST_VALID_FLAG;
248 				filter->priority = attr->priority;
249 				PMD_DRV_LOG(DEBUG,
250 					    "Creating a priority flow\n");
251 			}
252 			if (rte_is_broadcast_ether_addr(&eth_mask->src)) {
253 				src = &eth_spec->src;
254 				if (!rte_is_valid_assigned_ether_addr(src)) {
255 					rte_flow_error_set(error,
256 							   EINVAL,
257 							   RTE_FLOW_ERROR_TYPE_ITEM,
258 							   item,
259 							   "SMAC is invalid");
260 					PMD_DRV_LOG(ERR,
261 						    "SMAC is invalid!\n");
262 					return -rte_errno;
263 				}
264 				rte_memcpy(filter->src_macaddr,
265 					   &eth_spec->src, RTE_ETHER_ADDR_LEN);
266 				en |= use_ntuple ?
267 					NTUPLE_FLTR_ALLOC_INPUT_EN_SRC_MACADDR :
268 					EM_FLOW_ALLOC_INPUT_EN_SRC_MACADDR;
269 				valid_flags |= inner ?
270 					BNXT_FLOW_L2_INNER_SRC_VALID_FLAG :
271 					BNXT_FLOW_L2_SRC_VALID_FLAG;
272 			} /*
273 			   * else {
274 			   *  PMD_DRV_LOG(ERR, "Handle this condition\n");
275 			   * }
276 			   */
277 			if (eth_mask->type) {
278 				filter->ethertype =
279 					rte_be_to_cpu_16(eth_spec->type);
280 				en |= en_ethertype;
281 			}
282 			if (inner)
283 				valid_flags |= BNXT_FLOW_PARSE_INNER_FLAG;
284 
285 			break;
286 		case RTE_FLOW_ITEM_TYPE_VLAN:
287 			vlan_spec = item->spec;
288 
289 			if (item->mask)
290 				vlan_mask = item->mask;
291 			else
292 				vlan_mask = &rte_flow_item_vlan_mask;
293 
294 			if (en & en_ethertype) {
295 				rte_flow_error_set(error, EINVAL,
296 						   RTE_FLOW_ERROR_TYPE_ITEM,
297 						   item,
298 						   "VLAN TPID matching is not"
299 						   " supported");
300 				return -rte_errno;
301 			}
302 			if (vlan_mask->tci &&
303 			    vlan_mask->tci == RTE_BE16(0x0fff)) {
304 				/* Only the VLAN ID can be matched. */
305 				filter->l2_ovlan =
306 					rte_be_to_cpu_16(vlan_spec->tci &
307 							 RTE_BE16(0x0fff));
308 				en |= EM_FLOW_ALLOC_INPUT_EN_OVLAN_VID;
309 			} else {
310 				rte_flow_error_set(error,
311 						   EINVAL,
312 						   RTE_FLOW_ERROR_TYPE_ITEM,
313 						   item,
314 						   "VLAN mask is invalid");
315 				return -rte_errno;
316 			}
317 			if (vlan_mask->inner_type &&
318 			    vlan_mask->inner_type != RTE_BE16(0xffff)) {
319 				rte_flow_error_set(error, EINVAL,
320 						   RTE_FLOW_ERROR_TYPE_ITEM,
321 						   item,
322 						   "inner ethertype mask not"
323 						   " valid");
324 				return -rte_errno;
325 			}
326 			if (vlan_mask->inner_type) {
327 				filter->ethertype =
328 					rte_be_to_cpu_16(vlan_spec->inner_type);
329 				en |= en_ethertype;
330 			}
331 
332 			break;
333 		case RTE_FLOW_ITEM_TYPE_IPV4:
334 			/* If mask is not involved, we could use EM filters. */
335 			ipv4_spec = item->spec;
336 
337 			if (!item->spec)
338 				break;
339 
340 			if (item->mask)
341 				ipv4_mask = item->mask;
342 			else
343 				ipv4_mask = &rte_flow_item_ipv4_mask;
344 
345 			/* Only IP DST and SRC fields are maskable. */
346 			if (ipv4_mask->hdr.version_ihl ||
347 			    ipv4_mask->hdr.type_of_service ||
348 			    ipv4_mask->hdr.total_length ||
349 			    ipv4_mask->hdr.packet_id ||
350 			    ipv4_mask->hdr.fragment_offset ||
351 			    ipv4_mask->hdr.time_to_live ||
352 			    ipv4_mask->hdr.next_proto_id ||
353 			    ipv4_mask->hdr.hdr_checksum) {
354 				rte_flow_error_set(error,
355 						   EINVAL,
356 						   RTE_FLOW_ERROR_TYPE_ITEM,
357 						   item,
358 						   "Invalid IPv4 mask.");
359 				return -rte_errno;
360 			}
361 
362 			filter->dst_ipaddr[0] = ipv4_spec->hdr.dst_addr;
363 			filter->src_ipaddr[0] = ipv4_spec->hdr.src_addr;
364 
365 			if (use_ntuple)
366 				en |= NTUPLE_FLTR_ALLOC_INPUT_EN_SRC_IPADDR |
367 					NTUPLE_FLTR_ALLOC_INPUT_EN_DST_IPADDR;
368 			else
369 				en |= EM_FLOW_ALLOC_INPUT_EN_SRC_IPADDR |
370 					EM_FLOW_ALLOC_INPUT_EN_DST_IPADDR;
371 
372 			if (ipv4_mask->hdr.src_addr) {
373 				filter->src_ipaddr_mask[0] =
374 					ipv4_mask->hdr.src_addr;
375 				en |= !use_ntuple ? 0 :
376 				     NTUPLE_FLTR_ALLOC_INPUT_EN_SRC_IPADDR_MASK;
377 			}
378 
379 			if (ipv4_mask->hdr.dst_addr) {
380 				filter->dst_ipaddr_mask[0] =
381 					ipv4_mask->hdr.dst_addr;
382 				en |= !use_ntuple ? 0 :
383 				     NTUPLE_FLTR_ALLOC_INPUT_EN_DST_IPADDR_MASK;
384 			}
385 
386 			filter->ip_addr_type = use_ntuple ?
387 			 HWRM_CFA_NTUPLE_FILTER_ALLOC_INPUT_IP_ADDR_TYPE_IPV4 :
388 			 HWRM_CFA_EM_FLOW_ALLOC_INPUT_IP_ADDR_TYPE_IPV4;
389 
390 			if (ipv4_spec->hdr.next_proto_id) {
391 				filter->ip_protocol =
392 					ipv4_spec->hdr.next_proto_id;
393 				if (use_ntuple)
394 					en |= NTUPLE_FLTR_ALLOC_IN_EN_IP_PROTO;
395 				else
396 					en |= EM_FLOW_ALLOC_INPUT_EN_IP_PROTO;
397 			}
398 			break;
399 		case RTE_FLOW_ITEM_TYPE_IPV6:
400 			ipv6_spec = item->spec;
401 
402 			if (!item->spec)
403 				break;
404 
405 			if (item->mask)
406 				ipv6_mask = item->mask;
407 			else
408 				ipv6_mask = &rte_flow_item_ipv6_mask;
409 
410 			/* Only IP DST and SRC fields are maskable. */
411 			if (ipv6_mask->hdr.vtc_flow ||
412 			    ipv6_mask->hdr.payload_len ||
413 			    ipv6_mask->hdr.proto ||
414 			    ipv6_mask->hdr.hop_limits) {
415 				rte_flow_error_set(error,
416 						   EINVAL,
417 						   RTE_FLOW_ERROR_TYPE_ITEM,
418 						   item,
419 						   "Invalid IPv6 mask.");
420 				return -rte_errno;
421 			}
422 
423 			if (use_ntuple)
424 				en |= NTUPLE_FLTR_ALLOC_INPUT_EN_SRC_IPADDR |
425 					NTUPLE_FLTR_ALLOC_INPUT_EN_DST_IPADDR;
426 			else
427 				en |= EM_FLOW_ALLOC_INPUT_EN_SRC_IPADDR |
428 					EM_FLOW_ALLOC_INPUT_EN_DST_IPADDR;
429 
430 			rte_memcpy(filter->src_ipaddr,
431 				   ipv6_spec->hdr.src_addr, 16);
432 			rte_memcpy(filter->dst_ipaddr,
433 				   ipv6_spec->hdr.dst_addr, 16);
434 
435 			if (!bnxt_check_zero_bytes(ipv6_mask->hdr.src_addr,
436 						   16)) {
437 				rte_memcpy(filter->src_ipaddr_mask,
438 					   ipv6_mask->hdr.src_addr, 16);
439 				en |= !use_ntuple ? 0 :
440 				    NTUPLE_FLTR_ALLOC_INPUT_EN_SRC_IPADDR_MASK;
441 			}
442 
443 			if (!bnxt_check_zero_bytes(ipv6_mask->hdr.dst_addr,
444 						   16)) {
445 				rte_memcpy(filter->dst_ipaddr_mask,
446 					   ipv6_mask->hdr.dst_addr, 16);
447 				en |= !use_ntuple ? 0 :
448 				     NTUPLE_FLTR_ALLOC_INPUT_EN_DST_IPADDR_MASK;
449 			}
450 
451 			filter->ip_addr_type = use_ntuple ?
452 				NTUPLE_FLTR_ALLOC_INPUT_IP_ADDR_TYPE_IPV6 :
453 				EM_FLOW_ALLOC_INPUT_IP_ADDR_TYPE_IPV6;
454 			break;
455 		case RTE_FLOW_ITEM_TYPE_TCP:
456 			tcp_spec = item->spec;
457 
458 			if (!item->spec)
459 				break;
460 
461 			if (item->mask)
462 				tcp_mask = item->mask;
463 			else
464 				tcp_mask = &rte_flow_item_tcp_mask;
465 
466 			/* Check TCP mask. Only DST & SRC ports are maskable */
467 			if (tcp_mask->hdr.sent_seq ||
468 			    tcp_mask->hdr.recv_ack ||
469 			    tcp_mask->hdr.data_off ||
470 			    tcp_mask->hdr.tcp_flags ||
471 			    tcp_mask->hdr.rx_win ||
472 			    tcp_mask->hdr.cksum ||
473 			    tcp_mask->hdr.tcp_urp) {
474 				rte_flow_error_set(error,
475 						   EINVAL,
476 						   RTE_FLOW_ERROR_TYPE_ITEM,
477 						   item,
478 						   "Invalid TCP mask");
479 				return -rte_errno;
480 			}
481 
482 			filter->src_port = tcp_spec->hdr.src_port;
483 			filter->dst_port = tcp_spec->hdr.dst_port;
484 
485 			if (use_ntuple)
486 				en |= NTUPLE_FLTR_ALLOC_INPUT_EN_SRC_PORT |
487 					NTUPLE_FLTR_ALLOC_INPUT_EN_DST_PORT;
488 			else
489 				en |= EM_FLOW_ALLOC_INPUT_EN_SRC_PORT |
490 					EM_FLOW_ALLOC_INPUT_EN_DST_PORT;
491 
492 			if (tcp_mask->hdr.dst_port) {
493 				filter->dst_port_mask = tcp_mask->hdr.dst_port;
494 				en |= !use_ntuple ? 0 :
495 				  NTUPLE_FLTR_ALLOC_INPUT_EN_DST_PORT_MASK;
496 			}
497 
498 			if (tcp_mask->hdr.src_port) {
499 				filter->src_port_mask = tcp_mask->hdr.src_port;
500 				en |= !use_ntuple ? 0 :
501 				  NTUPLE_FLTR_ALLOC_INPUT_EN_SRC_PORT_MASK;
502 			}
503 			break;
504 		case RTE_FLOW_ITEM_TYPE_UDP:
505 			udp_spec = item->spec;
506 
507 			if (!item->spec)
508 				break;
509 
510 			if (item->mask)
511 				udp_mask = item->mask;
512 			else
513 				udp_mask = &rte_flow_item_udp_mask;
514 
515 			if (udp_mask->hdr.dgram_len ||
516 			    udp_mask->hdr.dgram_cksum) {
517 				rte_flow_error_set(error,
518 						   EINVAL,
519 						   RTE_FLOW_ERROR_TYPE_ITEM,
520 						   item,
521 						   "Invalid UDP mask");
522 				return -rte_errno;
523 			}
524 
525 			filter->src_port = udp_spec->hdr.src_port;
526 			filter->dst_port = udp_spec->hdr.dst_port;
527 
528 			if (use_ntuple)
529 				en |= NTUPLE_FLTR_ALLOC_INPUT_EN_SRC_PORT |
530 					NTUPLE_FLTR_ALLOC_INPUT_EN_DST_PORT;
531 			else
532 				en |= EM_FLOW_ALLOC_INPUT_EN_SRC_PORT |
533 					EM_FLOW_ALLOC_INPUT_EN_DST_PORT;
534 
535 			if (udp_mask->hdr.dst_port) {
536 				filter->dst_port_mask = udp_mask->hdr.dst_port;
537 				en |= !use_ntuple ? 0 :
538 				  NTUPLE_FLTR_ALLOC_INPUT_EN_DST_PORT_MASK;
539 			}
540 
541 			if (udp_mask->hdr.src_port) {
542 				filter->src_port_mask = udp_mask->hdr.src_port;
543 				en |= !use_ntuple ? 0 :
544 				  NTUPLE_FLTR_ALLOC_INPUT_EN_SRC_PORT_MASK;
545 			}
546 			break;
547 		case RTE_FLOW_ITEM_TYPE_VXLAN:
548 			vxlan_spec = item->spec;
549 			vxlan_mask = item->mask;
550 			/* Check if VXLAN item is used to describe protocol.
551 			 * If yes, both spec and mask should be NULL.
552 			 * If no, both spec and mask shouldn't be NULL.
553 			 */
554 			if ((!vxlan_spec && vxlan_mask) ||
555 			    (vxlan_spec && !vxlan_mask)) {
556 				rte_flow_error_set(error,
557 						   EINVAL,
558 						   RTE_FLOW_ERROR_TYPE_ITEM,
559 						   item,
560 						   "Invalid VXLAN item");
561 				return -rte_errno;
562 			}
563 
564 			if (!vxlan_spec && !vxlan_mask) {
565 				filter->tunnel_type =
566 				CFA_NTUPLE_FILTER_ALLOC_REQ_TUNNEL_TYPE_VXLAN;
567 				break;
568 			}
569 
570 			if (vxlan_spec->rsvd1 || vxlan_spec->rsvd0[0] ||
571 			    vxlan_spec->rsvd0[1] || vxlan_spec->rsvd0[2] ||
572 			    vxlan_spec->flags != 0x8) {
573 				rte_flow_error_set(error,
574 						   EINVAL,
575 						   RTE_FLOW_ERROR_TYPE_ITEM,
576 						   item,
577 						   "Invalid VXLAN item");
578 				return -rte_errno;
579 			}
580 
581 			/* Check if VNI is masked. */
582 			if (vxlan_mask != NULL) {
583 				vni_masked =
584 					!!memcmp(vxlan_mask->vni, vni_mask,
585 						 RTE_DIM(vni_mask));
586 				if (vni_masked) {
587 					rte_flow_error_set
588 						(error,
589 						 EINVAL,
590 						 RTE_FLOW_ERROR_TYPE_ITEM,
591 						 item,
592 						 "Invalid VNI mask");
593 					return -rte_errno;
594 				}
595 
596 				rte_memcpy(((uint8_t *)&tenant_id_be + 1),
597 					   vxlan_spec->vni, 3);
598 				filter->vni =
599 					rte_be_to_cpu_32(tenant_id_be);
600 				filter->tunnel_type =
601 				 CFA_NTUPLE_FILTER_ALLOC_REQ_TUNNEL_TYPE_VXLAN;
602 			}
603 			break;
604 		case RTE_FLOW_ITEM_TYPE_NVGRE:
605 			nvgre_spec = item->spec;
606 			nvgre_mask = item->mask;
607 			/* Check if NVGRE item is used to describe protocol.
608 			 * If yes, both spec and mask should be NULL.
609 			 * If no, both spec and mask shouldn't be NULL.
610 			 */
611 			if ((!nvgre_spec && nvgre_mask) ||
612 			    (nvgre_spec && !nvgre_mask)) {
613 				rte_flow_error_set(error,
614 						   EINVAL,
615 						   RTE_FLOW_ERROR_TYPE_ITEM,
616 						   item,
617 						   "Invalid NVGRE item");
618 				return -rte_errno;
619 			}
620 
621 			if (!nvgre_spec && !nvgre_mask) {
622 				filter->tunnel_type =
623 				CFA_NTUPLE_FILTER_ALLOC_REQ_TUNNEL_TYPE_NVGRE;
624 				break;
625 			}
626 
627 			if (nvgre_spec->c_k_s_rsvd0_ver != 0x2000 ||
628 			    nvgre_spec->protocol != 0x6558) {
629 				rte_flow_error_set(error,
630 						   EINVAL,
631 						   RTE_FLOW_ERROR_TYPE_ITEM,
632 						   item,
633 						   "Invalid NVGRE item");
634 				return -rte_errno;
635 			}
636 
637 			if (nvgre_spec && nvgre_mask) {
638 				tni_masked =
639 					!!memcmp(nvgre_mask->tni, tni_mask,
640 						 RTE_DIM(tni_mask));
641 				if (tni_masked) {
642 					rte_flow_error_set
643 						(error,
644 						 EINVAL,
645 						 RTE_FLOW_ERROR_TYPE_ITEM,
646 						 item,
647 						 "Invalid TNI mask");
648 					return -rte_errno;
649 				}
650 				rte_memcpy(((uint8_t *)&tenant_id_be + 1),
651 					   nvgre_spec->tni, 3);
652 				filter->vni =
653 					rte_be_to_cpu_32(tenant_id_be);
654 				filter->tunnel_type =
655 				 CFA_NTUPLE_FILTER_ALLOC_REQ_TUNNEL_TYPE_NVGRE;
656 			}
657 			break;
658 
659 		case RTE_FLOW_ITEM_TYPE_GRE:
660 			gre_spec = (const struct rte_flow_item_gre *)item->spec;
661 			gre_mask = (const struct rte_flow_item_gre *)item->mask;
662 
663 			/*
664 			 *Check if GRE item is used to describe protocol.
665 			 * If yes, both spec and mask should be NULL.
666 			 * If no, both spec and mask shouldn't be NULL.
667 			 */
668 			if (!!gre_spec ^ !!gre_mask) {
669 				rte_flow_error_set(error, EINVAL,
670 						   RTE_FLOW_ERROR_TYPE_ITEM,
671 						   item,
672 						   "Invalid GRE item");
673 				return -rte_errno;
674 			}
675 
676 			if (!gre_spec && !gre_mask) {
677 				filter->tunnel_type =
678 				CFA_NTUPLE_FILTER_ALLOC_REQ_TUNNEL_TYPE_IPGRE;
679 				break;
680 			}
681 			break;
682 
683 		case RTE_FLOW_ITEM_TYPE_VF:
684 			vf_spec = item->spec;
685 			vf = vf_spec->id;
686 			if (!BNXT_PF(bp)) {
687 				rte_flow_error_set(error,
688 						   EINVAL,
689 						   RTE_FLOW_ERROR_TYPE_ITEM,
690 						   item,
691 						   "Configuring on a VF!");
692 				return -rte_errno;
693 			}
694 
695 			if (vf >= bp->pdev->max_vfs) {
696 				rte_flow_error_set(error,
697 						   EINVAL,
698 						   RTE_FLOW_ERROR_TYPE_ITEM,
699 						   item,
700 						   "Incorrect VF id!");
701 				return -rte_errno;
702 			}
703 
704 			if (!attr->transfer) {
705 				rte_flow_error_set(error,
706 						   ENOTSUP,
707 						   RTE_FLOW_ERROR_TYPE_ITEM,
708 						   item,
709 						   "Matching VF traffic without"
710 						   " affecting it (transfer attribute)"
711 						   " is unsupported");
712 				return -rte_errno;
713 			}
714 
715 			filter->mirror_vnic_id =
716 			dflt_vnic = bnxt_hwrm_func_qcfg_vf_dflt_vnic_id(bp, vf);
717 			if (dflt_vnic < 0) {
718 				/* This simply indicates there's no driver
719 				 * loaded. This is not an error.
720 				 */
721 				rte_flow_error_set
722 					(error,
723 					 EINVAL,
724 					 RTE_FLOW_ERROR_TYPE_ITEM,
725 					 item,
726 					 "Unable to get default VNIC for VF");
727 				return -rte_errno;
728 			}
729 
730 			filter->mirror_vnic_id = dflt_vnic;
731 			en |= NTUPLE_FLTR_ALLOC_INPUT_EN_MIRROR_VNIC_ID;
732 			break;
733 		default:
734 			break;
735 		}
736 		item++;
737 	}
738 	filter->enables = en;
739 	filter->valid_flags = valid_flags;
740 
741 	return 0;
742 }
743 
744 /* Parse attributes */
745 static int
746 bnxt_flow_parse_attr(const struct rte_flow_attr *attr,
747 		     struct rte_flow_error *error)
748 {
749 	/* Must be input direction */
750 	if (!attr->ingress) {
751 		rte_flow_error_set(error,
752 				   EINVAL,
753 				   RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
754 				   attr,
755 				   "Only support ingress.");
756 		return -rte_errno;
757 	}
758 
759 	/* Not supported */
760 	if (attr->egress) {
761 		rte_flow_error_set(error,
762 				   EINVAL,
763 				   RTE_FLOW_ERROR_TYPE_ATTR_EGRESS,
764 				   attr,
765 				   "No support for egress.");
766 		return -rte_errno;
767 	}
768 
769 	return 0;
770 }
771 
772 static struct bnxt_filter_info *
773 bnxt_find_matching_l2_filter(struct bnxt *bp, struct bnxt_filter_info *nf)
774 {
775 	struct bnxt_filter_info *mf, *f0;
776 	struct bnxt_vnic_info *vnic0;
777 	int i;
778 
779 	vnic0 = BNXT_GET_DEFAULT_VNIC(bp);
780 	f0 = STAILQ_FIRST(&vnic0->filter);
781 
782 	/* This flow has same DST MAC as the port/l2 filter. */
783 	if (memcmp(f0->l2_addr, nf->dst_macaddr, RTE_ETHER_ADDR_LEN) == 0)
784 		return f0;
785 
786 	for (i = bp->max_vnics - 1; i >= 0; i--) {
787 		struct bnxt_vnic_info *vnic = &bp->vnic_info[i];
788 
789 		if (vnic->fw_vnic_id == INVALID_VNIC_ID)
790 			continue;
791 
792 		STAILQ_FOREACH(mf, &vnic->filter, next) {
793 
794 			if (mf->matching_l2_fltr_ptr)
795 				continue;
796 
797 			if (mf->ethertype == nf->ethertype &&
798 			    mf->l2_ovlan == nf->l2_ovlan &&
799 			    mf->l2_ovlan_mask == nf->l2_ovlan_mask &&
800 			    mf->l2_ivlan == nf->l2_ivlan &&
801 			    mf->l2_ivlan_mask == nf->l2_ivlan_mask &&
802 			    !memcmp(mf->src_macaddr, nf->src_macaddr,
803 				    RTE_ETHER_ADDR_LEN) &&
804 			    !memcmp(mf->dst_macaddr, nf->dst_macaddr,
805 				    RTE_ETHER_ADDR_LEN))
806 				return mf;
807 		}
808 	}
809 	return NULL;
810 }
811 
812 static struct bnxt_filter_info *
813 bnxt_create_l2_filter(struct bnxt *bp, struct bnxt_filter_info *nf,
814 		      struct bnxt_vnic_info *vnic)
815 {
816 	struct bnxt_filter_info *filter1;
817 	int rc;
818 
819 	/* Alloc new L2 filter.
820 	 * This flow needs MAC filter which does not match any existing
821 	 * L2 filters.
822 	 */
823 	filter1 = bnxt_get_unused_filter(bp);
824 	if (filter1 == NULL)
825 		return NULL;
826 
827 	memcpy(filter1, nf, sizeof(*filter1));
828 
829 	filter1->flags = HWRM_CFA_L2_FILTER_ALLOC_INPUT_FLAGS_XDP_DISABLE;
830 	filter1->flags |= HWRM_CFA_L2_FILTER_ALLOC_INPUT_FLAGS_PATH_RX;
831 	if (nf->valid_flags & BNXT_FLOW_L2_SRC_VALID_FLAG ||
832 	    nf->valid_flags & BNXT_FLOW_L2_DST_VALID_FLAG) {
833 		filter1->flags |=
834 			HWRM_CFA_L2_FILTER_ALLOC_INPUT_FLAGS_OUTERMOST;
835 		PMD_DRV_LOG(DEBUG, "Create Outer filter\n");
836 	}
837 
838 	if (nf->filter_type == HWRM_CFA_L2_FILTER &&
839 	    (nf->valid_flags & BNXT_FLOW_L2_SRC_VALID_FLAG ||
840 	     nf->valid_flags & BNXT_FLOW_L2_INNER_SRC_VALID_FLAG)) {
841 		PMD_DRV_LOG(DEBUG, "Create L2 filter for SRC MAC\n");
842 		filter1->flags |=
843 			HWRM_CFA_L2_FILTER_ALLOC_INPUT_FLAGS_SOURCE_VALID;
844 		memcpy(filter1->l2_addr, nf->src_macaddr, RTE_ETHER_ADDR_LEN);
845 	} else {
846 		PMD_DRV_LOG(DEBUG, "Create L2 filter for DST MAC\n");
847 		memcpy(filter1->l2_addr, nf->dst_macaddr, RTE_ETHER_ADDR_LEN);
848 	}
849 
850 	if (nf->priority &&
851 	    (nf->valid_flags & BNXT_FLOW_L2_DST_VALID_FLAG ||
852 	     nf->valid_flags & BNXT_FLOW_L2_INNER_DST_VALID_FLAG)) {
853 		/* Tell the FW where to place the filter in the table. */
854 		if (nf->priority > 65535) {
855 			filter1->pri_hint =
856 			HWRM_CFA_L2_FILTER_ALLOC_INPUT_PRI_HINT_BELOW_FILTER;
857 			/* This will place the filter in TCAM */
858 			filter1->l2_filter_id_hint = (uint64_t)-1;
859 		}
860 	}
861 
862 	if (nf->valid_flags & (BNXT_FLOW_L2_DST_VALID_FLAG |
863 			       BNXT_FLOW_L2_SRC_VALID_FLAG |
864 			       BNXT_FLOW_L2_INNER_SRC_VALID_FLAG |
865 			       BNXT_FLOW_L2_INNER_DST_VALID_FLAG)) {
866 		filter1->enables =
867 			HWRM_CFA_L2_FILTER_ALLOC_INPUT_ENABLES_L2_ADDR |
868 			L2_FILTER_ALLOC_INPUT_EN_L2_ADDR_MASK;
869 		memset(filter1->l2_addr_mask, 0xff, RTE_ETHER_ADDR_LEN);
870 	}
871 
872 	if (nf->valid_flags & BNXT_FLOW_L2_DROP_FLAG) {
873 		filter1->flags |=
874 			HWRM_CFA_L2_FILTER_ALLOC_INPUT_FLAGS_DROP;
875 		if (nf->ethertype == RTE_ETHER_TYPE_IPV4) {
876 			/* Num VLANs for drop filter will/should be 0.
877 			 * If the req is memset to 0, then the count will
878 			 * be automatically set to 0.
879 			 */
880 			if (nf->valid_flags & BNXT_FLOW_PARSE_INNER_FLAG) {
881 				filter1->enables |=
882 					L2_FILTER_ALLOC_INPUT_EN_T_NUM_VLANS;
883 			} else {
884 				filter1->enables |=
885 					L2_FILTER_ALLOC_INPUT_EN_NUM_VLANS;
886 				filter1->flags |=
887 				HWRM_CFA_L2_FILTER_ALLOC_INPUT_FLAGS_OUTERMOST;
888 			}
889 		}
890 	}
891 
892 	rc = bnxt_hwrm_set_l2_filter(bp, vnic->fw_vnic_id,
893 				     filter1);
894 	if (rc) {
895 		bnxt_free_filter(bp, filter1);
896 		return NULL;
897 	}
898 	return filter1;
899 }
900 
901 struct bnxt_filter_info *
902 bnxt_get_l2_filter(struct bnxt *bp, struct bnxt_filter_info *nf,
903 		   struct bnxt_vnic_info *vnic)
904 {
905 	struct bnxt_filter_info *l2_filter = NULL;
906 
907 	l2_filter = bnxt_find_matching_l2_filter(bp, nf);
908 	if (l2_filter) {
909 		l2_filter->l2_ref_cnt++;
910 	} else {
911 		l2_filter = bnxt_create_l2_filter(bp, nf, vnic);
912 		if (l2_filter) {
913 			STAILQ_INSERT_TAIL(&vnic->filter, l2_filter, next);
914 			l2_filter->vnic = vnic;
915 		}
916 	}
917 	nf->matching_l2_fltr_ptr = l2_filter;
918 
919 	return l2_filter;
920 }
921 
922 static int bnxt_vnic_prep(struct bnxt *bp, struct bnxt_vnic_info *vnic)
923 {
924 	struct rte_eth_conf *dev_conf = &bp->eth_dev->data->dev_conf;
925 	uint64_t rx_offloads = dev_conf->rxmode.offloads;
926 	int rc;
927 
928 	rc = bnxt_vnic_grp_alloc(bp, vnic);
929 	if (rc)
930 		goto ret;
931 
932 	rc = bnxt_hwrm_vnic_alloc(bp, vnic);
933 	if (rc) {
934 		PMD_DRV_LOG(ERR, "HWRM vnic alloc failure rc: %x\n", rc);
935 		goto ret;
936 	}
937 	bp->nr_vnics++;
938 
939 	/* RSS context is required only when there is more than one RSS ring */
940 	if (vnic->rx_queue_cnt > 1) {
941 		rc = bnxt_hwrm_vnic_ctx_alloc(bp, vnic, 0 /* ctx_idx 0 */);
942 		if (rc) {
943 			PMD_DRV_LOG(ERR,
944 				    "HWRM vnic ctx alloc failure: %x\n", rc);
945 			goto ret;
946 		}
947 	} else {
948 		PMD_DRV_LOG(DEBUG, "No RSS context required\n");
949 	}
950 
951 	if (rx_offloads & DEV_RX_OFFLOAD_VLAN_STRIP)
952 		vnic->vlan_strip = true;
953 	else
954 		vnic->vlan_strip = false;
955 
956 	rc = bnxt_hwrm_vnic_cfg(bp, vnic);
957 	if (rc)
958 		goto ret;
959 
960 	bnxt_hwrm_vnic_plcmode_cfg(bp, vnic);
961 
962 ret:
963 	return rc;
964 }
965 
966 static int match_vnic_rss_cfg(struct bnxt *bp,
967 			      struct bnxt_vnic_info *vnic,
968 			      const struct rte_flow_action_rss *rss)
969 {
970 	unsigned int match = 0, i;
971 
972 	if (vnic->rx_queue_cnt != rss->queue_num)
973 		return -EINVAL;
974 
975 	for (i = 0; i < rss->queue_num; i++) {
976 		if (!bp->rx_queues[rss->queue[i]]->vnic->rx_queue_cnt &&
977 		    !bp->rx_queues[rss->queue[i]]->rx_started)
978 			return -EINVAL;
979 	}
980 
981 	for (i = 0; i < vnic->rx_queue_cnt; i++) {
982 		int j;
983 
984 		for (j = 0; j < vnic->rx_queue_cnt; j++) {
985 			if (bp->grp_info[rss->queue[i]].fw_grp_id ==
986 			    vnic->fw_grp_ids[j])
987 				match++;
988 		}
989 	}
990 
991 	if (match != vnic->rx_queue_cnt) {
992 		PMD_DRV_LOG(ERR,
993 			    "VNIC queue count %d vs queues matched %d\n",
994 			    match, vnic->rx_queue_cnt);
995 		return -EINVAL;
996 	}
997 
998 	return 0;
999 }
1000 
1001 static void
1002 bnxt_update_filter_flags_en(struct bnxt_filter_info *filter,
1003 			    struct bnxt_filter_info *filter1,
1004 			    int use_ntuple)
1005 {
1006 	if (!use_ntuple &&
1007 	    !(filter->valid_flags &
1008 	      ~(BNXT_FLOW_L2_DST_VALID_FLAG |
1009 		BNXT_FLOW_L2_SRC_VALID_FLAG |
1010 		BNXT_FLOW_L2_INNER_SRC_VALID_FLAG |
1011 		BNXT_FLOW_L2_INNER_DST_VALID_FLAG |
1012 		BNXT_FLOW_L2_DROP_FLAG |
1013 		BNXT_FLOW_PARSE_INNER_FLAG))) {
1014 		filter->flags = filter1->flags;
1015 		filter->enables = filter1->enables;
1016 		filter->filter_type = HWRM_CFA_L2_FILTER;
1017 		memcpy(filter->l2_addr, filter1->l2_addr, RTE_ETHER_ADDR_LEN);
1018 		memset(filter->l2_addr_mask, 0xff, RTE_ETHER_ADDR_LEN);
1019 		filter->pri_hint = filter1->pri_hint;
1020 		filter->l2_filter_id_hint = filter1->l2_filter_id_hint;
1021 	}
1022 	filter->fw_l2_filter_id = filter1->fw_l2_filter_id;
1023 	filter->l2_ref_cnt = filter1->l2_ref_cnt;
1024 	filter->flow_id = filter1->flow_id;
1025 	PMD_DRV_LOG(DEBUG,
1026 		"l2_filter: %p fw_l2_filter_id %" PRIx64 " l2_ref_cnt %u\n",
1027 		filter1, filter->fw_l2_filter_id, filter->l2_ref_cnt);
1028 }
1029 
1030 static int
1031 bnxt_validate_and_parse_flow(struct rte_eth_dev *dev,
1032 			     const struct rte_flow_item pattern[],
1033 			     const struct rte_flow_action actions[],
1034 			     const struct rte_flow_attr *attr,
1035 			     struct rte_flow_error *error,
1036 			     struct bnxt_filter_info *filter)
1037 {
1038 	const struct rte_flow_action *act =
1039 		bnxt_flow_non_void_action(actions);
1040 	struct bnxt *bp = dev->data->dev_private;
1041 	struct rte_eth_conf *dev_conf = &bp->eth_dev->data->dev_conf;
1042 	struct bnxt_vnic_info *vnic = NULL, *vnic0 = NULL;
1043 	const struct rte_flow_action_queue *act_q;
1044 	const struct rte_flow_action_vf *act_vf;
1045 	struct bnxt_filter_info *filter1 = NULL;
1046 	const struct rte_flow_action_rss *rss;
1047 	struct bnxt_rx_queue *rxq = NULL;
1048 	int dflt_vnic, vnic_id;
1049 	unsigned int rss_idx;
1050 	uint32_t vf = 0, i;
1051 	int rc, use_ntuple;
1052 
1053 	rc =
1054 	bnxt_validate_and_parse_flow_type(bp, attr, pattern, error, filter);
1055 	if (rc != 0)
1056 		goto ret;
1057 
1058 	rc = bnxt_flow_parse_attr(attr, error);
1059 	if (rc != 0)
1060 		goto ret;
1061 
1062 	/* Since we support ingress attribute only - right now. */
1063 	if (filter->filter_type == HWRM_CFA_EM_FILTER)
1064 		filter->flags = HWRM_CFA_EM_FLOW_ALLOC_INPUT_FLAGS_PATH_RX;
1065 
1066 	use_ntuple = bnxt_filter_type_check(pattern, error);
1067 
1068 start:
1069 	switch (act->type) {
1070 	case RTE_FLOW_ACTION_TYPE_QUEUE:
1071 		/* Allow this flow. Redirect to a VNIC. */
1072 		act_q = (const struct rte_flow_action_queue *)act->conf;
1073 		if (!act_q->index || act_q->index >= bp->rx_nr_rings) {
1074 			rte_flow_error_set(error,
1075 					   EINVAL,
1076 					   RTE_FLOW_ERROR_TYPE_ACTION,
1077 					   act,
1078 					   "Invalid queue ID.");
1079 			rc = -rte_errno;
1080 			goto ret;
1081 		}
1082 		PMD_DRV_LOG(DEBUG, "Queue index %d\n", act_q->index);
1083 
1084 		if (use_ntuple && !BNXT_RFS_NEEDS_VNIC(bp)) {
1085 			filter->flags =
1086 				HWRM_CFA_NTUPLE_FILTER_ALLOC_INPUT_FLAGS_DEST_RFS_RING_IDX;
1087 			filter->dst_id = act_q->index;
1088 			goto skip_vnic_alloc;
1089 		}
1090 
1091 		vnic_id = attr->group;
1092 		if (!vnic_id) {
1093 			PMD_DRV_LOG(DEBUG, "Group id is 0\n");
1094 			vnic_id = act_q->index;
1095 		}
1096 
1097 		BNXT_VALID_VNIC_OR_RET(bp, vnic_id);
1098 
1099 		vnic = &bp->vnic_info[vnic_id];
1100 		if (vnic->rx_queue_cnt) {
1101 			if (vnic->start_grp_id != act_q->index) {
1102 				PMD_DRV_LOG(ERR,
1103 					    "VNIC already in use\n");
1104 				rte_flow_error_set(error,
1105 						   EINVAL,
1106 						   RTE_FLOW_ERROR_TYPE_ACTION,
1107 						   act,
1108 						   "VNIC already in use");
1109 				rc = -rte_errno;
1110 				goto ret;
1111 			}
1112 			goto use_vnic;
1113 		}
1114 
1115 		rxq = bp->rx_queues[act_q->index];
1116 
1117 		if (!(dev_conf->rxmode.mq_mode & ETH_MQ_RX_RSS) && rxq &&
1118 		    vnic->fw_vnic_id != INVALID_HW_RING_ID)
1119 			goto use_vnic;
1120 
1121 		if (!rxq) {
1122 			PMD_DRV_LOG(ERR,
1123 				    "Queue invalid or used with other VNIC\n");
1124 			rte_flow_error_set(error,
1125 					   EINVAL,
1126 					   RTE_FLOW_ERROR_TYPE_ACTION,
1127 					   act,
1128 					   "Queue invalid queue or in use");
1129 			rc = -rte_errno;
1130 			goto ret;
1131 		}
1132 
1133 		rxq->vnic = vnic;
1134 		rxq->rx_started = 1;
1135 		vnic->rx_queue_cnt++;
1136 		vnic->start_grp_id = act_q->index;
1137 		vnic->end_grp_id = act_q->index;
1138 		vnic->func_default = 0;	//This is not a default VNIC.
1139 
1140 		PMD_DRV_LOG(DEBUG, "VNIC found\n");
1141 
1142 		rc = bnxt_vnic_prep(bp, vnic);
1143 		if (rc)  {
1144 			rte_flow_error_set(error,
1145 					   EINVAL,
1146 					   RTE_FLOW_ERROR_TYPE_ACTION,
1147 					   act,
1148 					   "VNIC prep fail");
1149 			rc = -rte_errno;
1150 			goto ret;
1151 		}
1152 
1153 		PMD_DRV_LOG(DEBUG,
1154 			    "vnic[%d] = %p vnic->fw_grp_ids = %p\n",
1155 			    act_q->index, vnic, vnic->fw_grp_ids);
1156 
1157 use_vnic:
1158 		vnic->ff_pool_idx = vnic_id;
1159 		PMD_DRV_LOG(DEBUG,
1160 			    "Setting vnic ff_idx %d\n", vnic->ff_pool_idx);
1161 		filter->dst_id = vnic->fw_vnic_id;
1162 skip_vnic_alloc:
1163 		/* For ntuple filter, create the L2 filter with default VNIC.
1164 		 * The user specified redirect queue will be set while creating
1165 		 * the ntuple filter in hardware.
1166 		 */
1167 		vnic0 = BNXT_GET_DEFAULT_VNIC(bp);
1168 		if (use_ntuple)
1169 			filter1 = bnxt_get_l2_filter(bp, filter, vnic0);
1170 		else
1171 			filter1 = bnxt_get_l2_filter(bp, filter, vnic);
1172 		if (filter1 == NULL) {
1173 			rte_flow_error_set(error,
1174 					   ENOSPC,
1175 					   RTE_FLOW_ERROR_TYPE_ACTION,
1176 					   act,
1177 					   "Filter not available");
1178 			rc = -rte_errno;
1179 			goto ret;
1180 		}
1181 
1182 		PMD_DRV_LOG(DEBUG, "new fltr: %p l2fltr: %p l2_ref_cnt: %d\n",
1183 			    filter, filter1, filter1->l2_ref_cnt);
1184 		bnxt_update_filter_flags_en(filter, filter1, use_ntuple);
1185 		break;
1186 	case RTE_FLOW_ACTION_TYPE_DROP:
1187 		vnic0 = &bp->vnic_info[0];
1188 		filter->dst_id = vnic0->fw_vnic_id;
1189 		filter->valid_flags |= BNXT_FLOW_L2_DROP_FLAG;
1190 		filter1 = bnxt_get_l2_filter(bp, filter, vnic0);
1191 		if (filter1 == NULL) {
1192 			rte_flow_error_set(error,
1193 					   ENOSPC,
1194 					   RTE_FLOW_ERROR_TYPE_ACTION,
1195 					   act,
1196 					   "Filter not available");
1197 			rc = -rte_errno;
1198 			goto ret;
1199 		}
1200 
1201 		if (filter->filter_type == HWRM_CFA_EM_FILTER)
1202 			filter->flags =
1203 				HWRM_CFA_EM_FLOW_ALLOC_INPUT_FLAGS_DROP;
1204 		else if (filter->filter_type == HWRM_CFA_NTUPLE_FILTER)
1205 			filter->flags =
1206 				HWRM_CFA_NTUPLE_FILTER_ALLOC_INPUT_FLAGS_DROP;
1207 
1208 		bnxt_update_filter_flags_en(filter, filter1, use_ntuple);
1209 		break;
1210 	case RTE_FLOW_ACTION_TYPE_COUNT:
1211 		vnic0 = &bp->vnic_info[0];
1212 		filter1 = bnxt_get_l2_filter(bp, filter, vnic0);
1213 		if (filter1 == NULL) {
1214 			rte_flow_error_set(error,
1215 					   ENOSPC,
1216 					   RTE_FLOW_ERROR_TYPE_ACTION,
1217 					   act,
1218 					   "New filter not available");
1219 			rc = -rte_errno;
1220 			goto ret;
1221 		}
1222 
1223 		filter->fw_l2_filter_id = filter1->fw_l2_filter_id;
1224 		filter->flow_id = filter1->flow_id;
1225 		filter->flags = HWRM_CFA_NTUPLE_FILTER_ALLOC_INPUT_FLAGS_METER;
1226 		break;
1227 	case RTE_FLOW_ACTION_TYPE_VF:
1228 		act_vf = (const struct rte_flow_action_vf *)act->conf;
1229 		vf = act_vf->id;
1230 
1231 		if (filter->tunnel_type ==
1232 		    CFA_NTUPLE_FILTER_ALLOC_REQ_TUNNEL_TYPE_VXLAN ||
1233 		    filter->tunnel_type ==
1234 		    CFA_NTUPLE_FILTER_ALLOC_REQ_TUNNEL_TYPE_IPGRE) {
1235 			/* If issued on a VF, ensure id is 0 and is trusted */
1236 			if (BNXT_VF(bp)) {
1237 				if (!BNXT_VF_IS_TRUSTED(bp) || vf) {
1238 					rte_flow_error_set(error, EINVAL,
1239 						RTE_FLOW_ERROR_TYPE_ACTION,
1240 						act,
1241 						"Incorrect VF");
1242 					rc = -rte_errno;
1243 					goto ret;
1244 				}
1245 			}
1246 
1247 			filter->enables |= filter->tunnel_type;
1248 			filter->filter_type = HWRM_CFA_TUNNEL_REDIRECT_FILTER;
1249 			goto done;
1250 		}
1251 
1252 		if (vf >= bp->pdev->max_vfs) {
1253 			rte_flow_error_set(error,
1254 					   EINVAL,
1255 					   RTE_FLOW_ERROR_TYPE_ACTION,
1256 					   act,
1257 					   "Incorrect VF id!");
1258 			rc = -rte_errno;
1259 			goto ret;
1260 		}
1261 
1262 		filter->mirror_vnic_id =
1263 		dflt_vnic = bnxt_hwrm_func_qcfg_vf_dflt_vnic_id(bp, vf);
1264 		if (dflt_vnic < 0) {
1265 			/* This simply indicates there's no driver loaded.
1266 			 * This is not an error.
1267 			 */
1268 			rte_flow_error_set(error,
1269 					   EINVAL,
1270 					   RTE_FLOW_ERROR_TYPE_ACTION,
1271 					   act,
1272 					   "Unable to get default VNIC for VF");
1273 			rc = -rte_errno;
1274 			goto ret;
1275 		}
1276 
1277 		filter->mirror_vnic_id = dflt_vnic;
1278 		filter->enables |= NTUPLE_FLTR_ALLOC_INPUT_EN_MIRROR_VNIC_ID;
1279 
1280 		vnic0 = &bp->vnic_info[0];
1281 		filter1 = bnxt_get_l2_filter(bp, filter, vnic0);
1282 		if (filter1 == NULL) {
1283 			rte_flow_error_set(error,
1284 					   ENOSPC,
1285 					   RTE_FLOW_ERROR_TYPE_ACTION,
1286 					   act,
1287 					   "New filter not available");
1288 			rc = -rte_errno;
1289 			goto ret;
1290 		}
1291 
1292 		filter->fw_l2_filter_id = filter1->fw_l2_filter_id;
1293 		filter->flow_id = filter1->flow_id;
1294 		break;
1295 	case RTE_FLOW_ACTION_TYPE_RSS:
1296 		rss = (const struct rte_flow_action_rss *)act->conf;
1297 
1298 		vnic_id = attr->group;
1299 
1300 		BNXT_VALID_VNIC_OR_RET(bp, vnic_id);
1301 		vnic = &bp->vnic_info[vnic_id];
1302 
1303 		/* Check if requested RSS config matches RSS config of VNIC
1304 		 * only if it is not a fresh VNIC configuration.
1305 		 * Otherwise the existing VNIC configuration can be used.
1306 		 */
1307 		if (vnic->rx_queue_cnt) {
1308 			rc = match_vnic_rss_cfg(bp, vnic, rss);
1309 			if (rc) {
1310 				PMD_DRV_LOG(ERR,
1311 					    "VNIC and RSS config mismatch\n");
1312 				rte_flow_error_set(error,
1313 						   EINVAL,
1314 						   RTE_FLOW_ERROR_TYPE_ACTION,
1315 						   act,
1316 						   "VNIC and RSS cfg mismatch");
1317 				rc = -rte_errno;
1318 				goto ret;
1319 			}
1320 			goto vnic_found;
1321 		}
1322 
1323 		for (i = 0; i < rss->queue_num; i++) {
1324 			PMD_DRV_LOG(DEBUG, "RSS action Queue %d\n",
1325 				    rss->queue[i]);
1326 
1327 			if (!rss->queue[i] ||
1328 			    rss->queue[i] >= bp->rx_nr_rings ||
1329 			    !bp->rx_queues[rss->queue[i]]) {
1330 				rte_flow_error_set(error,
1331 						   EINVAL,
1332 						   RTE_FLOW_ERROR_TYPE_ACTION,
1333 						   act,
1334 						   "Invalid queue ID for RSS");
1335 				rc = -rte_errno;
1336 				goto ret;
1337 			}
1338 			rxq = bp->rx_queues[rss->queue[i]];
1339 
1340 			if (bp->vnic_info[0].fw_grp_ids[rss->queue[i]] !=
1341 			    INVALID_HW_RING_ID) {
1342 				PMD_DRV_LOG(ERR,
1343 					    "queue active with other VNIC\n");
1344 				rte_flow_error_set(error,
1345 						   EINVAL,
1346 						   RTE_FLOW_ERROR_TYPE_ACTION,
1347 						   act,
1348 						   "Invalid queue ID for RSS");
1349 				rc = -rte_errno;
1350 				goto ret;
1351 			}
1352 
1353 			rxq->vnic = vnic;
1354 			rxq->rx_started = 1;
1355 			vnic->rx_queue_cnt++;
1356 		}
1357 
1358 		vnic->start_grp_id = rss->queue[0];
1359 		vnic->end_grp_id = rss->queue[rss->queue_num - 1];
1360 		vnic->func_default = 0;	//This is not a default VNIC.
1361 
1362 		rc = bnxt_vnic_prep(bp, vnic);
1363 		if (rc) {
1364 			rte_flow_error_set(error,
1365 					   EINVAL,
1366 					   RTE_FLOW_ERROR_TYPE_ACTION,
1367 					   act,
1368 					   "VNIC prep fail");
1369 			rc = -rte_errno;
1370 			goto ret;
1371 		}
1372 
1373 		PMD_DRV_LOG(DEBUG,
1374 			    "vnic[%d] = %p vnic->fw_grp_ids = %p\n",
1375 			    vnic_id, vnic, vnic->fw_grp_ids);
1376 
1377 		vnic->ff_pool_idx = vnic_id;
1378 		PMD_DRV_LOG(DEBUG,
1379 			    "Setting vnic ff_pool_idx %d\n", vnic->ff_pool_idx);
1380 
1381 		/* This can be done only after vnic_grp_alloc is done. */
1382 		for (i = 0; i < vnic->rx_queue_cnt; i++) {
1383 			vnic->fw_grp_ids[i] =
1384 				bp->grp_info[rss->queue[i]].fw_grp_id;
1385 			/* Make sure vnic0 does not use these rings. */
1386 			bp->vnic_info[0].fw_grp_ids[rss->queue[i]] =
1387 				INVALID_HW_RING_ID;
1388 		}
1389 
1390 		for (rss_idx = 0; rss_idx < HW_HASH_INDEX_SIZE; ) {
1391 			for (i = 0; i < vnic->rx_queue_cnt; i++)
1392 				vnic->rss_table[rss_idx++] =
1393 					vnic->fw_grp_ids[i];
1394 		}
1395 
1396 		/* Configure RSS only if the queue count is > 1 */
1397 		if (vnic->rx_queue_cnt > 1) {
1398 			vnic->hash_type =
1399 				bnxt_rte_to_hwrm_hash_types(rss->types);
1400 			vnic->hash_mode =
1401 			bnxt_rte_to_hwrm_hash_level(bp, rss->types, rss->level);
1402 
1403 			if (!rss->key_len) {
1404 				/* If hash key has not been specified,
1405 				 * use random hash key.
1406 				 */
1407 				prandom_bytes(vnic->rss_hash_key,
1408 					      HW_HASH_KEY_SIZE);
1409 			} else {
1410 				if (rss->key_len > HW_HASH_KEY_SIZE)
1411 					memcpy(vnic->rss_hash_key,
1412 					       rss->key,
1413 					       HW_HASH_KEY_SIZE);
1414 				else
1415 					memcpy(vnic->rss_hash_key,
1416 					       rss->key,
1417 					       rss->key_len);
1418 			}
1419 			bnxt_hwrm_vnic_rss_cfg(bp, vnic);
1420 		} else {
1421 			PMD_DRV_LOG(DEBUG, "No RSS config required\n");
1422 		}
1423 
1424 vnic_found:
1425 		filter->dst_id = vnic->fw_vnic_id;
1426 		filter1 = bnxt_get_l2_filter(bp, filter, vnic);
1427 		if (filter1 == NULL) {
1428 			rte_flow_error_set(error,
1429 					   ENOSPC,
1430 					   RTE_FLOW_ERROR_TYPE_ACTION,
1431 					   act,
1432 					   "New filter not available");
1433 			rc = -rte_errno;
1434 			goto ret;
1435 		}
1436 
1437 		PMD_DRV_LOG(DEBUG, "L2 filter created\n");
1438 		bnxt_update_filter_flags_en(filter, filter1, use_ntuple);
1439 		break;
1440 	case RTE_FLOW_ACTION_TYPE_MARK:
1441 		if (bp->mark_table == NULL) {
1442 			rte_flow_error_set(error,
1443 					   ENOMEM,
1444 					   RTE_FLOW_ERROR_TYPE_ACTION,
1445 					   act,
1446 					   "Mark table not allocated.");
1447 			rc = -rte_errno;
1448 			goto ret;
1449 		}
1450 
1451 		if (bp->flags & BNXT_FLAG_RX_VECTOR_PKT_MODE) {
1452 			PMD_DRV_LOG(DEBUG,
1453 				    "Disabling vector processing for mark\n");
1454 			bp->eth_dev->rx_pkt_burst = bnxt_recv_pkts;
1455 			bp->flags &= ~BNXT_FLAG_RX_VECTOR_PKT_MODE;
1456 		}
1457 
1458 		filter->valid_flags |= BNXT_FLOW_MARK_FLAG;
1459 		filter->mark = ((const struct rte_flow_action_mark *)
1460 				act->conf)->id;
1461 		PMD_DRV_LOG(DEBUG, "Mark the flow %d\n", filter->mark);
1462 		break;
1463 	default:
1464 		rte_flow_error_set(error,
1465 				   EINVAL,
1466 				   RTE_FLOW_ERROR_TYPE_ACTION,
1467 				   act,
1468 				   "Invalid action.");
1469 		rc = -rte_errno;
1470 		goto ret;
1471 	}
1472 
1473 done:
1474 	act = bnxt_flow_non_void_action(++act);
1475 	while (act->type != RTE_FLOW_ACTION_TYPE_END)
1476 		goto start;
1477 
1478 	return rc;
1479 ret:
1480 
1481 	if (filter1) {
1482 		bnxt_hwrm_clear_l2_filter(bp, filter1);
1483 		bnxt_free_filter(bp, filter1);
1484 	}
1485 
1486 	if (rte_errno)  {
1487 		if (vnic && STAILQ_EMPTY(&vnic->filter))
1488 			vnic->rx_queue_cnt = 0;
1489 
1490 		if (rxq && !vnic->rx_queue_cnt)
1491 			rxq->vnic = &bp->vnic_info[0];
1492 	}
1493 	return -rte_errno;
1494 }
1495 
1496 static
1497 struct bnxt_vnic_info *find_matching_vnic(struct bnxt *bp,
1498 					  struct bnxt_filter_info *filter)
1499 {
1500 	struct bnxt_vnic_info *vnic = NULL;
1501 	unsigned int i;
1502 
1503 	for (i = 0; i < bp->max_vnics; i++) {
1504 		vnic = &bp->vnic_info[i];
1505 		if (vnic->fw_vnic_id != INVALID_VNIC_ID &&
1506 		    filter->dst_id == vnic->fw_vnic_id) {
1507 			PMD_DRV_LOG(DEBUG, "Found matching VNIC Id %d\n",
1508 				    vnic->ff_pool_idx);
1509 			return vnic;
1510 		}
1511 	}
1512 	return NULL;
1513 }
1514 
1515 static int
1516 bnxt_flow_validate(struct rte_eth_dev *dev,
1517 		   const struct rte_flow_attr *attr,
1518 		   const struct rte_flow_item pattern[],
1519 		   const struct rte_flow_action actions[],
1520 		   struct rte_flow_error *error)
1521 {
1522 	struct bnxt *bp = dev->data->dev_private;
1523 	struct bnxt_vnic_info *vnic = NULL;
1524 	struct bnxt_filter_info *filter;
1525 	int ret = 0;
1526 
1527 	bnxt_acquire_flow_lock(bp);
1528 	ret = bnxt_flow_args_validate(attr, pattern, actions, error);
1529 	if (ret != 0) {
1530 		bnxt_release_flow_lock(bp);
1531 		return ret;
1532 	}
1533 
1534 	filter = bnxt_get_unused_filter(bp);
1535 	if (filter == NULL) {
1536 		PMD_DRV_LOG(ERR, "Not enough resources for a new flow.\n");
1537 		bnxt_release_flow_lock(bp);
1538 		return -ENOMEM;
1539 	}
1540 
1541 	ret = bnxt_validate_and_parse_flow(dev, pattern, actions, attr,
1542 					   error, filter);
1543 	if (ret)
1544 		goto exit;
1545 
1546 	vnic = find_matching_vnic(bp, filter);
1547 	if (vnic) {
1548 		if (STAILQ_EMPTY(&vnic->filter)) {
1549 			rte_free(vnic->fw_grp_ids);
1550 			bnxt_hwrm_vnic_ctx_free(bp, vnic);
1551 			bnxt_hwrm_vnic_free(bp, vnic);
1552 			vnic->rx_queue_cnt = 0;
1553 			PMD_DRV_LOG(DEBUG, "Free VNIC\n");
1554 		}
1555 	}
1556 
1557 	if (filter->filter_type == HWRM_CFA_EM_FILTER)
1558 		bnxt_hwrm_clear_em_filter(bp, filter);
1559 	else if (filter->filter_type == HWRM_CFA_NTUPLE_FILTER)
1560 		bnxt_hwrm_clear_ntuple_filter(bp, filter);
1561 	else
1562 		bnxt_hwrm_clear_l2_filter(bp, filter);
1563 
1564 exit:
1565 	/* No need to hold on to this filter if we are just validating flow */
1566 	bnxt_free_filter(bp, filter);
1567 	bnxt_release_flow_lock(bp);
1568 
1569 	return ret;
1570 }
1571 
1572 static void
1573 bnxt_update_filter(struct bnxt *bp, struct bnxt_filter_info *old_filter,
1574 		   struct bnxt_filter_info *new_filter)
1575 {
1576 	/* Clear the new L2 filter that was created in the previous step in
1577 	 * bnxt_validate_and_parse_flow. For L2 filters, we will use the new
1578 	 * filter which points to the new destination queue and so we clear
1579 	 * the previous L2 filter. For ntuple filters, we are going to reuse
1580 	 * the old L2 filter and create new NTUPLE filter with this new
1581 	 * destination queue subsequently during bnxt_flow_create. So we
1582 	 * decrement the ref cnt of the L2 filter that would've been bumped
1583 	 * up previously in bnxt_validate_and_parse_flow as the old n-tuple
1584 	 * filter that was referencing it will be deleted now.
1585 	 */
1586 	bnxt_hwrm_clear_l2_filter(bp, old_filter);
1587 	if (new_filter->filter_type == HWRM_CFA_L2_FILTER) {
1588 		bnxt_hwrm_set_l2_filter(bp, new_filter->dst_id, new_filter);
1589 	} else {
1590 		if (new_filter->filter_type == HWRM_CFA_EM_FILTER)
1591 			bnxt_hwrm_clear_em_filter(bp, old_filter);
1592 		if (new_filter->filter_type == HWRM_CFA_NTUPLE_FILTER)
1593 			bnxt_hwrm_clear_ntuple_filter(bp, old_filter);
1594 	}
1595 }
1596 
1597 static int
1598 bnxt_match_filter(struct bnxt *bp, struct bnxt_filter_info *nf)
1599 {
1600 	struct bnxt_filter_info *mf;
1601 	struct rte_flow *flow;
1602 	int i;
1603 
1604 	for (i = bp->max_vnics - 1; i >= 0; i--) {
1605 		struct bnxt_vnic_info *vnic = &bp->vnic_info[i];
1606 
1607 		if (vnic->fw_vnic_id == INVALID_VNIC_ID)
1608 			continue;
1609 
1610 		STAILQ_FOREACH(flow, &vnic->flow_list, next) {
1611 			mf = flow->filter;
1612 
1613 			if (mf->filter_type == nf->filter_type &&
1614 			    mf->flags == nf->flags &&
1615 			    mf->src_port == nf->src_port &&
1616 			    mf->src_port_mask == nf->src_port_mask &&
1617 			    mf->dst_port == nf->dst_port &&
1618 			    mf->dst_port_mask == nf->dst_port_mask &&
1619 			    mf->ip_protocol == nf->ip_protocol &&
1620 			    mf->ip_addr_type == nf->ip_addr_type &&
1621 			    mf->ethertype == nf->ethertype &&
1622 			    mf->vni == nf->vni &&
1623 			    mf->tunnel_type == nf->tunnel_type &&
1624 			    mf->l2_ovlan == nf->l2_ovlan &&
1625 			    mf->l2_ovlan_mask == nf->l2_ovlan_mask &&
1626 			    mf->l2_ivlan == nf->l2_ivlan &&
1627 			    mf->l2_ivlan_mask == nf->l2_ivlan_mask &&
1628 			    !memcmp(mf->l2_addr, nf->l2_addr,
1629 				    RTE_ETHER_ADDR_LEN) &&
1630 			    !memcmp(mf->l2_addr_mask, nf->l2_addr_mask,
1631 				    RTE_ETHER_ADDR_LEN) &&
1632 			    !memcmp(mf->src_macaddr, nf->src_macaddr,
1633 				    RTE_ETHER_ADDR_LEN) &&
1634 			    !memcmp(mf->dst_macaddr, nf->dst_macaddr,
1635 				    RTE_ETHER_ADDR_LEN) &&
1636 			    !memcmp(mf->src_ipaddr, nf->src_ipaddr,
1637 				    sizeof(nf->src_ipaddr)) &&
1638 			    !memcmp(mf->src_ipaddr_mask, nf->src_ipaddr_mask,
1639 				    sizeof(nf->src_ipaddr_mask)) &&
1640 			    !memcmp(mf->dst_ipaddr, nf->dst_ipaddr,
1641 				    sizeof(nf->dst_ipaddr)) &&
1642 			    !memcmp(mf->dst_ipaddr_mask, nf->dst_ipaddr_mask,
1643 				    sizeof(nf->dst_ipaddr_mask))) {
1644 				if (mf->dst_id == nf->dst_id)
1645 					return -EEXIST;
1646 				/* Free the old filter, update flow
1647 				 * with new filter
1648 				 */
1649 				bnxt_update_filter(bp, mf, nf);
1650 				STAILQ_REMOVE(&vnic->filter, mf,
1651 					      bnxt_filter_info, next);
1652 				STAILQ_INSERT_TAIL(&vnic->filter, nf, next);
1653 				bnxt_free_filter(bp, mf);
1654 				flow->filter = nf;
1655 				return -EXDEV;
1656 			}
1657 		}
1658 	}
1659 	return 0;
1660 }
1661 
1662 static void
1663 bnxt_setup_flow_counter(struct bnxt *bp)
1664 {
1665 	if (bp->fw_cap & BNXT_FW_CAP_ADV_FLOW_COUNTERS &&
1666 	    !(bp->flags & BNXT_FLAG_FC_THREAD) && BNXT_FLOW_XSTATS_EN(bp)) {
1667 		rte_eal_alarm_set(US_PER_S * BNXT_FC_TIMER,
1668 				  bnxt_flow_cnt_alarm_cb,
1669 				  (void *)bp);
1670 		bp->flags |= BNXT_FLAG_FC_THREAD;
1671 	}
1672 }
1673 
1674 void bnxt_flow_cnt_alarm_cb(void *arg)
1675 {
1676 	int rc = 0;
1677 	struct bnxt *bp = arg;
1678 
1679 	if (!bp->flow_stat->rx_fc_out_tbl.va) {
1680 		PMD_DRV_LOG(ERR, "bp->flow_stat->rx_fc_out_tbl.va is NULL?\n");
1681 		bnxt_cancel_fc_thread(bp);
1682 		return;
1683 	}
1684 
1685 	if (!bp->flow_stat->flow_count) {
1686 		bnxt_cancel_fc_thread(bp);
1687 		return;
1688 	}
1689 
1690 	if (!bp->eth_dev->data->dev_started) {
1691 		bnxt_cancel_fc_thread(bp);
1692 		return;
1693 	}
1694 
1695 	rc = bnxt_flow_stats_req(bp);
1696 	if (rc) {
1697 		PMD_DRV_LOG(ERR, "Flow stat alarm not rescheduled.\n");
1698 		return;
1699 	}
1700 
1701 	rte_eal_alarm_set(US_PER_S * BNXT_FC_TIMER,
1702 			  bnxt_flow_cnt_alarm_cb,
1703 			  (void *)bp);
1704 }
1705 
1706 
1707 static struct rte_flow *
1708 bnxt_flow_create(struct rte_eth_dev *dev,
1709 		 const struct rte_flow_attr *attr,
1710 		 const struct rte_flow_item pattern[],
1711 		 const struct rte_flow_action actions[],
1712 		 struct rte_flow_error *error)
1713 {
1714 	struct bnxt *bp = dev->data->dev_private;
1715 	struct bnxt_vnic_info *vnic = NULL;
1716 	struct bnxt_filter_info *filter;
1717 	bool update_flow = false;
1718 	struct rte_flow *flow;
1719 	int ret = 0;
1720 	uint32_t tun_type, flow_id;
1721 
1722 	if (BNXT_VF(bp) && !BNXT_VF_IS_TRUSTED(bp)) {
1723 		rte_flow_error_set(error, EINVAL,
1724 				   RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
1725 				   "Failed to create flow, Not a Trusted VF!");
1726 		return NULL;
1727 	}
1728 
1729 	if (!dev->data->dev_started) {
1730 		rte_flow_error_set(error,
1731 				   EINVAL,
1732 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
1733 				   NULL,
1734 				   "Device must be started");
1735 		return NULL;
1736 	}
1737 
1738 	flow = rte_zmalloc("bnxt_flow", sizeof(struct rte_flow), 0);
1739 	if (!flow) {
1740 		rte_flow_error_set(error, ENOMEM,
1741 				   RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
1742 				   "Failed to allocate memory");
1743 		return flow;
1744 	}
1745 
1746 	bnxt_acquire_flow_lock(bp);
1747 	ret = bnxt_flow_args_validate(attr, pattern, actions, error);
1748 	if (ret != 0) {
1749 		PMD_DRV_LOG(ERR, "Not a validate flow.\n");
1750 		goto free_flow;
1751 	}
1752 
1753 	filter = bnxt_get_unused_filter(bp);
1754 	if (filter == NULL) {
1755 		rte_flow_error_set(error, ENOSPC,
1756 				   RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
1757 				   "Not enough resources for a new flow");
1758 		goto free_flow;
1759 	}
1760 
1761 	ret = bnxt_validate_and_parse_flow(dev, pattern, actions, attr,
1762 					   error, filter);
1763 	if (ret != 0)
1764 		goto free_filter;
1765 
1766 	ret = bnxt_match_filter(bp, filter);
1767 	if (ret == -EEXIST) {
1768 		PMD_DRV_LOG(DEBUG, "Flow already exists.\n");
1769 		/* Clear the filter that was created as part of
1770 		 * validate_and_parse_flow() above
1771 		 */
1772 		bnxt_hwrm_clear_l2_filter(bp, filter);
1773 		goto free_filter;
1774 	} else if (ret == -EXDEV) {
1775 		PMD_DRV_LOG(DEBUG, "Flow with same pattern exists\n");
1776 		PMD_DRV_LOG(DEBUG, "Updating with different destination\n");
1777 		update_flow = true;
1778 	}
1779 
1780 	/* If tunnel redirection to a VF/PF is specified then only tunnel_type
1781 	 * is set and enable is set to the tunnel type. Issue hwrm cmd directly
1782 	 * in such a case.
1783 	 */
1784 	if (filter->filter_type == HWRM_CFA_TUNNEL_REDIRECT_FILTER &&
1785 	    filter->enables == filter->tunnel_type) {
1786 		ret = bnxt_hwrm_tunnel_redirect_query(bp, &tun_type);
1787 		if (ret) {
1788 			rte_flow_error_set(error, -ret,
1789 					   RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
1790 					   "Unable to query tunnel to VF");
1791 			goto free_filter;
1792 		}
1793 		if (tun_type == (1U << filter->tunnel_type)) {
1794 			ret =
1795 			bnxt_hwrm_tunnel_redirect_free(bp,
1796 						       filter->tunnel_type);
1797 			if (ret) {
1798 				PMD_DRV_LOG(ERR,
1799 					    "Unable to free existing tunnel\n");
1800 				rte_flow_error_set(error, -ret,
1801 						   RTE_FLOW_ERROR_TYPE_HANDLE,
1802 						   NULL,
1803 						   "Unable to free preexisting "
1804 						   "tunnel on VF");
1805 				goto free_filter;
1806 			}
1807 		}
1808 		ret = bnxt_hwrm_tunnel_redirect(bp, filter->tunnel_type);
1809 		if (ret) {
1810 			rte_flow_error_set(error, -ret,
1811 					   RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
1812 					   "Unable to redirect tunnel to VF");
1813 			goto free_filter;
1814 		}
1815 		vnic = &bp->vnic_info[0];
1816 		goto done;
1817 	}
1818 
1819 	if (filter->filter_type == HWRM_CFA_EM_FILTER) {
1820 		filter->enables |=
1821 			HWRM_CFA_EM_FLOW_ALLOC_INPUT_ENABLES_L2_FILTER_ID;
1822 		ret = bnxt_hwrm_set_em_filter(bp, filter->dst_id, filter);
1823 		if (ret != 0) {
1824 			rte_flow_error_set(error, -ret,
1825 					   RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
1826 					   "Failed to create EM filter");
1827 			goto free_filter;
1828 		}
1829 	}
1830 
1831 	if (filter->filter_type == HWRM_CFA_NTUPLE_FILTER) {
1832 		filter->enables |=
1833 			HWRM_CFA_NTUPLE_FILTER_ALLOC_INPUT_ENABLES_L2_FILTER_ID;
1834 		ret = bnxt_hwrm_set_ntuple_filter(bp, filter->dst_id, filter);
1835 		if (ret != 0) {
1836 			rte_flow_error_set(error, -ret,
1837 					   RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
1838 					   "Failed to create ntuple filter");
1839 			goto free_filter;
1840 		}
1841 	}
1842 
1843 	if (BNXT_RFS_NEEDS_VNIC(bp))
1844 		vnic = find_matching_vnic(bp, filter);
1845 	else
1846 		vnic = BNXT_GET_DEFAULT_VNIC(bp);
1847 done:
1848 	if (!ret || update_flow) {
1849 		flow->filter = filter;
1850 		flow->vnic = vnic;
1851 		if (update_flow) {
1852 			ret = -EXDEV;
1853 			goto free_flow;
1854 		}
1855 
1856 		if (filter->valid_flags & BNXT_FLOW_MARK_FLAG) {
1857 			PMD_DRV_LOG(DEBUG,
1858 				    "Mark action: mark id 0x%x, flow id 0x%x\n",
1859 				    filter->mark, filter->flow_id);
1860 
1861 			/* TCAM and EM should be 16-bit only.
1862 			 * Other modes not supported.
1863 			 */
1864 			flow_id = filter->flow_id & BNXT_FLOW_ID_MASK;
1865 			if (bp->mark_table[flow_id].valid) {
1866 				rte_flow_error_set(error, EEXIST,
1867 						   RTE_FLOW_ERROR_TYPE_HANDLE,
1868 						   NULL,
1869 						   "Flow with mark id exists");
1870 				bnxt_clear_one_vnic_filter(bp, filter);
1871 				goto free_filter;
1872 			}
1873 			bp->mark_table[flow_id].valid = true;
1874 			bp->mark_table[flow_id].mark_id = filter->mark;
1875 		}
1876 
1877 		STAILQ_INSERT_TAIL(&vnic->filter, filter, next);
1878 		STAILQ_INSERT_TAIL(&vnic->flow_list, flow, next);
1879 
1880 		if (BNXT_FLOW_XSTATS_EN(bp))
1881 			bp->flow_stat->flow_count++;
1882 		bnxt_release_flow_lock(bp);
1883 		bnxt_setup_flow_counter(bp);
1884 		PMD_DRV_LOG(DEBUG, "Successfully created flow.\n");
1885 		return flow;
1886 	}
1887 
1888 free_filter:
1889 	bnxt_free_filter(bp, filter);
1890 free_flow:
1891 	if (ret == -EEXIST)
1892 		rte_flow_error_set(error, ret,
1893 				   RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
1894 				   "Matching Flow exists.");
1895 	else if (ret == -EXDEV)
1896 		rte_flow_error_set(error, 0,
1897 				   RTE_FLOW_ERROR_TYPE_NONE, NULL,
1898 				   "Flow with pattern exists, updating destination queue");
1899 	else if (!rte_errno)
1900 		rte_flow_error_set(error, -ret,
1901 				   RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
1902 				   "Failed to create flow.");
1903 	rte_free(flow);
1904 	flow = NULL;
1905 	bnxt_release_flow_lock(bp);
1906 	return flow;
1907 }
1908 
1909 static int bnxt_handle_tunnel_redirect_destroy(struct bnxt *bp,
1910 					       struct bnxt_filter_info *filter,
1911 					       struct rte_flow_error *error)
1912 {
1913 	uint16_t tun_dst_fid;
1914 	uint32_t tun_type;
1915 	int ret = 0;
1916 
1917 	ret = bnxt_hwrm_tunnel_redirect_query(bp, &tun_type);
1918 	if (ret) {
1919 		rte_flow_error_set(error, -ret,
1920 				   RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
1921 				   "Unable to query tunnel to VF");
1922 		return ret;
1923 	}
1924 	if (tun_type == (1U << filter->tunnel_type)) {
1925 		ret = bnxt_hwrm_tunnel_redirect_info(bp, filter->tunnel_type,
1926 						     &tun_dst_fid);
1927 		if (ret) {
1928 			rte_flow_error_set(error, -ret,
1929 					   RTE_FLOW_ERROR_TYPE_HANDLE,
1930 					   NULL,
1931 					   "tunnel_redirect info cmd fail");
1932 			return ret;
1933 		}
1934 		PMD_DRV_LOG(INFO, "Pre-existing tunnel fid = %x vf->fid = %x\n",
1935 			    tun_dst_fid + bp->first_vf_id, bp->fw_fid);
1936 
1937 		/* Tunnel doesn't belong to this VF, so don't send HWRM
1938 		 * cmd, just delete the flow from driver
1939 		 */
1940 		if (bp->fw_fid != (tun_dst_fid + bp->first_vf_id))
1941 			PMD_DRV_LOG(ERR,
1942 				    "Tunnel does not belong to this VF, skip hwrm_tunnel_redirect_free\n");
1943 		else
1944 			ret = bnxt_hwrm_tunnel_redirect_free(bp,
1945 							filter->tunnel_type);
1946 	}
1947 	return ret;
1948 }
1949 
1950 static int
1951 _bnxt_flow_destroy(struct bnxt *bp,
1952 		   struct rte_flow *flow,
1953 		    struct rte_flow_error *error)
1954 {
1955 	struct bnxt_filter_info *filter;
1956 	struct bnxt_vnic_info *vnic;
1957 	int ret = 0;
1958 	uint32_t flow_id;
1959 
1960 	filter = flow->filter;
1961 	vnic = flow->vnic;
1962 
1963 	if (filter->filter_type == HWRM_CFA_TUNNEL_REDIRECT_FILTER &&
1964 	    filter->enables == filter->tunnel_type) {
1965 		ret = bnxt_handle_tunnel_redirect_destroy(bp, filter, error);
1966 		if (!ret)
1967 			goto done;
1968 		else
1969 			return ret;
1970 	}
1971 
1972 	ret = bnxt_match_filter(bp, filter);
1973 	if (ret == 0)
1974 		PMD_DRV_LOG(ERR, "Could not find matching flow\n");
1975 
1976 	if (filter->valid_flags & BNXT_FLOW_MARK_FLAG) {
1977 		flow_id = filter->flow_id & BNXT_FLOW_ID_MASK;
1978 		memset(&bp->mark_table[flow_id], 0,
1979 		       sizeof(bp->mark_table[flow_id]));
1980 		filter->flow_id = 0;
1981 	}
1982 
1983 	ret = bnxt_clear_one_vnic_filter(bp, filter);
1984 
1985 done:
1986 	if (!ret) {
1987 		/* If it is a L2 drop filter, when the filter is created,
1988 		 * the FW updates the BC/MC records.
1989 		 * Once this filter is removed, issue the set_rx_mask command
1990 		 * to reset the BC/MC records in the HW to the settings
1991 		 * before the drop counter is created.
1992 		 */
1993 		if (filter->valid_flags & BNXT_FLOW_L2_DROP_FLAG)
1994 			bnxt_set_rx_mask_no_vlan(bp, &bp->vnic_info[0]);
1995 
1996 		STAILQ_REMOVE(&vnic->filter, filter, bnxt_filter_info, next);
1997 		bnxt_free_filter(bp, filter);
1998 		STAILQ_REMOVE(&vnic->flow_list, flow, rte_flow, next);
1999 		rte_free(flow);
2000 		if (BNXT_FLOW_XSTATS_EN(bp))
2001 			bp->flow_stat->flow_count--;
2002 
2003 		/* If this was the last flow associated with this vnic,
2004 		 * switch the queue back to RSS pool.
2005 		 */
2006 		if (vnic && !vnic->func_default &&
2007 		    STAILQ_EMPTY(&vnic->flow_list)) {
2008 			rte_free(vnic->fw_grp_ids);
2009 			if (vnic->rx_queue_cnt > 1)
2010 				bnxt_hwrm_vnic_ctx_free(bp, vnic);
2011 
2012 			bnxt_hwrm_vnic_free(bp, vnic);
2013 			vnic->rx_queue_cnt = 0;
2014 		}
2015 	} else {
2016 		rte_flow_error_set(error, -ret,
2017 				   RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
2018 				   "Failed to destroy flow.");
2019 	}
2020 
2021 	return ret;
2022 }
2023 
2024 static int
2025 bnxt_flow_destroy(struct rte_eth_dev *dev,
2026 		  struct rte_flow *flow,
2027 		  struct rte_flow_error *error)
2028 {
2029 	struct bnxt *bp = dev->data->dev_private;
2030 	int ret = 0;
2031 
2032 	bnxt_acquire_flow_lock(bp);
2033 	if (!flow) {
2034 		rte_flow_error_set(error, EINVAL,
2035 				   RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
2036 				   "Invalid flow: failed to destroy flow.");
2037 		bnxt_release_flow_lock(bp);
2038 		return -EINVAL;
2039 	}
2040 
2041 	if (!flow->filter) {
2042 		rte_flow_error_set(error, EINVAL,
2043 				   RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
2044 				   "Invalid flow: failed to destroy flow.");
2045 		bnxt_release_flow_lock(bp);
2046 		return -EINVAL;
2047 	}
2048 	ret = _bnxt_flow_destroy(bp, flow, error);
2049 	bnxt_release_flow_lock(bp);
2050 
2051 	return ret;
2052 }
2053 
2054 void bnxt_cancel_fc_thread(struct bnxt *bp)
2055 {
2056 	bp->flags &= ~BNXT_FLAG_FC_THREAD;
2057 	rte_eal_alarm_cancel(bnxt_flow_cnt_alarm_cb, (void *)bp);
2058 }
2059 
2060 static int
2061 bnxt_flow_flush(struct rte_eth_dev *dev, struct rte_flow_error *error)
2062 {
2063 	struct bnxt *bp = dev->data->dev_private;
2064 	struct bnxt_vnic_info *vnic;
2065 	struct rte_flow *flow;
2066 	unsigned int i;
2067 	int ret = 0;
2068 
2069 	bnxt_acquire_flow_lock(bp);
2070 	for (i = 0; i < bp->max_vnics; i++) {
2071 		vnic = &bp->vnic_info[i];
2072 		if (vnic && vnic->fw_vnic_id == INVALID_VNIC_ID)
2073 			continue;
2074 
2075 		while (!STAILQ_EMPTY(&vnic->flow_list)) {
2076 			flow = STAILQ_FIRST(&vnic->flow_list);
2077 
2078 			if (!flow->filter)
2079 				continue;
2080 
2081 			ret = _bnxt_flow_destroy(bp, flow, error);
2082 			if (ret)
2083 				break;
2084 		}
2085 	}
2086 
2087 	bnxt_cancel_fc_thread(bp);
2088 	bnxt_release_flow_lock(bp);
2089 
2090 	return ret;
2091 }
2092 
2093 const struct rte_flow_ops bnxt_flow_ops = {
2094 	.validate = bnxt_flow_validate,
2095 	.create = bnxt_flow_create,
2096 	.destroy = bnxt_flow_destroy,
2097 	.flush = bnxt_flow_flush,
2098 };
2099