xref: /freebsd/sys/dev/cxgbe/iw_cxgbe/cm.c (revision ff0ba87247820afbdfdc1b307c803f7923d0e4d3)
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/route.h>
46 
47 #include <netinet/in_systm.h>
48 #include <netinet/in_pcb.h>
49 #include <netinet/ip.h>
50 #include <netinet/ip_var.h>
51 #include <netinet/tcp_var.h>
52 #include <netinet/tcp.h>
53 #include <netinet/tcpip.h>
54 
55 #include <netinet/toecore.h>
56 
57 struct sge_iq;
58 struct rss_header;
59 #include <linux/types.h>
60 #include "offload.h"
61 #include "tom/t4_tom.h"
62 
63 #define TOEPCB(so)  ((struct toepcb *)(so_sototcpcb((so))->t_toe))
64 
65 #include "iw_cxgbe.h"
66 #include <linux/module.h>
67 #include <linux/workqueue.h>
68 #include <linux/notifier.h>
69 #include <linux/inetdevice.h>
70 #include <linux/if_vlan.h>
71 #include <net/netevent.h>
72 
73 static spinlock_t req_lock;
74 static TAILQ_HEAD(c4iw_ep_list, c4iw_ep_common) req_list;
75 static struct work_struct c4iw_task;
76 static struct workqueue_struct *c4iw_taskq;
77 static LIST_HEAD(timeout_list);
78 static spinlock_t timeout_lock;
79 
80 static void process_req(struct work_struct *ctx);
81 static void start_ep_timer(struct c4iw_ep *ep);
82 static void stop_ep_timer(struct c4iw_ep *ep);
83 static int set_tcpinfo(struct c4iw_ep *ep);
84 static enum c4iw_ep_state state_read(struct c4iw_ep_common *epc);
85 static void __state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state tostate);
86 static void state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state tostate);
87 static void *alloc_ep(int size, gfp_t flags);
88 void __free_ep(struct c4iw_ep_common *epc);
89 static struct rtentry * find_route(__be32 local_ip, __be32 peer_ip, __be16 local_port,
90 		__be16 peer_port, u8 tos);
91 static int close_socket(struct c4iw_ep_common *epc, int close);
92 static int shutdown_socket(struct c4iw_ep_common *epc);
93 static void abort_socket(struct c4iw_ep *ep);
94 static void send_mpa_req(struct c4iw_ep *ep);
95 static int send_mpa_reject(struct c4iw_ep *ep, const void *pdata, u8 plen);
96 static int send_mpa_reply(struct c4iw_ep *ep, const void *pdata, u8 plen);
97 static void close_complete_upcall(struct c4iw_ep *ep, int status);
98 static int abort_connection(struct c4iw_ep *ep);
99 static void peer_close_upcall(struct c4iw_ep *ep);
100 static void peer_abort_upcall(struct c4iw_ep *ep);
101 static void connect_reply_upcall(struct c4iw_ep *ep, int status);
102 static void connect_request_upcall(struct c4iw_ep *ep);
103 static void established_upcall(struct c4iw_ep *ep);
104 static void process_mpa_reply(struct c4iw_ep *ep);
105 static void process_mpa_request(struct c4iw_ep *ep);
106 static void process_peer_close(struct c4iw_ep *ep);
107 static void process_conn_error(struct c4iw_ep *ep);
108 static void process_close_complete(struct c4iw_ep *ep);
109 static void ep_timeout(unsigned long arg);
110 static void init_sock(struct c4iw_ep_common *epc);
111 static void process_data(struct c4iw_ep *ep);
112 static void process_connected(struct c4iw_ep *ep);
113 static struct socket * dequeue_socket(struct socket *head, struct sockaddr_in **remote, struct c4iw_ep *child_ep);
114 static void process_newconn(struct c4iw_ep *parent_ep);
115 static int c4iw_so_upcall(struct socket *so, void *arg, int waitflag);
116 static void process_socket_event(struct c4iw_ep *ep);
117 static void release_ep_resources(struct c4iw_ep *ep);
118 
119 #define START_EP_TIMER(ep) \
120     do { \
121 	    CTR3(KTR_IW_CXGBE, "start_ep_timer (%s:%d) ep %p", \
122 		__func__, __LINE__, (ep)); \
123 	    start_ep_timer(ep); \
124     } while (0)
125 
126 #define STOP_EP_TIMER(ep) \
127     do { \
128 	    CTR3(KTR_IW_CXGBE, "stop_ep_timer (%s:%d) ep %p", \
129 		__func__, __LINE__, (ep)); \
130 	    stop_ep_timer(ep); \
131     } while (0)
132 
133 #ifdef KTR
134 static char *states[] = {
135 	"idle",
136 	"listen",
137 	"connecting",
138 	"mpa_wait_req",
139 	"mpa_req_sent",
140 	"mpa_req_rcvd",
141 	"mpa_rep_sent",
142 	"fpdu_mode",
143 	"aborting",
144 	"closing",
145 	"moribund",
146 	"dead",
147 	NULL,
148 };
149 #endif
150 
151 static void
152 process_req(struct work_struct *ctx)
153 {
154 	struct c4iw_ep_common *epc;
155 
156 	spin_lock(&req_lock);
157 	while (!TAILQ_EMPTY(&req_list)) {
158 		epc = TAILQ_FIRST(&req_list);
159 		TAILQ_REMOVE(&req_list, epc, entry);
160 		epc->entry.tqe_prev = NULL;
161 		spin_unlock(&req_lock);
162 		if (epc->so)
163 			process_socket_event((struct c4iw_ep *)epc);
164 		c4iw_put_ep(epc);
165 		spin_lock(&req_lock);
166 	}
167 	spin_unlock(&req_lock);
168 }
169 
170 /*
171  * XXX: doesn't belong here in the iWARP driver.
172  * XXX: assumes that the connection was offloaded by cxgbe/t4_tom if TF_TOE is
173  *      set.  Is this a valid assumption for active open?
174  */
175 static int
176 set_tcpinfo(struct c4iw_ep *ep)
177 {
178 	struct socket *so = ep->com.so;
179 	struct inpcb *inp = sotoinpcb(so);
180 	struct tcpcb *tp;
181 	struct toepcb *toep;
182 	int rc = 0;
183 
184 	INP_WLOCK(inp);
185 	tp = intotcpcb(inp);
186 	if ((tp->t_flags & TF_TOE) == 0) {
187 		rc = EINVAL;
188 		log(LOG_ERR, "%s: connection not offloaded (so %p, ep %p)\n",
189 		    __func__, so, ep);
190 		goto done;
191 	}
192 	toep = TOEPCB(so);
193 
194 	ep->hwtid = toep->tid;
195 	ep->snd_seq = tp->snd_nxt;
196 	ep->rcv_seq = tp->rcv_nxt;
197 	ep->emss = max(tp->t_maxseg, 128);
198 done:
199 	INP_WUNLOCK(inp);
200 	return (rc);
201 
202 }
203 
204 static struct rtentry *
205 find_route(__be32 local_ip, __be32 peer_ip, __be16 local_port,
206 		__be16 peer_port, u8 tos)
207 {
208 	struct route iproute;
209 	struct sockaddr_in *dst = (struct sockaddr_in *)&iproute.ro_dst;
210 
211 	CTR5(KTR_IW_CXGBE, "%s:frtB %x, %x, %d, %d", __func__, local_ip,
212 	    peer_ip, ntohs(local_port), ntohs(peer_port));
213 	bzero(&iproute, sizeof iproute);
214 	dst->sin_family = AF_INET;
215 	dst->sin_len = sizeof *dst;
216 	dst->sin_addr.s_addr = peer_ip;
217 
218 	rtalloc(&iproute);
219 	CTR2(KTR_IW_CXGBE, "%s:frtE %p", __func__, (uint64_t)iproute.ro_rt);
220 	return iproute.ro_rt;
221 }
222 
223 static int
224 close_socket(struct c4iw_ep_common *epc, int close)
225 {
226 	struct socket *so = epc->so;
227 	int rc;
228 
229 	CTR4(KTR_IW_CXGBE, "%s: so %p, ep %p, state %s", __func__, epc, so,
230 	    states[epc->state]);
231 
232 	SOCK_LOCK(so);
233 	soupcall_clear(so, SO_RCV);
234 	SOCK_UNLOCK(so);
235 
236 	if (close)
237                 rc = soclose(so);
238         else
239                 rc = soshutdown(so, SHUT_WR | SHUT_RD);
240 	epc->so = NULL;
241 
242 	return (rc);
243 }
244 
245 static int
246 shutdown_socket(struct c4iw_ep_common *epc)
247 {
248 
249 	CTR4(KTR_IW_CXGBE, "%s: so %p, ep %p, state %s", __func__, epc->so, epc,
250 	    states[epc->state]);
251 
252 	return (soshutdown(epc->so, SHUT_WR));
253 }
254 
255 static void
256 abort_socket(struct c4iw_ep *ep)
257 {
258 	struct sockopt sopt;
259 	int rc;
260 	struct linger l;
261 
262 	CTR4(KTR_IW_CXGBE, "%s ep %p so %p state %s", __func__, ep, ep->com.so,
263 	    states[ep->com.state]);
264 
265 	l.l_onoff = 1;
266 	l.l_linger = 0;
267 
268 	/* linger_time of 0 forces RST to be sent */
269 	sopt.sopt_dir = SOPT_SET;
270 	sopt.sopt_level = SOL_SOCKET;
271 	sopt.sopt_name = SO_LINGER;
272 	sopt.sopt_val = (caddr_t)&l;
273 	sopt.sopt_valsize = sizeof l;
274 	sopt.sopt_td = NULL;
275 	rc = sosetopt(ep->com.so, &sopt);
276 	if (rc) {
277 		log(LOG_ERR, "%s: can't set linger to 0, no RST! err %d\n",
278 		    __func__, rc);
279 	}
280 }
281 
282 static void
283 process_peer_close(struct c4iw_ep *ep)
284 {
285 	struct c4iw_qp_attributes attrs;
286 	int disconnect = 1;
287 	int release = 0;
288 
289 	CTR4(KTR_IW_CXGBE, "%s:ppcB ep %p so %p state %s", __func__, ep,
290 	    ep->com.so, states[ep->com.state]);
291 
292 	mutex_lock(&ep->com.mutex);
293 	switch (ep->com.state) {
294 
295 		case MPA_REQ_WAIT:
296 			CTR2(KTR_IW_CXGBE, "%s:ppc1 %p MPA_REQ_WAIT CLOSING",
297 			    __func__, ep);
298 			__state_set(&ep->com, CLOSING);
299 			break;
300 
301 		case MPA_REQ_SENT:
302 			CTR2(KTR_IW_CXGBE, "%s:ppc2 %p MPA_REQ_SENT CLOSING",
303 			    __func__, ep);
304 			__state_set(&ep->com, DEAD);
305 			connect_reply_upcall(ep, -ECONNABORTED);
306 
307 			disconnect = 0;
308 			STOP_EP_TIMER(ep);
309 			close_socket(&ep->com, 0);
310 			ep->com.cm_id->rem_ref(ep->com.cm_id);
311 			ep->com.cm_id = NULL;
312 			ep->com.qp = NULL;
313 			release = 1;
314 			break;
315 
316 		case MPA_REQ_RCVD:
317 
318 			/*
319 			 * We're gonna mark this puppy DEAD, but keep
320 			 * the reference on it until the ULP accepts or
321 			 * rejects the CR.
322 			 */
323 			CTR2(KTR_IW_CXGBE, "%s:ppc3 %p MPA_REQ_RCVD CLOSING",
324 			    __func__, ep);
325 			__state_set(&ep->com, CLOSING);
326 			c4iw_get_ep(&ep->com);
327 			break;
328 
329 		case MPA_REP_SENT:
330 			CTR2(KTR_IW_CXGBE, "%s:ppc4 %p MPA_REP_SENT CLOSING",
331 			    __func__, ep);
332 			__state_set(&ep->com, CLOSING);
333 			break;
334 
335 		case FPDU_MODE:
336 			CTR2(KTR_IW_CXGBE, "%s:ppc5 %p FPDU_MODE CLOSING",
337 			    __func__, ep);
338 			START_EP_TIMER(ep);
339 			__state_set(&ep->com, CLOSING);
340 			attrs.next_state = C4IW_QP_STATE_CLOSING;
341 			c4iw_modify_qp(ep->com.dev, ep->com.qp,
342 					C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
343 			peer_close_upcall(ep);
344 			break;
345 
346 		case ABORTING:
347 			CTR2(KTR_IW_CXGBE, "%s:ppc6 %p ABORTING (disconn)",
348 			    __func__, ep);
349 			disconnect = 0;
350 			break;
351 
352 		case CLOSING:
353 			CTR2(KTR_IW_CXGBE, "%s:ppc7 %p CLOSING MORIBUND",
354 			    __func__, ep);
355 			__state_set(&ep->com, MORIBUND);
356 			disconnect = 0;
357 			break;
358 
359 		case MORIBUND:
360 			CTR2(KTR_IW_CXGBE, "%s:ppc8 %p MORIBUND DEAD", __func__,
361 			    ep);
362 			STOP_EP_TIMER(ep);
363 			if (ep->com.cm_id && ep->com.qp) {
364 				attrs.next_state = C4IW_QP_STATE_IDLE;
365 				c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
366 						C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
367 			}
368 			close_socket(&ep->com, 0);
369 			close_complete_upcall(ep, 0);
370 			__state_set(&ep->com, DEAD);
371 			release = 1;
372 			disconnect = 0;
373 			break;
374 
375 		case DEAD:
376 			CTR2(KTR_IW_CXGBE, "%s:ppc9 %p DEAD (disconn)",
377 			    __func__, ep);
378 			disconnect = 0;
379 			break;
380 
381 		default:
382 			panic("%s: ep %p state %d", __func__, ep,
383 			    ep->com.state);
384 			break;
385 	}
386 
387 	mutex_unlock(&ep->com.mutex);
388 
389 	if (disconnect) {
390 
391 		CTR2(KTR_IW_CXGBE, "%s:ppca %p", __func__, ep);
392 		c4iw_ep_disconnect(ep, 0, M_NOWAIT);
393 	}
394 	if (release) {
395 
396 		CTR2(KTR_IW_CXGBE, "%s:ppcb %p", __func__, ep);
397 		c4iw_put_ep(&ep->com);
398 	}
399 	CTR2(KTR_IW_CXGBE, "%s:ppcE %p", __func__, ep);
400 	return;
401 }
402 
403 static void
404 process_conn_error(struct c4iw_ep *ep)
405 {
406 	struct c4iw_qp_attributes attrs;
407 	int ret;
408 	int state;
409 
410 	state = state_read(&ep->com);
411 	CTR5(KTR_IW_CXGBE, "%s:pceB ep %p so %p so->so_error %u state %s",
412 	    __func__, ep, ep->com.so, ep->com.so->so_error,
413 	    states[ep->com.state]);
414 
415 	switch (state) {
416 
417 		case MPA_REQ_WAIT:
418 			STOP_EP_TIMER(ep);
419 			break;
420 
421 		case MPA_REQ_SENT:
422 			STOP_EP_TIMER(ep);
423 			connect_reply_upcall(ep, -ECONNRESET);
424 			break;
425 
426 		case MPA_REP_SENT:
427 			ep->com.rpl_err = ECONNRESET;
428 			CTR1(KTR_IW_CXGBE, "waking up ep %p", ep);
429 			break;
430 
431 		case MPA_REQ_RCVD:
432 
433 			/*
434 			 * We're gonna mark this puppy DEAD, but keep
435 			 * the reference on it until the ULP accepts or
436 			 * rejects the CR.
437 			 */
438 			c4iw_get_ep(&ep->com);
439 			break;
440 
441 		case MORIBUND:
442 		case CLOSING:
443 			STOP_EP_TIMER(ep);
444 			/*FALLTHROUGH*/
445 		case FPDU_MODE:
446 
447 			if (ep->com.cm_id && ep->com.qp) {
448 
449 				attrs.next_state = C4IW_QP_STATE_ERROR;
450 				ret = c4iw_modify_qp(ep->com.qp->rhp,
451 					ep->com.qp, C4IW_QP_ATTR_NEXT_STATE,
452 					&attrs, 1);
453 				if (ret)
454 					log(LOG_ERR,
455 							"%s - qp <- error failed!\n",
456 							__func__);
457 			}
458 			peer_abort_upcall(ep);
459 			break;
460 
461 		case ABORTING:
462 			break;
463 
464 		case DEAD:
465 			CTR2(KTR_IW_CXGBE, "%s so_error %d IN DEAD STATE!!!!",
466 			    __func__, ep->com.so->so_error);
467 			return;
468 
469 		default:
470 			panic("%s: ep %p state %d", __func__, ep, state);
471 			break;
472 	}
473 
474 	if (state != ABORTING) {
475 
476 		CTR2(KTR_IW_CXGBE, "%s:pce1 %p", __func__, ep);
477 		close_socket(&ep->com, 1);
478 		state_set(&ep->com, DEAD);
479 		c4iw_put_ep(&ep->com);
480 	}
481 	CTR2(KTR_IW_CXGBE, "%s:pceE %p", __func__, ep);
482 	return;
483 }
484 
485 static void
486 process_close_complete(struct c4iw_ep *ep)
487 {
488 	struct c4iw_qp_attributes attrs;
489 	int release = 0;
490 
491 	CTR4(KTR_IW_CXGBE, "%s:pccB ep %p so %p state %s", __func__, ep,
492 	    ep->com.so, states[ep->com.state]);
493 
494 	/* The cm_id may be null if we failed to connect */
495 	mutex_lock(&ep->com.mutex);
496 
497 	switch (ep->com.state) {
498 
499 		case CLOSING:
500 			CTR2(KTR_IW_CXGBE, "%s:pcc1 %p CLOSING MORIBUND",
501 			    __func__, ep);
502 			__state_set(&ep->com, MORIBUND);
503 			break;
504 
505 		case MORIBUND:
506 			CTR2(KTR_IW_CXGBE, "%s:pcc1 %p MORIBUND DEAD", __func__,
507 			    ep);
508 			STOP_EP_TIMER(ep);
509 
510 			if ((ep->com.cm_id) && (ep->com.qp)) {
511 
512 				CTR2(KTR_IW_CXGBE, "%s:pcc2 %p QP_STATE_IDLE",
513 				    __func__, ep);
514 				attrs.next_state = C4IW_QP_STATE_IDLE;
515 				c4iw_modify_qp(ep->com.dev,
516 						ep->com.qp,
517 						C4IW_QP_ATTR_NEXT_STATE,
518 						&attrs, 1);
519 			}
520 
521 			if (ep->parent_ep) {
522 
523 				CTR2(KTR_IW_CXGBE, "%s:pcc3 %p", __func__, ep);
524 				close_socket(&ep->com, 1);
525 			}
526 			else {
527 
528 				CTR2(KTR_IW_CXGBE, "%s:pcc4 %p", __func__, ep);
529 				close_socket(&ep->com, 0);
530 			}
531 			close_complete_upcall(ep, 0);
532 			__state_set(&ep->com, DEAD);
533 			release = 1;
534 			break;
535 
536 		case ABORTING:
537 			CTR2(KTR_IW_CXGBE, "%s:pcc5 %p ABORTING", __func__, ep);
538 			break;
539 
540 		case DEAD:
541 		default:
542 			CTR2(KTR_IW_CXGBE, "%s:pcc6 %p DEAD", __func__, ep);
543 			panic("%s:pcc6 %p DEAD", __func__, ep);
544 			break;
545 	}
546 	mutex_unlock(&ep->com.mutex);
547 
548 	if (release) {
549 
550 		CTR2(KTR_IW_CXGBE, "%s:pcc7 %p", __func__, ep);
551 		c4iw_put_ep(&ep->com);
552 	}
553 	CTR2(KTR_IW_CXGBE, "%s:pccE %p", __func__, ep);
554 	return;
555 }
556 
557 static void
558 init_sock(struct c4iw_ep_common *epc)
559 {
560 	int rc;
561 	struct sockopt sopt;
562 	struct socket *so = epc->so;
563 	int on = 1;
564 
565 	SOCK_LOCK(so);
566 	soupcall_set(so, SO_RCV, c4iw_so_upcall, epc);
567 	so->so_state |= SS_NBIO;
568 	SOCK_UNLOCK(so);
569 	sopt.sopt_dir = SOPT_SET;
570 	sopt.sopt_level = IPPROTO_TCP;
571 	sopt.sopt_name = TCP_NODELAY;
572 	sopt.sopt_val = (caddr_t)&on;
573 	sopt.sopt_valsize = sizeof on;
574 	sopt.sopt_td = NULL;
575 	rc = sosetopt(so, &sopt);
576 	if (rc) {
577 		log(LOG_ERR, "%s: can't set TCP_NODELAY on so %p (%d)\n",
578 		    __func__, so, rc);
579 	}
580 }
581 
582 static void
583 process_data(struct c4iw_ep *ep)
584 {
585 	struct sockaddr_in *local, *remote;
586 
587 	CTR5(KTR_IW_CXGBE, "%s: so %p, ep %p, state %s, sbused %d", __func__,
588 	    ep->com.so, ep, states[ep->com.state], sbused(&ep->com.so->so_rcv));
589 
590 	switch (state_read(&ep->com)) {
591 	case MPA_REQ_SENT:
592 		process_mpa_reply(ep);
593 		break;
594 	case MPA_REQ_WAIT:
595 		in_getsockaddr(ep->com.so, (struct sockaddr **)&local);
596 		in_getpeeraddr(ep->com.so, (struct sockaddr **)&remote);
597 		ep->com.local_addr = *local;
598 		ep->com.remote_addr = *remote;
599 		free(local, M_SONAME);
600 		free(remote, M_SONAME);
601 		process_mpa_request(ep);
602 		break;
603 	default:
604 		if (sbused(&ep->com.so->so_rcv))
605 			log(LOG_ERR, "%s: Unexpected streaming data. ep %p, "
606 			    "state %d, so %p, so_state 0x%x, sbused %u\n",
607 			    __func__, ep, state_read(&ep->com), ep->com.so,
608 			    ep->com.so->so_state, sbused(&ep->com.so->so_rcv));
609 		break;
610 	}
611 }
612 
613 static void
614 process_connected(struct c4iw_ep *ep)
615 {
616 
617 	if ((ep->com.so->so_state & SS_ISCONNECTED) && !ep->com.so->so_error)
618 		send_mpa_req(ep);
619 	else {
620 		connect_reply_upcall(ep, -ep->com.so->so_error);
621 		close_socket(&ep->com, 0);
622 		state_set(&ep->com, DEAD);
623 		c4iw_put_ep(&ep->com);
624 	}
625 }
626 
627 static struct socket *
628 dequeue_socket(struct socket *head, struct sockaddr_in **remote,
629     struct c4iw_ep *child_ep)
630 {
631 	struct socket *so;
632 
633 	ACCEPT_LOCK();
634 	so = TAILQ_FIRST(&head->so_comp);
635 	if (!so) {
636 		ACCEPT_UNLOCK();
637 		return (NULL);
638 	}
639 	TAILQ_REMOVE(&head->so_comp, so, so_list);
640 	head->so_qlen--;
641 	SOCK_LOCK(so);
642 	so->so_qstate &= ~SQ_COMP;
643 	so->so_head = NULL;
644 	soref(so);
645 	soupcall_set(so, SO_RCV, c4iw_so_upcall, child_ep);
646 	so->so_state |= SS_NBIO;
647 	SOCK_UNLOCK(so);
648 	ACCEPT_UNLOCK();
649 	soaccept(so, (struct sockaddr **)remote);
650 
651 	return (so);
652 }
653 
654 static void
655 process_newconn(struct c4iw_ep *parent_ep)
656 {
657 	struct socket *child_so;
658 	struct c4iw_ep *child_ep;
659 	struct sockaddr_in *remote;
660 
661 	child_ep = alloc_ep(sizeof(*child_ep), M_NOWAIT);
662 	if (!child_ep) {
663 		CTR3(KTR_IW_CXGBE, "%s: parent so %p, parent ep %p, ENOMEM",
664 		    __func__, parent_ep->com.so, parent_ep);
665 		log(LOG_ERR, "%s: failed to allocate ep entry\n", __func__);
666 		return;
667 	}
668 
669 	child_so = dequeue_socket(parent_ep->com.so, &remote, child_ep);
670 	if (!child_so) {
671 		CTR4(KTR_IW_CXGBE,
672 		    "%s: parent so %p, parent ep %p, child ep %p, dequeue err",
673 		    __func__, parent_ep->com.so, parent_ep, child_ep);
674 		log(LOG_ERR, "%s: failed to dequeue child socket\n", __func__);
675 		__free_ep(&child_ep->com);
676 		return;
677 
678 	}
679 
680 	CTR5(KTR_IW_CXGBE,
681 	    "%s: parent so %p, parent ep %p, child so %p, child ep %p",
682 	     __func__, parent_ep->com.so, parent_ep, child_so, child_ep);
683 
684 	child_ep->com.local_addr = parent_ep->com.local_addr;
685 	child_ep->com.remote_addr = *remote;
686 	child_ep->com.dev = parent_ep->com.dev;
687 	child_ep->com.so = child_so;
688 	child_ep->com.cm_id = NULL;
689 	child_ep->com.thread = parent_ep->com.thread;
690 	child_ep->parent_ep = parent_ep;
691 
692 	free(remote, M_SONAME);
693 	c4iw_get_ep(&parent_ep->com);
694 	child_ep->parent_ep = parent_ep;
695 	init_timer(&child_ep->timer);
696 	state_set(&child_ep->com, MPA_REQ_WAIT);
697 	START_EP_TIMER(child_ep);
698 
699 	/* maybe the request has already been queued up on the socket... */
700 	process_mpa_request(child_ep);
701 }
702 
703 static int
704 c4iw_so_upcall(struct socket *so, void *arg, int waitflag)
705 {
706 	struct c4iw_ep *ep = arg;
707 
708 	spin_lock(&req_lock);
709 
710 	CTR6(KTR_IW_CXGBE,
711 	    "%s: so %p, so_state 0x%x, ep %p, ep_state %s, tqe_prev %p",
712 	    __func__, so, so->so_state, ep, states[ep->com.state],
713 	    ep->com.entry.tqe_prev);
714 
715 	if (ep && ep->com.so && !ep->com.entry.tqe_prev) {
716 		KASSERT(ep->com.so == so, ("%s: XXX review.", __func__));
717 		c4iw_get_ep(&ep->com);
718 		TAILQ_INSERT_TAIL(&req_list, &ep->com, entry);
719 		queue_work(c4iw_taskq, &c4iw_task);
720 	}
721 
722 	spin_unlock(&req_lock);
723 	return (SU_OK);
724 }
725 
726 static void
727 process_socket_event(struct c4iw_ep *ep)
728 {
729 	int state = state_read(&ep->com);
730 	struct socket *so = ep->com.so;
731 
732 	CTR6(KTR_IW_CXGBE, "process_socket_event: so %p, so_state 0x%x, "
733 	    "so_err %d, sb_state 0x%x, ep %p, ep_state %s", so, so->so_state,
734 	    so->so_error, so->so_rcv.sb_state, ep, states[state]);
735 
736 	if (state == CONNECTING) {
737 		process_connected(ep);
738 		return;
739 	}
740 
741 	if (state == LISTEN) {
742 		process_newconn(ep);
743 		return;
744 	}
745 
746 	/* connection error */
747 	if (so->so_error) {
748 		process_conn_error(ep);
749 		return;
750 	}
751 
752 	/* peer close */
753 	if ((so->so_rcv.sb_state & SBS_CANTRCVMORE) && state < CLOSING) {
754 		process_peer_close(ep);
755 		return;
756 	}
757 
758 	/* close complete */
759 	if (so->so_state & SS_ISDISCONNECTED) {
760 		process_close_complete(ep);
761 		return;
762 	}
763 
764 	/* rx data */
765 	process_data(ep);
766 }
767 
768 SYSCTL_NODE(_hw, OID_AUTO, iw_cxgbe, CTLFLAG_RD, 0, "iw_cxgbe driver parameters");
769 
770 int db_delay_usecs = 1;
771 SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, db_delay_usecs, CTLFLAG_RWTUN, &db_delay_usecs, 0,
772 		"Usecs to delay awaiting db fifo to drain");
773 
774 static int dack_mode = 1;
775 SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, dack_mode, CTLFLAG_RWTUN, &dack_mode, 0,
776 		"Delayed ack mode (default = 1)");
777 
778 int c4iw_max_read_depth = 8;
779 SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, c4iw_max_read_depth, CTLFLAG_RWTUN, &c4iw_max_read_depth, 0,
780 		"Per-connection max ORD/IRD (default = 8)");
781 
782 static int enable_tcp_timestamps;
783 SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, enable_tcp_timestamps, CTLFLAG_RWTUN, &enable_tcp_timestamps, 0,
784 		"Enable tcp timestamps (default = 0)");
785 
786 static int enable_tcp_sack;
787 SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, enable_tcp_sack, CTLFLAG_RWTUN, &enable_tcp_sack, 0,
788 		"Enable tcp SACK (default = 0)");
789 
790 static int enable_tcp_window_scaling = 1;
791 SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, enable_tcp_window_scaling, CTLFLAG_RWTUN, &enable_tcp_window_scaling, 0,
792 		"Enable tcp window scaling (default = 1)");
793 
794 int c4iw_debug = 1;
795 SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, c4iw_debug, CTLFLAG_RWTUN, &c4iw_debug, 0,
796 		"Enable debug logging (default = 0)");
797 
798 static int peer2peer;
799 SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, peer2peer, CTLFLAG_RWTUN, &peer2peer, 0,
800 		"Support peer2peer ULPs (default = 0)");
801 
802 static int p2p_type = FW_RI_INIT_P2PTYPE_READ_REQ;
803 SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, p2p_type, CTLFLAG_RWTUN, &p2p_type, 0,
804 		"RDMAP opcode to use for the RTR message: 1 = RDMA_READ 0 = RDMA_WRITE (default 1)");
805 
806 static int ep_timeout_secs = 60;
807 SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, ep_timeout_secs, CTLFLAG_RWTUN, &ep_timeout_secs, 0,
808 		"CM Endpoint operation timeout in seconds (default = 60)");
809 
810 static int mpa_rev = 1;
811 #ifdef IW_CM_MPAV2
812 SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, mpa_rev, CTLFLAG_RWTUN, &mpa_rev, 0,
813 		"MPA Revision, 0 supports amso1100, 1 is RFC0544 spec compliant, 2 is IETF MPA Peer Connect Draft compliant (default = 1)");
814 #else
815 SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, mpa_rev, CTLFLAG_RWTUN, &mpa_rev, 0,
816 		"MPA Revision, 0 supports amso1100, 1 is RFC0544 spec compliant (default = 1)");
817 #endif
818 
819 static int markers_enabled;
820 SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, markers_enabled, CTLFLAG_RWTUN, &markers_enabled, 0,
821 		"Enable MPA MARKERS (default(0) = disabled)");
822 
823 static int crc_enabled = 1;
824 SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, crc_enabled, CTLFLAG_RWTUN, &crc_enabled, 0,
825 		"Enable MPA CRC (default(1) = enabled)");
826 
827 static int rcv_win = 256 * 1024;
828 SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, rcv_win, CTLFLAG_RWTUN, &rcv_win, 0,
829 		"TCP receive window in bytes (default = 256KB)");
830 
831 static int snd_win = 128 * 1024;
832 SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, snd_win, CTLFLAG_RWTUN, &snd_win, 0,
833 		"TCP send window in bytes (default = 128KB)");
834 
835 int db_fc_threshold = 2000;
836 SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, db_fc_threshold, CTLFLAG_RWTUN, &db_fc_threshold, 0,
837 		"QP count/threshold that triggers automatic");
838 
839 static void
840 start_ep_timer(struct c4iw_ep *ep)
841 {
842 
843 	if (timer_pending(&ep->timer)) {
844 		CTR2(KTR_IW_CXGBE, "%s: ep %p, already started", __func__, ep);
845 		printk(KERN_ERR "%s timer already started! ep %p\n", __func__,
846 		    ep);
847 		return;
848 	}
849 	clear_bit(TIMEOUT, &ep->com.flags);
850 	c4iw_get_ep(&ep->com);
851 	ep->timer.expires = jiffies + ep_timeout_secs * HZ;
852 	ep->timer.data = (unsigned long)ep;
853 	ep->timer.function = ep_timeout;
854 	add_timer(&ep->timer);
855 }
856 
857 static void
858 stop_ep_timer(struct c4iw_ep *ep)
859 {
860 
861 	del_timer_sync(&ep->timer);
862 	if (!test_and_set_bit(TIMEOUT, &ep->com.flags)) {
863 		c4iw_put_ep(&ep->com);
864 	}
865 }
866 
867 static enum
868 c4iw_ep_state state_read(struct c4iw_ep_common *epc)
869 {
870 	enum c4iw_ep_state state;
871 
872 	mutex_lock(&epc->mutex);
873 	state = epc->state;
874 	mutex_unlock(&epc->mutex);
875 
876 	return (state);
877 }
878 
879 static void
880 __state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state new)
881 {
882 
883 	epc->state = new;
884 }
885 
886 static void
887 state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state new)
888 {
889 
890 	mutex_lock(&epc->mutex);
891 	__state_set(epc, new);
892 	mutex_unlock(&epc->mutex);
893 }
894 
895 static void *
896 alloc_ep(int size, gfp_t gfp)
897 {
898 	struct c4iw_ep_common *epc;
899 
900 	epc = kzalloc(size, gfp);
901 	if (epc == NULL)
902 		return (NULL);
903 
904 	kref_init(&epc->kref);
905 	mutex_init(&epc->mutex);
906 	c4iw_init_wr_wait(&epc->wr_wait);
907 
908 	return (epc);
909 }
910 
911 void
912 __free_ep(struct c4iw_ep_common *epc)
913 {
914 	CTR2(KTR_IW_CXGBE, "%s:feB %p", __func__, epc);
915 	KASSERT(!epc->so, ("%s warning ep->so %p \n", __func__, epc->so));
916 	KASSERT(!epc->entry.tqe_prev, ("%s epc %p still on req list!\n", __func__, epc));
917 	free(epc, M_DEVBUF);
918 	CTR2(KTR_IW_CXGBE, "%s:feE %p", __func__, epc);
919 }
920 
921 void _c4iw_free_ep(struct kref *kref)
922 {
923 	struct c4iw_ep *ep;
924 	struct c4iw_ep_common *epc;
925 
926 	ep = container_of(kref, struct c4iw_ep, com.kref);
927 	epc = &ep->com;
928 	KASSERT(!epc->so, ("%s ep->so %p", __func__, epc->so));
929 	KASSERT(!epc->entry.tqe_prev, ("%s epc %p still on req list",
930 	    __func__, epc));
931 	kfree(ep);
932 }
933 
934 static void release_ep_resources(struct c4iw_ep *ep)
935 {
936 	CTR2(KTR_IW_CXGBE, "%s:rerB %p", __func__, ep);
937 	set_bit(RELEASE_RESOURCES, &ep->com.flags);
938 	c4iw_put_ep(&ep->com);
939 	CTR2(KTR_IW_CXGBE, "%s:rerE %p", __func__, ep);
940 }
941 
942 static void
943 send_mpa_req(struct c4iw_ep *ep)
944 {
945 	int mpalen;
946 	struct mpa_message *mpa;
947 	struct mpa_v2_conn_params mpa_v2_params;
948 	struct mbuf *m;
949 	char mpa_rev_to_use = mpa_rev;
950 	int err;
951 
952 	if (ep->retry_with_mpa_v1)
953 		mpa_rev_to_use = 1;
954 	mpalen = sizeof(*mpa) + ep->plen;
955 	if (mpa_rev_to_use == 2)
956 		mpalen += sizeof(struct mpa_v2_conn_params);
957 
958 	if (mpalen > MHLEN)
959 		CXGBE_UNIMPLEMENTED(__func__);
960 
961 	m = m_gethdr(M_NOWAIT, MT_DATA);
962 	if (m == NULL) {
963 		connect_reply_upcall(ep, -ENOMEM);
964 		return;
965 	}
966 
967 	mpa = mtod(m, struct mpa_message *);
968 	m->m_len = mpalen;
969 	m->m_pkthdr.len = mpalen;
970 	memcpy(mpa->key, MPA_KEY_REQ, sizeof(mpa->key));
971 	mpa->flags = (crc_enabled ? MPA_CRC : 0) |
972 		(markers_enabled ? MPA_MARKERS : 0) |
973 		(mpa_rev_to_use == 2 ? MPA_ENHANCED_RDMA_CONN : 0);
974 	mpa->private_data_size = htons(ep->plen);
975 	mpa->revision = mpa_rev_to_use;
976 
977 	if (mpa_rev_to_use == 1) {
978 		ep->tried_with_mpa_v1 = 1;
979 		ep->retry_with_mpa_v1 = 0;
980 	}
981 
982 	if (mpa_rev_to_use == 2) {
983 		mpa->private_data_size +=
984 			htons(sizeof(struct mpa_v2_conn_params));
985 		mpa_v2_params.ird = htons((u16)ep->ird);
986 		mpa_v2_params.ord = htons((u16)ep->ord);
987 
988 		if (peer2peer) {
989 			mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL);
990 
991 			if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE) {
992 				mpa_v2_params.ord |=
993 				    htons(MPA_V2_RDMA_WRITE_RTR);
994 			} else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ) {
995 				mpa_v2_params.ord |=
996 					htons(MPA_V2_RDMA_READ_RTR);
997 			}
998 		}
999 		memcpy(mpa->private_data, &mpa_v2_params,
1000 			sizeof(struct mpa_v2_conn_params));
1001 
1002 		if (ep->plen) {
1003 
1004 			memcpy(mpa->private_data +
1005 				sizeof(struct mpa_v2_conn_params),
1006 				ep->mpa_pkt + sizeof(*mpa), ep->plen);
1007 		}
1008 	} else {
1009 
1010 		if (ep->plen)
1011 			memcpy(mpa->private_data,
1012 					ep->mpa_pkt + sizeof(*mpa), ep->plen);
1013 		CTR2(KTR_IW_CXGBE, "%s:smr7 %p", __func__, ep);
1014 	}
1015 
1016 	err = sosend(ep->com.so, NULL, NULL, m, NULL, MSG_DONTWAIT, ep->com.thread);
1017 	if (err) {
1018 		connect_reply_upcall(ep, -ENOMEM);
1019 		return;
1020 	}
1021 
1022 	START_EP_TIMER(ep);
1023 	state_set(&ep->com, MPA_REQ_SENT);
1024 	ep->mpa_attr.initiator = 1;
1025 }
1026 
1027 static int send_mpa_reject(struct c4iw_ep *ep, const void *pdata, u8 plen)
1028 {
1029 	int mpalen ;
1030 	struct mpa_message *mpa;
1031 	struct mpa_v2_conn_params mpa_v2_params;
1032 	struct mbuf *m;
1033 	int err;
1034 
1035 	CTR4(KTR_IW_CXGBE, "%s:smrejB %p %u %d", __func__, ep, ep->hwtid,
1036 	    ep->plen);
1037 
1038 	mpalen = sizeof(*mpa) + plen;
1039 
1040 	if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
1041 
1042 		mpalen += sizeof(struct mpa_v2_conn_params);
1043 		CTR4(KTR_IW_CXGBE, "%s:smrej1 %p %u %d", __func__, ep,
1044 		    ep->mpa_attr.version, mpalen);
1045 	}
1046 
1047 	if (mpalen > MHLEN)
1048 		CXGBE_UNIMPLEMENTED(__func__);
1049 
1050 	m = m_gethdr(M_NOWAIT, MT_DATA);
1051 	if (m == NULL) {
1052 
1053 		printf("%s - cannot alloc mbuf!\n", __func__);
1054 		CTR2(KTR_IW_CXGBE, "%s:smrej2 %p", __func__, ep);
1055 		return (-ENOMEM);
1056 	}
1057 
1058 
1059 	mpa = mtod(m, struct mpa_message *);
1060 	m->m_len = mpalen;
1061 	m->m_pkthdr.len = mpalen;
1062 	memset(mpa, 0, sizeof(*mpa));
1063 	memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
1064 	mpa->flags = MPA_REJECT;
1065 	mpa->revision = mpa_rev;
1066 	mpa->private_data_size = htons(plen);
1067 
1068 	if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
1069 
1070 		mpa->flags |= MPA_ENHANCED_RDMA_CONN;
1071 		mpa->private_data_size +=
1072 			htons(sizeof(struct mpa_v2_conn_params));
1073 		mpa_v2_params.ird = htons(((u16)ep->ird) |
1074 				(peer2peer ? MPA_V2_PEER2PEER_MODEL :
1075 				 0));
1076 		mpa_v2_params.ord = htons(((u16)ep->ord) | (peer2peer ?
1077 					(p2p_type ==
1078 					 FW_RI_INIT_P2PTYPE_RDMA_WRITE ?
1079 					 MPA_V2_RDMA_WRITE_RTR : p2p_type ==
1080 					 FW_RI_INIT_P2PTYPE_READ_REQ ?
1081 					 MPA_V2_RDMA_READ_RTR : 0) : 0));
1082 		memcpy(mpa->private_data, &mpa_v2_params,
1083 				sizeof(struct mpa_v2_conn_params));
1084 
1085 		if (ep->plen)
1086 			memcpy(mpa->private_data +
1087 					sizeof(struct mpa_v2_conn_params), pdata, plen);
1088 		CTR5(KTR_IW_CXGBE, "%s:smrej3 %p %d %d %d", __func__, ep,
1089 		    mpa_v2_params.ird, mpa_v2_params.ord, ep->plen);
1090 	} else
1091 		if (plen)
1092 			memcpy(mpa->private_data, pdata, plen);
1093 
1094 	err = sosend(ep->com.so, NULL, NULL, m, NULL, MSG_DONTWAIT, ep->com.thread);
1095 	if (!err)
1096 		ep->snd_seq += mpalen;
1097 	CTR4(KTR_IW_CXGBE, "%s:smrejE %p %u %d", __func__, ep, ep->hwtid, err);
1098 	return err;
1099 }
1100 
1101 static int send_mpa_reply(struct c4iw_ep *ep, const void *pdata, u8 plen)
1102 {
1103 	int mpalen;
1104 	struct mpa_message *mpa;
1105 	struct mbuf *m;
1106 	struct mpa_v2_conn_params mpa_v2_params;
1107 	int err;
1108 
1109 	CTR2(KTR_IW_CXGBE, "%s:smrepB %p", __func__, ep);
1110 
1111 	mpalen = sizeof(*mpa) + plen;
1112 
1113 	if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
1114 
1115 		CTR3(KTR_IW_CXGBE, "%s:smrep1 %p %d", __func__, ep,
1116 		    ep->mpa_attr.version);
1117 		mpalen += sizeof(struct mpa_v2_conn_params);
1118 	}
1119 
1120 	if (mpalen > MHLEN)
1121 		CXGBE_UNIMPLEMENTED(__func__);
1122 
1123 	m = m_gethdr(M_NOWAIT, MT_DATA);
1124 	if (m == NULL) {
1125 
1126 		CTR2(KTR_IW_CXGBE, "%s:smrep2 %p", __func__, ep);
1127 		printf("%s - cannot alloc mbuf!\n", __func__);
1128 		return (-ENOMEM);
1129 	}
1130 
1131 
1132 	mpa = mtod(m, struct mpa_message *);
1133 	m->m_len = mpalen;
1134 	m->m_pkthdr.len = mpalen;
1135 	memset(mpa, 0, sizeof(*mpa));
1136 	memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
1137 	mpa->flags = (ep->mpa_attr.crc_enabled ? MPA_CRC : 0) |
1138 		(markers_enabled ? MPA_MARKERS : 0);
1139 	mpa->revision = ep->mpa_attr.version;
1140 	mpa->private_data_size = htons(plen);
1141 
1142 	if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
1143 
1144 		mpa->flags |= MPA_ENHANCED_RDMA_CONN;
1145 		mpa->private_data_size +=
1146 			htons(sizeof(struct mpa_v2_conn_params));
1147 		mpa_v2_params.ird = htons((u16)ep->ird);
1148 		mpa_v2_params.ord = htons((u16)ep->ord);
1149 		CTR5(KTR_IW_CXGBE, "%s:smrep3 %p %d %d %d", __func__, ep,
1150 		    ep->mpa_attr.version, mpa_v2_params.ird, mpa_v2_params.ord);
1151 
1152 		if (peer2peer && (ep->mpa_attr.p2p_type !=
1153 			FW_RI_INIT_P2PTYPE_DISABLED)) {
1154 
1155 			mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL);
1156 
1157 			if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE) {
1158 
1159 				mpa_v2_params.ord |=
1160 					htons(MPA_V2_RDMA_WRITE_RTR);
1161 				CTR5(KTR_IW_CXGBE, "%s:smrep4 %p %d %d %d",
1162 				    __func__, ep, p2p_type, mpa_v2_params.ird,
1163 				    mpa_v2_params.ord);
1164 			}
1165 			else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ) {
1166 
1167 				mpa_v2_params.ord |=
1168 					htons(MPA_V2_RDMA_READ_RTR);
1169 				CTR5(KTR_IW_CXGBE, "%s:smrep5 %p %d %d %d",
1170 				    __func__, ep, p2p_type, mpa_v2_params.ird,
1171 				    mpa_v2_params.ord);
1172 			}
1173 		}
1174 
1175 		memcpy(mpa->private_data, &mpa_v2_params,
1176 			sizeof(struct mpa_v2_conn_params));
1177 
1178 		if (ep->plen)
1179 			memcpy(mpa->private_data +
1180 				sizeof(struct mpa_v2_conn_params), pdata, plen);
1181 	} else
1182 		if (plen)
1183 			memcpy(mpa->private_data, pdata, plen);
1184 
1185 	state_set(&ep->com, MPA_REP_SENT);
1186 	ep->snd_seq += mpalen;
1187 	err = sosend(ep->com.so, NULL, NULL, m, NULL, MSG_DONTWAIT,
1188 			ep->com.thread);
1189 	CTR3(KTR_IW_CXGBE, "%s:smrepE %p %d", __func__, ep, err);
1190 	return err;
1191 }
1192 
1193 
1194 
1195 static void close_complete_upcall(struct c4iw_ep *ep, int status)
1196 {
1197 	struct iw_cm_event event;
1198 
1199 	CTR2(KTR_IW_CXGBE, "%s:ccuB %p", __func__, ep);
1200 	memset(&event, 0, sizeof(event));
1201 	event.event = IW_CM_EVENT_CLOSE;
1202 	event.status = status;
1203 
1204 	if (ep->com.cm_id) {
1205 
1206 		CTR2(KTR_IW_CXGBE, "%s:ccu1 %1", __func__, ep);
1207 		ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1208 		ep->com.cm_id->rem_ref(ep->com.cm_id);
1209 		ep->com.cm_id = NULL;
1210 		ep->com.qp = NULL;
1211 		set_bit(CLOSE_UPCALL, &ep->com.history);
1212 	}
1213 	CTR2(KTR_IW_CXGBE, "%s:ccuE %p", __func__, ep);
1214 }
1215 
1216 static int abort_connection(struct c4iw_ep *ep)
1217 {
1218 	int err;
1219 
1220 	CTR2(KTR_IW_CXGBE, "%s:abB %p", __func__, ep);
1221 	close_complete_upcall(ep, -ECONNRESET);
1222 	state_set(&ep->com, ABORTING);
1223 	abort_socket(ep);
1224 	err = close_socket(&ep->com, 0);
1225 	set_bit(ABORT_CONN, &ep->com.history);
1226 	CTR2(KTR_IW_CXGBE, "%s:abE %p", __func__, ep);
1227 	return err;
1228 }
1229 
1230 static void peer_close_upcall(struct c4iw_ep *ep)
1231 {
1232 	struct iw_cm_event event;
1233 
1234 	CTR2(KTR_IW_CXGBE, "%s:pcuB %p", __func__, ep);
1235 	memset(&event, 0, sizeof(event));
1236 	event.event = IW_CM_EVENT_DISCONNECT;
1237 
1238 	if (ep->com.cm_id) {
1239 
1240 		CTR2(KTR_IW_CXGBE, "%s:pcu1 %p", __func__, ep);
1241 		ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1242 		set_bit(DISCONN_UPCALL, &ep->com.history);
1243 	}
1244 	CTR2(KTR_IW_CXGBE, "%s:pcuE %p", __func__, ep);
1245 }
1246 
1247 static void peer_abort_upcall(struct c4iw_ep *ep)
1248 {
1249 	struct iw_cm_event event;
1250 
1251 	CTR2(KTR_IW_CXGBE, "%s:pauB %p", __func__, ep);
1252 	memset(&event, 0, sizeof(event));
1253 	event.event = IW_CM_EVENT_CLOSE;
1254 	event.status = -ECONNRESET;
1255 
1256 	if (ep->com.cm_id) {
1257 
1258 		CTR2(KTR_IW_CXGBE, "%s:pau1 %p", __func__, ep);
1259 		ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1260 		ep->com.cm_id->rem_ref(ep->com.cm_id);
1261 		ep->com.cm_id = NULL;
1262 		ep->com.qp = NULL;
1263 		set_bit(ABORT_UPCALL, &ep->com.history);
1264 	}
1265 	CTR2(KTR_IW_CXGBE, "%s:pauE %p", __func__, ep);
1266 }
1267 
1268 static void connect_reply_upcall(struct c4iw_ep *ep, int status)
1269 {
1270 	struct iw_cm_event event;
1271 
1272 	CTR3(KTR_IW_CXGBE, "%s:cruB %p", __func__, ep, status);
1273 	memset(&event, 0, sizeof(event));
1274 	event.event = IW_CM_EVENT_CONNECT_REPLY;
1275 	event.status = (status ==-ECONNABORTED)?-ECONNRESET: status;
1276 	event.local_addr = ep->com.local_addr;
1277 	event.remote_addr = ep->com.remote_addr;
1278 
1279 	if ((status == 0) || (status == -ECONNREFUSED)) {
1280 
1281 		if (!ep->tried_with_mpa_v1) {
1282 
1283 			CTR2(KTR_IW_CXGBE, "%s:cru1 %p", __func__, ep);
1284 			/* this means MPA_v2 is used */
1285 			event.private_data_len = ep->plen -
1286 				sizeof(struct mpa_v2_conn_params);
1287 			event.private_data = ep->mpa_pkt +
1288 				sizeof(struct mpa_message) +
1289 				sizeof(struct mpa_v2_conn_params);
1290 		} else {
1291 
1292 			CTR2(KTR_IW_CXGBE, "%s:cru2 %p", __func__, ep);
1293 			/* this means MPA_v1 is used */
1294 			event.private_data_len = ep->plen;
1295 			event.private_data = ep->mpa_pkt +
1296 				sizeof(struct mpa_message);
1297 		}
1298 	}
1299 
1300 	if (ep->com.cm_id) {
1301 
1302 		CTR2(KTR_IW_CXGBE, "%s:cru3 %p", __func__, ep);
1303 		set_bit(CONN_RPL_UPCALL, &ep->com.history);
1304 		ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1305 	}
1306 
1307 	if(status == -ECONNABORTED) {
1308 
1309 		CTR3(KTR_IW_CXGBE, "%s:cruE %p %d", __func__, ep, status);
1310 		return;
1311 	}
1312 
1313 	if (status < 0) {
1314 
1315 		CTR3(KTR_IW_CXGBE, "%s:cru4 %p %d", __func__, ep, status);
1316 		ep->com.cm_id->rem_ref(ep->com.cm_id);
1317 		ep->com.cm_id = NULL;
1318 		ep->com.qp = NULL;
1319 	}
1320 
1321 	CTR2(KTR_IW_CXGBE, "%s:cruE %p", __func__, ep);
1322 }
1323 
1324 static void connect_request_upcall(struct c4iw_ep *ep)
1325 {
1326 	struct iw_cm_event event;
1327 
1328 	CTR3(KTR_IW_CXGBE, "%s: ep %p, mpa_v1 %d", __func__, ep,
1329 	    ep->tried_with_mpa_v1);
1330 
1331 	memset(&event, 0, sizeof(event));
1332 	event.event = IW_CM_EVENT_CONNECT_REQUEST;
1333 	event.local_addr = ep->com.local_addr;
1334 	event.remote_addr = ep->com.remote_addr;
1335 	event.provider_data = ep;
1336 	event.so = ep->com.so;
1337 
1338 	if (!ep->tried_with_mpa_v1) {
1339 		/* this means MPA_v2 is used */
1340 #ifdef IW_CM_MPAV2
1341 		event.ord = ep->ord;
1342 		event.ird = ep->ird;
1343 #endif
1344 		event.private_data_len = ep->plen -
1345 			sizeof(struct mpa_v2_conn_params);
1346 		event.private_data = ep->mpa_pkt + sizeof(struct mpa_message) +
1347 			sizeof(struct mpa_v2_conn_params);
1348 	} else {
1349 
1350 		/* this means MPA_v1 is used. Send max supported */
1351 #ifdef IW_CM_MPAV2
1352 		event.ord = c4iw_max_read_depth;
1353 		event.ird = c4iw_max_read_depth;
1354 #endif
1355 		event.private_data_len = ep->plen;
1356 		event.private_data = ep->mpa_pkt + sizeof(struct mpa_message);
1357 	}
1358 
1359 	c4iw_get_ep(&ep->com);
1360 	ep->parent_ep->com.cm_id->event_handler(ep->parent_ep->com.cm_id,
1361 	    &event);
1362 	set_bit(CONNREQ_UPCALL, &ep->com.history);
1363 	c4iw_put_ep(&ep->parent_ep->com);
1364 }
1365 
1366 static void established_upcall(struct c4iw_ep *ep)
1367 {
1368 	struct iw_cm_event event;
1369 
1370 	CTR2(KTR_IW_CXGBE, "%s:euB %p", __func__, ep);
1371 	memset(&event, 0, sizeof(event));
1372 	event.event = IW_CM_EVENT_ESTABLISHED;
1373 #ifdef IW_CM_MPAV2
1374 	event.ird = ep->ird;
1375 	event.ord = ep->ord;
1376 #endif
1377 	if (ep->com.cm_id) {
1378 
1379 		CTR2(KTR_IW_CXGBE, "%s:eu1 %p", __func__, ep);
1380 		ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1381 		set_bit(ESTAB_UPCALL, &ep->com.history);
1382 	}
1383 	CTR2(KTR_IW_CXGBE, "%s:euE %p", __func__, ep);
1384 }
1385 
1386 
1387 
1388 static void process_mpa_reply(struct c4iw_ep *ep)
1389 {
1390 	struct mpa_message *mpa;
1391 	struct mpa_v2_conn_params *mpa_v2_params;
1392 	u16 plen;
1393 	u16 resp_ird, resp_ord;
1394 	u8 rtr_mismatch = 0, insuff_ird = 0;
1395 	struct c4iw_qp_attributes attrs;
1396 	enum c4iw_qp_attr_mask mask;
1397 	int err;
1398 	struct mbuf *top, *m;
1399 	int flags = MSG_DONTWAIT;
1400 	struct uio uio;
1401 
1402 	CTR2(KTR_IW_CXGBE, "%s:pmrB %p", __func__, ep);
1403 
1404 	/*
1405 	 * Stop mpa timer.  If it expired, then the state has
1406 	 * changed and we bail since ep_timeout already aborted
1407 	 * the connection.
1408 	 */
1409 	STOP_EP_TIMER(ep);
1410 	if (state_read(&ep->com) != MPA_REQ_SENT)
1411 		return;
1412 
1413 	uio.uio_resid = 1000000;
1414 	uio.uio_td = ep->com.thread;
1415 	err = soreceive(ep->com.so, NULL, &uio, &top, NULL, &flags);
1416 
1417 	if (err) {
1418 
1419 		if (err == EWOULDBLOCK) {
1420 
1421 			CTR2(KTR_IW_CXGBE, "%s:pmr1 %p", __func__, ep);
1422 			START_EP_TIMER(ep);
1423 			return;
1424 		}
1425 		err = -err;
1426 		CTR2(KTR_IW_CXGBE, "%s:pmr2 %p", __func__, ep);
1427 		goto err;
1428 	}
1429 
1430 	if (ep->com.so->so_rcv.sb_mb) {
1431 
1432 		CTR2(KTR_IW_CXGBE, "%s:pmr3 %p", __func__, ep);
1433 		printf("%s data after soreceive called! so %p sb_mb %p top %p\n",
1434 		       __func__, ep->com.so, ep->com.so->so_rcv.sb_mb, top);
1435 	}
1436 
1437 	m = top;
1438 
1439 	do {
1440 
1441 		CTR2(KTR_IW_CXGBE, "%s:pmr4 %p", __func__, ep);
1442 		/*
1443 		 * If we get more than the supported amount of private data
1444 		 * then we must fail this connection.
1445 		 */
1446 		if (ep->mpa_pkt_len + m->m_len > sizeof(ep->mpa_pkt)) {
1447 
1448 			CTR3(KTR_IW_CXGBE, "%s:pmr5 %p %d", __func__, ep,
1449 			    ep->mpa_pkt_len + m->m_len);
1450 			err = (-EINVAL);
1451 			goto err;
1452 		}
1453 
1454 		/*
1455 		 * copy the new data into our accumulation buffer.
1456 		 */
1457 		m_copydata(m, 0, m->m_len, &(ep->mpa_pkt[ep->mpa_pkt_len]));
1458 		ep->mpa_pkt_len += m->m_len;
1459 		if (!m->m_next)
1460 			m = m->m_nextpkt;
1461 		else
1462 			m = m->m_next;
1463 	} while (m);
1464 
1465 	m_freem(top);
1466 	/*
1467 	 * if we don't even have the mpa message, then bail.
1468 	 */
1469 	if (ep->mpa_pkt_len < sizeof(*mpa))
1470 		return;
1471 	mpa = (struct mpa_message *) ep->mpa_pkt;
1472 
1473 	/* Validate MPA header. */
1474 	if (mpa->revision > mpa_rev) {
1475 
1476 		CTR4(KTR_IW_CXGBE, "%s:pmr6 %p %d %d", __func__, ep,
1477 		    mpa->revision, mpa_rev);
1478 		printk(KERN_ERR MOD "%s MPA version mismatch. Local = %d, "
1479 				" Received = %d\n", __func__, mpa_rev, mpa->revision);
1480 		err = -EPROTO;
1481 		goto err;
1482 	}
1483 
1484 	if (memcmp(mpa->key, MPA_KEY_REP, sizeof(mpa->key))) {
1485 
1486 		CTR2(KTR_IW_CXGBE, "%s:pmr7 %p", __func__, ep);
1487 		err = -EPROTO;
1488 		goto err;
1489 	}
1490 
1491 	plen = ntohs(mpa->private_data_size);
1492 
1493 	/*
1494 	 * Fail if there's too much private data.
1495 	 */
1496 	if (plen > MPA_MAX_PRIVATE_DATA) {
1497 
1498 		CTR2(KTR_IW_CXGBE, "%s:pmr8 %p", __func__, ep);
1499 		err = -EPROTO;
1500 		goto err;
1501 	}
1502 
1503 	/*
1504 	 * If plen does not account for pkt size
1505 	 */
1506 	if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
1507 
1508 		CTR2(KTR_IW_CXGBE, "%s:pmr9 %p", __func__, ep);
1509 		err = -EPROTO;
1510 		goto err;
1511 	}
1512 
1513 	ep->plen = (u8) plen;
1514 
1515 	/*
1516 	 * If we don't have all the pdata yet, then bail.
1517 	 * We'll continue process when more data arrives.
1518 	 */
1519 	if (ep->mpa_pkt_len < (sizeof(*mpa) + plen)) {
1520 
1521 		CTR2(KTR_IW_CXGBE, "%s:pmra %p", __func__, ep);
1522 		return;
1523 	}
1524 
1525 	if (mpa->flags & MPA_REJECT) {
1526 
1527 		CTR2(KTR_IW_CXGBE, "%s:pmrb %p", __func__, ep);
1528 		err = -ECONNREFUSED;
1529 		goto err;
1530 	}
1531 
1532 	/*
1533 	 * If we get here we have accumulated the entire mpa
1534 	 * start reply message including private data. And
1535 	 * the MPA header is valid.
1536 	 */
1537 	state_set(&ep->com, FPDU_MODE);
1538 	ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1539 	ep->mpa_attr.recv_marker_enabled = markers_enabled;
1540 	ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1541 	ep->mpa_attr.version = mpa->revision;
1542 	ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1543 
1544 	if (mpa->revision == 2) {
1545 
1546 		CTR2(KTR_IW_CXGBE, "%s:pmrc %p", __func__, ep);
1547 		ep->mpa_attr.enhanced_rdma_conn =
1548 			mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
1549 
1550 		if (ep->mpa_attr.enhanced_rdma_conn) {
1551 
1552 			CTR2(KTR_IW_CXGBE, "%s:pmrd %p", __func__, ep);
1553 			mpa_v2_params = (struct mpa_v2_conn_params *)
1554 				(ep->mpa_pkt + sizeof(*mpa));
1555 			resp_ird = ntohs(mpa_v2_params->ird) &
1556 				MPA_V2_IRD_ORD_MASK;
1557 			resp_ord = ntohs(mpa_v2_params->ord) &
1558 				MPA_V2_IRD_ORD_MASK;
1559 
1560 			/*
1561 			 * This is a double-check. Ideally, below checks are
1562 			 * not required since ird/ord stuff has been taken
1563 			 * care of in c4iw_accept_cr
1564 			 */
1565 			if ((ep->ird < resp_ord) || (ep->ord > resp_ird)) {
1566 
1567 				CTR2(KTR_IW_CXGBE, "%s:pmre %p", __func__, ep);
1568 				err = -ENOMEM;
1569 				ep->ird = resp_ord;
1570 				ep->ord = resp_ird;
1571 				insuff_ird = 1;
1572 			}
1573 
1574 			if (ntohs(mpa_v2_params->ird) &
1575 				MPA_V2_PEER2PEER_MODEL) {
1576 
1577 				CTR2(KTR_IW_CXGBE, "%s:pmrf %p", __func__, ep);
1578 				if (ntohs(mpa_v2_params->ord) &
1579 					MPA_V2_RDMA_WRITE_RTR) {
1580 
1581 					CTR2(KTR_IW_CXGBE, "%s:pmrg %p", __func__, ep);
1582 					ep->mpa_attr.p2p_type =
1583 						FW_RI_INIT_P2PTYPE_RDMA_WRITE;
1584 				}
1585 				else if (ntohs(mpa_v2_params->ord) &
1586 					MPA_V2_RDMA_READ_RTR) {
1587 
1588 					CTR2(KTR_IW_CXGBE, "%s:pmrh %p", __func__, ep);
1589 					ep->mpa_attr.p2p_type =
1590 						FW_RI_INIT_P2PTYPE_READ_REQ;
1591 				}
1592 			}
1593 		}
1594 	} else {
1595 
1596 		CTR2(KTR_IW_CXGBE, "%s:pmri %p", __func__, ep);
1597 
1598 		if (mpa->revision == 1) {
1599 
1600 			CTR2(KTR_IW_CXGBE, "%s:pmrj %p", __func__, ep);
1601 
1602 			if (peer2peer) {
1603 
1604 				CTR2(KTR_IW_CXGBE, "%s:pmrk %p", __func__, ep);
1605 				ep->mpa_attr.p2p_type = p2p_type;
1606 			}
1607 		}
1608 	}
1609 
1610 	if (set_tcpinfo(ep)) {
1611 
1612 		CTR2(KTR_IW_CXGBE, "%s:pmrl %p", __func__, ep);
1613 		printf("%s set_tcpinfo error\n", __func__);
1614 		goto err;
1615 	}
1616 
1617 	CTR6(KTR_IW_CXGBE, "%s - crc_enabled = %d, recv_marker_enabled = %d, "
1618 	    "xmit_marker_enabled = %d, version = %d p2p_type = %d", __func__,
1619 	    ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
1620 	    ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version,
1621 	    ep->mpa_attr.p2p_type);
1622 
1623 	/*
1624 	 * If responder's RTR does not match with that of initiator, assign
1625 	 * FW_RI_INIT_P2PTYPE_DISABLED in mpa attributes so that RTR is not
1626 	 * generated when moving QP to RTS state.
1627 	 * A TERM message will be sent after QP has moved to RTS state
1628 	 */
1629 	if ((ep->mpa_attr.version == 2) && peer2peer &&
1630 		(ep->mpa_attr.p2p_type != p2p_type)) {
1631 
1632 		CTR2(KTR_IW_CXGBE, "%s:pmrm %p", __func__, ep);
1633 		ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1634 		rtr_mismatch = 1;
1635 	}
1636 
1637 
1638 	//ep->ofld_txq = TOEPCB(ep->com.so)->ofld_txq;
1639 	attrs.mpa_attr = ep->mpa_attr;
1640 	attrs.max_ird = ep->ird;
1641 	attrs.max_ord = ep->ord;
1642 	attrs.llp_stream_handle = ep;
1643 	attrs.next_state = C4IW_QP_STATE_RTS;
1644 
1645 	mask = C4IW_QP_ATTR_NEXT_STATE |
1646 		C4IW_QP_ATTR_LLP_STREAM_HANDLE | C4IW_QP_ATTR_MPA_ATTR |
1647 		C4IW_QP_ATTR_MAX_IRD | C4IW_QP_ATTR_MAX_ORD;
1648 
1649 	/* bind QP and TID with INIT_WR */
1650 	err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp, mask, &attrs, 1);
1651 
1652 	if (err) {
1653 
1654 		CTR2(KTR_IW_CXGBE, "%s:pmrn %p", __func__, ep);
1655 		goto err;
1656 	}
1657 
1658 	/*
1659 	 * If responder's RTR requirement did not match with what initiator
1660 	 * supports, generate TERM message
1661 	 */
1662 	if (rtr_mismatch) {
1663 
1664 		CTR2(KTR_IW_CXGBE, "%s:pmro %p", __func__, ep);
1665 		printk(KERN_ERR "%s: RTR mismatch, sending TERM\n", __func__);
1666 		attrs.layer_etype = LAYER_MPA | DDP_LLP;
1667 		attrs.ecode = MPA_NOMATCH_RTR;
1668 		attrs.next_state = C4IW_QP_STATE_TERMINATE;
1669 		err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1670 			C4IW_QP_ATTR_NEXT_STATE, &attrs, 0);
1671 		err = -ENOMEM;
1672 		goto out;
1673 	}
1674 
1675 	/*
1676 	 * Generate TERM if initiator IRD is not sufficient for responder
1677 	 * provided ORD. Currently, we do the same behaviour even when
1678 	 * responder provided IRD is also not sufficient as regards to
1679 	 * initiator ORD.
1680 	 */
1681 	if (insuff_ird) {
1682 
1683 		CTR2(KTR_IW_CXGBE, "%s:pmrp %p", __func__, ep);
1684 		printk(KERN_ERR "%s: Insufficient IRD, sending TERM\n",
1685 				__func__);
1686 		attrs.layer_etype = LAYER_MPA | DDP_LLP;
1687 		attrs.ecode = MPA_INSUFF_IRD;
1688 		attrs.next_state = C4IW_QP_STATE_TERMINATE;
1689 		err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1690 			C4IW_QP_ATTR_NEXT_STATE, &attrs, 0);
1691 		err = -ENOMEM;
1692 		goto out;
1693 	}
1694 	goto out;
1695 err:
1696 	state_set(&ep->com, ABORTING);
1697 	abort_connection(ep);
1698 out:
1699 	connect_reply_upcall(ep, err);
1700 	CTR2(KTR_IW_CXGBE, "%s:pmrE %p", __func__, ep);
1701 	return;
1702 }
1703 
1704 static void
1705 process_mpa_request(struct c4iw_ep *ep)
1706 {
1707 	struct mpa_message *mpa;
1708 	u16 plen;
1709 	int flags = MSG_DONTWAIT;
1710 	int rc;
1711 	struct iovec iov;
1712 	struct uio uio;
1713 	enum c4iw_ep_state state = state_read(&ep->com);
1714 
1715 	CTR3(KTR_IW_CXGBE, "%s: ep %p, state %s", __func__, ep, states[state]);
1716 
1717 	if (state != MPA_REQ_WAIT)
1718 		return;
1719 
1720 	iov.iov_base = &ep->mpa_pkt[ep->mpa_pkt_len];
1721 	iov.iov_len = sizeof(ep->mpa_pkt) - ep->mpa_pkt_len;
1722 	uio.uio_iov = &iov;
1723 	uio.uio_iovcnt = 1;
1724 	uio.uio_offset = 0;
1725 	uio.uio_resid = sizeof(ep->mpa_pkt) - ep->mpa_pkt_len;
1726 	uio.uio_segflg = UIO_SYSSPACE;
1727 	uio.uio_rw = UIO_READ;
1728 	uio.uio_td = NULL; /* uio.uio_td = ep->com.thread; */
1729 
1730 	rc = soreceive(ep->com.so, NULL, &uio, NULL, NULL, &flags);
1731 	if (rc == EAGAIN)
1732 		return;
1733 	else if (rc) {
1734 abort:
1735 		STOP_EP_TIMER(ep);
1736 		abort_connection(ep);
1737 		return;
1738 	}
1739 	KASSERT(uio.uio_offset > 0, ("%s: sorecieve on so %p read no data",
1740 	    __func__, ep->com.so));
1741 	ep->mpa_pkt_len += uio.uio_offset;
1742 
1743 	/*
1744 	 * If we get more than the supported amount of private data then we must
1745 	 * fail this connection.  XXX: check so_rcv->sb_cc, or peek with another
1746 	 * soreceive, or increase the size of mpa_pkt by 1 and abort if the last
1747 	 * byte is filled by the soreceive above.
1748 	 */
1749 
1750 	/* Don't even have the MPA message.  Wait for more data to arrive. */
1751 	if (ep->mpa_pkt_len < sizeof(*mpa))
1752 		return;
1753 	mpa = (struct mpa_message *) ep->mpa_pkt;
1754 
1755 	/*
1756 	 * Validate MPA Header.
1757 	 */
1758 	if (mpa->revision > mpa_rev) {
1759 		log(LOG_ERR, "%s: MPA version mismatch. Local = %d,"
1760 		    " Received = %d\n", __func__, mpa_rev, mpa->revision);
1761 		goto abort;
1762 	}
1763 
1764 	if (memcmp(mpa->key, MPA_KEY_REQ, sizeof(mpa->key)))
1765 		goto abort;
1766 
1767 	/*
1768 	 * Fail if there's too much private data.
1769 	 */
1770 	plen = ntohs(mpa->private_data_size);
1771 	if (plen > MPA_MAX_PRIVATE_DATA)
1772 		goto abort;
1773 
1774 	/*
1775 	 * If plen does not account for pkt size
1776 	 */
1777 	if (ep->mpa_pkt_len > (sizeof(*mpa) + plen))
1778 		goto abort;
1779 
1780 	ep->plen = (u8) plen;
1781 
1782 	/*
1783 	 * If we don't have all the pdata yet, then bail.
1784 	 */
1785 	if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
1786 		return;
1787 
1788 	/*
1789 	 * If we get here we have accumulated the entire mpa
1790 	 * start reply message including private data.
1791 	 */
1792 	ep->mpa_attr.initiator = 0;
1793 	ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1794 	ep->mpa_attr.recv_marker_enabled = markers_enabled;
1795 	ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1796 	ep->mpa_attr.version = mpa->revision;
1797 	if (mpa->revision == 1)
1798 		ep->tried_with_mpa_v1 = 1;
1799 	ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1800 
1801 	if (mpa->revision == 2) {
1802 		ep->mpa_attr.enhanced_rdma_conn =
1803 		    mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
1804 		if (ep->mpa_attr.enhanced_rdma_conn) {
1805 			struct mpa_v2_conn_params *mpa_v2_params;
1806 			u16 ird, ord;
1807 
1808 			mpa_v2_params = (void *)&ep->mpa_pkt[sizeof(*mpa)];
1809 			ird = ntohs(mpa_v2_params->ird);
1810 			ord = ntohs(mpa_v2_params->ord);
1811 
1812 			ep->ird = ird & MPA_V2_IRD_ORD_MASK;
1813 			ep->ord = ord & MPA_V2_IRD_ORD_MASK;
1814 			if (ird & MPA_V2_PEER2PEER_MODEL && peer2peer) {
1815 				if (ord & MPA_V2_RDMA_WRITE_RTR) {
1816 					ep->mpa_attr.p2p_type =
1817 					    FW_RI_INIT_P2PTYPE_RDMA_WRITE;
1818 				} else if (ord & MPA_V2_RDMA_READ_RTR) {
1819 					ep->mpa_attr.p2p_type =
1820 					    FW_RI_INIT_P2PTYPE_READ_REQ;
1821 				}
1822 			}
1823 		}
1824 	} else if (mpa->revision == 1 && peer2peer)
1825 		ep->mpa_attr.p2p_type = p2p_type;
1826 
1827 	if (set_tcpinfo(ep))
1828 		goto abort;
1829 
1830 	CTR5(KTR_IW_CXGBE, "%s: crc_enabled = %d, recv_marker_enabled = %d, "
1831 	    "xmit_marker_enabled = %d, version = %d", __func__,
1832 	    ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
1833 	    ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version);
1834 
1835 	state_set(&ep->com, MPA_REQ_RCVD);
1836 	STOP_EP_TIMER(ep);
1837 
1838 	/* drive upcall */
1839 	mutex_lock(&ep->parent_ep->com.mutex);
1840 	if (ep->parent_ep->com.state != DEAD)
1841 		connect_request_upcall(ep);
1842 	else
1843 		abort_connection(ep);
1844 	mutex_unlock(&ep->parent_ep->com.mutex);
1845 }
1846 
1847 /*
1848  * Upcall from the adapter indicating data has been transmitted.
1849  * For us its just the single MPA request or reply.  We can now free
1850  * the skb holding the mpa message.
1851  */
1852 int c4iw_reject_cr(struct iw_cm_id *cm_id, const void *pdata, u8 pdata_len)
1853 {
1854 	int err;
1855 	struct c4iw_ep *ep = to_ep(cm_id);
1856 	CTR2(KTR_IW_CXGBE, "%s:crcB %p", __func__, ep);
1857 
1858 	if (state_read(&ep->com) == DEAD) {
1859 
1860 		CTR2(KTR_IW_CXGBE, "%s:crc1 %p", __func__, ep);
1861 		c4iw_put_ep(&ep->com);
1862 		return -ECONNRESET;
1863 	}
1864 	set_bit(ULP_REJECT, &ep->com.history);
1865 	BUG_ON(state_read(&ep->com) != MPA_REQ_RCVD);
1866 
1867 	if (mpa_rev == 0) {
1868 
1869 		CTR2(KTR_IW_CXGBE, "%s:crc2 %p", __func__, ep);
1870 		abort_connection(ep);
1871 	}
1872 	else {
1873 
1874 		CTR2(KTR_IW_CXGBE, "%s:crc3 %p", __func__, ep);
1875 		err = send_mpa_reject(ep, pdata, pdata_len);
1876 		err = soshutdown(ep->com.so, 3);
1877 	}
1878 	c4iw_put_ep(&ep->com);
1879 	CTR2(KTR_IW_CXGBE, "%s:crc4 %p", __func__, ep);
1880 	return 0;
1881 }
1882 
1883 int c4iw_accept_cr(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
1884 {
1885 	int err;
1886 	struct c4iw_qp_attributes attrs;
1887 	enum c4iw_qp_attr_mask mask;
1888 	struct c4iw_ep *ep = to_ep(cm_id);
1889 	struct c4iw_dev *h = to_c4iw_dev(cm_id->device);
1890 	struct c4iw_qp *qp = get_qhp(h, conn_param->qpn);
1891 
1892 	CTR2(KTR_IW_CXGBE, "%s:cacB %p", __func__, ep);
1893 
1894 	if (state_read(&ep->com) == DEAD) {
1895 
1896 		CTR2(KTR_IW_CXGBE, "%s:cac1 %p", __func__, ep);
1897 		err = -ECONNRESET;
1898 		goto err;
1899 	}
1900 
1901 	BUG_ON(state_read(&ep->com) != MPA_REQ_RCVD);
1902 	BUG_ON(!qp);
1903 
1904 	set_bit(ULP_ACCEPT, &ep->com.history);
1905 
1906 	if ((conn_param->ord > c4iw_max_read_depth) ||
1907 		(conn_param->ird > c4iw_max_read_depth)) {
1908 
1909 		CTR2(KTR_IW_CXGBE, "%s:cac2 %p", __func__, ep);
1910 		abort_connection(ep);
1911 		err = -EINVAL;
1912 		goto err;
1913 	}
1914 
1915 	if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
1916 
1917 		CTR2(KTR_IW_CXGBE, "%s:cac3 %p", __func__, ep);
1918 
1919 		if (conn_param->ord > ep->ird) {
1920 
1921 			CTR2(KTR_IW_CXGBE, "%s:cac4 %p", __func__, ep);
1922 			ep->ird = conn_param->ird;
1923 			ep->ord = conn_param->ord;
1924 			send_mpa_reject(ep, conn_param->private_data,
1925 					conn_param->private_data_len);
1926 			abort_connection(ep);
1927 			err = -ENOMEM;
1928 			goto err;
1929 		}
1930 
1931 		if (conn_param->ird > ep->ord) {
1932 
1933 			CTR2(KTR_IW_CXGBE, "%s:cac5 %p", __func__, ep);
1934 
1935 			if (!ep->ord) {
1936 
1937 				CTR2(KTR_IW_CXGBE, "%s:cac6 %p", __func__, ep);
1938 				conn_param->ird = 1;
1939 			}
1940 			else {
1941 				CTR2(KTR_IW_CXGBE, "%s:cac7 %p", __func__, ep);
1942 				abort_connection(ep);
1943 				err = -ENOMEM;
1944 				goto err;
1945 			}
1946 		}
1947 
1948 	}
1949 	ep->ird = conn_param->ird;
1950 	ep->ord = conn_param->ord;
1951 
1952 	if (ep->mpa_attr.version != 2) {
1953 
1954 		CTR2(KTR_IW_CXGBE, "%s:cac8 %p", __func__, ep);
1955 
1956 		if (peer2peer && ep->ird == 0) {
1957 
1958 			CTR2(KTR_IW_CXGBE, "%s:cac9 %p", __func__, ep);
1959 			ep->ird = 1;
1960 		}
1961 	}
1962 
1963 
1964 	cm_id->add_ref(cm_id);
1965 	ep->com.cm_id = cm_id;
1966 	ep->com.qp = qp;
1967 	//ep->ofld_txq = TOEPCB(ep->com.so)->ofld_txq;
1968 
1969 	/* bind QP to EP and move to RTS */
1970 	attrs.mpa_attr = ep->mpa_attr;
1971 	attrs.max_ird = ep->ird;
1972 	attrs.max_ord = ep->ord;
1973 	attrs.llp_stream_handle = ep;
1974 	attrs.next_state = C4IW_QP_STATE_RTS;
1975 
1976 	/* bind QP and TID with INIT_WR */
1977 	mask = C4IW_QP_ATTR_NEXT_STATE |
1978 		C4IW_QP_ATTR_LLP_STREAM_HANDLE |
1979 		C4IW_QP_ATTR_MPA_ATTR |
1980 		C4IW_QP_ATTR_MAX_IRD |
1981 		C4IW_QP_ATTR_MAX_ORD;
1982 
1983 	err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp, mask, &attrs, 1);
1984 
1985 	if (err) {
1986 
1987 		CTR2(KTR_IW_CXGBE, "%s:caca %p", __func__, ep);
1988 		goto err1;
1989 	}
1990 	err = send_mpa_reply(ep, conn_param->private_data,
1991 			conn_param->private_data_len);
1992 
1993 	if (err) {
1994 
1995 		CTR2(KTR_IW_CXGBE, "%s:caca %p", __func__, ep);
1996 		goto err1;
1997 	}
1998 
1999 	state_set(&ep->com, FPDU_MODE);
2000 	established_upcall(ep);
2001 	c4iw_put_ep(&ep->com);
2002 	CTR2(KTR_IW_CXGBE, "%s:cacE %p", __func__, ep);
2003 	return 0;
2004 err1:
2005 	ep->com.cm_id = NULL;
2006 	ep->com.qp = NULL;
2007 	cm_id->rem_ref(cm_id);
2008 err:
2009 	c4iw_put_ep(&ep->com);
2010 	CTR2(KTR_IW_CXGBE, "%s:cacE err %p", __func__, ep);
2011 	return err;
2012 }
2013 
2014 
2015 
2016 int c4iw_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
2017 {
2018 	int err = 0;
2019 	struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
2020 	struct c4iw_ep *ep = NULL;
2021 	struct rtentry *rt;
2022 	struct toedev *tdev;
2023 
2024 	CTR2(KTR_IW_CXGBE, "%s:ccB %p", __func__, cm_id);
2025 
2026 	if ((conn_param->ord > c4iw_max_read_depth) ||
2027 		(conn_param->ird > c4iw_max_read_depth)) {
2028 
2029 		CTR2(KTR_IW_CXGBE, "%s:cc1 %p", __func__, cm_id);
2030 		err = -EINVAL;
2031 		goto out;
2032 	}
2033 	ep = alloc_ep(sizeof(*ep), M_NOWAIT);
2034 
2035 	if (!ep) {
2036 
2037 		CTR2(KTR_IW_CXGBE, "%s:cc2 %p", __func__, cm_id);
2038 		printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__);
2039 		err = -ENOMEM;
2040 		goto out;
2041 	}
2042 	init_timer(&ep->timer);
2043 	ep->plen = conn_param->private_data_len;
2044 
2045 	if (ep->plen) {
2046 
2047 		CTR2(KTR_IW_CXGBE, "%s:cc3 %p", __func__, ep);
2048 		memcpy(ep->mpa_pkt + sizeof(struct mpa_message),
2049 				conn_param->private_data, ep->plen);
2050 	}
2051 	ep->ird = conn_param->ird;
2052 	ep->ord = conn_param->ord;
2053 
2054 	if (peer2peer && ep->ord == 0) {
2055 
2056 		CTR2(KTR_IW_CXGBE, "%s:cc4 %p", __func__, ep);
2057 		ep->ord = 1;
2058 	}
2059 
2060 	cm_id->add_ref(cm_id);
2061 	ep->com.dev = dev;
2062 	ep->com.cm_id = cm_id;
2063 	ep->com.qp = get_qhp(dev, conn_param->qpn);
2064 
2065 	if (!ep->com.qp) {
2066 
2067 		CTR2(KTR_IW_CXGBE, "%s:cc5 %p", __func__, ep);
2068 		err = -EINVAL;
2069 		goto fail2;
2070 	}
2071 	ep->com.thread = curthread;
2072 	ep->com.so = cm_id->so;
2073 
2074 	init_sock(&ep->com);
2075 
2076 	/* find a route */
2077 	rt = find_route(
2078 		cm_id->local_addr.sin_addr.s_addr,
2079 		cm_id->remote_addr.sin_addr.s_addr,
2080 		cm_id->local_addr.sin_port,
2081 		cm_id->remote_addr.sin_port, 0);
2082 
2083 	if (!rt) {
2084 
2085 		CTR2(KTR_IW_CXGBE, "%s:cc7 %p", __func__, ep);
2086 		printk(KERN_ERR MOD "%s - cannot find route.\n", __func__);
2087 		err = -EHOSTUNREACH;
2088 		goto fail2;
2089 	}
2090 
2091 	if (!(rt->rt_ifp->if_capenable & IFCAP_TOE)) {
2092 
2093 		CTR2(KTR_IW_CXGBE, "%s:cc8 %p", __func__, ep);
2094 		printf("%s - interface not TOE capable.\n", __func__);
2095 		close_socket(&ep->com, 0);
2096 		err = -ENOPROTOOPT;
2097 		goto fail3;
2098 	}
2099 	tdev = TOEDEV(rt->rt_ifp);
2100 
2101 	if (tdev == NULL) {
2102 
2103 		CTR2(KTR_IW_CXGBE, "%s:cc9 %p", __func__, ep);
2104 		printf("%s - No toedev for interface.\n", __func__);
2105 		goto fail3;
2106 	}
2107 	RTFREE(rt);
2108 
2109 	state_set(&ep->com, CONNECTING);
2110 	ep->tos = 0;
2111 	ep->com.local_addr = cm_id->local_addr;
2112 	ep->com.remote_addr = cm_id->remote_addr;
2113 	err = soconnect(ep->com.so, (struct sockaddr *)&ep->com.remote_addr,
2114 		ep->com.thread);
2115 
2116 	if (!err) {
2117 		CTR2(KTR_IW_CXGBE, "%s:cca %p", __func__, ep);
2118 		goto out;
2119 	} else {
2120 		close_socket(&ep->com, 0);
2121 		goto fail2;
2122 	}
2123 
2124 fail3:
2125 	CTR2(KTR_IW_CXGBE, "%s:ccb %p", __func__, ep);
2126 	RTFREE(rt);
2127 fail2:
2128 	cm_id->rem_ref(cm_id);
2129 	c4iw_put_ep(&ep->com);
2130 out:
2131 	CTR2(KTR_IW_CXGBE, "%s:ccE %p", __func__, ep);
2132 	return err;
2133 }
2134 
2135 /*
2136  * iwcm->create_listen.  Returns -errno on failure.
2137  */
2138 int
2139 c4iw_create_listen(struct iw_cm_id *cm_id, int backlog)
2140 {
2141 	int rc;
2142 	struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
2143 	struct c4iw_listen_ep *ep;
2144 	struct socket *so = cm_id->so;
2145 
2146 	ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
2147 	CTR5(KTR_IW_CXGBE, "%s: cm_id %p, lso %p, ep %p, inp %p", __func__,
2148 	    cm_id, so, ep, so->so_pcb);
2149 	if (ep == NULL) {
2150 		log(LOG_ERR, "%s: failed to alloc memory for endpoint\n",
2151 		    __func__);
2152 		rc = ENOMEM;
2153 		goto failed;
2154 	}
2155 
2156 	cm_id->add_ref(cm_id);
2157 	ep->com.cm_id = cm_id;
2158 	ep->com.dev = dev;
2159 	ep->backlog = backlog;
2160 	ep->com.local_addr = cm_id->local_addr;
2161 	ep->com.thread = curthread;
2162 	state_set(&ep->com, LISTEN);
2163 	ep->com.so = so;
2164 	init_sock(&ep->com);
2165 
2166 	rc = solisten(so, ep->backlog, ep->com.thread);
2167 	if (rc != 0) {
2168 		log(LOG_ERR, "%s: failed to start listener: %d\n", __func__,
2169 		    rc);
2170 		close_socket(&ep->com, 0);
2171 		cm_id->rem_ref(cm_id);
2172 		c4iw_put_ep(&ep->com);
2173 		goto failed;
2174 	}
2175 
2176 	cm_id->provider_data = ep;
2177 	return (0);
2178 
2179 failed:
2180 	CTR3(KTR_IW_CXGBE, "%s: cm_id %p, FAILED (%d)", __func__, cm_id, rc);
2181 	return (-rc);
2182 }
2183 
2184 int
2185 c4iw_destroy_listen(struct iw_cm_id *cm_id)
2186 {
2187 	int rc;
2188 	struct c4iw_listen_ep *ep = to_listen_ep(cm_id);
2189 
2190 	CTR4(KTR_IW_CXGBE, "%s: cm_id %p, so %p, inp %p", __func__, cm_id,
2191 	    cm_id->so, cm_id->so->so_pcb);
2192 
2193 	state_set(&ep->com, DEAD);
2194 	rc = close_socket(&ep->com, 0);
2195 	cm_id->rem_ref(cm_id);
2196 	c4iw_put_ep(&ep->com);
2197 
2198 	return (rc);
2199 }
2200 
2201 int c4iw_ep_disconnect(struct c4iw_ep *ep, int abrupt, gfp_t gfp)
2202 {
2203 	int ret = 0;
2204 	int close = 0;
2205 	int fatal = 0;
2206 	struct c4iw_rdev *rdev;
2207 
2208 	mutex_lock(&ep->com.mutex);
2209 
2210 	CTR2(KTR_IW_CXGBE, "%s:cedB %p", __func__, ep);
2211 
2212 	rdev = &ep->com.dev->rdev;
2213 
2214 	if (c4iw_fatal_error(rdev)) {
2215 
2216 		CTR2(KTR_IW_CXGBE, "%s:ced1 %p", __func__, ep);
2217 		fatal = 1;
2218 		close_complete_upcall(ep, -EIO);
2219 		ep->com.state = DEAD;
2220 	}
2221 	CTR3(KTR_IW_CXGBE, "%s:ced2 %p %s", __func__, ep,
2222 	    states[ep->com.state]);
2223 
2224 	switch (ep->com.state) {
2225 
2226 		case MPA_REQ_WAIT:
2227 		case MPA_REQ_SENT:
2228 		case MPA_REQ_RCVD:
2229 		case MPA_REP_SENT:
2230 		case FPDU_MODE:
2231 			close = 1;
2232 			if (abrupt)
2233 				ep->com.state = ABORTING;
2234 			else {
2235 				ep->com.state = CLOSING;
2236 				START_EP_TIMER(ep);
2237 			}
2238 			set_bit(CLOSE_SENT, &ep->com.flags);
2239 			break;
2240 
2241 		case CLOSING:
2242 
2243 			if (!test_and_set_bit(CLOSE_SENT, &ep->com.flags)) {
2244 
2245 				close = 1;
2246 				if (abrupt) {
2247 					STOP_EP_TIMER(ep);
2248 					ep->com.state = ABORTING;
2249 				} else
2250 					ep->com.state = MORIBUND;
2251 			}
2252 			break;
2253 
2254 		case MORIBUND:
2255 		case ABORTING:
2256 		case DEAD:
2257 			CTR3(KTR_IW_CXGBE,
2258 			    "%s ignoring disconnect ep %p state %u", __func__,
2259 			    ep, ep->com.state);
2260 			break;
2261 
2262 		default:
2263 			BUG();
2264 			break;
2265 	}
2266 
2267 	mutex_unlock(&ep->com.mutex);
2268 
2269 	if (close) {
2270 
2271 		CTR2(KTR_IW_CXGBE, "%s:ced3 %p", __func__, ep);
2272 
2273 		if (abrupt) {
2274 
2275 			CTR2(KTR_IW_CXGBE, "%s:ced4 %p", __func__, ep);
2276 			set_bit(EP_DISC_ABORT, &ep->com.history);
2277 			ret = abort_connection(ep);
2278 		} else {
2279 
2280 			CTR2(KTR_IW_CXGBE, "%s:ced5 %p", __func__, ep);
2281 			set_bit(EP_DISC_CLOSE, &ep->com.history);
2282 
2283 			if (!ep->parent_ep)
2284 				__state_set(&ep->com, MORIBUND);
2285 			ret = shutdown_socket(&ep->com);
2286 		}
2287 
2288 		if (ret) {
2289 
2290 			fatal = 1;
2291 		}
2292 	}
2293 
2294 	if (fatal) {
2295 
2296 		release_ep_resources(ep);
2297 		CTR2(KTR_IW_CXGBE, "%s:ced6 %p", __func__, ep);
2298 	}
2299 	CTR2(KTR_IW_CXGBE, "%s:cedE %p", __func__, ep);
2300 	return ret;
2301 }
2302 
2303 #ifdef C4IW_EP_REDIRECT
2304 int c4iw_ep_redirect(void *ctx, struct dst_entry *old, struct dst_entry *new,
2305 		struct l2t_entry *l2t)
2306 {
2307 	struct c4iw_ep *ep = ctx;
2308 
2309 	if (ep->dst != old)
2310 		return 0;
2311 
2312 	PDBG("%s ep %p redirect to dst %p l2t %p\n", __func__, ep, new,
2313 			l2t);
2314 	dst_hold(new);
2315 	cxgb4_l2t_release(ep->l2t);
2316 	ep->l2t = l2t;
2317 	dst_release(old);
2318 	ep->dst = new;
2319 	return 1;
2320 }
2321 #endif
2322 
2323 
2324 
2325 static void ep_timeout(unsigned long arg)
2326 {
2327 	struct c4iw_ep *ep = (struct c4iw_ep *)arg;
2328 	int kickit = 0;
2329 
2330 	CTR2(KTR_IW_CXGBE, "%s:etB %p", __func__, ep);
2331 	spin_lock(&timeout_lock);
2332 
2333 	if (!test_and_set_bit(TIMEOUT, &ep->com.flags)) {
2334 
2335 		list_add_tail(&ep->entry, &timeout_list);
2336 		kickit = 1;
2337 	}
2338 	spin_unlock(&timeout_lock);
2339 
2340 	if (kickit) {
2341 
2342 		CTR2(KTR_IW_CXGBE, "%s:et1 %p", __func__, ep);
2343 		queue_work(c4iw_taskq, &c4iw_task);
2344 	}
2345 	CTR2(KTR_IW_CXGBE, "%s:etE %p", __func__, ep);
2346 }
2347 
2348 static int fw6_wr_rpl(struct adapter *sc, const __be64 *rpl)
2349 {
2350 	uint64_t val = be64toh(*rpl);
2351 	int ret;
2352 	struct c4iw_wr_wait *wr_waitp;
2353 
2354 	ret = (int)((val >> 8) & 0xff);
2355 	wr_waitp = (struct c4iw_wr_wait *)rpl[1];
2356 	CTR3(KTR_IW_CXGBE, "%s wr_waitp %p ret %u", __func__, wr_waitp, ret);
2357 	if (wr_waitp)
2358 		c4iw_wake_up(wr_waitp, ret ? -ret : 0);
2359 
2360 	return (0);
2361 }
2362 
2363 static int fw6_cqe_handler(struct adapter *sc, const __be64 *rpl)
2364 {
2365 	struct t4_cqe cqe =*(const struct t4_cqe *)(&rpl[0]);
2366 
2367 	CTR2(KTR_IW_CXGBE, "%s rpl %p", __func__, rpl);
2368 	c4iw_ev_dispatch(sc->iwarp_softc, &cqe);
2369 
2370 	return (0);
2371 }
2372 
2373 static int terminate(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m)
2374 {
2375 
2376 	struct adapter *sc = iq->adapter;
2377 
2378 	const struct cpl_rdma_terminate *rpl = (const void *)(rss + 1);
2379 	unsigned int tid = GET_TID(rpl);
2380 	struct c4iw_qp_attributes attrs;
2381 	struct toepcb *toep = lookup_tid(sc, tid);
2382 	struct socket *so = inp_inpcbtosocket(toep->inp);
2383 	struct c4iw_ep *ep = so->so_rcv.sb_upcallarg;
2384 
2385 	CTR2(KTR_IW_CXGBE, "%s:tB %p %d", __func__, ep);
2386 
2387 	if (ep && ep->com.qp) {
2388 
2389 		printk(KERN_WARNING MOD "TERM received tid %u qpid %u\n", tid,
2390 				ep->com.qp->wq.sq.qid);
2391 		attrs.next_state = C4IW_QP_STATE_TERMINATE;
2392 		c4iw_modify_qp(ep->com.dev, ep->com.qp, C4IW_QP_ATTR_NEXT_STATE, &attrs,
2393 				1);
2394 	} else
2395 		printk(KERN_WARNING MOD "TERM received tid %u no ep/qp\n", tid);
2396 	CTR2(KTR_IW_CXGBE, "%s:tE %p %d", __func__, ep);
2397 
2398 	return 0;
2399 }
2400 
2401 	void
2402 c4iw_cm_init_cpl(struct adapter *sc)
2403 {
2404 
2405 	t4_register_cpl_handler(sc, CPL_RDMA_TERMINATE, terminate);
2406 	t4_register_fw_msg_handler(sc, FW6_TYPE_WR_RPL, fw6_wr_rpl);
2407 	t4_register_fw_msg_handler(sc, FW6_TYPE_CQE, fw6_cqe_handler);
2408 	t4_register_an_handler(sc, c4iw_ev_handler);
2409 }
2410 
2411 	void
2412 c4iw_cm_term_cpl(struct adapter *sc)
2413 {
2414 
2415 	t4_register_cpl_handler(sc, CPL_RDMA_TERMINATE, NULL);
2416 	t4_register_fw_msg_handler(sc, FW6_TYPE_WR_RPL, NULL);
2417 	t4_register_fw_msg_handler(sc, FW6_TYPE_CQE, NULL);
2418 }
2419 
2420 int __init c4iw_cm_init(void)
2421 {
2422 
2423 	TAILQ_INIT(&req_list);
2424 	spin_lock_init(&req_lock);
2425 	INIT_LIST_HEAD(&timeout_list);
2426 	spin_lock_init(&timeout_lock);
2427 
2428 	INIT_WORK(&c4iw_task, process_req);
2429 
2430 	c4iw_taskq = create_singlethread_workqueue("iw_cxgbe");
2431 	if (!c4iw_taskq)
2432 		return -ENOMEM;
2433 
2434 
2435 	return 0;
2436 }
2437 
2438 void __exit c4iw_cm_term(void)
2439 {
2440 	WARN_ON(!TAILQ_EMPTY(&req_list));
2441 	WARN_ON(!list_empty(&timeout_list));
2442 	flush_workqueue(c4iw_taskq);
2443 	destroy_workqueue(c4iw_taskq);
2444 }
2445 #endif
2446