1/* 2 * Copyright (c) 2006 Oracle. 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 */ 33#include <linux/kernel.h> 34#include <linux/in.h> 35#include <linux/slab.h> 36#include <linux/vmalloc.h> 37 38#include "rds.h" 39#include "iw.h" 40 41/* 42 * Set the selected protocol version 43 */ 44static void rds_iw_set_protocol(struct rds_connection *conn, unsigned int version) 45{ 46 conn->c_version = version; 47} 48 49/* 50 * Set up flow control 51 */ 52static void rds_iw_set_flow_control(struct rds_connection *conn, u32 credits) 53{ 54 struct rds_iw_connection *ic = conn->c_transport_data; 55 56 if (rds_iw_sysctl_flow_control && credits != 0) { 57 /* We're doing flow control */ 58 ic->i_flowctl = 1; 59 rds_iw_send_add_credits(conn, credits); 60 } else { 61 ic->i_flowctl = 0; 62 } 63} 64 65/* 66 * Connection established. 67 * We get here for both outgoing and incoming connection. 68 */ 69void rds_iw_cm_connect_complete(struct rds_connection *conn, struct rdma_cm_event *event) 70{ 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; 74 int err; 75 76 if (event->param.conn.private_data_len) { 77 dp = event->param.conn.private_data; 78 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)); 83 } 84 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); 88 if (err) 89 printk(KERN_ERR "rds_iw_update_ipaddr failed (%d)\n", err); 90 rds_iw_add_conn(rds_iwdev, conn); 91 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); 96 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" : ""); 102 103 rds_connect_complete(conn); 104} 105 106static 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) 110{ 111 struct rds_iw_connection *ic = conn->c_transport_data; 112 113 memset(conn_param, 0, sizeof(struct rdma_conn_param)); 114 conn_param->responder_resources = 1; 115 conn_param->initiator_depth = 1; 116 117 if (dp) { 118 memset(dp, 0, sizeof(*dp)); 119 dp->dp_saddr = conn->c_laddr; 120 dp->dp_daddr = conn->c_faddr; 121 dp->dp_protocol_major = RDS_PROTOCOL_MAJOR(protocol_version); 122 dp->dp_protocol_minor = RDS_PROTOCOL_MINOR(protocol_version); 123 dp->dp_protocol_minor_mask = cpu_to_be16(RDS_IW_SUPPORTED_PROTOCOLS); 124 dp->dp_ack_seq = rds_iw_piggyb_ack(ic); 125 126 /* Advertise flow control */ 127 if (ic->i_flowctl) { 128 unsigned int credits; 129 130 credits = IB_GET_POST_CREDITS(atomic_read(&ic->i_credits)); 131 dp->dp_credit = cpu_to_be32(credits); 132 atomic_sub(IB_SET_POST_CREDITS(credits), &ic->i_credits); 133 } 134 135 conn_param->private_data = dp; 136 conn_param->private_data_len = sizeof(*dp); 137 } 138} 139 140static void rds_iw_cq_event_handler(struct ib_event *event, void *data) 141{ 142 rdsdebug("event %u data %p\n", event->event, data); 143} 144 145static void rds_iw_qp_event_handler(struct ib_event *event, void *data) 146{ 147 struct rds_connection *conn = data; 148 struct rds_iw_connection *ic = conn->c_transport_data; 149 150 rdsdebug("conn %p ic %p event %u\n", conn, ic, event->event); 151 152 switch (event->event) { 153 case IB_EVENT_COMM_EST: 154 rdma_notify(ic->i_cm_id, IB_EVENT_COMM_EST); 155 break; 156 case IB_EVENT_QP_REQ_ERR: 157 case IB_EVENT_QP_FATAL: 158 default: 159 rdsdebug("Fatal QP Event %u " 160 "- connection %pI4->%pI4, reconnecting\n", 161 event->event, &conn->c_laddr, 162 &conn->c_faddr); 163 rds_conn_drop(conn); 164 break; 165 } 166} 167 168/* 169 * Create a QP 170 */ 171static int rds_iw_init_qp_attrs(struct ib_qp_init_attr *attr, 172 struct rds_iw_device *rds_iwdev, 173 struct rds_iw_work_ring *send_ring, 174 void (*send_cq_handler)(struct ib_cq *, void *), 175 struct rds_iw_work_ring *recv_ring, 176 void (*recv_cq_handler)(struct ib_cq *, void *), 177 void *context) 178{ 179 struct ib_device *dev = rds_iwdev->dev; 180 unsigned int send_size, recv_size; 181 int ret; 182 183 /* The offset of 1 is to accomodate the additional ACK WR. */ 184 send_size = min_t(unsigned int, rds_iwdev->max_wrs, rds_iw_sysctl_max_send_wr + 1); 185 recv_size = min_t(unsigned int, rds_iwdev->max_wrs, rds_iw_sysctl_max_recv_wr + 1); 186 rds_iw_ring_resize(send_ring, send_size - 1); 187 rds_iw_ring_resize(recv_ring, recv_size - 1); 188 189 memset(attr, 0, sizeof(*attr)); 190 attr->event_handler = rds_iw_qp_event_handler; 191 attr->qp_context = context; 192 attr->cap.max_send_wr = send_size; 193 attr->cap.max_recv_wr = recv_size; 194 attr->cap.max_send_sge = rds_iwdev->max_sge; 195 attr->cap.max_recv_sge = RDS_IW_RECV_SGE; 196 attr->sq_sig_type = IB_SIGNAL_REQ_WR; 197 attr->qp_type = IB_QPT_RC; 198 199 attr->send_cq = ib_create_cq(dev, send_cq_handler, 200 rds_iw_cq_event_handler, 201 context, send_size, 0); 202 if (IS_ERR(attr->send_cq)) { 203 ret = PTR_ERR(attr->send_cq); 204 attr->send_cq = NULL; 205 rdsdebug("ib_create_cq send failed: %d\n", ret); 206 goto out; 207 } 208 209 attr->recv_cq = ib_create_cq(dev, recv_cq_handler, 210 rds_iw_cq_event_handler, 211 context, recv_size, 0); 212 if (IS_ERR(attr->recv_cq)) { 213 ret = PTR_ERR(attr->recv_cq); 214 attr->recv_cq = NULL; 215 rdsdebug("ib_create_cq send failed: %d\n", ret); 216 goto out; 217 } 218 219 ret = ib_req_notify_cq(attr->send_cq, IB_CQ_NEXT_COMP); 220 if (ret) { 221 rdsdebug("ib_req_notify_cq send failed: %d\n", ret); 222 goto out; 223 } 224 225 ret = ib_req_notify_cq(attr->recv_cq, IB_CQ_SOLICITED); 226 if (ret) { 227 rdsdebug("ib_req_notify_cq recv failed: %d\n", ret); 228 goto out; 229 } 230 231out: 232 if (ret) { 233 if (attr->send_cq) 234 ib_destroy_cq(attr->send_cq); 235 if (attr->recv_cq) 236 ib_destroy_cq(attr->recv_cq); 237 } 238 return ret; 239} 240 241/* 242 * This needs to be very careful to not leave IS_ERR pointers around for 243 * cleanup to trip over. 244 */ 245static int rds_iw_setup_qp(struct rds_connection *conn) 246{ 247 struct rds_iw_connection *ic = conn->c_transport_data; 248 struct ib_device *dev = ic->i_cm_id->device; 249 struct ib_qp_init_attr attr; 250 struct rds_iw_device *rds_iwdev; 251 int ret; 252 253 /* rds_iw_add_one creates a rds_iw_device object per IB device, 254 * and allocates a protection domain, memory range and MR pool 255 * for each. If that fails for any reason, it will not register 256 * the rds_iwdev at all. 257 */ 258 rds_iwdev = ib_get_client_data(dev, &rds_iw_client); 259 if (rds_iwdev == NULL) { 260 if (printk_ratelimit()) 261 printk(KERN_NOTICE "RDS/IW: No client_data for device %s\n", 262 dev->name); 263 return -EOPNOTSUPP; 264 } 265 266 /* Protection domain and memory range */ 267 ic->i_pd = rds_iwdev->pd; 268 ic->i_mr = rds_iwdev->mr; 269 270 ret = rds_iw_init_qp_attrs(&attr, rds_iwdev, 271 &ic->i_send_ring, rds_iw_send_cq_comp_handler, 272 &ic->i_recv_ring, rds_iw_recv_cq_comp_handler, 273 conn); 274 if (ret < 0) 275 goto out; 276 277 ic->i_send_cq = attr.send_cq; 278 ic->i_recv_cq = attr.recv_cq; 279 280 ret = rdma_create_qp(ic->i_cm_id, ic->i_pd, &attr); 281 if (ret) { 282 rdsdebug("rdma_create_qp failed: %d\n", ret); 283 goto out; 284 } 285 286 ic->i_send_hdrs = ib_dma_alloc_coherent(dev, 287 ic->i_send_ring.w_nr * 288 sizeof(struct rds_header), 289 &ic->i_send_hdrs_dma, GFP_KERNEL); 290 if (ic->i_send_hdrs == NULL) { 291 ret = -ENOMEM; 292 rdsdebug("ib_dma_alloc_coherent send failed\n"); 293 goto out; 294 } 295 296 ic->i_recv_hdrs = ib_dma_alloc_coherent(dev, 297 ic->i_recv_ring.w_nr * 298 sizeof(struct rds_header), 299 &ic->i_recv_hdrs_dma, GFP_KERNEL); 300 if (ic->i_recv_hdrs == NULL) { 301 ret = -ENOMEM; 302 rdsdebug("ib_dma_alloc_coherent recv failed\n"); 303 goto out; 304 } 305 306 ic->i_ack = ib_dma_alloc_coherent(dev, sizeof(struct rds_header), 307 &ic->i_ack_dma, GFP_KERNEL); 308 if (ic->i_ack == NULL) { 309 ret = -ENOMEM; 310 rdsdebug("ib_dma_alloc_coherent ack failed\n"); 311 goto out; 312 } 313 314 ic->i_sends = vmalloc(ic->i_send_ring.w_nr * sizeof(struct rds_iw_send_work)); 315 if (ic->i_sends == NULL) { 316 ret = -ENOMEM; 317 rdsdebug("send allocation failed\n"); 318 goto out; 319 } 320 rds_iw_send_init_ring(ic); 321 322 ic->i_recvs = vmalloc(ic->i_recv_ring.w_nr * sizeof(struct rds_iw_recv_work)); 323 if (ic->i_recvs == NULL) { 324 ret = -ENOMEM; 325 rdsdebug("recv allocation failed\n"); 326 goto out; 327 } 328 329 rds_iw_recv_init_ring(ic); 330 rds_iw_recv_init_ack(ic); 331 332 /* Post receive buffers - as a side effect, this will update 333 * the posted credit count. */ 334 rds_iw_recv_refill(conn, GFP_KERNEL, GFP_HIGHUSER, 1); 335 336 rdsdebug("conn %p pd %p mr %p cq %p %p\n", conn, ic->i_pd, ic->i_mr, 337 ic->i_send_cq, ic->i_recv_cq); 338 339out: 340 return ret; 341} 342 343static u32 rds_iw_protocol_compatible(const struct rds_iw_connect_private *dp) 344{ 345 u16 common; 346 u32 version = 0; 347 348 /* rdma_cm private data is odd - when there is any private data in the 349 * request, we will be given a pretty large buffer without telling us the 350 * original size. The only way to tell the difference is by looking at 351 * the contents, which are initialized to zero. 352 * If the protocol version fields aren't set, this is a connection attempt 353 * from an older version. This could could be 3.0 or 2.0 - we can't tell. 354 * We really should have changed this for OFED 1.3 :-( */ 355 if (dp->dp_protocol_major == 0) 356 return RDS_PROTOCOL_3_0; 357 358 common = be16_to_cpu(dp->dp_protocol_minor_mask) & RDS_IW_SUPPORTED_PROTOCOLS; 359 if (dp->dp_protocol_major == 3 && common) { 360 version = RDS_PROTOCOL_3_0; 361 while ((common >>= 1) != 0) 362 version++; 363 } else if (printk_ratelimit()) { 364 printk(KERN_NOTICE "RDS: Connection from %pI4 using " 365 "incompatible protocol version %u.%u\n", 366 &dp->dp_saddr, 367 dp->dp_protocol_major, 368 dp->dp_protocol_minor); 369 } 370 return version; 371} 372 373int rds_iw_cm_handle_connect(struct rdma_cm_id *cm_id, 374 struct rdma_cm_event *event) 375{ 376 const struct rds_iw_connect_private *dp = event->param.conn.private_data; 377 struct rds_iw_connect_private dp_rep; 378 struct rds_connection *conn = NULL; 379 struct rds_iw_connection *ic = NULL; 380 struct rdma_conn_param conn_param; 381 struct rds_iw_device *rds_iwdev; 382 u32 version; 383 int err, destroy = 1; 384 385 /* Check whether the remote protocol version matches ours. */ 386 version = rds_iw_protocol_compatible(dp); 387 if (!version) 388 goto out; 389 390 rdsdebug("saddr %pI4 daddr %pI4 RDSv%u.%u\n", 391 &dp->dp_saddr, &dp->dp_daddr, 392 RDS_PROTOCOL_MAJOR(version), RDS_PROTOCOL_MINOR(version)); 393 394 conn = rds_conn_create(dp->dp_daddr, dp->dp_saddr, &rds_iw_transport, 395 GFP_KERNEL); 396 if (IS_ERR(conn)) { 397 rdsdebug("rds_conn_create failed (%ld)\n", PTR_ERR(conn)); 398 conn = NULL; 399 goto out; 400 } 401 402 /* 403 * The connection request may occur while the 404 * previous connection exist, e.g. in case of failover. 405 * But as connections may be initiated simultaneously 406 * by both hosts, we have a random backoff mechanism - 407 * see the comment above rds_queue_reconnect() 408 */ 409 mutex_lock(&conn->c_cm_lock); 410 if (!rds_conn_transition(conn, RDS_CONN_DOWN, RDS_CONN_CONNECTING)) { 411 if (rds_conn_state(conn) == RDS_CONN_UP) { 412 rdsdebug("incoming connect while connecting\n"); 413 rds_conn_drop(conn); 414 rds_iw_stats_inc(s_iw_listen_closed_stale); 415 } else 416 if (rds_conn_state(conn) == RDS_CONN_CONNECTING) { 417 /* Wait and see - our connect may still be succeeding */ 418 rds_iw_stats_inc(s_iw_connect_raced); 419 } 420 mutex_unlock(&conn->c_cm_lock); 421 goto out; 422 } 423 424 ic = conn->c_transport_data; 425 426 rds_iw_set_protocol(conn, version); 427 rds_iw_set_flow_control(conn, be32_to_cpu(dp->dp_credit)); 428 429 /* If the peer gave us the last packet it saw, process this as if 430 * we had received a regular ACK. */ 431 if (dp->dp_ack_seq) 432 rds_send_drop_acked(conn, be64_to_cpu(dp->dp_ack_seq), NULL); 433 434 BUG_ON(cm_id->context); 435 BUG_ON(ic->i_cm_id); 436 437 ic->i_cm_id = cm_id; 438 cm_id->context = conn; 439 440 rds_iwdev = ib_get_client_data(cm_id->device, &rds_iw_client); 441 ic->i_dma_local_lkey = rds_iwdev->dma_local_lkey; 442 443 /* We got halfway through setting up the ib_connection, if we 444 * fail now, we have to take the long route out of this mess. */ 445 destroy = 0; 446 447 err = rds_iw_setup_qp(conn); 448 if (err) { 449 rds_iw_conn_error(conn, "rds_iw_setup_qp failed (%d)\n", err); 450 mutex_unlock(&conn->c_cm_lock); 451 goto out; 452 } 453 454 rds_iw_cm_fill_conn_param(conn, &conn_param, &dp_rep, version); 455 456 /* rdma_accept() calls rdma_reject() internally if it fails */ 457 err = rdma_accept(cm_id, &conn_param); 458 mutex_unlock(&conn->c_cm_lock); 459 if (err) { 460 rds_iw_conn_error(conn, "rdma_accept failed (%d)\n", err); 461 goto out; 462 } 463 464 return 0; 465 466out: 467 rdma_reject(cm_id, NULL, 0); 468 return destroy; 469} 470 471 472int rds_iw_cm_initiate_connect(struct rdma_cm_id *cm_id) 473{ 474 struct rds_connection *conn = cm_id->context; 475 struct rds_iw_connection *ic = conn->c_transport_data; 476 struct rdma_conn_param conn_param; 477 struct rds_iw_connect_private dp; 478 int ret; 479 480 /* If the peer doesn't do protocol negotiation, we must 481 * default to RDSv3.0 */ 482 rds_iw_set_protocol(conn, RDS_PROTOCOL_3_0); 483 ic->i_flowctl = rds_iw_sysctl_flow_control; /* advertise flow control */ 484 485 ret = rds_iw_setup_qp(conn); 486 if (ret) { 487 rds_iw_conn_error(conn, "rds_iw_setup_qp failed (%d)\n", ret); 488 goto out; 489 } 490 491 rds_iw_cm_fill_conn_param(conn, &conn_param, &dp, RDS_PROTOCOL_VERSION); 492 493 ret = rdma_connect(cm_id, &conn_param); 494 if (ret) 495 rds_iw_conn_error(conn, "rdma_connect failed (%d)\n", ret); 496 497out: 498 /* Beware - returning non-zero tells the rdma_cm to destroy 499 * the cm_id. We should certainly not do it as long as we still 500 * "own" the cm_id. */ 501 if (ret) { 502 struct rds_iw_connection *ic = conn->c_transport_data; 503 504 if (ic->i_cm_id == cm_id) 505 ret = 0; 506 } 507 return ret; 508} 509 510int rds_iw_conn_connect(struct rds_connection *conn) 511{ 512 struct rds_iw_connection *ic = conn->c_transport_data; 513 struct rds_iw_device *rds_iwdev; 514 struct sockaddr_in src, dest; 515 int ret; 516 517 /* delegate cm event handler to rdma_transport */ 518 ic->i_cm_id = rdma_create_id(rds_rdma_cm_event_handler, conn, 519 RDMA_PS_TCP); 520 if (IS_ERR(ic->i_cm_id)) { 521 ret = PTR_ERR(ic->i_cm_id); 522 ic->i_cm_id = NULL; 523 rdsdebug("rdma_create_id() failed: %d\n", ret); 524 goto out; 525 } 526 527 rdsdebug("created cm id %p for conn %p\n", ic->i_cm_id, conn); 528 529 src.sin_family = AF_INET; 530 src.sin_addr.s_addr = (__force u32)conn->c_laddr; 531 src.sin_port = (__force u16)htons(0); 532 533 /* First, bind to the local address and device. */ 534 ret = rdma_bind_addr(ic->i_cm_id, (struct sockaddr *) &src); 535 if (ret) { 536 rdsdebug("rdma_bind_addr(%pI4) failed: %d\n", 537 &conn->c_laddr, ret); 538 rdma_destroy_id(ic->i_cm_id); 539 ic->i_cm_id = NULL; 540 goto out; 541 } 542 543 rds_iwdev = ib_get_client_data(ic->i_cm_id->device, &rds_iw_client); 544 ic->i_dma_local_lkey = rds_iwdev->dma_local_lkey; 545 546 dest.sin_family = AF_INET; 547 dest.sin_addr.s_addr = (__force u32)conn->c_faddr; 548 dest.sin_port = (__force u16)htons(RDS_PORT); 549 550 ret = rdma_resolve_addr(ic->i_cm_id, (struct sockaddr *)&src, 551 (struct sockaddr *)&dest, 552 RDS_RDMA_RESOLVE_TIMEOUT_MS); 553 if (ret) { 554 rdsdebug("addr resolve failed for cm id %p: %d\n", ic->i_cm_id, 555 ret); 556 rdma_destroy_id(ic->i_cm_id); 557 ic->i_cm_id = NULL; 558 } 559 560out: 561 return ret; 562} 563 564/* 565 * This is so careful about only cleaning up resources that were built up 566 * so that it can be called at any point during startup. In fact it 567 * can be called multiple times for a given connection. 568 */ 569void rds_iw_conn_shutdown(struct rds_connection *conn) 570{ 571 struct rds_iw_connection *ic = conn->c_transport_data; 572 int err = 0; 573 struct ib_qp_attr qp_attr; 574 575 rdsdebug("cm %p pd %p cq %p %p qp %p\n", ic->i_cm_id, 576 ic->i_pd, ic->i_send_cq, ic->i_recv_cq, 577 ic->i_cm_id ? ic->i_cm_id->qp : NULL); 578 579 if (ic->i_cm_id) { 580 struct ib_device *dev = ic->i_cm_id->device; 581 582 rdsdebug("disconnecting cm %p\n", ic->i_cm_id); 583 err = rdma_disconnect(ic->i_cm_id); 584 if (err) { 585 /* Actually this may happen quite frequently, when 586 * an outgoing connect raced with an incoming connect. 587 */ 588 rdsdebug("rds_iw_conn_shutdown: failed to disconnect," 589 " cm: %p err %d\n", ic->i_cm_id, err); 590 } 591 592 if (ic->i_cm_id->qp) { 593 qp_attr.qp_state = IB_QPS_ERR; 594 ib_modify_qp(ic->i_cm_id->qp, &qp_attr, IB_QP_STATE); 595 } 596 597 wait_event(rds_iw_ring_empty_wait, 598 rds_iw_ring_empty(&ic->i_send_ring) && 599 rds_iw_ring_empty(&ic->i_recv_ring)); 600 601 if (ic->i_send_hdrs) 602 ib_dma_free_coherent(dev, 603 ic->i_send_ring.w_nr * 604 sizeof(struct rds_header), 605 ic->i_send_hdrs, 606 ic->i_send_hdrs_dma); 607 608 if (ic->i_recv_hdrs) 609 ib_dma_free_coherent(dev, 610 ic->i_recv_ring.w_nr * 611 sizeof(struct rds_header), 612 ic->i_recv_hdrs, 613 ic->i_recv_hdrs_dma); 614 615 if (ic->i_ack) 616 ib_dma_free_coherent(dev, sizeof(struct rds_header), 617 ic->i_ack, ic->i_ack_dma); 618 619 if (ic->i_sends) 620 rds_iw_send_clear_ring(ic); 621 if (ic->i_recvs) 622 rds_iw_recv_clear_ring(ic); 623 624 if (ic->i_cm_id->qp) 625 rdma_destroy_qp(ic->i_cm_id); 626 if (ic->i_send_cq) 627 ib_destroy_cq(ic->i_send_cq); 628 if (ic->i_recv_cq) 629 ib_destroy_cq(ic->i_recv_cq); 630 631 /* 632 * If associated with an rds_iw_device: 633 * Move connection back to the nodev list. 634 * Remove cm_id from the device cm_id list. 635 */ 636 if (ic->rds_iwdev) 637 rds_iw_remove_conn(ic->rds_iwdev, conn); 638 639 rdma_destroy_id(ic->i_cm_id); 640 641 ic->i_cm_id = NULL; 642 ic->i_pd = NULL; 643 ic->i_mr = NULL; 644 ic->i_send_cq = NULL; 645 ic->i_recv_cq = NULL; 646 ic->i_send_hdrs = NULL; 647 ic->i_recv_hdrs = NULL; 648 ic->i_ack = NULL; 649 } 650 BUG_ON(ic->rds_iwdev); 651 652 /* Clear pending transmit */ 653 if (ic->i_rm) { 654 rds_message_put(ic->i_rm); 655 ic->i_rm = NULL; 656 } 657 658 /* Clear the ACK state */ 659 clear_bit(IB_ACK_IN_FLIGHT, &ic->i_ack_flags); 660#ifdef KERNEL_HAS_ATOMIC64 661 atomic64_set(&ic->i_ack_next, 0); 662#else 663 ic->i_ack_next = 0; 664#endif 665 ic->i_ack_recv = 0; 666 667 /* Clear flow control state */ 668 ic->i_flowctl = 0; 669 atomic_set(&ic->i_credits, 0); 670 671 rds_iw_ring_init(&ic->i_send_ring, rds_iw_sysctl_max_send_wr); 672 rds_iw_ring_init(&ic->i_recv_ring, rds_iw_sysctl_max_recv_wr); 673 674 if (ic->i_iwinc) { 675 rds_inc_put(&ic->i_iwinc->ii_inc); 676 ic->i_iwinc = NULL; 677 } 678 679 vfree(ic->i_sends); 680 ic->i_sends = NULL; 681 vfree(ic->i_recvs); 682 ic->i_recvs = NULL; 683 rdsdebug("shutdown complete\n"); 684} 685 686int rds_iw_conn_alloc(struct rds_connection *conn, gfp_t gfp) 687{ 688 struct rds_iw_connection *ic; 689 unsigned long flags; 690 691 ic = kzalloc(sizeof(struct rds_iw_connection), GFP_KERNEL); 692 if (ic == NULL) 693 return -ENOMEM; 694 695 INIT_LIST_HEAD(&ic->iw_node); 696 tasklet_init(&ic->i_recv_tasklet, rds_iw_recv_tasklet_fn, 697 (unsigned long) ic); 698 mutex_init(&ic->i_recv_mutex); 699#ifndef KERNEL_HAS_ATOMIC64 700 spin_lock_init(&ic->i_ack_lock); 701#endif 702 703 /* 704 * rds_iw_conn_shutdown() waits for these to be emptied so they 705 * must be initialized before it can be called. 706 */ 707 rds_iw_ring_init(&ic->i_send_ring, rds_iw_sysctl_max_send_wr); 708 rds_iw_ring_init(&ic->i_recv_ring, rds_iw_sysctl_max_recv_wr); 709 710 ic->conn = conn; 711 conn->c_transport_data = ic; 712 713 spin_lock_irqsave(&iw_nodev_conns_lock, flags); 714 list_add_tail(&ic->iw_node, &iw_nodev_conns); 715 spin_unlock_irqrestore(&iw_nodev_conns_lock, flags); 716 717 718 rdsdebug("conn %p conn ic %p\n", conn, conn->c_transport_data); 719 return 0; 720} 721 722/* 723 * Free a connection. Connection must be shut down and not set for reconnect. 724 */ 725void rds_iw_conn_free(void *arg) 726{ 727 struct rds_iw_connection *ic = arg; 728 spinlock_t *lock_ptr; 729 730 rdsdebug("ic %p\n", ic); 731 732 /* 733 * Conn is either on a dev's list or on the nodev list. 734 * A race with shutdown() or connect() would cause problems 735 * (since rds_iwdev would change) but that should never happen. 736 */ 737 lock_ptr = ic->rds_iwdev ? &ic->rds_iwdev->spinlock : &iw_nodev_conns_lock; 738 739 spin_lock_irq(lock_ptr); 740 list_del(&ic->iw_node); 741 spin_unlock_irq(lock_ptr); 742 743 kfree(ic); 744} 745 746/* 747 * An error occurred on the connection 748 */ 749void 750__rds_iw_conn_error(struct rds_connection *conn, const char *fmt, ...) 751{ 752 va_list ap; 753 754 rds_conn_drop(conn); 755 756 va_start(ap, fmt); 757 vprintk(fmt, ap); 758 va_end(ap); 759} 760