xref: /openbmc/linux/drivers/infiniband/hw/irdma/cm.c (revision 85250a24)
1 // SPDX-License-Identifier: GPL-2.0 or Linux-OpenIB
2 /* Copyright (c) 2015 - 2021 Intel Corporation */
3 #include "main.h"
4 #include "trace.h"
5 
6 static void irdma_cm_post_event(struct irdma_cm_event *event);
7 static void irdma_disconnect_worker(struct work_struct *work);
8 
9 /**
10  * irdma_free_sqbuf - put back puda buffer if refcount is 0
11  * @vsi: The VSI structure of the device
12  * @bufp: puda buffer to free
13  */
14 void irdma_free_sqbuf(struct irdma_sc_vsi *vsi, void *bufp)
15 {
16 	struct irdma_puda_buf *buf = bufp;
17 	struct irdma_puda_rsrc *ilq = vsi->ilq;
18 
19 	if (refcount_dec_and_test(&buf->refcount))
20 		irdma_puda_ret_bufpool(ilq, buf);
21 }
22 
23 /**
24  * irdma_record_ird_ord - Record IRD/ORD passed in
25  * @cm_node: connection's node
26  * @conn_ird: connection IRD
27  * @conn_ord: connection ORD
28  */
29 static void irdma_record_ird_ord(struct irdma_cm_node *cm_node, u32 conn_ird,
30 				 u32 conn_ord)
31 {
32 	if (conn_ird > cm_node->dev->hw_attrs.max_hw_ird)
33 		conn_ird = cm_node->dev->hw_attrs.max_hw_ird;
34 
35 	if (conn_ord > cm_node->dev->hw_attrs.max_hw_ord)
36 		conn_ord = cm_node->dev->hw_attrs.max_hw_ord;
37 	else if (!conn_ord && cm_node->send_rdma0_op == SEND_RDMA_READ_ZERO)
38 		conn_ord = 1;
39 	cm_node->ird_size = conn_ird;
40 	cm_node->ord_size = conn_ord;
41 }
42 
43 /**
44  * irdma_copy_ip_ntohl - copy IP address from  network to host
45  * @dst: IP address in host order
46  * @src: IP address in network order (big endian)
47  */
48 void irdma_copy_ip_ntohl(u32 *dst, __be32 *src)
49 {
50 	*dst++ = ntohl(*src++);
51 	*dst++ = ntohl(*src++);
52 	*dst++ = ntohl(*src++);
53 	*dst = ntohl(*src);
54 }
55 
56 /**
57  * irdma_copy_ip_htonl - copy IP address from host to network order
58  * @dst: IP address in network order (big endian)
59  * @src: IP address in host order
60  */
61 void irdma_copy_ip_htonl(__be32 *dst, u32 *src)
62 {
63 	*dst++ = htonl(*src++);
64 	*dst++ = htonl(*src++);
65 	*dst++ = htonl(*src++);
66 	*dst = htonl(*src);
67 }
68 
69 /**
70  * irdma_get_addr_info
71  * @cm_node: contains ip/tcp info
72  * @cm_info: to get a copy of the cm_node ip/tcp info
73  */
74 static void irdma_get_addr_info(struct irdma_cm_node *cm_node,
75 				struct irdma_cm_info *cm_info)
76 {
77 	memset(cm_info, 0, sizeof(*cm_info));
78 	cm_info->ipv4 = cm_node->ipv4;
79 	cm_info->vlan_id = cm_node->vlan_id;
80 	memcpy(cm_info->loc_addr, cm_node->loc_addr, sizeof(cm_info->loc_addr));
81 	memcpy(cm_info->rem_addr, cm_node->rem_addr, sizeof(cm_info->rem_addr));
82 	cm_info->loc_port = cm_node->loc_port;
83 	cm_info->rem_port = cm_node->rem_port;
84 }
85 
86 /**
87  * irdma_fill_sockaddr4 - fill in addr info for IPv4 connection
88  * @cm_node: connection's node
89  * @event: upper layer's cm event
90  */
91 static inline void irdma_fill_sockaddr4(struct irdma_cm_node *cm_node,
92 					struct iw_cm_event *event)
93 {
94 	struct sockaddr_in *laddr = (struct sockaddr_in *)&event->local_addr;
95 	struct sockaddr_in *raddr = (struct sockaddr_in *)&event->remote_addr;
96 
97 	laddr->sin_family = AF_INET;
98 	raddr->sin_family = AF_INET;
99 
100 	laddr->sin_port = htons(cm_node->loc_port);
101 	raddr->sin_port = htons(cm_node->rem_port);
102 
103 	laddr->sin_addr.s_addr = htonl(cm_node->loc_addr[0]);
104 	raddr->sin_addr.s_addr = htonl(cm_node->rem_addr[0]);
105 }
106 
107 /**
108  * irdma_fill_sockaddr6 - fill in addr info for IPv6 connection
109  * @cm_node: connection's node
110  * @event: upper layer's cm event
111  */
112 static inline void irdma_fill_sockaddr6(struct irdma_cm_node *cm_node,
113 					struct iw_cm_event *event)
114 {
115 	struct sockaddr_in6 *laddr6 = (struct sockaddr_in6 *)&event->local_addr;
116 	struct sockaddr_in6 *raddr6 = (struct sockaddr_in6 *)&event->remote_addr;
117 
118 	laddr6->sin6_family = AF_INET6;
119 	raddr6->sin6_family = AF_INET6;
120 
121 	laddr6->sin6_port = htons(cm_node->loc_port);
122 	raddr6->sin6_port = htons(cm_node->rem_port);
123 
124 	irdma_copy_ip_htonl(laddr6->sin6_addr.in6_u.u6_addr32,
125 			    cm_node->loc_addr);
126 	irdma_copy_ip_htonl(raddr6->sin6_addr.in6_u.u6_addr32,
127 			    cm_node->rem_addr);
128 }
129 
130 /**
131  * irdma_get_cmevent_info - for cm event upcall
132  * @cm_node: connection's node
133  * @cm_id: upper layers cm struct for the event
134  * @event: upper layer's cm event
135  */
136 static inline void irdma_get_cmevent_info(struct irdma_cm_node *cm_node,
137 					  struct iw_cm_id *cm_id,
138 					  struct iw_cm_event *event)
139 {
140 	memcpy(&event->local_addr, &cm_id->m_local_addr,
141 	       sizeof(event->local_addr));
142 	memcpy(&event->remote_addr, &cm_id->m_remote_addr,
143 	       sizeof(event->remote_addr));
144 	if (cm_node) {
145 		event->private_data = cm_node->pdata_buf;
146 		event->private_data_len = (u8)cm_node->pdata.size;
147 		event->ird = cm_node->ird_size;
148 		event->ord = cm_node->ord_size;
149 	}
150 }
151 
152 /**
153  * irdma_send_cm_event - upcall cm's event handler
154  * @cm_node: connection's node
155  * @cm_id: upper layer's cm info struct
156  * @type: Event type to indicate
157  * @status: status for the event type
158  */
159 static int irdma_send_cm_event(struct irdma_cm_node *cm_node,
160 			       struct iw_cm_id *cm_id,
161 			       enum iw_cm_event_type type, int status)
162 {
163 	struct iw_cm_event event = {};
164 
165 	event.event = type;
166 	event.status = status;
167 	trace_irdma_send_cm_event(cm_node, cm_id, type, status,
168 				  __builtin_return_address(0));
169 
170 	ibdev_dbg(&cm_node->iwdev->ibdev,
171 		  "CM: cm_node %p cm_id=%p state=%d accel=%d event_type=%d status=%d\n",
172 		  cm_node, cm_id, cm_node->accelerated, cm_node->state, type,
173 		  status);
174 
175 	switch (type) {
176 	case IW_CM_EVENT_CONNECT_REQUEST:
177 		if (cm_node->ipv4)
178 			irdma_fill_sockaddr4(cm_node, &event);
179 		else
180 			irdma_fill_sockaddr6(cm_node, &event);
181 		event.provider_data = cm_node;
182 		event.private_data = cm_node->pdata_buf;
183 		event.private_data_len = (u8)cm_node->pdata.size;
184 		event.ird = cm_node->ird_size;
185 		break;
186 	case IW_CM_EVENT_CONNECT_REPLY:
187 		irdma_get_cmevent_info(cm_node, cm_id, &event);
188 		break;
189 	case IW_CM_EVENT_ESTABLISHED:
190 		event.ird = cm_node->ird_size;
191 		event.ord = cm_node->ord_size;
192 		break;
193 	case IW_CM_EVENT_DISCONNECT:
194 	case IW_CM_EVENT_CLOSE:
195 		/* Wait if we are in RTS but havent issued the iwcm event upcall */
196 		if (!cm_node->accelerated)
197 			wait_for_completion(&cm_node->establish_comp);
198 		break;
199 	default:
200 		return -EINVAL;
201 	}
202 
203 	return cm_id->event_handler(cm_id, &event);
204 }
205 
206 /**
207  * irdma_timer_list_prep - add connection nodes to a list to perform timer tasks
208  * @cm_core: cm's core
209  * @timer_list: a timer list to which cm_node will be selected
210  */
211 static void irdma_timer_list_prep(struct irdma_cm_core *cm_core,
212 				  struct list_head *timer_list)
213 {
214 	struct irdma_cm_node *cm_node;
215 	int bkt;
216 
217 	hash_for_each_rcu(cm_core->cm_hash_tbl, bkt, cm_node, list) {
218 		if ((cm_node->close_entry || cm_node->send_entry) &&
219 		    refcount_inc_not_zero(&cm_node->refcnt))
220 			list_add(&cm_node->timer_entry, timer_list);
221 	}
222 }
223 
224 /**
225  * irdma_create_event - create cm event
226  * @cm_node: connection's node
227  * @type: Event type to generate
228  */
229 static struct irdma_cm_event *irdma_create_event(struct irdma_cm_node *cm_node,
230 						 enum irdma_cm_event_type type)
231 {
232 	struct irdma_cm_event *event;
233 
234 	if (!cm_node->cm_id)
235 		return NULL;
236 
237 	event = kzalloc(sizeof(*event), GFP_ATOMIC);
238 
239 	if (!event)
240 		return NULL;
241 
242 	event->type = type;
243 	event->cm_node = cm_node;
244 	memcpy(event->cm_info.rem_addr, cm_node->rem_addr,
245 	       sizeof(event->cm_info.rem_addr));
246 	memcpy(event->cm_info.loc_addr, cm_node->loc_addr,
247 	       sizeof(event->cm_info.loc_addr));
248 	event->cm_info.rem_port = cm_node->rem_port;
249 	event->cm_info.loc_port = cm_node->loc_port;
250 	event->cm_info.cm_id = cm_node->cm_id;
251 	ibdev_dbg(&cm_node->iwdev->ibdev,
252 		  "CM: node=%p event=%p type=%u dst=%pI4 src=%pI4\n", cm_node,
253 		  event, type, event->cm_info.loc_addr,
254 		  event->cm_info.rem_addr);
255 	trace_irdma_create_event(cm_node, type, __builtin_return_address(0));
256 	irdma_cm_post_event(event);
257 
258 	return event;
259 }
260 
261 /**
262  * irdma_free_retrans_entry - free send entry
263  * @cm_node: connection's node
264  */
265 static void irdma_free_retrans_entry(struct irdma_cm_node *cm_node)
266 {
267 	struct irdma_device *iwdev = cm_node->iwdev;
268 	struct irdma_timer_entry *send_entry;
269 
270 	send_entry = cm_node->send_entry;
271 	if (!send_entry)
272 		return;
273 
274 	cm_node->send_entry = NULL;
275 	irdma_free_sqbuf(&iwdev->vsi, send_entry->sqbuf);
276 	kfree(send_entry);
277 	refcount_dec(&cm_node->refcnt);
278 }
279 
280 /**
281  * irdma_cleanup_retrans_entry - free send entry with lock
282  * @cm_node: connection's node
283  */
284 static void irdma_cleanup_retrans_entry(struct irdma_cm_node *cm_node)
285 {
286 	unsigned long flags;
287 
288 	spin_lock_irqsave(&cm_node->retrans_list_lock, flags);
289 	irdma_free_retrans_entry(cm_node);
290 	spin_unlock_irqrestore(&cm_node->retrans_list_lock, flags);
291 }
292 
293 /**
294  * irdma_form_ah_cm_frame - get a free packet and build frame with address handle
295  * @cm_node: connection's node ionfo to use in frame
296  * @options: pointer to options info
297  * @hdr: pointer mpa header
298  * @pdata: pointer to private data
299  * @flags:  indicates FIN or ACK
300  */
301 static struct irdma_puda_buf *irdma_form_ah_cm_frame(struct irdma_cm_node *cm_node,
302 						     struct irdma_kmem_info *options,
303 						     struct irdma_kmem_info *hdr,
304 						     struct irdma_mpa_priv_info *pdata,
305 						     u8 flags)
306 {
307 	struct irdma_puda_buf *sqbuf;
308 	struct irdma_sc_vsi *vsi = &cm_node->iwdev->vsi;
309 	u8 *buf;
310 	struct tcphdr *tcph;
311 	u16 pktsize;
312 	u32 opts_len = 0;
313 	u32 pd_len = 0;
314 	u32 hdr_len = 0;
315 
316 	if (!cm_node->ah || !cm_node->ah->ah_info.ah_valid) {
317 		ibdev_dbg(&cm_node->iwdev->ibdev, "CM: AH invalid\n");
318 		return NULL;
319 	}
320 
321 	sqbuf = irdma_puda_get_bufpool(vsi->ilq);
322 	if (!sqbuf) {
323 		ibdev_dbg(&cm_node->iwdev->ibdev, "CM: SQ buf NULL\n");
324 		return NULL;
325 	}
326 
327 	sqbuf->ah_id = cm_node->ah->ah_info.ah_idx;
328 	buf = sqbuf->mem.va;
329 	if (options)
330 		opts_len = (u32)options->size;
331 
332 	if (hdr)
333 		hdr_len = hdr->size;
334 
335 	if (pdata)
336 		pd_len = pdata->size;
337 
338 	pktsize = sizeof(*tcph) + opts_len + hdr_len + pd_len;
339 
340 	memset(buf, 0, pktsize);
341 
342 	sqbuf->totallen = pktsize;
343 	sqbuf->tcphlen = sizeof(*tcph) + opts_len;
344 	sqbuf->scratch = cm_node;
345 
346 	tcph = (struct tcphdr *)buf;
347 	buf += sizeof(*tcph);
348 
349 	tcph->source = htons(cm_node->loc_port);
350 	tcph->dest = htons(cm_node->rem_port);
351 	tcph->seq = htonl(cm_node->tcp_cntxt.loc_seq_num);
352 
353 	if (flags & SET_ACK) {
354 		cm_node->tcp_cntxt.loc_ack_num = cm_node->tcp_cntxt.rcv_nxt;
355 		tcph->ack_seq = htonl(cm_node->tcp_cntxt.loc_ack_num);
356 		tcph->ack = 1;
357 	} else {
358 		tcph->ack_seq = 0;
359 	}
360 
361 	if (flags & SET_SYN) {
362 		cm_node->tcp_cntxt.loc_seq_num++;
363 		tcph->syn = 1;
364 	} else {
365 		cm_node->tcp_cntxt.loc_seq_num += hdr_len + pd_len;
366 	}
367 
368 	if (flags & SET_FIN) {
369 		cm_node->tcp_cntxt.loc_seq_num++;
370 		tcph->fin = 1;
371 	}
372 
373 	if (flags & SET_RST)
374 		tcph->rst = 1;
375 
376 	tcph->doff = (u16)((sizeof(*tcph) + opts_len + 3) >> 2);
377 	sqbuf->tcphlen = tcph->doff << 2;
378 	tcph->window = htons(cm_node->tcp_cntxt.rcv_wnd);
379 	tcph->urg_ptr = 0;
380 
381 	if (opts_len) {
382 		memcpy(buf, options->addr, opts_len);
383 		buf += opts_len;
384 	}
385 
386 	if (hdr_len) {
387 		memcpy(buf, hdr->addr, hdr_len);
388 		buf += hdr_len;
389 	}
390 
391 	if (pdata && pdata->addr)
392 		memcpy(buf, pdata->addr, pdata->size);
393 
394 	refcount_set(&sqbuf->refcount, 1);
395 
396 	print_hex_dump_debug("ILQ: TRANSMIT ILQ BUFFER", DUMP_PREFIX_OFFSET,
397 			     16, 8, sqbuf->mem.va, sqbuf->totallen, false);
398 
399 	return sqbuf;
400 }
401 
402 /**
403  * irdma_form_uda_cm_frame - get a free packet and build frame full tcpip packet
404  * @cm_node: connection's node ionfo to use in frame
405  * @options: pointer to options info
406  * @hdr: pointer mpa header
407  * @pdata: pointer to private data
408  * @flags:  indicates FIN or ACK
409  */
410 static struct irdma_puda_buf *irdma_form_uda_cm_frame(struct irdma_cm_node *cm_node,
411 						      struct irdma_kmem_info *options,
412 						      struct irdma_kmem_info *hdr,
413 						      struct irdma_mpa_priv_info *pdata,
414 						      u8 flags)
415 {
416 	struct irdma_puda_buf *sqbuf;
417 	struct irdma_sc_vsi *vsi = &cm_node->iwdev->vsi;
418 	u8 *buf;
419 
420 	struct tcphdr *tcph;
421 	struct iphdr *iph;
422 	struct ipv6hdr *ip6h;
423 	struct ethhdr *ethh;
424 	u16 pktsize;
425 	u16 eth_hlen = ETH_HLEN;
426 	u32 opts_len = 0;
427 	u32 pd_len = 0;
428 	u32 hdr_len = 0;
429 
430 	u16 vtag;
431 
432 	sqbuf = irdma_puda_get_bufpool(vsi->ilq);
433 	if (!sqbuf)
434 		return NULL;
435 
436 	buf = sqbuf->mem.va;
437 
438 	if (options)
439 		opts_len = (u32)options->size;
440 
441 	if (hdr)
442 		hdr_len = hdr->size;
443 
444 	if (pdata)
445 		pd_len = pdata->size;
446 
447 	if (cm_node->vlan_id < VLAN_N_VID)
448 		eth_hlen += 4;
449 
450 	if (cm_node->ipv4)
451 		pktsize = sizeof(*iph) + sizeof(*tcph);
452 	else
453 		pktsize = sizeof(*ip6h) + sizeof(*tcph);
454 	pktsize += opts_len + hdr_len + pd_len;
455 
456 	memset(buf, 0, eth_hlen + pktsize);
457 
458 	sqbuf->totallen = pktsize + eth_hlen;
459 	sqbuf->maclen = eth_hlen;
460 	sqbuf->tcphlen = sizeof(*tcph) + opts_len;
461 	sqbuf->scratch = cm_node;
462 
463 	ethh = (struct ethhdr *)buf;
464 	buf += eth_hlen;
465 
466 	if (cm_node->do_lpb)
467 		sqbuf->do_lpb = true;
468 
469 	if (cm_node->ipv4) {
470 		sqbuf->ipv4 = true;
471 
472 		iph = (struct iphdr *)buf;
473 		buf += sizeof(*iph);
474 		tcph = (struct tcphdr *)buf;
475 		buf += sizeof(*tcph);
476 
477 		ether_addr_copy(ethh->h_dest, cm_node->rem_mac);
478 		ether_addr_copy(ethh->h_source, cm_node->loc_mac);
479 		if (cm_node->vlan_id < VLAN_N_VID) {
480 			((struct vlan_ethhdr *)ethh)->h_vlan_proto =
481 				htons(ETH_P_8021Q);
482 			vtag = (cm_node->user_pri << VLAN_PRIO_SHIFT) |
483 			       cm_node->vlan_id;
484 			((struct vlan_ethhdr *)ethh)->h_vlan_TCI = htons(vtag);
485 
486 			((struct vlan_ethhdr *)ethh)->h_vlan_encapsulated_proto =
487 				htons(ETH_P_IP);
488 		} else {
489 			ethh->h_proto = htons(ETH_P_IP);
490 		}
491 
492 		iph->version = IPVERSION;
493 		iph->ihl = 5; /* 5 * 4Byte words, IP headr len */
494 		iph->tos = cm_node->tos;
495 		iph->tot_len = htons(pktsize);
496 		iph->id = htons(++cm_node->tcp_cntxt.loc_id);
497 
498 		iph->frag_off = htons(0x4000);
499 		iph->ttl = 0x40;
500 		iph->protocol = IPPROTO_TCP;
501 		iph->saddr = htonl(cm_node->loc_addr[0]);
502 		iph->daddr = htonl(cm_node->rem_addr[0]);
503 	} else {
504 		sqbuf->ipv4 = false;
505 		ip6h = (struct ipv6hdr *)buf;
506 		buf += sizeof(*ip6h);
507 		tcph = (struct tcphdr *)buf;
508 		buf += sizeof(*tcph);
509 
510 		ether_addr_copy(ethh->h_dest, cm_node->rem_mac);
511 		ether_addr_copy(ethh->h_source, cm_node->loc_mac);
512 		if (cm_node->vlan_id < VLAN_N_VID) {
513 			((struct vlan_ethhdr *)ethh)->h_vlan_proto =
514 				htons(ETH_P_8021Q);
515 			vtag = (cm_node->user_pri << VLAN_PRIO_SHIFT) |
516 			       cm_node->vlan_id;
517 			((struct vlan_ethhdr *)ethh)->h_vlan_TCI = htons(vtag);
518 			((struct vlan_ethhdr *)ethh)->h_vlan_encapsulated_proto =
519 				htons(ETH_P_IPV6);
520 		} else {
521 			ethh->h_proto = htons(ETH_P_IPV6);
522 		}
523 		ip6h->version = 6;
524 		ip6h->priority = cm_node->tos >> 4;
525 		ip6h->flow_lbl[0] = cm_node->tos << 4;
526 		ip6h->flow_lbl[1] = 0;
527 		ip6h->flow_lbl[2] = 0;
528 		ip6h->payload_len = htons(pktsize - sizeof(*ip6h));
529 		ip6h->nexthdr = 6;
530 		ip6h->hop_limit = 128;
531 		irdma_copy_ip_htonl(ip6h->saddr.in6_u.u6_addr32,
532 				    cm_node->loc_addr);
533 		irdma_copy_ip_htonl(ip6h->daddr.in6_u.u6_addr32,
534 				    cm_node->rem_addr);
535 	}
536 
537 	tcph->source = htons(cm_node->loc_port);
538 	tcph->dest = htons(cm_node->rem_port);
539 	tcph->seq = htonl(cm_node->tcp_cntxt.loc_seq_num);
540 
541 	if (flags & SET_ACK) {
542 		cm_node->tcp_cntxt.loc_ack_num = cm_node->tcp_cntxt.rcv_nxt;
543 		tcph->ack_seq = htonl(cm_node->tcp_cntxt.loc_ack_num);
544 		tcph->ack = 1;
545 	} else {
546 		tcph->ack_seq = 0;
547 	}
548 
549 	if (flags & SET_SYN) {
550 		cm_node->tcp_cntxt.loc_seq_num++;
551 		tcph->syn = 1;
552 	} else {
553 		cm_node->tcp_cntxt.loc_seq_num += hdr_len + pd_len;
554 	}
555 
556 	if (flags & SET_FIN) {
557 		cm_node->tcp_cntxt.loc_seq_num++;
558 		tcph->fin = 1;
559 	}
560 
561 	if (flags & SET_RST)
562 		tcph->rst = 1;
563 
564 	tcph->doff = (u16)((sizeof(*tcph) + opts_len + 3) >> 2);
565 	sqbuf->tcphlen = tcph->doff << 2;
566 	tcph->window = htons(cm_node->tcp_cntxt.rcv_wnd);
567 	tcph->urg_ptr = 0;
568 
569 	if (opts_len) {
570 		memcpy(buf, options->addr, opts_len);
571 		buf += opts_len;
572 	}
573 
574 	if (hdr_len) {
575 		memcpy(buf, hdr->addr, hdr_len);
576 		buf += hdr_len;
577 	}
578 
579 	if (pdata && pdata->addr)
580 		memcpy(buf, pdata->addr, pdata->size);
581 
582 	refcount_set(&sqbuf->refcount, 1);
583 
584 	print_hex_dump_debug("ILQ: TRANSMIT ILQ BUFFER", DUMP_PREFIX_OFFSET,
585 			     16, 8, sqbuf->mem.va, sqbuf->totallen, false);
586 	return sqbuf;
587 }
588 
589 /**
590  * irdma_send_reset - Send RST packet
591  * @cm_node: connection's node
592  */
593 int irdma_send_reset(struct irdma_cm_node *cm_node)
594 {
595 	struct irdma_puda_buf *sqbuf;
596 	int flags = SET_RST | SET_ACK;
597 
598 	trace_irdma_send_reset(cm_node, 0, __builtin_return_address(0));
599 	sqbuf = cm_node->cm_core->form_cm_frame(cm_node, NULL, NULL, NULL,
600 						flags);
601 	if (!sqbuf)
602 		return -ENOMEM;
603 
604 	ibdev_dbg(&cm_node->iwdev->ibdev,
605 		  "CM: caller: %pS cm_node %p cm_id=%p accel=%d state=%d rem_port=0x%04x, loc_port=0x%04x rem_addr=%pI4 loc_addr=%pI4\n",
606 		  __builtin_return_address(0), cm_node, cm_node->cm_id,
607 		  cm_node->accelerated, cm_node->state, cm_node->rem_port,
608 		  cm_node->loc_port, cm_node->rem_addr, cm_node->loc_addr);
609 
610 	return irdma_schedule_cm_timer(cm_node, sqbuf, IRDMA_TIMER_TYPE_SEND, 0,
611 				       1);
612 }
613 
614 /**
615  * irdma_active_open_err - send event for active side cm error
616  * @cm_node: connection's node
617  * @reset: Flag to send reset or not
618  */
619 static void irdma_active_open_err(struct irdma_cm_node *cm_node, bool reset)
620 {
621 	trace_irdma_active_open_err(cm_node, reset,
622 				    __builtin_return_address(0));
623 	irdma_cleanup_retrans_entry(cm_node);
624 	cm_node->cm_core->stats_connect_errs++;
625 	if (reset) {
626 		ibdev_dbg(&cm_node->iwdev->ibdev,
627 			  "CM: cm_node=%p state=%d\n", cm_node,
628 			  cm_node->state);
629 		refcount_inc(&cm_node->refcnt);
630 		irdma_send_reset(cm_node);
631 	}
632 
633 	cm_node->state = IRDMA_CM_STATE_CLOSED;
634 	irdma_create_event(cm_node, IRDMA_CM_EVENT_ABORTED);
635 }
636 
637 /**
638  * irdma_passive_open_err - handle passive side cm error
639  * @cm_node: connection's node
640  * @reset: send reset or just free cm_node
641  */
642 static void irdma_passive_open_err(struct irdma_cm_node *cm_node, bool reset)
643 {
644 	irdma_cleanup_retrans_entry(cm_node);
645 	cm_node->cm_core->stats_passive_errs++;
646 	cm_node->state = IRDMA_CM_STATE_CLOSED;
647 	ibdev_dbg(&cm_node->iwdev->ibdev, "CM: cm_node=%p state =%d\n",
648 		  cm_node, cm_node->state);
649 	trace_irdma_passive_open_err(cm_node, reset,
650 				     __builtin_return_address(0));
651 	if (reset)
652 		irdma_send_reset(cm_node);
653 	else
654 		irdma_rem_ref_cm_node(cm_node);
655 }
656 
657 /**
658  * irdma_event_connect_error - to create connect error event
659  * @event: cm information for connect event
660  */
661 static void irdma_event_connect_error(struct irdma_cm_event *event)
662 {
663 	struct irdma_qp *iwqp;
664 	struct iw_cm_id *cm_id;
665 
666 	cm_id = event->cm_node->cm_id;
667 	if (!cm_id)
668 		return;
669 
670 	iwqp = cm_id->provider_data;
671 
672 	if (!iwqp || !iwqp->iwdev)
673 		return;
674 
675 	iwqp->cm_id = NULL;
676 	cm_id->provider_data = NULL;
677 	irdma_send_cm_event(event->cm_node, cm_id, IW_CM_EVENT_CONNECT_REPLY,
678 			    -ECONNRESET);
679 	irdma_rem_ref_cm_node(event->cm_node);
680 }
681 
682 /**
683  * irdma_process_options - process options from TCP header
684  * @cm_node: connection's node
685  * @optionsloc: point to start of options
686  * @optionsize: size of all options
687  * @syn_pkt: flag if syn packet
688  */
689 static int irdma_process_options(struct irdma_cm_node *cm_node, u8 *optionsloc,
690 				 u32 optionsize, u32 syn_pkt)
691 {
692 	u32 tmp;
693 	u32 offset = 0;
694 	union all_known_options *all_options;
695 	char got_mss_option = 0;
696 
697 	while (offset < optionsize) {
698 		all_options = (union all_known_options *)(optionsloc + offset);
699 		switch (all_options->base.optionnum) {
700 		case OPTION_NUM_EOL:
701 			offset = optionsize;
702 			break;
703 		case OPTION_NUM_NONE:
704 			offset += 1;
705 			continue;
706 		case OPTION_NUM_MSS:
707 			ibdev_dbg(&cm_node->iwdev->ibdev,
708 				  "CM: MSS Length: %d Offset: %d Size: %d\n",
709 				  all_options->mss.len, offset, optionsize);
710 			got_mss_option = 1;
711 			if (all_options->mss.len != 4)
712 				return -EINVAL;
713 			tmp = ntohs(all_options->mss.mss);
714 			if ((cm_node->ipv4 &&
715 			     (tmp + IRDMA_MTU_TO_MSS_IPV4) < IRDMA_MIN_MTU_IPV4) ||
716 			    (!cm_node->ipv4 &&
717 			     (tmp + IRDMA_MTU_TO_MSS_IPV6) < IRDMA_MIN_MTU_IPV6))
718 				return -EINVAL;
719 			if (tmp < cm_node->tcp_cntxt.mss)
720 				cm_node->tcp_cntxt.mss = tmp;
721 			break;
722 		case OPTION_NUM_WINDOW_SCALE:
723 			cm_node->tcp_cntxt.snd_wscale =
724 				all_options->windowscale.shiftcount;
725 			break;
726 		default:
727 			ibdev_dbg(&cm_node->iwdev->ibdev,
728 				  "CM: Unsupported TCP Option: %x\n",
729 				  all_options->base.optionnum);
730 			break;
731 		}
732 		offset += all_options->base.len;
733 	}
734 	if (!got_mss_option && syn_pkt)
735 		cm_node->tcp_cntxt.mss = IRDMA_CM_DEFAULT_MSS;
736 
737 	return 0;
738 }
739 
740 /**
741  * irdma_handle_tcp_options - setup TCP context info after parsing TCP options
742  * @cm_node: connection's node
743  * @tcph: pointer tcp header
744  * @optionsize: size of options rcvd
745  * @passive: active or passive flag
746  */
747 static int irdma_handle_tcp_options(struct irdma_cm_node *cm_node,
748 				    struct tcphdr *tcph, int optionsize,
749 				    int passive)
750 {
751 	u8 *optionsloc = (u8 *)&tcph[1];
752 	int ret;
753 
754 	if (optionsize) {
755 		ret = irdma_process_options(cm_node, optionsloc, optionsize,
756 					    (u32)tcph->syn);
757 		if (ret) {
758 			ibdev_dbg(&cm_node->iwdev->ibdev,
759 				  "CM: Node %p, Sending Reset\n", cm_node);
760 			if (passive)
761 				irdma_passive_open_err(cm_node, true);
762 			else
763 				irdma_active_open_err(cm_node, true);
764 			return ret;
765 		}
766 	}
767 
768 	cm_node->tcp_cntxt.snd_wnd = ntohs(tcph->window)
769 				     << cm_node->tcp_cntxt.snd_wscale;
770 
771 	if (cm_node->tcp_cntxt.snd_wnd > cm_node->tcp_cntxt.max_snd_wnd)
772 		cm_node->tcp_cntxt.max_snd_wnd = cm_node->tcp_cntxt.snd_wnd;
773 
774 	return 0;
775 }
776 
777 /**
778  * irdma_build_mpa_v1 - build a MPA V1 frame
779  * @cm_node: connection's node
780  * @start_addr: address where to build frame
781  * @mpa_key: to do read0 or write0
782  */
783 static void irdma_build_mpa_v1(struct irdma_cm_node *cm_node, void *start_addr,
784 			       u8 mpa_key)
785 {
786 	struct ietf_mpa_v1 *mpa_frame = start_addr;
787 
788 	switch (mpa_key) {
789 	case MPA_KEY_REQUEST:
790 		memcpy(mpa_frame->key, IEFT_MPA_KEY_REQ, IETF_MPA_KEY_SIZE);
791 		break;
792 	case MPA_KEY_REPLY:
793 		memcpy(mpa_frame->key, IEFT_MPA_KEY_REP, IETF_MPA_KEY_SIZE);
794 		break;
795 	default:
796 		break;
797 	}
798 	mpa_frame->flags = IETF_MPA_FLAGS_CRC;
799 	mpa_frame->rev = cm_node->mpa_frame_rev;
800 	mpa_frame->priv_data_len = htons(cm_node->pdata.size);
801 }
802 
803 /**
804  * irdma_build_mpa_v2 - build a MPA V2 frame
805  * @cm_node: connection's node
806  * @start_addr: buffer start address
807  * @mpa_key: to do read0 or write0
808  */
809 static void irdma_build_mpa_v2(struct irdma_cm_node *cm_node, void *start_addr,
810 			       u8 mpa_key)
811 {
812 	struct ietf_mpa_v2 *mpa_frame = start_addr;
813 	struct ietf_rtr_msg *rtr_msg = &mpa_frame->rtr_msg;
814 	u16 ctrl_ird, ctrl_ord;
815 
816 	/* initialize the upper 5 bytes of the frame */
817 	irdma_build_mpa_v1(cm_node, start_addr, mpa_key);
818 	mpa_frame->flags |= IETF_MPA_V2_FLAG;
819 	if (cm_node->iwdev->iw_ooo) {
820 		mpa_frame->flags |= IETF_MPA_FLAGS_MARKERS;
821 		cm_node->rcv_mark_en = true;
822 	}
823 	mpa_frame->priv_data_len = cpu_to_be16(be16_to_cpu(mpa_frame->priv_data_len) +
824 					       IETF_RTR_MSG_SIZE);
825 
826 	/* initialize RTR msg */
827 	if (cm_node->mpav2_ird_ord == IETF_NO_IRD_ORD) {
828 		ctrl_ird = IETF_NO_IRD_ORD;
829 		ctrl_ord = IETF_NO_IRD_ORD;
830 	} else {
831 		ctrl_ird = (cm_node->ird_size > IETF_NO_IRD_ORD) ?
832 				   IETF_NO_IRD_ORD :
833 				   cm_node->ird_size;
834 		ctrl_ord = (cm_node->ord_size > IETF_NO_IRD_ORD) ?
835 				   IETF_NO_IRD_ORD :
836 				   cm_node->ord_size;
837 	}
838 	ctrl_ird |= IETF_PEER_TO_PEER;
839 
840 	switch (mpa_key) {
841 	case MPA_KEY_REQUEST:
842 		ctrl_ord |= IETF_RDMA0_WRITE;
843 		ctrl_ord |= IETF_RDMA0_READ;
844 		break;
845 	case MPA_KEY_REPLY:
846 		switch (cm_node->send_rdma0_op) {
847 		case SEND_RDMA_WRITE_ZERO:
848 			ctrl_ord |= IETF_RDMA0_WRITE;
849 			break;
850 		case SEND_RDMA_READ_ZERO:
851 			ctrl_ord |= IETF_RDMA0_READ;
852 			break;
853 		}
854 		break;
855 	default:
856 		break;
857 	}
858 	rtr_msg->ctrl_ird = htons(ctrl_ird);
859 	rtr_msg->ctrl_ord = htons(ctrl_ord);
860 }
861 
862 /**
863  * irdma_cm_build_mpa_frame - build mpa frame for mpa version 1 or version 2
864  * @cm_node: connection's node
865  * @mpa: mpa: data buffer
866  * @mpa_key: to do read0 or write0
867  */
868 static int irdma_cm_build_mpa_frame(struct irdma_cm_node *cm_node,
869 				    struct irdma_kmem_info *mpa, u8 mpa_key)
870 {
871 	int hdr_len = 0;
872 
873 	switch (cm_node->mpa_frame_rev) {
874 	case IETF_MPA_V1:
875 		hdr_len = sizeof(struct ietf_mpa_v1);
876 		irdma_build_mpa_v1(cm_node, mpa->addr, mpa_key);
877 		break;
878 	case IETF_MPA_V2:
879 		hdr_len = sizeof(struct ietf_mpa_v2);
880 		irdma_build_mpa_v2(cm_node, mpa->addr, mpa_key);
881 		break;
882 	default:
883 		break;
884 	}
885 
886 	return hdr_len;
887 }
888 
889 /**
890  * irdma_send_mpa_request - active node send mpa request to passive node
891  * @cm_node: connection's node
892  */
893 static int irdma_send_mpa_request(struct irdma_cm_node *cm_node)
894 {
895 	struct irdma_puda_buf *sqbuf;
896 
897 	cm_node->mpa_hdr.addr = &cm_node->mpa_v2_frame;
898 	cm_node->mpa_hdr.size = irdma_cm_build_mpa_frame(cm_node,
899 							 &cm_node->mpa_hdr,
900 							 MPA_KEY_REQUEST);
901 	if (!cm_node->mpa_hdr.size) {
902 		ibdev_dbg(&cm_node->iwdev->ibdev,
903 			  "CM: mpa size = %d\n", cm_node->mpa_hdr.size);
904 		return -EINVAL;
905 	}
906 
907 	sqbuf = cm_node->cm_core->form_cm_frame(cm_node, NULL,
908 						&cm_node->mpa_hdr,
909 						&cm_node->pdata, SET_ACK);
910 	if (!sqbuf)
911 		return -ENOMEM;
912 
913 	return irdma_schedule_cm_timer(cm_node, sqbuf, IRDMA_TIMER_TYPE_SEND, 1,
914 				       0);
915 }
916 
917 /**
918  * irdma_send_mpa_reject -
919  * @cm_node: connection's node
920  * @pdata: reject data for connection
921  * @plen: length of reject data
922  */
923 static int irdma_send_mpa_reject(struct irdma_cm_node *cm_node,
924 				 const void *pdata, u8 plen)
925 {
926 	struct irdma_puda_buf *sqbuf;
927 	struct irdma_mpa_priv_info priv_info;
928 
929 	cm_node->mpa_hdr.addr = &cm_node->mpa_v2_frame;
930 	cm_node->mpa_hdr.size = irdma_cm_build_mpa_frame(cm_node,
931 							 &cm_node->mpa_hdr,
932 							 MPA_KEY_REPLY);
933 
934 	cm_node->mpa_frame.flags |= IETF_MPA_FLAGS_REJECT;
935 	priv_info.addr = pdata;
936 	priv_info.size = plen;
937 
938 	sqbuf = cm_node->cm_core->form_cm_frame(cm_node, NULL,
939 						&cm_node->mpa_hdr, &priv_info,
940 						SET_ACK | SET_FIN);
941 	if (!sqbuf)
942 		return -ENOMEM;
943 
944 	cm_node->state = IRDMA_CM_STATE_FIN_WAIT1;
945 
946 	return irdma_schedule_cm_timer(cm_node, sqbuf, IRDMA_TIMER_TYPE_SEND, 1,
947 				       0);
948 }
949 
950 /**
951  * irdma_negotiate_mpa_v2_ird_ord - negotiate MPAv2 IRD/ORD
952  * @cm_node: connection's node
953  * @buf: Data pointer
954  */
955 static int irdma_negotiate_mpa_v2_ird_ord(struct irdma_cm_node *cm_node,
956 					  u8 *buf)
957 {
958 	struct ietf_mpa_v2 *mpa_v2_frame;
959 	struct ietf_rtr_msg *rtr_msg;
960 	u16 ird_size;
961 	u16 ord_size;
962 	u16 ctrl_ord;
963 	u16 ctrl_ird;
964 
965 	mpa_v2_frame = (struct ietf_mpa_v2 *)buf;
966 	rtr_msg = &mpa_v2_frame->rtr_msg;
967 
968 	/* parse rtr message */
969 	ctrl_ord = ntohs(rtr_msg->ctrl_ord);
970 	ctrl_ird = ntohs(rtr_msg->ctrl_ird);
971 	ird_size = ctrl_ird & IETF_NO_IRD_ORD;
972 	ord_size = ctrl_ord & IETF_NO_IRD_ORD;
973 
974 	if (!(ctrl_ird & IETF_PEER_TO_PEER))
975 		return -EOPNOTSUPP;
976 
977 	if (ird_size == IETF_NO_IRD_ORD || ord_size == IETF_NO_IRD_ORD) {
978 		cm_node->mpav2_ird_ord = IETF_NO_IRD_ORD;
979 		goto negotiate_done;
980 	}
981 
982 	if (cm_node->state != IRDMA_CM_STATE_MPAREQ_SENT) {
983 		/* responder */
984 		if (!ord_size && (ctrl_ord & IETF_RDMA0_READ))
985 			cm_node->ird_size = 1;
986 		if (cm_node->ord_size > ird_size)
987 			cm_node->ord_size = ird_size;
988 	} else {
989 		/* initiator */
990 		if (!ird_size && (ctrl_ord & IETF_RDMA0_READ))
991 			/* Remote peer doesn't support RDMA0_READ */
992 			return -EOPNOTSUPP;
993 
994 		if (cm_node->ord_size > ird_size)
995 			cm_node->ord_size = ird_size;
996 
997 		if (cm_node->ird_size < ord_size)
998 		/* no resources available */
999 			return -EINVAL;
1000 	}
1001 
1002 negotiate_done:
1003 	if (ctrl_ord & IETF_RDMA0_READ)
1004 		cm_node->send_rdma0_op = SEND_RDMA_READ_ZERO;
1005 	else if (ctrl_ord & IETF_RDMA0_WRITE)
1006 		cm_node->send_rdma0_op = SEND_RDMA_WRITE_ZERO;
1007 	else
1008 		/* Not supported RDMA0 operation */
1009 		return -EOPNOTSUPP;
1010 
1011 	ibdev_dbg(&cm_node->iwdev->ibdev,
1012 		  "CM: MPAV2 Negotiated ORD: %d, IRD: %d\n",
1013 		  cm_node->ord_size, cm_node->ird_size);
1014 	trace_irdma_negotiate_mpa_v2(cm_node);
1015 	return 0;
1016 }
1017 
1018 /**
1019  * irdma_parse_mpa - process an IETF MPA frame
1020  * @cm_node: connection's node
1021  * @buf: Data pointer
1022  * @type: to return accept or reject
1023  * @len: Len of mpa buffer
1024  */
1025 static int irdma_parse_mpa(struct irdma_cm_node *cm_node, u8 *buf, u32 *type,
1026 			   u32 len)
1027 {
1028 	struct ietf_mpa_v1 *mpa_frame;
1029 	int mpa_hdr_len, priv_data_len, ret;
1030 
1031 	*type = IRDMA_MPA_REQUEST_ACCEPT;
1032 
1033 	if (len < sizeof(struct ietf_mpa_v1)) {
1034 		ibdev_dbg(&cm_node->iwdev->ibdev,
1035 			  "CM: ietf buffer small (%x)\n", len);
1036 		return -EINVAL;
1037 	}
1038 
1039 	mpa_frame = (struct ietf_mpa_v1 *)buf;
1040 	mpa_hdr_len = sizeof(struct ietf_mpa_v1);
1041 	priv_data_len = ntohs(mpa_frame->priv_data_len);
1042 
1043 	if (priv_data_len > IETF_MAX_PRIV_DATA_LEN) {
1044 		ibdev_dbg(&cm_node->iwdev->ibdev,
1045 			  "CM: private_data too big %d\n", priv_data_len);
1046 		return -EOVERFLOW;
1047 	}
1048 
1049 	if (mpa_frame->rev != IETF_MPA_V1 && mpa_frame->rev != IETF_MPA_V2) {
1050 		ibdev_dbg(&cm_node->iwdev->ibdev,
1051 			  "CM: unsupported mpa rev = %d\n", mpa_frame->rev);
1052 		return -EINVAL;
1053 	}
1054 
1055 	if (mpa_frame->rev > cm_node->mpa_frame_rev) {
1056 		ibdev_dbg(&cm_node->iwdev->ibdev, "CM: rev %d\n",
1057 			  mpa_frame->rev);
1058 		return -EINVAL;
1059 	}
1060 
1061 	cm_node->mpa_frame_rev = mpa_frame->rev;
1062 	if (cm_node->state != IRDMA_CM_STATE_MPAREQ_SENT) {
1063 		if (memcmp(mpa_frame->key, IEFT_MPA_KEY_REQ,
1064 			   IETF_MPA_KEY_SIZE)) {
1065 			ibdev_dbg(&cm_node->iwdev->ibdev,
1066 				  "CM: Unexpected MPA Key received\n");
1067 			return -EINVAL;
1068 		}
1069 	} else {
1070 		if (memcmp(mpa_frame->key, IEFT_MPA_KEY_REP,
1071 			   IETF_MPA_KEY_SIZE)) {
1072 			ibdev_dbg(&cm_node->iwdev->ibdev,
1073 				  "CM: Unexpected MPA Key received\n");
1074 			return -EINVAL;
1075 		}
1076 	}
1077 
1078 	if (priv_data_len + mpa_hdr_len > len) {
1079 		ibdev_dbg(&cm_node->iwdev->ibdev,
1080 			  "CM: ietf buffer len(%x + %x != %x)\n",
1081 			  priv_data_len, mpa_hdr_len, len);
1082 		return -EOVERFLOW;
1083 	}
1084 
1085 	if (len > IRDMA_MAX_CM_BUF) {
1086 		ibdev_dbg(&cm_node->iwdev->ibdev,
1087 			  "CM: ietf buffer large len = %d\n", len);
1088 		return -EOVERFLOW;
1089 	}
1090 
1091 	switch (mpa_frame->rev) {
1092 	case IETF_MPA_V2:
1093 		mpa_hdr_len += IETF_RTR_MSG_SIZE;
1094 		ret = irdma_negotiate_mpa_v2_ird_ord(cm_node, buf);
1095 		if (ret)
1096 			return ret;
1097 		break;
1098 	case IETF_MPA_V1:
1099 	default:
1100 		break;
1101 	}
1102 
1103 	memcpy(cm_node->pdata_buf, buf + mpa_hdr_len, priv_data_len);
1104 	cm_node->pdata.size = priv_data_len;
1105 
1106 	if (mpa_frame->flags & IETF_MPA_FLAGS_REJECT)
1107 		*type = IRDMA_MPA_REQUEST_REJECT;
1108 
1109 	if (mpa_frame->flags & IETF_MPA_FLAGS_MARKERS)
1110 		cm_node->snd_mark_en = true;
1111 
1112 	return 0;
1113 }
1114 
1115 /**
1116  * irdma_schedule_cm_timer
1117  * @cm_node: connection's node
1118  * @sqbuf: buffer to send
1119  * @type: if it is send or close
1120  * @send_retrans: if rexmits to be done
1121  * @close_when_complete: is cm_node to be removed
1122  *
1123  * note - cm_node needs to be protected before calling this. Encase in:
1124  *		irdma_rem_ref_cm_node(cm_core, cm_node);
1125  *		irdma_schedule_cm_timer(...)
1126  *		refcount_inc(&cm_node->refcnt);
1127  */
1128 int irdma_schedule_cm_timer(struct irdma_cm_node *cm_node,
1129 			    struct irdma_puda_buf *sqbuf,
1130 			    enum irdma_timer_type type, int send_retrans,
1131 			    int close_when_complete)
1132 {
1133 	struct irdma_sc_vsi *vsi = &cm_node->iwdev->vsi;
1134 	struct irdma_cm_core *cm_core = cm_node->cm_core;
1135 	struct irdma_timer_entry *new_send;
1136 	u32 was_timer_set;
1137 	unsigned long flags;
1138 
1139 	new_send = kzalloc(sizeof(*new_send), GFP_ATOMIC);
1140 	if (!new_send) {
1141 		if (type != IRDMA_TIMER_TYPE_CLOSE)
1142 			irdma_free_sqbuf(vsi, sqbuf);
1143 		return -ENOMEM;
1144 	}
1145 
1146 	new_send->retrycount = IRDMA_DEFAULT_RETRYS;
1147 	new_send->retranscount = IRDMA_DEFAULT_RETRANS;
1148 	new_send->sqbuf = sqbuf;
1149 	new_send->timetosend = jiffies;
1150 	new_send->type = type;
1151 	new_send->send_retrans = send_retrans;
1152 	new_send->close_when_complete = close_when_complete;
1153 
1154 	if (type == IRDMA_TIMER_TYPE_CLOSE) {
1155 		new_send->timetosend += (HZ / 10);
1156 		if (cm_node->close_entry) {
1157 			kfree(new_send);
1158 			ibdev_dbg(&cm_node->iwdev->ibdev,
1159 				  "CM: already close entry\n");
1160 			return -EINVAL;
1161 		}
1162 
1163 		cm_node->close_entry = new_send;
1164 	} else { /* type == IRDMA_TIMER_TYPE_SEND */
1165 		spin_lock_irqsave(&cm_node->retrans_list_lock, flags);
1166 		cm_node->send_entry = new_send;
1167 		refcount_inc(&cm_node->refcnt);
1168 		spin_unlock_irqrestore(&cm_node->retrans_list_lock, flags);
1169 		new_send->timetosend = jiffies + IRDMA_RETRY_TIMEOUT;
1170 
1171 		refcount_inc(&sqbuf->refcount);
1172 		irdma_puda_send_buf(vsi->ilq, sqbuf);
1173 		if (!send_retrans) {
1174 			irdma_cleanup_retrans_entry(cm_node);
1175 			if (close_when_complete)
1176 				irdma_rem_ref_cm_node(cm_node);
1177 			return 0;
1178 		}
1179 	}
1180 
1181 	spin_lock_irqsave(&cm_core->ht_lock, flags);
1182 	was_timer_set = timer_pending(&cm_core->tcp_timer);
1183 
1184 	if (!was_timer_set) {
1185 		cm_core->tcp_timer.expires = new_send->timetosend;
1186 		add_timer(&cm_core->tcp_timer);
1187 	}
1188 	spin_unlock_irqrestore(&cm_core->ht_lock, flags);
1189 
1190 	return 0;
1191 }
1192 
1193 /**
1194  * irdma_retrans_expired - Could not rexmit the packet
1195  * @cm_node: connection's node
1196  */
1197 static void irdma_retrans_expired(struct irdma_cm_node *cm_node)
1198 {
1199 	enum irdma_cm_node_state state = cm_node->state;
1200 
1201 	cm_node->state = IRDMA_CM_STATE_CLOSED;
1202 	switch (state) {
1203 	case IRDMA_CM_STATE_SYN_RCVD:
1204 	case IRDMA_CM_STATE_CLOSING:
1205 		irdma_rem_ref_cm_node(cm_node);
1206 		break;
1207 	case IRDMA_CM_STATE_FIN_WAIT1:
1208 	case IRDMA_CM_STATE_LAST_ACK:
1209 		irdma_send_reset(cm_node);
1210 		break;
1211 	default:
1212 		refcount_inc(&cm_node->refcnt);
1213 		irdma_send_reset(cm_node);
1214 		irdma_create_event(cm_node, IRDMA_CM_EVENT_ABORTED);
1215 		break;
1216 	}
1217 }
1218 
1219 /**
1220  * irdma_handle_close_entry - for handling retry/timeouts
1221  * @cm_node: connection's node
1222  * @rem_node: flag for remove cm_node
1223  */
1224 static void irdma_handle_close_entry(struct irdma_cm_node *cm_node,
1225 				     u32 rem_node)
1226 {
1227 	struct irdma_timer_entry *close_entry = cm_node->close_entry;
1228 	struct irdma_qp *iwqp;
1229 	unsigned long flags;
1230 
1231 	if (!close_entry)
1232 		return;
1233 	iwqp = (struct irdma_qp *)close_entry->sqbuf;
1234 	if (iwqp) {
1235 		spin_lock_irqsave(&iwqp->lock, flags);
1236 		if (iwqp->cm_id) {
1237 			iwqp->hw_tcp_state = IRDMA_TCP_STATE_CLOSED;
1238 			iwqp->hw_iwarp_state = IRDMA_QP_STATE_ERROR;
1239 			iwqp->last_aeq = IRDMA_AE_RESET_SENT;
1240 			iwqp->ibqp_state = IB_QPS_ERR;
1241 			spin_unlock_irqrestore(&iwqp->lock, flags);
1242 			irdma_cm_disconn(iwqp);
1243 		} else {
1244 			spin_unlock_irqrestore(&iwqp->lock, flags);
1245 		}
1246 	} else if (rem_node) {
1247 		/* TIME_WAIT state */
1248 		irdma_rem_ref_cm_node(cm_node);
1249 	}
1250 
1251 	kfree(close_entry);
1252 	cm_node->close_entry = NULL;
1253 }
1254 
1255 /**
1256  * irdma_cm_timer_tick - system's timer expired callback
1257  * @t: Pointer to timer_list
1258  */
1259 static void irdma_cm_timer_tick(struct timer_list *t)
1260 {
1261 	unsigned long nexttimeout = jiffies + IRDMA_LONG_TIME;
1262 	struct irdma_cm_node *cm_node;
1263 	struct irdma_timer_entry *send_entry, *close_entry;
1264 	struct list_head *list_core_temp;
1265 	struct list_head *list_node;
1266 	struct irdma_cm_core *cm_core = from_timer(cm_core, t, tcp_timer);
1267 	struct irdma_sc_vsi *vsi;
1268 	u32 settimer = 0;
1269 	unsigned long timetosend;
1270 	unsigned long flags;
1271 	struct list_head timer_list;
1272 
1273 	INIT_LIST_HEAD(&timer_list);
1274 
1275 	rcu_read_lock();
1276 	irdma_timer_list_prep(cm_core, &timer_list);
1277 	rcu_read_unlock();
1278 
1279 	list_for_each_safe (list_node, list_core_temp, &timer_list) {
1280 		cm_node = container_of(list_node, struct irdma_cm_node,
1281 				       timer_entry);
1282 		close_entry = cm_node->close_entry;
1283 
1284 		if (close_entry) {
1285 			if (time_after(close_entry->timetosend, jiffies)) {
1286 				if (nexttimeout > close_entry->timetosend ||
1287 				    !settimer) {
1288 					nexttimeout = close_entry->timetosend;
1289 					settimer = 1;
1290 				}
1291 			} else {
1292 				irdma_handle_close_entry(cm_node, 1);
1293 			}
1294 		}
1295 
1296 		spin_lock_irqsave(&cm_node->retrans_list_lock, flags);
1297 
1298 		send_entry = cm_node->send_entry;
1299 		if (!send_entry)
1300 			goto done;
1301 		if (time_after(send_entry->timetosend, jiffies)) {
1302 			if (cm_node->state != IRDMA_CM_STATE_OFFLOADED) {
1303 				if (nexttimeout > send_entry->timetosend ||
1304 				    !settimer) {
1305 					nexttimeout = send_entry->timetosend;
1306 					settimer = 1;
1307 				}
1308 			} else {
1309 				irdma_free_retrans_entry(cm_node);
1310 			}
1311 			goto done;
1312 		}
1313 
1314 		if (cm_node->state == IRDMA_CM_STATE_OFFLOADED ||
1315 		    cm_node->state == IRDMA_CM_STATE_CLOSED) {
1316 			irdma_free_retrans_entry(cm_node);
1317 			goto done;
1318 		}
1319 
1320 		if (!send_entry->retranscount || !send_entry->retrycount) {
1321 			irdma_free_retrans_entry(cm_node);
1322 
1323 			spin_unlock_irqrestore(&cm_node->retrans_list_lock,
1324 					       flags);
1325 			irdma_retrans_expired(cm_node);
1326 			cm_node->state = IRDMA_CM_STATE_CLOSED;
1327 			spin_lock_irqsave(&cm_node->retrans_list_lock, flags);
1328 			goto done;
1329 		}
1330 		spin_unlock_irqrestore(&cm_node->retrans_list_lock, flags);
1331 
1332 		vsi = &cm_node->iwdev->vsi;
1333 		if (!cm_node->ack_rcvd) {
1334 			refcount_inc(&send_entry->sqbuf->refcount);
1335 			irdma_puda_send_buf(vsi->ilq, send_entry->sqbuf);
1336 			cm_node->cm_core->stats_pkt_retrans++;
1337 		}
1338 
1339 		spin_lock_irqsave(&cm_node->retrans_list_lock, flags);
1340 		if (send_entry->send_retrans) {
1341 			send_entry->retranscount--;
1342 			timetosend = (IRDMA_RETRY_TIMEOUT <<
1343 				      (IRDMA_DEFAULT_RETRANS -
1344 				       send_entry->retranscount));
1345 
1346 			send_entry->timetosend = jiffies +
1347 			    min(timetosend, IRDMA_MAX_TIMEOUT);
1348 			if (nexttimeout > send_entry->timetosend || !settimer) {
1349 				nexttimeout = send_entry->timetosend;
1350 				settimer = 1;
1351 			}
1352 		} else {
1353 			int close_when_complete;
1354 
1355 			close_when_complete = send_entry->close_when_complete;
1356 			irdma_free_retrans_entry(cm_node);
1357 			if (close_when_complete)
1358 				irdma_rem_ref_cm_node(cm_node);
1359 		}
1360 done:
1361 		spin_unlock_irqrestore(&cm_node->retrans_list_lock, flags);
1362 		irdma_rem_ref_cm_node(cm_node);
1363 	}
1364 
1365 	if (settimer) {
1366 		spin_lock_irqsave(&cm_core->ht_lock, flags);
1367 		if (!timer_pending(&cm_core->tcp_timer)) {
1368 			cm_core->tcp_timer.expires = nexttimeout;
1369 			add_timer(&cm_core->tcp_timer);
1370 		}
1371 		spin_unlock_irqrestore(&cm_core->ht_lock, flags);
1372 	}
1373 }
1374 
1375 /**
1376  * irdma_send_syn - send SYN packet
1377  * @cm_node: connection's node
1378  * @sendack: flag to set ACK bit or not
1379  */
1380 int irdma_send_syn(struct irdma_cm_node *cm_node, u32 sendack)
1381 {
1382 	struct irdma_puda_buf *sqbuf;
1383 	int flags = SET_SYN;
1384 	char optionsbuf[sizeof(struct option_mss) +
1385 			sizeof(struct option_windowscale) +
1386 			sizeof(struct option_base) + TCP_OPTIONS_PADDING];
1387 	struct irdma_kmem_info opts;
1388 	int optionssize = 0;
1389 	/* Sending MSS option */
1390 	union all_known_options *options;
1391 
1392 	opts.addr = optionsbuf;
1393 	if (!cm_node)
1394 		return -EINVAL;
1395 
1396 	options = (union all_known_options *)&optionsbuf[optionssize];
1397 	options->mss.optionnum = OPTION_NUM_MSS;
1398 	options->mss.len = sizeof(struct option_mss);
1399 	options->mss.mss = htons(cm_node->tcp_cntxt.mss);
1400 	optionssize += sizeof(struct option_mss);
1401 
1402 	options = (union all_known_options *)&optionsbuf[optionssize];
1403 	options->windowscale.optionnum = OPTION_NUM_WINDOW_SCALE;
1404 	options->windowscale.len = sizeof(struct option_windowscale);
1405 	options->windowscale.shiftcount = cm_node->tcp_cntxt.rcv_wscale;
1406 	optionssize += sizeof(struct option_windowscale);
1407 	options = (union all_known_options *)&optionsbuf[optionssize];
1408 	options->eol = OPTION_NUM_EOL;
1409 	optionssize += 1;
1410 
1411 	if (sendack)
1412 		flags |= SET_ACK;
1413 
1414 	opts.size = optionssize;
1415 
1416 	sqbuf = cm_node->cm_core->form_cm_frame(cm_node, &opts, NULL, NULL,
1417 						flags);
1418 	if (!sqbuf)
1419 		return -ENOMEM;
1420 
1421 	return irdma_schedule_cm_timer(cm_node, sqbuf, IRDMA_TIMER_TYPE_SEND, 1,
1422 				       0);
1423 }
1424 
1425 /**
1426  * irdma_send_ack - Send ACK packet
1427  * @cm_node: connection's node
1428  */
1429 void irdma_send_ack(struct irdma_cm_node *cm_node)
1430 {
1431 	struct irdma_puda_buf *sqbuf;
1432 	struct irdma_sc_vsi *vsi = &cm_node->iwdev->vsi;
1433 
1434 	sqbuf = cm_node->cm_core->form_cm_frame(cm_node, NULL, NULL, NULL,
1435 						SET_ACK);
1436 	if (sqbuf)
1437 		irdma_puda_send_buf(vsi->ilq, sqbuf);
1438 }
1439 
1440 /**
1441  * irdma_send_fin - Send FIN pkt
1442  * @cm_node: connection's node
1443  */
1444 static int irdma_send_fin(struct irdma_cm_node *cm_node)
1445 {
1446 	struct irdma_puda_buf *sqbuf;
1447 
1448 	sqbuf = cm_node->cm_core->form_cm_frame(cm_node, NULL, NULL, NULL,
1449 						SET_ACK | SET_FIN);
1450 	if (!sqbuf)
1451 		return -ENOMEM;
1452 
1453 	return irdma_schedule_cm_timer(cm_node, sqbuf, IRDMA_TIMER_TYPE_SEND, 1,
1454 				       0);
1455 }
1456 
1457 /**
1458  * irdma_find_listener - find a cm node listening on this addr-port pair
1459  * @cm_core: cm's core
1460  * @dst_addr: listener ip addr
1461  * @dst_port: listener tcp port num
1462  * @vlan_id: virtual LAN ID
1463  * @listener_state: state to match with listen node's
1464  */
1465 static struct irdma_cm_listener *
1466 irdma_find_listener(struct irdma_cm_core *cm_core, u32 *dst_addr, u16 dst_port,
1467 		    u16 vlan_id, enum irdma_cm_listener_state listener_state)
1468 {
1469 	struct irdma_cm_listener *listen_node;
1470 	static const u32 ip_zero[4] = { 0, 0, 0, 0 };
1471 	u32 listen_addr[4];
1472 	u16 listen_port;
1473 	unsigned long flags;
1474 
1475 	/* walk list and find cm_node associated with this session ID */
1476 	spin_lock_irqsave(&cm_core->listen_list_lock, flags);
1477 	list_for_each_entry (listen_node, &cm_core->listen_list, list) {
1478 		memcpy(listen_addr, listen_node->loc_addr, sizeof(listen_addr));
1479 		listen_port = listen_node->loc_port;
1480 		if (listen_port != dst_port ||
1481 		    !(listener_state & listen_node->listener_state))
1482 			continue;
1483 		/* compare node pair, return node handle if a match */
1484 		if (!memcmp(listen_addr, ip_zero, sizeof(listen_addr)) ||
1485 		    (!memcmp(listen_addr, dst_addr, sizeof(listen_addr)) &&
1486 		     vlan_id == listen_node->vlan_id)) {
1487 			refcount_inc(&listen_node->refcnt);
1488 			spin_unlock_irqrestore(&cm_core->listen_list_lock,
1489 					       flags);
1490 			trace_irdma_find_listener(listen_node);
1491 			return listen_node;
1492 		}
1493 	}
1494 	spin_unlock_irqrestore(&cm_core->listen_list_lock, flags);
1495 
1496 	return NULL;
1497 }
1498 
1499 /**
1500  * irdma_del_multiple_qhash - Remove qhash and child listens
1501  * @iwdev: iWarp device
1502  * @cm_info: CM info for parent listen node
1503  * @cm_parent_listen_node: The parent listen node
1504  */
1505 static int irdma_del_multiple_qhash(struct irdma_device *iwdev,
1506 				    struct irdma_cm_info *cm_info,
1507 				    struct irdma_cm_listener *cm_parent_listen_node)
1508 {
1509 	struct irdma_cm_listener *child_listen_node;
1510 	struct list_head *pos, *tpos;
1511 	unsigned long flags;
1512 	int ret = -EINVAL;
1513 
1514 	spin_lock_irqsave(&iwdev->cm_core.listen_list_lock, flags);
1515 	list_for_each_safe (pos, tpos,
1516 			    &cm_parent_listen_node->child_listen_list) {
1517 		child_listen_node = list_entry(pos, struct irdma_cm_listener,
1518 					       child_listen_list);
1519 		if (child_listen_node->ipv4)
1520 			ibdev_dbg(&iwdev->ibdev,
1521 				  "CM: removing child listen for IP=%pI4, port=%d, vlan=%d\n",
1522 				  child_listen_node->loc_addr,
1523 				  child_listen_node->loc_port,
1524 				  child_listen_node->vlan_id);
1525 		else
1526 			ibdev_dbg(&iwdev->ibdev,
1527 				  "CM: removing child listen for IP=%pI6, port=%d, vlan=%d\n",
1528 				  child_listen_node->loc_addr,
1529 				  child_listen_node->loc_port,
1530 				  child_listen_node->vlan_id);
1531 		trace_irdma_del_multiple_qhash(child_listen_node);
1532 		list_del(pos);
1533 		memcpy(cm_info->loc_addr, child_listen_node->loc_addr,
1534 		       sizeof(cm_info->loc_addr));
1535 		cm_info->vlan_id = child_listen_node->vlan_id;
1536 		if (child_listen_node->qhash_set) {
1537 			ret = irdma_manage_qhash(iwdev, cm_info,
1538 						 IRDMA_QHASH_TYPE_TCP_SYN,
1539 						 IRDMA_QHASH_MANAGE_TYPE_DELETE,
1540 						 NULL, false);
1541 			child_listen_node->qhash_set = false;
1542 		} else {
1543 			ret = 0;
1544 		}
1545 		ibdev_dbg(&iwdev->ibdev,
1546 			  "CM: Child listen node freed = %p\n",
1547 			  child_listen_node);
1548 		kfree(child_listen_node);
1549 		cm_parent_listen_node->cm_core->stats_listen_nodes_destroyed++;
1550 	}
1551 	spin_unlock_irqrestore(&iwdev->cm_core.listen_list_lock, flags);
1552 
1553 	return ret;
1554 }
1555 
1556 /**
1557  * irdma_netdev_vlan_ipv6 - Gets the netdev and mac
1558  * @addr: local IPv6 address
1559  * @vlan_id: vlan id for the given IPv6 address
1560  * @mac: mac address for the given IPv6 address
1561  *
1562  * Returns the net_device of the IPv6 address and also sets the
1563  * vlan id and mac for that address.
1564  */
1565 struct net_device *irdma_netdev_vlan_ipv6(u32 *addr, u16 *vlan_id, u8 *mac)
1566 {
1567 	struct net_device *ip_dev = NULL;
1568 	struct in6_addr laddr6;
1569 
1570 	if (!IS_ENABLED(CONFIG_IPV6))
1571 		return NULL;
1572 
1573 	irdma_copy_ip_htonl(laddr6.in6_u.u6_addr32, addr);
1574 	if (vlan_id)
1575 		*vlan_id = 0xFFFF;	/* Match rdma_vlan_dev_vlan_id() */
1576 	if (mac)
1577 		eth_zero_addr(mac);
1578 
1579 	rcu_read_lock();
1580 	for_each_netdev_rcu (&init_net, ip_dev) {
1581 		if (ipv6_chk_addr(&init_net, &laddr6, ip_dev, 1)) {
1582 			if (vlan_id)
1583 				*vlan_id = rdma_vlan_dev_vlan_id(ip_dev);
1584 			if (ip_dev->dev_addr && mac)
1585 				ether_addr_copy(mac, ip_dev->dev_addr);
1586 			break;
1587 		}
1588 	}
1589 	rcu_read_unlock();
1590 
1591 	return ip_dev;
1592 }
1593 
1594 /**
1595  * irdma_get_vlan_ipv4 - Returns the vlan_id for IPv4 address
1596  * @addr: local IPv4 address
1597  */
1598 u16 irdma_get_vlan_ipv4(u32 *addr)
1599 {
1600 	struct net_device *netdev;
1601 	u16 vlan_id = 0xFFFF;
1602 
1603 	netdev = ip_dev_find(&init_net, htonl(addr[0]));
1604 	if (netdev) {
1605 		vlan_id = rdma_vlan_dev_vlan_id(netdev);
1606 		dev_put(netdev);
1607 	}
1608 
1609 	return vlan_id;
1610 }
1611 
1612 /**
1613  * irdma_add_mqh_6 - Adds multiple qhashes for IPv6
1614  * @iwdev: iWarp device
1615  * @cm_info: CM info for parent listen node
1616  * @cm_parent_listen_node: The parent listen node
1617  *
1618  * Adds a qhash and a child listen node for every IPv6 address
1619  * on the adapter and adds the associated qhash filter
1620  */
1621 static int irdma_add_mqh_6(struct irdma_device *iwdev,
1622 			   struct irdma_cm_info *cm_info,
1623 			   struct irdma_cm_listener *cm_parent_listen_node)
1624 {
1625 	struct net_device *ip_dev;
1626 	struct inet6_dev *idev;
1627 	struct inet6_ifaddr *ifp, *tmp;
1628 	struct irdma_cm_listener *child_listen_node;
1629 	unsigned long flags;
1630 	int ret = 0;
1631 
1632 	rtnl_lock();
1633 	for_each_netdev(&init_net, ip_dev) {
1634 		if (!(ip_dev->flags & IFF_UP))
1635 			continue;
1636 
1637 		if (((rdma_vlan_dev_vlan_id(ip_dev) >= VLAN_N_VID) ||
1638 		     (rdma_vlan_dev_real_dev(ip_dev) != iwdev->netdev)) &&
1639 		    ip_dev != iwdev->netdev)
1640 			continue;
1641 
1642 		idev = __in6_dev_get(ip_dev);
1643 		if (!idev) {
1644 			ibdev_dbg(&iwdev->ibdev, "CM: idev == NULL\n");
1645 			break;
1646 		}
1647 		list_for_each_entry_safe (ifp, tmp, &idev->addr_list, if_list) {
1648 			ibdev_dbg(&iwdev->ibdev, "CM: IP=%pI6, vlan_id=%d, MAC=%pM\n",
1649 				  &ifp->addr, rdma_vlan_dev_vlan_id(ip_dev),
1650 				  ip_dev->dev_addr);
1651 			child_listen_node = kzalloc(sizeof(*child_listen_node), GFP_KERNEL);
1652 			ibdev_dbg(&iwdev->ibdev, "CM: Allocating child listener %p\n",
1653 				  child_listen_node);
1654 			if (!child_listen_node) {
1655 				ibdev_dbg(&iwdev->ibdev, "CM: listener memory allocation\n");
1656 				ret = -ENOMEM;
1657 				goto exit;
1658 			}
1659 
1660 			cm_info->vlan_id = rdma_vlan_dev_vlan_id(ip_dev);
1661 			cm_parent_listen_node->vlan_id = cm_info->vlan_id;
1662 			memcpy(child_listen_node, cm_parent_listen_node,
1663 			       sizeof(*child_listen_node));
1664 			irdma_copy_ip_ntohl(child_listen_node->loc_addr,
1665 					    ifp->addr.in6_u.u6_addr32);
1666 			memcpy(cm_info->loc_addr, child_listen_node->loc_addr,
1667 			       sizeof(cm_info->loc_addr));
1668 			ret = irdma_manage_qhash(iwdev, cm_info,
1669 						 IRDMA_QHASH_TYPE_TCP_SYN,
1670 						 IRDMA_QHASH_MANAGE_TYPE_ADD,
1671 						 NULL, true);
1672 			if (ret) {
1673 				kfree(child_listen_node);
1674 				continue;
1675 			}
1676 
1677 			trace_irdma_add_mqh_6(iwdev, child_listen_node,
1678 					      ip_dev->dev_addr);
1679 
1680 			child_listen_node->qhash_set = true;
1681 			spin_lock_irqsave(&iwdev->cm_core.listen_list_lock, flags);
1682 			list_add(&child_listen_node->child_listen_list,
1683 				 &cm_parent_listen_node->child_listen_list);
1684 			spin_unlock_irqrestore(&iwdev->cm_core.listen_list_lock, flags);
1685 			cm_parent_listen_node->cm_core->stats_listen_nodes_created++;
1686 		}
1687 	}
1688 exit:
1689 	rtnl_unlock();
1690 
1691 	return ret;
1692 }
1693 
1694 /**
1695  * irdma_add_mqh_4 - Adds multiple qhashes for IPv4
1696  * @iwdev: iWarp device
1697  * @cm_info: CM info for parent listen node
1698  * @cm_parent_listen_node: The parent listen node
1699  *
1700  * Adds a qhash and a child listen node for every IPv4 address
1701  * on the adapter and adds the associated qhash filter
1702  */
1703 static int irdma_add_mqh_4(struct irdma_device *iwdev,
1704 			   struct irdma_cm_info *cm_info,
1705 			   struct irdma_cm_listener *cm_parent_listen_node)
1706 {
1707 	struct net_device *ip_dev;
1708 	struct in_device *idev;
1709 	struct irdma_cm_listener *child_listen_node;
1710 	unsigned long flags;
1711 	const struct in_ifaddr *ifa;
1712 	int ret = 0;
1713 
1714 	rtnl_lock();
1715 	for_each_netdev(&init_net, ip_dev) {
1716 		if (!(ip_dev->flags & IFF_UP))
1717 			continue;
1718 
1719 		if (((rdma_vlan_dev_vlan_id(ip_dev) >= VLAN_N_VID) ||
1720 		     (rdma_vlan_dev_real_dev(ip_dev) != iwdev->netdev)) &&
1721 		    ip_dev != iwdev->netdev)
1722 			continue;
1723 
1724 		idev = in_dev_get(ip_dev);
1725 		in_dev_for_each_ifa_rtnl(ifa, idev) {
1726 			ibdev_dbg(&iwdev->ibdev,
1727 				  "CM: Allocating child CM Listener forIP=%pI4, vlan_id=%d, MAC=%pM\n",
1728 				  &ifa->ifa_address, rdma_vlan_dev_vlan_id(ip_dev),
1729 				  ip_dev->dev_addr);
1730 			child_listen_node = kzalloc(sizeof(*child_listen_node), GFP_KERNEL);
1731 			cm_parent_listen_node->cm_core->stats_listen_nodes_created++;
1732 			ibdev_dbg(&iwdev->ibdev, "CM: Allocating child listener %p\n",
1733 				  child_listen_node);
1734 			if (!child_listen_node) {
1735 				ibdev_dbg(&iwdev->ibdev, "CM: listener memory allocation\n");
1736 				in_dev_put(idev);
1737 				ret = -ENOMEM;
1738 				goto exit;
1739 			}
1740 
1741 			cm_info->vlan_id = rdma_vlan_dev_vlan_id(ip_dev);
1742 			cm_parent_listen_node->vlan_id = cm_info->vlan_id;
1743 			memcpy(child_listen_node, cm_parent_listen_node,
1744 			       sizeof(*child_listen_node));
1745 			child_listen_node->loc_addr[0] =
1746 				ntohl(ifa->ifa_address);
1747 			memcpy(cm_info->loc_addr, child_listen_node->loc_addr,
1748 			       sizeof(cm_info->loc_addr));
1749 			ret = irdma_manage_qhash(iwdev, cm_info,
1750 						 IRDMA_QHASH_TYPE_TCP_SYN,
1751 						 IRDMA_QHASH_MANAGE_TYPE_ADD,
1752 						 NULL, true);
1753 			if (ret) {
1754 				kfree(child_listen_node);
1755 				cm_parent_listen_node->cm_core
1756 					->stats_listen_nodes_created--;
1757 				continue;
1758 			}
1759 
1760 			trace_irdma_add_mqh_4(iwdev, child_listen_node,
1761 					      ip_dev->dev_addr);
1762 
1763 			child_listen_node->qhash_set = true;
1764 			spin_lock_irqsave(&iwdev->cm_core.listen_list_lock,
1765 					  flags);
1766 			list_add(&child_listen_node->child_listen_list,
1767 				 &cm_parent_listen_node->child_listen_list);
1768 			spin_unlock_irqrestore(&iwdev->cm_core.listen_list_lock, flags);
1769 		}
1770 		in_dev_put(idev);
1771 	}
1772 exit:
1773 	rtnl_unlock();
1774 
1775 	return ret;
1776 }
1777 
1778 /**
1779  * irdma_add_mqh - Adds multiple qhashes
1780  * @iwdev: iWarp device
1781  * @cm_info: CM info for parent listen node
1782  * @cm_listen_node: The parent listen node
1783  */
1784 static int irdma_add_mqh(struct irdma_device *iwdev,
1785 			 struct irdma_cm_info *cm_info,
1786 			 struct irdma_cm_listener *cm_listen_node)
1787 {
1788 	if (cm_info->ipv4)
1789 		return irdma_add_mqh_4(iwdev, cm_info, cm_listen_node);
1790 	else
1791 		return irdma_add_mqh_6(iwdev, cm_info, cm_listen_node);
1792 }
1793 
1794 /**
1795  * irdma_reset_list_prep - add connection nodes slated for reset to list
1796  * @cm_core: cm's core
1797  * @listener: pointer to listener node
1798  * @reset_list: a list to which cm_node will be selected
1799  */
1800 static void irdma_reset_list_prep(struct irdma_cm_core *cm_core,
1801 				  struct irdma_cm_listener *listener,
1802 				  struct list_head *reset_list)
1803 {
1804 	struct irdma_cm_node *cm_node;
1805 	int bkt;
1806 
1807 	hash_for_each_rcu(cm_core->cm_hash_tbl, bkt, cm_node, list) {
1808 		if (cm_node->listener == listener &&
1809 		    !cm_node->accelerated &&
1810 		    refcount_inc_not_zero(&cm_node->refcnt))
1811 			list_add(&cm_node->reset_entry, reset_list);
1812 	}
1813 }
1814 
1815 /**
1816  * irdma_dec_refcnt_listen - delete listener and associated cm nodes
1817  * @cm_core: cm's core
1818  * @listener: pointer to listener node
1819  * @free_hanging_nodes: to free associated cm_nodes
1820  * @apbvt_del: flag to delete the apbvt
1821  */
1822 static int irdma_dec_refcnt_listen(struct irdma_cm_core *cm_core,
1823 				   struct irdma_cm_listener *listener,
1824 				   int free_hanging_nodes, bool apbvt_del)
1825 {
1826 	int err;
1827 	struct list_head *list_pos;
1828 	struct list_head *list_temp;
1829 	struct irdma_cm_node *cm_node;
1830 	struct list_head reset_list;
1831 	struct irdma_cm_info nfo;
1832 	enum irdma_cm_node_state old_state;
1833 	unsigned long flags;
1834 
1835 	trace_irdma_dec_refcnt_listen(listener, __builtin_return_address(0));
1836 	/* free non-accelerated child nodes for this listener */
1837 	INIT_LIST_HEAD(&reset_list);
1838 	if (free_hanging_nodes) {
1839 		rcu_read_lock();
1840 		irdma_reset_list_prep(cm_core, listener, &reset_list);
1841 		rcu_read_unlock();
1842 	}
1843 
1844 	list_for_each_safe (list_pos, list_temp, &reset_list) {
1845 		cm_node = container_of(list_pos, struct irdma_cm_node,
1846 				       reset_entry);
1847 		if (cm_node->state >= IRDMA_CM_STATE_FIN_WAIT1) {
1848 			irdma_rem_ref_cm_node(cm_node);
1849 			continue;
1850 		}
1851 
1852 		irdma_cleanup_retrans_entry(cm_node);
1853 		err = irdma_send_reset(cm_node);
1854 		if (err) {
1855 			cm_node->state = IRDMA_CM_STATE_CLOSED;
1856 			ibdev_dbg(&cm_node->iwdev->ibdev,
1857 				  "CM: send reset failed\n");
1858 		} else {
1859 			old_state = cm_node->state;
1860 			cm_node->state = IRDMA_CM_STATE_LISTENER_DESTROYED;
1861 			if (old_state != IRDMA_CM_STATE_MPAREQ_RCVD)
1862 				irdma_rem_ref_cm_node(cm_node);
1863 		}
1864 	}
1865 
1866 	if (refcount_dec_and_test(&listener->refcnt)) {
1867 		spin_lock_irqsave(&cm_core->listen_list_lock, flags);
1868 		list_del(&listener->list);
1869 		spin_unlock_irqrestore(&cm_core->listen_list_lock, flags);
1870 
1871 		if (apbvt_del)
1872 			irdma_del_apbvt(listener->iwdev,
1873 					listener->apbvt_entry);
1874 		memcpy(nfo.loc_addr, listener->loc_addr, sizeof(nfo.loc_addr));
1875 		nfo.loc_port = listener->loc_port;
1876 		nfo.ipv4 = listener->ipv4;
1877 		nfo.vlan_id = listener->vlan_id;
1878 		nfo.user_pri = listener->user_pri;
1879 		nfo.qh_qpid = listener->iwdev->vsi.ilq->qp_id;
1880 
1881 		if (!list_empty(&listener->child_listen_list)) {
1882 			irdma_del_multiple_qhash(listener->iwdev, &nfo,
1883 						 listener);
1884 		} else {
1885 			if (listener->qhash_set)
1886 				irdma_manage_qhash(listener->iwdev,
1887 						   &nfo,
1888 						   IRDMA_QHASH_TYPE_TCP_SYN,
1889 						   IRDMA_QHASH_MANAGE_TYPE_DELETE,
1890 						   NULL, false);
1891 		}
1892 
1893 		cm_core->stats_listen_destroyed++;
1894 		cm_core->stats_listen_nodes_destroyed++;
1895 		ibdev_dbg(&listener->iwdev->ibdev,
1896 			  "CM: loc_port=0x%04x loc_addr=%pI4 cm_listen_node=%p cm_id=%p qhash_set=%d vlan_id=%d apbvt_del=%d\n",
1897 			  listener->loc_port, listener->loc_addr, listener,
1898 			  listener->cm_id, listener->qhash_set,
1899 			  listener->vlan_id, apbvt_del);
1900 		kfree(listener);
1901 		listener = NULL;
1902 		return 0;
1903 	}
1904 
1905 	return -EINVAL;
1906 }
1907 
1908 /**
1909  * irdma_cm_del_listen - delete a listener
1910  * @cm_core: cm's core
1911  * @listener: passive connection's listener
1912  * @apbvt_del: flag to delete apbvt
1913  */
1914 static int irdma_cm_del_listen(struct irdma_cm_core *cm_core,
1915 			       struct irdma_cm_listener *listener,
1916 			       bool apbvt_del)
1917 {
1918 	listener->listener_state = IRDMA_CM_LISTENER_PASSIVE_STATE;
1919 	listener->cm_id = NULL;
1920 
1921 	return irdma_dec_refcnt_listen(cm_core, listener, 1, apbvt_del);
1922 }
1923 
1924 /**
1925  * irdma_addr_resolve_neigh - resolve neighbor address
1926  * @iwdev: iwarp device structure
1927  * @src_ip: local ip address
1928  * @dst_ip: remote ip address
1929  * @arpindex: if there is an arp entry
1930  */
1931 static int irdma_addr_resolve_neigh(struct irdma_device *iwdev, u32 src_ip,
1932 				    u32 dst_ip, int arpindex)
1933 {
1934 	struct rtable *rt;
1935 	struct neighbour *neigh;
1936 	int rc = arpindex;
1937 	__be32 dst_ipaddr = htonl(dst_ip);
1938 	__be32 src_ipaddr = htonl(src_ip);
1939 
1940 	rt = ip_route_output(&init_net, dst_ipaddr, src_ipaddr, 0, 0);
1941 	if (IS_ERR(rt)) {
1942 		ibdev_dbg(&iwdev->ibdev, "CM: ip_route_output fail\n");
1943 		return -EINVAL;
1944 	}
1945 
1946 	neigh = dst_neigh_lookup(&rt->dst, &dst_ipaddr);
1947 	if (!neigh)
1948 		goto exit;
1949 
1950 	if (neigh->nud_state & NUD_VALID)
1951 		rc = irdma_add_arp(iwdev->rf, &dst_ip, true, neigh->ha);
1952 	else
1953 		neigh_event_send(neigh, NULL);
1954 	if (neigh)
1955 		neigh_release(neigh);
1956 exit:
1957 	ip_rt_put(rt);
1958 
1959 	return rc;
1960 }
1961 
1962 /**
1963  * irdma_get_dst_ipv6 - get destination cache entry via ipv6 lookup
1964  * @src_addr: local ipv6 sock address
1965  * @dst_addr: destination ipv6 sock address
1966  */
1967 static struct dst_entry *irdma_get_dst_ipv6(struct sockaddr_in6 *src_addr,
1968 					    struct sockaddr_in6 *dst_addr)
1969 {
1970 	struct dst_entry *dst = NULL;
1971 
1972 	if ((IS_ENABLED(CONFIG_IPV6))) {
1973 		struct flowi6 fl6 = {};
1974 
1975 		fl6.daddr = dst_addr->sin6_addr;
1976 		fl6.saddr = src_addr->sin6_addr;
1977 		if (ipv6_addr_type(&fl6.daddr) & IPV6_ADDR_LINKLOCAL)
1978 			fl6.flowi6_oif = dst_addr->sin6_scope_id;
1979 
1980 		dst = ip6_route_output(&init_net, NULL, &fl6);
1981 	}
1982 
1983 	return dst;
1984 }
1985 
1986 /**
1987  * irdma_addr_resolve_neigh_ipv6 - resolve neighbor ipv6 address
1988  * @iwdev: iwarp device structure
1989  * @src: local ip address
1990  * @dest: remote ip address
1991  * @arpindex: if there is an arp entry
1992  */
1993 static int irdma_addr_resolve_neigh_ipv6(struct irdma_device *iwdev, u32 *src,
1994 					 u32 *dest, int arpindex)
1995 {
1996 	struct neighbour *neigh;
1997 	int rc = arpindex;
1998 	struct dst_entry *dst;
1999 	struct sockaddr_in6 dst_addr = {};
2000 	struct sockaddr_in6 src_addr = {};
2001 
2002 	dst_addr.sin6_family = AF_INET6;
2003 	irdma_copy_ip_htonl(dst_addr.sin6_addr.in6_u.u6_addr32, dest);
2004 	src_addr.sin6_family = AF_INET6;
2005 	irdma_copy_ip_htonl(src_addr.sin6_addr.in6_u.u6_addr32, src);
2006 	dst = irdma_get_dst_ipv6(&src_addr, &dst_addr);
2007 	if (!dst || dst->error) {
2008 		if (dst) {
2009 			dst_release(dst);
2010 			ibdev_dbg(&iwdev->ibdev,
2011 				  "CM: ip6_route_output returned dst->error = %d\n",
2012 				  dst->error);
2013 		}
2014 		return -EINVAL;
2015 	}
2016 
2017 	neigh = dst_neigh_lookup(dst, dst_addr.sin6_addr.in6_u.u6_addr32);
2018 	if (!neigh)
2019 		goto exit;
2020 
2021 	ibdev_dbg(&iwdev->ibdev, "CM: dst_neigh_lookup MAC=%pM\n",
2022 		  neigh->ha);
2023 
2024 	trace_irdma_addr_resolve(iwdev, neigh->ha);
2025 
2026 	if (neigh->nud_state & NUD_VALID)
2027 		rc = irdma_add_arp(iwdev->rf, dest, false, neigh->ha);
2028 	else
2029 		neigh_event_send(neigh, NULL);
2030 	if (neigh)
2031 		neigh_release(neigh);
2032 exit:
2033 	dst_release(dst);
2034 
2035 	return rc;
2036 }
2037 
2038 /**
2039  * irdma_find_node - find a cm node that matches the reference cm node
2040  * @cm_core: cm's core
2041  * @rem_port: remote tcp port num
2042  * @rem_addr: remote ip addr
2043  * @loc_port: local tcp port num
2044  * @loc_addr: local ip addr
2045  * @vlan_id: local VLAN ID
2046  */
2047 struct irdma_cm_node *irdma_find_node(struct irdma_cm_core *cm_core,
2048 				      u16 rem_port, u32 *rem_addr, u16 loc_port,
2049 				      u32 *loc_addr, u16 vlan_id)
2050 {
2051 	struct irdma_cm_node *cm_node;
2052 	u32 key = (rem_port << 16) | loc_port;
2053 
2054 	rcu_read_lock();
2055 	hash_for_each_possible_rcu(cm_core->cm_hash_tbl, cm_node, list, key) {
2056 		if (cm_node->vlan_id == vlan_id &&
2057 		    cm_node->loc_port == loc_port && cm_node->rem_port == rem_port &&
2058 		    !memcmp(cm_node->loc_addr, loc_addr, sizeof(cm_node->loc_addr)) &&
2059 		    !memcmp(cm_node->rem_addr, rem_addr, sizeof(cm_node->rem_addr))) {
2060 			if (!refcount_inc_not_zero(&cm_node->refcnt))
2061 				goto exit;
2062 			rcu_read_unlock();
2063 			trace_irdma_find_node(cm_node, 0, NULL);
2064 			return cm_node;
2065 		}
2066 	}
2067 
2068 exit:
2069 	rcu_read_unlock();
2070 
2071 	/* no owner node */
2072 	return NULL;
2073 }
2074 
2075 /**
2076  * irdma_add_hte_node - add a cm node to the hash table
2077  * @cm_core: cm's core
2078  * @cm_node: connection's node
2079  */
2080 static void irdma_add_hte_node(struct irdma_cm_core *cm_core,
2081 			       struct irdma_cm_node *cm_node)
2082 {
2083 	unsigned long flags;
2084 	u32 key = (cm_node->rem_port << 16) | cm_node->loc_port;
2085 
2086 	spin_lock_irqsave(&cm_core->ht_lock, flags);
2087 	hash_add_rcu(cm_core->cm_hash_tbl, &cm_node->list, key);
2088 	spin_unlock_irqrestore(&cm_core->ht_lock, flags);
2089 }
2090 
2091 /**
2092  * irdma_ipv4_is_lpb - check if loopback
2093  * @loc_addr: local addr to compare
2094  * @rem_addr: remote address
2095  */
2096 bool irdma_ipv4_is_lpb(u32 loc_addr, u32 rem_addr)
2097 {
2098 	return ipv4_is_loopback(htonl(rem_addr)) || (loc_addr == rem_addr);
2099 }
2100 
2101 /**
2102  * irdma_ipv6_is_lpb - check if loopback
2103  * @loc_addr: local addr to compare
2104  * @rem_addr: remote address
2105  */
2106 bool irdma_ipv6_is_lpb(u32 *loc_addr, u32 *rem_addr)
2107 {
2108 	struct in6_addr raddr6;
2109 
2110 	irdma_copy_ip_htonl(raddr6.in6_u.u6_addr32, rem_addr);
2111 
2112 	return !memcmp(loc_addr, rem_addr, 16) || ipv6_addr_loopback(&raddr6);
2113 }
2114 
2115 /**
2116  * irdma_cm_create_ah - create a cm address handle
2117  * @cm_node: The connection manager node to create AH for
2118  * @wait: Provides option to wait for ah creation or not
2119  */
2120 static int irdma_cm_create_ah(struct irdma_cm_node *cm_node, bool wait)
2121 {
2122 	struct irdma_ah_info ah_info = {};
2123 	struct irdma_device *iwdev = cm_node->iwdev;
2124 
2125 	ether_addr_copy(ah_info.mac_addr, iwdev->netdev->dev_addr);
2126 
2127 	ah_info.hop_ttl = 0x40;
2128 	ah_info.tc_tos = cm_node->tos;
2129 	ah_info.vsi = &iwdev->vsi;
2130 
2131 	if (cm_node->ipv4) {
2132 		ah_info.ipv4_valid = true;
2133 		ah_info.dest_ip_addr[0] = cm_node->rem_addr[0];
2134 		ah_info.src_ip_addr[0] = cm_node->loc_addr[0];
2135 		ah_info.do_lpbk = irdma_ipv4_is_lpb(ah_info.src_ip_addr[0],
2136 						    ah_info.dest_ip_addr[0]);
2137 	} else {
2138 		memcpy(ah_info.dest_ip_addr, cm_node->rem_addr,
2139 		       sizeof(ah_info.dest_ip_addr));
2140 		memcpy(ah_info.src_ip_addr, cm_node->loc_addr,
2141 		       sizeof(ah_info.src_ip_addr));
2142 		ah_info.do_lpbk = irdma_ipv6_is_lpb(ah_info.src_ip_addr,
2143 						    ah_info.dest_ip_addr);
2144 	}
2145 
2146 	ah_info.vlan_tag = cm_node->vlan_id;
2147 	if (cm_node->vlan_id < VLAN_N_VID) {
2148 		ah_info.insert_vlan_tag = 1;
2149 		ah_info.vlan_tag |= cm_node->user_pri << VLAN_PRIO_SHIFT;
2150 	}
2151 
2152 	ah_info.dst_arpindex =
2153 		irdma_arp_table(iwdev->rf, ah_info.dest_ip_addr,
2154 				ah_info.ipv4_valid, NULL, IRDMA_ARP_RESOLVE);
2155 
2156 	if (irdma_puda_create_ah(&iwdev->rf->sc_dev, &ah_info, wait,
2157 				 IRDMA_PUDA_RSRC_TYPE_ILQ, cm_node,
2158 				 &cm_node->ah))
2159 		return -ENOMEM;
2160 
2161 	trace_irdma_create_ah(cm_node);
2162 	return 0;
2163 }
2164 
2165 /**
2166  * irdma_cm_free_ah - free a cm address handle
2167  * @cm_node: The connection manager node to create AH for
2168  */
2169 static void irdma_cm_free_ah(struct irdma_cm_node *cm_node)
2170 {
2171 	struct irdma_device *iwdev = cm_node->iwdev;
2172 
2173 	trace_irdma_cm_free_ah(cm_node);
2174 	irdma_puda_free_ah(&iwdev->rf->sc_dev, cm_node->ah);
2175 	cm_node->ah = NULL;
2176 }
2177 
2178 /**
2179  * irdma_make_cm_node - create a new instance of a cm node
2180  * @cm_core: cm's core
2181  * @iwdev: iwarp device structure
2182  * @cm_info: quad info for connection
2183  * @listener: passive connection's listener
2184  */
2185 static struct irdma_cm_node *
2186 irdma_make_cm_node(struct irdma_cm_core *cm_core, struct irdma_device *iwdev,
2187 		   struct irdma_cm_info *cm_info,
2188 		   struct irdma_cm_listener *listener)
2189 {
2190 	struct irdma_cm_node *cm_node;
2191 	int oldarpindex;
2192 	int arpindex;
2193 	struct net_device *netdev = iwdev->netdev;
2194 
2195 	/* create an hte and cm_node for this instance */
2196 	cm_node = kzalloc(sizeof(*cm_node), GFP_ATOMIC);
2197 	if (!cm_node)
2198 		return NULL;
2199 
2200 	/* set our node specific transport info */
2201 	cm_node->ipv4 = cm_info->ipv4;
2202 	cm_node->vlan_id = cm_info->vlan_id;
2203 	if (cm_node->vlan_id >= VLAN_N_VID && iwdev->dcb_vlan_mode)
2204 		cm_node->vlan_id = 0;
2205 	cm_node->tos = cm_info->tos;
2206 	cm_node->user_pri = cm_info->user_pri;
2207 	if (listener) {
2208 		if (listener->tos != cm_info->tos)
2209 			ibdev_warn(&iwdev->ibdev,
2210 				   "application TOS[%d] and remote client TOS[%d] mismatch\n",
2211 				   listener->tos, cm_info->tos);
2212 		if (iwdev->vsi.dscp_mode) {
2213 			cm_node->user_pri = listener->user_pri;
2214 		} else {
2215 			cm_node->tos = max(listener->tos, cm_info->tos);
2216 			cm_node->user_pri = rt_tos2priority(cm_node->tos);
2217 		}
2218 		ibdev_dbg(&iwdev->ibdev,
2219 			  "DCB: listener: TOS:[%d] UP:[%d]\n", cm_node->tos,
2220 			  cm_node->user_pri);
2221 		trace_irdma_listener_tos(iwdev, cm_node->tos,
2222 					 cm_node->user_pri);
2223 	}
2224 	memcpy(cm_node->loc_addr, cm_info->loc_addr, sizeof(cm_node->loc_addr));
2225 	memcpy(cm_node->rem_addr, cm_info->rem_addr, sizeof(cm_node->rem_addr));
2226 	cm_node->loc_port = cm_info->loc_port;
2227 	cm_node->rem_port = cm_info->rem_port;
2228 
2229 	cm_node->mpa_frame_rev = IRDMA_CM_DEFAULT_MPA_VER;
2230 	cm_node->send_rdma0_op = SEND_RDMA_READ_ZERO;
2231 	cm_node->iwdev = iwdev;
2232 	cm_node->dev = &iwdev->rf->sc_dev;
2233 
2234 	cm_node->ird_size = cm_node->dev->hw_attrs.max_hw_ird;
2235 	cm_node->ord_size = cm_node->dev->hw_attrs.max_hw_ord;
2236 
2237 	cm_node->listener = listener;
2238 	cm_node->cm_id = cm_info->cm_id;
2239 	ether_addr_copy(cm_node->loc_mac, netdev->dev_addr);
2240 	spin_lock_init(&cm_node->retrans_list_lock);
2241 	cm_node->ack_rcvd = false;
2242 
2243 	init_completion(&cm_node->establish_comp);
2244 	refcount_set(&cm_node->refcnt, 1);
2245 	/* associate our parent CM core */
2246 	cm_node->cm_core = cm_core;
2247 	cm_node->tcp_cntxt.loc_id = IRDMA_CM_DEFAULT_LOCAL_ID;
2248 	cm_node->tcp_cntxt.rcv_wscale = iwdev->rcv_wscale;
2249 	cm_node->tcp_cntxt.rcv_wnd = iwdev->rcv_wnd >> cm_node->tcp_cntxt.rcv_wscale;
2250 	if (cm_node->ipv4) {
2251 		cm_node->tcp_cntxt.loc_seq_num = secure_tcp_seq(htonl(cm_node->loc_addr[0]),
2252 								htonl(cm_node->rem_addr[0]),
2253 								htons(cm_node->loc_port),
2254 								htons(cm_node->rem_port));
2255 		cm_node->tcp_cntxt.mss = iwdev->vsi.mtu - IRDMA_MTU_TO_MSS_IPV4;
2256 	} else if (IS_ENABLED(CONFIG_IPV6)) {
2257 		__be32 loc[4] = {
2258 			htonl(cm_node->loc_addr[0]), htonl(cm_node->loc_addr[1]),
2259 			htonl(cm_node->loc_addr[2]), htonl(cm_node->loc_addr[3])
2260 		};
2261 		__be32 rem[4] = {
2262 			htonl(cm_node->rem_addr[0]), htonl(cm_node->rem_addr[1]),
2263 			htonl(cm_node->rem_addr[2]), htonl(cm_node->rem_addr[3])
2264 		};
2265 		cm_node->tcp_cntxt.loc_seq_num = secure_tcpv6_seq(loc, rem,
2266 								  htons(cm_node->loc_port),
2267 								  htons(cm_node->rem_port));
2268 		cm_node->tcp_cntxt.mss = iwdev->vsi.mtu - IRDMA_MTU_TO_MSS_IPV6;
2269 	}
2270 
2271 	if ((cm_node->ipv4 &&
2272 	     irdma_ipv4_is_lpb(cm_node->loc_addr[0], cm_node->rem_addr[0])) ||
2273 	    (!cm_node->ipv4 &&
2274 	     irdma_ipv6_is_lpb(cm_node->loc_addr, cm_node->rem_addr))) {
2275 		cm_node->do_lpb = true;
2276 		arpindex = irdma_arp_table(iwdev->rf, cm_node->rem_addr,
2277 					   cm_node->ipv4, NULL,
2278 					   IRDMA_ARP_RESOLVE);
2279 	} else {
2280 		oldarpindex = irdma_arp_table(iwdev->rf, cm_node->rem_addr,
2281 					      cm_node->ipv4, NULL,
2282 					      IRDMA_ARP_RESOLVE);
2283 		if (cm_node->ipv4)
2284 			arpindex = irdma_addr_resolve_neigh(iwdev,
2285 							    cm_info->loc_addr[0],
2286 							    cm_info->rem_addr[0],
2287 							    oldarpindex);
2288 		else if (IS_ENABLED(CONFIG_IPV6))
2289 			arpindex = irdma_addr_resolve_neigh_ipv6(iwdev,
2290 								 cm_info->loc_addr,
2291 								 cm_info->rem_addr,
2292 								 oldarpindex);
2293 		else
2294 			arpindex = -EINVAL;
2295 	}
2296 
2297 	if (arpindex < 0)
2298 		goto err;
2299 
2300 	ether_addr_copy(cm_node->rem_mac,
2301 			iwdev->rf->arp_table[arpindex].mac_addr);
2302 	irdma_add_hte_node(cm_core, cm_node);
2303 	cm_core->stats_nodes_created++;
2304 	return cm_node;
2305 
2306 err:
2307 	kfree(cm_node);
2308 
2309 	return NULL;
2310 }
2311 
2312 static void irdma_destroy_connection(struct irdma_cm_node *cm_node)
2313 {
2314 	struct irdma_cm_core *cm_core = cm_node->cm_core;
2315 	struct irdma_qp *iwqp;
2316 	struct irdma_cm_info nfo;
2317 
2318 	/* if the node is destroyed before connection was accelerated */
2319 	if (!cm_node->accelerated && cm_node->accept_pend) {
2320 		ibdev_dbg(&cm_node->iwdev->ibdev,
2321 			  "CM: node destroyed before established\n");
2322 		atomic_dec(&cm_node->listener->pend_accepts_cnt);
2323 	}
2324 	if (cm_node->close_entry)
2325 		irdma_handle_close_entry(cm_node, 0);
2326 	if (cm_node->listener) {
2327 		irdma_dec_refcnt_listen(cm_core, cm_node->listener, 0, true);
2328 	} else {
2329 		if (cm_node->apbvt_set) {
2330 			irdma_del_apbvt(cm_node->iwdev, cm_node->apbvt_entry);
2331 			cm_node->apbvt_set = 0;
2332 		}
2333 		irdma_get_addr_info(cm_node, &nfo);
2334 		if (cm_node->qhash_set) {
2335 			nfo.qh_qpid = cm_node->iwdev->vsi.ilq->qp_id;
2336 			irdma_manage_qhash(cm_node->iwdev, &nfo,
2337 					   IRDMA_QHASH_TYPE_TCP_ESTABLISHED,
2338 					   IRDMA_QHASH_MANAGE_TYPE_DELETE, NULL,
2339 					   false);
2340 			cm_node->qhash_set = 0;
2341 		}
2342 	}
2343 
2344 	iwqp = cm_node->iwqp;
2345 	if (iwqp) {
2346 		cm_node->cm_id->rem_ref(cm_node->cm_id);
2347 		cm_node->cm_id = NULL;
2348 		iwqp->cm_id = NULL;
2349 		irdma_qp_rem_ref(&iwqp->ibqp);
2350 		cm_node->iwqp = NULL;
2351 	} else if (cm_node->qhash_set) {
2352 		irdma_get_addr_info(cm_node, &nfo);
2353 		nfo.qh_qpid = cm_node->iwdev->vsi.ilq->qp_id;
2354 		irdma_manage_qhash(cm_node->iwdev, &nfo,
2355 				   IRDMA_QHASH_TYPE_TCP_ESTABLISHED,
2356 				   IRDMA_QHASH_MANAGE_TYPE_DELETE, NULL, false);
2357 		cm_node->qhash_set = 0;
2358 	}
2359 
2360 	cm_core->cm_free_ah(cm_node);
2361 }
2362 
2363 /**
2364  * irdma_rem_ref_cm_node - destroy an instance of a cm node
2365  * @cm_node: connection's node
2366  */
2367 void irdma_rem_ref_cm_node(struct irdma_cm_node *cm_node)
2368 {
2369 	struct irdma_cm_core *cm_core = cm_node->cm_core;
2370 	unsigned long flags;
2371 
2372 	trace_irdma_rem_ref_cm_node(cm_node, 0, __builtin_return_address(0));
2373 	spin_lock_irqsave(&cm_core->ht_lock, flags);
2374 
2375 	if (!refcount_dec_and_test(&cm_node->refcnt)) {
2376 		spin_unlock_irqrestore(&cm_core->ht_lock, flags);
2377 		return;
2378 	}
2379 	if (cm_node->iwqp) {
2380 		cm_node->iwqp->cm_node = NULL;
2381 		cm_node->iwqp->cm_id = NULL;
2382 	}
2383 	hash_del_rcu(&cm_node->list);
2384 	cm_node->cm_core->stats_nodes_destroyed++;
2385 
2386 	spin_unlock_irqrestore(&cm_core->ht_lock, flags);
2387 
2388 	irdma_destroy_connection(cm_node);
2389 
2390 	kfree_rcu(cm_node, rcu_head);
2391 }
2392 
2393 /**
2394  * irdma_handle_fin_pkt - FIN packet received
2395  * @cm_node: connection's node
2396  */
2397 static void irdma_handle_fin_pkt(struct irdma_cm_node *cm_node)
2398 {
2399 	switch (cm_node->state) {
2400 	case IRDMA_CM_STATE_SYN_RCVD:
2401 	case IRDMA_CM_STATE_SYN_SENT:
2402 	case IRDMA_CM_STATE_ESTABLISHED:
2403 	case IRDMA_CM_STATE_MPAREJ_RCVD:
2404 		cm_node->tcp_cntxt.rcv_nxt++;
2405 		irdma_cleanup_retrans_entry(cm_node);
2406 		cm_node->state = IRDMA_CM_STATE_LAST_ACK;
2407 		irdma_send_fin(cm_node);
2408 		break;
2409 	case IRDMA_CM_STATE_MPAREQ_SENT:
2410 		irdma_create_event(cm_node, IRDMA_CM_EVENT_ABORTED);
2411 		cm_node->tcp_cntxt.rcv_nxt++;
2412 		irdma_cleanup_retrans_entry(cm_node);
2413 		cm_node->state = IRDMA_CM_STATE_CLOSED;
2414 		refcount_inc(&cm_node->refcnt);
2415 		irdma_send_reset(cm_node);
2416 		break;
2417 	case IRDMA_CM_STATE_FIN_WAIT1:
2418 		cm_node->tcp_cntxt.rcv_nxt++;
2419 		irdma_cleanup_retrans_entry(cm_node);
2420 		cm_node->state = IRDMA_CM_STATE_CLOSING;
2421 		irdma_send_ack(cm_node);
2422 		/*
2423 		 * Wait for ACK as this is simultaneous close.
2424 		 * After we receive ACK, do not send anything.
2425 		 * Just rm the node.
2426 		 */
2427 		break;
2428 	case IRDMA_CM_STATE_FIN_WAIT2:
2429 		cm_node->tcp_cntxt.rcv_nxt++;
2430 		irdma_cleanup_retrans_entry(cm_node);
2431 		cm_node->state = IRDMA_CM_STATE_TIME_WAIT;
2432 		irdma_send_ack(cm_node);
2433 		irdma_schedule_cm_timer(cm_node, NULL, IRDMA_TIMER_TYPE_CLOSE,
2434 					1, 0);
2435 		break;
2436 	case IRDMA_CM_STATE_TIME_WAIT:
2437 		cm_node->tcp_cntxt.rcv_nxt++;
2438 		irdma_cleanup_retrans_entry(cm_node);
2439 		cm_node->state = IRDMA_CM_STATE_CLOSED;
2440 		irdma_rem_ref_cm_node(cm_node);
2441 		break;
2442 	case IRDMA_CM_STATE_OFFLOADED:
2443 	default:
2444 		ibdev_dbg(&cm_node->iwdev->ibdev,
2445 			  "CM: bad state node state = %d\n", cm_node->state);
2446 		break;
2447 	}
2448 }
2449 
2450 /**
2451  * irdma_handle_rst_pkt - process received RST packet
2452  * @cm_node: connection's node
2453  * @rbuf: receive buffer
2454  */
2455 static void irdma_handle_rst_pkt(struct irdma_cm_node *cm_node,
2456 				 struct irdma_puda_buf *rbuf)
2457 {
2458 	ibdev_dbg(&cm_node->iwdev->ibdev,
2459 		  "CM: caller: %pS cm_node=%p state=%d rem_port=0x%04x loc_port=0x%04x rem_addr=%pI4 loc_addr=%pI4\n",
2460 		  __builtin_return_address(0), cm_node, cm_node->state,
2461 		  cm_node->rem_port, cm_node->loc_port, cm_node->rem_addr,
2462 		  cm_node->loc_addr);
2463 
2464 	irdma_cleanup_retrans_entry(cm_node);
2465 	switch (cm_node->state) {
2466 	case IRDMA_CM_STATE_SYN_SENT:
2467 	case IRDMA_CM_STATE_MPAREQ_SENT:
2468 		switch (cm_node->mpa_frame_rev) {
2469 		case IETF_MPA_V2:
2470 			/* Drop down to MPA_V1*/
2471 			cm_node->mpa_frame_rev = IETF_MPA_V1;
2472 			/* send a syn and goto syn sent state */
2473 			cm_node->state = IRDMA_CM_STATE_SYN_SENT;
2474 			if (irdma_send_syn(cm_node, 0))
2475 				irdma_active_open_err(cm_node, false);
2476 			break;
2477 		case IETF_MPA_V1:
2478 		default:
2479 			irdma_active_open_err(cm_node, false);
2480 			break;
2481 		}
2482 		break;
2483 	case IRDMA_CM_STATE_MPAREQ_RCVD:
2484 		atomic_inc(&cm_node->passive_state);
2485 		break;
2486 	case IRDMA_CM_STATE_ESTABLISHED:
2487 	case IRDMA_CM_STATE_SYN_RCVD:
2488 	case IRDMA_CM_STATE_LISTENING:
2489 		irdma_passive_open_err(cm_node, false);
2490 		break;
2491 	case IRDMA_CM_STATE_OFFLOADED:
2492 		irdma_active_open_err(cm_node, false);
2493 		break;
2494 	case IRDMA_CM_STATE_CLOSED:
2495 		break;
2496 	case IRDMA_CM_STATE_FIN_WAIT2:
2497 	case IRDMA_CM_STATE_FIN_WAIT1:
2498 	case IRDMA_CM_STATE_LAST_ACK:
2499 	case IRDMA_CM_STATE_TIME_WAIT:
2500 		cm_node->state = IRDMA_CM_STATE_CLOSED;
2501 		irdma_rem_ref_cm_node(cm_node);
2502 		break;
2503 	default:
2504 		break;
2505 	}
2506 }
2507 
2508 /**
2509  * irdma_handle_rcv_mpa - Process a recv'd mpa buffer
2510  * @cm_node: connection's node
2511  * @rbuf: receive buffer
2512  */
2513 static void irdma_handle_rcv_mpa(struct irdma_cm_node *cm_node,
2514 				 struct irdma_puda_buf *rbuf)
2515 {
2516 	int err;
2517 	int datasize = rbuf->datalen;
2518 	u8 *dataloc = rbuf->data;
2519 
2520 	enum irdma_cm_event_type type = IRDMA_CM_EVENT_UNKNOWN;
2521 	u32 res_type;
2522 
2523 	err = irdma_parse_mpa(cm_node, dataloc, &res_type, datasize);
2524 	if (err) {
2525 		if (cm_node->state == IRDMA_CM_STATE_MPAREQ_SENT)
2526 			irdma_active_open_err(cm_node, true);
2527 		else
2528 			irdma_passive_open_err(cm_node, true);
2529 		return;
2530 	}
2531 
2532 	switch (cm_node->state) {
2533 	case IRDMA_CM_STATE_ESTABLISHED:
2534 		if (res_type == IRDMA_MPA_REQUEST_REJECT)
2535 			ibdev_dbg(&cm_node->iwdev->ibdev,
2536 				  "CM: state for reject\n");
2537 		cm_node->state = IRDMA_CM_STATE_MPAREQ_RCVD;
2538 		type = IRDMA_CM_EVENT_MPA_REQ;
2539 		irdma_send_ack(cm_node); /* ACK received MPA request */
2540 		atomic_set(&cm_node->passive_state,
2541 			   IRDMA_PASSIVE_STATE_INDICATED);
2542 		break;
2543 	case IRDMA_CM_STATE_MPAREQ_SENT:
2544 		irdma_cleanup_retrans_entry(cm_node);
2545 		if (res_type == IRDMA_MPA_REQUEST_REJECT) {
2546 			type = IRDMA_CM_EVENT_MPA_REJECT;
2547 			cm_node->state = IRDMA_CM_STATE_MPAREJ_RCVD;
2548 		} else {
2549 			type = IRDMA_CM_EVENT_CONNECTED;
2550 			cm_node->state = IRDMA_CM_STATE_OFFLOADED;
2551 		}
2552 		irdma_send_ack(cm_node);
2553 		break;
2554 	default:
2555 		ibdev_dbg(&cm_node->iwdev->ibdev,
2556 			  "CM: wrong cm_node state =%d\n", cm_node->state);
2557 		break;
2558 	}
2559 	irdma_create_event(cm_node, type);
2560 }
2561 
2562 /**
2563  * irdma_check_syn - Check for error on received syn ack
2564  * @cm_node: connection's node
2565  * @tcph: pointer tcp header
2566  */
2567 static int irdma_check_syn(struct irdma_cm_node *cm_node, struct tcphdr *tcph)
2568 {
2569 	if (ntohl(tcph->ack_seq) != cm_node->tcp_cntxt.loc_seq_num) {
2570 		irdma_active_open_err(cm_node, true);
2571 		return 1;
2572 	}
2573 
2574 	return 0;
2575 }
2576 
2577 /**
2578  * irdma_check_seq - check seq numbers if OK
2579  * @cm_node: connection's node
2580  * @tcph: pointer tcp header
2581  */
2582 static int irdma_check_seq(struct irdma_cm_node *cm_node, struct tcphdr *tcph)
2583 {
2584 	u32 seq;
2585 	u32 ack_seq;
2586 	u32 loc_seq_num = cm_node->tcp_cntxt.loc_seq_num;
2587 	u32 rcv_nxt = cm_node->tcp_cntxt.rcv_nxt;
2588 	u32 rcv_wnd;
2589 	int err = 0;
2590 
2591 	seq = ntohl(tcph->seq);
2592 	ack_seq = ntohl(tcph->ack_seq);
2593 	rcv_wnd = cm_node->tcp_cntxt.rcv_wnd;
2594 	if (ack_seq != loc_seq_num ||
2595 	    !between(seq, rcv_nxt, (rcv_nxt + rcv_wnd)))
2596 		err = -1;
2597 	if (err)
2598 		ibdev_dbg(&cm_node->iwdev->ibdev,
2599 			  "CM: seq number err\n");
2600 
2601 	return err;
2602 }
2603 
2604 void irdma_add_conn_est_qh(struct irdma_cm_node *cm_node)
2605 {
2606 	struct irdma_cm_info nfo;
2607 
2608 	irdma_get_addr_info(cm_node, &nfo);
2609 	nfo.qh_qpid = cm_node->iwdev->vsi.ilq->qp_id;
2610 	irdma_manage_qhash(cm_node->iwdev, &nfo,
2611 			   IRDMA_QHASH_TYPE_TCP_ESTABLISHED,
2612 			   IRDMA_QHASH_MANAGE_TYPE_ADD,
2613 			   cm_node, false);
2614 	cm_node->qhash_set = true;
2615 }
2616 
2617 /**
2618  * irdma_handle_syn_pkt - is for Passive node
2619  * @cm_node: connection's node
2620  * @rbuf: receive buffer
2621  */
2622 static void irdma_handle_syn_pkt(struct irdma_cm_node *cm_node,
2623 				 struct irdma_puda_buf *rbuf)
2624 {
2625 	struct tcphdr *tcph = (struct tcphdr *)rbuf->tcph;
2626 	int err;
2627 	u32 inc_sequence;
2628 	int optionsize;
2629 
2630 	optionsize = (tcph->doff << 2) - sizeof(struct tcphdr);
2631 	inc_sequence = ntohl(tcph->seq);
2632 
2633 	switch (cm_node->state) {
2634 	case IRDMA_CM_STATE_SYN_SENT:
2635 	case IRDMA_CM_STATE_MPAREQ_SENT:
2636 		/* Rcvd syn on active open connection */
2637 		irdma_active_open_err(cm_node, 1);
2638 		break;
2639 	case IRDMA_CM_STATE_LISTENING:
2640 		/* Passive OPEN */
2641 		if (atomic_read(&cm_node->listener->pend_accepts_cnt) >
2642 		    cm_node->listener->backlog) {
2643 			cm_node->cm_core->stats_backlog_drops++;
2644 			irdma_passive_open_err(cm_node, false);
2645 			break;
2646 		}
2647 		err = irdma_handle_tcp_options(cm_node, tcph, optionsize, 1);
2648 		if (err) {
2649 			irdma_passive_open_err(cm_node, false);
2650 			/* drop pkt */
2651 			break;
2652 		}
2653 		err = cm_node->cm_core->cm_create_ah(cm_node, false);
2654 		if (err) {
2655 			irdma_passive_open_err(cm_node, false);
2656 			/* drop pkt */
2657 			break;
2658 		}
2659 		cm_node->tcp_cntxt.rcv_nxt = inc_sequence + 1;
2660 		cm_node->accept_pend = 1;
2661 		atomic_inc(&cm_node->listener->pend_accepts_cnt);
2662 
2663 		cm_node->state = IRDMA_CM_STATE_SYN_RCVD;
2664 		break;
2665 	case IRDMA_CM_STATE_CLOSED:
2666 		irdma_cleanup_retrans_entry(cm_node);
2667 		refcount_inc(&cm_node->refcnt);
2668 		irdma_send_reset(cm_node);
2669 		break;
2670 	case IRDMA_CM_STATE_OFFLOADED:
2671 	case IRDMA_CM_STATE_ESTABLISHED:
2672 	case IRDMA_CM_STATE_FIN_WAIT1:
2673 	case IRDMA_CM_STATE_FIN_WAIT2:
2674 	case IRDMA_CM_STATE_MPAREQ_RCVD:
2675 	case IRDMA_CM_STATE_LAST_ACK:
2676 	case IRDMA_CM_STATE_CLOSING:
2677 	case IRDMA_CM_STATE_UNKNOWN:
2678 	default:
2679 		break;
2680 	}
2681 }
2682 
2683 /**
2684  * irdma_handle_synack_pkt - Process SYN+ACK packet (active side)
2685  * @cm_node: connection's node
2686  * @rbuf: receive buffer
2687  */
2688 static void irdma_handle_synack_pkt(struct irdma_cm_node *cm_node,
2689 				    struct irdma_puda_buf *rbuf)
2690 {
2691 	struct tcphdr *tcph = (struct tcphdr *)rbuf->tcph;
2692 	int err;
2693 	u32 inc_sequence;
2694 	int optionsize;
2695 
2696 	optionsize = (tcph->doff << 2) - sizeof(struct tcphdr);
2697 	inc_sequence = ntohl(tcph->seq);
2698 	switch (cm_node->state) {
2699 	case IRDMA_CM_STATE_SYN_SENT:
2700 		irdma_cleanup_retrans_entry(cm_node);
2701 		/* active open */
2702 		if (irdma_check_syn(cm_node, tcph)) {
2703 			ibdev_dbg(&cm_node->iwdev->ibdev,
2704 				  "CM: check syn fail\n");
2705 			return;
2706 		}
2707 		cm_node->tcp_cntxt.rem_ack_num = ntohl(tcph->ack_seq);
2708 		/* setup options */
2709 		err = irdma_handle_tcp_options(cm_node, tcph, optionsize, 0);
2710 		if (err) {
2711 			ibdev_dbg(&cm_node->iwdev->ibdev,
2712 				  "CM: cm_node=%p tcp_options failed\n",
2713 				  cm_node);
2714 			break;
2715 		}
2716 		irdma_cleanup_retrans_entry(cm_node);
2717 		cm_node->tcp_cntxt.rcv_nxt = inc_sequence + 1;
2718 		irdma_send_ack(cm_node); /* ACK  for the syn_ack */
2719 		err = irdma_send_mpa_request(cm_node);
2720 		if (err) {
2721 			ibdev_dbg(&cm_node->iwdev->ibdev,
2722 				  "CM: cm_node=%p irdma_send_mpa_request failed\n",
2723 				  cm_node);
2724 			break;
2725 		}
2726 		cm_node->state = IRDMA_CM_STATE_MPAREQ_SENT;
2727 		break;
2728 	case IRDMA_CM_STATE_MPAREQ_RCVD:
2729 		irdma_passive_open_err(cm_node, true);
2730 		break;
2731 	case IRDMA_CM_STATE_LISTENING:
2732 		cm_node->tcp_cntxt.loc_seq_num = ntohl(tcph->ack_seq);
2733 		irdma_cleanup_retrans_entry(cm_node);
2734 		cm_node->state = IRDMA_CM_STATE_CLOSED;
2735 		irdma_send_reset(cm_node);
2736 		break;
2737 	case IRDMA_CM_STATE_CLOSED:
2738 		cm_node->tcp_cntxt.loc_seq_num = ntohl(tcph->ack_seq);
2739 		irdma_cleanup_retrans_entry(cm_node);
2740 		refcount_inc(&cm_node->refcnt);
2741 		irdma_send_reset(cm_node);
2742 		break;
2743 	case IRDMA_CM_STATE_ESTABLISHED:
2744 	case IRDMA_CM_STATE_FIN_WAIT1:
2745 	case IRDMA_CM_STATE_FIN_WAIT2:
2746 	case IRDMA_CM_STATE_LAST_ACK:
2747 	case IRDMA_CM_STATE_OFFLOADED:
2748 	case IRDMA_CM_STATE_CLOSING:
2749 	case IRDMA_CM_STATE_UNKNOWN:
2750 	case IRDMA_CM_STATE_MPAREQ_SENT:
2751 	default:
2752 		break;
2753 	}
2754 }
2755 
2756 /**
2757  * irdma_handle_ack_pkt - process packet with ACK
2758  * @cm_node: connection's node
2759  * @rbuf: receive buffer
2760  */
2761 static int irdma_handle_ack_pkt(struct irdma_cm_node *cm_node,
2762 				struct irdma_puda_buf *rbuf)
2763 {
2764 	struct tcphdr *tcph = (struct tcphdr *)rbuf->tcph;
2765 	u32 inc_sequence;
2766 	int ret;
2767 	int optionsize;
2768 	u32 datasize = rbuf->datalen;
2769 
2770 	optionsize = (tcph->doff << 2) - sizeof(struct tcphdr);
2771 
2772 	if (irdma_check_seq(cm_node, tcph))
2773 		return -EINVAL;
2774 
2775 	inc_sequence = ntohl(tcph->seq);
2776 	switch (cm_node->state) {
2777 	case IRDMA_CM_STATE_SYN_RCVD:
2778 		irdma_cleanup_retrans_entry(cm_node);
2779 		ret = irdma_handle_tcp_options(cm_node, tcph, optionsize, 1);
2780 		if (ret)
2781 			return ret;
2782 		cm_node->tcp_cntxt.rem_ack_num = ntohl(tcph->ack_seq);
2783 		cm_node->state = IRDMA_CM_STATE_ESTABLISHED;
2784 		if (datasize) {
2785 			cm_node->tcp_cntxt.rcv_nxt = inc_sequence + datasize;
2786 			irdma_handle_rcv_mpa(cm_node, rbuf);
2787 		}
2788 		break;
2789 	case IRDMA_CM_STATE_ESTABLISHED:
2790 		irdma_cleanup_retrans_entry(cm_node);
2791 		if (datasize) {
2792 			cm_node->tcp_cntxt.rcv_nxt = inc_sequence + datasize;
2793 			irdma_handle_rcv_mpa(cm_node, rbuf);
2794 		}
2795 		break;
2796 	case IRDMA_CM_STATE_MPAREQ_SENT:
2797 		cm_node->tcp_cntxt.rem_ack_num = ntohl(tcph->ack_seq);
2798 		if (datasize) {
2799 			cm_node->tcp_cntxt.rcv_nxt = inc_sequence + datasize;
2800 			cm_node->ack_rcvd = false;
2801 			irdma_handle_rcv_mpa(cm_node, rbuf);
2802 		} else {
2803 			cm_node->ack_rcvd = true;
2804 		}
2805 		break;
2806 	case IRDMA_CM_STATE_LISTENING:
2807 		irdma_cleanup_retrans_entry(cm_node);
2808 		cm_node->state = IRDMA_CM_STATE_CLOSED;
2809 		irdma_send_reset(cm_node);
2810 		break;
2811 	case IRDMA_CM_STATE_CLOSED:
2812 		irdma_cleanup_retrans_entry(cm_node);
2813 		refcount_inc(&cm_node->refcnt);
2814 		irdma_send_reset(cm_node);
2815 		break;
2816 	case IRDMA_CM_STATE_LAST_ACK:
2817 	case IRDMA_CM_STATE_CLOSING:
2818 		irdma_cleanup_retrans_entry(cm_node);
2819 		cm_node->state = IRDMA_CM_STATE_CLOSED;
2820 		irdma_rem_ref_cm_node(cm_node);
2821 		break;
2822 	case IRDMA_CM_STATE_FIN_WAIT1:
2823 		irdma_cleanup_retrans_entry(cm_node);
2824 		cm_node->state = IRDMA_CM_STATE_FIN_WAIT2;
2825 		break;
2826 	case IRDMA_CM_STATE_SYN_SENT:
2827 	case IRDMA_CM_STATE_FIN_WAIT2:
2828 	case IRDMA_CM_STATE_OFFLOADED:
2829 	case IRDMA_CM_STATE_MPAREQ_RCVD:
2830 	case IRDMA_CM_STATE_UNKNOWN:
2831 	default:
2832 		irdma_cleanup_retrans_entry(cm_node);
2833 		break;
2834 	}
2835 
2836 	return 0;
2837 }
2838 
2839 /**
2840  * irdma_process_pkt - process cm packet
2841  * @cm_node: connection's node
2842  * @rbuf: receive buffer
2843  */
2844 static void irdma_process_pkt(struct irdma_cm_node *cm_node,
2845 			      struct irdma_puda_buf *rbuf)
2846 {
2847 	enum irdma_tcpip_pkt_type pkt_type = IRDMA_PKT_TYPE_UNKNOWN;
2848 	struct tcphdr *tcph = (struct tcphdr *)rbuf->tcph;
2849 	u32 fin_set = 0;
2850 	int err;
2851 
2852 	if (tcph->rst) {
2853 		pkt_type = IRDMA_PKT_TYPE_RST;
2854 	} else if (tcph->syn) {
2855 		pkt_type = IRDMA_PKT_TYPE_SYN;
2856 		if (tcph->ack)
2857 			pkt_type = IRDMA_PKT_TYPE_SYNACK;
2858 	} else if (tcph->ack) {
2859 		pkt_type = IRDMA_PKT_TYPE_ACK;
2860 	}
2861 	if (tcph->fin)
2862 		fin_set = 1;
2863 
2864 	switch (pkt_type) {
2865 	case IRDMA_PKT_TYPE_SYN:
2866 		irdma_handle_syn_pkt(cm_node, rbuf);
2867 		break;
2868 	case IRDMA_PKT_TYPE_SYNACK:
2869 		irdma_handle_synack_pkt(cm_node, rbuf);
2870 		break;
2871 	case IRDMA_PKT_TYPE_ACK:
2872 		err = irdma_handle_ack_pkt(cm_node, rbuf);
2873 		if (fin_set && !err)
2874 			irdma_handle_fin_pkt(cm_node);
2875 		break;
2876 	case IRDMA_PKT_TYPE_RST:
2877 		irdma_handle_rst_pkt(cm_node, rbuf);
2878 		break;
2879 	default:
2880 		if (fin_set &&
2881 		    (!irdma_check_seq(cm_node, (struct tcphdr *)rbuf->tcph)))
2882 			irdma_handle_fin_pkt(cm_node);
2883 		break;
2884 	}
2885 }
2886 
2887 /**
2888  * irdma_make_listen_node - create a listen node with params
2889  * @cm_core: cm's core
2890  * @iwdev: iwarp device structure
2891  * @cm_info: quad info for connection
2892  */
2893 static struct irdma_cm_listener *
2894 irdma_make_listen_node(struct irdma_cm_core *cm_core,
2895 		       struct irdma_device *iwdev,
2896 		       struct irdma_cm_info *cm_info)
2897 {
2898 	struct irdma_cm_listener *listener;
2899 	unsigned long flags;
2900 
2901 	/* cannot have multiple matching listeners */
2902 	listener = irdma_find_listener(cm_core, cm_info->loc_addr,
2903 				       cm_info->loc_port, cm_info->vlan_id,
2904 				       IRDMA_CM_LISTENER_EITHER_STATE);
2905 	if (listener &&
2906 	    listener->listener_state == IRDMA_CM_LISTENER_ACTIVE_STATE) {
2907 		refcount_dec(&listener->refcnt);
2908 		return NULL;
2909 	}
2910 
2911 	if (!listener) {
2912 		/* create a CM listen node
2913 		 * 1/2 node to compare incoming traffic to
2914 		 */
2915 		listener = kzalloc(sizeof(*listener), GFP_KERNEL);
2916 		if (!listener)
2917 			return NULL;
2918 		cm_core->stats_listen_nodes_created++;
2919 		memcpy(listener->loc_addr, cm_info->loc_addr,
2920 		       sizeof(listener->loc_addr));
2921 		listener->loc_port = cm_info->loc_port;
2922 
2923 		INIT_LIST_HEAD(&listener->child_listen_list);
2924 
2925 		refcount_set(&listener->refcnt, 1);
2926 	} else {
2927 		listener->reused_node = 1;
2928 	}
2929 
2930 	listener->cm_id = cm_info->cm_id;
2931 	listener->ipv4 = cm_info->ipv4;
2932 	listener->vlan_id = cm_info->vlan_id;
2933 	atomic_set(&listener->pend_accepts_cnt, 0);
2934 	listener->cm_core = cm_core;
2935 	listener->iwdev = iwdev;
2936 
2937 	listener->backlog = cm_info->backlog;
2938 	listener->listener_state = IRDMA_CM_LISTENER_ACTIVE_STATE;
2939 
2940 	if (!listener->reused_node) {
2941 		spin_lock_irqsave(&cm_core->listen_list_lock, flags);
2942 		list_add(&listener->list, &cm_core->listen_list);
2943 		spin_unlock_irqrestore(&cm_core->listen_list_lock, flags);
2944 	}
2945 
2946 	return listener;
2947 }
2948 
2949 /**
2950  * irdma_create_cm_node - make a connection node with params
2951  * @cm_core: cm's core
2952  * @iwdev: iwarp device structure
2953  * @conn_param: connection parameters
2954  * @cm_info: quad info for connection
2955  * @caller_cm_node: pointer to cm_node structure to return
2956  */
2957 static int irdma_create_cm_node(struct irdma_cm_core *cm_core,
2958 				struct irdma_device *iwdev,
2959 				struct iw_cm_conn_param *conn_param,
2960 				struct irdma_cm_info *cm_info,
2961 				struct irdma_cm_node **caller_cm_node)
2962 {
2963 	struct irdma_cm_node *cm_node;
2964 	u16 private_data_len = conn_param->private_data_len;
2965 	const void *private_data = conn_param->private_data;
2966 
2967 	/* create a CM connection node */
2968 	cm_node = irdma_make_cm_node(cm_core, iwdev, cm_info, NULL);
2969 	if (!cm_node)
2970 		return -ENOMEM;
2971 
2972 	/* set our node side to client (active) side */
2973 	cm_node->tcp_cntxt.client = 1;
2974 	cm_node->tcp_cntxt.rcv_wscale = IRDMA_CM_DEFAULT_RCV_WND_SCALE;
2975 
2976 	irdma_record_ird_ord(cm_node, conn_param->ird, conn_param->ord);
2977 
2978 	cm_node->pdata.size = private_data_len;
2979 	cm_node->pdata.addr = cm_node->pdata_buf;
2980 
2981 	memcpy(cm_node->pdata_buf, private_data, private_data_len);
2982 	*caller_cm_node = cm_node;
2983 
2984 	return 0;
2985 }
2986 
2987 /**
2988  * irdma_cm_reject - reject and teardown a connection
2989  * @cm_node: connection's node
2990  * @pdata: ptr to private data for reject
2991  * @plen: size of private data
2992  */
2993 static int irdma_cm_reject(struct irdma_cm_node *cm_node, const void *pdata,
2994 			   u8 plen)
2995 {
2996 	int ret;
2997 	int passive_state;
2998 
2999 	if (cm_node->tcp_cntxt.client)
3000 		return 0;
3001 
3002 	irdma_cleanup_retrans_entry(cm_node);
3003 
3004 	passive_state = atomic_add_return(1, &cm_node->passive_state);
3005 	if (passive_state == IRDMA_SEND_RESET_EVENT) {
3006 		cm_node->state = IRDMA_CM_STATE_CLOSED;
3007 		irdma_rem_ref_cm_node(cm_node);
3008 		return 0;
3009 	}
3010 
3011 	if (cm_node->state == IRDMA_CM_STATE_LISTENER_DESTROYED) {
3012 		irdma_rem_ref_cm_node(cm_node);
3013 		return 0;
3014 	}
3015 
3016 	ret = irdma_send_mpa_reject(cm_node, pdata, plen);
3017 	if (!ret)
3018 		return 0;
3019 
3020 	cm_node->state = IRDMA_CM_STATE_CLOSED;
3021 	if (irdma_send_reset(cm_node))
3022 		ibdev_dbg(&cm_node->iwdev->ibdev,
3023 			  "CM: send reset failed\n");
3024 
3025 	return ret;
3026 }
3027 
3028 /**
3029  * irdma_cm_close - close of cm connection
3030  * @cm_node: connection's node
3031  */
3032 static int irdma_cm_close(struct irdma_cm_node *cm_node)
3033 {
3034 	switch (cm_node->state) {
3035 	case IRDMA_CM_STATE_SYN_RCVD:
3036 	case IRDMA_CM_STATE_SYN_SENT:
3037 	case IRDMA_CM_STATE_ONE_SIDE_ESTABLISHED:
3038 	case IRDMA_CM_STATE_ESTABLISHED:
3039 	case IRDMA_CM_STATE_ACCEPTING:
3040 	case IRDMA_CM_STATE_MPAREQ_SENT:
3041 	case IRDMA_CM_STATE_MPAREQ_RCVD:
3042 		irdma_cleanup_retrans_entry(cm_node);
3043 		irdma_send_reset(cm_node);
3044 		break;
3045 	case IRDMA_CM_STATE_CLOSE_WAIT:
3046 		cm_node->state = IRDMA_CM_STATE_LAST_ACK;
3047 		irdma_send_fin(cm_node);
3048 		break;
3049 	case IRDMA_CM_STATE_FIN_WAIT1:
3050 	case IRDMA_CM_STATE_FIN_WAIT2:
3051 	case IRDMA_CM_STATE_LAST_ACK:
3052 	case IRDMA_CM_STATE_TIME_WAIT:
3053 	case IRDMA_CM_STATE_CLOSING:
3054 		return -EINVAL;
3055 	case IRDMA_CM_STATE_LISTENING:
3056 		irdma_cleanup_retrans_entry(cm_node);
3057 		irdma_send_reset(cm_node);
3058 		break;
3059 	case IRDMA_CM_STATE_MPAREJ_RCVD:
3060 	case IRDMA_CM_STATE_UNKNOWN:
3061 	case IRDMA_CM_STATE_INITED:
3062 	case IRDMA_CM_STATE_CLOSED:
3063 	case IRDMA_CM_STATE_LISTENER_DESTROYED:
3064 		irdma_rem_ref_cm_node(cm_node);
3065 		break;
3066 	case IRDMA_CM_STATE_OFFLOADED:
3067 		if (cm_node->send_entry)
3068 			ibdev_dbg(&cm_node->iwdev->ibdev,
3069 				  "CM: CM send_entry in OFFLOADED state\n");
3070 		irdma_rem_ref_cm_node(cm_node);
3071 		break;
3072 	}
3073 
3074 	return 0;
3075 }
3076 
3077 /**
3078  * irdma_receive_ilq - recv an ETHERNET packet, and process it
3079  * through CM
3080  * @vsi: VSI structure of dev
3081  * @rbuf: receive buffer
3082  */
3083 void irdma_receive_ilq(struct irdma_sc_vsi *vsi, struct irdma_puda_buf *rbuf)
3084 {
3085 	struct irdma_cm_node *cm_node;
3086 	struct irdma_cm_listener *listener;
3087 	struct iphdr *iph;
3088 	struct ipv6hdr *ip6h;
3089 	struct tcphdr *tcph;
3090 	struct irdma_cm_info cm_info = {};
3091 	struct irdma_device *iwdev = vsi->back_vsi;
3092 	struct irdma_cm_core *cm_core = &iwdev->cm_core;
3093 	struct vlan_ethhdr *ethh;
3094 	u16 vtag;
3095 
3096 	/* if vlan, then maclen = 18 else 14 */
3097 	iph = (struct iphdr *)rbuf->iph;
3098 	print_hex_dump_debug("ILQ: RECEIVE ILQ BUFFER", DUMP_PREFIX_OFFSET,
3099 			     16, 8, rbuf->mem.va, rbuf->totallen, false);
3100 	if (iwdev->rf->sc_dev.hw_attrs.uk_attrs.hw_rev >= IRDMA_GEN_2) {
3101 		if (rbuf->vlan_valid) {
3102 			vtag = rbuf->vlan_id;
3103 			cm_info.user_pri = (vtag & VLAN_PRIO_MASK) >>
3104 					   VLAN_PRIO_SHIFT;
3105 			cm_info.vlan_id = vtag & VLAN_VID_MASK;
3106 		} else {
3107 			cm_info.vlan_id = 0xFFFF;
3108 		}
3109 	} else {
3110 		ethh = rbuf->mem.va;
3111 
3112 		if (ethh->h_vlan_proto == htons(ETH_P_8021Q)) {
3113 			vtag = ntohs(ethh->h_vlan_TCI);
3114 			cm_info.user_pri = (vtag & VLAN_PRIO_MASK) >>
3115 					   VLAN_PRIO_SHIFT;
3116 			cm_info.vlan_id = vtag & VLAN_VID_MASK;
3117 			ibdev_dbg(&cm_core->iwdev->ibdev,
3118 				  "CM: vlan_id=%d\n", cm_info.vlan_id);
3119 		} else {
3120 			cm_info.vlan_id = 0xFFFF;
3121 		}
3122 	}
3123 	tcph = (struct tcphdr *)rbuf->tcph;
3124 
3125 	if (rbuf->ipv4) {
3126 		cm_info.loc_addr[0] = ntohl(iph->daddr);
3127 		cm_info.rem_addr[0] = ntohl(iph->saddr);
3128 		cm_info.ipv4 = true;
3129 		cm_info.tos = iph->tos;
3130 	} else {
3131 		ip6h = (struct ipv6hdr *)rbuf->iph;
3132 		irdma_copy_ip_ntohl(cm_info.loc_addr,
3133 				    ip6h->daddr.in6_u.u6_addr32);
3134 		irdma_copy_ip_ntohl(cm_info.rem_addr,
3135 				    ip6h->saddr.in6_u.u6_addr32);
3136 		cm_info.ipv4 = false;
3137 		cm_info.tos = (ip6h->priority << 4) | (ip6h->flow_lbl[0] >> 4);
3138 	}
3139 	cm_info.loc_port = ntohs(tcph->dest);
3140 	cm_info.rem_port = ntohs(tcph->source);
3141 	cm_node = irdma_find_node(cm_core, cm_info.rem_port, cm_info.rem_addr,
3142 				  cm_info.loc_port, cm_info.loc_addr, cm_info.vlan_id);
3143 
3144 	if (!cm_node) {
3145 		/* Only type of packet accepted are for the
3146 		 * PASSIVE open (syn only)
3147 		 */
3148 		if (!tcph->syn || tcph->ack)
3149 			return;
3150 
3151 		listener = irdma_find_listener(cm_core,
3152 					       cm_info.loc_addr,
3153 					       cm_info.loc_port,
3154 					       cm_info.vlan_id,
3155 					       IRDMA_CM_LISTENER_ACTIVE_STATE);
3156 		if (!listener) {
3157 			cm_info.cm_id = NULL;
3158 			ibdev_dbg(&cm_core->iwdev->ibdev,
3159 				  "CM: no listener found\n");
3160 			return;
3161 		}
3162 
3163 		cm_info.cm_id = listener->cm_id;
3164 		cm_node = irdma_make_cm_node(cm_core, iwdev, &cm_info,
3165 					     listener);
3166 		if (!cm_node) {
3167 			ibdev_dbg(&cm_core->iwdev->ibdev,
3168 				  "CM: allocate node failed\n");
3169 			refcount_dec(&listener->refcnt);
3170 			return;
3171 		}
3172 
3173 		if (!tcph->rst && !tcph->fin) {
3174 			cm_node->state = IRDMA_CM_STATE_LISTENING;
3175 		} else {
3176 			irdma_rem_ref_cm_node(cm_node);
3177 			return;
3178 		}
3179 
3180 		refcount_inc(&cm_node->refcnt);
3181 	} else if (cm_node->state == IRDMA_CM_STATE_OFFLOADED) {
3182 		irdma_rem_ref_cm_node(cm_node);
3183 		return;
3184 	}
3185 
3186 	irdma_process_pkt(cm_node, rbuf);
3187 	irdma_rem_ref_cm_node(cm_node);
3188 }
3189 
3190 static int irdma_add_qh(struct irdma_cm_node *cm_node, bool active)
3191 {
3192 	if (!active)
3193 		irdma_add_conn_est_qh(cm_node);
3194 	return 0;
3195 }
3196 
3197 static void irdma_cm_free_ah_nop(struct irdma_cm_node *cm_node)
3198 {
3199 }
3200 
3201 /**
3202  * irdma_setup_cm_core - setup top level instance of a cm core
3203  * @iwdev: iwarp device structure
3204  * @rdma_ver: HW version
3205  */
3206 int irdma_setup_cm_core(struct irdma_device *iwdev, u8 rdma_ver)
3207 {
3208 	struct irdma_cm_core *cm_core = &iwdev->cm_core;
3209 
3210 	cm_core->iwdev = iwdev;
3211 	cm_core->dev = &iwdev->rf->sc_dev;
3212 
3213 	/* Handles CM event work items send to Iwarp core */
3214 	cm_core->event_wq = alloc_ordered_workqueue("iwarp-event-wq", 0);
3215 	if (!cm_core->event_wq)
3216 		return -ENOMEM;
3217 
3218 	INIT_LIST_HEAD(&cm_core->listen_list);
3219 
3220 	timer_setup(&cm_core->tcp_timer, irdma_cm_timer_tick, 0);
3221 
3222 	spin_lock_init(&cm_core->ht_lock);
3223 	spin_lock_init(&cm_core->listen_list_lock);
3224 	spin_lock_init(&cm_core->apbvt_lock);
3225 	switch (rdma_ver) {
3226 	case IRDMA_GEN_1:
3227 		cm_core->form_cm_frame = irdma_form_uda_cm_frame;
3228 		cm_core->cm_create_ah = irdma_add_qh;
3229 		cm_core->cm_free_ah = irdma_cm_free_ah_nop;
3230 		break;
3231 	case IRDMA_GEN_2:
3232 	default:
3233 		cm_core->form_cm_frame = irdma_form_ah_cm_frame;
3234 		cm_core->cm_create_ah = irdma_cm_create_ah;
3235 		cm_core->cm_free_ah = irdma_cm_free_ah;
3236 	}
3237 
3238 	return 0;
3239 }
3240 
3241 /**
3242  * irdma_cleanup_cm_core - deallocate a top level instance of a
3243  * cm core
3244  * @cm_core: cm's core
3245  */
3246 void irdma_cleanup_cm_core(struct irdma_cm_core *cm_core)
3247 {
3248 	if (!cm_core)
3249 		return;
3250 
3251 	del_timer_sync(&cm_core->tcp_timer);
3252 
3253 	destroy_workqueue(cm_core->event_wq);
3254 	cm_core->dev->ws_reset(&cm_core->iwdev->vsi);
3255 }
3256 
3257 /**
3258  * irdma_init_tcp_ctx - setup qp context
3259  * @cm_node: connection's node
3260  * @tcp_info: offload info for tcp
3261  * @iwqp: associate qp for the connection
3262  */
3263 static void irdma_init_tcp_ctx(struct irdma_cm_node *cm_node,
3264 			       struct irdma_tcp_offload_info *tcp_info,
3265 			       struct irdma_qp *iwqp)
3266 {
3267 	tcp_info->ipv4 = cm_node->ipv4;
3268 	tcp_info->drop_ooo_seg = !iwqp->iwdev->iw_ooo;
3269 	tcp_info->wscale = true;
3270 	tcp_info->ignore_tcp_opt = true;
3271 	tcp_info->ignore_tcp_uns_opt = true;
3272 	tcp_info->no_nagle = false;
3273 
3274 	tcp_info->ttl = IRDMA_DEFAULT_TTL;
3275 	tcp_info->rtt_var = IRDMA_DEFAULT_RTT_VAR;
3276 	tcp_info->ss_thresh = IRDMA_DEFAULT_SS_THRESH;
3277 	tcp_info->rexmit_thresh = IRDMA_DEFAULT_REXMIT_THRESH;
3278 
3279 	tcp_info->tcp_state = IRDMA_TCP_STATE_ESTABLISHED;
3280 	tcp_info->snd_wscale = cm_node->tcp_cntxt.snd_wscale;
3281 	tcp_info->rcv_wscale = cm_node->tcp_cntxt.rcv_wscale;
3282 
3283 	tcp_info->snd_nxt = cm_node->tcp_cntxt.loc_seq_num;
3284 	tcp_info->snd_wnd = cm_node->tcp_cntxt.snd_wnd;
3285 	tcp_info->rcv_nxt = cm_node->tcp_cntxt.rcv_nxt;
3286 	tcp_info->snd_max = cm_node->tcp_cntxt.loc_seq_num;
3287 
3288 	tcp_info->snd_una = cm_node->tcp_cntxt.loc_seq_num;
3289 	tcp_info->cwnd = 2 * cm_node->tcp_cntxt.mss;
3290 	tcp_info->snd_wl1 = cm_node->tcp_cntxt.rcv_nxt;
3291 	tcp_info->snd_wl2 = cm_node->tcp_cntxt.loc_seq_num;
3292 	tcp_info->max_snd_window = cm_node->tcp_cntxt.max_snd_wnd;
3293 	tcp_info->rcv_wnd = cm_node->tcp_cntxt.rcv_wnd
3294 			    << cm_node->tcp_cntxt.rcv_wscale;
3295 
3296 	tcp_info->flow_label = 0;
3297 	tcp_info->snd_mss = (u32)cm_node->tcp_cntxt.mss;
3298 	tcp_info->tos = cm_node->tos;
3299 	if (cm_node->vlan_id < VLAN_N_VID) {
3300 		tcp_info->insert_vlan_tag = true;
3301 		tcp_info->vlan_tag = cm_node->vlan_id;
3302 		tcp_info->vlan_tag |= cm_node->user_pri << VLAN_PRIO_SHIFT;
3303 	}
3304 	if (cm_node->ipv4) {
3305 		tcp_info->src_port = cm_node->loc_port;
3306 		tcp_info->dst_port = cm_node->rem_port;
3307 
3308 		tcp_info->dest_ip_addr[3] = cm_node->rem_addr[0];
3309 		tcp_info->local_ipaddr[3] = cm_node->loc_addr[0];
3310 		tcp_info->arp_idx = (u16)irdma_arp_table(iwqp->iwdev->rf,
3311 							 &tcp_info->dest_ip_addr[3],
3312 							 true, NULL,
3313 							 IRDMA_ARP_RESOLVE);
3314 	} else {
3315 		tcp_info->src_port = cm_node->loc_port;
3316 		tcp_info->dst_port = cm_node->rem_port;
3317 		memcpy(tcp_info->dest_ip_addr, cm_node->rem_addr,
3318 		       sizeof(tcp_info->dest_ip_addr));
3319 		memcpy(tcp_info->local_ipaddr, cm_node->loc_addr,
3320 		       sizeof(tcp_info->local_ipaddr));
3321 
3322 		tcp_info->arp_idx = (u16)irdma_arp_table(iwqp->iwdev->rf,
3323 							 &tcp_info->dest_ip_addr[0],
3324 							 false, NULL,
3325 							 IRDMA_ARP_RESOLVE);
3326 	}
3327 }
3328 
3329 /**
3330  * irdma_cm_init_tsa_conn - setup qp for RTS
3331  * @iwqp: associate qp for the connection
3332  * @cm_node: connection's node
3333  */
3334 static void irdma_cm_init_tsa_conn(struct irdma_qp *iwqp,
3335 				   struct irdma_cm_node *cm_node)
3336 {
3337 	struct irdma_iwarp_offload_info *iwarp_info;
3338 	struct irdma_qp_host_ctx_info *ctx_info;
3339 
3340 	iwarp_info = &iwqp->iwarp_info;
3341 	ctx_info = &iwqp->ctx_info;
3342 
3343 	ctx_info->tcp_info = &iwqp->tcp_info;
3344 	ctx_info->send_cq_num = iwqp->iwscq->sc_cq.cq_uk.cq_id;
3345 	ctx_info->rcv_cq_num = iwqp->iwrcq->sc_cq.cq_uk.cq_id;
3346 
3347 	iwarp_info->ord_size = cm_node->ord_size;
3348 	iwarp_info->ird_size = cm_node->ird_size;
3349 	iwarp_info->rd_en = true;
3350 	iwarp_info->rdmap_ver = 1;
3351 	iwarp_info->ddp_ver = 1;
3352 	iwarp_info->pd_id = iwqp->iwpd->sc_pd.pd_id;
3353 
3354 	ctx_info->tcp_info_valid = true;
3355 	ctx_info->iwarp_info_valid = true;
3356 	ctx_info->user_pri = cm_node->user_pri;
3357 
3358 	irdma_init_tcp_ctx(cm_node, &iwqp->tcp_info, iwqp);
3359 	if (cm_node->snd_mark_en) {
3360 		iwarp_info->snd_mark_en = true;
3361 		iwarp_info->snd_mark_offset = (iwqp->tcp_info.snd_nxt & SNDMARKER_SEQNMASK) +
3362 					       cm_node->lsmm_size;
3363 	}
3364 
3365 	cm_node->state = IRDMA_CM_STATE_OFFLOADED;
3366 	iwqp->tcp_info.tcp_state = IRDMA_TCP_STATE_ESTABLISHED;
3367 	iwqp->tcp_info.src_mac_addr_idx = iwqp->iwdev->mac_ip_table_idx;
3368 
3369 	if (cm_node->rcv_mark_en) {
3370 		iwarp_info->rcv_mark_en = true;
3371 		iwarp_info->align_hdrs = true;
3372 	}
3373 
3374 	irdma_sc_qp_setctx(&iwqp->sc_qp, iwqp->host_ctx.va, ctx_info);
3375 
3376 	/* once tcp_info is set, no need to do it again */
3377 	ctx_info->tcp_info_valid = false;
3378 	ctx_info->iwarp_info_valid = false;
3379 }
3380 
3381 /**
3382  * irdma_cm_disconn - when a connection is being closed
3383  * @iwqp: associated qp for the connection
3384  */
3385 void irdma_cm_disconn(struct irdma_qp *iwqp)
3386 {
3387 	struct irdma_device *iwdev = iwqp->iwdev;
3388 	struct disconn_work *work;
3389 	unsigned long flags;
3390 
3391 	work = kzalloc(sizeof(*work), GFP_ATOMIC);
3392 	if (!work)
3393 		return;
3394 
3395 	spin_lock_irqsave(&iwdev->rf->qptable_lock, flags);
3396 	if (!iwdev->rf->qp_table[iwqp->ibqp.qp_num]) {
3397 		spin_unlock_irqrestore(&iwdev->rf->qptable_lock, flags);
3398 		ibdev_dbg(&iwdev->ibdev,
3399 			  "CM: qp_id %d is already freed\n",
3400 			  iwqp->ibqp.qp_num);
3401 		kfree(work);
3402 		return;
3403 	}
3404 	irdma_qp_add_ref(&iwqp->ibqp);
3405 	spin_unlock_irqrestore(&iwdev->rf->qptable_lock, flags);
3406 
3407 	work->iwqp = iwqp;
3408 	INIT_WORK(&work->work, irdma_disconnect_worker);
3409 	queue_work(iwdev->cleanup_wq, &work->work);
3410 }
3411 
3412 /**
3413  * irdma_qp_disconnect - free qp and close cm
3414  * @iwqp: associate qp for the connection
3415  */
3416 static void irdma_qp_disconnect(struct irdma_qp *iwqp)
3417 {
3418 	struct irdma_device *iwdev = iwqp->iwdev;
3419 
3420 	iwqp->active_conn = 0;
3421 	/* close the CM node down if it is still active */
3422 	ibdev_dbg(&iwdev->ibdev, "CM: Call close API\n");
3423 	irdma_cm_close(iwqp->cm_node);
3424 }
3425 
3426 /**
3427  * irdma_cm_disconn_true - called by worker thread to disconnect qp
3428  * @iwqp: associate qp for the connection
3429  */
3430 static void irdma_cm_disconn_true(struct irdma_qp *iwqp)
3431 {
3432 	struct iw_cm_id *cm_id;
3433 	struct irdma_device *iwdev;
3434 	struct irdma_sc_qp *qp = &iwqp->sc_qp;
3435 	u16 last_ae;
3436 	u8 original_hw_tcp_state;
3437 	u8 original_ibqp_state;
3438 	int disconn_status = 0;
3439 	int issue_disconn = 0;
3440 	int issue_close = 0;
3441 	int issue_flush = 0;
3442 	unsigned long flags;
3443 	int err;
3444 
3445 	iwdev = iwqp->iwdev;
3446 	spin_lock_irqsave(&iwqp->lock, flags);
3447 	if (rdma_protocol_roce(&iwdev->ibdev, 1)) {
3448 		struct ib_qp_attr attr;
3449 
3450 		if (iwqp->flush_issued || iwqp->sc_qp.qp_uk.destroy_pending) {
3451 			spin_unlock_irqrestore(&iwqp->lock, flags);
3452 			return;
3453 		}
3454 
3455 		spin_unlock_irqrestore(&iwqp->lock, flags);
3456 
3457 		attr.qp_state = IB_QPS_ERR;
3458 		irdma_modify_qp_roce(&iwqp->ibqp, &attr, IB_QP_STATE, NULL);
3459 		irdma_ib_qp_event(iwqp, qp->event_type);
3460 		return;
3461 	}
3462 
3463 	cm_id = iwqp->cm_id;
3464 	original_hw_tcp_state = iwqp->hw_tcp_state;
3465 	original_ibqp_state = iwqp->ibqp_state;
3466 	last_ae = iwqp->last_aeq;
3467 
3468 	if (qp->term_flags) {
3469 		issue_disconn = 1;
3470 		issue_close = 1;
3471 		iwqp->cm_id = NULL;
3472 		irdma_terminate_del_timer(qp);
3473 		if (!iwqp->flush_issued) {
3474 			iwqp->flush_issued = 1;
3475 			issue_flush = 1;
3476 		}
3477 	} else if ((original_hw_tcp_state == IRDMA_TCP_STATE_CLOSE_WAIT) ||
3478 		   ((original_ibqp_state == IB_QPS_RTS) &&
3479 		    (last_ae == IRDMA_AE_LLP_CONNECTION_RESET))) {
3480 		issue_disconn = 1;
3481 		if (last_ae == IRDMA_AE_LLP_CONNECTION_RESET)
3482 			disconn_status = -ECONNRESET;
3483 	}
3484 
3485 	if (original_hw_tcp_state == IRDMA_TCP_STATE_CLOSED ||
3486 	    original_hw_tcp_state == IRDMA_TCP_STATE_TIME_WAIT ||
3487 	    last_ae == IRDMA_AE_RDMAP_ROE_BAD_LLP_CLOSE ||
3488 	    last_ae == IRDMA_AE_BAD_CLOSE ||
3489 	    last_ae == IRDMA_AE_LLP_CONNECTION_RESET || iwdev->rf->reset || !cm_id) {
3490 		issue_close = 1;
3491 		iwqp->cm_id = NULL;
3492 		qp->term_flags = 0;
3493 		if (!iwqp->flush_issued) {
3494 			iwqp->flush_issued = 1;
3495 			issue_flush = 1;
3496 		}
3497 	}
3498 
3499 	spin_unlock_irqrestore(&iwqp->lock, flags);
3500 	if (issue_flush && !iwqp->sc_qp.qp_uk.destroy_pending) {
3501 		irdma_flush_wqes(iwqp, IRDMA_FLUSH_SQ | IRDMA_FLUSH_RQ |
3502 				 IRDMA_FLUSH_WAIT);
3503 
3504 		if (qp->term_flags)
3505 			irdma_ib_qp_event(iwqp, qp->event_type);
3506 	}
3507 
3508 	if (!cm_id || !cm_id->event_handler)
3509 		return;
3510 
3511 	spin_lock_irqsave(&iwdev->cm_core.ht_lock, flags);
3512 	if (!iwqp->cm_node) {
3513 		spin_unlock_irqrestore(&iwdev->cm_core.ht_lock, flags);
3514 		return;
3515 	}
3516 	refcount_inc(&iwqp->cm_node->refcnt);
3517 
3518 	spin_unlock_irqrestore(&iwdev->cm_core.ht_lock, flags);
3519 
3520 	if (issue_disconn) {
3521 		err = irdma_send_cm_event(iwqp->cm_node, cm_id,
3522 					  IW_CM_EVENT_DISCONNECT,
3523 					  disconn_status);
3524 		if (err)
3525 			ibdev_dbg(&iwdev->ibdev,
3526 				  "CM: disconnect event failed: - cm_id = %p\n",
3527 				  cm_id);
3528 	}
3529 	if (issue_close) {
3530 		cm_id->provider_data = iwqp;
3531 		err = irdma_send_cm_event(iwqp->cm_node, cm_id,
3532 					  IW_CM_EVENT_CLOSE, 0);
3533 		if (err)
3534 			ibdev_dbg(&iwdev->ibdev,
3535 				  "CM: close event failed: - cm_id = %p\n",
3536 				  cm_id);
3537 		irdma_qp_disconnect(iwqp);
3538 	}
3539 	irdma_rem_ref_cm_node(iwqp->cm_node);
3540 }
3541 
3542 /**
3543  * irdma_disconnect_worker - worker for connection close
3544  * @work: points or disconn structure
3545  */
3546 static void irdma_disconnect_worker(struct work_struct *work)
3547 {
3548 	struct disconn_work *dwork = container_of(work, struct disconn_work, work);
3549 	struct irdma_qp *iwqp = dwork->iwqp;
3550 
3551 	kfree(dwork);
3552 	irdma_cm_disconn_true(iwqp);
3553 	irdma_qp_rem_ref(&iwqp->ibqp);
3554 }
3555 
3556 /**
3557  * irdma_free_lsmm_rsrc - free lsmm memory and deregister
3558  * @iwqp: associate qp for the connection
3559  */
3560 void irdma_free_lsmm_rsrc(struct irdma_qp *iwqp)
3561 {
3562 	struct irdma_device *iwdev;
3563 
3564 	iwdev = iwqp->iwdev;
3565 
3566 	if (iwqp->ietf_mem.va) {
3567 		if (iwqp->lsmm_mr)
3568 			iwdev->ibdev.ops.dereg_mr(iwqp->lsmm_mr, NULL);
3569 		dma_free_coherent(iwdev->rf->sc_dev.hw->device,
3570 				  iwqp->ietf_mem.size, iwqp->ietf_mem.va,
3571 				  iwqp->ietf_mem.pa);
3572 		iwqp->ietf_mem.va = NULL;
3573 		iwqp->ietf_mem.va = NULL;
3574 	}
3575 }
3576 
3577 /**
3578  * irdma_accept - registered call for connection to be accepted
3579  * @cm_id: cm information for passive connection
3580  * @conn_param: accpet parameters
3581  */
3582 int irdma_accept(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
3583 {
3584 	struct ib_qp *ibqp;
3585 	struct irdma_qp *iwqp;
3586 	struct irdma_device *iwdev;
3587 	struct irdma_sc_dev *dev;
3588 	struct irdma_cm_node *cm_node;
3589 	struct ib_qp_attr attr = {};
3590 	int passive_state;
3591 	struct ib_mr *ibmr;
3592 	struct irdma_pd *iwpd;
3593 	u16 buf_len = 0;
3594 	struct irdma_kmem_info accept;
3595 	u64 tagged_offset;
3596 	int wait_ret;
3597 	int ret = 0;
3598 
3599 	ibqp = irdma_get_qp(cm_id->device, conn_param->qpn);
3600 	if (!ibqp)
3601 		return -EINVAL;
3602 
3603 	iwqp = to_iwqp(ibqp);
3604 	iwdev = iwqp->iwdev;
3605 	dev = &iwdev->rf->sc_dev;
3606 	cm_node = cm_id->provider_data;
3607 
3608 	if (((struct sockaddr_in *)&cm_id->local_addr)->sin_family == AF_INET) {
3609 		cm_node->ipv4 = true;
3610 		cm_node->vlan_id = irdma_get_vlan_ipv4(cm_node->loc_addr);
3611 	} else {
3612 		cm_node->ipv4 = false;
3613 		irdma_netdev_vlan_ipv6(cm_node->loc_addr, &cm_node->vlan_id,
3614 				       NULL);
3615 	}
3616 	ibdev_dbg(&iwdev->ibdev, "CM: Accept vlan_id=%d\n",
3617 		  cm_node->vlan_id);
3618 
3619 	trace_irdma_accept(cm_node, 0, NULL);
3620 
3621 	if (cm_node->state == IRDMA_CM_STATE_LISTENER_DESTROYED) {
3622 		ret = -EINVAL;
3623 		goto error;
3624 	}
3625 
3626 	passive_state = atomic_add_return(1, &cm_node->passive_state);
3627 	if (passive_state == IRDMA_SEND_RESET_EVENT) {
3628 		ret = -ECONNRESET;
3629 		goto error;
3630 	}
3631 
3632 	buf_len = conn_param->private_data_len + IRDMA_MAX_IETF_SIZE;
3633 	iwqp->ietf_mem.size = ALIGN(buf_len, 1);
3634 	iwqp->ietf_mem.va = dma_alloc_coherent(dev->hw->device,
3635 					       iwqp->ietf_mem.size,
3636 					       &iwqp->ietf_mem.pa, GFP_KERNEL);
3637 	if (!iwqp->ietf_mem.va) {
3638 		ret = -ENOMEM;
3639 		goto error;
3640 	}
3641 
3642 	cm_node->pdata.size = conn_param->private_data_len;
3643 	accept.addr = iwqp->ietf_mem.va;
3644 	accept.size = irdma_cm_build_mpa_frame(cm_node, &accept, MPA_KEY_REPLY);
3645 	memcpy((u8 *)accept.addr + accept.size, conn_param->private_data,
3646 	       conn_param->private_data_len);
3647 
3648 	if (cm_node->dev->ws_add(iwqp->sc_qp.vsi, cm_node->user_pri)) {
3649 		ret = -ENOMEM;
3650 		goto error;
3651 	}
3652 	iwqp->sc_qp.user_pri = cm_node->user_pri;
3653 	irdma_qp_add_qos(&iwqp->sc_qp);
3654 	/* setup our first outgoing iWarp send WQE (the IETF frame response) */
3655 	iwpd = iwqp->iwpd;
3656 	tagged_offset = (uintptr_t)iwqp->ietf_mem.va;
3657 	ibmr = irdma_reg_phys_mr(&iwpd->ibpd, iwqp->ietf_mem.pa, buf_len,
3658 				 IB_ACCESS_LOCAL_WRITE, &tagged_offset);
3659 	if (IS_ERR(ibmr)) {
3660 		ret = -ENOMEM;
3661 		goto error;
3662 	}
3663 
3664 	ibmr->pd = &iwpd->ibpd;
3665 	ibmr->device = iwpd->ibpd.device;
3666 	iwqp->lsmm_mr = ibmr;
3667 	if (iwqp->page)
3668 		iwqp->sc_qp.qp_uk.sq_base = kmap_local_page(iwqp->page);
3669 
3670 	cm_node->lsmm_size = accept.size + conn_param->private_data_len;
3671 	irdma_sc_send_lsmm(&iwqp->sc_qp, iwqp->ietf_mem.va, cm_node->lsmm_size,
3672 			   ibmr->lkey);
3673 
3674 	if (iwqp->page)
3675 		kunmap_local(iwqp->sc_qp.qp_uk.sq_base);
3676 
3677 	iwqp->cm_id = cm_id;
3678 	cm_node->cm_id = cm_id;
3679 
3680 	cm_id->provider_data = iwqp;
3681 	iwqp->active_conn = 0;
3682 	iwqp->cm_node = cm_node;
3683 	cm_node->iwqp = iwqp;
3684 	irdma_cm_init_tsa_conn(iwqp, cm_node);
3685 	irdma_qp_add_ref(&iwqp->ibqp);
3686 	cm_id->add_ref(cm_id);
3687 
3688 	attr.qp_state = IB_QPS_RTS;
3689 	cm_node->qhash_set = false;
3690 	cm_node->cm_core->cm_free_ah(cm_node);
3691 
3692 	irdma_modify_qp(&iwqp->ibqp, &attr, IB_QP_STATE, NULL);
3693 	if (dev->hw_attrs.uk_attrs.feature_flags & IRDMA_FEATURE_RTS_AE) {
3694 		wait_ret = wait_event_interruptible_timeout(iwqp->waitq,
3695 							    iwqp->rts_ae_rcvd,
3696 							    IRDMA_MAX_TIMEOUT);
3697 		if (!wait_ret) {
3698 			ibdev_dbg(&iwdev->ibdev,
3699 				  "CM: Slow Connection: cm_node=%p, loc_port=%d, rem_port=%d, cm_id=%p\n",
3700 				  cm_node, cm_node->loc_port,
3701 				  cm_node->rem_port, cm_node->cm_id);
3702 			ret = -ECONNRESET;
3703 			goto error;
3704 		}
3705 	}
3706 
3707 	irdma_send_cm_event(cm_node, cm_id, IW_CM_EVENT_ESTABLISHED, 0);
3708 	cm_node->accelerated = true;
3709 	complete(&cm_node->establish_comp);
3710 
3711 	if (cm_node->accept_pend) {
3712 		atomic_dec(&cm_node->listener->pend_accepts_cnt);
3713 		cm_node->accept_pend = 0;
3714 	}
3715 
3716 	ibdev_dbg(&iwdev->ibdev,
3717 		  "CM: rem_port=0x%04x, loc_port=0x%04x rem_addr=%pI4 loc_addr=%pI4 cm_node=%p cm_id=%p qp_id = %d\n\n",
3718 		  cm_node->rem_port, cm_node->loc_port, cm_node->rem_addr,
3719 		  cm_node->loc_addr, cm_node, cm_id, ibqp->qp_num);
3720 	cm_node->cm_core->stats_accepts++;
3721 
3722 	return 0;
3723 error:
3724 	irdma_free_lsmm_rsrc(iwqp);
3725 	irdma_rem_ref_cm_node(cm_node);
3726 
3727 	return ret;
3728 }
3729 
3730 /**
3731  * irdma_reject - registered call for connection to be rejected
3732  * @cm_id: cm information for passive connection
3733  * @pdata: private data to be sent
3734  * @pdata_len: private data length
3735  */
3736 int irdma_reject(struct iw_cm_id *cm_id, const void *pdata, u8 pdata_len)
3737 {
3738 	struct irdma_device *iwdev;
3739 	struct irdma_cm_node *cm_node;
3740 
3741 	cm_node = cm_id->provider_data;
3742 	cm_node->pdata.size = pdata_len;
3743 
3744 	trace_irdma_reject(cm_node, 0, NULL);
3745 
3746 	iwdev = to_iwdev(cm_id->device);
3747 	if (!iwdev)
3748 		return -EINVAL;
3749 
3750 	cm_node->cm_core->stats_rejects++;
3751 
3752 	if (pdata_len + sizeof(struct ietf_mpa_v2) > IRDMA_MAX_CM_BUF)
3753 		return -EINVAL;
3754 
3755 	return irdma_cm_reject(cm_node, pdata, pdata_len);
3756 }
3757 
3758 /**
3759  * irdma_connect - registered call for connection to be established
3760  * @cm_id: cm information for passive connection
3761  * @conn_param: Information about the connection
3762  */
3763 int irdma_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
3764 {
3765 	struct ib_qp *ibqp;
3766 	struct irdma_qp *iwqp;
3767 	struct irdma_device *iwdev;
3768 	struct irdma_cm_node *cm_node;
3769 	struct irdma_cm_info cm_info;
3770 	struct sockaddr_in *laddr;
3771 	struct sockaddr_in *raddr;
3772 	struct sockaddr_in6 *laddr6;
3773 	struct sockaddr_in6 *raddr6;
3774 	int ret = 0;
3775 
3776 	ibqp = irdma_get_qp(cm_id->device, conn_param->qpn);
3777 	if (!ibqp)
3778 		return -EINVAL;
3779 	iwqp = to_iwqp(ibqp);
3780 	if (!iwqp)
3781 		return -EINVAL;
3782 	iwdev = iwqp->iwdev;
3783 	if (!iwdev)
3784 		return -EINVAL;
3785 
3786 	laddr = (struct sockaddr_in *)&cm_id->m_local_addr;
3787 	raddr = (struct sockaddr_in *)&cm_id->m_remote_addr;
3788 	laddr6 = (struct sockaddr_in6 *)&cm_id->m_local_addr;
3789 	raddr6 = (struct sockaddr_in6 *)&cm_id->m_remote_addr;
3790 
3791 	if (!(laddr->sin_port) || !(raddr->sin_port))
3792 		return -EINVAL;
3793 
3794 	iwqp->active_conn = 1;
3795 	iwqp->cm_id = NULL;
3796 	cm_id->provider_data = iwqp;
3797 
3798 	/* set up the connection params for the node */
3799 	if (cm_id->remote_addr.ss_family == AF_INET) {
3800 		if (iwdev->vsi.mtu < IRDMA_MIN_MTU_IPV4)
3801 			return -EINVAL;
3802 
3803 		cm_info.ipv4 = true;
3804 		memset(cm_info.loc_addr, 0, sizeof(cm_info.loc_addr));
3805 		memset(cm_info.rem_addr, 0, sizeof(cm_info.rem_addr));
3806 		cm_info.loc_addr[0] = ntohl(laddr->sin_addr.s_addr);
3807 		cm_info.rem_addr[0] = ntohl(raddr->sin_addr.s_addr);
3808 		cm_info.loc_port = ntohs(laddr->sin_port);
3809 		cm_info.rem_port = ntohs(raddr->sin_port);
3810 		cm_info.vlan_id = irdma_get_vlan_ipv4(cm_info.loc_addr);
3811 	} else {
3812 		if (iwdev->vsi.mtu < IRDMA_MIN_MTU_IPV6)
3813 			return -EINVAL;
3814 
3815 		cm_info.ipv4 = false;
3816 		irdma_copy_ip_ntohl(cm_info.loc_addr,
3817 				    laddr6->sin6_addr.in6_u.u6_addr32);
3818 		irdma_copy_ip_ntohl(cm_info.rem_addr,
3819 				    raddr6->sin6_addr.in6_u.u6_addr32);
3820 		cm_info.loc_port = ntohs(laddr6->sin6_port);
3821 		cm_info.rem_port = ntohs(raddr6->sin6_port);
3822 		irdma_netdev_vlan_ipv6(cm_info.loc_addr, &cm_info.vlan_id,
3823 				       NULL);
3824 	}
3825 	cm_info.cm_id = cm_id;
3826 	cm_info.qh_qpid = iwdev->vsi.ilq->qp_id;
3827 	cm_info.tos = cm_id->tos;
3828 	if (iwdev->vsi.dscp_mode)
3829 		cm_info.user_pri =
3830 			iwqp->sc_qp.vsi->dscp_map[irdma_tos2dscp(cm_info.tos)];
3831 	else
3832 		cm_info.user_pri = rt_tos2priority(cm_id->tos);
3833 
3834 	if (iwqp->sc_qp.dev->ws_add(iwqp->sc_qp.vsi, cm_info.user_pri))
3835 		return -ENOMEM;
3836 	iwqp->sc_qp.user_pri = cm_info.user_pri;
3837 	irdma_qp_add_qos(&iwqp->sc_qp);
3838 	ibdev_dbg(&iwdev->ibdev, "DCB: TOS:[%d] UP:[%d]\n", cm_id->tos,
3839 		  cm_info.user_pri);
3840 
3841 	trace_irdma_dcb_tos(iwdev, cm_id->tos, cm_info.user_pri);
3842 
3843 	ret = irdma_create_cm_node(&iwdev->cm_core, iwdev, conn_param, &cm_info,
3844 				   &cm_node);
3845 	if (ret)
3846 		return ret;
3847 	ret = cm_node->cm_core->cm_create_ah(cm_node, true);
3848 	if (ret)
3849 		goto err;
3850 	if (irdma_manage_qhash(iwdev, &cm_info,
3851 			       IRDMA_QHASH_TYPE_TCP_ESTABLISHED,
3852 			       IRDMA_QHASH_MANAGE_TYPE_ADD, NULL, true)) {
3853 		ret = -EINVAL;
3854 		goto err;
3855 	}
3856 	cm_node->qhash_set = true;
3857 
3858 	cm_node->apbvt_entry = irdma_add_apbvt(iwdev, cm_info.loc_port);
3859 	if (!cm_node->apbvt_entry) {
3860 		ret = -EINVAL;
3861 		goto err;
3862 	}
3863 
3864 	cm_node->apbvt_set = true;
3865 	iwqp->cm_node = cm_node;
3866 	cm_node->iwqp = iwqp;
3867 	iwqp->cm_id = cm_id;
3868 	irdma_qp_add_ref(&iwqp->ibqp);
3869 	cm_id->add_ref(cm_id);
3870 
3871 	if (cm_node->state != IRDMA_CM_STATE_OFFLOADED) {
3872 		cm_node->state = IRDMA_CM_STATE_SYN_SENT;
3873 		ret = irdma_send_syn(cm_node, 0);
3874 		if (ret)
3875 			goto err;
3876 	}
3877 
3878 	ibdev_dbg(&iwdev->ibdev,
3879 		  "CM: rem_port=0x%04x, loc_port=0x%04x rem_addr=%pI4 loc_addr=%pI4 cm_node=%p cm_id=%p qp_id = %d\n\n",
3880 		  cm_node->rem_port, cm_node->loc_port, cm_node->rem_addr,
3881 		  cm_node->loc_addr, cm_node, cm_id, ibqp->qp_num);
3882 
3883 	trace_irdma_connect(cm_node, 0, NULL);
3884 
3885 	return 0;
3886 
3887 err:
3888 	if (cm_info.ipv4)
3889 		ibdev_dbg(&iwdev->ibdev,
3890 			  "CM: connect() FAILED: dest addr=%pI4",
3891 			  cm_info.rem_addr);
3892 	else
3893 		ibdev_dbg(&iwdev->ibdev,
3894 			  "CM: connect() FAILED: dest addr=%pI6",
3895 			  cm_info.rem_addr);
3896 	irdma_rem_ref_cm_node(cm_node);
3897 	iwdev->cm_core.stats_connect_errs++;
3898 
3899 	return ret;
3900 }
3901 
3902 /**
3903  * irdma_create_listen - registered call creating listener
3904  * @cm_id: cm information for passive connection
3905  * @backlog: to max accept pending count
3906  */
3907 int irdma_create_listen(struct iw_cm_id *cm_id, int backlog)
3908 {
3909 	struct irdma_device *iwdev;
3910 	struct irdma_cm_listener *cm_listen_node;
3911 	struct irdma_cm_info cm_info = {};
3912 	struct sockaddr_in *laddr;
3913 	struct sockaddr_in6 *laddr6;
3914 	bool wildcard = false;
3915 	int err;
3916 
3917 	iwdev = to_iwdev(cm_id->device);
3918 	if (!iwdev)
3919 		return -EINVAL;
3920 
3921 	laddr = (struct sockaddr_in *)&cm_id->m_local_addr;
3922 	laddr6 = (struct sockaddr_in6 *)&cm_id->m_local_addr;
3923 	cm_info.qh_qpid = iwdev->vsi.ilq->qp_id;
3924 
3925 	if (laddr->sin_family == AF_INET) {
3926 		if (iwdev->vsi.mtu < IRDMA_MIN_MTU_IPV4)
3927 			return -EINVAL;
3928 
3929 		cm_info.ipv4 = true;
3930 		cm_info.loc_addr[0] = ntohl(laddr->sin_addr.s_addr);
3931 		cm_info.loc_port = ntohs(laddr->sin_port);
3932 
3933 		if (laddr->sin_addr.s_addr != htonl(INADDR_ANY)) {
3934 			cm_info.vlan_id = irdma_get_vlan_ipv4(cm_info.loc_addr);
3935 		} else {
3936 			cm_info.vlan_id = 0xFFFF;
3937 			wildcard = true;
3938 		}
3939 	} else {
3940 		if (iwdev->vsi.mtu < IRDMA_MIN_MTU_IPV6)
3941 			return -EINVAL;
3942 
3943 		cm_info.ipv4 = false;
3944 		irdma_copy_ip_ntohl(cm_info.loc_addr,
3945 				    laddr6->sin6_addr.in6_u.u6_addr32);
3946 		cm_info.loc_port = ntohs(laddr6->sin6_port);
3947 		if (ipv6_addr_type(&laddr6->sin6_addr) != IPV6_ADDR_ANY) {
3948 			irdma_netdev_vlan_ipv6(cm_info.loc_addr,
3949 					       &cm_info.vlan_id, NULL);
3950 		} else {
3951 			cm_info.vlan_id = 0xFFFF;
3952 			wildcard = true;
3953 		}
3954 	}
3955 
3956 	if (cm_info.vlan_id >= VLAN_N_VID && iwdev->dcb_vlan_mode)
3957 		cm_info.vlan_id = 0;
3958 	cm_info.backlog = backlog;
3959 	cm_info.cm_id = cm_id;
3960 
3961 	trace_irdma_create_listen(iwdev, &cm_info);
3962 
3963 	cm_listen_node = irdma_make_listen_node(&iwdev->cm_core, iwdev,
3964 						&cm_info);
3965 	if (!cm_listen_node) {
3966 		ibdev_dbg(&iwdev->ibdev,
3967 			  "CM: cm_listen_node == NULL\n");
3968 		return -ENOMEM;
3969 	}
3970 
3971 	cm_id->provider_data = cm_listen_node;
3972 
3973 	cm_listen_node->tos = cm_id->tos;
3974 	if (iwdev->vsi.dscp_mode)
3975 		cm_listen_node->user_pri =
3976 			iwdev->vsi.dscp_map[irdma_tos2dscp(cm_id->tos)];
3977 	else
3978 		cm_listen_node->user_pri = rt_tos2priority(cm_id->tos);
3979 	cm_info.user_pri = cm_listen_node->user_pri;
3980 	if (!cm_listen_node->reused_node) {
3981 		if (wildcard) {
3982 			err = irdma_add_mqh(iwdev, &cm_info, cm_listen_node);
3983 			if (err)
3984 				goto error;
3985 		} else {
3986 			err = irdma_manage_qhash(iwdev, &cm_info,
3987 						 IRDMA_QHASH_TYPE_TCP_SYN,
3988 						 IRDMA_QHASH_MANAGE_TYPE_ADD,
3989 						 NULL, true);
3990 			if (err)
3991 				goto error;
3992 
3993 			cm_listen_node->qhash_set = true;
3994 		}
3995 
3996 		cm_listen_node->apbvt_entry = irdma_add_apbvt(iwdev,
3997 							      cm_info.loc_port);
3998 		if (!cm_listen_node->apbvt_entry)
3999 			goto error;
4000 	}
4001 	cm_id->add_ref(cm_id);
4002 	cm_listen_node->cm_core->stats_listen_created++;
4003 	ibdev_dbg(&iwdev->ibdev,
4004 		  "CM: loc_port=0x%04x loc_addr=%pI4 cm_listen_node=%p cm_id=%p qhash_set=%d vlan_id=%d\n",
4005 		  cm_listen_node->loc_port, cm_listen_node->loc_addr,
4006 		  cm_listen_node, cm_listen_node->cm_id,
4007 		  cm_listen_node->qhash_set, cm_listen_node->vlan_id);
4008 
4009 	return 0;
4010 
4011 error:
4012 
4013 	irdma_cm_del_listen(&iwdev->cm_core, cm_listen_node, false);
4014 
4015 	return -EINVAL;
4016 }
4017 
4018 /**
4019  * irdma_destroy_listen - registered call to destroy listener
4020  * @cm_id: cm information for passive connection
4021  */
4022 int irdma_destroy_listen(struct iw_cm_id *cm_id)
4023 {
4024 	struct irdma_device *iwdev;
4025 
4026 	iwdev = to_iwdev(cm_id->device);
4027 	if (cm_id->provider_data)
4028 		irdma_cm_del_listen(&iwdev->cm_core, cm_id->provider_data,
4029 				    true);
4030 	else
4031 		ibdev_dbg(&iwdev->ibdev,
4032 			  "CM: cm_id->provider_data was NULL\n");
4033 
4034 	cm_id->rem_ref(cm_id);
4035 
4036 	return 0;
4037 }
4038 
4039 /**
4040  * irdma_teardown_list_prep - add conn nodes slated for tear down to list
4041  * @cm_core: cm's core
4042  * @teardown_list: a list to which cm_node will be selected
4043  * @ipaddr: pointer to ip address
4044  * @nfo: pointer to cm_info structure instance
4045  * @disconnect_all: flag indicating disconnect all QPs
4046  */
4047 static void irdma_teardown_list_prep(struct irdma_cm_core *cm_core,
4048 				     struct list_head *teardown_list,
4049 				     u32 *ipaddr,
4050 				     struct irdma_cm_info *nfo,
4051 				     bool disconnect_all)
4052 {
4053 	struct irdma_cm_node *cm_node;
4054 	int bkt;
4055 
4056 	hash_for_each_rcu(cm_core->cm_hash_tbl, bkt, cm_node, list) {
4057 		if ((disconnect_all ||
4058 		     (nfo->vlan_id == cm_node->vlan_id &&
4059 		      !memcmp(cm_node->loc_addr, ipaddr, nfo->ipv4 ? 4 : 16))) &&
4060 		    refcount_inc_not_zero(&cm_node->refcnt))
4061 			list_add(&cm_node->teardown_entry, teardown_list);
4062 	}
4063 }
4064 
4065 /**
4066  * irdma_cm_event_connected - handle connected active node
4067  * @event: the info for cm_node of connection
4068  */
4069 static void irdma_cm_event_connected(struct irdma_cm_event *event)
4070 {
4071 	struct irdma_qp *iwqp;
4072 	struct irdma_device *iwdev;
4073 	struct irdma_cm_node *cm_node;
4074 	struct irdma_sc_dev *dev;
4075 	struct ib_qp_attr attr = {};
4076 	struct iw_cm_id *cm_id;
4077 	int status;
4078 	bool read0;
4079 	int wait_ret = 0;
4080 
4081 	cm_node = event->cm_node;
4082 	cm_id = cm_node->cm_id;
4083 	iwqp = cm_id->provider_data;
4084 	iwdev = iwqp->iwdev;
4085 	dev = &iwdev->rf->sc_dev;
4086 	if (iwqp->sc_qp.qp_uk.destroy_pending) {
4087 		status = -ETIMEDOUT;
4088 		goto error;
4089 	}
4090 
4091 	irdma_cm_init_tsa_conn(iwqp, cm_node);
4092 	read0 = (cm_node->send_rdma0_op == SEND_RDMA_READ_ZERO);
4093 	if (iwqp->page)
4094 		iwqp->sc_qp.qp_uk.sq_base = kmap_local_page(iwqp->page);
4095 	irdma_sc_send_rtt(&iwqp->sc_qp, read0);
4096 	if (iwqp->page)
4097 		kunmap_local(iwqp->sc_qp.qp_uk.sq_base);
4098 
4099 	attr.qp_state = IB_QPS_RTS;
4100 	cm_node->qhash_set = false;
4101 	irdma_modify_qp(&iwqp->ibqp, &attr, IB_QP_STATE, NULL);
4102 	if (dev->hw_attrs.uk_attrs.feature_flags & IRDMA_FEATURE_RTS_AE) {
4103 		wait_ret = wait_event_interruptible_timeout(iwqp->waitq,
4104 							    iwqp->rts_ae_rcvd,
4105 							    IRDMA_MAX_TIMEOUT);
4106 		if (!wait_ret)
4107 			ibdev_dbg(&iwdev->ibdev,
4108 				  "CM: Slow Connection: cm_node=%p, loc_port=%d, rem_port=%d, cm_id=%p\n",
4109 				  cm_node, cm_node->loc_port,
4110 				  cm_node->rem_port, cm_node->cm_id);
4111 	}
4112 
4113 	irdma_send_cm_event(cm_node, cm_id, IW_CM_EVENT_CONNECT_REPLY, 0);
4114 	cm_node->accelerated = true;
4115 	complete(&cm_node->establish_comp);
4116 	cm_node->cm_core->cm_free_ah(cm_node);
4117 	return;
4118 
4119 error:
4120 	iwqp->cm_id = NULL;
4121 	cm_id->provider_data = NULL;
4122 	irdma_send_cm_event(event->cm_node, cm_id, IW_CM_EVENT_CONNECT_REPLY,
4123 			    status);
4124 	irdma_rem_ref_cm_node(event->cm_node);
4125 }
4126 
4127 /**
4128  * irdma_cm_event_reset - handle reset
4129  * @event: the info for cm_node of connection
4130  */
4131 static void irdma_cm_event_reset(struct irdma_cm_event *event)
4132 {
4133 	struct irdma_cm_node *cm_node = event->cm_node;
4134 	struct iw_cm_id *cm_id = cm_node->cm_id;
4135 	struct irdma_qp *iwqp;
4136 
4137 	if (!cm_id)
4138 		return;
4139 
4140 	iwqp = cm_id->provider_data;
4141 	if (!iwqp)
4142 		return;
4143 
4144 	ibdev_dbg(&cm_node->iwdev->ibdev,
4145 		  "CM: reset event %p - cm_id = %p\n", event->cm_node, cm_id);
4146 	iwqp->cm_id = NULL;
4147 
4148 	irdma_send_cm_event(cm_node, cm_node->cm_id, IW_CM_EVENT_DISCONNECT,
4149 			    -ECONNRESET);
4150 	irdma_send_cm_event(cm_node, cm_node->cm_id, IW_CM_EVENT_CLOSE, 0);
4151 }
4152 
4153 /**
4154  * irdma_cm_event_handler - send event to cm upper layer
4155  * @work: pointer of cm event info.
4156  */
4157 static void irdma_cm_event_handler(struct work_struct *work)
4158 {
4159 	struct irdma_cm_event *event = container_of(work, struct irdma_cm_event, event_work);
4160 	struct irdma_cm_node *cm_node;
4161 
4162 	if (!event || !event->cm_node || !event->cm_node->cm_core)
4163 		return;
4164 
4165 	cm_node = event->cm_node;
4166 	trace_irdma_cm_event_handler(cm_node, event->type, NULL);
4167 
4168 	switch (event->type) {
4169 	case IRDMA_CM_EVENT_MPA_REQ:
4170 		irdma_send_cm_event(cm_node, cm_node->cm_id,
4171 				    IW_CM_EVENT_CONNECT_REQUEST, 0);
4172 		break;
4173 	case IRDMA_CM_EVENT_RESET:
4174 		irdma_cm_event_reset(event);
4175 		break;
4176 	case IRDMA_CM_EVENT_CONNECTED:
4177 		if (!event->cm_node->cm_id ||
4178 		    event->cm_node->state != IRDMA_CM_STATE_OFFLOADED)
4179 			break;
4180 		irdma_cm_event_connected(event);
4181 		break;
4182 	case IRDMA_CM_EVENT_MPA_REJECT:
4183 		if (!event->cm_node->cm_id ||
4184 		    cm_node->state == IRDMA_CM_STATE_OFFLOADED)
4185 			break;
4186 		irdma_send_cm_event(cm_node, cm_node->cm_id,
4187 				    IW_CM_EVENT_CONNECT_REPLY, -ECONNREFUSED);
4188 		break;
4189 	case IRDMA_CM_EVENT_ABORTED:
4190 		if (!event->cm_node->cm_id ||
4191 		    event->cm_node->state == IRDMA_CM_STATE_OFFLOADED)
4192 			break;
4193 		irdma_event_connect_error(event);
4194 		break;
4195 	default:
4196 		ibdev_dbg(&cm_node->iwdev->ibdev,
4197 			  "CM: bad event type = %d\n", event->type);
4198 		break;
4199 	}
4200 
4201 	irdma_rem_ref_cm_node(event->cm_node);
4202 	kfree(event);
4203 }
4204 
4205 /**
4206  * irdma_cm_post_event - queue event request for worker thread
4207  * @event: cm node's info for up event call
4208  */
4209 static void irdma_cm_post_event(struct irdma_cm_event *event)
4210 {
4211 	refcount_inc(&event->cm_node->refcnt);
4212 	INIT_WORK(&event->event_work, irdma_cm_event_handler);
4213 	queue_work(event->cm_node->cm_core->event_wq, &event->event_work);
4214 }
4215 
4216 /**
4217  * irdma_cm_teardown_connections - teardown QPs
4218  * @iwdev: device pointer
4219  * @ipaddr: Pointer to IPv4 or IPv6 address
4220  * @nfo: Connection info
4221  * @disconnect_all: flag indicating disconnect all QPs
4222  *
4223  * teardown QPs where source or destination addr matches ip addr
4224  */
4225 void irdma_cm_teardown_connections(struct irdma_device *iwdev, u32 *ipaddr,
4226 				   struct irdma_cm_info *nfo,
4227 				   bool disconnect_all)
4228 {
4229 	struct irdma_cm_core *cm_core = &iwdev->cm_core;
4230 	struct list_head *list_core_temp;
4231 	struct list_head *list_node;
4232 	struct irdma_cm_node *cm_node;
4233 	struct list_head teardown_list;
4234 	struct ib_qp_attr attr;
4235 
4236 	INIT_LIST_HEAD(&teardown_list);
4237 
4238 	rcu_read_lock();
4239 	irdma_teardown_list_prep(cm_core, &teardown_list, ipaddr, nfo, disconnect_all);
4240 	rcu_read_unlock();
4241 
4242 	list_for_each_safe (list_node, list_core_temp, &teardown_list) {
4243 		cm_node = container_of(list_node, struct irdma_cm_node,
4244 				       teardown_entry);
4245 		attr.qp_state = IB_QPS_ERR;
4246 		irdma_modify_qp(&cm_node->iwqp->ibqp, &attr, IB_QP_STATE, NULL);
4247 		if (iwdev->rf->reset)
4248 			irdma_cm_disconn(cm_node->iwqp);
4249 		irdma_rem_ref_cm_node(cm_node);
4250 	}
4251 }
4252 
4253 /**
4254  * irdma_qhash_ctrl - enable/disable qhash for list
4255  * @iwdev: device pointer
4256  * @parent_listen_node: parent listen node
4257  * @nfo: cm info node
4258  * @ipaddr: Pointer to IPv4 or IPv6 address
4259  * @ipv4: flag indicating IPv4 when true
4260  * @ifup: flag indicating interface up when true
4261  *
4262  * Enables or disables the qhash for the node in the child
4263  * listen list that matches ipaddr. If no matching IP was found
4264  * it will allocate and add a new child listen node to the
4265  * parent listen node. The listen_list_lock is assumed to be
4266  * held when called.
4267  */
4268 static void irdma_qhash_ctrl(struct irdma_device *iwdev,
4269 			     struct irdma_cm_listener *parent_listen_node,
4270 			     struct irdma_cm_info *nfo, u32 *ipaddr, bool ipv4,
4271 			     bool ifup)
4272 {
4273 	struct list_head *child_listen_list = &parent_listen_node->child_listen_list;
4274 	struct irdma_cm_listener *child_listen_node;
4275 	struct list_head *pos, *tpos;
4276 	bool node_allocated = false;
4277 	enum irdma_quad_hash_manage_type op = ifup ?
4278 					      IRDMA_QHASH_MANAGE_TYPE_ADD :
4279 					      IRDMA_QHASH_MANAGE_TYPE_DELETE;
4280 	int err;
4281 
4282 	list_for_each_safe (pos, tpos, child_listen_list) {
4283 		child_listen_node = list_entry(pos, struct irdma_cm_listener,
4284 					       child_listen_list);
4285 		if (!memcmp(child_listen_node->loc_addr, ipaddr, ipv4 ? 4 : 16))
4286 			goto set_qhash;
4287 	}
4288 
4289 	/* if not found then add a child listener if interface is going up */
4290 	if (!ifup)
4291 		return;
4292 	child_listen_node = kmemdup(parent_listen_node,
4293 				    sizeof(*child_listen_node), GFP_ATOMIC);
4294 	if (!child_listen_node)
4295 		return;
4296 
4297 	node_allocated = true;
4298 	memcpy(child_listen_node->loc_addr, ipaddr, ipv4 ? 4 : 16);
4299 
4300 set_qhash:
4301 	memcpy(nfo->loc_addr, child_listen_node->loc_addr,
4302 	       sizeof(nfo->loc_addr));
4303 	nfo->vlan_id = child_listen_node->vlan_id;
4304 	err = irdma_manage_qhash(iwdev, nfo, IRDMA_QHASH_TYPE_TCP_SYN, op, NULL,
4305 				 false);
4306 	if (!err) {
4307 		child_listen_node->qhash_set = ifup;
4308 		if (node_allocated)
4309 			list_add(&child_listen_node->child_listen_list,
4310 				 &parent_listen_node->child_listen_list);
4311 	} else if (node_allocated) {
4312 		kfree(child_listen_node);
4313 	}
4314 }
4315 
4316 /**
4317  * irdma_if_notify - process an ifdown on an interface
4318  * @iwdev: device pointer
4319  * @netdev: network device structure
4320  * @ipaddr: Pointer to IPv4 or IPv6 address
4321  * @ipv4: flag indicating IPv4 when true
4322  * @ifup: flag indicating interface up when true
4323  */
4324 void irdma_if_notify(struct irdma_device *iwdev, struct net_device *netdev,
4325 		     u32 *ipaddr, bool ipv4, bool ifup)
4326 {
4327 	struct irdma_cm_core *cm_core = &iwdev->cm_core;
4328 	unsigned long flags;
4329 	struct irdma_cm_listener *listen_node;
4330 	static const u32 ip_zero[4] = { 0, 0, 0, 0 };
4331 	struct irdma_cm_info nfo = {};
4332 	u16 vlan_id = rdma_vlan_dev_vlan_id(netdev);
4333 	enum irdma_quad_hash_manage_type op = ifup ?
4334 					      IRDMA_QHASH_MANAGE_TYPE_ADD :
4335 					      IRDMA_QHASH_MANAGE_TYPE_DELETE;
4336 
4337 	nfo.vlan_id = vlan_id;
4338 	nfo.ipv4 = ipv4;
4339 	nfo.qh_qpid = 1;
4340 
4341 	/* Disable or enable qhash for listeners */
4342 	spin_lock_irqsave(&cm_core->listen_list_lock, flags);
4343 	list_for_each_entry (listen_node, &cm_core->listen_list, list) {
4344 		if (vlan_id != listen_node->vlan_id ||
4345 		    (memcmp(listen_node->loc_addr, ipaddr, ipv4 ? 4 : 16) &&
4346 		     memcmp(listen_node->loc_addr, ip_zero, ipv4 ? 4 : 16)))
4347 			continue;
4348 
4349 		memcpy(nfo.loc_addr, listen_node->loc_addr,
4350 		       sizeof(nfo.loc_addr));
4351 		nfo.loc_port = listen_node->loc_port;
4352 		nfo.user_pri = listen_node->user_pri;
4353 		if (!list_empty(&listen_node->child_listen_list)) {
4354 			irdma_qhash_ctrl(iwdev, listen_node, &nfo, ipaddr, ipv4,
4355 					 ifup);
4356 		} else if (memcmp(listen_node->loc_addr, ip_zero,
4357 				  ipv4 ? 4 : 16)) {
4358 			if (!irdma_manage_qhash(iwdev, &nfo,
4359 						IRDMA_QHASH_TYPE_TCP_SYN, op,
4360 						NULL, false))
4361 				listen_node->qhash_set = ifup;
4362 		}
4363 	}
4364 	spin_unlock_irqrestore(&cm_core->listen_list_lock, flags);
4365 
4366 	/* disconnect any connected qp's on ifdown */
4367 	if (!ifup)
4368 		irdma_cm_teardown_connections(iwdev, ipaddr, &nfo, false);
4369 }
4370