net/rds/ib_cm.c

   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 #include <linux/ratelimit.h>
  38
  39 #include "rds.h"
  40 #include "ib.h"
  41
  42 /*
  43  * Set the selected protocol version
  44  */
  45 static void rds_ib_set_protocol(struct rds_connection *conn, unsigned int version)
  46 {
  47         conn->c_version = version;
  48 }
  49
  50 /*
  51  * Set up flow control
  52  */
  53 static void rds_ib_set_flow_control(struct rds_connection *conn, u32 credits)
  54 {
  55         struct rds_ib_connection *ic = conn->c_transport_data;
  56
  57         if (rds_ib_sysctl_flow_control && credits != 0) {
  58                 /* We're doing flow control */
  59                 ic->i_flowctl = 1;
  60                 rds_ib_send_add_credits(conn, credits);
  61         } else {
  62                 ic->i_flowctl = 0;
  63         }
  64 }
  65
  66 /*
  67  * Tune RNR behavior. Without flow control, we use a rather
  68  * low timeout, but not the absolute minimum - this should
  69  * be tunable.
  70  *
  71  * We already set the RNR retry count to 7 (which is the
  72  * smallest infinite number :-) above.
  73  * If flow control is off, we want to change this back to 0
  74  * so that we learn quickly when our credit accounting is
  75  * buggy.
  76  *
  77  * Caller passes in a qp_attr pointer - don't waste stack spacv
  78  * by allocation this twice.
  79  */
  80 static void
  81 rds_ib_tune_rnr(struct rds_ib_connection *ic, struct ib_qp_attr *attr)
  82 {
  83         int ret;
  84
  85         attr->min_rnr_timer = IB_RNR_TIMER_000_32;
  86         ret = ib_modify_qp(ic->i_cm_id->qp, attr, IB_QP_MIN_RNR_TIMER);
  87         if (ret)
  88                 printk(KERN_NOTICE "ib_modify_qp(IB_QP_MIN_RNR_TIMER): err=%d\n", -ret);
  89 }
  90
  91 /*
  92  * Connection established.
  93  * We get here for both outgoing and incoming connection.
  94  */
  95 void rds_ib_cm_connect_complete(struct rds_connection *conn, struct rdma_cm_event *event)
  96 {
  97         const struct rds_ib_connect_private *dp = NULL;
  98         struct rds_ib_connection *ic = conn->c_transport_data;
  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         if (conn->c_version < RDS_PROTOCOL(3,1)) {
 115                 printk(KERN_NOTICE "RDS/IB: Connection to %pI4 version %u.%u failed,"
 116                        " no longer supported\n",
 117                        &conn->c_faddr,
 118                        RDS_PROTOCOL_MAJOR(conn->c_version),
 119                        RDS_PROTOCOL_MINOR(conn->c_version));
 120                 rds_conn_destroy(conn);
 121                 return;
 122         } else {
 123                 printk(KERN_NOTICE "RDS/IB: connected to %pI4 version %u.%u%s\n",
 124                        &conn->c_faddr,
 125                        RDS_PROTOCOL_MAJOR(conn->c_version),
 126                        RDS_PROTOCOL_MINOR(conn->c_version),
 127                        ic->i_flowctl ? ", flow control" : "");
 128         }
 129
 130         /*
 131          * Init rings and fill recv. this needs to wait until protocol negotiation
 132          * is complete, since ring layout is different from 3.0 to 3.1.
 133          */
 134         rds_ib_send_init_ring(ic);
 135         rds_ib_recv_init_ring(ic);
 136         /* Post receive buffers - as a side effect, this will update
 137          * the posted credit count. */
 138         rds_ib_recv_refill(conn, 1, GFP_KERNEL);
 139
 140         /* Tune RNR behavior */
 141         rds_ib_tune_rnr(ic, &qp_attr);
 142
 143         qp_attr.qp_state = IB_QPS_RTS;
 144         err = ib_modify_qp(ic->i_cm_id->qp, &qp_attr, IB_QP_STATE);
 145         if (err)
 146                 printk(KERN_NOTICE "ib_modify_qp(IB_QP_STATE, RTS): err=%d\n", err);
 147
 148         /* update ib_device with this local ipaddr */
 149         err = rds_ib_update_ipaddr(ic->rds_ibdev, conn->c_laddr);
 150         if (err)
 151                 printk(KERN_ERR "rds_ib_update_ipaddr failed (%d)\n",
 152                         err);
 153
 154         /* If the peer gave us the last packet it saw, process this as if
 155          * we had received a regular ACK. */
 156         if (dp) {
 157                 /* dp structure start is not guaranteed to be 8 bytes aligned.
 158                  * Since dp_ack_seq is 64-bit extended load operations can be
 159                  * used so go through get_unaligned to avoid unaligned errors.
 160                  */
 161                 __be64 dp_ack_seq = get_unaligned(&dp->dp_ack_seq);
 162
 163                 if (dp_ack_seq)
 164                         rds_send_drop_acked(conn, be64_to_cpu(dp_ack_seq),
 165                                             NULL);
 166         }
 167
 168         rds_connect_complete(conn);
 169 }
 170
 171 static void rds_ib_cm_fill_conn_param(struct rds_connection *conn,
 172                         struct rdma_conn_param *conn_param,
 173                         struct rds_ib_connect_private *dp,
 174                         u32 protocol_version,
 175                         u32 max_responder_resources,
 176                         u32 max_initiator_depth)
 177 {
 178         struct rds_ib_connection *ic = conn->c_transport_data;
 179         struct rds_ib_device *rds_ibdev = ic->rds_ibdev;
 180
 181         memset(conn_param, 0, sizeof(struct rdma_conn_param));
 182
 183         conn_param->responder_resources =
 184                 min_t(u32, rds_ibdev->max_responder_resources, max_responder_resources);
 185         conn_param->initiator_depth =
 186                 min_t(u32, rds_ibdev->max_initiator_depth, max_initiator_depth);
 187         conn_param->retry_count = min_t(unsigned int, rds_ib_retry_count, 7);
 188         conn_param->rnr_retry_count = 7;
 189
 190         if (dp) {
 191                 memset(dp, 0, sizeof(*dp));
 192                 dp->dp_saddr = conn->c_laddr;
 193                 dp->dp_daddr = conn->c_faddr;
 194                 dp->dp_protocol_major = RDS_PROTOCOL_MAJOR(protocol_version);
 195                 dp->dp_protocol_minor = RDS_PROTOCOL_MINOR(protocol_version);
 196                 dp->dp_protocol_minor_mask = cpu_to_be16(RDS_IB_SUPPORTED_PROTOCOLS);
 197                 dp->dp_ack_seq = cpu_to_be64(rds_ib_piggyb_ack(ic));
 198
 199                 /* Advertise flow control */
 200                 if (ic->i_flowctl) {
 201                         unsigned int credits;
 202
 203                         credits = IB_GET_POST_CREDITS(atomic_read(&ic->i_credits));
 204                         dp->dp_credit = cpu_to_be32(credits);
 205                         atomic_sub(IB_SET_POST_CREDITS(credits), &ic->i_credits);
 206                 }
 207
 208                 conn_param->private_data = dp;
 209                 conn_param->private_data_len = sizeof(*dp);
 210         }
 211 }
 212
 213 static void rds_ib_cq_event_handler(struct ib_event *event, void *data)
 214 {
 215         rdsdebug("event %u (%s) data %p\n",
 216                  event->event, ib_event_msg(event->event), data);
 217 }
 218
 219 /* Plucking the oldest entry from the ring can be done concurrently with
 220  * the thread refilling the ring.  Each ring operation is protected by
 221  * spinlocks and the transient state of refilling doesn't change the
 222  * recording of which entry is oldest.
 223  *
 224  * This relies on IB only calling one cq comp_handler for each cq so that
 225  * there will only be one caller of rds_recv_incoming() per RDS connection.
 226  */
 227 static void rds_ib_cq_comp_handler_recv(struct ib_cq *cq, void *context)
 228 {
 229         struct rds_connection *conn = context;
 230         struct rds_ib_connection *ic = conn->c_transport_data;
 231
 232         rdsdebug("conn %p cq %p\n", conn, cq);
 233
 234         rds_ib_stats_inc(s_ib_evt_handler_call);
 235
 236         tasklet_schedule(&ic->i_recv_tasklet);
 237 }
 238
 239 static void poll_scq(struct rds_ib_connection *ic, struct ib_cq *cq,
 240                      struct ib_wc *wcs)
 241 {
 242         int nr, i;
 243         struct ib_wc *wc;
 244
 245         while ((nr = ib_poll_cq(cq, RDS_IB_WC_MAX, wcs)) > 0) {
 246                 for (i = 0; i < nr; i++) {
 247                         wc = wcs + i;
 248                         rdsdebug("wc wr_id 0x%llx status %u byte_len %u imm_data %u\n",
 249                                  (unsigned long long)wc->wr_id, wc->status,
 250                                  wc->byte_len, be32_to_cpu(wc->ex.imm_data));
 251
 252                         if (wc->wr_id <= ic->i_send_ring.w_nr ||
 253                             wc->wr_id == RDS_IB_ACK_WR_ID)
 254                                 rds_ib_send_cqe_handler(ic, wc);
 255                         else
 256                                 rds_ib_mr_cqe_handler(ic, wc);
 257
 258                 }
 259         }
 260 }
 261
 262 static void rds_ib_tasklet_fn_send(unsigned long data)
 263 {
 264         struct rds_ib_connection *ic = (struct rds_ib_connection *)data;
 265         struct rds_connection *conn = ic->conn;
 266
 267         rds_ib_stats_inc(s_ib_tasklet_call);
 268
 269         poll_scq(ic, ic->i_send_cq, ic->i_send_wc);
 270         ib_req_notify_cq(ic->i_send_cq, IB_CQ_NEXT_COMP);
 271         poll_scq(ic, ic->i_send_cq, ic->i_send_wc);
 272
 273         if (rds_conn_up(conn) &&
 274             (!test_bit(RDS_LL_SEND_FULL, &conn->c_flags) ||
 275             test_bit(0, &conn->c_map_queued)))
 276                 rds_send_xmit(ic->conn);
 277 }
 278
 279 static void poll_rcq(struct rds_ib_connection *ic, struct ib_cq *cq,
 280                      struct ib_wc *wcs,
 281                      struct rds_ib_ack_state *ack_state)
 282 {
 283         int nr, i;
 284         struct ib_wc *wc;
 285
 286         while ((nr = ib_poll_cq(cq, RDS_IB_WC_MAX, wcs)) > 0) {
 287                 for (i = 0; i < nr; i++) {
 288                         wc = wcs + i;
 289                         rdsdebug("wc wr_id 0x%llx status %u byte_len %u imm_data %u\n",
 290                                  (unsigned long long)wc->wr_id, wc->status,
 291                                  wc->byte_len, be32_to_cpu(wc->ex.imm_data));
 292
 293                         rds_ib_recv_cqe_handler(ic, wc, ack_state);
 294                 }
 295         }
 296 }
 297
 298 static void rds_ib_tasklet_fn_recv(unsigned long data)
 299 {
 300         struct rds_ib_connection *ic = (struct rds_ib_connection *)data;
 301         struct rds_connection *conn = ic->conn;
 302         struct rds_ib_device *rds_ibdev = ic->rds_ibdev;
 303         struct rds_ib_ack_state state;
 304
 305         if (!rds_ibdev)
 306                 rds_conn_drop(conn);
 307
 308         rds_ib_stats_inc(s_ib_tasklet_call);
 309
 310         memset(&state, 0, sizeof(state));
 311         poll_rcq(ic, ic->i_recv_cq, ic->i_recv_wc, &state);
 312         ib_req_notify_cq(ic->i_recv_cq, IB_CQ_SOLICITED);
 313         poll_rcq(ic, ic->i_recv_cq, ic->i_recv_wc, &state);
 314
 315         if (state.ack_next_valid)
 316                 rds_ib_set_ack(ic, state.ack_next, state.ack_required);
 317         if (state.ack_recv_valid && state.ack_recv > ic->i_ack_recv) {
 318                 rds_send_drop_acked(conn, state.ack_recv, NULL);
 319                 ic->i_ack_recv = state.ack_recv;
 320         }
 321
 322         if (rds_conn_up(conn))
 323                 rds_ib_attempt_ack(ic);
 324 }
 325
 326 static void rds_ib_qp_event_handler(struct ib_event *event, void *data)
 327 {
 328         struct rds_connection *conn = data;
 329         struct rds_ib_connection *ic = conn->c_transport_data;
 330
 331         rdsdebug("conn %p ic %p event %u (%s)\n", conn, ic, event->event,
 332                  ib_event_msg(event->event));
 333
 334         switch (event->event) {
 335         case IB_EVENT_COMM_EST:
 336                 rdma_notify(ic->i_cm_id, IB_EVENT_COMM_EST);
 337                 break;
 338         default:
 339                 rdsdebug("Fatal QP Event %u (%s) "
 340                         "- connection %pI4->%pI4, reconnecting\n",
 341                         event->event, ib_event_msg(event->event),
 342                         &conn->c_laddr, &conn->c_faddr);
 343                 rds_conn_drop(conn);
 344                 break;
 345         }
 346 }
 347
 348 static void rds_ib_cq_comp_handler_send(struct ib_cq *cq, void *context)
 349 {
 350         struct rds_connection *conn = context;
 351         struct rds_ib_connection *ic = conn->c_transport_data;
 352
 353         rdsdebug("conn %p cq %p\n", conn, cq);
 354
 355         rds_ib_stats_inc(s_ib_evt_handler_call);
 356
 357         tasklet_schedule(&ic->i_send_tasklet);
 358 }
 359
 360 /*
 361  * This needs to be very careful to not leave IS_ERR pointers around for
 362  * cleanup to trip over.
 363  */
 364 static int rds_ib_setup_qp(struct rds_connection *conn)
 365 {
 366         struct rds_ib_connection *ic = conn->c_transport_data;
 367         struct ib_device *dev = ic->i_cm_id->device;
 368         struct ib_qp_init_attr attr;
 369         struct ib_cq_init_attr cq_attr = {};
 370         struct rds_ib_device *rds_ibdev;
 371         int ret, fr_queue_space;
 372
 373         /*
 374          * It's normal to see a null device if an incoming connection races
 375          * with device removal, so we don't print a warning.
 376          */
 377         rds_ibdev = rds_ib_get_client_data(dev);
 378         if (!rds_ibdev)
 379                 return -EOPNOTSUPP;
 380
 381         /* The fr_queue_space is currently set to 512, to add extra space on
 382          * completion queue and send queue. This extra space is used for FRMR
 383          * registration and invalidation work requests
 384          */
 385         fr_queue_space = (rds_ibdev->use_fastreg ? RDS_IB_DEFAULT_FR_WR : 0);
 386
 387         /* add the conn now so that connection establishment has the dev */
 388         rds_ib_add_conn(rds_ibdev, conn);
 389
 390         if (rds_ibdev->max_wrs < ic->i_send_ring.w_nr + 1)
 391                 rds_ib_ring_resize(&ic->i_send_ring, rds_ibdev->max_wrs - 1);
 392         if (rds_ibdev->max_wrs < ic->i_recv_ring.w_nr + 1)
 393                 rds_ib_ring_resize(&ic->i_recv_ring, rds_ibdev->max_wrs - 1);
 394
 395         /* Protection domain and memory range */
 396         ic->i_pd = rds_ibdev->pd;
 397
 398         cq_attr.cqe = ic->i_send_ring.w_nr + fr_queue_space + 1;
 399
 400         ic->i_send_cq = ib_create_cq(dev, rds_ib_cq_comp_handler_send,
 401                                      rds_ib_cq_event_handler, conn,
 402                                      &cq_attr);
 403         if (IS_ERR(ic->i_send_cq)) {
 404                 ret = PTR_ERR(ic->i_send_cq);
 405                 ic->i_send_cq = NULL;
 406                 rdsdebug("ib_create_cq send failed: %d\n", ret);
 407                 goto out;
 408         }
 409
 410         cq_attr.cqe = ic->i_recv_ring.w_nr;
 411         ic->i_recv_cq = ib_create_cq(dev, rds_ib_cq_comp_handler_recv,
 412                                      rds_ib_cq_event_handler, conn,
 413                                      &cq_attr);
 414         if (IS_ERR(ic->i_recv_cq)) {
 415                 ret = PTR_ERR(ic->i_recv_cq);
 416                 ic->i_recv_cq = NULL;
 417                 rdsdebug("ib_create_cq recv failed: %d\n", ret);
 418                 goto out;
 419         }
 420
 421         ret = ib_req_notify_cq(ic->i_send_cq, IB_CQ_NEXT_COMP);
 422         if (ret) {
 423                 rdsdebug("ib_req_notify_cq send failed: %d\n", ret);
 424                 goto out;
 425         }
 426
 427         ret = ib_req_notify_cq(ic->i_recv_cq, IB_CQ_SOLICITED);
 428         if (ret) {
 429                 rdsdebug("ib_req_notify_cq recv failed: %d\n", ret);
 430                 goto out;
 431         }
 432
 433         /* XXX negotiate max send/recv with remote? */
 434         memset(&attr, 0, sizeof(attr));
 435         attr.event_handler = rds_ib_qp_event_handler;
 436         attr.qp_context = conn;
 437         /* + 1 to allow for the single ack message */
 438         attr.cap.max_send_wr = ic->i_send_ring.w_nr + fr_queue_space + 1;
 439         attr.cap.max_recv_wr = ic->i_recv_ring.w_nr + 1;
 440         attr.cap.max_send_sge = rds_ibdev->max_sge;
 441         attr.cap.max_recv_sge = RDS_IB_RECV_SGE;
 442         attr.sq_sig_type = IB_SIGNAL_REQ_WR;
 443         attr.qp_type = IB_QPT_RC;
 444         attr.send_cq = ic->i_send_cq;
 445         attr.recv_cq = ic->i_recv_cq;
 446         atomic_set(&ic->i_fastreg_wrs, RDS_IB_DEFAULT_FR_WR);
 447
 448         /*
 449          * XXX this can fail if max_*_wr is too large?  Are we supposed
 450          * to back off until we get a value that the hardware can support?
 451          */
 452         ret = rdma_create_qp(ic->i_cm_id, ic->i_pd, &attr);
 453         if (ret) {
 454                 rdsdebug("rdma_create_qp failed: %d\n", ret);
 455                 goto out;
 456         }
 457
 458         ic->i_send_hdrs = ib_dma_alloc_coherent(dev,
 459                                            ic->i_send_ring.w_nr *
 460                                                 sizeof(struct rds_header),
 461                                            &ic->i_send_hdrs_dma, GFP_KERNEL);
 462         if (!ic->i_send_hdrs) {
 463                 ret = -ENOMEM;
 464                 rdsdebug("ib_dma_alloc_coherent send failed\n");
 465                 goto out;
 466         }
 467
 468         ic->i_recv_hdrs = ib_dma_alloc_coherent(dev,
 469                                            ic->i_recv_ring.w_nr *
 470                                                 sizeof(struct rds_header),
 471                                            &ic->i_recv_hdrs_dma, GFP_KERNEL);
 472         if (!ic->i_recv_hdrs) {
 473                 ret = -ENOMEM;
 474                 rdsdebug("ib_dma_alloc_coherent recv failed\n");
 475                 goto out;
 476         }
 477
 478         ic->i_ack = ib_dma_alloc_coherent(dev, sizeof(struct rds_header),
 479                                        &ic->i_ack_dma, GFP_KERNEL);
 480         if (!ic->i_ack) {
 481                 ret = -ENOMEM;
 482                 rdsdebug("ib_dma_alloc_coherent ack failed\n");
 483                 goto out;
 484         }
 485
 486         ic->i_sends = vzalloc_node(ic->i_send_ring.w_nr * sizeof(struct rds_ib_send_work),
 487                                    ibdev_to_node(dev));
 488         if (!ic->i_sends) {
 489                 ret = -ENOMEM;
 490                 rdsdebug("send allocation failed\n");
 491                 goto out;
 492         }
 493
 494         ic->i_recvs = vzalloc_node(ic->i_recv_ring.w_nr * sizeof(struct rds_ib_recv_work),
 495                                    ibdev_to_node(dev));
 496         if (!ic->i_recvs) {
 497                 ret = -ENOMEM;
 498                 rdsdebug("recv allocation failed\n");
 499                 goto out;
 500         }
 501
 502         rds_ib_recv_init_ack(ic);
 503
 504         rdsdebug("conn %p pd %p cq %p %p\n", conn, ic->i_pd,
 505                  ic->i_send_cq, ic->i_recv_cq);
 506
 507 out:
 508         rds_ib_dev_put(rds_ibdev);
 509         return ret;
 510 }
 511
 512 static u32 rds_ib_protocol_compatible(struct rdma_cm_event *event)
 513 {
 514         const struct rds_ib_connect_private *dp = event->param.conn.private_data;
 515         u16 common;
 516         u32 version = 0;
 517
 518         /*
 519          * rdma_cm private data is odd - when there is any private data in the
 520          * request, we will be given a pretty large buffer without telling us the
 521          * original size. The only way to tell the difference is by looking at
 522          * the contents, which are initialized to zero.
 523          * If the protocol version fields aren't set, this is a connection attempt
 524          * from an older version. This could could be 3.0 or 2.0 - we can't tell.
 525          * We really should have changed this for OFED 1.3 :-(
 526          */
 527
 528         /* Be paranoid. RDS always has privdata */
 529         if (!event->param.conn.private_data_len) {
 530                 printk(KERN_NOTICE "RDS incoming connection has no private data, "
 531                         "rejecting\n");
 532                 return 0;
 533         }
 534
 535         /* Even if len is crap *now* I still want to check it. -ASG */
 536         if (event->param.conn.private_data_len < sizeof (*dp) ||
 537             dp->dp_protocol_major == 0)
 538                 return RDS_PROTOCOL_3_0;
 539
 540         common = be16_to_cpu(dp->dp_protocol_minor_mask) & RDS_IB_SUPPORTED_PROTOCOLS;
 541         if (dp->dp_protocol_major == 3 && common) {
 542                 version = RDS_PROTOCOL_3_0;
 543                 while ((common >>= 1) != 0)
 544                         version++;
 545         } else
 546                 printk_ratelimited(KERN_NOTICE "RDS: Connection from %pI4 using incompatible protocol version %u.%u\n",
 547                                 &dp->dp_saddr,
 548                                 dp->dp_protocol_major,
 549                                 dp->dp_protocol_minor);
 550         return version;
 551 }
 552
 553 int rds_ib_cm_handle_connect(struct rdma_cm_id *cm_id,
 554                                     struct rdma_cm_event *event)
 555 {
 556         __be64 lguid = cm_id->route.path_rec->sgid.global.interface_id;
 557         __be64 fguid = cm_id->route.path_rec->dgid.global.interface_id;
 558         const struct rds_ib_connect_private *dp = event->param.conn.private_data;
 559         struct rds_ib_connect_private dp_rep;
 560         struct rds_connection *conn = NULL;
 561         struct rds_ib_connection *ic = NULL;
 562         struct rdma_conn_param conn_param;
 563         u32 version;
 564         int err = 1, destroy = 1;
 565
 566         /* Check whether the remote protocol version matches ours. */
 567         version = rds_ib_protocol_compatible(event);
 568         if (!version)
 569                 goto out;
 570
 571         rdsdebug("saddr %pI4 daddr %pI4 RDSv%u.%u lguid 0x%llx fguid "
 572                  "0x%llx\n", &dp->dp_saddr, &dp->dp_daddr,
 573                  RDS_PROTOCOL_MAJOR(version), RDS_PROTOCOL_MINOR(version),
 574                  (unsigned long long)be64_to_cpu(lguid),
 575                  (unsigned long long)be64_to_cpu(fguid));
 576
 577         /* RDS/IB is not currently netns aware, thus init_net */
 578         conn = rds_conn_create(&init_net, dp->dp_daddr, dp->dp_saddr,
 579                                &rds_ib_transport, GFP_KERNEL);
 580         if (IS_ERR(conn)) {
 581                 rdsdebug("rds_conn_create failed (%ld)\n", PTR_ERR(conn));
 582                 conn = NULL;
 583                 goto out;
 584         }
 585
 586         /*
 587          * The connection request may occur while the
 588          * previous connection exist, e.g. in case of failover.
 589          * But as connections may be initiated simultaneously
 590          * by both hosts, we have a random backoff mechanism -
 591          * see the comment above rds_queue_reconnect()
 592          */
 593         mutex_lock(&conn->c_cm_lock);
 594         if (!rds_conn_transition(conn, RDS_CONN_DOWN, RDS_CONN_CONNECTING)) {
 595                 if (rds_conn_state(conn) == RDS_CONN_UP) {
 596                         rdsdebug("incoming connect while connecting\n");
 597                         rds_conn_drop(conn);
 598                         rds_ib_stats_inc(s_ib_listen_closed_stale);
 599                 } else
 600                 if (rds_conn_state(conn) == RDS_CONN_CONNECTING) {
 601                         /* Wait and see - our connect may still be succeeding */
 602                         rds_ib_stats_inc(s_ib_connect_raced);
 603                 }
 604                 goto out;
 605         }
 606
 607         ic = conn->c_transport_data;
 608
 609         rds_ib_set_protocol(conn, version);
 610         rds_ib_set_flow_control(conn, be32_to_cpu(dp->dp_credit));
 611
 612         /* If the peer gave us the last packet it saw, process this as if
 613          * we had received a regular ACK. */
 614         if (dp->dp_ack_seq)
 615                 rds_send_drop_acked(conn, be64_to_cpu(dp->dp_ack_seq), NULL);
 616
 617         BUG_ON(cm_id->context);
 618         BUG_ON(ic->i_cm_id);
 619
 620         ic->i_cm_id = cm_id;
 621         cm_id->context = conn;
 622
 623         /* We got halfway through setting up the ib_connection, if we
 624          * fail now, we have to take the long route out of this mess. */
 625         destroy = 0;
 626
 627         err = rds_ib_setup_qp(conn);
 628         if (err) {
 629                 rds_ib_conn_error(conn, "rds_ib_setup_qp failed (%d)\n", err);
 630                 goto out;
 631         }
 632
 633         rds_ib_cm_fill_conn_param(conn, &conn_param, &dp_rep, version,
 634                 event->param.conn.responder_resources,
 635                 event->param.conn.initiator_depth);
 636
 637         /* rdma_accept() calls rdma_reject() internally if it fails */
 638         err = rdma_accept(cm_id, &conn_param);
 639         if (err)
 640                 rds_ib_conn_error(conn, "rdma_accept failed (%d)\n", err);
 641
 642 out:
 643         if (conn)
 644                 mutex_unlock(&conn->c_cm_lock);
 645         if (err)
 646                 rdma_reject(cm_id, NULL, 0);
 647         return destroy;
 648 }
 649
 650
 651 int rds_ib_cm_initiate_connect(struct rdma_cm_id *cm_id)
 652 {
 653         struct rds_connection *conn = cm_id->context;
 654         struct rds_ib_connection *ic = conn->c_transport_data;
 655         struct rdma_conn_param conn_param;
 656         struct rds_ib_connect_private dp;
 657         int ret;
 658
 659         /* If the peer doesn't do protocol negotiation, we must
 660          * default to RDSv3.0 */
 661         rds_ib_set_protocol(conn, RDS_PROTOCOL_3_0);
 662         ic->i_flowctl = rds_ib_sysctl_flow_control;     /* advertise flow control */
 663
 664         ret = rds_ib_setup_qp(conn);
 665         if (ret) {
 666                 rds_ib_conn_error(conn, "rds_ib_setup_qp failed (%d)\n", ret);
 667                 goto out;
 668         }
 669
 670         rds_ib_cm_fill_conn_param(conn, &conn_param, &dp, RDS_PROTOCOL_VERSION,
 671                 UINT_MAX, UINT_MAX);
 672         ret = rdma_connect(cm_id, &conn_param);
 673         if (ret)
 674                 rds_ib_conn_error(conn, "rdma_connect failed (%d)\n", ret);
 675
 676 out:
 677         /* Beware - returning non-zero tells the rdma_cm to destroy
 678          * the cm_id. We should certainly not do it as long as we still
 679          * "own" the cm_id. */
 680         if (ret) {
 681                 if (ic->i_cm_id == cm_id)
 682                         ret = 0;
 683         }
 684         return ret;
 685 }
 686
 687 int rds_ib_conn_connect(struct rds_connection *conn)
 688 {
 689         struct rds_ib_connection *ic = conn->c_transport_data;
 690         struct sockaddr_in src, dest;
 691         int ret;
 692
 693         /* XXX I wonder what affect the port space has */
 694         /* delegate cm event handler to rdma_transport */
 695         ic->i_cm_id = rdma_create_id(&init_net, rds_rdma_cm_event_handler, conn,
 696                                      RDMA_PS_TCP, IB_QPT_RC);
 697         if (IS_ERR(ic->i_cm_id)) {
 698                 ret = PTR_ERR(ic->i_cm_id);
 699                 ic->i_cm_id = NULL;
 700                 rdsdebug("rdma_create_id() failed: %d\n", ret);
 701                 goto out;
 702         }
 703
 704         rdsdebug("created cm id %p for conn %p\n", ic->i_cm_id, conn);
 705
 706         src.sin_family = AF_INET;
 707         src.sin_addr.s_addr = (__force u32)conn->c_laddr;
 708         src.sin_port = (__force u16)htons(0);
 709
 710         dest.sin_family = AF_INET;
 711         dest.sin_addr.s_addr = (__force u32)conn->c_faddr;
 712         dest.sin_port = (__force u16)htons(RDS_PORT);
 713
 714         ret = rdma_resolve_addr(ic->i_cm_id, (struct sockaddr *)&src,
 715                                 (struct sockaddr *)&dest,
 716                                 RDS_RDMA_RESOLVE_TIMEOUT_MS);
 717         if (ret) {
 718                 rdsdebug("addr resolve failed for cm id %p: %d\n", ic->i_cm_id,
 719                          ret);
 720                 rdma_destroy_id(ic->i_cm_id);
 721                 ic->i_cm_id = NULL;
 722         }
 723
 724 out:
 725         return ret;
 726 }
 727
 728 /*
 729  * This is so careful about only cleaning up resources that were built up
 730  * so that it can be called at any point during startup.  In fact it
 731  * can be called multiple times for a given connection.
 732  */
 733 void rds_ib_conn_shutdown(struct rds_connection *conn)
 734 {
 735         struct rds_ib_connection *ic = conn->c_transport_data;
 736         int err = 0;
 737
 738         rdsdebug("cm %p pd %p cq %p %p qp %p\n", ic->i_cm_id,
 739                  ic->i_pd, ic->i_send_cq, ic->i_recv_cq,
 740                  ic->i_cm_id ? ic->i_cm_id->qp : NULL);
 741
 742         if (ic->i_cm_id) {
 743                 struct ib_device *dev = ic->i_cm_id->device;
 744
 745                 rdsdebug("disconnecting cm %p\n", ic->i_cm_id);
 746                 err = rdma_disconnect(ic->i_cm_id);
 747                 if (err) {
 748                         /* Actually this may happen quite frequently, when
 749                          * an outgoing connect raced with an incoming connect.
 750                          */
 751                         rdsdebug("failed to disconnect, cm: %p err %d\n",
 752                                 ic->i_cm_id, err);
 753                 }
 754
 755                 /*
 756                  * We want to wait for tx and rx completion to finish
 757                  * before we tear down the connection, but we have to be
 758                  * careful not to get stuck waiting on a send ring that
 759                  * only has unsignaled sends in it.  We've shutdown new
 760                  * sends before getting here so by waiting for signaled
 761                  * sends to complete we're ensured that there will be no
 762                  * more tx processing.
 763                  */
 764                 wait_event(rds_ib_ring_empty_wait,
 765                            rds_ib_ring_empty(&ic->i_recv_ring) &&
 766                            (atomic_read(&ic->i_signaled_sends) == 0) &&
 767                            (atomic_read(&ic->i_fastreg_wrs) == RDS_IB_DEFAULT_FR_WR));
 768                 tasklet_kill(&ic->i_send_tasklet);
 769                 tasklet_kill(&ic->i_recv_tasklet);
 770
 771                 /* first destroy the ib state that generates callbacks */
 772                 if (ic->i_cm_id->qp)
 773                         rdma_destroy_qp(ic->i_cm_id);
 774                 if (ic->i_send_cq)
 775                         ib_destroy_cq(ic->i_send_cq);
 776                 if (ic->i_recv_cq)
 777                         ib_destroy_cq(ic->i_recv_cq);
 778
 779                 /* then free the resources that ib callbacks use */
 780                 if (ic->i_send_hdrs)
 781                         ib_dma_free_coherent(dev,
 782                                            ic->i_send_ring.w_nr *
 783                                                 sizeof(struct rds_header),
 784                                            ic->i_send_hdrs,
 785                                            ic->i_send_hdrs_dma);
 786
 787                 if (ic->i_recv_hdrs)
 788                         ib_dma_free_coherent(dev,
 789                                            ic->i_recv_ring.w_nr *
 790                                                 sizeof(struct rds_header),
 791                                            ic->i_recv_hdrs,
 792                                            ic->i_recv_hdrs_dma);
 793
 794                 if (ic->i_ack)
 795                         ib_dma_free_coherent(dev, sizeof(struct rds_header),
 796                                              ic->i_ack, ic->i_ack_dma);
 797
 798                 if (ic->i_sends)
 799                         rds_ib_send_clear_ring(ic);
 800                 if (ic->i_recvs)
 801                         rds_ib_recv_clear_ring(ic);
 802
 803                 rdma_destroy_id(ic->i_cm_id);
 804
 805                 /*
 806                  * Move connection back to the nodev list.
 807                  */
 808                 if (ic->rds_ibdev)
 809                         rds_ib_remove_conn(ic->rds_ibdev, conn);
 810
 811                 ic->i_cm_id = NULL;
 812                 ic->i_pd = NULL;
 813                 ic->i_send_cq = NULL;
 814                 ic->i_recv_cq = NULL;
 815                 ic->i_send_hdrs = NULL;
 816                 ic->i_recv_hdrs = NULL;
 817                 ic->i_ack = NULL;
 818         }
 819         BUG_ON(ic->rds_ibdev);
 820
 821         /* Clear pending transmit */
 822         if (ic->i_data_op) {
 823                 struct rds_message *rm;
 824
 825                 rm = container_of(ic->i_data_op, struct rds_message, data);
 826                 rds_message_put(rm);
 827                 ic->i_data_op = NULL;
 828         }
 829
 830         /* Clear the ACK state */
 831         clear_bit(IB_ACK_IN_FLIGHT, &ic->i_ack_flags);
 832 #ifdef KERNEL_HAS_ATOMIC64
 833         atomic64_set(&ic->i_ack_next, 0);
 834 #else
 835         ic->i_ack_next = 0;
 836 #endif
 837         ic->i_ack_recv = 0;
 838
 839         /* Clear flow control state */
 840         ic->i_flowctl = 0;
 841         atomic_set(&ic->i_credits, 0);
 842
 843         rds_ib_ring_init(&ic->i_send_ring, rds_ib_sysctl_max_send_wr);
 844         rds_ib_ring_init(&ic->i_recv_ring, rds_ib_sysctl_max_recv_wr);
 845
 846         if (ic->i_ibinc) {
 847                 rds_inc_put(&ic->i_ibinc->ii_inc);
 848                 ic->i_ibinc = NULL;
 849         }
 850
 851         vfree(ic->i_sends);
 852         ic->i_sends = NULL;
 853         vfree(ic->i_recvs);
 854         ic->i_recvs = NULL;
 855 }
 856
 857 int rds_ib_conn_alloc(struct rds_connection *conn, gfp_t gfp)
 858 {
 859         struct rds_ib_connection *ic;
 860         unsigned long flags;
 861         int ret;
 862
 863         /* XXX too lazy? */
 864         ic = kzalloc(sizeof(struct rds_ib_connection), gfp);
 865         if (!ic)
 866                 return -ENOMEM;
 867
 868         ret = rds_ib_recv_alloc_caches(ic);
 869         if (ret) {
 870                 kfree(ic);
 871                 return ret;
 872         }
 873
 874         INIT_LIST_HEAD(&ic->ib_node);
 875         tasklet_init(&ic->i_send_tasklet, rds_ib_tasklet_fn_send,
 876                      (unsigned long)ic);
 877         tasklet_init(&ic->i_recv_tasklet, rds_ib_tasklet_fn_recv,
 878                      (unsigned long)ic);
 879         mutex_init(&ic->i_recv_mutex);
 880 #ifndef KERNEL_HAS_ATOMIC64
 881         spin_lock_init(&ic->i_ack_lock);
 882 #endif
 883         atomic_set(&ic->i_signaled_sends, 0);
 884
 885         /*
 886          * rds_ib_conn_shutdown() waits for these to be emptied so they
 887          * must be initialized before it can be called.
 888          */
 889         rds_ib_ring_init(&ic->i_send_ring, rds_ib_sysctl_max_send_wr);
 890         rds_ib_ring_init(&ic->i_recv_ring, rds_ib_sysctl_max_recv_wr);
 891
 892         ic->conn = conn;
 893         conn->c_transport_data = ic;
 894
 895         spin_lock_irqsave(&ib_nodev_conns_lock, flags);
 896         list_add_tail(&ic->ib_node, &ib_nodev_conns);
 897         spin_unlock_irqrestore(&ib_nodev_conns_lock, flags);
 898
 899
 900         rdsdebug("conn %p conn ic %p\n", conn, conn->c_transport_data);
 901         return 0;
 902 }
 903
 904 /*
 905  * Free a connection. Connection must be shut down and not set for reconnect.
 906  */
 907 void rds_ib_conn_free(void *arg)
 908 {
 909         struct rds_ib_connection *ic = arg;
 910         spinlock_t      *lock_ptr;
 911
 912         rdsdebug("ic %p\n", ic);
 913
 914         /*
 915          * Conn is either on a dev's list or on the nodev list.
 916          * A race with shutdown() or connect() would cause problems
 917          * (since rds_ibdev would change) but that should never happen.
 918          */
 919         lock_ptr = ic->rds_ibdev ? &ic->rds_ibdev->spinlock : &ib_nodev_conns_lock;
 920
 921         spin_lock_irq(lock_ptr);
 922         list_del(&ic->ib_node);
 923         spin_unlock_irq(lock_ptr);
 924
 925         rds_ib_recv_free_caches(ic);
 926
 927         kfree(ic);
 928 }
 929
 930
 931 /*
 932  * An error occurred on the connection
 933  */
 934 void
 935 __rds_ib_conn_error(struct rds_connection *conn, const char *fmt, ...)
 936 {
 937         va_list ap;
 938
 939         rds_conn_drop(conn);
 940
 941         va_start(ap, fmt);
 942         vprintk(fmt, ap);
 943         va_end(ap);
 944 }