2 * Copyright (c) 2006 Oracle. 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
33 #include <linux/kernel.h>
35 #include <linux/slab.h>
36 #include <linux/vmalloc.h>
42 * Set the selected protocol version
44 static void rds_iw_set_protocol(struct rds_connection *conn, unsigned int version)
46 conn->c_version = version;
52 static void rds_iw_set_flow_control(struct rds_connection *conn, u32 credits)
54 struct rds_iw_connection *ic = conn->c_transport_data;
56 if (rds_iw_sysctl_flow_control && credits != 0) {
57 /* We're doing flow control */
59 rds_iw_send_add_credits(conn, credits);
66 * Connection established.
67 * We get here for both outgoing and incoming connection.
69 void rds_iw_cm_connect_complete(struct rds_connection *conn, struct rdma_cm_event *event)
71 const struct rds_iw_connect_private *dp = NULL;
72 struct rds_iw_connection *ic = conn->c_transport_data;
73 struct rds_iw_device *rds_iwdev;
76 if (event->param.conn.private_data_len) {
77 dp = event->param.conn.private_data;
79 rds_iw_set_protocol(conn,
80 RDS_PROTOCOL(dp->dp_protocol_major,
81 dp->dp_protocol_minor));
82 rds_iw_set_flow_control(conn, be32_to_cpu(dp->dp_credit));
85 /* update ib_device with this local ipaddr & conn */
86 rds_iwdev = ib_get_client_data(ic->i_cm_id->device, &rds_iw_client);
87 err = rds_iw_update_cm_id(rds_iwdev, ic->i_cm_id);
89 printk(KERN_ERR "rds_iw_update_ipaddr failed (%d)\n", err);
90 rds_iw_add_conn(rds_iwdev, conn);
92 /* If the peer gave us the last packet it saw, process this as if
93 * we had received a regular ACK. */
94 if (dp && dp->dp_ack_seq)
95 rds_send_drop_acked(conn, be64_to_cpu(dp->dp_ack_seq), NULL);
97 printk(KERN_NOTICE "RDS/IW: connected to %pI4<->%pI4 version %u.%u%s\n",
98 &conn->c_laddr, &conn->c_faddr,
99 RDS_PROTOCOL_MAJOR(conn->c_version),
100 RDS_PROTOCOL_MINOR(conn->c_version),
101 ic->i_flowctl ? ", flow control" : "");
103 rds_connect_complete(conn);
106 static void rds_iw_cm_fill_conn_param(struct rds_connection *conn,
107 struct rdma_conn_param *conn_param,
108 struct rds_iw_connect_private *dp,
109 u32 protocol_version)
111 struct rds_iw_connection *ic = conn->c_transport_data;
113 memset(conn_param, 0, sizeof(struct rdma_conn_param));
114 /* XXX tune these? */
115 conn_param->responder_resources = 1;
116 conn_param->initiator_depth = 1;
119 memset(dp, 0, sizeof(*dp));
120 dp->dp_saddr = conn->c_laddr;
121 dp->dp_daddr = conn->c_faddr;
122 dp->dp_protocol_major = RDS_PROTOCOL_MAJOR(protocol_version);
123 dp->dp_protocol_minor = RDS_PROTOCOL_MINOR(protocol_version);
124 dp->dp_protocol_minor_mask = cpu_to_be16(RDS_IW_SUPPORTED_PROTOCOLS);
125 dp->dp_ack_seq = rds_iw_piggyb_ack(ic);
127 /* Advertise flow control */
129 unsigned int credits;
131 credits = IB_GET_POST_CREDITS(atomic_read(&ic->i_credits));
132 dp->dp_credit = cpu_to_be32(credits);
133 atomic_sub(IB_SET_POST_CREDITS(credits), &ic->i_credits);
136 conn_param->private_data = dp;
137 conn_param->private_data_len = sizeof(*dp);
141 static void rds_iw_cq_event_handler(struct ib_event *event, void *data)
143 rdsdebug("event %u data %p\n", event->event, data);
146 static void rds_iw_qp_event_handler(struct ib_event *event, void *data)
148 struct rds_connection *conn = data;
149 struct rds_iw_connection *ic = conn->c_transport_data;
151 rdsdebug("conn %p ic %p event %u\n", conn, ic, event->event);
153 switch (event->event) {
154 case IB_EVENT_COMM_EST:
155 rdma_notify(ic->i_cm_id, IB_EVENT_COMM_EST);
157 case IB_EVENT_QP_REQ_ERR:
158 case IB_EVENT_QP_FATAL:
160 rds_iw_conn_error(conn, "RDS/IW: Fatal QP Event %u - connection %pI4->%pI4...reconnecting\n",
161 event->event, &conn->c_laddr,
170 static int rds_iw_init_qp_attrs(struct ib_qp_init_attr *attr,
171 struct rds_iw_device *rds_iwdev,
172 struct rds_iw_work_ring *send_ring,
173 void (*send_cq_handler)(struct ib_cq *, void *),
174 struct rds_iw_work_ring *recv_ring,
175 void (*recv_cq_handler)(struct ib_cq *, void *),
178 struct ib_device *dev = rds_iwdev->dev;
179 unsigned int send_size, recv_size;
182 /* The offset of 1 is to accomodate the additional ACK WR. */
183 send_size = min_t(unsigned int, rds_iwdev->max_wrs, rds_iw_sysctl_max_send_wr + 1);
184 recv_size = min_t(unsigned int, rds_iwdev->max_wrs, rds_iw_sysctl_max_recv_wr + 1);
185 rds_iw_ring_resize(send_ring, send_size - 1);
186 rds_iw_ring_resize(recv_ring, recv_size - 1);
188 memset(attr, 0, sizeof(*attr));
189 attr->event_handler = rds_iw_qp_event_handler;
190 attr->qp_context = context;
191 attr->cap.max_send_wr = send_size;
192 attr->cap.max_recv_wr = recv_size;
193 attr->cap.max_send_sge = rds_iwdev->max_sge;
194 attr->cap.max_recv_sge = RDS_IW_RECV_SGE;
195 attr->sq_sig_type = IB_SIGNAL_REQ_WR;
196 attr->qp_type = IB_QPT_RC;
198 attr->send_cq = ib_create_cq(dev, send_cq_handler,
199 rds_iw_cq_event_handler,
200 context, send_size, 0);
201 if (IS_ERR(attr->send_cq)) {
202 ret = PTR_ERR(attr->send_cq);
203 attr->send_cq = NULL;
204 rdsdebug("ib_create_cq send failed: %d\n", ret);
208 attr->recv_cq = ib_create_cq(dev, recv_cq_handler,
209 rds_iw_cq_event_handler,
210 context, recv_size, 0);
211 if (IS_ERR(attr->recv_cq)) {
212 ret = PTR_ERR(attr->recv_cq);
213 attr->recv_cq = NULL;
214 rdsdebug("ib_create_cq send failed: %d\n", ret);
218 ret = ib_req_notify_cq(attr->send_cq, IB_CQ_NEXT_COMP);
220 rdsdebug("ib_req_notify_cq send failed: %d\n", ret);
224 ret = ib_req_notify_cq(attr->recv_cq, IB_CQ_SOLICITED);
226 rdsdebug("ib_req_notify_cq recv failed: %d\n", ret);
233 ib_destroy_cq(attr->send_cq);
235 ib_destroy_cq(attr->recv_cq);
241 * This needs to be very careful to not leave IS_ERR pointers around for
242 * cleanup to trip over.
244 static int rds_iw_setup_qp(struct rds_connection *conn)
246 struct rds_iw_connection *ic = conn->c_transport_data;
247 struct ib_device *dev = ic->i_cm_id->device;
248 struct ib_qp_init_attr attr;
249 struct rds_iw_device *rds_iwdev;
252 /* rds_iw_add_one creates a rds_iw_device object per IB device,
253 * and allocates a protection domain, memory range and MR pool
254 * for each. If that fails for any reason, it will not register
255 * the rds_iwdev at all.
257 rds_iwdev = ib_get_client_data(dev, &rds_iw_client);
258 if (rds_iwdev == NULL) {
259 if (printk_ratelimit())
260 printk(KERN_NOTICE "RDS/IW: No client_data for device %s\n",
265 /* Protection domain and memory range */
266 ic->i_pd = rds_iwdev->pd;
267 ic->i_mr = rds_iwdev->mr;
269 ret = rds_iw_init_qp_attrs(&attr, rds_iwdev,
270 &ic->i_send_ring, rds_iw_send_cq_comp_handler,
271 &ic->i_recv_ring, rds_iw_recv_cq_comp_handler,
276 ic->i_send_cq = attr.send_cq;
277 ic->i_recv_cq = attr.recv_cq;
280 * XXX this can fail if max_*_wr is too large? Are we supposed
281 * to back off until we get a value that the hardware can support?
283 ret = rdma_create_qp(ic->i_cm_id, ic->i_pd, &attr);
285 rdsdebug("rdma_create_qp failed: %d\n", ret);
289 ic->i_send_hdrs = ib_dma_alloc_coherent(dev,
290 ic->i_send_ring.w_nr *
291 sizeof(struct rds_header),
292 &ic->i_send_hdrs_dma, GFP_KERNEL);
293 if (ic->i_send_hdrs == NULL) {
295 rdsdebug("ib_dma_alloc_coherent send failed\n");
299 ic->i_recv_hdrs = ib_dma_alloc_coherent(dev,
300 ic->i_recv_ring.w_nr *
301 sizeof(struct rds_header),
302 &ic->i_recv_hdrs_dma, GFP_KERNEL);
303 if (ic->i_recv_hdrs == NULL) {
305 rdsdebug("ib_dma_alloc_coherent recv failed\n");
309 ic->i_ack = ib_dma_alloc_coherent(dev, sizeof(struct rds_header),
310 &ic->i_ack_dma, GFP_KERNEL);
311 if (ic->i_ack == NULL) {
313 rdsdebug("ib_dma_alloc_coherent ack failed\n");
317 ic->i_sends = vmalloc(ic->i_send_ring.w_nr * sizeof(struct rds_iw_send_work));
318 if (ic->i_sends == NULL) {
320 rdsdebug("send allocation failed\n");
323 rds_iw_send_init_ring(ic);
325 ic->i_recvs = vmalloc(ic->i_recv_ring.w_nr * sizeof(struct rds_iw_recv_work));
326 if (ic->i_recvs == NULL) {
328 rdsdebug("recv allocation failed\n");
332 rds_iw_recv_init_ring(ic);
333 rds_iw_recv_init_ack(ic);
335 /* Post receive buffers - as a side effect, this will update
336 * the posted credit count. */
337 rds_iw_recv_refill(conn, GFP_KERNEL, GFP_HIGHUSER, 1);
339 rdsdebug("conn %p pd %p mr %p cq %p %p\n", conn, ic->i_pd, ic->i_mr,
340 ic->i_send_cq, ic->i_recv_cq);
346 static u32 rds_iw_protocol_compatible(const struct rds_iw_connect_private *dp)
351 /* rdma_cm private data is odd - when there is any private data in the
352 * request, we will be given a pretty large buffer without telling us the
353 * original size. The only way to tell the difference is by looking at
354 * the contents, which are initialized to zero.
355 * If the protocol version fields aren't set, this is a connection attempt
356 * from an older version. This could could be 3.0 or 2.0 - we can't tell.
357 * We really should have changed this for OFED 1.3 :-( */
358 if (dp->dp_protocol_major == 0)
359 return RDS_PROTOCOL_3_0;
361 common = be16_to_cpu(dp->dp_protocol_minor_mask) & RDS_IW_SUPPORTED_PROTOCOLS;
362 if (dp->dp_protocol_major == 3 && common) {
363 version = RDS_PROTOCOL_3_0;
364 while ((common >>= 1) != 0)
366 } else if (printk_ratelimit()) {
367 printk(KERN_NOTICE "RDS: Connection from %pI4 using "
368 "incompatible protocol version %u.%u\n",
370 dp->dp_protocol_major,
371 dp->dp_protocol_minor);
376 int rds_iw_cm_handle_connect(struct rdma_cm_id *cm_id,
377 struct rdma_cm_event *event)
379 const struct rds_iw_connect_private *dp = event->param.conn.private_data;
380 struct rds_iw_connect_private dp_rep;
381 struct rds_connection *conn = NULL;
382 struct rds_iw_connection *ic = NULL;
383 struct rdma_conn_param conn_param;
384 struct rds_iw_device *rds_iwdev;
386 int err, destroy = 1;
388 /* Check whether the remote protocol version matches ours. */
389 version = rds_iw_protocol_compatible(dp);
393 rdsdebug("saddr %pI4 daddr %pI4 RDSv%u.%u\n",
394 &dp->dp_saddr, &dp->dp_daddr,
395 RDS_PROTOCOL_MAJOR(version), RDS_PROTOCOL_MINOR(version));
397 conn = rds_conn_create(dp->dp_daddr, dp->dp_saddr, &rds_iw_transport,
400 rdsdebug("rds_conn_create failed (%ld)\n", PTR_ERR(conn));
406 * The connection request may occur while the
407 * previous connection exist, e.g. in case of failover.
408 * But as connections may be initiated simultaneously
409 * by both hosts, we have a random backoff mechanism -
410 * see the comment above rds_queue_reconnect()
412 mutex_lock(&conn->c_cm_lock);
413 if (!rds_conn_transition(conn, RDS_CONN_DOWN, RDS_CONN_CONNECTING)) {
414 if (rds_conn_state(conn) == RDS_CONN_UP) {
415 rdsdebug("incoming connect while connecting\n");
417 rds_iw_stats_inc(s_iw_listen_closed_stale);
419 if (rds_conn_state(conn) == RDS_CONN_CONNECTING) {
420 /* Wait and see - our connect may still be succeeding */
421 rds_iw_stats_inc(s_iw_connect_raced);
423 mutex_unlock(&conn->c_cm_lock);
427 ic = conn->c_transport_data;
429 rds_iw_set_protocol(conn, version);
430 rds_iw_set_flow_control(conn, be32_to_cpu(dp->dp_credit));
432 /* If the peer gave us the last packet it saw, process this as if
433 * we had received a regular ACK. */
435 rds_send_drop_acked(conn, be64_to_cpu(dp->dp_ack_seq), NULL);
437 BUG_ON(cm_id->context);
441 cm_id->context = conn;
443 rds_iwdev = ib_get_client_data(cm_id->device, &rds_iw_client);
444 ic->i_dma_local_lkey = rds_iwdev->dma_local_lkey;
446 /* We got halfway through setting up the ib_connection, if we
447 * fail now, we have to take the long route out of this mess. */
450 err = rds_iw_setup_qp(conn);
452 rds_iw_conn_error(conn, "rds_iw_setup_qp failed (%d)\n", err);
456 rds_iw_cm_fill_conn_param(conn, &conn_param, &dp_rep, version);
458 /* rdma_accept() calls rdma_reject() internally if it fails */
459 err = rdma_accept(cm_id, &conn_param);
460 mutex_unlock(&conn->c_cm_lock);
462 rds_iw_conn_error(conn, "rdma_accept failed (%d)\n", err);
469 rdma_reject(cm_id, NULL, 0);
474 int rds_iw_cm_initiate_connect(struct rdma_cm_id *cm_id)
476 struct rds_connection *conn = cm_id->context;
477 struct rds_iw_connection *ic = conn->c_transport_data;
478 struct rdma_conn_param conn_param;
479 struct rds_iw_connect_private dp;
482 /* If the peer doesn't do protocol negotiation, we must
483 * default to RDSv3.0 */
484 rds_iw_set_protocol(conn, RDS_PROTOCOL_3_0);
485 ic->i_flowctl = rds_iw_sysctl_flow_control; /* advertise flow control */
487 ret = rds_iw_setup_qp(conn);
489 rds_iw_conn_error(conn, "rds_iw_setup_qp failed (%d)\n", ret);
493 rds_iw_cm_fill_conn_param(conn, &conn_param, &dp, RDS_PROTOCOL_VERSION);
495 ret = rdma_connect(cm_id, &conn_param);
497 rds_iw_conn_error(conn, "rdma_connect failed (%d)\n", ret);
500 /* Beware - returning non-zero tells the rdma_cm to destroy
501 * the cm_id. We should certainly not do it as long as we still
502 * "own" the cm_id. */
504 struct rds_iw_connection *ic = conn->c_transport_data;
506 if (ic->i_cm_id == cm_id)
512 int rds_iw_conn_connect(struct rds_connection *conn)
514 struct rds_iw_connection *ic = conn->c_transport_data;
515 struct rds_iw_device *rds_iwdev;
516 struct sockaddr_in src, dest;
519 /* XXX I wonder what affect the port space has */
520 /* delegate cm event handler to rdma_transport */
521 ic->i_cm_id = rdma_create_id(rds_rdma_cm_event_handler, conn,
523 if (IS_ERR(ic->i_cm_id)) {
524 ret = PTR_ERR(ic->i_cm_id);
526 rdsdebug("rdma_create_id() failed: %d\n", ret);
530 rdsdebug("created cm id %p for conn %p\n", ic->i_cm_id, conn);
532 src.sin_family = AF_INET;
533 src.sin_addr.s_addr = (__force u32)conn->c_laddr;
534 src.sin_port = (__force u16)htons(0);
536 /* First, bind to the local address and device. */
537 ret = rdma_bind_addr(ic->i_cm_id, (struct sockaddr *) &src);
539 rdsdebug("rdma_bind_addr(%pI4) failed: %d\n",
540 &conn->c_laddr, ret);
541 rdma_destroy_id(ic->i_cm_id);
546 rds_iwdev = ib_get_client_data(ic->i_cm_id->device, &rds_iw_client);
547 ic->i_dma_local_lkey = rds_iwdev->dma_local_lkey;
549 dest.sin_family = AF_INET;
550 dest.sin_addr.s_addr = (__force u32)conn->c_faddr;
551 dest.sin_port = (__force u16)htons(RDS_PORT);
553 ret = rdma_resolve_addr(ic->i_cm_id, (struct sockaddr *)&src,
554 (struct sockaddr *)&dest,
555 RDS_RDMA_RESOLVE_TIMEOUT_MS);
557 rdsdebug("addr resolve failed for cm id %p: %d\n", ic->i_cm_id,
559 rdma_destroy_id(ic->i_cm_id);
568 * This is so careful about only cleaning up resources that were built up
569 * so that it can be called at any point during startup. In fact it
570 * can be called multiple times for a given connection.
572 void rds_iw_conn_shutdown(struct rds_connection *conn)
574 struct rds_iw_connection *ic = conn->c_transport_data;
576 struct ib_qp_attr qp_attr;
578 rdsdebug("cm %p pd %p cq %p %p qp %p\n", ic->i_cm_id,
579 ic->i_pd, ic->i_send_cq, ic->i_recv_cq,
580 ic->i_cm_id ? ic->i_cm_id->qp : NULL);
583 struct ib_device *dev = ic->i_cm_id->device;
585 rdsdebug("disconnecting cm %p\n", ic->i_cm_id);
586 err = rdma_disconnect(ic->i_cm_id);
588 /* Actually this may happen quite frequently, when
589 * an outgoing connect raced with an incoming connect.
591 rdsdebug("rds_iw_conn_shutdown: failed to disconnect,"
592 " cm: %p err %d\n", ic->i_cm_id, err);
595 if (ic->i_cm_id->qp) {
596 qp_attr.qp_state = IB_QPS_ERR;
597 ib_modify_qp(ic->i_cm_id->qp, &qp_attr, IB_QP_STATE);
600 wait_event(rds_iw_ring_empty_wait,
601 rds_iw_ring_empty(&ic->i_send_ring) &&
602 rds_iw_ring_empty(&ic->i_recv_ring));
605 ib_dma_free_coherent(dev,
606 ic->i_send_ring.w_nr *
607 sizeof(struct rds_header),
609 ic->i_send_hdrs_dma);
612 ib_dma_free_coherent(dev,
613 ic->i_recv_ring.w_nr *
614 sizeof(struct rds_header),
616 ic->i_recv_hdrs_dma);
619 ib_dma_free_coherent(dev, sizeof(struct rds_header),
620 ic->i_ack, ic->i_ack_dma);
623 rds_iw_send_clear_ring(ic);
625 rds_iw_recv_clear_ring(ic);
628 rdma_destroy_qp(ic->i_cm_id);
630 ib_destroy_cq(ic->i_send_cq);
632 ib_destroy_cq(ic->i_recv_cq);
635 * If associated with an rds_iw_device:
636 * Move connection back to the nodev list.
637 * Remove cm_id from the device cm_id list.
640 rds_iw_remove_conn(ic->rds_iwdev, conn);
642 rdma_destroy_id(ic->i_cm_id);
647 ic->i_send_cq = NULL;
648 ic->i_recv_cq = NULL;
649 ic->i_send_hdrs = NULL;
650 ic->i_recv_hdrs = NULL;
653 BUG_ON(ic->rds_iwdev);
655 /* Clear pending transmit */
657 rds_message_put(ic->i_rm);
661 /* Clear the ACK state */
662 clear_bit(IB_ACK_IN_FLIGHT, &ic->i_ack_flags);
663 #ifdef KERNEL_HAS_ATOMIC64
664 atomic64_set(&ic->i_ack_next, 0);
670 /* Clear flow control state */
672 atomic_set(&ic->i_credits, 0);
674 rds_iw_ring_init(&ic->i_send_ring, rds_iw_sysctl_max_send_wr);
675 rds_iw_ring_init(&ic->i_recv_ring, rds_iw_sysctl_max_recv_wr);
678 rds_inc_put(&ic->i_iwinc->ii_inc);
686 rdsdebug("shutdown complete\n");
689 int rds_iw_conn_alloc(struct rds_connection *conn, gfp_t gfp)
691 struct rds_iw_connection *ic;
695 ic = kzalloc(sizeof(struct rds_iw_connection), GFP_KERNEL);
699 INIT_LIST_HEAD(&ic->iw_node);
700 tasklet_init(&ic->i_recv_tasklet, rds_iw_recv_tasklet_fn,
702 mutex_init(&ic->i_recv_mutex);
703 #ifndef KERNEL_HAS_ATOMIC64
704 spin_lock_init(&ic->i_ack_lock);
708 * rds_iw_conn_shutdown() waits for these to be emptied so they
709 * must be initialized before it can be called.
711 rds_iw_ring_init(&ic->i_send_ring, rds_iw_sysctl_max_send_wr);
712 rds_iw_ring_init(&ic->i_recv_ring, rds_iw_sysctl_max_recv_wr);
715 conn->c_transport_data = ic;
717 spin_lock_irqsave(&iw_nodev_conns_lock, flags);
718 list_add_tail(&ic->iw_node, &iw_nodev_conns);
719 spin_unlock_irqrestore(&iw_nodev_conns_lock, flags);
722 rdsdebug("conn %p conn ic %p\n", conn, conn->c_transport_data);
727 * Free a connection. Connection must be shut down and not set for reconnect.
729 void rds_iw_conn_free(void *arg)
731 struct rds_iw_connection *ic = arg;
732 spinlock_t *lock_ptr;
734 rdsdebug("ic %p\n", ic);
737 * Conn is either on a dev's list or on the nodev list.
738 * A race with shutdown() or connect() would cause problems
739 * (since rds_iwdev would change) but that should never happen.
741 lock_ptr = ic->rds_iwdev ? &ic->rds_iwdev->spinlock : &iw_nodev_conns_lock;
743 spin_lock_irq(lock_ptr);
744 list_del(&ic->iw_node);
745 spin_unlock_irq(lock_ptr);
751 * An error occurred on the connection
754 __rds_iw_conn_error(struct rds_connection *conn, const char *fmt, ...)