xref: /freebsd/sys/dev/cxgbe/iw_cxgbe/cm.c (revision 4c9e27bd0a5f7fda85b0c0bf750575aee300a172)
1 /*
2  * Copyright (c) 2009-2013 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 <sys/cdefs.h>
33 __FBSDID("$FreeBSD$");
34 
35 #include "opt_inet.h"
36 
37 #ifdef TCP_OFFLOAD
38 #include <sys/types.h>
39 #include <sys/malloc.h>
40 #include <sys/socket.h>
41 #include <sys/socketvar.h>
42 #include <sys/sockio.h>
43 #include <sys/taskqueue.h>
44 #include <netinet/in.h>
45 #include <net/neighbour.h>
46 #include <net/route.h>
47 
48 #include <netinet/in_systm.h>
49 #include <netinet/in_pcb.h>
50 #include <netinet/ip.h>
51 #include <netinet/ip_var.h>
52 #include <netinet/tcp_var.h>
53 #include <netinet/tcp.h>
54 #include <netinet/tcpip.h>
55 
56 #include <netinet/toecore.h>
57 
58 struct sge_iq;
59 struct rss_header;
60 #include <linux/types.h>
61 #include "offload.h"
62 #include "tom/t4_tom.h"
63 
64 #define TOEPCB(so)  ((struct toepcb *)(so_sototcpcb((so))->t_toe))
65 
66 #include "iw_cxgbe.h"
67 #include <linux/module.h>
68 #include <linux/workqueue.h>
69 #include <linux/notifier.h>
70 #include <linux/inetdevice.h>
71 #include <linux/if_vlan.h>
72 #include <net/netevent.h>
73 
74 static spinlock_t req_lock;
75 static TAILQ_HEAD(c4iw_ep_list, c4iw_ep_common) req_list;
76 static struct work_struct c4iw_task;
77 static struct workqueue_struct *c4iw_taskq;
78 static LIST_HEAD(timeout_list);
79 static spinlock_t timeout_lock;
80 
81 static void process_req(struct work_struct *ctx);
82 static void start_ep_timer(struct c4iw_ep *ep);
83 static void stop_ep_timer(struct c4iw_ep *ep);
84 static int set_tcpinfo(struct c4iw_ep *ep);
85 static enum c4iw_ep_state state_read(struct c4iw_ep_common *epc);
86 static void __state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state tostate);
87 static void state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state tostate);
88 static void *alloc_ep(int size, gfp_t flags);
89 void __free_ep(struct c4iw_ep_common *epc);
90 static struct rtentry * find_route(__be32 local_ip, __be32 peer_ip, __be16 local_port,
91 		__be16 peer_port, u8 tos);
92 static int close_socket(struct c4iw_ep_common *epc, int close);
93 static int shutdown_socket(struct c4iw_ep_common *epc);
94 static void abort_socket(struct c4iw_ep *ep);
95 static void send_mpa_req(struct c4iw_ep *ep);
96 static int send_mpa_reject(struct c4iw_ep *ep, const void *pdata, u8 plen);
97 static int send_mpa_reply(struct c4iw_ep *ep, const void *pdata, u8 plen);
98 static void close_complete_upcall(struct c4iw_ep *ep);
99 static int abort_connection(struct c4iw_ep *ep);
100 static void peer_close_upcall(struct c4iw_ep *ep);
101 static void peer_abort_upcall(struct c4iw_ep *ep);
102 static void connect_reply_upcall(struct c4iw_ep *ep, int status);
103 static void connect_request_upcall(struct c4iw_ep *ep);
104 static void established_upcall(struct c4iw_ep *ep);
105 static void process_mpa_reply(struct c4iw_ep *ep);
106 static void process_mpa_request(struct c4iw_ep *ep);
107 static void process_peer_close(struct c4iw_ep *ep);
108 static void process_conn_error(struct c4iw_ep *ep);
109 static void process_close_complete(struct c4iw_ep *ep);
110 static void ep_timeout(unsigned long arg);
111 static void init_sock(struct c4iw_ep_common *epc);
112 static void process_data(struct c4iw_ep *ep);
113 static void process_connected(struct c4iw_ep *ep);
114 static struct socket * dequeue_socket(struct socket *head, struct sockaddr_in **remote, struct c4iw_ep *child_ep);
115 static void process_newconn(struct c4iw_ep *parent_ep);
116 static int c4iw_so_upcall(struct socket *so, void *arg, int waitflag);
117 static void process_socket_event(struct c4iw_ep *ep);
118 static void release_ep_resources(struct c4iw_ep *ep);
119 
120 #define START_EP_TIMER(ep) \
121     do { \
122 	    CTR3(KTR_IW_CXGBE, "start_ep_timer (%s:%d) ep %p", \
123 		__func__, __LINE__, (ep)); \
124 	    start_ep_timer(ep); \
125     } while (0)
126 
127 #define STOP_EP_TIMER(ep) \
128     do { \
129 	    CTR3(KTR_IW_CXGBE, "stop_ep_timer (%s:%d) ep %p", \
130 		__func__, __LINE__, (ep)); \
131 	    stop_ep_timer(ep); \
132     } while (0)
133 
134 #ifdef KTR
135 static char *states[] = {
136 	"idle",
137 	"listen",
138 	"connecting",
139 	"mpa_wait_req",
140 	"mpa_req_sent",
141 	"mpa_req_rcvd",
142 	"mpa_rep_sent",
143 	"fpdu_mode",
144 	"aborting",
145 	"closing",
146 	"moribund",
147 	"dead",
148 	NULL,
149 };
150 #endif
151 
152 static void
153 process_req(struct work_struct *ctx)
154 {
155 	struct c4iw_ep_common *epc;
156 
157 	spin_lock(&req_lock);
158 	while (!TAILQ_EMPTY(&req_list)) {
159 		epc = TAILQ_FIRST(&req_list);
160 		TAILQ_REMOVE(&req_list, epc, entry);
161 		epc->entry.tqe_prev = NULL;
162 		spin_unlock(&req_lock);
163 		if (epc->so)
164 			process_socket_event((struct c4iw_ep *)epc);
165 		c4iw_put_ep(epc);
166 		spin_lock(&req_lock);
167 	}
168 	spin_unlock(&req_lock);
169 }
170 
171 /*
172  * XXX: doesn't belong here in the iWARP driver.
173  * XXX: assumes that the connection was offloaded by cxgbe/t4_tom if TF_TOE is
174  *      set.  Is this a valid assumption for active open?
175  */
176 static int
177 set_tcpinfo(struct c4iw_ep *ep)
178 {
179 	struct socket *so = ep->com.so;
180 	struct inpcb *inp = sotoinpcb(so);
181 	struct tcpcb *tp;
182 	struct toepcb *toep;
183 	int rc = 0;
184 
185 	INP_WLOCK(inp);
186 	tp = intotcpcb(inp);
187 	if ((tp->t_flags & TF_TOE) == 0) {
188 		rc = EINVAL;
189 		log(LOG_ERR, "%s: connection not offloaded (so %p, ep %p)\n",
190 		    __func__, so, ep);
191 		goto done;
192 	}
193 	toep = TOEPCB(so);
194 
195 	ep->hwtid = toep->tid;
196 	ep->snd_seq = tp->snd_nxt;
197 	ep->rcv_seq = tp->rcv_nxt;
198 	ep->emss = max(tp->t_maxseg, 128);
199 done:
200 	INP_WUNLOCK(inp);
201 	return (rc);
202 
203 }
204 
205 static struct rtentry *
206 find_route(__be32 local_ip, __be32 peer_ip, __be16 local_port,
207 		__be16 peer_port, u8 tos)
208 {
209 	struct route iproute;
210 	struct sockaddr_in *dst = (struct sockaddr_in *)&iproute.ro_dst;
211 
212 	CTR5(KTR_IW_CXGBE, "%s:frtB %x, %x, %d, %d", __func__, local_ip,
213 	    peer_ip, ntohs(local_port), ntohs(peer_port));
214 	bzero(&iproute, sizeof iproute);
215 	dst->sin_family = AF_INET;
216 	dst->sin_len = sizeof *dst;
217 	dst->sin_addr.s_addr = peer_ip;
218 
219 	rtalloc(&iproute);
220 	CTR2(KTR_IW_CXGBE, "%s:frtE %p", __func__, (uint64_t)iproute.ro_rt);
221 	return iproute.ro_rt;
222 }
223 
224 static int
225 close_socket(struct c4iw_ep_common *epc, int close)
226 {
227 	struct socket *so = epc->so;
228 	int rc;
229 
230 	CTR4(KTR_IW_CXGBE, "%s: so %p, ep %p, state %s", __func__, epc, so,
231 	    states[epc->state]);
232 
233 	SOCK_LOCK(so);
234 	soupcall_clear(so, SO_RCV);
235 	SOCK_UNLOCK(so);
236 
237 	if (close)
238                 rc = soclose(so);
239         else
240                 rc = soshutdown(so, SHUT_WR | SHUT_RD);
241 	epc->so = NULL;
242 
243 	return (rc);
244 }
245 
246 static int
247 shutdown_socket(struct c4iw_ep_common *epc)
248 {
249 
250 	CTR4(KTR_IW_CXGBE, "%s: so %p, ep %p, state %s", __func__, epc->so, epc,
251 	    states[epc->state]);
252 
253 	return (soshutdown(epc->so, SHUT_WR));
254 }
255 
256 static void
257 abort_socket(struct c4iw_ep *ep)
258 {
259 	struct sockopt sopt;
260 	int rc;
261 	struct linger l;
262 
263 	CTR4(KTR_IW_CXGBE, "%s ep %p so %p state %s", __func__, ep, ep->com.so,
264 	    states[ep->com.state]);
265 
266 	l.l_onoff = 1;
267 	l.l_linger = 0;
268 
269 	/* linger_time of 0 forces RST to be sent */
270 	sopt.sopt_dir = SOPT_SET;
271 	sopt.sopt_level = SOL_SOCKET;
272 	sopt.sopt_name = SO_LINGER;
273 	sopt.sopt_val = (caddr_t)&l;
274 	sopt.sopt_valsize = sizeof l;
275 	sopt.sopt_td = NULL;
276 	rc = sosetopt(ep->com.so, &sopt);
277 	if (rc) {
278 		log(LOG_ERR, "%s: can't set linger to 0, no RST! err %d\n",
279 		    __func__, rc);
280 	}
281 }
282 
283 static void
284 process_peer_close(struct c4iw_ep *ep)
285 {
286 	struct c4iw_qp_attributes attrs;
287 	int disconnect = 1;
288 	int release = 0;
289 
290 	CTR4(KTR_IW_CXGBE, "%s:ppcB ep %p so %p state %s", __func__, ep,
291 	    ep->com.so, states[ep->com.state]);
292 
293 	mutex_lock(&ep->com.mutex);
294 	switch (ep->com.state) {
295 
296 		case MPA_REQ_WAIT:
297 			CTR2(KTR_IW_CXGBE, "%s:ppc1 %p MPA_REQ_WAIT CLOSING",
298 			    __func__, ep);
299 			__state_set(&ep->com, CLOSING);
300 			break;
301 
302 		case MPA_REQ_SENT:
303 			CTR2(KTR_IW_CXGBE, "%s:ppc2 %p MPA_REQ_SENT CLOSING",
304 			    __func__, ep);
305 			__state_set(&ep->com, DEAD);
306 			connect_reply_upcall(ep, -ECONNABORTED);
307 
308 			disconnect = 0;
309 			STOP_EP_TIMER(ep);
310 			close_socket(&ep->com, 0);
311 			ep->com.cm_id->rem_ref(ep->com.cm_id);
312 			ep->com.cm_id = NULL;
313 			ep->com.qp = NULL;
314 			release = 1;
315 			break;
316 
317 		case MPA_REQ_RCVD:
318 
319 			/*
320 			 * We're gonna mark this puppy DEAD, but keep
321 			 * the reference on it until the ULP accepts or
322 			 * rejects the CR.
323 			 */
324 			CTR2(KTR_IW_CXGBE, "%s:ppc3 %p MPA_REQ_RCVD CLOSING",
325 			    __func__, ep);
326 			__state_set(&ep->com, CLOSING);
327 			c4iw_get_ep(&ep->com);
328 			break;
329 
330 		case MPA_REP_SENT:
331 			CTR2(KTR_IW_CXGBE, "%s:ppc4 %p MPA_REP_SENT CLOSING",
332 			    __func__, ep);
333 			__state_set(&ep->com, CLOSING);
334 			break;
335 
336 		case FPDU_MODE:
337 			CTR2(KTR_IW_CXGBE, "%s:ppc5 %p FPDU_MODE CLOSING",
338 			    __func__, ep);
339 			START_EP_TIMER(ep);
340 			__state_set(&ep->com, CLOSING);
341 			attrs.next_state = C4IW_QP_STATE_CLOSING;
342 			c4iw_modify_qp(ep->com.dev, ep->com.qp,
343 					C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
344 			peer_close_upcall(ep);
345 			break;
346 
347 		case ABORTING:
348 			CTR2(KTR_IW_CXGBE, "%s:ppc6 %p ABORTING (disconn)",
349 			    __func__, ep);
350 			disconnect = 0;
351 			break;
352 
353 		case CLOSING:
354 			CTR2(KTR_IW_CXGBE, "%s:ppc7 %p CLOSING MORIBUND",
355 			    __func__, ep);
356 			__state_set(&ep->com, MORIBUND);
357 			disconnect = 0;
358 			break;
359 
360 		case MORIBUND:
361 			CTR2(KTR_IW_CXGBE, "%s:ppc8 %p MORIBUND DEAD", __func__,
362 			    ep);
363 			STOP_EP_TIMER(ep);
364 			if (ep->com.cm_id && ep->com.qp) {
365 				attrs.next_state = C4IW_QP_STATE_IDLE;
366 				c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
367 						C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
368 			}
369 			close_socket(&ep->com, 0);
370 			close_complete_upcall(ep);
371 			__state_set(&ep->com, DEAD);
372 			release = 1;
373 			disconnect = 0;
374 			break;
375 
376 		case DEAD:
377 			CTR2(KTR_IW_CXGBE, "%s:ppc9 %p DEAD (disconn)",
378 			    __func__, ep);
379 			disconnect = 0;
380 			break;
381 
382 		default:
383 			panic("%s: ep %p state %d", __func__, ep,
384 			    ep->com.state);
385 			break;
386 	}
387 
388 	mutex_unlock(&ep->com.mutex);
389 
390 	if (disconnect) {
391 
392 		CTR2(KTR_IW_CXGBE, "%s:ppca %p", __func__, ep);
393 		c4iw_ep_disconnect(ep, 0, M_NOWAIT);
394 	}
395 	if (release) {
396 
397 		CTR2(KTR_IW_CXGBE, "%s:ppcb %p", __func__, ep);
398 		c4iw_put_ep(&ep->com);
399 	}
400 	CTR2(KTR_IW_CXGBE, "%s:ppcE %p", __func__, ep);
401 	return;
402 }
403 
404 static void
405 process_conn_error(struct c4iw_ep *ep)
406 {
407 	struct c4iw_qp_attributes attrs;
408 	int ret;
409 	int state;
410 
411 	state = state_read(&ep->com);
412 	CTR5(KTR_IW_CXGBE, "%s:pceB ep %p so %p so->so_error %u state %s",
413 	    __func__, ep, ep->com.so, ep->com.so->so_error,
414 	    states[ep->com.state]);
415 
416 	switch (state) {
417 
418 		case MPA_REQ_WAIT:
419 			STOP_EP_TIMER(ep);
420 			break;
421 
422 		case MPA_REQ_SENT:
423 			STOP_EP_TIMER(ep);
424 			connect_reply_upcall(ep, -ECONNRESET);
425 			break;
426 
427 		case MPA_REP_SENT:
428 			ep->com.rpl_err = ECONNRESET;
429 			CTR1(KTR_IW_CXGBE, "waking up ep %p", ep);
430 			break;
431 
432 		case MPA_REQ_RCVD:
433 
434 			/*
435 			 * We're gonna mark this puppy DEAD, but keep
436 			 * the reference on it until the ULP accepts or
437 			 * rejects the CR.
438 			 */
439 			c4iw_get_ep(&ep->com);
440 			break;
441 
442 		case MORIBUND:
443 		case CLOSING:
444 			STOP_EP_TIMER(ep);
445 			/*FALLTHROUGH*/
446 		case FPDU_MODE:
447 
448 			if (ep->com.cm_id && ep->com.qp) {
449 
450 				attrs.next_state = C4IW_QP_STATE_ERROR;
451 				ret = c4iw_modify_qp(ep->com.qp->rhp,
452 					ep->com.qp, C4IW_QP_ATTR_NEXT_STATE,
453 					&attrs, 1);
454 				if (ret)
455 					log(LOG_ERR,
456 							"%s - qp <- error failed!\n",
457 							__func__);
458 			}
459 			peer_abort_upcall(ep);
460 			break;
461 
462 		case ABORTING:
463 			break;
464 
465 		case DEAD:
466 			CTR2(KTR_IW_CXGBE, "%s so_error %d IN DEAD STATE!!!!",
467 			    __func__, ep->com.so->so_error);
468 			return;
469 
470 		default:
471 			panic("%s: ep %p state %d", __func__, ep, state);
472 			break;
473 	}
474 
475 	if (state != ABORTING) {
476 
477 		CTR2(KTR_IW_CXGBE, "%s:pce1 %p", __func__, ep);
478 		close_socket(&ep->com, 0);
479 		state_set(&ep->com, DEAD);
480 		c4iw_put_ep(&ep->com);
481 	}
482 	CTR2(KTR_IW_CXGBE, "%s:pceE %p", __func__, ep);
483 	return;
484 }
485 
486 static void
487 process_close_complete(struct c4iw_ep *ep)
488 {
489 	struct c4iw_qp_attributes attrs;
490 	int release = 0;
491 
492 	CTR4(KTR_IW_CXGBE, "%s:pccB ep %p so %p state %s", __func__, ep,
493 	    ep->com.so, states[ep->com.state]);
494 
495 	/* The cm_id may be null if we failed to connect */
496 	mutex_lock(&ep->com.mutex);
497 
498 	switch (ep->com.state) {
499 
500 		case CLOSING:
501 			CTR2(KTR_IW_CXGBE, "%s:pcc1 %p CLOSING MORIBUND",
502 			    __func__, ep);
503 			__state_set(&ep->com, MORIBUND);
504 			break;
505 
506 		case MORIBUND:
507 			CTR2(KTR_IW_CXGBE, "%s:pcc1 %p MORIBUND DEAD", __func__,
508 			    ep);
509 			STOP_EP_TIMER(ep);
510 
511 			if ((ep->com.cm_id) && (ep->com.qp)) {
512 
513 				CTR2(KTR_IW_CXGBE, "%s:pcc2 %p QP_STATE_IDLE",
514 				    __func__, ep);
515 				attrs.next_state = C4IW_QP_STATE_IDLE;
516 				c4iw_modify_qp(ep->com.dev,
517 						ep->com.qp,
518 						C4IW_QP_ATTR_NEXT_STATE,
519 						&attrs, 1);
520 			}
521 
522 			if (ep->parent_ep) {
523 
524 				CTR2(KTR_IW_CXGBE, "%s:pcc3 %p", __func__, ep);
525 				close_socket(&ep->com, 1);
526 			}
527 			else {
528 
529 				CTR2(KTR_IW_CXGBE, "%s:pcc4 %p", __func__, ep);
530 				close_socket(&ep->com, 0);
531 			}
532 			close_complete_upcall(ep);
533 			__state_set(&ep->com, DEAD);
534 			release = 1;
535 			break;
536 
537 		case ABORTING:
538 			CTR2(KTR_IW_CXGBE, "%s:pcc5 %p ABORTING", __func__, ep);
539 			break;
540 
541 		case DEAD:
542 		default:
543 			CTR2(KTR_IW_CXGBE, "%s:pcc6 %p DEAD", __func__, ep);
544 			panic("%s:pcc6 %p DEAD", __func__, ep);
545 			break;
546 	}
547 	mutex_unlock(&ep->com.mutex);
548 
549 	if (release) {
550 
551 		CTR2(KTR_IW_CXGBE, "%s:pcc7 %p", __func__, ep);
552 		c4iw_put_ep(&ep->com);
553 	}
554 	CTR2(KTR_IW_CXGBE, "%s:pccE %p", __func__, ep);
555 	return;
556 }
557 
558 static void
559 init_sock(struct c4iw_ep_common *epc)
560 {
561 	int rc;
562 	struct sockopt sopt;
563 	struct socket *so = epc->so;
564 	int on = 1;
565 
566 	SOCK_LOCK(so);
567 	soupcall_set(so, SO_RCV, c4iw_so_upcall, epc);
568 	so->so_state |= SS_NBIO;
569 	SOCK_UNLOCK(so);
570 	sopt.sopt_dir = SOPT_SET;
571 	sopt.sopt_level = IPPROTO_TCP;
572 	sopt.sopt_name = TCP_NODELAY;
573 	sopt.sopt_val = (caddr_t)&on;
574 	sopt.sopt_valsize = sizeof on;
575 	sopt.sopt_td = NULL;
576 	rc = sosetopt(so, &sopt);
577 	if (rc) {
578 		log(LOG_ERR, "%s: can't set TCP_NODELAY on so %p (%d)\n",
579 		    __func__, so, rc);
580 	}
581 }
582 
583 static void
584 process_data(struct c4iw_ep *ep)
585 {
586 	struct sockaddr_in *local, *remote;
587 
588 	CTR5(KTR_IW_CXGBE, "%s: so %p, ep %p, state %s, sb_cc %d", __func__,
589 	    ep->com.so, ep, states[ep->com.state], ep->com.so->so_rcv.sb_cc);
590 
591 	switch (state_read(&ep->com)) {
592 	case MPA_REQ_SENT:
593 		process_mpa_reply(ep);
594 		break;
595 	case MPA_REQ_WAIT:
596 		in_getsockaddr(ep->com.so, (struct sockaddr **)&local);
597 		in_getpeeraddr(ep->com.so, (struct sockaddr **)&remote);
598 		ep->com.local_addr = *local;
599 		ep->com.remote_addr = *remote;
600 		free(local, M_SONAME);
601 		free(remote, M_SONAME);
602 		process_mpa_request(ep);
603 		break;
604 	default:
605 		if (ep->com.so->so_rcv.sb_cc)
606 			log(LOG_ERR, "%s: Unexpected streaming data.  "
607 			    "ep %p, state %d, so %p, so_state 0x%x, sb_cc %u\n",
608 			    __func__, ep, state_read(&ep->com), ep->com.so,
609 			    ep->com.so->so_state, ep->com.so->so_rcv.sb_cc);
610 		break;
611 	}
612 }
613 
614 static void
615 process_connected(struct c4iw_ep *ep)
616 {
617 
618 	if ((ep->com.so->so_state & SS_ISCONNECTED) && !ep->com.so->so_error)
619 		send_mpa_req(ep);
620 	else {
621 		connect_reply_upcall(ep, -ep->com.so->so_error);
622 		close_socket(&ep->com, 0);
623 		state_set(&ep->com, DEAD);
624 		c4iw_put_ep(&ep->com);
625 	}
626 }
627 
628 static struct socket *
629 dequeue_socket(struct socket *head, struct sockaddr_in **remote,
630     struct c4iw_ep *child_ep)
631 {
632 	struct socket *so;
633 
634 	ACCEPT_LOCK();
635 	so = TAILQ_FIRST(&head->so_comp);
636 	if (!so) {
637 		ACCEPT_UNLOCK();
638 		return (NULL);
639 	}
640 	TAILQ_REMOVE(&head->so_comp, so, so_list);
641 	head->so_qlen--;
642 	SOCK_LOCK(so);
643 	so->so_qstate &= ~SQ_COMP;
644 	so->so_head = NULL;
645 	soref(so);
646 	soupcall_set(so, SO_RCV, c4iw_so_upcall, child_ep);
647 	so->so_state |= SS_NBIO;
648 	SOCK_UNLOCK(so);
649 	ACCEPT_UNLOCK();
650 	soaccept(so, (struct sockaddr **)remote);
651 
652 	return (so);
653 }
654 
655 static void
656 process_newconn(struct c4iw_ep *parent_ep)
657 {
658 	struct socket *child_so;
659 	struct c4iw_ep *child_ep;
660 	struct sockaddr_in *remote;
661 
662 	child_ep = alloc_ep(sizeof(*child_ep), M_NOWAIT);
663 	if (!child_ep) {
664 		CTR3(KTR_IW_CXGBE, "%s: parent so %p, parent ep %p, ENOMEM",
665 		    __func__, parent_ep->com.so, parent_ep);
666 		log(LOG_ERR, "%s: failed to allocate ep entry\n", __func__);
667 		return;
668 	}
669 
670 	child_so = dequeue_socket(parent_ep->com.so, &remote, child_ep);
671 	if (!child_so) {
672 		CTR4(KTR_IW_CXGBE,
673 		    "%s: parent so %p, parent ep %p, child ep %p, dequeue err",
674 		    __func__, parent_ep->com.so, parent_ep, child_ep);
675 		log(LOG_ERR, "%s: failed to dequeue child socket\n", __func__);
676 		__free_ep(&child_ep->com);
677 		return;
678 
679 	}
680 
681 	CTR5(KTR_IW_CXGBE,
682 	    "%s: parent so %p, parent ep %p, child so %p, child ep %p",
683 	     __func__, parent_ep->com.so, parent_ep, child_so, child_ep);
684 
685 	child_ep->com.local_addr = parent_ep->com.local_addr;
686 	child_ep->com.remote_addr = *remote;
687 	child_ep->com.dev = parent_ep->com.dev;
688 	child_ep->com.so = child_so;
689 	child_ep->com.cm_id = NULL;
690 	child_ep->com.thread = parent_ep->com.thread;
691 	child_ep->parent_ep = parent_ep;
692 
693 	free(remote, M_SONAME);
694 	c4iw_get_ep(&parent_ep->com);
695 	child_ep->parent_ep = parent_ep;
696 	init_timer(&child_ep->timer);
697 	state_set(&child_ep->com, MPA_REQ_WAIT);
698 	START_EP_TIMER(child_ep);
699 
700 	/* maybe the request has already been queued up on the socket... */
701 	process_mpa_request(child_ep);
702 }
703 
704 static int
705 c4iw_so_upcall(struct socket *so, void *arg, int waitflag)
706 {
707 	struct c4iw_ep *ep = arg;
708 
709 	spin_lock(&req_lock);
710 
711 	CTR6(KTR_IW_CXGBE,
712 	    "%s: so %p, so_state 0x%x, ep %p, ep_state %s, tqe_prev %p",
713 	    __func__, so, so->so_state, ep, states[ep->com.state],
714 	    ep->com.entry.tqe_prev);
715 
716 	if (ep && ep->com.so && !ep->com.entry.tqe_prev) {
717 		KASSERT(ep->com.so == so, ("%s: XXX review.", __func__));
718 		c4iw_get_ep(&ep->com);
719 		TAILQ_INSERT_TAIL(&req_list, &ep->com, entry);
720 		queue_work(c4iw_taskq, &c4iw_task);
721 	}
722 
723 	spin_unlock(&req_lock);
724 	return (SU_OK);
725 }
726 
727 static void
728 process_socket_event(struct c4iw_ep *ep)
729 {
730 	int state = state_read(&ep->com);
731 	struct socket *so = ep->com.so;
732 
733 	CTR6(KTR_IW_CXGBE, "process_socket_event: so %p, so_state 0x%x, "
734 	    "so_err %d, sb_state 0x%x, ep %p, ep_state %s", so, so->so_state,
735 	    so->so_error, so->so_rcv.sb_state, ep, states[state]);
736 
737 	if (state == CONNECTING) {
738 		process_connected(ep);
739 		return;
740 	}
741 
742 	if (state == LISTEN) {
743 		process_newconn(ep);
744 		return;
745 	}
746 
747 	/* connection error */
748 	if (so->so_error) {
749 		process_conn_error(ep);
750 		return;
751 	}
752 
753 	/* peer close */
754 	if ((so->so_rcv.sb_state & SBS_CANTRCVMORE) && state < CLOSING) {
755 		process_peer_close(ep);
756 		return;
757 	}
758 
759 	/* close complete */
760 	if (so->so_state & SS_ISDISCONNECTED) {
761 		process_close_complete(ep);
762 		return;
763 	}
764 
765 	/* rx data */
766 	process_data(ep);
767 }
768 
769 SYSCTL_NODE(_hw, OID_AUTO, iw_cxgbe, CTLFLAG_RD, 0, "iw_cxgbe driver parameters");
770 
771 int db_delay_usecs = 1;
772 TUNABLE_INT("hw.iw_cxgbe.db_delay_usecs", &db_delay_usecs);
773 SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, db_delay_usecs, CTLFLAG_RW, &db_delay_usecs, 0,
774 		"Usecs to delay awaiting db fifo to drain");
775 
776 static int dack_mode = 1;
777 TUNABLE_INT("hw.iw_cxgbe.dack_mode", &dack_mode);
778 SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, dack_mode, CTLFLAG_RW, &dack_mode, 0,
779 		"Delayed ack mode (default = 1)");
780 
781 int c4iw_max_read_depth = 8;
782 TUNABLE_INT("hw.iw_cxgbe.c4iw_max_read_depth", &c4iw_max_read_depth);
783 SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, c4iw_max_read_depth, CTLFLAG_RW, &c4iw_max_read_depth, 0,
784 		"Per-connection max ORD/IRD (default = 8)");
785 
786 static int enable_tcp_timestamps;
787 TUNABLE_INT("hw.iw_cxgbe.enable_tcp_timestamps", &enable_tcp_timestamps);
788 SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, enable_tcp_timestamps, CTLFLAG_RW, &enable_tcp_timestamps, 0,
789 		"Enable tcp timestamps (default = 0)");
790 
791 static int enable_tcp_sack;
792 TUNABLE_INT("hw.iw_cxgbe.enable_tcp_sack", &enable_tcp_sack);
793 SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, enable_tcp_sack, CTLFLAG_RW, &enable_tcp_sack, 0,
794 		"Enable tcp SACK (default = 0)");
795 
796 static int enable_tcp_window_scaling = 1;
797 TUNABLE_INT("hw.iw_cxgbe.enable_tcp_window_scaling", &enable_tcp_window_scaling);
798 SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, enable_tcp_window_scaling, CTLFLAG_RW, &enable_tcp_window_scaling, 0,
799 		"Enable tcp window scaling (default = 1)");
800 
801 int c4iw_debug = 1;
802 TUNABLE_INT("hw.iw_cxgbe.c4iw_debug", &c4iw_debug);
803 SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, c4iw_debug, CTLFLAG_RW, &c4iw_debug, 0,
804 		"Enable debug logging (default = 0)");
805 
806 static int peer2peer;
807 TUNABLE_INT("hw.iw_cxgbe.peer2peer", &peer2peer);
808 SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, peer2peer, CTLFLAG_RW, &peer2peer, 0,
809 		"Support peer2peer ULPs (default = 0)");
810 
811 static int p2p_type = FW_RI_INIT_P2PTYPE_READ_REQ;
812 TUNABLE_INT("hw.iw_cxgbe.p2p_type", &p2p_type);
813 SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, p2p_type, CTLFLAG_RW, &p2p_type, 0,
814 		"RDMAP opcode to use for the RTR message: 1 = RDMA_READ 0 = RDMA_WRITE (default 1)");
815 
816 static int ep_timeout_secs = 60;
817 TUNABLE_INT("hw.iw_cxgbe.ep_timeout_secs", &ep_timeout_secs);
818 SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, ep_timeout_secs, CTLFLAG_RW, &ep_timeout_secs, 0,
819 		"CM Endpoint operation timeout in seconds (default = 60)");
820 
821 static int mpa_rev = 1;
822 TUNABLE_INT("hw.iw_cxgbe.mpa_rev", &mpa_rev);
823 #ifdef IW_CM_MPAV2
824 SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, mpa_rev, CTLFLAG_RW, &mpa_rev, 0,
825 		"MPA Revision, 0 supports amso1100, 1 is RFC0544 spec compliant, 2 is IETF MPA Peer Connect Draft compliant (default = 1)");
826 #else
827 SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, mpa_rev, CTLFLAG_RW, &mpa_rev, 0,
828 		"MPA Revision, 0 supports amso1100, 1 is RFC0544 spec compliant (default = 1)");
829 #endif
830 
831 static int markers_enabled;
832 TUNABLE_INT("hw.iw_cxgbe.markers_enabled", &markers_enabled);
833 SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, markers_enabled, CTLFLAG_RW, &markers_enabled, 0,
834 		"Enable MPA MARKERS (default(0) = disabled)");
835 
836 static int crc_enabled = 1;
837 TUNABLE_INT("hw.iw_cxgbe.crc_enabled", &crc_enabled);
838 SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, crc_enabled, CTLFLAG_RW, &crc_enabled, 0,
839 		"Enable MPA CRC (default(1) = enabled)");
840 
841 static int rcv_win = 256 * 1024;
842 TUNABLE_INT("hw.iw_cxgbe.rcv_win", &rcv_win);
843 SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, rcv_win, CTLFLAG_RW, &rcv_win, 0,
844 		"TCP receive window in bytes (default = 256KB)");
845 
846 static int snd_win = 128 * 1024;
847 TUNABLE_INT("hw.iw_cxgbe.snd_win", &snd_win);
848 SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, snd_win, CTLFLAG_RW, &snd_win, 0,
849 		"TCP send window in bytes (default = 128KB)");
850 
851 int db_fc_threshold = 2000;
852 TUNABLE_INT("hw.iw_cxgbe.db_fc_threshold", &db_fc_threshold);
853 SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, db_fc_threshold, CTLFLAG_RW, &db_fc_threshold, 0,
854 		"QP count/threshold that triggers automatic");
855 
856 static void
857 start_ep_timer(struct c4iw_ep *ep)
858 {
859 
860 	if (timer_pending(&ep->timer)) {
861 		CTR2(KTR_IW_CXGBE, "%s: ep %p, already started", __func__, ep);
862 		printk(KERN_ERR "%s timer already started! ep %p\n", __func__,
863 		    ep);
864 		return;
865 	}
866 	clear_bit(TIMEOUT, &ep->com.flags);
867 	c4iw_get_ep(&ep->com);
868 	ep->timer.expires = jiffies + ep_timeout_secs * HZ;
869 	ep->timer.data = (unsigned long)ep;
870 	ep->timer.function = ep_timeout;
871 	add_timer(&ep->timer);
872 }
873 
874 static void
875 stop_ep_timer(struct c4iw_ep *ep)
876 {
877 
878 	del_timer_sync(&ep->timer);
879 	if (!test_and_set_bit(TIMEOUT, &ep->com.flags)) {
880 		c4iw_put_ep(&ep->com);
881 	}
882 }
883 
884 static enum
885 c4iw_ep_state state_read(struct c4iw_ep_common *epc)
886 {
887 	enum c4iw_ep_state state;
888 
889 	mutex_lock(&epc->mutex);
890 	state = epc->state;
891 	mutex_unlock(&epc->mutex);
892 
893 	return (state);
894 }
895 
896 static void
897 __state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state new)
898 {
899 
900 	epc->state = new;
901 }
902 
903 static void
904 state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state new)
905 {
906 
907 	mutex_lock(&epc->mutex);
908 	__state_set(epc, new);
909 	mutex_unlock(&epc->mutex);
910 }
911 
912 static void *
913 alloc_ep(int size, gfp_t gfp)
914 {
915 	struct c4iw_ep_common *epc;
916 
917 	epc = kzalloc(size, gfp);
918 	if (epc == NULL)
919 		return (NULL);
920 
921 	kref_init(&epc->kref);
922 	mutex_init(&epc->mutex);
923 	c4iw_init_wr_wait(&epc->wr_wait);
924 
925 	return (epc);
926 }
927 
928 void
929 __free_ep(struct c4iw_ep_common *epc)
930 {
931 	CTR2(KTR_IW_CXGBE, "%s:feB %p", __func__, epc);
932 	KASSERT(!epc->so, ("%s warning ep->so %p \n", __func__, epc->so));
933 	KASSERT(!epc->entry.tqe_prev, ("%s epc %p still on req list!\n", __func__, epc));
934 	free(epc, M_DEVBUF);
935 	CTR2(KTR_IW_CXGBE, "%s:feE %p", __func__, epc);
936 }
937 
938 void _c4iw_free_ep(struct kref *kref)
939 {
940 	struct c4iw_ep *ep;
941 	struct c4iw_ep_common *epc;
942 
943 	ep = container_of(kref, struct c4iw_ep, com.kref);
944 	epc = &ep->com;
945 	KASSERT(!epc->so, ("%s ep->so %p", __func__, epc->so));
946 	KASSERT(!epc->entry.tqe_prev, ("%s epc %p still on req list",
947 	    __func__, epc));
948 	kfree(ep);
949 }
950 
951 static void release_ep_resources(struct c4iw_ep *ep)
952 {
953 	CTR2(KTR_IW_CXGBE, "%s:rerB %p", __func__, ep);
954 	set_bit(RELEASE_RESOURCES, &ep->com.flags);
955 	c4iw_put_ep(&ep->com);
956 	CTR2(KTR_IW_CXGBE, "%s:rerE %p", __func__, ep);
957 }
958 
959 static void
960 send_mpa_req(struct c4iw_ep *ep)
961 {
962 	int mpalen;
963 	struct mpa_message *mpa;
964 	struct mpa_v2_conn_params mpa_v2_params;
965 	struct mbuf *m;
966 	char mpa_rev_to_use = mpa_rev;
967 	int err;
968 
969 	if (ep->retry_with_mpa_v1)
970 		mpa_rev_to_use = 1;
971 	mpalen = sizeof(*mpa) + ep->plen;
972 	if (mpa_rev_to_use == 2)
973 		mpalen += sizeof(struct mpa_v2_conn_params);
974 
975 	if (mpalen > MHLEN)
976 		CXGBE_UNIMPLEMENTED(__func__);
977 
978 	m = m_gethdr(M_NOWAIT, MT_DATA);
979 	if (m == NULL) {
980 		connect_reply_upcall(ep, -ENOMEM);
981 		return;
982 	}
983 
984 	mpa = mtod(m, struct mpa_message *);
985 	m->m_len = mpalen;
986 	m->m_pkthdr.len = mpalen;
987 	memcpy(mpa->key, MPA_KEY_REQ, sizeof(mpa->key));
988 	mpa->flags = (crc_enabled ? MPA_CRC : 0) |
989 		(markers_enabled ? MPA_MARKERS : 0) |
990 		(mpa_rev_to_use == 2 ? MPA_ENHANCED_RDMA_CONN : 0);
991 	mpa->private_data_size = htons(ep->plen);
992 	mpa->revision = mpa_rev_to_use;
993 
994 	if (mpa_rev_to_use == 1) {
995 		ep->tried_with_mpa_v1 = 1;
996 		ep->retry_with_mpa_v1 = 0;
997 	}
998 
999 	if (mpa_rev_to_use == 2) {
1000 		mpa->private_data_size +=
1001 			htons(sizeof(struct mpa_v2_conn_params));
1002 		mpa_v2_params.ird = htons((u16)ep->ird);
1003 		mpa_v2_params.ord = htons((u16)ep->ord);
1004 
1005 		if (peer2peer) {
1006 			mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL);
1007 
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 		}
1016 		memcpy(mpa->private_data, &mpa_v2_params,
1017 			sizeof(struct mpa_v2_conn_params));
1018 
1019 		if (ep->plen) {
1020 
1021 			memcpy(mpa->private_data +
1022 				sizeof(struct mpa_v2_conn_params),
1023 				ep->mpa_pkt + sizeof(*mpa), ep->plen);
1024 		}
1025 	} else {
1026 
1027 		if (ep->plen)
1028 			memcpy(mpa->private_data,
1029 					ep->mpa_pkt + sizeof(*mpa), ep->plen);
1030 		CTR2(KTR_IW_CXGBE, "%s:smr7 %p", __func__, ep);
1031 	}
1032 
1033 	err = sosend(ep->com.so, NULL, NULL, m, NULL, MSG_DONTWAIT, ep->com.thread);
1034 	if (err) {
1035 		connect_reply_upcall(ep, -ENOMEM);
1036 		return;
1037 	}
1038 
1039 	START_EP_TIMER(ep);
1040 	state_set(&ep->com, MPA_REQ_SENT);
1041 	ep->mpa_attr.initiator = 1;
1042 }
1043 
1044 static int send_mpa_reject(struct c4iw_ep *ep, const void *pdata, u8 plen)
1045 {
1046 	int mpalen ;
1047 	struct mpa_message *mpa;
1048 	struct mpa_v2_conn_params mpa_v2_params;
1049 	struct mbuf *m;
1050 	int err;
1051 
1052 	CTR4(KTR_IW_CXGBE, "%s:smrejB %p %u %d", __func__, ep, ep->hwtid,
1053 	    ep->plen);
1054 
1055 	mpalen = sizeof(*mpa) + plen;
1056 
1057 	if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
1058 
1059 		mpalen += sizeof(struct mpa_v2_conn_params);
1060 		CTR4(KTR_IW_CXGBE, "%s:smrej1 %p %u %d", __func__, ep,
1061 		    ep->mpa_attr.version, mpalen);
1062 	}
1063 
1064 	if (mpalen > MHLEN)
1065 		CXGBE_UNIMPLEMENTED(__func__);
1066 
1067 	m = m_gethdr(M_NOWAIT, MT_DATA);
1068 	if (m == NULL) {
1069 
1070 		printf("%s - cannot alloc mbuf!\n", __func__);
1071 		CTR2(KTR_IW_CXGBE, "%s:smrej2 %p", __func__, ep);
1072 		return (-ENOMEM);
1073 	}
1074 
1075 
1076 	mpa = mtod(m, struct mpa_message *);
1077 	m->m_len = mpalen;
1078 	m->m_pkthdr.len = mpalen;
1079 	memset(mpa, 0, sizeof(*mpa));
1080 	memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
1081 	mpa->flags = MPA_REJECT;
1082 	mpa->revision = mpa_rev;
1083 	mpa->private_data_size = htons(plen);
1084 
1085 	if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
1086 
1087 		mpa->flags |= MPA_ENHANCED_RDMA_CONN;
1088 		mpa->private_data_size +=
1089 			htons(sizeof(struct mpa_v2_conn_params));
1090 		mpa_v2_params.ird = htons(((u16)ep->ird) |
1091 				(peer2peer ? MPA_V2_PEER2PEER_MODEL :
1092 				 0));
1093 		mpa_v2_params.ord = htons(((u16)ep->ord) | (peer2peer ?
1094 					(p2p_type ==
1095 					 FW_RI_INIT_P2PTYPE_RDMA_WRITE ?
1096 					 MPA_V2_RDMA_WRITE_RTR : p2p_type ==
1097 					 FW_RI_INIT_P2PTYPE_READ_REQ ?
1098 					 MPA_V2_RDMA_READ_RTR : 0) : 0));
1099 		memcpy(mpa->private_data, &mpa_v2_params,
1100 				sizeof(struct mpa_v2_conn_params));
1101 
1102 		if (ep->plen)
1103 			memcpy(mpa->private_data +
1104 					sizeof(struct mpa_v2_conn_params), pdata, plen);
1105 		CTR5(KTR_IW_CXGBE, "%s:smrej3 %p %d %d %d", __func__, ep,
1106 		    mpa_v2_params.ird, mpa_v2_params.ord, ep->plen);
1107 	} else
1108 		if (plen)
1109 			memcpy(mpa->private_data, pdata, plen);
1110 
1111 	err = sosend(ep->com.so, NULL, NULL, m, NULL, MSG_DONTWAIT, ep->com.thread);
1112 	if (!err)
1113 		ep->snd_seq += mpalen;
1114 	CTR4(KTR_IW_CXGBE, "%s:smrejE %p %u %d", __func__, ep, ep->hwtid, err);
1115 	return err;
1116 }
1117 
1118 static int send_mpa_reply(struct c4iw_ep *ep, const void *pdata, u8 plen)
1119 {
1120 	int mpalen;
1121 	struct mpa_message *mpa;
1122 	struct mbuf *m;
1123 	struct mpa_v2_conn_params mpa_v2_params;
1124 	int err;
1125 
1126 	CTR2(KTR_IW_CXGBE, "%s:smrepB %p", __func__, ep);
1127 
1128 	mpalen = sizeof(*mpa) + plen;
1129 
1130 	if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
1131 
1132 		CTR3(KTR_IW_CXGBE, "%s:smrep1 %p %d", __func__, ep,
1133 		    ep->mpa_attr.version);
1134 		mpalen += sizeof(struct mpa_v2_conn_params);
1135 	}
1136 
1137 	if (mpalen > MHLEN)
1138 		CXGBE_UNIMPLEMENTED(__func__);
1139 
1140 	m = m_gethdr(M_NOWAIT, MT_DATA);
1141 	if (m == NULL) {
1142 
1143 		CTR2(KTR_IW_CXGBE, "%s:smrep2 %p", __func__, ep);
1144 		printf("%s - cannot alloc mbuf!\n", __func__);
1145 		return (-ENOMEM);
1146 	}
1147 
1148 
1149 	mpa = mtod(m, struct mpa_message *);
1150 	m->m_len = mpalen;
1151 	m->m_pkthdr.len = mpalen;
1152 	memset(mpa, 0, sizeof(*mpa));
1153 	memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
1154 	mpa->flags = (ep->mpa_attr.crc_enabled ? MPA_CRC : 0) |
1155 		(markers_enabled ? MPA_MARKERS : 0);
1156 	mpa->revision = ep->mpa_attr.version;
1157 	mpa->private_data_size = htons(plen);
1158 
1159 	if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
1160 
1161 		mpa->flags |= MPA_ENHANCED_RDMA_CONN;
1162 		mpa->private_data_size +=
1163 			htons(sizeof(struct mpa_v2_conn_params));
1164 		mpa_v2_params.ird = htons((u16)ep->ird);
1165 		mpa_v2_params.ord = htons((u16)ep->ord);
1166 		CTR5(KTR_IW_CXGBE, "%s:smrep3 %p %d %d %d", __func__, ep,
1167 		    ep->mpa_attr.version, mpa_v2_params.ird, mpa_v2_params.ord);
1168 
1169 		if (peer2peer && (ep->mpa_attr.p2p_type !=
1170 			FW_RI_INIT_P2PTYPE_DISABLED)) {
1171 
1172 			mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL);
1173 
1174 			if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE) {
1175 
1176 				mpa_v2_params.ord |=
1177 					htons(MPA_V2_RDMA_WRITE_RTR);
1178 				CTR5(KTR_IW_CXGBE, "%s:smrep4 %p %d %d %d",
1179 				    __func__, ep, p2p_type, mpa_v2_params.ird,
1180 				    mpa_v2_params.ord);
1181 			}
1182 			else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ) {
1183 
1184 				mpa_v2_params.ord |=
1185 					htons(MPA_V2_RDMA_READ_RTR);
1186 				CTR5(KTR_IW_CXGBE, "%s:smrep5 %p %d %d %d",
1187 				    __func__, ep, p2p_type, mpa_v2_params.ird,
1188 				    mpa_v2_params.ord);
1189 			}
1190 		}
1191 
1192 		memcpy(mpa->private_data, &mpa_v2_params,
1193 			sizeof(struct mpa_v2_conn_params));
1194 
1195 		if (ep->plen)
1196 			memcpy(mpa->private_data +
1197 				sizeof(struct mpa_v2_conn_params), pdata, plen);
1198 	} else
1199 		if (plen)
1200 			memcpy(mpa->private_data, pdata, plen);
1201 
1202 	state_set(&ep->com, MPA_REP_SENT);
1203 	ep->snd_seq += mpalen;
1204 	err = sosend(ep->com.so, NULL, NULL, m, NULL, MSG_DONTWAIT,
1205 			ep->com.thread);
1206 	CTR3(KTR_IW_CXGBE, "%s:smrepE %p %d", __func__, ep, err);
1207 	return err;
1208 }
1209 
1210 
1211 
1212 static void close_complete_upcall(struct c4iw_ep *ep)
1213 {
1214 	struct iw_cm_event event;
1215 
1216 	CTR2(KTR_IW_CXGBE, "%s:ccuB %p", __func__, ep);
1217 	memset(&event, 0, sizeof(event));
1218 	event.event = IW_CM_EVENT_CLOSE;
1219 
1220 	if (ep->com.cm_id) {
1221 
1222 		CTR2(KTR_IW_CXGBE, "%s:ccu1 %1", __func__, ep);
1223 		ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1224 		ep->com.cm_id->rem_ref(ep->com.cm_id);
1225 		ep->com.cm_id = NULL;
1226 		ep->com.qp = NULL;
1227 		set_bit(CLOSE_UPCALL, &ep->com.history);
1228 	}
1229 	CTR2(KTR_IW_CXGBE, "%s:ccuE %p", __func__, ep);
1230 }
1231 
1232 static int abort_connection(struct c4iw_ep *ep)
1233 {
1234 	int err;
1235 
1236 	CTR2(KTR_IW_CXGBE, "%s:abB %p", __func__, ep);
1237 	close_complete_upcall(ep);
1238 	state_set(&ep->com, ABORTING);
1239 	abort_socket(ep);
1240 	err = close_socket(&ep->com, 0);
1241 	set_bit(ABORT_CONN, &ep->com.history);
1242 	CTR2(KTR_IW_CXGBE, "%s:abE %p", __func__, ep);
1243 	return err;
1244 }
1245 
1246 static void peer_close_upcall(struct c4iw_ep *ep)
1247 {
1248 	struct iw_cm_event event;
1249 
1250 	CTR2(KTR_IW_CXGBE, "%s:pcuB %p", __func__, ep);
1251 	memset(&event, 0, sizeof(event));
1252 	event.event = IW_CM_EVENT_DISCONNECT;
1253 
1254 	if (ep->com.cm_id) {
1255 
1256 		CTR2(KTR_IW_CXGBE, "%s:pcu1 %p", __func__, ep);
1257 		ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1258 		set_bit(DISCONN_UPCALL, &ep->com.history);
1259 	}
1260 	CTR2(KTR_IW_CXGBE, "%s:pcuE %p", __func__, ep);
1261 }
1262 
1263 static void peer_abort_upcall(struct c4iw_ep *ep)
1264 {
1265 	struct iw_cm_event event;
1266 
1267 	CTR2(KTR_IW_CXGBE, "%s:pauB %p", __func__, ep);
1268 	memset(&event, 0, sizeof(event));
1269 	event.event = IW_CM_EVENT_CLOSE;
1270 	event.status = -ECONNRESET;
1271 
1272 	if (ep->com.cm_id) {
1273 
1274 		CTR2(KTR_IW_CXGBE, "%s:pau1 %p", __func__, ep);
1275 		ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1276 		ep->com.cm_id->rem_ref(ep->com.cm_id);
1277 		ep->com.cm_id = NULL;
1278 		ep->com.qp = NULL;
1279 		set_bit(ABORT_UPCALL, &ep->com.history);
1280 	}
1281 	CTR2(KTR_IW_CXGBE, "%s:pauE %p", __func__, ep);
1282 }
1283 
1284 static void connect_reply_upcall(struct c4iw_ep *ep, int status)
1285 {
1286 	struct iw_cm_event event;
1287 
1288 	CTR3(KTR_IW_CXGBE, "%s:cruB %p", __func__, ep, status);
1289 	memset(&event, 0, sizeof(event));
1290 	event.event = IW_CM_EVENT_CONNECT_REPLY;
1291 	event.status = (status ==-ECONNABORTED)?-ECONNRESET: status;
1292 	event.local_addr = ep->com.local_addr;
1293 	event.remote_addr = ep->com.remote_addr;
1294 
1295 	if ((status == 0) || (status == -ECONNREFUSED)) {
1296 
1297 		if (!ep->tried_with_mpa_v1) {
1298 
1299 			CTR2(KTR_IW_CXGBE, "%s:cru1 %p", __func__, ep);
1300 			/* this means MPA_v2 is used */
1301 			event.private_data_len = ep->plen -
1302 				sizeof(struct mpa_v2_conn_params);
1303 			event.private_data = ep->mpa_pkt +
1304 				sizeof(struct mpa_message) +
1305 				sizeof(struct mpa_v2_conn_params);
1306 		} else {
1307 
1308 			CTR2(KTR_IW_CXGBE, "%s:cru2 %p", __func__, ep);
1309 			/* this means MPA_v1 is used */
1310 			event.private_data_len = ep->plen;
1311 			event.private_data = ep->mpa_pkt +
1312 				sizeof(struct mpa_message);
1313 		}
1314 	}
1315 
1316 	if (ep->com.cm_id) {
1317 
1318 		CTR2(KTR_IW_CXGBE, "%s:cru3 %p", __func__, ep);
1319 		set_bit(CONN_RPL_UPCALL, &ep->com.history);
1320 		ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1321 	}
1322 
1323 	if(status == -ECONNABORTED) {
1324 
1325 		CTR3(KTR_IW_CXGBE, "%s:cruE %p %d", __func__, ep, status);
1326 		return;
1327 	}
1328 
1329 	if (status < 0) {
1330 
1331 		CTR3(KTR_IW_CXGBE, "%s:cru4 %p %d", __func__, ep, status);
1332 		ep->com.cm_id->rem_ref(ep->com.cm_id);
1333 		ep->com.cm_id = NULL;
1334 		ep->com.qp = NULL;
1335 	}
1336 
1337 	CTR2(KTR_IW_CXGBE, "%s:cruE %p", __func__, ep);
1338 }
1339 
1340 static void connect_request_upcall(struct c4iw_ep *ep)
1341 {
1342 	struct iw_cm_event event;
1343 
1344 	CTR3(KTR_IW_CXGBE, "%s: ep %p, mpa_v1 %d", __func__, ep,
1345 	    ep->tried_with_mpa_v1);
1346 
1347 	memset(&event, 0, sizeof(event));
1348 	event.event = IW_CM_EVENT_CONNECT_REQUEST;
1349 	event.local_addr = ep->com.local_addr;
1350 	event.remote_addr = ep->com.remote_addr;
1351 	event.provider_data = ep;
1352 	event.so = ep->com.so;
1353 
1354 	if (!ep->tried_with_mpa_v1) {
1355 		/* this means MPA_v2 is used */
1356 #ifdef IW_CM_MPAV2
1357 		event.ord = ep->ord;
1358 		event.ird = ep->ird;
1359 #endif
1360 		event.private_data_len = ep->plen -
1361 			sizeof(struct mpa_v2_conn_params);
1362 		event.private_data = ep->mpa_pkt + sizeof(struct mpa_message) +
1363 			sizeof(struct mpa_v2_conn_params);
1364 	} else {
1365 
1366 		/* this means MPA_v1 is used. Send max supported */
1367 #ifdef IW_CM_MPAV2
1368 		event.ord = c4iw_max_read_depth;
1369 		event.ird = c4iw_max_read_depth;
1370 #endif
1371 		event.private_data_len = ep->plen;
1372 		event.private_data = ep->mpa_pkt + sizeof(struct mpa_message);
1373 	}
1374 
1375 	c4iw_get_ep(&ep->com);
1376 	ep->parent_ep->com.cm_id->event_handler(ep->parent_ep->com.cm_id,
1377 	    &event);
1378 	set_bit(CONNREQ_UPCALL, &ep->com.history);
1379 	c4iw_put_ep(&ep->parent_ep->com);
1380 }
1381 
1382 static void established_upcall(struct c4iw_ep *ep)
1383 {
1384 	struct iw_cm_event event;
1385 
1386 	CTR2(KTR_IW_CXGBE, "%s:euB %p", __func__, ep);
1387 	memset(&event, 0, sizeof(event));
1388 	event.event = IW_CM_EVENT_ESTABLISHED;
1389 #ifdef IW_CM_MPAV2
1390 	event.ird = ep->ird;
1391 	event.ord = ep->ord;
1392 #endif
1393 	if (ep->com.cm_id) {
1394 
1395 		CTR2(KTR_IW_CXGBE, "%s:eu1 %p", __func__, ep);
1396 		ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1397 		set_bit(ESTAB_UPCALL, &ep->com.history);
1398 	}
1399 	CTR2(KTR_IW_CXGBE, "%s:euE %p", __func__, ep);
1400 }
1401 
1402 
1403 
1404 static void process_mpa_reply(struct c4iw_ep *ep)
1405 {
1406 	struct mpa_message *mpa;
1407 	struct mpa_v2_conn_params *mpa_v2_params;
1408 	u16 plen;
1409 	u16 resp_ird, resp_ord;
1410 	u8 rtr_mismatch = 0, insuff_ird = 0;
1411 	struct c4iw_qp_attributes attrs;
1412 	enum c4iw_qp_attr_mask mask;
1413 	int err;
1414 	struct mbuf *top, *m;
1415 	int flags = MSG_DONTWAIT;
1416 	struct uio uio;
1417 
1418 	CTR2(KTR_IW_CXGBE, "%s:pmrB %p", __func__, ep);
1419 
1420 	/*
1421 	 * Stop mpa timer.  If it expired, then the state has
1422 	 * changed and we bail since ep_timeout already aborted
1423 	 * the connection.
1424 	 */
1425 	STOP_EP_TIMER(ep);
1426 	if (state_read(&ep->com) != MPA_REQ_SENT)
1427 		return;
1428 
1429 	uio.uio_resid = 1000000;
1430 	uio.uio_td = ep->com.thread;
1431 	err = soreceive(ep->com.so, NULL, &uio, &top, NULL, &flags);
1432 
1433 	if (err) {
1434 
1435 		if (err == EWOULDBLOCK) {
1436 
1437 			CTR2(KTR_IW_CXGBE, "%s:pmr1 %p", __func__, ep);
1438 			START_EP_TIMER(ep);
1439 			return;
1440 		}
1441 		err = -err;
1442 		CTR2(KTR_IW_CXGBE, "%s:pmr2 %p", __func__, ep);
1443 		goto err;
1444 	}
1445 
1446 	if (ep->com.so->so_rcv.sb_mb) {
1447 
1448 		CTR2(KTR_IW_CXGBE, "%s:pmr3 %p", __func__, ep);
1449 		printf("%s data after soreceive called! so %p sb_mb %p top %p\n",
1450 		       __func__, ep->com.so, ep->com.so->so_rcv.sb_mb, top);
1451 	}
1452 
1453 	m = top;
1454 
1455 	do {
1456 
1457 		CTR2(KTR_IW_CXGBE, "%s:pmr4 %p", __func__, ep);
1458 		/*
1459 		 * If we get more than the supported amount of private data
1460 		 * then we must fail this connection.
1461 		 */
1462 		if (ep->mpa_pkt_len + m->m_len > sizeof(ep->mpa_pkt)) {
1463 
1464 			CTR3(KTR_IW_CXGBE, "%s:pmr5 %p %d", __func__, ep,
1465 			    ep->mpa_pkt_len + m->m_len);
1466 			err = (-EINVAL);
1467 			goto err;
1468 		}
1469 
1470 		/*
1471 		 * copy the new data into our accumulation buffer.
1472 		 */
1473 		m_copydata(m, 0, m->m_len, &(ep->mpa_pkt[ep->mpa_pkt_len]));
1474 		ep->mpa_pkt_len += m->m_len;
1475 		if (!m->m_next)
1476 			m = m->m_nextpkt;
1477 		else
1478 			m = m->m_next;
1479 	} while (m);
1480 
1481 	m_freem(top);
1482 	/*
1483 	 * if we don't even have the mpa message, then bail.
1484 	 */
1485 	if (ep->mpa_pkt_len < sizeof(*mpa))
1486 		return;
1487 	mpa = (struct mpa_message *) ep->mpa_pkt;
1488 
1489 	/* Validate MPA header. */
1490 	if (mpa->revision > mpa_rev) {
1491 
1492 		CTR4(KTR_IW_CXGBE, "%s:pmr6 %p %d %d", __func__, ep,
1493 		    mpa->revision, mpa_rev);
1494 		printk(KERN_ERR MOD "%s MPA version mismatch. Local = %d, "
1495 				" Received = %d\n", __func__, mpa_rev, mpa->revision);
1496 		err = -EPROTO;
1497 		goto err;
1498 	}
1499 
1500 	if (memcmp(mpa->key, MPA_KEY_REP, sizeof(mpa->key))) {
1501 
1502 		CTR2(KTR_IW_CXGBE, "%s:pmr7 %p", __func__, ep);
1503 		err = -EPROTO;
1504 		goto err;
1505 	}
1506 
1507 	plen = ntohs(mpa->private_data_size);
1508 
1509 	/*
1510 	 * Fail if there's too much private data.
1511 	 */
1512 	if (plen > MPA_MAX_PRIVATE_DATA) {
1513 
1514 		CTR2(KTR_IW_CXGBE, "%s:pmr8 %p", __func__, ep);
1515 		err = -EPROTO;
1516 		goto err;
1517 	}
1518 
1519 	/*
1520 	 * If plen does not account for pkt size
1521 	 */
1522 	if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
1523 
1524 		CTR2(KTR_IW_CXGBE, "%s:pmr9 %p", __func__, ep);
1525 		err = -EPROTO;
1526 		goto err;
1527 	}
1528 
1529 	ep->plen = (u8) plen;
1530 
1531 	/*
1532 	 * If we don't have all the pdata yet, then bail.
1533 	 * We'll continue process when more data arrives.
1534 	 */
1535 	if (ep->mpa_pkt_len < (sizeof(*mpa) + plen)) {
1536 
1537 		CTR2(KTR_IW_CXGBE, "%s:pmra %p", __func__, ep);
1538 		return;
1539 	}
1540 
1541 	if (mpa->flags & MPA_REJECT) {
1542 
1543 		CTR2(KTR_IW_CXGBE, "%s:pmrb %p", __func__, ep);
1544 		err = -ECONNREFUSED;
1545 		goto err;
1546 	}
1547 
1548 	/*
1549 	 * If we get here we have accumulated the entire mpa
1550 	 * start reply message including private data. And
1551 	 * the MPA header is valid.
1552 	 */
1553 	state_set(&ep->com, FPDU_MODE);
1554 	ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1555 	ep->mpa_attr.recv_marker_enabled = markers_enabled;
1556 	ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1557 	ep->mpa_attr.version = mpa->revision;
1558 	ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1559 
1560 	if (mpa->revision == 2) {
1561 
1562 		CTR2(KTR_IW_CXGBE, "%s:pmrc %p", __func__, ep);
1563 		ep->mpa_attr.enhanced_rdma_conn =
1564 			mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
1565 
1566 		if (ep->mpa_attr.enhanced_rdma_conn) {
1567 
1568 			CTR2(KTR_IW_CXGBE, "%s:pmrd %p", __func__, ep);
1569 			mpa_v2_params = (struct mpa_v2_conn_params *)
1570 				(ep->mpa_pkt + sizeof(*mpa));
1571 			resp_ird = ntohs(mpa_v2_params->ird) &
1572 				MPA_V2_IRD_ORD_MASK;
1573 			resp_ord = ntohs(mpa_v2_params->ord) &
1574 				MPA_V2_IRD_ORD_MASK;
1575 
1576 			/*
1577 			 * This is a double-check. Ideally, below checks are
1578 			 * not required since ird/ord stuff has been taken
1579 			 * care of in c4iw_accept_cr
1580 			 */
1581 			if ((ep->ird < resp_ord) || (ep->ord > resp_ird)) {
1582 
1583 				CTR2(KTR_IW_CXGBE, "%s:pmre %p", __func__, ep);
1584 				err = -ENOMEM;
1585 				ep->ird = resp_ord;
1586 				ep->ord = resp_ird;
1587 				insuff_ird = 1;
1588 			}
1589 
1590 			if (ntohs(mpa_v2_params->ird) &
1591 				MPA_V2_PEER2PEER_MODEL) {
1592 
1593 				CTR2(KTR_IW_CXGBE, "%s:pmrf %p", __func__, ep);
1594 				if (ntohs(mpa_v2_params->ord) &
1595 					MPA_V2_RDMA_WRITE_RTR) {
1596 
1597 					CTR2(KTR_IW_CXGBE, "%s:pmrg %p", __func__, ep);
1598 					ep->mpa_attr.p2p_type =
1599 						FW_RI_INIT_P2PTYPE_RDMA_WRITE;
1600 				}
1601 				else if (ntohs(mpa_v2_params->ord) &
1602 					MPA_V2_RDMA_READ_RTR) {
1603 
1604 					CTR2(KTR_IW_CXGBE, "%s:pmrh %p", __func__, ep);
1605 					ep->mpa_attr.p2p_type =
1606 						FW_RI_INIT_P2PTYPE_READ_REQ;
1607 				}
1608 			}
1609 		}
1610 	} else {
1611 
1612 		CTR2(KTR_IW_CXGBE, "%s:pmri %p", __func__, ep);
1613 
1614 		if (mpa->revision == 1) {
1615 
1616 			CTR2(KTR_IW_CXGBE, "%s:pmrj %p", __func__, ep);
1617 
1618 			if (peer2peer) {
1619 
1620 				CTR2(KTR_IW_CXGBE, "%s:pmrk %p", __func__, ep);
1621 				ep->mpa_attr.p2p_type = p2p_type;
1622 			}
1623 		}
1624 	}
1625 
1626 	if (set_tcpinfo(ep)) {
1627 
1628 		CTR2(KTR_IW_CXGBE, "%s:pmrl %p", __func__, ep);
1629 		printf("%s set_tcpinfo error\n", __func__);
1630 		goto err;
1631 	}
1632 
1633 	CTR6(KTR_IW_CXGBE, "%s - crc_enabled = %d, recv_marker_enabled = %d, "
1634 	    "xmit_marker_enabled = %d, version = %d p2p_type = %d", __func__,
1635 	    ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
1636 	    ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version,
1637 	    ep->mpa_attr.p2p_type);
1638 
1639 	/*
1640 	 * If responder's RTR does not match with that of initiator, assign
1641 	 * FW_RI_INIT_P2PTYPE_DISABLED in mpa attributes so that RTR is not
1642 	 * generated when moving QP to RTS state.
1643 	 * A TERM message will be sent after QP has moved to RTS state
1644 	 */
1645 	if ((ep->mpa_attr.version == 2) && peer2peer &&
1646 		(ep->mpa_attr.p2p_type != p2p_type)) {
1647 
1648 		CTR2(KTR_IW_CXGBE, "%s:pmrm %p", __func__, ep);
1649 		ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1650 		rtr_mismatch = 1;
1651 	}
1652 
1653 
1654 	//ep->ofld_txq = TOEPCB(ep->com.so)->ofld_txq;
1655 	attrs.mpa_attr = ep->mpa_attr;
1656 	attrs.max_ird = ep->ird;
1657 	attrs.max_ord = ep->ord;
1658 	attrs.llp_stream_handle = ep;
1659 	attrs.next_state = C4IW_QP_STATE_RTS;
1660 
1661 	mask = C4IW_QP_ATTR_NEXT_STATE |
1662 		C4IW_QP_ATTR_LLP_STREAM_HANDLE | C4IW_QP_ATTR_MPA_ATTR |
1663 		C4IW_QP_ATTR_MAX_IRD | C4IW_QP_ATTR_MAX_ORD;
1664 
1665 	/* bind QP and TID with INIT_WR */
1666 	err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp, mask, &attrs, 1);
1667 
1668 	if (err) {
1669 
1670 		CTR2(KTR_IW_CXGBE, "%s:pmrn %p", __func__, ep);
1671 		goto err;
1672 	}
1673 
1674 	/*
1675 	 * If responder's RTR requirement did not match with what initiator
1676 	 * supports, generate TERM message
1677 	 */
1678 	if (rtr_mismatch) {
1679 
1680 		CTR2(KTR_IW_CXGBE, "%s:pmro %p", __func__, ep);
1681 		printk(KERN_ERR "%s: RTR mismatch, sending TERM\n", __func__);
1682 		attrs.layer_etype = LAYER_MPA | DDP_LLP;
1683 		attrs.ecode = MPA_NOMATCH_RTR;
1684 		attrs.next_state = C4IW_QP_STATE_TERMINATE;
1685 		err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1686 			C4IW_QP_ATTR_NEXT_STATE, &attrs, 0);
1687 		err = -ENOMEM;
1688 		goto out;
1689 	}
1690 
1691 	/*
1692 	 * Generate TERM if initiator IRD is not sufficient for responder
1693 	 * provided ORD. Currently, we do the same behaviour even when
1694 	 * responder provided IRD is also not sufficient as regards to
1695 	 * initiator ORD.
1696 	 */
1697 	if (insuff_ird) {
1698 
1699 		CTR2(KTR_IW_CXGBE, "%s:pmrp %p", __func__, ep);
1700 		printk(KERN_ERR "%s: Insufficient IRD, sending TERM\n",
1701 				__func__);
1702 		attrs.layer_etype = LAYER_MPA | DDP_LLP;
1703 		attrs.ecode = MPA_INSUFF_IRD;
1704 		attrs.next_state = C4IW_QP_STATE_TERMINATE;
1705 		err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1706 			C4IW_QP_ATTR_NEXT_STATE, &attrs, 0);
1707 		err = -ENOMEM;
1708 		goto out;
1709 	}
1710 	goto out;
1711 err:
1712 	state_set(&ep->com, ABORTING);
1713 	abort_connection(ep);
1714 out:
1715 	connect_reply_upcall(ep, err);
1716 	CTR2(KTR_IW_CXGBE, "%s:pmrE %p", __func__, ep);
1717 	return;
1718 }
1719 
1720 static void
1721 process_mpa_request(struct c4iw_ep *ep)
1722 {
1723 	struct mpa_message *mpa;
1724 	u16 plen;
1725 	int flags = MSG_DONTWAIT;
1726 	int rc;
1727 	struct iovec iov;
1728 	struct uio uio;
1729 	enum c4iw_ep_state state = state_read(&ep->com);
1730 
1731 	CTR3(KTR_IW_CXGBE, "%s: ep %p, state %s", __func__, ep, states[state]);
1732 
1733 	if (state != MPA_REQ_WAIT)
1734 		return;
1735 
1736 	iov.iov_base = &ep->mpa_pkt[ep->mpa_pkt_len];
1737 	iov.iov_len = sizeof(ep->mpa_pkt) - ep->mpa_pkt_len;
1738 	uio.uio_iov = &iov;
1739 	uio.uio_iovcnt = 1;
1740 	uio.uio_offset = 0;
1741 	uio.uio_resid = sizeof(ep->mpa_pkt) - ep->mpa_pkt_len;
1742 	uio.uio_segflg = UIO_SYSSPACE;
1743 	uio.uio_rw = UIO_READ;
1744 	uio.uio_td = NULL; /* uio.uio_td = ep->com.thread; */
1745 
1746 	rc = soreceive(ep->com.so, NULL, &uio, NULL, NULL, &flags);
1747 	if (rc == EAGAIN)
1748 		return;
1749 	else if (rc) {
1750 abort:
1751 		STOP_EP_TIMER(ep);
1752 		abort_connection(ep);
1753 		return;
1754 	}
1755 	KASSERT(uio.uio_offset > 0, ("%s: sorecieve on so %p read no data",
1756 	    __func__, ep->com.so));
1757 	ep->mpa_pkt_len += uio.uio_offset;
1758 
1759 	/*
1760 	 * If we get more than the supported amount of private data then we must
1761 	 * fail this connection.  XXX: check so_rcv->sb_cc, or peek with another
1762 	 * soreceive, or increase the size of mpa_pkt by 1 and abort if the last
1763 	 * byte is filled by the soreceive above.
1764 	 */
1765 
1766 	/* Don't even have the MPA message.  Wait for more data to arrive. */
1767 	if (ep->mpa_pkt_len < sizeof(*mpa))
1768 		return;
1769 	mpa = (struct mpa_message *) ep->mpa_pkt;
1770 
1771 	/*
1772 	 * Validate MPA Header.
1773 	 */
1774 	if (mpa->revision > mpa_rev) {
1775 		log(LOG_ERR, "%s: MPA version mismatch. Local = %d,"
1776 		    " Received = %d\n", __func__, mpa_rev, mpa->revision);
1777 		goto abort;
1778 	}
1779 
1780 	if (memcmp(mpa->key, MPA_KEY_REQ, sizeof(mpa->key)))
1781 		goto abort;
1782 
1783 	/*
1784 	 * Fail if there's too much private data.
1785 	 */
1786 	plen = ntohs(mpa->private_data_size);
1787 	if (plen > MPA_MAX_PRIVATE_DATA)
1788 		goto abort;
1789 
1790 	/*
1791 	 * If plen does not account for pkt size
1792 	 */
1793 	if (ep->mpa_pkt_len > (sizeof(*mpa) + plen))
1794 		goto abort;
1795 
1796 	ep->plen = (u8) plen;
1797 
1798 	/*
1799 	 * If we don't have all the pdata yet, then bail.
1800 	 */
1801 	if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
1802 		return;
1803 
1804 	/*
1805 	 * If we get here we have accumulated the entire mpa
1806 	 * start reply message including private data.
1807 	 */
1808 	ep->mpa_attr.initiator = 0;
1809 	ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1810 	ep->mpa_attr.recv_marker_enabled = markers_enabled;
1811 	ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1812 	ep->mpa_attr.version = mpa->revision;
1813 	if (mpa->revision == 1)
1814 		ep->tried_with_mpa_v1 = 1;
1815 	ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1816 
1817 	if (mpa->revision == 2) {
1818 		ep->mpa_attr.enhanced_rdma_conn =
1819 		    mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
1820 		if (ep->mpa_attr.enhanced_rdma_conn) {
1821 			struct mpa_v2_conn_params *mpa_v2_params;
1822 			u16 ird, ord;
1823 
1824 			mpa_v2_params = (void *)&ep->mpa_pkt[sizeof(*mpa)];
1825 			ird = ntohs(mpa_v2_params->ird);
1826 			ord = ntohs(mpa_v2_params->ord);
1827 
1828 			ep->ird = ird & MPA_V2_IRD_ORD_MASK;
1829 			ep->ord = ord & MPA_V2_IRD_ORD_MASK;
1830 			if (ird & MPA_V2_PEER2PEER_MODEL && peer2peer) {
1831 				if (ord & MPA_V2_RDMA_WRITE_RTR) {
1832 					ep->mpa_attr.p2p_type =
1833 					    FW_RI_INIT_P2PTYPE_RDMA_WRITE;
1834 				} else if (ord & MPA_V2_RDMA_READ_RTR) {
1835 					ep->mpa_attr.p2p_type =
1836 					    FW_RI_INIT_P2PTYPE_READ_REQ;
1837 				}
1838 			}
1839 		}
1840 	} else if (mpa->revision == 1 && peer2peer)
1841 		ep->mpa_attr.p2p_type = p2p_type;
1842 
1843 	if (set_tcpinfo(ep))
1844 		goto abort;
1845 
1846 	CTR5(KTR_IW_CXGBE, "%s: crc_enabled = %d, recv_marker_enabled = %d, "
1847 	    "xmit_marker_enabled = %d, version = %d", __func__,
1848 	    ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
1849 	    ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version);
1850 
1851 	state_set(&ep->com, MPA_REQ_RCVD);
1852 	STOP_EP_TIMER(ep);
1853 
1854 	/* drive upcall */
1855 	mutex_lock(&ep->parent_ep->com.mutex);
1856 	if (ep->parent_ep->com.state != DEAD)
1857 		connect_request_upcall(ep);
1858 	else
1859 		abort_connection(ep);
1860 	mutex_unlock(&ep->parent_ep->com.mutex);
1861 }
1862 
1863 /*
1864  * Upcall from the adapter indicating data has been transmitted.
1865  * For us its just the single MPA request or reply.  We can now free
1866  * the skb holding the mpa message.
1867  */
1868 int c4iw_reject_cr(struct iw_cm_id *cm_id, const void *pdata, u8 pdata_len)
1869 {
1870 	int err;
1871 	struct c4iw_ep *ep = to_ep(cm_id);
1872 	CTR2(KTR_IW_CXGBE, "%s:crcB %p", __func__, ep);
1873 
1874 	if (state_read(&ep->com) == DEAD) {
1875 
1876 		CTR2(KTR_IW_CXGBE, "%s:crc1 %p", __func__, ep);
1877 		c4iw_put_ep(&ep->com);
1878 		return -ECONNRESET;
1879 	}
1880 	set_bit(ULP_REJECT, &ep->com.history);
1881 	BUG_ON(state_read(&ep->com) != MPA_REQ_RCVD);
1882 
1883 	if (mpa_rev == 0) {
1884 
1885 		CTR2(KTR_IW_CXGBE, "%s:crc2 %p", __func__, ep);
1886 		abort_connection(ep);
1887 	}
1888 	else {
1889 
1890 		CTR2(KTR_IW_CXGBE, "%s:crc3 %p", __func__, ep);
1891 		err = send_mpa_reject(ep, pdata, pdata_len);
1892 		err = soshutdown(ep->com.so, 3);
1893 	}
1894 	c4iw_put_ep(&ep->com);
1895 	CTR2(KTR_IW_CXGBE, "%s:crc4 %p", __func__, ep);
1896 	return 0;
1897 }
1898 
1899 int c4iw_accept_cr(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
1900 {
1901 	int err;
1902 	struct c4iw_qp_attributes attrs;
1903 	enum c4iw_qp_attr_mask mask;
1904 	struct c4iw_ep *ep = to_ep(cm_id);
1905 	struct c4iw_dev *h = to_c4iw_dev(cm_id->device);
1906 	struct c4iw_qp *qp = get_qhp(h, conn_param->qpn);
1907 
1908 	CTR2(KTR_IW_CXGBE, "%s:cacB %p", __func__, ep);
1909 
1910 	if (state_read(&ep->com) == DEAD) {
1911 
1912 		CTR2(KTR_IW_CXGBE, "%s:cac1 %p", __func__, ep);
1913 		err = -ECONNRESET;
1914 		goto err;
1915 	}
1916 
1917 	BUG_ON(state_read(&ep->com) != MPA_REQ_RCVD);
1918 	BUG_ON(!qp);
1919 
1920 	set_bit(ULP_ACCEPT, &ep->com.history);
1921 
1922 	if ((conn_param->ord > c4iw_max_read_depth) ||
1923 		(conn_param->ird > c4iw_max_read_depth)) {
1924 
1925 		CTR2(KTR_IW_CXGBE, "%s:cac2 %p", __func__, ep);
1926 		abort_connection(ep);
1927 		err = -EINVAL;
1928 		goto err;
1929 	}
1930 
1931 	if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
1932 
1933 		CTR2(KTR_IW_CXGBE, "%s:cac3 %p", __func__, ep);
1934 
1935 		if (conn_param->ord > ep->ird) {
1936 
1937 			CTR2(KTR_IW_CXGBE, "%s:cac4 %p", __func__, ep);
1938 			ep->ird = conn_param->ird;
1939 			ep->ord = conn_param->ord;
1940 			send_mpa_reject(ep, conn_param->private_data,
1941 					conn_param->private_data_len);
1942 			abort_connection(ep);
1943 			err = -ENOMEM;
1944 			goto err;
1945 		}
1946 
1947 		if (conn_param->ird > ep->ord) {
1948 
1949 			CTR2(KTR_IW_CXGBE, "%s:cac5 %p", __func__, ep);
1950 
1951 			if (!ep->ord) {
1952 
1953 				CTR2(KTR_IW_CXGBE, "%s:cac6 %p", __func__, ep);
1954 				conn_param->ird = 1;
1955 			}
1956 			else {
1957 				CTR2(KTR_IW_CXGBE, "%s:cac7 %p", __func__, ep);
1958 				abort_connection(ep);
1959 				err = -ENOMEM;
1960 				goto err;
1961 			}
1962 		}
1963 
1964 	}
1965 	ep->ird = conn_param->ird;
1966 	ep->ord = conn_param->ord;
1967 
1968 	if (ep->mpa_attr.version != 2) {
1969 
1970 		CTR2(KTR_IW_CXGBE, "%s:cac8 %p", __func__, ep);
1971 
1972 		if (peer2peer && ep->ird == 0) {
1973 
1974 			CTR2(KTR_IW_CXGBE, "%s:cac9 %p", __func__, ep);
1975 			ep->ird = 1;
1976 		}
1977 	}
1978 
1979 
1980 	cm_id->add_ref(cm_id);
1981 	ep->com.cm_id = cm_id;
1982 	ep->com.qp = qp;
1983 	//ep->ofld_txq = TOEPCB(ep->com.so)->ofld_txq;
1984 
1985 	/* bind QP to EP and move to RTS */
1986 	attrs.mpa_attr = ep->mpa_attr;
1987 	attrs.max_ird = ep->ird;
1988 	attrs.max_ord = ep->ord;
1989 	attrs.llp_stream_handle = ep;
1990 	attrs.next_state = C4IW_QP_STATE_RTS;
1991 
1992 	/* bind QP and TID with INIT_WR */
1993 	mask = C4IW_QP_ATTR_NEXT_STATE |
1994 		C4IW_QP_ATTR_LLP_STREAM_HANDLE |
1995 		C4IW_QP_ATTR_MPA_ATTR |
1996 		C4IW_QP_ATTR_MAX_IRD |
1997 		C4IW_QP_ATTR_MAX_ORD;
1998 
1999 	err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp, mask, &attrs, 1);
2000 
2001 	if (err) {
2002 
2003 		CTR2(KTR_IW_CXGBE, "%s:caca %p", __func__, ep);
2004 		goto err1;
2005 	}
2006 	err = send_mpa_reply(ep, conn_param->private_data,
2007 			conn_param->private_data_len);
2008 
2009 	if (err) {
2010 
2011 		CTR2(KTR_IW_CXGBE, "%s:caca %p", __func__, ep);
2012 		goto err1;
2013 	}
2014 
2015 	state_set(&ep->com, FPDU_MODE);
2016 	established_upcall(ep);
2017 	c4iw_put_ep(&ep->com);
2018 	CTR2(KTR_IW_CXGBE, "%s:cacE %p", __func__, ep);
2019 	return 0;
2020 err1:
2021 	ep->com.cm_id = NULL;
2022 	ep->com.qp = NULL;
2023 	cm_id->rem_ref(cm_id);
2024 err:
2025 	c4iw_put_ep(&ep->com);
2026 	CTR2(KTR_IW_CXGBE, "%s:cacE err %p", __func__, ep);
2027 	return err;
2028 }
2029 
2030 
2031 
2032 int c4iw_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
2033 {
2034 	int err = 0;
2035 	struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
2036 	struct c4iw_ep *ep = NULL;
2037 	struct rtentry *rt;
2038 	struct toedev *tdev;
2039 
2040 	CTR2(KTR_IW_CXGBE, "%s:ccB %p", __func__, cm_id);
2041 
2042 	if ((conn_param->ord > c4iw_max_read_depth) ||
2043 		(conn_param->ird > c4iw_max_read_depth)) {
2044 
2045 		CTR2(KTR_IW_CXGBE, "%s:cc1 %p", __func__, cm_id);
2046 		err = -EINVAL;
2047 		goto out;
2048 	}
2049 	ep = alloc_ep(sizeof(*ep), M_NOWAIT);
2050 
2051 	if (!ep) {
2052 
2053 		CTR2(KTR_IW_CXGBE, "%s:cc2 %p", __func__, cm_id);
2054 		printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__);
2055 		err = -ENOMEM;
2056 		goto out;
2057 	}
2058 	init_timer(&ep->timer);
2059 	ep->plen = conn_param->private_data_len;
2060 
2061 	if (ep->plen) {
2062 
2063 		CTR2(KTR_IW_CXGBE, "%s:cc3 %p", __func__, ep);
2064 		memcpy(ep->mpa_pkt + sizeof(struct mpa_message),
2065 				conn_param->private_data, ep->plen);
2066 	}
2067 	ep->ird = conn_param->ird;
2068 	ep->ord = conn_param->ord;
2069 
2070 	if (peer2peer && ep->ord == 0) {
2071 
2072 		CTR2(KTR_IW_CXGBE, "%s:cc4 %p", __func__, ep);
2073 		ep->ord = 1;
2074 	}
2075 
2076 	cm_id->add_ref(cm_id);
2077 	ep->com.dev = dev;
2078 	ep->com.cm_id = cm_id;
2079 	ep->com.qp = get_qhp(dev, conn_param->qpn);
2080 
2081 	if (!ep->com.qp) {
2082 
2083 		CTR2(KTR_IW_CXGBE, "%s:cc5 %p", __func__, ep);
2084 		err = -EINVAL;
2085 		goto fail2;
2086 	}
2087 	ep->com.thread = curthread;
2088 	ep->com.so = cm_id->so;
2089 
2090 	init_sock(&ep->com);
2091 
2092 	/* find a route */
2093 	rt = find_route(
2094 		cm_id->local_addr.sin_addr.s_addr,
2095 		cm_id->remote_addr.sin_addr.s_addr,
2096 		cm_id->local_addr.sin_port,
2097 		cm_id->remote_addr.sin_port, 0);
2098 
2099 	if (!rt) {
2100 
2101 		CTR2(KTR_IW_CXGBE, "%s:cc7 %p", __func__, ep);
2102 		printk(KERN_ERR MOD "%s - cannot find route.\n", __func__);
2103 		err = -EHOSTUNREACH;
2104 		goto fail3;
2105 	}
2106 
2107 
2108 	if (!(rt->rt_ifp->if_flags & IFCAP_TOE)) {
2109 
2110 		CTR2(KTR_IW_CXGBE, "%s:cc8 %p", __func__, ep);
2111 		printf("%s - interface not TOE capable.\n", __func__);
2112 		goto fail3;
2113 	}
2114 	tdev = TOEDEV(rt->rt_ifp);
2115 
2116 	if (tdev == NULL) {
2117 
2118 		CTR2(KTR_IW_CXGBE, "%s:cc9 %p", __func__, ep);
2119 		printf("%s - No toedev for interface.\n", __func__);
2120 		goto fail3;
2121 	}
2122 	RTFREE(rt);
2123 
2124 	state_set(&ep->com, CONNECTING);
2125 	ep->tos = 0;
2126 	ep->com.local_addr = cm_id->local_addr;
2127 	ep->com.remote_addr = cm_id->remote_addr;
2128 	err = soconnect(ep->com.so, (struct sockaddr *)&ep->com.remote_addr,
2129 		ep->com.thread);
2130 
2131 	if (!err) {
2132 
2133 		CTR2(KTR_IW_CXGBE, "%s:cca %p", __func__, ep);
2134 		goto out;
2135 	}
2136 
2137 fail3:
2138 	CTR2(KTR_IW_CXGBE, "%s:ccb %p", __func__, ep);
2139 	RTFREE(rt);
2140 fail2:
2141 	cm_id->rem_ref(cm_id);
2142 	c4iw_put_ep(&ep->com);
2143 out:
2144 	CTR2(KTR_IW_CXGBE, "%s:ccE %p", __func__, ep);
2145 	return err;
2146 }
2147 
2148 /*
2149  * iwcm->create_listen.  Returns -errno on failure.
2150  */
2151 int
2152 c4iw_create_listen(struct iw_cm_id *cm_id, int backlog)
2153 {
2154 	int rc;
2155 	struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
2156 	struct c4iw_listen_ep *ep;
2157 	struct socket *so = cm_id->so;
2158 
2159 	ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
2160 	CTR5(KTR_IW_CXGBE, "%s: cm_id %p, lso %p, ep %p, inp %p", __func__,
2161 	    cm_id, so, ep, so->so_pcb);
2162 	if (ep == NULL) {
2163 		log(LOG_ERR, "%s: failed to alloc memory for endpoint\n",
2164 		    __func__);
2165 		rc = ENOMEM;
2166 		goto failed;
2167 	}
2168 
2169 	cm_id->add_ref(cm_id);
2170 	ep->com.cm_id = cm_id;
2171 	ep->com.dev = dev;
2172 	ep->backlog = backlog;
2173 	ep->com.local_addr = cm_id->local_addr;
2174 	ep->com.thread = curthread;
2175 	state_set(&ep->com, LISTEN);
2176 	ep->com.so = so;
2177 	init_sock(&ep->com);
2178 
2179 	rc = solisten(so, ep->backlog, ep->com.thread);
2180 	if (rc != 0) {
2181 		log(LOG_ERR, "%s: failed to start listener: %d\n", __func__,
2182 		    rc);
2183 		close_socket(&ep->com, 0);
2184 		cm_id->rem_ref(cm_id);
2185 		c4iw_put_ep(&ep->com);
2186 		goto failed;
2187 	}
2188 
2189 	cm_id->provider_data = ep;
2190 	return (0);
2191 
2192 failed:
2193 	CTR3(KTR_IW_CXGBE, "%s: cm_id %p, FAILED (%d)", __func__, cm_id, rc);
2194 	return (-rc);
2195 }
2196 
2197 int
2198 c4iw_destroy_listen(struct iw_cm_id *cm_id)
2199 {
2200 	int rc;
2201 	struct c4iw_listen_ep *ep = to_listen_ep(cm_id);
2202 
2203 	CTR4(KTR_IW_CXGBE, "%s: cm_id %p, so %p, inp %p", __func__, cm_id,
2204 	    cm_id->so, cm_id->so->so_pcb);
2205 
2206 	state_set(&ep->com, DEAD);
2207 	rc = close_socket(&ep->com, 0);
2208 	cm_id->rem_ref(cm_id);
2209 	c4iw_put_ep(&ep->com);
2210 
2211 	return (rc);
2212 }
2213 
2214 int c4iw_ep_disconnect(struct c4iw_ep *ep, int abrupt, gfp_t gfp)
2215 {
2216 	int ret = 0;
2217 	int close = 0;
2218 	int fatal = 0;
2219 	struct c4iw_rdev *rdev;
2220 
2221 	mutex_lock(&ep->com.mutex);
2222 
2223 	CTR2(KTR_IW_CXGBE, "%s:cedB %p", __func__, ep);
2224 
2225 	rdev = &ep->com.dev->rdev;
2226 
2227 	if (c4iw_fatal_error(rdev)) {
2228 
2229 		CTR2(KTR_IW_CXGBE, "%s:ced1 %p", __func__, ep);
2230 		fatal = 1;
2231 		close_complete_upcall(ep);
2232 		ep->com.state = DEAD;
2233 	}
2234 	CTR3(KTR_IW_CXGBE, "%s:ced2 %p %s", __func__, ep,
2235 	    states[ep->com.state]);
2236 
2237 	switch (ep->com.state) {
2238 
2239 		case MPA_REQ_WAIT:
2240 		case MPA_REQ_SENT:
2241 		case MPA_REQ_RCVD:
2242 		case MPA_REP_SENT:
2243 		case FPDU_MODE:
2244 			close = 1;
2245 			if (abrupt)
2246 				ep->com.state = ABORTING;
2247 			else {
2248 				ep->com.state = CLOSING;
2249 				START_EP_TIMER(ep);
2250 			}
2251 			set_bit(CLOSE_SENT, &ep->com.flags);
2252 			break;
2253 
2254 		case CLOSING:
2255 
2256 			if (!test_and_set_bit(CLOSE_SENT, &ep->com.flags)) {
2257 
2258 				close = 1;
2259 				if (abrupt) {
2260 					STOP_EP_TIMER(ep);
2261 					ep->com.state = ABORTING;
2262 				} else
2263 					ep->com.state = MORIBUND;
2264 			}
2265 			break;
2266 
2267 		case MORIBUND:
2268 		case ABORTING:
2269 		case DEAD:
2270 			CTR3(KTR_IW_CXGBE,
2271 			    "%s ignoring disconnect ep %p state %u", __func__,
2272 			    ep, ep->com.state);
2273 			break;
2274 
2275 		default:
2276 			BUG();
2277 			break;
2278 	}
2279 
2280 	mutex_unlock(&ep->com.mutex);
2281 
2282 	if (close) {
2283 
2284 		CTR2(KTR_IW_CXGBE, "%s:ced3 %p", __func__, ep);
2285 
2286 		if (abrupt) {
2287 
2288 			CTR2(KTR_IW_CXGBE, "%s:ced4 %p", __func__, ep);
2289 			set_bit(EP_DISC_ABORT, &ep->com.history);
2290 			ret = abort_connection(ep);
2291 		} else {
2292 
2293 			CTR2(KTR_IW_CXGBE, "%s:ced5 %p", __func__, ep);
2294 			set_bit(EP_DISC_CLOSE, &ep->com.history);
2295 
2296 			if (!ep->parent_ep)
2297 				__state_set(&ep->com, MORIBUND);
2298 			ret = shutdown_socket(&ep->com);
2299 		}
2300 
2301 		if (ret) {
2302 
2303 			fatal = 1;
2304 		}
2305 	}
2306 
2307 	if (fatal) {
2308 
2309 		release_ep_resources(ep);
2310 		CTR2(KTR_IW_CXGBE, "%s:ced6 %p", __func__, ep);
2311 	}
2312 	CTR2(KTR_IW_CXGBE, "%s:cedE %p", __func__, ep);
2313 	return ret;
2314 }
2315 
2316 #ifdef C4IW_EP_REDIRECT
2317 int c4iw_ep_redirect(void *ctx, struct dst_entry *old, struct dst_entry *new,
2318 		struct l2t_entry *l2t)
2319 {
2320 	struct c4iw_ep *ep = ctx;
2321 
2322 	if (ep->dst != old)
2323 		return 0;
2324 
2325 	PDBG("%s ep %p redirect to dst %p l2t %p\n", __func__, ep, new,
2326 			l2t);
2327 	dst_hold(new);
2328 	cxgb4_l2t_release(ep->l2t);
2329 	ep->l2t = l2t;
2330 	dst_release(old);
2331 	ep->dst = new;
2332 	return 1;
2333 }
2334 #endif
2335 
2336 
2337 
2338 static void ep_timeout(unsigned long arg)
2339 {
2340 	struct c4iw_ep *ep = (struct c4iw_ep *)arg;
2341 	int kickit = 0;
2342 
2343 	CTR2(KTR_IW_CXGBE, "%s:etB %p", __func__, ep);
2344 	spin_lock(&timeout_lock);
2345 
2346 	if (!test_and_set_bit(TIMEOUT, &ep->com.flags)) {
2347 
2348 		list_add_tail(&ep->entry, &timeout_list);
2349 		kickit = 1;
2350 	}
2351 	spin_unlock(&timeout_lock);
2352 
2353 	if (kickit) {
2354 
2355 		CTR2(KTR_IW_CXGBE, "%s:et1 %p", __func__, ep);
2356 		queue_work(c4iw_taskq, &c4iw_task);
2357 	}
2358 	CTR2(KTR_IW_CXGBE, "%s:etE %p", __func__, ep);
2359 }
2360 
2361 static int fw6_wr_rpl(struct adapter *sc, const __be64 *rpl)
2362 {
2363 	uint64_t val = be64toh(*rpl);
2364 	int ret;
2365 	struct c4iw_wr_wait *wr_waitp;
2366 
2367 	ret = (int)((val >> 8) & 0xff);
2368 	wr_waitp = (struct c4iw_wr_wait *)rpl[1];
2369 	CTR3(KTR_IW_CXGBE, "%s wr_waitp %p ret %u", __func__, wr_waitp, ret);
2370 	if (wr_waitp)
2371 		c4iw_wake_up(wr_waitp, ret ? -ret : 0);
2372 
2373 	return (0);
2374 }
2375 
2376 static int fw6_cqe_handler(struct adapter *sc, const __be64 *rpl)
2377 {
2378 	struct t4_cqe cqe =*(const struct t4_cqe *)(&rpl[0]);
2379 
2380 	CTR2(KTR_IW_CXGBE, "%s rpl %p", __func__, rpl);
2381 	c4iw_ev_dispatch(sc->iwarp_softc, &cqe);
2382 
2383 	return (0);
2384 }
2385 
2386 static int terminate(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m)
2387 {
2388 
2389 	struct adapter *sc = iq->adapter;
2390 
2391 	const struct cpl_rdma_terminate *rpl = (const void *)(rss + 1);
2392 	unsigned int tid = GET_TID(rpl);
2393 	struct c4iw_qp_attributes attrs;
2394 	struct toepcb *toep = lookup_tid(sc, tid);
2395 	struct socket *so = inp_inpcbtosocket(toep->inp);
2396 	struct c4iw_ep *ep = so->so_rcv.sb_upcallarg;
2397 
2398 	CTR2(KTR_IW_CXGBE, "%s:tB %p %d", __func__, ep);
2399 
2400 	if (ep && ep->com.qp) {
2401 
2402 		printk(KERN_WARNING MOD "TERM received tid %u qpid %u\n", tid,
2403 				ep->com.qp->wq.sq.qid);
2404 		attrs.next_state = C4IW_QP_STATE_TERMINATE;
2405 		c4iw_modify_qp(ep->com.dev, ep->com.qp, C4IW_QP_ATTR_NEXT_STATE, &attrs,
2406 				1);
2407 	} else
2408 		printk(KERN_WARNING MOD "TERM received tid %u no ep/qp\n", tid);
2409 	CTR2(KTR_IW_CXGBE, "%s:tE %p %d", __func__, ep);
2410 
2411 	return 0;
2412 }
2413 
2414 	void
2415 c4iw_cm_init_cpl(struct adapter *sc)
2416 {
2417 
2418 	t4_register_cpl_handler(sc, CPL_RDMA_TERMINATE, terminate);
2419 	t4_register_fw_msg_handler(sc, FW6_TYPE_WR_RPL, fw6_wr_rpl);
2420 	t4_register_fw_msg_handler(sc, FW6_TYPE_CQE, fw6_cqe_handler);
2421 	t4_register_an_handler(sc, c4iw_ev_handler);
2422 }
2423 
2424 	void
2425 c4iw_cm_term_cpl(struct adapter *sc)
2426 {
2427 
2428 	t4_register_cpl_handler(sc, CPL_RDMA_TERMINATE, NULL);
2429 	t4_register_fw_msg_handler(sc, FW6_TYPE_WR_RPL, NULL);
2430 	t4_register_fw_msg_handler(sc, FW6_TYPE_CQE, NULL);
2431 }
2432 
2433 int __init c4iw_cm_init(void)
2434 {
2435 
2436 	TAILQ_INIT(&req_list);
2437 	spin_lock_init(&req_lock);
2438 	INIT_LIST_HEAD(&timeout_list);
2439 	spin_lock_init(&timeout_lock);
2440 
2441 	INIT_WORK(&c4iw_task, process_req);
2442 
2443 	c4iw_taskq = create_singlethread_workqueue("iw_cxgbe");
2444 	if (!c4iw_taskq)
2445 		return -ENOMEM;
2446 
2447 
2448 	return 0;
2449 }
2450 
2451 void __exit c4iw_cm_term(void)
2452 {
2453 	WARN_ON(!TAILQ_EMPTY(&req_list));
2454 	WARN_ON(!list_empty(&timeout_list));
2455 	flush_workqueue(c4iw_taskq);
2456 	destroy_workqueue(c4iw_taskq);
2457 }
2458 #endif
2459