xref: /linux/drivers/target/iscsi/cxgbit/cxgbit_cm.c (revision c532de5a67a70f8533d495f8f2aaa9a0491c3ad0)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (c) 2016 Chelsio Communications, Inc.
4  */
5 
6 #include <linux/module.h>
7 #include <linux/list.h>
8 #include <linux/workqueue.h>
9 #include <linux/skbuff.h>
10 #include <linux/timer.h>
11 #include <linux/notifier.h>
12 #include <linux/inetdevice.h>
13 #include <linux/ip.h>
14 #include <linux/tcp.h>
15 #include <linux/if_vlan.h>
16 
17 #include <net/neighbour.h>
18 #include <net/netevent.h>
19 #include <net/route.h>
20 #include <net/tcp.h>
21 #include <net/ip6_route.h>
22 #include <net/addrconf.h>
23 
24 #include <libcxgb_cm.h>
25 #include "cxgbit.h"
26 #include "clip_tbl.h"
27 
28 static void cxgbit_init_wr_wait(struct cxgbit_wr_wait *wr_waitp)
29 {
30 	wr_waitp->ret = 0;
31 	reinit_completion(&wr_waitp->completion);
32 }
33 
34 static void
35 cxgbit_wake_up(struct cxgbit_wr_wait *wr_waitp, const char *func, u8 ret)
36 {
37 	if (ret == CPL_ERR_NONE)
38 		wr_waitp->ret = 0;
39 	else
40 		wr_waitp->ret = -EIO;
41 
42 	if (wr_waitp->ret)
43 		pr_err("%s: err:%u", func, ret);
44 
45 	complete(&wr_waitp->completion);
46 }
47 
48 static int
49 cxgbit_wait_for_reply(struct cxgbit_device *cdev,
50 		      struct cxgbit_wr_wait *wr_waitp, u32 tid, u32 timeout,
51 		      const char *func)
52 {
53 	int ret;
54 
55 	if (!test_bit(CDEV_STATE_UP, &cdev->flags)) {
56 		wr_waitp->ret = -EIO;
57 		goto out;
58 	}
59 
60 	ret = wait_for_completion_timeout(&wr_waitp->completion, timeout * HZ);
61 	if (!ret) {
62 		pr_info("%s - Device %s not responding tid %u\n",
63 			func, pci_name(cdev->lldi.pdev), tid);
64 		wr_waitp->ret = -ETIMEDOUT;
65 	}
66 out:
67 	if (wr_waitp->ret)
68 		pr_info("%s: FW reply %d tid %u\n",
69 			pci_name(cdev->lldi.pdev), wr_waitp->ret, tid);
70 	return wr_waitp->ret;
71 }
72 
73 static int cxgbit_np_hashfn(const struct cxgbit_np *cnp)
74 {
75 	return ((unsigned long)cnp >> 10) & (NP_INFO_HASH_SIZE - 1);
76 }
77 
78 static struct np_info *
79 cxgbit_np_hash_add(struct cxgbit_device *cdev, struct cxgbit_np *cnp,
80 		   unsigned int stid)
81 {
82 	struct np_info *p = kzalloc(sizeof(*p), GFP_KERNEL);
83 
84 	if (p) {
85 		int bucket = cxgbit_np_hashfn(cnp);
86 
87 		p->cnp = cnp;
88 		p->stid = stid;
89 		spin_lock(&cdev->np_lock);
90 		p->next = cdev->np_hash_tab[bucket];
91 		cdev->np_hash_tab[bucket] = p;
92 		spin_unlock(&cdev->np_lock);
93 	}
94 
95 	return p;
96 }
97 
98 static int
99 cxgbit_np_hash_find(struct cxgbit_device *cdev, struct cxgbit_np *cnp)
100 {
101 	int stid = -1, bucket = cxgbit_np_hashfn(cnp);
102 	struct np_info *p;
103 
104 	spin_lock(&cdev->np_lock);
105 	for (p = cdev->np_hash_tab[bucket]; p; p = p->next) {
106 		if (p->cnp == cnp) {
107 			stid = p->stid;
108 			break;
109 		}
110 	}
111 	spin_unlock(&cdev->np_lock);
112 
113 	return stid;
114 }
115 
116 static int cxgbit_np_hash_del(struct cxgbit_device *cdev, struct cxgbit_np *cnp)
117 {
118 	int stid = -1, bucket = cxgbit_np_hashfn(cnp);
119 	struct np_info *p, **prev = &cdev->np_hash_tab[bucket];
120 
121 	spin_lock(&cdev->np_lock);
122 	for (p = *prev; p; prev = &p->next, p = p->next) {
123 		if (p->cnp == cnp) {
124 			stid = p->stid;
125 			*prev = p->next;
126 			kfree(p);
127 			break;
128 		}
129 	}
130 	spin_unlock(&cdev->np_lock);
131 
132 	return stid;
133 }
134 
135 void _cxgbit_free_cnp(struct kref *kref)
136 {
137 	struct cxgbit_np *cnp;
138 
139 	cnp = container_of(kref, struct cxgbit_np, kref);
140 	kfree(cnp);
141 }
142 
143 static int
144 cxgbit_create_server6(struct cxgbit_device *cdev, unsigned int stid,
145 		      struct cxgbit_np *cnp)
146 {
147 	struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)
148 				     &cnp->com.local_addr;
149 	int addr_type;
150 	int ret;
151 
152 	pr_debug("%s: dev = %s; stid = %u; sin6_port = %u\n",
153 		 __func__, cdev->lldi.ports[0]->name, stid, sin6->sin6_port);
154 
155 	addr_type = ipv6_addr_type((const struct in6_addr *)
156 				   &sin6->sin6_addr);
157 	if (addr_type != IPV6_ADDR_ANY) {
158 		ret = cxgb4_clip_get(cdev->lldi.ports[0],
159 				     (const u32 *)&sin6->sin6_addr.s6_addr, 1);
160 		if (ret) {
161 			pr_err("Unable to find clip table entry. laddr %pI6. Error:%d.\n",
162 			       sin6->sin6_addr.s6_addr, ret);
163 			return -ENOMEM;
164 		}
165 	}
166 
167 	cxgbit_get_cnp(cnp);
168 	cxgbit_init_wr_wait(&cnp->com.wr_wait);
169 
170 	ret = cxgb4_create_server6(cdev->lldi.ports[0],
171 				   stid, &sin6->sin6_addr,
172 				   sin6->sin6_port,
173 				   cdev->lldi.rxq_ids[0]);
174 	if (!ret)
175 		ret = cxgbit_wait_for_reply(cdev, &cnp->com.wr_wait,
176 					    0, 10, __func__);
177 	else if (ret > 0)
178 		ret = net_xmit_errno(ret);
179 	else
180 		cxgbit_put_cnp(cnp);
181 
182 	if (ret) {
183 		if (ret != -ETIMEDOUT)
184 			cxgb4_clip_release(cdev->lldi.ports[0],
185 				   (const u32 *)&sin6->sin6_addr.s6_addr, 1);
186 
187 		pr_err("create server6 err %d stid %d laddr %pI6 lport %d\n",
188 		       ret, stid, sin6->sin6_addr.s6_addr,
189 		       ntohs(sin6->sin6_port));
190 	}
191 
192 	return ret;
193 }
194 
195 static int
196 cxgbit_create_server4(struct cxgbit_device *cdev, unsigned int stid,
197 		      struct cxgbit_np *cnp)
198 {
199 	struct sockaddr_in *sin = (struct sockaddr_in *)
200 				   &cnp->com.local_addr;
201 	int ret;
202 
203 	pr_debug("%s: dev = %s; stid = %u; sin_port = %u\n",
204 		 __func__, cdev->lldi.ports[0]->name, stid, sin->sin_port);
205 
206 	cxgbit_get_cnp(cnp);
207 	cxgbit_init_wr_wait(&cnp->com.wr_wait);
208 
209 	ret = cxgb4_create_server(cdev->lldi.ports[0],
210 				  stid, sin->sin_addr.s_addr,
211 				  sin->sin_port, 0,
212 				  cdev->lldi.rxq_ids[0]);
213 	if (!ret)
214 		ret = cxgbit_wait_for_reply(cdev,
215 					    &cnp->com.wr_wait,
216 					    0, 10, __func__);
217 	else if (ret > 0)
218 		ret = net_xmit_errno(ret);
219 	else
220 		cxgbit_put_cnp(cnp);
221 
222 	if (ret)
223 		pr_err("create server failed err %d stid %d laddr %pI4 lport %d\n",
224 		       ret, stid, &sin->sin_addr, ntohs(sin->sin_port));
225 	return ret;
226 }
227 
228 struct cxgbit_device *cxgbit_find_device(struct net_device *ndev, u8 *port_id)
229 {
230 	struct cxgbit_device *cdev;
231 	u8 i;
232 
233 	list_for_each_entry(cdev, &cdev_list_head, list) {
234 		struct cxgb4_lld_info *lldi = &cdev->lldi;
235 
236 		for (i = 0; i < lldi->nports; i++) {
237 			if (lldi->ports[i] == ndev) {
238 				if (port_id)
239 					*port_id = i;
240 				return cdev;
241 			}
242 		}
243 	}
244 
245 	return NULL;
246 }
247 
248 static struct net_device *cxgbit_get_real_dev(struct net_device *ndev)
249 {
250 	if (ndev->priv_flags & IFF_BONDING) {
251 		pr_err("Bond devices are not supported. Interface:%s\n",
252 		       ndev->name);
253 		return NULL;
254 	}
255 
256 	if (is_vlan_dev(ndev))
257 		return vlan_dev_real_dev(ndev);
258 
259 	return ndev;
260 }
261 
262 static struct net_device *cxgbit_ipv4_netdev(__be32 saddr)
263 {
264 	struct net_device *ndev;
265 
266 	ndev = __ip_dev_find(&init_net, saddr, false);
267 	if (!ndev)
268 		return NULL;
269 
270 	return cxgbit_get_real_dev(ndev);
271 }
272 
273 static struct net_device *cxgbit_ipv6_netdev(struct in6_addr *addr6)
274 {
275 	struct net_device *ndev = NULL;
276 	bool found = false;
277 
278 	if (IS_ENABLED(CONFIG_IPV6)) {
279 		for_each_netdev_rcu(&init_net, ndev)
280 			if (ipv6_chk_addr(&init_net, addr6, ndev, 1)) {
281 				found = true;
282 				break;
283 			}
284 	}
285 	if (!found)
286 		return NULL;
287 	return cxgbit_get_real_dev(ndev);
288 }
289 
290 static struct cxgbit_device *cxgbit_find_np_cdev(struct cxgbit_np *cnp)
291 {
292 	struct sockaddr_storage *sockaddr = &cnp->com.local_addr;
293 	int ss_family = sockaddr->ss_family;
294 	struct net_device *ndev = NULL;
295 	struct cxgbit_device *cdev = NULL;
296 
297 	rcu_read_lock();
298 	if (ss_family == AF_INET) {
299 		struct sockaddr_in *sin;
300 
301 		sin = (struct sockaddr_in *)sockaddr;
302 		ndev = cxgbit_ipv4_netdev(sin->sin_addr.s_addr);
303 	} else if (ss_family == AF_INET6) {
304 		struct sockaddr_in6 *sin6;
305 
306 		sin6 = (struct sockaddr_in6 *)sockaddr;
307 		ndev = cxgbit_ipv6_netdev(&sin6->sin6_addr);
308 	}
309 	if (!ndev)
310 		goto out;
311 
312 	cdev = cxgbit_find_device(ndev, NULL);
313 out:
314 	rcu_read_unlock();
315 	return cdev;
316 }
317 
318 static bool cxgbit_inaddr_any(struct cxgbit_np *cnp)
319 {
320 	struct sockaddr_storage *sockaddr = &cnp->com.local_addr;
321 	int ss_family = sockaddr->ss_family;
322 	int addr_type;
323 
324 	if (ss_family == AF_INET) {
325 		struct sockaddr_in *sin;
326 
327 		sin = (struct sockaddr_in *)sockaddr;
328 		if (sin->sin_addr.s_addr == htonl(INADDR_ANY))
329 			return true;
330 	} else if (ss_family == AF_INET6) {
331 		struct sockaddr_in6 *sin6;
332 
333 		sin6 = (struct sockaddr_in6 *)sockaddr;
334 		addr_type = ipv6_addr_type((const struct in6_addr *)
335 				&sin6->sin6_addr);
336 		if (addr_type == IPV6_ADDR_ANY)
337 			return true;
338 	}
339 	return false;
340 }
341 
342 static int
343 __cxgbit_setup_cdev_np(struct cxgbit_device *cdev, struct cxgbit_np *cnp)
344 {
345 	int stid, ret;
346 	int ss_family = cnp->com.local_addr.ss_family;
347 
348 	if (!test_bit(CDEV_STATE_UP, &cdev->flags))
349 		return -EINVAL;
350 
351 	stid = cxgb4_alloc_stid(cdev->lldi.tids, ss_family, cnp);
352 	if (stid < 0)
353 		return -EINVAL;
354 
355 	if (!cxgbit_np_hash_add(cdev, cnp, stid)) {
356 		cxgb4_free_stid(cdev->lldi.tids, stid, ss_family);
357 		return -EINVAL;
358 	}
359 
360 	if (ss_family == AF_INET)
361 		ret = cxgbit_create_server4(cdev, stid, cnp);
362 	else
363 		ret = cxgbit_create_server6(cdev, stid, cnp);
364 
365 	if (ret) {
366 		if (ret != -ETIMEDOUT)
367 			cxgb4_free_stid(cdev->lldi.tids, stid,
368 					ss_family);
369 		cxgbit_np_hash_del(cdev, cnp);
370 		return ret;
371 	}
372 	return ret;
373 }
374 
375 static int cxgbit_setup_cdev_np(struct cxgbit_np *cnp)
376 {
377 	struct cxgbit_device *cdev;
378 	int ret = -1;
379 
380 	mutex_lock(&cdev_list_lock);
381 	cdev = cxgbit_find_np_cdev(cnp);
382 	if (!cdev)
383 		goto out;
384 
385 	if (cxgbit_np_hash_find(cdev, cnp) >= 0)
386 		goto out;
387 
388 	if (__cxgbit_setup_cdev_np(cdev, cnp))
389 		goto out;
390 
391 	cnp->com.cdev = cdev;
392 	ret = 0;
393 out:
394 	mutex_unlock(&cdev_list_lock);
395 	return ret;
396 }
397 
398 static int cxgbit_setup_all_np(struct cxgbit_np *cnp)
399 {
400 	struct cxgbit_device *cdev;
401 	int ret;
402 	u32 count = 0;
403 
404 	mutex_lock(&cdev_list_lock);
405 	list_for_each_entry(cdev, &cdev_list_head, list) {
406 		if (cxgbit_np_hash_find(cdev, cnp) >= 0) {
407 			mutex_unlock(&cdev_list_lock);
408 			return -1;
409 		}
410 	}
411 
412 	list_for_each_entry(cdev, &cdev_list_head, list) {
413 		ret = __cxgbit_setup_cdev_np(cdev, cnp);
414 		if (ret == -ETIMEDOUT)
415 			break;
416 		if (ret != 0)
417 			continue;
418 		count++;
419 	}
420 	mutex_unlock(&cdev_list_lock);
421 
422 	return count ? 0 : -1;
423 }
424 
425 int cxgbit_setup_np(struct iscsi_np *np, struct sockaddr_storage *ksockaddr)
426 {
427 	struct cxgbit_np *cnp;
428 	int ret;
429 
430 	if ((ksockaddr->ss_family != AF_INET) &&
431 	    (ksockaddr->ss_family != AF_INET6))
432 		return -EINVAL;
433 
434 	cnp = kzalloc(sizeof(*cnp), GFP_KERNEL);
435 	if (!cnp)
436 		return -ENOMEM;
437 
438 	init_waitqueue_head(&cnp->accept_wait);
439 	init_completion(&cnp->com.wr_wait.completion);
440 	init_completion(&cnp->accept_comp);
441 	INIT_LIST_HEAD(&cnp->np_accept_list);
442 	spin_lock_init(&cnp->np_accept_lock);
443 	kref_init(&cnp->kref);
444 	memcpy(&np->np_sockaddr, ksockaddr,
445 	       sizeof(struct sockaddr_storage));
446 	memcpy(&cnp->com.local_addr, &np->np_sockaddr,
447 	       sizeof(cnp->com.local_addr));
448 
449 	cnp->np = np;
450 	cnp->com.cdev = NULL;
451 
452 	if (cxgbit_inaddr_any(cnp))
453 		ret = cxgbit_setup_all_np(cnp);
454 	else
455 		ret = cxgbit_setup_cdev_np(cnp);
456 
457 	if (ret) {
458 		cxgbit_put_cnp(cnp);
459 		return -EINVAL;
460 	}
461 
462 	np->np_context = cnp;
463 	cnp->com.state = CSK_STATE_LISTEN;
464 	return 0;
465 }
466 
467 static void
468 cxgbit_set_conn_info(struct iscsi_np *np, struct iscsit_conn *conn,
469 		     struct cxgbit_sock *csk)
470 {
471 	conn->login_family = np->np_sockaddr.ss_family;
472 	conn->login_sockaddr = csk->com.remote_addr;
473 	conn->local_sockaddr = csk->com.local_addr;
474 }
475 
476 int cxgbit_accept_np(struct iscsi_np *np, struct iscsit_conn *conn)
477 {
478 	struct cxgbit_np *cnp = np->np_context;
479 	struct cxgbit_sock *csk;
480 	int ret = 0;
481 
482 accept_wait:
483 	ret = wait_for_completion_interruptible(&cnp->accept_comp);
484 	if (ret)
485 		return -ENODEV;
486 
487 	spin_lock_bh(&np->np_thread_lock);
488 	if (np->np_thread_state >= ISCSI_NP_THREAD_RESET) {
489 		spin_unlock_bh(&np->np_thread_lock);
490 		/**
491 		 * No point in stalling here when np_thread
492 		 * is in state RESET/SHUTDOWN/EXIT - bail
493 		 **/
494 		return -ENODEV;
495 	}
496 	spin_unlock_bh(&np->np_thread_lock);
497 
498 	spin_lock_bh(&cnp->np_accept_lock);
499 	if (list_empty(&cnp->np_accept_list)) {
500 		spin_unlock_bh(&cnp->np_accept_lock);
501 		goto accept_wait;
502 	}
503 
504 	csk = list_first_entry(&cnp->np_accept_list,
505 			       struct cxgbit_sock,
506 			       accept_node);
507 
508 	list_del_init(&csk->accept_node);
509 	spin_unlock_bh(&cnp->np_accept_lock);
510 	conn->context = csk;
511 	csk->conn = conn;
512 
513 	cxgbit_set_conn_info(np, conn, csk);
514 	return 0;
515 }
516 
517 static int
518 __cxgbit_free_cdev_np(struct cxgbit_device *cdev, struct cxgbit_np *cnp)
519 {
520 	int stid, ret;
521 	bool ipv6 = false;
522 
523 	stid = cxgbit_np_hash_del(cdev, cnp);
524 	if (stid < 0)
525 		return -EINVAL;
526 	if (!test_bit(CDEV_STATE_UP, &cdev->flags))
527 		return -EINVAL;
528 
529 	if (cnp->np->np_sockaddr.ss_family == AF_INET6)
530 		ipv6 = true;
531 
532 	cxgbit_get_cnp(cnp);
533 	cxgbit_init_wr_wait(&cnp->com.wr_wait);
534 	ret = cxgb4_remove_server(cdev->lldi.ports[0], stid,
535 				  cdev->lldi.rxq_ids[0], ipv6);
536 
537 	if (ret > 0)
538 		ret = net_xmit_errno(ret);
539 
540 	if (ret) {
541 		cxgbit_put_cnp(cnp);
542 		return ret;
543 	}
544 
545 	ret = cxgbit_wait_for_reply(cdev, &cnp->com.wr_wait,
546 				    0, 10, __func__);
547 	if (ret == -ETIMEDOUT)
548 		return ret;
549 
550 	if (ipv6 && cnp->com.cdev) {
551 		struct sockaddr_in6 *sin6;
552 
553 		sin6 = (struct sockaddr_in6 *)&cnp->com.local_addr;
554 		cxgb4_clip_release(cdev->lldi.ports[0],
555 				   (const u32 *)&sin6->sin6_addr.s6_addr,
556 				   1);
557 	}
558 
559 	cxgb4_free_stid(cdev->lldi.tids, stid,
560 			cnp->com.local_addr.ss_family);
561 	return 0;
562 }
563 
564 static void cxgbit_free_all_np(struct cxgbit_np *cnp)
565 {
566 	struct cxgbit_device *cdev;
567 	int ret;
568 
569 	mutex_lock(&cdev_list_lock);
570 	list_for_each_entry(cdev, &cdev_list_head, list) {
571 		ret = __cxgbit_free_cdev_np(cdev, cnp);
572 		if (ret == -ETIMEDOUT)
573 			break;
574 	}
575 	mutex_unlock(&cdev_list_lock);
576 }
577 
578 static void cxgbit_free_cdev_np(struct cxgbit_np *cnp)
579 {
580 	struct cxgbit_device *cdev;
581 	bool found = false;
582 
583 	mutex_lock(&cdev_list_lock);
584 	list_for_each_entry(cdev, &cdev_list_head, list) {
585 		if (cdev == cnp->com.cdev) {
586 			found = true;
587 			break;
588 		}
589 	}
590 	if (!found)
591 		goto out;
592 
593 	__cxgbit_free_cdev_np(cdev, cnp);
594 out:
595 	mutex_unlock(&cdev_list_lock);
596 }
597 
598 static void __cxgbit_free_conn(struct cxgbit_sock *csk);
599 
600 void cxgbit_free_np(struct iscsi_np *np)
601 {
602 	struct cxgbit_np *cnp = np->np_context;
603 	struct cxgbit_sock *csk, *tmp;
604 
605 	cnp->com.state = CSK_STATE_DEAD;
606 	if (cnp->com.cdev)
607 		cxgbit_free_cdev_np(cnp);
608 	else
609 		cxgbit_free_all_np(cnp);
610 
611 	spin_lock_bh(&cnp->np_accept_lock);
612 	list_for_each_entry_safe(csk, tmp, &cnp->np_accept_list, accept_node) {
613 		list_del_init(&csk->accept_node);
614 		__cxgbit_free_conn(csk);
615 	}
616 	spin_unlock_bh(&cnp->np_accept_lock);
617 
618 	np->np_context = NULL;
619 	cxgbit_put_cnp(cnp);
620 }
621 
622 static void cxgbit_send_halfclose(struct cxgbit_sock *csk)
623 {
624 	struct sk_buff *skb;
625 	u32 len = roundup(sizeof(struct cpl_close_con_req), 16);
626 
627 	skb = alloc_skb(len, GFP_ATOMIC);
628 	if (!skb)
629 		return;
630 
631 	cxgb_mk_close_con_req(skb, len, csk->tid, csk->txq_idx,
632 			      NULL, NULL);
633 
634 	cxgbit_skcb_flags(skb) |= SKCBF_TX_FLAG_COMPL;
635 	__skb_queue_tail(&csk->txq, skb);
636 	cxgbit_push_tx_frames(csk);
637 }
638 
639 static void cxgbit_arp_failure_discard(void *handle, struct sk_buff *skb)
640 {
641 	struct cxgbit_sock *csk = handle;
642 
643 	pr_debug("%s cxgbit_device %p\n", __func__, handle);
644 	kfree_skb(skb);
645 	cxgbit_put_csk(csk);
646 }
647 
648 static void cxgbit_abort_arp_failure(void *handle, struct sk_buff *skb)
649 {
650 	struct cxgbit_device *cdev = handle;
651 	struct cpl_abort_req *req = cplhdr(skb);
652 
653 	pr_debug("%s cdev %p\n", __func__, cdev);
654 	req->cmd = CPL_ABORT_NO_RST;
655 	cxgbit_ofld_send(cdev, skb);
656 }
657 
658 static int cxgbit_send_abort_req(struct cxgbit_sock *csk)
659 {
660 	struct sk_buff *skb;
661 	u32 len = roundup(sizeof(struct cpl_abort_req), 16);
662 
663 	pr_debug("%s: csk %p tid %u; state %d\n",
664 		 __func__, csk, csk->tid, csk->com.state);
665 
666 	__skb_queue_purge(&csk->txq);
667 
668 	if (!test_and_set_bit(CSK_TX_DATA_SENT, &csk->com.flags))
669 		cxgbit_send_tx_flowc_wr(csk);
670 
671 	skb = __skb_dequeue(&csk->skbq);
672 	cxgb_mk_abort_req(skb, len, csk->tid, csk->txq_idx,
673 			  csk->com.cdev, cxgbit_abort_arp_failure);
674 
675 	return cxgbit_l2t_send(csk->com.cdev, skb, csk->l2t);
676 }
677 
678 static void
679 __cxgbit_abort_conn(struct cxgbit_sock *csk, struct sk_buff *skb)
680 {
681 	__kfree_skb(skb);
682 
683 	if (csk->com.state != CSK_STATE_ESTABLISHED)
684 		goto no_abort;
685 
686 	set_bit(CSK_ABORT_RPL_WAIT, &csk->com.flags);
687 	csk->com.state = CSK_STATE_ABORTING;
688 
689 	cxgbit_send_abort_req(csk);
690 
691 	return;
692 
693 no_abort:
694 	cxgbit_wake_up(&csk->com.wr_wait, __func__, CPL_ERR_NONE);
695 	cxgbit_put_csk(csk);
696 }
697 
698 void cxgbit_abort_conn(struct cxgbit_sock *csk)
699 {
700 	struct sk_buff *skb = alloc_skb(0, GFP_KERNEL | __GFP_NOFAIL);
701 
702 	cxgbit_get_csk(csk);
703 	cxgbit_init_wr_wait(&csk->com.wr_wait);
704 
705 	spin_lock_bh(&csk->lock);
706 	if (csk->lock_owner) {
707 		cxgbit_skcb_rx_backlog_fn(skb) = __cxgbit_abort_conn;
708 		__skb_queue_tail(&csk->backlogq, skb);
709 	} else {
710 		__cxgbit_abort_conn(csk, skb);
711 	}
712 	spin_unlock_bh(&csk->lock);
713 
714 	cxgbit_wait_for_reply(csk->com.cdev, &csk->com.wr_wait,
715 			      csk->tid, 600, __func__);
716 }
717 
718 static void __cxgbit_free_conn(struct cxgbit_sock *csk)
719 {
720 	struct iscsit_conn *conn = csk->conn;
721 	bool release = false;
722 
723 	pr_debug("%s: state %d\n",
724 		 __func__, csk->com.state);
725 
726 	spin_lock_bh(&csk->lock);
727 	switch (csk->com.state) {
728 	case CSK_STATE_ESTABLISHED:
729 		if (conn && (conn->conn_state == TARG_CONN_STATE_IN_LOGOUT)) {
730 			csk->com.state = CSK_STATE_CLOSING;
731 			cxgbit_send_halfclose(csk);
732 		} else {
733 			csk->com.state = CSK_STATE_ABORTING;
734 			cxgbit_send_abort_req(csk);
735 		}
736 		break;
737 	case CSK_STATE_CLOSING:
738 		csk->com.state = CSK_STATE_MORIBUND;
739 		cxgbit_send_halfclose(csk);
740 		break;
741 	case CSK_STATE_DEAD:
742 		release = true;
743 		break;
744 	default:
745 		pr_err("%s: csk %p; state %d\n",
746 		       __func__, csk, csk->com.state);
747 	}
748 	spin_unlock_bh(&csk->lock);
749 
750 	if (release)
751 		cxgbit_put_csk(csk);
752 }
753 
754 void cxgbit_free_conn(struct iscsit_conn *conn)
755 {
756 	__cxgbit_free_conn(conn->context);
757 }
758 
759 static void cxgbit_set_emss(struct cxgbit_sock *csk, u16 opt)
760 {
761 	csk->emss = csk->com.cdev->lldi.mtus[TCPOPT_MSS_G(opt)] -
762 			((csk->com.remote_addr.ss_family == AF_INET) ?
763 			sizeof(struct iphdr) : sizeof(struct ipv6hdr)) -
764 			sizeof(struct tcphdr);
765 	csk->mss = csk->emss;
766 	if (TCPOPT_TSTAMP_G(opt))
767 		csk->emss -= round_up(TCPOLEN_TIMESTAMP, 4);
768 	if (csk->emss < 128)
769 		csk->emss = 128;
770 	if (csk->emss & 7)
771 		pr_info("Warning: misaligned mtu idx %u mss %u emss=%u\n",
772 			TCPOPT_MSS_G(opt), csk->mss, csk->emss);
773 	pr_debug("%s mss_idx %u mss %u emss=%u\n", __func__, TCPOPT_MSS_G(opt),
774 		 csk->mss, csk->emss);
775 }
776 
777 static void cxgbit_free_skb(struct cxgbit_sock *csk)
778 {
779 	struct sk_buff *skb;
780 
781 	__skb_queue_purge(&csk->txq);
782 	__skb_queue_purge(&csk->rxq);
783 	__skb_queue_purge(&csk->backlogq);
784 	__skb_queue_purge(&csk->ppodq);
785 	__skb_queue_purge(&csk->skbq);
786 
787 	while ((skb = cxgbit_sock_dequeue_wr(csk)))
788 		kfree_skb(skb);
789 
790 	__kfree_skb(csk->lro_hskb);
791 }
792 
793 void _cxgbit_free_csk(struct kref *kref)
794 {
795 	struct cxgbit_sock *csk;
796 	struct cxgbit_device *cdev;
797 
798 	csk = container_of(kref, struct cxgbit_sock, kref);
799 
800 	pr_debug("%s csk %p state %d\n", __func__, csk, csk->com.state);
801 
802 	if (csk->com.local_addr.ss_family == AF_INET6) {
803 		struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)
804 					     &csk->com.local_addr;
805 		cxgb4_clip_release(csk->com.cdev->lldi.ports[0],
806 				   (const u32 *)
807 				   &sin6->sin6_addr.s6_addr, 1);
808 	}
809 
810 	cxgb4_remove_tid(csk->com.cdev->lldi.tids, 0, csk->tid,
811 			 csk->com.local_addr.ss_family);
812 	dst_release(csk->dst);
813 	cxgb4_l2t_release(csk->l2t);
814 
815 	cdev = csk->com.cdev;
816 	spin_lock_bh(&cdev->cskq.lock);
817 	list_del(&csk->list);
818 	spin_unlock_bh(&cdev->cskq.lock);
819 
820 	cxgbit_free_skb(csk);
821 	cxgbit_put_cnp(csk->cnp);
822 	cxgbit_put_cdev(cdev);
823 
824 	kfree(csk);
825 }
826 
827 static void cxgbit_set_tcp_window(struct cxgbit_sock *csk, struct port_info *pi)
828 {
829 	unsigned int linkspeed;
830 	u8 scale;
831 
832 	linkspeed = pi->link_cfg.speed;
833 	scale = linkspeed / SPEED_10000;
834 
835 #define CXGBIT_10G_RCV_WIN (256 * 1024)
836 	csk->rcv_win = CXGBIT_10G_RCV_WIN;
837 	if (scale)
838 		csk->rcv_win *= scale;
839 	csk->rcv_win = min(csk->rcv_win, RCV_BUFSIZ_M << 10);
840 
841 #define CXGBIT_10G_SND_WIN (256 * 1024)
842 	csk->snd_win = CXGBIT_10G_SND_WIN;
843 	if (scale)
844 		csk->snd_win *= scale;
845 	csk->snd_win = min(csk->snd_win, 512U * 1024);
846 
847 	pr_debug("%s snd_win %d rcv_win %d\n",
848 		 __func__, csk->snd_win, csk->rcv_win);
849 }
850 
851 #ifdef CONFIG_CHELSIO_T4_DCB
852 static u8 cxgbit_get_iscsi_dcb_state(struct net_device *ndev)
853 {
854 	return ndev->dcbnl_ops->getstate(ndev);
855 }
856 
857 static int cxgbit_select_priority(int pri_mask)
858 {
859 	if (!pri_mask)
860 		return 0;
861 
862 	return (ffs(pri_mask) - 1);
863 }
864 
865 static u8 cxgbit_get_iscsi_dcb_priority(struct net_device *ndev, u16 local_port)
866 {
867 	int ret;
868 	u8 caps;
869 
870 	struct dcb_app iscsi_dcb_app = {
871 		.protocol = local_port
872 	};
873 
874 	ret = (int)ndev->dcbnl_ops->getcap(ndev, DCB_CAP_ATTR_DCBX, &caps);
875 
876 	if (ret)
877 		return 0;
878 
879 	if (caps & DCB_CAP_DCBX_VER_IEEE) {
880 		iscsi_dcb_app.selector = IEEE_8021QAZ_APP_SEL_STREAM;
881 		ret = dcb_ieee_getapp_mask(ndev, &iscsi_dcb_app);
882 		if (!ret) {
883 			iscsi_dcb_app.selector = IEEE_8021QAZ_APP_SEL_ANY;
884 			ret = dcb_ieee_getapp_mask(ndev, &iscsi_dcb_app);
885 		}
886 	} else if (caps & DCB_CAP_DCBX_VER_CEE) {
887 		iscsi_dcb_app.selector = DCB_APP_IDTYPE_PORTNUM;
888 
889 		ret = dcb_getapp(ndev, &iscsi_dcb_app);
890 	}
891 
892 	pr_info("iSCSI priority is set to %u\n", cxgbit_select_priority(ret));
893 
894 	return cxgbit_select_priority(ret);
895 }
896 #endif
897 
898 static int
899 cxgbit_offload_init(struct cxgbit_sock *csk, int iptype, __u8 *peer_ip,
900 		    u16 local_port, struct dst_entry *dst,
901 		    struct cxgbit_device *cdev)
902 {
903 	struct neighbour *n;
904 	int ret, step;
905 	struct net_device *ndev;
906 	u16 rxq_idx, port_id;
907 #ifdef CONFIG_CHELSIO_T4_DCB
908 	u8 priority = 0;
909 #endif
910 
911 	n = dst_neigh_lookup(dst, peer_ip);
912 	if (!n)
913 		return -ENODEV;
914 
915 	rcu_read_lock();
916 	if (!(n->nud_state & NUD_VALID))
917 		neigh_event_send(n, NULL);
918 
919 	ret = -ENOMEM;
920 	if (n->dev->flags & IFF_LOOPBACK) {
921 		if (iptype == 4)
922 			ndev = cxgbit_ipv4_netdev(*(__be32 *)peer_ip);
923 		else if (IS_ENABLED(CONFIG_IPV6))
924 			ndev = cxgbit_ipv6_netdev((struct in6_addr *)peer_ip);
925 		else
926 			ndev = NULL;
927 
928 		if (!ndev) {
929 			ret = -ENODEV;
930 			goto out;
931 		}
932 
933 		csk->l2t = cxgb4_l2t_get(cdev->lldi.l2t,
934 					 n, ndev, 0);
935 		if (!csk->l2t)
936 			goto out;
937 		csk->mtu = ndev->mtu;
938 		csk->tx_chan = cxgb4_port_chan(ndev);
939 		csk->smac_idx =
940 			       ((struct port_info *)netdev_priv(ndev))->smt_idx;
941 		step = cdev->lldi.ntxq /
942 			cdev->lldi.nchan;
943 		csk->txq_idx = cxgb4_port_idx(ndev) * step;
944 		step = cdev->lldi.nrxq /
945 			cdev->lldi.nchan;
946 		csk->ctrlq_idx = cxgb4_port_idx(ndev);
947 		csk->rss_qid = cdev->lldi.rxq_ids[
948 				cxgb4_port_idx(ndev) * step];
949 		csk->port_id = cxgb4_port_idx(ndev);
950 		cxgbit_set_tcp_window(csk,
951 				      (struct port_info *)netdev_priv(ndev));
952 	} else {
953 		ndev = cxgbit_get_real_dev(n->dev);
954 		if (!ndev) {
955 			ret = -ENODEV;
956 			goto out;
957 		}
958 
959 #ifdef CONFIG_CHELSIO_T4_DCB
960 		if (cxgbit_get_iscsi_dcb_state(ndev))
961 			priority = cxgbit_get_iscsi_dcb_priority(ndev,
962 								 local_port);
963 
964 		csk->dcb_priority = priority;
965 
966 		csk->l2t = cxgb4_l2t_get(cdev->lldi.l2t, n, ndev, priority);
967 #else
968 		csk->l2t = cxgb4_l2t_get(cdev->lldi.l2t, n, ndev, 0);
969 #endif
970 		if (!csk->l2t)
971 			goto out;
972 		port_id = cxgb4_port_idx(ndev);
973 		csk->mtu = dst_mtu(dst);
974 		csk->tx_chan = cxgb4_port_chan(ndev);
975 		csk->smac_idx =
976 			       ((struct port_info *)netdev_priv(ndev))->smt_idx;
977 		step = cdev->lldi.ntxq /
978 			cdev->lldi.nports;
979 		csk->txq_idx = (port_id * step) +
980 				(cdev->selectq[port_id][0]++ % step);
981 		csk->ctrlq_idx = cxgb4_port_idx(ndev);
982 		step = cdev->lldi.nrxq /
983 			cdev->lldi.nports;
984 		rxq_idx = (port_id * step) +
985 				(cdev->selectq[port_id][1]++ % step);
986 		csk->rss_qid = cdev->lldi.rxq_ids[rxq_idx];
987 		csk->port_id = port_id;
988 		cxgbit_set_tcp_window(csk,
989 				      (struct port_info *)netdev_priv(ndev));
990 	}
991 	ret = 0;
992 out:
993 	rcu_read_unlock();
994 	neigh_release(n);
995 	return ret;
996 }
997 
998 int cxgbit_ofld_send(struct cxgbit_device *cdev, struct sk_buff *skb)
999 {
1000 	int ret = 0;
1001 
1002 	if (!test_bit(CDEV_STATE_UP, &cdev->flags)) {
1003 		kfree_skb(skb);
1004 		pr_err("%s - device not up - dropping\n", __func__);
1005 		return -EIO;
1006 	}
1007 
1008 	ret = cxgb4_ofld_send(cdev->lldi.ports[0], skb);
1009 	if (ret < 0)
1010 		kfree_skb(skb);
1011 	return ret < 0 ? ret : 0;
1012 }
1013 
1014 static void cxgbit_release_tid(struct cxgbit_device *cdev, u32 tid)
1015 {
1016 	u32 len = roundup(sizeof(struct cpl_tid_release), 16);
1017 	struct sk_buff *skb;
1018 
1019 	skb = alloc_skb(len, GFP_ATOMIC);
1020 	if (!skb)
1021 		return;
1022 
1023 	cxgb_mk_tid_release(skb, len, tid, 0);
1024 	cxgbit_ofld_send(cdev, skb);
1025 }
1026 
1027 int
1028 cxgbit_l2t_send(struct cxgbit_device *cdev, struct sk_buff *skb,
1029 		struct l2t_entry *l2e)
1030 {
1031 	int ret = 0;
1032 
1033 	if (!test_bit(CDEV_STATE_UP, &cdev->flags)) {
1034 		kfree_skb(skb);
1035 		pr_err("%s - device not up - dropping\n", __func__);
1036 		return -EIO;
1037 	}
1038 
1039 	ret = cxgb4_l2t_send(cdev->lldi.ports[0], skb, l2e);
1040 	if (ret < 0)
1041 		kfree_skb(skb);
1042 	return ret < 0 ? ret : 0;
1043 }
1044 
1045 static void cxgbit_send_rx_credits(struct cxgbit_sock *csk, struct sk_buff *skb)
1046 {
1047 	if (csk->com.state != CSK_STATE_ESTABLISHED) {
1048 		__kfree_skb(skb);
1049 		return;
1050 	}
1051 
1052 	cxgbit_ofld_send(csk->com.cdev, skb);
1053 }
1054 
1055 /*
1056  * CPL connection rx data ack: host ->
1057  * Send RX credits through an RX_DATA_ACK CPL message.
1058  * Returns the number of credits sent.
1059  */
1060 int cxgbit_rx_data_ack(struct cxgbit_sock *csk)
1061 {
1062 	struct sk_buff *skb;
1063 	u32 len = roundup(sizeof(struct cpl_rx_data_ack), 16);
1064 	u32 credit_dack;
1065 
1066 	skb = alloc_skb(len, GFP_KERNEL);
1067 	if (!skb)
1068 		return -1;
1069 
1070 	credit_dack = RX_DACK_CHANGE_F | RX_DACK_MODE_V(3) |
1071 		      RX_CREDITS_V(csk->rx_credits);
1072 
1073 	cxgb_mk_rx_data_ack(skb, len, csk->tid, csk->ctrlq_idx,
1074 			    credit_dack);
1075 
1076 	csk->rx_credits = 0;
1077 
1078 	spin_lock_bh(&csk->lock);
1079 	if (csk->lock_owner) {
1080 		cxgbit_skcb_rx_backlog_fn(skb) = cxgbit_send_rx_credits;
1081 		__skb_queue_tail(&csk->backlogq, skb);
1082 		spin_unlock_bh(&csk->lock);
1083 		return 0;
1084 	}
1085 
1086 	cxgbit_send_rx_credits(csk, skb);
1087 	spin_unlock_bh(&csk->lock);
1088 
1089 	return 0;
1090 }
1091 
1092 #define FLOWC_WR_NPARAMS_MIN    9
1093 #define FLOWC_WR_NPARAMS_MAX	11
1094 static int cxgbit_alloc_csk_skb(struct cxgbit_sock *csk)
1095 {
1096 	struct sk_buff *skb;
1097 	u32 len, flowclen;
1098 	u8 i;
1099 
1100 	flowclen = offsetof(struct fw_flowc_wr,
1101 			    mnemval[FLOWC_WR_NPARAMS_MAX]);
1102 
1103 	len = max_t(u32, sizeof(struct cpl_abort_req),
1104 		    sizeof(struct cpl_abort_rpl));
1105 
1106 	len = max(len, flowclen);
1107 	len = roundup(len, 16);
1108 
1109 	for (i = 0; i < 3; i++) {
1110 		skb = alloc_skb(len, GFP_ATOMIC);
1111 		if (!skb)
1112 			goto out;
1113 		__skb_queue_tail(&csk->skbq, skb);
1114 	}
1115 
1116 	skb = alloc_skb(LRO_SKB_MIN_HEADROOM, GFP_ATOMIC);
1117 	if (!skb)
1118 		goto out;
1119 
1120 	memset(skb->data, 0, LRO_SKB_MIN_HEADROOM);
1121 	csk->lro_hskb = skb;
1122 
1123 	return 0;
1124 out:
1125 	__skb_queue_purge(&csk->skbq);
1126 	return -ENOMEM;
1127 }
1128 
1129 static void
1130 cxgbit_pass_accept_rpl(struct cxgbit_sock *csk, struct cpl_pass_accept_req *req)
1131 {
1132 	struct sk_buff *skb;
1133 	const struct tcphdr *tcph;
1134 	struct cpl_t5_pass_accept_rpl *rpl5;
1135 	struct cxgb4_lld_info *lldi = &csk->com.cdev->lldi;
1136 	unsigned int len = roundup(sizeof(*rpl5), 16);
1137 	unsigned int mtu_idx;
1138 	u64 opt0;
1139 	u32 opt2, hlen;
1140 	u32 wscale;
1141 	u32 win;
1142 
1143 	pr_debug("%s csk %p tid %u\n", __func__, csk, csk->tid);
1144 
1145 	skb = alloc_skb(len, GFP_ATOMIC);
1146 	if (!skb) {
1147 		cxgbit_put_csk(csk);
1148 		return;
1149 	}
1150 
1151 	rpl5 = __skb_put_zero(skb, len);
1152 
1153 	INIT_TP_WR(rpl5, csk->tid);
1154 	OPCODE_TID(rpl5) = cpu_to_be32(MK_OPCODE_TID(CPL_PASS_ACCEPT_RPL,
1155 						     csk->tid));
1156 	cxgb_best_mtu(csk->com.cdev->lldi.mtus, csk->mtu, &mtu_idx,
1157 		      req->tcpopt.tstamp,
1158 		      (csk->com.remote_addr.ss_family == AF_INET) ? 0 : 1);
1159 	wscale = cxgb_compute_wscale(csk->rcv_win);
1160 	/*
1161 	 * Specify the largest window that will fit in opt0. The
1162 	 * remainder will be specified in the rx_data_ack.
1163 	 */
1164 	win = csk->rcv_win >> 10;
1165 	if (win > RCV_BUFSIZ_M)
1166 		win = RCV_BUFSIZ_M;
1167 	opt0 =  TCAM_BYPASS_F |
1168 		WND_SCALE_V(wscale) |
1169 		MSS_IDX_V(mtu_idx) |
1170 		L2T_IDX_V(csk->l2t->idx) |
1171 		TX_CHAN_V(csk->tx_chan) |
1172 		SMAC_SEL_V(csk->smac_idx) |
1173 		DSCP_V(csk->tos >> 2) |
1174 		ULP_MODE_V(ULP_MODE_ISCSI) |
1175 		RCV_BUFSIZ_V(win);
1176 
1177 	opt2 = RX_CHANNEL_V(0) |
1178 		RSS_QUEUE_VALID_F | RSS_QUEUE_V(csk->rss_qid);
1179 
1180 	if (!is_t5(lldi->adapter_type))
1181 		opt2 |= RX_FC_DISABLE_F;
1182 
1183 	if (req->tcpopt.tstamp)
1184 		opt2 |= TSTAMPS_EN_F;
1185 	if (req->tcpopt.sack)
1186 		opt2 |= SACK_EN_F;
1187 	if (wscale)
1188 		opt2 |= WND_SCALE_EN_F;
1189 
1190 	hlen = ntohl(req->hdr_len);
1191 
1192 	if (is_t5(lldi->adapter_type))
1193 		tcph = (struct tcphdr *)((u8 *)(req + 1) +
1194 		       ETH_HDR_LEN_G(hlen) + IP_HDR_LEN_G(hlen));
1195 	else
1196 		tcph = (struct tcphdr *)((u8 *)(req + 1) +
1197 		       T6_ETH_HDR_LEN_G(hlen) + T6_IP_HDR_LEN_G(hlen));
1198 
1199 	if (tcph->ece && tcph->cwr)
1200 		opt2 |= CCTRL_ECN_V(1);
1201 
1202 	opt2 |= CONG_CNTRL_V(CONG_ALG_NEWRENO);
1203 
1204 	opt2 |= T5_ISS_F;
1205 	rpl5->iss = cpu_to_be32((get_random_u32() & ~7UL) - 1);
1206 
1207 	opt2 |= T5_OPT_2_VALID_F;
1208 
1209 	rpl5->opt0 = cpu_to_be64(opt0);
1210 	rpl5->opt2 = cpu_to_be32(opt2);
1211 	set_wr_txq(skb, CPL_PRIORITY_SETUP, csk->ctrlq_idx);
1212 	t4_set_arp_err_handler(skb, csk, cxgbit_arp_failure_discard);
1213 	cxgbit_l2t_send(csk->com.cdev, skb, csk->l2t);
1214 }
1215 
1216 static void
1217 cxgbit_pass_accept_req(struct cxgbit_device *cdev, struct sk_buff *skb)
1218 {
1219 	struct cxgbit_sock *csk = NULL;
1220 	struct cxgbit_np *cnp;
1221 	struct cpl_pass_accept_req *req = cplhdr(skb);
1222 	unsigned int stid = PASS_OPEN_TID_G(ntohl(req->tos_stid));
1223 	struct tid_info *t = cdev->lldi.tids;
1224 	unsigned int tid = GET_TID(req);
1225 	u16 peer_mss = ntohs(req->tcpopt.mss);
1226 	unsigned short hdrs;
1227 
1228 	struct dst_entry *dst;
1229 	__u8 local_ip[16], peer_ip[16];
1230 	__be16 local_port, peer_port;
1231 	int ret;
1232 	int iptype;
1233 
1234 	pr_debug("%s: cdev = %p; stid = %u; tid = %u\n",
1235 		 __func__, cdev, stid, tid);
1236 
1237 	cnp = lookup_stid(t, stid);
1238 	if (!cnp) {
1239 		pr_err("%s connect request on invalid stid %d\n",
1240 		       __func__, stid);
1241 		goto rel_skb;
1242 	}
1243 
1244 	if (cnp->com.state != CSK_STATE_LISTEN) {
1245 		pr_err("%s - listening parent not in CSK_STATE_LISTEN\n",
1246 		       __func__);
1247 		goto reject;
1248 	}
1249 
1250 	csk = lookup_tid(t, tid);
1251 	if (csk) {
1252 		pr_err("%s csk not null tid %u\n",
1253 		       __func__, tid);
1254 		goto rel_skb;
1255 	}
1256 
1257 	cxgb_get_4tuple(req, cdev->lldi.adapter_type, &iptype, local_ip,
1258 			peer_ip, &local_port, &peer_port);
1259 
1260 	/* Find output route */
1261 	if (iptype == 4)  {
1262 		pr_debug("%s parent sock %p tid %u laddr %pI4 raddr %pI4 "
1263 			 "lport %d rport %d peer_mss %d\n"
1264 			 , __func__, cnp, tid,
1265 			 local_ip, peer_ip, ntohs(local_port),
1266 			 ntohs(peer_port), peer_mss);
1267 		dst = cxgb_find_route(&cdev->lldi, cxgbit_get_real_dev,
1268 				      *(__be32 *)local_ip,
1269 				      *(__be32 *)peer_ip,
1270 				      local_port, peer_port,
1271 				      PASS_OPEN_TOS_G(ntohl(req->tos_stid)));
1272 	} else {
1273 		pr_debug("%s parent sock %p tid %u laddr %pI6 raddr %pI6 "
1274 			 "lport %d rport %d peer_mss %d\n"
1275 			 , __func__, cnp, tid,
1276 			 local_ip, peer_ip, ntohs(local_port),
1277 			 ntohs(peer_port), peer_mss);
1278 		dst = cxgb_find_route6(&cdev->lldi, cxgbit_get_real_dev,
1279 				       local_ip, peer_ip,
1280 				       local_port, peer_port,
1281 				       PASS_OPEN_TOS_G(ntohl(req->tos_stid)),
1282 				       ((struct sockaddr_in6 *)
1283 					&cnp->com.local_addr)->sin6_scope_id);
1284 	}
1285 	if (!dst) {
1286 		pr_err("%s - failed to find dst entry!\n",
1287 		       __func__);
1288 		goto reject;
1289 	}
1290 
1291 	csk = kzalloc(sizeof(*csk), GFP_ATOMIC);
1292 	if (!csk) {
1293 		dst_release(dst);
1294 		goto rel_skb;
1295 	}
1296 
1297 	ret = cxgbit_offload_init(csk, iptype, peer_ip, ntohs(local_port),
1298 				  dst, cdev);
1299 	if (ret) {
1300 		pr_err("%s - failed to allocate l2t entry!\n",
1301 		       __func__);
1302 		dst_release(dst);
1303 		kfree(csk);
1304 		goto reject;
1305 	}
1306 
1307 	kref_init(&csk->kref);
1308 	init_completion(&csk->com.wr_wait.completion);
1309 
1310 	INIT_LIST_HEAD(&csk->accept_node);
1311 
1312 	hdrs = (iptype == 4 ? sizeof(struct iphdr) : sizeof(struct ipv6hdr)) +
1313 		sizeof(struct tcphdr) +	(req->tcpopt.tstamp ? 12 : 0);
1314 	if (peer_mss && csk->mtu > (peer_mss + hdrs))
1315 		csk->mtu = peer_mss + hdrs;
1316 
1317 	csk->com.state = CSK_STATE_CONNECTING;
1318 	csk->com.cdev = cdev;
1319 	csk->cnp = cnp;
1320 	csk->tos = PASS_OPEN_TOS_G(ntohl(req->tos_stid));
1321 	csk->dst = dst;
1322 	csk->tid = tid;
1323 	csk->wr_cred = cdev->lldi.wr_cred -
1324 			DIV_ROUND_UP(sizeof(struct cpl_abort_req), 16);
1325 	csk->wr_max_cred = csk->wr_cred;
1326 	csk->wr_una_cred = 0;
1327 
1328 	if (iptype == 4) {
1329 		struct sockaddr_in *sin = (struct sockaddr_in *)
1330 					  &csk->com.local_addr;
1331 		sin->sin_family = AF_INET;
1332 		sin->sin_port = local_port;
1333 		sin->sin_addr.s_addr = *(__be32 *)local_ip;
1334 
1335 		sin = (struct sockaddr_in *)&csk->com.remote_addr;
1336 		sin->sin_family = AF_INET;
1337 		sin->sin_port = peer_port;
1338 		sin->sin_addr.s_addr = *(__be32 *)peer_ip;
1339 	} else {
1340 		struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)
1341 					    &csk->com.local_addr;
1342 
1343 		sin6->sin6_family = PF_INET6;
1344 		sin6->sin6_port = local_port;
1345 		memcpy(sin6->sin6_addr.s6_addr, local_ip, 16);
1346 		cxgb4_clip_get(cdev->lldi.ports[0],
1347 			       (const u32 *)&sin6->sin6_addr.s6_addr,
1348 			       1);
1349 
1350 		sin6 = (struct sockaddr_in6 *)&csk->com.remote_addr;
1351 		sin6->sin6_family = PF_INET6;
1352 		sin6->sin6_port = peer_port;
1353 		memcpy(sin6->sin6_addr.s6_addr, peer_ip, 16);
1354 	}
1355 
1356 	skb_queue_head_init(&csk->rxq);
1357 	skb_queue_head_init(&csk->txq);
1358 	skb_queue_head_init(&csk->ppodq);
1359 	skb_queue_head_init(&csk->backlogq);
1360 	skb_queue_head_init(&csk->skbq);
1361 	cxgbit_sock_reset_wr_list(csk);
1362 	spin_lock_init(&csk->lock);
1363 	init_waitqueue_head(&csk->waitq);
1364 	csk->lock_owner = false;
1365 
1366 	if (cxgbit_alloc_csk_skb(csk)) {
1367 		dst_release(dst);
1368 		kfree(csk);
1369 		goto rel_skb;
1370 	}
1371 
1372 	cxgbit_get_cnp(cnp);
1373 	cxgbit_get_cdev(cdev);
1374 
1375 	spin_lock(&cdev->cskq.lock);
1376 	list_add_tail(&csk->list, &cdev->cskq.list);
1377 	spin_unlock(&cdev->cskq.lock);
1378 	cxgb4_insert_tid(t, csk, tid, csk->com.local_addr.ss_family);
1379 	cxgbit_pass_accept_rpl(csk, req);
1380 	goto rel_skb;
1381 
1382 reject:
1383 	cxgbit_release_tid(cdev, tid);
1384 rel_skb:
1385 	__kfree_skb(skb);
1386 }
1387 
1388 static u32
1389 cxgbit_tx_flowc_wr_credits(struct cxgbit_sock *csk, u32 *nparamsp,
1390 			   u32 *flowclenp)
1391 {
1392 	u32 nparams, flowclen16, flowclen;
1393 
1394 	nparams = FLOWC_WR_NPARAMS_MIN;
1395 
1396 	if (csk->snd_wscale)
1397 		nparams++;
1398 
1399 #ifdef CONFIG_CHELSIO_T4_DCB
1400 	nparams++;
1401 #endif
1402 	flowclen = offsetof(struct fw_flowc_wr, mnemval[nparams]);
1403 	flowclen16 = DIV_ROUND_UP(flowclen, 16);
1404 	flowclen = flowclen16 * 16;
1405 	/*
1406 	 * Return the number of 16-byte credits used by the flowc request.
1407 	 * Pass back the nparams and actual flowc length if requested.
1408 	 */
1409 	if (nparamsp)
1410 		*nparamsp = nparams;
1411 	if (flowclenp)
1412 		*flowclenp = flowclen;
1413 	return flowclen16;
1414 }
1415 
1416 u32 cxgbit_send_tx_flowc_wr(struct cxgbit_sock *csk)
1417 {
1418 	struct cxgbit_device *cdev = csk->com.cdev;
1419 	struct fw_flowc_wr *flowc;
1420 	u32 nparams, flowclen16, flowclen;
1421 	struct sk_buff *skb;
1422 	u8 index;
1423 
1424 #ifdef CONFIG_CHELSIO_T4_DCB
1425 	u16 vlan = ((struct l2t_entry *)csk->l2t)->vlan;
1426 #endif
1427 
1428 	flowclen16 = cxgbit_tx_flowc_wr_credits(csk, &nparams, &flowclen);
1429 
1430 	skb = __skb_dequeue(&csk->skbq);
1431 	flowc = __skb_put_zero(skb, flowclen);
1432 
1433 	flowc->op_to_nparams = cpu_to_be32(FW_WR_OP_V(FW_FLOWC_WR) |
1434 					   FW_FLOWC_WR_NPARAMS_V(nparams));
1435 	flowc->flowid_len16 = cpu_to_be32(FW_WR_LEN16_V(flowclen16) |
1436 					  FW_WR_FLOWID_V(csk->tid));
1437 	flowc->mnemval[0].mnemonic = FW_FLOWC_MNEM_PFNVFN;
1438 	flowc->mnemval[0].val = cpu_to_be32(FW_PFVF_CMD_PFN_V
1439 					    (csk->com.cdev->lldi.pf));
1440 	flowc->mnemval[1].mnemonic = FW_FLOWC_MNEM_CH;
1441 	flowc->mnemval[1].val = cpu_to_be32(csk->tx_chan);
1442 	flowc->mnemval[2].mnemonic = FW_FLOWC_MNEM_PORT;
1443 	flowc->mnemval[2].val = cpu_to_be32(csk->tx_chan);
1444 	flowc->mnemval[3].mnemonic = FW_FLOWC_MNEM_IQID;
1445 	flowc->mnemval[3].val = cpu_to_be32(csk->rss_qid);
1446 	flowc->mnemval[4].mnemonic = FW_FLOWC_MNEM_SNDNXT;
1447 	flowc->mnemval[4].val = cpu_to_be32(csk->snd_nxt);
1448 	flowc->mnemval[5].mnemonic = FW_FLOWC_MNEM_RCVNXT;
1449 	flowc->mnemval[5].val = cpu_to_be32(csk->rcv_nxt);
1450 	flowc->mnemval[6].mnemonic = FW_FLOWC_MNEM_SNDBUF;
1451 	flowc->mnemval[6].val = cpu_to_be32(csk->snd_win);
1452 	flowc->mnemval[7].mnemonic = FW_FLOWC_MNEM_MSS;
1453 	flowc->mnemval[7].val = cpu_to_be32(csk->emss);
1454 
1455 	flowc->mnemval[8].mnemonic = FW_FLOWC_MNEM_TXDATAPLEN_MAX;
1456 	if (test_bit(CDEV_ISO_ENABLE, &cdev->flags))
1457 		flowc->mnemval[8].val = cpu_to_be32(CXGBIT_MAX_ISO_PAYLOAD);
1458 	else
1459 		flowc->mnemval[8].val = cpu_to_be32(16384);
1460 
1461 	index = 9;
1462 
1463 	if (csk->snd_wscale) {
1464 		flowc->mnemval[index].mnemonic = FW_FLOWC_MNEM_RCV_SCALE;
1465 		flowc->mnemval[index].val = cpu_to_be32(csk->snd_wscale);
1466 		index++;
1467 	}
1468 
1469 #ifdef CONFIG_CHELSIO_T4_DCB
1470 	flowc->mnemval[index].mnemonic = FW_FLOWC_MNEM_DCBPRIO;
1471 	if (vlan == VLAN_NONE) {
1472 		pr_warn("csk %u without VLAN Tag on DCB Link\n", csk->tid);
1473 		flowc->mnemval[index].val = cpu_to_be32(0);
1474 	} else
1475 		flowc->mnemval[index].val = cpu_to_be32(
1476 				(vlan & VLAN_PRIO_MASK) >> VLAN_PRIO_SHIFT);
1477 #endif
1478 
1479 	pr_debug("%s: csk %p; tx_chan = %u; rss_qid = %u; snd_seq = %u;"
1480 		 " rcv_seq = %u; snd_win = %u; emss = %u\n",
1481 		 __func__, csk, csk->tx_chan, csk->rss_qid, csk->snd_nxt,
1482 		 csk->rcv_nxt, csk->snd_win, csk->emss);
1483 	set_wr_txq(skb, CPL_PRIORITY_DATA, csk->txq_idx);
1484 	cxgbit_ofld_send(csk->com.cdev, skb);
1485 	return flowclen16;
1486 }
1487 
1488 static int
1489 cxgbit_send_tcb_skb(struct cxgbit_sock *csk, struct sk_buff *skb)
1490 {
1491 	spin_lock_bh(&csk->lock);
1492 	if (unlikely(csk->com.state != CSK_STATE_ESTABLISHED)) {
1493 		spin_unlock_bh(&csk->lock);
1494 		pr_err("%s: csk 0x%p, tid %u, state %u\n",
1495 		       __func__, csk, csk->tid, csk->com.state);
1496 		__kfree_skb(skb);
1497 		return -1;
1498 	}
1499 
1500 	cxgbit_get_csk(csk);
1501 	cxgbit_init_wr_wait(&csk->com.wr_wait);
1502 	cxgbit_ofld_send(csk->com.cdev, skb);
1503 	spin_unlock_bh(&csk->lock);
1504 
1505 	return 0;
1506 }
1507 
1508 int cxgbit_setup_conn_digest(struct cxgbit_sock *csk)
1509 {
1510 	struct sk_buff *skb;
1511 	struct cpl_set_tcb_field *req;
1512 	u8 hcrc = csk->submode & CXGBIT_SUBMODE_HCRC;
1513 	u8 dcrc = csk->submode & CXGBIT_SUBMODE_DCRC;
1514 	unsigned int len = roundup(sizeof(*req), 16);
1515 	int ret;
1516 
1517 	skb = alloc_skb(len, GFP_KERNEL);
1518 	if (!skb)
1519 		return -ENOMEM;
1520 
1521 	/*  set up ulp submode */
1522 	req = __skb_put_zero(skb, len);
1523 
1524 	INIT_TP_WR(req, csk->tid);
1525 	OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_SET_TCB_FIELD, csk->tid));
1526 	req->reply_ctrl = htons(NO_REPLY_V(0) | QUEUENO_V(csk->rss_qid));
1527 	req->word_cookie = htons(0);
1528 	req->mask = cpu_to_be64(0x3 << 4);
1529 	req->val = cpu_to_be64(((hcrc ? ULP_CRC_HEADER : 0) |
1530 				(dcrc ? ULP_CRC_DATA : 0)) << 4);
1531 	set_wr_txq(skb, CPL_PRIORITY_CONTROL, csk->ctrlq_idx);
1532 
1533 	if (cxgbit_send_tcb_skb(csk, skb))
1534 		return -1;
1535 
1536 	ret = cxgbit_wait_for_reply(csk->com.cdev,
1537 				    &csk->com.wr_wait,
1538 				    csk->tid, 5, __func__);
1539 	if (ret)
1540 		return -1;
1541 
1542 	return 0;
1543 }
1544 
1545 int cxgbit_setup_conn_pgidx(struct cxgbit_sock *csk, u32 pg_idx)
1546 {
1547 	struct sk_buff *skb;
1548 	struct cpl_set_tcb_field *req;
1549 	unsigned int len = roundup(sizeof(*req), 16);
1550 	int ret;
1551 
1552 	skb = alloc_skb(len, GFP_KERNEL);
1553 	if (!skb)
1554 		return -ENOMEM;
1555 
1556 	req = __skb_put_zero(skb, len);
1557 
1558 	INIT_TP_WR(req, csk->tid);
1559 	OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_SET_TCB_FIELD, csk->tid));
1560 	req->reply_ctrl = htons(NO_REPLY_V(0) | QUEUENO_V(csk->rss_qid));
1561 	req->word_cookie = htons(0);
1562 	req->mask = cpu_to_be64(0x3 << 8);
1563 	req->val = cpu_to_be64(pg_idx << 8);
1564 	set_wr_txq(skb, CPL_PRIORITY_CONTROL, csk->ctrlq_idx);
1565 
1566 	if (cxgbit_send_tcb_skb(csk, skb))
1567 		return -1;
1568 
1569 	ret = cxgbit_wait_for_reply(csk->com.cdev,
1570 				    &csk->com.wr_wait,
1571 				    csk->tid, 5, __func__);
1572 	if (ret)
1573 		return -1;
1574 
1575 	return 0;
1576 }
1577 
1578 static void
1579 cxgbit_pass_open_rpl(struct cxgbit_device *cdev, struct sk_buff *skb)
1580 {
1581 	struct cpl_pass_open_rpl *rpl = cplhdr(skb);
1582 	struct tid_info *t = cdev->lldi.tids;
1583 	unsigned int stid = GET_TID(rpl);
1584 	struct cxgbit_np *cnp = lookup_stid(t, stid);
1585 
1586 	pr_debug("%s: cnp = %p; stid = %u; status = %d\n",
1587 		 __func__, cnp, stid, rpl->status);
1588 
1589 	if (!cnp) {
1590 		pr_info("%s stid %d lookup failure\n", __func__, stid);
1591 		goto rel_skb;
1592 	}
1593 
1594 	cxgbit_wake_up(&cnp->com.wr_wait, __func__, rpl->status);
1595 	cxgbit_put_cnp(cnp);
1596 rel_skb:
1597 	__kfree_skb(skb);
1598 }
1599 
1600 static void
1601 cxgbit_close_listsrv_rpl(struct cxgbit_device *cdev, struct sk_buff *skb)
1602 {
1603 	struct cpl_close_listsvr_rpl *rpl = cplhdr(skb);
1604 	struct tid_info *t = cdev->lldi.tids;
1605 	unsigned int stid = GET_TID(rpl);
1606 	struct cxgbit_np *cnp = lookup_stid(t, stid);
1607 
1608 	pr_debug("%s: cnp = %p; stid = %u; status = %d\n",
1609 		 __func__, cnp, stid, rpl->status);
1610 
1611 	if (!cnp) {
1612 		pr_info("%s stid %d lookup failure\n", __func__, stid);
1613 		goto rel_skb;
1614 	}
1615 
1616 	cxgbit_wake_up(&cnp->com.wr_wait, __func__, rpl->status);
1617 	cxgbit_put_cnp(cnp);
1618 rel_skb:
1619 	__kfree_skb(skb);
1620 }
1621 
1622 static void
1623 cxgbit_pass_establish(struct cxgbit_device *cdev, struct sk_buff *skb)
1624 {
1625 	struct cpl_pass_establish *req = cplhdr(skb);
1626 	struct tid_info *t = cdev->lldi.tids;
1627 	unsigned int tid = GET_TID(req);
1628 	struct cxgbit_sock *csk;
1629 	struct cxgbit_np *cnp;
1630 	u16 tcp_opt = be16_to_cpu(req->tcp_opt);
1631 	u32 snd_isn = be32_to_cpu(req->snd_isn);
1632 	u32 rcv_isn = be32_to_cpu(req->rcv_isn);
1633 
1634 	csk = lookup_tid(t, tid);
1635 	if (unlikely(!csk)) {
1636 		pr_err("can't find connection for tid %u.\n", tid);
1637 		goto rel_skb;
1638 	}
1639 	cnp = csk->cnp;
1640 
1641 	pr_debug("%s: csk %p; tid %u; cnp %p\n",
1642 		 __func__, csk, tid, cnp);
1643 
1644 	csk->write_seq = snd_isn;
1645 	csk->snd_una = snd_isn;
1646 	csk->snd_nxt = snd_isn;
1647 
1648 	csk->rcv_nxt = rcv_isn;
1649 
1650 	csk->snd_wscale = TCPOPT_SND_WSCALE_G(tcp_opt);
1651 	cxgbit_set_emss(csk, tcp_opt);
1652 	dst_confirm(csk->dst);
1653 	csk->com.state = CSK_STATE_ESTABLISHED;
1654 	spin_lock_bh(&cnp->np_accept_lock);
1655 	list_add_tail(&csk->accept_node, &cnp->np_accept_list);
1656 	spin_unlock_bh(&cnp->np_accept_lock);
1657 	complete(&cnp->accept_comp);
1658 rel_skb:
1659 	__kfree_skb(skb);
1660 }
1661 
1662 static void cxgbit_queue_rx_skb(struct cxgbit_sock *csk, struct sk_buff *skb)
1663 {
1664 	cxgbit_skcb_flags(skb) = 0;
1665 	spin_lock_bh(&csk->rxq.lock);
1666 	__skb_queue_tail(&csk->rxq, skb);
1667 	spin_unlock_bh(&csk->rxq.lock);
1668 	wake_up(&csk->waitq);
1669 }
1670 
1671 static void cxgbit_peer_close(struct cxgbit_sock *csk, struct sk_buff *skb)
1672 {
1673 	pr_debug("%s: csk %p; tid %u; state %d\n",
1674 		 __func__, csk, csk->tid, csk->com.state);
1675 
1676 	switch (csk->com.state) {
1677 	case CSK_STATE_ESTABLISHED:
1678 		csk->com.state = CSK_STATE_CLOSING;
1679 		cxgbit_queue_rx_skb(csk, skb);
1680 		return;
1681 	case CSK_STATE_CLOSING:
1682 		/* simultaneous close */
1683 		csk->com.state = CSK_STATE_MORIBUND;
1684 		break;
1685 	case CSK_STATE_MORIBUND:
1686 		csk->com.state = CSK_STATE_DEAD;
1687 		cxgbit_put_csk(csk);
1688 		break;
1689 	case CSK_STATE_ABORTING:
1690 		break;
1691 	default:
1692 		pr_info("%s: cpl_peer_close in bad state %d\n",
1693 			__func__, csk->com.state);
1694 	}
1695 
1696 	__kfree_skb(skb);
1697 }
1698 
1699 static void cxgbit_close_con_rpl(struct cxgbit_sock *csk, struct sk_buff *skb)
1700 {
1701 	pr_debug("%s: csk %p; tid %u; state %d\n",
1702 		 __func__, csk, csk->tid, csk->com.state);
1703 
1704 	switch (csk->com.state) {
1705 	case CSK_STATE_CLOSING:
1706 		csk->com.state = CSK_STATE_MORIBUND;
1707 		break;
1708 	case CSK_STATE_MORIBUND:
1709 		csk->com.state = CSK_STATE_DEAD;
1710 		cxgbit_put_csk(csk);
1711 		break;
1712 	case CSK_STATE_ABORTING:
1713 	case CSK_STATE_DEAD:
1714 		break;
1715 	default:
1716 		pr_info("%s: cpl_close_con_rpl in bad state %d\n",
1717 			__func__, csk->com.state);
1718 	}
1719 
1720 	__kfree_skb(skb);
1721 }
1722 
1723 static void cxgbit_abort_req_rss(struct cxgbit_sock *csk, struct sk_buff *skb)
1724 {
1725 	struct cpl_abort_req_rss *hdr = cplhdr(skb);
1726 	unsigned int tid = GET_TID(hdr);
1727 	struct sk_buff *rpl_skb;
1728 	bool release = false;
1729 	bool wakeup_thread = false;
1730 	u32 len = roundup(sizeof(struct cpl_abort_rpl), 16);
1731 
1732 	pr_debug("%s: csk %p; tid %u; state %d\n",
1733 		 __func__, csk, tid, csk->com.state);
1734 
1735 	if (cxgb_is_neg_adv(hdr->status)) {
1736 		pr_err("%s: got neg advise %d on tid %u\n",
1737 		       __func__, hdr->status, tid);
1738 		goto rel_skb;
1739 	}
1740 
1741 	switch (csk->com.state) {
1742 	case CSK_STATE_CONNECTING:
1743 	case CSK_STATE_MORIBUND:
1744 		csk->com.state = CSK_STATE_DEAD;
1745 		release = true;
1746 		break;
1747 	case CSK_STATE_ESTABLISHED:
1748 		csk->com.state = CSK_STATE_DEAD;
1749 		wakeup_thread = true;
1750 		break;
1751 	case CSK_STATE_CLOSING:
1752 		csk->com.state = CSK_STATE_DEAD;
1753 		if (!csk->conn)
1754 			release = true;
1755 		break;
1756 	case CSK_STATE_ABORTING:
1757 		break;
1758 	default:
1759 		pr_info("%s: cpl_abort_req_rss in bad state %d\n",
1760 			__func__, csk->com.state);
1761 		csk->com.state = CSK_STATE_DEAD;
1762 	}
1763 
1764 	__skb_queue_purge(&csk->txq);
1765 
1766 	if (!test_and_set_bit(CSK_TX_DATA_SENT, &csk->com.flags))
1767 		cxgbit_send_tx_flowc_wr(csk);
1768 
1769 	rpl_skb = __skb_dequeue(&csk->skbq);
1770 
1771 	cxgb_mk_abort_rpl(rpl_skb, len, csk->tid, csk->txq_idx);
1772 	cxgbit_ofld_send(csk->com.cdev, rpl_skb);
1773 
1774 	if (wakeup_thread) {
1775 		cxgbit_queue_rx_skb(csk, skb);
1776 		return;
1777 	}
1778 
1779 	if (release)
1780 		cxgbit_put_csk(csk);
1781 rel_skb:
1782 	__kfree_skb(skb);
1783 }
1784 
1785 static void cxgbit_abort_rpl_rss(struct cxgbit_sock *csk, struct sk_buff *skb)
1786 {
1787 	struct cpl_abort_rpl_rss *rpl = cplhdr(skb);
1788 
1789 	pr_debug("%s: csk %p; tid %u; state %d\n",
1790 		 __func__, csk, csk->tid, csk->com.state);
1791 
1792 	switch (csk->com.state) {
1793 	case CSK_STATE_ABORTING:
1794 		csk->com.state = CSK_STATE_DEAD;
1795 		if (test_bit(CSK_ABORT_RPL_WAIT, &csk->com.flags))
1796 			cxgbit_wake_up(&csk->com.wr_wait, __func__,
1797 				       rpl->status);
1798 		cxgbit_put_csk(csk);
1799 		break;
1800 	default:
1801 		pr_info("%s: cpl_abort_rpl_rss in state %d\n",
1802 			__func__, csk->com.state);
1803 	}
1804 
1805 	__kfree_skb(skb);
1806 }
1807 
1808 static bool cxgbit_credit_err(const struct cxgbit_sock *csk)
1809 {
1810 	const struct sk_buff *skb = csk->wr_pending_head;
1811 	u32 credit = 0;
1812 
1813 	if (unlikely(csk->wr_cred > csk->wr_max_cred)) {
1814 		pr_err("csk 0x%p, tid %u, credit %u > %u\n",
1815 		       csk, csk->tid, csk->wr_cred, csk->wr_max_cred);
1816 		return true;
1817 	}
1818 
1819 	while (skb) {
1820 		credit += (__force u32)skb->csum;
1821 		skb = cxgbit_skcb_tx_wr_next(skb);
1822 	}
1823 
1824 	if (unlikely((csk->wr_cred + credit) != csk->wr_max_cred)) {
1825 		pr_err("csk 0x%p, tid %u, credit %u + %u != %u.\n",
1826 		       csk, csk->tid, csk->wr_cred,
1827 		       credit, csk->wr_max_cred);
1828 
1829 		return true;
1830 	}
1831 
1832 	return false;
1833 }
1834 
1835 static void cxgbit_fw4_ack(struct cxgbit_sock *csk, struct sk_buff *skb)
1836 {
1837 	struct cpl_fw4_ack *rpl = (struct cpl_fw4_ack *)cplhdr(skb);
1838 	u32 credits = rpl->credits;
1839 	u32 snd_una = ntohl(rpl->snd_una);
1840 
1841 	csk->wr_cred += credits;
1842 	if (csk->wr_una_cred > (csk->wr_max_cred - csk->wr_cred))
1843 		csk->wr_una_cred = csk->wr_max_cred - csk->wr_cred;
1844 
1845 	while (credits) {
1846 		struct sk_buff *p = cxgbit_sock_peek_wr(csk);
1847 		u32 csum;
1848 
1849 		if (unlikely(!p)) {
1850 			pr_err("csk 0x%p,%u, cr %u,%u+%u, empty.\n",
1851 			       csk, csk->tid, credits,
1852 			       csk->wr_cred, csk->wr_una_cred);
1853 			break;
1854 		}
1855 
1856 		csum = (__force u32)p->csum;
1857 		if (unlikely(credits < csum)) {
1858 			pr_warn("csk 0x%p,%u, cr %u,%u+%u, < %u.\n",
1859 				csk,  csk->tid,
1860 				credits, csk->wr_cred, csk->wr_una_cred,
1861 				csum);
1862 			p->csum = (__force __wsum)(csum - credits);
1863 			break;
1864 		}
1865 
1866 		cxgbit_sock_dequeue_wr(csk);
1867 		credits -= csum;
1868 		kfree_skb(p);
1869 	}
1870 
1871 	if (unlikely(cxgbit_credit_err(csk))) {
1872 		cxgbit_queue_rx_skb(csk, skb);
1873 		return;
1874 	}
1875 
1876 	if (rpl->seq_vld & CPL_FW4_ACK_FLAGS_SEQVAL) {
1877 		if (unlikely(before(snd_una, csk->snd_una))) {
1878 			pr_warn("csk 0x%p,%u, snd_una %u/%u.",
1879 				csk, csk->tid, snd_una,
1880 				csk->snd_una);
1881 			goto rel_skb;
1882 		}
1883 
1884 		if (csk->snd_una != snd_una) {
1885 			csk->snd_una = snd_una;
1886 			dst_confirm(csk->dst);
1887 		}
1888 	}
1889 
1890 	if (skb_queue_len(&csk->txq))
1891 		cxgbit_push_tx_frames(csk);
1892 
1893 rel_skb:
1894 	__kfree_skb(skb);
1895 }
1896 
1897 static void cxgbit_set_tcb_rpl(struct cxgbit_device *cdev, struct sk_buff *skb)
1898 {
1899 	struct cxgbit_sock *csk;
1900 	struct cpl_set_tcb_rpl *rpl = (struct cpl_set_tcb_rpl *)skb->data;
1901 	unsigned int tid = GET_TID(rpl);
1902 	struct cxgb4_lld_info *lldi = &cdev->lldi;
1903 	struct tid_info *t = lldi->tids;
1904 
1905 	csk = lookup_tid(t, tid);
1906 	if (unlikely(!csk)) {
1907 		pr_err("can't find connection for tid %u.\n", tid);
1908 		goto rel_skb;
1909 	} else {
1910 		cxgbit_wake_up(&csk->com.wr_wait, __func__, rpl->status);
1911 	}
1912 
1913 	cxgbit_put_csk(csk);
1914 rel_skb:
1915 	__kfree_skb(skb);
1916 }
1917 
1918 static void cxgbit_rx_data(struct cxgbit_device *cdev, struct sk_buff *skb)
1919 {
1920 	struct cxgbit_sock *csk;
1921 	struct cpl_rx_data *cpl = cplhdr(skb);
1922 	unsigned int tid = GET_TID(cpl);
1923 	struct cxgb4_lld_info *lldi = &cdev->lldi;
1924 	struct tid_info *t = lldi->tids;
1925 
1926 	csk = lookup_tid(t, tid);
1927 	if (unlikely(!csk)) {
1928 		pr_err("can't find conn. for tid %u.\n", tid);
1929 		goto rel_skb;
1930 	}
1931 
1932 	cxgbit_queue_rx_skb(csk, skb);
1933 	return;
1934 rel_skb:
1935 	__kfree_skb(skb);
1936 }
1937 
1938 static void
1939 __cxgbit_process_rx_cpl(struct cxgbit_sock *csk, struct sk_buff *skb)
1940 {
1941 	spin_lock(&csk->lock);
1942 	if (csk->lock_owner) {
1943 		__skb_queue_tail(&csk->backlogq, skb);
1944 		spin_unlock(&csk->lock);
1945 		return;
1946 	}
1947 
1948 	cxgbit_skcb_rx_backlog_fn(skb)(csk, skb);
1949 	spin_unlock(&csk->lock);
1950 }
1951 
1952 static void cxgbit_process_rx_cpl(struct cxgbit_sock *csk, struct sk_buff *skb)
1953 {
1954 	cxgbit_get_csk(csk);
1955 	__cxgbit_process_rx_cpl(csk, skb);
1956 	cxgbit_put_csk(csk);
1957 }
1958 
1959 static void cxgbit_rx_cpl(struct cxgbit_device *cdev, struct sk_buff *skb)
1960 {
1961 	struct cxgbit_sock *csk;
1962 	struct cpl_tx_data *cpl = cplhdr(skb);
1963 	struct cxgb4_lld_info *lldi = &cdev->lldi;
1964 	struct tid_info *t = lldi->tids;
1965 	unsigned int tid = GET_TID(cpl);
1966 	u8 opcode = cxgbit_skcb_rx_opcode(skb);
1967 	bool ref = true;
1968 
1969 	switch (opcode) {
1970 	case CPL_FW4_ACK:
1971 			cxgbit_skcb_rx_backlog_fn(skb) = cxgbit_fw4_ack;
1972 			ref = false;
1973 			break;
1974 	case CPL_PEER_CLOSE:
1975 			cxgbit_skcb_rx_backlog_fn(skb) = cxgbit_peer_close;
1976 			break;
1977 	case CPL_CLOSE_CON_RPL:
1978 			cxgbit_skcb_rx_backlog_fn(skb) = cxgbit_close_con_rpl;
1979 			break;
1980 	case CPL_ABORT_REQ_RSS:
1981 			cxgbit_skcb_rx_backlog_fn(skb) = cxgbit_abort_req_rss;
1982 			break;
1983 	case CPL_ABORT_RPL_RSS:
1984 			cxgbit_skcb_rx_backlog_fn(skb) = cxgbit_abort_rpl_rss;
1985 			break;
1986 	default:
1987 		goto rel_skb;
1988 	}
1989 
1990 	csk = lookup_tid(t, tid);
1991 	if (unlikely(!csk)) {
1992 		pr_err("can't find conn. for tid %u.\n", tid);
1993 		goto rel_skb;
1994 	}
1995 
1996 	if (ref)
1997 		cxgbit_process_rx_cpl(csk, skb);
1998 	else
1999 		__cxgbit_process_rx_cpl(csk, skb);
2000 
2001 	return;
2002 rel_skb:
2003 	__kfree_skb(skb);
2004 }
2005 
2006 cxgbit_cplhandler_func cxgbit_cplhandlers[NUM_CPL_CMDS] = {
2007 	[CPL_PASS_OPEN_RPL]	= cxgbit_pass_open_rpl,
2008 	[CPL_CLOSE_LISTSRV_RPL] = cxgbit_close_listsrv_rpl,
2009 	[CPL_PASS_ACCEPT_REQ]	= cxgbit_pass_accept_req,
2010 	[CPL_PASS_ESTABLISH]	= cxgbit_pass_establish,
2011 	[CPL_SET_TCB_RPL]	= cxgbit_set_tcb_rpl,
2012 	[CPL_RX_DATA]		= cxgbit_rx_data,
2013 	[CPL_FW4_ACK]		= cxgbit_rx_cpl,
2014 	[CPL_PEER_CLOSE]	= cxgbit_rx_cpl,
2015 	[CPL_CLOSE_CON_RPL]	= cxgbit_rx_cpl,
2016 	[CPL_ABORT_REQ_RSS]	= cxgbit_rx_cpl,
2017 	[CPL_ABORT_RPL_RSS]	= cxgbit_rx_cpl,
2018 };
2019