cm.c revision 271127
1/*
2 * Copyright (c) 2009-2013 Chelsio, Inc. All rights reserved.
3 *
4 * This software is available to you under a choice of one of two
5 * licenses.  You may choose to be licensed under the terms of the GNU
6 * General Public License (GPL) Version 2, available from the file
7 * COPYING in the main directory of this source tree, or the
8 * OpenIB.org BSD license below:
9 *
10 *     Redistribution and use in source and binary forms, with or
11 *     without modification, are permitted provided that the following
12 *     conditions are met:
13 *
14 *      - Redistributions of source code must retain the above
15 *	  copyright notice, this list of conditions and the following
16 *	  disclaimer.
17 *
18 *      - Redistributions in binary form must reproduce the above
19 *	  copyright notice, this list of conditions and the following
20 *	  disclaimer in the documentation and/or other materials
21 *	  provided with the distribution.
22 *
23 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
24 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
25 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
26 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
27 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
28 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
29 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
30 * SOFTWARE.
31 */
32#include <sys/cdefs.h>
33__FBSDID("$FreeBSD: stable/10/sys/dev/cxgbe/iw_cxgbe/cm.c 271127 2014-09-04 20:12:36Z hselasky $");
34
35#include "opt_inet.h"
36
37#ifdef TCP_OFFLOAD
38#include <sys/types.h>
39#include <sys/malloc.h>
40#include <sys/socket.h>
41#include <sys/socketvar.h>
42#include <sys/sockio.h>
43#include <sys/taskqueue.h>
44#include <netinet/in.h>
45#include <net/route.h>
46
47#include <netinet/in_systm.h>
48#include <netinet/in_pcb.h>
49#include <netinet/ip.h>
50#include <netinet/ip_var.h>
51#include <netinet/tcp_var.h>
52#include <netinet/tcp.h>
53#include <netinet/tcpip.h>
54
55#include <netinet/toecore.h>
56
57struct sge_iq;
58struct rss_header;
59#include <linux/types.h>
60#include "offload.h"
61#include "tom/t4_tom.h"
62
63#define TOEPCB(so)  ((struct toepcb *)(so_sototcpcb((so))->t_toe))
64
65#include "iw_cxgbe.h"
66#include <linux/module.h>
67#include <linux/workqueue.h>
68#include <linux/notifier.h>
69#include <linux/inetdevice.h>
70#include <linux/if_vlan.h>
71#include <net/netevent.h>
72
73static spinlock_t req_lock;
74static TAILQ_HEAD(c4iw_ep_list, c4iw_ep_common) req_list;
75static struct work_struct c4iw_task;
76static struct workqueue_struct *c4iw_taskq;
77static LIST_HEAD(timeout_list);
78static spinlock_t timeout_lock;
79
80static void process_req(struct work_struct *ctx);
81static void start_ep_timer(struct c4iw_ep *ep);
82static void stop_ep_timer(struct c4iw_ep *ep);
83static int set_tcpinfo(struct c4iw_ep *ep);
84static enum c4iw_ep_state state_read(struct c4iw_ep_common *epc);
85static void __state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state tostate);
86static void state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state tostate);
87static void *alloc_ep(int size, gfp_t flags);
88void __free_ep(struct c4iw_ep_common *epc);
89static struct rtentry * find_route(__be32 local_ip, __be32 peer_ip, __be16 local_port,
90		__be16 peer_port, u8 tos);
91static int close_socket(struct c4iw_ep_common *epc, int close);
92static int shutdown_socket(struct c4iw_ep_common *epc);
93static void abort_socket(struct c4iw_ep *ep);
94static void send_mpa_req(struct c4iw_ep *ep);
95static int send_mpa_reject(struct c4iw_ep *ep, const void *pdata, u8 plen);
96static int send_mpa_reply(struct c4iw_ep *ep, const void *pdata, u8 plen);
97static void close_complete_upcall(struct c4iw_ep *ep);
98static int abort_connection(struct c4iw_ep *ep);
99static void peer_close_upcall(struct c4iw_ep *ep);
100static void peer_abort_upcall(struct c4iw_ep *ep);
101static void connect_reply_upcall(struct c4iw_ep *ep, int status);
102static void connect_request_upcall(struct c4iw_ep *ep);
103static void established_upcall(struct c4iw_ep *ep);
104static void process_mpa_reply(struct c4iw_ep *ep);
105static void process_mpa_request(struct c4iw_ep *ep);
106static void process_peer_close(struct c4iw_ep *ep);
107static void process_conn_error(struct c4iw_ep *ep);
108static void process_close_complete(struct c4iw_ep *ep);
109static void ep_timeout(unsigned long arg);
110static void init_sock(struct c4iw_ep_common *epc);
111static void process_data(struct c4iw_ep *ep);
112static void process_connected(struct c4iw_ep *ep);
113static struct socket * dequeue_socket(struct socket *head, struct sockaddr_in **remote, struct c4iw_ep *child_ep);
114static void process_newconn(struct c4iw_ep *parent_ep);
115static int c4iw_so_upcall(struct socket *so, void *arg, int waitflag);
116static void process_socket_event(struct c4iw_ep *ep);
117static void release_ep_resources(struct c4iw_ep *ep);
118
119#define START_EP_TIMER(ep) \
120    do { \
121	    CTR3(KTR_IW_CXGBE, "start_ep_timer (%s:%d) ep %p", \
122		__func__, __LINE__, (ep)); \
123	    start_ep_timer(ep); \
124    } while (0)
125
126#define STOP_EP_TIMER(ep) \
127    do { \
128	    CTR3(KTR_IW_CXGBE, "stop_ep_timer (%s:%d) ep %p", \
129		__func__, __LINE__, (ep)); \
130	    stop_ep_timer(ep); \
131    } while (0)
132
133#ifdef KTR
134static char *states[] = {
135	"idle",
136	"listen",
137	"connecting",
138	"mpa_wait_req",
139	"mpa_req_sent",
140	"mpa_req_rcvd",
141	"mpa_rep_sent",
142	"fpdu_mode",
143	"aborting",
144	"closing",
145	"moribund",
146	"dead",
147	NULL,
148};
149#endif
150
151static void
152process_req(struct work_struct *ctx)
153{
154	struct c4iw_ep_common *epc;
155
156	spin_lock(&req_lock);
157	while (!TAILQ_EMPTY(&req_list)) {
158		epc = TAILQ_FIRST(&req_list);
159		TAILQ_REMOVE(&req_list, epc, entry);
160		epc->entry.tqe_prev = NULL;
161		spin_unlock(&req_lock);
162		if (epc->so)
163			process_socket_event((struct c4iw_ep *)epc);
164		c4iw_put_ep(epc);
165		spin_lock(&req_lock);
166	}
167	spin_unlock(&req_lock);
168}
169
170/*
171 * XXX: doesn't belong here in the iWARP driver.
172 * XXX: assumes that the connection was offloaded by cxgbe/t4_tom if TF_TOE is
173 *      set.  Is this a valid assumption for active open?
174 */
175static int
176set_tcpinfo(struct c4iw_ep *ep)
177{
178	struct socket *so = ep->com.so;
179	struct inpcb *inp = sotoinpcb(so);
180	struct tcpcb *tp;
181	struct toepcb *toep;
182	int rc = 0;
183
184	INP_WLOCK(inp);
185	tp = intotcpcb(inp);
186	if ((tp->t_flags & TF_TOE) == 0) {
187		rc = EINVAL;
188		log(LOG_ERR, "%s: connection not offloaded (so %p, ep %p)\n",
189		    __func__, so, ep);
190		goto done;
191	}
192	toep = TOEPCB(so);
193
194	ep->hwtid = toep->tid;
195	ep->snd_seq = tp->snd_nxt;
196	ep->rcv_seq = tp->rcv_nxt;
197	ep->emss = max(tp->t_maxseg, 128);
198done:
199	INP_WUNLOCK(inp);
200	return (rc);
201
202}
203
204static struct rtentry *
205find_route(__be32 local_ip, __be32 peer_ip, __be16 local_port,
206		__be16 peer_port, u8 tos)
207{
208	struct route iproute;
209	struct sockaddr_in *dst = (struct sockaddr_in *)&iproute.ro_dst;
210
211	CTR5(KTR_IW_CXGBE, "%s:frtB %x, %x, %d, %d", __func__, local_ip,
212	    peer_ip, ntohs(local_port), ntohs(peer_port));
213	bzero(&iproute, sizeof iproute);
214	dst->sin_family = AF_INET;
215	dst->sin_len = sizeof *dst;
216	dst->sin_addr.s_addr = peer_ip;
217
218	rtalloc(&iproute);
219	CTR2(KTR_IW_CXGBE, "%s:frtE %p", __func__, (uint64_t)iproute.ro_rt);
220	return iproute.ro_rt;
221}
222
223static int
224close_socket(struct c4iw_ep_common *epc, int close)
225{
226	struct socket *so = epc->so;
227	int rc;
228
229	CTR4(KTR_IW_CXGBE, "%s: so %p, ep %p, state %s", __func__, epc, so,
230	    states[epc->state]);
231
232	SOCK_LOCK(so);
233	soupcall_clear(so, SO_RCV);
234	SOCK_UNLOCK(so);
235
236	if (close)
237                rc = soclose(so);
238        else
239                rc = soshutdown(so, SHUT_WR | SHUT_RD);
240	epc->so = NULL;
241
242	return (rc);
243}
244
245static int
246shutdown_socket(struct c4iw_ep_common *epc)
247{
248
249	CTR4(KTR_IW_CXGBE, "%s: so %p, ep %p, state %s", __func__, epc->so, epc,
250	    states[epc->state]);
251
252	return (soshutdown(epc->so, SHUT_WR));
253}
254
255static void
256abort_socket(struct c4iw_ep *ep)
257{
258	struct sockopt sopt;
259	int rc;
260	struct linger l;
261
262	CTR4(KTR_IW_CXGBE, "%s ep %p so %p state %s", __func__, ep, ep->com.so,
263	    states[ep->com.state]);
264
265	l.l_onoff = 1;
266	l.l_linger = 0;
267
268	/* linger_time of 0 forces RST to be sent */
269	sopt.sopt_dir = SOPT_SET;
270	sopt.sopt_level = SOL_SOCKET;
271	sopt.sopt_name = SO_LINGER;
272	sopt.sopt_val = (caddr_t)&l;
273	sopt.sopt_valsize = sizeof l;
274	sopt.sopt_td = NULL;
275	rc = sosetopt(ep->com.so, &sopt);
276	if (rc) {
277		log(LOG_ERR, "%s: can't set linger to 0, no RST! err %d\n",
278		    __func__, rc);
279	}
280}
281
282static void
283process_peer_close(struct c4iw_ep *ep)
284{
285	struct c4iw_qp_attributes attrs;
286	int disconnect = 1;
287	int release = 0;
288
289	CTR4(KTR_IW_CXGBE, "%s:ppcB ep %p so %p state %s", __func__, ep,
290	    ep->com.so, states[ep->com.state]);
291
292	mutex_lock(&ep->com.mutex);
293	switch (ep->com.state) {
294
295		case MPA_REQ_WAIT:
296			CTR2(KTR_IW_CXGBE, "%s:ppc1 %p MPA_REQ_WAIT CLOSING",
297			    __func__, ep);
298			__state_set(&ep->com, CLOSING);
299			break;
300
301		case MPA_REQ_SENT:
302			CTR2(KTR_IW_CXGBE, "%s:ppc2 %p MPA_REQ_SENT CLOSING",
303			    __func__, ep);
304			__state_set(&ep->com, DEAD);
305			connect_reply_upcall(ep, -ECONNABORTED);
306
307			disconnect = 0;
308			STOP_EP_TIMER(ep);
309			close_socket(&ep->com, 0);
310			ep->com.cm_id->rem_ref(ep->com.cm_id);
311			ep->com.cm_id = NULL;
312			ep->com.qp = NULL;
313			release = 1;
314			break;
315
316		case MPA_REQ_RCVD:
317
318			/*
319			 * We're gonna mark this puppy DEAD, but keep
320			 * the reference on it until the ULP accepts or
321			 * rejects the CR.
322			 */
323			CTR2(KTR_IW_CXGBE, "%s:ppc3 %p MPA_REQ_RCVD CLOSING",
324			    __func__, ep);
325			__state_set(&ep->com, CLOSING);
326			c4iw_get_ep(&ep->com);
327			break;
328
329		case MPA_REP_SENT:
330			CTR2(KTR_IW_CXGBE, "%s:ppc4 %p MPA_REP_SENT CLOSING",
331			    __func__, ep);
332			__state_set(&ep->com, CLOSING);
333			break;
334
335		case FPDU_MODE:
336			CTR2(KTR_IW_CXGBE, "%s:ppc5 %p FPDU_MODE CLOSING",
337			    __func__, ep);
338			START_EP_TIMER(ep);
339			__state_set(&ep->com, CLOSING);
340			attrs.next_state = C4IW_QP_STATE_CLOSING;
341			c4iw_modify_qp(ep->com.dev, ep->com.qp,
342					C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
343			peer_close_upcall(ep);
344			break;
345
346		case ABORTING:
347			CTR2(KTR_IW_CXGBE, "%s:ppc6 %p ABORTING (disconn)",
348			    __func__, ep);
349			disconnect = 0;
350			break;
351
352		case CLOSING:
353			CTR2(KTR_IW_CXGBE, "%s:ppc7 %p CLOSING MORIBUND",
354			    __func__, ep);
355			__state_set(&ep->com, MORIBUND);
356			disconnect = 0;
357			break;
358
359		case MORIBUND:
360			CTR2(KTR_IW_CXGBE, "%s:ppc8 %p MORIBUND DEAD", __func__,
361			    ep);
362			STOP_EP_TIMER(ep);
363			if (ep->com.cm_id && ep->com.qp) {
364				attrs.next_state = C4IW_QP_STATE_IDLE;
365				c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
366						C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
367			}
368			close_socket(&ep->com, 0);
369			close_complete_upcall(ep);
370			__state_set(&ep->com, DEAD);
371			release = 1;
372			disconnect = 0;
373			break;
374
375		case DEAD:
376			CTR2(KTR_IW_CXGBE, "%s:ppc9 %p DEAD (disconn)",
377			    __func__, ep);
378			disconnect = 0;
379			break;
380
381		default:
382			panic("%s: ep %p state %d", __func__, ep,
383			    ep->com.state);
384			break;
385	}
386
387	mutex_unlock(&ep->com.mutex);
388
389	if (disconnect) {
390
391		CTR2(KTR_IW_CXGBE, "%s:ppca %p", __func__, ep);
392		c4iw_ep_disconnect(ep, 0, M_NOWAIT);
393	}
394	if (release) {
395
396		CTR2(KTR_IW_CXGBE, "%s:ppcb %p", __func__, ep);
397		c4iw_put_ep(&ep->com);
398	}
399	CTR2(KTR_IW_CXGBE, "%s:ppcE %p", __func__, ep);
400	return;
401}
402
403static void
404process_conn_error(struct c4iw_ep *ep)
405{
406	struct c4iw_qp_attributes attrs;
407	int ret;
408	int state;
409
410	state = state_read(&ep->com);
411	CTR5(KTR_IW_CXGBE, "%s:pceB ep %p so %p so->so_error %u state %s",
412	    __func__, ep, ep->com.so, ep->com.so->so_error,
413	    states[ep->com.state]);
414
415	switch (state) {
416
417		case MPA_REQ_WAIT:
418			STOP_EP_TIMER(ep);
419			break;
420
421		case MPA_REQ_SENT:
422			STOP_EP_TIMER(ep);
423			connect_reply_upcall(ep, -ECONNRESET);
424			break;
425
426		case MPA_REP_SENT:
427			ep->com.rpl_err = ECONNRESET;
428			CTR1(KTR_IW_CXGBE, "waking up ep %p", ep);
429			break;
430
431		case MPA_REQ_RCVD:
432
433			/*
434			 * We're gonna mark this puppy DEAD, but keep
435			 * the reference on it until the ULP accepts or
436			 * rejects the CR.
437			 */
438			c4iw_get_ep(&ep->com);
439			break;
440
441		case MORIBUND:
442		case CLOSING:
443			STOP_EP_TIMER(ep);
444			/*FALLTHROUGH*/
445		case FPDU_MODE:
446
447			if (ep->com.cm_id && ep->com.qp) {
448
449				attrs.next_state = C4IW_QP_STATE_ERROR;
450				ret = c4iw_modify_qp(ep->com.qp->rhp,
451					ep->com.qp, C4IW_QP_ATTR_NEXT_STATE,
452					&attrs, 1);
453				if (ret)
454					log(LOG_ERR,
455							"%s - qp <- error failed!\n",
456							__func__);
457			}
458			peer_abort_upcall(ep);
459			break;
460
461		case ABORTING:
462			break;
463
464		case DEAD:
465			CTR2(KTR_IW_CXGBE, "%s so_error %d IN DEAD STATE!!!!",
466			    __func__, ep->com.so->so_error);
467			return;
468
469		default:
470			panic("%s: ep %p state %d", __func__, ep, state);
471			break;
472	}
473
474	if (state != ABORTING) {
475
476		CTR2(KTR_IW_CXGBE, "%s:pce1 %p", __func__, ep);
477		close_socket(&ep->com, 0);
478		state_set(&ep->com, DEAD);
479		c4iw_put_ep(&ep->com);
480	}
481	CTR2(KTR_IW_CXGBE, "%s:pceE %p", __func__, ep);
482	return;
483}
484
485static void
486process_close_complete(struct c4iw_ep *ep)
487{
488	struct c4iw_qp_attributes attrs;
489	int release = 0;
490
491	CTR4(KTR_IW_CXGBE, "%s:pccB ep %p so %p state %s", __func__, ep,
492	    ep->com.so, states[ep->com.state]);
493
494	/* The cm_id may be null if we failed to connect */
495	mutex_lock(&ep->com.mutex);
496
497	switch (ep->com.state) {
498
499		case CLOSING:
500			CTR2(KTR_IW_CXGBE, "%s:pcc1 %p CLOSING MORIBUND",
501			    __func__, ep);
502			__state_set(&ep->com, MORIBUND);
503			break;
504
505		case MORIBUND:
506			CTR2(KTR_IW_CXGBE, "%s:pcc1 %p MORIBUND DEAD", __func__,
507			    ep);
508			STOP_EP_TIMER(ep);
509
510			if ((ep->com.cm_id) && (ep->com.qp)) {
511
512				CTR2(KTR_IW_CXGBE, "%s:pcc2 %p QP_STATE_IDLE",
513				    __func__, ep);
514				attrs.next_state = C4IW_QP_STATE_IDLE;
515				c4iw_modify_qp(ep->com.dev,
516						ep->com.qp,
517						C4IW_QP_ATTR_NEXT_STATE,
518						&attrs, 1);
519			}
520
521			if (ep->parent_ep) {
522
523				CTR2(KTR_IW_CXGBE, "%s:pcc3 %p", __func__, ep);
524				close_socket(&ep->com, 1);
525			}
526			else {
527
528				CTR2(KTR_IW_CXGBE, "%s:pcc4 %p", __func__, ep);
529				close_socket(&ep->com, 0);
530			}
531			close_complete_upcall(ep);
532			__state_set(&ep->com, DEAD);
533			release = 1;
534			break;
535
536		case ABORTING:
537			CTR2(KTR_IW_CXGBE, "%s:pcc5 %p ABORTING", __func__, ep);
538			break;
539
540		case DEAD:
541		default:
542			CTR2(KTR_IW_CXGBE, "%s:pcc6 %p DEAD", __func__, ep);
543			panic("%s:pcc6 %p DEAD", __func__, ep);
544			break;
545	}
546	mutex_unlock(&ep->com.mutex);
547
548	if (release) {
549
550		CTR2(KTR_IW_CXGBE, "%s:pcc7 %p", __func__, ep);
551		c4iw_put_ep(&ep->com);
552	}
553	CTR2(KTR_IW_CXGBE, "%s:pccE %p", __func__, ep);
554	return;
555}
556
557static void
558init_sock(struct c4iw_ep_common *epc)
559{
560	int rc;
561	struct sockopt sopt;
562	struct socket *so = epc->so;
563	int on = 1;
564
565	SOCK_LOCK(so);
566	soupcall_set(so, SO_RCV, c4iw_so_upcall, epc);
567	so->so_state |= SS_NBIO;
568	SOCK_UNLOCK(so);
569	sopt.sopt_dir = SOPT_SET;
570	sopt.sopt_level = IPPROTO_TCP;
571	sopt.sopt_name = TCP_NODELAY;
572	sopt.sopt_val = (caddr_t)&on;
573	sopt.sopt_valsize = sizeof on;
574	sopt.sopt_td = NULL;
575	rc = sosetopt(so, &sopt);
576	if (rc) {
577		log(LOG_ERR, "%s: can't set TCP_NODELAY on so %p (%d)\n",
578		    __func__, so, rc);
579	}
580}
581
582static void
583process_data(struct c4iw_ep *ep)
584{
585	struct sockaddr_in *local, *remote;
586
587	CTR5(KTR_IW_CXGBE, "%s: so %p, ep %p, state %s, sb_cc %d", __func__,
588	    ep->com.so, ep, states[ep->com.state], ep->com.so->so_rcv.sb_cc);
589
590	switch (state_read(&ep->com)) {
591	case MPA_REQ_SENT:
592		process_mpa_reply(ep);
593		break;
594	case MPA_REQ_WAIT:
595		in_getsockaddr(ep->com.so, (struct sockaddr **)&local);
596		in_getpeeraddr(ep->com.so, (struct sockaddr **)&remote);
597		ep->com.local_addr = *local;
598		ep->com.remote_addr = *remote;
599		free(local, M_SONAME);
600		free(remote, M_SONAME);
601		process_mpa_request(ep);
602		break;
603	default:
604		if (ep->com.so->so_rcv.sb_cc)
605			log(LOG_ERR, "%s: Unexpected streaming data.  "
606			    "ep %p, state %d, so %p, so_state 0x%x, sb_cc %u\n",
607			    __func__, ep, state_read(&ep->com), ep->com.so,
608			    ep->com.so->so_state, ep->com.so->so_rcv.sb_cc);
609		break;
610	}
611}
612
613static void
614process_connected(struct c4iw_ep *ep)
615{
616
617	if ((ep->com.so->so_state & SS_ISCONNECTED) && !ep->com.so->so_error)
618		send_mpa_req(ep);
619	else {
620		connect_reply_upcall(ep, -ep->com.so->so_error);
621		close_socket(&ep->com, 0);
622		state_set(&ep->com, DEAD);
623		c4iw_put_ep(&ep->com);
624	}
625}
626
627static struct socket *
628dequeue_socket(struct socket *head, struct sockaddr_in **remote,
629    struct c4iw_ep *child_ep)
630{
631	struct socket *so;
632
633	ACCEPT_LOCK();
634	so = TAILQ_FIRST(&head->so_comp);
635	if (!so) {
636		ACCEPT_UNLOCK();
637		return (NULL);
638	}
639	TAILQ_REMOVE(&head->so_comp, so, so_list);
640	head->so_qlen--;
641	SOCK_LOCK(so);
642	so->so_qstate &= ~SQ_COMP;
643	so->so_head = NULL;
644	soref(so);
645	soupcall_set(so, SO_RCV, c4iw_so_upcall, child_ep);
646	so->so_state |= SS_NBIO;
647	SOCK_UNLOCK(so);
648	ACCEPT_UNLOCK();
649	soaccept(so, (struct sockaddr **)remote);
650
651	return (so);
652}
653
654static void
655process_newconn(struct c4iw_ep *parent_ep)
656{
657	struct socket *child_so;
658	struct c4iw_ep *child_ep;
659	struct sockaddr_in *remote;
660
661	child_ep = alloc_ep(sizeof(*child_ep), M_NOWAIT);
662	if (!child_ep) {
663		CTR3(KTR_IW_CXGBE, "%s: parent so %p, parent ep %p, ENOMEM",
664		    __func__, parent_ep->com.so, parent_ep);
665		log(LOG_ERR, "%s: failed to allocate ep entry\n", __func__);
666		return;
667	}
668
669	child_so = dequeue_socket(parent_ep->com.so, &remote, child_ep);
670	if (!child_so) {
671		CTR4(KTR_IW_CXGBE,
672		    "%s: parent so %p, parent ep %p, child ep %p, dequeue err",
673		    __func__, parent_ep->com.so, parent_ep, child_ep);
674		log(LOG_ERR, "%s: failed to dequeue child socket\n", __func__);
675		__free_ep(&child_ep->com);
676		return;
677
678	}
679
680	CTR5(KTR_IW_CXGBE,
681	    "%s: parent so %p, parent ep %p, child so %p, child ep %p",
682	     __func__, parent_ep->com.so, parent_ep, child_so, child_ep);
683
684	child_ep->com.local_addr = parent_ep->com.local_addr;
685	child_ep->com.remote_addr = *remote;
686	child_ep->com.dev = parent_ep->com.dev;
687	child_ep->com.so = child_so;
688	child_ep->com.cm_id = NULL;
689	child_ep->com.thread = parent_ep->com.thread;
690	child_ep->parent_ep = parent_ep;
691
692	free(remote, M_SONAME);
693	c4iw_get_ep(&parent_ep->com);
694	child_ep->parent_ep = parent_ep;
695	init_timer(&child_ep->timer);
696	state_set(&child_ep->com, MPA_REQ_WAIT);
697	START_EP_TIMER(child_ep);
698
699	/* maybe the request has already been queued up on the socket... */
700	process_mpa_request(child_ep);
701}
702
703static int
704c4iw_so_upcall(struct socket *so, void *arg, int waitflag)
705{
706	struct c4iw_ep *ep = arg;
707
708	spin_lock(&req_lock);
709
710	CTR6(KTR_IW_CXGBE,
711	    "%s: so %p, so_state 0x%x, ep %p, ep_state %s, tqe_prev %p",
712	    __func__, so, so->so_state, ep, states[ep->com.state],
713	    ep->com.entry.tqe_prev);
714
715	if (ep && ep->com.so && !ep->com.entry.tqe_prev) {
716		KASSERT(ep->com.so == so, ("%s: XXX review.", __func__));
717		c4iw_get_ep(&ep->com);
718		TAILQ_INSERT_TAIL(&req_list, &ep->com, entry);
719		queue_work(c4iw_taskq, &c4iw_task);
720	}
721
722	spin_unlock(&req_lock);
723	return (SU_OK);
724}
725
726static void
727process_socket_event(struct c4iw_ep *ep)
728{
729	int state = state_read(&ep->com);
730	struct socket *so = ep->com.so;
731
732	CTR6(KTR_IW_CXGBE, "process_socket_event: so %p, so_state 0x%x, "
733	    "so_err %d, sb_state 0x%x, ep %p, ep_state %s", so, so->so_state,
734	    so->so_error, so->so_rcv.sb_state, ep, states[state]);
735
736	if (state == CONNECTING) {
737		process_connected(ep);
738		return;
739	}
740
741	if (state == LISTEN) {
742		process_newconn(ep);
743		return;
744	}
745
746	/* connection error */
747	if (so->so_error) {
748		process_conn_error(ep);
749		return;
750	}
751
752	/* peer close */
753	if ((so->so_rcv.sb_state & SBS_CANTRCVMORE) && state < CLOSING) {
754		process_peer_close(ep);
755		return;
756	}
757
758	/* close complete */
759	if (so->so_state & SS_ISDISCONNECTED) {
760		process_close_complete(ep);
761		return;
762	}
763
764	/* rx data */
765	process_data(ep);
766}
767
768SYSCTL_NODE(_hw, OID_AUTO, iw_cxgbe, CTLFLAG_RD, 0, "iw_cxgbe driver parameters");
769
770int db_delay_usecs = 1;
771TUNABLE_INT("hw.iw_cxgbe.db_delay_usecs", &db_delay_usecs);
772SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, db_delay_usecs, CTLFLAG_RW, &db_delay_usecs, 0,
773		"Usecs to delay awaiting db fifo to drain");
774
775static int dack_mode = 1;
776TUNABLE_INT("hw.iw_cxgbe.dack_mode", &dack_mode);
777SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, dack_mode, CTLFLAG_RW, &dack_mode, 0,
778		"Delayed ack mode (default = 1)");
779
780int c4iw_max_read_depth = 8;
781TUNABLE_INT("hw.iw_cxgbe.c4iw_max_read_depth", &c4iw_max_read_depth);
782SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, c4iw_max_read_depth, CTLFLAG_RW, &c4iw_max_read_depth, 0,
783		"Per-connection max ORD/IRD (default = 8)");
784
785static int enable_tcp_timestamps;
786TUNABLE_INT("hw.iw_cxgbe.enable_tcp_timestamps", &enable_tcp_timestamps);
787SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, enable_tcp_timestamps, CTLFLAG_RW, &enable_tcp_timestamps, 0,
788		"Enable tcp timestamps (default = 0)");
789
790static int enable_tcp_sack;
791TUNABLE_INT("hw.iw_cxgbe.enable_tcp_sack", &enable_tcp_sack);
792SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, enable_tcp_sack, CTLFLAG_RW, &enable_tcp_sack, 0,
793		"Enable tcp SACK (default = 0)");
794
795static int enable_tcp_window_scaling = 1;
796TUNABLE_INT("hw.iw_cxgbe.enable_tcp_window_scaling", &enable_tcp_window_scaling);
797SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, enable_tcp_window_scaling, CTLFLAG_RW, &enable_tcp_window_scaling, 0,
798		"Enable tcp window scaling (default = 1)");
799
800int c4iw_debug = 1;
801TUNABLE_INT("hw.iw_cxgbe.c4iw_debug", &c4iw_debug);
802SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, c4iw_debug, CTLFLAG_RW, &c4iw_debug, 0,
803		"Enable debug logging (default = 0)");
804
805static int peer2peer;
806TUNABLE_INT("hw.iw_cxgbe.peer2peer", &peer2peer);
807SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, peer2peer, CTLFLAG_RW, &peer2peer, 0,
808		"Support peer2peer ULPs (default = 0)");
809
810static int p2p_type = FW_RI_INIT_P2PTYPE_READ_REQ;
811TUNABLE_INT("hw.iw_cxgbe.p2p_type", &p2p_type);
812SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, p2p_type, CTLFLAG_RW, &p2p_type, 0,
813		"RDMAP opcode to use for the RTR message: 1 = RDMA_READ 0 = RDMA_WRITE (default 1)");
814
815static int ep_timeout_secs = 60;
816TUNABLE_INT("hw.iw_cxgbe.ep_timeout_secs", &ep_timeout_secs);
817SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, ep_timeout_secs, CTLFLAG_RW, &ep_timeout_secs, 0,
818		"CM Endpoint operation timeout in seconds (default = 60)");
819
820static int mpa_rev = 1;
821TUNABLE_INT("hw.iw_cxgbe.mpa_rev", &mpa_rev);
822#ifdef IW_CM_MPAV2
823SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, mpa_rev, CTLFLAG_RW, &mpa_rev, 0,
824		"MPA Revision, 0 supports amso1100, 1 is RFC0544 spec compliant, 2 is IETF MPA Peer Connect Draft compliant (default = 1)");
825#else
826SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, mpa_rev, CTLFLAG_RW, &mpa_rev, 0,
827		"MPA Revision, 0 supports amso1100, 1 is RFC0544 spec compliant (default = 1)");
828#endif
829
830static int markers_enabled;
831TUNABLE_INT("hw.iw_cxgbe.markers_enabled", &markers_enabled);
832SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, markers_enabled, CTLFLAG_RW, &markers_enabled, 0,
833		"Enable MPA MARKERS (default(0) = disabled)");
834
835static int crc_enabled = 1;
836TUNABLE_INT("hw.iw_cxgbe.crc_enabled", &crc_enabled);
837SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, crc_enabled, CTLFLAG_RW, &crc_enabled, 0,
838		"Enable MPA CRC (default(1) = enabled)");
839
840static int rcv_win = 256 * 1024;
841TUNABLE_INT("hw.iw_cxgbe.rcv_win", &rcv_win);
842SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, rcv_win, CTLFLAG_RW, &rcv_win, 0,
843		"TCP receive window in bytes (default = 256KB)");
844
845static int snd_win = 128 * 1024;
846TUNABLE_INT("hw.iw_cxgbe.snd_win", &snd_win);
847SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, snd_win, CTLFLAG_RW, &snd_win, 0,
848		"TCP send window in bytes (default = 128KB)");
849
850int db_fc_threshold = 2000;
851TUNABLE_INT("hw.iw_cxgbe.db_fc_threshold", &db_fc_threshold);
852SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, db_fc_threshold, CTLFLAG_RW, &db_fc_threshold, 0,
853		"QP count/threshold that triggers automatic");
854
855static void
856start_ep_timer(struct c4iw_ep *ep)
857{
858
859	if (timer_pending(&ep->timer)) {
860		CTR2(KTR_IW_CXGBE, "%s: ep %p, already started", __func__, ep);
861		printk(KERN_ERR "%s timer already started! ep %p\n", __func__,
862		    ep);
863		return;
864	}
865	clear_bit(TIMEOUT, &ep->com.flags);
866	c4iw_get_ep(&ep->com);
867	ep->timer.expires = jiffies + ep_timeout_secs * HZ;
868	ep->timer.data = (unsigned long)ep;
869	ep->timer.function = ep_timeout;
870	add_timer(&ep->timer);
871}
872
873static void
874stop_ep_timer(struct c4iw_ep *ep)
875{
876
877	del_timer_sync(&ep->timer);
878	if (!test_and_set_bit(TIMEOUT, &ep->com.flags)) {
879		c4iw_put_ep(&ep->com);
880	}
881}
882
883static enum
884c4iw_ep_state state_read(struct c4iw_ep_common *epc)
885{
886	enum c4iw_ep_state state;
887
888	mutex_lock(&epc->mutex);
889	state = epc->state;
890	mutex_unlock(&epc->mutex);
891
892	return (state);
893}
894
895static void
896__state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state new)
897{
898
899	epc->state = new;
900}
901
902static void
903state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state new)
904{
905
906	mutex_lock(&epc->mutex);
907	__state_set(epc, new);
908	mutex_unlock(&epc->mutex);
909}
910
911static void *
912alloc_ep(int size, gfp_t gfp)
913{
914	struct c4iw_ep_common *epc;
915
916	epc = kzalloc(size, gfp);
917	if (epc == NULL)
918		return (NULL);
919
920	kref_init(&epc->kref);
921	mutex_init(&epc->mutex);
922	c4iw_init_wr_wait(&epc->wr_wait);
923
924	return (epc);
925}
926
927void
928__free_ep(struct c4iw_ep_common *epc)
929{
930	CTR2(KTR_IW_CXGBE, "%s:feB %p", __func__, epc);
931	KASSERT(!epc->so, ("%s warning ep->so %p \n", __func__, epc->so));
932	KASSERT(!epc->entry.tqe_prev, ("%s epc %p still on req list!\n", __func__, epc));
933	free(epc, M_DEVBUF);
934	CTR2(KTR_IW_CXGBE, "%s:feE %p", __func__, epc);
935}
936
937void _c4iw_free_ep(struct kref *kref)
938{
939	struct c4iw_ep *ep;
940	struct c4iw_ep_common *epc;
941
942	ep = container_of(kref, struct c4iw_ep, com.kref);
943	epc = &ep->com;
944	KASSERT(!epc->so, ("%s ep->so %p", __func__, epc->so));
945	KASSERT(!epc->entry.tqe_prev, ("%s epc %p still on req list",
946	    __func__, epc));
947	kfree(ep);
948}
949
950static void release_ep_resources(struct c4iw_ep *ep)
951{
952	CTR2(KTR_IW_CXGBE, "%s:rerB %p", __func__, ep);
953	set_bit(RELEASE_RESOURCES, &ep->com.flags);
954	c4iw_put_ep(&ep->com);
955	CTR2(KTR_IW_CXGBE, "%s:rerE %p", __func__, ep);
956}
957
958static void
959send_mpa_req(struct c4iw_ep *ep)
960{
961	int mpalen;
962	struct mpa_message *mpa;
963	struct mpa_v2_conn_params mpa_v2_params;
964	struct mbuf *m;
965	char mpa_rev_to_use = mpa_rev;
966	int err;
967
968	if (ep->retry_with_mpa_v1)
969		mpa_rev_to_use = 1;
970	mpalen = sizeof(*mpa) + ep->plen;
971	if (mpa_rev_to_use == 2)
972		mpalen += sizeof(struct mpa_v2_conn_params);
973
974	if (mpalen > MHLEN)
975		CXGBE_UNIMPLEMENTED(__func__);
976
977	m = m_gethdr(M_NOWAIT, MT_DATA);
978	if (m == NULL) {
979		connect_reply_upcall(ep, -ENOMEM);
980		return;
981	}
982
983	mpa = mtod(m, struct mpa_message *);
984	m->m_len = mpalen;
985	m->m_pkthdr.len = mpalen;
986	memcpy(mpa->key, MPA_KEY_REQ, sizeof(mpa->key));
987	mpa->flags = (crc_enabled ? MPA_CRC : 0) |
988		(markers_enabled ? MPA_MARKERS : 0) |
989		(mpa_rev_to_use == 2 ? MPA_ENHANCED_RDMA_CONN : 0);
990	mpa->private_data_size = htons(ep->plen);
991	mpa->revision = mpa_rev_to_use;
992
993	if (mpa_rev_to_use == 1) {
994		ep->tried_with_mpa_v1 = 1;
995		ep->retry_with_mpa_v1 = 0;
996	}
997
998	if (mpa_rev_to_use == 2) {
999		mpa->private_data_size +=
1000			htons(sizeof(struct mpa_v2_conn_params));
1001		mpa_v2_params.ird = htons((u16)ep->ird);
1002		mpa_v2_params.ord = htons((u16)ep->ord);
1003
1004		if (peer2peer) {
1005			mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL);
1006
1007			if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE) {
1008				mpa_v2_params.ord |=
1009				    htons(MPA_V2_RDMA_WRITE_RTR);
1010			} else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ) {
1011				mpa_v2_params.ord |=
1012					htons(MPA_V2_RDMA_READ_RTR);
1013			}
1014		}
1015		memcpy(mpa->private_data, &mpa_v2_params,
1016			sizeof(struct mpa_v2_conn_params));
1017
1018		if (ep->plen) {
1019
1020			memcpy(mpa->private_data +
1021				sizeof(struct mpa_v2_conn_params),
1022				ep->mpa_pkt + sizeof(*mpa), ep->plen);
1023		}
1024	} else {
1025
1026		if (ep->plen)
1027			memcpy(mpa->private_data,
1028					ep->mpa_pkt + sizeof(*mpa), ep->plen);
1029		CTR2(KTR_IW_CXGBE, "%s:smr7 %p", __func__, ep);
1030	}
1031
1032	err = sosend(ep->com.so, NULL, NULL, m, NULL, MSG_DONTWAIT, ep->com.thread);
1033	if (err) {
1034		connect_reply_upcall(ep, -ENOMEM);
1035		return;
1036	}
1037
1038	START_EP_TIMER(ep);
1039	state_set(&ep->com, MPA_REQ_SENT);
1040	ep->mpa_attr.initiator = 1;
1041}
1042
1043static int send_mpa_reject(struct c4iw_ep *ep, const void *pdata, u8 plen)
1044{
1045	int mpalen ;
1046	struct mpa_message *mpa;
1047	struct mpa_v2_conn_params mpa_v2_params;
1048	struct mbuf *m;
1049	int err;
1050
1051	CTR4(KTR_IW_CXGBE, "%s:smrejB %p %u %d", __func__, ep, ep->hwtid,
1052	    ep->plen);
1053
1054	mpalen = sizeof(*mpa) + plen;
1055
1056	if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
1057
1058		mpalen += sizeof(struct mpa_v2_conn_params);
1059		CTR4(KTR_IW_CXGBE, "%s:smrej1 %p %u %d", __func__, ep,
1060		    ep->mpa_attr.version, mpalen);
1061	}
1062
1063	if (mpalen > MHLEN)
1064		CXGBE_UNIMPLEMENTED(__func__);
1065
1066	m = m_gethdr(M_NOWAIT, MT_DATA);
1067	if (m == NULL) {
1068
1069		printf("%s - cannot alloc mbuf!\n", __func__);
1070		CTR2(KTR_IW_CXGBE, "%s:smrej2 %p", __func__, ep);
1071		return (-ENOMEM);
1072	}
1073
1074
1075	mpa = mtod(m, struct mpa_message *);
1076	m->m_len = mpalen;
1077	m->m_pkthdr.len = mpalen;
1078	memset(mpa, 0, sizeof(*mpa));
1079	memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
1080	mpa->flags = MPA_REJECT;
1081	mpa->revision = mpa_rev;
1082	mpa->private_data_size = htons(plen);
1083
1084	if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
1085
1086		mpa->flags |= MPA_ENHANCED_RDMA_CONN;
1087		mpa->private_data_size +=
1088			htons(sizeof(struct mpa_v2_conn_params));
1089		mpa_v2_params.ird = htons(((u16)ep->ird) |
1090				(peer2peer ? MPA_V2_PEER2PEER_MODEL :
1091				 0));
1092		mpa_v2_params.ord = htons(((u16)ep->ord) | (peer2peer ?
1093					(p2p_type ==
1094					 FW_RI_INIT_P2PTYPE_RDMA_WRITE ?
1095					 MPA_V2_RDMA_WRITE_RTR : p2p_type ==
1096					 FW_RI_INIT_P2PTYPE_READ_REQ ?
1097					 MPA_V2_RDMA_READ_RTR : 0) : 0));
1098		memcpy(mpa->private_data, &mpa_v2_params,
1099				sizeof(struct mpa_v2_conn_params));
1100
1101		if (ep->plen)
1102			memcpy(mpa->private_data +
1103					sizeof(struct mpa_v2_conn_params), pdata, plen);
1104		CTR5(KTR_IW_CXGBE, "%s:smrej3 %p %d %d %d", __func__, ep,
1105		    mpa_v2_params.ird, mpa_v2_params.ord, ep->plen);
1106	} else
1107		if (plen)
1108			memcpy(mpa->private_data, pdata, plen);
1109
1110	err = sosend(ep->com.so, NULL, NULL, m, NULL, MSG_DONTWAIT, ep->com.thread);
1111	if (!err)
1112		ep->snd_seq += mpalen;
1113	CTR4(KTR_IW_CXGBE, "%s:smrejE %p %u %d", __func__, ep, ep->hwtid, err);
1114	return err;
1115}
1116
1117static int send_mpa_reply(struct c4iw_ep *ep, const void *pdata, u8 plen)
1118{
1119	int mpalen;
1120	struct mpa_message *mpa;
1121	struct mbuf *m;
1122	struct mpa_v2_conn_params mpa_v2_params;
1123	int err;
1124
1125	CTR2(KTR_IW_CXGBE, "%s:smrepB %p", __func__, ep);
1126
1127	mpalen = sizeof(*mpa) + plen;
1128
1129	if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
1130
1131		CTR3(KTR_IW_CXGBE, "%s:smrep1 %p %d", __func__, ep,
1132		    ep->mpa_attr.version);
1133		mpalen += sizeof(struct mpa_v2_conn_params);
1134	}
1135
1136	if (mpalen > MHLEN)
1137		CXGBE_UNIMPLEMENTED(__func__);
1138
1139	m = m_gethdr(M_NOWAIT, MT_DATA);
1140	if (m == NULL) {
1141
1142		CTR2(KTR_IW_CXGBE, "%s:smrep2 %p", __func__, ep);
1143		printf("%s - cannot alloc mbuf!\n", __func__);
1144		return (-ENOMEM);
1145	}
1146
1147
1148	mpa = mtod(m, struct mpa_message *);
1149	m->m_len = mpalen;
1150	m->m_pkthdr.len = mpalen;
1151	memset(mpa, 0, sizeof(*mpa));
1152	memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
1153	mpa->flags = (ep->mpa_attr.crc_enabled ? MPA_CRC : 0) |
1154		(markers_enabled ? MPA_MARKERS : 0);
1155	mpa->revision = ep->mpa_attr.version;
1156	mpa->private_data_size = htons(plen);
1157
1158	if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
1159
1160		mpa->flags |= MPA_ENHANCED_RDMA_CONN;
1161		mpa->private_data_size +=
1162			htons(sizeof(struct mpa_v2_conn_params));
1163		mpa_v2_params.ird = htons((u16)ep->ird);
1164		mpa_v2_params.ord = htons((u16)ep->ord);
1165		CTR5(KTR_IW_CXGBE, "%s:smrep3 %p %d %d %d", __func__, ep,
1166		    ep->mpa_attr.version, mpa_v2_params.ird, mpa_v2_params.ord);
1167
1168		if (peer2peer && (ep->mpa_attr.p2p_type !=
1169			FW_RI_INIT_P2PTYPE_DISABLED)) {
1170
1171			mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL);
1172
1173			if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE) {
1174
1175				mpa_v2_params.ord |=
1176					htons(MPA_V2_RDMA_WRITE_RTR);
1177				CTR5(KTR_IW_CXGBE, "%s:smrep4 %p %d %d %d",
1178				    __func__, ep, p2p_type, mpa_v2_params.ird,
1179				    mpa_v2_params.ord);
1180			}
1181			else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ) {
1182
1183				mpa_v2_params.ord |=
1184					htons(MPA_V2_RDMA_READ_RTR);
1185				CTR5(KTR_IW_CXGBE, "%s:smrep5 %p %d %d %d",
1186				    __func__, ep, p2p_type, mpa_v2_params.ird,
1187				    mpa_v2_params.ord);
1188			}
1189		}
1190
1191		memcpy(mpa->private_data, &mpa_v2_params,
1192			sizeof(struct mpa_v2_conn_params));
1193
1194		if (ep->plen)
1195			memcpy(mpa->private_data +
1196				sizeof(struct mpa_v2_conn_params), pdata, plen);
1197	} else
1198		if (plen)
1199			memcpy(mpa->private_data, pdata, plen);
1200
1201	state_set(&ep->com, MPA_REP_SENT);
1202	ep->snd_seq += mpalen;
1203	err = sosend(ep->com.so, NULL, NULL, m, NULL, MSG_DONTWAIT,
1204			ep->com.thread);
1205	CTR3(KTR_IW_CXGBE, "%s:smrepE %p %d", __func__, ep, err);
1206	return err;
1207}
1208
1209
1210
1211static void close_complete_upcall(struct c4iw_ep *ep)
1212{
1213	struct iw_cm_event event;
1214
1215	CTR2(KTR_IW_CXGBE, "%s:ccuB %p", __func__, ep);
1216	memset(&event, 0, sizeof(event));
1217	event.event = IW_CM_EVENT_CLOSE;
1218
1219	if (ep->com.cm_id) {
1220
1221		CTR2(KTR_IW_CXGBE, "%s:ccu1 %1", __func__, ep);
1222		ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1223		ep->com.cm_id->rem_ref(ep->com.cm_id);
1224		ep->com.cm_id = NULL;
1225		ep->com.qp = NULL;
1226		set_bit(CLOSE_UPCALL, &ep->com.history);
1227	}
1228	CTR2(KTR_IW_CXGBE, "%s:ccuE %p", __func__, ep);
1229}
1230
1231static int abort_connection(struct c4iw_ep *ep)
1232{
1233	int err;
1234
1235	CTR2(KTR_IW_CXGBE, "%s:abB %p", __func__, ep);
1236	close_complete_upcall(ep);
1237	state_set(&ep->com, ABORTING);
1238	abort_socket(ep);
1239	err = close_socket(&ep->com, 0);
1240	set_bit(ABORT_CONN, &ep->com.history);
1241	CTR2(KTR_IW_CXGBE, "%s:abE %p", __func__, ep);
1242	return err;
1243}
1244
1245static void peer_close_upcall(struct c4iw_ep *ep)
1246{
1247	struct iw_cm_event event;
1248
1249	CTR2(KTR_IW_CXGBE, "%s:pcuB %p", __func__, ep);
1250	memset(&event, 0, sizeof(event));
1251	event.event = IW_CM_EVENT_DISCONNECT;
1252
1253	if (ep->com.cm_id) {
1254
1255		CTR2(KTR_IW_CXGBE, "%s:pcu1 %p", __func__, ep);
1256		ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1257		set_bit(DISCONN_UPCALL, &ep->com.history);
1258	}
1259	CTR2(KTR_IW_CXGBE, "%s:pcuE %p", __func__, ep);
1260}
1261
1262static void peer_abort_upcall(struct c4iw_ep *ep)
1263{
1264	struct iw_cm_event event;
1265
1266	CTR2(KTR_IW_CXGBE, "%s:pauB %p", __func__, ep);
1267	memset(&event, 0, sizeof(event));
1268	event.event = IW_CM_EVENT_CLOSE;
1269	event.status = -ECONNRESET;
1270
1271	if (ep->com.cm_id) {
1272
1273		CTR2(KTR_IW_CXGBE, "%s:pau1 %p", __func__, ep);
1274		ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1275		ep->com.cm_id->rem_ref(ep->com.cm_id);
1276		ep->com.cm_id = NULL;
1277		ep->com.qp = NULL;
1278		set_bit(ABORT_UPCALL, &ep->com.history);
1279	}
1280	CTR2(KTR_IW_CXGBE, "%s:pauE %p", __func__, ep);
1281}
1282
1283static void connect_reply_upcall(struct c4iw_ep *ep, int status)
1284{
1285	struct iw_cm_event event;
1286
1287	CTR3(KTR_IW_CXGBE, "%s:cruB %p", __func__, ep, status);
1288	memset(&event, 0, sizeof(event));
1289	event.event = IW_CM_EVENT_CONNECT_REPLY;
1290	event.status = (status ==-ECONNABORTED)?-ECONNRESET: status;
1291	event.local_addr = ep->com.local_addr;
1292	event.remote_addr = ep->com.remote_addr;
1293
1294	if ((status == 0) || (status == -ECONNREFUSED)) {
1295
1296		if (!ep->tried_with_mpa_v1) {
1297
1298			CTR2(KTR_IW_CXGBE, "%s:cru1 %p", __func__, ep);
1299			/* this means MPA_v2 is used */
1300			event.private_data_len = ep->plen -
1301				sizeof(struct mpa_v2_conn_params);
1302			event.private_data = ep->mpa_pkt +
1303				sizeof(struct mpa_message) +
1304				sizeof(struct mpa_v2_conn_params);
1305		} else {
1306
1307			CTR2(KTR_IW_CXGBE, "%s:cru2 %p", __func__, ep);
1308			/* this means MPA_v1 is used */
1309			event.private_data_len = ep->plen;
1310			event.private_data = ep->mpa_pkt +
1311				sizeof(struct mpa_message);
1312		}
1313	}
1314
1315	if (ep->com.cm_id) {
1316
1317		CTR2(KTR_IW_CXGBE, "%s:cru3 %p", __func__, ep);
1318		set_bit(CONN_RPL_UPCALL, &ep->com.history);
1319		ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1320	}
1321
1322	if(status == -ECONNABORTED) {
1323
1324		CTR3(KTR_IW_CXGBE, "%s:cruE %p %d", __func__, ep, status);
1325		return;
1326	}
1327
1328	if (status < 0) {
1329
1330		CTR3(KTR_IW_CXGBE, "%s:cru4 %p %d", __func__, ep, status);
1331		ep->com.cm_id->rem_ref(ep->com.cm_id);
1332		ep->com.cm_id = NULL;
1333		ep->com.qp = NULL;
1334	}
1335
1336	CTR2(KTR_IW_CXGBE, "%s:cruE %p", __func__, ep);
1337}
1338
1339static void connect_request_upcall(struct c4iw_ep *ep)
1340{
1341	struct iw_cm_event event;
1342
1343	CTR3(KTR_IW_CXGBE, "%s: ep %p, mpa_v1 %d", __func__, ep,
1344	    ep->tried_with_mpa_v1);
1345
1346	memset(&event, 0, sizeof(event));
1347	event.event = IW_CM_EVENT_CONNECT_REQUEST;
1348	event.local_addr = ep->com.local_addr;
1349	event.remote_addr = ep->com.remote_addr;
1350	event.provider_data = ep;
1351	event.so = ep->com.so;
1352
1353	if (!ep->tried_with_mpa_v1) {
1354		/* this means MPA_v2 is used */
1355#ifdef IW_CM_MPAV2
1356		event.ord = ep->ord;
1357		event.ird = ep->ird;
1358#endif
1359		event.private_data_len = ep->plen -
1360			sizeof(struct mpa_v2_conn_params);
1361		event.private_data = ep->mpa_pkt + sizeof(struct mpa_message) +
1362			sizeof(struct mpa_v2_conn_params);
1363	} else {
1364
1365		/* this means MPA_v1 is used. Send max supported */
1366#ifdef IW_CM_MPAV2
1367		event.ord = c4iw_max_read_depth;
1368		event.ird = c4iw_max_read_depth;
1369#endif
1370		event.private_data_len = ep->plen;
1371		event.private_data = ep->mpa_pkt + sizeof(struct mpa_message);
1372	}
1373
1374	c4iw_get_ep(&ep->com);
1375	ep->parent_ep->com.cm_id->event_handler(ep->parent_ep->com.cm_id,
1376	    &event);
1377	set_bit(CONNREQ_UPCALL, &ep->com.history);
1378	c4iw_put_ep(&ep->parent_ep->com);
1379}
1380
1381static void established_upcall(struct c4iw_ep *ep)
1382{
1383	struct iw_cm_event event;
1384
1385	CTR2(KTR_IW_CXGBE, "%s:euB %p", __func__, ep);
1386	memset(&event, 0, sizeof(event));
1387	event.event = IW_CM_EVENT_ESTABLISHED;
1388#ifdef IW_CM_MPAV2
1389	event.ird = ep->ird;
1390	event.ord = ep->ord;
1391#endif
1392	if (ep->com.cm_id) {
1393
1394		CTR2(KTR_IW_CXGBE, "%s:eu1 %p", __func__, ep);
1395		ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1396		set_bit(ESTAB_UPCALL, &ep->com.history);
1397	}
1398	CTR2(KTR_IW_CXGBE, "%s:euE %p", __func__, ep);
1399}
1400
1401
1402
1403static void process_mpa_reply(struct c4iw_ep *ep)
1404{
1405	struct mpa_message *mpa;
1406	struct mpa_v2_conn_params *mpa_v2_params;
1407	u16 plen;
1408	u16 resp_ird, resp_ord;
1409	u8 rtr_mismatch = 0, insuff_ird = 0;
1410	struct c4iw_qp_attributes attrs;
1411	enum c4iw_qp_attr_mask mask;
1412	int err;
1413	struct mbuf *top, *m;
1414	int flags = MSG_DONTWAIT;
1415	struct uio uio;
1416
1417	CTR2(KTR_IW_CXGBE, "%s:pmrB %p", __func__, ep);
1418
1419	/*
1420	 * Stop mpa timer.  If it expired, then the state has
1421	 * changed and we bail since ep_timeout already aborted
1422	 * the connection.
1423	 */
1424	STOP_EP_TIMER(ep);
1425	if (state_read(&ep->com) != MPA_REQ_SENT)
1426		return;
1427
1428	uio.uio_resid = 1000000;
1429	uio.uio_td = ep->com.thread;
1430	err = soreceive(ep->com.so, NULL, &uio, &top, NULL, &flags);
1431
1432	if (err) {
1433
1434		if (err == EWOULDBLOCK) {
1435
1436			CTR2(KTR_IW_CXGBE, "%s:pmr1 %p", __func__, ep);
1437			START_EP_TIMER(ep);
1438			return;
1439		}
1440		err = -err;
1441		CTR2(KTR_IW_CXGBE, "%s:pmr2 %p", __func__, ep);
1442		goto err;
1443	}
1444
1445	if (ep->com.so->so_rcv.sb_mb) {
1446
1447		CTR2(KTR_IW_CXGBE, "%s:pmr3 %p", __func__, ep);
1448		printf("%s data after soreceive called! so %p sb_mb %p top %p\n",
1449		       __func__, ep->com.so, ep->com.so->so_rcv.sb_mb, top);
1450	}
1451
1452	m = top;
1453
1454	do {
1455
1456		CTR2(KTR_IW_CXGBE, "%s:pmr4 %p", __func__, ep);
1457		/*
1458		 * If we get more than the supported amount of private data
1459		 * then we must fail this connection.
1460		 */
1461		if (ep->mpa_pkt_len + m->m_len > sizeof(ep->mpa_pkt)) {
1462
1463			CTR3(KTR_IW_CXGBE, "%s:pmr5 %p %d", __func__, ep,
1464			    ep->mpa_pkt_len + m->m_len);
1465			err = (-EINVAL);
1466			goto err;
1467		}
1468
1469		/*
1470		 * copy the new data into our accumulation buffer.
1471		 */
1472		m_copydata(m, 0, m->m_len, &(ep->mpa_pkt[ep->mpa_pkt_len]));
1473		ep->mpa_pkt_len += m->m_len;
1474		if (!m->m_next)
1475			m = m->m_nextpkt;
1476		else
1477			m = m->m_next;
1478	} while (m);
1479
1480	m_freem(top);
1481	/*
1482	 * if we don't even have the mpa message, then bail.
1483	 */
1484	if (ep->mpa_pkt_len < sizeof(*mpa))
1485		return;
1486	mpa = (struct mpa_message *) ep->mpa_pkt;
1487
1488	/* Validate MPA header. */
1489	if (mpa->revision > mpa_rev) {
1490
1491		CTR4(KTR_IW_CXGBE, "%s:pmr6 %p %d %d", __func__, ep,
1492		    mpa->revision, mpa_rev);
1493		printk(KERN_ERR MOD "%s MPA version mismatch. Local = %d, "
1494				" Received = %d\n", __func__, mpa_rev, mpa->revision);
1495		err = -EPROTO;
1496		goto err;
1497	}
1498
1499	if (memcmp(mpa->key, MPA_KEY_REP, sizeof(mpa->key))) {
1500
1501		CTR2(KTR_IW_CXGBE, "%s:pmr7 %p", __func__, ep);
1502		err = -EPROTO;
1503		goto err;
1504	}
1505
1506	plen = ntohs(mpa->private_data_size);
1507
1508	/*
1509	 * Fail if there's too much private data.
1510	 */
1511	if (plen > MPA_MAX_PRIVATE_DATA) {
1512
1513		CTR2(KTR_IW_CXGBE, "%s:pmr8 %p", __func__, ep);
1514		err = -EPROTO;
1515		goto err;
1516	}
1517
1518	/*
1519	 * If plen does not account for pkt size
1520	 */
1521	if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
1522
1523		CTR2(KTR_IW_CXGBE, "%s:pmr9 %p", __func__, ep);
1524		err = -EPROTO;
1525		goto err;
1526	}
1527
1528	ep->plen = (u8) plen;
1529
1530	/*
1531	 * If we don't have all the pdata yet, then bail.
1532	 * We'll continue process when more data arrives.
1533	 */
1534	if (ep->mpa_pkt_len < (sizeof(*mpa) + plen)) {
1535
1536		CTR2(KTR_IW_CXGBE, "%s:pmra %p", __func__, ep);
1537		return;
1538	}
1539
1540	if (mpa->flags & MPA_REJECT) {
1541
1542		CTR2(KTR_IW_CXGBE, "%s:pmrb %p", __func__, ep);
1543		err = -ECONNREFUSED;
1544		goto err;
1545	}
1546
1547	/*
1548	 * If we get here we have accumulated the entire mpa
1549	 * start reply message including private data. And
1550	 * the MPA header is valid.
1551	 */
1552	state_set(&ep->com, FPDU_MODE);
1553	ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1554	ep->mpa_attr.recv_marker_enabled = markers_enabled;
1555	ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1556	ep->mpa_attr.version = mpa->revision;
1557	ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1558
1559	if (mpa->revision == 2) {
1560
1561		CTR2(KTR_IW_CXGBE, "%s:pmrc %p", __func__, ep);
1562		ep->mpa_attr.enhanced_rdma_conn =
1563			mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
1564
1565		if (ep->mpa_attr.enhanced_rdma_conn) {
1566
1567			CTR2(KTR_IW_CXGBE, "%s:pmrd %p", __func__, ep);
1568			mpa_v2_params = (struct mpa_v2_conn_params *)
1569				(ep->mpa_pkt + sizeof(*mpa));
1570			resp_ird = ntohs(mpa_v2_params->ird) &
1571				MPA_V2_IRD_ORD_MASK;
1572			resp_ord = ntohs(mpa_v2_params->ord) &
1573				MPA_V2_IRD_ORD_MASK;
1574
1575			/*
1576			 * This is a double-check. Ideally, below checks are
1577			 * not required since ird/ord stuff has been taken
1578			 * care of in c4iw_accept_cr
1579			 */
1580			if ((ep->ird < resp_ord) || (ep->ord > resp_ird)) {
1581
1582				CTR2(KTR_IW_CXGBE, "%s:pmre %p", __func__, ep);
1583				err = -ENOMEM;
1584				ep->ird = resp_ord;
1585				ep->ord = resp_ird;
1586				insuff_ird = 1;
1587			}
1588
1589			if (ntohs(mpa_v2_params->ird) &
1590				MPA_V2_PEER2PEER_MODEL) {
1591
1592				CTR2(KTR_IW_CXGBE, "%s:pmrf %p", __func__, ep);
1593				if (ntohs(mpa_v2_params->ord) &
1594					MPA_V2_RDMA_WRITE_RTR) {
1595
1596					CTR2(KTR_IW_CXGBE, "%s:pmrg %p", __func__, ep);
1597					ep->mpa_attr.p2p_type =
1598						FW_RI_INIT_P2PTYPE_RDMA_WRITE;
1599				}
1600				else if (ntohs(mpa_v2_params->ord) &
1601					MPA_V2_RDMA_READ_RTR) {
1602
1603					CTR2(KTR_IW_CXGBE, "%s:pmrh %p", __func__, ep);
1604					ep->mpa_attr.p2p_type =
1605						FW_RI_INIT_P2PTYPE_READ_REQ;
1606				}
1607			}
1608		}
1609	} else {
1610
1611		CTR2(KTR_IW_CXGBE, "%s:pmri %p", __func__, ep);
1612
1613		if (mpa->revision == 1) {
1614
1615			CTR2(KTR_IW_CXGBE, "%s:pmrj %p", __func__, ep);
1616
1617			if (peer2peer) {
1618
1619				CTR2(KTR_IW_CXGBE, "%s:pmrk %p", __func__, ep);
1620				ep->mpa_attr.p2p_type = p2p_type;
1621			}
1622		}
1623	}
1624
1625	if (set_tcpinfo(ep)) {
1626
1627		CTR2(KTR_IW_CXGBE, "%s:pmrl %p", __func__, ep);
1628		printf("%s set_tcpinfo error\n", __func__);
1629		goto err;
1630	}
1631
1632	CTR6(KTR_IW_CXGBE, "%s - crc_enabled = %d, recv_marker_enabled = %d, "
1633	    "xmit_marker_enabled = %d, version = %d p2p_type = %d", __func__,
1634	    ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
1635	    ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version,
1636	    ep->mpa_attr.p2p_type);
1637
1638	/*
1639	 * If responder's RTR does not match with that of initiator, assign
1640	 * FW_RI_INIT_P2PTYPE_DISABLED in mpa attributes so that RTR is not
1641	 * generated when moving QP to RTS state.
1642	 * A TERM message will be sent after QP has moved to RTS state
1643	 */
1644	if ((ep->mpa_attr.version == 2) && peer2peer &&
1645		(ep->mpa_attr.p2p_type != p2p_type)) {
1646
1647		CTR2(KTR_IW_CXGBE, "%s:pmrm %p", __func__, ep);
1648		ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1649		rtr_mismatch = 1;
1650	}
1651
1652
1653	//ep->ofld_txq = TOEPCB(ep->com.so)->ofld_txq;
1654	attrs.mpa_attr = ep->mpa_attr;
1655	attrs.max_ird = ep->ird;
1656	attrs.max_ord = ep->ord;
1657	attrs.llp_stream_handle = ep;
1658	attrs.next_state = C4IW_QP_STATE_RTS;
1659
1660	mask = C4IW_QP_ATTR_NEXT_STATE |
1661		C4IW_QP_ATTR_LLP_STREAM_HANDLE | C4IW_QP_ATTR_MPA_ATTR |
1662		C4IW_QP_ATTR_MAX_IRD | C4IW_QP_ATTR_MAX_ORD;
1663
1664	/* bind QP and TID with INIT_WR */
1665	err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp, mask, &attrs, 1);
1666
1667	if (err) {
1668
1669		CTR2(KTR_IW_CXGBE, "%s:pmrn %p", __func__, ep);
1670		goto err;
1671	}
1672
1673	/*
1674	 * If responder's RTR requirement did not match with what initiator
1675	 * supports, generate TERM message
1676	 */
1677	if (rtr_mismatch) {
1678
1679		CTR2(KTR_IW_CXGBE, "%s:pmro %p", __func__, ep);
1680		printk(KERN_ERR "%s: RTR mismatch, sending TERM\n", __func__);
1681		attrs.layer_etype = LAYER_MPA | DDP_LLP;
1682		attrs.ecode = MPA_NOMATCH_RTR;
1683		attrs.next_state = C4IW_QP_STATE_TERMINATE;
1684		err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1685			C4IW_QP_ATTR_NEXT_STATE, &attrs, 0);
1686		err = -ENOMEM;
1687		goto out;
1688	}
1689
1690	/*
1691	 * Generate TERM if initiator IRD is not sufficient for responder
1692	 * provided ORD. Currently, we do the same behaviour even when
1693	 * responder provided IRD is also not sufficient as regards to
1694	 * initiator ORD.
1695	 */
1696	if (insuff_ird) {
1697
1698		CTR2(KTR_IW_CXGBE, "%s:pmrp %p", __func__, ep);
1699		printk(KERN_ERR "%s: Insufficient IRD, sending TERM\n",
1700				__func__);
1701		attrs.layer_etype = LAYER_MPA | DDP_LLP;
1702		attrs.ecode = MPA_INSUFF_IRD;
1703		attrs.next_state = C4IW_QP_STATE_TERMINATE;
1704		err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1705			C4IW_QP_ATTR_NEXT_STATE, &attrs, 0);
1706		err = -ENOMEM;
1707		goto out;
1708	}
1709	goto out;
1710err:
1711	state_set(&ep->com, ABORTING);
1712	abort_connection(ep);
1713out:
1714	connect_reply_upcall(ep, err);
1715	CTR2(KTR_IW_CXGBE, "%s:pmrE %p", __func__, ep);
1716	return;
1717}
1718
1719static void
1720process_mpa_request(struct c4iw_ep *ep)
1721{
1722	struct mpa_message *mpa;
1723	u16 plen;
1724	int flags = MSG_DONTWAIT;
1725	int rc;
1726	struct iovec iov;
1727	struct uio uio;
1728	enum c4iw_ep_state state = state_read(&ep->com);
1729
1730	CTR3(KTR_IW_CXGBE, "%s: ep %p, state %s", __func__, ep, states[state]);
1731
1732	if (state != MPA_REQ_WAIT)
1733		return;
1734
1735	iov.iov_base = &ep->mpa_pkt[ep->mpa_pkt_len];
1736	iov.iov_len = sizeof(ep->mpa_pkt) - ep->mpa_pkt_len;
1737	uio.uio_iov = &iov;
1738	uio.uio_iovcnt = 1;
1739	uio.uio_offset = 0;
1740	uio.uio_resid = sizeof(ep->mpa_pkt) - ep->mpa_pkt_len;
1741	uio.uio_segflg = UIO_SYSSPACE;
1742	uio.uio_rw = UIO_READ;
1743	uio.uio_td = NULL; /* uio.uio_td = ep->com.thread; */
1744
1745	rc = soreceive(ep->com.so, NULL, &uio, NULL, NULL, &flags);
1746	if (rc == EAGAIN)
1747		return;
1748	else if (rc) {
1749abort:
1750		STOP_EP_TIMER(ep);
1751		abort_connection(ep);
1752		return;
1753	}
1754	KASSERT(uio.uio_offset > 0, ("%s: sorecieve on so %p read no data",
1755	    __func__, ep->com.so));
1756	ep->mpa_pkt_len += uio.uio_offset;
1757
1758	/*
1759	 * If we get more than the supported amount of private data then we must
1760	 * fail this connection.  XXX: check so_rcv->sb_cc, or peek with another
1761	 * soreceive, or increase the size of mpa_pkt by 1 and abort if the last
1762	 * byte is filled by the soreceive above.
1763	 */
1764
1765	/* Don't even have the MPA message.  Wait for more data to arrive. */
1766	if (ep->mpa_pkt_len < sizeof(*mpa))
1767		return;
1768	mpa = (struct mpa_message *) ep->mpa_pkt;
1769
1770	/*
1771	 * Validate MPA Header.
1772	 */
1773	if (mpa->revision > mpa_rev) {
1774		log(LOG_ERR, "%s: MPA version mismatch. Local = %d,"
1775		    " Received = %d\n", __func__, mpa_rev, mpa->revision);
1776		goto abort;
1777	}
1778
1779	if (memcmp(mpa->key, MPA_KEY_REQ, sizeof(mpa->key)))
1780		goto abort;
1781
1782	/*
1783	 * Fail if there's too much private data.
1784	 */
1785	plen = ntohs(mpa->private_data_size);
1786	if (plen > MPA_MAX_PRIVATE_DATA)
1787		goto abort;
1788
1789	/*
1790	 * If plen does not account for pkt size
1791	 */
1792	if (ep->mpa_pkt_len > (sizeof(*mpa) + plen))
1793		goto abort;
1794
1795	ep->plen = (u8) plen;
1796
1797	/*
1798	 * If we don't have all the pdata yet, then bail.
1799	 */
1800	if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
1801		return;
1802
1803	/*
1804	 * If we get here we have accumulated the entire mpa
1805	 * start reply message including private data.
1806	 */
1807	ep->mpa_attr.initiator = 0;
1808	ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1809	ep->mpa_attr.recv_marker_enabled = markers_enabled;
1810	ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1811	ep->mpa_attr.version = mpa->revision;
1812	if (mpa->revision == 1)
1813		ep->tried_with_mpa_v1 = 1;
1814	ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1815
1816	if (mpa->revision == 2) {
1817		ep->mpa_attr.enhanced_rdma_conn =
1818		    mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
1819		if (ep->mpa_attr.enhanced_rdma_conn) {
1820			struct mpa_v2_conn_params *mpa_v2_params;
1821			u16 ird, ord;
1822
1823			mpa_v2_params = (void *)&ep->mpa_pkt[sizeof(*mpa)];
1824			ird = ntohs(mpa_v2_params->ird);
1825			ord = ntohs(mpa_v2_params->ord);
1826
1827			ep->ird = ird & MPA_V2_IRD_ORD_MASK;
1828			ep->ord = ord & MPA_V2_IRD_ORD_MASK;
1829			if (ird & MPA_V2_PEER2PEER_MODEL && peer2peer) {
1830				if (ord & MPA_V2_RDMA_WRITE_RTR) {
1831					ep->mpa_attr.p2p_type =
1832					    FW_RI_INIT_P2PTYPE_RDMA_WRITE;
1833				} else if (ord & MPA_V2_RDMA_READ_RTR) {
1834					ep->mpa_attr.p2p_type =
1835					    FW_RI_INIT_P2PTYPE_READ_REQ;
1836				}
1837			}
1838		}
1839	} else if (mpa->revision == 1 && peer2peer)
1840		ep->mpa_attr.p2p_type = p2p_type;
1841
1842	if (set_tcpinfo(ep))
1843		goto abort;
1844
1845	CTR5(KTR_IW_CXGBE, "%s: crc_enabled = %d, recv_marker_enabled = %d, "
1846	    "xmit_marker_enabled = %d, version = %d", __func__,
1847	    ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
1848	    ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version);
1849
1850	state_set(&ep->com, MPA_REQ_RCVD);
1851	STOP_EP_TIMER(ep);
1852
1853	/* drive upcall */
1854	mutex_lock(&ep->parent_ep->com.mutex);
1855	if (ep->parent_ep->com.state != DEAD)
1856		connect_request_upcall(ep);
1857	else
1858		abort_connection(ep);
1859	mutex_unlock(&ep->parent_ep->com.mutex);
1860}
1861
1862/*
1863 * Upcall from the adapter indicating data has been transmitted.
1864 * For us its just the single MPA request or reply.  We can now free
1865 * the skb holding the mpa message.
1866 */
1867int c4iw_reject_cr(struct iw_cm_id *cm_id, const void *pdata, u8 pdata_len)
1868{
1869	int err;
1870	struct c4iw_ep *ep = to_ep(cm_id);
1871	CTR2(KTR_IW_CXGBE, "%s:crcB %p", __func__, ep);
1872
1873	if (state_read(&ep->com) == DEAD) {
1874
1875		CTR2(KTR_IW_CXGBE, "%s:crc1 %p", __func__, ep);
1876		c4iw_put_ep(&ep->com);
1877		return -ECONNRESET;
1878	}
1879	set_bit(ULP_REJECT, &ep->com.history);
1880	BUG_ON(state_read(&ep->com) != MPA_REQ_RCVD);
1881
1882	if (mpa_rev == 0) {
1883
1884		CTR2(KTR_IW_CXGBE, "%s:crc2 %p", __func__, ep);
1885		abort_connection(ep);
1886	}
1887	else {
1888
1889		CTR2(KTR_IW_CXGBE, "%s:crc3 %p", __func__, ep);
1890		err = send_mpa_reject(ep, pdata, pdata_len);
1891		err = soshutdown(ep->com.so, 3);
1892	}
1893	c4iw_put_ep(&ep->com);
1894	CTR2(KTR_IW_CXGBE, "%s:crc4 %p", __func__, ep);
1895	return 0;
1896}
1897
1898int c4iw_accept_cr(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
1899{
1900	int err;
1901	struct c4iw_qp_attributes attrs;
1902	enum c4iw_qp_attr_mask mask;
1903	struct c4iw_ep *ep = to_ep(cm_id);
1904	struct c4iw_dev *h = to_c4iw_dev(cm_id->device);
1905	struct c4iw_qp *qp = get_qhp(h, conn_param->qpn);
1906
1907	CTR2(KTR_IW_CXGBE, "%s:cacB %p", __func__, ep);
1908
1909	if (state_read(&ep->com) == DEAD) {
1910
1911		CTR2(KTR_IW_CXGBE, "%s:cac1 %p", __func__, ep);
1912		err = -ECONNRESET;
1913		goto err;
1914	}
1915
1916	BUG_ON(state_read(&ep->com) != MPA_REQ_RCVD);
1917	BUG_ON(!qp);
1918
1919	set_bit(ULP_ACCEPT, &ep->com.history);
1920
1921	if ((conn_param->ord > c4iw_max_read_depth) ||
1922		(conn_param->ird > c4iw_max_read_depth)) {
1923
1924		CTR2(KTR_IW_CXGBE, "%s:cac2 %p", __func__, ep);
1925		abort_connection(ep);
1926		err = -EINVAL;
1927		goto err;
1928	}
1929
1930	if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
1931
1932		CTR2(KTR_IW_CXGBE, "%s:cac3 %p", __func__, ep);
1933
1934		if (conn_param->ord > ep->ird) {
1935
1936			CTR2(KTR_IW_CXGBE, "%s:cac4 %p", __func__, ep);
1937			ep->ird = conn_param->ird;
1938			ep->ord = conn_param->ord;
1939			send_mpa_reject(ep, conn_param->private_data,
1940					conn_param->private_data_len);
1941			abort_connection(ep);
1942			err = -ENOMEM;
1943			goto err;
1944		}
1945
1946		if (conn_param->ird > ep->ord) {
1947
1948			CTR2(KTR_IW_CXGBE, "%s:cac5 %p", __func__, ep);
1949
1950			if (!ep->ord) {
1951
1952				CTR2(KTR_IW_CXGBE, "%s:cac6 %p", __func__, ep);
1953				conn_param->ird = 1;
1954			}
1955			else {
1956				CTR2(KTR_IW_CXGBE, "%s:cac7 %p", __func__, ep);
1957				abort_connection(ep);
1958				err = -ENOMEM;
1959				goto err;
1960			}
1961		}
1962
1963	}
1964	ep->ird = conn_param->ird;
1965	ep->ord = conn_param->ord;
1966
1967	if (ep->mpa_attr.version != 2) {
1968
1969		CTR2(KTR_IW_CXGBE, "%s:cac8 %p", __func__, ep);
1970
1971		if (peer2peer && ep->ird == 0) {
1972
1973			CTR2(KTR_IW_CXGBE, "%s:cac9 %p", __func__, ep);
1974			ep->ird = 1;
1975		}
1976	}
1977
1978
1979	cm_id->add_ref(cm_id);
1980	ep->com.cm_id = cm_id;
1981	ep->com.qp = qp;
1982	//ep->ofld_txq = TOEPCB(ep->com.so)->ofld_txq;
1983
1984	/* bind QP to EP and move to RTS */
1985	attrs.mpa_attr = ep->mpa_attr;
1986	attrs.max_ird = ep->ird;
1987	attrs.max_ord = ep->ord;
1988	attrs.llp_stream_handle = ep;
1989	attrs.next_state = C4IW_QP_STATE_RTS;
1990
1991	/* bind QP and TID with INIT_WR */
1992	mask = C4IW_QP_ATTR_NEXT_STATE |
1993		C4IW_QP_ATTR_LLP_STREAM_HANDLE |
1994		C4IW_QP_ATTR_MPA_ATTR |
1995		C4IW_QP_ATTR_MAX_IRD |
1996		C4IW_QP_ATTR_MAX_ORD;
1997
1998	err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp, mask, &attrs, 1);
1999
2000	if (err) {
2001
2002		CTR2(KTR_IW_CXGBE, "%s:caca %p", __func__, ep);
2003		goto err1;
2004	}
2005	err = send_mpa_reply(ep, conn_param->private_data,
2006			conn_param->private_data_len);
2007
2008	if (err) {
2009
2010		CTR2(KTR_IW_CXGBE, "%s:caca %p", __func__, ep);
2011		goto err1;
2012	}
2013
2014	state_set(&ep->com, FPDU_MODE);
2015	established_upcall(ep);
2016	c4iw_put_ep(&ep->com);
2017	CTR2(KTR_IW_CXGBE, "%s:cacE %p", __func__, ep);
2018	return 0;
2019err1:
2020	ep->com.cm_id = NULL;
2021	ep->com.qp = NULL;
2022	cm_id->rem_ref(cm_id);
2023err:
2024	c4iw_put_ep(&ep->com);
2025	CTR2(KTR_IW_CXGBE, "%s:cacE err %p", __func__, ep);
2026	return err;
2027}
2028
2029
2030
2031int c4iw_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
2032{
2033	int err = 0;
2034	struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
2035	struct c4iw_ep *ep = NULL;
2036	struct rtentry *rt;
2037	struct toedev *tdev;
2038
2039	CTR2(KTR_IW_CXGBE, "%s:ccB %p", __func__, cm_id);
2040
2041	if ((conn_param->ord > c4iw_max_read_depth) ||
2042		(conn_param->ird > c4iw_max_read_depth)) {
2043
2044		CTR2(KTR_IW_CXGBE, "%s:cc1 %p", __func__, cm_id);
2045		err = -EINVAL;
2046		goto out;
2047	}
2048	ep = alloc_ep(sizeof(*ep), M_NOWAIT);
2049
2050	if (!ep) {
2051
2052		CTR2(KTR_IW_CXGBE, "%s:cc2 %p", __func__, cm_id);
2053		printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__);
2054		err = -ENOMEM;
2055		goto out;
2056	}
2057	init_timer(&ep->timer);
2058	ep->plen = conn_param->private_data_len;
2059
2060	if (ep->plen) {
2061
2062		CTR2(KTR_IW_CXGBE, "%s:cc3 %p", __func__, ep);
2063		memcpy(ep->mpa_pkt + sizeof(struct mpa_message),
2064				conn_param->private_data, ep->plen);
2065	}
2066	ep->ird = conn_param->ird;
2067	ep->ord = conn_param->ord;
2068
2069	if (peer2peer && ep->ord == 0) {
2070
2071		CTR2(KTR_IW_CXGBE, "%s:cc4 %p", __func__, ep);
2072		ep->ord = 1;
2073	}
2074
2075	cm_id->add_ref(cm_id);
2076	ep->com.dev = dev;
2077	ep->com.cm_id = cm_id;
2078	ep->com.qp = get_qhp(dev, conn_param->qpn);
2079
2080	if (!ep->com.qp) {
2081
2082		CTR2(KTR_IW_CXGBE, "%s:cc5 %p", __func__, ep);
2083		err = -EINVAL;
2084		goto fail2;
2085	}
2086	ep->com.thread = curthread;
2087	ep->com.so = cm_id->so;
2088
2089	init_sock(&ep->com);
2090
2091	/* find a route */
2092	rt = find_route(
2093		cm_id->local_addr.sin_addr.s_addr,
2094		cm_id->remote_addr.sin_addr.s_addr,
2095		cm_id->local_addr.sin_port,
2096		cm_id->remote_addr.sin_port, 0);
2097
2098	if (!rt) {
2099
2100		CTR2(KTR_IW_CXGBE, "%s:cc7 %p", __func__, ep);
2101		printk(KERN_ERR MOD "%s - cannot find route.\n", __func__);
2102		err = -EHOSTUNREACH;
2103		goto fail3;
2104	}
2105
2106
2107	if (!(rt->rt_ifp->if_flags & IFCAP_TOE)) {
2108
2109		CTR2(KTR_IW_CXGBE, "%s:cc8 %p", __func__, ep);
2110		printf("%s - interface not TOE capable.\n", __func__);
2111		goto fail3;
2112	}
2113	tdev = TOEDEV(rt->rt_ifp);
2114
2115	if (tdev == NULL) {
2116
2117		CTR2(KTR_IW_CXGBE, "%s:cc9 %p", __func__, ep);
2118		printf("%s - No toedev for interface.\n", __func__);
2119		goto fail3;
2120	}
2121	RTFREE(rt);
2122
2123	state_set(&ep->com, CONNECTING);
2124	ep->tos = 0;
2125	ep->com.local_addr = cm_id->local_addr;
2126	ep->com.remote_addr = cm_id->remote_addr;
2127	err = soconnect(ep->com.so, (struct sockaddr *)&ep->com.remote_addr,
2128		ep->com.thread);
2129
2130	if (!err) {
2131
2132		CTR2(KTR_IW_CXGBE, "%s:cca %p", __func__, ep);
2133		goto out;
2134	}
2135
2136fail3:
2137	CTR2(KTR_IW_CXGBE, "%s:ccb %p", __func__, ep);
2138	RTFREE(rt);
2139fail2:
2140	cm_id->rem_ref(cm_id);
2141	c4iw_put_ep(&ep->com);
2142out:
2143	CTR2(KTR_IW_CXGBE, "%s:ccE %p", __func__, ep);
2144	return err;
2145}
2146
2147/*
2148 * iwcm->create_listen.  Returns -errno on failure.
2149 */
2150int
2151c4iw_create_listen(struct iw_cm_id *cm_id, int backlog)
2152{
2153	int rc;
2154	struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
2155	struct c4iw_listen_ep *ep;
2156	struct socket *so = cm_id->so;
2157
2158	ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
2159	CTR5(KTR_IW_CXGBE, "%s: cm_id %p, lso %p, ep %p, inp %p", __func__,
2160	    cm_id, so, ep, so->so_pcb);
2161	if (ep == NULL) {
2162		log(LOG_ERR, "%s: failed to alloc memory for endpoint\n",
2163		    __func__);
2164		rc = ENOMEM;
2165		goto failed;
2166	}
2167
2168	cm_id->add_ref(cm_id);
2169	ep->com.cm_id = cm_id;
2170	ep->com.dev = dev;
2171	ep->backlog = backlog;
2172	ep->com.local_addr = cm_id->local_addr;
2173	ep->com.thread = curthread;
2174	state_set(&ep->com, LISTEN);
2175	ep->com.so = so;
2176	init_sock(&ep->com);
2177
2178	rc = solisten(so, ep->backlog, ep->com.thread);
2179	if (rc != 0) {
2180		log(LOG_ERR, "%s: failed to start listener: %d\n", __func__,
2181		    rc);
2182		close_socket(&ep->com, 0);
2183		cm_id->rem_ref(cm_id);
2184		c4iw_put_ep(&ep->com);
2185		goto failed;
2186	}
2187
2188	cm_id->provider_data = ep;
2189	return (0);
2190
2191failed:
2192	CTR3(KTR_IW_CXGBE, "%s: cm_id %p, FAILED (%d)", __func__, cm_id, rc);
2193	return (-rc);
2194}
2195
2196int
2197c4iw_destroy_listen(struct iw_cm_id *cm_id)
2198{
2199	int rc;
2200	struct c4iw_listen_ep *ep = to_listen_ep(cm_id);
2201
2202	CTR4(KTR_IW_CXGBE, "%s: cm_id %p, so %p, inp %p", __func__, cm_id,
2203	    cm_id->so, cm_id->so->so_pcb);
2204
2205	state_set(&ep->com, DEAD);
2206	rc = close_socket(&ep->com, 0);
2207	cm_id->rem_ref(cm_id);
2208	c4iw_put_ep(&ep->com);
2209
2210	return (rc);
2211}
2212
2213int c4iw_ep_disconnect(struct c4iw_ep *ep, int abrupt, gfp_t gfp)
2214{
2215	int ret = 0;
2216	int close = 0;
2217	int fatal = 0;
2218	struct c4iw_rdev *rdev;
2219
2220	mutex_lock(&ep->com.mutex);
2221
2222	CTR2(KTR_IW_CXGBE, "%s:cedB %p", __func__, ep);
2223
2224	rdev = &ep->com.dev->rdev;
2225
2226	if (c4iw_fatal_error(rdev)) {
2227
2228		CTR2(KTR_IW_CXGBE, "%s:ced1 %p", __func__, ep);
2229		fatal = 1;
2230		close_complete_upcall(ep);
2231		ep->com.state = DEAD;
2232	}
2233	CTR3(KTR_IW_CXGBE, "%s:ced2 %p %s", __func__, ep,
2234	    states[ep->com.state]);
2235
2236	switch (ep->com.state) {
2237
2238		case MPA_REQ_WAIT:
2239		case MPA_REQ_SENT:
2240		case MPA_REQ_RCVD:
2241		case MPA_REP_SENT:
2242		case FPDU_MODE:
2243			close = 1;
2244			if (abrupt)
2245				ep->com.state = ABORTING;
2246			else {
2247				ep->com.state = CLOSING;
2248				START_EP_TIMER(ep);
2249			}
2250			set_bit(CLOSE_SENT, &ep->com.flags);
2251			break;
2252
2253		case CLOSING:
2254
2255			if (!test_and_set_bit(CLOSE_SENT, &ep->com.flags)) {
2256
2257				close = 1;
2258				if (abrupt) {
2259					STOP_EP_TIMER(ep);
2260					ep->com.state = ABORTING;
2261				} else
2262					ep->com.state = MORIBUND;
2263			}
2264			break;
2265
2266		case MORIBUND:
2267		case ABORTING:
2268		case DEAD:
2269			CTR3(KTR_IW_CXGBE,
2270			    "%s ignoring disconnect ep %p state %u", __func__,
2271			    ep, ep->com.state);
2272			break;
2273
2274		default:
2275			BUG();
2276			break;
2277	}
2278
2279	mutex_unlock(&ep->com.mutex);
2280
2281	if (close) {
2282
2283		CTR2(KTR_IW_CXGBE, "%s:ced3 %p", __func__, ep);
2284
2285		if (abrupt) {
2286
2287			CTR2(KTR_IW_CXGBE, "%s:ced4 %p", __func__, ep);
2288			set_bit(EP_DISC_ABORT, &ep->com.history);
2289			ret = abort_connection(ep);
2290		} else {
2291
2292			CTR2(KTR_IW_CXGBE, "%s:ced5 %p", __func__, ep);
2293			set_bit(EP_DISC_CLOSE, &ep->com.history);
2294
2295			if (!ep->parent_ep)
2296				__state_set(&ep->com, MORIBUND);
2297			ret = shutdown_socket(&ep->com);
2298		}
2299
2300		if (ret) {
2301
2302			fatal = 1;
2303		}
2304	}
2305
2306	if (fatal) {
2307
2308		release_ep_resources(ep);
2309		CTR2(KTR_IW_CXGBE, "%s:ced6 %p", __func__, ep);
2310	}
2311	CTR2(KTR_IW_CXGBE, "%s:cedE %p", __func__, ep);
2312	return ret;
2313}
2314
2315#ifdef C4IW_EP_REDIRECT
2316int c4iw_ep_redirect(void *ctx, struct dst_entry *old, struct dst_entry *new,
2317		struct l2t_entry *l2t)
2318{
2319	struct c4iw_ep *ep = ctx;
2320
2321	if (ep->dst != old)
2322		return 0;
2323
2324	PDBG("%s ep %p redirect to dst %p l2t %p\n", __func__, ep, new,
2325			l2t);
2326	dst_hold(new);
2327	cxgb4_l2t_release(ep->l2t);
2328	ep->l2t = l2t;
2329	dst_release(old);
2330	ep->dst = new;
2331	return 1;
2332}
2333#endif
2334
2335
2336
2337static void ep_timeout(unsigned long arg)
2338{
2339	struct c4iw_ep *ep = (struct c4iw_ep *)arg;
2340	int kickit = 0;
2341
2342	CTR2(KTR_IW_CXGBE, "%s:etB %p", __func__, ep);
2343	spin_lock(&timeout_lock);
2344
2345	if (!test_and_set_bit(TIMEOUT, &ep->com.flags)) {
2346
2347		list_add_tail(&ep->entry, &timeout_list);
2348		kickit = 1;
2349	}
2350	spin_unlock(&timeout_lock);
2351
2352	if (kickit) {
2353
2354		CTR2(KTR_IW_CXGBE, "%s:et1 %p", __func__, ep);
2355		queue_work(c4iw_taskq, &c4iw_task);
2356	}
2357	CTR2(KTR_IW_CXGBE, "%s:etE %p", __func__, ep);
2358}
2359
2360static int fw6_wr_rpl(struct adapter *sc, const __be64 *rpl)
2361{
2362	uint64_t val = be64toh(*rpl);
2363	int ret;
2364	struct c4iw_wr_wait *wr_waitp;
2365
2366	ret = (int)((val >> 8) & 0xff);
2367	wr_waitp = (struct c4iw_wr_wait *)rpl[1];
2368	CTR3(KTR_IW_CXGBE, "%s wr_waitp %p ret %u", __func__, wr_waitp, ret);
2369	if (wr_waitp)
2370		c4iw_wake_up(wr_waitp, ret ? -ret : 0);
2371
2372	return (0);
2373}
2374
2375static int fw6_cqe_handler(struct adapter *sc, const __be64 *rpl)
2376{
2377	struct t4_cqe cqe =*(const struct t4_cqe *)(&rpl[0]);
2378
2379	CTR2(KTR_IW_CXGBE, "%s rpl %p", __func__, rpl);
2380	c4iw_ev_dispatch(sc->iwarp_softc, &cqe);
2381
2382	return (0);
2383}
2384
2385static int terminate(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m)
2386{
2387
2388	struct adapter *sc = iq->adapter;
2389
2390	const struct cpl_rdma_terminate *rpl = (const void *)(rss + 1);
2391	unsigned int tid = GET_TID(rpl);
2392	struct c4iw_qp_attributes attrs;
2393	struct toepcb *toep = lookup_tid(sc, tid);
2394	struct socket *so = inp_inpcbtosocket(toep->inp);
2395	struct c4iw_ep *ep = so->so_rcv.sb_upcallarg;
2396
2397	CTR2(KTR_IW_CXGBE, "%s:tB %p %d", __func__, ep);
2398
2399	if (ep && ep->com.qp) {
2400
2401		printk(KERN_WARNING MOD "TERM received tid %u qpid %u\n", tid,
2402				ep->com.qp->wq.sq.qid);
2403		attrs.next_state = C4IW_QP_STATE_TERMINATE;
2404		c4iw_modify_qp(ep->com.dev, ep->com.qp, C4IW_QP_ATTR_NEXT_STATE, &attrs,
2405				1);
2406	} else
2407		printk(KERN_WARNING MOD "TERM received tid %u no ep/qp\n", tid);
2408	CTR2(KTR_IW_CXGBE, "%s:tE %p %d", __func__, ep);
2409
2410	return 0;
2411}
2412
2413	void
2414c4iw_cm_init_cpl(struct adapter *sc)
2415{
2416
2417	t4_register_cpl_handler(sc, CPL_RDMA_TERMINATE, terminate);
2418	t4_register_fw_msg_handler(sc, FW6_TYPE_WR_RPL, fw6_wr_rpl);
2419	t4_register_fw_msg_handler(sc, FW6_TYPE_CQE, fw6_cqe_handler);
2420	t4_register_an_handler(sc, c4iw_ev_handler);
2421}
2422
2423	void
2424c4iw_cm_term_cpl(struct adapter *sc)
2425{
2426
2427	t4_register_cpl_handler(sc, CPL_RDMA_TERMINATE, NULL);
2428	t4_register_fw_msg_handler(sc, FW6_TYPE_WR_RPL, NULL);
2429	t4_register_fw_msg_handler(sc, FW6_TYPE_CQE, NULL);
2430}
2431
2432int __init c4iw_cm_init(void)
2433{
2434
2435	TAILQ_INIT(&req_list);
2436	spin_lock_init(&req_lock);
2437	INIT_LIST_HEAD(&timeout_list);
2438	spin_lock_init(&timeout_lock);
2439
2440	INIT_WORK(&c4iw_task, process_req);
2441
2442	c4iw_taskq = create_singlethread_workqueue("iw_cxgbe");
2443	if (!c4iw_taskq)
2444		return -ENOMEM;
2445
2446
2447	return 0;
2448}
2449
2450void __exit c4iw_cm_term(void)
2451{
2452	WARN_ON(!TAILQ_EMPTY(&req_list));
2453	WARN_ON(!list_empty(&timeout_list));
2454	flush_workqueue(c4iw_taskq);
2455	destroy_workqueue(c4iw_taskq);
2456}
2457#endif
2458