2 * Copyright (c) 2009-2013 Chelsio, Inc. All rights reserved.
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:
10 * Redistribution and use in source and binary forms, with or
11 * without modification, are permitted provided that the following
14 * - Redistributions of source code must retain the above
15 * copyright notice, this list of conditions and the following
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.
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
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD$");
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>
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>
55 #include <netinet/toecore.h>
59 #include <linux/types.h>
61 #include "tom/t4_tom.h"
63 #define TOEPCB(so) ((struct toepcb *)(so_sototcpcb((so))->t_toe))
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>
73 static spinlock_t req_lock;
74 static TAILQ_HEAD(c4iw_ep_list, c4iw_ep_common) req_list;
75 static struct work_struct c4iw_task;
76 static struct workqueue_struct *c4iw_taskq;
77 static LIST_HEAD(timeout_list);
78 static spinlock_t timeout_lock;
80 static void process_req(struct work_struct *ctx);
81 static void start_ep_timer(struct c4iw_ep *ep);
82 static void stop_ep_timer(struct c4iw_ep *ep);
83 static int set_tcpinfo(struct c4iw_ep *ep);
84 static enum c4iw_ep_state state_read(struct c4iw_ep_common *epc);
85 static void __state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state tostate);
86 static void state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state tostate);
87 static void *alloc_ep(int size, gfp_t flags);
88 void __free_ep(struct c4iw_ep_common *epc);
89 static struct rtentry * find_route(__be32 local_ip, __be32 peer_ip, __be16 local_port,
90 __be16 peer_port, u8 tos);
91 static int close_socket(struct c4iw_ep_common *epc, int close);
92 static int shutdown_socket(struct c4iw_ep_common *epc);
93 static void abort_socket(struct c4iw_ep *ep);
94 static void send_mpa_req(struct c4iw_ep *ep);
95 static int send_mpa_reject(struct c4iw_ep *ep, const void *pdata, u8 plen);
96 static int send_mpa_reply(struct c4iw_ep *ep, const void *pdata, u8 plen);
97 static void close_complete_upcall(struct c4iw_ep *ep, int status);
98 static int abort_connection(struct c4iw_ep *ep);
99 static void peer_close_upcall(struct c4iw_ep *ep);
100 static void peer_abort_upcall(struct c4iw_ep *ep);
101 static void connect_reply_upcall(struct c4iw_ep *ep, int status);
102 static void connect_request_upcall(struct c4iw_ep *ep);
103 static void established_upcall(struct c4iw_ep *ep);
104 static void process_mpa_reply(struct c4iw_ep *ep);
105 static void process_mpa_request(struct c4iw_ep *ep);
106 static void process_peer_close(struct c4iw_ep *ep);
107 static void process_conn_error(struct c4iw_ep *ep);
108 static void process_close_complete(struct c4iw_ep *ep);
109 static void ep_timeout(unsigned long arg);
110 static void init_sock(struct c4iw_ep_common *epc);
111 static void process_data(struct c4iw_ep *ep);
112 static void process_connected(struct c4iw_ep *ep);
113 static struct socket * dequeue_socket(struct socket *head, struct sockaddr_in **remote, struct c4iw_ep *child_ep);
114 static void process_newconn(struct c4iw_ep *parent_ep);
115 static int c4iw_so_upcall(struct socket *so, void *arg, int waitflag);
116 static void process_socket_event(struct c4iw_ep *ep);
117 static void release_ep_resources(struct c4iw_ep *ep);
119 #define START_EP_TIMER(ep) \
121 CTR3(KTR_IW_CXGBE, "start_ep_timer (%s:%d) ep %p", \
122 __func__, __LINE__, (ep)); \
123 start_ep_timer(ep); \
126 #define STOP_EP_TIMER(ep) \
128 CTR3(KTR_IW_CXGBE, "stop_ep_timer (%s:%d) ep %p", \
129 __func__, __LINE__, (ep)); \
134 static char *states[] = {
152 process_req(struct work_struct *ctx)
154 struct c4iw_ep_common *epc;
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);
163 process_socket_event((struct c4iw_ep *)epc);
165 spin_lock(&req_lock);
167 spin_unlock(&req_lock);
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?
176 set_tcpinfo(struct c4iw_ep *ep)
178 struct socket *so = ep->com.so;
179 struct inpcb *inp = sotoinpcb(so);
186 if ((tp->t_flags & TF_TOE) == 0) {
188 log(LOG_ERR, "%s: connection not offloaded (so %p, ep %p)\n",
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);
204 static struct rtentry *
205 find_route(__be32 local_ip, __be32 peer_ip, __be16 local_port,
206 __be16 peer_port, u8 tos)
208 struct route iproute;
209 struct sockaddr_in *dst = (struct sockaddr_in *)&iproute.ro_dst;
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;
219 CTR2(KTR_IW_CXGBE, "%s:frtE %p", __func__, (uint64_t)iproute.ro_rt);
220 return iproute.ro_rt;
224 close_socket(struct c4iw_ep_common *epc, int close)
226 struct socket *so = epc->so;
229 CTR4(KTR_IW_CXGBE, "%s: so %p, ep %p, state %s", __func__, epc, so,
233 soupcall_clear(so, SO_RCV);
239 rc = soshutdown(so, SHUT_WR | SHUT_RD);
246 shutdown_socket(struct c4iw_ep_common *epc)
249 CTR4(KTR_IW_CXGBE, "%s: so %p, ep %p, state %s", __func__, epc->so, epc,
252 return (soshutdown(epc->so, SHUT_WR));
256 abort_socket(struct c4iw_ep *ep)
262 CTR4(KTR_IW_CXGBE, "%s ep %p so %p state %s", __func__, ep, ep->com.so,
263 states[ep->com.state]);
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;
275 rc = sosetopt(ep->com.so, &sopt);
277 log(LOG_ERR, "%s: can't set linger to 0, no RST! err %d\n",
283 process_peer_close(struct c4iw_ep *ep)
285 struct c4iw_qp_attributes attrs;
289 CTR4(KTR_IW_CXGBE, "%s:ppcB ep %p so %p state %s", __func__, ep,
290 ep->com.so, states[ep->com.state]);
292 mutex_lock(&ep->com.mutex);
293 switch (ep->com.state) {
296 CTR2(KTR_IW_CXGBE, "%s:ppc1 %p MPA_REQ_WAIT CLOSING",
298 __state_set(&ep->com, CLOSING);
302 CTR2(KTR_IW_CXGBE, "%s:ppc2 %p MPA_REQ_SENT CLOSING",
304 __state_set(&ep->com, DEAD);
305 connect_reply_upcall(ep, -ECONNABORTED);
309 close_socket(&ep->com, 0);
310 ep->com.cm_id->rem_ref(ep->com.cm_id);
311 ep->com.cm_id = NULL;
319 * We're gonna mark this puppy DEAD, but keep
320 * the reference on it until the ULP accepts or
323 CTR2(KTR_IW_CXGBE, "%s:ppc3 %p MPA_REQ_RCVD CLOSING",
325 __state_set(&ep->com, CLOSING);
326 c4iw_get_ep(&ep->com);
330 CTR2(KTR_IW_CXGBE, "%s:ppc4 %p MPA_REP_SENT CLOSING",
332 __state_set(&ep->com, CLOSING);
336 CTR2(KTR_IW_CXGBE, "%s:ppc5 %p FPDU_MODE CLOSING",
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);
347 CTR2(KTR_IW_CXGBE, "%s:ppc6 %p ABORTING (disconn)",
353 CTR2(KTR_IW_CXGBE, "%s:ppc7 %p CLOSING MORIBUND",
355 __state_set(&ep->com, MORIBUND);
360 CTR2(KTR_IW_CXGBE, "%s:ppc8 %p MORIBUND DEAD", __func__,
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);
368 close_socket(&ep->com, 0);
369 close_complete_upcall(ep, 0);
370 __state_set(&ep->com, DEAD);
376 CTR2(KTR_IW_CXGBE, "%s:ppc9 %p DEAD (disconn)",
382 panic("%s: ep %p state %d", __func__, ep,
387 mutex_unlock(&ep->com.mutex);
391 CTR2(KTR_IW_CXGBE, "%s:ppca %p", __func__, ep);
392 c4iw_ep_disconnect(ep, 0, M_NOWAIT);
396 CTR2(KTR_IW_CXGBE, "%s:ppcb %p", __func__, ep);
397 c4iw_put_ep(&ep->com);
399 CTR2(KTR_IW_CXGBE, "%s:ppcE %p", __func__, ep);
404 process_conn_error(struct c4iw_ep *ep)
406 struct c4iw_qp_attributes attrs;
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]);
423 connect_reply_upcall(ep, -ECONNRESET);
427 ep->com.rpl_err = ECONNRESET;
428 CTR1(KTR_IW_CXGBE, "waking up ep %p", ep);
434 * We're gonna mark this puppy DEAD, but keep
435 * the reference on it until the ULP accepts or
438 c4iw_get_ep(&ep->com);
447 if (ep->com.cm_id && ep->com.qp) {
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,
455 "%s - qp <- error failed!\n",
458 peer_abort_upcall(ep);
465 CTR2(KTR_IW_CXGBE, "%s so_error %d IN DEAD STATE!!!!",
466 __func__, ep->com.so->so_error);
470 panic("%s: ep %p state %d", __func__, ep, state);
474 if (state != ABORTING) {
476 CTR2(KTR_IW_CXGBE, "%s:pce1 %p", __func__, ep);
477 close_socket(&ep->com, 1);
478 state_set(&ep->com, DEAD);
479 c4iw_put_ep(&ep->com);
481 CTR2(KTR_IW_CXGBE, "%s:pceE %p", __func__, ep);
486 process_close_complete(struct c4iw_ep *ep)
488 struct c4iw_qp_attributes attrs;
491 CTR4(KTR_IW_CXGBE, "%s:pccB ep %p so %p state %s", __func__, ep,
492 ep->com.so, states[ep->com.state]);
494 /* The cm_id may be null if we failed to connect */
495 mutex_lock(&ep->com.mutex);
497 switch (ep->com.state) {
500 CTR2(KTR_IW_CXGBE, "%s:pcc1 %p CLOSING MORIBUND",
502 __state_set(&ep->com, MORIBUND);
506 CTR2(KTR_IW_CXGBE, "%s:pcc1 %p MORIBUND DEAD", __func__,
510 if ((ep->com.cm_id) && (ep->com.qp)) {
512 CTR2(KTR_IW_CXGBE, "%s:pcc2 %p QP_STATE_IDLE",
514 attrs.next_state = C4IW_QP_STATE_IDLE;
515 c4iw_modify_qp(ep->com.dev,
517 C4IW_QP_ATTR_NEXT_STATE,
523 CTR2(KTR_IW_CXGBE, "%s:pcc3 %p", __func__, ep);
524 close_socket(&ep->com, 1);
528 CTR2(KTR_IW_CXGBE, "%s:pcc4 %p", __func__, ep);
529 close_socket(&ep->com, 0);
531 close_complete_upcall(ep, 0);
532 __state_set(&ep->com, DEAD);
537 CTR2(KTR_IW_CXGBE, "%s:pcc5 %p ABORTING", __func__, ep);
542 CTR2(KTR_IW_CXGBE, "%s:pcc6 %p DEAD", __func__, ep);
543 panic("%s:pcc6 %p DEAD", __func__, ep);
546 mutex_unlock(&ep->com.mutex);
550 CTR2(KTR_IW_CXGBE, "%s:pcc7 %p", __func__, ep);
551 c4iw_put_ep(&ep->com);
553 CTR2(KTR_IW_CXGBE, "%s:pccE %p", __func__, ep);
558 init_sock(struct c4iw_ep_common *epc)
562 struct socket *so = epc->so;
566 soupcall_set(so, SO_RCV, c4iw_so_upcall, epc);
567 so->so_state |= SS_NBIO;
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;
575 rc = sosetopt(so, &sopt);
577 log(LOG_ERR, "%s: can't set TCP_NODELAY on so %p (%d)\n",
583 process_data(struct c4iw_ep *ep)
585 struct sockaddr_in *local, *remote;
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);
590 switch (state_read(&ep->com)) {
592 process_mpa_reply(ep);
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);
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);
614 process_connected(struct c4iw_ep *ep)
617 if ((ep->com.so->so_state & SS_ISCONNECTED) && !ep->com.so->so_error)
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);
627 static struct socket *
628 dequeue_socket(struct socket *head, struct sockaddr_in **remote,
629 struct c4iw_ep *child_ep)
634 so = TAILQ_FIRST(&head->so_comp);
639 TAILQ_REMOVE(&head->so_comp, so, so_list);
642 so->so_qstate &= ~SQ_COMP;
645 soupcall_set(so, SO_RCV, c4iw_so_upcall, child_ep);
646 so->so_state |= SS_NBIO;
649 soaccept(so, (struct sockaddr **)remote);
655 process_newconn(struct c4iw_ep *parent_ep)
657 struct socket *child_so;
658 struct c4iw_ep *child_ep;
659 struct sockaddr_in *remote;
661 child_ep = alloc_ep(sizeof(*child_ep), M_NOWAIT);
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__);
669 child_so = dequeue_socket(parent_ep->com.so, &remote, child_ep);
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);
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);
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;
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);
699 /* maybe the request has already been queued up on the socket... */
700 process_mpa_request(child_ep);
704 c4iw_so_upcall(struct socket *so, void *arg, int waitflag)
706 struct c4iw_ep *ep = arg;
708 spin_lock(&req_lock);
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);
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);
722 spin_unlock(&req_lock);
727 process_socket_event(struct c4iw_ep *ep)
729 int state = state_read(&ep->com);
730 struct socket *so = ep->com.so;
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]);
736 if (state == CONNECTING) {
737 process_connected(ep);
741 if (state == LISTEN) {
746 /* connection error */
748 process_conn_error(ep);
753 if ((so->so_rcv.sb_state & SBS_CANTRCVMORE) && state < CLOSING) {
754 process_peer_close(ep);
759 if (so->so_state & SS_ISDISCONNECTED) {
760 process_close_complete(ep);
768 SYSCTL_NODE(_hw, OID_AUTO, iw_cxgbe, CTLFLAG_RD, 0, "iw_cxgbe driver parameters");
770 int db_delay_usecs = 1;
771 TUNABLE_INT("hw.iw_cxgbe.db_delay_usecs", &db_delay_usecs);
772 SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, db_delay_usecs, CTLFLAG_RW, &db_delay_usecs, 0,
773 "Usecs to delay awaiting db fifo to drain");
775 static int dack_mode = 1;
776 TUNABLE_INT("hw.iw_cxgbe.dack_mode", &dack_mode);
777 SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, dack_mode, CTLFLAG_RW, &dack_mode, 0,
778 "Delayed ack mode (default = 1)");
780 int c4iw_max_read_depth = 8;
781 TUNABLE_INT("hw.iw_cxgbe.c4iw_max_read_depth", &c4iw_max_read_depth);
782 SYSCTL_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)");
785 static int enable_tcp_timestamps;
786 TUNABLE_INT("hw.iw_cxgbe.enable_tcp_timestamps", &enable_tcp_timestamps);
787 SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, enable_tcp_timestamps, CTLFLAG_RW, &enable_tcp_timestamps, 0,
788 "Enable tcp timestamps (default = 0)");
790 static int enable_tcp_sack;
791 TUNABLE_INT("hw.iw_cxgbe.enable_tcp_sack", &enable_tcp_sack);
792 SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, enable_tcp_sack, CTLFLAG_RW, &enable_tcp_sack, 0,
793 "Enable tcp SACK (default = 0)");
795 static int enable_tcp_window_scaling = 1;
796 TUNABLE_INT("hw.iw_cxgbe.enable_tcp_window_scaling", &enable_tcp_window_scaling);
797 SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, enable_tcp_window_scaling, CTLFLAG_RW, &enable_tcp_window_scaling, 0,
798 "Enable tcp window scaling (default = 1)");
801 TUNABLE_INT("hw.iw_cxgbe.c4iw_debug", &c4iw_debug);
802 SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, c4iw_debug, CTLFLAG_RW, &c4iw_debug, 0,
803 "Enable debug logging (default = 0)");
805 static int peer2peer;
806 TUNABLE_INT("hw.iw_cxgbe.peer2peer", &peer2peer);
807 SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, peer2peer, CTLFLAG_RW, &peer2peer, 0,
808 "Support peer2peer ULPs (default = 0)");
810 static int p2p_type = FW_RI_INIT_P2PTYPE_READ_REQ;
811 TUNABLE_INT("hw.iw_cxgbe.p2p_type", &p2p_type);
812 SYSCTL_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)");
815 static int ep_timeout_secs = 60;
816 TUNABLE_INT("hw.iw_cxgbe.ep_timeout_secs", &ep_timeout_secs);
817 SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, ep_timeout_secs, CTLFLAG_RW, &ep_timeout_secs, 0,
818 "CM Endpoint operation timeout in seconds (default = 60)");
820 static int mpa_rev = 1;
821 TUNABLE_INT("hw.iw_cxgbe.mpa_rev", &mpa_rev);
823 SYSCTL_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)");
826 SYSCTL_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)");
830 static int markers_enabled;
831 TUNABLE_INT("hw.iw_cxgbe.markers_enabled", &markers_enabled);
832 SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, markers_enabled, CTLFLAG_RW, &markers_enabled, 0,
833 "Enable MPA MARKERS (default(0) = disabled)");
835 static int crc_enabled = 1;
836 TUNABLE_INT("hw.iw_cxgbe.crc_enabled", &crc_enabled);
837 SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, crc_enabled, CTLFLAG_RW, &crc_enabled, 0,
838 "Enable MPA CRC (default(1) = enabled)");
840 static int rcv_win = 256 * 1024;
841 TUNABLE_INT("hw.iw_cxgbe.rcv_win", &rcv_win);
842 SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, rcv_win, CTLFLAG_RW, &rcv_win, 0,
843 "TCP receive window in bytes (default = 256KB)");
845 static int snd_win = 128 * 1024;
846 TUNABLE_INT("hw.iw_cxgbe.snd_win", &snd_win);
847 SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, snd_win, CTLFLAG_RW, &snd_win, 0,
848 "TCP send window in bytes (default = 128KB)");
850 int db_fc_threshold = 2000;
851 TUNABLE_INT("hw.iw_cxgbe.db_fc_threshold", &db_fc_threshold);
852 SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, db_fc_threshold, CTLFLAG_RW, &db_fc_threshold, 0,
853 "QP count/threshold that triggers automatic");
856 start_ep_timer(struct c4iw_ep *ep)
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__,
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);
874 stop_ep_timer(struct c4iw_ep *ep)
877 del_timer_sync(&ep->timer);
878 if (!test_and_set_bit(TIMEOUT, &ep->com.flags)) {
879 c4iw_put_ep(&ep->com);
884 c4iw_ep_state state_read(struct c4iw_ep_common *epc)
886 enum c4iw_ep_state state;
888 mutex_lock(&epc->mutex);
890 mutex_unlock(&epc->mutex);
896 __state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state new)
903 state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state new)
906 mutex_lock(&epc->mutex);
907 __state_set(epc, new);
908 mutex_unlock(&epc->mutex);
912 alloc_ep(int size, gfp_t gfp)
914 struct c4iw_ep_common *epc;
916 epc = kzalloc(size, gfp);
920 kref_init(&epc->kref);
921 mutex_init(&epc->mutex);
922 c4iw_init_wr_wait(&epc->wr_wait);
928 __free_ep(struct c4iw_ep_common *epc)
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));
934 CTR2(KTR_IW_CXGBE, "%s:feE %p", __func__, epc);
937 void _c4iw_free_ep(struct kref *kref)
940 struct c4iw_ep_common *epc;
942 ep = container_of(kref, struct c4iw_ep, com.kref);
944 KASSERT(!epc->so, ("%s ep->so %p", __func__, epc->so));
945 KASSERT(!epc->entry.tqe_prev, ("%s epc %p still on req list",
950 static void release_ep_resources(struct c4iw_ep *ep)
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);
959 send_mpa_req(struct c4iw_ep *ep)
962 struct mpa_message *mpa;
963 struct mpa_v2_conn_params mpa_v2_params;
965 char mpa_rev_to_use = mpa_rev;
968 if (ep->retry_with_mpa_v1)
970 mpalen = sizeof(*mpa) + ep->plen;
971 if (mpa_rev_to_use == 2)
972 mpalen += sizeof(struct mpa_v2_conn_params);
974 mpa = malloc(mpalen, M_CXGBE, M_NOWAIT);
977 connect_reply_upcall(ep, -ENOMEM);
981 memset(mpa, 0, mpalen);
982 memcpy(mpa->key, MPA_KEY_REQ, sizeof(mpa->key));
983 mpa->flags = (crc_enabled ? MPA_CRC : 0) |
984 (markers_enabled ? MPA_MARKERS : 0) |
985 (mpa_rev_to_use == 2 ? MPA_ENHANCED_RDMA_CONN : 0);
986 mpa->private_data_size = htons(ep->plen);
987 mpa->revision = mpa_rev_to_use;
989 if (mpa_rev_to_use == 1) {
990 ep->tried_with_mpa_v1 = 1;
991 ep->retry_with_mpa_v1 = 0;
994 if (mpa_rev_to_use == 2) {
995 mpa->private_data_size +=
996 htons(sizeof(struct mpa_v2_conn_params));
997 mpa_v2_params.ird = htons((u16)ep->ird);
998 mpa_v2_params.ord = htons((u16)ep->ord);
1001 mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL);
1003 if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE) {
1004 mpa_v2_params.ord |=
1005 htons(MPA_V2_RDMA_WRITE_RTR);
1006 } else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ) {
1007 mpa_v2_params.ord |=
1008 htons(MPA_V2_RDMA_READ_RTR);
1011 memcpy(mpa->private_data, &mpa_v2_params,
1012 sizeof(struct mpa_v2_conn_params));
1016 memcpy(mpa->private_data +
1017 sizeof(struct mpa_v2_conn_params),
1018 ep->mpa_pkt + sizeof(*mpa), ep->plen);
1023 memcpy(mpa->private_data,
1024 ep->mpa_pkt + sizeof(*mpa), ep->plen);
1025 CTR2(KTR_IW_CXGBE, "%s:smr7 %p", __func__, ep);
1028 m = m_getm(NULL, mpalen, M_NOWAIT, MT_DATA);
1033 m_copyback(m, 0, mpalen, (void *)mpa);
1036 err = sosend(ep->com.so, NULL, NULL, m, NULL, MSG_DONTWAIT,
1042 state_set(&ep->com, MPA_REQ_SENT);
1043 ep->mpa_attr.initiator = 1;
1046 static int send_mpa_reject(struct c4iw_ep *ep, const void *pdata, u8 plen)
1049 struct mpa_message *mpa;
1050 struct mpa_v2_conn_params mpa_v2_params;
1054 CTR4(KTR_IW_CXGBE, "%s:smrejB %p %u %d", __func__, ep, ep->hwtid,
1057 mpalen = sizeof(*mpa) + plen;
1059 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
1061 mpalen += sizeof(struct mpa_v2_conn_params);
1062 CTR4(KTR_IW_CXGBE, "%s:smrej1 %p %u %d", __func__, ep,
1063 ep->mpa_attr.version, mpalen);
1066 mpa = malloc(mpalen, M_CXGBE, M_NOWAIT);
1070 memset(mpa, 0, mpalen);
1071 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
1072 mpa->flags = MPA_REJECT;
1073 mpa->revision = mpa_rev;
1074 mpa->private_data_size = htons(plen);
1076 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
1078 mpa->flags |= MPA_ENHANCED_RDMA_CONN;
1079 mpa->private_data_size +=
1080 htons(sizeof(struct mpa_v2_conn_params));
1081 mpa_v2_params.ird = htons(((u16)ep->ird) |
1082 (peer2peer ? MPA_V2_PEER2PEER_MODEL :
1084 mpa_v2_params.ord = htons(((u16)ep->ord) | (peer2peer ?
1086 FW_RI_INIT_P2PTYPE_RDMA_WRITE ?
1087 MPA_V2_RDMA_WRITE_RTR : p2p_type ==
1088 FW_RI_INIT_P2PTYPE_READ_REQ ?
1089 MPA_V2_RDMA_READ_RTR : 0) : 0));
1090 memcpy(mpa->private_data, &mpa_v2_params,
1091 sizeof(struct mpa_v2_conn_params));
1094 memcpy(mpa->private_data +
1095 sizeof(struct mpa_v2_conn_params), pdata, plen);
1096 CTR5(KTR_IW_CXGBE, "%s:smrej3 %p %d %d %d", __func__, ep,
1097 mpa_v2_params.ird, mpa_v2_params.ord, ep->plen);
1100 memcpy(mpa->private_data, pdata, plen);
1102 m = m_getm(NULL, mpalen, M_NOWAIT, MT_DATA);
1107 m_copyback(m, 0, mpalen, (void *)mpa);
1110 err = -sosend(ep->com.so, NULL, NULL, m, NULL, MSG_DONTWAIT, ep->com.thread);
1112 ep->snd_seq += mpalen;
1113 CTR4(KTR_IW_CXGBE, "%s:smrejE %p %u %d", __func__, ep, ep->hwtid, err);
1117 static int send_mpa_reply(struct c4iw_ep *ep, const void *pdata, u8 plen)
1120 struct mpa_message *mpa;
1122 struct mpa_v2_conn_params mpa_v2_params;
1125 CTR2(KTR_IW_CXGBE, "%s:smrepB %p", __func__, ep);
1127 mpalen = sizeof(*mpa) + plen;
1129 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
1131 CTR3(KTR_IW_CXGBE, "%s:smrep1 %p %d", __func__, ep,
1132 ep->mpa_attr.version);
1133 mpalen += sizeof(struct mpa_v2_conn_params);
1136 mpa = malloc(mpalen, M_CXGBE, M_NOWAIT);
1140 memset(mpa, 0, sizeof(*mpa));
1141 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
1142 mpa->flags = (ep->mpa_attr.crc_enabled ? MPA_CRC : 0) |
1143 (markers_enabled ? MPA_MARKERS : 0);
1144 mpa->revision = ep->mpa_attr.version;
1145 mpa->private_data_size = htons(plen);
1147 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
1149 mpa->flags |= MPA_ENHANCED_RDMA_CONN;
1150 mpa->private_data_size +=
1151 htons(sizeof(struct mpa_v2_conn_params));
1152 mpa_v2_params.ird = htons((u16)ep->ird);
1153 mpa_v2_params.ord = htons((u16)ep->ord);
1154 CTR5(KTR_IW_CXGBE, "%s:smrep3 %p %d %d %d", __func__, ep,
1155 ep->mpa_attr.version, mpa_v2_params.ird, mpa_v2_params.ord);
1157 if (peer2peer && (ep->mpa_attr.p2p_type !=
1158 FW_RI_INIT_P2PTYPE_DISABLED)) {
1160 mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL);
1162 if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE) {
1164 mpa_v2_params.ord |=
1165 htons(MPA_V2_RDMA_WRITE_RTR);
1166 CTR5(KTR_IW_CXGBE, "%s:smrep4 %p %d %d %d",
1167 __func__, ep, p2p_type, mpa_v2_params.ird,
1170 else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ) {
1172 mpa_v2_params.ord |=
1173 htons(MPA_V2_RDMA_READ_RTR);
1174 CTR5(KTR_IW_CXGBE, "%s:smrep5 %p %d %d %d",
1175 __func__, ep, p2p_type, mpa_v2_params.ird,
1180 memcpy(mpa->private_data, &mpa_v2_params,
1181 sizeof(struct mpa_v2_conn_params));
1184 memcpy(mpa->private_data +
1185 sizeof(struct mpa_v2_conn_params), pdata, plen);
1188 memcpy(mpa->private_data, pdata, plen);
1190 m = m_getm(NULL, mpalen, M_NOWAIT, MT_DATA);
1195 m_copyback(m, 0, mpalen, (void *)mpa);
1199 state_set(&ep->com, MPA_REP_SENT);
1200 ep->snd_seq += mpalen;
1201 err = -sosend(ep->com.so, NULL, NULL, m, NULL, MSG_DONTWAIT,
1203 CTR3(KTR_IW_CXGBE, "%s:smrepE %p %d", __func__, ep, err);
1209 static void close_complete_upcall(struct c4iw_ep *ep, int status)
1211 struct iw_cm_event event;
1213 CTR2(KTR_IW_CXGBE, "%s:ccuB %p", __func__, ep);
1214 memset(&event, 0, sizeof(event));
1215 event.event = IW_CM_EVENT_CLOSE;
1216 event.status = status;
1218 if (ep->com.cm_id) {
1220 CTR2(KTR_IW_CXGBE, "%s:ccu1 %1", __func__, ep);
1221 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1222 ep->com.cm_id->rem_ref(ep->com.cm_id);
1223 ep->com.cm_id = NULL;
1225 set_bit(CLOSE_UPCALL, &ep->com.history);
1227 CTR2(KTR_IW_CXGBE, "%s:ccuE %p", __func__, ep);
1230 static int abort_connection(struct c4iw_ep *ep)
1234 CTR2(KTR_IW_CXGBE, "%s:abB %p", __func__, ep);
1235 close_complete_upcall(ep, -ECONNRESET);
1236 state_set(&ep->com, ABORTING);
1238 err = close_socket(&ep->com, 0);
1239 set_bit(ABORT_CONN, &ep->com.history);
1240 CTR2(KTR_IW_CXGBE, "%s:abE %p", __func__, ep);
1244 static void peer_close_upcall(struct c4iw_ep *ep)
1246 struct iw_cm_event event;
1248 CTR2(KTR_IW_CXGBE, "%s:pcuB %p", __func__, ep);
1249 memset(&event, 0, sizeof(event));
1250 event.event = IW_CM_EVENT_DISCONNECT;
1252 if (ep->com.cm_id) {
1254 CTR2(KTR_IW_CXGBE, "%s:pcu1 %p", __func__, ep);
1255 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1256 set_bit(DISCONN_UPCALL, &ep->com.history);
1258 CTR2(KTR_IW_CXGBE, "%s:pcuE %p", __func__, ep);
1261 static void peer_abort_upcall(struct c4iw_ep *ep)
1263 struct iw_cm_event event;
1265 CTR2(KTR_IW_CXGBE, "%s:pauB %p", __func__, ep);
1266 memset(&event, 0, sizeof(event));
1267 event.event = IW_CM_EVENT_CLOSE;
1268 event.status = -ECONNRESET;
1270 if (ep->com.cm_id) {
1272 CTR2(KTR_IW_CXGBE, "%s:pau1 %p", __func__, ep);
1273 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1274 ep->com.cm_id->rem_ref(ep->com.cm_id);
1275 ep->com.cm_id = NULL;
1277 set_bit(ABORT_UPCALL, &ep->com.history);
1279 CTR2(KTR_IW_CXGBE, "%s:pauE %p", __func__, ep);
1282 static void connect_reply_upcall(struct c4iw_ep *ep, int status)
1284 struct iw_cm_event event;
1286 CTR3(KTR_IW_CXGBE, "%s:cruB %p", __func__, ep, status);
1287 memset(&event, 0, sizeof(event));
1288 event.event = IW_CM_EVENT_CONNECT_REPLY;
1289 event.status = (status ==-ECONNABORTED)?-ECONNRESET: status;
1290 event.local_addr = ep->com.local_addr;
1291 event.remote_addr = ep->com.remote_addr;
1293 if ((status == 0) || (status == -ECONNREFUSED)) {
1295 if (!ep->tried_with_mpa_v1) {
1297 CTR2(KTR_IW_CXGBE, "%s:cru1 %p", __func__, ep);
1298 /* this means MPA_v2 is used */
1299 event.private_data_len = ep->plen -
1300 sizeof(struct mpa_v2_conn_params);
1301 event.private_data = ep->mpa_pkt +
1302 sizeof(struct mpa_message) +
1303 sizeof(struct mpa_v2_conn_params);
1306 CTR2(KTR_IW_CXGBE, "%s:cru2 %p", __func__, ep);
1307 /* this means MPA_v1 is used */
1308 event.private_data_len = ep->plen;
1309 event.private_data = ep->mpa_pkt +
1310 sizeof(struct mpa_message);
1314 if (ep->com.cm_id) {
1316 CTR2(KTR_IW_CXGBE, "%s:cru3 %p", __func__, ep);
1317 set_bit(CONN_RPL_UPCALL, &ep->com.history);
1318 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1321 if(status == -ECONNABORTED) {
1323 CTR3(KTR_IW_CXGBE, "%s:cruE %p %d", __func__, ep, status);
1329 CTR3(KTR_IW_CXGBE, "%s:cru4 %p %d", __func__, ep, status);
1330 ep->com.cm_id->rem_ref(ep->com.cm_id);
1331 ep->com.cm_id = NULL;
1335 CTR2(KTR_IW_CXGBE, "%s:cruE %p", __func__, ep);
1338 static void connect_request_upcall(struct c4iw_ep *ep)
1340 struct iw_cm_event event;
1342 CTR3(KTR_IW_CXGBE, "%s: ep %p, mpa_v1 %d", __func__, ep,
1343 ep->tried_with_mpa_v1);
1345 memset(&event, 0, sizeof(event));
1346 event.event = IW_CM_EVENT_CONNECT_REQUEST;
1347 event.local_addr = ep->com.local_addr;
1348 event.remote_addr = ep->com.remote_addr;
1349 event.provider_data = ep;
1350 event.so = ep->com.so;
1352 if (!ep->tried_with_mpa_v1) {
1353 /* this means MPA_v2 is used */
1355 event.ord = ep->ord;
1356 event.ird = ep->ird;
1358 event.private_data_len = ep->plen -
1359 sizeof(struct mpa_v2_conn_params);
1360 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message) +
1361 sizeof(struct mpa_v2_conn_params);
1364 /* this means MPA_v1 is used. Send max supported */
1366 event.ord = c4iw_max_read_depth;
1367 event.ird = c4iw_max_read_depth;
1369 event.private_data_len = ep->plen;
1370 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message);
1373 c4iw_get_ep(&ep->com);
1374 ep->parent_ep->com.cm_id->event_handler(ep->parent_ep->com.cm_id,
1376 set_bit(CONNREQ_UPCALL, &ep->com.history);
1377 c4iw_put_ep(&ep->parent_ep->com);
1380 static void established_upcall(struct c4iw_ep *ep)
1382 struct iw_cm_event event;
1384 CTR2(KTR_IW_CXGBE, "%s:euB %p", __func__, ep);
1385 memset(&event, 0, sizeof(event));
1386 event.event = IW_CM_EVENT_ESTABLISHED;
1388 event.ird = ep->ird;
1389 event.ord = ep->ord;
1391 if (ep->com.cm_id) {
1393 CTR2(KTR_IW_CXGBE, "%s:eu1 %p", __func__, ep);
1394 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1395 set_bit(ESTAB_UPCALL, &ep->com.history);
1397 CTR2(KTR_IW_CXGBE, "%s:euE %p", __func__, ep);
1402 static void process_mpa_reply(struct c4iw_ep *ep)
1404 struct mpa_message *mpa;
1405 struct mpa_v2_conn_params *mpa_v2_params;
1407 u16 resp_ird, resp_ord;
1408 u8 rtr_mismatch = 0, insuff_ird = 0;
1409 struct c4iw_qp_attributes attrs;
1410 enum c4iw_qp_attr_mask mask;
1412 struct mbuf *top, *m;
1413 int flags = MSG_DONTWAIT;
1416 CTR2(KTR_IW_CXGBE, "%s:pmrB %p", __func__, ep);
1419 * Stop mpa timer. If it expired, then the state has
1420 * changed and we bail since ep_timeout already aborted
1424 if (state_read(&ep->com) != MPA_REQ_SENT)
1427 uio.uio_resid = 1000000;
1428 uio.uio_td = ep->com.thread;
1429 err = soreceive(ep->com.so, NULL, &uio, &top, NULL, &flags);
1433 if (err == EWOULDBLOCK) {
1435 CTR2(KTR_IW_CXGBE, "%s:pmr1 %p", __func__, ep);
1440 CTR2(KTR_IW_CXGBE, "%s:pmr2 %p", __func__, ep);
1444 if (ep->com.so->so_rcv.sb_mb) {
1446 CTR2(KTR_IW_CXGBE, "%s:pmr3 %p", __func__, ep);
1447 printf("%s data after soreceive called! so %p sb_mb %p top %p\n",
1448 __func__, ep->com.so, ep->com.so->so_rcv.sb_mb, top);
1455 CTR2(KTR_IW_CXGBE, "%s:pmr4 %p", __func__, ep);
1457 * If we get more than the supported amount of private data
1458 * then we must fail this connection.
1460 if (ep->mpa_pkt_len + m->m_len > sizeof(ep->mpa_pkt)) {
1462 CTR3(KTR_IW_CXGBE, "%s:pmr5 %p %d", __func__, ep,
1463 ep->mpa_pkt_len + m->m_len);
1469 * copy the new data into our accumulation buffer.
1471 m_copydata(m, 0, m->m_len, &(ep->mpa_pkt[ep->mpa_pkt_len]));
1472 ep->mpa_pkt_len += m->m_len;
1481 * if we don't even have the mpa message, then bail.
1483 if (ep->mpa_pkt_len < sizeof(*mpa))
1485 mpa = (struct mpa_message *) ep->mpa_pkt;
1487 /* Validate MPA header. */
1488 if (mpa->revision > mpa_rev) {
1490 CTR4(KTR_IW_CXGBE, "%s:pmr6 %p %d %d", __func__, ep,
1491 mpa->revision, mpa_rev);
1492 printk(KERN_ERR MOD "%s MPA version mismatch. Local = %d, "
1493 " Received = %d\n", __func__, mpa_rev, mpa->revision);
1498 if (memcmp(mpa->key, MPA_KEY_REP, sizeof(mpa->key))) {
1500 CTR2(KTR_IW_CXGBE, "%s:pmr7 %p", __func__, ep);
1505 plen = ntohs(mpa->private_data_size);
1508 * Fail if there's too much private data.
1510 if (plen > MPA_MAX_PRIVATE_DATA) {
1512 CTR2(KTR_IW_CXGBE, "%s:pmr8 %p", __func__, ep);
1518 * If plen does not account for pkt size
1520 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
1522 CTR2(KTR_IW_CXGBE, "%s:pmr9 %p", __func__, ep);
1527 ep->plen = (u8) plen;
1530 * If we don't have all the pdata yet, then bail.
1531 * We'll continue process when more data arrives.
1533 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen)) {
1535 CTR2(KTR_IW_CXGBE, "%s:pmra %p", __func__, ep);
1539 if (mpa->flags & MPA_REJECT) {
1541 CTR2(KTR_IW_CXGBE, "%s:pmrb %p", __func__, ep);
1542 err = -ECONNREFUSED;
1547 * If we get here we have accumulated the entire mpa
1548 * start reply message including private data. And
1549 * the MPA header is valid.
1551 state_set(&ep->com, FPDU_MODE);
1552 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1553 ep->mpa_attr.recv_marker_enabled = markers_enabled;
1554 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1555 ep->mpa_attr.version = mpa->revision;
1556 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1558 if (mpa->revision == 2) {
1560 CTR2(KTR_IW_CXGBE, "%s:pmrc %p", __func__, ep);
1561 ep->mpa_attr.enhanced_rdma_conn =
1562 mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
1564 if (ep->mpa_attr.enhanced_rdma_conn) {
1566 CTR2(KTR_IW_CXGBE, "%s:pmrd %p", __func__, ep);
1567 mpa_v2_params = (struct mpa_v2_conn_params *)
1568 (ep->mpa_pkt + sizeof(*mpa));
1569 resp_ird = ntohs(mpa_v2_params->ird) &
1570 MPA_V2_IRD_ORD_MASK;
1571 resp_ord = ntohs(mpa_v2_params->ord) &
1572 MPA_V2_IRD_ORD_MASK;
1575 * This is a double-check. Ideally, below checks are
1576 * not required since ird/ord stuff has been taken
1577 * care of in c4iw_accept_cr
1579 if ((ep->ird < resp_ord) || (ep->ord > resp_ird)) {
1581 CTR2(KTR_IW_CXGBE, "%s:pmre %p", __func__, ep);
1588 if (ntohs(mpa_v2_params->ird) &
1589 MPA_V2_PEER2PEER_MODEL) {
1591 CTR2(KTR_IW_CXGBE, "%s:pmrf %p", __func__, ep);
1592 if (ntohs(mpa_v2_params->ord) &
1593 MPA_V2_RDMA_WRITE_RTR) {
1595 CTR2(KTR_IW_CXGBE, "%s:pmrg %p", __func__, ep);
1596 ep->mpa_attr.p2p_type =
1597 FW_RI_INIT_P2PTYPE_RDMA_WRITE;
1599 else if (ntohs(mpa_v2_params->ord) &
1600 MPA_V2_RDMA_READ_RTR) {
1602 CTR2(KTR_IW_CXGBE, "%s:pmrh %p", __func__, ep);
1603 ep->mpa_attr.p2p_type =
1604 FW_RI_INIT_P2PTYPE_READ_REQ;
1610 CTR2(KTR_IW_CXGBE, "%s:pmri %p", __func__, ep);
1612 if (mpa->revision == 1) {
1614 CTR2(KTR_IW_CXGBE, "%s:pmrj %p", __func__, ep);
1618 CTR2(KTR_IW_CXGBE, "%s:pmrk %p", __func__, ep);
1619 ep->mpa_attr.p2p_type = p2p_type;
1624 if (set_tcpinfo(ep)) {
1626 CTR2(KTR_IW_CXGBE, "%s:pmrl %p", __func__, ep);
1627 printf("%s set_tcpinfo error\n", __func__);
1631 CTR6(KTR_IW_CXGBE, "%s - crc_enabled = %d, recv_marker_enabled = %d, "
1632 "xmit_marker_enabled = %d, version = %d p2p_type = %d", __func__,
1633 ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
1634 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version,
1635 ep->mpa_attr.p2p_type);
1638 * If responder's RTR does not match with that of initiator, assign
1639 * FW_RI_INIT_P2PTYPE_DISABLED in mpa attributes so that RTR is not
1640 * generated when moving QP to RTS state.
1641 * A TERM message will be sent after QP has moved to RTS state
1643 if ((ep->mpa_attr.version == 2) && peer2peer &&
1644 (ep->mpa_attr.p2p_type != p2p_type)) {
1646 CTR2(KTR_IW_CXGBE, "%s:pmrm %p", __func__, ep);
1647 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1652 //ep->ofld_txq = TOEPCB(ep->com.so)->ofld_txq;
1653 attrs.mpa_attr = ep->mpa_attr;
1654 attrs.max_ird = ep->ird;
1655 attrs.max_ord = ep->ord;
1656 attrs.llp_stream_handle = ep;
1657 attrs.next_state = C4IW_QP_STATE_RTS;
1659 mask = C4IW_QP_ATTR_NEXT_STATE |
1660 C4IW_QP_ATTR_LLP_STREAM_HANDLE | C4IW_QP_ATTR_MPA_ATTR |
1661 C4IW_QP_ATTR_MAX_IRD | C4IW_QP_ATTR_MAX_ORD;
1663 /* bind QP and TID with INIT_WR */
1664 err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp, mask, &attrs, 1);
1668 CTR2(KTR_IW_CXGBE, "%s:pmrn %p", __func__, ep);
1673 * If responder's RTR requirement did not match with what initiator
1674 * supports, generate TERM message
1678 CTR2(KTR_IW_CXGBE, "%s:pmro %p", __func__, ep);
1679 printk(KERN_ERR "%s: RTR mismatch, sending TERM\n", __func__);
1680 attrs.layer_etype = LAYER_MPA | DDP_LLP;
1681 attrs.ecode = MPA_NOMATCH_RTR;
1682 attrs.next_state = C4IW_QP_STATE_TERMINATE;
1683 err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1684 C4IW_QP_ATTR_NEXT_STATE, &attrs, 0);
1690 * Generate TERM if initiator IRD is not sufficient for responder
1691 * provided ORD. Currently, we do the same behaviour even when
1692 * responder provided IRD is also not sufficient as regards to
1697 CTR2(KTR_IW_CXGBE, "%s:pmrp %p", __func__, ep);
1698 printk(KERN_ERR "%s: Insufficient IRD, sending TERM\n",
1700 attrs.layer_etype = LAYER_MPA | DDP_LLP;
1701 attrs.ecode = MPA_INSUFF_IRD;
1702 attrs.next_state = C4IW_QP_STATE_TERMINATE;
1703 err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1704 C4IW_QP_ATTR_NEXT_STATE, &attrs, 0);
1710 state_set(&ep->com, ABORTING);
1711 abort_connection(ep);
1713 connect_reply_upcall(ep, err);
1714 CTR2(KTR_IW_CXGBE, "%s:pmrE %p", __func__, ep);
1719 process_mpa_request(struct c4iw_ep *ep)
1721 struct mpa_message *mpa;
1723 int flags = MSG_DONTWAIT;
1727 enum c4iw_ep_state state = state_read(&ep->com);
1729 CTR3(KTR_IW_CXGBE, "%s: ep %p, state %s", __func__, ep, states[state]);
1731 if (state != MPA_REQ_WAIT)
1734 iov.iov_base = &ep->mpa_pkt[ep->mpa_pkt_len];
1735 iov.iov_len = sizeof(ep->mpa_pkt) - ep->mpa_pkt_len;
1739 uio.uio_resid = sizeof(ep->mpa_pkt) - ep->mpa_pkt_len;
1740 uio.uio_segflg = UIO_SYSSPACE;
1741 uio.uio_rw = UIO_READ;
1742 uio.uio_td = NULL; /* uio.uio_td = ep->com.thread; */
1744 rc = soreceive(ep->com.so, NULL, &uio, NULL, NULL, &flags);
1750 abort_connection(ep);
1753 KASSERT(uio.uio_offset > 0, ("%s: sorecieve on so %p read no data",
1754 __func__, ep->com.so));
1755 ep->mpa_pkt_len += uio.uio_offset;
1758 * If we get more than the supported amount of private data then we must
1759 * fail this connection. XXX: check so_rcv->sb_cc, or peek with another
1760 * soreceive, or increase the size of mpa_pkt by 1 and abort if the last
1761 * byte is filled by the soreceive above.
1764 /* Don't even have the MPA message. Wait for more data to arrive. */
1765 if (ep->mpa_pkt_len < sizeof(*mpa))
1767 mpa = (struct mpa_message *) ep->mpa_pkt;
1770 * Validate MPA Header.
1772 if (mpa->revision > mpa_rev) {
1773 log(LOG_ERR, "%s: MPA version mismatch. Local = %d,"
1774 " Received = %d\n", __func__, mpa_rev, mpa->revision);
1778 if (memcmp(mpa->key, MPA_KEY_REQ, sizeof(mpa->key)))
1782 * Fail if there's too much private data.
1784 plen = ntohs(mpa->private_data_size);
1785 if (plen > MPA_MAX_PRIVATE_DATA)
1789 * If plen does not account for pkt size
1791 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen))
1794 ep->plen = (u8) plen;
1797 * If we don't have all the pdata yet, then bail.
1799 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
1803 * If we get here we have accumulated the entire mpa
1804 * start reply message including private data.
1806 ep->mpa_attr.initiator = 0;
1807 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1808 ep->mpa_attr.recv_marker_enabled = markers_enabled;
1809 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1810 ep->mpa_attr.version = mpa->revision;
1811 if (mpa->revision == 1)
1812 ep->tried_with_mpa_v1 = 1;
1813 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1815 if (mpa->revision == 2) {
1816 ep->mpa_attr.enhanced_rdma_conn =
1817 mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
1818 if (ep->mpa_attr.enhanced_rdma_conn) {
1819 struct mpa_v2_conn_params *mpa_v2_params;
1822 mpa_v2_params = (void *)&ep->mpa_pkt[sizeof(*mpa)];
1823 ird = ntohs(mpa_v2_params->ird);
1824 ord = ntohs(mpa_v2_params->ord);
1826 ep->ird = ird & MPA_V2_IRD_ORD_MASK;
1827 ep->ord = ord & MPA_V2_IRD_ORD_MASK;
1828 if (ird & MPA_V2_PEER2PEER_MODEL && peer2peer) {
1829 if (ord & MPA_V2_RDMA_WRITE_RTR) {
1830 ep->mpa_attr.p2p_type =
1831 FW_RI_INIT_P2PTYPE_RDMA_WRITE;
1832 } else if (ord & MPA_V2_RDMA_READ_RTR) {
1833 ep->mpa_attr.p2p_type =
1834 FW_RI_INIT_P2PTYPE_READ_REQ;
1838 } else if (mpa->revision == 1 && peer2peer)
1839 ep->mpa_attr.p2p_type = p2p_type;
1841 if (set_tcpinfo(ep))
1844 CTR5(KTR_IW_CXGBE, "%s: crc_enabled = %d, recv_marker_enabled = %d, "
1845 "xmit_marker_enabled = %d, version = %d", __func__,
1846 ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
1847 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version);
1849 state_set(&ep->com, MPA_REQ_RCVD);
1853 mutex_lock(&ep->parent_ep->com.mutex);
1854 if (ep->parent_ep->com.state != DEAD)
1855 connect_request_upcall(ep);
1857 abort_connection(ep);
1858 mutex_unlock(&ep->parent_ep->com.mutex);
1862 * Upcall from the adapter indicating data has been transmitted.
1863 * For us its just the single MPA request or reply. We can now free
1864 * the skb holding the mpa message.
1866 int c4iw_reject_cr(struct iw_cm_id *cm_id, const void *pdata, u8 pdata_len)
1869 struct c4iw_ep *ep = to_ep(cm_id);
1870 CTR2(KTR_IW_CXGBE, "%s:crcB %p", __func__, ep);
1872 if (state_read(&ep->com) == DEAD) {
1874 CTR2(KTR_IW_CXGBE, "%s:crc1 %p", __func__, ep);
1875 c4iw_put_ep(&ep->com);
1878 set_bit(ULP_REJECT, &ep->com.history);
1879 BUG_ON(state_read(&ep->com) != MPA_REQ_RCVD);
1883 CTR2(KTR_IW_CXGBE, "%s:crc2 %p", __func__, ep);
1884 abort_connection(ep);
1888 CTR2(KTR_IW_CXGBE, "%s:crc3 %p", __func__, ep);
1889 err = send_mpa_reject(ep, pdata, pdata_len);
1890 err = soshutdown(ep->com.so, 3);
1892 c4iw_put_ep(&ep->com);
1893 CTR2(KTR_IW_CXGBE, "%s:crc4 %p", __func__, ep);
1897 int c4iw_accept_cr(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
1900 struct c4iw_qp_attributes attrs;
1901 enum c4iw_qp_attr_mask mask;
1902 struct c4iw_ep *ep = to_ep(cm_id);
1903 struct c4iw_dev *h = to_c4iw_dev(cm_id->device);
1904 struct c4iw_qp *qp = get_qhp(h, conn_param->qpn);
1906 CTR2(KTR_IW_CXGBE, "%s:cacB %p", __func__, ep);
1908 if (state_read(&ep->com) == DEAD) {
1910 CTR2(KTR_IW_CXGBE, "%s:cac1 %p", __func__, ep);
1915 BUG_ON(state_read(&ep->com) != MPA_REQ_RCVD);
1918 set_bit(ULP_ACCEPT, &ep->com.history);
1920 if ((conn_param->ord > c4iw_max_read_depth) ||
1921 (conn_param->ird > c4iw_max_read_depth)) {
1923 CTR2(KTR_IW_CXGBE, "%s:cac2 %p", __func__, ep);
1924 abort_connection(ep);
1929 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
1931 CTR2(KTR_IW_CXGBE, "%s:cac3 %p", __func__, ep);
1933 if (conn_param->ord > ep->ird) {
1935 CTR2(KTR_IW_CXGBE, "%s:cac4 %p", __func__, ep);
1936 ep->ird = conn_param->ird;
1937 ep->ord = conn_param->ord;
1938 send_mpa_reject(ep, conn_param->private_data,
1939 conn_param->private_data_len);
1940 abort_connection(ep);
1945 if (conn_param->ird > ep->ord) {
1947 CTR2(KTR_IW_CXGBE, "%s:cac5 %p", __func__, ep);
1951 CTR2(KTR_IW_CXGBE, "%s:cac6 %p", __func__, ep);
1952 conn_param->ird = 1;
1955 CTR2(KTR_IW_CXGBE, "%s:cac7 %p", __func__, ep);
1956 abort_connection(ep);
1963 ep->ird = conn_param->ird;
1964 ep->ord = conn_param->ord;
1966 if (ep->mpa_attr.version != 2) {
1968 CTR2(KTR_IW_CXGBE, "%s:cac8 %p", __func__, ep);
1970 if (peer2peer && ep->ird == 0) {
1972 CTR2(KTR_IW_CXGBE, "%s:cac9 %p", __func__, ep);
1978 cm_id->add_ref(cm_id);
1979 ep->com.cm_id = cm_id;
1981 //ep->ofld_txq = TOEPCB(ep->com.so)->ofld_txq;
1983 /* bind QP to EP and move to RTS */
1984 attrs.mpa_attr = ep->mpa_attr;
1985 attrs.max_ird = ep->ird;
1986 attrs.max_ord = ep->ord;
1987 attrs.llp_stream_handle = ep;
1988 attrs.next_state = C4IW_QP_STATE_RTS;
1990 /* bind QP and TID with INIT_WR */
1991 mask = C4IW_QP_ATTR_NEXT_STATE |
1992 C4IW_QP_ATTR_LLP_STREAM_HANDLE |
1993 C4IW_QP_ATTR_MPA_ATTR |
1994 C4IW_QP_ATTR_MAX_IRD |
1995 C4IW_QP_ATTR_MAX_ORD;
1997 err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp, mask, &attrs, 1);
2001 CTR2(KTR_IW_CXGBE, "%s:caca %p", __func__, ep);
2004 err = send_mpa_reply(ep, conn_param->private_data,
2005 conn_param->private_data_len);
2009 CTR2(KTR_IW_CXGBE, "%s:caca %p", __func__, ep);
2013 state_set(&ep->com, FPDU_MODE);
2014 established_upcall(ep);
2015 c4iw_put_ep(&ep->com);
2016 CTR2(KTR_IW_CXGBE, "%s:cacE %p", __func__, ep);
2019 ep->com.cm_id = NULL;
2021 cm_id->rem_ref(cm_id);
2023 c4iw_put_ep(&ep->com);
2024 CTR2(KTR_IW_CXGBE, "%s:cacE err %p", __func__, ep);
2030 int c4iw_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
2033 struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
2034 struct c4iw_ep *ep = NULL;
2036 struct toedev *tdev;
2038 CTR2(KTR_IW_CXGBE, "%s:ccB %p", __func__, cm_id);
2040 if ((conn_param->ord > c4iw_max_read_depth) ||
2041 (conn_param->ird > c4iw_max_read_depth)) {
2043 CTR2(KTR_IW_CXGBE, "%s:cc1 %p", __func__, cm_id);
2047 ep = alloc_ep(sizeof(*ep), M_NOWAIT);
2051 CTR2(KTR_IW_CXGBE, "%s:cc2 %p", __func__, cm_id);
2052 printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__);
2056 init_timer(&ep->timer);
2057 ep->plen = conn_param->private_data_len;
2061 CTR2(KTR_IW_CXGBE, "%s:cc3 %p", __func__, ep);
2062 memcpy(ep->mpa_pkt + sizeof(struct mpa_message),
2063 conn_param->private_data, ep->plen);
2065 ep->ird = conn_param->ird;
2066 ep->ord = conn_param->ord;
2068 if (peer2peer && ep->ord == 0) {
2070 CTR2(KTR_IW_CXGBE, "%s:cc4 %p", __func__, ep);
2074 cm_id->add_ref(cm_id);
2076 ep->com.cm_id = cm_id;
2077 ep->com.qp = get_qhp(dev, conn_param->qpn);
2081 CTR2(KTR_IW_CXGBE, "%s:cc5 %p", __func__, ep);
2085 ep->com.thread = curthread;
2086 ep->com.so = cm_id->so;
2088 init_sock(&ep->com);
2092 cm_id->local_addr.sin_addr.s_addr,
2093 cm_id->remote_addr.sin_addr.s_addr,
2094 cm_id->local_addr.sin_port,
2095 cm_id->remote_addr.sin_port, 0);
2099 CTR2(KTR_IW_CXGBE, "%s:cc7 %p", __func__, ep);
2100 printk(KERN_ERR MOD "%s - cannot find route.\n", __func__);
2101 err = -EHOSTUNREACH;
2105 if (!(rt->rt_ifp->if_capenable & IFCAP_TOE)) {
2107 CTR2(KTR_IW_CXGBE, "%s:cc8 %p", __func__, ep);
2108 printf("%s - interface not TOE capable.\n", __func__);
2109 close_socket(&ep->com, 0);
2113 tdev = TOEDEV(rt->rt_ifp);
2117 CTR2(KTR_IW_CXGBE, "%s:cc9 %p", __func__, ep);
2118 printf("%s - No toedev for interface.\n", __func__);
2123 state_set(&ep->com, CONNECTING);
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,
2131 CTR2(KTR_IW_CXGBE, "%s:cca %p", __func__, ep);
2134 close_socket(&ep->com, 0);
2139 CTR2(KTR_IW_CXGBE, "%s:ccb %p", __func__, ep);
2142 cm_id->rem_ref(cm_id);
2143 c4iw_put_ep(&ep->com);
2145 CTR2(KTR_IW_CXGBE, "%s:ccE %p", __func__, ep);
2150 * iwcm->create_listen. Returns -errno on failure.
2153 c4iw_create_listen(struct iw_cm_id *cm_id, int backlog)
2156 struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
2157 struct c4iw_listen_ep *ep;
2158 struct socket *so = cm_id->so;
2160 ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
2161 CTR5(KTR_IW_CXGBE, "%s: cm_id %p, lso %p, ep %p, inp %p", __func__,
2162 cm_id, so, ep, so->so_pcb);
2164 log(LOG_ERR, "%s: failed to alloc memory for endpoint\n",
2170 cm_id->add_ref(cm_id);
2171 ep->com.cm_id = cm_id;
2173 ep->backlog = backlog;
2174 ep->com.local_addr = cm_id->local_addr;
2175 ep->com.thread = curthread;
2176 state_set(&ep->com, LISTEN);
2178 init_sock(&ep->com);
2180 rc = solisten(so, ep->backlog, ep->com.thread);
2182 log(LOG_ERR, "%s: failed to start listener: %d\n", __func__,
2184 close_socket(&ep->com, 0);
2185 cm_id->rem_ref(cm_id);
2186 c4iw_put_ep(&ep->com);
2190 cm_id->provider_data = ep;
2194 CTR3(KTR_IW_CXGBE, "%s: cm_id %p, FAILED (%d)", __func__, cm_id, rc);
2199 c4iw_destroy_listen(struct iw_cm_id *cm_id)
2202 struct c4iw_listen_ep *ep = to_listen_ep(cm_id);
2204 CTR4(KTR_IW_CXGBE, "%s: cm_id %p, so %p, inp %p", __func__, cm_id,
2205 cm_id->so, cm_id->so->so_pcb);
2207 state_set(&ep->com, DEAD);
2208 rc = close_socket(&ep->com, 0);
2209 cm_id->rem_ref(cm_id);
2210 c4iw_put_ep(&ep->com);
2215 int c4iw_ep_disconnect(struct c4iw_ep *ep, int abrupt, gfp_t gfp)
2220 struct c4iw_rdev *rdev;
2222 mutex_lock(&ep->com.mutex);
2224 CTR2(KTR_IW_CXGBE, "%s:cedB %p", __func__, ep);
2226 rdev = &ep->com.dev->rdev;
2228 if (c4iw_fatal_error(rdev)) {
2230 CTR2(KTR_IW_CXGBE, "%s:ced1 %p", __func__, ep);
2232 close_complete_upcall(ep, -EIO);
2233 ep->com.state = DEAD;
2235 CTR3(KTR_IW_CXGBE, "%s:ced2 %p %s", __func__, ep,
2236 states[ep->com.state]);
2238 switch (ep->com.state) {
2247 ep->com.state = ABORTING;
2249 ep->com.state = CLOSING;
2252 set_bit(CLOSE_SENT, &ep->com.flags);
2257 if (!test_and_set_bit(CLOSE_SENT, &ep->com.flags)) {
2262 ep->com.state = ABORTING;
2264 ep->com.state = MORIBUND;
2272 "%s ignoring disconnect ep %p state %u", __func__,
2281 mutex_unlock(&ep->com.mutex);
2285 CTR2(KTR_IW_CXGBE, "%s:ced3 %p", __func__, ep);
2289 CTR2(KTR_IW_CXGBE, "%s:ced4 %p", __func__, ep);
2290 set_bit(EP_DISC_ABORT, &ep->com.history);
2291 ret = abort_connection(ep);
2294 CTR2(KTR_IW_CXGBE, "%s:ced5 %p", __func__, ep);
2295 set_bit(EP_DISC_CLOSE, &ep->com.history);
2298 __state_set(&ep->com, MORIBUND);
2299 ret = shutdown_socket(&ep->com);
2310 release_ep_resources(ep);
2311 CTR2(KTR_IW_CXGBE, "%s:ced6 %p", __func__, ep);
2313 CTR2(KTR_IW_CXGBE, "%s:cedE %p", __func__, ep);
2317 #ifdef C4IW_EP_REDIRECT
2318 int c4iw_ep_redirect(void *ctx, struct dst_entry *old, struct dst_entry *new,
2319 struct l2t_entry *l2t)
2321 struct c4iw_ep *ep = ctx;
2326 PDBG("%s ep %p redirect to dst %p l2t %p\n", __func__, ep, new,
2329 cxgb4_l2t_release(ep->l2t);
2339 static void ep_timeout(unsigned long arg)
2341 struct c4iw_ep *ep = (struct c4iw_ep *)arg;
2344 CTR2(KTR_IW_CXGBE, "%s:etB %p", __func__, ep);
2345 spin_lock(&timeout_lock);
2347 if (!test_and_set_bit(TIMEOUT, &ep->com.flags)) {
2349 list_add_tail(&ep->entry, &timeout_list);
2352 spin_unlock(&timeout_lock);
2356 CTR2(KTR_IW_CXGBE, "%s:et1 %p", __func__, ep);
2357 queue_work(c4iw_taskq, &c4iw_task);
2359 CTR2(KTR_IW_CXGBE, "%s:etE %p", __func__, ep);
2362 static int fw6_wr_rpl(struct adapter *sc, const __be64 *rpl)
2364 uint64_t val = be64toh(*rpl);
2366 struct c4iw_wr_wait *wr_waitp;
2368 ret = (int)((val >> 8) & 0xff);
2369 wr_waitp = (struct c4iw_wr_wait *)rpl[1];
2370 CTR3(KTR_IW_CXGBE, "%s wr_waitp %p ret %u", __func__, wr_waitp, ret);
2372 c4iw_wake_up(wr_waitp, ret ? -ret : 0);
2377 static int fw6_cqe_handler(struct adapter *sc, const __be64 *rpl)
2379 struct t4_cqe cqe =*(const struct t4_cqe *)(&rpl[0]);
2381 CTR2(KTR_IW_CXGBE, "%s rpl %p", __func__, rpl);
2382 c4iw_ev_dispatch(sc->iwarp_softc, &cqe);
2387 static int terminate(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m)
2390 struct adapter *sc = iq->adapter;
2392 const struct cpl_rdma_terminate *rpl = (const void *)(rss + 1);
2393 unsigned int tid = GET_TID(rpl);
2394 struct c4iw_qp_attributes attrs;
2395 struct toepcb *toep = lookup_tid(sc, tid);
2396 struct socket *so = inp_inpcbtosocket(toep->inp);
2397 struct c4iw_ep *ep = so->so_rcv.sb_upcallarg;
2399 CTR2(KTR_IW_CXGBE, "%s:tB %p %d", __func__, ep);
2401 if (ep && ep->com.qp) {
2403 printk(KERN_WARNING MOD "TERM received tid %u qpid %u\n", tid,
2404 ep->com.qp->wq.sq.qid);
2405 attrs.next_state = C4IW_QP_STATE_TERMINATE;
2406 c4iw_modify_qp(ep->com.dev, ep->com.qp, C4IW_QP_ATTR_NEXT_STATE, &attrs,
2409 printk(KERN_WARNING MOD "TERM received tid %u no ep/qp\n", tid);
2410 CTR2(KTR_IW_CXGBE, "%s:tE %p %d", __func__, ep);
2416 c4iw_cm_init_cpl(struct adapter *sc)
2419 t4_register_cpl_handler(sc, CPL_RDMA_TERMINATE, terminate);
2420 t4_register_fw_msg_handler(sc, FW6_TYPE_WR_RPL, fw6_wr_rpl);
2421 t4_register_fw_msg_handler(sc, FW6_TYPE_CQE, fw6_cqe_handler);
2422 t4_register_an_handler(sc, c4iw_ev_handler);
2426 c4iw_cm_term_cpl(struct adapter *sc)
2429 t4_register_cpl_handler(sc, CPL_RDMA_TERMINATE, NULL);
2430 t4_register_fw_msg_handler(sc, FW6_TYPE_WR_RPL, NULL);
2431 t4_register_fw_msg_handler(sc, FW6_TYPE_CQE, NULL);
2434 int __init c4iw_cm_init(void)
2437 TAILQ_INIT(&req_list);
2438 spin_lock_init(&req_lock);
2439 INIT_LIST_HEAD(&timeout_list);
2440 spin_lock_init(&timeout_lock);
2442 INIT_WORK(&c4iw_task, process_req);
2444 c4iw_taskq = create_singlethread_workqueue("iw_cxgbe");
2452 void __exit c4iw_cm_term(void)
2454 WARN_ON(!TAILQ_EMPTY(&req_list));
2455 WARN_ON(!list_empty(&timeout_list));
2456 flush_workqueue(c4iw_taskq);
2457 destroy_workqueue(c4iw_taskq);