xref: /freebsd/sys/dev/cxgbe/iw_cxgbe/cm.c (revision 0a36787e4c1fa0cf77dcf83be0867178476e372b)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 2009-2013, 2016 Chelsio, Inc. All rights reserved.
5  *
6  * This software is available to you under a choice of one of two
7  * licenses.  You may choose to be licensed under the terms of the GNU
8  * General Public License (GPL) Version 2, available from the file
9  * COPYING in the main directory of this source tree, or the
10  * OpenIB.org BSD license below:
11  *
12  *     Redistribution and use in source and binary forms, with or
13  *     without modification, are permitted provided that the following
14  *     conditions are met:
15  *
16  *      - Redistributions of source code must retain the above
17  *	  copyright notice, this list of conditions and the following
18  *	  disclaimer.
19  *
20  *      - Redistributions in binary form must reproduce the above
21  *	  copyright notice, this list of conditions and the following
22  *	  disclaimer in the documentation and/or other materials
23  *	  provided with the distribution.
24  *
25  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
26  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
27  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
28  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
29  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
30  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
31  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
32  * SOFTWARE.
33  */
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
36 
37 #include "opt_inet.h"
38 
39 #ifdef TCP_OFFLOAD
40 #include <sys/types.h>
41 #include <sys/malloc.h>
42 #include <sys/socket.h>
43 #include <sys/socketvar.h>
44 #include <sys/sockio.h>
45 #include <sys/taskqueue.h>
46 #include <netinet/in.h>
47 #include <net/route.h>
48 #include <net/route/nhop.h>
49 
50 #include <netinet/in_systm.h>
51 #include <netinet/in_pcb.h>
52 #include <netinet6/in6_pcb.h>
53 #include <netinet/ip.h>
54 #include <netinet/in_fib.h>
55 #include <netinet6/in6_fib.h>
56 #include <netinet6/scope6_var.h>
57 #include <netinet/ip_var.h>
58 #include <netinet/tcp_var.h>
59 #include <netinet/tcp.h>
60 #include <netinet/tcpip.h>
61 
62 #include <netinet/toecore.h>
63 
64 struct sge_iq;
65 struct rss_header;
66 struct cpl_set_tcb_rpl;
67 #include <linux/types.h>
68 #include "offload.h"
69 #include "tom/t4_tom.h"
70 
71 #define TOEPCB(so)  ((struct toepcb *)(so_sototcpcb((so))->t_toe))
72 
73 #include "iw_cxgbe.h"
74 #include <linux/module.h>
75 #include <linux/workqueue.h>
76 #include <linux/if_vlan.h>
77 #include <net/netevent.h>
78 #include <rdma/rdma_cm.h>
79 
80 static spinlock_t req_lock;
81 static TAILQ_HEAD(c4iw_ep_list, c4iw_ep_common) req_list;
82 static struct work_struct c4iw_task;
83 static struct workqueue_struct *c4iw_taskq;
84 static LIST_HEAD(err_cqe_list);
85 static spinlock_t err_cqe_lock;
86 static LIST_HEAD(listen_port_list);
87 static DEFINE_MUTEX(listen_port_mutex);
88 
89 static void process_req(struct work_struct *ctx);
90 static void start_ep_timer(struct c4iw_ep *ep);
91 static int stop_ep_timer(struct c4iw_ep *ep);
92 static int set_tcpinfo(struct c4iw_ep *ep);
93 static void process_timeout(struct c4iw_ep *ep);
94 static void process_err_cqes(void);
95 static void *alloc_ep(int size, gfp_t flags);
96 static void close_socket(struct socket *so);
97 static int send_mpa_req(struct c4iw_ep *ep);
98 static int send_mpa_reject(struct c4iw_ep *ep, const void *pdata, u8 plen);
99 static int send_mpa_reply(struct c4iw_ep *ep, const void *pdata, u8 plen);
100 static void close_complete_upcall(struct c4iw_ep *ep, int status);
101 static int send_abort(struct c4iw_ep *ep);
102 static void peer_close_upcall(struct c4iw_ep *ep);
103 static void peer_abort_upcall(struct c4iw_ep *ep);
104 static void connect_reply_upcall(struct c4iw_ep *ep, int status);
105 static int connect_request_upcall(struct c4iw_ep *ep);
106 static void established_upcall(struct c4iw_ep *ep);
107 static int process_mpa_reply(struct c4iw_ep *ep);
108 static int process_mpa_request(struct c4iw_ep *ep);
109 static void process_peer_close(struct c4iw_ep *ep);
110 static void process_conn_error(struct c4iw_ep *ep);
111 static void process_close_complete(struct c4iw_ep *ep);
112 static void ep_timeout(unsigned long arg);
113 static void setiwsockopt(struct socket *so);
114 static void init_iwarp_socket(struct socket *so, void *arg);
115 static void uninit_iwarp_socket(struct socket *so);
116 static void process_data(struct c4iw_ep *ep);
117 static void process_connected(struct c4iw_ep *ep);
118 static int c4iw_so_upcall(struct socket *so, void *arg, int waitflag);
119 static void process_socket_event(struct c4iw_ep *ep);
120 static void release_ep_resources(struct c4iw_ep *ep);
121 static int process_terminate(struct c4iw_ep *ep);
122 static int terminate(struct sge_iq *iq, const struct rss_header *rss,
123     struct mbuf *m);
124 static int add_ep_to_req_list(struct c4iw_ep *ep, int ep_events);
125 static struct listen_port_info *
126 add_ep_to_listenlist(struct c4iw_listen_ep *lep);
127 static int rem_ep_from_listenlist(struct c4iw_listen_ep *lep);
128 static struct c4iw_listen_ep *
129 find_real_listen_ep(struct c4iw_listen_ep *master_lep, struct socket *so);
130 static int get_ifnet_from_raddr(struct sockaddr_storage *raddr,
131 		struct ifnet **ifp);
132 static void process_newconn(struct c4iw_listen_ep *master_lep,
133 		struct socket *new_so);
134 #define START_EP_TIMER(ep) \
135     do { \
136 	    CTR3(KTR_IW_CXGBE, "start_ep_timer (%s:%d) ep %p", \
137 		__func__, __LINE__, (ep)); \
138 	    start_ep_timer(ep); \
139     } while (0)
140 
141 #define STOP_EP_TIMER(ep) \
142     ({ \
143 	    CTR3(KTR_IW_CXGBE, "stop_ep_timer (%s:%d) ep %p", \
144 		__func__, __LINE__, (ep)); \
145 	    stop_ep_timer(ep); \
146     })
147 
148 #define GET_LOCAL_ADDR(pladdr, so) \
149 	do { \
150 		struct sockaddr_storage *__a = NULL; \
151 		struct  inpcb *__inp = sotoinpcb(so); \
152 		KASSERT(__inp != NULL, \
153 		   ("GET_LOCAL_ADDR(%s):so:%p, inp = NULL", __func__, so)); \
154 		if (__inp->inp_vflag & INP_IPV4) \
155 			in_getsockaddr(so, (struct sockaddr **)&__a); \
156 		else \
157 			in6_getsockaddr(so, (struct sockaddr **)&__a); \
158 		*(pladdr) = *__a; \
159 		free(__a, M_SONAME); \
160 	} while (0)
161 
162 #define GET_REMOTE_ADDR(praddr, so) \
163 	do { \
164 		struct sockaddr_storage *__a = NULL; \
165 		struct  inpcb *__inp = sotoinpcb(so); \
166 		KASSERT(__inp != NULL, \
167 		   ("GET_REMOTE_ADDR(%s):so:%p, inp = NULL", __func__, so)); \
168 		if (__inp->inp_vflag & INP_IPV4) \
169 			in_getpeeraddr(so, (struct sockaddr **)&__a); \
170 		else \
171 			in6_getpeeraddr(so, (struct sockaddr **)&__a); \
172 		*(praddr) = *__a; \
173 		free(__a, M_SONAME); \
174 	} while (0)
175 
176 static char *states[] = {
177 	"idle",
178 	"listen",
179 	"connecting",
180 	"mpa_wait_req",
181 	"mpa_req_sent",
182 	"mpa_req_rcvd",
183 	"mpa_rep_sent",
184 	"fpdu_mode",
185 	"aborting",
186 	"closing",
187 	"moribund",
188 	"dead",
189 	NULL,
190 };
191 
192 static void deref_cm_id(struct c4iw_ep_common *epc)
193 {
194       epc->cm_id->rem_ref(epc->cm_id);
195       epc->cm_id = NULL;
196       set_bit(CM_ID_DEREFED, &epc->history);
197 }
198 
199 static void ref_cm_id(struct c4iw_ep_common *epc)
200 {
201       set_bit(CM_ID_REFED, &epc->history);
202       epc->cm_id->add_ref(epc->cm_id);
203 }
204 
205 static void deref_qp(struct c4iw_ep *ep)
206 {
207 	c4iw_qp_rem_ref(&ep->com.qp->ibqp);
208 	clear_bit(QP_REFERENCED, &ep->com.flags);
209 	set_bit(QP_DEREFED, &ep->com.history);
210 }
211 
212 static void ref_qp(struct c4iw_ep *ep)
213 {
214 	set_bit(QP_REFERENCED, &ep->com.flags);
215 	set_bit(QP_REFED, &ep->com.history);
216 	c4iw_qp_add_ref(&ep->com.qp->ibqp);
217 }
218 /* allocated per TCP port while listening */
219 struct listen_port_info {
220 	uint16_t port_num; /* TCP port address */
221 	struct list_head list; /* belongs to listen_port_list */
222 	struct list_head lep_list; /* per port lep list */
223 	uint32_t refcnt; /* number of lep's listening */
224 };
225 
226 /*
227  * Following two lists are used to manage INADDR_ANY listeners:
228  * 1)listen_port_list
229  * 2)lep_list
230  *
231  * Below is the INADDR_ANY listener lists overview on a system with a two port
232  * adapter:
233  *   |------------------|
234  *   |listen_port_list  |
235  *   |------------------|
236  *            |
237  *            |              |-----------|       |-----------|
238  *            |              | port_num:X|       | port_num:X|
239  *            |--------------|-list------|-------|-list------|-------....
240  *                           | lep_list----|     | lep_list----|
241  *                           | refcnt    | |     | refcnt    | |
242  *                           |           | |     |           | |
243  *                           |           | |     |           | |
244  *                           |-----------| |     |-----------| |
245  *                                         |                   |
246  *                                         |                   |
247  *                                         |                   |
248  *                                         |                   |         lep1                  lep2
249  *                                         |                   |    |----------------|    |----------------|
250  *                                         |                   |----| listen_ep_list |----| listen_ep_list |
251  *                                         |                        |----------------|    |----------------|
252  *                                         |
253  *                                         |
254  *                                         |        lep1                  lep2
255  *                                         |   |----------------|    |----------------|
256  *                                         |---| listen_ep_list |----| listen_ep_list |
257  *                                             |----------------|    |----------------|
258  *
259  * Because of two port adapter, the number of lep's are two(lep1 & lep2) for
260  * each TCP port number.
261  *
262  * Here 'lep1' is always marked as Master lep, because solisten() is always
263  * called through first lep.
264  *
265  */
266 static struct listen_port_info *
267 add_ep_to_listenlist(struct c4iw_listen_ep *lep)
268 {
269 	uint16_t port;
270 	struct listen_port_info *port_info = NULL;
271 	struct sockaddr_storage *laddr = &lep->com.local_addr;
272 
273 	port = (laddr->ss_family == AF_INET) ?
274 		((struct sockaddr_in *)laddr)->sin_port :
275 		((struct sockaddr_in6 *)laddr)->sin6_port;
276 
277 	mutex_lock(&listen_port_mutex);
278 
279 	list_for_each_entry(port_info, &listen_port_list, list)
280 		if (port_info->port_num == port)
281 			goto found_port;
282 
283 	port_info = malloc(sizeof(*port_info), M_CXGBE, M_WAITOK);
284 	port_info->port_num = port;
285 	port_info->refcnt    = 0;
286 
287 	list_add_tail(&port_info->list, &listen_port_list);
288 	INIT_LIST_HEAD(&port_info->lep_list);
289 
290 found_port:
291 	port_info->refcnt++;
292 	list_add_tail(&lep->listen_ep_list, &port_info->lep_list);
293 	mutex_unlock(&listen_port_mutex);
294 	return port_info;
295 }
296 
297 static int
298 rem_ep_from_listenlist(struct c4iw_listen_ep *lep)
299 {
300 	uint16_t port;
301 	struct listen_port_info *port_info = NULL;
302 	struct sockaddr_storage *laddr = &lep->com.local_addr;
303 	int refcnt = 0;
304 
305 	port = (laddr->ss_family == AF_INET) ?
306 		((struct sockaddr_in *)laddr)->sin_port :
307 		((struct sockaddr_in6 *)laddr)->sin6_port;
308 
309 	mutex_lock(&listen_port_mutex);
310 
311 	/* get the port_info structure based on the lep's port address */
312 	list_for_each_entry(port_info, &listen_port_list, list) {
313 		if (port_info->port_num == port) {
314 			port_info->refcnt--;
315 			refcnt = port_info->refcnt;
316 			/* remove the current lep from the listen list */
317 			list_del(&lep->listen_ep_list);
318 			if (port_info->refcnt == 0) {
319 				/* Remove this entry from the list as there
320 				 * are no more listeners for this port_num.
321 				 */
322 				list_del(&port_info->list);
323 				kfree(port_info);
324 			}
325 			break;
326 		}
327 	}
328 	mutex_unlock(&listen_port_mutex);
329 	return refcnt;
330 }
331 
332 /*
333  * Find the lep that belongs to the ifnet on which the SYN frame was received.
334  */
335 struct c4iw_listen_ep *
336 find_real_listen_ep(struct c4iw_listen_ep *master_lep, struct socket *so)
337 {
338 	struct adapter *adap = NULL;
339 	struct c4iw_listen_ep *lep = NULL;
340 	struct ifnet *ifp = NULL, *hw_ifp = NULL;
341 	struct listen_port_info *port_info = NULL;
342 	int i = 0, found_portinfo = 0, found_lep = 0;
343 	uint16_t port;
344 
345 	/*
346 	 * STEP 1: Figure out 'ifp' of the physical interface, not pseudo
347 	 * interfaces like vlan, lagg, etc..
348 	 * TBD: lagg support, lagg + vlan support.
349 	 */
350 	ifp = TOEPCB(so)->l2te->ifp;
351 	if (ifp->if_type == IFT_L2VLAN) {
352 		hw_ifp = VLAN_TRUNKDEV(ifp);
353 		if (hw_ifp == NULL) {
354 			CTR4(KTR_IW_CXGBE, "%s: Failed to get parent ifnet of "
355 				"vlan ifnet %p, sock %p, master_lep %p",
356 				__func__, ifp, so, master_lep);
357 			return (NULL);
358 		}
359 	} else
360 		hw_ifp = ifp;
361 
362 	/* STEP 2: Find 'port_info' with listener local port address. */
363 	port = (master_lep->com.local_addr.ss_family == AF_INET) ?
364 		((struct sockaddr_in *)&master_lep->com.local_addr)->sin_port :
365 		((struct sockaddr_in6 *)&master_lep->com.local_addr)->sin6_port;
366 
367 
368 	mutex_lock(&listen_port_mutex);
369 	list_for_each_entry(port_info, &listen_port_list, list)
370 		if (port_info->port_num == port) {
371 			found_portinfo =1;
372 			break;
373 		}
374 	if (!found_portinfo)
375 		goto out;
376 
377 	/* STEP 3: Traverse through list of lep's that are bound to the current
378 	 * TCP port address and find the lep that belongs to the ifnet on which
379 	 * the SYN frame was received.
380 	 */
381 	list_for_each_entry(lep, &port_info->lep_list, listen_ep_list) {
382 		adap = lep->com.dev->rdev.adap;
383 		for_each_port(adap, i) {
384 			if (hw_ifp == adap->port[i]->vi[0].ifp) {
385 				found_lep =1;
386 				goto out;
387 			}
388 		}
389 	}
390 out:
391 	mutex_unlock(&listen_port_mutex);
392 	return found_lep ? lep : (NULL);
393 }
394 
395 static void process_timeout(struct c4iw_ep *ep)
396 {
397 	struct c4iw_qp_attributes attrs = {0};
398 	int abort = 1;
399 
400 	CTR4(KTR_IW_CXGBE, "%s ep :%p, tid:%u, state %d", __func__,
401 			ep, ep->hwtid, ep->com.state);
402 	set_bit(TIMEDOUT, &ep->com.history);
403 	switch (ep->com.state) {
404 	case MPA_REQ_SENT:
405 		connect_reply_upcall(ep, -ETIMEDOUT);
406 		break;
407 	case MPA_REQ_WAIT:
408 	case MPA_REQ_RCVD:
409 	case MPA_REP_SENT:
410 	case FPDU_MODE:
411 		break;
412 	case CLOSING:
413 	case MORIBUND:
414 		if (ep->com.cm_id && ep->com.qp) {
415 			attrs.next_state = C4IW_QP_STATE_ERROR;
416 			c4iw_modify_qp(ep->com.dev, ep->com.qp,
417 					C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
418 		}
419 		close_complete_upcall(ep, -ETIMEDOUT);
420 		break;
421 	case ABORTING:
422 	case DEAD:
423 		/*
424 		 * These states are expected if the ep timed out at the same
425 		 * time as another thread was calling stop_ep_timer().
426 		 * So we silently do nothing for these states.
427 		 */
428 		abort = 0;
429 		break;
430 	default:
431 		CTR4(KTR_IW_CXGBE, "%s unexpected state ep %p tid %u state %u"
432 				, __func__, ep, ep->hwtid, ep->com.state);
433 		abort = 0;
434 	}
435 	if (abort)
436 		c4iw_ep_disconnect(ep, 1, GFP_KERNEL);
437 	c4iw_put_ep(&ep->com);
438 	return;
439 }
440 
441 struct cqe_list_entry {
442 	struct list_head entry;
443 	struct c4iw_dev *rhp;
444 	struct t4_cqe err_cqe;
445 };
446 
447 static void
448 process_err_cqes(void)
449 {
450 	unsigned long flag;
451 	struct cqe_list_entry *cle;
452 
453 	spin_lock_irqsave(&err_cqe_lock, flag);
454 	while (!list_empty(&err_cqe_list)) {
455 		struct list_head *tmp;
456 		tmp = err_cqe_list.next;
457 		list_del(tmp);
458 		tmp->next = tmp->prev = NULL;
459 		spin_unlock_irqrestore(&err_cqe_lock, flag);
460 		cle = list_entry(tmp, struct cqe_list_entry, entry);
461 		c4iw_ev_dispatch(cle->rhp, &cle->err_cqe);
462 		free(cle, M_CXGBE);
463 		spin_lock_irqsave(&err_cqe_lock, flag);
464 	}
465 	spin_unlock_irqrestore(&err_cqe_lock, flag);
466 
467 	return;
468 }
469 
470 static void
471 process_req(struct work_struct *ctx)
472 {
473 	struct c4iw_ep_common *epc;
474 	unsigned long flag;
475 	int ep_events;
476 
477 	process_err_cqes();
478 	spin_lock_irqsave(&req_lock, flag);
479 	while (!TAILQ_EMPTY(&req_list)) {
480 		epc = TAILQ_FIRST(&req_list);
481 		TAILQ_REMOVE(&req_list, epc, entry);
482 		epc->entry.tqe_prev = NULL;
483 		ep_events = epc->ep_events;
484 		epc->ep_events = 0;
485 		spin_unlock_irqrestore(&req_lock, flag);
486 		mutex_lock(&epc->mutex);
487 		CTR5(KTR_IW_CXGBE, "%s: so %p, ep %p, ep_state %s events 0x%x",
488 		    __func__, epc->so, epc, states[epc->state], ep_events);
489 		if (ep_events & C4IW_EVENT_TERM)
490 			process_terminate((struct c4iw_ep *)epc);
491 		if (ep_events & C4IW_EVENT_TIMEOUT)
492 			process_timeout((struct c4iw_ep *)epc);
493 		if (ep_events & C4IW_EVENT_SOCKET)
494 			process_socket_event((struct c4iw_ep *)epc);
495 		mutex_unlock(&epc->mutex);
496 		c4iw_put_ep(epc);
497 		process_err_cqes();
498 		spin_lock_irqsave(&req_lock, flag);
499 	}
500 	spin_unlock_irqrestore(&req_lock, flag);
501 }
502 
503 /*
504  * XXX: doesn't belong here in the iWARP driver.
505  * XXX: assumes that the connection was offloaded by cxgbe/t4_tom if TF_TOE is
506  *      set.  Is this a valid assumption for active open?
507  */
508 static int
509 set_tcpinfo(struct c4iw_ep *ep)
510 {
511 	struct socket *so = ep->com.so;
512 	struct inpcb *inp = sotoinpcb(so);
513 	struct tcpcb *tp;
514 	struct toepcb *toep;
515 	int rc = 0;
516 
517 	INP_WLOCK(inp);
518 	tp = intotcpcb(inp);
519 	if ((tp->t_flags & TF_TOE) == 0) {
520 		rc = EINVAL;
521 		log(LOG_ERR, "%s: connection not offloaded (so %p, ep %p)\n",
522 		    __func__, so, ep);
523 		goto done;
524 	}
525 	toep = TOEPCB(so);
526 
527 	ep->hwtid = toep->tid;
528 	ep->snd_seq = tp->snd_nxt;
529 	ep->rcv_seq = tp->rcv_nxt;
530 done:
531 	INP_WUNLOCK(inp);
532 	return (rc);
533 
534 }
535 static int
536 get_ifnet_from_raddr(struct sockaddr_storage *raddr, struct ifnet **ifp)
537 {
538 	int err = 0;
539 	struct nhop_object *nh;
540 
541 	if (raddr->ss_family == AF_INET) {
542 		struct sockaddr_in *raddr4 = (struct sockaddr_in *)raddr;
543 
544 		nh = fib4_lookup(RT_DEFAULT_FIB, raddr4->sin_addr, 0,
545 				NHR_NONE, 0);
546 	} else {
547 		struct sockaddr_in6 *raddr6 = (struct sockaddr_in6 *)raddr;
548 		struct in6_addr addr6;
549 		uint32_t scopeid;
550 
551 		memset(&addr6, 0, sizeof(addr6));
552 		in6_splitscope((struct in6_addr *)&raddr6->sin6_addr,
553 					&addr6, &scopeid);
554 		nh = fib6_lookup(RT_DEFAULT_FIB, &addr6, scopeid,
555 				NHR_NONE, 0);
556 	}
557 
558 	if (nh == NULL)
559 		err = EHOSTUNREACH;
560 	else
561 		*ifp = nh->nh_ifp;
562 	CTR2(KTR_IW_CXGBE, "%s: return: %d", __func__, err);
563 	return err;
564 }
565 
566 static void
567 close_socket(struct socket *so)
568 {
569 	uninit_iwarp_socket(so);
570 	soclose(so);
571 }
572 
573 static void
574 process_peer_close(struct c4iw_ep *ep)
575 {
576 	struct c4iw_qp_attributes attrs = {0};
577 	int disconnect = 1;
578 	int release = 0;
579 
580 	CTR4(KTR_IW_CXGBE, "%s:ppcB ep %p so %p state %s", __func__, ep,
581 	    ep->com.so, states[ep->com.state]);
582 
583 	switch (ep->com.state) {
584 
585 		case MPA_REQ_WAIT:
586 			CTR2(KTR_IW_CXGBE, "%s:ppc1 %p MPA_REQ_WAIT DEAD",
587 			    __func__, ep);
588 			/* Fallthrough */
589 		case MPA_REQ_SENT:
590 			CTR2(KTR_IW_CXGBE, "%s:ppc2 %p MPA_REQ_SENT DEAD",
591 			    __func__, ep);
592 			ep->com.state = DEAD;
593 			connect_reply_upcall(ep, -ECONNABORTED);
594 
595 			disconnect = 0;
596 			STOP_EP_TIMER(ep);
597 			close_socket(ep->com.so);
598 			deref_cm_id(&ep->com);
599 			release = 1;
600 			break;
601 
602 		case MPA_REQ_RCVD:
603 
604 			/*
605 			 * We're gonna mark this puppy DEAD, but keep
606 			 * the reference on it until the ULP accepts or
607 			 * rejects the CR.
608 			 */
609 			CTR2(KTR_IW_CXGBE, "%s:ppc3 %p MPA_REQ_RCVD CLOSING",
610 			    __func__, ep);
611 			ep->com.state = CLOSING;
612 			break;
613 
614 		case MPA_REP_SENT:
615 			CTR2(KTR_IW_CXGBE, "%s:ppc4 %p MPA_REP_SENT CLOSING",
616 			    __func__, ep);
617 			ep->com.state = CLOSING;
618 			break;
619 
620 		case FPDU_MODE:
621 			CTR2(KTR_IW_CXGBE, "%s:ppc5 %p FPDU_MODE CLOSING",
622 			    __func__, ep);
623 			START_EP_TIMER(ep);
624 			ep->com.state = CLOSING;
625 			attrs.next_state = C4IW_QP_STATE_CLOSING;
626 			c4iw_modify_qp(ep->com.dev, ep->com.qp,
627 					C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
628 			peer_close_upcall(ep);
629 			break;
630 
631 		case ABORTING:
632 			CTR2(KTR_IW_CXGBE, "%s:ppc6 %p ABORTING (disconn)",
633 			    __func__, ep);
634 			disconnect = 0;
635 			break;
636 
637 		case CLOSING:
638 			CTR2(KTR_IW_CXGBE, "%s:ppc7 %p CLOSING MORIBUND",
639 			    __func__, ep);
640 			ep->com.state = MORIBUND;
641 			disconnect = 0;
642 			break;
643 
644 		case MORIBUND:
645 			CTR2(KTR_IW_CXGBE, "%s:ppc8 %p MORIBUND DEAD", __func__,
646 			    ep);
647 			STOP_EP_TIMER(ep);
648 			if (ep->com.cm_id && ep->com.qp) {
649 				attrs.next_state = C4IW_QP_STATE_IDLE;
650 				c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
651 						C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
652 			}
653 			close_socket(ep->com.so);
654 			close_complete_upcall(ep, 0);
655 			ep->com.state = DEAD;
656 			release = 1;
657 			disconnect = 0;
658 			break;
659 
660 		case DEAD:
661 			CTR2(KTR_IW_CXGBE, "%s:ppc9 %p DEAD (disconn)",
662 			    __func__, ep);
663 			disconnect = 0;
664 			break;
665 
666 		default:
667 			panic("%s: ep %p state %d", __func__, ep,
668 			    ep->com.state);
669 			break;
670 	}
671 
672 
673 	if (disconnect) {
674 
675 		CTR2(KTR_IW_CXGBE, "%s:ppca %p", __func__, ep);
676 		c4iw_ep_disconnect(ep, 0, M_NOWAIT);
677 	}
678 	if (release) {
679 
680 		CTR2(KTR_IW_CXGBE, "%s:ppcb %p", __func__, ep);
681 		c4iw_put_ep(&ep->com);
682 	}
683 	CTR2(KTR_IW_CXGBE, "%s:ppcE %p", __func__, ep);
684 	return;
685 }
686 
687 static void
688 process_conn_error(struct c4iw_ep *ep)
689 {
690 	struct c4iw_qp_attributes attrs = {0};
691 	int ret;
692 	int state;
693 
694 	state = ep->com.state;
695 	CTR5(KTR_IW_CXGBE, "%s:pceB ep %p so %p so->so_error %u state %s",
696 	    __func__, ep, ep->com.so, ep->com.so->so_error,
697 	    states[ep->com.state]);
698 
699 	switch (state) {
700 
701 		case MPA_REQ_WAIT:
702 			STOP_EP_TIMER(ep);
703 			c4iw_put_ep(&ep->parent_ep->com);
704 			break;
705 
706 		case MPA_REQ_SENT:
707 			STOP_EP_TIMER(ep);
708 			connect_reply_upcall(ep, -ECONNRESET);
709 			break;
710 
711 		case MPA_REP_SENT:
712 			ep->com.rpl_err = ECONNRESET;
713 			CTR1(KTR_IW_CXGBE, "waking up ep %p", ep);
714 			break;
715 
716 		case MPA_REQ_RCVD:
717 			break;
718 
719 		case MORIBUND:
720 		case CLOSING:
721 			STOP_EP_TIMER(ep);
722 			/*FALLTHROUGH*/
723 		case FPDU_MODE:
724 
725 			if (ep->com.cm_id && ep->com.qp) {
726 
727 				attrs.next_state = C4IW_QP_STATE_ERROR;
728 				ret = c4iw_modify_qp(ep->com.qp->rhp,
729 					ep->com.qp, C4IW_QP_ATTR_NEXT_STATE,
730 					&attrs, 1);
731 				if (ret)
732 					log(LOG_ERR,
733 							"%s - qp <- error failed!\n",
734 							__func__);
735 			}
736 			peer_abort_upcall(ep);
737 			break;
738 
739 		case ABORTING:
740 			break;
741 
742 		case DEAD:
743 			CTR2(KTR_IW_CXGBE, "%s so_error %d IN DEAD STATE!!!!",
744 			    __func__, ep->com.so->so_error);
745 			return;
746 
747 		default:
748 			panic("%s: ep %p state %d", __func__, ep, state);
749 			break;
750 	}
751 
752 	if (state != ABORTING) {
753 		close_socket(ep->com.so);
754 		ep->com.state = DEAD;
755 		c4iw_put_ep(&ep->com);
756 	}
757 	CTR2(KTR_IW_CXGBE, "%s:pceE %p", __func__, ep);
758 	return;
759 }
760 
761 static void
762 process_close_complete(struct c4iw_ep *ep)
763 {
764 	struct c4iw_qp_attributes attrs = {0};
765 	int release = 0;
766 
767 	CTR4(KTR_IW_CXGBE, "%s:pccB ep %p so %p state %s", __func__, ep,
768 	    ep->com.so, states[ep->com.state]);
769 
770 	/* The cm_id may be null if we failed to connect */
771 	set_bit(CLOSE_CON_RPL, &ep->com.history);
772 
773 	switch (ep->com.state) {
774 
775 		case CLOSING:
776 			CTR2(KTR_IW_CXGBE, "%s:pcc1 %p CLOSING MORIBUND",
777 			    __func__, ep);
778 			ep->com.state = MORIBUND;
779 			break;
780 
781 		case MORIBUND:
782 			CTR2(KTR_IW_CXGBE, "%s:pcc1 %p MORIBUND DEAD", __func__,
783 			    ep);
784 			STOP_EP_TIMER(ep);
785 
786 			if ((ep->com.cm_id) && (ep->com.qp)) {
787 
788 				CTR2(KTR_IW_CXGBE, "%s:pcc2 %p QP_STATE_IDLE",
789 				    __func__, ep);
790 				attrs.next_state = C4IW_QP_STATE_IDLE;
791 				c4iw_modify_qp(ep->com.dev,
792 						ep->com.qp,
793 						C4IW_QP_ATTR_NEXT_STATE,
794 						&attrs, 1);
795 			}
796 
797 			close_socket(ep->com.so);
798 			close_complete_upcall(ep, 0);
799 			ep->com.state = DEAD;
800 			release = 1;
801 			break;
802 
803 		case ABORTING:
804 			CTR2(KTR_IW_CXGBE, "%s:pcc5 %p ABORTING", __func__, ep);
805 			break;
806 
807 		case DEAD:
808 			CTR2(KTR_IW_CXGBE, "%s:pcc6 %p DEAD", __func__, ep);
809 			break;
810 		default:
811 			CTR2(KTR_IW_CXGBE, "%s:pcc7 %p unknown ep state",
812 					__func__, ep);
813 			panic("%s:pcc6 %p unknown ep state", __func__, ep);
814 			break;
815 	}
816 
817 	if (release) {
818 
819 		CTR2(KTR_IW_CXGBE, "%s:pcc8 %p", __func__, ep);
820 		release_ep_resources(ep);
821 	}
822 	CTR2(KTR_IW_CXGBE, "%s:pccE %p", __func__, ep);
823 	return;
824 }
825 
826 static void
827 setiwsockopt(struct socket *so)
828 {
829 	int rc;
830 	struct sockopt sopt;
831 	int on = 1;
832 
833 	sopt.sopt_dir = SOPT_SET;
834 	sopt.sopt_level = IPPROTO_TCP;
835 	sopt.sopt_name = TCP_NODELAY;
836 	sopt.sopt_val = (caddr_t)&on;
837 	sopt.sopt_valsize = sizeof on;
838 	sopt.sopt_td = NULL;
839 	rc = -sosetopt(so, &sopt);
840 	if (rc) {
841 		log(LOG_ERR, "%s: can't set TCP_NODELAY on so %p (%d)\n",
842 		    __func__, so, rc);
843 	}
844 }
845 
846 static void
847 init_iwarp_socket(struct socket *so, void *arg)
848 {
849 	if (SOLISTENING(so)) {
850 		SOLISTEN_LOCK(so);
851 		solisten_upcall_set(so, c4iw_so_upcall, arg);
852 		so->so_state |= SS_NBIO;
853 		SOLISTEN_UNLOCK(so);
854 	} else {
855 		SOCKBUF_LOCK(&so->so_rcv);
856 		soupcall_set(so, SO_RCV, c4iw_so_upcall, arg);
857 		so->so_state |= SS_NBIO;
858 		SOCKBUF_UNLOCK(&so->so_rcv);
859 	}
860 }
861 
862 static void
863 uninit_iwarp_socket(struct socket *so)
864 {
865 	if (SOLISTENING(so)) {
866 		SOLISTEN_LOCK(so);
867 		solisten_upcall_set(so, NULL, NULL);
868 		SOLISTEN_UNLOCK(so);
869 	} else {
870 		SOCKBUF_LOCK(&so->so_rcv);
871 		soupcall_clear(so, SO_RCV);
872 		SOCKBUF_UNLOCK(&so->so_rcv);
873 	}
874 }
875 
876 static void
877 process_data(struct c4iw_ep *ep)
878 {
879 	int ret = 0;
880 	int disconnect = 0;
881 	struct c4iw_qp_attributes attrs = {0};
882 
883 	CTR5(KTR_IW_CXGBE, "%s: so %p, ep %p, state %s, sbused %d", __func__,
884 	    ep->com.so, ep, states[ep->com.state], sbused(&ep->com.so->so_rcv));
885 
886 	switch (ep->com.state) {
887 	case MPA_REQ_SENT:
888 		disconnect = process_mpa_reply(ep);
889 		break;
890 	case MPA_REQ_WAIT:
891 		disconnect = process_mpa_request(ep);
892 		if (disconnect)
893 			/* Refered in process_newconn() */
894 			c4iw_put_ep(&ep->parent_ep->com);
895 		break;
896 	case FPDU_MODE:
897 		MPASS(ep->com.qp != NULL);
898 		attrs.next_state = C4IW_QP_STATE_TERMINATE;
899 		ret = c4iw_modify_qp(ep->com.dev, ep->com.qp,
900 					C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
901 		if (ret != -EINPROGRESS)
902 			disconnect = 1;
903 		break;
904 	default:
905 		log(LOG_ERR, "%s: Unexpected streaming data. ep %p, "
906 			    "state %d, so %p, so_state 0x%x, sbused %u\n",
907 			    __func__, ep, ep->com.state, ep->com.so,
908 			    ep->com.so->so_state, sbused(&ep->com.so->so_rcv));
909 		break;
910 	}
911 	if (disconnect)
912 		c4iw_ep_disconnect(ep, disconnect == 2, GFP_KERNEL);
913 
914 }
915 
916 static void
917 process_connected(struct c4iw_ep *ep)
918 {
919 	struct socket *so = ep->com.so;
920 
921 	if ((so->so_state & SS_ISCONNECTED) && !so->so_error) {
922 		if (send_mpa_req(ep))
923 			goto err;
924 	} else {
925 		connect_reply_upcall(ep, -so->so_error);
926 		goto err;
927 	}
928 	return;
929 err:
930 	close_socket(so);
931 	ep->com.state = DEAD;
932 	c4iw_put_ep(&ep->com);
933 	return;
934 }
935 
936 static inline int c4iw_zero_addr(struct sockaddr *addr)
937 {
938 	struct in6_addr *ip6;
939 
940 	if (addr->sa_family == AF_INET)
941 		return IN_ZERONET(
942 			ntohl(((struct sockaddr_in *)addr)->sin_addr.s_addr));
943 	else {
944 		ip6 = &((struct sockaddr_in6 *) addr)->sin6_addr;
945 		return (ip6->s6_addr32[0] | ip6->s6_addr32[1] |
946 				ip6->s6_addr32[2] | ip6->s6_addr32[3]) == 0;
947 	}
948 }
949 
950 static inline int c4iw_loopback_addr(struct sockaddr *addr)
951 {
952 	if (addr->sa_family == AF_INET)
953 		return IN_LOOPBACK(
954 			ntohl(((struct sockaddr_in *) addr)->sin_addr.s_addr));
955 	else
956 		return IN6_IS_ADDR_LOOPBACK(
957 				&((struct sockaddr_in6 *) addr)->sin6_addr);
958 }
959 
960 static inline int c4iw_any_addr(struct sockaddr *addr)
961 {
962 	return c4iw_zero_addr(addr) || c4iw_loopback_addr(addr);
963 }
964 
965 static void
966 process_newconn(struct c4iw_listen_ep *master_lep, struct socket *new_so)
967 {
968 	struct c4iw_listen_ep *real_lep = NULL;
969 	struct c4iw_ep *new_ep = NULL;
970 	struct sockaddr_in *remote = NULL;
971 	int ret = 0;
972 
973 	MPASS(new_so != NULL);
974 
975 	if (c4iw_any_addr((struct sockaddr *)&master_lep->com.local_addr)) {
976 		/* Here we need to find the 'real_lep' that belongs to the
977 		 * incomming socket's network interface, such that the newly
978 		 * created 'ep' can be attached to the real 'lep'.
979 		 */
980 		real_lep = find_real_listen_ep(master_lep, new_so);
981 		if (real_lep == NULL) {
982 			CTR2(KTR_IW_CXGBE, "%s: Could not find the real listen "
983 					"ep for sock: %p", __func__, new_so);
984 			log(LOG_ERR,"%s: Could not find the real listen ep for "
985 					"sock: %p\n", __func__, new_so);
986 			/* FIXME: properly free the 'new_so' in failure case.
987 			 * Use of soabort() and  soclose() are not legal
988 			 * here(before soaccept()).
989 			 */
990 			return;
991 		}
992 	} else /* for Non-Wildcard address, master_lep is always the real_lep */
993 		real_lep = master_lep;
994 
995 	new_ep = alloc_ep(sizeof(*new_ep), GFP_KERNEL);
996 
997 	CTR6(KTR_IW_CXGBE, "%s: master_lep %p, real_lep: %p, new ep %p, "
998 	    "listening so %p, new so %p", __func__, master_lep, real_lep,
999 	    new_ep, master_lep->com.so, new_so);
1000 
1001 	new_ep->com.dev = real_lep->com.dev;
1002 	new_ep->com.so = new_so;
1003 	new_ep->com.cm_id = NULL;
1004 	new_ep->com.thread = real_lep->com.thread;
1005 	new_ep->parent_ep = real_lep;
1006 
1007 	GET_LOCAL_ADDR(&new_ep->com.local_addr, new_so);
1008 	GET_REMOTE_ADDR(&new_ep->com.remote_addr, new_so);
1009 	c4iw_get_ep(&real_lep->com);
1010 	init_timer(&new_ep->timer);
1011 	new_ep->com.state = MPA_REQ_WAIT;
1012 
1013 	setiwsockopt(new_so);
1014 	ret = soaccept(new_so, (struct sockaddr **)&remote);
1015 	if (ret != 0) {
1016 		CTR4(KTR_IW_CXGBE,
1017 				"%s:listen sock:%p, new sock:%p, ret:%d",
1018 				__func__, master_lep->com.so, new_so, ret);
1019 		if (remote != NULL)
1020 			free(remote, M_SONAME);
1021 		soclose(new_so);
1022 		c4iw_put_ep(&new_ep->com);
1023 		c4iw_put_ep(&real_lep->com);
1024 		return;
1025 	}
1026 	free(remote, M_SONAME);
1027 
1028 	START_EP_TIMER(new_ep);
1029 
1030 	/* MPA request might have been queued up on the socket already, so we
1031 	 * initialize the socket/upcall_handler under lock to prevent processing
1032 	 * MPA request on another thread(via process_req()) simultaniously.
1033 	 */
1034 	c4iw_get_ep(&new_ep->com); /* Dereferenced at the end below, this is to
1035 				      avoid freeing of ep before ep unlock. */
1036 	mutex_lock(&new_ep->com.mutex);
1037 	init_iwarp_socket(new_so, &new_ep->com);
1038 
1039 	ret = process_mpa_request(new_ep);
1040 	if (ret) {
1041 		/* ABORT */
1042 		c4iw_ep_disconnect(new_ep, 1, GFP_KERNEL);
1043 		c4iw_put_ep(&real_lep->com);
1044 	}
1045 	mutex_unlock(&new_ep->com.mutex);
1046 	c4iw_put_ep(&new_ep->com);
1047 	return;
1048 }
1049 
1050 static int
1051 add_ep_to_req_list(struct c4iw_ep *ep, int new_ep_event)
1052 {
1053 	unsigned long flag;
1054 
1055 	spin_lock_irqsave(&req_lock, flag);
1056 	if (ep && ep->com.so) {
1057 		ep->com.ep_events |= new_ep_event;
1058 		if (!ep->com.entry.tqe_prev) {
1059 			c4iw_get_ep(&ep->com);
1060 			TAILQ_INSERT_TAIL(&req_list, &ep->com, entry);
1061 			queue_work(c4iw_taskq, &c4iw_task);
1062 		}
1063 	}
1064 	spin_unlock_irqrestore(&req_lock, flag);
1065 
1066 	return (0);
1067 }
1068 
1069 static int
1070 c4iw_so_upcall(struct socket *so, void *arg, int waitflag)
1071 {
1072 	struct c4iw_ep *ep = arg;
1073 
1074 	CTR6(KTR_IW_CXGBE,
1075 	    "%s: so %p, so_state 0x%x, ep %p, ep_state %s, tqe_prev %p",
1076 	    __func__, so, so->so_state, ep, states[ep->com.state],
1077 	    ep->com.entry.tqe_prev);
1078 
1079 	MPASS(ep->com.so == so);
1080 	/*
1081 	 * Wake up any threads waiting in rdma_init()/rdma_fini(),
1082 	 * with locks held.
1083 	 */
1084 	if (so->so_error || (ep->com.dev->rdev.flags & T4_FATAL_ERROR))
1085 		c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
1086 	add_ep_to_req_list(ep, C4IW_EVENT_SOCKET);
1087 
1088 	return (SU_OK);
1089 }
1090 
1091 
1092 static int
1093 terminate(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m)
1094 {
1095 	struct adapter *sc = iq->adapter;
1096 	const struct cpl_rdma_terminate *cpl = mtod(m, const void *);
1097 	unsigned int tid = GET_TID(cpl);
1098 	struct toepcb *toep = lookup_tid(sc, tid);
1099 	struct socket *so;
1100 	struct c4iw_ep *ep;
1101 
1102 	INP_WLOCK(toep->inp);
1103 	so = inp_inpcbtosocket(toep->inp);
1104 	ep = so->so_rcv.sb_upcallarg;
1105 	INP_WUNLOCK(toep->inp);
1106 
1107 	CTR3(KTR_IW_CXGBE, "%s: so %p, ep %p", __func__, so, ep);
1108 	add_ep_to_req_list(ep, C4IW_EVENT_TERM);
1109 
1110 	return 0;
1111 }
1112 
1113 static void
1114 process_socket_event(struct c4iw_ep *ep)
1115 {
1116 	int state = ep->com.state;
1117 	struct socket *so = ep->com.so;
1118 
1119 	if (ep->com.state == DEAD) {
1120 		CTR3(KTR_IW_CXGBE, "%s: Pending socket event discarded "
1121 			"ep %p ep_state %s", __func__, ep, states[state]);
1122 		return;
1123 	}
1124 
1125 	CTR6(KTR_IW_CXGBE, "process_socket_event: so %p, so_state 0x%x, "
1126 	    "so_err %d, sb_state 0x%x, ep %p, ep_state %s", so, so->so_state,
1127 	    so->so_error, so->so_rcv.sb_state, ep, states[state]);
1128 
1129 	if (state == CONNECTING) {
1130 		process_connected(ep);
1131 		return;
1132 	}
1133 
1134 	if (state == LISTEN) {
1135 		struct c4iw_listen_ep *lep = (struct c4iw_listen_ep *)ep;
1136 		struct socket *listen_so = so, *new_so = NULL;
1137 		int error = 0;
1138 
1139 		SOLISTEN_LOCK(listen_so);
1140 		do {
1141 			error = solisten_dequeue(listen_so, &new_so,
1142 						SOCK_NONBLOCK);
1143 			if (error) {
1144 				CTR4(KTR_IW_CXGBE, "%s: lep %p listen_so %p "
1145 					"error %d", __func__, lep, listen_so,
1146 					error);
1147 				return;
1148 			}
1149 			process_newconn(lep, new_so);
1150 
1151 			/* solisten_dequeue() unlocks while return, so aquire
1152 			 * lock again for sol_qlen and also for next iteration.
1153 			 */
1154 			SOLISTEN_LOCK(listen_so);
1155 		} while (listen_so->sol_qlen);
1156 		SOLISTEN_UNLOCK(listen_so);
1157 
1158 		return;
1159 	}
1160 
1161 	/* connection error */
1162 	if (so->so_error) {
1163 		process_conn_error(ep);
1164 		return;
1165 	}
1166 
1167 	/* peer close */
1168 	if ((so->so_rcv.sb_state & SBS_CANTRCVMORE) && state <= CLOSING) {
1169 		process_peer_close(ep);
1170 		/*
1171 		 * check whether socket disconnect event is pending before
1172 		 * returning. Fallthrough if yes.
1173 		 */
1174 		if (!(so->so_state & SS_ISDISCONNECTED))
1175 			return;
1176 	}
1177 
1178 	/* close complete */
1179 	if (so->so_state & SS_ISDISCONNECTED) {
1180 		process_close_complete(ep);
1181 		return;
1182 	}
1183 
1184 	/* rx data */
1185 	if (sbused(&ep->com.so->so_rcv)) {
1186 		process_data(ep);
1187 		return;
1188 	}
1189 
1190 	/* Socket events for 'MPA Request Received' and 'Close Complete'
1191 	 * were already processed earlier in their previous events handlers.
1192 	 * Hence, these socket events are skipped.
1193 	 * And any other socket events must have handled above.
1194 	 */
1195 	MPASS((ep->com.state == MPA_REQ_RCVD) || (ep->com.state == MORIBUND));
1196 
1197 	if ((ep->com.state != MPA_REQ_RCVD) && (ep->com.state != MORIBUND))
1198 		log(LOG_ERR, "%s: Unprocessed socket event so %p, "
1199 		"so_state 0x%x, so_err %d, sb_state 0x%x, ep %p, ep_state %s\n",
1200 		__func__, so, so->so_state, so->so_error, so->so_rcv.sb_state,
1201 			ep, states[state]);
1202 
1203 }
1204 
1205 SYSCTL_NODE(_hw, OID_AUTO, iw_cxgbe, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
1206     "iw_cxgbe driver parameters");
1207 
1208 static int dack_mode = 0;
1209 SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, dack_mode, CTLFLAG_RWTUN, &dack_mode, 0,
1210 		"Delayed ack mode (default = 0)");
1211 
1212 int c4iw_max_read_depth = 8;
1213 SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, c4iw_max_read_depth, CTLFLAG_RWTUN, &c4iw_max_read_depth, 0,
1214 		"Per-connection max ORD/IRD (default = 8)");
1215 
1216 static int enable_tcp_timestamps;
1217 SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, enable_tcp_timestamps, CTLFLAG_RWTUN, &enable_tcp_timestamps, 0,
1218 		"Enable tcp timestamps (default = 0)");
1219 
1220 static int enable_tcp_sack;
1221 SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, enable_tcp_sack, CTLFLAG_RWTUN, &enable_tcp_sack, 0,
1222 		"Enable tcp SACK (default = 0)");
1223 
1224 static int enable_tcp_window_scaling = 1;
1225 SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, enable_tcp_window_scaling, CTLFLAG_RWTUN, &enable_tcp_window_scaling, 0,
1226 		"Enable tcp window scaling (default = 1)");
1227 
1228 int c4iw_debug = 0;
1229 SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, c4iw_debug, CTLFLAG_RWTUN, &c4iw_debug, 0,
1230 		"Enable debug logging (default = 0)");
1231 
1232 static int peer2peer = 1;
1233 SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, peer2peer, CTLFLAG_RWTUN, &peer2peer, 0,
1234 		"Support peer2peer ULPs (default = 1)");
1235 
1236 static int p2p_type = FW_RI_INIT_P2PTYPE_READ_REQ;
1237 SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, p2p_type, CTLFLAG_RWTUN, &p2p_type, 0,
1238 		"RDMAP opcode to use for the RTR message: 1 = RDMA_READ 0 = RDMA_WRITE (default 1)");
1239 
1240 static int ep_timeout_secs = 60;
1241 SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, ep_timeout_secs, CTLFLAG_RWTUN, &ep_timeout_secs, 0,
1242 		"CM Endpoint operation timeout in seconds (default = 60)");
1243 
1244 static int mpa_rev = 1;
1245 SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, mpa_rev, CTLFLAG_RWTUN, &mpa_rev, 0,
1246 		"MPA Revision, 0 supports amso1100, 1 is RFC5044 spec compliant, 2 is IETF MPA Peer Connect Draft compliant (default = 1)");
1247 
1248 static int markers_enabled;
1249 SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, markers_enabled, CTLFLAG_RWTUN, &markers_enabled, 0,
1250 		"Enable MPA MARKERS (default(0) = disabled)");
1251 
1252 static int crc_enabled = 1;
1253 SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, crc_enabled, CTLFLAG_RWTUN, &crc_enabled, 0,
1254 		"Enable MPA CRC (default(1) = enabled)");
1255 
1256 static int rcv_win = 256 * 1024;
1257 SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, rcv_win, CTLFLAG_RWTUN, &rcv_win, 0,
1258 		"TCP receive window in bytes (default = 256KB)");
1259 
1260 static int snd_win = 128 * 1024;
1261 SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, snd_win, CTLFLAG_RWTUN, &snd_win, 0,
1262 		"TCP send window in bytes (default = 128KB)");
1263 
1264 int use_dsgl = 1;
1265 SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, use_dsgl, CTLFLAG_RWTUN, &use_dsgl, 0,
1266 		"Use DSGL for PBL/FastReg (default=1)");
1267 
1268 int inline_threshold = 128;
1269 SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, inline_threshold, CTLFLAG_RWTUN, &inline_threshold, 0,
1270 		"inline vs dsgl threshold (default=128)");
1271 
1272 static int reuseaddr = 0;
1273 SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, reuseaddr, CTLFLAG_RWTUN, &reuseaddr, 0,
1274 		"Enable SO_REUSEADDR & SO_REUSEPORT socket options on all iWARP client connections(default = 0)");
1275 
1276 static void
1277 start_ep_timer(struct c4iw_ep *ep)
1278 {
1279 
1280 	if (timer_pending(&ep->timer)) {
1281 		CTR2(KTR_IW_CXGBE, "%s: ep %p, already started", __func__, ep);
1282 		printk(KERN_ERR "%s timer already started! ep %p\n", __func__,
1283 		    ep);
1284 		return;
1285 	}
1286 	clear_bit(TIMEOUT, &ep->com.flags);
1287 	c4iw_get_ep(&ep->com);
1288 	ep->timer.expires = jiffies + ep_timeout_secs * HZ;
1289 	ep->timer.data = (unsigned long)ep;
1290 	ep->timer.function = ep_timeout;
1291 	add_timer(&ep->timer);
1292 }
1293 
1294 static int
1295 stop_ep_timer(struct c4iw_ep *ep)
1296 {
1297 
1298 	del_timer_sync(&ep->timer);
1299 	if (!test_and_set_bit(TIMEOUT, &ep->com.flags)) {
1300 		c4iw_put_ep(&ep->com);
1301 		return 0;
1302 	}
1303 	return 1;
1304 }
1305 
1306 static void *
1307 alloc_ep(int size, gfp_t gfp)
1308 {
1309 	struct c4iw_ep_common *epc;
1310 
1311 	epc = kzalloc(size, gfp);
1312 	if (epc == NULL)
1313 		return (NULL);
1314 
1315 	kref_init(&epc->kref);
1316 	mutex_init(&epc->mutex);
1317 	c4iw_init_wr_wait(&epc->wr_wait);
1318 
1319 	return (epc);
1320 }
1321 
1322 void _c4iw_free_ep(struct kref *kref)
1323 {
1324 	struct c4iw_ep *ep;
1325 	struct c4iw_ep_common *epc;
1326 
1327 	ep = container_of(kref, struct c4iw_ep, com.kref);
1328 	epc = &ep->com;
1329 	KASSERT(!epc->entry.tqe_prev, ("%s epc %p still on req list",
1330 	    __func__, epc));
1331 	if (test_bit(QP_REFERENCED, &ep->com.flags))
1332 		deref_qp(ep);
1333 	CTR4(KTR_IW_CXGBE, "%s: ep %p, history 0x%lx, flags 0x%lx",
1334 	    __func__, ep, epc->history, epc->flags);
1335 	kfree(ep);
1336 }
1337 
1338 static void release_ep_resources(struct c4iw_ep *ep)
1339 {
1340 	CTR2(KTR_IW_CXGBE, "%s:rerB %p", __func__, ep);
1341 	set_bit(RELEASE_RESOURCES, &ep->com.flags);
1342 	c4iw_put_ep(&ep->com);
1343 	CTR2(KTR_IW_CXGBE, "%s:rerE %p", __func__, ep);
1344 }
1345 
1346 static int
1347 send_mpa_req(struct c4iw_ep *ep)
1348 {
1349 	int mpalen;
1350 	struct mpa_message *mpa;
1351 	struct mpa_v2_conn_params mpa_v2_params;
1352 	struct mbuf *m;
1353 	char mpa_rev_to_use = mpa_rev;
1354 	int err = 0;
1355 
1356 	if (ep->retry_with_mpa_v1)
1357 		mpa_rev_to_use = 1;
1358 	mpalen = sizeof(*mpa) + ep->plen;
1359 	if (mpa_rev_to_use == 2)
1360 		mpalen += sizeof(struct mpa_v2_conn_params);
1361 
1362 	mpa = malloc(mpalen, M_CXGBE, M_NOWAIT);
1363 	if (mpa == NULL) {
1364 		err = -ENOMEM;
1365 		CTR3(KTR_IW_CXGBE, "%s:smr1 ep: %p , error: %d",
1366 				__func__, ep, err);
1367 		goto err;
1368 	}
1369 
1370 	memset(mpa, 0, mpalen);
1371 	memcpy(mpa->key, MPA_KEY_REQ, sizeof(mpa->key));
1372 	mpa->flags = (crc_enabled ? MPA_CRC : 0) |
1373 		(markers_enabled ? MPA_MARKERS : 0) |
1374 		(mpa_rev_to_use == 2 ? MPA_ENHANCED_RDMA_CONN : 0);
1375 	mpa->private_data_size = htons(ep->plen);
1376 	mpa->revision = mpa_rev_to_use;
1377 
1378 	if (mpa_rev_to_use == 1) {
1379 		ep->tried_with_mpa_v1 = 1;
1380 		ep->retry_with_mpa_v1 = 0;
1381 	}
1382 
1383 	if (mpa_rev_to_use == 2) {
1384 		mpa->private_data_size = htons(ntohs(mpa->private_data_size) +
1385 					    sizeof(struct mpa_v2_conn_params));
1386 		mpa_v2_params.ird = htons((u16)ep->ird);
1387 		mpa_v2_params.ord = htons((u16)ep->ord);
1388 
1389 		if (peer2peer) {
1390 			mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL);
1391 
1392 			if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE) {
1393 				mpa_v2_params.ord |=
1394 				    htons(MPA_V2_RDMA_WRITE_RTR);
1395 			} else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ) {
1396 				mpa_v2_params.ord |=
1397 					htons(MPA_V2_RDMA_READ_RTR);
1398 			}
1399 		}
1400 		memcpy(mpa->private_data, &mpa_v2_params,
1401 			sizeof(struct mpa_v2_conn_params));
1402 
1403 		if (ep->plen) {
1404 
1405 			memcpy(mpa->private_data +
1406 				sizeof(struct mpa_v2_conn_params),
1407 				ep->mpa_pkt + sizeof(*mpa), ep->plen);
1408 		}
1409 	} else {
1410 
1411 		if (ep->plen)
1412 			memcpy(mpa->private_data,
1413 					ep->mpa_pkt + sizeof(*mpa), ep->plen);
1414 		CTR2(KTR_IW_CXGBE, "%s:smr7 %p", __func__, ep);
1415 	}
1416 
1417 	m = m_getm(NULL, mpalen, M_NOWAIT, MT_DATA);
1418 	if (m == NULL) {
1419 		err = -ENOMEM;
1420 		CTR3(KTR_IW_CXGBE, "%s:smr2 ep: %p , error: %d",
1421 				__func__, ep, err);
1422 		free(mpa, M_CXGBE);
1423 		goto err;
1424 	}
1425 	m_copyback(m, 0, mpalen, (void *)mpa);
1426 	free(mpa, M_CXGBE);
1427 
1428 	err = -sosend(ep->com.so, NULL, NULL, m, NULL, MSG_DONTWAIT,
1429 			ep->com.thread);
1430 	if (err) {
1431 		CTR3(KTR_IW_CXGBE, "%s:smr3 ep: %p , error: %d",
1432 				__func__, ep, err);
1433 		goto err;
1434 	}
1435 
1436 	START_EP_TIMER(ep);
1437 	ep->com.state = MPA_REQ_SENT;
1438 	ep->mpa_attr.initiator = 1;
1439 	CTR3(KTR_IW_CXGBE, "%s:smrE %p, error: %d", __func__, ep, err);
1440 	return 0;
1441 err:
1442 	connect_reply_upcall(ep, err);
1443 	CTR3(KTR_IW_CXGBE, "%s:smrE %p, error: %d", __func__, ep, err);
1444 	return err;
1445 }
1446 
1447 static int send_mpa_reject(struct c4iw_ep *ep, const void *pdata, u8 plen)
1448 {
1449 	int mpalen ;
1450 	struct mpa_message *mpa;
1451 	struct mpa_v2_conn_params mpa_v2_params;
1452 	struct mbuf *m;
1453 	int err;
1454 
1455 	CTR4(KTR_IW_CXGBE, "%s:smrejB %p %u %d", __func__, ep, ep->hwtid,
1456 	    ep->plen);
1457 
1458 	mpalen = sizeof(*mpa) + plen;
1459 
1460 	if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
1461 
1462 		mpalen += sizeof(struct mpa_v2_conn_params);
1463 		CTR4(KTR_IW_CXGBE, "%s:smrej1 %p %u %d", __func__, ep,
1464 		    ep->mpa_attr.version, mpalen);
1465 	}
1466 
1467 	mpa = malloc(mpalen, M_CXGBE, M_NOWAIT);
1468 	if (mpa == NULL)
1469 		return (-ENOMEM);
1470 
1471 	memset(mpa, 0, mpalen);
1472 	memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
1473 	mpa->flags = MPA_REJECT;
1474 	mpa->revision = mpa_rev;
1475 	mpa->private_data_size = htons(plen);
1476 
1477 	if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
1478 
1479 		mpa->flags |= MPA_ENHANCED_RDMA_CONN;
1480 		mpa->private_data_size = htons(ntohs(mpa->private_data_size) +
1481 					    sizeof(struct mpa_v2_conn_params));
1482 		mpa_v2_params.ird = htons(((u16)ep->ird) |
1483 				(peer2peer ? MPA_V2_PEER2PEER_MODEL :
1484 				 0));
1485 		mpa_v2_params.ord = htons(((u16)ep->ord) | (peer2peer ?
1486 					(p2p_type ==
1487 					 FW_RI_INIT_P2PTYPE_RDMA_WRITE ?
1488 					 MPA_V2_RDMA_WRITE_RTR : p2p_type ==
1489 					 FW_RI_INIT_P2PTYPE_READ_REQ ?
1490 					 MPA_V2_RDMA_READ_RTR : 0) : 0));
1491 		memcpy(mpa->private_data, &mpa_v2_params,
1492 				sizeof(struct mpa_v2_conn_params));
1493 
1494 		if (ep->plen)
1495 			memcpy(mpa->private_data +
1496 				sizeof(struct mpa_v2_conn_params), pdata, plen);
1497 		CTR5(KTR_IW_CXGBE, "%s:smrej3 %p %d %d %d", __func__, ep,
1498 		    mpa_v2_params.ird, mpa_v2_params.ord, ep->plen);
1499 	} else
1500 		if (plen)
1501 			memcpy(mpa->private_data, pdata, plen);
1502 
1503 	m = m_getm(NULL, mpalen, M_NOWAIT, MT_DATA);
1504 	if (m == NULL) {
1505 		free(mpa, M_CXGBE);
1506 		return (-ENOMEM);
1507 	}
1508 	m_copyback(m, 0, mpalen, (void *)mpa);
1509 	free(mpa, M_CXGBE);
1510 
1511 	err = -sosend(ep->com.so, NULL, NULL, m, NULL, MSG_DONTWAIT, ep->com.thread);
1512 	if (!err)
1513 		ep->snd_seq += mpalen;
1514 	CTR4(KTR_IW_CXGBE, "%s:smrejE %p %u %d", __func__, ep, ep->hwtid, err);
1515 	return err;
1516 }
1517 
1518 static int send_mpa_reply(struct c4iw_ep *ep, const void *pdata, u8 plen)
1519 {
1520 	int mpalen;
1521 	struct mpa_message *mpa;
1522 	struct mbuf *m;
1523 	struct mpa_v2_conn_params mpa_v2_params;
1524 	int err;
1525 
1526 	CTR2(KTR_IW_CXGBE, "%s:smrepB %p", __func__, ep);
1527 
1528 	mpalen = sizeof(*mpa) + plen;
1529 
1530 	if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
1531 
1532 		CTR3(KTR_IW_CXGBE, "%s:smrep1 %p %d", __func__, ep,
1533 		    ep->mpa_attr.version);
1534 		mpalen += sizeof(struct mpa_v2_conn_params);
1535 	}
1536 
1537 	mpa = malloc(mpalen, M_CXGBE, M_NOWAIT);
1538 	if (mpa == NULL)
1539 		return (-ENOMEM);
1540 
1541 	memset(mpa, 0, sizeof(*mpa));
1542 	memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
1543 	mpa->flags = (ep->mpa_attr.crc_enabled ? MPA_CRC : 0) |
1544 		(markers_enabled ? MPA_MARKERS : 0);
1545 	mpa->revision = ep->mpa_attr.version;
1546 	mpa->private_data_size = htons(plen);
1547 
1548 	if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
1549 
1550 		mpa->flags |= MPA_ENHANCED_RDMA_CONN;
1551 		mpa->private_data_size +=
1552 			htons(sizeof(struct mpa_v2_conn_params));
1553 		mpa_v2_params.ird = htons((u16)ep->ird);
1554 		mpa_v2_params.ord = htons((u16)ep->ord);
1555 		CTR5(KTR_IW_CXGBE, "%s:smrep3 %p %d %d %d", __func__, ep,
1556 		    ep->mpa_attr.version, mpa_v2_params.ird, mpa_v2_params.ord);
1557 
1558 		if (peer2peer && (ep->mpa_attr.p2p_type !=
1559 			FW_RI_INIT_P2PTYPE_DISABLED)) {
1560 
1561 			mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL);
1562 
1563 			if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE) {
1564 
1565 				mpa_v2_params.ord |=
1566 					htons(MPA_V2_RDMA_WRITE_RTR);
1567 				CTR5(KTR_IW_CXGBE, "%s:smrep4 %p %d %d %d",
1568 				    __func__, ep, p2p_type, mpa_v2_params.ird,
1569 				    mpa_v2_params.ord);
1570 			}
1571 			else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ) {
1572 
1573 				mpa_v2_params.ord |=
1574 					htons(MPA_V2_RDMA_READ_RTR);
1575 				CTR5(KTR_IW_CXGBE, "%s:smrep5 %p %d %d %d",
1576 				    __func__, ep, p2p_type, mpa_v2_params.ird,
1577 				    mpa_v2_params.ord);
1578 			}
1579 		}
1580 
1581 		memcpy(mpa->private_data, &mpa_v2_params,
1582 			sizeof(struct mpa_v2_conn_params));
1583 
1584 		if (ep->plen)
1585 			memcpy(mpa->private_data +
1586 				sizeof(struct mpa_v2_conn_params), pdata, plen);
1587 	} else
1588 		if (plen)
1589 			memcpy(mpa->private_data, pdata, plen);
1590 
1591 	m = m_getm(NULL, mpalen, M_NOWAIT, MT_DATA);
1592 	if (m == NULL) {
1593 		free(mpa, M_CXGBE);
1594 		return (-ENOMEM);
1595 	}
1596 	m_copyback(m, 0, mpalen, (void *)mpa);
1597 	free(mpa, M_CXGBE);
1598 
1599 
1600 	ep->com.state = MPA_REP_SENT;
1601 	ep->snd_seq += mpalen;
1602 	err = -sosend(ep->com.so, NULL, NULL, m, NULL, MSG_DONTWAIT,
1603 			ep->com.thread);
1604 	CTR3(KTR_IW_CXGBE, "%s:smrepE %p %d", __func__, ep, err);
1605 	return err;
1606 }
1607 
1608 
1609 
1610 static void close_complete_upcall(struct c4iw_ep *ep, int status)
1611 {
1612 	struct iw_cm_event event;
1613 
1614 	CTR2(KTR_IW_CXGBE, "%s:ccuB %p", __func__, ep);
1615 	memset(&event, 0, sizeof(event));
1616 	event.event = IW_CM_EVENT_CLOSE;
1617 	event.status = status;
1618 
1619 	if (ep->com.cm_id) {
1620 
1621 		CTR2(KTR_IW_CXGBE, "%s:ccu1 %1", __func__, ep);
1622 		ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1623 		deref_cm_id(&ep->com);
1624 		set_bit(CLOSE_UPCALL, &ep->com.history);
1625 	}
1626 	CTR2(KTR_IW_CXGBE, "%s:ccuE %p", __func__, ep);
1627 }
1628 
1629 static int
1630 send_abort(struct c4iw_ep *ep)
1631 {
1632 	struct socket *so = ep->com.so;
1633 	struct sockopt sopt;
1634 	int rc;
1635 	struct linger l;
1636 
1637 	CTR5(KTR_IW_CXGBE, "%s ep %p so %p state %s tid %d", __func__, ep, so,
1638 	    states[ep->com.state], ep->hwtid);
1639 
1640 	l.l_onoff = 1;
1641 	l.l_linger = 0;
1642 
1643 	/* linger_time of 0 forces RST to be sent */
1644 	sopt.sopt_dir = SOPT_SET;
1645 	sopt.sopt_level = SOL_SOCKET;
1646 	sopt.sopt_name = SO_LINGER;
1647 	sopt.sopt_val = (caddr_t)&l;
1648 	sopt.sopt_valsize = sizeof l;
1649 	sopt.sopt_td = NULL;
1650 	rc = -sosetopt(so, &sopt);
1651 	if (rc != 0) {
1652 		log(LOG_ERR, "%s: sosetopt(%p, linger = 0) failed with %d.\n",
1653 		    __func__, so, rc);
1654 	}
1655 
1656 	uninit_iwarp_socket(so);
1657 	soclose(so);
1658 	set_bit(ABORT_CONN, &ep->com.history);
1659 
1660 	/*
1661 	 * TBD: iw_cxgbe driver should receive ABORT reply for every ABORT
1662 	 * request it has sent. But the current TOE driver is not propagating
1663 	 * this ABORT reply event (via do_abort_rpl) to iw_cxgbe. So as a work-
1664 	 * around de-refererece 'ep' here instead of doing it in abort_rpl()
1665 	 * handler(not yet implemented) of iw_cxgbe driver.
1666 	 */
1667 	release_ep_resources(ep);
1668 	ep->com.state = DEAD;
1669 
1670 	return (0);
1671 }
1672 
1673 static void peer_close_upcall(struct c4iw_ep *ep)
1674 {
1675 	struct iw_cm_event event;
1676 
1677 	CTR2(KTR_IW_CXGBE, "%s:pcuB %p", __func__, ep);
1678 	memset(&event, 0, sizeof(event));
1679 	event.event = IW_CM_EVENT_DISCONNECT;
1680 
1681 	if (ep->com.cm_id) {
1682 
1683 		CTR2(KTR_IW_CXGBE, "%s:pcu1 %p", __func__, ep);
1684 		ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1685 		set_bit(DISCONN_UPCALL, &ep->com.history);
1686 	}
1687 	CTR2(KTR_IW_CXGBE, "%s:pcuE %p", __func__, ep);
1688 }
1689 
1690 static void peer_abort_upcall(struct c4iw_ep *ep)
1691 {
1692 	struct iw_cm_event event;
1693 
1694 	CTR2(KTR_IW_CXGBE, "%s:pauB %p", __func__, ep);
1695 	memset(&event, 0, sizeof(event));
1696 	event.event = IW_CM_EVENT_CLOSE;
1697 	event.status = -ECONNRESET;
1698 
1699 	if (ep->com.cm_id) {
1700 
1701 		CTR2(KTR_IW_CXGBE, "%s:pau1 %p", __func__, ep);
1702 		ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1703 		deref_cm_id(&ep->com);
1704 		set_bit(ABORT_UPCALL, &ep->com.history);
1705 	}
1706 	CTR2(KTR_IW_CXGBE, "%s:pauE %p", __func__, ep);
1707 }
1708 
1709 static void connect_reply_upcall(struct c4iw_ep *ep, int status)
1710 {
1711 	struct iw_cm_event event;
1712 
1713 	CTR3(KTR_IW_CXGBE, "%s:cruB %p, status: %d", __func__, ep, status);
1714 	memset(&event, 0, sizeof(event));
1715 	event.event = IW_CM_EVENT_CONNECT_REPLY;
1716 	event.status = ((status == -ECONNABORTED) || (status == -EPIPE)) ?
1717 					-ECONNRESET : status;
1718 	event.local_addr = ep->com.local_addr;
1719 	event.remote_addr = ep->com.remote_addr;
1720 
1721 	if ((status == 0) || (status == -ECONNREFUSED)) {
1722 
1723 		if (!ep->tried_with_mpa_v1) {
1724 
1725 			CTR2(KTR_IW_CXGBE, "%s:cru1 %p", __func__, ep);
1726 			/* this means MPA_v2 is used */
1727 			event.ord = ep->ird;
1728 			event.ird = ep->ord;
1729 			event.private_data_len = ep->plen -
1730 				sizeof(struct mpa_v2_conn_params);
1731 			event.private_data = ep->mpa_pkt +
1732 				sizeof(struct mpa_message) +
1733 				sizeof(struct mpa_v2_conn_params);
1734 		} else {
1735 
1736 			CTR2(KTR_IW_CXGBE, "%s:cru2 %p", __func__, ep);
1737 			/* this means MPA_v1 is used */
1738 			event.ord = c4iw_max_read_depth;
1739 			event.ird = c4iw_max_read_depth;
1740 			event.private_data_len = ep->plen;
1741 			event.private_data = ep->mpa_pkt +
1742 				sizeof(struct mpa_message);
1743 		}
1744 	}
1745 
1746 	if (ep->com.cm_id) {
1747 
1748 		CTR2(KTR_IW_CXGBE, "%s:cru3 %p", __func__, ep);
1749 		set_bit(CONN_RPL_UPCALL, &ep->com.history);
1750 		ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1751 	}
1752 
1753 	if(status == -ECONNABORTED) {
1754 
1755 		CTR3(KTR_IW_CXGBE, "%s:cruE %p %d", __func__, ep, status);
1756 		return;
1757 	}
1758 
1759 	if (status < 0) {
1760 
1761 		CTR3(KTR_IW_CXGBE, "%s:cru4 %p %d", __func__, ep, status);
1762 		deref_cm_id(&ep->com);
1763 	}
1764 
1765 	CTR2(KTR_IW_CXGBE, "%s:cruE %p", __func__, ep);
1766 }
1767 
1768 static int connect_request_upcall(struct c4iw_ep *ep)
1769 {
1770 	struct iw_cm_event event;
1771 	int ret;
1772 
1773 	CTR3(KTR_IW_CXGBE, "%s: ep %p, mpa_v1 %d", __func__, ep,
1774 	    ep->tried_with_mpa_v1);
1775 
1776 	memset(&event, 0, sizeof(event));
1777 	event.event = IW_CM_EVENT_CONNECT_REQUEST;
1778 	event.local_addr = ep->com.local_addr;
1779 	event.remote_addr = ep->com.remote_addr;
1780 	event.provider_data = ep;
1781 
1782 	if (!ep->tried_with_mpa_v1) {
1783 		/* this means MPA_v2 is used */
1784 		event.ord = ep->ord;
1785 		event.ird = ep->ird;
1786 		event.private_data_len = ep->plen -
1787 			sizeof(struct mpa_v2_conn_params);
1788 		event.private_data = ep->mpa_pkt + sizeof(struct mpa_message) +
1789 			sizeof(struct mpa_v2_conn_params);
1790 	} else {
1791 
1792 		/* this means MPA_v1 is used. Send max supported */
1793 		event.ord = c4iw_max_read_depth;
1794 		event.ird = c4iw_max_read_depth;
1795 		event.private_data_len = ep->plen;
1796 		event.private_data = ep->mpa_pkt + sizeof(struct mpa_message);
1797 	}
1798 
1799 	c4iw_get_ep(&ep->com);
1800 	ret = ep->parent_ep->com.cm_id->event_handler(ep->parent_ep->com.cm_id,
1801 	    &event);
1802 	if(ret) {
1803 		CTR3(KTR_IW_CXGBE, "%s: ep %p, Failure while notifying event to"
1804 			" IWCM, err:%d", __func__, ep, ret);
1805 		c4iw_put_ep(&ep->com);
1806 	} else
1807 		/* Dereference parent_ep only in success case.
1808 		 * In case of failure, parent_ep is dereferenced by the caller
1809 		 * of process_mpa_request().
1810 		 */
1811 		c4iw_put_ep(&ep->parent_ep->com);
1812 
1813 	set_bit(CONNREQ_UPCALL, &ep->com.history);
1814 	return ret;
1815 }
1816 
1817 static void established_upcall(struct c4iw_ep *ep)
1818 {
1819 	struct iw_cm_event event;
1820 
1821 	CTR2(KTR_IW_CXGBE, "%s:euB %p", __func__, ep);
1822 	memset(&event, 0, sizeof(event));
1823 	event.event = IW_CM_EVENT_ESTABLISHED;
1824 	event.ird = ep->ord;
1825 	event.ord = ep->ird;
1826 
1827 	if (ep->com.cm_id) {
1828 
1829 		CTR2(KTR_IW_CXGBE, "%s:eu1 %p", __func__, ep);
1830 		ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1831 		set_bit(ESTAB_UPCALL, &ep->com.history);
1832 	}
1833 	CTR2(KTR_IW_CXGBE, "%s:euE %p", __func__, ep);
1834 }
1835 
1836 
1837 #define RELAXED_IRD_NEGOTIATION 1
1838 
1839 /*
1840  * process_mpa_reply - process streaming mode MPA reply
1841  *
1842  * Returns:
1843  *
1844  * 0 upon success indicating a connect request was delivered to the ULP
1845  * or the mpa request is incomplete but valid so far.
1846  *
1847  * 1 if a failure requires the caller to close the connection.
1848  *
1849  * 2 if a failure requires the caller to abort the connection.
1850  */
1851 static int process_mpa_reply(struct c4iw_ep *ep)
1852 {
1853 	struct mpa_message *mpa;
1854 	struct mpa_v2_conn_params *mpa_v2_params;
1855 	u16 plen;
1856 	u16 resp_ird, resp_ord;
1857 	u8 rtr_mismatch = 0, insuff_ird = 0;
1858 	struct c4iw_qp_attributes attrs = {0};
1859 	enum c4iw_qp_attr_mask mask;
1860 	int err;
1861 	struct mbuf *top, *m;
1862 	int flags = MSG_DONTWAIT;
1863 	struct uio uio;
1864 	int disconnect = 0;
1865 
1866 	CTR2(KTR_IW_CXGBE, "%s:pmrB %p", __func__, ep);
1867 
1868 	/*
1869 	 * Stop mpa timer.  If it expired, then
1870 	 * we ignore the MPA reply.  process_timeout()
1871 	 * will abort the connection.
1872 	 */
1873 	if (STOP_EP_TIMER(ep))
1874 		return 0;
1875 
1876 	uio.uio_resid = 1000000;
1877 	uio.uio_td = ep->com.thread;
1878 	err = soreceive(ep->com.so, NULL, &uio, &top, NULL, &flags);
1879 
1880 	if (err) {
1881 
1882 		if (err == EWOULDBLOCK) {
1883 
1884 			CTR2(KTR_IW_CXGBE, "%s:pmr1 %p", __func__, ep);
1885 			START_EP_TIMER(ep);
1886 			return 0;
1887 		}
1888 		err = -err;
1889 		CTR2(KTR_IW_CXGBE, "%s:pmr2 %p", __func__, ep);
1890 		goto err;
1891 	}
1892 
1893 	if (ep->com.so->so_rcv.sb_mb) {
1894 
1895 		CTR2(KTR_IW_CXGBE, "%s:pmr3 %p", __func__, ep);
1896 		printf("%s data after soreceive called! so %p sb_mb %p top %p\n",
1897 		       __func__, ep->com.so, ep->com.so->so_rcv.sb_mb, top);
1898 	}
1899 
1900 	m = top;
1901 
1902 	do {
1903 
1904 		CTR2(KTR_IW_CXGBE, "%s:pmr4 %p", __func__, ep);
1905 		/*
1906 		 * If we get more than the supported amount of private data
1907 		 * then we must fail this connection.
1908 		 */
1909 		if (ep->mpa_pkt_len + m->m_len > sizeof(ep->mpa_pkt)) {
1910 
1911 			CTR3(KTR_IW_CXGBE, "%s:pmr5 %p %d", __func__, ep,
1912 			    ep->mpa_pkt_len + m->m_len);
1913 			err = (-EINVAL);
1914 			goto err_stop_timer;
1915 		}
1916 
1917 		/*
1918 		 * copy the new data into our accumulation buffer.
1919 		 */
1920 		m_copydata(m, 0, m->m_len, &(ep->mpa_pkt[ep->mpa_pkt_len]));
1921 		ep->mpa_pkt_len += m->m_len;
1922 		if (!m->m_next)
1923 			m = m->m_nextpkt;
1924 		else
1925 			m = m->m_next;
1926 	} while (m);
1927 
1928 	m_freem(top);
1929 	/*
1930 	 * if we don't even have the mpa message, then bail.
1931 	 */
1932 	if (ep->mpa_pkt_len < sizeof(*mpa)) {
1933 		return 0;
1934 	}
1935 	mpa = (struct mpa_message *) ep->mpa_pkt;
1936 
1937 	/* Validate MPA header. */
1938 	if (mpa->revision > mpa_rev) {
1939 
1940 		CTR4(KTR_IW_CXGBE, "%s:pmr6 %p %d %d", __func__, ep,
1941 		    mpa->revision, mpa_rev);
1942 		printk(KERN_ERR MOD "%s MPA version mismatch. Local = %d, "
1943 				" Received = %d\n", __func__, mpa_rev, mpa->revision);
1944 		err = -EPROTO;
1945 		goto err_stop_timer;
1946 	}
1947 
1948 	if (memcmp(mpa->key, MPA_KEY_REP, sizeof(mpa->key))) {
1949 
1950 		CTR2(KTR_IW_CXGBE, "%s:pmr7 %p", __func__, ep);
1951 		err = -EPROTO;
1952 		goto err_stop_timer;
1953 	}
1954 
1955 	plen = ntohs(mpa->private_data_size);
1956 
1957 	/*
1958 	 * Fail if there's too much private data.
1959 	 */
1960 	if (plen > MPA_MAX_PRIVATE_DATA) {
1961 
1962 		CTR2(KTR_IW_CXGBE, "%s:pmr8 %p", __func__, ep);
1963 		err = -EPROTO;
1964 		goto err_stop_timer;
1965 	}
1966 
1967 	/*
1968 	 * If plen does not account for pkt size
1969 	 */
1970 	if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
1971 
1972 		CTR2(KTR_IW_CXGBE, "%s:pmr9 %p", __func__, ep);
1973 		STOP_EP_TIMER(ep);
1974 		err = -EPROTO;
1975 		goto err_stop_timer;
1976 	}
1977 
1978 	ep->plen = (u8) plen;
1979 
1980 	/*
1981 	 * If we don't have all the pdata yet, then bail.
1982 	 * We'll continue process when more data arrives.
1983 	 */
1984 	if (ep->mpa_pkt_len < (sizeof(*mpa) + plen)) {
1985 
1986 		CTR2(KTR_IW_CXGBE, "%s:pmra %p", __func__, ep);
1987 		return 0;
1988 	}
1989 
1990 	if (mpa->flags & MPA_REJECT) {
1991 
1992 		CTR2(KTR_IW_CXGBE, "%s:pmrb %p", __func__, ep);
1993 		err = -ECONNREFUSED;
1994 		goto err_stop_timer;
1995 	}
1996 
1997 	/*
1998 	 * If we get here we have accumulated the entire mpa
1999 	 * start reply message including private data. And
2000 	 * the MPA header is valid.
2001 	 */
2002 	ep->com.state = FPDU_MODE;
2003 	ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
2004 	ep->mpa_attr.recv_marker_enabled = markers_enabled;
2005 	ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
2006 	ep->mpa_attr.version = mpa->revision;
2007 	ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
2008 
2009 	if (mpa->revision == 2) {
2010 
2011 		CTR2(KTR_IW_CXGBE, "%s:pmrc %p", __func__, ep);
2012 		ep->mpa_attr.enhanced_rdma_conn =
2013 			mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
2014 
2015 		if (ep->mpa_attr.enhanced_rdma_conn) {
2016 
2017 			CTR2(KTR_IW_CXGBE, "%s:pmrd %p", __func__, ep);
2018 			mpa_v2_params = (struct mpa_v2_conn_params *)
2019 				(ep->mpa_pkt + sizeof(*mpa));
2020 			resp_ird = ntohs(mpa_v2_params->ird) &
2021 				MPA_V2_IRD_ORD_MASK;
2022 			resp_ord = ntohs(mpa_v2_params->ord) &
2023 				MPA_V2_IRD_ORD_MASK;
2024 
2025 			/*
2026 			 * This is a double-check. Ideally, below checks are
2027 			 * not required since ird/ord stuff has been taken
2028 			 * care of in c4iw_accept_cr
2029 			 */
2030 			if (ep->ird < resp_ord) {
2031 				if (RELAXED_IRD_NEGOTIATION && resp_ord <=
2032 				   ep->com.dev->rdev.adap->params.max_ordird_qp)
2033 					ep->ird = resp_ord;
2034 				else
2035 					insuff_ird = 1;
2036 			} else if (ep->ird > resp_ord) {
2037 				ep->ird = resp_ord;
2038 			}
2039 			if (ep->ord > resp_ird) {
2040 				if (RELAXED_IRD_NEGOTIATION)
2041 					ep->ord = resp_ird;
2042 				else
2043 					insuff_ird = 1;
2044 			}
2045 			if (insuff_ird) {
2046 				err = -ENOMEM;
2047 				ep->ird = resp_ord;
2048 				ep->ord = resp_ird;
2049 			}
2050 
2051 			if (ntohs(mpa_v2_params->ird) &
2052 				MPA_V2_PEER2PEER_MODEL) {
2053 
2054 				CTR2(KTR_IW_CXGBE, "%s:pmrf %p", __func__, ep);
2055 				if (ntohs(mpa_v2_params->ord) &
2056 					MPA_V2_RDMA_WRITE_RTR) {
2057 
2058 					CTR2(KTR_IW_CXGBE, "%s:pmrg %p", __func__, ep);
2059 					ep->mpa_attr.p2p_type =
2060 						FW_RI_INIT_P2PTYPE_RDMA_WRITE;
2061 				}
2062 				else if (ntohs(mpa_v2_params->ord) &
2063 					MPA_V2_RDMA_READ_RTR) {
2064 
2065 					CTR2(KTR_IW_CXGBE, "%s:pmrh %p", __func__, ep);
2066 					ep->mpa_attr.p2p_type =
2067 						FW_RI_INIT_P2PTYPE_READ_REQ;
2068 				}
2069 			}
2070 		}
2071 	} else {
2072 
2073 		CTR2(KTR_IW_CXGBE, "%s:pmri %p", __func__, ep);
2074 
2075 		if (mpa->revision == 1) {
2076 
2077 			CTR2(KTR_IW_CXGBE, "%s:pmrj %p", __func__, ep);
2078 
2079 			if (peer2peer) {
2080 
2081 				CTR2(KTR_IW_CXGBE, "%s:pmrk %p", __func__, ep);
2082 				ep->mpa_attr.p2p_type = p2p_type;
2083 			}
2084 		}
2085 	}
2086 
2087 	if (set_tcpinfo(ep)) {
2088 
2089 		CTR2(KTR_IW_CXGBE, "%s:pmrl %p", __func__, ep);
2090 		printf("%s set_tcpinfo error\n", __func__);
2091 		err = -ECONNRESET;
2092 		goto err;
2093 	}
2094 
2095 	CTR6(KTR_IW_CXGBE, "%s - crc_enabled = %d, recv_marker_enabled = %d, "
2096 	    "xmit_marker_enabled = %d, version = %d p2p_type = %d", __func__,
2097 	    ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
2098 	    ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version,
2099 	    ep->mpa_attr.p2p_type);
2100 
2101 	/*
2102 	 * If responder's RTR does not match with that of initiator, assign
2103 	 * FW_RI_INIT_P2PTYPE_DISABLED in mpa attributes so that RTR is not
2104 	 * generated when moving QP to RTS state.
2105 	 * A TERM message will be sent after QP has moved to RTS state
2106 	 */
2107 	if ((ep->mpa_attr.version == 2) && peer2peer &&
2108 		(ep->mpa_attr.p2p_type != p2p_type)) {
2109 
2110 		CTR2(KTR_IW_CXGBE, "%s:pmrm %p", __func__, ep);
2111 		ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
2112 		rtr_mismatch = 1;
2113 	}
2114 
2115 
2116 	//ep->ofld_txq = TOEPCB(ep->com.so)->ofld_txq;
2117 	attrs.mpa_attr = ep->mpa_attr;
2118 	attrs.max_ird = ep->ird;
2119 	attrs.max_ord = ep->ord;
2120 	attrs.llp_stream_handle = ep;
2121 	attrs.next_state = C4IW_QP_STATE_RTS;
2122 
2123 	mask = C4IW_QP_ATTR_NEXT_STATE |
2124 		C4IW_QP_ATTR_LLP_STREAM_HANDLE | C4IW_QP_ATTR_MPA_ATTR |
2125 		C4IW_QP_ATTR_MAX_IRD | C4IW_QP_ATTR_MAX_ORD;
2126 
2127 	/* bind QP and TID with INIT_WR */
2128 	err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp, mask, &attrs, 1);
2129 
2130 	if (err) {
2131 
2132 		CTR2(KTR_IW_CXGBE, "%s:pmrn %p", __func__, ep);
2133 		goto err;
2134 	}
2135 
2136 	/*
2137 	 * If responder's RTR requirement did not match with what initiator
2138 	 * supports, generate TERM message
2139 	 */
2140 	if (rtr_mismatch) {
2141 
2142 		CTR2(KTR_IW_CXGBE, "%s:pmro %p", __func__, ep);
2143 		printk(KERN_ERR "%s: RTR mismatch, sending TERM\n", __func__);
2144 		attrs.layer_etype = LAYER_MPA | DDP_LLP;
2145 		attrs.ecode = MPA_NOMATCH_RTR;
2146 		attrs.next_state = C4IW_QP_STATE_TERMINATE;
2147 		attrs.send_term = 1;
2148 		err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2149 			C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2150 		err = -ENOMEM;
2151 		disconnect = 1;
2152 		goto out;
2153 	}
2154 
2155 	/*
2156 	 * Generate TERM if initiator IRD is not sufficient for responder
2157 	 * provided ORD. Currently, we do the same behaviour even when
2158 	 * responder provided IRD is also not sufficient as regards to
2159 	 * initiator ORD.
2160 	 */
2161 	if (insuff_ird) {
2162 
2163 		CTR2(KTR_IW_CXGBE, "%s:pmrp %p", __func__, ep);
2164 		printk(KERN_ERR "%s: Insufficient IRD, sending TERM\n",
2165 				__func__);
2166 		attrs.layer_etype = LAYER_MPA | DDP_LLP;
2167 		attrs.ecode = MPA_INSUFF_IRD;
2168 		attrs.next_state = C4IW_QP_STATE_TERMINATE;
2169 		attrs.send_term = 1;
2170 		err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2171 			C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2172 		err = -ENOMEM;
2173 		disconnect = 1;
2174 		goto out;
2175 	}
2176 	goto out;
2177 err_stop_timer:
2178 	STOP_EP_TIMER(ep);
2179 err:
2180 	disconnect = 2;
2181 out:
2182 	connect_reply_upcall(ep, err);
2183 	CTR2(KTR_IW_CXGBE, "%s:pmrE %p", __func__, ep);
2184 	return disconnect;
2185 }
2186 
2187 /*
2188  * process_mpa_request - process streaming mode MPA request
2189  *
2190  * Returns:
2191  *
2192  * 0 upon success indicating a connect request was delivered to the ULP
2193  * or the mpa request is incomplete but valid so far.
2194  *
2195  * 1 if a failure requires the caller to close the connection.
2196  *
2197  * 2 if a failure requires the caller to abort the connection.
2198  */
2199 static int
2200 process_mpa_request(struct c4iw_ep *ep)
2201 {
2202 	struct mpa_message *mpa;
2203 	struct mpa_v2_conn_params *mpa_v2_params;
2204 	u16 plen;
2205 	int flags = MSG_DONTWAIT;
2206 	int rc;
2207 	struct iovec iov;
2208 	struct uio uio;
2209 	enum c4iw_ep_state state = ep->com.state;
2210 
2211 	CTR3(KTR_IW_CXGBE, "%s: ep %p, state %s", __func__, ep, states[state]);
2212 
2213 	if (state != MPA_REQ_WAIT)
2214 		return 0;
2215 
2216 	iov.iov_base = &ep->mpa_pkt[ep->mpa_pkt_len];
2217 	iov.iov_len = sizeof(ep->mpa_pkt) - ep->mpa_pkt_len;
2218 	uio.uio_iov = &iov;
2219 	uio.uio_iovcnt = 1;
2220 	uio.uio_offset = 0;
2221 	uio.uio_resid = sizeof(ep->mpa_pkt) - ep->mpa_pkt_len;
2222 	uio.uio_segflg = UIO_SYSSPACE;
2223 	uio.uio_rw = UIO_READ;
2224 	uio.uio_td = NULL; /* uio.uio_td = ep->com.thread; */
2225 
2226 	rc = soreceive(ep->com.so, NULL, &uio, NULL, NULL, &flags);
2227 	if (rc == EAGAIN)
2228 		return 0;
2229 	else if (rc)
2230 		goto err_stop_timer;
2231 
2232 	KASSERT(uio.uio_offset > 0, ("%s: sorecieve on so %p read no data",
2233 	    __func__, ep->com.so));
2234 	ep->mpa_pkt_len += uio.uio_offset;
2235 
2236 	/*
2237 	 * If we get more than the supported amount of private data then we must
2238 	 * fail this connection.  XXX: check so_rcv->sb_cc, or peek with another
2239 	 * soreceive, or increase the size of mpa_pkt by 1 and abort if the last
2240 	 * byte is filled by the soreceive above.
2241 	 */
2242 
2243 	/* Don't even have the MPA message.  Wait for more data to arrive. */
2244 	if (ep->mpa_pkt_len < sizeof(*mpa))
2245 		return 0;
2246 	mpa = (struct mpa_message *) ep->mpa_pkt;
2247 
2248 	/*
2249 	 * Validate MPA Header.
2250 	 */
2251 	if (mpa->revision > mpa_rev) {
2252 		log(LOG_ERR, "%s: MPA version mismatch. Local = %d,"
2253 		    " Received = %d\n", __func__, mpa_rev, mpa->revision);
2254 		goto err_stop_timer;
2255 	}
2256 
2257 	if (memcmp(mpa->key, MPA_KEY_REQ, sizeof(mpa->key)))
2258 		goto err_stop_timer;
2259 
2260 	/*
2261 	 * Fail if there's too much private data.
2262 	 */
2263 	plen = ntohs(mpa->private_data_size);
2264 	if (plen > MPA_MAX_PRIVATE_DATA)
2265 		goto err_stop_timer;
2266 
2267 	/*
2268 	 * If plen does not account for pkt size
2269 	 */
2270 	if (ep->mpa_pkt_len > (sizeof(*mpa) + plen))
2271 		goto err_stop_timer;
2272 
2273 	ep->plen = (u8) plen;
2274 
2275 	/*
2276 	 * If we don't have all the pdata yet, then bail.
2277 	 */
2278 	if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
2279 		return 0;
2280 
2281 	/*
2282 	 * If we get here we have accumulated the entire mpa
2283 	 * start reply message including private data.
2284 	 */
2285 	ep->mpa_attr.initiator = 0;
2286 	ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
2287 	ep->mpa_attr.recv_marker_enabled = markers_enabled;
2288 	ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
2289 	ep->mpa_attr.version = mpa->revision;
2290 	if (mpa->revision == 1)
2291 		ep->tried_with_mpa_v1 = 1;
2292 	ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
2293 
2294 	if (mpa->revision == 2) {
2295 		ep->mpa_attr.enhanced_rdma_conn =
2296 		    mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
2297 		if (ep->mpa_attr.enhanced_rdma_conn) {
2298 			mpa_v2_params = (struct mpa_v2_conn_params *)
2299 				(ep->mpa_pkt + sizeof(*mpa));
2300 			ep->ird = ntohs(mpa_v2_params->ird) &
2301 				MPA_V2_IRD_ORD_MASK;
2302 			ep->ird = min_t(u32, ep->ird,
2303 					cur_max_read_depth(ep->com.dev));
2304 			ep->ord = ntohs(mpa_v2_params->ord) &
2305 				MPA_V2_IRD_ORD_MASK;
2306 			ep->ord = min_t(u32, ep->ord,
2307 					cur_max_read_depth(ep->com.dev));
2308 			CTR3(KTR_IW_CXGBE, "%s initiator ird %u ord %u",
2309 				 __func__, ep->ird, ep->ord);
2310 			if (ntohs(mpa_v2_params->ird) & MPA_V2_PEER2PEER_MODEL)
2311 				if (peer2peer) {
2312 					if (ntohs(mpa_v2_params->ord) &
2313 							MPA_V2_RDMA_WRITE_RTR)
2314 						ep->mpa_attr.p2p_type =
2315 						FW_RI_INIT_P2PTYPE_RDMA_WRITE;
2316 					else if (ntohs(mpa_v2_params->ord) &
2317 							MPA_V2_RDMA_READ_RTR)
2318 						ep->mpa_attr.p2p_type =
2319 						FW_RI_INIT_P2PTYPE_READ_REQ;
2320 				}
2321 		}
2322 	} else if (mpa->revision == 1 && peer2peer)
2323 		ep->mpa_attr.p2p_type = p2p_type;
2324 
2325 	if (set_tcpinfo(ep))
2326 		goto err_stop_timer;
2327 
2328 	CTR5(KTR_IW_CXGBE, "%s: crc_enabled = %d, recv_marker_enabled = %d, "
2329 	    "xmit_marker_enabled = %d, version = %d", __func__,
2330 	    ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
2331 	    ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version);
2332 
2333 	ep->com.state = MPA_REQ_RCVD;
2334 	STOP_EP_TIMER(ep);
2335 
2336 	/* drive upcall */
2337 	if (ep->parent_ep->com.state != DEAD)
2338 		if (connect_request_upcall(ep))
2339 			goto err_out;
2340 	return 0;
2341 
2342 err_stop_timer:
2343 	STOP_EP_TIMER(ep);
2344 err_out:
2345 	return 2;
2346 }
2347 
2348 /*
2349  * Upcall from the adapter indicating data has been transmitted.
2350  * For us its just the single MPA request or reply.  We can now free
2351  * the skb holding the mpa message.
2352  */
2353 int c4iw_reject_cr(struct iw_cm_id *cm_id, const void *pdata, u8 pdata_len)
2354 {
2355 	int err;
2356 	struct c4iw_ep *ep = to_ep(cm_id);
2357 	int abort = 0;
2358 
2359 	mutex_lock(&ep->com.mutex);
2360 	CTR2(KTR_IW_CXGBE, "%s:crcB %p", __func__, ep);
2361 
2362 	if ((ep->com.state == DEAD) ||
2363 			(ep->com.state != MPA_REQ_RCVD)) {
2364 
2365 		CTR2(KTR_IW_CXGBE, "%s:crc1 %p", __func__, ep);
2366 		mutex_unlock(&ep->com.mutex);
2367 		c4iw_put_ep(&ep->com);
2368 		return -ECONNRESET;
2369 	}
2370 	set_bit(ULP_REJECT, &ep->com.history);
2371 
2372 	if (mpa_rev == 0) {
2373 
2374 		CTR2(KTR_IW_CXGBE, "%s:crc2 %p", __func__, ep);
2375 		abort = 1;
2376 	}
2377 	else {
2378 
2379 		CTR2(KTR_IW_CXGBE, "%s:crc3 %p", __func__, ep);
2380 		abort = send_mpa_reject(ep, pdata, pdata_len);
2381 	}
2382 	STOP_EP_TIMER(ep);
2383 	err = c4iw_ep_disconnect(ep, abort != 0, GFP_KERNEL);
2384 	mutex_unlock(&ep->com.mutex);
2385 	c4iw_put_ep(&ep->com);
2386 	CTR3(KTR_IW_CXGBE, "%s:crc4 %p, err: %d", __func__, ep, err);
2387 	return 0;
2388 }
2389 
2390 int c4iw_accept_cr(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
2391 {
2392 	int err;
2393 	struct c4iw_qp_attributes attrs = {0};
2394 	enum c4iw_qp_attr_mask mask;
2395 	struct c4iw_ep *ep = to_ep(cm_id);
2396 	struct c4iw_dev *h = to_c4iw_dev(cm_id->device);
2397 	struct c4iw_qp *qp = get_qhp(h, conn_param->qpn);
2398 	int abort = 0;
2399 
2400 	mutex_lock(&ep->com.mutex);
2401 	CTR2(KTR_IW_CXGBE, "%s:cacB %p", __func__, ep);
2402 
2403 	if ((ep->com.state == DEAD) ||
2404 			(ep->com.state != MPA_REQ_RCVD)) {
2405 
2406 		CTR2(KTR_IW_CXGBE, "%s:cac1 %p", __func__, ep);
2407 		err = -ECONNRESET;
2408 		goto err_out;
2409 	}
2410 
2411 	BUG_ON(!qp);
2412 
2413 	set_bit(ULP_ACCEPT, &ep->com.history);
2414 
2415 	if ((conn_param->ord > c4iw_max_read_depth) ||
2416 		(conn_param->ird > c4iw_max_read_depth)) {
2417 
2418 		CTR2(KTR_IW_CXGBE, "%s:cac2 %p", __func__, ep);
2419 		err = -EINVAL;
2420 		goto err_abort;
2421 	}
2422 
2423 	if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
2424 
2425 		CTR2(KTR_IW_CXGBE, "%s:cac3 %p", __func__, ep);
2426 
2427 		if (conn_param->ord > ep->ird) {
2428 			if (RELAXED_IRD_NEGOTIATION) {
2429 				conn_param->ord = ep->ird;
2430 			} else {
2431 				ep->ird = conn_param->ird;
2432 				ep->ord = conn_param->ord;
2433 				send_mpa_reject(ep, conn_param->private_data,
2434 						conn_param->private_data_len);
2435 				err = -ENOMEM;
2436 				goto err_abort;
2437 			}
2438 		}
2439 		if (conn_param->ird < ep->ord) {
2440 			if (RELAXED_IRD_NEGOTIATION &&
2441 			    ep->ord <= h->rdev.adap->params.max_ordird_qp) {
2442 				conn_param->ird = ep->ord;
2443 			} else {
2444 				err = -ENOMEM;
2445 				goto err_abort;
2446 			}
2447 		}
2448 	}
2449 	ep->ird = conn_param->ird;
2450 	ep->ord = conn_param->ord;
2451 
2452 	if (ep->mpa_attr.version == 1) {
2453 		if (peer2peer && ep->ird == 0)
2454 			ep->ird = 1;
2455 	} else {
2456 		if (peer2peer &&
2457 		    (ep->mpa_attr.p2p_type != FW_RI_INIT_P2PTYPE_DISABLED) &&
2458 		    (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ) && ep->ird == 0)
2459 			ep->ird = 1;
2460 	}
2461 
2462 	CTR4(KTR_IW_CXGBE, "%s %d ird %d ord %d", __func__, __LINE__,
2463 			ep->ird, ep->ord);
2464 
2465 	ep->com.cm_id = cm_id;
2466 	ref_cm_id(&ep->com);
2467 	ep->com.qp = qp;
2468 	ref_qp(ep);
2469 	//ep->ofld_txq = TOEPCB(ep->com.so)->ofld_txq;
2470 
2471 	/* bind QP to EP and move to RTS */
2472 	attrs.mpa_attr = ep->mpa_attr;
2473 	attrs.max_ird = ep->ird;
2474 	attrs.max_ord = ep->ord;
2475 	attrs.llp_stream_handle = ep;
2476 	attrs.next_state = C4IW_QP_STATE_RTS;
2477 
2478 	/* bind QP and TID with INIT_WR */
2479 	mask = C4IW_QP_ATTR_NEXT_STATE |
2480 		C4IW_QP_ATTR_LLP_STREAM_HANDLE |
2481 		C4IW_QP_ATTR_MPA_ATTR |
2482 		C4IW_QP_ATTR_MAX_IRD |
2483 		C4IW_QP_ATTR_MAX_ORD;
2484 
2485 	err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp, mask, &attrs, 1);
2486 	if (err) {
2487 		CTR3(KTR_IW_CXGBE, "%s:caca %p, err: %d", __func__, ep, err);
2488 		goto err_defef_cm_id;
2489 	}
2490 
2491 	err = send_mpa_reply(ep, conn_param->private_data,
2492 			conn_param->private_data_len);
2493 	if (err) {
2494 		CTR3(KTR_IW_CXGBE, "%s:cacb %p, err: %d", __func__, ep, err);
2495 		goto err_defef_cm_id;
2496 	}
2497 
2498 	ep->com.state = FPDU_MODE;
2499 	established_upcall(ep);
2500 	mutex_unlock(&ep->com.mutex);
2501 	c4iw_put_ep(&ep->com);
2502 	CTR2(KTR_IW_CXGBE, "%s:cacE %p", __func__, ep);
2503 	return 0;
2504 err_defef_cm_id:
2505 	deref_cm_id(&ep->com);
2506 err_abort:
2507 	abort = 1;
2508 err_out:
2509 	if (abort)
2510 		c4iw_ep_disconnect(ep, 1, GFP_KERNEL);
2511 	mutex_unlock(&ep->com.mutex);
2512 	c4iw_put_ep(&ep->com);
2513 	CTR2(KTR_IW_CXGBE, "%s:cacE err %p", __func__, ep);
2514 	return err;
2515 }
2516 
2517 static int
2518 c4iw_sock_create(struct sockaddr_storage *laddr, struct socket **so)
2519 {
2520 	int ret;
2521 	int size, on;
2522 	struct socket *sock = NULL;
2523 	struct sockopt sopt;
2524 
2525 	ret = sock_create_kern(laddr->ss_family,
2526 			SOCK_STREAM, IPPROTO_TCP, &sock);
2527 	if (ret) {
2528 		CTR2(KTR_IW_CXGBE, "%s:Failed to create TCP socket. err %d",
2529 				__func__, ret);
2530 		return ret;
2531 	}
2532 
2533 	if (reuseaddr) {
2534 		bzero(&sopt, sizeof(struct sockopt));
2535 		sopt.sopt_dir = SOPT_SET;
2536 		sopt.sopt_level = SOL_SOCKET;
2537 		sopt.sopt_name = SO_REUSEADDR;
2538 		on = 1;
2539 		sopt.sopt_val = &on;
2540 		sopt.sopt_valsize = sizeof(on);
2541 		ret = -sosetopt(sock, &sopt);
2542 		if (ret != 0) {
2543 			log(LOG_ERR, "%s: sosetopt(%p, SO_REUSEADDR) "
2544 				"failed with %d.\n", __func__, sock, ret);
2545 		}
2546 		bzero(&sopt, sizeof(struct sockopt));
2547 		sopt.sopt_dir = SOPT_SET;
2548 		sopt.sopt_level = SOL_SOCKET;
2549 		sopt.sopt_name = SO_REUSEPORT;
2550 		on = 1;
2551 		sopt.sopt_val = &on;
2552 		sopt.sopt_valsize = sizeof(on);
2553 		ret = -sosetopt(sock, &sopt);
2554 		if (ret != 0) {
2555 			log(LOG_ERR, "%s: sosetopt(%p, SO_REUSEPORT) "
2556 				"failed with %d.\n", __func__, sock, ret);
2557 		}
2558 	}
2559 
2560 	ret = -sobind(sock, (struct sockaddr *)laddr, curthread);
2561 	if (ret) {
2562 		CTR2(KTR_IW_CXGBE, "%s:Failed to bind socket. err %p",
2563 				__func__, ret);
2564 		sock_release(sock);
2565 		return ret;
2566 	}
2567 
2568 	size = laddr->ss_family == AF_INET6 ?
2569 		sizeof(struct sockaddr_in6) : sizeof(struct sockaddr_in);
2570 	ret = sock_getname(sock, (struct sockaddr *)laddr, &size, 0);
2571 	if (ret) {
2572 		CTR2(KTR_IW_CXGBE, "%s:sock_getname failed. err %p",
2573 				__func__, ret);
2574 		sock_release(sock);
2575 		return ret;
2576 	}
2577 
2578 	*so = sock;
2579 	return 0;
2580 }
2581 
2582 int c4iw_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
2583 {
2584 	int err = 0;
2585 	struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
2586 	struct c4iw_ep *ep = NULL;
2587 	struct ifnet    *nh_ifp;        /* Logical egress interface */
2588 	struct epoch_tracker et;
2589 #ifdef VIMAGE
2590 	struct rdma_cm_id *rdma_id = (struct rdma_cm_id*)cm_id->context;
2591 	struct vnet *vnet = rdma_id->route.addr.dev_addr.net;
2592 #endif
2593 
2594 	CTR2(KTR_IW_CXGBE, "%s:ccB %p", __func__, cm_id);
2595 
2596 
2597 	if ((conn_param->ord > c4iw_max_read_depth) ||
2598 		(conn_param->ird > c4iw_max_read_depth)) {
2599 
2600 		CTR2(KTR_IW_CXGBE, "%s:cc1 %p", __func__, cm_id);
2601 		err = -EINVAL;
2602 		goto out;
2603 	}
2604 	ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
2605 	cm_id->provider_data = ep;
2606 
2607 	init_timer(&ep->timer);
2608 	ep->plen = conn_param->private_data_len;
2609 
2610 	if (ep->plen) {
2611 
2612 		CTR2(KTR_IW_CXGBE, "%s:cc3 %p", __func__, ep);
2613 		memcpy(ep->mpa_pkt + sizeof(struct mpa_message),
2614 				conn_param->private_data, ep->plen);
2615 	}
2616 	ep->ird = conn_param->ird;
2617 	ep->ord = conn_param->ord;
2618 
2619 	if (peer2peer && ep->ord == 0) {
2620 
2621 		CTR2(KTR_IW_CXGBE, "%s:cc4 %p", __func__, ep);
2622 		ep->ord = 1;
2623 	}
2624 
2625 	ep->com.dev = dev;
2626 	ep->com.cm_id = cm_id;
2627 	ref_cm_id(&ep->com);
2628 	ep->com.qp = get_qhp(dev, conn_param->qpn);
2629 
2630 	if (!ep->com.qp) {
2631 
2632 		CTR2(KTR_IW_CXGBE, "%s:cc5 %p", __func__, ep);
2633 		err = -EINVAL;
2634 		goto fail;
2635 	}
2636 	ref_qp(ep);
2637 	ep->com.thread = curthread;
2638 
2639 	NET_EPOCH_ENTER(et);
2640 	CURVNET_SET(vnet);
2641 	err = get_ifnet_from_raddr(&cm_id->remote_addr, &nh_ifp);
2642 	CURVNET_RESTORE();
2643 	NET_EPOCH_EXIT(et);
2644 
2645 	if (err) {
2646 
2647 		CTR2(KTR_IW_CXGBE, "%s:cc7 %p", __func__, ep);
2648 		printk(KERN_ERR MOD "%s - cannot find route.\n", __func__);
2649 		err = EHOSTUNREACH;
2650 		return err;
2651 	}
2652 
2653 	if (!(nh_ifp->if_capenable & IFCAP_TOE) ||
2654 	    TOEDEV(nh_ifp) == NULL) {
2655 		err = -ENOPROTOOPT;
2656 		goto fail;
2657 	}
2658 	ep->com.state = CONNECTING;
2659 	ep->tos = 0;
2660 	ep->com.local_addr = cm_id->local_addr;
2661 	ep->com.remote_addr = cm_id->remote_addr;
2662 
2663 	err = c4iw_sock_create(&cm_id->local_addr, &ep->com.so);
2664 	if (err)
2665 		goto fail;
2666 
2667 	setiwsockopt(ep->com.so);
2668 	init_iwarp_socket(ep->com.so, &ep->com);
2669 	err = -soconnect(ep->com.so, (struct sockaddr *)&ep->com.remote_addr,
2670 		ep->com.thread);
2671 	if (err)
2672 		goto fail_free_so;
2673 	CTR2(KTR_IW_CXGBE, "%s:ccE, ep %p", __func__, ep);
2674 	return 0;
2675 
2676 fail_free_so:
2677 	uninit_iwarp_socket(ep->com.so);
2678 	ep->com.state = DEAD;
2679 	sock_release(ep->com.so);
2680 fail:
2681 	deref_cm_id(&ep->com);
2682 	c4iw_put_ep(&ep->com);
2683 	ep = NULL;
2684 out:
2685 	CTR2(KTR_IW_CXGBE, "%s:ccE Error %d", __func__, err);
2686 	return err;
2687 }
2688 
2689 /*
2690  * iwcm->create_listen.  Returns -errno on failure.
2691  */
2692 int
2693 c4iw_create_listen(struct iw_cm_id *cm_id, int backlog)
2694 {
2695 	struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
2696 	struct c4iw_listen_ep *lep = NULL;
2697 	struct listen_port_info *port_info = NULL;
2698 	int rc = 0;
2699 
2700 	CTR3(KTR_IW_CXGBE, "%s: cm_id %p, backlog %s", __func__, cm_id,
2701 			backlog);
2702 	if (c4iw_fatal_error(&dev->rdev)) {
2703 		CTR2(KTR_IW_CXGBE, "%s: cm_id %p, fatal error", __func__,
2704 			       cm_id);
2705 		return -EIO;
2706 	}
2707 	lep = alloc_ep(sizeof(*lep), GFP_KERNEL);
2708 	lep->com.cm_id = cm_id;
2709 	ref_cm_id(&lep->com);
2710 	lep->com.dev = dev;
2711 	lep->backlog = backlog;
2712 	lep->com.local_addr = cm_id->local_addr;
2713 	lep->com.thread = curthread;
2714 	cm_id->provider_data = lep;
2715 	lep->com.state = LISTEN;
2716 
2717 	/* In case of INDADDR_ANY, ibcore creates cmid for each device and
2718 	 * invokes iw_cxgbe listener callbacks assuming that iw_cxgbe creates
2719 	 * HW listeners for each device seperately. But toecore expects single
2720 	 * solisten() call with INADDR_ANY address to create HW listeners on
2721 	 * all devices for a given port number. So iw_cxgbe driver calls
2722 	 * solisten() only once for INADDR_ANY(usually done at first time
2723 	 * listener callback from ibcore). And all the subsequent INADDR_ANY
2724 	 * listener callbacks from ibcore(for the same port address) do not
2725 	 * invoke solisten() as first listener callback has already created
2726 	 * listeners for all other devices(via solisten).
2727 	 */
2728 	if (c4iw_any_addr((struct sockaddr *)&lep->com.local_addr)) {
2729 		port_info = add_ep_to_listenlist(lep);
2730 		/* skip solisten() if refcnt > 1, as the listeners were
2731 		 * alredy created by 'Master lep'
2732 		 */
2733 		if (port_info->refcnt > 1) {
2734 			/* As there will be only one listener socket for a TCP
2735 			 * port, copy Master lep's socket pointer to other lep's
2736 			 * that are belonging to same TCP port.
2737 			 */
2738 			struct c4iw_listen_ep *head_lep =
2739 					container_of(port_info->lep_list.next,
2740 					struct c4iw_listen_ep, listen_ep_list);
2741 			lep->com.so =  head_lep->com.so;
2742 			goto out;
2743 		}
2744 	}
2745 	rc = c4iw_sock_create(&cm_id->local_addr, &lep->com.so);
2746 	if (rc) {
2747 		CTR2(KTR_IW_CXGBE, "%s:Failed to create socket. err %d",
2748 				__func__, rc);
2749 		goto fail;
2750 	}
2751 
2752 	rc = -solisten(lep->com.so, backlog, curthread);
2753 	if (rc) {
2754 		CTR3(KTR_IW_CXGBE, "%s:Failed to listen on sock:%p. err %d",
2755 				__func__, lep->com.so, rc);
2756 		goto fail_free_so;
2757 	}
2758 	init_iwarp_socket(lep->com.so, &lep->com);
2759 out:
2760 	return 0;
2761 
2762 fail_free_so:
2763 	sock_release(lep->com.so);
2764 fail:
2765 	if (port_info)
2766 		rem_ep_from_listenlist(lep);
2767 	deref_cm_id(&lep->com);
2768 	c4iw_put_ep(&lep->com);
2769 	return rc;
2770 }
2771 
2772 int
2773 c4iw_destroy_listen(struct iw_cm_id *cm_id)
2774 {
2775 	struct c4iw_listen_ep *lep = to_listen_ep(cm_id);
2776 
2777 	mutex_lock(&lep->com.mutex);
2778 	CTR3(KTR_IW_CXGBE, "%s: cm_id %p, state %s", __func__, cm_id,
2779 	    states[lep->com.state]);
2780 
2781 	lep->com.state = DEAD;
2782 	if (c4iw_any_addr((struct sockaddr *)&lep->com.local_addr)) {
2783 		/* if no refcount then close listen socket */
2784 		if (!rem_ep_from_listenlist(lep))
2785 			close_socket(lep->com.so);
2786 	} else
2787 		close_socket(lep->com.so);
2788 	deref_cm_id(&lep->com);
2789 	mutex_unlock(&lep->com.mutex);
2790 	c4iw_put_ep(&lep->com);
2791 	return 0;
2792 }
2793 
2794 int __c4iw_ep_disconnect(struct c4iw_ep *ep, int abrupt, gfp_t gfp)
2795 {
2796 	int ret;
2797 	mutex_lock(&ep->com.mutex);
2798 	ret = c4iw_ep_disconnect(ep, abrupt, gfp);
2799 	mutex_unlock(&ep->com.mutex);
2800 	return ret;
2801 }
2802 
2803 int c4iw_ep_disconnect(struct c4iw_ep *ep, int abrupt, gfp_t gfp)
2804 {
2805 	int ret = 0;
2806 	int close = 0;
2807 	struct c4iw_rdev *rdev;
2808 
2809 
2810 	CTR2(KTR_IW_CXGBE, "%s:cedB %p", __func__, ep);
2811 
2812 	rdev = &ep->com.dev->rdev;
2813 
2814 	if (c4iw_fatal_error(rdev)) {
2815 		CTR3(KTR_IW_CXGBE, "%s:ced1 fatal error %p %s", __func__, ep,
2816 					states[ep->com.state]);
2817 		if (ep->com.state != DEAD) {
2818 			send_abort(ep);
2819 			ep->com.state = DEAD;
2820 		}
2821 		close_complete_upcall(ep, -ECONNRESET);
2822 		return ECONNRESET;
2823 	}
2824 	CTR3(KTR_IW_CXGBE, "%s:ced2 %p %s", __func__, ep,
2825 	    states[ep->com.state]);
2826 
2827 	/*
2828 	 * Ref the ep here in case we have fatal errors causing the
2829 	 * ep to be released and freed.
2830 	 */
2831 	c4iw_get_ep(&ep->com);
2832 	switch (ep->com.state) {
2833 
2834 		case MPA_REQ_WAIT:
2835 		case MPA_REQ_SENT:
2836 		case MPA_REQ_RCVD:
2837 		case MPA_REP_SENT:
2838 		case FPDU_MODE:
2839 			close = 1;
2840 			if (abrupt)
2841 				ep->com.state = ABORTING;
2842 			else {
2843 				ep->com.state = CLOSING;
2844 				START_EP_TIMER(ep);
2845 			}
2846 			set_bit(CLOSE_SENT, &ep->com.flags);
2847 			break;
2848 
2849 		case CLOSING:
2850 
2851 			if (!test_and_set_bit(CLOSE_SENT, &ep->com.flags)) {
2852 
2853 				close = 1;
2854 				if (abrupt) {
2855 					STOP_EP_TIMER(ep);
2856 					ep->com.state = ABORTING;
2857 				} else
2858 					ep->com.state = MORIBUND;
2859 			}
2860 			break;
2861 
2862 		case MORIBUND:
2863 		case ABORTING:
2864 		case DEAD:
2865 			CTR3(KTR_IW_CXGBE,
2866 			    "%s ignoring disconnect ep %p state %u", __func__,
2867 			    ep, ep->com.state);
2868 			break;
2869 
2870 		default:
2871 			BUG();
2872 			break;
2873 	}
2874 
2875 
2876 	if (close) {
2877 
2878 		CTR2(KTR_IW_CXGBE, "%s:ced3 %p", __func__, ep);
2879 
2880 		if (abrupt) {
2881 
2882 			CTR2(KTR_IW_CXGBE, "%s:ced4 %p", __func__, ep);
2883 			set_bit(EP_DISC_ABORT, &ep->com.history);
2884 			close_complete_upcall(ep, -ECONNRESET);
2885 			send_abort(ep);
2886 		} else {
2887 
2888 			CTR2(KTR_IW_CXGBE, "%s:ced5 %p", __func__, ep);
2889 			set_bit(EP_DISC_CLOSE, &ep->com.history);
2890 
2891 			if (!ep->parent_ep)
2892 				ep->com.state = MORIBUND;
2893 
2894 			CURVNET_SET(ep->com.so->so_vnet);
2895 			ret = sodisconnect(ep->com.so);
2896 			CURVNET_RESTORE();
2897 			if (ret) {
2898 				CTR2(KTR_IW_CXGBE, "%s:ced6 %p", __func__, ep);
2899 				STOP_EP_TIMER(ep);
2900 				send_abort(ep);
2901 				ep->com.state = DEAD;
2902 				close_complete_upcall(ep, -ECONNRESET);
2903 				set_bit(EP_DISC_FAIL, &ep->com.history);
2904 				if (ep->com.qp) {
2905 					struct c4iw_qp_attributes attrs = {0};
2906 
2907 					attrs.next_state = C4IW_QP_STATE_ERROR;
2908 					ret = c4iw_modify_qp(
2909 							ep->com.dev, ep->com.qp,
2910 							C4IW_QP_ATTR_NEXT_STATE,
2911 							&attrs, 1);
2912 					CTR3(KTR_IW_CXGBE, "%s:ced7 %p ret %d",
2913 						__func__, ep, ret);
2914 				}
2915 			}
2916 		}
2917 	}
2918 	c4iw_put_ep(&ep->com);
2919 	CTR2(KTR_IW_CXGBE, "%s:cedE %p", __func__, ep);
2920 	return ret;
2921 }
2922 
2923 #ifdef C4IW_EP_REDIRECT
2924 int c4iw_ep_redirect(void *ctx, struct dst_entry *old, struct dst_entry *new,
2925 		struct l2t_entry *l2t)
2926 {
2927 	struct c4iw_ep *ep = ctx;
2928 
2929 	if (ep->dst != old)
2930 		return 0;
2931 
2932 	PDBG("%s ep %p redirect to dst %p l2t %p\n", __func__, ep, new,
2933 			l2t);
2934 	dst_hold(new);
2935 	cxgb4_l2t_release(ep->l2t);
2936 	ep->l2t = l2t;
2937 	dst_release(old);
2938 	ep->dst = new;
2939 	return 1;
2940 }
2941 #endif
2942 
2943 
2944 
2945 static void ep_timeout(unsigned long arg)
2946 {
2947 	struct c4iw_ep *ep = (struct c4iw_ep *)arg;
2948 
2949 	if (!test_and_set_bit(TIMEOUT, &ep->com.flags)) {
2950 
2951 		/*
2952 		 * Only insert if it is not already on the list.
2953 		 */
2954 		if (!(ep->com.ep_events & C4IW_EVENT_TIMEOUT)) {
2955 			CTR2(KTR_IW_CXGBE, "%s:et1 %p", __func__, ep);
2956 			add_ep_to_req_list(ep, C4IW_EVENT_TIMEOUT);
2957 		}
2958 	}
2959 }
2960 
2961 static int fw6_wr_rpl(struct adapter *sc, const __be64 *rpl)
2962 {
2963 	uint64_t val = be64toh(*rpl);
2964 	int ret;
2965 	struct c4iw_wr_wait *wr_waitp;
2966 
2967 	ret = (int)((val >> 8) & 0xff);
2968 	wr_waitp = (struct c4iw_wr_wait *)rpl[1];
2969 	CTR3(KTR_IW_CXGBE, "%s wr_waitp %p ret %u", __func__, wr_waitp, ret);
2970 	if (wr_waitp)
2971 		c4iw_wake_up(wr_waitp, ret ? -ret : 0);
2972 
2973 	return (0);
2974 }
2975 
2976 static int fw6_cqe_handler(struct adapter *sc, const __be64 *rpl)
2977 {
2978 	struct cqe_list_entry *cle;
2979 	unsigned long flag;
2980 
2981 	cle = malloc(sizeof(*cle), M_CXGBE, M_NOWAIT);
2982 	cle->rhp = sc->iwarp_softc;
2983 	cle->err_cqe = *(const struct t4_cqe *)(&rpl[0]);
2984 
2985 	spin_lock_irqsave(&err_cqe_lock, flag);
2986 	list_add_tail(&cle->entry, &err_cqe_list);
2987 	queue_work(c4iw_taskq, &c4iw_task);
2988 	spin_unlock_irqrestore(&err_cqe_lock, flag);
2989 
2990 	return (0);
2991 }
2992 
2993 static int
2994 process_terminate(struct c4iw_ep *ep)
2995 {
2996 	struct c4iw_qp_attributes attrs = {0};
2997 
2998 	CTR2(KTR_IW_CXGBE, "%s:tB %p %d", __func__, ep);
2999 
3000 	if (ep && ep->com.qp) {
3001 
3002 		printk(KERN_WARNING MOD "TERM received tid %u qpid %u\n",
3003 				ep->hwtid, ep->com.qp->wq.sq.qid);
3004 		attrs.next_state = C4IW_QP_STATE_TERMINATE;
3005 		c4iw_modify_qp(ep->com.dev, ep->com.qp, C4IW_QP_ATTR_NEXT_STATE, &attrs,
3006 				1);
3007 	} else
3008 		printk(KERN_WARNING MOD "TERM received tid %u no ep/qp\n",
3009 								ep->hwtid);
3010 	CTR2(KTR_IW_CXGBE, "%s:tE %p %d", __func__, ep);
3011 
3012 	return 0;
3013 }
3014 
3015 int __init c4iw_cm_init(void)
3016 {
3017 
3018 	t4_register_cpl_handler(CPL_RDMA_TERMINATE, terminate);
3019 	t4_register_fw_msg_handler(FW6_TYPE_WR_RPL, fw6_wr_rpl);
3020 	t4_register_fw_msg_handler(FW6_TYPE_CQE, fw6_cqe_handler);
3021 	t4_register_an_handler(c4iw_ev_handler);
3022 
3023 	TAILQ_INIT(&req_list);
3024 	spin_lock_init(&req_lock);
3025 	INIT_LIST_HEAD(&err_cqe_list);
3026 	spin_lock_init(&err_cqe_lock);
3027 
3028 	INIT_WORK(&c4iw_task, process_req);
3029 
3030 	c4iw_taskq = create_singlethread_workqueue("iw_cxgbe");
3031 	if (!c4iw_taskq)
3032 		return -ENOMEM;
3033 
3034 	return 0;
3035 }
3036 
3037 void __exit c4iw_cm_term(void)
3038 {
3039 	WARN_ON(!TAILQ_EMPTY(&req_list));
3040 	WARN_ON(!list_empty(&err_cqe_list));
3041 	flush_workqueue(c4iw_taskq);
3042 	destroy_workqueue(c4iw_taskq);
3043 
3044 	t4_register_cpl_handler(CPL_RDMA_TERMINATE, NULL);
3045 	t4_register_fw_msg_handler(FW6_TYPE_WR_RPL, NULL);
3046 	t4_register_fw_msg_handler(FW6_TYPE_CQE, NULL);
3047 	t4_register_an_handler(NULL);
3048 }
3049 #endif
3050