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 "ib.h" 40 41 /* 42 * Set the selected protocol version 43 */ 44 static void rds_ib_set_protocol(struct rds_connection *conn, unsigned int version) 45 { 46 conn->c_version = version; 47 } 48 49 /* 50 * Set up flow control 51 */ 52 static void rds_ib_set_flow_control(struct rds_connection *conn, u32 credits) 53 { 54 struct rds_ib_connection *ic = conn->c_transport_data; 55 56 if (rds_ib_sysctl_flow_control && credits != 0) { 57 /* We're doing flow control */ 58 ic->i_flowctl = 1; 59 rds_ib_send_add_credits(conn, credits); 60 } else { 61 ic->i_flowctl = 0; 62 } 63 } 64 65 /* 66 * Tune RNR behavior. Without flow control, we use a rather 67 * low timeout, but not the absolute minimum - this should 68 * be tunable. 69 * 70 * We already set the RNR retry count to 7 (which is the 71 * smallest infinite number :-) above. 72 * If flow control is off, we want to change this back to 0 73 * so that we learn quickly when our credit accounting is 74 * buggy. 75 * 76 * Caller passes in a qp_attr pointer - don't waste stack spacv 77 * by allocation this twice. 78 */ 79 static void 80 rds_ib_tune_rnr(struct rds_ib_connection *ic, struct ib_qp_attr *attr) 81 { 82 int ret; 83 84 attr->min_rnr_timer = IB_RNR_TIMER_000_32; 85 ret = ib_modify_qp(ic->i_cm_id->qp, attr, IB_QP_MIN_RNR_TIMER); 86 if (ret) 87 printk(KERN_NOTICE "ib_modify_qp(IB_QP_MIN_RNR_TIMER): err=%d\n", -ret); 88 } 89 90 /* 91 * Connection established. 92 * We get here for both outgoing and incoming connection. 93 */ 94 void rds_ib_cm_connect_complete(struct rds_connection *conn, struct rdma_cm_event *event) 95 { 96 const struct rds_ib_connect_private *dp = NULL; 97 struct rds_ib_connection *ic = conn->c_transport_data; 98 struct rds_ib_device *rds_ibdev; 99 struct ib_qp_attr qp_attr; 100 int err; 101 102 if (event->param.conn.private_data_len >= sizeof(*dp)) { 103 dp = event->param.conn.private_data; 104 105 /* make sure it isn't empty data */ 106 if (dp->dp_protocol_major) { 107 rds_ib_set_protocol(conn, 108 RDS_PROTOCOL(dp->dp_protocol_major, 109 dp->dp_protocol_minor)); 110 rds_ib_set_flow_control(conn, be32_to_cpu(dp->dp_credit)); 111 } 112 } 113 114 printk(KERN_NOTICE "RDS/IB: connected to %pI4 version %u.%u%s\n", 115 &conn->c_faddr, 116 RDS_PROTOCOL_MAJOR(conn->c_version), 117 RDS_PROTOCOL_MINOR(conn->c_version), 118 ic->i_flowctl ? ", flow control" : ""); 119 120 /* 121 * Init rings and fill recv. this needs to wait until protocol negotiation 122 * is complete, since ring layout is different from 3.0 to 3.1. 123 */ 124 rds_ib_send_init_ring(ic); 125 rds_ib_recv_init_ring(ic); 126 /* Post receive buffers - as a side effect, this will update 127 * the posted credit count. */ 128 rds_ib_recv_refill(conn, GFP_KERNEL, GFP_HIGHUSER, 1); 129 130 /* Tune RNR behavior */ 131 rds_ib_tune_rnr(ic, &qp_attr); 132 133 qp_attr.qp_state = IB_QPS_RTS; 134 err = ib_modify_qp(ic->i_cm_id->qp, &qp_attr, IB_QP_STATE); 135 if (err) 136 printk(KERN_NOTICE "ib_modify_qp(IB_QP_STATE, RTS): err=%d\n", err); 137 138 /* update ib_device with this local ipaddr & conn */ 139 rds_ibdev = ib_get_client_data(ic->i_cm_id->device, &rds_ib_client); 140 err = rds_ib_update_ipaddr(rds_ibdev, conn->c_laddr); 141 if (err) 142 printk(KERN_ERR "rds_ib_update_ipaddr failed (%d)\n", err); 143 rds_ib_add_conn(rds_ibdev, conn); 144 145 /* If the peer gave us the last packet it saw, process this as if 146 * we had received a regular ACK. */ 147 if (dp && dp->dp_ack_seq) 148 rds_send_drop_acked(conn, be64_to_cpu(dp->dp_ack_seq), NULL); 149 150 rds_connect_complete(conn); 151 } 152 153 static void rds_ib_cm_fill_conn_param(struct rds_connection *conn, 154 struct rdma_conn_param *conn_param, 155 struct rds_ib_connect_private *dp, 156 u32 protocol_version) 157 { 158 memset(conn_param, 0, sizeof(struct rdma_conn_param)); 159 /* XXX tune these? */ 160 conn_param->responder_resources = 1; 161 conn_param->initiator_depth = 1; 162 conn_param->retry_count = min_t(unsigned int, rds_ib_retry_count, 7); 163 conn_param->rnr_retry_count = 7; 164 165 if (dp) { 166 struct rds_ib_connection *ic = conn->c_transport_data; 167 168 memset(dp, 0, sizeof(*dp)); 169 dp->dp_saddr = conn->c_laddr; 170 dp->dp_daddr = conn->c_faddr; 171 dp->dp_protocol_major = RDS_PROTOCOL_MAJOR(protocol_version); 172 dp->dp_protocol_minor = RDS_PROTOCOL_MINOR(protocol_version); 173 dp->dp_protocol_minor_mask = cpu_to_be16(RDS_IB_SUPPORTED_PROTOCOLS); 174 dp->dp_ack_seq = rds_ib_piggyb_ack(ic); 175 176 /* Advertise flow control */ 177 if (ic->i_flowctl) { 178 unsigned int credits; 179 180 credits = IB_GET_POST_CREDITS(atomic_read(&ic->i_credits)); 181 dp->dp_credit = cpu_to_be32(credits); 182 atomic_sub(IB_SET_POST_CREDITS(credits), &ic->i_credits); 183 } 184 185 conn_param->private_data = dp; 186 conn_param->private_data_len = sizeof(*dp); 187 } 188 } 189 190 static void rds_ib_cq_event_handler(struct ib_event *event, void *data) 191 { 192 rdsdebug("event %u data %p\n", event->event, data); 193 } 194 195 static void rds_ib_qp_event_handler(struct ib_event *event, void *data) 196 { 197 struct rds_connection *conn = data; 198 struct rds_ib_connection *ic = conn->c_transport_data; 199 200 rdsdebug("conn %p ic %p event %u\n", conn, ic, event->event); 201 202 switch (event->event) { 203 case IB_EVENT_COMM_EST: 204 rdma_notify(ic->i_cm_id, IB_EVENT_COMM_EST); 205 break; 206 default: 207 rdsdebug("Fatal QP Event %u " 208 "- connection %pI4->%pI4, reconnecting\n", 209 event->event, &conn->c_laddr, &conn->c_faddr); 210 rds_conn_drop(conn); 211 break; 212 } 213 } 214 215 /* 216 * This needs to be very careful to not leave IS_ERR pointers around for 217 * cleanup to trip over. 218 */ 219 static int rds_ib_setup_qp(struct rds_connection *conn) 220 { 221 struct rds_ib_connection *ic = conn->c_transport_data; 222 struct ib_device *dev = ic->i_cm_id->device; 223 struct ib_qp_init_attr attr; 224 struct rds_ib_device *rds_ibdev; 225 int ret; 226 227 /* rds_ib_add_one creates a rds_ib_device object per IB device, 228 * and allocates a protection domain, memory range and FMR pool 229 * for each. If that fails for any reason, it will not register 230 * the rds_ibdev at all. 231 */ 232 rds_ibdev = ib_get_client_data(dev, &rds_ib_client); 233 if (rds_ibdev == NULL) { 234 if (printk_ratelimit()) 235 printk(KERN_NOTICE "RDS/IB: No client_data for device %s\n", 236 dev->name); 237 return -EOPNOTSUPP; 238 } 239 240 if (rds_ibdev->max_wrs < ic->i_send_ring.w_nr + 1) 241 rds_ib_ring_resize(&ic->i_send_ring, rds_ibdev->max_wrs - 1); 242 if (rds_ibdev->max_wrs < ic->i_recv_ring.w_nr + 1) 243 rds_ib_ring_resize(&ic->i_recv_ring, rds_ibdev->max_wrs - 1); 244 245 /* Protection domain and memory range */ 246 ic->i_pd = rds_ibdev->pd; 247 ic->i_mr = rds_ibdev->mr; 248 249 ic->i_send_cq = ib_create_cq(dev, rds_ib_send_cq_comp_handler, 250 rds_ib_cq_event_handler, conn, 251 ic->i_send_ring.w_nr + 1, 0); 252 if (IS_ERR(ic->i_send_cq)) { 253 ret = PTR_ERR(ic->i_send_cq); 254 ic->i_send_cq = NULL; 255 rdsdebug("ib_create_cq send failed: %d\n", ret); 256 goto out; 257 } 258 259 ic->i_recv_cq = ib_create_cq(dev, rds_ib_recv_cq_comp_handler, 260 rds_ib_cq_event_handler, conn, 261 ic->i_recv_ring.w_nr, 0); 262 if (IS_ERR(ic->i_recv_cq)) { 263 ret = PTR_ERR(ic->i_recv_cq); 264 ic->i_recv_cq = NULL; 265 rdsdebug("ib_create_cq recv failed: %d\n", ret); 266 goto out; 267 } 268 269 ret = ib_req_notify_cq(ic->i_send_cq, IB_CQ_NEXT_COMP); 270 if (ret) { 271 rdsdebug("ib_req_notify_cq send failed: %d\n", ret); 272 goto out; 273 } 274 275 ret = ib_req_notify_cq(ic->i_recv_cq, IB_CQ_SOLICITED); 276 if (ret) { 277 rdsdebug("ib_req_notify_cq recv failed: %d\n", ret); 278 goto out; 279 } 280 281 /* XXX negotiate max send/recv with remote? */ 282 memset(&attr, 0, sizeof(attr)); 283 attr.event_handler = rds_ib_qp_event_handler; 284 attr.qp_context = conn; 285 /* + 1 to allow for the single ack message */ 286 attr.cap.max_send_wr = ic->i_send_ring.w_nr + 1; 287 attr.cap.max_recv_wr = ic->i_recv_ring.w_nr + 1; 288 attr.cap.max_send_sge = rds_ibdev->max_sge; 289 attr.cap.max_recv_sge = RDS_IB_RECV_SGE; 290 attr.sq_sig_type = IB_SIGNAL_REQ_WR; 291 attr.qp_type = IB_QPT_RC; 292 attr.send_cq = ic->i_send_cq; 293 attr.recv_cq = ic->i_recv_cq; 294 295 /* 296 * XXX this can fail if max_*_wr is too large? Are we supposed 297 * to back off until we get a value that the hardware can support? 298 */ 299 ret = rdma_create_qp(ic->i_cm_id, ic->i_pd, &attr); 300 if (ret) { 301 rdsdebug("rdma_create_qp failed: %d\n", ret); 302 goto out; 303 } 304 305 ic->i_send_hdrs = ib_dma_alloc_coherent(dev, 306 ic->i_send_ring.w_nr * 307 sizeof(struct rds_header), 308 &ic->i_send_hdrs_dma, GFP_KERNEL); 309 if (ic->i_send_hdrs == NULL) { 310 ret = -ENOMEM; 311 rdsdebug("ib_dma_alloc_coherent send failed\n"); 312 goto out; 313 } 314 315 ic->i_recv_hdrs = ib_dma_alloc_coherent(dev, 316 ic->i_recv_ring.w_nr * 317 sizeof(struct rds_header), 318 &ic->i_recv_hdrs_dma, GFP_KERNEL); 319 if (ic->i_recv_hdrs == NULL) { 320 ret = -ENOMEM; 321 rdsdebug("ib_dma_alloc_coherent recv failed\n"); 322 goto out; 323 } 324 325 ic->i_ack = ib_dma_alloc_coherent(dev, sizeof(struct rds_header), 326 &ic->i_ack_dma, GFP_KERNEL); 327 if (ic->i_ack == NULL) { 328 ret = -ENOMEM; 329 rdsdebug("ib_dma_alloc_coherent ack failed\n"); 330 goto out; 331 } 332 333 ic->i_sends = vmalloc(ic->i_send_ring.w_nr * sizeof(struct rds_ib_send_work)); 334 if (ic->i_sends == NULL) { 335 ret = -ENOMEM; 336 rdsdebug("send allocation failed\n"); 337 goto out; 338 } 339 memset(ic->i_sends, 0, ic->i_send_ring.w_nr * sizeof(struct rds_ib_send_work)); 340 341 ic->i_recvs = vmalloc(ic->i_recv_ring.w_nr * sizeof(struct rds_ib_recv_work)); 342 if (ic->i_recvs == NULL) { 343 ret = -ENOMEM; 344 rdsdebug("recv allocation failed\n"); 345 goto out; 346 } 347 memset(ic->i_recvs, 0, ic->i_recv_ring.w_nr * sizeof(struct rds_ib_recv_work)); 348 349 rds_ib_recv_init_ack(ic); 350 351 rdsdebug("conn %p pd %p mr %p cq %p %p\n", conn, ic->i_pd, ic->i_mr, 352 ic->i_send_cq, ic->i_recv_cq); 353 354 out: 355 return ret; 356 } 357 358 static u32 rds_ib_protocol_compatible(struct rdma_cm_event *event) 359 { 360 const struct rds_ib_connect_private *dp = event->param.conn.private_data; 361 u16 common; 362 u32 version = 0; 363 364 /* 365 * rdma_cm private data is odd - when there is any private data in the 366 * request, we will be given a pretty large buffer without telling us the 367 * original size. The only way to tell the difference is by looking at 368 * the contents, which are initialized to zero. 369 * If the protocol version fields aren't set, this is a connection attempt 370 * from an older version. This could could be 3.0 or 2.0 - we can't tell. 371 * We really should have changed this for OFED 1.3 :-( 372 */ 373 374 /* Be paranoid. RDS always has privdata */ 375 if (!event->param.conn.private_data_len) { 376 printk(KERN_NOTICE "RDS incoming connection has no private data, " 377 "rejecting\n"); 378 return 0; 379 } 380 381 /* Even if len is crap *now* I still want to check it. -ASG */ 382 if (event->param.conn.private_data_len < sizeof (*dp) || 383 dp->dp_protocol_major == 0) 384 return RDS_PROTOCOL_3_0; 385 386 common = be16_to_cpu(dp->dp_protocol_minor_mask) & RDS_IB_SUPPORTED_PROTOCOLS; 387 if (dp->dp_protocol_major == 3 && common) { 388 version = RDS_PROTOCOL_3_0; 389 while ((common >>= 1) != 0) 390 version++; 391 } else if (printk_ratelimit()) { 392 printk(KERN_NOTICE "RDS: Connection from %pI4 using " 393 "incompatible protocol version %u.%u\n", 394 &dp->dp_saddr, 395 dp->dp_protocol_major, 396 dp->dp_protocol_minor); 397 } 398 return version; 399 } 400 401 int rds_ib_cm_handle_connect(struct rdma_cm_id *cm_id, 402 struct rdma_cm_event *event) 403 { 404 __be64 lguid = cm_id->route.path_rec->sgid.global.interface_id; 405 __be64 fguid = cm_id->route.path_rec->dgid.global.interface_id; 406 const struct rds_ib_connect_private *dp = event->param.conn.private_data; 407 struct rds_ib_connect_private dp_rep; 408 struct rds_connection *conn = NULL; 409 struct rds_ib_connection *ic = NULL; 410 struct rdma_conn_param conn_param; 411 u32 version; 412 int err, destroy = 1; 413 414 /* Check whether the remote protocol version matches ours. */ 415 version = rds_ib_protocol_compatible(event); 416 if (!version) 417 goto out; 418 419 rdsdebug("saddr %pI4 daddr %pI4 RDSv%u.%u lguid 0x%llx fguid " 420 "0x%llx\n", &dp->dp_saddr, &dp->dp_daddr, 421 RDS_PROTOCOL_MAJOR(version), RDS_PROTOCOL_MINOR(version), 422 (unsigned long long)be64_to_cpu(lguid), 423 (unsigned long long)be64_to_cpu(fguid)); 424 425 conn = rds_conn_create(dp->dp_daddr, dp->dp_saddr, &rds_ib_transport, 426 GFP_KERNEL); 427 if (IS_ERR(conn)) { 428 rdsdebug("rds_conn_create failed (%ld)\n", PTR_ERR(conn)); 429 conn = NULL; 430 goto out; 431 } 432 433 /* 434 * The connection request may occur while the 435 * previous connection exist, e.g. in case of failover. 436 * But as connections may be initiated simultaneously 437 * by both hosts, we have a random backoff mechanism - 438 * see the comment above rds_queue_reconnect() 439 */ 440 mutex_lock(&conn->c_cm_lock); 441 if (!rds_conn_transition(conn, RDS_CONN_DOWN, RDS_CONN_CONNECTING)) { 442 if (rds_conn_state(conn) == RDS_CONN_UP) { 443 rdsdebug("incoming connect while connecting\n"); 444 rds_conn_drop(conn); 445 rds_ib_stats_inc(s_ib_listen_closed_stale); 446 } else 447 if (rds_conn_state(conn) == RDS_CONN_CONNECTING) { 448 /* Wait and see - our connect may still be succeeding */ 449 rds_ib_stats_inc(s_ib_connect_raced); 450 } 451 mutex_unlock(&conn->c_cm_lock); 452 goto out; 453 } 454 455 ic = conn->c_transport_data; 456 457 rds_ib_set_protocol(conn, version); 458 rds_ib_set_flow_control(conn, be32_to_cpu(dp->dp_credit)); 459 460 /* If the peer gave us the last packet it saw, process this as if 461 * we had received a regular ACK. */ 462 if (dp->dp_ack_seq) 463 rds_send_drop_acked(conn, be64_to_cpu(dp->dp_ack_seq), NULL); 464 465 BUG_ON(cm_id->context); 466 BUG_ON(ic->i_cm_id); 467 468 ic->i_cm_id = cm_id; 469 cm_id->context = conn; 470 471 /* We got halfway through setting up the ib_connection, if we 472 * fail now, we have to take the long route out of this mess. */ 473 destroy = 0; 474 475 err = rds_ib_setup_qp(conn); 476 if (err) { 477 rds_ib_conn_error(conn, "rds_ib_setup_qp failed (%d)\n", err); 478 mutex_unlock(&conn->c_cm_lock); 479 goto out; 480 } 481 482 rds_ib_cm_fill_conn_param(conn, &conn_param, &dp_rep, version); 483 484 /* rdma_accept() calls rdma_reject() internally if it fails */ 485 err = rdma_accept(cm_id, &conn_param); 486 mutex_unlock(&conn->c_cm_lock); 487 if (err) { 488 rds_ib_conn_error(conn, "rdma_accept failed (%d)\n", err); 489 goto out; 490 } 491 492 return 0; 493 494 out: 495 rdma_reject(cm_id, NULL, 0); 496 return destroy; 497 } 498 499 500 int rds_ib_cm_initiate_connect(struct rdma_cm_id *cm_id) 501 { 502 struct rds_connection *conn = cm_id->context; 503 struct rds_ib_connection *ic = conn->c_transport_data; 504 struct rdma_conn_param conn_param; 505 struct rds_ib_connect_private dp; 506 int ret; 507 508 /* If the peer doesn't do protocol negotiation, we must 509 * default to RDSv3.0 */ 510 rds_ib_set_protocol(conn, RDS_PROTOCOL_3_0); 511 ic->i_flowctl = rds_ib_sysctl_flow_control; /* advertise flow control */ 512 513 ret = rds_ib_setup_qp(conn); 514 if (ret) { 515 rds_ib_conn_error(conn, "rds_ib_setup_qp failed (%d)\n", ret); 516 goto out; 517 } 518 519 rds_ib_cm_fill_conn_param(conn, &conn_param, &dp, RDS_PROTOCOL_VERSION); 520 521 ret = rdma_connect(cm_id, &conn_param); 522 if (ret) 523 rds_ib_conn_error(conn, "rdma_connect failed (%d)\n", ret); 524 525 out: 526 /* Beware - returning non-zero tells the rdma_cm to destroy 527 * the cm_id. We should certainly not do it as long as we still 528 * "own" the cm_id. */ 529 if (ret) { 530 if (ic->i_cm_id == cm_id) 531 ret = 0; 532 } 533 return ret; 534 } 535 536 int rds_ib_conn_connect(struct rds_connection *conn) 537 { 538 struct rds_ib_connection *ic = conn->c_transport_data; 539 struct sockaddr_in src, dest; 540 int ret; 541 542 /* XXX I wonder what affect the port space has */ 543 /* delegate cm event handler to rdma_transport */ 544 ic->i_cm_id = rdma_create_id(rds_rdma_cm_event_handler, conn, 545 RDMA_PS_TCP); 546 if (IS_ERR(ic->i_cm_id)) { 547 ret = PTR_ERR(ic->i_cm_id); 548 ic->i_cm_id = NULL; 549 rdsdebug("rdma_create_id() failed: %d\n", ret); 550 goto out; 551 } 552 553 rdsdebug("created cm id %p for conn %p\n", ic->i_cm_id, conn); 554 555 src.sin_family = AF_INET; 556 src.sin_addr.s_addr = (__force u32)conn->c_laddr; 557 src.sin_port = (__force u16)htons(0); 558 559 dest.sin_family = AF_INET; 560 dest.sin_addr.s_addr = (__force u32)conn->c_faddr; 561 dest.sin_port = (__force u16)htons(RDS_PORT); 562 563 ret = rdma_resolve_addr(ic->i_cm_id, (struct sockaddr *)&src, 564 (struct sockaddr *)&dest, 565 RDS_RDMA_RESOLVE_TIMEOUT_MS); 566 if (ret) { 567 rdsdebug("addr resolve failed for cm id %p: %d\n", ic->i_cm_id, 568 ret); 569 rdma_destroy_id(ic->i_cm_id); 570 ic->i_cm_id = NULL; 571 } 572 573 out: 574 return ret; 575 } 576 577 /* 578 * This is so careful about only cleaning up resources that were built up 579 * so that it can be called at any point during startup. In fact it 580 * can be called multiple times for a given connection. 581 */ 582 void rds_ib_conn_shutdown(struct rds_connection *conn) 583 { 584 struct rds_ib_connection *ic = conn->c_transport_data; 585 int err = 0; 586 587 rdsdebug("cm %p pd %p cq %p %p qp %p\n", ic->i_cm_id, 588 ic->i_pd, ic->i_send_cq, ic->i_recv_cq, 589 ic->i_cm_id ? ic->i_cm_id->qp : NULL); 590 591 if (ic->i_cm_id) { 592 struct ib_device *dev = ic->i_cm_id->device; 593 594 rdsdebug("disconnecting cm %p\n", ic->i_cm_id); 595 err = rdma_disconnect(ic->i_cm_id); 596 if (err) { 597 /* Actually this may happen quite frequently, when 598 * an outgoing connect raced with an incoming connect. 599 */ 600 rdsdebug("failed to disconnect, cm: %p err %d\n", 601 ic->i_cm_id, err); 602 } 603 604 wait_event(rds_ib_ring_empty_wait, 605 rds_ib_ring_empty(&ic->i_send_ring) && 606 rds_ib_ring_empty(&ic->i_recv_ring)); 607 608 if (ic->i_send_hdrs) 609 ib_dma_free_coherent(dev, 610 ic->i_send_ring.w_nr * 611 sizeof(struct rds_header), 612 ic->i_send_hdrs, 613 ic->i_send_hdrs_dma); 614 615 if (ic->i_recv_hdrs) 616 ib_dma_free_coherent(dev, 617 ic->i_recv_ring.w_nr * 618 sizeof(struct rds_header), 619 ic->i_recv_hdrs, 620 ic->i_recv_hdrs_dma); 621 622 if (ic->i_ack) 623 ib_dma_free_coherent(dev, sizeof(struct rds_header), 624 ic->i_ack, ic->i_ack_dma); 625 626 if (ic->i_sends) 627 rds_ib_send_clear_ring(ic); 628 if (ic->i_recvs) 629 rds_ib_recv_clear_ring(ic); 630 631 if (ic->i_cm_id->qp) 632 rdma_destroy_qp(ic->i_cm_id); 633 if (ic->i_send_cq) 634 ib_destroy_cq(ic->i_send_cq); 635 if (ic->i_recv_cq) 636 ib_destroy_cq(ic->i_recv_cq); 637 rdma_destroy_id(ic->i_cm_id); 638 639 /* 640 * Move connection back to the nodev list. 641 */ 642 if (ic->rds_ibdev) 643 rds_ib_remove_conn(ic->rds_ibdev, conn); 644 645 ic->i_cm_id = NULL; 646 ic->i_pd = NULL; 647 ic->i_mr = NULL; 648 ic->i_send_cq = NULL; 649 ic->i_recv_cq = NULL; 650 ic->i_send_hdrs = NULL; 651 ic->i_recv_hdrs = NULL; 652 ic->i_ack = NULL; 653 } 654 BUG_ON(ic->rds_ibdev); 655 656 /* Clear pending transmit */ 657 if (ic->i_rm) { 658 rds_message_put(ic->i_rm); 659 ic->i_rm = NULL; 660 } 661 662 /* Clear the ACK state */ 663 clear_bit(IB_ACK_IN_FLIGHT, &ic->i_ack_flags); 664 #ifdef KERNEL_HAS_ATOMIC64 665 atomic64_set(&ic->i_ack_next, 0); 666 #else 667 ic->i_ack_next = 0; 668 #endif 669 ic->i_ack_recv = 0; 670 671 /* Clear flow control state */ 672 ic->i_flowctl = 0; 673 atomic_set(&ic->i_credits, 0); 674 675 rds_ib_ring_init(&ic->i_send_ring, rds_ib_sysctl_max_send_wr); 676 rds_ib_ring_init(&ic->i_recv_ring, rds_ib_sysctl_max_recv_wr); 677 678 if (ic->i_ibinc) { 679 rds_inc_put(&ic->i_ibinc->ii_inc); 680 ic->i_ibinc = NULL; 681 } 682 683 vfree(ic->i_sends); 684 ic->i_sends = NULL; 685 vfree(ic->i_recvs); 686 ic->i_recvs = NULL; 687 } 688 689 int rds_ib_conn_alloc(struct rds_connection *conn, gfp_t gfp) 690 { 691 struct rds_ib_connection *ic; 692 unsigned long flags; 693 694 /* XXX too lazy? */ 695 ic = kzalloc(sizeof(struct rds_ib_connection), GFP_KERNEL); 696 if (ic == NULL) 697 return -ENOMEM; 698 699 INIT_LIST_HEAD(&ic->ib_node); 700 tasklet_init(&ic->i_recv_tasklet, rds_ib_recv_tasklet_fn, 701 (unsigned long) ic); 702 mutex_init(&ic->i_recv_mutex); 703 #ifndef KERNEL_HAS_ATOMIC64 704 spin_lock_init(&ic->i_ack_lock); 705 #endif 706 707 /* 708 * rds_ib_conn_shutdown() waits for these to be emptied so they 709 * must be initialized before it can be called. 710 */ 711 rds_ib_ring_init(&ic->i_send_ring, rds_ib_sysctl_max_send_wr); 712 rds_ib_ring_init(&ic->i_recv_ring, rds_ib_sysctl_max_recv_wr); 713 714 ic->conn = conn; 715 conn->c_transport_data = ic; 716 717 spin_lock_irqsave(&ib_nodev_conns_lock, flags); 718 list_add_tail(&ic->ib_node, &ib_nodev_conns); 719 spin_unlock_irqrestore(&ib_nodev_conns_lock, flags); 720 721 722 rdsdebug("conn %p conn ic %p\n", conn, conn->c_transport_data); 723 return 0; 724 } 725 726 /* 727 * Free a connection. Connection must be shut down and not set for reconnect. 728 */ 729 void rds_ib_conn_free(void *arg) 730 { 731 struct rds_ib_connection *ic = arg; 732 spinlock_t *lock_ptr; 733 734 rdsdebug("ic %p\n", ic); 735 736 /* 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_ibdev would change) but that should never happen. 740 */ 741 lock_ptr = ic->rds_ibdev ? &ic->rds_ibdev->spinlock : &ib_nodev_conns_lock; 742 743 spin_lock_irq(lock_ptr); 744 list_del(&ic->ib_node); 745 spin_unlock_irq(lock_ptr); 746 747 kfree(ic); 748 } 749 750 751 /* 752 * An error occurred on the connection 753 */ 754 void 755 __rds_ib_conn_error(struct rds_connection *conn, const char *fmt, ...) 756 { 757 va_list ap; 758 759 rds_conn_drop(conn); 760 761 va_start(ap, fmt); 762 vprintk(fmt, ap); 763 va_end(ap); 764 } 765