xref: /dpdk/drivers/net/bnxt/bnxt_flow.c (revision dc348f2e81a94dd3b8a32c2f882483227796905d)
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(const struct rte_flow_attr *attr,
130 				  const struct rte_flow_item pattern[],
131 				  struct rte_flow_error *error,
132 				  struct bnxt_filter_info *filter)
133 {
134 	const struct rte_flow_item *item = bnxt_flow_non_void_item(pattern);
135 	const struct rte_flow_item_vlan *vlan_spec, *vlan_mask;
136 	const struct rte_flow_item_ipv4 *ipv4_spec, *ipv4_mask;
137 	const struct rte_flow_item_ipv6 *ipv6_spec, *ipv6_mask;
138 	const struct rte_flow_item_tcp *tcp_spec, *tcp_mask;
139 	const struct rte_flow_item_udp *udp_spec, *udp_mask;
140 	const struct rte_flow_item_eth *eth_spec, *eth_mask;
141 	const struct rte_ether_addr *dst, *src;
142 	const struct rte_flow_item_nvgre *nvgre_spec;
143 	const struct rte_flow_item_nvgre *nvgre_mask;
144 	const struct rte_flow_item_gre *gre_spec;
145 	const struct rte_flow_item_gre *gre_mask;
146 	const struct rte_flow_item_vxlan *vxlan_spec;
147 	const struct rte_flow_item_vxlan *vxlan_mask;
148 	uint8_t vni_mask[] = {0xFF, 0xFF, 0xFF};
149 	uint8_t tni_mask[] = {0xFF, 0xFF, 0xFF};
150 	uint32_t tenant_id_be = 0, valid_flags = 0;
151 	bool vni_masked = 0;
152 	bool tni_masked = 0;
153 	uint32_t en_ethertype;
154 	uint8_t inner = 0;
155 	uint32_t en = 0;
156 	int use_ntuple;
157 
158 	use_ntuple = bnxt_filter_type_check(pattern, error);
159 	if (use_ntuple < 0)
160 		return use_ntuple;
161 	PMD_DRV_LOG(DEBUG, "Use NTUPLE %d\n", use_ntuple);
162 
163 	filter->filter_type = use_ntuple ?
164 		HWRM_CFA_NTUPLE_FILTER : HWRM_CFA_L2_FILTER;
165 	en_ethertype = use_ntuple ?
166 		NTUPLE_FLTR_ALLOC_INPUT_EN_ETHERTYPE :
167 		EM_FLOW_ALLOC_INPUT_EN_ETHERTYPE;
168 
169 	while (item->type != RTE_FLOW_ITEM_TYPE_END) {
170 		if (item->last) {
171 			/* last or range is NOT supported as match criteria */
172 			rte_flow_error_set(error, EINVAL,
173 					   RTE_FLOW_ERROR_TYPE_ITEM,
174 					   item,
175 					   "No support for range");
176 			return -rte_errno;
177 		}
178 
179 		switch (item->type) {
180 		case RTE_FLOW_ITEM_TYPE_ANY:
181 			inner =
182 			((const struct rte_flow_item_any *)item->spec)->num > 3;
183 			if (inner)
184 				PMD_DRV_LOG(DEBUG, "Parse inner header\n");
185 			break;
186 		case RTE_FLOW_ITEM_TYPE_ETH:
187 			if (!item->spec)
188 				break;
189 
190 			eth_spec = item->spec;
191 
192 			if (item->mask)
193 				eth_mask = item->mask;
194 			else
195 				eth_mask = &rte_flow_item_eth_mask;
196 
197 			/* Source MAC address mask cannot be partially set.
198 			 * Should be All 0's or all 1's.
199 			 * Destination MAC address mask must not be partially
200 			 * set. Should be all 1's or all 0's.
201 			 */
202 			if ((!rte_is_zero_ether_addr(&eth_mask->src) &&
203 			     !rte_is_broadcast_ether_addr(&eth_mask->src)) ||
204 			    (!rte_is_zero_ether_addr(&eth_mask->dst) &&
205 			     !rte_is_broadcast_ether_addr(&eth_mask->dst))) {
206 				rte_flow_error_set(error,
207 						   EINVAL,
208 						   RTE_FLOW_ERROR_TYPE_ITEM,
209 						   item,
210 						   "MAC_addr mask not valid");
211 				return -rte_errno;
212 			}
213 
214 			/* Mask is not allowed. Only exact matches are */
215 			if (eth_mask->type &&
216 			    eth_mask->type != RTE_BE16(0xffff)) {
217 				rte_flow_error_set(error, EINVAL,
218 						   RTE_FLOW_ERROR_TYPE_ITEM,
219 						   item,
220 						   "ethertype mask not valid");
221 				return -rte_errno;
222 			}
223 
224 			if (rte_is_broadcast_ether_addr(&eth_mask->dst)) {
225 				dst = &eth_spec->dst;
226 				if (!rte_is_valid_assigned_ether_addr(dst)) {
227 					rte_flow_error_set(error,
228 							   EINVAL,
229 							   RTE_FLOW_ERROR_TYPE_ITEM,
230 							   item,
231 							   "DMAC is invalid");
232 					PMD_DRV_LOG(ERR,
233 						    "DMAC is invalid!\n");
234 					return -rte_errno;
235 				}
236 				rte_memcpy(filter->dst_macaddr,
237 					   &eth_spec->dst, RTE_ETHER_ADDR_LEN);
238 				en |= use_ntuple ?
239 					NTUPLE_FLTR_ALLOC_INPUT_EN_DST_MACADDR :
240 					EM_FLOW_ALLOC_INPUT_EN_DST_MACADDR;
241 				valid_flags |= inner ?
242 					BNXT_FLOW_L2_INNER_DST_VALID_FLAG :
243 					BNXT_FLOW_L2_DST_VALID_FLAG;
244 				filter->priority = attr->priority;
245 				PMD_DRV_LOG(DEBUG,
246 					    "Creating a priority flow\n");
247 			}
248 			if (rte_is_broadcast_ether_addr(&eth_mask->src)) {
249 				src = &eth_spec->src;
250 				if (!rte_is_valid_assigned_ether_addr(src)) {
251 					rte_flow_error_set(error,
252 							   EINVAL,
253 							   RTE_FLOW_ERROR_TYPE_ITEM,
254 							   item,
255 							   "SMAC is invalid");
256 					PMD_DRV_LOG(ERR,
257 						    "SMAC is invalid!\n");
258 					return -rte_errno;
259 				}
260 				rte_memcpy(filter->src_macaddr,
261 					   &eth_spec->src, RTE_ETHER_ADDR_LEN);
262 				en |= use_ntuple ?
263 					NTUPLE_FLTR_ALLOC_INPUT_EN_SRC_MACADDR :
264 					EM_FLOW_ALLOC_INPUT_EN_SRC_MACADDR;
265 				valid_flags |= inner ?
266 					BNXT_FLOW_L2_INNER_SRC_VALID_FLAG :
267 					BNXT_FLOW_L2_SRC_VALID_FLAG;
268 			} /*
269 			   * else {
270 			   *  PMD_DRV_LOG(ERR, "Handle this condition\n");
271 			   * }
272 			   */
273 			if (eth_mask->type) {
274 				filter->ethertype =
275 					rte_be_to_cpu_16(eth_spec->type);
276 				en |= en_ethertype;
277 			}
278 			if (inner)
279 				valid_flags |= BNXT_FLOW_PARSE_INNER_FLAG;
280 
281 			break;
282 		case RTE_FLOW_ITEM_TYPE_VLAN:
283 			vlan_spec = item->spec;
284 
285 			if (item->mask)
286 				vlan_mask = item->mask;
287 			else
288 				vlan_mask = &rte_flow_item_vlan_mask;
289 
290 			if (en & en_ethertype) {
291 				rte_flow_error_set(error, EINVAL,
292 						   RTE_FLOW_ERROR_TYPE_ITEM,
293 						   item,
294 						   "VLAN TPID matching is not"
295 						   " supported");
296 				return -rte_errno;
297 			}
298 			if (vlan_mask->tci &&
299 			    vlan_mask->tci == RTE_BE16(0x0fff)) {
300 				/* Only the VLAN ID can be matched. */
301 				filter->l2_ovlan =
302 					rte_be_to_cpu_16(vlan_spec->tci &
303 							 RTE_BE16(0x0fff));
304 				en |= EM_FLOW_ALLOC_INPUT_EN_OVLAN_VID;
305 			} else {
306 				rte_flow_error_set(error,
307 						   EINVAL,
308 						   RTE_FLOW_ERROR_TYPE_ITEM,
309 						   item,
310 						   "VLAN mask is invalid");
311 				return -rte_errno;
312 			}
313 			if (vlan_mask->inner_type &&
314 			    vlan_mask->inner_type != RTE_BE16(0xffff)) {
315 				rte_flow_error_set(error, EINVAL,
316 						   RTE_FLOW_ERROR_TYPE_ITEM,
317 						   item,
318 						   "inner ethertype mask not"
319 						   " valid");
320 				return -rte_errno;
321 			}
322 			if (vlan_mask->inner_type) {
323 				filter->ethertype =
324 					rte_be_to_cpu_16(vlan_spec->inner_type);
325 				en |= en_ethertype;
326 			}
327 
328 			break;
329 		case RTE_FLOW_ITEM_TYPE_IPV4:
330 			/* If mask is not involved, we could use EM filters. */
331 			ipv4_spec = item->spec;
332 
333 			if (!item->spec)
334 				break;
335 
336 			if (item->mask)
337 				ipv4_mask = item->mask;
338 			else
339 				ipv4_mask = &rte_flow_item_ipv4_mask;
340 
341 			/* Only IP DST and SRC fields are maskable. */
342 			if (ipv4_mask->hdr.version_ihl ||
343 			    ipv4_mask->hdr.type_of_service ||
344 			    ipv4_mask->hdr.total_length ||
345 			    ipv4_mask->hdr.packet_id ||
346 			    ipv4_mask->hdr.fragment_offset ||
347 			    ipv4_mask->hdr.time_to_live ||
348 			    ipv4_mask->hdr.next_proto_id ||
349 			    ipv4_mask->hdr.hdr_checksum) {
350 				rte_flow_error_set(error,
351 						   EINVAL,
352 						   RTE_FLOW_ERROR_TYPE_ITEM,
353 						   item,
354 						   "Invalid IPv4 mask.");
355 				return -rte_errno;
356 			}
357 
358 			filter->dst_ipaddr[0] = ipv4_spec->hdr.dst_addr;
359 			filter->src_ipaddr[0] = ipv4_spec->hdr.src_addr;
360 
361 			if (use_ntuple)
362 				en |= NTUPLE_FLTR_ALLOC_INPUT_EN_SRC_IPADDR |
363 					NTUPLE_FLTR_ALLOC_INPUT_EN_DST_IPADDR;
364 			else
365 				en |= EM_FLOW_ALLOC_INPUT_EN_SRC_IPADDR |
366 					EM_FLOW_ALLOC_INPUT_EN_DST_IPADDR;
367 
368 			if (ipv4_mask->hdr.src_addr) {
369 				filter->src_ipaddr_mask[0] =
370 					ipv4_mask->hdr.src_addr;
371 				en |= !use_ntuple ? 0 :
372 				     NTUPLE_FLTR_ALLOC_INPUT_EN_SRC_IPADDR_MASK;
373 			}
374 
375 			if (ipv4_mask->hdr.dst_addr) {
376 				filter->dst_ipaddr_mask[0] =
377 					ipv4_mask->hdr.dst_addr;
378 				en |= !use_ntuple ? 0 :
379 				     NTUPLE_FLTR_ALLOC_INPUT_EN_DST_IPADDR_MASK;
380 			}
381 
382 			filter->ip_addr_type = use_ntuple ?
383 			 HWRM_CFA_NTUPLE_FILTER_ALLOC_INPUT_IP_ADDR_TYPE_IPV4 :
384 			 HWRM_CFA_EM_FLOW_ALLOC_INPUT_IP_ADDR_TYPE_IPV4;
385 
386 			if (ipv4_spec->hdr.next_proto_id) {
387 				filter->ip_protocol =
388 					ipv4_spec->hdr.next_proto_id;
389 				if (use_ntuple)
390 					en |= NTUPLE_FLTR_ALLOC_IN_EN_IP_PROTO;
391 				else
392 					en |= EM_FLOW_ALLOC_INPUT_EN_IP_PROTO;
393 			}
394 			break;
395 		case RTE_FLOW_ITEM_TYPE_IPV6:
396 			ipv6_spec = item->spec;
397 
398 			if (!item->spec)
399 				break;
400 
401 			if (item->mask)
402 				ipv6_mask = item->mask;
403 			else
404 				ipv6_mask = &rte_flow_item_ipv6_mask;
405 
406 			/* Only IP DST and SRC fields are maskable. */
407 			if (ipv6_mask->hdr.vtc_flow ||
408 			    ipv6_mask->hdr.payload_len ||
409 			    ipv6_mask->hdr.proto ||
410 			    ipv6_mask->hdr.hop_limits) {
411 				rte_flow_error_set(error,
412 						   EINVAL,
413 						   RTE_FLOW_ERROR_TYPE_ITEM,
414 						   item,
415 						   "Invalid IPv6 mask.");
416 				return -rte_errno;
417 			}
418 
419 			if (use_ntuple)
420 				en |= NTUPLE_FLTR_ALLOC_INPUT_EN_SRC_IPADDR |
421 					NTUPLE_FLTR_ALLOC_INPUT_EN_DST_IPADDR;
422 			else
423 				en |= EM_FLOW_ALLOC_INPUT_EN_SRC_IPADDR |
424 					EM_FLOW_ALLOC_INPUT_EN_DST_IPADDR;
425 
426 			rte_memcpy(filter->src_ipaddr,
427 				   ipv6_spec->hdr.src_addr, 16);
428 			rte_memcpy(filter->dst_ipaddr,
429 				   ipv6_spec->hdr.dst_addr, 16);
430 
431 			if (!bnxt_check_zero_bytes(ipv6_mask->hdr.src_addr,
432 						   16)) {
433 				rte_memcpy(filter->src_ipaddr_mask,
434 					   ipv6_mask->hdr.src_addr, 16);
435 				en |= !use_ntuple ? 0 :
436 				    NTUPLE_FLTR_ALLOC_INPUT_EN_SRC_IPADDR_MASK;
437 			}
438 
439 			if (!bnxt_check_zero_bytes(ipv6_mask->hdr.dst_addr,
440 						   16)) {
441 				rte_memcpy(filter->dst_ipaddr_mask,
442 					   ipv6_mask->hdr.dst_addr, 16);
443 				en |= !use_ntuple ? 0 :
444 				     NTUPLE_FLTR_ALLOC_INPUT_EN_DST_IPADDR_MASK;
445 			}
446 
447 			filter->ip_addr_type = use_ntuple ?
448 				NTUPLE_FLTR_ALLOC_INPUT_IP_ADDR_TYPE_IPV6 :
449 				EM_FLOW_ALLOC_INPUT_IP_ADDR_TYPE_IPV6;
450 			break;
451 		case RTE_FLOW_ITEM_TYPE_TCP:
452 			tcp_spec = item->spec;
453 
454 			if (!item->spec)
455 				break;
456 
457 			if (item->mask)
458 				tcp_mask = item->mask;
459 			else
460 				tcp_mask = &rte_flow_item_tcp_mask;
461 
462 			/* Check TCP mask. Only DST & SRC ports are maskable */
463 			if (tcp_mask->hdr.sent_seq ||
464 			    tcp_mask->hdr.recv_ack ||
465 			    tcp_mask->hdr.data_off ||
466 			    tcp_mask->hdr.tcp_flags ||
467 			    tcp_mask->hdr.rx_win ||
468 			    tcp_mask->hdr.cksum ||
469 			    tcp_mask->hdr.tcp_urp) {
470 				rte_flow_error_set(error,
471 						   EINVAL,
472 						   RTE_FLOW_ERROR_TYPE_ITEM,
473 						   item,
474 						   "Invalid TCP mask");
475 				return -rte_errno;
476 			}
477 
478 			filter->src_port = tcp_spec->hdr.src_port;
479 			filter->dst_port = tcp_spec->hdr.dst_port;
480 
481 			if (use_ntuple)
482 				en |= NTUPLE_FLTR_ALLOC_INPUT_EN_SRC_PORT |
483 					NTUPLE_FLTR_ALLOC_INPUT_EN_DST_PORT;
484 			else
485 				en |= EM_FLOW_ALLOC_INPUT_EN_SRC_PORT |
486 					EM_FLOW_ALLOC_INPUT_EN_DST_PORT;
487 
488 			if (tcp_mask->hdr.dst_port) {
489 				filter->dst_port_mask = tcp_mask->hdr.dst_port;
490 				en |= !use_ntuple ? 0 :
491 				  NTUPLE_FLTR_ALLOC_INPUT_EN_DST_PORT_MASK;
492 			}
493 
494 			if (tcp_mask->hdr.src_port) {
495 				filter->src_port_mask = tcp_mask->hdr.src_port;
496 				en |= !use_ntuple ? 0 :
497 				  NTUPLE_FLTR_ALLOC_INPUT_EN_SRC_PORT_MASK;
498 			}
499 			break;
500 		case RTE_FLOW_ITEM_TYPE_UDP:
501 			udp_spec = item->spec;
502 
503 			if (!item->spec)
504 				break;
505 
506 			if (item->mask)
507 				udp_mask = item->mask;
508 			else
509 				udp_mask = &rte_flow_item_udp_mask;
510 
511 			if (udp_mask->hdr.dgram_len ||
512 			    udp_mask->hdr.dgram_cksum) {
513 				rte_flow_error_set(error,
514 						   EINVAL,
515 						   RTE_FLOW_ERROR_TYPE_ITEM,
516 						   item,
517 						   "Invalid UDP mask");
518 				return -rte_errno;
519 			}
520 
521 			filter->src_port = udp_spec->hdr.src_port;
522 			filter->dst_port = udp_spec->hdr.dst_port;
523 
524 			if (use_ntuple)
525 				en |= NTUPLE_FLTR_ALLOC_INPUT_EN_SRC_PORT |
526 					NTUPLE_FLTR_ALLOC_INPUT_EN_DST_PORT;
527 			else
528 				en |= EM_FLOW_ALLOC_INPUT_EN_SRC_PORT |
529 					EM_FLOW_ALLOC_INPUT_EN_DST_PORT;
530 
531 			if (udp_mask->hdr.dst_port) {
532 				filter->dst_port_mask = udp_mask->hdr.dst_port;
533 				en |= !use_ntuple ? 0 :
534 				  NTUPLE_FLTR_ALLOC_INPUT_EN_DST_PORT_MASK;
535 			}
536 
537 			if (udp_mask->hdr.src_port) {
538 				filter->src_port_mask = udp_mask->hdr.src_port;
539 				en |= !use_ntuple ? 0 :
540 				  NTUPLE_FLTR_ALLOC_INPUT_EN_SRC_PORT_MASK;
541 			}
542 			break;
543 		case RTE_FLOW_ITEM_TYPE_VXLAN:
544 			vxlan_spec = item->spec;
545 			vxlan_mask = item->mask;
546 			/* Check if VXLAN item is used to describe protocol.
547 			 * If yes, both spec and mask should be NULL.
548 			 * If no, both spec and mask shouldn't be NULL.
549 			 */
550 			if ((!vxlan_spec && vxlan_mask) ||
551 			    (vxlan_spec && !vxlan_mask)) {
552 				rte_flow_error_set(error,
553 						   EINVAL,
554 						   RTE_FLOW_ERROR_TYPE_ITEM,
555 						   item,
556 						   "Invalid VXLAN item");
557 				return -rte_errno;
558 			}
559 
560 			if (!vxlan_spec && !vxlan_mask) {
561 				filter->tunnel_type =
562 				CFA_NTUPLE_FILTER_ALLOC_REQ_TUNNEL_TYPE_VXLAN;
563 				break;
564 			}
565 
566 			if (vxlan_spec->rsvd1 || vxlan_spec->rsvd0[0] ||
567 			    vxlan_spec->rsvd0[1] || vxlan_spec->rsvd0[2] ||
568 			    vxlan_spec->flags != 0x8) {
569 				rte_flow_error_set(error,
570 						   EINVAL,
571 						   RTE_FLOW_ERROR_TYPE_ITEM,
572 						   item,
573 						   "Invalid VXLAN item");
574 				return -rte_errno;
575 			}
576 
577 			/* Check if VNI is masked. */
578 			if (vxlan_mask != NULL) {
579 				vni_masked =
580 					!!memcmp(vxlan_mask->vni, vni_mask,
581 						 RTE_DIM(vni_mask));
582 				if (vni_masked) {
583 					rte_flow_error_set
584 						(error,
585 						 EINVAL,
586 						 RTE_FLOW_ERROR_TYPE_ITEM,
587 						 item,
588 						 "Invalid VNI mask");
589 					return -rte_errno;
590 				}
591 
592 				rte_memcpy(((uint8_t *)&tenant_id_be + 1),
593 					   vxlan_spec->vni, 3);
594 				filter->vni =
595 					rte_be_to_cpu_32(tenant_id_be);
596 				filter->tunnel_type =
597 				 CFA_NTUPLE_FILTER_ALLOC_REQ_TUNNEL_TYPE_VXLAN;
598 			}
599 			break;
600 		case RTE_FLOW_ITEM_TYPE_NVGRE:
601 			nvgre_spec = item->spec;
602 			nvgre_mask = item->mask;
603 			/* Check if NVGRE item is used to describe protocol.
604 			 * If yes, both spec and mask should be NULL.
605 			 * If no, both spec and mask shouldn't be NULL.
606 			 */
607 			if ((!nvgre_spec && nvgre_mask) ||
608 			    (nvgre_spec && !nvgre_mask)) {
609 				rte_flow_error_set(error,
610 						   EINVAL,
611 						   RTE_FLOW_ERROR_TYPE_ITEM,
612 						   item,
613 						   "Invalid NVGRE item");
614 				return -rte_errno;
615 			}
616 
617 			if (!nvgre_spec && !nvgre_mask) {
618 				filter->tunnel_type =
619 				CFA_NTUPLE_FILTER_ALLOC_REQ_TUNNEL_TYPE_NVGRE;
620 				break;
621 			}
622 
623 			if (nvgre_spec->c_k_s_rsvd0_ver != 0x2000 ||
624 			    nvgre_spec->protocol != 0x6558) {
625 				rte_flow_error_set(error,
626 						   EINVAL,
627 						   RTE_FLOW_ERROR_TYPE_ITEM,
628 						   item,
629 						   "Invalid NVGRE item");
630 				return -rte_errno;
631 			}
632 
633 			if (nvgre_spec && nvgre_mask) {
634 				tni_masked =
635 					!!memcmp(nvgre_mask->tni, tni_mask,
636 						 RTE_DIM(tni_mask));
637 				if (tni_masked) {
638 					rte_flow_error_set
639 						(error,
640 						 EINVAL,
641 						 RTE_FLOW_ERROR_TYPE_ITEM,
642 						 item,
643 						 "Invalid TNI mask");
644 					return -rte_errno;
645 				}
646 				rte_memcpy(((uint8_t *)&tenant_id_be + 1),
647 					   nvgre_spec->tni, 3);
648 				filter->vni =
649 					rte_be_to_cpu_32(tenant_id_be);
650 				filter->tunnel_type =
651 				 CFA_NTUPLE_FILTER_ALLOC_REQ_TUNNEL_TYPE_NVGRE;
652 			}
653 			break;
654 
655 		case RTE_FLOW_ITEM_TYPE_GRE:
656 			gre_spec = (const struct rte_flow_item_gre *)item->spec;
657 			gre_mask = (const struct rte_flow_item_gre *)item->mask;
658 
659 			/*
660 			 *Check if GRE item is used to describe protocol.
661 			 * If yes, both spec and mask should be NULL.
662 			 * If no, both spec and mask shouldn't be NULL.
663 			 */
664 			if (!!gre_spec ^ !!gre_mask) {
665 				rte_flow_error_set(error, EINVAL,
666 						   RTE_FLOW_ERROR_TYPE_ITEM,
667 						   item,
668 						   "Invalid GRE item");
669 				return -rte_errno;
670 			}
671 
672 			if (!gre_spec && !gre_mask) {
673 				filter->tunnel_type =
674 				CFA_NTUPLE_FILTER_ALLOC_REQ_TUNNEL_TYPE_IPGRE;
675 				break;
676 			}
677 			break;
678 
679 		default:
680 			break;
681 		}
682 		item++;
683 	}
684 	filter->enables = en;
685 	filter->valid_flags = valid_flags;
686 
687 	/* Items parsed but no filter to create in HW. */
688 	if (filter->enables == 0 && filter->valid_flags == 0)
689 		filter->filter_type = HWRM_CFA_CONFIG;
690 
691 	return 0;
692 }
693 
694 /* Parse attributes */
695 static int
696 bnxt_flow_parse_attr(const struct rte_flow_attr *attr,
697 		     struct rte_flow_error *error)
698 {
699 	/* Must be input direction */
700 	if (!attr->ingress) {
701 		rte_flow_error_set(error,
702 				   EINVAL,
703 				   RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
704 				   attr,
705 				   "Only support ingress.");
706 		return -rte_errno;
707 	}
708 
709 	/* Not supported */
710 	if (attr->egress) {
711 		rte_flow_error_set(error,
712 				   EINVAL,
713 				   RTE_FLOW_ERROR_TYPE_ATTR_EGRESS,
714 				   attr,
715 				   "No support for egress.");
716 		return -rte_errno;
717 	}
718 
719 	return 0;
720 }
721 
722 static struct bnxt_filter_info *
723 bnxt_find_matching_l2_filter(struct bnxt *bp, struct bnxt_filter_info *nf)
724 {
725 	struct bnxt_filter_info *mf, *f0;
726 	struct bnxt_vnic_info *vnic0;
727 	int i;
728 
729 	vnic0 = BNXT_GET_DEFAULT_VNIC(bp);
730 	f0 = STAILQ_FIRST(&vnic0->filter);
731 
732 	/* This flow has same DST MAC as the port/l2 filter. */
733 	if (memcmp(f0->l2_addr, nf->dst_macaddr, RTE_ETHER_ADDR_LEN) == 0)
734 		return f0;
735 
736 	for (i = bp->max_vnics - 1; i >= 0; i--) {
737 		struct bnxt_vnic_info *vnic = &bp->vnic_info[i];
738 
739 		if (vnic->fw_vnic_id == INVALID_VNIC_ID)
740 			continue;
741 
742 		STAILQ_FOREACH(mf, &vnic->filter, next) {
743 
744 			if (mf->matching_l2_fltr_ptr)
745 				continue;
746 
747 			if (mf->ethertype == nf->ethertype &&
748 			    mf->l2_ovlan == nf->l2_ovlan &&
749 			    mf->l2_ovlan_mask == nf->l2_ovlan_mask &&
750 			    mf->l2_ivlan == nf->l2_ivlan &&
751 			    mf->l2_ivlan_mask == nf->l2_ivlan_mask &&
752 			    !memcmp(mf->src_macaddr, nf->src_macaddr,
753 				    RTE_ETHER_ADDR_LEN) &&
754 			    !memcmp(mf->dst_macaddr, nf->dst_macaddr,
755 				    RTE_ETHER_ADDR_LEN))
756 				return mf;
757 		}
758 	}
759 	return NULL;
760 }
761 
762 static struct bnxt_filter_info *
763 bnxt_create_l2_filter(struct bnxt *bp, struct bnxt_filter_info *nf,
764 		      struct bnxt_vnic_info *vnic)
765 {
766 	struct bnxt_filter_info *filter1;
767 	int rc;
768 
769 	/* Alloc new L2 filter.
770 	 * This flow needs MAC filter which does not match any existing
771 	 * L2 filters.
772 	 */
773 	filter1 = bnxt_get_unused_filter(bp);
774 	if (filter1 == NULL)
775 		return NULL;
776 
777 	memcpy(filter1, nf, sizeof(*filter1));
778 
779 	filter1->flags = HWRM_CFA_L2_FILTER_ALLOC_INPUT_FLAGS_XDP_DISABLE;
780 	filter1->flags |= HWRM_CFA_L2_FILTER_ALLOC_INPUT_FLAGS_PATH_RX;
781 	if (nf->valid_flags & BNXT_FLOW_L2_SRC_VALID_FLAG ||
782 	    nf->valid_flags & BNXT_FLOW_L2_DST_VALID_FLAG) {
783 		filter1->flags |=
784 			HWRM_CFA_L2_FILTER_ALLOC_INPUT_FLAGS_OUTERMOST;
785 		PMD_DRV_LOG(DEBUG, "Create Outer filter\n");
786 	}
787 
788 	if (nf->filter_type == HWRM_CFA_L2_FILTER &&
789 	    (nf->valid_flags & BNXT_FLOW_L2_SRC_VALID_FLAG ||
790 	     nf->valid_flags & BNXT_FLOW_L2_INNER_SRC_VALID_FLAG)) {
791 		PMD_DRV_LOG(DEBUG, "Create L2 filter for SRC MAC\n");
792 		filter1->flags |=
793 			HWRM_CFA_L2_FILTER_ALLOC_INPUT_FLAGS_SOURCE_VALID;
794 		memcpy(filter1->l2_addr, nf->src_macaddr, RTE_ETHER_ADDR_LEN);
795 	} else {
796 		PMD_DRV_LOG(DEBUG, "Create L2 filter for DST MAC\n");
797 		memcpy(filter1->l2_addr, nf->dst_macaddr, RTE_ETHER_ADDR_LEN);
798 	}
799 
800 	if (nf->priority &&
801 	    (nf->valid_flags & BNXT_FLOW_L2_DST_VALID_FLAG ||
802 	     nf->valid_flags & BNXT_FLOW_L2_INNER_DST_VALID_FLAG)) {
803 		/* Tell the FW where to place the filter in the table. */
804 		if (nf->priority > 65535) {
805 			filter1->pri_hint =
806 			HWRM_CFA_L2_FILTER_ALLOC_INPUT_PRI_HINT_BELOW_FILTER;
807 			/* This will place the filter in TCAM */
808 			filter1->l2_filter_id_hint = (uint64_t)-1;
809 		}
810 	}
811 
812 	if (nf->valid_flags & (BNXT_FLOW_L2_DST_VALID_FLAG |
813 			       BNXT_FLOW_L2_SRC_VALID_FLAG |
814 			       BNXT_FLOW_L2_INNER_SRC_VALID_FLAG |
815 			       BNXT_FLOW_L2_INNER_DST_VALID_FLAG)) {
816 		filter1->enables =
817 			HWRM_CFA_L2_FILTER_ALLOC_INPUT_ENABLES_L2_ADDR |
818 			L2_FILTER_ALLOC_INPUT_EN_L2_ADDR_MASK;
819 		memset(filter1->l2_addr_mask, 0xff, RTE_ETHER_ADDR_LEN);
820 	}
821 
822 	if (nf->valid_flags & BNXT_FLOW_L2_DROP_FLAG) {
823 		filter1->flags |=
824 			HWRM_CFA_L2_FILTER_ALLOC_INPUT_FLAGS_DROP;
825 		if (nf->ethertype == RTE_ETHER_TYPE_IPV4) {
826 			/* Num VLANs for drop filter will/should be 0.
827 			 * If the req is memset to 0, then the count will
828 			 * be automatically set to 0.
829 			 */
830 			if (nf->valid_flags & BNXT_FLOW_PARSE_INNER_FLAG) {
831 				filter1->enables |=
832 					L2_FILTER_ALLOC_INPUT_EN_T_NUM_VLANS;
833 			} else {
834 				filter1->enables |=
835 					L2_FILTER_ALLOC_INPUT_EN_NUM_VLANS;
836 				filter1->flags |=
837 				HWRM_CFA_L2_FILTER_ALLOC_INPUT_FLAGS_OUTERMOST;
838 			}
839 		}
840 	}
841 
842 	rc = bnxt_hwrm_set_l2_filter(bp, vnic->fw_vnic_id,
843 				     filter1);
844 	if (rc) {
845 		bnxt_free_filter(bp, filter1);
846 		return NULL;
847 	}
848 	return filter1;
849 }
850 
851 struct bnxt_filter_info *
852 bnxt_get_l2_filter(struct bnxt *bp, struct bnxt_filter_info *nf,
853 		   struct bnxt_vnic_info *vnic)
854 {
855 	struct bnxt_filter_info *l2_filter = NULL;
856 
857 	l2_filter = bnxt_find_matching_l2_filter(bp, nf);
858 	if (l2_filter) {
859 		l2_filter->l2_ref_cnt++;
860 	} else {
861 		l2_filter = bnxt_create_l2_filter(bp, nf, vnic);
862 		if (l2_filter) {
863 			STAILQ_INSERT_TAIL(&vnic->filter, l2_filter, next);
864 			l2_filter->vnic = vnic;
865 		}
866 	}
867 	nf->matching_l2_fltr_ptr = l2_filter;
868 
869 	return l2_filter;
870 }
871 
872 static void bnxt_vnic_cleanup(struct bnxt *bp, struct bnxt_vnic_info *vnic)
873 {
874 	if (vnic->rx_queue_cnt > 1)
875 		bnxt_hwrm_vnic_ctx_free(bp, vnic);
876 
877 	bnxt_hwrm_vnic_free(bp, vnic);
878 
879 	rte_free(vnic->fw_grp_ids);
880 	vnic->fw_grp_ids = NULL;
881 
882 	vnic->rx_queue_cnt = 0;
883 }
884 
885 static int bnxt_vnic_prep(struct bnxt *bp, struct bnxt_vnic_info *vnic,
886 			  const struct rte_flow_action *act,
887 			  struct rte_flow_error *error)
888 {
889 	struct rte_eth_conf *dev_conf = &bp->eth_dev->data->dev_conf;
890 	uint64_t rx_offloads = dev_conf->rxmode.offloads;
891 	int rc;
892 
893 	if (bp->nr_vnics > bp->max_vnics - 1)
894 		return rte_flow_error_set(error, EINVAL,
895 					  RTE_FLOW_ERROR_TYPE_ATTR_GROUP,
896 					  NULL,
897 					  "Group id is invalid");
898 
899 	rc = bnxt_vnic_grp_alloc(bp, vnic);
900 	if (rc)
901 		return rte_flow_error_set(error, -rc,
902 					  RTE_FLOW_ERROR_TYPE_ACTION,
903 					  act,
904 					  "Failed to alloc VNIC group");
905 
906 	rc = bnxt_hwrm_vnic_alloc(bp, vnic);
907 	if (rc) {
908 		rte_flow_error_set(error, -rc,
909 				   RTE_FLOW_ERROR_TYPE_ACTION,
910 				   act,
911 				   "Failed to alloc VNIC");
912 		goto ret;
913 	}
914 
915 	/* RSS context is required only when there is more than one RSS ring */
916 	if (vnic->rx_queue_cnt > 1) {
917 		rc = bnxt_hwrm_vnic_ctx_alloc(bp, vnic, 0);
918 		if (rc) {
919 			rte_flow_error_set(error, -rc,
920 					   RTE_FLOW_ERROR_TYPE_ACTION,
921 					   act,
922 					   "Failed to alloc VNIC context");
923 			goto ret;
924 		}
925 	}
926 
927 	if (rx_offloads & RTE_ETH_RX_OFFLOAD_VLAN_STRIP)
928 		vnic->vlan_strip = true;
929 	else
930 		vnic->vlan_strip = false;
931 
932 	rc = bnxt_hwrm_vnic_cfg(bp, vnic);
933 	if (rc) {
934 		rte_flow_error_set(error, -rc,
935 				   RTE_FLOW_ERROR_TYPE_ACTION,
936 				   act,
937 				   "Failed to configure VNIC");
938 		goto ret;
939 	}
940 
941 	rc = bnxt_hwrm_vnic_plcmode_cfg(bp, vnic);
942 	if (rc) {
943 		rte_flow_error_set(error, -rc,
944 				   RTE_FLOW_ERROR_TYPE_ACTION,
945 				   act,
946 				   "Failed to configure VNIC plcmode");
947 		goto ret;
948 	}
949 
950 	bp->nr_vnics++;
951 
952 	return 0;
953 
954 ret:
955 	bnxt_vnic_cleanup(bp, vnic);
956 	return rc;
957 }
958 
959 static int match_vnic_rss_cfg(struct bnxt *bp,
960 			      struct bnxt_vnic_info *vnic,
961 			      const struct rte_flow_action_rss *rss)
962 {
963 	unsigned int match = 0, i;
964 
965 	if (vnic->rx_queue_cnt != rss->queue_num)
966 		return -EINVAL;
967 
968 	for (i = 0; i < rss->queue_num; i++) {
969 		if (!bp->rx_queues[rss->queue[i]]->vnic->rx_queue_cnt &&
970 		    !bp->rx_queues[rss->queue[i]]->rx_started)
971 			return -EINVAL;
972 	}
973 
974 	for (i = 0; i < vnic->rx_queue_cnt; i++) {
975 		int j;
976 
977 		for (j = 0; j < vnic->rx_queue_cnt; j++) {
978 			if (bp->grp_info[rss->queue[i]].fw_grp_id ==
979 			    vnic->fw_grp_ids[j])
980 				match++;
981 		}
982 	}
983 
984 	if (match != vnic->rx_queue_cnt) {
985 		PMD_DRV_LOG(ERR,
986 			    "VNIC queue count %d vs queues matched %d\n",
987 			    match, vnic->rx_queue_cnt);
988 		return -EINVAL;
989 	}
990 
991 	return 0;
992 }
993 
994 static void
995 bnxt_update_filter_flags_en(struct bnxt_filter_info *filter,
996 			    struct bnxt_filter_info *filter1,
997 			    int use_ntuple)
998 {
999 	if (!use_ntuple &&
1000 	    !(filter->valid_flags &
1001 	      ~(BNXT_FLOW_L2_DST_VALID_FLAG |
1002 		BNXT_FLOW_L2_SRC_VALID_FLAG |
1003 		BNXT_FLOW_L2_INNER_SRC_VALID_FLAG |
1004 		BNXT_FLOW_L2_INNER_DST_VALID_FLAG |
1005 		BNXT_FLOW_L2_DROP_FLAG |
1006 		BNXT_FLOW_PARSE_INNER_FLAG))) {
1007 		filter->flags = filter1->flags;
1008 		filter->enables = filter1->enables;
1009 		filter->filter_type = HWRM_CFA_L2_FILTER;
1010 		memcpy(filter->l2_addr, filter1->l2_addr, RTE_ETHER_ADDR_LEN);
1011 		memset(filter->l2_addr_mask, 0xff, RTE_ETHER_ADDR_LEN);
1012 		filter->pri_hint = filter1->pri_hint;
1013 		filter->l2_filter_id_hint = filter1->l2_filter_id_hint;
1014 	}
1015 	filter->fw_l2_filter_id = filter1->fw_l2_filter_id;
1016 	filter->l2_ref_cnt = filter1->l2_ref_cnt;
1017 	filter->flow_id = filter1->flow_id;
1018 	PMD_DRV_LOG(DEBUG,
1019 		"l2_filter: %p fw_l2_filter_id %" PRIx64 " l2_ref_cnt %u\n",
1020 		filter1, filter->fw_l2_filter_id, filter->l2_ref_cnt);
1021 }
1022 
1023 static int
1024 bnxt_validate_rss_action(const struct rte_flow_action actions[])
1025 {
1026 	for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
1027 		switch (actions->type) {
1028 		case RTE_FLOW_ACTION_TYPE_VOID:
1029 			break;
1030 		case RTE_FLOW_ACTION_TYPE_RSS:
1031 			break;
1032 		default:
1033 			return -ENOTSUP;
1034 		}
1035 	}
1036 
1037 	return 0;
1038 }
1039 
1040 static int
1041 bnxt_get_vnic(struct bnxt *bp, uint32_t group)
1042 {
1043 	int vnic_id = 0;
1044 
1045 	/* For legacy NS3 based implementations,
1046 	 * group_id will be mapped to a VNIC ID.
1047 	 */
1048 	if (BNXT_STINGRAY(bp))
1049 		vnic_id = group;
1050 
1051 	/* Non NS3 cases, group_id will be ignored.
1052 	 * Setting will be configured on default VNIC.
1053 	 */
1054 	return vnic_id;
1055 }
1056 
1057 static int
1058 bnxt_vnic_rss_cfg_update(struct bnxt *bp,
1059 			 struct bnxt_vnic_info *vnic,
1060 			 const struct rte_flow_action *act,
1061 			 struct rte_flow_error *error)
1062 {
1063 	const struct rte_flow_action_rss *rss;
1064 	unsigned int rss_idx, i, j, fw_idx;
1065 	uint16_t hash_type;
1066 	uint64_t types;
1067 	int rc;
1068 
1069 	rss = (const struct rte_flow_action_rss *)act->conf;
1070 
1071 	/* must specify either all the Rx queues created by application or zero queues */
1072 	if (rss->queue_num && vnic->rx_queue_cnt != rss->queue_num) {
1073 		rte_flow_error_set(error,
1074 				   EINVAL,
1075 				   RTE_FLOW_ERROR_TYPE_ACTION,
1076 				   act,
1077 				   "Incorrect RXQ count");
1078 		rc = -rte_errno;
1079 		goto ret;
1080 	}
1081 
1082 	/* Validate Rx queues */
1083 	for (i = 0; i < rss->queue_num; i++) {
1084 		PMD_DRV_LOG(DEBUG, "RSS action Queue %d\n", rss->queue[i]);
1085 
1086 		if (rss->queue[i] >= bp->rx_nr_rings ||
1087 		    !bp->rx_queues[rss->queue[i]]) {
1088 			rte_flow_error_set(error,
1089 					   EINVAL,
1090 					   RTE_FLOW_ERROR_TYPE_ACTION,
1091 					   act,
1092 					   "Invalid queue ID for RSS");
1093 			rc = -rte_errno;
1094 			goto ret;
1095 		}
1096 	}
1097 
1098 	/* Duplicate queue ids are not supported. */
1099 	for (i = 0; i < rss->queue_num; i++) {
1100 		for (j = i + 1; j < rss->queue_num; j++) {
1101 			if (rss->queue[i] == rss->queue[j]) {
1102 				rte_flow_error_set(error,
1103 						   EINVAL,
1104 						   RTE_FLOW_ERROR_TYPE_ACTION,
1105 						   act,
1106 						   "Duplicate queue ID for RSS");
1107 				rc = -rte_errno;
1108 				goto ret;
1109 			}
1110 		}
1111 	}
1112 
1113 	/* Currently only Toeplitz hash is supported. */
1114 	if (rss->func != RTE_ETH_HASH_FUNCTION_DEFAULT &&
1115 	    rss->func != RTE_ETH_HASH_FUNCTION_TOEPLITZ) {
1116 		rte_flow_error_set(error,
1117 				   ENOTSUP,
1118 				   RTE_FLOW_ERROR_TYPE_ACTION,
1119 				   act,
1120 				   "Unsupported RSS hash function");
1121 		rc = -rte_errno;
1122 		goto ret;
1123 	}
1124 
1125 	/* key_len should match the hash key supported by hardware */
1126 	if (rss->key_len != 0 && rss->key_len != HW_HASH_KEY_SIZE) {
1127 		rte_flow_error_set(error,
1128 				   EINVAL,
1129 				   RTE_FLOW_ERROR_TYPE_ACTION,
1130 				   act,
1131 				   "Incorrect hash key parameters");
1132 		rc = -rte_errno;
1133 		goto ret;
1134 	}
1135 
1136 	/* Currently RSS hash on inner and outer headers are supported.
1137 	 * 0 => Default (innermost RSS) setting
1138 	 * 1 => Outermost
1139 	 */
1140 	if (rss->level > 1) {
1141 		rte_flow_error_set(error,
1142 				   ENOTSUP,
1143 				   RTE_FLOW_ERROR_TYPE_ACTION,
1144 				   act,
1145 				   "Unsupported hash level");
1146 		rc = -rte_errno;
1147 		goto ret;
1148 	}
1149 
1150 	if ((rss->queue_num == 0 && rss->queue != NULL) ||
1151 	    (rss->queue_num != 0 && rss->queue == NULL)) {
1152 		rte_flow_error_set(error,
1153 				   EINVAL,
1154 				   RTE_FLOW_ERROR_TYPE_ACTION,
1155 				   act,
1156 				   "Invalid queue config specified");
1157 		rc = -rte_errno;
1158 		goto ret;
1159 	}
1160 
1161 	/* If RSS types is 0, use a best effort configuration */
1162 	types = rss->types ? rss->types : RTE_ETH_RSS_IPV4 | RTE_ETH_RSS_IPV6;
1163 
1164 	hash_type = bnxt_rte_to_hwrm_hash_types(types);
1165 
1166 	/* If requested types can't be supported, leave existing settings */
1167 	if (hash_type)
1168 		vnic->hash_type = hash_type;
1169 
1170 	vnic->hash_mode =
1171 		bnxt_rte_to_hwrm_hash_level(bp, rss->types, rss->level);
1172 
1173 	/* Update RSS key only if key_len != 0 */
1174 	if (rss->key_len != 0)
1175 		memcpy(vnic->rss_hash_key, rss->key, rss->key_len);
1176 
1177 	if (rss->queue_num == 0)
1178 		goto skip_rss_table;
1179 
1180 	/* Prepare the indirection table */
1181 	for (rss_idx = 0, fw_idx = 0; rss_idx < HW_HASH_INDEX_SIZE;
1182 	     rss_idx++, fw_idx++) {
1183 		uint8_t *rxq_state = bp->eth_dev->data->rx_queue_state;
1184 		struct bnxt_rx_queue *rxq;
1185 		uint32_t idx;
1186 
1187 		for (i = 0; i < bp->rx_cp_nr_rings; i++) {
1188 			idx = rss->queue[fw_idx % rss->queue_num];
1189 			if (rxq_state[idx] != RTE_ETH_QUEUE_STATE_STOPPED)
1190 				break;
1191 			fw_idx++;
1192 		}
1193 
1194 		if (i == bp->rx_cp_nr_rings)
1195 			return 0;
1196 
1197 		if (BNXT_CHIP_P5(bp)) {
1198 			rxq = bp->rx_queues[idx];
1199 			vnic->rss_table[rss_idx * 2] =
1200 				rxq->rx_ring->rx_ring_struct->fw_ring_id;
1201 			vnic->rss_table[rss_idx * 2 + 1] =
1202 				rxq->cp_ring->cp_ring_struct->fw_ring_id;
1203 		} else {
1204 			vnic->rss_table[rss_idx] = vnic->fw_grp_ids[idx];
1205 		}
1206 	}
1207 
1208 skip_rss_table:
1209 	rc = bnxt_hwrm_vnic_rss_cfg(bp, vnic);
1210 	if (rc != 0) {
1211 		rte_flow_error_set(error,
1212 				   -rc,
1213 				   RTE_FLOW_ERROR_TYPE_ACTION,
1214 				   act,
1215 				   "VNIC RSS configure failed");
1216 		rc = -rte_errno;
1217 		goto ret;
1218 	}
1219 ret:
1220 	return rc;
1221 }
1222 
1223 static int
1224 bnxt_validate_and_parse_flow(struct rte_eth_dev *dev,
1225 			     const struct rte_flow_item pattern[],
1226 			     const struct rte_flow_action actions[],
1227 			     const struct rte_flow_attr *attr,
1228 			     struct rte_flow_error *error,
1229 			     struct bnxt_filter_info *filter)
1230 {
1231 	const struct rte_flow_action *act =
1232 		bnxt_flow_non_void_action(actions);
1233 	struct bnxt *bp = dev->data->dev_private;
1234 	struct rte_eth_conf *dev_conf = &bp->eth_dev->data->dev_conf;
1235 	struct bnxt_vnic_info *vnic = NULL, *vnic0 = NULL;
1236 	const struct rte_flow_action_queue *act_q;
1237 	const struct rte_flow_action_vf *act_vf;
1238 	struct bnxt_filter_info *filter1 = NULL;
1239 	const struct rte_flow_action_rss *rss;
1240 	struct bnxt_rx_queue *rxq = NULL;
1241 	int dflt_vnic, vnic_id;
1242 	unsigned int rss_idx;
1243 	uint32_t vf = 0, i;
1244 	int rc, use_ntuple;
1245 
1246 	rc =
1247 	bnxt_validate_and_parse_flow_type(attr, pattern, error, filter);
1248 	if (rc != 0)
1249 		goto ret;
1250 
1251 	rc = bnxt_flow_parse_attr(attr, error);
1252 	if (rc != 0)
1253 		goto ret;
1254 
1255 	/* Since we support ingress attribute only - right now. */
1256 	if (filter->filter_type == HWRM_CFA_EM_FILTER)
1257 		filter->flags = HWRM_CFA_EM_FLOW_ALLOC_INPUT_FLAGS_PATH_RX;
1258 
1259 	use_ntuple = bnxt_filter_type_check(pattern, error);
1260 
1261 start:
1262 	switch (act->type) {
1263 	case RTE_FLOW_ACTION_TYPE_QUEUE:
1264 		/* Allow this flow. Redirect to a VNIC. */
1265 		act_q = (const struct rte_flow_action_queue *)act->conf;
1266 		if (!act_q->index || act_q->index >= bp->rx_nr_rings) {
1267 			rte_flow_error_set(error,
1268 					   EINVAL,
1269 					   RTE_FLOW_ERROR_TYPE_ACTION,
1270 					   act,
1271 					   "Invalid queue ID.");
1272 			rc = -rte_errno;
1273 			goto ret;
1274 		}
1275 		PMD_DRV_LOG(DEBUG, "Queue index %d\n", act_q->index);
1276 
1277 		vnic_id = attr->group;
1278 		if (!vnic_id) {
1279 			PMD_DRV_LOG(DEBUG, "Group id is 0\n");
1280 			vnic_id = act_q->index;
1281 		}
1282 
1283 		BNXT_VALID_VNIC_OR_RET(bp, vnic_id);
1284 
1285 		vnic = &bp->vnic_info[vnic_id];
1286 		if (vnic->rx_queue_cnt) {
1287 			if (vnic->start_grp_id != act_q->index) {
1288 				PMD_DRV_LOG(ERR,
1289 					    "VNIC already in use\n");
1290 				rte_flow_error_set(error,
1291 						   EINVAL,
1292 						   RTE_FLOW_ERROR_TYPE_ACTION,
1293 						   act,
1294 						   "VNIC already in use");
1295 				rc = -rte_errno;
1296 				goto ret;
1297 			}
1298 			goto use_vnic;
1299 		}
1300 
1301 		rxq = bp->rx_queues[act_q->index];
1302 
1303 		if (!(dev_conf->rxmode.mq_mode & RTE_ETH_MQ_RX_RSS) && rxq &&
1304 		    vnic->fw_vnic_id != INVALID_HW_RING_ID)
1305 			goto use_vnic;
1306 
1307 		if (!rxq) {
1308 			PMD_DRV_LOG(ERR,
1309 				    "Queue invalid or used with other VNIC\n");
1310 			rte_flow_error_set(error,
1311 					   EINVAL,
1312 					   RTE_FLOW_ERROR_TYPE_ACTION,
1313 					   act,
1314 					   "Queue invalid queue or in use");
1315 			rc = -rte_errno;
1316 			goto ret;
1317 		}
1318 
1319 		rxq->vnic = vnic;
1320 		rxq->rx_started = 1;
1321 		vnic->rx_queue_cnt++;
1322 		vnic->start_grp_id = act_q->index;
1323 		vnic->end_grp_id = act_q->index;
1324 		vnic->func_default = 0;	//This is not a default VNIC.
1325 
1326 		PMD_DRV_LOG(DEBUG, "VNIC found\n");
1327 
1328 		rc = bnxt_vnic_prep(bp, vnic, act, error);
1329 		if (rc)
1330 			goto ret;
1331 
1332 		PMD_DRV_LOG(DEBUG,
1333 			    "vnic[%d] = %p vnic->fw_grp_ids = %p\n",
1334 			    act_q->index, vnic, vnic->fw_grp_ids);
1335 
1336 use_vnic:
1337 		vnic->ff_pool_idx = vnic_id;
1338 		PMD_DRV_LOG(DEBUG,
1339 			    "Setting vnic ff_idx %d\n", vnic->ff_pool_idx);
1340 		filter->dst_id = vnic->fw_vnic_id;
1341 
1342 		/* For ntuple filter, create the L2 filter with default VNIC.
1343 		 * The user specified redirect queue will be set while creating
1344 		 * the ntuple filter in hardware.
1345 		 */
1346 		vnic0 = BNXT_GET_DEFAULT_VNIC(bp);
1347 		if (use_ntuple)
1348 			filter1 = bnxt_get_l2_filter(bp, filter, vnic0);
1349 		else
1350 			filter1 = bnxt_get_l2_filter(bp, filter, vnic);
1351 		if (filter1 == NULL) {
1352 			rte_flow_error_set(error,
1353 					   ENOSPC,
1354 					   RTE_FLOW_ERROR_TYPE_ACTION,
1355 					   act,
1356 					   "Filter not available");
1357 			rc = -rte_errno;
1358 			goto ret;
1359 		}
1360 
1361 		PMD_DRV_LOG(DEBUG, "new fltr: %p l2fltr: %p l2_ref_cnt: %d\n",
1362 			    filter, filter1, filter1->l2_ref_cnt);
1363 		bnxt_update_filter_flags_en(filter, filter1, use_ntuple);
1364 		break;
1365 	case RTE_FLOW_ACTION_TYPE_DROP:
1366 		vnic0 = &bp->vnic_info[0];
1367 		filter->dst_id = vnic0->fw_vnic_id;
1368 		filter->valid_flags |= BNXT_FLOW_L2_DROP_FLAG;
1369 		filter1 = bnxt_get_l2_filter(bp, filter, vnic0);
1370 		if (filter1 == NULL) {
1371 			rte_flow_error_set(error,
1372 					   ENOSPC,
1373 					   RTE_FLOW_ERROR_TYPE_ACTION,
1374 					   act,
1375 					   "Filter not available");
1376 			rc = -rte_errno;
1377 			goto ret;
1378 		}
1379 
1380 		if (filter->filter_type == HWRM_CFA_EM_FILTER)
1381 			filter->flags =
1382 				HWRM_CFA_EM_FLOW_ALLOC_INPUT_FLAGS_DROP;
1383 		else if (filter->filter_type == HWRM_CFA_NTUPLE_FILTER)
1384 			filter->flags =
1385 				HWRM_CFA_NTUPLE_FILTER_ALLOC_INPUT_FLAGS_DROP;
1386 
1387 		bnxt_update_filter_flags_en(filter, filter1, use_ntuple);
1388 		break;
1389 	case RTE_FLOW_ACTION_TYPE_VF:
1390 		act_vf = (const struct rte_flow_action_vf *)act->conf;
1391 		vf = act_vf->id;
1392 
1393 		if (filter->tunnel_type ==
1394 		    CFA_NTUPLE_FILTER_ALLOC_REQ_TUNNEL_TYPE_VXLAN ||
1395 		    filter->tunnel_type ==
1396 		    CFA_NTUPLE_FILTER_ALLOC_REQ_TUNNEL_TYPE_IPGRE) {
1397 			/* If issued on a VF, ensure id is 0 and is trusted */
1398 			if (BNXT_VF(bp)) {
1399 				if (!BNXT_VF_IS_TRUSTED(bp) || vf) {
1400 					rte_flow_error_set(error, EINVAL,
1401 						RTE_FLOW_ERROR_TYPE_ACTION,
1402 						act,
1403 						"Incorrect VF");
1404 					rc = -rte_errno;
1405 					goto ret;
1406 				}
1407 			}
1408 
1409 			filter->enables |= filter->tunnel_type;
1410 			filter->filter_type = HWRM_CFA_TUNNEL_REDIRECT_FILTER;
1411 			goto done;
1412 		}
1413 
1414 		if (vf >= bp->pdev->max_vfs) {
1415 			rte_flow_error_set(error,
1416 					   EINVAL,
1417 					   RTE_FLOW_ERROR_TYPE_ACTION,
1418 					   act,
1419 					   "Incorrect VF id!");
1420 			rc = -rte_errno;
1421 			goto ret;
1422 		}
1423 
1424 		filter->mirror_vnic_id =
1425 		dflt_vnic = bnxt_hwrm_func_qcfg_vf_dflt_vnic_id(bp, vf);
1426 		if (dflt_vnic < 0) {
1427 			/* This simply indicates there's no driver loaded.
1428 			 * This is not an error.
1429 			 */
1430 			rte_flow_error_set(error,
1431 					   EINVAL,
1432 					   RTE_FLOW_ERROR_TYPE_ACTION,
1433 					   act,
1434 					   "Unable to get default VNIC for VF");
1435 			rc = -rte_errno;
1436 			goto ret;
1437 		}
1438 
1439 		filter->mirror_vnic_id = dflt_vnic;
1440 		filter->enables |= NTUPLE_FLTR_ALLOC_INPUT_EN_MIRROR_VNIC_ID;
1441 
1442 		vnic0 = &bp->vnic_info[0];
1443 		filter1 = bnxt_get_l2_filter(bp, filter, vnic0);
1444 		if (filter1 == NULL) {
1445 			rte_flow_error_set(error,
1446 					   ENOSPC,
1447 					   RTE_FLOW_ERROR_TYPE_ACTION,
1448 					   act,
1449 					   "New filter not available");
1450 			rc = -rte_errno;
1451 			goto ret;
1452 		}
1453 
1454 		filter->fw_l2_filter_id = filter1->fw_l2_filter_id;
1455 		filter->flow_id = filter1->flow_id;
1456 		break;
1457 	case RTE_FLOW_ACTION_TYPE_RSS:
1458 		rc = bnxt_validate_rss_action(actions);
1459 		if (rc != 0) {
1460 			rte_flow_error_set(error,
1461 					   EINVAL,
1462 					   RTE_FLOW_ERROR_TYPE_ACTION,
1463 					   act,
1464 					   "Invalid actions specified with RSS");
1465 			rc = -rte_errno;
1466 			goto ret;
1467 		}
1468 
1469 		rss = (const struct rte_flow_action_rss *)act->conf;
1470 
1471 		vnic_id = bnxt_get_vnic(bp, attr->group);
1472 
1473 		BNXT_VALID_VNIC_OR_RET(bp, vnic_id);
1474 		vnic = &bp->vnic_info[vnic_id];
1475 
1476 		/*
1477 		 * For non NS3 cases, rte_flow_items will not be considered
1478 		 * for RSS updates.
1479 		 */
1480 		if (filter->filter_type == HWRM_CFA_CONFIG) {
1481 			/* RSS config update requested */
1482 			rc = bnxt_vnic_rss_cfg_update(bp, vnic, act, error);
1483 			if (rc != 0)
1484 				goto ret;
1485 
1486 			filter->dst_id = vnic->fw_vnic_id;
1487 			break;
1488 		}
1489 
1490 		/* Check if requested RSS config matches RSS config of VNIC
1491 		 * only if it is not a fresh VNIC configuration.
1492 		 * Otherwise the existing VNIC configuration can be used.
1493 		 */
1494 		if (vnic->rx_queue_cnt) {
1495 			rc = match_vnic_rss_cfg(bp, vnic, rss);
1496 			if (rc) {
1497 				PMD_DRV_LOG(ERR,
1498 					    "VNIC and RSS config mismatch\n");
1499 				rte_flow_error_set(error,
1500 						   EINVAL,
1501 						   RTE_FLOW_ERROR_TYPE_ACTION,
1502 						   act,
1503 						   "VNIC and RSS cfg mismatch");
1504 				rc = -rte_errno;
1505 				goto ret;
1506 			}
1507 			goto vnic_found;
1508 		}
1509 
1510 		for (i = 0; i < rss->queue_num; i++) {
1511 			PMD_DRV_LOG(DEBUG, "RSS action Queue %d\n",
1512 				    rss->queue[i]);
1513 
1514 			if (!rss->queue[i] ||
1515 			    rss->queue[i] >= bp->rx_nr_rings ||
1516 			    !bp->rx_queues[rss->queue[i]]) {
1517 				rte_flow_error_set(error,
1518 						   EINVAL,
1519 						   RTE_FLOW_ERROR_TYPE_ACTION,
1520 						   act,
1521 						   "Invalid queue ID for RSS");
1522 				rc = -rte_errno;
1523 				goto ret;
1524 			}
1525 			rxq = bp->rx_queues[rss->queue[i]];
1526 
1527 			if (bp->vnic_info[0].fw_grp_ids[rss->queue[i]] !=
1528 			    INVALID_HW_RING_ID) {
1529 				PMD_DRV_LOG(ERR,
1530 					    "queue active with other VNIC\n");
1531 				rte_flow_error_set(error,
1532 						   EINVAL,
1533 						   RTE_FLOW_ERROR_TYPE_ACTION,
1534 						   act,
1535 						   "Invalid queue ID for RSS");
1536 				rc = -rte_errno;
1537 				goto ret;
1538 			}
1539 
1540 			rxq->vnic = vnic;
1541 			rxq->rx_started = 1;
1542 			vnic->rx_queue_cnt++;
1543 		}
1544 
1545 		vnic->start_grp_id = rss->queue[0];
1546 		vnic->end_grp_id = rss->queue[rss->queue_num - 1];
1547 		vnic->func_default = 0;	//This is not a default VNIC.
1548 
1549 		rc = bnxt_vnic_prep(bp, vnic, act, error);
1550 		if (rc)
1551 			goto ret;
1552 
1553 		PMD_DRV_LOG(DEBUG,
1554 			    "vnic[%d] = %p vnic->fw_grp_ids = %p\n",
1555 			    vnic_id, vnic, vnic->fw_grp_ids);
1556 
1557 		vnic->ff_pool_idx = vnic_id;
1558 		PMD_DRV_LOG(DEBUG,
1559 			    "Setting vnic ff_pool_idx %d\n", vnic->ff_pool_idx);
1560 
1561 		/* This can be done only after vnic_grp_alloc is done. */
1562 		for (i = 0; i < vnic->rx_queue_cnt; i++) {
1563 			vnic->fw_grp_ids[i] =
1564 				bp->grp_info[rss->queue[i]].fw_grp_id;
1565 			/* Make sure vnic0 does not use these rings. */
1566 			bp->vnic_info[0].fw_grp_ids[rss->queue[i]] =
1567 				INVALID_HW_RING_ID;
1568 		}
1569 
1570 		for (rss_idx = 0; rss_idx < HW_HASH_INDEX_SIZE; ) {
1571 			for (i = 0; i < vnic->rx_queue_cnt; i++)
1572 				vnic->rss_table[rss_idx++] =
1573 					vnic->fw_grp_ids[i];
1574 		}
1575 
1576 		/* Configure RSS only if the queue count is > 1 */
1577 		if (vnic->rx_queue_cnt > 1) {
1578 			vnic->hash_type =
1579 				bnxt_rte_to_hwrm_hash_types(rss->types);
1580 			vnic->hash_mode =
1581 			bnxt_rte_to_hwrm_hash_level(bp, rss->types, rss->level);
1582 
1583 			if (!rss->key_len) {
1584 				/* If hash key has not been specified,
1585 				 * use random hash key.
1586 				 */
1587 				bnxt_prandom_bytes(vnic->rss_hash_key,
1588 						   HW_HASH_KEY_SIZE);
1589 			} else {
1590 				if (rss->key_len > HW_HASH_KEY_SIZE)
1591 					memcpy(vnic->rss_hash_key,
1592 					       rss->key,
1593 					       HW_HASH_KEY_SIZE);
1594 				else
1595 					memcpy(vnic->rss_hash_key,
1596 					       rss->key,
1597 					       rss->key_len);
1598 			}
1599 			bnxt_hwrm_vnic_rss_cfg(bp, vnic);
1600 		} else {
1601 			PMD_DRV_LOG(DEBUG, "No RSS config required\n");
1602 		}
1603 
1604 vnic_found:
1605 		filter->dst_id = vnic->fw_vnic_id;
1606 		filter1 = bnxt_get_l2_filter(bp, filter, vnic);
1607 		if (filter1 == NULL) {
1608 			rte_flow_error_set(error,
1609 					   ENOSPC,
1610 					   RTE_FLOW_ERROR_TYPE_ACTION,
1611 					   act,
1612 					   "New filter not available");
1613 			rc = -rte_errno;
1614 			goto ret;
1615 		}
1616 
1617 		PMD_DRV_LOG(DEBUG, "L2 filter created\n");
1618 		bnxt_update_filter_flags_en(filter, filter1, use_ntuple);
1619 		break;
1620 	case RTE_FLOW_ACTION_TYPE_MARK:
1621 		if (bp->mark_table == NULL) {
1622 			rte_flow_error_set(error,
1623 					   ENOMEM,
1624 					   RTE_FLOW_ERROR_TYPE_ACTION,
1625 					   act,
1626 					   "Mark table not allocated.");
1627 			rc = -rte_errno;
1628 			goto ret;
1629 		}
1630 
1631 		if (bp->flags & BNXT_FLAG_RX_VECTOR_PKT_MODE) {
1632 			PMD_DRV_LOG(DEBUG,
1633 				    "Disabling vector processing for mark\n");
1634 			bp->eth_dev->rx_pkt_burst = bnxt_recv_pkts;
1635 			bp->flags &= ~BNXT_FLAG_RX_VECTOR_PKT_MODE;
1636 		}
1637 
1638 		filter->valid_flags |= BNXT_FLOW_MARK_FLAG;
1639 		filter->mark = ((const struct rte_flow_action_mark *)
1640 				act->conf)->id;
1641 		PMD_DRV_LOG(DEBUG, "Mark the flow %d\n", filter->mark);
1642 		break;
1643 	default:
1644 		rte_flow_error_set(error,
1645 				   EINVAL,
1646 				   RTE_FLOW_ERROR_TYPE_ACTION,
1647 				   act,
1648 				   "Invalid action.");
1649 		rc = -rte_errno;
1650 		goto ret;
1651 	}
1652 
1653 done:
1654 	act = bnxt_flow_non_void_action(++act);
1655 	while (act->type != RTE_FLOW_ACTION_TYPE_END)
1656 		goto start;
1657 
1658 	return rc;
1659 ret:
1660 
1661 	if (filter1) {
1662 		bnxt_hwrm_clear_l2_filter(bp, filter1);
1663 		bnxt_free_filter(bp, filter1);
1664 	}
1665 
1666 	if (rte_errno)  {
1667 		if (vnic && STAILQ_EMPTY(&vnic->filter))
1668 			vnic->rx_queue_cnt = 0;
1669 
1670 		if (rxq && !vnic->rx_queue_cnt)
1671 			rxq->vnic = &bp->vnic_info[0];
1672 	}
1673 	return -rte_errno;
1674 }
1675 
1676 static
1677 struct bnxt_vnic_info *find_matching_vnic(struct bnxt *bp,
1678 					  struct bnxt_filter_info *filter)
1679 {
1680 	struct bnxt_vnic_info *vnic = NULL;
1681 	unsigned int i;
1682 
1683 	for (i = 0; i < bp->max_vnics; i++) {
1684 		vnic = &bp->vnic_info[i];
1685 		if (vnic->fw_vnic_id != INVALID_VNIC_ID &&
1686 		    filter->dst_id == vnic->fw_vnic_id) {
1687 			PMD_DRV_LOG(DEBUG, "Found matching VNIC Id %d\n",
1688 				    vnic->ff_pool_idx);
1689 			return vnic;
1690 		}
1691 	}
1692 	return NULL;
1693 }
1694 
1695 static int
1696 bnxt_flow_validate(struct rte_eth_dev *dev,
1697 		   const struct rte_flow_attr *attr,
1698 		   const struct rte_flow_item pattern[],
1699 		   const struct rte_flow_action actions[],
1700 		   struct rte_flow_error *error)
1701 {
1702 	struct bnxt *bp = dev->data->dev_private;
1703 	struct bnxt_vnic_info *vnic = NULL;
1704 	struct bnxt_filter_info *filter;
1705 	int ret = 0;
1706 
1707 	bnxt_acquire_flow_lock(bp);
1708 	ret = bnxt_flow_args_validate(attr, pattern, actions, error);
1709 	if (ret != 0) {
1710 		bnxt_release_flow_lock(bp);
1711 		return ret;
1712 	}
1713 
1714 	filter = bnxt_get_unused_filter(bp);
1715 	if (filter == NULL) {
1716 		rte_flow_error_set(error, ENOSPC,
1717 				   RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
1718 				   "Not enough resources for a new flow");
1719 		bnxt_release_flow_lock(bp);
1720 		return -ENOSPC;
1721 	}
1722 
1723 	ret = bnxt_validate_and_parse_flow(dev, pattern, actions, attr,
1724 					   error, filter);
1725 	if (ret)
1726 		goto exit;
1727 
1728 	vnic = find_matching_vnic(bp, filter);
1729 	if (vnic) {
1730 		if (STAILQ_EMPTY(&vnic->filter)) {
1731 			bnxt_vnic_cleanup(bp, vnic);
1732 			bp->nr_vnics--;
1733 			PMD_DRV_LOG(DEBUG, "Free VNIC\n");
1734 		}
1735 	}
1736 
1737 	if (filter->filter_type == HWRM_CFA_EM_FILTER)
1738 		bnxt_hwrm_clear_em_filter(bp, filter);
1739 	else if (filter->filter_type == HWRM_CFA_NTUPLE_FILTER)
1740 		bnxt_hwrm_clear_ntuple_filter(bp, filter);
1741 	else
1742 		bnxt_hwrm_clear_l2_filter(bp, filter);
1743 
1744 exit:
1745 	/* No need to hold on to this filter if we are just validating flow */
1746 	bnxt_free_filter(bp, filter);
1747 	bnxt_release_flow_lock(bp);
1748 
1749 	return ret;
1750 }
1751 
1752 static void
1753 bnxt_update_filter(struct bnxt *bp, struct bnxt_filter_info *old_filter,
1754 		   struct bnxt_filter_info *new_filter)
1755 {
1756 	/* Clear the new L2 filter that was created in the previous step in
1757 	 * bnxt_validate_and_parse_flow. For L2 filters, we will use the new
1758 	 * filter which points to the new destination queue and so we clear
1759 	 * the previous L2 filter. For ntuple filters, we are going to reuse
1760 	 * the old L2 filter and create new NTUPLE filter with this new
1761 	 * destination queue subsequently during bnxt_flow_create. So we
1762 	 * decrement the ref cnt of the L2 filter that would've been bumped
1763 	 * up previously in bnxt_validate_and_parse_flow as the old n-tuple
1764 	 * filter that was referencing it will be deleted now.
1765 	 */
1766 	bnxt_hwrm_clear_l2_filter(bp, old_filter);
1767 	if (new_filter->filter_type == HWRM_CFA_L2_FILTER) {
1768 		bnxt_hwrm_set_l2_filter(bp, new_filter->dst_id, new_filter);
1769 	} else {
1770 		if (new_filter->filter_type == HWRM_CFA_EM_FILTER)
1771 			bnxt_hwrm_clear_em_filter(bp, old_filter);
1772 		if (new_filter->filter_type == HWRM_CFA_NTUPLE_FILTER)
1773 			bnxt_hwrm_clear_ntuple_filter(bp, old_filter);
1774 	}
1775 }
1776 
1777 static int
1778 bnxt_match_filter(struct bnxt *bp, struct bnxt_filter_info *nf)
1779 {
1780 	struct bnxt_filter_info *mf;
1781 	struct rte_flow *flow;
1782 	int i;
1783 
1784 	for (i = bp->max_vnics - 1; i >= 0; i--) {
1785 		struct bnxt_vnic_info *vnic = &bp->vnic_info[i];
1786 
1787 		if (vnic->fw_vnic_id == INVALID_VNIC_ID)
1788 			continue;
1789 
1790 		STAILQ_FOREACH(flow, &vnic->flow_list, next) {
1791 			mf = flow->filter;
1792 
1793 			if (mf->filter_type == nf->filter_type &&
1794 			    mf->flags == nf->flags &&
1795 			    mf->src_port == nf->src_port &&
1796 			    mf->src_port_mask == nf->src_port_mask &&
1797 			    mf->dst_port == nf->dst_port &&
1798 			    mf->dst_port_mask == nf->dst_port_mask &&
1799 			    mf->ip_protocol == nf->ip_protocol &&
1800 			    mf->ip_addr_type == nf->ip_addr_type &&
1801 			    mf->ethertype == nf->ethertype &&
1802 			    mf->vni == nf->vni &&
1803 			    mf->tunnel_type == nf->tunnel_type &&
1804 			    mf->l2_ovlan == nf->l2_ovlan &&
1805 			    mf->l2_ovlan_mask == nf->l2_ovlan_mask &&
1806 			    mf->l2_ivlan == nf->l2_ivlan &&
1807 			    mf->l2_ivlan_mask == nf->l2_ivlan_mask &&
1808 			    !memcmp(mf->l2_addr, nf->l2_addr,
1809 				    RTE_ETHER_ADDR_LEN) &&
1810 			    !memcmp(mf->l2_addr_mask, nf->l2_addr_mask,
1811 				    RTE_ETHER_ADDR_LEN) &&
1812 			    !memcmp(mf->src_macaddr, nf->src_macaddr,
1813 				    RTE_ETHER_ADDR_LEN) &&
1814 			    !memcmp(mf->dst_macaddr, nf->dst_macaddr,
1815 				    RTE_ETHER_ADDR_LEN) &&
1816 			    !memcmp(mf->src_ipaddr, nf->src_ipaddr,
1817 				    sizeof(nf->src_ipaddr)) &&
1818 			    !memcmp(mf->src_ipaddr_mask, nf->src_ipaddr_mask,
1819 				    sizeof(nf->src_ipaddr_mask)) &&
1820 			    !memcmp(mf->dst_ipaddr, nf->dst_ipaddr,
1821 				    sizeof(nf->dst_ipaddr)) &&
1822 			    !memcmp(mf->dst_ipaddr_mask, nf->dst_ipaddr_mask,
1823 				    sizeof(nf->dst_ipaddr_mask))) {
1824 				if (mf->dst_id == nf->dst_id)
1825 					return -EEXIST;
1826 				/* Free the old filter, update flow
1827 				 * with new filter
1828 				 */
1829 				bnxt_update_filter(bp, mf, nf);
1830 				STAILQ_REMOVE(&vnic->filter, mf,
1831 					      bnxt_filter_info, next);
1832 				STAILQ_INSERT_TAIL(&vnic->filter, nf, next);
1833 				bnxt_free_filter(bp, mf);
1834 				flow->filter = nf;
1835 				return -EXDEV;
1836 			}
1837 		}
1838 	}
1839 	return 0;
1840 }
1841 
1842 static void
1843 bnxt_setup_flow_counter(struct bnxt *bp)
1844 {
1845 	if (bp->fw_cap & BNXT_FW_CAP_ADV_FLOW_COUNTERS &&
1846 	    !(bp->flags & BNXT_FLAG_FC_THREAD) && BNXT_FLOW_XSTATS_EN(bp)) {
1847 		rte_eal_alarm_set(US_PER_S * BNXT_FC_TIMER,
1848 				  bnxt_flow_cnt_alarm_cb,
1849 				  (void *)bp);
1850 		bp->flags |= BNXT_FLAG_FC_THREAD;
1851 	}
1852 }
1853 
1854 void bnxt_flow_cnt_alarm_cb(void *arg)
1855 {
1856 	int rc = 0;
1857 	struct bnxt *bp = arg;
1858 
1859 	if (!bp->flow_stat->rx_fc_out_tbl.va) {
1860 		PMD_DRV_LOG(ERR, "bp->flow_stat->rx_fc_out_tbl.va is NULL?\n");
1861 		bnxt_cancel_fc_thread(bp);
1862 		return;
1863 	}
1864 
1865 	if (!bp->flow_stat->flow_count) {
1866 		bnxt_cancel_fc_thread(bp);
1867 		return;
1868 	}
1869 
1870 	if (!bp->eth_dev->data->dev_started) {
1871 		bnxt_cancel_fc_thread(bp);
1872 		return;
1873 	}
1874 
1875 	rc = bnxt_flow_stats_req(bp);
1876 	if (rc) {
1877 		PMD_DRV_LOG(ERR, "Flow stat alarm not rescheduled.\n");
1878 		return;
1879 	}
1880 
1881 	rte_eal_alarm_set(US_PER_S * BNXT_FC_TIMER,
1882 			  bnxt_flow_cnt_alarm_cb,
1883 			  (void *)bp);
1884 }
1885 
1886 
1887 static struct rte_flow *
1888 bnxt_flow_create(struct rte_eth_dev *dev,
1889 		 const struct rte_flow_attr *attr,
1890 		 const struct rte_flow_item pattern[],
1891 		 const struct rte_flow_action actions[],
1892 		 struct rte_flow_error *error)
1893 {
1894 	struct bnxt *bp = dev->data->dev_private;
1895 	struct bnxt_vnic_info *vnic = NULL;
1896 	struct bnxt_filter_info *filter;
1897 	bool update_flow = false;
1898 	struct rte_flow *flow;
1899 	int ret = 0;
1900 	uint32_t tun_type, flow_id;
1901 
1902 	if (BNXT_VF(bp) && !BNXT_VF_IS_TRUSTED(bp)) {
1903 		rte_flow_error_set(error, EINVAL,
1904 				   RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
1905 				   "Failed to create flow, Not a Trusted VF!");
1906 		return NULL;
1907 	}
1908 
1909 	if (!dev->data->dev_started) {
1910 		rte_flow_error_set(error,
1911 				   EINVAL,
1912 				   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
1913 				   NULL,
1914 				   "Device must be started");
1915 		return NULL;
1916 	}
1917 
1918 	flow = rte_zmalloc("bnxt_flow", sizeof(struct rte_flow), 0);
1919 	if (!flow) {
1920 		rte_flow_error_set(error, ENOMEM,
1921 				   RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
1922 				   "Failed to allocate memory");
1923 		return flow;
1924 	}
1925 
1926 	bnxt_acquire_flow_lock(bp);
1927 	ret = bnxt_flow_args_validate(attr, pattern, actions, error);
1928 	if (ret != 0) {
1929 		PMD_DRV_LOG(ERR, "Not a validate flow.\n");
1930 		goto free_flow;
1931 	}
1932 
1933 	filter = bnxt_get_unused_filter(bp);
1934 	if (filter == NULL) {
1935 		rte_flow_error_set(error, ENOSPC,
1936 				   RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
1937 				   "Not enough resources for a new flow");
1938 		goto free_flow;
1939 	}
1940 
1941 	ret = bnxt_validate_and_parse_flow(dev, pattern, actions, attr,
1942 					   error, filter);
1943 	if (ret != 0)
1944 		goto free_filter;
1945 
1946 	ret = bnxt_match_filter(bp, filter);
1947 	if (ret == -EEXIST) {
1948 		PMD_DRV_LOG(DEBUG, "Flow already exists.\n");
1949 		/* Clear the filter that was created as part of
1950 		 * validate_and_parse_flow() above
1951 		 */
1952 		bnxt_hwrm_clear_l2_filter(bp, filter);
1953 		goto free_filter;
1954 	} else if (ret == -EXDEV) {
1955 		PMD_DRV_LOG(DEBUG, "Flow with same pattern exists\n");
1956 		PMD_DRV_LOG(DEBUG, "Updating with different destination\n");
1957 		update_flow = true;
1958 	}
1959 
1960 	/* If tunnel redirection to a VF/PF is specified then only tunnel_type
1961 	 * is set and enable is set to the tunnel type. Issue hwrm cmd directly
1962 	 * in such a case.
1963 	 */
1964 	if (filter->filter_type == HWRM_CFA_TUNNEL_REDIRECT_FILTER &&
1965 	    filter->enables == filter->tunnel_type) {
1966 		ret = bnxt_hwrm_tunnel_redirect_query(bp, &tun_type);
1967 		if (ret) {
1968 			rte_flow_error_set(error, -ret,
1969 					   RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
1970 					   "Unable to query tunnel to VF");
1971 			goto free_filter;
1972 		}
1973 		if (tun_type == (1U << filter->tunnel_type)) {
1974 			ret =
1975 			bnxt_hwrm_tunnel_redirect_free(bp,
1976 						       filter->tunnel_type);
1977 			if (ret) {
1978 				PMD_DRV_LOG(ERR,
1979 					    "Unable to free existing tunnel\n");
1980 				rte_flow_error_set(error, -ret,
1981 						   RTE_FLOW_ERROR_TYPE_HANDLE,
1982 						   NULL,
1983 						   "Unable to free preexisting "
1984 						   "tunnel on VF");
1985 				goto free_filter;
1986 			}
1987 		}
1988 		ret = bnxt_hwrm_tunnel_redirect(bp, filter->tunnel_type);
1989 		if (ret) {
1990 			rte_flow_error_set(error, -ret,
1991 					   RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
1992 					   "Unable to redirect tunnel to VF");
1993 			goto free_filter;
1994 		}
1995 		vnic = &bp->vnic_info[0];
1996 		goto done;
1997 	}
1998 
1999 	if (filter->filter_type == HWRM_CFA_EM_FILTER) {
2000 		filter->enables |=
2001 			HWRM_CFA_EM_FLOW_ALLOC_INPUT_ENABLES_L2_FILTER_ID;
2002 		ret = bnxt_hwrm_set_em_filter(bp, filter->dst_id, filter);
2003 		if (ret != 0) {
2004 			rte_flow_error_set(error, -ret,
2005 					   RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
2006 					   "Failed to create EM filter");
2007 			goto free_filter;
2008 		}
2009 	}
2010 
2011 	if (filter->filter_type == HWRM_CFA_NTUPLE_FILTER) {
2012 		filter->enables |=
2013 			HWRM_CFA_NTUPLE_FILTER_ALLOC_INPUT_ENABLES_L2_FILTER_ID;
2014 		ret = bnxt_hwrm_set_ntuple_filter(bp, filter->dst_id, filter);
2015 		if (ret != 0) {
2016 			rte_flow_error_set(error, -ret,
2017 					   RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
2018 					   "Failed to create ntuple filter");
2019 			goto free_filter;
2020 		}
2021 	}
2022 
2023 	vnic = find_matching_vnic(bp, filter);
2024 done:
2025 	if (!ret || update_flow) {
2026 		flow->filter = filter;
2027 		flow->vnic = vnic;
2028 		if (update_flow) {
2029 			ret = -EXDEV;
2030 			goto free_flow;
2031 		}
2032 
2033 		if (filter->valid_flags & BNXT_FLOW_MARK_FLAG) {
2034 			PMD_DRV_LOG(DEBUG,
2035 				    "Mark action: mark id 0x%x, flow id 0x%x\n",
2036 				    filter->mark, filter->flow_id);
2037 
2038 			/* TCAM and EM should be 16-bit only.
2039 			 * Other modes not supported.
2040 			 */
2041 			flow_id = filter->flow_id & BNXT_FLOW_ID_MASK;
2042 			if (bp->mark_table[flow_id].valid) {
2043 				rte_flow_error_set(error, EEXIST,
2044 						   RTE_FLOW_ERROR_TYPE_HANDLE,
2045 						   NULL,
2046 						   "Flow with mark id exists");
2047 				bnxt_clear_one_vnic_filter(bp, filter);
2048 				goto free_filter;
2049 			}
2050 			bp->mark_table[flow_id].valid = true;
2051 			bp->mark_table[flow_id].mark_id = filter->mark;
2052 		}
2053 
2054 		STAILQ_INSERT_TAIL(&vnic->filter, filter, next);
2055 		STAILQ_INSERT_TAIL(&vnic->flow_list, flow, next);
2056 
2057 		if (BNXT_FLOW_XSTATS_EN(bp))
2058 			bp->flow_stat->flow_count++;
2059 		bnxt_release_flow_lock(bp);
2060 		bnxt_setup_flow_counter(bp);
2061 		PMD_DRV_LOG(DEBUG, "Successfully created flow.\n");
2062 		return flow;
2063 	}
2064 
2065 free_filter:
2066 	bnxt_free_filter(bp, filter);
2067 free_flow:
2068 	if (ret == -EEXIST)
2069 		rte_flow_error_set(error, ret,
2070 				   RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
2071 				   "Matching Flow exists.");
2072 	else if (ret == -EXDEV)
2073 		rte_flow_error_set(error, 0,
2074 				   RTE_FLOW_ERROR_TYPE_NONE, NULL,
2075 				   "Flow with pattern exists, updating destination queue");
2076 	else if (!rte_errno)
2077 		rte_flow_error_set(error, -ret,
2078 				   RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
2079 				   "Failed to create flow.");
2080 	rte_free(flow);
2081 	flow = NULL;
2082 	bnxt_release_flow_lock(bp);
2083 	return flow;
2084 }
2085 
2086 static int bnxt_handle_tunnel_redirect_destroy(struct bnxt *bp,
2087 					       struct bnxt_filter_info *filter,
2088 					       struct rte_flow_error *error)
2089 {
2090 	uint16_t tun_dst_fid;
2091 	uint32_t tun_type;
2092 	int ret = 0;
2093 
2094 	ret = bnxt_hwrm_tunnel_redirect_query(bp, &tun_type);
2095 	if (ret) {
2096 		rte_flow_error_set(error, -ret,
2097 				   RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
2098 				   "Unable to query tunnel to VF");
2099 		return ret;
2100 	}
2101 	if (tun_type == (1U << filter->tunnel_type)) {
2102 		ret = bnxt_hwrm_tunnel_redirect_info(bp, filter->tunnel_type,
2103 						     &tun_dst_fid);
2104 		if (ret) {
2105 			rte_flow_error_set(error, -ret,
2106 					   RTE_FLOW_ERROR_TYPE_HANDLE,
2107 					   NULL,
2108 					   "tunnel_redirect info cmd fail");
2109 			return ret;
2110 		}
2111 		PMD_DRV_LOG(INFO, "Pre-existing tunnel fid = %x vf->fid = %x\n",
2112 			    tun_dst_fid + bp->first_vf_id, bp->fw_fid);
2113 
2114 		/* Tunnel doesn't belong to this VF, so don't send HWRM
2115 		 * cmd, just delete the flow from driver
2116 		 */
2117 		if (bp->fw_fid != (tun_dst_fid + bp->first_vf_id)) {
2118 			PMD_DRV_LOG(ERR,
2119 				    "Tunnel does not belong to this VF, skip hwrm_tunnel_redirect_free\n");
2120 		} else {
2121 			ret = bnxt_hwrm_tunnel_redirect_free(bp,
2122 							filter->tunnel_type);
2123 			if (ret) {
2124 				rte_flow_error_set(error, -ret,
2125 						   RTE_FLOW_ERROR_TYPE_HANDLE,
2126 						   NULL,
2127 						   "Unable to free tunnel redirection");
2128 				return ret;
2129 			}
2130 		}
2131 	}
2132 	return ret;
2133 }
2134 
2135 static int
2136 _bnxt_flow_destroy(struct bnxt *bp,
2137 		   struct rte_flow *flow,
2138 		    struct rte_flow_error *error)
2139 {
2140 	struct bnxt_filter_info *filter;
2141 	struct bnxt_vnic_info *vnic;
2142 	int ret = 0;
2143 	uint32_t flow_id;
2144 
2145 	filter = flow->filter;
2146 	vnic = flow->vnic;
2147 
2148 	if (filter->filter_type == HWRM_CFA_TUNNEL_REDIRECT_FILTER &&
2149 	    filter->enables == filter->tunnel_type) {
2150 		ret = bnxt_handle_tunnel_redirect_destroy(bp, filter, error);
2151 		if (!ret)
2152 			goto done;
2153 		else
2154 			return ret;
2155 	}
2156 
2157 	/* For config type, there is no filter in HW. Finish cleanup here */
2158 	if (filter->filter_type == HWRM_CFA_CONFIG)
2159 		goto done;
2160 
2161 	ret = bnxt_match_filter(bp, filter);
2162 	if (ret == 0)
2163 		PMD_DRV_LOG(ERR, "Could not find matching flow\n");
2164 
2165 	if (filter->valid_flags & BNXT_FLOW_MARK_FLAG) {
2166 		flow_id = filter->flow_id & BNXT_FLOW_ID_MASK;
2167 		memset(&bp->mark_table[flow_id], 0,
2168 		       sizeof(bp->mark_table[flow_id]));
2169 		filter->flow_id = 0;
2170 	}
2171 
2172 	ret = bnxt_clear_one_vnic_filter(bp, filter);
2173 
2174 done:
2175 	if (!ret) {
2176 		/* If it is a L2 drop filter, when the filter is created,
2177 		 * the FW updates the BC/MC records.
2178 		 * Once this filter is removed, issue the set_rx_mask command
2179 		 * to reset the BC/MC records in the HW to the settings
2180 		 * before the drop counter is created.
2181 		 */
2182 		if (filter->valid_flags & BNXT_FLOW_L2_DROP_FLAG)
2183 			bnxt_set_rx_mask_no_vlan(bp, &bp->vnic_info[0]);
2184 
2185 		STAILQ_REMOVE(&vnic->filter, filter, bnxt_filter_info, next);
2186 		bnxt_free_filter(bp, filter);
2187 		STAILQ_REMOVE(&vnic->flow_list, flow, rte_flow, next);
2188 		rte_free(flow);
2189 		if (BNXT_FLOW_XSTATS_EN(bp))
2190 			bp->flow_stat->flow_count--;
2191 
2192 		/* If this was the last flow associated with this vnic,
2193 		 * switch the queue back to RSS pool.
2194 		 */
2195 		if (vnic && !vnic->func_default &&
2196 		    STAILQ_EMPTY(&vnic->flow_list)) {
2197 			bnxt_vnic_cleanup(bp, vnic);
2198 			bp->nr_vnics--;
2199 		}
2200 	} else {
2201 		rte_flow_error_set(error, -ret,
2202 				   RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
2203 				   "Failed to destroy flow.");
2204 	}
2205 
2206 	return ret;
2207 }
2208 
2209 static int
2210 bnxt_flow_destroy(struct rte_eth_dev *dev,
2211 		  struct rte_flow *flow,
2212 		  struct rte_flow_error *error)
2213 {
2214 	struct bnxt *bp = dev->data->dev_private;
2215 	int ret = 0;
2216 
2217 	bnxt_acquire_flow_lock(bp);
2218 	if (!flow) {
2219 		rte_flow_error_set(error, EINVAL,
2220 				   RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
2221 				   "Invalid flow: failed to destroy flow.");
2222 		bnxt_release_flow_lock(bp);
2223 		return -EINVAL;
2224 	}
2225 
2226 	if (!flow->filter) {
2227 		rte_flow_error_set(error, EINVAL,
2228 				   RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
2229 				   "Invalid flow: failed to destroy flow.");
2230 		bnxt_release_flow_lock(bp);
2231 		return -EINVAL;
2232 	}
2233 	ret = _bnxt_flow_destroy(bp, flow, error);
2234 	bnxt_release_flow_lock(bp);
2235 
2236 	return ret;
2237 }
2238 
2239 void bnxt_cancel_fc_thread(struct bnxt *bp)
2240 {
2241 	bp->flags &= ~BNXT_FLAG_FC_THREAD;
2242 	rte_eal_alarm_cancel(bnxt_flow_cnt_alarm_cb, (void *)bp);
2243 }
2244 
2245 static int
2246 bnxt_flow_flush(struct rte_eth_dev *dev, struct rte_flow_error *error)
2247 {
2248 	struct bnxt *bp = dev->data->dev_private;
2249 	struct bnxt_vnic_info *vnic;
2250 	struct rte_flow *flow;
2251 	unsigned int i;
2252 	int ret = 0;
2253 
2254 	bnxt_acquire_flow_lock(bp);
2255 	for (i = 0; i < bp->max_vnics; i++) {
2256 		vnic = &bp->vnic_info[i];
2257 		if (vnic && vnic->fw_vnic_id == INVALID_VNIC_ID)
2258 			continue;
2259 
2260 		while (!STAILQ_EMPTY(&vnic->flow_list)) {
2261 			flow = STAILQ_FIRST(&vnic->flow_list);
2262 
2263 			if (!flow->filter)
2264 				continue;
2265 
2266 			ret = _bnxt_flow_destroy(bp, flow, error);
2267 			if (ret)
2268 				break;
2269 		}
2270 	}
2271 
2272 	bnxt_cancel_fc_thread(bp);
2273 	bnxt_release_flow_lock(bp);
2274 
2275 	return ret;
2276 }
2277 
2278 const struct rte_flow_ops bnxt_flow_ops = {
2279 	.validate = bnxt_flow_validate,
2280 	.create = bnxt_flow_create,
2281 	.destroy = bnxt_flow_destroy,
2282 	.flush = bnxt_flow_flush,
2283 };
2284