xref: /openbmc/linux/drivers/infiniband/hw/cxgb4/cm.c (revision 0e6774ec)
1 /*
2  * Copyright (c) 2009-2014 Chelsio, Inc. All rights reserved.
3  *
4  * This software is available to you under a choice of one of two
5  * licenses.  You may choose to be licensed under the terms of the GNU
6  * General Public License (GPL) Version 2, available from the file
7  * COPYING in the main directory of this source tree, or the
8  * OpenIB.org BSD license below:
9  *
10  *     Redistribution and use in source and binary forms, with or
11  *     without modification, are permitted provided that the following
12  *     conditions are met:
13  *
14  *      - Redistributions of source code must retain the above
15  *	  copyright notice, this list of conditions and the following
16  *	  disclaimer.
17  *
18  *      - Redistributions in binary form must reproduce the above
19  *	  copyright notice, this list of conditions and the following
20  *	  disclaimer in the documentation and/or other materials
21  *	  provided with the distribution.
22  *
23  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
24  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
25  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
26  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
27  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
28  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
29  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
30  * SOFTWARE.
31  */
32 #include <linux/module.h>
33 #include <linux/list.h>
34 #include <linux/workqueue.h>
35 #include <linux/skbuff.h>
36 #include <linux/timer.h>
37 #include <linux/notifier.h>
38 #include <linux/inetdevice.h>
39 #include <linux/ip.h>
40 #include <linux/tcp.h>
41 #include <linux/if_vlan.h>
42 
43 #include <net/neighbour.h>
44 #include <net/netevent.h>
45 #include <net/route.h>
46 #include <net/tcp.h>
47 #include <net/ip6_route.h>
48 #include <net/addrconf.h>
49 
50 #include <rdma/ib_addr.h>
51 
52 #include <libcxgb_cm.h>
53 #include "iw_cxgb4.h"
54 #include "clip_tbl.h"
55 
56 static char *states[] = {
57 	"idle",
58 	"listen",
59 	"connecting",
60 	"mpa_wait_req",
61 	"mpa_req_sent",
62 	"mpa_req_rcvd",
63 	"mpa_rep_sent",
64 	"fpdu_mode",
65 	"aborting",
66 	"closing",
67 	"moribund",
68 	"dead",
69 	NULL,
70 };
71 
72 static int nocong;
73 module_param(nocong, int, 0644);
74 MODULE_PARM_DESC(nocong, "Turn of congestion control (default=0)");
75 
76 static int enable_ecn;
77 module_param(enable_ecn, int, 0644);
78 MODULE_PARM_DESC(enable_ecn, "Enable ECN (default=0/disabled)");
79 
80 static int dack_mode;
81 module_param(dack_mode, int, 0644);
82 MODULE_PARM_DESC(dack_mode, "Delayed ack mode (default=0)");
83 
84 uint c4iw_max_read_depth = 32;
85 module_param(c4iw_max_read_depth, int, 0644);
86 MODULE_PARM_DESC(c4iw_max_read_depth,
87 		 "Per-connection max ORD/IRD (default=32)");
88 
89 static int enable_tcp_timestamps;
90 module_param(enable_tcp_timestamps, int, 0644);
91 MODULE_PARM_DESC(enable_tcp_timestamps, "Enable tcp timestamps (default=0)");
92 
93 static int enable_tcp_sack;
94 module_param(enable_tcp_sack, int, 0644);
95 MODULE_PARM_DESC(enable_tcp_sack, "Enable tcp SACK (default=0)");
96 
97 static int enable_tcp_window_scaling = 1;
98 module_param(enable_tcp_window_scaling, int, 0644);
99 MODULE_PARM_DESC(enable_tcp_window_scaling,
100 		 "Enable tcp window scaling (default=1)");
101 
102 static int peer2peer = 1;
103 module_param(peer2peer, int, 0644);
104 MODULE_PARM_DESC(peer2peer, "Support peer2peer ULPs (default=1)");
105 
106 static int p2p_type = FW_RI_INIT_P2PTYPE_READ_REQ;
107 module_param(p2p_type, int, 0644);
108 MODULE_PARM_DESC(p2p_type, "RDMAP opcode to use for the RTR message: "
109 			   "1=RDMA_READ 0=RDMA_WRITE (default 1)");
110 
111 static int ep_timeout_secs = 60;
112 module_param(ep_timeout_secs, int, 0644);
113 MODULE_PARM_DESC(ep_timeout_secs, "CM Endpoint operation timeout "
114 				   "in seconds (default=60)");
115 
116 static int mpa_rev = 2;
117 module_param(mpa_rev, int, 0644);
118 MODULE_PARM_DESC(mpa_rev, "MPA Revision, 0 supports amso1100, "
119 		"1 is RFC5044 spec compliant, 2 is IETF MPA Peer Connect Draft"
120 		" compliant (default=2)");
121 
122 static int markers_enabled;
123 module_param(markers_enabled, int, 0644);
124 MODULE_PARM_DESC(markers_enabled, "Enable MPA MARKERS (default(0)=disabled)");
125 
126 static int crc_enabled = 1;
127 module_param(crc_enabled, int, 0644);
128 MODULE_PARM_DESC(crc_enabled, "Enable MPA CRC (default(1)=enabled)");
129 
130 static int rcv_win = 256 * 1024;
131 module_param(rcv_win, int, 0644);
132 MODULE_PARM_DESC(rcv_win, "TCP receive window in bytes (default=256KB)");
133 
134 static int snd_win = 128 * 1024;
135 module_param(snd_win, int, 0644);
136 MODULE_PARM_DESC(snd_win, "TCP send window in bytes (default=128KB)");
137 
138 static struct workqueue_struct *workq;
139 
140 static struct sk_buff_head rxq;
141 
142 static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp);
143 static void ep_timeout(struct timer_list *t);
144 static void connect_reply_upcall(struct c4iw_ep *ep, int status);
145 static int sched(struct c4iw_dev *dev, struct sk_buff *skb);
146 
147 static LIST_HEAD(timeout_list);
148 static DEFINE_SPINLOCK(timeout_lock);
149 
150 static void deref_cm_id(struct c4iw_ep_common *epc)
151 {
152 	epc->cm_id->rem_ref(epc->cm_id);
153 	epc->cm_id = NULL;
154 	set_bit(CM_ID_DEREFED, &epc->history);
155 }
156 
157 static void ref_cm_id(struct c4iw_ep_common *epc)
158 {
159 	set_bit(CM_ID_REFED, &epc->history);
160 	epc->cm_id->add_ref(epc->cm_id);
161 }
162 
163 static void deref_qp(struct c4iw_ep *ep)
164 {
165 	c4iw_qp_rem_ref(&ep->com.qp->ibqp);
166 	clear_bit(QP_REFERENCED, &ep->com.flags);
167 	set_bit(QP_DEREFED, &ep->com.history);
168 }
169 
170 static void ref_qp(struct c4iw_ep *ep)
171 {
172 	set_bit(QP_REFERENCED, &ep->com.flags);
173 	set_bit(QP_REFED, &ep->com.history);
174 	c4iw_qp_add_ref(&ep->com.qp->ibqp);
175 }
176 
177 static void start_ep_timer(struct c4iw_ep *ep)
178 {
179 	pr_debug("ep %p\n", ep);
180 	if (timer_pending(&ep->timer)) {
181 		pr_err("%s timer already started! ep %p\n",
182 		       __func__, ep);
183 		return;
184 	}
185 	clear_bit(TIMEOUT, &ep->com.flags);
186 	c4iw_get_ep(&ep->com);
187 	ep->timer.expires = jiffies + ep_timeout_secs * HZ;
188 	add_timer(&ep->timer);
189 }
190 
191 static int stop_ep_timer(struct c4iw_ep *ep)
192 {
193 	pr_debug("ep %p stopping\n", ep);
194 	del_timer_sync(&ep->timer);
195 	if (!test_and_set_bit(TIMEOUT, &ep->com.flags)) {
196 		c4iw_put_ep(&ep->com);
197 		return 0;
198 	}
199 	return 1;
200 }
201 
202 static int c4iw_l2t_send(struct c4iw_rdev *rdev, struct sk_buff *skb,
203 		  struct l2t_entry *l2e)
204 {
205 	int	error = 0;
206 
207 	if (c4iw_fatal_error(rdev)) {
208 		kfree_skb(skb);
209 		pr_err("%s - device in error state - dropping\n", __func__);
210 		return -EIO;
211 	}
212 	error = cxgb4_l2t_send(rdev->lldi.ports[0], skb, l2e);
213 	if (error < 0)
214 		kfree_skb(skb);
215 	else if (error == NET_XMIT_DROP)
216 		return -ENOMEM;
217 	return error < 0 ? error : 0;
218 }
219 
220 int c4iw_ofld_send(struct c4iw_rdev *rdev, struct sk_buff *skb)
221 {
222 	int	error = 0;
223 
224 	if (c4iw_fatal_error(rdev)) {
225 		kfree_skb(skb);
226 		pr_err("%s - device in error state - dropping\n", __func__);
227 		return -EIO;
228 	}
229 	error = cxgb4_ofld_send(rdev->lldi.ports[0], skb);
230 	if (error < 0)
231 		kfree_skb(skb);
232 	return error < 0 ? error : 0;
233 }
234 
235 static void release_tid(struct c4iw_rdev *rdev, u32 hwtid, struct sk_buff *skb)
236 {
237 	u32 len = roundup(sizeof(struct cpl_tid_release), 16);
238 
239 	skb = get_skb(skb, len, GFP_KERNEL);
240 	if (!skb)
241 		return;
242 
243 	cxgb_mk_tid_release(skb, len, hwtid, 0);
244 	c4iw_ofld_send(rdev, skb);
245 	return;
246 }
247 
248 static void set_emss(struct c4iw_ep *ep, u16 opt)
249 {
250 	ep->emss = ep->com.dev->rdev.lldi.mtus[TCPOPT_MSS_G(opt)] -
251 		   ((AF_INET == ep->com.remote_addr.ss_family) ?
252 		    sizeof(struct iphdr) : sizeof(struct ipv6hdr)) -
253 		   sizeof(struct tcphdr);
254 	ep->mss = ep->emss;
255 	if (TCPOPT_TSTAMP_G(opt))
256 		ep->emss -= round_up(TCPOLEN_TIMESTAMP, 4);
257 	if (ep->emss < 128)
258 		ep->emss = 128;
259 	if (ep->emss & 7)
260 		pr_debug("Warning: misaligned mtu idx %u mss %u emss=%u\n",
261 			 TCPOPT_MSS_G(opt), ep->mss, ep->emss);
262 	pr_debug("mss_idx %u mss %u emss=%u\n", TCPOPT_MSS_G(opt), ep->mss,
263 		 ep->emss);
264 }
265 
266 static enum c4iw_ep_state state_read(struct c4iw_ep_common *epc)
267 {
268 	enum c4iw_ep_state state;
269 
270 	mutex_lock(&epc->mutex);
271 	state = epc->state;
272 	mutex_unlock(&epc->mutex);
273 	return state;
274 }
275 
276 static void __state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state new)
277 {
278 	epc->state = new;
279 }
280 
281 static void state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state new)
282 {
283 	mutex_lock(&epc->mutex);
284 	pr_debug("%s -> %s\n", states[epc->state], states[new]);
285 	__state_set(epc, new);
286 	mutex_unlock(&epc->mutex);
287 	return;
288 }
289 
290 static int alloc_ep_skb_list(struct sk_buff_head *ep_skb_list, int size)
291 {
292 	struct sk_buff *skb;
293 	unsigned int i;
294 	size_t len;
295 
296 	len = roundup(sizeof(union cpl_wr_size), 16);
297 	for (i = 0; i < size; i++) {
298 		skb = alloc_skb(len, GFP_KERNEL);
299 		if (!skb)
300 			goto fail;
301 		skb_queue_tail(ep_skb_list, skb);
302 	}
303 	return 0;
304 fail:
305 	skb_queue_purge(ep_skb_list);
306 	return -ENOMEM;
307 }
308 
309 static void *alloc_ep(int size, gfp_t gfp)
310 {
311 	struct c4iw_ep_common *epc;
312 
313 	epc = kzalloc(size, gfp);
314 	if (epc) {
315 		epc->wr_waitp = c4iw_alloc_wr_wait(gfp);
316 		if (!epc->wr_waitp) {
317 			kfree(epc);
318 			epc = NULL;
319 			goto out;
320 		}
321 		kref_init(&epc->kref);
322 		mutex_init(&epc->mutex);
323 		c4iw_init_wr_wait(epc->wr_waitp);
324 	}
325 	pr_debug("alloc ep %p\n", epc);
326 out:
327 	return epc;
328 }
329 
330 static void remove_ep_tid(struct c4iw_ep *ep)
331 {
332 	unsigned long flags;
333 
334 	xa_lock_irqsave(&ep->com.dev->hwtids, flags);
335 	__xa_erase(&ep->com.dev->hwtids, ep->hwtid);
336 	if (xa_empty(&ep->com.dev->hwtids))
337 		wake_up(&ep->com.dev->wait);
338 	xa_unlock_irqrestore(&ep->com.dev->hwtids, flags);
339 }
340 
341 static int insert_ep_tid(struct c4iw_ep *ep)
342 {
343 	unsigned long flags;
344 	int err;
345 
346 	xa_lock_irqsave(&ep->com.dev->hwtids, flags);
347 	err = __xa_insert(&ep->com.dev->hwtids, ep->hwtid, ep, GFP_KERNEL);
348 	xa_unlock_irqrestore(&ep->com.dev->hwtids, flags);
349 
350 	return err;
351 }
352 
353 /*
354  * Atomically lookup the ep ptr given the tid and grab a reference on the ep.
355  */
356 static struct c4iw_ep *get_ep_from_tid(struct c4iw_dev *dev, unsigned int tid)
357 {
358 	struct c4iw_ep *ep;
359 	unsigned long flags;
360 
361 	xa_lock_irqsave(&dev->hwtids, flags);
362 	ep = xa_load(&dev->hwtids, tid);
363 	if (ep)
364 		c4iw_get_ep(&ep->com);
365 	xa_unlock_irqrestore(&dev->hwtids, flags);
366 	return ep;
367 }
368 
369 /*
370  * Atomically lookup the ep ptr given the stid and grab a reference on the ep.
371  */
372 static struct c4iw_listen_ep *get_ep_from_stid(struct c4iw_dev *dev,
373 					       unsigned int stid)
374 {
375 	struct c4iw_listen_ep *ep;
376 	unsigned long flags;
377 
378 	xa_lock_irqsave(&dev->stids, flags);
379 	ep = xa_load(&dev->stids, stid);
380 	if (ep)
381 		c4iw_get_ep(&ep->com);
382 	xa_unlock_irqrestore(&dev->stids, flags);
383 	return ep;
384 }
385 
386 void _c4iw_free_ep(struct kref *kref)
387 {
388 	struct c4iw_ep *ep;
389 
390 	ep = container_of(kref, struct c4iw_ep, com.kref);
391 	pr_debug("ep %p state %s\n", ep, states[ep->com.state]);
392 	if (test_bit(QP_REFERENCED, &ep->com.flags))
393 		deref_qp(ep);
394 	if (test_bit(RELEASE_RESOURCES, &ep->com.flags)) {
395 		if (ep->com.remote_addr.ss_family == AF_INET6) {
396 			struct sockaddr_in6 *sin6 =
397 					(struct sockaddr_in6 *)
398 					&ep->com.local_addr;
399 
400 			cxgb4_clip_release(
401 					ep->com.dev->rdev.lldi.ports[0],
402 					(const u32 *)&sin6->sin6_addr.s6_addr,
403 					1);
404 		}
405 		cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, ep->hwtid,
406 				 ep->com.local_addr.ss_family);
407 		dst_release(ep->dst);
408 		cxgb4_l2t_release(ep->l2t);
409 		kfree_skb(ep->mpa_skb);
410 	}
411 	if (!skb_queue_empty(&ep->com.ep_skb_list))
412 		skb_queue_purge(&ep->com.ep_skb_list);
413 	c4iw_put_wr_wait(ep->com.wr_waitp);
414 	kfree(ep);
415 }
416 
417 static void release_ep_resources(struct c4iw_ep *ep)
418 {
419 	set_bit(RELEASE_RESOURCES, &ep->com.flags);
420 
421 	/*
422 	 * If we have a hwtid, then remove it from the idr table
423 	 * so lookups will no longer find this endpoint.  Otherwise
424 	 * we have a race where one thread finds the ep ptr just
425 	 * before the other thread is freeing the ep memory.
426 	 */
427 	if (ep->hwtid != -1)
428 		remove_ep_tid(ep);
429 	c4iw_put_ep(&ep->com);
430 }
431 
432 static int status2errno(int status)
433 {
434 	switch (status) {
435 	case CPL_ERR_NONE:
436 		return 0;
437 	case CPL_ERR_CONN_RESET:
438 		return -ECONNRESET;
439 	case CPL_ERR_ARP_MISS:
440 		return -EHOSTUNREACH;
441 	case CPL_ERR_CONN_TIMEDOUT:
442 		return -ETIMEDOUT;
443 	case CPL_ERR_TCAM_FULL:
444 		return -ENOMEM;
445 	case CPL_ERR_CONN_EXIST:
446 		return -EADDRINUSE;
447 	default:
448 		return -EIO;
449 	}
450 }
451 
452 /*
453  * Try and reuse skbs already allocated...
454  */
455 static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp)
456 {
457 	if (skb && !skb_is_nonlinear(skb) && !skb_cloned(skb)) {
458 		skb_trim(skb, 0);
459 		skb_get(skb);
460 		skb_reset_transport_header(skb);
461 	} else {
462 		skb = alloc_skb(len, gfp);
463 		if (!skb)
464 			return NULL;
465 	}
466 	t4_set_arp_err_handler(skb, NULL, NULL);
467 	return skb;
468 }
469 
470 static struct net_device *get_real_dev(struct net_device *egress_dev)
471 {
472 	return rdma_vlan_dev_real_dev(egress_dev) ? : egress_dev;
473 }
474 
475 static void arp_failure_discard(void *handle, struct sk_buff *skb)
476 {
477 	pr_err("ARP failure\n");
478 	kfree_skb(skb);
479 }
480 
481 static void mpa_start_arp_failure(void *handle, struct sk_buff *skb)
482 {
483 	pr_err("ARP failure during MPA Negotiation - Closing Connection\n");
484 }
485 
486 enum {
487 	NUM_FAKE_CPLS = 2,
488 	FAKE_CPL_PUT_EP_SAFE = NUM_CPL_CMDS + 0,
489 	FAKE_CPL_PASS_PUT_EP_SAFE = NUM_CPL_CMDS + 1,
490 };
491 
492 static int _put_ep_safe(struct c4iw_dev *dev, struct sk_buff *skb)
493 {
494 	struct c4iw_ep *ep;
495 
496 	ep = *((struct c4iw_ep **)(skb->cb + 2 * sizeof(void *)));
497 	release_ep_resources(ep);
498 	return 0;
499 }
500 
501 static int _put_pass_ep_safe(struct c4iw_dev *dev, struct sk_buff *skb)
502 {
503 	struct c4iw_ep *ep;
504 
505 	ep = *((struct c4iw_ep **)(skb->cb + 2 * sizeof(void *)));
506 	c4iw_put_ep(&ep->parent_ep->com);
507 	release_ep_resources(ep);
508 	return 0;
509 }
510 
511 /*
512  * Fake up a special CPL opcode and call sched() so process_work() will call
513  * _put_ep_safe() in a safe context to free the ep resources.  This is needed
514  * because ARP error handlers are called in an ATOMIC context, and
515  * _c4iw_free_ep() needs to block.
516  */
517 static void queue_arp_failure_cpl(struct c4iw_ep *ep, struct sk_buff *skb,
518 				  int cpl)
519 {
520 	struct cpl_act_establish *rpl = cplhdr(skb);
521 
522 	/* Set our special ARP_FAILURE opcode */
523 	rpl->ot.opcode = cpl;
524 
525 	/*
526 	 * Save ep in the skb->cb area, after where sched() will save the dev
527 	 * ptr.
528 	 */
529 	*((struct c4iw_ep **)(skb->cb + 2 * sizeof(void *))) = ep;
530 	sched(ep->com.dev, skb);
531 }
532 
533 /* Handle an ARP failure for an accept */
534 static void pass_accept_rpl_arp_failure(void *handle, struct sk_buff *skb)
535 {
536 	struct c4iw_ep *ep = handle;
537 
538 	pr_err("ARP failure during accept - tid %u - dropping connection\n",
539 	       ep->hwtid);
540 
541 	__state_set(&ep->com, DEAD);
542 	queue_arp_failure_cpl(ep, skb, FAKE_CPL_PASS_PUT_EP_SAFE);
543 }
544 
545 /*
546  * Handle an ARP failure for an active open.
547  */
548 static void act_open_req_arp_failure(void *handle, struct sk_buff *skb)
549 {
550 	struct c4iw_ep *ep = handle;
551 
552 	pr_err("ARP failure during connect\n");
553 	connect_reply_upcall(ep, -EHOSTUNREACH);
554 	__state_set(&ep->com, DEAD);
555 	if (ep->com.remote_addr.ss_family == AF_INET6) {
556 		struct sockaddr_in6 *sin6 =
557 			(struct sockaddr_in6 *)&ep->com.local_addr;
558 		cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
559 				   (const u32 *)&sin6->sin6_addr.s6_addr, 1);
560 	}
561 	xa_erase_irq(&ep->com.dev->atids, ep->atid);
562 	cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
563 	queue_arp_failure_cpl(ep, skb, FAKE_CPL_PUT_EP_SAFE);
564 }
565 
566 /*
567  * Handle an ARP failure for a CPL_ABORT_REQ.  Change it into a no RST variant
568  * and send it along.
569  */
570 static void abort_arp_failure(void *handle, struct sk_buff *skb)
571 {
572 	int ret;
573 	struct c4iw_ep *ep = handle;
574 	struct c4iw_rdev *rdev = &ep->com.dev->rdev;
575 	struct cpl_abort_req *req = cplhdr(skb);
576 
577 	pr_debug("rdev %p\n", rdev);
578 	req->cmd = CPL_ABORT_NO_RST;
579 	skb_get(skb);
580 	ret = c4iw_ofld_send(rdev, skb);
581 	if (ret) {
582 		__state_set(&ep->com, DEAD);
583 		queue_arp_failure_cpl(ep, skb, FAKE_CPL_PUT_EP_SAFE);
584 	} else
585 		kfree_skb(skb);
586 }
587 
588 static int send_flowc(struct c4iw_ep *ep)
589 {
590 	struct fw_flowc_wr *flowc;
591 	struct sk_buff *skb = skb_dequeue(&ep->com.ep_skb_list);
592 	u16 vlan = ep->l2t->vlan;
593 	int nparams;
594 	int flowclen, flowclen16;
595 
596 	if (WARN_ON(!skb))
597 		return -ENOMEM;
598 
599 	if (vlan == CPL_L2T_VLAN_NONE)
600 		nparams = 9;
601 	else
602 		nparams = 10;
603 
604 	flowclen = offsetof(struct fw_flowc_wr, mnemval[nparams]);
605 	flowclen16 = DIV_ROUND_UP(flowclen, 16);
606 	flowclen = flowclen16 * 16;
607 
608 	flowc = __skb_put(skb, flowclen);
609 	memset(flowc, 0, flowclen);
610 
611 	flowc->op_to_nparams = cpu_to_be32(FW_WR_OP_V(FW_FLOWC_WR) |
612 					   FW_FLOWC_WR_NPARAMS_V(nparams));
613 	flowc->flowid_len16 = cpu_to_be32(FW_WR_LEN16_V(flowclen16) |
614 					  FW_WR_FLOWID_V(ep->hwtid));
615 
616 	flowc->mnemval[0].mnemonic = FW_FLOWC_MNEM_PFNVFN;
617 	flowc->mnemval[0].val = cpu_to_be32(FW_PFVF_CMD_PFN_V
618 					    (ep->com.dev->rdev.lldi.pf));
619 	flowc->mnemval[1].mnemonic = FW_FLOWC_MNEM_CH;
620 	flowc->mnemval[1].val = cpu_to_be32(ep->tx_chan);
621 	flowc->mnemval[2].mnemonic = FW_FLOWC_MNEM_PORT;
622 	flowc->mnemval[2].val = cpu_to_be32(ep->tx_chan);
623 	flowc->mnemval[3].mnemonic = FW_FLOWC_MNEM_IQID;
624 	flowc->mnemval[3].val = cpu_to_be32(ep->rss_qid);
625 	flowc->mnemval[4].mnemonic = FW_FLOWC_MNEM_SNDNXT;
626 	flowc->mnemval[4].val = cpu_to_be32(ep->snd_seq);
627 	flowc->mnemval[5].mnemonic = FW_FLOWC_MNEM_RCVNXT;
628 	flowc->mnemval[5].val = cpu_to_be32(ep->rcv_seq);
629 	flowc->mnemval[6].mnemonic = FW_FLOWC_MNEM_SNDBUF;
630 	flowc->mnemval[6].val = cpu_to_be32(ep->snd_win);
631 	flowc->mnemval[7].mnemonic = FW_FLOWC_MNEM_MSS;
632 	flowc->mnemval[7].val = cpu_to_be32(ep->emss);
633 	flowc->mnemval[8].mnemonic = FW_FLOWC_MNEM_RCV_SCALE;
634 	flowc->mnemval[8].val = cpu_to_be32(ep->snd_wscale);
635 	if (nparams == 10) {
636 		u16 pri;
637 		pri = (vlan & VLAN_PRIO_MASK) >> VLAN_PRIO_SHIFT;
638 		flowc->mnemval[9].mnemonic = FW_FLOWC_MNEM_SCHEDCLASS;
639 		flowc->mnemval[9].val = cpu_to_be32(pri);
640 	}
641 
642 	set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
643 	return c4iw_ofld_send(&ep->com.dev->rdev, skb);
644 }
645 
646 static int send_halfclose(struct c4iw_ep *ep)
647 {
648 	struct sk_buff *skb = skb_dequeue(&ep->com.ep_skb_list);
649 	u32 wrlen = roundup(sizeof(struct cpl_close_con_req), 16);
650 
651 	pr_debug("ep %p tid %u\n", ep, ep->hwtid);
652 	if (WARN_ON(!skb))
653 		return -ENOMEM;
654 
655 	cxgb_mk_close_con_req(skb, wrlen, ep->hwtid, ep->txq_idx,
656 			      NULL, arp_failure_discard);
657 
658 	return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
659 }
660 
661 static void read_tcb(struct c4iw_ep *ep)
662 {
663 	struct sk_buff *skb;
664 	struct cpl_get_tcb *req;
665 	int wrlen = roundup(sizeof(*req), 16);
666 
667 	skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
668 	if (WARN_ON(!skb))
669 		return;
670 
671 	set_wr_txq(skb, CPL_PRIORITY_CONTROL, ep->ctrlq_idx);
672 	req = (struct cpl_get_tcb *) skb_put(skb, wrlen);
673 	memset(req, 0, wrlen);
674 	INIT_TP_WR(req, ep->hwtid);
675 	OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_GET_TCB, ep->hwtid));
676 	req->reply_ctrl = htons(REPLY_CHAN_V(0) | QUEUENO_V(ep->rss_qid));
677 
678 	/*
679 	 * keep a ref on the ep so the tcb is not unlocked before this
680 	 * cpl completes. The ref is released in read_tcb_rpl().
681 	 */
682 	c4iw_get_ep(&ep->com);
683 	if (WARN_ON(c4iw_ofld_send(&ep->com.dev->rdev, skb)))
684 		c4iw_put_ep(&ep->com);
685 }
686 
687 static int send_abort_req(struct c4iw_ep *ep)
688 {
689 	u32 wrlen = roundup(sizeof(struct cpl_abort_req), 16);
690 	struct sk_buff *req_skb = skb_dequeue(&ep->com.ep_skb_list);
691 
692 	pr_debug("ep %p tid %u\n", ep, ep->hwtid);
693 	if (WARN_ON(!req_skb))
694 		return -ENOMEM;
695 
696 	cxgb_mk_abort_req(req_skb, wrlen, ep->hwtid, ep->txq_idx,
697 			  ep, abort_arp_failure);
698 
699 	return c4iw_l2t_send(&ep->com.dev->rdev, req_skb, ep->l2t);
700 }
701 
702 static int send_abort(struct c4iw_ep *ep)
703 {
704 	if (!ep->com.qp || !ep->com.qp->srq) {
705 		send_abort_req(ep);
706 		return 0;
707 	}
708 	set_bit(ABORT_REQ_IN_PROGRESS, &ep->com.flags);
709 	read_tcb(ep);
710 	return 0;
711 }
712 
713 static int send_connect(struct c4iw_ep *ep)
714 {
715 	struct cpl_act_open_req *req = NULL;
716 	struct cpl_t5_act_open_req *t5req = NULL;
717 	struct cpl_t6_act_open_req *t6req = NULL;
718 	struct cpl_act_open_req6 *req6 = NULL;
719 	struct cpl_t5_act_open_req6 *t5req6 = NULL;
720 	struct cpl_t6_act_open_req6 *t6req6 = NULL;
721 	struct sk_buff *skb;
722 	u64 opt0;
723 	u32 opt2;
724 	unsigned int mtu_idx;
725 	u32 wscale;
726 	int win, sizev4, sizev6, wrlen;
727 	struct sockaddr_in *la = (struct sockaddr_in *)
728 				 &ep->com.local_addr;
729 	struct sockaddr_in *ra = (struct sockaddr_in *)
730 				 &ep->com.remote_addr;
731 	struct sockaddr_in6 *la6 = (struct sockaddr_in6 *)
732 				   &ep->com.local_addr;
733 	struct sockaddr_in6 *ra6 = (struct sockaddr_in6 *)
734 				   &ep->com.remote_addr;
735 	int ret;
736 	enum chip_type adapter_type = ep->com.dev->rdev.lldi.adapter_type;
737 	u32 isn = (get_random_u32() & ~7UL) - 1;
738 	struct net_device *netdev;
739 	u64 params;
740 
741 	netdev = ep->com.dev->rdev.lldi.ports[0];
742 
743 	switch (CHELSIO_CHIP_VERSION(adapter_type)) {
744 	case CHELSIO_T4:
745 		sizev4 = sizeof(struct cpl_act_open_req);
746 		sizev6 = sizeof(struct cpl_act_open_req6);
747 		break;
748 	case CHELSIO_T5:
749 		sizev4 = sizeof(struct cpl_t5_act_open_req);
750 		sizev6 = sizeof(struct cpl_t5_act_open_req6);
751 		break;
752 	case CHELSIO_T6:
753 		sizev4 = sizeof(struct cpl_t6_act_open_req);
754 		sizev6 = sizeof(struct cpl_t6_act_open_req6);
755 		break;
756 	default:
757 		pr_err("T%d Chip is not supported\n",
758 		       CHELSIO_CHIP_VERSION(adapter_type));
759 		return -EINVAL;
760 	}
761 
762 	wrlen = (ep->com.remote_addr.ss_family == AF_INET) ?
763 			roundup(sizev4, 16) :
764 			roundup(sizev6, 16);
765 
766 	pr_debug("ep %p atid %u\n", ep, ep->atid);
767 
768 	skb = get_skb(NULL, wrlen, GFP_KERNEL);
769 	if (!skb) {
770 		pr_err("%s - failed to alloc skb\n", __func__);
771 		return -ENOMEM;
772 	}
773 	set_wr_txq(skb, CPL_PRIORITY_SETUP, ep->ctrlq_idx);
774 
775 	cxgb_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx,
776 		      enable_tcp_timestamps,
777 		      (ep->com.remote_addr.ss_family == AF_INET) ? 0 : 1);
778 	wscale = cxgb_compute_wscale(rcv_win);
779 
780 	/*
781 	 * Specify the largest window that will fit in opt0. The
782 	 * remainder will be specified in the rx_data_ack.
783 	 */
784 	win = ep->rcv_win >> 10;
785 	if (win > RCV_BUFSIZ_M)
786 		win = RCV_BUFSIZ_M;
787 
788 	opt0 = (nocong ? NO_CONG_F : 0) |
789 	       KEEP_ALIVE_F |
790 	       DELACK_F |
791 	       WND_SCALE_V(wscale) |
792 	       MSS_IDX_V(mtu_idx) |
793 	       L2T_IDX_V(ep->l2t->idx) |
794 	       TX_CHAN_V(ep->tx_chan) |
795 	       SMAC_SEL_V(ep->smac_idx) |
796 	       DSCP_V(ep->tos >> 2) |
797 	       ULP_MODE_V(ULP_MODE_TCPDDP) |
798 	       RCV_BUFSIZ_V(win);
799 	opt2 = RX_CHANNEL_V(0) |
800 	       CCTRL_ECN_V(enable_ecn) |
801 	       RSS_QUEUE_VALID_F | RSS_QUEUE_V(ep->rss_qid);
802 	if (enable_tcp_timestamps)
803 		opt2 |= TSTAMPS_EN_F;
804 	if (enable_tcp_sack)
805 		opt2 |= SACK_EN_F;
806 	if (wscale && enable_tcp_window_scaling)
807 		opt2 |= WND_SCALE_EN_F;
808 	if (CHELSIO_CHIP_VERSION(adapter_type) > CHELSIO_T4) {
809 		if (peer2peer)
810 			isn += 4;
811 
812 		opt2 |= T5_OPT_2_VALID_F;
813 		opt2 |= CONG_CNTRL_V(CONG_ALG_TAHOE);
814 		opt2 |= T5_ISS_F;
815 	}
816 
817 	params = cxgb4_select_ntuple(netdev, ep->l2t);
818 
819 	if (ep->com.remote_addr.ss_family == AF_INET6)
820 		cxgb4_clip_get(ep->com.dev->rdev.lldi.ports[0],
821 			       (const u32 *)&la6->sin6_addr.s6_addr, 1);
822 
823 	t4_set_arp_err_handler(skb, ep, act_open_req_arp_failure);
824 
825 	if (ep->com.remote_addr.ss_family == AF_INET) {
826 		switch (CHELSIO_CHIP_VERSION(adapter_type)) {
827 		case CHELSIO_T4:
828 			req = skb_put(skb, wrlen);
829 			INIT_TP_WR(req, 0);
830 			break;
831 		case CHELSIO_T5:
832 			t5req = skb_put(skb, wrlen);
833 			INIT_TP_WR(t5req, 0);
834 			req = (struct cpl_act_open_req *)t5req;
835 			break;
836 		case CHELSIO_T6:
837 			t6req = skb_put(skb, wrlen);
838 			INIT_TP_WR(t6req, 0);
839 			req = (struct cpl_act_open_req *)t6req;
840 			t5req = (struct cpl_t5_act_open_req *)t6req;
841 			break;
842 		default:
843 			pr_err("T%d Chip is not supported\n",
844 			       CHELSIO_CHIP_VERSION(adapter_type));
845 			ret = -EINVAL;
846 			goto clip_release;
847 		}
848 
849 		OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_ACT_OPEN_REQ,
850 					((ep->rss_qid<<14) | ep->atid)));
851 		req->local_port = la->sin_port;
852 		req->peer_port = ra->sin_port;
853 		req->local_ip = la->sin_addr.s_addr;
854 		req->peer_ip = ra->sin_addr.s_addr;
855 		req->opt0 = cpu_to_be64(opt0);
856 
857 		if (is_t4(ep->com.dev->rdev.lldi.adapter_type)) {
858 			req->params = cpu_to_be32(params);
859 			req->opt2 = cpu_to_be32(opt2);
860 		} else {
861 			if (is_t5(ep->com.dev->rdev.lldi.adapter_type)) {
862 				t5req->params =
863 					  cpu_to_be64(FILTER_TUPLE_V(params));
864 				t5req->rsvd = cpu_to_be32(isn);
865 				pr_debug("snd_isn %u\n", t5req->rsvd);
866 				t5req->opt2 = cpu_to_be32(opt2);
867 			} else {
868 				t6req->params =
869 					  cpu_to_be64(FILTER_TUPLE_V(params));
870 				t6req->rsvd = cpu_to_be32(isn);
871 				pr_debug("snd_isn %u\n", t6req->rsvd);
872 				t6req->opt2 = cpu_to_be32(opt2);
873 			}
874 		}
875 	} else {
876 		switch (CHELSIO_CHIP_VERSION(adapter_type)) {
877 		case CHELSIO_T4:
878 			req6 = skb_put(skb, wrlen);
879 			INIT_TP_WR(req6, 0);
880 			break;
881 		case CHELSIO_T5:
882 			t5req6 = skb_put(skb, wrlen);
883 			INIT_TP_WR(t5req6, 0);
884 			req6 = (struct cpl_act_open_req6 *)t5req6;
885 			break;
886 		case CHELSIO_T6:
887 			t6req6 = skb_put(skb, wrlen);
888 			INIT_TP_WR(t6req6, 0);
889 			req6 = (struct cpl_act_open_req6 *)t6req6;
890 			t5req6 = (struct cpl_t5_act_open_req6 *)t6req6;
891 			break;
892 		default:
893 			pr_err("T%d Chip is not supported\n",
894 			       CHELSIO_CHIP_VERSION(adapter_type));
895 			ret = -EINVAL;
896 			goto clip_release;
897 		}
898 
899 		OPCODE_TID(req6) = cpu_to_be32(MK_OPCODE_TID(CPL_ACT_OPEN_REQ6,
900 					((ep->rss_qid<<14)|ep->atid)));
901 		req6->local_port = la6->sin6_port;
902 		req6->peer_port = ra6->sin6_port;
903 		req6->local_ip_hi = *((__be64 *)(la6->sin6_addr.s6_addr));
904 		req6->local_ip_lo = *((__be64 *)(la6->sin6_addr.s6_addr + 8));
905 		req6->peer_ip_hi = *((__be64 *)(ra6->sin6_addr.s6_addr));
906 		req6->peer_ip_lo = *((__be64 *)(ra6->sin6_addr.s6_addr + 8));
907 		req6->opt0 = cpu_to_be64(opt0);
908 
909 		if (is_t4(ep->com.dev->rdev.lldi.adapter_type)) {
910 			req6->params = cpu_to_be32(cxgb4_select_ntuple(netdev,
911 								      ep->l2t));
912 			req6->opt2 = cpu_to_be32(opt2);
913 		} else {
914 			if (is_t5(ep->com.dev->rdev.lldi.adapter_type)) {
915 				t5req6->params =
916 					    cpu_to_be64(FILTER_TUPLE_V(params));
917 				t5req6->rsvd = cpu_to_be32(isn);
918 				pr_debug("snd_isn %u\n", t5req6->rsvd);
919 				t5req6->opt2 = cpu_to_be32(opt2);
920 			} else {
921 				t6req6->params =
922 					    cpu_to_be64(FILTER_TUPLE_V(params));
923 				t6req6->rsvd = cpu_to_be32(isn);
924 				pr_debug("snd_isn %u\n", t6req6->rsvd);
925 				t6req6->opt2 = cpu_to_be32(opt2);
926 			}
927 
928 		}
929 	}
930 
931 	set_bit(ACT_OPEN_REQ, &ep->com.history);
932 	ret = c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
933 clip_release:
934 	if (ret && ep->com.remote_addr.ss_family == AF_INET6)
935 		cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
936 				   (const u32 *)&la6->sin6_addr.s6_addr, 1);
937 	return ret;
938 }
939 
940 static int send_mpa_req(struct c4iw_ep *ep, struct sk_buff *skb,
941 			u8 mpa_rev_to_use)
942 {
943 	int mpalen, wrlen, ret;
944 	struct fw_ofld_tx_data_wr *req;
945 	struct mpa_message *mpa;
946 	struct mpa_v2_conn_params mpa_v2_params;
947 
948 	pr_debug("ep %p tid %u pd_len %d\n",
949 		 ep, ep->hwtid, ep->plen);
950 
951 	mpalen = sizeof(*mpa) + ep->plen;
952 	if (mpa_rev_to_use == 2)
953 		mpalen += sizeof(struct mpa_v2_conn_params);
954 	wrlen = roundup(mpalen + sizeof(*req), 16);
955 	skb = get_skb(skb, wrlen, GFP_KERNEL);
956 	if (!skb) {
957 		connect_reply_upcall(ep, -ENOMEM);
958 		return -ENOMEM;
959 	}
960 	set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
961 
962 	req = skb_put_zero(skb, wrlen);
963 	req->op_to_immdlen = cpu_to_be32(
964 		FW_WR_OP_V(FW_OFLD_TX_DATA_WR) |
965 		FW_WR_COMPL_F |
966 		FW_WR_IMMDLEN_V(mpalen));
967 	req->flowid_len16 = cpu_to_be32(
968 		FW_WR_FLOWID_V(ep->hwtid) |
969 		FW_WR_LEN16_V(wrlen >> 4));
970 	req->plen = cpu_to_be32(mpalen);
971 	req->tunnel_to_proxy = cpu_to_be32(
972 		FW_OFLD_TX_DATA_WR_FLUSH_F |
973 		FW_OFLD_TX_DATA_WR_SHOVE_F);
974 
975 	mpa = (struct mpa_message *)(req + 1);
976 	memcpy(mpa->key, MPA_KEY_REQ, sizeof(mpa->key));
977 
978 	mpa->flags = 0;
979 	if (crc_enabled)
980 		mpa->flags |= MPA_CRC;
981 	if (markers_enabled) {
982 		mpa->flags |= MPA_MARKERS;
983 		ep->mpa_attr.recv_marker_enabled = 1;
984 	} else {
985 		ep->mpa_attr.recv_marker_enabled = 0;
986 	}
987 	if (mpa_rev_to_use == 2)
988 		mpa->flags |= MPA_ENHANCED_RDMA_CONN;
989 
990 	mpa->private_data_size = htons(ep->plen);
991 	mpa->revision = mpa_rev_to_use;
992 	if (mpa_rev_to_use == 1) {
993 		ep->tried_with_mpa_v1 = 1;
994 		ep->retry_with_mpa_v1 = 0;
995 	}
996 
997 	if (mpa_rev_to_use == 2) {
998 		mpa->private_data_size =
999 			htons(ntohs(mpa->private_data_size) +
1000 			      sizeof(struct mpa_v2_conn_params));
1001 		pr_debug("initiator ird %u ord %u\n", ep->ird,
1002 			 ep->ord);
1003 		mpa_v2_params.ird = htons((u16)ep->ird);
1004 		mpa_v2_params.ord = htons((u16)ep->ord);
1005 
1006 		if (peer2peer) {
1007 			mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL);
1008 			if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE)
1009 				mpa_v2_params.ord |=
1010 					htons(MPA_V2_RDMA_WRITE_RTR);
1011 			else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ)
1012 				mpa_v2_params.ord |=
1013 					htons(MPA_V2_RDMA_READ_RTR);
1014 		}
1015 		memcpy(mpa->private_data, &mpa_v2_params,
1016 		       sizeof(struct mpa_v2_conn_params));
1017 
1018 		if (ep->plen)
1019 			memcpy(mpa->private_data +
1020 			       sizeof(struct mpa_v2_conn_params),
1021 			       ep->mpa_pkt + sizeof(*mpa), ep->plen);
1022 	} else
1023 		if (ep->plen)
1024 			memcpy(mpa->private_data,
1025 					ep->mpa_pkt + sizeof(*mpa), ep->plen);
1026 
1027 	/*
1028 	 * Reference the mpa skb.  This ensures the data area
1029 	 * will remain in memory until the hw acks the tx.
1030 	 * Function fw4_ack() will deref it.
1031 	 */
1032 	skb_get(skb);
1033 	t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
1034 	ep->mpa_skb = skb;
1035 	ret = c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
1036 	if (ret)
1037 		return ret;
1038 	start_ep_timer(ep);
1039 	__state_set(&ep->com, MPA_REQ_SENT);
1040 	ep->mpa_attr.initiator = 1;
1041 	ep->snd_seq += mpalen;
1042 	return ret;
1043 }
1044 
1045 static int send_mpa_reject(struct c4iw_ep *ep, const void *pdata, u8 plen)
1046 {
1047 	int mpalen, wrlen;
1048 	struct fw_ofld_tx_data_wr *req;
1049 	struct mpa_message *mpa;
1050 	struct sk_buff *skb;
1051 	struct mpa_v2_conn_params mpa_v2_params;
1052 
1053 	pr_debug("ep %p tid %u pd_len %d\n",
1054 		 ep, ep->hwtid, ep->plen);
1055 
1056 	mpalen = sizeof(*mpa) + plen;
1057 	if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn)
1058 		mpalen += sizeof(struct mpa_v2_conn_params);
1059 	wrlen = roundup(mpalen + sizeof(*req), 16);
1060 
1061 	skb = get_skb(NULL, wrlen, GFP_KERNEL);
1062 	if (!skb) {
1063 		pr_err("%s - cannot alloc skb!\n", __func__);
1064 		return -ENOMEM;
1065 	}
1066 	set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
1067 
1068 	req = skb_put_zero(skb, wrlen);
1069 	req->op_to_immdlen = cpu_to_be32(
1070 		FW_WR_OP_V(FW_OFLD_TX_DATA_WR) |
1071 		FW_WR_COMPL_F |
1072 		FW_WR_IMMDLEN_V(mpalen));
1073 	req->flowid_len16 = cpu_to_be32(
1074 		FW_WR_FLOWID_V(ep->hwtid) |
1075 		FW_WR_LEN16_V(wrlen >> 4));
1076 	req->plen = cpu_to_be32(mpalen);
1077 	req->tunnel_to_proxy = cpu_to_be32(
1078 		FW_OFLD_TX_DATA_WR_FLUSH_F |
1079 		FW_OFLD_TX_DATA_WR_SHOVE_F);
1080 
1081 	mpa = (struct mpa_message *)(req + 1);
1082 	memset(mpa, 0, sizeof(*mpa));
1083 	memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
1084 	mpa->flags = MPA_REJECT;
1085 	mpa->revision = ep->mpa_attr.version;
1086 	mpa->private_data_size = htons(plen);
1087 
1088 	if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
1089 		mpa->flags |= MPA_ENHANCED_RDMA_CONN;
1090 		mpa->private_data_size =
1091 			htons(ntohs(mpa->private_data_size) +
1092 			      sizeof(struct mpa_v2_conn_params));
1093 		mpa_v2_params.ird = htons(((u16)ep->ird) |
1094 					  (peer2peer ? MPA_V2_PEER2PEER_MODEL :
1095 					   0));
1096 		mpa_v2_params.ord = htons(((u16)ep->ord) | (peer2peer ?
1097 					  (p2p_type ==
1098 					   FW_RI_INIT_P2PTYPE_RDMA_WRITE ?
1099 					   MPA_V2_RDMA_WRITE_RTR : p2p_type ==
1100 					   FW_RI_INIT_P2PTYPE_READ_REQ ?
1101 					   MPA_V2_RDMA_READ_RTR : 0) : 0));
1102 		memcpy(mpa->private_data, &mpa_v2_params,
1103 		       sizeof(struct mpa_v2_conn_params));
1104 
1105 		if (ep->plen)
1106 			memcpy(mpa->private_data +
1107 			       sizeof(struct mpa_v2_conn_params), pdata, plen);
1108 	} else
1109 		if (plen)
1110 			memcpy(mpa->private_data, pdata, plen);
1111 
1112 	/*
1113 	 * Reference the mpa skb again.  This ensures the data area
1114 	 * will remain in memory until the hw acks the tx.
1115 	 * Function fw4_ack() will deref it.
1116 	 */
1117 	skb_get(skb);
1118 	set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
1119 	t4_set_arp_err_handler(skb, NULL, mpa_start_arp_failure);
1120 	ep->mpa_skb = skb;
1121 	ep->snd_seq += mpalen;
1122 	return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
1123 }
1124 
1125 static int send_mpa_reply(struct c4iw_ep *ep, const void *pdata, u8 plen)
1126 {
1127 	int mpalen, wrlen;
1128 	struct fw_ofld_tx_data_wr *req;
1129 	struct mpa_message *mpa;
1130 	struct sk_buff *skb;
1131 	struct mpa_v2_conn_params mpa_v2_params;
1132 
1133 	pr_debug("ep %p tid %u pd_len %d\n",
1134 		 ep, ep->hwtid, ep->plen);
1135 
1136 	mpalen = sizeof(*mpa) + plen;
1137 	if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn)
1138 		mpalen += sizeof(struct mpa_v2_conn_params);
1139 	wrlen = roundup(mpalen + sizeof(*req), 16);
1140 
1141 	skb = get_skb(NULL, wrlen, GFP_KERNEL);
1142 	if (!skb) {
1143 		pr_err("%s - cannot alloc skb!\n", __func__);
1144 		return -ENOMEM;
1145 	}
1146 	set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
1147 
1148 	req = skb_put_zero(skb, wrlen);
1149 	req->op_to_immdlen = cpu_to_be32(
1150 		FW_WR_OP_V(FW_OFLD_TX_DATA_WR) |
1151 		FW_WR_COMPL_F |
1152 		FW_WR_IMMDLEN_V(mpalen));
1153 	req->flowid_len16 = cpu_to_be32(
1154 		FW_WR_FLOWID_V(ep->hwtid) |
1155 		FW_WR_LEN16_V(wrlen >> 4));
1156 	req->plen = cpu_to_be32(mpalen);
1157 	req->tunnel_to_proxy = cpu_to_be32(
1158 		FW_OFLD_TX_DATA_WR_FLUSH_F |
1159 		FW_OFLD_TX_DATA_WR_SHOVE_F);
1160 
1161 	mpa = (struct mpa_message *)(req + 1);
1162 	memset(mpa, 0, sizeof(*mpa));
1163 	memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
1164 	mpa->flags = 0;
1165 	if (ep->mpa_attr.crc_enabled)
1166 		mpa->flags |= MPA_CRC;
1167 	if (ep->mpa_attr.recv_marker_enabled)
1168 		mpa->flags |= MPA_MARKERS;
1169 	mpa->revision = ep->mpa_attr.version;
1170 	mpa->private_data_size = htons(plen);
1171 
1172 	if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
1173 		mpa->flags |= MPA_ENHANCED_RDMA_CONN;
1174 		mpa->private_data_size =
1175 			htons(ntohs(mpa->private_data_size) +
1176 			      sizeof(struct mpa_v2_conn_params));
1177 		mpa_v2_params.ird = htons((u16)ep->ird);
1178 		mpa_v2_params.ord = htons((u16)ep->ord);
1179 		if (peer2peer && (ep->mpa_attr.p2p_type !=
1180 					FW_RI_INIT_P2PTYPE_DISABLED)) {
1181 			mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL);
1182 
1183 			if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE)
1184 				mpa_v2_params.ord |=
1185 					htons(MPA_V2_RDMA_WRITE_RTR);
1186 			else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ)
1187 				mpa_v2_params.ord |=
1188 					htons(MPA_V2_RDMA_READ_RTR);
1189 		}
1190 
1191 		memcpy(mpa->private_data, &mpa_v2_params,
1192 		       sizeof(struct mpa_v2_conn_params));
1193 
1194 		if (ep->plen)
1195 			memcpy(mpa->private_data +
1196 			       sizeof(struct mpa_v2_conn_params), pdata, plen);
1197 	} else
1198 		if (plen)
1199 			memcpy(mpa->private_data, pdata, plen);
1200 
1201 	/*
1202 	 * Reference the mpa skb.  This ensures the data area
1203 	 * will remain in memory until the hw acks the tx.
1204 	 * Function fw4_ack() will deref it.
1205 	 */
1206 	skb_get(skb);
1207 	t4_set_arp_err_handler(skb, NULL, mpa_start_arp_failure);
1208 	ep->mpa_skb = skb;
1209 	__state_set(&ep->com, MPA_REP_SENT);
1210 	ep->snd_seq += mpalen;
1211 	return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
1212 }
1213 
1214 static int act_establish(struct c4iw_dev *dev, struct sk_buff *skb)
1215 {
1216 	struct c4iw_ep *ep;
1217 	struct cpl_act_establish *req = cplhdr(skb);
1218 	unsigned short tcp_opt = ntohs(req->tcp_opt);
1219 	unsigned int tid = GET_TID(req);
1220 	unsigned int atid = TID_TID_G(ntohl(req->tos_atid));
1221 	struct tid_info *t = dev->rdev.lldi.tids;
1222 	int ret;
1223 
1224 	ep = lookup_atid(t, atid);
1225 	if (!ep)
1226 		return -EINVAL;
1227 
1228 	pr_debug("ep %p tid %u snd_isn %u rcv_isn %u\n", ep, tid,
1229 		 be32_to_cpu(req->snd_isn), be32_to_cpu(req->rcv_isn));
1230 
1231 	mutex_lock(&ep->com.mutex);
1232 	dst_confirm(ep->dst);
1233 
1234 	/* setup the hwtid for this connection */
1235 	ep->hwtid = tid;
1236 	cxgb4_insert_tid(t, ep, tid, ep->com.local_addr.ss_family);
1237 	insert_ep_tid(ep);
1238 
1239 	ep->snd_seq = be32_to_cpu(req->snd_isn);
1240 	ep->rcv_seq = be32_to_cpu(req->rcv_isn);
1241 	ep->snd_wscale = TCPOPT_SND_WSCALE_G(tcp_opt);
1242 
1243 	set_emss(ep, tcp_opt);
1244 
1245 	/* dealloc the atid */
1246 	xa_erase_irq(&ep->com.dev->atids, atid);
1247 	cxgb4_free_atid(t, atid);
1248 	set_bit(ACT_ESTAB, &ep->com.history);
1249 
1250 	/* start MPA negotiation */
1251 	ret = send_flowc(ep);
1252 	if (ret)
1253 		goto err;
1254 	if (ep->retry_with_mpa_v1)
1255 		ret = send_mpa_req(ep, skb, 1);
1256 	else
1257 		ret = send_mpa_req(ep, skb, mpa_rev);
1258 	if (ret)
1259 		goto err;
1260 	mutex_unlock(&ep->com.mutex);
1261 	return 0;
1262 err:
1263 	mutex_unlock(&ep->com.mutex);
1264 	connect_reply_upcall(ep, -ENOMEM);
1265 	c4iw_ep_disconnect(ep, 0, GFP_KERNEL);
1266 	return 0;
1267 }
1268 
1269 static void close_complete_upcall(struct c4iw_ep *ep, int status)
1270 {
1271 	struct iw_cm_event event;
1272 
1273 	pr_debug("ep %p tid %u\n", ep, ep->hwtid);
1274 	memset(&event, 0, sizeof(event));
1275 	event.event = IW_CM_EVENT_CLOSE;
1276 	event.status = status;
1277 	if (ep->com.cm_id) {
1278 		pr_debug("close complete delivered ep %p cm_id %p tid %u\n",
1279 			 ep, ep->com.cm_id, ep->hwtid);
1280 		ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1281 		deref_cm_id(&ep->com);
1282 		set_bit(CLOSE_UPCALL, &ep->com.history);
1283 	}
1284 }
1285 
1286 static void peer_close_upcall(struct c4iw_ep *ep)
1287 {
1288 	struct iw_cm_event event;
1289 
1290 	pr_debug("ep %p tid %u\n", ep, ep->hwtid);
1291 	memset(&event, 0, sizeof(event));
1292 	event.event = IW_CM_EVENT_DISCONNECT;
1293 	if (ep->com.cm_id) {
1294 		pr_debug("peer close delivered ep %p cm_id %p tid %u\n",
1295 			 ep, ep->com.cm_id, ep->hwtid);
1296 		ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1297 		set_bit(DISCONN_UPCALL, &ep->com.history);
1298 	}
1299 }
1300 
1301 static void peer_abort_upcall(struct c4iw_ep *ep)
1302 {
1303 	struct iw_cm_event event;
1304 
1305 	pr_debug("ep %p tid %u\n", ep, ep->hwtid);
1306 	memset(&event, 0, sizeof(event));
1307 	event.event = IW_CM_EVENT_CLOSE;
1308 	event.status = -ECONNRESET;
1309 	if (ep->com.cm_id) {
1310 		pr_debug("abort delivered ep %p cm_id %p tid %u\n", ep,
1311 			 ep->com.cm_id, ep->hwtid);
1312 		ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1313 		deref_cm_id(&ep->com);
1314 		set_bit(ABORT_UPCALL, &ep->com.history);
1315 	}
1316 }
1317 
1318 static void connect_reply_upcall(struct c4iw_ep *ep, int status)
1319 {
1320 	struct iw_cm_event event;
1321 
1322 	pr_debug("ep %p tid %u status %d\n",
1323 		 ep, ep->hwtid, status);
1324 	memset(&event, 0, sizeof(event));
1325 	event.event = IW_CM_EVENT_CONNECT_REPLY;
1326 	event.status = status;
1327 	memcpy(&event.local_addr, &ep->com.local_addr,
1328 	       sizeof(ep->com.local_addr));
1329 	memcpy(&event.remote_addr, &ep->com.remote_addr,
1330 	       sizeof(ep->com.remote_addr));
1331 
1332 	if ((status == 0) || (status == -ECONNREFUSED)) {
1333 		if (!ep->tried_with_mpa_v1) {
1334 			/* this means MPA_v2 is used */
1335 			event.ord = ep->ird;
1336 			event.ird = ep->ord;
1337 			event.private_data_len = ep->plen -
1338 				sizeof(struct mpa_v2_conn_params);
1339 			event.private_data = ep->mpa_pkt +
1340 				sizeof(struct mpa_message) +
1341 				sizeof(struct mpa_v2_conn_params);
1342 		} else {
1343 			/* this means MPA_v1 is used */
1344 			event.ord = cur_max_read_depth(ep->com.dev);
1345 			event.ird = cur_max_read_depth(ep->com.dev);
1346 			event.private_data_len = ep->plen;
1347 			event.private_data = ep->mpa_pkt +
1348 				sizeof(struct mpa_message);
1349 		}
1350 	}
1351 
1352 	pr_debug("ep %p tid %u status %d\n", ep,
1353 		 ep->hwtid, status);
1354 	set_bit(CONN_RPL_UPCALL, &ep->com.history);
1355 	ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1356 
1357 	if (status < 0)
1358 		deref_cm_id(&ep->com);
1359 }
1360 
1361 static int connect_request_upcall(struct c4iw_ep *ep)
1362 {
1363 	struct iw_cm_event event;
1364 	int ret;
1365 
1366 	pr_debug("ep %p tid %u\n", ep, ep->hwtid);
1367 	memset(&event, 0, sizeof(event));
1368 	event.event = IW_CM_EVENT_CONNECT_REQUEST;
1369 	memcpy(&event.local_addr, &ep->com.local_addr,
1370 	       sizeof(ep->com.local_addr));
1371 	memcpy(&event.remote_addr, &ep->com.remote_addr,
1372 	       sizeof(ep->com.remote_addr));
1373 	event.provider_data = ep;
1374 	if (!ep->tried_with_mpa_v1) {
1375 		/* this means MPA_v2 is used */
1376 		event.ord = ep->ord;
1377 		event.ird = ep->ird;
1378 		event.private_data_len = ep->plen -
1379 			sizeof(struct mpa_v2_conn_params);
1380 		event.private_data = ep->mpa_pkt + sizeof(struct mpa_message) +
1381 			sizeof(struct mpa_v2_conn_params);
1382 	} else {
1383 		/* this means MPA_v1 is used. Send max supported */
1384 		event.ord = cur_max_read_depth(ep->com.dev);
1385 		event.ird = cur_max_read_depth(ep->com.dev);
1386 		event.private_data_len = ep->plen;
1387 		event.private_data = ep->mpa_pkt + sizeof(struct mpa_message);
1388 	}
1389 	c4iw_get_ep(&ep->com);
1390 	ret = ep->parent_ep->com.cm_id->event_handler(ep->parent_ep->com.cm_id,
1391 						      &event);
1392 	if (ret)
1393 		c4iw_put_ep(&ep->com);
1394 	set_bit(CONNREQ_UPCALL, &ep->com.history);
1395 	c4iw_put_ep(&ep->parent_ep->com);
1396 	return ret;
1397 }
1398 
1399 static void established_upcall(struct c4iw_ep *ep)
1400 {
1401 	struct iw_cm_event event;
1402 
1403 	pr_debug("ep %p tid %u\n", ep, ep->hwtid);
1404 	memset(&event, 0, sizeof(event));
1405 	event.event = IW_CM_EVENT_ESTABLISHED;
1406 	event.ird = ep->ord;
1407 	event.ord = ep->ird;
1408 	if (ep->com.cm_id) {
1409 		pr_debug("ep %p tid %u\n", ep, ep->hwtid);
1410 		ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1411 		set_bit(ESTAB_UPCALL, &ep->com.history);
1412 	}
1413 }
1414 
1415 static int update_rx_credits(struct c4iw_ep *ep, u32 credits)
1416 {
1417 	struct sk_buff *skb;
1418 	u32 wrlen = roundup(sizeof(struct cpl_rx_data_ack), 16);
1419 	u32 credit_dack;
1420 
1421 	pr_debug("ep %p tid %u credits %u\n",
1422 		 ep, ep->hwtid, credits);
1423 	skb = get_skb(NULL, wrlen, GFP_KERNEL);
1424 	if (!skb) {
1425 		pr_err("update_rx_credits - cannot alloc skb!\n");
1426 		return 0;
1427 	}
1428 
1429 	/*
1430 	 * If we couldn't specify the entire rcv window at connection setup
1431 	 * due to the limit in the number of bits in the RCV_BUFSIZ field,
1432 	 * then add the overage in to the credits returned.
1433 	 */
1434 	if (ep->rcv_win > RCV_BUFSIZ_M * 1024)
1435 		credits += ep->rcv_win - RCV_BUFSIZ_M * 1024;
1436 
1437 	credit_dack = credits | RX_FORCE_ACK_F | RX_DACK_CHANGE_F |
1438 		      RX_DACK_MODE_V(dack_mode);
1439 
1440 	cxgb_mk_rx_data_ack(skb, wrlen, ep->hwtid, ep->ctrlq_idx,
1441 			    credit_dack);
1442 
1443 	c4iw_ofld_send(&ep->com.dev->rdev, skb);
1444 	return credits;
1445 }
1446 
1447 #define RELAXED_IRD_NEGOTIATION 1
1448 
1449 /*
1450  * process_mpa_reply - process streaming mode MPA reply
1451  *
1452  * Returns:
1453  *
1454  * 0 upon success indicating a connect request was delivered to the ULP
1455  * or the mpa request is incomplete but valid so far.
1456  *
1457  * 1 if a failure requires the caller to close the connection.
1458  *
1459  * 2 if a failure requires the caller to abort the connection.
1460  */
1461 static int process_mpa_reply(struct c4iw_ep *ep, struct sk_buff *skb)
1462 {
1463 	struct mpa_message *mpa;
1464 	struct mpa_v2_conn_params *mpa_v2_params;
1465 	u16 plen;
1466 	u16 resp_ird, resp_ord;
1467 	u8 rtr_mismatch = 0, insuff_ird = 0;
1468 	struct c4iw_qp_attributes attrs;
1469 	enum c4iw_qp_attr_mask mask;
1470 	int err;
1471 	int disconnect = 0;
1472 
1473 	pr_debug("ep %p tid %u\n", ep, ep->hwtid);
1474 
1475 	/*
1476 	 * If we get more than the supported amount of private data
1477 	 * then we must fail this connection.
1478 	 */
1479 	if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
1480 		err = -EINVAL;
1481 		goto err_stop_timer;
1482 	}
1483 
1484 	/*
1485 	 * copy the new data into our accumulation buffer.
1486 	 */
1487 	skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
1488 				  skb->len);
1489 	ep->mpa_pkt_len += skb->len;
1490 
1491 	/*
1492 	 * if we don't even have the mpa message, then bail.
1493 	 */
1494 	if (ep->mpa_pkt_len < sizeof(*mpa))
1495 		return 0;
1496 	mpa = (struct mpa_message *) ep->mpa_pkt;
1497 
1498 	/* Validate MPA header. */
1499 	if (mpa->revision > mpa_rev) {
1500 		pr_err("%s MPA version mismatch. Local = %d, Received = %d\n",
1501 		       __func__, mpa_rev, mpa->revision);
1502 		err = -EPROTO;
1503 		goto err_stop_timer;
1504 	}
1505 	if (memcmp(mpa->key, MPA_KEY_REP, sizeof(mpa->key))) {
1506 		err = -EPROTO;
1507 		goto err_stop_timer;
1508 	}
1509 
1510 	plen = ntohs(mpa->private_data_size);
1511 
1512 	/*
1513 	 * Fail if there's too much private data.
1514 	 */
1515 	if (plen > MPA_MAX_PRIVATE_DATA) {
1516 		err = -EPROTO;
1517 		goto err_stop_timer;
1518 	}
1519 
1520 	/*
1521 	 * If plen does not account for pkt size
1522 	 */
1523 	if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
1524 		err = -EPROTO;
1525 		goto err_stop_timer;
1526 	}
1527 
1528 	ep->plen = (u8) plen;
1529 
1530 	/*
1531 	 * If we don't have all the pdata yet, then bail.
1532 	 * We'll continue process when more data arrives.
1533 	 */
1534 	if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
1535 		return 0;
1536 
1537 	if (mpa->flags & MPA_REJECT) {
1538 		err = -ECONNREFUSED;
1539 		goto err_stop_timer;
1540 	}
1541 
1542 	/*
1543 	 * Stop mpa timer.  If it expired, then
1544 	 * we ignore the MPA reply.  process_timeout()
1545 	 * will abort the connection.
1546 	 */
1547 	if (stop_ep_timer(ep))
1548 		return 0;
1549 
1550 	/*
1551 	 * If we get here we have accumulated the entire mpa
1552 	 * start reply message including private data. And
1553 	 * the MPA header is valid.
1554 	 */
1555 	__state_set(&ep->com, FPDU_MODE);
1556 	ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1557 	ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1558 	ep->mpa_attr.version = mpa->revision;
1559 	ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1560 
1561 	if (mpa->revision == 2) {
1562 		ep->mpa_attr.enhanced_rdma_conn =
1563 			mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
1564 		if (ep->mpa_attr.enhanced_rdma_conn) {
1565 			mpa_v2_params = (struct mpa_v2_conn_params *)
1566 				(ep->mpa_pkt + sizeof(*mpa));
1567 			resp_ird = ntohs(mpa_v2_params->ird) &
1568 				MPA_V2_IRD_ORD_MASK;
1569 			resp_ord = ntohs(mpa_v2_params->ord) &
1570 				MPA_V2_IRD_ORD_MASK;
1571 			pr_debug("responder ird %u ord %u ep ird %u ord %u\n",
1572 				 resp_ird, resp_ord, ep->ird, ep->ord);
1573 
1574 			/*
1575 			 * This is a double-check. Ideally, below checks are
1576 			 * not required since ird/ord stuff has been taken
1577 			 * care of in c4iw_accept_cr
1578 			 */
1579 			if (ep->ird < resp_ord) {
1580 				if (RELAXED_IRD_NEGOTIATION && resp_ord <=
1581 				    ep->com.dev->rdev.lldi.max_ordird_qp)
1582 					ep->ird = resp_ord;
1583 				else
1584 					insuff_ird = 1;
1585 			} else if (ep->ird > resp_ord) {
1586 				ep->ird = resp_ord;
1587 			}
1588 			if (ep->ord > resp_ird) {
1589 				if (RELAXED_IRD_NEGOTIATION)
1590 					ep->ord = resp_ird;
1591 				else
1592 					insuff_ird = 1;
1593 			}
1594 			if (insuff_ird) {
1595 				err = -ENOMEM;
1596 				ep->ird = resp_ord;
1597 				ep->ord = resp_ird;
1598 			}
1599 
1600 			if (ntohs(mpa_v2_params->ird) &
1601 					MPA_V2_PEER2PEER_MODEL) {
1602 				if (ntohs(mpa_v2_params->ord) &
1603 						MPA_V2_RDMA_WRITE_RTR)
1604 					ep->mpa_attr.p2p_type =
1605 						FW_RI_INIT_P2PTYPE_RDMA_WRITE;
1606 				else if (ntohs(mpa_v2_params->ord) &
1607 						MPA_V2_RDMA_READ_RTR)
1608 					ep->mpa_attr.p2p_type =
1609 						FW_RI_INIT_P2PTYPE_READ_REQ;
1610 			}
1611 		}
1612 	} else if (mpa->revision == 1)
1613 		if (peer2peer)
1614 			ep->mpa_attr.p2p_type = p2p_type;
1615 
1616 	pr_debug("crc_enabled=%d, recv_marker_enabled=%d, xmit_marker_enabled=%d, version=%d p2p_type=%d local-p2p_type = %d\n",
1617 		 ep->mpa_attr.crc_enabled,
1618 		 ep->mpa_attr.recv_marker_enabled,
1619 		 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version,
1620 		 ep->mpa_attr.p2p_type, p2p_type);
1621 
1622 	/*
1623 	 * If responder's RTR does not match with that of initiator, assign
1624 	 * FW_RI_INIT_P2PTYPE_DISABLED in mpa attributes so that RTR is not
1625 	 * generated when moving QP to RTS state.
1626 	 * A TERM message will be sent after QP has moved to RTS state
1627 	 */
1628 	if ((ep->mpa_attr.version == 2) && peer2peer &&
1629 			(ep->mpa_attr.p2p_type != p2p_type)) {
1630 		ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1631 		rtr_mismatch = 1;
1632 	}
1633 
1634 	attrs.mpa_attr = ep->mpa_attr;
1635 	attrs.max_ird = ep->ird;
1636 	attrs.max_ord = ep->ord;
1637 	attrs.llp_stream_handle = ep;
1638 	attrs.next_state = C4IW_QP_STATE_RTS;
1639 
1640 	mask = C4IW_QP_ATTR_NEXT_STATE |
1641 	    C4IW_QP_ATTR_LLP_STREAM_HANDLE | C4IW_QP_ATTR_MPA_ATTR |
1642 	    C4IW_QP_ATTR_MAX_IRD | C4IW_QP_ATTR_MAX_ORD;
1643 
1644 	/* bind QP and TID with INIT_WR */
1645 	err = c4iw_modify_qp(ep->com.qp->rhp,
1646 			     ep->com.qp, mask, &attrs, 1);
1647 	if (err)
1648 		goto err;
1649 
1650 	/*
1651 	 * If responder's RTR requirement did not match with what initiator
1652 	 * supports, generate TERM message
1653 	 */
1654 	if (rtr_mismatch) {
1655 		pr_err("%s: RTR mismatch, sending TERM\n", __func__);
1656 		attrs.layer_etype = LAYER_MPA | DDP_LLP;
1657 		attrs.ecode = MPA_NOMATCH_RTR;
1658 		attrs.next_state = C4IW_QP_STATE_TERMINATE;
1659 		attrs.send_term = 1;
1660 		err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1661 				C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
1662 		err = -ENOMEM;
1663 		disconnect = 1;
1664 		goto out;
1665 	}
1666 
1667 	/*
1668 	 * Generate TERM if initiator IRD is not sufficient for responder
1669 	 * provided ORD. Currently, we do the same behaviour even when
1670 	 * responder provided IRD is also not sufficient as regards to
1671 	 * initiator ORD.
1672 	 */
1673 	if (insuff_ird) {
1674 		pr_err("%s: Insufficient IRD, sending TERM\n", __func__);
1675 		attrs.layer_etype = LAYER_MPA | DDP_LLP;
1676 		attrs.ecode = MPA_INSUFF_IRD;
1677 		attrs.next_state = C4IW_QP_STATE_TERMINATE;
1678 		attrs.send_term = 1;
1679 		err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1680 				C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
1681 		err = -ENOMEM;
1682 		disconnect = 1;
1683 		goto out;
1684 	}
1685 	goto out;
1686 err_stop_timer:
1687 	stop_ep_timer(ep);
1688 err:
1689 	disconnect = 2;
1690 out:
1691 	connect_reply_upcall(ep, err);
1692 	return disconnect;
1693 }
1694 
1695 /*
1696  * process_mpa_request - process streaming mode MPA request
1697  *
1698  * Returns:
1699  *
1700  * 0 upon success indicating a connect request was delivered to the ULP
1701  * or the mpa request is incomplete but valid so far.
1702  *
1703  * 1 if a failure requires the caller to close the connection.
1704  *
1705  * 2 if a failure requires the caller to abort the connection.
1706  */
1707 static int process_mpa_request(struct c4iw_ep *ep, struct sk_buff *skb)
1708 {
1709 	struct mpa_message *mpa;
1710 	struct mpa_v2_conn_params *mpa_v2_params;
1711 	u16 plen;
1712 
1713 	pr_debug("ep %p tid %u\n", ep, ep->hwtid);
1714 
1715 	/*
1716 	 * If we get more than the supported amount of private data
1717 	 * then we must fail this connection.
1718 	 */
1719 	if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt))
1720 		goto err_stop_timer;
1721 
1722 	pr_debug("enter (%s line %u)\n", __FILE__, __LINE__);
1723 
1724 	/*
1725 	 * Copy the new data into our accumulation buffer.
1726 	 */
1727 	skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
1728 				  skb->len);
1729 	ep->mpa_pkt_len += skb->len;
1730 
1731 	/*
1732 	 * If we don't even have the mpa message, then bail.
1733 	 * We'll continue process when more data arrives.
1734 	 */
1735 	if (ep->mpa_pkt_len < sizeof(*mpa))
1736 		return 0;
1737 
1738 	pr_debug("enter (%s line %u)\n", __FILE__, __LINE__);
1739 	mpa = (struct mpa_message *) ep->mpa_pkt;
1740 
1741 	/*
1742 	 * Validate MPA Header.
1743 	 */
1744 	if (mpa->revision > mpa_rev) {
1745 		pr_err("%s MPA version mismatch. Local = %d, Received = %d\n",
1746 		       __func__, mpa_rev, mpa->revision);
1747 		goto err_stop_timer;
1748 	}
1749 
1750 	if (memcmp(mpa->key, MPA_KEY_REQ, sizeof(mpa->key)))
1751 		goto err_stop_timer;
1752 
1753 	plen = ntohs(mpa->private_data_size);
1754 
1755 	/*
1756 	 * Fail if there's too much private data.
1757 	 */
1758 	if (plen > MPA_MAX_PRIVATE_DATA)
1759 		goto err_stop_timer;
1760 
1761 	/*
1762 	 * If plen does not account for pkt size
1763 	 */
1764 	if (ep->mpa_pkt_len > (sizeof(*mpa) + plen))
1765 		goto err_stop_timer;
1766 	ep->plen = (u8) plen;
1767 
1768 	/*
1769 	 * If we don't have all the pdata yet, then bail.
1770 	 */
1771 	if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
1772 		return 0;
1773 
1774 	/*
1775 	 * If we get here we have accumulated the entire mpa
1776 	 * start reply message including private data.
1777 	 */
1778 	ep->mpa_attr.initiator = 0;
1779 	ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1780 	ep->mpa_attr.recv_marker_enabled = markers_enabled;
1781 	ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1782 	ep->mpa_attr.version = mpa->revision;
1783 	if (mpa->revision == 1)
1784 		ep->tried_with_mpa_v1 = 1;
1785 	ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1786 
1787 	if (mpa->revision == 2) {
1788 		ep->mpa_attr.enhanced_rdma_conn =
1789 			mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
1790 		if (ep->mpa_attr.enhanced_rdma_conn) {
1791 			mpa_v2_params = (struct mpa_v2_conn_params *)
1792 				(ep->mpa_pkt + sizeof(*mpa));
1793 			ep->ird = ntohs(mpa_v2_params->ird) &
1794 				MPA_V2_IRD_ORD_MASK;
1795 			ep->ird = min_t(u32, ep->ird,
1796 					cur_max_read_depth(ep->com.dev));
1797 			ep->ord = ntohs(mpa_v2_params->ord) &
1798 				MPA_V2_IRD_ORD_MASK;
1799 			ep->ord = min_t(u32, ep->ord,
1800 					cur_max_read_depth(ep->com.dev));
1801 			pr_debug("initiator ird %u ord %u\n",
1802 				 ep->ird, ep->ord);
1803 			if (ntohs(mpa_v2_params->ird) & MPA_V2_PEER2PEER_MODEL)
1804 				if (peer2peer) {
1805 					if (ntohs(mpa_v2_params->ord) &
1806 							MPA_V2_RDMA_WRITE_RTR)
1807 						ep->mpa_attr.p2p_type =
1808 						FW_RI_INIT_P2PTYPE_RDMA_WRITE;
1809 					else if (ntohs(mpa_v2_params->ord) &
1810 							MPA_V2_RDMA_READ_RTR)
1811 						ep->mpa_attr.p2p_type =
1812 						FW_RI_INIT_P2PTYPE_READ_REQ;
1813 				}
1814 		}
1815 	} else if (mpa->revision == 1)
1816 		if (peer2peer)
1817 			ep->mpa_attr.p2p_type = p2p_type;
1818 
1819 	pr_debug("crc_enabled=%d, recv_marker_enabled=%d, xmit_marker_enabled=%d, version=%d p2p_type=%d\n",
1820 		 ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
1821 		 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version,
1822 		 ep->mpa_attr.p2p_type);
1823 
1824 	__state_set(&ep->com, MPA_REQ_RCVD);
1825 
1826 	/* drive upcall */
1827 	mutex_lock_nested(&ep->parent_ep->com.mutex, SINGLE_DEPTH_NESTING);
1828 	if (ep->parent_ep->com.state != DEAD) {
1829 		if (connect_request_upcall(ep))
1830 			goto err_unlock_parent;
1831 	} else {
1832 		goto err_unlock_parent;
1833 	}
1834 	mutex_unlock(&ep->parent_ep->com.mutex);
1835 	return 0;
1836 
1837 err_unlock_parent:
1838 	mutex_unlock(&ep->parent_ep->com.mutex);
1839 	goto err_out;
1840 err_stop_timer:
1841 	(void)stop_ep_timer(ep);
1842 err_out:
1843 	return 2;
1844 }
1845 
1846 static int rx_data(struct c4iw_dev *dev, struct sk_buff *skb)
1847 {
1848 	struct c4iw_ep *ep;
1849 	struct cpl_rx_data *hdr = cplhdr(skb);
1850 	unsigned int dlen = ntohs(hdr->len);
1851 	unsigned int tid = GET_TID(hdr);
1852 	__u8 status = hdr->status;
1853 	int disconnect = 0;
1854 
1855 	ep = get_ep_from_tid(dev, tid);
1856 	if (!ep)
1857 		return 0;
1858 	pr_debug("ep %p tid %u dlen %u\n", ep, ep->hwtid, dlen);
1859 	skb_pull(skb, sizeof(*hdr));
1860 	skb_trim(skb, dlen);
1861 	mutex_lock(&ep->com.mutex);
1862 
1863 	switch (ep->com.state) {
1864 	case MPA_REQ_SENT:
1865 		update_rx_credits(ep, dlen);
1866 		ep->rcv_seq += dlen;
1867 		disconnect = process_mpa_reply(ep, skb);
1868 		break;
1869 	case MPA_REQ_WAIT:
1870 		update_rx_credits(ep, dlen);
1871 		ep->rcv_seq += dlen;
1872 		disconnect = process_mpa_request(ep, skb);
1873 		break;
1874 	case FPDU_MODE: {
1875 		struct c4iw_qp_attributes attrs;
1876 
1877 		update_rx_credits(ep, dlen);
1878 		if (status)
1879 			pr_err("%s Unexpected streaming data." \
1880 			       " qpid %u ep %p state %d tid %u status %d\n",
1881 			       __func__, ep->com.qp->wq.sq.qid, ep,
1882 			       ep->com.state, ep->hwtid, status);
1883 		attrs.next_state = C4IW_QP_STATE_TERMINATE;
1884 		c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1885 			       C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
1886 		disconnect = 1;
1887 		break;
1888 	}
1889 	default:
1890 		break;
1891 	}
1892 	mutex_unlock(&ep->com.mutex);
1893 	if (disconnect)
1894 		c4iw_ep_disconnect(ep, disconnect == 2, GFP_KERNEL);
1895 	c4iw_put_ep(&ep->com);
1896 	return 0;
1897 }
1898 
1899 static void complete_cached_srq_buffers(struct c4iw_ep *ep, u32 srqidx)
1900 {
1901 	enum chip_type adapter_type;
1902 
1903 	adapter_type = ep->com.dev->rdev.lldi.adapter_type;
1904 
1905 	/*
1906 	 * If this TCB had a srq buffer cached, then we must complete
1907 	 * it. For user mode, that means saving the srqidx in the
1908 	 * user/kernel status page for this qp.  For kernel mode, just
1909 	 * synthesize the CQE now.
1910 	 */
1911 	if (CHELSIO_CHIP_VERSION(adapter_type) > CHELSIO_T5 && srqidx) {
1912 		if (ep->com.qp->ibqp.uobject)
1913 			t4_set_wq_in_error(&ep->com.qp->wq, srqidx);
1914 		else
1915 			c4iw_flush_srqidx(ep->com.qp, srqidx);
1916 	}
1917 }
1918 
1919 static int abort_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
1920 {
1921 	u32 srqidx;
1922 	struct c4iw_ep *ep;
1923 	struct cpl_abort_rpl_rss6 *rpl = cplhdr(skb);
1924 	int release = 0;
1925 	unsigned int tid = GET_TID(rpl);
1926 
1927 	ep = get_ep_from_tid(dev, tid);
1928 	if (!ep) {
1929 		pr_warn("Abort rpl to freed endpoint\n");
1930 		return 0;
1931 	}
1932 
1933 	if (ep->com.qp && ep->com.qp->srq) {
1934 		srqidx = ABORT_RSS_SRQIDX_G(be32_to_cpu(rpl->srqidx_status));
1935 		complete_cached_srq_buffers(ep, srqidx ? srqidx : ep->srqe_idx);
1936 	}
1937 
1938 	pr_debug("ep %p tid %u\n", ep, ep->hwtid);
1939 	mutex_lock(&ep->com.mutex);
1940 	switch (ep->com.state) {
1941 	case ABORTING:
1942 		c4iw_wake_up_noref(ep->com.wr_waitp, -ECONNRESET);
1943 		__state_set(&ep->com, DEAD);
1944 		release = 1;
1945 		break;
1946 	default:
1947 		pr_err("%s ep %p state %d\n", __func__, ep, ep->com.state);
1948 		break;
1949 	}
1950 	mutex_unlock(&ep->com.mutex);
1951 
1952 	if (release) {
1953 		close_complete_upcall(ep, -ECONNRESET);
1954 		release_ep_resources(ep);
1955 	}
1956 	c4iw_put_ep(&ep->com);
1957 	return 0;
1958 }
1959 
1960 static int send_fw_act_open_req(struct c4iw_ep *ep, unsigned int atid)
1961 {
1962 	struct sk_buff *skb;
1963 	struct fw_ofld_connection_wr *req;
1964 	unsigned int mtu_idx;
1965 	u32 wscale;
1966 	struct sockaddr_in *sin;
1967 	int win;
1968 
1969 	skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
1970 	if (!skb)
1971 		return -ENOMEM;
1972 
1973 	req = __skb_put_zero(skb, sizeof(*req));
1974 	req->op_compl = htonl(WR_OP_V(FW_OFLD_CONNECTION_WR));
1975 	req->len16_pkd = htonl(FW_WR_LEN16_V(DIV_ROUND_UP(sizeof(*req), 16)));
1976 	req->le.filter = cpu_to_be32(cxgb4_select_ntuple(
1977 				     ep->com.dev->rdev.lldi.ports[0],
1978 				     ep->l2t));
1979 	sin = (struct sockaddr_in *)&ep->com.local_addr;
1980 	req->le.lport = sin->sin_port;
1981 	req->le.u.ipv4.lip = sin->sin_addr.s_addr;
1982 	sin = (struct sockaddr_in *)&ep->com.remote_addr;
1983 	req->le.pport = sin->sin_port;
1984 	req->le.u.ipv4.pip = sin->sin_addr.s_addr;
1985 	req->tcb.t_state_to_astid =
1986 			htonl(FW_OFLD_CONNECTION_WR_T_STATE_V(TCP_SYN_SENT) |
1987 			FW_OFLD_CONNECTION_WR_ASTID_V(atid));
1988 	req->tcb.cplrxdataack_cplpassacceptrpl =
1989 			htons(FW_OFLD_CONNECTION_WR_CPLRXDATAACK_F);
1990 	req->tcb.tx_max = (__force __be32) jiffies;
1991 	req->tcb.rcv_adv = htons(1);
1992 	cxgb_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx,
1993 		      enable_tcp_timestamps,
1994 		      (ep->com.remote_addr.ss_family == AF_INET) ? 0 : 1);
1995 	wscale = cxgb_compute_wscale(rcv_win);
1996 
1997 	/*
1998 	 * Specify the largest window that will fit in opt0. The
1999 	 * remainder will be specified in the rx_data_ack.
2000 	 */
2001 	win = ep->rcv_win >> 10;
2002 	if (win > RCV_BUFSIZ_M)
2003 		win = RCV_BUFSIZ_M;
2004 
2005 	req->tcb.opt0 = (__force __be64) (TCAM_BYPASS_F |
2006 		(nocong ? NO_CONG_F : 0) |
2007 		KEEP_ALIVE_F |
2008 		DELACK_F |
2009 		WND_SCALE_V(wscale) |
2010 		MSS_IDX_V(mtu_idx) |
2011 		L2T_IDX_V(ep->l2t->idx) |
2012 		TX_CHAN_V(ep->tx_chan) |
2013 		SMAC_SEL_V(ep->smac_idx) |
2014 		DSCP_V(ep->tos >> 2) |
2015 		ULP_MODE_V(ULP_MODE_TCPDDP) |
2016 		RCV_BUFSIZ_V(win));
2017 	req->tcb.opt2 = (__force __be32) (PACE_V(1) |
2018 		TX_QUEUE_V(ep->com.dev->rdev.lldi.tx_modq[ep->tx_chan]) |
2019 		RX_CHANNEL_V(0) |
2020 		CCTRL_ECN_V(enable_ecn) |
2021 		RSS_QUEUE_VALID_F | RSS_QUEUE_V(ep->rss_qid));
2022 	if (enable_tcp_timestamps)
2023 		req->tcb.opt2 |= (__force __be32)TSTAMPS_EN_F;
2024 	if (enable_tcp_sack)
2025 		req->tcb.opt2 |= (__force __be32)SACK_EN_F;
2026 	if (wscale && enable_tcp_window_scaling)
2027 		req->tcb.opt2 |= (__force __be32)WND_SCALE_EN_F;
2028 	req->tcb.opt0 = cpu_to_be64((__force u64)req->tcb.opt0);
2029 	req->tcb.opt2 = cpu_to_be32((__force u32)req->tcb.opt2);
2030 	set_wr_txq(skb, CPL_PRIORITY_CONTROL, ep->ctrlq_idx);
2031 	set_bit(ACT_OFLD_CONN, &ep->com.history);
2032 	return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
2033 }
2034 
2035 /*
2036  * Some of the error codes above implicitly indicate that there is no TID
2037  * allocated with the result of an ACT_OPEN.  We use this predicate to make
2038  * that explicit.
2039  */
2040 static inline int act_open_has_tid(int status)
2041 {
2042 	return (status != CPL_ERR_TCAM_PARITY &&
2043 		status != CPL_ERR_TCAM_MISS &&
2044 		status != CPL_ERR_TCAM_FULL &&
2045 		status != CPL_ERR_CONN_EXIST_SYNRECV &&
2046 		status != CPL_ERR_CONN_EXIST);
2047 }
2048 
2049 static char *neg_adv_str(unsigned int status)
2050 {
2051 	switch (status) {
2052 	case CPL_ERR_RTX_NEG_ADVICE:
2053 		return "Retransmit timeout";
2054 	case CPL_ERR_PERSIST_NEG_ADVICE:
2055 		return "Persist timeout";
2056 	case CPL_ERR_KEEPALV_NEG_ADVICE:
2057 		return "Keepalive timeout";
2058 	default:
2059 		return "Unknown";
2060 	}
2061 }
2062 
2063 static void set_tcp_window(struct c4iw_ep *ep, struct port_info *pi)
2064 {
2065 	ep->snd_win = snd_win;
2066 	ep->rcv_win = rcv_win;
2067 	pr_debug("snd_win %d rcv_win %d\n",
2068 		 ep->snd_win, ep->rcv_win);
2069 }
2070 
2071 #define ACT_OPEN_RETRY_COUNT 2
2072 
2073 static int import_ep(struct c4iw_ep *ep, int iptype, __u8 *peer_ip,
2074 		     struct dst_entry *dst, struct c4iw_dev *cdev,
2075 		     bool clear_mpa_v1, enum chip_type adapter_type, u8 tos)
2076 {
2077 	struct neighbour *n;
2078 	int err, step;
2079 	struct net_device *pdev;
2080 
2081 	n = dst_neigh_lookup(dst, peer_ip);
2082 	if (!n)
2083 		return -ENODEV;
2084 
2085 	rcu_read_lock();
2086 	err = -ENOMEM;
2087 	if (n->dev->flags & IFF_LOOPBACK) {
2088 		if (iptype == 4)
2089 			pdev = ip_dev_find(&init_net, *(__be32 *)peer_ip);
2090 		else if (IS_ENABLED(CONFIG_IPV6))
2091 			for_each_netdev(&init_net, pdev) {
2092 				if (ipv6_chk_addr(&init_net,
2093 						  (struct in6_addr *)peer_ip,
2094 						  pdev, 1))
2095 					break;
2096 			}
2097 		else
2098 			pdev = NULL;
2099 
2100 		if (!pdev) {
2101 			err = -ENODEV;
2102 			goto out;
2103 		}
2104 		ep->l2t = cxgb4_l2t_get(cdev->rdev.lldi.l2t,
2105 					n, pdev, rt_tos2priority(tos));
2106 		if (!ep->l2t) {
2107 			dev_put(pdev);
2108 			goto out;
2109 		}
2110 		ep->mtu = pdev->mtu;
2111 		ep->tx_chan = cxgb4_port_chan(pdev);
2112 		ep->smac_idx = ((struct port_info *)netdev_priv(pdev))->smt_idx;
2113 		step = cdev->rdev.lldi.ntxq /
2114 			cdev->rdev.lldi.nchan;
2115 		ep->txq_idx = cxgb4_port_idx(pdev) * step;
2116 		step = cdev->rdev.lldi.nrxq /
2117 			cdev->rdev.lldi.nchan;
2118 		ep->ctrlq_idx = cxgb4_port_idx(pdev);
2119 		ep->rss_qid = cdev->rdev.lldi.rxq_ids[
2120 			cxgb4_port_idx(pdev) * step];
2121 		set_tcp_window(ep, (struct port_info *)netdev_priv(pdev));
2122 		dev_put(pdev);
2123 	} else {
2124 		pdev = get_real_dev(n->dev);
2125 		ep->l2t = cxgb4_l2t_get(cdev->rdev.lldi.l2t,
2126 					n, pdev, rt_tos2priority(tos));
2127 		if (!ep->l2t)
2128 			goto out;
2129 		ep->mtu = dst_mtu(dst);
2130 		ep->tx_chan = cxgb4_port_chan(pdev);
2131 		ep->smac_idx = ((struct port_info *)netdev_priv(pdev))->smt_idx;
2132 		step = cdev->rdev.lldi.ntxq /
2133 			cdev->rdev.lldi.nchan;
2134 		ep->txq_idx = cxgb4_port_idx(pdev) * step;
2135 		ep->ctrlq_idx = cxgb4_port_idx(pdev);
2136 		step = cdev->rdev.lldi.nrxq /
2137 			cdev->rdev.lldi.nchan;
2138 		ep->rss_qid = cdev->rdev.lldi.rxq_ids[
2139 			cxgb4_port_idx(pdev) * step];
2140 		set_tcp_window(ep, (struct port_info *)netdev_priv(pdev));
2141 
2142 		if (clear_mpa_v1) {
2143 			ep->retry_with_mpa_v1 = 0;
2144 			ep->tried_with_mpa_v1 = 0;
2145 		}
2146 	}
2147 	err = 0;
2148 out:
2149 	rcu_read_unlock();
2150 
2151 	neigh_release(n);
2152 
2153 	return err;
2154 }
2155 
2156 static int c4iw_reconnect(struct c4iw_ep *ep)
2157 {
2158 	int err = 0;
2159 	int size = 0;
2160 	struct sockaddr_in *laddr = (struct sockaddr_in *)
2161 				    &ep->com.cm_id->m_local_addr;
2162 	struct sockaddr_in *raddr = (struct sockaddr_in *)
2163 				    &ep->com.cm_id->m_remote_addr;
2164 	struct sockaddr_in6 *laddr6 = (struct sockaddr_in6 *)
2165 				      &ep->com.cm_id->m_local_addr;
2166 	struct sockaddr_in6 *raddr6 = (struct sockaddr_in6 *)
2167 				      &ep->com.cm_id->m_remote_addr;
2168 	int iptype;
2169 	__u8 *ra;
2170 
2171 	pr_debug("qp %p cm_id %p\n", ep->com.qp, ep->com.cm_id);
2172 	c4iw_init_wr_wait(ep->com.wr_waitp);
2173 
2174 	/* When MPA revision is different on nodes, the node with MPA_rev=2
2175 	 * tries to reconnect with MPA_rev 1 for the same EP through
2176 	 * c4iw_reconnect(), where the same EP is assigned with new tid for
2177 	 * further connection establishment. As we are using the same EP pointer
2178 	 * for reconnect, few skbs are used during the previous c4iw_connect(),
2179 	 * which leaves the EP with inadequate skbs for further
2180 	 * c4iw_reconnect(), Further causing a crash due to an empty
2181 	 * skb_list() during peer_abort(). Allocate skbs which is already used.
2182 	 */
2183 	size = (CN_MAX_CON_BUF - skb_queue_len(&ep->com.ep_skb_list));
2184 	if (alloc_ep_skb_list(&ep->com.ep_skb_list, size)) {
2185 		err = -ENOMEM;
2186 		goto fail1;
2187 	}
2188 
2189 	/*
2190 	 * Allocate an active TID to initiate a TCP connection.
2191 	 */
2192 	ep->atid = cxgb4_alloc_atid(ep->com.dev->rdev.lldi.tids, ep);
2193 	if (ep->atid == -1) {
2194 		pr_err("%s - cannot alloc atid\n", __func__);
2195 		err = -ENOMEM;
2196 		goto fail2;
2197 	}
2198 	err = xa_insert_irq(&ep->com.dev->atids, ep->atid, ep, GFP_KERNEL);
2199 	if (err)
2200 		goto fail2a;
2201 
2202 	/* find a route */
2203 	if (ep->com.cm_id->m_local_addr.ss_family == AF_INET) {
2204 		ep->dst = cxgb_find_route(&ep->com.dev->rdev.lldi, get_real_dev,
2205 					  laddr->sin_addr.s_addr,
2206 					  raddr->sin_addr.s_addr,
2207 					  laddr->sin_port,
2208 					  raddr->sin_port, ep->com.cm_id->tos);
2209 		iptype = 4;
2210 		ra = (__u8 *)&raddr->sin_addr;
2211 	} else {
2212 		ep->dst = cxgb_find_route6(&ep->com.dev->rdev.lldi,
2213 					   get_real_dev,
2214 					   laddr6->sin6_addr.s6_addr,
2215 					   raddr6->sin6_addr.s6_addr,
2216 					   laddr6->sin6_port,
2217 					   raddr6->sin6_port,
2218 					   ep->com.cm_id->tos,
2219 					   raddr6->sin6_scope_id);
2220 		iptype = 6;
2221 		ra = (__u8 *)&raddr6->sin6_addr;
2222 	}
2223 	if (!ep->dst) {
2224 		pr_err("%s - cannot find route\n", __func__);
2225 		err = -EHOSTUNREACH;
2226 		goto fail3;
2227 	}
2228 	err = import_ep(ep, iptype, ra, ep->dst, ep->com.dev, false,
2229 			ep->com.dev->rdev.lldi.adapter_type,
2230 			ep->com.cm_id->tos);
2231 	if (err) {
2232 		pr_err("%s - cannot alloc l2e\n", __func__);
2233 		goto fail4;
2234 	}
2235 
2236 	pr_debug("txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n",
2237 		 ep->txq_idx, ep->tx_chan, ep->smac_idx, ep->rss_qid,
2238 		 ep->l2t->idx);
2239 
2240 	state_set(&ep->com, CONNECTING);
2241 	ep->tos = ep->com.cm_id->tos;
2242 
2243 	/* send connect request to rnic */
2244 	err = send_connect(ep);
2245 	if (!err)
2246 		goto out;
2247 
2248 	cxgb4_l2t_release(ep->l2t);
2249 fail4:
2250 	dst_release(ep->dst);
2251 fail3:
2252 	xa_erase_irq(&ep->com.dev->atids, ep->atid);
2253 fail2a:
2254 	cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
2255 fail2:
2256 	/*
2257 	 * remember to send notification to upper layer.
2258 	 * We are in here so the upper layer is not aware that this is
2259 	 * re-connect attempt and so, upper layer is still waiting for
2260 	 * response of 1st connect request.
2261 	 */
2262 	connect_reply_upcall(ep, -ECONNRESET);
2263 fail1:
2264 	c4iw_put_ep(&ep->com);
2265 out:
2266 	return err;
2267 }
2268 
2269 static int act_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2270 {
2271 	struct c4iw_ep *ep;
2272 	struct cpl_act_open_rpl *rpl = cplhdr(skb);
2273 	unsigned int atid = TID_TID_G(AOPEN_ATID_G(
2274 				      ntohl(rpl->atid_status)));
2275 	struct tid_info *t = dev->rdev.lldi.tids;
2276 	int status = AOPEN_STATUS_G(ntohl(rpl->atid_status));
2277 	struct sockaddr_in *la;
2278 	struct sockaddr_in *ra;
2279 	struct sockaddr_in6 *la6;
2280 	struct sockaddr_in6 *ra6;
2281 	int ret = 0;
2282 
2283 	ep = lookup_atid(t, atid);
2284 	if (!ep)
2285 		return -EINVAL;
2286 
2287 	la = (struct sockaddr_in *)&ep->com.local_addr;
2288 	ra = (struct sockaddr_in *)&ep->com.remote_addr;
2289 	la6 = (struct sockaddr_in6 *)&ep->com.local_addr;
2290 	ra6 = (struct sockaddr_in6 *)&ep->com.remote_addr;
2291 
2292 	pr_debug("ep %p atid %u status %u errno %d\n", ep, atid,
2293 		 status, status2errno(status));
2294 
2295 	if (cxgb_is_neg_adv(status)) {
2296 		pr_debug("Connection problems for atid %u status %u (%s)\n",
2297 			 atid, status, neg_adv_str(status));
2298 		ep->stats.connect_neg_adv++;
2299 		mutex_lock(&dev->rdev.stats.lock);
2300 		dev->rdev.stats.neg_adv++;
2301 		mutex_unlock(&dev->rdev.stats.lock);
2302 		return 0;
2303 	}
2304 
2305 	set_bit(ACT_OPEN_RPL, &ep->com.history);
2306 
2307 	/*
2308 	 * Log interesting failures.
2309 	 */
2310 	switch (status) {
2311 	case CPL_ERR_CONN_RESET:
2312 	case CPL_ERR_CONN_TIMEDOUT:
2313 		break;
2314 	case CPL_ERR_TCAM_FULL:
2315 		mutex_lock(&dev->rdev.stats.lock);
2316 		dev->rdev.stats.tcam_full++;
2317 		mutex_unlock(&dev->rdev.stats.lock);
2318 		if (ep->com.local_addr.ss_family == AF_INET &&
2319 		    dev->rdev.lldi.enable_fw_ofld_conn) {
2320 			ret = send_fw_act_open_req(ep, TID_TID_G(AOPEN_ATID_G(
2321 						   ntohl(rpl->atid_status))));
2322 			if (ret)
2323 				goto fail;
2324 			return 0;
2325 		}
2326 		break;
2327 	case CPL_ERR_CONN_EXIST:
2328 		if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
2329 			set_bit(ACT_RETRY_INUSE, &ep->com.history);
2330 			if (ep->com.remote_addr.ss_family == AF_INET6) {
2331 				struct sockaddr_in6 *sin6 =
2332 						(struct sockaddr_in6 *)
2333 						&ep->com.local_addr;
2334 				cxgb4_clip_release(
2335 						ep->com.dev->rdev.lldi.ports[0],
2336 						(const u32 *)
2337 						&sin6->sin6_addr.s6_addr, 1);
2338 			}
2339 			xa_erase_irq(&ep->com.dev->atids, atid);
2340 			cxgb4_free_atid(t, atid);
2341 			dst_release(ep->dst);
2342 			cxgb4_l2t_release(ep->l2t);
2343 			c4iw_reconnect(ep);
2344 			return 0;
2345 		}
2346 		break;
2347 	default:
2348 		if (ep->com.local_addr.ss_family == AF_INET) {
2349 			pr_info("Active open failure - atid %u status %u errno %d %pI4:%u->%pI4:%u\n",
2350 				atid, status, status2errno(status),
2351 				&la->sin_addr.s_addr, ntohs(la->sin_port),
2352 				&ra->sin_addr.s_addr, ntohs(ra->sin_port));
2353 		} else {
2354 			pr_info("Active open failure - atid %u status %u errno %d %pI6:%u->%pI6:%u\n",
2355 				atid, status, status2errno(status),
2356 				la6->sin6_addr.s6_addr, ntohs(la6->sin6_port),
2357 				ra6->sin6_addr.s6_addr, ntohs(ra6->sin6_port));
2358 		}
2359 		break;
2360 	}
2361 
2362 fail:
2363 	connect_reply_upcall(ep, status2errno(status));
2364 	state_set(&ep->com, DEAD);
2365 
2366 	if (ep->com.remote_addr.ss_family == AF_INET6) {
2367 		struct sockaddr_in6 *sin6 =
2368 			(struct sockaddr_in6 *)&ep->com.local_addr;
2369 		cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
2370 				   (const u32 *)&sin6->sin6_addr.s6_addr, 1);
2371 	}
2372 	if (status && act_open_has_tid(status))
2373 		cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, GET_TID(rpl),
2374 				 ep->com.local_addr.ss_family);
2375 
2376 	xa_erase_irq(&ep->com.dev->atids, atid);
2377 	cxgb4_free_atid(t, atid);
2378 	dst_release(ep->dst);
2379 	cxgb4_l2t_release(ep->l2t);
2380 	c4iw_put_ep(&ep->com);
2381 
2382 	return 0;
2383 }
2384 
2385 static int pass_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2386 {
2387 	struct cpl_pass_open_rpl *rpl = cplhdr(skb);
2388 	unsigned int stid = GET_TID(rpl);
2389 	struct c4iw_listen_ep *ep = get_ep_from_stid(dev, stid);
2390 
2391 	if (!ep) {
2392 		pr_warn("%s stid %d lookup failure!\n", __func__, stid);
2393 		goto out;
2394 	}
2395 	pr_debug("ep %p status %d error %d\n", ep,
2396 		 rpl->status, status2errno(rpl->status));
2397 	c4iw_wake_up_noref(ep->com.wr_waitp, status2errno(rpl->status));
2398 	c4iw_put_ep(&ep->com);
2399 out:
2400 	return 0;
2401 }
2402 
2403 static int close_listsrv_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2404 {
2405 	struct cpl_close_listsvr_rpl *rpl = cplhdr(skb);
2406 	unsigned int stid = GET_TID(rpl);
2407 	struct c4iw_listen_ep *ep = get_ep_from_stid(dev, stid);
2408 
2409 	if (!ep) {
2410 		pr_warn("%s stid %d lookup failure!\n", __func__, stid);
2411 		goto out;
2412 	}
2413 	pr_debug("ep %p\n", ep);
2414 	c4iw_wake_up_noref(ep->com.wr_waitp, status2errno(rpl->status));
2415 	c4iw_put_ep(&ep->com);
2416 out:
2417 	return 0;
2418 }
2419 
2420 static int accept_cr(struct c4iw_ep *ep, struct sk_buff *skb,
2421 		     struct cpl_pass_accept_req *req)
2422 {
2423 	struct cpl_pass_accept_rpl *rpl;
2424 	unsigned int mtu_idx;
2425 	u64 opt0;
2426 	u32 opt2;
2427 	u32 wscale;
2428 	struct cpl_t5_pass_accept_rpl *rpl5 = NULL;
2429 	int win;
2430 	enum chip_type adapter_type = ep->com.dev->rdev.lldi.adapter_type;
2431 
2432 	pr_debug("ep %p tid %u\n", ep, ep->hwtid);
2433 	cxgb_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx,
2434 		      enable_tcp_timestamps && req->tcpopt.tstamp,
2435 		      (ep->com.remote_addr.ss_family == AF_INET) ? 0 : 1);
2436 	wscale = cxgb_compute_wscale(rcv_win);
2437 
2438 	/*
2439 	 * Specify the largest window that will fit in opt0. The
2440 	 * remainder will be specified in the rx_data_ack.
2441 	 */
2442 	win = ep->rcv_win >> 10;
2443 	if (win > RCV_BUFSIZ_M)
2444 		win = RCV_BUFSIZ_M;
2445 	opt0 = (nocong ? NO_CONG_F : 0) |
2446 	       KEEP_ALIVE_F |
2447 	       DELACK_F |
2448 	       WND_SCALE_V(wscale) |
2449 	       MSS_IDX_V(mtu_idx) |
2450 	       L2T_IDX_V(ep->l2t->idx) |
2451 	       TX_CHAN_V(ep->tx_chan) |
2452 	       SMAC_SEL_V(ep->smac_idx) |
2453 	       DSCP_V(ep->tos >> 2) |
2454 	       ULP_MODE_V(ULP_MODE_TCPDDP) |
2455 	       RCV_BUFSIZ_V(win);
2456 	opt2 = RX_CHANNEL_V(0) |
2457 	       RSS_QUEUE_VALID_F | RSS_QUEUE_V(ep->rss_qid);
2458 
2459 	if (enable_tcp_timestamps && req->tcpopt.tstamp)
2460 		opt2 |= TSTAMPS_EN_F;
2461 	if (enable_tcp_sack && req->tcpopt.sack)
2462 		opt2 |= SACK_EN_F;
2463 	if (wscale && enable_tcp_window_scaling)
2464 		opt2 |= WND_SCALE_EN_F;
2465 	if (enable_ecn) {
2466 		const struct tcphdr *tcph;
2467 		u32 hlen = ntohl(req->hdr_len);
2468 
2469 		if (CHELSIO_CHIP_VERSION(adapter_type) <= CHELSIO_T5)
2470 			tcph = (const void *)(req + 1) + ETH_HDR_LEN_G(hlen) +
2471 				IP_HDR_LEN_G(hlen);
2472 		else
2473 			tcph = (const void *)(req + 1) +
2474 				T6_ETH_HDR_LEN_G(hlen) + T6_IP_HDR_LEN_G(hlen);
2475 		if (tcph->ece && tcph->cwr)
2476 			opt2 |= CCTRL_ECN_V(1);
2477 	}
2478 
2479 	if (!is_t4(adapter_type)) {
2480 		u32 isn = (get_random_u32() & ~7UL) - 1;
2481 
2482 		skb = get_skb(skb, roundup(sizeof(*rpl5), 16), GFP_KERNEL);
2483 		rpl5 = __skb_put_zero(skb, roundup(sizeof(*rpl5), 16));
2484 		rpl = (void *)rpl5;
2485 		INIT_TP_WR_CPL(rpl5, CPL_PASS_ACCEPT_RPL, ep->hwtid);
2486 		opt2 |= T5_OPT_2_VALID_F;
2487 		opt2 |= CONG_CNTRL_V(CONG_ALG_TAHOE);
2488 		opt2 |= T5_ISS_F;
2489 		if (peer2peer)
2490 			isn += 4;
2491 		rpl5->iss = cpu_to_be32(isn);
2492 		pr_debug("iss %u\n", be32_to_cpu(rpl5->iss));
2493 	} else {
2494 		skb = get_skb(skb, sizeof(*rpl), GFP_KERNEL);
2495 		rpl = __skb_put_zero(skb, sizeof(*rpl));
2496 		INIT_TP_WR_CPL(rpl, CPL_PASS_ACCEPT_RPL, ep->hwtid);
2497 	}
2498 
2499 	rpl->opt0 = cpu_to_be64(opt0);
2500 	rpl->opt2 = cpu_to_be32(opt2);
2501 	set_wr_txq(skb, CPL_PRIORITY_SETUP, ep->ctrlq_idx);
2502 	t4_set_arp_err_handler(skb, ep, pass_accept_rpl_arp_failure);
2503 
2504 	return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
2505 }
2506 
2507 static void reject_cr(struct c4iw_dev *dev, u32 hwtid, struct sk_buff *skb)
2508 {
2509 	pr_debug("c4iw_dev %p tid %u\n", dev, hwtid);
2510 	skb_trim(skb, sizeof(struct cpl_tid_release));
2511 	release_tid(&dev->rdev, hwtid, skb);
2512 	return;
2513 }
2514 
2515 static int pass_accept_req(struct c4iw_dev *dev, struct sk_buff *skb)
2516 {
2517 	struct c4iw_ep *child_ep = NULL, *parent_ep;
2518 	struct cpl_pass_accept_req *req = cplhdr(skb);
2519 	unsigned int stid = PASS_OPEN_TID_G(ntohl(req->tos_stid));
2520 	struct tid_info *t = dev->rdev.lldi.tids;
2521 	unsigned int hwtid = GET_TID(req);
2522 	struct dst_entry *dst;
2523 	__u8 local_ip[16], peer_ip[16];
2524 	__be16 local_port, peer_port;
2525 	struct sockaddr_in6 *sin6;
2526 	int err;
2527 	u16 peer_mss = ntohs(req->tcpopt.mss);
2528 	int iptype;
2529 	unsigned short hdrs;
2530 	u8 tos;
2531 
2532 	parent_ep = (struct c4iw_ep *)get_ep_from_stid(dev, stid);
2533 	if (!parent_ep) {
2534 		pr_err("%s connect request on invalid stid %d\n",
2535 		       __func__, stid);
2536 		goto reject;
2537 	}
2538 
2539 	if (state_read(&parent_ep->com) != LISTEN) {
2540 		pr_err("%s - listening ep not in LISTEN\n", __func__);
2541 		goto reject;
2542 	}
2543 
2544 	if (parent_ep->com.cm_id->tos_set)
2545 		tos = parent_ep->com.cm_id->tos;
2546 	else
2547 		tos = PASS_OPEN_TOS_G(ntohl(req->tos_stid));
2548 
2549 	cxgb_get_4tuple(req, parent_ep->com.dev->rdev.lldi.adapter_type,
2550 			&iptype, local_ip, peer_ip, &local_port, &peer_port);
2551 
2552 	/* Find output route */
2553 	if (iptype == 4)  {
2554 		pr_debug("parent ep %p hwtid %u laddr %pI4 raddr %pI4 lport %d rport %d peer_mss %d\n"
2555 			 , parent_ep, hwtid,
2556 			 local_ip, peer_ip, ntohs(local_port),
2557 			 ntohs(peer_port), peer_mss);
2558 		dst = cxgb_find_route(&dev->rdev.lldi, get_real_dev,
2559 				      *(__be32 *)local_ip, *(__be32 *)peer_ip,
2560 				      local_port, peer_port, tos);
2561 	} else {
2562 		pr_debug("parent ep %p hwtid %u laddr %pI6 raddr %pI6 lport %d rport %d peer_mss %d\n"
2563 			 , parent_ep, hwtid,
2564 			 local_ip, peer_ip, ntohs(local_port),
2565 			 ntohs(peer_port), peer_mss);
2566 		dst = cxgb_find_route6(&dev->rdev.lldi, get_real_dev,
2567 				local_ip, peer_ip, local_port, peer_port,
2568 				tos,
2569 				((struct sockaddr_in6 *)
2570 				 &parent_ep->com.local_addr)->sin6_scope_id);
2571 	}
2572 	if (!dst) {
2573 		pr_err("%s - failed to find dst entry!\n", __func__);
2574 		goto reject;
2575 	}
2576 
2577 	child_ep = alloc_ep(sizeof(*child_ep), GFP_KERNEL);
2578 	if (!child_ep) {
2579 		pr_err("%s - failed to allocate ep entry!\n", __func__);
2580 		dst_release(dst);
2581 		goto reject;
2582 	}
2583 
2584 	err = import_ep(child_ep, iptype, peer_ip, dst, dev, false,
2585 			parent_ep->com.dev->rdev.lldi.adapter_type, tos);
2586 	if (err) {
2587 		pr_err("%s - failed to allocate l2t entry!\n", __func__);
2588 		dst_release(dst);
2589 		kfree(child_ep);
2590 		goto reject;
2591 	}
2592 
2593 	hdrs = ((iptype == 4) ? sizeof(struct iphdr) : sizeof(struct ipv6hdr)) +
2594 	       sizeof(struct tcphdr) +
2595 	       ((enable_tcp_timestamps && req->tcpopt.tstamp) ? 12 : 0);
2596 	if (peer_mss && child_ep->mtu > (peer_mss + hdrs))
2597 		child_ep->mtu = peer_mss + hdrs;
2598 
2599 	skb_queue_head_init(&child_ep->com.ep_skb_list);
2600 	if (alloc_ep_skb_list(&child_ep->com.ep_skb_list, CN_MAX_CON_BUF))
2601 		goto fail;
2602 
2603 	state_set(&child_ep->com, CONNECTING);
2604 	child_ep->com.dev = dev;
2605 	child_ep->com.cm_id = NULL;
2606 
2607 	if (iptype == 4) {
2608 		struct sockaddr_in *sin = (struct sockaddr_in *)
2609 			&child_ep->com.local_addr;
2610 
2611 		sin->sin_family = AF_INET;
2612 		sin->sin_port = local_port;
2613 		sin->sin_addr.s_addr = *(__be32 *)local_ip;
2614 
2615 		sin = (struct sockaddr_in *)&child_ep->com.local_addr;
2616 		sin->sin_family = AF_INET;
2617 		sin->sin_port = ((struct sockaddr_in *)
2618 				 &parent_ep->com.local_addr)->sin_port;
2619 		sin->sin_addr.s_addr = *(__be32 *)local_ip;
2620 
2621 		sin = (struct sockaddr_in *)&child_ep->com.remote_addr;
2622 		sin->sin_family = AF_INET;
2623 		sin->sin_port = peer_port;
2624 		sin->sin_addr.s_addr = *(__be32 *)peer_ip;
2625 	} else {
2626 		sin6 = (struct sockaddr_in6 *)&child_ep->com.local_addr;
2627 		sin6->sin6_family = PF_INET6;
2628 		sin6->sin6_port = local_port;
2629 		memcpy(sin6->sin6_addr.s6_addr, local_ip, 16);
2630 
2631 		sin6 = (struct sockaddr_in6 *)&child_ep->com.local_addr;
2632 		sin6->sin6_family = PF_INET6;
2633 		sin6->sin6_port = ((struct sockaddr_in6 *)
2634 				   &parent_ep->com.local_addr)->sin6_port;
2635 		memcpy(sin6->sin6_addr.s6_addr, local_ip, 16);
2636 
2637 		sin6 = (struct sockaddr_in6 *)&child_ep->com.remote_addr;
2638 		sin6->sin6_family = PF_INET6;
2639 		sin6->sin6_port = peer_port;
2640 		memcpy(sin6->sin6_addr.s6_addr, peer_ip, 16);
2641 	}
2642 
2643 	c4iw_get_ep(&parent_ep->com);
2644 	child_ep->parent_ep = parent_ep;
2645 	child_ep->tos = tos;
2646 	child_ep->dst = dst;
2647 	child_ep->hwtid = hwtid;
2648 
2649 	pr_debug("tx_chan %u smac_idx %u rss_qid %u\n",
2650 		 child_ep->tx_chan, child_ep->smac_idx, child_ep->rss_qid);
2651 
2652 	timer_setup(&child_ep->timer, ep_timeout, 0);
2653 	cxgb4_insert_tid(t, child_ep, hwtid,
2654 			 child_ep->com.local_addr.ss_family);
2655 	insert_ep_tid(child_ep);
2656 	if (accept_cr(child_ep, skb, req)) {
2657 		c4iw_put_ep(&parent_ep->com);
2658 		release_ep_resources(child_ep);
2659 	} else {
2660 		set_bit(PASS_ACCEPT_REQ, &child_ep->com.history);
2661 	}
2662 	if (iptype == 6) {
2663 		sin6 = (struct sockaddr_in6 *)&child_ep->com.local_addr;
2664 		cxgb4_clip_get(child_ep->com.dev->rdev.lldi.ports[0],
2665 			       (const u32 *)&sin6->sin6_addr.s6_addr, 1);
2666 	}
2667 	goto out;
2668 fail:
2669 	c4iw_put_ep(&child_ep->com);
2670 reject:
2671 	reject_cr(dev, hwtid, skb);
2672 out:
2673 	if (parent_ep)
2674 		c4iw_put_ep(&parent_ep->com);
2675 	return 0;
2676 }
2677 
2678 static int pass_establish(struct c4iw_dev *dev, struct sk_buff *skb)
2679 {
2680 	struct c4iw_ep *ep;
2681 	struct cpl_pass_establish *req = cplhdr(skb);
2682 	unsigned int tid = GET_TID(req);
2683 	int ret;
2684 	u16 tcp_opt = ntohs(req->tcp_opt);
2685 
2686 	ep = get_ep_from_tid(dev, tid);
2687 	if (!ep)
2688 		return 0;
2689 
2690 	pr_debug("ep %p tid %u\n", ep, ep->hwtid);
2691 	ep->snd_seq = be32_to_cpu(req->snd_isn);
2692 	ep->rcv_seq = be32_to_cpu(req->rcv_isn);
2693 	ep->snd_wscale = TCPOPT_SND_WSCALE_G(tcp_opt);
2694 
2695 	pr_debug("ep %p hwtid %u tcp_opt 0x%02x\n", ep, tid, tcp_opt);
2696 
2697 	set_emss(ep, tcp_opt);
2698 
2699 	dst_confirm(ep->dst);
2700 	mutex_lock(&ep->com.mutex);
2701 	ep->com.state = MPA_REQ_WAIT;
2702 	start_ep_timer(ep);
2703 	set_bit(PASS_ESTAB, &ep->com.history);
2704 	ret = send_flowc(ep);
2705 	mutex_unlock(&ep->com.mutex);
2706 	if (ret)
2707 		c4iw_ep_disconnect(ep, 1, GFP_KERNEL);
2708 	c4iw_put_ep(&ep->com);
2709 
2710 	return 0;
2711 }
2712 
2713 static int peer_close(struct c4iw_dev *dev, struct sk_buff *skb)
2714 {
2715 	struct cpl_peer_close *hdr = cplhdr(skb);
2716 	struct c4iw_ep *ep;
2717 	struct c4iw_qp_attributes attrs;
2718 	int disconnect = 1;
2719 	int release = 0;
2720 	unsigned int tid = GET_TID(hdr);
2721 	int ret;
2722 
2723 	ep = get_ep_from_tid(dev, tid);
2724 	if (!ep)
2725 		return 0;
2726 
2727 	pr_debug("ep %p tid %u\n", ep, ep->hwtid);
2728 	dst_confirm(ep->dst);
2729 
2730 	set_bit(PEER_CLOSE, &ep->com.history);
2731 	mutex_lock(&ep->com.mutex);
2732 	switch (ep->com.state) {
2733 	case MPA_REQ_WAIT:
2734 		__state_set(&ep->com, CLOSING);
2735 		break;
2736 	case MPA_REQ_SENT:
2737 		__state_set(&ep->com, CLOSING);
2738 		connect_reply_upcall(ep, -ECONNRESET);
2739 		break;
2740 	case MPA_REQ_RCVD:
2741 
2742 		/*
2743 		 * We're gonna mark this puppy DEAD, but keep
2744 		 * the reference on it until the ULP accepts or
2745 		 * rejects the CR. Also wake up anyone waiting
2746 		 * in rdma connection migration (see c4iw_accept_cr()).
2747 		 */
2748 		__state_set(&ep->com, CLOSING);
2749 		pr_debug("waking up ep %p tid %u\n", ep, ep->hwtid);
2750 		c4iw_wake_up_noref(ep->com.wr_waitp, -ECONNRESET);
2751 		break;
2752 	case MPA_REP_SENT:
2753 		__state_set(&ep->com, CLOSING);
2754 		pr_debug("waking up ep %p tid %u\n", ep, ep->hwtid);
2755 		c4iw_wake_up_noref(ep->com.wr_waitp, -ECONNRESET);
2756 		break;
2757 	case FPDU_MODE:
2758 		start_ep_timer(ep);
2759 		__state_set(&ep->com, CLOSING);
2760 		attrs.next_state = C4IW_QP_STATE_CLOSING;
2761 		ret = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2762 				       C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2763 		if (ret != -ECONNRESET) {
2764 			peer_close_upcall(ep);
2765 			disconnect = 1;
2766 		}
2767 		break;
2768 	case ABORTING:
2769 		disconnect = 0;
2770 		break;
2771 	case CLOSING:
2772 		__state_set(&ep->com, MORIBUND);
2773 		disconnect = 0;
2774 		break;
2775 	case MORIBUND:
2776 		(void)stop_ep_timer(ep);
2777 		if (ep->com.cm_id && ep->com.qp) {
2778 			attrs.next_state = C4IW_QP_STATE_IDLE;
2779 			c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2780 				       C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2781 		}
2782 		close_complete_upcall(ep, 0);
2783 		__state_set(&ep->com, DEAD);
2784 		release = 1;
2785 		disconnect = 0;
2786 		break;
2787 	case DEAD:
2788 		disconnect = 0;
2789 		break;
2790 	default:
2791 		WARN_ONCE(1, "Bad endpoint state %u\n", ep->com.state);
2792 	}
2793 	mutex_unlock(&ep->com.mutex);
2794 	if (disconnect)
2795 		c4iw_ep_disconnect(ep, 0, GFP_KERNEL);
2796 	if (release)
2797 		release_ep_resources(ep);
2798 	c4iw_put_ep(&ep->com);
2799 	return 0;
2800 }
2801 
2802 static void finish_peer_abort(struct c4iw_dev *dev, struct c4iw_ep *ep)
2803 {
2804 	complete_cached_srq_buffers(ep, ep->srqe_idx);
2805 	if (ep->com.cm_id && ep->com.qp) {
2806 		struct c4iw_qp_attributes attrs;
2807 
2808 		attrs.next_state = C4IW_QP_STATE_ERROR;
2809 		c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2810 			       C4IW_QP_ATTR_NEXT_STATE,	&attrs, 1);
2811 	}
2812 	peer_abort_upcall(ep);
2813 	release_ep_resources(ep);
2814 	c4iw_put_ep(&ep->com);
2815 }
2816 
2817 static int peer_abort(struct c4iw_dev *dev, struct sk_buff *skb)
2818 {
2819 	struct cpl_abort_req_rss6 *req = cplhdr(skb);
2820 	struct c4iw_ep *ep;
2821 	struct sk_buff *rpl_skb;
2822 	struct c4iw_qp_attributes attrs;
2823 	int ret;
2824 	int release = 0;
2825 	unsigned int tid = GET_TID(req);
2826 	u8 status;
2827 	u32 srqidx;
2828 
2829 	u32 len = roundup(sizeof(struct cpl_abort_rpl), 16);
2830 
2831 	ep = get_ep_from_tid(dev, tid);
2832 	if (!ep)
2833 		return 0;
2834 
2835 	status = ABORT_RSS_STATUS_G(be32_to_cpu(req->srqidx_status));
2836 
2837 	if (cxgb_is_neg_adv(status)) {
2838 		pr_debug("Negative advice on abort- tid %u status %d (%s)\n",
2839 			 ep->hwtid, status, neg_adv_str(status));
2840 		ep->stats.abort_neg_adv++;
2841 		mutex_lock(&dev->rdev.stats.lock);
2842 		dev->rdev.stats.neg_adv++;
2843 		mutex_unlock(&dev->rdev.stats.lock);
2844 		goto deref_ep;
2845 	}
2846 
2847 	pr_debug("ep %p tid %u state %u\n", ep, ep->hwtid,
2848 		 ep->com.state);
2849 	set_bit(PEER_ABORT, &ep->com.history);
2850 
2851 	/*
2852 	 * Wake up any threads in rdma_init() or rdma_fini().
2853 	 * However, this is not needed if com state is just
2854 	 * MPA_REQ_SENT
2855 	 */
2856 	if (ep->com.state != MPA_REQ_SENT)
2857 		c4iw_wake_up_noref(ep->com.wr_waitp, -ECONNRESET);
2858 
2859 	mutex_lock(&ep->com.mutex);
2860 	switch (ep->com.state) {
2861 	case CONNECTING:
2862 		c4iw_put_ep(&ep->parent_ep->com);
2863 		break;
2864 	case MPA_REQ_WAIT:
2865 		(void)stop_ep_timer(ep);
2866 		break;
2867 	case MPA_REQ_SENT:
2868 		(void)stop_ep_timer(ep);
2869 		if (status != CPL_ERR_CONN_RESET || mpa_rev == 1 ||
2870 		    (mpa_rev == 2 && ep->tried_with_mpa_v1))
2871 			connect_reply_upcall(ep, -ECONNRESET);
2872 		else {
2873 			/*
2874 			 * we just don't send notification upwards because we
2875 			 * want to retry with mpa_v1 without upper layers even
2876 			 * knowing it.
2877 			 *
2878 			 * do some housekeeping so as to re-initiate the
2879 			 * connection
2880 			 */
2881 			pr_info("%s: mpa_rev=%d. Retrying with mpav1\n",
2882 				__func__, mpa_rev);
2883 			ep->retry_with_mpa_v1 = 1;
2884 		}
2885 		break;
2886 	case MPA_REP_SENT:
2887 		break;
2888 	case MPA_REQ_RCVD:
2889 		break;
2890 	case MORIBUND:
2891 	case CLOSING:
2892 		stop_ep_timer(ep);
2893 		fallthrough;
2894 	case FPDU_MODE:
2895 		if (ep->com.qp && ep->com.qp->srq) {
2896 			srqidx = ABORT_RSS_SRQIDX_G(
2897 					be32_to_cpu(req->srqidx_status));
2898 			if (srqidx) {
2899 				complete_cached_srq_buffers(ep, srqidx);
2900 			} else {
2901 				/* Hold ep ref until finish_peer_abort() */
2902 				c4iw_get_ep(&ep->com);
2903 				__state_set(&ep->com, ABORTING);
2904 				set_bit(PEER_ABORT_IN_PROGRESS, &ep->com.flags);
2905 				read_tcb(ep);
2906 				break;
2907 
2908 			}
2909 		}
2910 
2911 		if (ep->com.cm_id && ep->com.qp) {
2912 			attrs.next_state = C4IW_QP_STATE_ERROR;
2913 			ret = c4iw_modify_qp(ep->com.qp->rhp,
2914 				     ep->com.qp, C4IW_QP_ATTR_NEXT_STATE,
2915 				     &attrs, 1);
2916 			if (ret)
2917 				pr_err("%s - qp <- error failed!\n", __func__);
2918 		}
2919 		peer_abort_upcall(ep);
2920 		break;
2921 	case ABORTING:
2922 		break;
2923 	case DEAD:
2924 		pr_warn("%s PEER_ABORT IN DEAD STATE!!!!\n", __func__);
2925 		mutex_unlock(&ep->com.mutex);
2926 		goto deref_ep;
2927 	default:
2928 		WARN_ONCE(1, "Bad endpoint state %u\n", ep->com.state);
2929 		break;
2930 	}
2931 	dst_confirm(ep->dst);
2932 	if (ep->com.state != ABORTING) {
2933 		__state_set(&ep->com, DEAD);
2934 		/* we don't release if we want to retry with mpa_v1 */
2935 		if (!ep->retry_with_mpa_v1)
2936 			release = 1;
2937 	}
2938 	mutex_unlock(&ep->com.mutex);
2939 
2940 	rpl_skb = skb_dequeue(&ep->com.ep_skb_list);
2941 	if (WARN_ON(!rpl_skb)) {
2942 		release = 1;
2943 		goto out;
2944 	}
2945 
2946 	cxgb_mk_abort_rpl(rpl_skb, len, ep->hwtid, ep->txq_idx);
2947 
2948 	c4iw_ofld_send(&ep->com.dev->rdev, rpl_skb);
2949 out:
2950 	if (release)
2951 		release_ep_resources(ep);
2952 	else if (ep->retry_with_mpa_v1) {
2953 		if (ep->com.remote_addr.ss_family == AF_INET6) {
2954 			struct sockaddr_in6 *sin6 =
2955 					(struct sockaddr_in6 *)
2956 					&ep->com.local_addr;
2957 			cxgb4_clip_release(
2958 					ep->com.dev->rdev.lldi.ports[0],
2959 					(const u32 *)&sin6->sin6_addr.s6_addr,
2960 					1);
2961 		}
2962 		xa_erase_irq(&ep->com.dev->hwtids, ep->hwtid);
2963 		cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, ep->hwtid,
2964 				 ep->com.local_addr.ss_family);
2965 		dst_release(ep->dst);
2966 		cxgb4_l2t_release(ep->l2t);
2967 		c4iw_reconnect(ep);
2968 	}
2969 
2970 deref_ep:
2971 	c4iw_put_ep(&ep->com);
2972 	/* Dereferencing ep, referenced in peer_abort_intr() */
2973 	c4iw_put_ep(&ep->com);
2974 	return 0;
2975 }
2976 
2977 static int close_con_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2978 {
2979 	struct c4iw_ep *ep;
2980 	struct c4iw_qp_attributes attrs;
2981 	struct cpl_close_con_rpl *rpl = cplhdr(skb);
2982 	int release = 0;
2983 	unsigned int tid = GET_TID(rpl);
2984 
2985 	ep = get_ep_from_tid(dev, tid);
2986 	if (!ep)
2987 		return 0;
2988 
2989 	pr_debug("ep %p tid %u\n", ep, ep->hwtid);
2990 
2991 	/* The cm_id may be null if we failed to connect */
2992 	mutex_lock(&ep->com.mutex);
2993 	set_bit(CLOSE_CON_RPL, &ep->com.history);
2994 	switch (ep->com.state) {
2995 	case CLOSING:
2996 		__state_set(&ep->com, MORIBUND);
2997 		break;
2998 	case MORIBUND:
2999 		(void)stop_ep_timer(ep);
3000 		if ((ep->com.cm_id) && (ep->com.qp)) {
3001 			attrs.next_state = C4IW_QP_STATE_IDLE;
3002 			c4iw_modify_qp(ep->com.qp->rhp,
3003 					     ep->com.qp,
3004 					     C4IW_QP_ATTR_NEXT_STATE,
3005 					     &attrs, 1);
3006 		}
3007 		close_complete_upcall(ep, 0);
3008 		__state_set(&ep->com, DEAD);
3009 		release = 1;
3010 		break;
3011 	case ABORTING:
3012 	case DEAD:
3013 		break;
3014 	default:
3015 		WARN_ONCE(1, "Bad endpoint state %u\n", ep->com.state);
3016 		break;
3017 	}
3018 	mutex_unlock(&ep->com.mutex);
3019 	if (release)
3020 		release_ep_resources(ep);
3021 	c4iw_put_ep(&ep->com);
3022 	return 0;
3023 }
3024 
3025 static int terminate(struct c4iw_dev *dev, struct sk_buff *skb)
3026 {
3027 	struct cpl_rdma_terminate *rpl = cplhdr(skb);
3028 	unsigned int tid = GET_TID(rpl);
3029 	struct c4iw_ep *ep;
3030 	struct c4iw_qp_attributes attrs;
3031 
3032 	ep = get_ep_from_tid(dev, tid);
3033 
3034 	if (ep) {
3035 		if (ep->com.qp) {
3036 			pr_warn("TERM received tid %u qpid %u\n", tid,
3037 				ep->com.qp->wq.sq.qid);
3038 			attrs.next_state = C4IW_QP_STATE_TERMINATE;
3039 			c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
3040 				       C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
3041 		}
3042 
3043 		/* As per draft-hilland-iwarp-verbs-v1.0, sec 6.2.3,
3044 		 * when entering the TERM state the RNIC MUST initiate a CLOSE.
3045 		 */
3046 		c4iw_ep_disconnect(ep, 1, GFP_KERNEL);
3047 		c4iw_put_ep(&ep->com);
3048 	} else
3049 		pr_warn("TERM received tid %u no ep/qp\n", tid);
3050 
3051 	return 0;
3052 }
3053 
3054 /*
3055  * Upcall from the adapter indicating data has been transmitted.
3056  * For us its just the single MPA request or reply.  We can now free
3057  * the skb holding the mpa message.
3058  */
3059 static int fw4_ack(struct c4iw_dev *dev, struct sk_buff *skb)
3060 {
3061 	struct c4iw_ep *ep;
3062 	struct cpl_fw4_ack *hdr = cplhdr(skb);
3063 	u8 credits = hdr->credits;
3064 	unsigned int tid = GET_TID(hdr);
3065 
3066 
3067 	ep = get_ep_from_tid(dev, tid);
3068 	if (!ep)
3069 		return 0;
3070 	pr_debug("ep %p tid %u credits %u\n",
3071 		 ep, ep->hwtid, credits);
3072 	if (credits == 0) {
3073 		pr_debug("0 credit ack ep %p tid %u state %u\n",
3074 			 ep, ep->hwtid, state_read(&ep->com));
3075 		goto out;
3076 	}
3077 
3078 	dst_confirm(ep->dst);
3079 	if (ep->mpa_skb) {
3080 		pr_debug("last streaming msg ack ep %p tid %u state %u initiator %u freeing skb\n",
3081 			 ep, ep->hwtid, state_read(&ep->com),
3082 			 ep->mpa_attr.initiator ? 1 : 0);
3083 		mutex_lock(&ep->com.mutex);
3084 		kfree_skb(ep->mpa_skb);
3085 		ep->mpa_skb = NULL;
3086 		if (test_bit(STOP_MPA_TIMER, &ep->com.flags))
3087 			stop_ep_timer(ep);
3088 		mutex_unlock(&ep->com.mutex);
3089 	}
3090 out:
3091 	c4iw_put_ep(&ep->com);
3092 	return 0;
3093 }
3094 
3095 int c4iw_reject_cr(struct iw_cm_id *cm_id, const void *pdata, u8 pdata_len)
3096 {
3097 	int abort;
3098 	struct c4iw_ep *ep = to_ep(cm_id);
3099 
3100 	pr_debug("ep %p tid %u\n", ep, ep->hwtid);
3101 
3102 	mutex_lock(&ep->com.mutex);
3103 	if (ep->com.state != MPA_REQ_RCVD) {
3104 		mutex_unlock(&ep->com.mutex);
3105 		c4iw_put_ep(&ep->com);
3106 		return -ECONNRESET;
3107 	}
3108 	set_bit(ULP_REJECT, &ep->com.history);
3109 	if (mpa_rev == 0)
3110 		abort = 1;
3111 	else
3112 		abort = send_mpa_reject(ep, pdata, pdata_len);
3113 	mutex_unlock(&ep->com.mutex);
3114 
3115 	stop_ep_timer(ep);
3116 	c4iw_ep_disconnect(ep, abort != 0, GFP_KERNEL);
3117 	c4iw_put_ep(&ep->com);
3118 	return 0;
3119 }
3120 
3121 int c4iw_accept_cr(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
3122 {
3123 	int err;
3124 	struct c4iw_qp_attributes attrs;
3125 	enum c4iw_qp_attr_mask mask;
3126 	struct c4iw_ep *ep = to_ep(cm_id);
3127 	struct c4iw_dev *h = to_c4iw_dev(cm_id->device);
3128 	struct c4iw_qp *qp = get_qhp(h, conn_param->qpn);
3129 	int abort = 0;
3130 
3131 	pr_debug("ep %p tid %u\n", ep, ep->hwtid);
3132 
3133 	mutex_lock(&ep->com.mutex);
3134 	if (ep->com.state != MPA_REQ_RCVD) {
3135 		err = -ECONNRESET;
3136 		goto err_out;
3137 	}
3138 
3139 	if (!qp) {
3140 		err = -EINVAL;
3141 		goto err_out;
3142 	}
3143 
3144 	set_bit(ULP_ACCEPT, &ep->com.history);
3145 	if ((conn_param->ord > cur_max_read_depth(ep->com.dev)) ||
3146 	    (conn_param->ird > cur_max_read_depth(ep->com.dev))) {
3147 		err = -EINVAL;
3148 		goto err_abort;
3149 	}
3150 
3151 	if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
3152 		if (conn_param->ord > ep->ird) {
3153 			if (RELAXED_IRD_NEGOTIATION) {
3154 				conn_param->ord = ep->ird;
3155 			} else {
3156 				ep->ird = conn_param->ird;
3157 				ep->ord = conn_param->ord;
3158 				send_mpa_reject(ep, conn_param->private_data,
3159 						conn_param->private_data_len);
3160 				err = -ENOMEM;
3161 				goto err_abort;
3162 			}
3163 		}
3164 		if (conn_param->ird < ep->ord) {
3165 			if (RELAXED_IRD_NEGOTIATION &&
3166 			    ep->ord <= h->rdev.lldi.max_ordird_qp) {
3167 				conn_param->ird = ep->ord;
3168 			} else {
3169 				err = -ENOMEM;
3170 				goto err_abort;
3171 			}
3172 		}
3173 	}
3174 	ep->ird = conn_param->ird;
3175 	ep->ord = conn_param->ord;
3176 
3177 	if (ep->mpa_attr.version == 1) {
3178 		if (peer2peer && ep->ird == 0)
3179 			ep->ird = 1;
3180 	} else {
3181 		if (peer2peer &&
3182 		    (ep->mpa_attr.p2p_type != FW_RI_INIT_P2PTYPE_DISABLED) &&
3183 		    (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ) && ep->ird == 0)
3184 			ep->ird = 1;
3185 	}
3186 
3187 	pr_debug("ird %d ord %d\n", ep->ird, ep->ord);
3188 
3189 	ep->com.cm_id = cm_id;
3190 	ref_cm_id(&ep->com);
3191 	ep->com.qp = qp;
3192 	ref_qp(ep);
3193 
3194 	/* bind QP to EP and move to RTS */
3195 	attrs.mpa_attr = ep->mpa_attr;
3196 	attrs.max_ird = ep->ird;
3197 	attrs.max_ord = ep->ord;
3198 	attrs.llp_stream_handle = ep;
3199 	attrs.next_state = C4IW_QP_STATE_RTS;
3200 
3201 	/* bind QP and TID with INIT_WR */
3202 	mask = C4IW_QP_ATTR_NEXT_STATE |
3203 			     C4IW_QP_ATTR_LLP_STREAM_HANDLE |
3204 			     C4IW_QP_ATTR_MPA_ATTR |
3205 			     C4IW_QP_ATTR_MAX_IRD |
3206 			     C4IW_QP_ATTR_MAX_ORD;
3207 
3208 	err = c4iw_modify_qp(ep->com.qp->rhp,
3209 			     ep->com.qp, mask, &attrs, 1);
3210 	if (err)
3211 		goto err_deref_cm_id;
3212 
3213 	set_bit(STOP_MPA_TIMER, &ep->com.flags);
3214 	err = send_mpa_reply(ep, conn_param->private_data,
3215 			     conn_param->private_data_len);
3216 	if (err)
3217 		goto err_deref_cm_id;
3218 
3219 	__state_set(&ep->com, FPDU_MODE);
3220 	established_upcall(ep);
3221 	mutex_unlock(&ep->com.mutex);
3222 	c4iw_put_ep(&ep->com);
3223 	return 0;
3224 err_deref_cm_id:
3225 	deref_cm_id(&ep->com);
3226 err_abort:
3227 	abort = 1;
3228 err_out:
3229 	mutex_unlock(&ep->com.mutex);
3230 	if (abort)
3231 		c4iw_ep_disconnect(ep, 1, GFP_KERNEL);
3232 	c4iw_put_ep(&ep->com);
3233 	return err;
3234 }
3235 
3236 static int pick_local_ipaddrs(struct c4iw_dev *dev, struct iw_cm_id *cm_id)
3237 {
3238 	struct in_device *ind;
3239 	int found = 0;
3240 	struct sockaddr_in *laddr = (struct sockaddr_in *)&cm_id->m_local_addr;
3241 	struct sockaddr_in *raddr = (struct sockaddr_in *)&cm_id->m_remote_addr;
3242 	const struct in_ifaddr *ifa;
3243 
3244 	ind = in_dev_get(dev->rdev.lldi.ports[0]);
3245 	if (!ind)
3246 		return -EADDRNOTAVAIL;
3247 	rcu_read_lock();
3248 	in_dev_for_each_ifa_rcu(ifa, ind) {
3249 		if (ifa->ifa_flags & IFA_F_SECONDARY)
3250 			continue;
3251 		laddr->sin_addr.s_addr = ifa->ifa_address;
3252 		raddr->sin_addr.s_addr = ifa->ifa_address;
3253 		found = 1;
3254 		break;
3255 	}
3256 	rcu_read_unlock();
3257 
3258 	in_dev_put(ind);
3259 	return found ? 0 : -EADDRNOTAVAIL;
3260 }
3261 
3262 static int get_lladdr(struct net_device *dev, struct in6_addr *addr,
3263 		      unsigned char banned_flags)
3264 {
3265 	struct inet6_dev *idev;
3266 	int err = -EADDRNOTAVAIL;
3267 
3268 	rcu_read_lock();
3269 	idev = __in6_dev_get(dev);
3270 	if (idev != NULL) {
3271 		struct inet6_ifaddr *ifp;
3272 
3273 		read_lock_bh(&idev->lock);
3274 		list_for_each_entry(ifp, &idev->addr_list, if_list) {
3275 			if (ifp->scope == IFA_LINK &&
3276 			    !(ifp->flags & banned_flags)) {
3277 				memcpy(addr, &ifp->addr, 16);
3278 				err = 0;
3279 				break;
3280 			}
3281 		}
3282 		read_unlock_bh(&idev->lock);
3283 	}
3284 	rcu_read_unlock();
3285 	return err;
3286 }
3287 
3288 static int pick_local_ip6addrs(struct c4iw_dev *dev, struct iw_cm_id *cm_id)
3289 {
3290 	struct in6_addr addr;
3291 	struct sockaddr_in6 *la6 = (struct sockaddr_in6 *)&cm_id->m_local_addr;
3292 	struct sockaddr_in6 *ra6 = (struct sockaddr_in6 *)&cm_id->m_remote_addr;
3293 
3294 	if (!get_lladdr(dev->rdev.lldi.ports[0], &addr, IFA_F_TENTATIVE)) {
3295 		memcpy(la6->sin6_addr.s6_addr, &addr, 16);
3296 		memcpy(ra6->sin6_addr.s6_addr, &addr, 16);
3297 		return 0;
3298 	}
3299 	return -EADDRNOTAVAIL;
3300 }
3301 
3302 int c4iw_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
3303 {
3304 	struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
3305 	struct c4iw_ep *ep;
3306 	int err = 0;
3307 	struct sockaddr_in *laddr;
3308 	struct sockaddr_in *raddr;
3309 	struct sockaddr_in6 *laddr6;
3310 	struct sockaddr_in6 *raddr6;
3311 	__u8 *ra;
3312 	int iptype;
3313 
3314 	if ((conn_param->ord > cur_max_read_depth(dev)) ||
3315 	    (conn_param->ird > cur_max_read_depth(dev))) {
3316 		err = -EINVAL;
3317 		goto out;
3318 	}
3319 	ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
3320 	if (!ep) {
3321 		pr_err("%s - cannot alloc ep\n", __func__);
3322 		err = -ENOMEM;
3323 		goto out;
3324 	}
3325 
3326 	skb_queue_head_init(&ep->com.ep_skb_list);
3327 	if (alloc_ep_skb_list(&ep->com.ep_skb_list, CN_MAX_CON_BUF)) {
3328 		err = -ENOMEM;
3329 		goto fail1;
3330 	}
3331 
3332 	timer_setup(&ep->timer, ep_timeout, 0);
3333 	ep->plen = conn_param->private_data_len;
3334 	if (ep->plen)
3335 		memcpy(ep->mpa_pkt + sizeof(struct mpa_message),
3336 		       conn_param->private_data, ep->plen);
3337 	ep->ird = conn_param->ird;
3338 	ep->ord = conn_param->ord;
3339 
3340 	if (peer2peer && ep->ord == 0)
3341 		ep->ord = 1;
3342 
3343 	ep->com.cm_id = cm_id;
3344 	ref_cm_id(&ep->com);
3345 	cm_id->provider_data = ep;
3346 	ep->com.dev = dev;
3347 	ep->com.qp = get_qhp(dev, conn_param->qpn);
3348 	if (!ep->com.qp) {
3349 		pr_warn("%s qpn 0x%x not found!\n", __func__, conn_param->qpn);
3350 		err = -EINVAL;
3351 		goto fail2;
3352 	}
3353 	ref_qp(ep);
3354 	pr_debug("qpn 0x%x qp %p cm_id %p\n", conn_param->qpn,
3355 		 ep->com.qp, cm_id);
3356 
3357 	/*
3358 	 * Allocate an active TID to initiate a TCP connection.
3359 	 */
3360 	ep->atid = cxgb4_alloc_atid(dev->rdev.lldi.tids, ep);
3361 	if (ep->atid == -1) {
3362 		pr_err("%s - cannot alloc atid\n", __func__);
3363 		err = -ENOMEM;
3364 		goto fail2;
3365 	}
3366 	err = xa_insert_irq(&dev->atids, ep->atid, ep, GFP_KERNEL);
3367 	if (err)
3368 		goto fail5;
3369 
3370 	memcpy(&ep->com.local_addr, &cm_id->m_local_addr,
3371 	       sizeof(ep->com.local_addr));
3372 	memcpy(&ep->com.remote_addr, &cm_id->m_remote_addr,
3373 	       sizeof(ep->com.remote_addr));
3374 
3375 	laddr = (struct sockaddr_in *)&ep->com.local_addr;
3376 	raddr = (struct sockaddr_in *)&ep->com.remote_addr;
3377 	laddr6 = (struct sockaddr_in6 *)&ep->com.local_addr;
3378 	raddr6 = (struct sockaddr_in6 *) &ep->com.remote_addr;
3379 
3380 	if (cm_id->m_remote_addr.ss_family == AF_INET) {
3381 		iptype = 4;
3382 		ra = (__u8 *)&raddr->sin_addr;
3383 
3384 		/*
3385 		 * Handle loopback requests to INADDR_ANY.
3386 		 */
3387 		if (raddr->sin_addr.s_addr == htonl(INADDR_ANY)) {
3388 			err = pick_local_ipaddrs(dev, cm_id);
3389 			if (err)
3390 				goto fail3;
3391 		}
3392 
3393 		/* find a route */
3394 		pr_debug("saddr %pI4 sport 0x%x raddr %pI4 rport 0x%x\n",
3395 			 &laddr->sin_addr, ntohs(laddr->sin_port),
3396 			 ra, ntohs(raddr->sin_port));
3397 		ep->dst = cxgb_find_route(&dev->rdev.lldi, get_real_dev,
3398 					  laddr->sin_addr.s_addr,
3399 					  raddr->sin_addr.s_addr,
3400 					  laddr->sin_port,
3401 					  raddr->sin_port, cm_id->tos);
3402 	} else {
3403 		iptype = 6;
3404 		ra = (__u8 *)&raddr6->sin6_addr;
3405 
3406 		/*
3407 		 * Handle loopback requests to INADDR_ANY.
3408 		 */
3409 		if (ipv6_addr_type(&raddr6->sin6_addr) == IPV6_ADDR_ANY) {
3410 			err = pick_local_ip6addrs(dev, cm_id);
3411 			if (err)
3412 				goto fail3;
3413 		}
3414 
3415 		/* find a route */
3416 		pr_debug("saddr %pI6 sport 0x%x raddr %pI6 rport 0x%x\n",
3417 			 laddr6->sin6_addr.s6_addr,
3418 			 ntohs(laddr6->sin6_port),
3419 			 raddr6->sin6_addr.s6_addr, ntohs(raddr6->sin6_port));
3420 		ep->dst = cxgb_find_route6(&dev->rdev.lldi, get_real_dev,
3421 					   laddr6->sin6_addr.s6_addr,
3422 					   raddr6->sin6_addr.s6_addr,
3423 					   laddr6->sin6_port,
3424 					   raddr6->sin6_port, cm_id->tos,
3425 					   raddr6->sin6_scope_id);
3426 	}
3427 	if (!ep->dst) {
3428 		pr_err("%s - cannot find route\n", __func__);
3429 		err = -EHOSTUNREACH;
3430 		goto fail3;
3431 	}
3432 
3433 	err = import_ep(ep, iptype, ra, ep->dst, ep->com.dev, true,
3434 			ep->com.dev->rdev.lldi.adapter_type, cm_id->tos);
3435 	if (err) {
3436 		pr_err("%s - cannot alloc l2e\n", __func__);
3437 		goto fail4;
3438 	}
3439 
3440 	pr_debug("txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n",
3441 		 ep->txq_idx, ep->tx_chan, ep->smac_idx, ep->rss_qid,
3442 		 ep->l2t->idx);
3443 
3444 	state_set(&ep->com, CONNECTING);
3445 	ep->tos = cm_id->tos;
3446 
3447 	/* send connect request to rnic */
3448 	err = send_connect(ep);
3449 	if (!err)
3450 		goto out;
3451 
3452 	cxgb4_l2t_release(ep->l2t);
3453 fail4:
3454 	dst_release(ep->dst);
3455 fail3:
3456 	xa_erase_irq(&ep->com.dev->atids, ep->atid);
3457 fail5:
3458 	cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
3459 fail2:
3460 	skb_queue_purge(&ep->com.ep_skb_list);
3461 	deref_cm_id(&ep->com);
3462 fail1:
3463 	c4iw_put_ep(&ep->com);
3464 out:
3465 	return err;
3466 }
3467 
3468 static int create_server6(struct c4iw_dev *dev, struct c4iw_listen_ep *ep)
3469 {
3470 	int err;
3471 	struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)
3472 				    &ep->com.local_addr;
3473 
3474 	if (ipv6_addr_type(&sin6->sin6_addr) != IPV6_ADDR_ANY) {
3475 		err = cxgb4_clip_get(ep->com.dev->rdev.lldi.ports[0],
3476 				     (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3477 		if (err)
3478 			return err;
3479 	}
3480 	c4iw_init_wr_wait(ep->com.wr_waitp);
3481 	err = cxgb4_create_server6(ep->com.dev->rdev.lldi.ports[0],
3482 				   ep->stid, &sin6->sin6_addr,
3483 				   sin6->sin6_port,
3484 				   ep->com.dev->rdev.lldi.rxq_ids[0]);
3485 	if (!err)
3486 		err = c4iw_wait_for_reply(&ep->com.dev->rdev,
3487 					  ep->com.wr_waitp,
3488 					  0, 0, __func__);
3489 	else if (err > 0)
3490 		err = net_xmit_errno(err);
3491 	if (err) {
3492 		cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
3493 				   (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3494 		pr_err("cxgb4_create_server6/filter failed err %d stid %d laddr %pI6 lport %d\n",
3495 		       err, ep->stid,
3496 		       sin6->sin6_addr.s6_addr, ntohs(sin6->sin6_port));
3497 	}
3498 	return err;
3499 }
3500 
3501 static int create_server4(struct c4iw_dev *dev, struct c4iw_listen_ep *ep)
3502 {
3503 	int err;
3504 	struct sockaddr_in *sin = (struct sockaddr_in *)
3505 				  &ep->com.local_addr;
3506 
3507 	if (dev->rdev.lldi.enable_fw_ofld_conn) {
3508 		do {
3509 			err = cxgb4_create_server_filter(
3510 				ep->com.dev->rdev.lldi.ports[0], ep->stid,
3511 				sin->sin_addr.s_addr, sin->sin_port, 0,
3512 				ep->com.dev->rdev.lldi.rxq_ids[0], 0, 0);
3513 			if (err == -EBUSY) {
3514 				if (c4iw_fatal_error(&ep->com.dev->rdev)) {
3515 					err = -EIO;
3516 					break;
3517 				}
3518 				set_current_state(TASK_UNINTERRUPTIBLE);
3519 				schedule_timeout(usecs_to_jiffies(100));
3520 			}
3521 		} while (err == -EBUSY);
3522 	} else {
3523 		c4iw_init_wr_wait(ep->com.wr_waitp);
3524 		err = cxgb4_create_server(ep->com.dev->rdev.lldi.ports[0],
3525 				ep->stid, sin->sin_addr.s_addr, sin->sin_port,
3526 				0, ep->com.dev->rdev.lldi.rxq_ids[0]);
3527 		if (!err)
3528 			err = c4iw_wait_for_reply(&ep->com.dev->rdev,
3529 						  ep->com.wr_waitp,
3530 						  0, 0, __func__);
3531 		else if (err > 0)
3532 			err = net_xmit_errno(err);
3533 	}
3534 	if (err)
3535 		pr_err("cxgb4_create_server/filter failed err %d stid %d laddr %pI4 lport %d\n"
3536 		       , err, ep->stid,
3537 		       &sin->sin_addr, ntohs(sin->sin_port));
3538 	return err;
3539 }
3540 
3541 int c4iw_create_listen(struct iw_cm_id *cm_id, int backlog)
3542 {
3543 	int err = 0;
3544 	struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
3545 	struct c4iw_listen_ep *ep;
3546 
3547 	might_sleep();
3548 
3549 	ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
3550 	if (!ep) {
3551 		pr_err("%s - cannot alloc ep\n", __func__);
3552 		err = -ENOMEM;
3553 		goto fail1;
3554 	}
3555 	skb_queue_head_init(&ep->com.ep_skb_list);
3556 	pr_debug("ep %p\n", ep);
3557 	ep->com.cm_id = cm_id;
3558 	ref_cm_id(&ep->com);
3559 	ep->com.dev = dev;
3560 	ep->backlog = backlog;
3561 	memcpy(&ep->com.local_addr, &cm_id->m_local_addr,
3562 	       sizeof(ep->com.local_addr));
3563 
3564 	/*
3565 	 * Allocate a server TID.
3566 	 */
3567 	if (dev->rdev.lldi.enable_fw_ofld_conn &&
3568 	    ep->com.local_addr.ss_family == AF_INET)
3569 		ep->stid = cxgb4_alloc_sftid(dev->rdev.lldi.tids,
3570 					     cm_id->m_local_addr.ss_family, ep);
3571 	else
3572 		ep->stid = cxgb4_alloc_stid(dev->rdev.lldi.tids,
3573 					    cm_id->m_local_addr.ss_family, ep);
3574 
3575 	if (ep->stid == -1) {
3576 		pr_err("%s - cannot alloc stid\n", __func__);
3577 		err = -ENOMEM;
3578 		goto fail2;
3579 	}
3580 	err = xa_insert_irq(&dev->stids, ep->stid, ep, GFP_KERNEL);
3581 	if (err)
3582 		goto fail3;
3583 
3584 	state_set(&ep->com, LISTEN);
3585 	if (ep->com.local_addr.ss_family == AF_INET)
3586 		err = create_server4(dev, ep);
3587 	else
3588 		err = create_server6(dev, ep);
3589 	if (!err) {
3590 		cm_id->provider_data = ep;
3591 		goto out;
3592 	}
3593 	xa_erase_irq(&ep->com.dev->stids, ep->stid);
3594 fail3:
3595 	cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid,
3596 			ep->com.local_addr.ss_family);
3597 fail2:
3598 	deref_cm_id(&ep->com);
3599 	c4iw_put_ep(&ep->com);
3600 fail1:
3601 out:
3602 	return err;
3603 }
3604 
3605 int c4iw_destroy_listen(struct iw_cm_id *cm_id)
3606 {
3607 	int err;
3608 	struct c4iw_listen_ep *ep = to_listen_ep(cm_id);
3609 
3610 	pr_debug("ep %p\n", ep);
3611 
3612 	might_sleep();
3613 	state_set(&ep->com, DEAD);
3614 	if (ep->com.dev->rdev.lldi.enable_fw_ofld_conn &&
3615 	    ep->com.local_addr.ss_family == AF_INET) {
3616 		err = cxgb4_remove_server_filter(
3617 			ep->com.dev->rdev.lldi.ports[0], ep->stid,
3618 			ep->com.dev->rdev.lldi.rxq_ids[0], false);
3619 	} else {
3620 		struct sockaddr_in6 *sin6;
3621 		c4iw_init_wr_wait(ep->com.wr_waitp);
3622 		err = cxgb4_remove_server(
3623 				ep->com.dev->rdev.lldi.ports[0], ep->stid,
3624 				ep->com.dev->rdev.lldi.rxq_ids[0],
3625 				ep->com.local_addr.ss_family == AF_INET6);
3626 		if (err)
3627 			goto done;
3628 		err = c4iw_wait_for_reply(&ep->com.dev->rdev, ep->com.wr_waitp,
3629 					  0, 0, __func__);
3630 		sin6 = (struct sockaddr_in6 *)&ep->com.local_addr;
3631 		cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
3632 				   (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3633 	}
3634 	xa_erase_irq(&ep->com.dev->stids, ep->stid);
3635 	cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid,
3636 			ep->com.local_addr.ss_family);
3637 done:
3638 	deref_cm_id(&ep->com);
3639 	c4iw_put_ep(&ep->com);
3640 	return err;
3641 }
3642 
3643 int c4iw_ep_disconnect(struct c4iw_ep *ep, int abrupt, gfp_t gfp)
3644 {
3645 	int ret = 0;
3646 	int close = 0;
3647 	int fatal = 0;
3648 	struct c4iw_rdev *rdev;
3649 
3650 	mutex_lock(&ep->com.mutex);
3651 
3652 	pr_debug("ep %p state %s, abrupt %d\n", ep,
3653 		 states[ep->com.state], abrupt);
3654 
3655 	/*
3656 	 * Ref the ep here in case we have fatal errors causing the
3657 	 * ep to be released and freed.
3658 	 */
3659 	c4iw_get_ep(&ep->com);
3660 
3661 	rdev = &ep->com.dev->rdev;
3662 	if (c4iw_fatal_error(rdev)) {
3663 		fatal = 1;
3664 		close_complete_upcall(ep, -EIO);
3665 		ep->com.state = DEAD;
3666 	}
3667 	switch (ep->com.state) {
3668 	case MPA_REQ_WAIT:
3669 	case MPA_REQ_SENT:
3670 	case MPA_REQ_RCVD:
3671 	case MPA_REP_SENT:
3672 	case FPDU_MODE:
3673 	case CONNECTING:
3674 		close = 1;
3675 		if (abrupt)
3676 			ep->com.state = ABORTING;
3677 		else {
3678 			ep->com.state = CLOSING;
3679 
3680 			/*
3681 			 * if we close before we see the fw4_ack() then we fix
3682 			 * up the timer state since we're reusing it.
3683 			 */
3684 			if (ep->mpa_skb &&
3685 			    test_bit(STOP_MPA_TIMER, &ep->com.flags)) {
3686 				clear_bit(STOP_MPA_TIMER, &ep->com.flags);
3687 				stop_ep_timer(ep);
3688 			}
3689 			start_ep_timer(ep);
3690 		}
3691 		set_bit(CLOSE_SENT, &ep->com.flags);
3692 		break;
3693 	case CLOSING:
3694 		if (!test_and_set_bit(CLOSE_SENT, &ep->com.flags)) {
3695 			close = 1;
3696 			if (abrupt) {
3697 				(void)stop_ep_timer(ep);
3698 				ep->com.state = ABORTING;
3699 			} else
3700 				ep->com.state = MORIBUND;
3701 		}
3702 		break;
3703 	case MORIBUND:
3704 	case ABORTING:
3705 	case DEAD:
3706 		pr_debug("ignoring disconnect ep %p state %u\n",
3707 			 ep, ep->com.state);
3708 		break;
3709 	default:
3710 		WARN_ONCE(1, "Bad endpoint state %u\n", ep->com.state);
3711 		break;
3712 	}
3713 
3714 	if (close) {
3715 		if (abrupt) {
3716 			set_bit(EP_DISC_ABORT, &ep->com.history);
3717 			ret = send_abort(ep);
3718 		} else {
3719 			set_bit(EP_DISC_CLOSE, &ep->com.history);
3720 			ret = send_halfclose(ep);
3721 		}
3722 		if (ret) {
3723 			set_bit(EP_DISC_FAIL, &ep->com.history);
3724 			if (!abrupt) {
3725 				stop_ep_timer(ep);
3726 				close_complete_upcall(ep, -EIO);
3727 			}
3728 			if (ep->com.qp) {
3729 				struct c4iw_qp_attributes attrs;
3730 
3731 				attrs.next_state = C4IW_QP_STATE_ERROR;
3732 				ret = c4iw_modify_qp(ep->com.qp->rhp,
3733 						     ep->com.qp,
3734 						     C4IW_QP_ATTR_NEXT_STATE,
3735 						     &attrs, 1);
3736 				if (ret)
3737 					pr_err("%s - qp <- error failed!\n",
3738 					       __func__);
3739 			}
3740 			fatal = 1;
3741 		}
3742 	}
3743 	mutex_unlock(&ep->com.mutex);
3744 	c4iw_put_ep(&ep->com);
3745 	if (fatal)
3746 		release_ep_resources(ep);
3747 	return ret;
3748 }
3749 
3750 static void active_ofld_conn_reply(struct c4iw_dev *dev, struct sk_buff *skb,
3751 			struct cpl_fw6_msg_ofld_connection_wr_rpl *req)
3752 {
3753 	struct c4iw_ep *ep;
3754 	int atid = be32_to_cpu(req->tid);
3755 
3756 	ep = (struct c4iw_ep *)lookup_atid(dev->rdev.lldi.tids,
3757 					   (__force u32) req->tid);
3758 	if (!ep)
3759 		return;
3760 
3761 	switch (req->retval) {
3762 	case FW_ENOMEM:
3763 		set_bit(ACT_RETRY_NOMEM, &ep->com.history);
3764 		if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
3765 			send_fw_act_open_req(ep, atid);
3766 			return;
3767 		}
3768 		fallthrough;
3769 	case FW_EADDRINUSE:
3770 		set_bit(ACT_RETRY_INUSE, &ep->com.history);
3771 		if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
3772 			send_fw_act_open_req(ep, atid);
3773 			return;
3774 		}
3775 		break;
3776 	default:
3777 		pr_info("%s unexpected ofld conn wr retval %d\n",
3778 		       __func__, req->retval);
3779 		break;
3780 	}
3781 	pr_err("active ofld_connect_wr failure %d atid %d\n",
3782 	       req->retval, atid);
3783 	mutex_lock(&dev->rdev.stats.lock);
3784 	dev->rdev.stats.act_ofld_conn_fails++;
3785 	mutex_unlock(&dev->rdev.stats.lock);
3786 	connect_reply_upcall(ep, status2errno(req->retval));
3787 	state_set(&ep->com, DEAD);
3788 	if (ep->com.remote_addr.ss_family == AF_INET6) {
3789 		struct sockaddr_in6 *sin6 =
3790 			(struct sockaddr_in6 *)&ep->com.local_addr;
3791 		cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
3792 				   (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3793 	}
3794 	xa_erase_irq(&dev->atids, atid);
3795 	cxgb4_free_atid(dev->rdev.lldi.tids, atid);
3796 	dst_release(ep->dst);
3797 	cxgb4_l2t_release(ep->l2t);
3798 	c4iw_put_ep(&ep->com);
3799 }
3800 
3801 static void passive_ofld_conn_reply(struct c4iw_dev *dev, struct sk_buff *skb,
3802 			struct cpl_fw6_msg_ofld_connection_wr_rpl *req)
3803 {
3804 	struct sk_buff *rpl_skb;
3805 	struct cpl_pass_accept_req *cpl;
3806 	int ret;
3807 
3808 	rpl_skb = (struct sk_buff *)(unsigned long)req->cookie;
3809 	if (req->retval) {
3810 		pr_err("%s passive open failure %d\n", __func__, req->retval);
3811 		mutex_lock(&dev->rdev.stats.lock);
3812 		dev->rdev.stats.pas_ofld_conn_fails++;
3813 		mutex_unlock(&dev->rdev.stats.lock);
3814 		kfree_skb(rpl_skb);
3815 	} else {
3816 		cpl = (struct cpl_pass_accept_req *)cplhdr(rpl_skb);
3817 		OPCODE_TID(cpl) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_REQ,
3818 					(__force u32) htonl(
3819 					(__force u32) req->tid)));
3820 		ret = pass_accept_req(dev, rpl_skb);
3821 		if (!ret)
3822 			kfree_skb(rpl_skb);
3823 	}
3824 	return;
3825 }
3826 
3827 static inline u64 t4_tcb_get_field64(__be64 *tcb, u16 word)
3828 {
3829 	u64 tlo = be64_to_cpu(tcb[((31 - word) / 2)]);
3830 	u64 thi = be64_to_cpu(tcb[((31 - word) / 2) - 1]);
3831 	u64 t;
3832 	u32 shift = 32;
3833 
3834 	t = (thi << shift) | (tlo >> shift);
3835 
3836 	return t;
3837 }
3838 
3839 static inline u32 t4_tcb_get_field32(__be64 *tcb, u16 word, u32 mask, u32 shift)
3840 {
3841 	u32 v;
3842 	u64 t = be64_to_cpu(tcb[(31 - word) / 2]);
3843 
3844 	if (word & 0x1)
3845 		shift += 32;
3846 	v = (t >> shift) & mask;
3847 	return v;
3848 }
3849 
3850 static int read_tcb_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
3851 {
3852 	struct cpl_get_tcb_rpl *rpl = cplhdr(skb);
3853 	__be64 *tcb = (__be64 *)(rpl + 1);
3854 	unsigned int tid = GET_TID(rpl);
3855 	struct c4iw_ep *ep;
3856 	u64 t_flags_64;
3857 	u32 rx_pdu_out;
3858 
3859 	ep = get_ep_from_tid(dev, tid);
3860 	if (!ep)
3861 		return 0;
3862 	/* Examine the TF_RX_PDU_OUT (bit 49 of the t_flags) in order to
3863 	 * determine if there's a rx PDU feedback event pending.
3864 	 *
3865 	 * If that bit is set, it means we'll need to re-read the TCB's
3866 	 * rq_start value. The final value is the one present in a TCB
3867 	 * with the TF_RX_PDU_OUT bit cleared.
3868 	 */
3869 
3870 	t_flags_64 = t4_tcb_get_field64(tcb, TCB_T_FLAGS_W);
3871 	rx_pdu_out = (t_flags_64 & TF_RX_PDU_OUT_V(1)) >> TF_RX_PDU_OUT_S;
3872 
3873 	c4iw_put_ep(&ep->com); /* from get_ep_from_tid() */
3874 	c4iw_put_ep(&ep->com); /* from read_tcb() */
3875 
3876 	/* If TF_RX_PDU_OUT bit is set, re-read the TCB */
3877 	if (rx_pdu_out) {
3878 		if (++ep->rx_pdu_out_cnt >= 2) {
3879 			WARN_ONCE(1, "tcb re-read() reached the guard limit, finishing the cleanup\n");
3880 			goto cleanup;
3881 		}
3882 		read_tcb(ep);
3883 		return 0;
3884 	}
3885 
3886 	ep->srqe_idx = t4_tcb_get_field32(tcb, TCB_RQ_START_W, TCB_RQ_START_M,
3887 					  TCB_RQ_START_S);
3888 cleanup:
3889 	pr_debug("ep %p tid %u %016x\n", ep, ep->hwtid, ep->srqe_idx);
3890 
3891 	if (test_bit(PEER_ABORT_IN_PROGRESS, &ep->com.flags))
3892 		finish_peer_abort(dev, ep);
3893 	else if (test_bit(ABORT_REQ_IN_PROGRESS, &ep->com.flags))
3894 		send_abort_req(ep);
3895 	else
3896 		WARN_ONCE(1, "unexpected state!");
3897 
3898 	return 0;
3899 }
3900 
3901 static int deferred_fw6_msg(struct c4iw_dev *dev, struct sk_buff *skb)
3902 {
3903 	struct cpl_fw6_msg *rpl = cplhdr(skb);
3904 	struct cpl_fw6_msg_ofld_connection_wr_rpl *req;
3905 
3906 	switch (rpl->type) {
3907 	case FW6_TYPE_CQE:
3908 		c4iw_ev_dispatch(dev, (struct t4_cqe *)&rpl->data[0]);
3909 		break;
3910 	case FW6_TYPE_OFLD_CONNECTION_WR_RPL:
3911 		req = (struct cpl_fw6_msg_ofld_connection_wr_rpl *)rpl->data;
3912 		switch (req->t_state) {
3913 		case TCP_SYN_SENT:
3914 			active_ofld_conn_reply(dev, skb, req);
3915 			break;
3916 		case TCP_SYN_RECV:
3917 			passive_ofld_conn_reply(dev, skb, req);
3918 			break;
3919 		default:
3920 			pr_err("%s unexpected ofld conn wr state %d\n",
3921 			       __func__, req->t_state);
3922 			break;
3923 		}
3924 		break;
3925 	}
3926 	return 0;
3927 }
3928 
3929 static void build_cpl_pass_accept_req(struct sk_buff *skb, int stid , u8 tos)
3930 {
3931 	__be32 l2info;
3932 	__be16 hdr_len, vlantag, len;
3933 	u16 eth_hdr_len;
3934 	int tcp_hdr_len, ip_hdr_len;
3935 	u8 intf;
3936 	struct cpl_rx_pkt *cpl = cplhdr(skb);
3937 	struct cpl_pass_accept_req *req;
3938 	struct tcp_options_received tmp_opt;
3939 	struct c4iw_dev *dev;
3940 	enum chip_type type;
3941 
3942 	dev = *((struct c4iw_dev **) (skb->cb + sizeof(void *)));
3943 	/* Store values from cpl_rx_pkt in temporary location. */
3944 	vlantag = cpl->vlan;
3945 	len = cpl->len;
3946 	l2info  = cpl->l2info;
3947 	hdr_len = cpl->hdr_len;
3948 	intf = cpl->iff;
3949 
3950 	__skb_pull(skb, sizeof(*req) + sizeof(struct rss_header));
3951 
3952 	/*
3953 	 * We need to parse the TCP options from SYN packet.
3954 	 * to generate cpl_pass_accept_req.
3955 	 */
3956 	memset(&tmp_opt, 0, sizeof(tmp_opt));
3957 	tcp_clear_options(&tmp_opt);
3958 	tcp_parse_options(&init_net, skb, &tmp_opt, 0, NULL);
3959 
3960 	req = __skb_push(skb, sizeof(*req));
3961 	memset(req, 0, sizeof(*req));
3962 	req->l2info = cpu_to_be16(SYN_INTF_V(intf) |
3963 			 SYN_MAC_IDX_V(RX_MACIDX_G(
3964 			 be32_to_cpu(l2info))) |
3965 			 SYN_XACT_MATCH_F);
3966 	type = dev->rdev.lldi.adapter_type;
3967 	tcp_hdr_len = RX_TCPHDR_LEN_G(be16_to_cpu(hdr_len));
3968 	ip_hdr_len = RX_IPHDR_LEN_G(be16_to_cpu(hdr_len));
3969 	req->hdr_len =
3970 		cpu_to_be32(SYN_RX_CHAN_V(RX_CHAN_G(be32_to_cpu(l2info))));
3971 	if (CHELSIO_CHIP_VERSION(type) <= CHELSIO_T5) {
3972 		eth_hdr_len = is_t4(type) ?
3973 				RX_ETHHDR_LEN_G(be32_to_cpu(l2info)) :
3974 				RX_T5_ETHHDR_LEN_G(be32_to_cpu(l2info));
3975 		req->hdr_len |= cpu_to_be32(TCP_HDR_LEN_V(tcp_hdr_len) |
3976 					    IP_HDR_LEN_V(ip_hdr_len) |
3977 					    ETH_HDR_LEN_V(eth_hdr_len));
3978 	} else { /* T6 and later */
3979 		eth_hdr_len = RX_T6_ETHHDR_LEN_G(be32_to_cpu(l2info));
3980 		req->hdr_len |= cpu_to_be32(T6_TCP_HDR_LEN_V(tcp_hdr_len) |
3981 					    T6_IP_HDR_LEN_V(ip_hdr_len) |
3982 					    T6_ETH_HDR_LEN_V(eth_hdr_len));
3983 	}
3984 	req->vlan = vlantag;
3985 	req->len = len;
3986 	req->tos_stid = cpu_to_be32(PASS_OPEN_TID_V(stid) |
3987 				    PASS_OPEN_TOS_V(tos));
3988 	req->tcpopt.mss = htons(tmp_opt.mss_clamp);
3989 	if (tmp_opt.wscale_ok)
3990 		req->tcpopt.wsf = tmp_opt.snd_wscale;
3991 	req->tcpopt.tstamp = tmp_opt.saw_tstamp;
3992 	if (tmp_opt.sack_ok)
3993 		req->tcpopt.sack = 1;
3994 	OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_REQ, 0));
3995 	return;
3996 }
3997 
3998 static void send_fw_pass_open_req(struct c4iw_dev *dev, struct sk_buff *skb,
3999 				  __be32 laddr, __be16 lport,
4000 				  __be32 raddr, __be16 rport,
4001 				  u32 rcv_isn, u32 filter, u16 window,
4002 				  u32 rss_qid, u8 port_id)
4003 {
4004 	struct sk_buff *req_skb;
4005 	struct fw_ofld_connection_wr *req;
4006 	struct cpl_pass_accept_req *cpl = cplhdr(skb);
4007 	int ret;
4008 
4009 	req_skb = alloc_skb(sizeof(struct fw_ofld_connection_wr), GFP_KERNEL);
4010 	if (!req_skb)
4011 		return;
4012 	req = __skb_put_zero(req_skb, sizeof(*req));
4013 	req->op_compl = htonl(WR_OP_V(FW_OFLD_CONNECTION_WR) | FW_WR_COMPL_F);
4014 	req->len16_pkd = htonl(FW_WR_LEN16_V(DIV_ROUND_UP(sizeof(*req), 16)));
4015 	req->le.version_cpl = htonl(FW_OFLD_CONNECTION_WR_CPL_F);
4016 	req->le.filter = (__force __be32) filter;
4017 	req->le.lport = lport;
4018 	req->le.pport = rport;
4019 	req->le.u.ipv4.lip = laddr;
4020 	req->le.u.ipv4.pip = raddr;
4021 	req->tcb.rcv_nxt = htonl(rcv_isn + 1);
4022 	req->tcb.rcv_adv = htons(window);
4023 	req->tcb.t_state_to_astid =
4024 		 htonl(FW_OFLD_CONNECTION_WR_T_STATE_V(TCP_SYN_RECV) |
4025 			FW_OFLD_CONNECTION_WR_RCV_SCALE_V(cpl->tcpopt.wsf) |
4026 			FW_OFLD_CONNECTION_WR_ASTID_V(
4027 			PASS_OPEN_TID_G(ntohl(cpl->tos_stid))));
4028 
4029 	/*
4030 	 * We store the qid in opt2 which will be used by the firmware
4031 	 * to send us the wr response.
4032 	 */
4033 	req->tcb.opt2 = htonl(RSS_QUEUE_V(rss_qid));
4034 
4035 	/*
4036 	 * We initialize the MSS index in TCB to 0xF.
4037 	 * So that when driver sends cpl_pass_accept_rpl
4038 	 * TCB picks up the correct value. If this was 0
4039 	 * TP will ignore any value > 0 for MSS index.
4040 	 */
4041 	req->tcb.opt0 = cpu_to_be64(MSS_IDX_V(0xF));
4042 	req->cookie = (uintptr_t)skb;
4043 
4044 	set_wr_txq(req_skb, CPL_PRIORITY_CONTROL, port_id);
4045 	ret = cxgb4_ofld_send(dev->rdev.lldi.ports[0], req_skb);
4046 	if (ret < 0) {
4047 		pr_err("%s - cxgb4_ofld_send error %d - dropping\n", __func__,
4048 		       ret);
4049 		kfree_skb(skb);
4050 		kfree_skb(req_skb);
4051 	}
4052 }
4053 
4054 /*
4055  * Handler for CPL_RX_PKT message. Need to handle cpl_rx_pkt
4056  * messages when a filter is being used instead of server to
4057  * redirect a syn packet. When packets hit filter they are redirected
4058  * to the offload queue and driver tries to establish the connection
4059  * using firmware work request.
4060  */
4061 static int rx_pkt(struct c4iw_dev *dev, struct sk_buff *skb)
4062 {
4063 	int stid;
4064 	unsigned int filter;
4065 	struct ethhdr *eh = NULL;
4066 	struct vlan_ethhdr *vlan_eh = NULL;
4067 	struct iphdr *iph;
4068 	struct tcphdr *tcph;
4069 	struct rss_header *rss = (void *)skb->data;
4070 	struct cpl_rx_pkt *cpl = (void *)skb->data;
4071 	struct cpl_pass_accept_req *req = (void *)(rss + 1);
4072 	struct l2t_entry *e;
4073 	struct dst_entry *dst;
4074 	struct c4iw_ep *lep = NULL;
4075 	u16 window;
4076 	struct port_info *pi;
4077 	struct net_device *pdev;
4078 	u16 rss_qid, eth_hdr_len;
4079 	int step;
4080 	struct neighbour *neigh;
4081 
4082 	/* Drop all non-SYN packets */
4083 	if (!(cpl->l2info & cpu_to_be32(RXF_SYN_F)))
4084 		goto reject;
4085 
4086 	/*
4087 	 * Drop all packets which did not hit the filter.
4088 	 * Unlikely to happen.
4089 	 */
4090 	if (!(rss->filter_hit && rss->filter_tid))
4091 		goto reject;
4092 
4093 	/*
4094 	 * Calculate the server tid from filter hit index from cpl_rx_pkt.
4095 	 */
4096 	stid = (__force int) cpu_to_be32((__force u32) rss->hash_val);
4097 
4098 	lep = (struct c4iw_ep *)get_ep_from_stid(dev, stid);
4099 	if (!lep) {
4100 		pr_warn("%s connect request on invalid stid %d\n",
4101 			__func__, stid);
4102 		goto reject;
4103 	}
4104 
4105 	switch (CHELSIO_CHIP_VERSION(dev->rdev.lldi.adapter_type)) {
4106 	case CHELSIO_T4:
4107 		eth_hdr_len = RX_ETHHDR_LEN_G(be32_to_cpu(cpl->l2info));
4108 		break;
4109 	case CHELSIO_T5:
4110 		eth_hdr_len = RX_T5_ETHHDR_LEN_G(be32_to_cpu(cpl->l2info));
4111 		break;
4112 	case CHELSIO_T6:
4113 		eth_hdr_len = RX_T6_ETHHDR_LEN_G(be32_to_cpu(cpl->l2info));
4114 		break;
4115 	default:
4116 		pr_err("T%d Chip is not supported\n",
4117 		       CHELSIO_CHIP_VERSION(dev->rdev.lldi.adapter_type));
4118 		goto reject;
4119 	}
4120 
4121 	if (eth_hdr_len == ETH_HLEN) {
4122 		eh = (struct ethhdr *)(req + 1);
4123 		iph = (struct iphdr *)(eh + 1);
4124 	} else {
4125 		vlan_eh = (struct vlan_ethhdr *)(req + 1);
4126 		iph = (struct iphdr *)(vlan_eh + 1);
4127 		__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), ntohs(cpl->vlan));
4128 	}
4129 
4130 	if (iph->version != 0x4)
4131 		goto reject;
4132 
4133 	tcph = (struct tcphdr *)(iph + 1);
4134 	skb_set_network_header(skb, (void *)iph - (void *)rss);
4135 	skb_set_transport_header(skb, (void *)tcph - (void *)rss);
4136 	skb_get(skb);
4137 
4138 	pr_debug("lip 0x%x lport %u pip 0x%x pport %u tos %d\n",
4139 		 ntohl(iph->daddr), ntohs(tcph->dest), ntohl(iph->saddr),
4140 		 ntohs(tcph->source), iph->tos);
4141 
4142 	dst = cxgb_find_route(&dev->rdev.lldi, get_real_dev,
4143 			      iph->daddr, iph->saddr, tcph->dest,
4144 			      tcph->source, iph->tos);
4145 	if (!dst) {
4146 		pr_err("%s - failed to find dst entry!\n", __func__);
4147 		goto reject;
4148 	}
4149 	neigh = dst_neigh_lookup_skb(dst, skb);
4150 
4151 	if (!neigh) {
4152 		pr_err("%s - failed to allocate neigh!\n", __func__);
4153 		goto free_dst;
4154 	}
4155 
4156 	if (neigh->dev->flags & IFF_LOOPBACK) {
4157 		pdev = ip_dev_find(&init_net, iph->daddr);
4158 		if (!pdev) {
4159 			pr_err("%s - failed to find device!\n", __func__);
4160 			goto free_dst;
4161 		}
4162 		e = cxgb4_l2t_get(dev->rdev.lldi.l2t, neigh,
4163 				    pdev, 0);
4164 		pi = (struct port_info *)netdev_priv(pdev);
4165 		dev_put(pdev);
4166 	} else {
4167 		pdev = get_real_dev(neigh->dev);
4168 		e = cxgb4_l2t_get(dev->rdev.lldi.l2t, neigh,
4169 					pdev, 0);
4170 		pi = (struct port_info *)netdev_priv(pdev);
4171 	}
4172 	neigh_release(neigh);
4173 	if (!e) {
4174 		pr_err("%s - failed to allocate l2t entry!\n",
4175 		       __func__);
4176 		goto free_dst;
4177 	}
4178 
4179 	step = dev->rdev.lldi.nrxq / dev->rdev.lldi.nchan;
4180 	rss_qid = dev->rdev.lldi.rxq_ids[pi->port_id * step];
4181 	window = (__force u16) htons((__force u16)tcph->window);
4182 
4183 	/* Calcuate filter portion for LE region. */
4184 	filter = (__force unsigned int) cpu_to_be32(cxgb4_select_ntuple(
4185 						    dev->rdev.lldi.ports[0],
4186 						    e));
4187 
4188 	/*
4189 	 * Synthesize the cpl_pass_accept_req. We have everything except the
4190 	 * TID. Once firmware sends a reply with TID we update the TID field
4191 	 * in cpl and pass it through the regular cpl_pass_accept_req path.
4192 	 */
4193 	build_cpl_pass_accept_req(skb, stid, iph->tos);
4194 	send_fw_pass_open_req(dev, skb, iph->daddr, tcph->dest, iph->saddr,
4195 			      tcph->source, ntohl(tcph->seq), filter, window,
4196 			      rss_qid, pi->port_id);
4197 	cxgb4_l2t_release(e);
4198 free_dst:
4199 	dst_release(dst);
4200 reject:
4201 	if (lep)
4202 		c4iw_put_ep(&lep->com);
4203 	return 0;
4204 }
4205 
4206 /*
4207  * These are the real handlers that are called from a
4208  * work queue.
4209  */
4210 static c4iw_handler_func work_handlers[NUM_CPL_CMDS + NUM_FAKE_CPLS] = {
4211 	[CPL_ACT_ESTABLISH] = act_establish,
4212 	[CPL_ACT_OPEN_RPL] = act_open_rpl,
4213 	[CPL_RX_DATA] = rx_data,
4214 	[CPL_ABORT_RPL_RSS] = abort_rpl,
4215 	[CPL_ABORT_RPL] = abort_rpl,
4216 	[CPL_PASS_OPEN_RPL] = pass_open_rpl,
4217 	[CPL_CLOSE_LISTSRV_RPL] = close_listsrv_rpl,
4218 	[CPL_PASS_ACCEPT_REQ] = pass_accept_req,
4219 	[CPL_PASS_ESTABLISH] = pass_establish,
4220 	[CPL_PEER_CLOSE] = peer_close,
4221 	[CPL_ABORT_REQ_RSS] = peer_abort,
4222 	[CPL_CLOSE_CON_RPL] = close_con_rpl,
4223 	[CPL_RDMA_TERMINATE] = terminate,
4224 	[CPL_FW4_ACK] = fw4_ack,
4225 	[CPL_GET_TCB_RPL] = read_tcb_rpl,
4226 	[CPL_FW6_MSG] = deferred_fw6_msg,
4227 	[CPL_RX_PKT] = rx_pkt,
4228 	[FAKE_CPL_PUT_EP_SAFE] = _put_ep_safe,
4229 	[FAKE_CPL_PASS_PUT_EP_SAFE] = _put_pass_ep_safe
4230 };
4231 
4232 static void process_timeout(struct c4iw_ep *ep)
4233 {
4234 	struct c4iw_qp_attributes attrs;
4235 	int abort = 1;
4236 
4237 	mutex_lock(&ep->com.mutex);
4238 	pr_debug("ep %p tid %u state %d\n", ep, ep->hwtid, ep->com.state);
4239 	set_bit(TIMEDOUT, &ep->com.history);
4240 	switch (ep->com.state) {
4241 	case MPA_REQ_SENT:
4242 		connect_reply_upcall(ep, -ETIMEDOUT);
4243 		break;
4244 	case MPA_REQ_WAIT:
4245 	case MPA_REQ_RCVD:
4246 	case MPA_REP_SENT:
4247 	case FPDU_MODE:
4248 		break;
4249 	case CLOSING:
4250 	case MORIBUND:
4251 		if (ep->com.cm_id && ep->com.qp) {
4252 			attrs.next_state = C4IW_QP_STATE_ERROR;
4253 			c4iw_modify_qp(ep->com.qp->rhp,
4254 				     ep->com.qp, C4IW_QP_ATTR_NEXT_STATE,
4255 				     &attrs, 1);
4256 		}
4257 		close_complete_upcall(ep, -ETIMEDOUT);
4258 		break;
4259 	case ABORTING:
4260 	case DEAD:
4261 
4262 		/*
4263 		 * These states are expected if the ep timed out at the same
4264 		 * time as another thread was calling stop_ep_timer().
4265 		 * So we silently do nothing for these states.
4266 		 */
4267 		abort = 0;
4268 		break;
4269 	default:
4270 		WARN(1, "%s unexpected state ep %p tid %u state %u\n",
4271 			__func__, ep, ep->hwtid, ep->com.state);
4272 		abort = 0;
4273 	}
4274 	mutex_unlock(&ep->com.mutex);
4275 	if (abort)
4276 		c4iw_ep_disconnect(ep, 1, GFP_KERNEL);
4277 	c4iw_put_ep(&ep->com);
4278 }
4279 
4280 static void process_timedout_eps(void)
4281 {
4282 	struct c4iw_ep *ep;
4283 
4284 	spin_lock_irq(&timeout_lock);
4285 	while (!list_empty(&timeout_list)) {
4286 		struct list_head *tmp;
4287 
4288 		tmp = timeout_list.next;
4289 		list_del(tmp);
4290 		tmp->next = NULL;
4291 		tmp->prev = NULL;
4292 		spin_unlock_irq(&timeout_lock);
4293 		ep = list_entry(tmp, struct c4iw_ep, entry);
4294 		process_timeout(ep);
4295 		spin_lock_irq(&timeout_lock);
4296 	}
4297 	spin_unlock_irq(&timeout_lock);
4298 }
4299 
4300 static void process_work(struct work_struct *work)
4301 {
4302 	struct sk_buff *skb = NULL;
4303 	struct c4iw_dev *dev;
4304 	struct cpl_act_establish *rpl;
4305 	unsigned int opcode;
4306 	int ret;
4307 
4308 	process_timedout_eps();
4309 	while ((skb = skb_dequeue(&rxq))) {
4310 		rpl = cplhdr(skb);
4311 		dev = *((struct c4iw_dev **) (skb->cb + sizeof(void *)));
4312 		opcode = rpl->ot.opcode;
4313 
4314 		if (opcode >= ARRAY_SIZE(work_handlers) ||
4315 		    !work_handlers[opcode]) {
4316 			pr_err("No handler for opcode 0x%x.\n", opcode);
4317 			kfree_skb(skb);
4318 		} else {
4319 			ret = work_handlers[opcode](dev, skb);
4320 			if (!ret)
4321 				kfree_skb(skb);
4322 		}
4323 		process_timedout_eps();
4324 	}
4325 }
4326 
4327 static DECLARE_WORK(skb_work, process_work);
4328 
4329 static void ep_timeout(struct timer_list *t)
4330 {
4331 	struct c4iw_ep *ep = from_timer(ep, t, timer);
4332 	int kickit = 0;
4333 
4334 	spin_lock(&timeout_lock);
4335 	if (!test_and_set_bit(TIMEOUT, &ep->com.flags)) {
4336 		/*
4337 		 * Only insert if it is not already on the list.
4338 		 */
4339 		if (!ep->entry.next) {
4340 			list_add_tail(&ep->entry, &timeout_list);
4341 			kickit = 1;
4342 		}
4343 	}
4344 	spin_unlock(&timeout_lock);
4345 	if (kickit)
4346 		queue_work(workq, &skb_work);
4347 }
4348 
4349 /*
4350  * All the CM events are handled on a work queue to have a safe context.
4351  */
4352 static int sched(struct c4iw_dev *dev, struct sk_buff *skb)
4353 {
4354 
4355 	/*
4356 	 * Save dev in the skb->cb area.
4357 	 */
4358 	*((struct c4iw_dev **) (skb->cb + sizeof(void *))) = dev;
4359 
4360 	/*
4361 	 * Queue the skb and schedule the worker thread.
4362 	 */
4363 	skb_queue_tail(&rxq, skb);
4364 	queue_work(workq, &skb_work);
4365 	return 0;
4366 }
4367 
4368 static int set_tcb_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
4369 {
4370 	struct cpl_set_tcb_rpl *rpl = cplhdr(skb);
4371 
4372 	if (rpl->status != CPL_ERR_NONE) {
4373 		pr_err("Unexpected SET_TCB_RPL status %u for tid %u\n",
4374 		       rpl->status, GET_TID(rpl));
4375 	}
4376 	kfree_skb(skb);
4377 	return 0;
4378 }
4379 
4380 static int fw6_msg(struct c4iw_dev *dev, struct sk_buff *skb)
4381 {
4382 	struct cpl_fw6_msg *rpl = cplhdr(skb);
4383 	struct c4iw_wr_wait *wr_waitp;
4384 	int ret;
4385 
4386 	pr_debug("type %u\n", rpl->type);
4387 
4388 	switch (rpl->type) {
4389 	case FW6_TYPE_WR_RPL:
4390 		ret = (int)((be64_to_cpu(rpl->data[0]) >> 8) & 0xff);
4391 		wr_waitp = (struct c4iw_wr_wait *)(__force unsigned long) rpl->data[1];
4392 		pr_debug("wr_waitp %p ret %u\n", wr_waitp, ret);
4393 		if (wr_waitp)
4394 			c4iw_wake_up_deref(wr_waitp, ret ? -ret : 0);
4395 		kfree_skb(skb);
4396 		break;
4397 	case FW6_TYPE_CQE:
4398 	case FW6_TYPE_OFLD_CONNECTION_WR_RPL:
4399 		sched(dev, skb);
4400 		break;
4401 	default:
4402 		pr_err("%s unexpected fw6 msg type %u\n",
4403 		       __func__, rpl->type);
4404 		kfree_skb(skb);
4405 		break;
4406 	}
4407 	return 0;
4408 }
4409 
4410 static int peer_abort_intr(struct c4iw_dev *dev, struct sk_buff *skb)
4411 {
4412 	struct cpl_abort_req_rss *req = cplhdr(skb);
4413 	struct c4iw_ep *ep;
4414 	unsigned int tid = GET_TID(req);
4415 
4416 	ep = get_ep_from_tid(dev, tid);
4417 	/* This EP will be dereferenced in peer_abort() */
4418 	if (!ep) {
4419 		pr_warn("Abort on non-existent endpoint, tid %d\n", tid);
4420 		kfree_skb(skb);
4421 		return 0;
4422 	}
4423 	if (cxgb_is_neg_adv(req->status)) {
4424 		pr_debug("Negative advice on abort- tid %u status %d (%s)\n",
4425 			 ep->hwtid, req->status,
4426 			 neg_adv_str(req->status));
4427 		goto out;
4428 	}
4429 	pr_debug("ep %p tid %u state %u\n", ep, ep->hwtid, ep->com.state);
4430 
4431 	c4iw_wake_up_noref(ep->com.wr_waitp, -ECONNRESET);
4432 out:
4433 	sched(dev, skb);
4434 	return 0;
4435 }
4436 
4437 /*
4438  * Most upcalls from the T4 Core go to sched() to
4439  * schedule the processing on a work queue.
4440  */
4441 c4iw_handler_func c4iw_handlers[NUM_CPL_CMDS] = {
4442 	[CPL_ACT_ESTABLISH] = sched,
4443 	[CPL_ACT_OPEN_RPL] = sched,
4444 	[CPL_RX_DATA] = sched,
4445 	[CPL_ABORT_RPL_RSS] = sched,
4446 	[CPL_ABORT_RPL] = sched,
4447 	[CPL_PASS_OPEN_RPL] = sched,
4448 	[CPL_CLOSE_LISTSRV_RPL] = sched,
4449 	[CPL_PASS_ACCEPT_REQ] = sched,
4450 	[CPL_PASS_ESTABLISH] = sched,
4451 	[CPL_PEER_CLOSE] = sched,
4452 	[CPL_CLOSE_CON_RPL] = sched,
4453 	[CPL_ABORT_REQ_RSS] = peer_abort_intr,
4454 	[CPL_RDMA_TERMINATE] = sched,
4455 	[CPL_FW4_ACK] = sched,
4456 	[CPL_SET_TCB_RPL] = set_tcb_rpl,
4457 	[CPL_GET_TCB_RPL] = sched,
4458 	[CPL_FW6_MSG] = fw6_msg,
4459 	[CPL_RX_PKT] = sched
4460 };
4461 
4462 int __init c4iw_cm_init(void)
4463 {
4464 	skb_queue_head_init(&rxq);
4465 
4466 	workq = alloc_ordered_workqueue("iw_cxgb4", WQ_MEM_RECLAIM);
4467 	if (!workq)
4468 		return -ENOMEM;
4469 
4470 	return 0;
4471 }
4472 
4473 void c4iw_cm_term(void)
4474 {
4475 	WARN_ON(!list_empty(&timeout_list));
4476 	destroy_workqueue(workq);
4477 }
4478