xref: /freebsd/sys/dev/cxgbe/iw_cxgbe/cm.c (revision bc96366c864c07ef352edb92017357917c75b36c)
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 	mpa = malloc(mpalen, M_CXGBE, M_NOWAIT);
959 	if (mpa == NULL) {
960 failed:
961 		connect_reply_upcall(ep, -ENOMEM);
962 		return;
963 	}
964 
965 	memset(mpa, 0, mpalen);
966 	memcpy(mpa->key, MPA_KEY_REQ, sizeof(mpa->key));
967 	mpa->flags = (crc_enabled ? MPA_CRC : 0) |
968 		(markers_enabled ? MPA_MARKERS : 0) |
969 		(mpa_rev_to_use == 2 ? MPA_ENHANCED_RDMA_CONN : 0);
970 	mpa->private_data_size = htons(ep->plen);
971 	mpa->revision = mpa_rev_to_use;
972 
973 	if (mpa_rev_to_use == 1) {
974 		ep->tried_with_mpa_v1 = 1;
975 		ep->retry_with_mpa_v1 = 0;
976 	}
977 
978 	if (mpa_rev_to_use == 2) {
979 		mpa->private_data_size +=
980 			htons(sizeof(struct mpa_v2_conn_params));
981 		mpa_v2_params.ird = htons((u16)ep->ird);
982 		mpa_v2_params.ord = htons((u16)ep->ord);
983 
984 		if (peer2peer) {
985 			mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL);
986 
987 			if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE) {
988 				mpa_v2_params.ord |=
989 				    htons(MPA_V2_RDMA_WRITE_RTR);
990 			} else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ) {
991 				mpa_v2_params.ord |=
992 					htons(MPA_V2_RDMA_READ_RTR);
993 			}
994 		}
995 		memcpy(mpa->private_data, &mpa_v2_params,
996 			sizeof(struct mpa_v2_conn_params));
997 
998 		if (ep->plen) {
999 
1000 			memcpy(mpa->private_data +
1001 				sizeof(struct mpa_v2_conn_params),
1002 				ep->mpa_pkt + sizeof(*mpa), ep->plen);
1003 		}
1004 	} else {
1005 
1006 		if (ep->plen)
1007 			memcpy(mpa->private_data,
1008 					ep->mpa_pkt + sizeof(*mpa), ep->plen);
1009 		CTR2(KTR_IW_CXGBE, "%s:smr7 %p", __func__, ep);
1010 	}
1011 
1012 	m = m_getm(NULL, mpalen, M_NOWAIT, MT_DATA);
1013 	if (m == NULL) {
1014 		free(mpa, M_CXGBE);
1015 		goto failed;
1016 	}
1017 	m_copyback(m, 0, mpalen, (void *)mpa);
1018 	free(mpa, M_CXGBE);
1019 
1020 	err = sosend(ep->com.so, NULL, NULL, m, NULL, MSG_DONTWAIT,
1021 	    ep->com.thread);
1022 	if (err)
1023 		goto failed;
1024 
1025 	START_EP_TIMER(ep);
1026 	state_set(&ep->com, MPA_REQ_SENT);
1027 	ep->mpa_attr.initiator = 1;
1028 }
1029 
1030 static int send_mpa_reject(struct c4iw_ep *ep, const void *pdata, u8 plen)
1031 {
1032 	int mpalen ;
1033 	struct mpa_message *mpa;
1034 	struct mpa_v2_conn_params mpa_v2_params;
1035 	struct mbuf *m;
1036 	int err;
1037 
1038 	CTR4(KTR_IW_CXGBE, "%s:smrejB %p %u %d", __func__, ep, ep->hwtid,
1039 	    ep->plen);
1040 
1041 	mpalen = sizeof(*mpa) + plen;
1042 
1043 	if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
1044 
1045 		mpalen += sizeof(struct mpa_v2_conn_params);
1046 		CTR4(KTR_IW_CXGBE, "%s:smrej1 %p %u %d", __func__, ep,
1047 		    ep->mpa_attr.version, mpalen);
1048 	}
1049 
1050 	mpa = malloc(mpalen, M_CXGBE, M_NOWAIT);
1051 	if (mpa == NULL)
1052 		return (-ENOMEM);
1053 
1054 	memset(mpa, 0, mpalen);
1055 	memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
1056 	mpa->flags = MPA_REJECT;
1057 	mpa->revision = mpa_rev;
1058 	mpa->private_data_size = htons(plen);
1059 
1060 	if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
1061 
1062 		mpa->flags |= MPA_ENHANCED_RDMA_CONN;
1063 		mpa->private_data_size +=
1064 			htons(sizeof(struct mpa_v2_conn_params));
1065 		mpa_v2_params.ird = htons(((u16)ep->ird) |
1066 				(peer2peer ? MPA_V2_PEER2PEER_MODEL :
1067 				 0));
1068 		mpa_v2_params.ord = htons(((u16)ep->ord) | (peer2peer ?
1069 					(p2p_type ==
1070 					 FW_RI_INIT_P2PTYPE_RDMA_WRITE ?
1071 					 MPA_V2_RDMA_WRITE_RTR : p2p_type ==
1072 					 FW_RI_INIT_P2PTYPE_READ_REQ ?
1073 					 MPA_V2_RDMA_READ_RTR : 0) : 0));
1074 		memcpy(mpa->private_data, &mpa_v2_params,
1075 				sizeof(struct mpa_v2_conn_params));
1076 
1077 		if (ep->plen)
1078 			memcpy(mpa->private_data +
1079 					sizeof(struct mpa_v2_conn_params), pdata, plen);
1080 		CTR5(KTR_IW_CXGBE, "%s:smrej3 %p %d %d %d", __func__, ep,
1081 		    mpa_v2_params.ird, mpa_v2_params.ord, ep->plen);
1082 	} else
1083 		if (plen)
1084 			memcpy(mpa->private_data, pdata, plen);
1085 
1086 	m = m_getm(NULL, mpalen, M_NOWAIT, MT_DATA);
1087 	if (m == NULL) {
1088 		free(mpa, M_CXGBE);
1089 		return (-ENOMEM);
1090 	}
1091 	m_copyback(m, 0, mpalen, (void *)mpa);
1092 	free(mpa, M_CXGBE);
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 	mpa = malloc(mpalen, M_CXGBE, M_NOWAIT);
1121 	if (mpa == NULL)
1122 		return (-ENOMEM);
1123 
1124 	memset(mpa, 0, sizeof(*mpa));
1125 	memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
1126 	mpa->flags = (ep->mpa_attr.crc_enabled ? MPA_CRC : 0) |
1127 		(markers_enabled ? MPA_MARKERS : 0);
1128 	mpa->revision = ep->mpa_attr.version;
1129 	mpa->private_data_size = htons(plen);
1130 
1131 	if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
1132 
1133 		mpa->flags |= MPA_ENHANCED_RDMA_CONN;
1134 		mpa->private_data_size +=
1135 			htons(sizeof(struct mpa_v2_conn_params));
1136 		mpa_v2_params.ird = htons((u16)ep->ird);
1137 		mpa_v2_params.ord = htons((u16)ep->ord);
1138 		CTR5(KTR_IW_CXGBE, "%s:smrep3 %p %d %d %d", __func__, ep,
1139 		    ep->mpa_attr.version, mpa_v2_params.ird, mpa_v2_params.ord);
1140 
1141 		if (peer2peer && (ep->mpa_attr.p2p_type !=
1142 			FW_RI_INIT_P2PTYPE_DISABLED)) {
1143 
1144 			mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL);
1145 
1146 			if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE) {
1147 
1148 				mpa_v2_params.ord |=
1149 					htons(MPA_V2_RDMA_WRITE_RTR);
1150 				CTR5(KTR_IW_CXGBE, "%s:smrep4 %p %d %d %d",
1151 				    __func__, ep, p2p_type, mpa_v2_params.ird,
1152 				    mpa_v2_params.ord);
1153 			}
1154 			else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ) {
1155 
1156 				mpa_v2_params.ord |=
1157 					htons(MPA_V2_RDMA_READ_RTR);
1158 				CTR5(KTR_IW_CXGBE, "%s:smrep5 %p %d %d %d",
1159 				    __func__, ep, p2p_type, mpa_v2_params.ird,
1160 				    mpa_v2_params.ord);
1161 			}
1162 		}
1163 
1164 		memcpy(mpa->private_data, &mpa_v2_params,
1165 			sizeof(struct mpa_v2_conn_params));
1166 
1167 		if (ep->plen)
1168 			memcpy(mpa->private_data +
1169 				sizeof(struct mpa_v2_conn_params), pdata, plen);
1170 	} else
1171 		if (plen)
1172 			memcpy(mpa->private_data, pdata, plen);
1173 
1174 	m = m_getm(NULL, mpalen, M_NOWAIT, MT_DATA);
1175 	if (m == NULL) {
1176 		free(mpa, M_CXGBE);
1177 		return (-ENOMEM);
1178 	}
1179 	m_copyback(m, 0, mpalen, (void *)mpa);
1180 	free(mpa, M_CXGBE);
1181 
1182 
1183 	state_set(&ep->com, MPA_REP_SENT);
1184 	ep->snd_seq += mpalen;
1185 	err = -sosend(ep->com.so, NULL, NULL, m, NULL, MSG_DONTWAIT,
1186 			ep->com.thread);
1187 	CTR3(KTR_IW_CXGBE, "%s:smrepE %p %d", __func__, ep, err);
1188 	return err;
1189 }
1190 
1191 
1192 
1193 static void close_complete_upcall(struct c4iw_ep *ep, int status)
1194 {
1195 	struct iw_cm_event event;
1196 
1197 	CTR2(KTR_IW_CXGBE, "%s:ccuB %p", __func__, ep);
1198 	memset(&event, 0, sizeof(event));
1199 	event.event = IW_CM_EVENT_CLOSE;
1200 	event.status = status;
1201 
1202 	if (ep->com.cm_id) {
1203 
1204 		CTR2(KTR_IW_CXGBE, "%s:ccu1 %1", __func__, ep);
1205 		ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1206 		ep->com.cm_id->rem_ref(ep->com.cm_id);
1207 		ep->com.cm_id = NULL;
1208 		ep->com.qp = NULL;
1209 		set_bit(CLOSE_UPCALL, &ep->com.history);
1210 	}
1211 	CTR2(KTR_IW_CXGBE, "%s:ccuE %p", __func__, ep);
1212 }
1213 
1214 static int abort_connection(struct c4iw_ep *ep)
1215 {
1216 	int err;
1217 
1218 	CTR2(KTR_IW_CXGBE, "%s:abB %p", __func__, ep);
1219 	close_complete_upcall(ep, -ECONNRESET);
1220 	state_set(&ep->com, ABORTING);
1221 	abort_socket(ep);
1222 	err = close_socket(&ep->com, 0);
1223 	set_bit(ABORT_CONN, &ep->com.history);
1224 	CTR2(KTR_IW_CXGBE, "%s:abE %p", __func__, ep);
1225 	return err;
1226 }
1227 
1228 static void peer_close_upcall(struct c4iw_ep *ep)
1229 {
1230 	struct iw_cm_event event;
1231 
1232 	CTR2(KTR_IW_CXGBE, "%s:pcuB %p", __func__, ep);
1233 	memset(&event, 0, sizeof(event));
1234 	event.event = IW_CM_EVENT_DISCONNECT;
1235 
1236 	if (ep->com.cm_id) {
1237 
1238 		CTR2(KTR_IW_CXGBE, "%s:pcu1 %p", __func__, ep);
1239 		ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1240 		set_bit(DISCONN_UPCALL, &ep->com.history);
1241 	}
1242 	CTR2(KTR_IW_CXGBE, "%s:pcuE %p", __func__, ep);
1243 }
1244 
1245 static void peer_abort_upcall(struct c4iw_ep *ep)
1246 {
1247 	struct iw_cm_event event;
1248 
1249 	CTR2(KTR_IW_CXGBE, "%s:pauB %p", __func__, ep);
1250 	memset(&event, 0, sizeof(event));
1251 	event.event = IW_CM_EVENT_CLOSE;
1252 	event.status = -ECONNRESET;
1253 
1254 	if (ep->com.cm_id) {
1255 
1256 		CTR2(KTR_IW_CXGBE, "%s:pau1 %p", __func__, ep);
1257 		ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1258 		ep->com.cm_id->rem_ref(ep->com.cm_id);
1259 		ep->com.cm_id = NULL;
1260 		ep->com.qp = NULL;
1261 		set_bit(ABORT_UPCALL, &ep->com.history);
1262 	}
1263 	CTR2(KTR_IW_CXGBE, "%s:pauE %p", __func__, ep);
1264 }
1265 
1266 static void connect_reply_upcall(struct c4iw_ep *ep, int status)
1267 {
1268 	struct iw_cm_event event;
1269 
1270 	CTR3(KTR_IW_CXGBE, "%s:cruB %p", __func__, ep, status);
1271 	memset(&event, 0, sizeof(event));
1272 	event.event = IW_CM_EVENT_CONNECT_REPLY;
1273 	event.status = (status ==-ECONNABORTED)?-ECONNRESET: status;
1274 	event.local_addr = ep->com.local_addr;
1275 	event.remote_addr = ep->com.remote_addr;
1276 
1277 	if ((status == 0) || (status == -ECONNREFUSED)) {
1278 
1279 		if (!ep->tried_with_mpa_v1) {
1280 
1281 			CTR2(KTR_IW_CXGBE, "%s:cru1 %p", __func__, ep);
1282 			/* this means MPA_v2 is used */
1283 			event.private_data_len = ep->plen -
1284 				sizeof(struct mpa_v2_conn_params);
1285 			event.private_data = ep->mpa_pkt +
1286 				sizeof(struct mpa_message) +
1287 				sizeof(struct mpa_v2_conn_params);
1288 		} else {
1289 
1290 			CTR2(KTR_IW_CXGBE, "%s:cru2 %p", __func__, ep);
1291 			/* this means MPA_v1 is used */
1292 			event.private_data_len = ep->plen;
1293 			event.private_data = ep->mpa_pkt +
1294 				sizeof(struct mpa_message);
1295 		}
1296 	}
1297 
1298 	if (ep->com.cm_id) {
1299 
1300 		CTR2(KTR_IW_CXGBE, "%s:cru3 %p", __func__, ep);
1301 		set_bit(CONN_RPL_UPCALL, &ep->com.history);
1302 		ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1303 	}
1304 
1305 	if(status == -ECONNABORTED) {
1306 
1307 		CTR3(KTR_IW_CXGBE, "%s:cruE %p %d", __func__, ep, status);
1308 		return;
1309 	}
1310 
1311 	if (status < 0) {
1312 
1313 		CTR3(KTR_IW_CXGBE, "%s:cru4 %p %d", __func__, ep, status);
1314 		ep->com.cm_id->rem_ref(ep->com.cm_id);
1315 		ep->com.cm_id = NULL;
1316 		ep->com.qp = NULL;
1317 	}
1318 
1319 	CTR2(KTR_IW_CXGBE, "%s:cruE %p", __func__, ep);
1320 }
1321 
1322 static void connect_request_upcall(struct c4iw_ep *ep)
1323 {
1324 	struct iw_cm_event event;
1325 
1326 	CTR3(KTR_IW_CXGBE, "%s: ep %p, mpa_v1 %d", __func__, ep,
1327 	    ep->tried_with_mpa_v1);
1328 
1329 	memset(&event, 0, sizeof(event));
1330 	event.event = IW_CM_EVENT_CONNECT_REQUEST;
1331 	event.local_addr = ep->com.local_addr;
1332 	event.remote_addr = ep->com.remote_addr;
1333 	event.provider_data = ep;
1334 	event.so = ep->com.so;
1335 
1336 	if (!ep->tried_with_mpa_v1) {
1337 		/* this means MPA_v2 is used */
1338 #ifdef IW_CM_MPAV2
1339 		event.ord = ep->ord;
1340 		event.ird = ep->ird;
1341 #endif
1342 		event.private_data_len = ep->plen -
1343 			sizeof(struct mpa_v2_conn_params);
1344 		event.private_data = ep->mpa_pkt + sizeof(struct mpa_message) +
1345 			sizeof(struct mpa_v2_conn_params);
1346 	} else {
1347 
1348 		/* this means MPA_v1 is used. Send max supported */
1349 #ifdef IW_CM_MPAV2
1350 		event.ord = c4iw_max_read_depth;
1351 		event.ird = c4iw_max_read_depth;
1352 #endif
1353 		event.private_data_len = ep->plen;
1354 		event.private_data = ep->mpa_pkt + sizeof(struct mpa_message);
1355 	}
1356 
1357 	c4iw_get_ep(&ep->com);
1358 	ep->parent_ep->com.cm_id->event_handler(ep->parent_ep->com.cm_id,
1359 	    &event);
1360 	set_bit(CONNREQ_UPCALL, &ep->com.history);
1361 	c4iw_put_ep(&ep->parent_ep->com);
1362 }
1363 
1364 static void established_upcall(struct c4iw_ep *ep)
1365 {
1366 	struct iw_cm_event event;
1367 
1368 	CTR2(KTR_IW_CXGBE, "%s:euB %p", __func__, ep);
1369 	memset(&event, 0, sizeof(event));
1370 	event.event = IW_CM_EVENT_ESTABLISHED;
1371 #ifdef IW_CM_MPAV2
1372 	event.ird = ep->ird;
1373 	event.ord = ep->ord;
1374 #endif
1375 	if (ep->com.cm_id) {
1376 
1377 		CTR2(KTR_IW_CXGBE, "%s:eu1 %p", __func__, ep);
1378 		ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1379 		set_bit(ESTAB_UPCALL, &ep->com.history);
1380 	}
1381 	CTR2(KTR_IW_CXGBE, "%s:euE %p", __func__, ep);
1382 }
1383 
1384 
1385 
1386 static void process_mpa_reply(struct c4iw_ep *ep)
1387 {
1388 	struct mpa_message *mpa;
1389 	struct mpa_v2_conn_params *mpa_v2_params;
1390 	u16 plen;
1391 	u16 resp_ird, resp_ord;
1392 	u8 rtr_mismatch = 0, insuff_ird = 0;
1393 	struct c4iw_qp_attributes attrs;
1394 	enum c4iw_qp_attr_mask mask;
1395 	int err;
1396 	struct mbuf *top, *m;
1397 	int flags = MSG_DONTWAIT;
1398 	struct uio uio;
1399 
1400 	CTR2(KTR_IW_CXGBE, "%s:pmrB %p", __func__, ep);
1401 
1402 	/*
1403 	 * Stop mpa timer.  If it expired, then the state has
1404 	 * changed and we bail since ep_timeout already aborted
1405 	 * the connection.
1406 	 */
1407 	STOP_EP_TIMER(ep);
1408 	if (state_read(&ep->com) != MPA_REQ_SENT)
1409 		return;
1410 
1411 	uio.uio_resid = 1000000;
1412 	uio.uio_td = ep->com.thread;
1413 	err = soreceive(ep->com.so, NULL, &uio, &top, NULL, &flags);
1414 
1415 	if (err) {
1416 
1417 		if (err == EWOULDBLOCK) {
1418 
1419 			CTR2(KTR_IW_CXGBE, "%s:pmr1 %p", __func__, ep);
1420 			START_EP_TIMER(ep);
1421 			return;
1422 		}
1423 		err = -err;
1424 		CTR2(KTR_IW_CXGBE, "%s:pmr2 %p", __func__, ep);
1425 		goto err;
1426 	}
1427 
1428 	if (ep->com.so->so_rcv.sb_mb) {
1429 
1430 		CTR2(KTR_IW_CXGBE, "%s:pmr3 %p", __func__, ep);
1431 		printf("%s data after soreceive called! so %p sb_mb %p top %p\n",
1432 		       __func__, ep->com.so, ep->com.so->so_rcv.sb_mb, top);
1433 	}
1434 
1435 	m = top;
1436 
1437 	do {
1438 
1439 		CTR2(KTR_IW_CXGBE, "%s:pmr4 %p", __func__, ep);
1440 		/*
1441 		 * If we get more than the supported amount of private data
1442 		 * then we must fail this connection.
1443 		 */
1444 		if (ep->mpa_pkt_len + m->m_len > sizeof(ep->mpa_pkt)) {
1445 
1446 			CTR3(KTR_IW_CXGBE, "%s:pmr5 %p %d", __func__, ep,
1447 			    ep->mpa_pkt_len + m->m_len);
1448 			err = (-EINVAL);
1449 			goto err;
1450 		}
1451 
1452 		/*
1453 		 * copy the new data into our accumulation buffer.
1454 		 */
1455 		m_copydata(m, 0, m->m_len, &(ep->mpa_pkt[ep->mpa_pkt_len]));
1456 		ep->mpa_pkt_len += m->m_len;
1457 		if (!m->m_next)
1458 			m = m->m_nextpkt;
1459 		else
1460 			m = m->m_next;
1461 	} while (m);
1462 
1463 	m_freem(top);
1464 	/*
1465 	 * if we don't even have the mpa message, then bail.
1466 	 */
1467 	if (ep->mpa_pkt_len < sizeof(*mpa))
1468 		return;
1469 	mpa = (struct mpa_message *) ep->mpa_pkt;
1470 
1471 	/* Validate MPA header. */
1472 	if (mpa->revision > mpa_rev) {
1473 
1474 		CTR4(KTR_IW_CXGBE, "%s:pmr6 %p %d %d", __func__, ep,
1475 		    mpa->revision, mpa_rev);
1476 		printk(KERN_ERR MOD "%s MPA version mismatch. Local = %d, "
1477 				" Received = %d\n", __func__, mpa_rev, mpa->revision);
1478 		err = -EPROTO;
1479 		goto err;
1480 	}
1481 
1482 	if (memcmp(mpa->key, MPA_KEY_REP, sizeof(mpa->key))) {
1483 
1484 		CTR2(KTR_IW_CXGBE, "%s:pmr7 %p", __func__, ep);
1485 		err = -EPROTO;
1486 		goto err;
1487 	}
1488 
1489 	plen = ntohs(mpa->private_data_size);
1490 
1491 	/*
1492 	 * Fail if there's too much private data.
1493 	 */
1494 	if (plen > MPA_MAX_PRIVATE_DATA) {
1495 
1496 		CTR2(KTR_IW_CXGBE, "%s:pmr8 %p", __func__, ep);
1497 		err = -EPROTO;
1498 		goto err;
1499 	}
1500 
1501 	/*
1502 	 * If plen does not account for pkt size
1503 	 */
1504 	if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
1505 
1506 		CTR2(KTR_IW_CXGBE, "%s:pmr9 %p", __func__, ep);
1507 		err = -EPROTO;
1508 		goto err;
1509 	}
1510 
1511 	ep->plen = (u8) plen;
1512 
1513 	/*
1514 	 * If we don't have all the pdata yet, then bail.
1515 	 * We'll continue process when more data arrives.
1516 	 */
1517 	if (ep->mpa_pkt_len < (sizeof(*mpa) + plen)) {
1518 
1519 		CTR2(KTR_IW_CXGBE, "%s:pmra %p", __func__, ep);
1520 		return;
1521 	}
1522 
1523 	if (mpa->flags & MPA_REJECT) {
1524 
1525 		CTR2(KTR_IW_CXGBE, "%s:pmrb %p", __func__, ep);
1526 		err = -ECONNREFUSED;
1527 		goto err;
1528 	}
1529 
1530 	/*
1531 	 * If we get here we have accumulated the entire mpa
1532 	 * start reply message including private data. And
1533 	 * the MPA header is valid.
1534 	 */
1535 	state_set(&ep->com, FPDU_MODE);
1536 	ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1537 	ep->mpa_attr.recv_marker_enabled = markers_enabled;
1538 	ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1539 	ep->mpa_attr.version = mpa->revision;
1540 	ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1541 
1542 	if (mpa->revision == 2) {
1543 
1544 		CTR2(KTR_IW_CXGBE, "%s:pmrc %p", __func__, ep);
1545 		ep->mpa_attr.enhanced_rdma_conn =
1546 			mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
1547 
1548 		if (ep->mpa_attr.enhanced_rdma_conn) {
1549 
1550 			CTR2(KTR_IW_CXGBE, "%s:pmrd %p", __func__, ep);
1551 			mpa_v2_params = (struct mpa_v2_conn_params *)
1552 				(ep->mpa_pkt + sizeof(*mpa));
1553 			resp_ird = ntohs(mpa_v2_params->ird) &
1554 				MPA_V2_IRD_ORD_MASK;
1555 			resp_ord = ntohs(mpa_v2_params->ord) &
1556 				MPA_V2_IRD_ORD_MASK;
1557 
1558 			/*
1559 			 * This is a double-check. Ideally, below checks are
1560 			 * not required since ird/ord stuff has been taken
1561 			 * care of in c4iw_accept_cr
1562 			 */
1563 			if ((ep->ird < resp_ord) || (ep->ord > resp_ird)) {
1564 
1565 				CTR2(KTR_IW_CXGBE, "%s:pmre %p", __func__, ep);
1566 				err = -ENOMEM;
1567 				ep->ird = resp_ord;
1568 				ep->ord = resp_ird;
1569 				insuff_ird = 1;
1570 			}
1571 
1572 			if (ntohs(mpa_v2_params->ird) &
1573 				MPA_V2_PEER2PEER_MODEL) {
1574 
1575 				CTR2(KTR_IW_CXGBE, "%s:pmrf %p", __func__, ep);
1576 				if (ntohs(mpa_v2_params->ord) &
1577 					MPA_V2_RDMA_WRITE_RTR) {
1578 
1579 					CTR2(KTR_IW_CXGBE, "%s:pmrg %p", __func__, ep);
1580 					ep->mpa_attr.p2p_type =
1581 						FW_RI_INIT_P2PTYPE_RDMA_WRITE;
1582 				}
1583 				else if (ntohs(mpa_v2_params->ord) &
1584 					MPA_V2_RDMA_READ_RTR) {
1585 
1586 					CTR2(KTR_IW_CXGBE, "%s:pmrh %p", __func__, ep);
1587 					ep->mpa_attr.p2p_type =
1588 						FW_RI_INIT_P2PTYPE_READ_REQ;
1589 				}
1590 			}
1591 		}
1592 	} else {
1593 
1594 		CTR2(KTR_IW_CXGBE, "%s:pmri %p", __func__, ep);
1595 
1596 		if (mpa->revision == 1) {
1597 
1598 			CTR2(KTR_IW_CXGBE, "%s:pmrj %p", __func__, ep);
1599 
1600 			if (peer2peer) {
1601 
1602 				CTR2(KTR_IW_CXGBE, "%s:pmrk %p", __func__, ep);
1603 				ep->mpa_attr.p2p_type = p2p_type;
1604 			}
1605 		}
1606 	}
1607 
1608 	if (set_tcpinfo(ep)) {
1609 
1610 		CTR2(KTR_IW_CXGBE, "%s:pmrl %p", __func__, ep);
1611 		printf("%s set_tcpinfo error\n", __func__);
1612 		goto err;
1613 	}
1614 
1615 	CTR6(KTR_IW_CXGBE, "%s - crc_enabled = %d, recv_marker_enabled = %d, "
1616 	    "xmit_marker_enabled = %d, version = %d p2p_type = %d", __func__,
1617 	    ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
1618 	    ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version,
1619 	    ep->mpa_attr.p2p_type);
1620 
1621 	/*
1622 	 * If responder's RTR does not match with that of initiator, assign
1623 	 * FW_RI_INIT_P2PTYPE_DISABLED in mpa attributes so that RTR is not
1624 	 * generated when moving QP to RTS state.
1625 	 * A TERM message will be sent after QP has moved to RTS state
1626 	 */
1627 	if ((ep->mpa_attr.version == 2) && peer2peer &&
1628 		(ep->mpa_attr.p2p_type != p2p_type)) {
1629 
1630 		CTR2(KTR_IW_CXGBE, "%s:pmrm %p", __func__, ep);
1631 		ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1632 		rtr_mismatch = 1;
1633 	}
1634 
1635 
1636 	//ep->ofld_txq = TOEPCB(ep->com.so)->ofld_txq;
1637 	attrs.mpa_attr = ep->mpa_attr;
1638 	attrs.max_ird = ep->ird;
1639 	attrs.max_ord = ep->ord;
1640 	attrs.llp_stream_handle = ep;
1641 	attrs.next_state = C4IW_QP_STATE_RTS;
1642 
1643 	mask = C4IW_QP_ATTR_NEXT_STATE |
1644 		C4IW_QP_ATTR_LLP_STREAM_HANDLE | C4IW_QP_ATTR_MPA_ATTR |
1645 		C4IW_QP_ATTR_MAX_IRD | C4IW_QP_ATTR_MAX_ORD;
1646 
1647 	/* bind QP and TID with INIT_WR */
1648 	err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp, mask, &attrs, 1);
1649 
1650 	if (err) {
1651 
1652 		CTR2(KTR_IW_CXGBE, "%s:pmrn %p", __func__, ep);
1653 		goto err;
1654 	}
1655 
1656 	/*
1657 	 * If responder's RTR requirement did not match with what initiator
1658 	 * supports, generate TERM message
1659 	 */
1660 	if (rtr_mismatch) {
1661 
1662 		CTR2(KTR_IW_CXGBE, "%s:pmro %p", __func__, ep);
1663 		printk(KERN_ERR "%s: RTR mismatch, sending TERM\n", __func__);
1664 		attrs.layer_etype = LAYER_MPA | DDP_LLP;
1665 		attrs.ecode = MPA_NOMATCH_RTR;
1666 		attrs.next_state = C4IW_QP_STATE_TERMINATE;
1667 		err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1668 			C4IW_QP_ATTR_NEXT_STATE, &attrs, 0);
1669 		err = -ENOMEM;
1670 		goto out;
1671 	}
1672 
1673 	/*
1674 	 * Generate TERM if initiator IRD is not sufficient for responder
1675 	 * provided ORD. Currently, we do the same behaviour even when
1676 	 * responder provided IRD is also not sufficient as regards to
1677 	 * initiator ORD.
1678 	 */
1679 	if (insuff_ird) {
1680 
1681 		CTR2(KTR_IW_CXGBE, "%s:pmrp %p", __func__, ep);
1682 		printk(KERN_ERR "%s: Insufficient IRD, sending TERM\n",
1683 				__func__);
1684 		attrs.layer_etype = LAYER_MPA | DDP_LLP;
1685 		attrs.ecode = MPA_INSUFF_IRD;
1686 		attrs.next_state = C4IW_QP_STATE_TERMINATE;
1687 		err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1688 			C4IW_QP_ATTR_NEXT_STATE, &attrs, 0);
1689 		err = -ENOMEM;
1690 		goto out;
1691 	}
1692 	goto out;
1693 err:
1694 	state_set(&ep->com, ABORTING);
1695 	abort_connection(ep);
1696 out:
1697 	connect_reply_upcall(ep, err);
1698 	CTR2(KTR_IW_CXGBE, "%s:pmrE %p", __func__, ep);
1699 	return;
1700 }
1701 
1702 static void
1703 process_mpa_request(struct c4iw_ep *ep)
1704 {
1705 	struct mpa_message *mpa;
1706 	u16 plen;
1707 	int flags = MSG_DONTWAIT;
1708 	int rc;
1709 	struct iovec iov;
1710 	struct uio uio;
1711 	enum c4iw_ep_state state = state_read(&ep->com);
1712 
1713 	CTR3(KTR_IW_CXGBE, "%s: ep %p, state %s", __func__, ep, states[state]);
1714 
1715 	if (state != MPA_REQ_WAIT)
1716 		return;
1717 
1718 	iov.iov_base = &ep->mpa_pkt[ep->mpa_pkt_len];
1719 	iov.iov_len = sizeof(ep->mpa_pkt) - ep->mpa_pkt_len;
1720 	uio.uio_iov = &iov;
1721 	uio.uio_iovcnt = 1;
1722 	uio.uio_offset = 0;
1723 	uio.uio_resid = sizeof(ep->mpa_pkt) - ep->mpa_pkt_len;
1724 	uio.uio_segflg = UIO_SYSSPACE;
1725 	uio.uio_rw = UIO_READ;
1726 	uio.uio_td = NULL; /* uio.uio_td = ep->com.thread; */
1727 
1728 	rc = soreceive(ep->com.so, NULL, &uio, NULL, NULL, &flags);
1729 	if (rc == EAGAIN)
1730 		return;
1731 	else if (rc) {
1732 abort:
1733 		STOP_EP_TIMER(ep);
1734 		abort_connection(ep);
1735 		return;
1736 	}
1737 	KASSERT(uio.uio_offset > 0, ("%s: sorecieve on so %p read no data",
1738 	    __func__, ep->com.so));
1739 	ep->mpa_pkt_len += uio.uio_offset;
1740 
1741 	/*
1742 	 * If we get more than the supported amount of private data then we must
1743 	 * fail this connection.  XXX: check so_rcv->sb_cc, or peek with another
1744 	 * soreceive, or increase the size of mpa_pkt by 1 and abort if the last
1745 	 * byte is filled by the soreceive above.
1746 	 */
1747 
1748 	/* Don't even have the MPA message.  Wait for more data to arrive. */
1749 	if (ep->mpa_pkt_len < sizeof(*mpa))
1750 		return;
1751 	mpa = (struct mpa_message *) ep->mpa_pkt;
1752 
1753 	/*
1754 	 * Validate MPA Header.
1755 	 */
1756 	if (mpa->revision > mpa_rev) {
1757 		log(LOG_ERR, "%s: MPA version mismatch. Local = %d,"
1758 		    " Received = %d\n", __func__, mpa_rev, mpa->revision);
1759 		goto abort;
1760 	}
1761 
1762 	if (memcmp(mpa->key, MPA_KEY_REQ, sizeof(mpa->key)))
1763 		goto abort;
1764 
1765 	/*
1766 	 * Fail if there's too much private data.
1767 	 */
1768 	plen = ntohs(mpa->private_data_size);
1769 	if (plen > MPA_MAX_PRIVATE_DATA)
1770 		goto abort;
1771 
1772 	/*
1773 	 * If plen does not account for pkt size
1774 	 */
1775 	if (ep->mpa_pkt_len > (sizeof(*mpa) + plen))
1776 		goto abort;
1777 
1778 	ep->plen = (u8) plen;
1779 
1780 	/*
1781 	 * If we don't have all the pdata yet, then bail.
1782 	 */
1783 	if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
1784 		return;
1785 
1786 	/*
1787 	 * If we get here we have accumulated the entire mpa
1788 	 * start reply message including private data.
1789 	 */
1790 	ep->mpa_attr.initiator = 0;
1791 	ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1792 	ep->mpa_attr.recv_marker_enabled = markers_enabled;
1793 	ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1794 	ep->mpa_attr.version = mpa->revision;
1795 	if (mpa->revision == 1)
1796 		ep->tried_with_mpa_v1 = 1;
1797 	ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1798 
1799 	if (mpa->revision == 2) {
1800 		ep->mpa_attr.enhanced_rdma_conn =
1801 		    mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
1802 		if (ep->mpa_attr.enhanced_rdma_conn) {
1803 			struct mpa_v2_conn_params *mpa_v2_params;
1804 			u16 ird, ord;
1805 
1806 			mpa_v2_params = (void *)&ep->mpa_pkt[sizeof(*mpa)];
1807 			ird = ntohs(mpa_v2_params->ird);
1808 			ord = ntohs(mpa_v2_params->ord);
1809 
1810 			ep->ird = ird & MPA_V2_IRD_ORD_MASK;
1811 			ep->ord = ord & MPA_V2_IRD_ORD_MASK;
1812 			if (ird & MPA_V2_PEER2PEER_MODEL && peer2peer) {
1813 				if (ord & MPA_V2_RDMA_WRITE_RTR) {
1814 					ep->mpa_attr.p2p_type =
1815 					    FW_RI_INIT_P2PTYPE_RDMA_WRITE;
1816 				} else if (ord & MPA_V2_RDMA_READ_RTR) {
1817 					ep->mpa_attr.p2p_type =
1818 					    FW_RI_INIT_P2PTYPE_READ_REQ;
1819 				}
1820 			}
1821 		}
1822 	} else if (mpa->revision == 1 && peer2peer)
1823 		ep->mpa_attr.p2p_type = p2p_type;
1824 
1825 	if (set_tcpinfo(ep))
1826 		goto abort;
1827 
1828 	CTR5(KTR_IW_CXGBE, "%s: crc_enabled = %d, recv_marker_enabled = %d, "
1829 	    "xmit_marker_enabled = %d, version = %d", __func__,
1830 	    ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
1831 	    ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version);
1832 
1833 	state_set(&ep->com, MPA_REQ_RCVD);
1834 	STOP_EP_TIMER(ep);
1835 
1836 	/* drive upcall */
1837 	mutex_lock(&ep->parent_ep->com.mutex);
1838 	if (ep->parent_ep->com.state != DEAD)
1839 		connect_request_upcall(ep);
1840 	else
1841 		abort_connection(ep);
1842 	mutex_unlock(&ep->parent_ep->com.mutex);
1843 }
1844 
1845 /*
1846  * Upcall from the adapter indicating data has been transmitted.
1847  * For us its just the single MPA request or reply.  We can now free
1848  * the skb holding the mpa message.
1849  */
1850 int c4iw_reject_cr(struct iw_cm_id *cm_id, const void *pdata, u8 pdata_len)
1851 {
1852 	int err;
1853 	struct c4iw_ep *ep = to_ep(cm_id);
1854 	CTR2(KTR_IW_CXGBE, "%s:crcB %p", __func__, ep);
1855 
1856 	if (state_read(&ep->com) == DEAD) {
1857 
1858 		CTR2(KTR_IW_CXGBE, "%s:crc1 %p", __func__, ep);
1859 		c4iw_put_ep(&ep->com);
1860 		return -ECONNRESET;
1861 	}
1862 	set_bit(ULP_REJECT, &ep->com.history);
1863 	BUG_ON(state_read(&ep->com) != MPA_REQ_RCVD);
1864 
1865 	if (mpa_rev == 0) {
1866 
1867 		CTR2(KTR_IW_CXGBE, "%s:crc2 %p", __func__, ep);
1868 		abort_connection(ep);
1869 	}
1870 	else {
1871 
1872 		CTR2(KTR_IW_CXGBE, "%s:crc3 %p", __func__, ep);
1873 		err = send_mpa_reject(ep, pdata, pdata_len);
1874 		err = soshutdown(ep->com.so, 3);
1875 	}
1876 	c4iw_put_ep(&ep->com);
1877 	CTR2(KTR_IW_CXGBE, "%s:crc4 %p", __func__, ep);
1878 	return 0;
1879 }
1880 
1881 int c4iw_accept_cr(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
1882 {
1883 	int err;
1884 	struct c4iw_qp_attributes attrs;
1885 	enum c4iw_qp_attr_mask mask;
1886 	struct c4iw_ep *ep = to_ep(cm_id);
1887 	struct c4iw_dev *h = to_c4iw_dev(cm_id->device);
1888 	struct c4iw_qp *qp = get_qhp(h, conn_param->qpn);
1889 
1890 	CTR2(KTR_IW_CXGBE, "%s:cacB %p", __func__, ep);
1891 
1892 	if (state_read(&ep->com) == DEAD) {
1893 
1894 		CTR2(KTR_IW_CXGBE, "%s:cac1 %p", __func__, ep);
1895 		err = -ECONNRESET;
1896 		goto err;
1897 	}
1898 
1899 	BUG_ON(state_read(&ep->com) != MPA_REQ_RCVD);
1900 	BUG_ON(!qp);
1901 
1902 	set_bit(ULP_ACCEPT, &ep->com.history);
1903 
1904 	if ((conn_param->ord > c4iw_max_read_depth) ||
1905 		(conn_param->ird > c4iw_max_read_depth)) {
1906 
1907 		CTR2(KTR_IW_CXGBE, "%s:cac2 %p", __func__, ep);
1908 		abort_connection(ep);
1909 		err = -EINVAL;
1910 		goto err;
1911 	}
1912 
1913 	if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
1914 
1915 		CTR2(KTR_IW_CXGBE, "%s:cac3 %p", __func__, ep);
1916 
1917 		if (conn_param->ord > ep->ird) {
1918 
1919 			CTR2(KTR_IW_CXGBE, "%s:cac4 %p", __func__, ep);
1920 			ep->ird = conn_param->ird;
1921 			ep->ord = conn_param->ord;
1922 			send_mpa_reject(ep, conn_param->private_data,
1923 					conn_param->private_data_len);
1924 			abort_connection(ep);
1925 			err = -ENOMEM;
1926 			goto err;
1927 		}
1928 
1929 		if (conn_param->ird > ep->ord) {
1930 
1931 			CTR2(KTR_IW_CXGBE, "%s:cac5 %p", __func__, ep);
1932 
1933 			if (!ep->ord) {
1934 
1935 				CTR2(KTR_IW_CXGBE, "%s:cac6 %p", __func__, ep);
1936 				conn_param->ird = 1;
1937 			}
1938 			else {
1939 				CTR2(KTR_IW_CXGBE, "%s:cac7 %p", __func__, ep);
1940 				abort_connection(ep);
1941 				err = -ENOMEM;
1942 				goto err;
1943 			}
1944 		}
1945 
1946 	}
1947 	ep->ird = conn_param->ird;
1948 	ep->ord = conn_param->ord;
1949 
1950 	if (ep->mpa_attr.version != 2) {
1951 
1952 		CTR2(KTR_IW_CXGBE, "%s:cac8 %p", __func__, ep);
1953 
1954 		if (peer2peer && ep->ird == 0) {
1955 
1956 			CTR2(KTR_IW_CXGBE, "%s:cac9 %p", __func__, ep);
1957 			ep->ird = 1;
1958 		}
1959 	}
1960 
1961 
1962 	cm_id->add_ref(cm_id);
1963 	ep->com.cm_id = cm_id;
1964 	ep->com.qp = qp;
1965 	//ep->ofld_txq = TOEPCB(ep->com.so)->ofld_txq;
1966 
1967 	/* bind QP to EP and move to RTS */
1968 	attrs.mpa_attr = ep->mpa_attr;
1969 	attrs.max_ird = ep->ird;
1970 	attrs.max_ord = ep->ord;
1971 	attrs.llp_stream_handle = ep;
1972 	attrs.next_state = C4IW_QP_STATE_RTS;
1973 
1974 	/* bind QP and TID with INIT_WR */
1975 	mask = C4IW_QP_ATTR_NEXT_STATE |
1976 		C4IW_QP_ATTR_LLP_STREAM_HANDLE |
1977 		C4IW_QP_ATTR_MPA_ATTR |
1978 		C4IW_QP_ATTR_MAX_IRD |
1979 		C4IW_QP_ATTR_MAX_ORD;
1980 
1981 	err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp, mask, &attrs, 1);
1982 
1983 	if (err) {
1984 
1985 		CTR2(KTR_IW_CXGBE, "%s:caca %p", __func__, ep);
1986 		goto err1;
1987 	}
1988 	err = send_mpa_reply(ep, conn_param->private_data,
1989 			conn_param->private_data_len);
1990 
1991 	if (err) {
1992 
1993 		CTR2(KTR_IW_CXGBE, "%s:caca %p", __func__, ep);
1994 		goto err1;
1995 	}
1996 
1997 	state_set(&ep->com, FPDU_MODE);
1998 	established_upcall(ep);
1999 	c4iw_put_ep(&ep->com);
2000 	CTR2(KTR_IW_CXGBE, "%s:cacE %p", __func__, ep);
2001 	return 0;
2002 err1:
2003 	ep->com.cm_id = NULL;
2004 	ep->com.qp = NULL;
2005 	cm_id->rem_ref(cm_id);
2006 err:
2007 	c4iw_put_ep(&ep->com);
2008 	CTR2(KTR_IW_CXGBE, "%s:cacE err %p", __func__, ep);
2009 	return err;
2010 }
2011 
2012 
2013 
2014 int c4iw_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
2015 {
2016 	int err = 0;
2017 	struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
2018 	struct c4iw_ep *ep = NULL;
2019 	struct rtentry *rt;
2020 	struct toedev *tdev;
2021 
2022 	CTR2(KTR_IW_CXGBE, "%s:ccB %p", __func__, cm_id);
2023 
2024 	if ((conn_param->ord > c4iw_max_read_depth) ||
2025 		(conn_param->ird > c4iw_max_read_depth)) {
2026 
2027 		CTR2(KTR_IW_CXGBE, "%s:cc1 %p", __func__, cm_id);
2028 		err = -EINVAL;
2029 		goto out;
2030 	}
2031 	ep = alloc_ep(sizeof(*ep), M_NOWAIT);
2032 
2033 	if (!ep) {
2034 
2035 		CTR2(KTR_IW_CXGBE, "%s:cc2 %p", __func__, cm_id);
2036 		printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__);
2037 		err = -ENOMEM;
2038 		goto out;
2039 	}
2040 	init_timer(&ep->timer);
2041 	ep->plen = conn_param->private_data_len;
2042 
2043 	if (ep->plen) {
2044 
2045 		CTR2(KTR_IW_CXGBE, "%s:cc3 %p", __func__, ep);
2046 		memcpy(ep->mpa_pkt + sizeof(struct mpa_message),
2047 				conn_param->private_data, ep->plen);
2048 	}
2049 	ep->ird = conn_param->ird;
2050 	ep->ord = conn_param->ord;
2051 
2052 	if (peer2peer && ep->ord == 0) {
2053 
2054 		CTR2(KTR_IW_CXGBE, "%s:cc4 %p", __func__, ep);
2055 		ep->ord = 1;
2056 	}
2057 
2058 	cm_id->add_ref(cm_id);
2059 	ep->com.dev = dev;
2060 	ep->com.cm_id = cm_id;
2061 	ep->com.qp = get_qhp(dev, conn_param->qpn);
2062 
2063 	if (!ep->com.qp) {
2064 
2065 		CTR2(KTR_IW_CXGBE, "%s:cc5 %p", __func__, ep);
2066 		err = -EINVAL;
2067 		goto fail2;
2068 	}
2069 	ep->com.thread = curthread;
2070 	ep->com.so = cm_id->so;
2071 
2072 	init_sock(&ep->com);
2073 
2074 	/* find a route */
2075 	rt = find_route(
2076 		cm_id->local_addr.sin_addr.s_addr,
2077 		cm_id->remote_addr.sin_addr.s_addr,
2078 		cm_id->local_addr.sin_port,
2079 		cm_id->remote_addr.sin_port, 0);
2080 
2081 	if (!rt) {
2082 
2083 		CTR2(KTR_IW_CXGBE, "%s:cc7 %p", __func__, ep);
2084 		printk(KERN_ERR MOD "%s - cannot find route.\n", __func__);
2085 		err = -EHOSTUNREACH;
2086 		goto fail2;
2087 	}
2088 
2089 	if (!(rt->rt_ifp->if_capenable & IFCAP_TOE)) {
2090 
2091 		CTR2(KTR_IW_CXGBE, "%s:cc8 %p", __func__, ep);
2092 		printf("%s - interface not TOE capable.\n", __func__);
2093 		close_socket(&ep->com, 0);
2094 		err = -ENOPROTOOPT;
2095 		goto fail3;
2096 	}
2097 	tdev = TOEDEV(rt->rt_ifp);
2098 
2099 	if (tdev == NULL) {
2100 
2101 		CTR2(KTR_IW_CXGBE, "%s:cc9 %p", __func__, ep);
2102 		printf("%s - No toedev for interface.\n", __func__);
2103 		goto fail3;
2104 	}
2105 	RTFREE(rt);
2106 
2107 	state_set(&ep->com, CONNECTING);
2108 	ep->tos = 0;
2109 	ep->com.local_addr = cm_id->local_addr;
2110 	ep->com.remote_addr = cm_id->remote_addr;
2111 	err = soconnect(ep->com.so, (struct sockaddr *)&ep->com.remote_addr,
2112 		ep->com.thread);
2113 
2114 	if (!err) {
2115 		CTR2(KTR_IW_CXGBE, "%s:cca %p", __func__, ep);
2116 		goto out;
2117 	} else {
2118 		close_socket(&ep->com, 0);
2119 		goto fail2;
2120 	}
2121 
2122 fail3:
2123 	CTR2(KTR_IW_CXGBE, "%s:ccb %p", __func__, ep);
2124 	RTFREE(rt);
2125 fail2:
2126 	cm_id->rem_ref(cm_id);
2127 	c4iw_put_ep(&ep->com);
2128 out:
2129 	CTR2(KTR_IW_CXGBE, "%s:ccE %p", __func__, ep);
2130 	return err;
2131 }
2132 
2133 /*
2134  * iwcm->create_listen.  Returns -errno on failure.
2135  */
2136 int
2137 c4iw_create_listen(struct iw_cm_id *cm_id, int backlog)
2138 {
2139 	int rc;
2140 	struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
2141 	struct c4iw_listen_ep *ep;
2142 	struct socket *so = cm_id->so;
2143 
2144 	ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
2145 	CTR5(KTR_IW_CXGBE, "%s: cm_id %p, lso %p, ep %p, inp %p", __func__,
2146 	    cm_id, so, ep, so->so_pcb);
2147 	if (ep == NULL) {
2148 		log(LOG_ERR, "%s: failed to alloc memory for endpoint\n",
2149 		    __func__);
2150 		rc = ENOMEM;
2151 		goto failed;
2152 	}
2153 
2154 	cm_id->add_ref(cm_id);
2155 	ep->com.cm_id = cm_id;
2156 	ep->com.dev = dev;
2157 	ep->backlog = backlog;
2158 	ep->com.local_addr = cm_id->local_addr;
2159 	ep->com.thread = curthread;
2160 	state_set(&ep->com, LISTEN);
2161 	ep->com.so = so;
2162 	init_sock(&ep->com);
2163 
2164 	rc = solisten(so, ep->backlog, ep->com.thread);
2165 	if (rc != 0) {
2166 		log(LOG_ERR, "%s: failed to start listener: %d\n", __func__,
2167 		    rc);
2168 		close_socket(&ep->com, 0);
2169 		cm_id->rem_ref(cm_id);
2170 		c4iw_put_ep(&ep->com);
2171 		goto failed;
2172 	}
2173 
2174 	cm_id->provider_data = ep;
2175 	return (0);
2176 
2177 failed:
2178 	CTR3(KTR_IW_CXGBE, "%s: cm_id %p, FAILED (%d)", __func__, cm_id, rc);
2179 	return (-rc);
2180 }
2181 
2182 int
2183 c4iw_destroy_listen(struct iw_cm_id *cm_id)
2184 {
2185 	int rc;
2186 	struct c4iw_listen_ep *ep = to_listen_ep(cm_id);
2187 
2188 	CTR4(KTR_IW_CXGBE, "%s: cm_id %p, so %p, inp %p", __func__, cm_id,
2189 	    cm_id->so, cm_id->so->so_pcb);
2190 
2191 	state_set(&ep->com, DEAD);
2192 	rc = close_socket(&ep->com, 0);
2193 	cm_id->rem_ref(cm_id);
2194 	c4iw_put_ep(&ep->com);
2195 
2196 	return (rc);
2197 }
2198 
2199 int c4iw_ep_disconnect(struct c4iw_ep *ep, int abrupt, gfp_t gfp)
2200 {
2201 	int ret = 0;
2202 	int close = 0;
2203 	int fatal = 0;
2204 	struct c4iw_rdev *rdev;
2205 
2206 	mutex_lock(&ep->com.mutex);
2207 
2208 	CTR2(KTR_IW_CXGBE, "%s:cedB %p", __func__, ep);
2209 
2210 	rdev = &ep->com.dev->rdev;
2211 
2212 	if (c4iw_fatal_error(rdev)) {
2213 
2214 		CTR2(KTR_IW_CXGBE, "%s:ced1 %p", __func__, ep);
2215 		fatal = 1;
2216 		close_complete_upcall(ep, -EIO);
2217 		ep->com.state = DEAD;
2218 	}
2219 	CTR3(KTR_IW_CXGBE, "%s:ced2 %p %s", __func__, ep,
2220 	    states[ep->com.state]);
2221 
2222 	switch (ep->com.state) {
2223 
2224 		case MPA_REQ_WAIT:
2225 		case MPA_REQ_SENT:
2226 		case MPA_REQ_RCVD:
2227 		case MPA_REP_SENT:
2228 		case FPDU_MODE:
2229 			close = 1;
2230 			if (abrupt)
2231 				ep->com.state = ABORTING;
2232 			else {
2233 				ep->com.state = CLOSING;
2234 				START_EP_TIMER(ep);
2235 			}
2236 			set_bit(CLOSE_SENT, &ep->com.flags);
2237 			break;
2238 
2239 		case CLOSING:
2240 
2241 			if (!test_and_set_bit(CLOSE_SENT, &ep->com.flags)) {
2242 
2243 				close = 1;
2244 				if (abrupt) {
2245 					STOP_EP_TIMER(ep);
2246 					ep->com.state = ABORTING;
2247 				} else
2248 					ep->com.state = MORIBUND;
2249 			}
2250 			break;
2251 
2252 		case MORIBUND:
2253 		case ABORTING:
2254 		case DEAD:
2255 			CTR3(KTR_IW_CXGBE,
2256 			    "%s ignoring disconnect ep %p state %u", __func__,
2257 			    ep, ep->com.state);
2258 			break;
2259 
2260 		default:
2261 			BUG();
2262 			break;
2263 	}
2264 
2265 	mutex_unlock(&ep->com.mutex);
2266 
2267 	if (close) {
2268 
2269 		CTR2(KTR_IW_CXGBE, "%s:ced3 %p", __func__, ep);
2270 
2271 		if (abrupt) {
2272 
2273 			CTR2(KTR_IW_CXGBE, "%s:ced4 %p", __func__, ep);
2274 			set_bit(EP_DISC_ABORT, &ep->com.history);
2275 			ret = abort_connection(ep);
2276 		} else {
2277 
2278 			CTR2(KTR_IW_CXGBE, "%s:ced5 %p", __func__, ep);
2279 			set_bit(EP_DISC_CLOSE, &ep->com.history);
2280 
2281 			if (!ep->parent_ep)
2282 				__state_set(&ep->com, MORIBUND);
2283 			ret = shutdown_socket(&ep->com);
2284 		}
2285 
2286 		if (ret) {
2287 
2288 			fatal = 1;
2289 		}
2290 	}
2291 
2292 	if (fatal) {
2293 
2294 		release_ep_resources(ep);
2295 		CTR2(KTR_IW_CXGBE, "%s:ced6 %p", __func__, ep);
2296 	}
2297 	CTR2(KTR_IW_CXGBE, "%s:cedE %p", __func__, ep);
2298 	return ret;
2299 }
2300 
2301 #ifdef C4IW_EP_REDIRECT
2302 int c4iw_ep_redirect(void *ctx, struct dst_entry *old, struct dst_entry *new,
2303 		struct l2t_entry *l2t)
2304 {
2305 	struct c4iw_ep *ep = ctx;
2306 
2307 	if (ep->dst != old)
2308 		return 0;
2309 
2310 	PDBG("%s ep %p redirect to dst %p l2t %p\n", __func__, ep, new,
2311 			l2t);
2312 	dst_hold(new);
2313 	cxgb4_l2t_release(ep->l2t);
2314 	ep->l2t = l2t;
2315 	dst_release(old);
2316 	ep->dst = new;
2317 	return 1;
2318 }
2319 #endif
2320 
2321 
2322 
2323 static void ep_timeout(unsigned long arg)
2324 {
2325 	struct c4iw_ep *ep = (struct c4iw_ep *)arg;
2326 	int kickit = 0;
2327 
2328 	CTR2(KTR_IW_CXGBE, "%s:etB %p", __func__, ep);
2329 	spin_lock(&timeout_lock);
2330 
2331 	if (!test_and_set_bit(TIMEOUT, &ep->com.flags)) {
2332 
2333 		list_add_tail(&ep->entry, &timeout_list);
2334 		kickit = 1;
2335 	}
2336 	spin_unlock(&timeout_lock);
2337 
2338 	if (kickit) {
2339 
2340 		CTR2(KTR_IW_CXGBE, "%s:et1 %p", __func__, ep);
2341 		queue_work(c4iw_taskq, &c4iw_task);
2342 	}
2343 	CTR2(KTR_IW_CXGBE, "%s:etE %p", __func__, ep);
2344 }
2345 
2346 static int fw6_wr_rpl(struct adapter *sc, const __be64 *rpl)
2347 {
2348 	uint64_t val = be64toh(*rpl);
2349 	int ret;
2350 	struct c4iw_wr_wait *wr_waitp;
2351 
2352 	ret = (int)((val >> 8) & 0xff);
2353 	wr_waitp = (struct c4iw_wr_wait *)rpl[1];
2354 	CTR3(KTR_IW_CXGBE, "%s wr_waitp %p ret %u", __func__, wr_waitp, ret);
2355 	if (wr_waitp)
2356 		c4iw_wake_up(wr_waitp, ret ? -ret : 0);
2357 
2358 	return (0);
2359 }
2360 
2361 static int fw6_cqe_handler(struct adapter *sc, const __be64 *rpl)
2362 {
2363 	struct t4_cqe cqe =*(const struct t4_cqe *)(&rpl[0]);
2364 
2365 	CTR2(KTR_IW_CXGBE, "%s rpl %p", __func__, rpl);
2366 	c4iw_ev_dispatch(sc->iwarp_softc, &cqe);
2367 
2368 	return (0);
2369 }
2370 
2371 static int terminate(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m)
2372 {
2373 
2374 	struct adapter *sc = iq->adapter;
2375 
2376 	const struct cpl_rdma_terminate *rpl = (const void *)(rss + 1);
2377 	unsigned int tid = GET_TID(rpl);
2378 	struct c4iw_qp_attributes attrs;
2379 	struct toepcb *toep = lookup_tid(sc, tid);
2380 	struct socket *so = inp_inpcbtosocket(toep->inp);
2381 	struct c4iw_ep *ep = so->so_rcv.sb_upcallarg;
2382 
2383 	CTR2(KTR_IW_CXGBE, "%s:tB %p %d", __func__, ep);
2384 
2385 	if (ep && ep->com.qp) {
2386 
2387 		printk(KERN_WARNING MOD "TERM received tid %u qpid %u\n", tid,
2388 				ep->com.qp->wq.sq.qid);
2389 		attrs.next_state = C4IW_QP_STATE_TERMINATE;
2390 		c4iw_modify_qp(ep->com.dev, ep->com.qp, C4IW_QP_ATTR_NEXT_STATE, &attrs,
2391 				1);
2392 	} else
2393 		printk(KERN_WARNING MOD "TERM received tid %u no ep/qp\n", tid);
2394 	CTR2(KTR_IW_CXGBE, "%s:tE %p %d", __func__, ep);
2395 
2396 	return 0;
2397 }
2398 
2399 	void
2400 c4iw_cm_init_cpl(struct adapter *sc)
2401 {
2402 
2403 	t4_register_cpl_handler(sc, CPL_RDMA_TERMINATE, terminate);
2404 	t4_register_fw_msg_handler(sc, FW6_TYPE_WR_RPL, fw6_wr_rpl);
2405 	t4_register_fw_msg_handler(sc, FW6_TYPE_CQE, fw6_cqe_handler);
2406 	t4_register_an_handler(sc, c4iw_ev_handler);
2407 }
2408 
2409 	void
2410 c4iw_cm_term_cpl(struct adapter *sc)
2411 {
2412 
2413 	t4_register_cpl_handler(sc, CPL_RDMA_TERMINATE, NULL);
2414 	t4_register_fw_msg_handler(sc, FW6_TYPE_WR_RPL, NULL);
2415 	t4_register_fw_msg_handler(sc, FW6_TYPE_CQE, NULL);
2416 }
2417 
2418 int __init c4iw_cm_init(void)
2419 {
2420 
2421 	TAILQ_INIT(&req_list);
2422 	spin_lock_init(&req_lock);
2423 	INIT_LIST_HEAD(&timeout_list);
2424 	spin_lock_init(&timeout_lock);
2425 
2426 	INIT_WORK(&c4iw_task, process_req);
2427 
2428 	c4iw_taskq = create_singlethread_workqueue("iw_cxgbe");
2429 	if (!c4iw_taskq)
2430 		return -ENOMEM;
2431 
2432 
2433 	return 0;
2434 }
2435 
2436 void __exit c4iw_cm_term(void)
2437 {
2438 	WARN_ON(!TAILQ_EMPTY(&req_list));
2439 	WARN_ON(!list_empty(&timeout_list));
2440 	flush_workqueue(c4iw_taskq);
2441 	destroy_workqueue(c4iw_taskq);
2442 }
2443 #endif
2444