drivers/infiniband/core/verbs.c

   1 /*
   2  * Copyright (c) 2004 Mellanox Technologies Ltd.  All rights reserved.
   3  * Copyright (c) 2004 Infinicon Corporation.  All rights reserved.
   4  * Copyright (c) 2004 Intel Corporation.  All rights reserved.
   5  * Copyright (c) 2004 Topspin Corporation.  All rights reserved.
   6  * Copyright (c) 2004 Voltaire Corporation.  All rights reserved.
   7  * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved.
   8  * Copyright (c) 2005, 2006 Cisco Systems.  All rights reserved.
   9  *
  10  * This software is available to you under a choice of one of two
  11  * licenses.  You may choose to be licensed under the terms of the GNU
  12  * General Public License (GPL) Version 2, available from the file
  13  * COPYING in the main directory of this source tree, or the
  14  * OpenIB.org BSD license below:
  15  *
  16  *     Redistribution and use in source and binary forms, with or
  17  *     without modification, are permitted provided that the following
  18  *     conditions are met:
  19  *
  20  *      - Redistributions of source code must retain the above
  21  *        copyright notice, this list of conditions and the following
  22  *        disclaimer.
  23  *
  24  *      - Redistributions in binary form must reproduce the above
  25  *        copyright notice, this list of conditions and the following
  26  *        disclaimer in the documentation and/or other materials
  27  *        provided with the distribution.
  28  *
  29  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
  30  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  31  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
  32  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
  33  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
  34  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
  35  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  36  * SOFTWARE.
  37  */
  38
  39 #include <linux/errno.h>
  40 #include <linux/err.h>
  41 #include <linux/export.h>
  42 #include <linux/string.h>
  43 #include <linux/slab.h>
  44 #include <linux/in.h>
  45 #include <linux/in6.h>
  46 #include <net/addrconf.h>
  47
  48 #include <rdma/ib_verbs.h>
  49 #include <rdma/ib_cache.h>
  50 #include <rdma/ib_addr.h>
  51
  52 #include "core_priv.h"
  53
  54 static const char * const ib_events[] = {
  55         [IB_EVENT_CQ_ERR]               = "CQ error",
  56         [IB_EVENT_QP_FATAL]             = "QP fatal error",
  57         [IB_EVENT_QP_REQ_ERR]           = "QP request error",
  58         [IB_EVENT_QP_ACCESS_ERR]        = "QP access error",
  59         [IB_EVENT_COMM_EST]             = "communication established",
  60         [IB_EVENT_SQ_DRAINED]           = "send queue drained",
  61         [IB_EVENT_PATH_MIG]             = "path migration successful",
  62         [IB_EVENT_PATH_MIG_ERR]         = "path migration error",
  63         [IB_EVENT_DEVICE_FATAL]         = "device fatal error",
  64         [IB_EVENT_PORT_ACTIVE]          = "port active",
  65         [IB_EVENT_PORT_ERR]             = "port error",
  66         [IB_EVENT_LID_CHANGE]           = "LID change",
  67         [IB_EVENT_PKEY_CHANGE]          = "P_key change",
  68         [IB_EVENT_SM_CHANGE]            = "SM change",
  69         [IB_EVENT_SRQ_ERR]              = "SRQ error",
  70         [IB_EVENT_SRQ_LIMIT_REACHED]    = "SRQ limit reached",
  71         [IB_EVENT_QP_LAST_WQE_REACHED]  = "last WQE reached",
  72         [IB_EVENT_CLIENT_REREGISTER]    = "client reregister",
  73         [IB_EVENT_GID_CHANGE]           = "GID changed",
  74 };
  75
  76 const char *__attribute_const__ ib_event_msg(enum ib_event_type event)
  77 {
  78         size_t index = event;
  79
  80         return (index < ARRAY_SIZE(ib_events) && ib_events[index]) ?
  81                         ib_events[index] : "unrecognized event";
  82 }
  83 EXPORT_SYMBOL(ib_event_msg);
  84
  85 static const char * const wc_statuses[] = {
  86         [IB_WC_SUCCESS]                 = "success",
  87         [IB_WC_LOC_LEN_ERR]             = "local length error",
  88         [IB_WC_LOC_QP_OP_ERR]           = "local QP operation error",
  89         [IB_WC_LOC_EEC_OP_ERR]          = "local EE context operation error",
  90         [IB_WC_LOC_PROT_ERR]            = "local protection error",
  91         [IB_WC_WR_FLUSH_ERR]            = "WR flushed",
  92         [IB_WC_MW_BIND_ERR]             = "memory management operation error",
  93         [IB_WC_BAD_RESP_ERR]            = "bad response error",
  94         [IB_WC_LOC_ACCESS_ERR]          = "local access error",
  95         [IB_WC_REM_INV_REQ_ERR]         = "invalid request error",
  96         [IB_WC_REM_ACCESS_ERR]          = "remote access error",
  97         [IB_WC_REM_OP_ERR]              = "remote operation error",
  98         [IB_WC_RETRY_EXC_ERR]           = "transport retry counter exceeded",
  99         [IB_WC_RNR_RETRY_EXC_ERR]       = "RNR retry counter exceeded",
 100         [IB_WC_LOC_RDD_VIOL_ERR]        = "local RDD violation error",
 101         [IB_WC_REM_INV_RD_REQ_ERR]      = "remote invalid RD request",
 102         [IB_WC_REM_ABORT_ERR]           = "operation aborted",
 103         [IB_WC_INV_EECN_ERR]            = "invalid EE context number",
 104         [IB_WC_INV_EEC_STATE_ERR]       = "invalid EE context state",
 105         [IB_WC_FATAL_ERR]               = "fatal error",
 106         [IB_WC_RESP_TIMEOUT_ERR]        = "response timeout error",
 107         [IB_WC_GENERAL_ERR]             = "general error",
 108 };
 109
 110 const char *__attribute_const__ ib_wc_status_msg(enum ib_wc_status status)
 111 {
 112         size_t index = status;
 113
 114         return (index < ARRAY_SIZE(wc_statuses) && wc_statuses[index]) ?
 115                         wc_statuses[index] : "unrecognized status";
 116 }
 117 EXPORT_SYMBOL(ib_wc_status_msg);
 118
 119 __attribute_const__ int ib_rate_to_mult(enum ib_rate rate)
 120 {
 121         switch (rate) {
 122         case IB_RATE_2_5_GBPS: return  1;
 123         case IB_RATE_5_GBPS:   return  2;
 124         case IB_RATE_10_GBPS:  return  4;
 125         case IB_RATE_20_GBPS:  return  8;
 126         case IB_RATE_30_GBPS:  return 12;
 127         case IB_RATE_40_GBPS:  return 16;
 128         case IB_RATE_60_GBPS:  return 24;
 129         case IB_RATE_80_GBPS:  return 32;
 130         case IB_RATE_120_GBPS: return 48;
 131         default:               return -1;
 132         }
 133 }
 134 EXPORT_SYMBOL(ib_rate_to_mult);
 135
 136 __attribute_const__ enum ib_rate mult_to_ib_rate(int mult)
 137 {
 138         switch (mult) {
 139         case 1:  return IB_RATE_2_5_GBPS;
 140         case 2:  return IB_RATE_5_GBPS;
 141         case 4:  return IB_RATE_10_GBPS;
 142         case 8:  return IB_RATE_20_GBPS;
 143         case 12: return IB_RATE_30_GBPS;
 144         case 16: return IB_RATE_40_GBPS;
 145         case 24: return IB_RATE_60_GBPS;
 146         case 32: return IB_RATE_80_GBPS;
 147         case 48: return IB_RATE_120_GBPS;
 148         default: return IB_RATE_PORT_CURRENT;
 149         }
 150 }
 151 EXPORT_SYMBOL(mult_to_ib_rate);
 152
 153 __attribute_const__ int ib_rate_to_mbps(enum ib_rate rate)
 154 {
 155         switch (rate) {
 156         case IB_RATE_2_5_GBPS: return 2500;
 157         case IB_RATE_5_GBPS:   return 5000;
 158         case IB_RATE_10_GBPS:  return 10000;
 159         case IB_RATE_20_GBPS:  return 20000;
 160         case IB_RATE_30_GBPS:  return 30000;
 161         case IB_RATE_40_GBPS:  return 40000;
 162         case IB_RATE_60_GBPS:  return 60000;
 163         case IB_RATE_80_GBPS:  return 80000;
 164         case IB_RATE_120_GBPS: return 120000;
 165         case IB_RATE_14_GBPS:  return 14062;
 166         case IB_RATE_56_GBPS:  return 56250;
 167         case IB_RATE_112_GBPS: return 112500;
 168         case IB_RATE_168_GBPS: return 168750;
 169         case IB_RATE_25_GBPS:  return 25781;
 170         case IB_RATE_100_GBPS: return 103125;
 171         case IB_RATE_200_GBPS: return 206250;
 172         case IB_RATE_300_GBPS: return 309375;
 173         default:               return -1;
 174         }
 175 }
 176 EXPORT_SYMBOL(ib_rate_to_mbps);
 177
 178 __attribute_const__ enum rdma_transport_type
 179 rdma_node_get_transport(enum rdma_node_type node_type)
 180 {
 181         switch (node_type) {
 182         case RDMA_NODE_IB_CA:
 183         case RDMA_NODE_IB_SWITCH:
 184         case RDMA_NODE_IB_ROUTER:
 185                 return RDMA_TRANSPORT_IB;
 186         case RDMA_NODE_RNIC:
 187                 return RDMA_TRANSPORT_IWARP;
 188         case RDMA_NODE_USNIC:
 189                 return RDMA_TRANSPORT_USNIC;
 190         case RDMA_NODE_USNIC_UDP:
 191                 return RDMA_TRANSPORT_USNIC_UDP;
 192         default:
 193                 BUG();
 194                 return 0;
 195         }
 196 }
 197 EXPORT_SYMBOL(rdma_node_get_transport);
 198
 199 enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device, u8 port_num)
 200 {
 201         if (device->get_link_layer)
 202                 return device->get_link_layer(device, port_num);
 203
 204         switch (rdma_node_get_transport(device->node_type)) {
 205         case RDMA_TRANSPORT_IB:
 206                 return IB_LINK_LAYER_INFINIBAND;
 207         case RDMA_TRANSPORT_IWARP:
 208         case RDMA_TRANSPORT_USNIC:
 209         case RDMA_TRANSPORT_USNIC_UDP:
 210                 return IB_LINK_LAYER_ETHERNET;
 211         default:
 212                 return IB_LINK_LAYER_UNSPECIFIED;
 213         }
 214 }
 215 EXPORT_SYMBOL(rdma_port_get_link_layer);
 216
 217 /* Protection domains */
 218
 219 /**
 220  * ib_alloc_pd - Allocates an unused protection domain.
 221  * @device: The device on which to allocate the protection domain.
 222  *
 223  * A protection domain object provides an association between QPs, shared
 224  * receive queues, address handles, memory regions, and memory windows.
 225  *
 226  * Every PD has a local_dma_lkey which can be used as the lkey value for local
 227  * memory operations.
 228  */
 229 struct ib_pd *ib_alloc_pd(struct ib_device *device)
 230 {
 231         struct ib_pd *pd;
 232
 233         pd = device->alloc_pd(device, NULL, NULL);
 234         if (IS_ERR(pd))
 235                 return pd;
 236
 237         pd->device = device;
 238         pd->uobject = NULL;
 239         pd->local_mr = NULL;
 240         atomic_set(&pd->usecnt, 0);
 241
 242         if (device->attrs.device_cap_flags & IB_DEVICE_LOCAL_DMA_LKEY)
 243                 pd->local_dma_lkey = device->local_dma_lkey;
 244         else {
 245                 struct ib_mr *mr;
 246
 247                 mr = ib_get_dma_mr(pd, IB_ACCESS_LOCAL_WRITE);
 248                 if (IS_ERR(mr)) {
 249                         ib_dealloc_pd(pd);
 250                         return (struct ib_pd *)mr;
 251                 }
 252
 253                 pd->local_mr = mr;
 254                 pd->local_dma_lkey = pd->local_mr->lkey;
 255         }
 256         return pd;
 257 }
 258 EXPORT_SYMBOL(ib_alloc_pd);
 259
 260 /**
 261  * ib_dealloc_pd - Deallocates a protection domain.
 262  * @pd: The protection domain to deallocate.
 263  *
 264  * It is an error to call this function while any resources in the pd still
 265  * exist.  The caller is responsible to synchronously destroy them and
 266  * guarantee no new allocations will happen.
 267  */
 268 void ib_dealloc_pd(struct ib_pd *pd)
 269 {
 270         int ret;
 271
 272         if (pd->local_mr) {
 273                 ret = ib_dereg_mr(pd->local_mr);
 274                 WARN_ON(ret);
 275                 pd->local_mr = NULL;
 276         }
 277
 278         /* uverbs manipulates usecnt with proper locking, while the kabi
 279            requires the caller to guarantee we can't race here. */
 280         WARN_ON(atomic_read(&pd->usecnt));
 281
 282         /* Making delalloc_pd a void return is a WIP, no driver should return
 283            an error here. */
 284         ret = pd->device->dealloc_pd(pd);
 285         WARN_ONCE(ret, "Infiniband HW driver failed dealloc_pd");
 286 }
 287 EXPORT_SYMBOL(ib_dealloc_pd);
 288
 289 /* Address handles */
 290
 291 struct ib_ah *ib_create_ah(struct ib_pd *pd, struct ib_ah_attr *ah_attr)
 292 {
 293         struct ib_ah *ah;
 294
 295         ah = pd->device->create_ah(pd, ah_attr);
 296
 297         if (!IS_ERR(ah)) {
 298                 ah->device  = pd->device;
 299                 ah->pd      = pd;
 300                 ah->uobject = NULL;
 301                 atomic_inc(&pd->usecnt);
 302         }
 303
 304         return ah;
 305 }
 306 EXPORT_SYMBOL(ib_create_ah);
 307
 308 static int ib_get_header_version(const union rdma_network_hdr *hdr)
 309 {
 310         const struct iphdr *ip4h = (struct iphdr *)&hdr->roce4grh;
 311         struct iphdr ip4h_checked;
 312         const struct ipv6hdr *ip6h = (struct ipv6hdr *)&hdr->ibgrh;
 313
 314         /* If it's IPv6, the version must be 6, otherwise, the first
 315          * 20 bytes (before the IPv4 header) are garbled.
 316          */
 317         if (ip6h->version != 6)
 318                 return (ip4h->version == 4) ? 4 : 0;
 319         /* version may be 6 or 4 because the first 20 bytes could be garbled */
 320
 321         /* RoCE v2 requires no options, thus header length
 322          * must be 5 words
 323          */
 324         if (ip4h->ihl != 5)
 325                 return 6;
 326
 327         /* Verify checksum.
 328          * We can't write on scattered buffers so we need to copy to
 329          * temp buffer.
 330          */
 331         memcpy(&ip4h_checked, ip4h, sizeof(ip4h_checked));
 332         ip4h_checked.check = 0;
 333         ip4h_checked.check = ip_fast_csum((u8 *)&ip4h_checked, 5);
 334         /* if IPv4 header checksum is OK, believe it */
 335         if (ip4h->check == ip4h_checked.check)
 336                 return 4;
 337         return 6;
 338 }
 339
 340 static enum rdma_network_type ib_get_net_type_by_grh(struct ib_device *device,
 341                                                      u8 port_num,
 342                                                      const struct ib_grh *grh)
 343 {
 344         int grh_version;
 345
 346         if (rdma_protocol_ib(device, port_num))
 347                 return RDMA_NETWORK_IB;
 348
 349         grh_version = ib_get_header_version((union rdma_network_hdr *)grh);
 350
 351         if (grh_version == 4)
 352                 return RDMA_NETWORK_IPV4;
 353
 354         if (grh->next_hdr == IPPROTO_UDP)
 355                 return RDMA_NETWORK_IPV6;
 356
 357         return RDMA_NETWORK_ROCE_V1;
 358 }
 359
 360 struct find_gid_index_context {
 361         u16 vlan_id;
 362         enum ib_gid_type gid_type;
 363 };
 364
 365 static bool find_gid_index(const union ib_gid *gid,
 366                            const struct ib_gid_attr *gid_attr,
 367                            void *context)
 368 {
 369         struct find_gid_index_context *ctx =
 370                 (struct find_gid_index_context *)context;
 371
 372         if (ctx->gid_type != gid_attr->gid_type)
 373                 return false;
 374
 375         if ((!!(ctx->vlan_id != 0xffff) == !is_vlan_dev(gid_attr->ndev)) ||
 376             (is_vlan_dev(gid_attr->ndev) &&
 377              vlan_dev_vlan_id(gid_attr->ndev) != ctx->vlan_id))
 378                 return false;
 379
 380         return true;
 381 }
 382
 383 static int get_sgid_index_from_eth(struct ib_device *device, u8 port_num,
 384                                    u16 vlan_id, const union ib_gid *sgid,
 385                                    enum ib_gid_type gid_type,
 386                                    u16 *gid_index)
 387 {
 388         struct find_gid_index_context context = {.vlan_id = vlan_id,
 389                                                  .gid_type = gid_type};
 390
 391         return ib_find_gid_by_filter(device, sgid, port_num, find_gid_index,
 392                                      &context, gid_index);
 393 }
 394
 395 static int get_gids_from_rdma_hdr(union rdma_network_hdr *hdr,
 396                                   enum rdma_network_type net_type,
 397                                   union ib_gid *sgid, union ib_gid *dgid)
 398 {
 399         struct sockaddr_in  src_in;
 400         struct sockaddr_in  dst_in;
 401         __be32 src_saddr, dst_saddr;
 402
 403         if (!sgid || !dgid)
 404                 return -EINVAL;
 405
 406         if (net_type == RDMA_NETWORK_IPV4) {
 407                 memcpy(&src_in.sin_addr.s_addr,
 408                        &hdr->roce4grh.saddr, 4);
 409                 memcpy(&dst_in.sin_addr.s_addr,
 410                        &hdr->roce4grh.daddr, 4);
 411                 src_saddr = src_in.sin_addr.s_addr;
 412                 dst_saddr = dst_in.sin_addr.s_addr;
 413                 ipv6_addr_set_v4mapped(src_saddr,
 414                                        (struct in6_addr *)sgid);
 415                 ipv6_addr_set_v4mapped(dst_saddr,
 416                                        (struct in6_addr *)dgid);
 417                 return 0;
 418         } else if (net_type == RDMA_NETWORK_IPV6 ||
 419                    net_type == RDMA_NETWORK_IB) {
 420                 *dgid = hdr->ibgrh.dgid;
 421                 *sgid = hdr->ibgrh.sgid;
 422                 return 0;
 423         } else {
 424                 return -EINVAL;
 425         }
 426 }
 427
 428 int ib_init_ah_from_wc(struct ib_device *device, u8 port_num,
 429                        const struct ib_wc *wc, const struct ib_grh *grh,
 430                        struct ib_ah_attr *ah_attr)
 431 {
 432         u32 flow_class;
 433         u16 gid_index;
 434         int ret;
 435         enum rdma_network_type net_type = RDMA_NETWORK_IB;
 436         enum ib_gid_type gid_type = IB_GID_TYPE_IB;
 437         int hoplimit = 0xff;
 438         union ib_gid dgid;
 439         union ib_gid sgid;
 440
 441         memset(ah_attr, 0, sizeof *ah_attr);
 442         if (rdma_cap_eth_ah(device, port_num)) {
 443                 if (wc->wc_flags & IB_WC_WITH_NETWORK_HDR_TYPE)
 444                         net_type = wc->network_hdr_type;
 445                 else
 446                         net_type = ib_get_net_type_by_grh(device, port_num, grh);
 447                 gid_type = ib_network_to_gid_type(net_type);
 448         }
 449         ret = get_gids_from_rdma_hdr((union rdma_network_hdr *)grh, net_type,
 450                                      &sgid, &dgid);
 451         if (ret)
 452                 return ret;
 453
 454         if (rdma_protocol_roce(device, port_num)) {
 455                 int if_index = 0;
 456                 u16 vlan_id = wc->wc_flags & IB_WC_WITH_VLAN ?
 457                                 wc->vlan_id : 0xffff;
 458                 struct net_device *idev;
 459                 struct net_device *resolved_dev;
 460
 461                 if (!(wc->wc_flags & IB_WC_GRH))
 462                         return -EPROTOTYPE;
 463
 464                 if (!device->get_netdev)
 465                         return -EOPNOTSUPP;
 466
 467                 idev = device->get_netdev(device, port_num);
 468                 if (!idev)
 469                         return -ENODEV;
 470
 471                 ret = rdma_addr_find_l2_eth_by_grh(&dgid, &sgid,
 472                                                    ah_attr->dmac,
 473                                                    wc->wc_flags & IB_WC_WITH_VLAN ?
 474                                                    NULL : &vlan_id,
 475                                                    &if_index, &hoplimit);
 476                 if (ret) {
 477                         dev_put(idev);
 478                         return ret;
 479                 }
 480
 481                 resolved_dev = dev_get_by_index(&init_net, if_index);
 482                 if (resolved_dev->flags & IFF_LOOPBACK) {
 483                         dev_put(resolved_dev);
 484                         resolved_dev = idev;
 485                         dev_hold(resolved_dev);
 486                 }
 487                 rcu_read_lock();
 488                 if (resolved_dev != idev && !rdma_is_upper_dev_rcu(idev,
 489                                                                    resolved_dev))
 490                         ret = -EHOSTUNREACH;
 491                 rcu_read_unlock();
 492                 dev_put(idev);
 493                 dev_put(resolved_dev);
 494                 if (ret)
 495                         return ret;
 496
 497                 ret = get_sgid_index_from_eth(device, port_num, vlan_id,
 498                                               &dgid, gid_type, &gid_index);
 499                 if (ret)
 500                         return ret;
 501         }
 502
 503         ah_attr->dlid = wc->slid;
 504         ah_attr->sl = wc->sl;
 505         ah_attr->src_path_bits = wc->dlid_path_bits;
 506         ah_attr->port_num = port_num;
 507
 508         if (wc->wc_flags & IB_WC_GRH) {
 509                 ah_attr->ah_flags = IB_AH_GRH;
 510                 ah_attr->grh.dgid = sgid;
 511
 512                 if (!rdma_cap_eth_ah(device, port_num)) {
 513                         ret = ib_find_cached_gid_by_port(device, &dgid,
 514                                                          IB_GID_TYPE_IB,
 515                                                          port_num, NULL,
 516                                                          &gid_index);
 517                         if (ret)
 518                                 return ret;
 519                 }
 520
 521                 ah_attr->grh.sgid_index = (u8) gid_index;
 522                 flow_class = be32_to_cpu(grh->version_tclass_flow);
 523                 ah_attr->grh.flow_label = flow_class & 0xFFFFF;
 524                 ah_attr->grh.hop_limit = hoplimit;
 525                 ah_attr->grh.traffic_class = (flow_class >> 20) & 0xFF;
 526         }
 527         return 0;
 528 }
 529 EXPORT_SYMBOL(ib_init_ah_from_wc);
 530
 531 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, const struct ib_wc *wc,
 532                                    const struct ib_grh *grh, u8 port_num)
 533 {
 534         struct ib_ah_attr ah_attr;
 535         int ret;
 536
 537         ret = ib_init_ah_from_wc(pd->device, port_num, wc, grh, &ah_attr);
 538         if (ret)
 539                 return ERR_PTR(ret);
 540
 541         return ib_create_ah(pd, &ah_attr);
 542 }
 543 EXPORT_SYMBOL(ib_create_ah_from_wc);
 544
 545 int ib_modify_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr)
 546 {
 547         return ah->device->modify_ah ?
 548                 ah->device->modify_ah(ah, ah_attr) :
 549                 -ENOSYS;
 550 }
 551 EXPORT_SYMBOL(ib_modify_ah);
 552
 553 int ib_query_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr)
 554 {
 555         return ah->device->query_ah ?
 556                 ah->device->query_ah(ah, ah_attr) :
 557                 -ENOSYS;
 558 }
 559 EXPORT_SYMBOL(ib_query_ah);
 560
 561 int ib_destroy_ah(struct ib_ah *ah)
 562 {
 563         struct ib_pd *pd;
 564         int ret;
 565
 566         pd = ah->pd;
 567         ret = ah->device->destroy_ah(ah);
 568         if (!ret)
 569                 atomic_dec(&pd->usecnt);
 570
 571         return ret;
 572 }
 573 EXPORT_SYMBOL(ib_destroy_ah);
 574
 575 /* Shared receive queues */
 576
 577 struct ib_srq *ib_create_srq(struct ib_pd *pd,
 578                              struct ib_srq_init_attr *srq_init_attr)
 579 {
 580         struct ib_srq *srq;
 581
 582         if (!pd->device->create_srq)
 583                 return ERR_PTR(-ENOSYS);
 584
 585         srq = pd->device->create_srq(pd, srq_init_attr, NULL);
 586
 587         if (!IS_ERR(srq)) {
 588                 srq->device        = pd->device;
 589                 srq->pd            = pd;
 590                 srq->uobject       = NULL;
 591                 srq->event_handler = srq_init_attr->event_handler;
 592                 srq->srq_context   = srq_init_attr->srq_context;
 593                 srq->srq_type      = srq_init_attr->srq_type;
 594                 if (srq->srq_type == IB_SRQT_XRC) {
 595                         srq->ext.xrc.xrcd = srq_init_attr->ext.xrc.xrcd;
 596                         srq->ext.xrc.cq   = srq_init_attr->ext.xrc.cq;
 597                         atomic_inc(&srq->ext.xrc.xrcd->usecnt);
 598                         atomic_inc(&srq->ext.xrc.cq->usecnt);
 599                 }
 600                 atomic_inc(&pd->usecnt);
 601                 atomic_set(&srq->usecnt, 0);
 602         }
 603
 604         return srq;
 605 }
 606 EXPORT_SYMBOL(ib_create_srq);
 607
 608 int ib_modify_srq(struct ib_srq *srq,
 609                   struct ib_srq_attr *srq_attr,
 610                   enum ib_srq_attr_mask srq_attr_mask)
 611 {
 612         return srq->device->modify_srq ?
 613                 srq->device->modify_srq(srq, srq_attr, srq_attr_mask, NULL) :
 614                 -ENOSYS;
 615 }
 616 EXPORT_SYMBOL(ib_modify_srq);
 617
 618 int ib_query_srq(struct ib_srq *srq,
 619                  struct ib_srq_attr *srq_attr)
 620 {
 621         return srq->device->query_srq ?
 622                 srq->device->query_srq(srq, srq_attr) : -ENOSYS;
 623 }
 624 EXPORT_SYMBOL(ib_query_srq);
 625
 626 int ib_destroy_srq(struct ib_srq *srq)
 627 {
 628         struct ib_pd *pd;
 629         enum ib_srq_type srq_type;
 630         struct ib_xrcd *uninitialized_var(xrcd);
 631         struct ib_cq *uninitialized_var(cq);
 632         int ret;
 633
 634         if (atomic_read(&srq->usecnt))
 635                 return -EBUSY;
 636
 637         pd = srq->pd;
 638         srq_type = srq->srq_type;
 639         if (srq_type == IB_SRQT_XRC) {
 640                 xrcd = srq->ext.xrc.xrcd;
 641                 cq = srq->ext.xrc.cq;
 642         }
 643
 644         ret = srq->device->destroy_srq(srq);
 645         if (!ret) {
 646                 atomic_dec(&pd->usecnt);
 647                 if (srq_type == IB_SRQT_XRC) {
 648                         atomic_dec(&xrcd->usecnt);
 649                         atomic_dec(&cq->usecnt);
 650                 }
 651         }
 652
 653         return ret;
 654 }
 655 EXPORT_SYMBOL(ib_destroy_srq);
 656
 657 /* Queue pairs */
 658
 659 static void __ib_shared_qp_event_handler(struct ib_event *event, void *context)
 660 {
 661         struct ib_qp *qp = context;
 662         unsigned long flags;
 663
 664         spin_lock_irqsave(&qp->device->event_handler_lock, flags);
 665         list_for_each_entry(event->element.qp, &qp->open_list, open_list)
 666                 if (event->element.qp->event_handler)
 667                         event->element.qp->event_handler(event, event->element.qp->qp_context);
 668         spin_unlock_irqrestore(&qp->device->event_handler_lock, flags);
 669 }
 670
 671 static void __ib_insert_xrcd_qp(struct ib_xrcd *xrcd, struct ib_qp *qp)
 672 {
 673         mutex_lock(&xrcd->tgt_qp_mutex);
 674         list_add(&qp->xrcd_list, &xrcd->tgt_qp_list);
 675         mutex_unlock(&xrcd->tgt_qp_mutex);
 676 }
 677
 678 static struct ib_qp *__ib_open_qp(struct ib_qp *real_qp,
 679                                   void (*event_handler)(struct ib_event *, void *),
 680                                   void *qp_context)
 681 {
 682         struct ib_qp *qp;
 683         unsigned long flags;
 684
 685         qp = kzalloc(sizeof *qp, GFP_KERNEL);
 686         if (!qp)
 687                 return ERR_PTR(-ENOMEM);
 688
 689         qp->real_qp = real_qp;
 690         atomic_inc(&real_qp->usecnt);
 691         qp->device = real_qp->device;
 692         qp->event_handler = event_handler;
 693         qp->qp_context = qp_context;
 694         qp->qp_num = real_qp->qp_num;
 695         qp->qp_type = real_qp->qp_type;
 696
 697         spin_lock_irqsave(&real_qp->device->event_handler_lock, flags);
 698         list_add(&qp->open_list, &real_qp->open_list);
 699         spin_unlock_irqrestore(&real_qp->device->event_handler_lock, flags);
 700
 701         return qp;
 702 }
 703
 704 struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
 705                          struct ib_qp_open_attr *qp_open_attr)
 706 {
 707         struct ib_qp *qp, *real_qp;
 708
 709         if (qp_open_attr->qp_type != IB_QPT_XRC_TGT)
 710                 return ERR_PTR(-EINVAL);
 711
 712         qp = ERR_PTR(-EINVAL);
 713         mutex_lock(&xrcd->tgt_qp_mutex);
 714         list_for_each_entry(real_qp, &xrcd->tgt_qp_list, xrcd_list) {
 715                 if (real_qp->qp_num == qp_open_attr->qp_num) {
 716                         qp = __ib_open_qp(real_qp, qp_open_attr->event_handler,
 717                                           qp_open_attr->qp_context);
 718                         break;
 719                 }
 720         }
 721         mutex_unlock(&xrcd->tgt_qp_mutex);
 722         return qp;
 723 }
 724 EXPORT_SYMBOL(ib_open_qp);
 725
 726 static struct ib_qp *ib_create_xrc_qp(struct ib_qp *qp,
 727                 struct ib_qp_init_attr *qp_init_attr)
 728 {
 729         struct ib_qp *real_qp = qp;
 730
 731         qp->event_handler = __ib_shared_qp_event_handler;
 732         qp->qp_context = qp;
 733         qp->pd = NULL;
 734         qp->send_cq = qp->recv_cq = NULL;
 735         qp->srq = NULL;
 736         qp->xrcd = qp_init_attr->xrcd;
 737         atomic_inc(&qp_init_attr->xrcd->usecnt);
 738         INIT_LIST_HEAD(&qp->open_list);
 739
 740         qp = __ib_open_qp(real_qp, qp_init_attr->event_handler,
 741                           qp_init_attr->qp_context);
 742         if (!IS_ERR(qp))
 743                 __ib_insert_xrcd_qp(qp_init_attr->xrcd, real_qp);
 744         else
 745                 real_qp->device->destroy_qp(real_qp);
 746         return qp;
 747 }
 748
 749 struct ib_qp *ib_create_qp(struct ib_pd *pd,
 750                            struct ib_qp_init_attr *qp_init_attr)
 751 {
 752         struct ib_device *device = pd ? pd->device : qp_init_attr->xrcd->device;
 753         struct ib_qp *qp;
 754
 755         qp = device->create_qp(pd, qp_init_attr, NULL);
 756         if (IS_ERR(qp))
 757                 return qp;
 758
 759         qp->device     = device;
 760         qp->real_qp    = qp;
 761         qp->uobject    = NULL;
 762         qp->qp_type    = qp_init_attr->qp_type;
 763
 764         atomic_set(&qp->usecnt, 0);
 765         qp->mrs_used = 0;
 766         spin_lock_init(&qp->mr_lock);
 767
 768         if (qp_init_attr->qp_type == IB_QPT_XRC_TGT)
 769                 return ib_create_xrc_qp(qp, qp_init_attr);
 770
 771         qp->event_handler = qp_init_attr->event_handler;
 772         qp->qp_context = qp_init_attr->qp_context;
 773         if (qp_init_attr->qp_type == IB_QPT_XRC_INI) {
 774                 qp->recv_cq = NULL;
 775                 qp->srq = NULL;
 776         } else {
 777                 qp->recv_cq = qp_init_attr->recv_cq;
 778                 atomic_inc(&qp_init_attr->recv_cq->usecnt);
 779                 qp->srq = qp_init_attr->srq;
 780                 if (qp->srq)
 781                         atomic_inc(&qp_init_attr->srq->usecnt);
 782         }
 783
 784         qp->pd      = pd;
 785         qp->send_cq = qp_init_attr->send_cq;
 786         qp->xrcd    = NULL;
 787
 788         atomic_inc(&pd->usecnt);
 789         atomic_inc(&qp_init_attr->send_cq->usecnt);
 790         return qp;
 791 }
 792 EXPORT_SYMBOL(ib_create_qp);
 793
 794 static const struct {
 795         int                     valid;
 796         enum ib_qp_attr_mask    req_param[IB_QPT_MAX];
 797         enum ib_qp_attr_mask    opt_param[IB_QPT_MAX];
 798 } qp_state_table[IB_QPS_ERR + 1][IB_QPS_ERR + 1] = {
 799         [IB_QPS_RESET] = {
 800                 [IB_QPS_RESET] = { .valid = 1 },
 801                 [IB_QPS_INIT]  = {
 802                         .valid = 1,
 803                         .req_param = {
 804                                 [IB_QPT_UD]  = (IB_QP_PKEY_INDEX                |
 805                                                 IB_QP_PORT                      |
 806                                                 IB_QP_QKEY),
 807                                 [IB_QPT_RAW_PACKET] = IB_QP_PORT,
 808                                 [IB_QPT_UC]  = (IB_QP_PKEY_INDEX                |
 809                                                 IB_QP_PORT                      |
 810                                                 IB_QP_ACCESS_FLAGS),
 811                                 [IB_QPT_RC]  = (IB_QP_PKEY_INDEX                |
 812                                                 IB_QP_PORT                      |
 813                                                 IB_QP_ACCESS_FLAGS),
 814                                 [IB_QPT_XRC_INI] = (IB_QP_PKEY_INDEX            |
 815                                                 IB_QP_PORT                      |
 816                                                 IB_QP_ACCESS_FLAGS),
 817                                 [IB_QPT_XRC_TGT] = (IB_QP_PKEY_INDEX            |
 818                                                 IB_QP_PORT                      |
 819                                                 IB_QP_ACCESS_FLAGS),
 820                                 [IB_QPT_SMI] = (IB_QP_PKEY_INDEX                |
 821                                                 IB_QP_QKEY),
 822                                 [IB_QPT_GSI] = (IB_QP_PKEY_INDEX                |
 823                                                 IB_QP_QKEY),
 824                         }
 825                 },
 826         },
 827         [IB_QPS_INIT]  = {
 828                 [IB_QPS_RESET] = { .valid = 1 },
 829                 [IB_QPS_ERR] =   { .valid = 1 },
 830                 [IB_QPS_INIT]  = {
 831                         .valid = 1,
 832                         .opt_param = {
 833                                 [IB_QPT_UD]  = (IB_QP_PKEY_INDEX                |
 834                                                 IB_QP_PORT                      |
 835                                                 IB_QP_QKEY),
 836                                 [IB_QPT_UC]  = (IB_QP_PKEY_INDEX                |
 837                                                 IB_QP_PORT                      |
 838                                                 IB_QP_ACCESS_FLAGS),
 839                                 [IB_QPT_RC]  = (IB_QP_PKEY_INDEX                |
 840                                                 IB_QP_PORT                      |
 841                                                 IB_QP_ACCESS_FLAGS),
 842                                 [IB_QPT_XRC_INI] = (IB_QP_PKEY_INDEX            |
 843                                                 IB_QP_PORT                      |
 844                                                 IB_QP_ACCESS_FLAGS),
 845                                 [IB_QPT_XRC_TGT] = (IB_QP_PKEY_INDEX            |
 846                                                 IB_QP_PORT                      |
 847                                                 IB_QP_ACCESS_FLAGS),
 848                                 [IB_QPT_SMI] = (IB_QP_PKEY_INDEX                |
 849                                                 IB_QP_QKEY),
 850                                 [IB_QPT_GSI] = (IB_QP_PKEY_INDEX                |
 851                                                 IB_QP_QKEY),
 852                         }
 853                 },
 854                 [IB_QPS_RTR]   = {
 855                         .valid = 1,
 856                         .req_param = {
 857                                 [IB_QPT_UC]  = (IB_QP_AV                        |
 858                                                 IB_QP_PATH_MTU                  |
 859                                                 IB_QP_DEST_QPN                  |
 860                                                 IB_QP_RQ_PSN),
 861                                 [IB_QPT_RC]  = (IB_QP_AV                        |
 862                                                 IB_QP_PATH_MTU                  |
 863                                                 IB_QP_DEST_QPN                  |
 864                                                 IB_QP_RQ_PSN                    |
 865                                                 IB_QP_MAX_DEST_RD_ATOMIC        |
 866                                                 IB_QP_MIN_RNR_TIMER),
 867                                 [IB_QPT_XRC_INI] = (IB_QP_AV                    |
 868                                                 IB_QP_PATH_MTU                  |
 869                                                 IB_QP_DEST_QPN                  |
 870                                                 IB_QP_RQ_PSN),
 871                                 [IB_QPT_XRC_TGT] = (IB_QP_AV                    |
 872                                                 IB_QP_PATH_MTU                  |
 873                                                 IB_QP_DEST_QPN                  |
 874                                                 IB_QP_RQ_PSN                    |
 875                                                 IB_QP_MAX_DEST_RD_ATOMIC        |
 876                                                 IB_QP_MIN_RNR_TIMER),
 877                         },
 878                         .opt_param = {
 879                                  [IB_QPT_UD]  = (IB_QP_PKEY_INDEX               |
 880                                                  IB_QP_QKEY),
 881                                  [IB_QPT_UC]  = (IB_QP_ALT_PATH                 |
 882                                                  IB_QP_ACCESS_FLAGS             |
 883                                                  IB_QP_PKEY_INDEX),
 884                                  [IB_QPT_RC]  = (IB_QP_ALT_PATH                 |
 885                                                  IB_QP_ACCESS_FLAGS             |
 886                                                  IB_QP_PKEY_INDEX),
 887                                  [IB_QPT_XRC_INI] = (IB_QP_ALT_PATH             |
 888                                                  IB_QP_ACCESS_FLAGS             |
 889                                                  IB_QP_PKEY_INDEX),
 890                                  [IB_QPT_XRC_TGT] = (IB_QP_ALT_PATH             |
 891                                                  IB_QP_ACCESS_FLAGS             |
 892                                                  IB_QP_PKEY_INDEX),
 893                                  [IB_QPT_SMI] = (IB_QP_PKEY_INDEX               |
 894                                                  IB_QP_QKEY),
 895                                  [IB_QPT_GSI] = (IB_QP_PKEY_INDEX               |
 896                                                  IB_QP_QKEY),
 897                          },
 898                 },
 899         },
 900         [IB_QPS_RTR]   = {
 901                 [IB_QPS_RESET] = { .valid = 1 },
 902                 [IB_QPS_ERR] =   { .valid = 1 },
 903                 [IB_QPS_RTS]   = {
 904                         .valid = 1,
 905                         .req_param = {
 906                                 [IB_QPT_UD]  = IB_QP_SQ_PSN,
 907                                 [IB_QPT_UC]  = IB_QP_SQ_PSN,
 908                                 [IB_QPT_RC]  = (IB_QP_TIMEOUT                   |
 909                                                 IB_QP_RETRY_CNT                 |
 910                                                 IB_QP_RNR_RETRY                 |
 911                                                 IB_QP_SQ_PSN                    |
 912                                                 IB_QP_MAX_QP_RD_ATOMIC),
 913                                 [IB_QPT_XRC_INI] = (IB_QP_TIMEOUT               |
 914                                                 IB_QP_RETRY_CNT                 |
 915                                                 IB_QP_RNR_RETRY                 |
 916                                                 IB_QP_SQ_PSN                    |
 917                                                 IB_QP_MAX_QP_RD_ATOMIC),
 918                                 [IB_QPT_XRC_TGT] = (IB_QP_TIMEOUT               |
 919                                                 IB_QP_SQ_PSN),
 920                                 [IB_QPT_SMI] = IB_QP_SQ_PSN,
 921                                 [IB_QPT_GSI] = IB_QP_SQ_PSN,
 922                         },
 923                         .opt_param = {
 924                                  [IB_QPT_UD]  = (IB_QP_CUR_STATE                |
 925                                                  IB_QP_QKEY),
 926                                  [IB_QPT_UC]  = (IB_QP_CUR_STATE                |
 927                                                  IB_QP_ALT_PATH                 |
 928                                                  IB_QP_ACCESS_FLAGS             |
 929                                                  IB_QP_PATH_MIG_STATE),
 930                                  [IB_QPT_RC]  = (IB_QP_CUR_STATE                |
 931                                                  IB_QP_ALT_PATH                 |
 932                                                  IB_QP_ACCESS_FLAGS             |
 933                                                  IB_QP_MIN_RNR_TIMER            |
 934                                                  IB_QP_PATH_MIG_STATE),
 935                                  [IB_QPT_XRC_INI] = (IB_QP_CUR_STATE            |
 936                                                  IB_QP_ALT_PATH                 |
 937                                                  IB_QP_ACCESS_FLAGS             |
 938                                                  IB_QP_PATH_MIG_STATE),
 939                                  [IB_QPT_XRC_TGT] = (IB_QP_CUR_STATE            |
 940                                                  IB_QP_ALT_PATH                 |
 941                                                  IB_QP_ACCESS_FLAGS             |
 942                                                  IB_QP_MIN_RNR_TIMER            |
 943                                                  IB_QP_PATH_MIG_STATE),
 944                                  [IB_QPT_SMI] = (IB_QP_CUR_STATE                |
 945                                                  IB_QP_QKEY),
 946                                  [IB_QPT_GSI] = (IB_QP_CUR_STATE                |
 947                                                  IB_QP_QKEY),
 948                          }
 949                 }
 950         },
 951         [IB_QPS_RTS]   = {
 952                 [IB_QPS_RESET] = { .valid = 1 },
 953                 [IB_QPS_ERR] =   { .valid = 1 },
 954                 [IB_QPS_RTS]   = {
 955                         .valid = 1,
 956                         .opt_param = {
 957                                 [IB_QPT_UD]  = (IB_QP_CUR_STATE                 |
 958                                                 IB_QP_QKEY),
 959                                 [IB_QPT_UC]  = (IB_QP_CUR_STATE                 |
 960                                                 IB_QP_ACCESS_FLAGS              |
 961                                                 IB_QP_ALT_PATH                  |
 962                                                 IB_QP_PATH_MIG_STATE),
 963                                 [IB_QPT_RC]  = (IB_QP_CUR_STATE                 |
 964                                                 IB_QP_ACCESS_FLAGS              |
 965                                                 IB_QP_ALT_PATH                  |
 966                                                 IB_QP_PATH_MIG_STATE            |
 967                                                 IB_QP_MIN_RNR_TIMER),
 968                                 [IB_QPT_XRC_INI] = (IB_QP_CUR_STATE             |
 969                                                 IB_QP_ACCESS_FLAGS              |
 970                                                 IB_QP_ALT_PATH                  |
 971                                                 IB_QP_PATH_MIG_STATE),
 972                                 [IB_QPT_XRC_TGT] = (IB_QP_CUR_STATE             |
 973                                                 IB_QP_ACCESS_FLAGS              |
 974                                                 IB_QP_ALT_PATH                  |
 975                                                 IB_QP_PATH_MIG_STATE            |
 976                                                 IB_QP_MIN_RNR_TIMER),
 977                                 [IB_QPT_SMI] = (IB_QP_CUR_STATE                 |
 978                                                 IB_QP_QKEY),
 979                                 [IB_QPT_GSI] = (IB_QP_CUR_STATE                 |
 980                                                 IB_QP_QKEY),
 981                         }
 982                 },
 983                 [IB_QPS_SQD]   = {
 984                         .valid = 1,
 985                         .opt_param = {
 986                                 [IB_QPT_UD]  = IB_QP_EN_SQD_ASYNC_NOTIFY,
 987                                 [IB_QPT_UC]  = IB_QP_EN_SQD_ASYNC_NOTIFY,
 988                                 [IB_QPT_RC]  = IB_QP_EN_SQD_ASYNC_NOTIFY,
 989                                 [IB_QPT_XRC_INI] = IB_QP_EN_SQD_ASYNC_NOTIFY,
 990                                 [IB_QPT_XRC_TGT] = IB_QP_EN_SQD_ASYNC_NOTIFY, /* ??? */
 991                                 [IB_QPT_SMI] = IB_QP_EN_SQD_ASYNC_NOTIFY,
 992                                 [IB_QPT_GSI] = IB_QP_EN_SQD_ASYNC_NOTIFY
 993                         }
 994                 },
 995         },
 996         [IB_QPS_SQD]   = {
 997                 [IB_QPS_RESET] = { .valid = 1 },
 998                 [IB_QPS_ERR] =   { .valid = 1 },
 999                 [IB_QPS_RTS]   = {
1000                         .valid = 1,
1001                         .opt_param = {
1002                                 [IB_QPT_UD]  = (IB_QP_CUR_STATE                 |
1003                                                 IB_QP_QKEY),
1004                                 [IB_QPT_UC]  = (IB_QP_CUR_STATE                 |
1005                                                 IB_QP_ALT_PATH                  |
1006                                                 IB_QP_ACCESS_FLAGS              |
1007                                                 IB_QP_PATH_MIG_STATE),
1008                                 [IB_QPT_RC]  = (IB_QP_CUR_STATE                 |
1009                                                 IB_QP_ALT_PATH                  |
1010                                                 IB_QP_ACCESS_FLAGS              |
1011                                                 IB_QP_MIN_RNR_TIMER             |
1012                                                 IB_QP_PATH_MIG_STATE),
1013                                 [IB_QPT_XRC_INI] = (IB_QP_CUR_STATE             |
1014                                                 IB_QP_ALT_PATH                  |
1015                                                 IB_QP_ACCESS_FLAGS              |
1016                                                 IB_QP_PATH_MIG_STATE),
1017                                 [IB_QPT_XRC_TGT] = (IB_QP_CUR_STATE             |
1018                                                 IB_QP_ALT_PATH                  |
1019                                                 IB_QP_ACCESS_FLAGS              |
1020                                                 IB_QP_MIN_RNR_TIMER             |
1021                                                 IB_QP_PATH_MIG_STATE),
1022                                 [IB_QPT_SMI] = (IB_QP_CUR_STATE                 |
1023                                                 IB_QP_QKEY),
1024                                 [IB_QPT_GSI] = (IB_QP_CUR_STATE                 |
1025                                                 IB_QP_QKEY),
1026                         }
1027                 },
1028                 [IB_QPS_SQD]   = {
1029                         .valid = 1,
1030                         .opt_param = {
1031                                 [IB_QPT_UD]  = (IB_QP_PKEY_INDEX                |
1032                                                 IB_QP_QKEY),
1033                                 [IB_QPT_UC]  = (IB_QP_AV                        |
1034                                                 IB_QP_ALT_PATH                  |
1035                                                 IB_QP_ACCESS_FLAGS              |
1036                                                 IB_QP_PKEY_INDEX                |
1037                                                 IB_QP_PATH_MIG_STATE),
1038                                 [IB_QPT_RC]  = (IB_QP_PORT                      |
1039                                                 IB_QP_AV                        |
1040                                                 IB_QP_TIMEOUT                   |
1041                                                 IB_QP_RETRY_CNT                 |
1042                                                 IB_QP_RNR_RETRY                 |
1043                                                 IB_QP_MAX_QP_RD_ATOMIC          |
1044                                                 IB_QP_MAX_DEST_RD_ATOMIC        |
1045                                                 IB_QP_ALT_PATH                  |
1046                                                 IB_QP_ACCESS_FLAGS              |
1047                                                 IB_QP_PKEY_INDEX                |
1048                                                 IB_QP_MIN_RNR_TIMER             |
1049                                                 IB_QP_PATH_MIG_STATE),
1050                                 [IB_QPT_XRC_INI] = (IB_QP_PORT                  |
1051                                                 IB_QP_AV                        |
1052                                                 IB_QP_TIMEOUT                   |
1053                                                 IB_QP_RETRY_CNT                 |
1054                                                 IB_QP_RNR_RETRY                 |
1055                                                 IB_QP_MAX_QP_RD_ATOMIC          |
1056                                                 IB_QP_ALT_PATH                  |
1057                                                 IB_QP_ACCESS_FLAGS              |
1058                                                 IB_QP_PKEY_INDEX                |
1059                                                 IB_QP_PATH_MIG_STATE),
1060                                 [IB_QPT_XRC_TGT] = (IB_QP_PORT                  |
1061                                                 IB_QP_AV                        |
1062                                                 IB_QP_TIMEOUT                   |
1063                                                 IB_QP_MAX_DEST_RD_ATOMIC        |
1064                                                 IB_QP_ALT_PATH                  |
1065                                                 IB_QP_ACCESS_FLAGS              |
1066                                                 IB_QP_PKEY_INDEX                |
1067                                                 IB_QP_MIN_RNR_TIMER             |
1068                                                 IB_QP_PATH_MIG_STATE),
1069                                 [IB_QPT_SMI] = (IB_QP_PKEY_INDEX                |
1070                                                 IB_QP_QKEY),
1071                                 [IB_QPT_GSI] = (IB_QP_PKEY_INDEX                |
1072                                                 IB_QP_QKEY),
1073                         }
1074                 }
1075         },
1076         [IB_QPS_SQE]   = {
1077                 [IB_QPS_RESET] = { .valid = 1 },
1078                 [IB_QPS_ERR] =   { .valid = 1 },
1079                 [IB_QPS_RTS]   = {
1080                         .valid = 1,
1081                         .opt_param = {
1082                                 [IB_QPT_UD]  = (IB_QP_CUR_STATE                 |
1083                                                 IB_QP_QKEY),
1084                                 [IB_QPT_UC]  = (IB_QP_CUR_STATE                 |
1085                                                 IB_QP_ACCESS_FLAGS),
1086                                 [IB_QPT_SMI] = (IB_QP_CUR_STATE                 |
1087                                                 IB_QP_QKEY),
1088                                 [IB_QPT_GSI] = (IB_QP_CUR_STATE                 |
1089                                                 IB_QP_QKEY),
1090                         }
1091                 }
1092         },
1093         [IB_QPS_ERR] = {
1094                 [IB_QPS_RESET] = { .valid = 1 },
1095                 [IB_QPS_ERR] =   { .valid = 1 }
1096         }
1097 };
1098
1099 int ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
1100                        enum ib_qp_type type, enum ib_qp_attr_mask mask,
1101                        enum rdma_link_layer ll)
1102 {
1103         enum ib_qp_attr_mask req_param, opt_param;
1104
1105         if (cur_state  < 0 || cur_state  > IB_QPS_ERR ||
1106             next_state < 0 || next_state > IB_QPS_ERR)
1107                 return 0;
1108
1109         if (mask & IB_QP_CUR_STATE  &&
1110             cur_state != IB_QPS_RTR && cur_state != IB_QPS_RTS &&
1111             cur_state != IB_QPS_SQD && cur_state != IB_QPS_SQE)
1112                 return 0;
1113
1114         if (!qp_state_table[cur_state][next_state].valid)
1115                 return 0;
1116
1117         req_param = qp_state_table[cur_state][next_state].req_param[type];
1118         opt_param = qp_state_table[cur_state][next_state].opt_param[type];
1119
1120         if ((mask & req_param) != req_param)
1121                 return 0;
1122
1123         if (mask & ~(req_param | opt_param | IB_QP_STATE))
1124                 return 0;
1125
1126         return 1;
1127 }
1128 EXPORT_SYMBOL(ib_modify_qp_is_ok);
1129
1130 int ib_resolve_eth_dmac(struct ib_qp *qp,
1131                         struct ib_qp_attr *qp_attr, int *qp_attr_mask)
1132 {
1133         int           ret = 0;
1134
1135         if (*qp_attr_mask & IB_QP_AV) {
1136                 if (qp_attr->ah_attr.port_num < rdma_start_port(qp->device) ||
1137                     qp_attr->ah_attr.port_num > rdma_end_port(qp->device))
1138                         return -EINVAL;
1139
1140                 if (!rdma_cap_eth_ah(qp->device, qp_attr->ah_attr.port_num))
1141                         return 0;
1142
1143                 if (rdma_link_local_addr((struct in6_addr *)qp_attr->ah_attr.grh.dgid.raw)) {
1144                         rdma_get_ll_mac((struct in6_addr *)qp_attr->ah_attr.grh.dgid.raw,
1145                                         qp_attr->ah_attr.dmac);
1146                 } else {
1147                         union ib_gid            sgid;
1148                         struct ib_gid_attr      sgid_attr;
1149                         int                     ifindex;
1150                         int                     hop_limit;
1151
1152                         ret = ib_query_gid(qp->device,
1153                                            qp_attr->ah_attr.port_num,
1154                                            qp_attr->ah_attr.grh.sgid_index,
1155                                            &sgid, &sgid_attr);
1156
1157                         if (ret || !sgid_attr.ndev) {
1158                                 if (!ret)
1159                                         ret = -ENXIO;
1160                                 goto out;
1161                         }
1162
1163                         ifindex = sgid_attr.ndev->ifindex;
1164
1165                         ret = rdma_addr_find_l2_eth_by_grh(&sgid,
1166                                                            &qp_attr->ah_attr.grh.dgid,
1167                                                            qp_attr->ah_attr.dmac,
1168                                                            NULL, &ifindex, &hop_limit);
1169
1170                         dev_put(sgid_attr.ndev);
1171
1172                         qp_attr->ah_attr.grh.hop_limit = hop_limit;
1173                 }
1174         }
1175 out:
1176         return ret;
1177 }
1178 EXPORT_SYMBOL(ib_resolve_eth_dmac);
1179
1180
1181 int ib_modify_qp(struct ib_qp *qp,
1182                  struct ib_qp_attr *qp_attr,
1183                  int qp_attr_mask)
1184 {
1185         int ret;
1186
1187         ret = ib_resolve_eth_dmac(qp, qp_attr, &qp_attr_mask);
1188         if (ret)
1189                 return ret;
1190
1191         return qp->device->modify_qp(qp->real_qp, qp_attr, qp_attr_mask, NULL);
1192 }
1193 EXPORT_SYMBOL(ib_modify_qp);
1194
1195 int ib_query_qp(struct ib_qp *qp,
1196                 struct ib_qp_attr *qp_attr,
1197                 int qp_attr_mask,
1198                 struct ib_qp_init_attr *qp_init_attr)
1199 {
1200         return qp->device->query_qp ?
1201                 qp->device->query_qp(qp->real_qp, qp_attr, qp_attr_mask, qp_init_attr) :
1202                 -ENOSYS;
1203 }
1204 EXPORT_SYMBOL(ib_query_qp);
1205
1206 int ib_close_qp(struct ib_qp *qp)
1207 {
1208         struct ib_qp *real_qp;
1209         unsigned long flags;
1210
1211         real_qp = qp->real_qp;
1212         if (real_qp == qp)
1213                 return -EINVAL;
1214
1215         spin_lock_irqsave(&real_qp->device->event_handler_lock, flags);
1216         list_del(&qp->open_list);
1217         spin_unlock_irqrestore(&real_qp->device->event_handler_lock, flags);
1218
1219         atomic_dec(&real_qp->usecnt);
1220         kfree(qp);
1221
1222         return 0;
1223 }
1224 EXPORT_SYMBOL(ib_close_qp);
1225
1226 static int __ib_destroy_shared_qp(struct ib_qp *qp)
1227 {
1228         struct ib_xrcd *xrcd;
1229         struct ib_qp *real_qp;
1230         int ret;
1231
1232         real_qp = qp->real_qp;
1233         xrcd = real_qp->xrcd;
1234
1235         mutex_lock(&xrcd->tgt_qp_mutex);
1236         ib_close_qp(qp);
1237         if (atomic_read(&real_qp->usecnt) == 0)
1238                 list_del(&real_qp->xrcd_list);
1239         else
1240                 real_qp = NULL;
1241         mutex_unlock(&xrcd->tgt_qp_mutex);
1242
1243         if (real_qp) {
1244                 ret = ib_destroy_qp(real_qp);
1245                 if (!ret)
1246                         atomic_dec(&xrcd->usecnt);
1247                 else
1248                         __ib_insert_xrcd_qp(xrcd, real_qp);
1249         }
1250
1251         return 0;
1252 }
1253
1254 int ib_destroy_qp(struct ib_qp *qp)
1255 {
1256         struct ib_pd *pd;
1257         struct ib_cq *scq, *rcq;
1258         struct ib_srq *srq;
1259         int ret;
1260
1261         WARN_ON_ONCE(qp->mrs_used > 0);
1262
1263         if (atomic_read(&qp->usecnt))
1264                 return -EBUSY;
1265
1266         if (qp->real_qp != qp)
1267                 return __ib_destroy_shared_qp(qp);
1268
1269         pd   = qp->pd;
1270         scq  = qp->send_cq;
1271         rcq  = qp->recv_cq;
1272         srq  = qp->srq;
1273
1274         ret = qp->device->destroy_qp(qp);
1275         if (!ret) {
1276                 if (pd)
1277                         atomic_dec(&pd->usecnt);
1278                 if (scq)
1279                         atomic_dec(&scq->usecnt);
1280                 if (rcq)
1281                         atomic_dec(&rcq->usecnt);
1282                 if (srq)
1283                         atomic_dec(&srq->usecnt);
1284         }
1285
1286         return ret;
1287 }
1288 EXPORT_SYMBOL(ib_destroy_qp);
1289
1290 /* Completion queues */
1291
1292 struct ib_cq *ib_create_cq(struct ib_device *device,
1293                            ib_comp_handler comp_handler,
1294                            void (*event_handler)(struct ib_event *, void *),
1295                            void *cq_context,
1296                            const struct ib_cq_init_attr *cq_attr)
1297 {
1298         struct ib_cq *cq;
1299
1300         cq = device->create_cq(device, cq_attr, NULL, NULL);
1301
1302         if (!IS_ERR(cq)) {
1303                 cq->device        = device;
1304                 cq->uobject       = NULL;
1305                 cq->comp_handler  = comp_handler;
1306                 cq->event_handler = event_handler;
1307                 cq->cq_context    = cq_context;
1308                 atomic_set(&cq->usecnt, 0);
1309         }
1310
1311         return cq;
1312 }
1313 EXPORT_SYMBOL(ib_create_cq);
1314
1315 int ib_modify_cq(struct ib_cq *cq, u16 cq_count, u16 cq_period)
1316 {
1317         return cq->device->modify_cq ?
1318                 cq->device->modify_cq(cq, cq_count, cq_period) : -ENOSYS;
1319 }
1320 EXPORT_SYMBOL(ib_modify_cq);
1321
1322 int ib_destroy_cq(struct ib_cq *cq)
1323 {
1324         if (atomic_read(&cq->usecnt))
1325                 return -EBUSY;
1326
1327         return cq->device->destroy_cq(cq);
1328 }
1329 EXPORT_SYMBOL(ib_destroy_cq);
1330
1331 int ib_resize_cq(struct ib_cq *cq, int cqe)
1332 {
1333         return cq->device->resize_cq ?
1334                 cq->device->resize_cq(cq, cqe, NULL) : -ENOSYS;
1335 }
1336 EXPORT_SYMBOL(ib_resize_cq);
1337
1338 /* Memory regions */
1339
1340 struct ib_mr *ib_get_dma_mr(struct ib_pd *pd, int mr_access_flags)
1341 {
1342         struct ib_mr *mr;
1343         int err;
1344
1345         err = ib_check_mr_access(mr_access_flags);
1346         if (err)
1347                 return ERR_PTR(err);
1348
1349         mr = pd->device->get_dma_mr(pd, mr_access_flags);
1350
1351         if (!IS_ERR(mr)) {
1352                 mr->device  = pd->device;
1353                 mr->pd      = pd;
1354                 mr->uobject = NULL;
1355                 atomic_inc(&pd->usecnt);
1356                 mr->need_inval = false;
1357         }
1358
1359         return mr;
1360 }
1361 EXPORT_SYMBOL(ib_get_dma_mr);
1362
1363 int ib_dereg_mr(struct ib_mr *mr)
1364 {
1365         struct ib_pd *pd = mr->pd;
1366         int ret;
1367
1368         ret = mr->device->dereg_mr(mr);
1369         if (!ret)
1370                 atomic_dec(&pd->usecnt);
1371
1372         return ret;
1373 }
1374 EXPORT_SYMBOL(ib_dereg_mr);
1375
1376 /**
1377  * ib_alloc_mr() - Allocates a memory region
1378  * @pd:            protection domain associated with the region
1379  * @mr_type:       memory region type
1380  * @max_num_sg:    maximum sg entries available for registration.
1381  *
1382  * Notes:
1383  * Memory registeration page/sg lists must not exceed max_num_sg.
1384  * For mr_type IB_MR_TYPE_MEM_REG, the total length cannot exceed
1385  * max_num_sg * used_page_size.
1386  *
1387  */
1388 struct ib_mr *ib_alloc_mr(struct ib_pd *pd,
1389                           enum ib_mr_type mr_type,
1390                           u32 max_num_sg)
1391 {
1392         struct ib_mr *mr;
1393
1394         if (!pd->device->alloc_mr)
1395                 return ERR_PTR(-ENOSYS);
1396
1397         mr = pd->device->alloc_mr(pd, mr_type, max_num_sg);
1398         if (!IS_ERR(mr)) {
1399                 mr->device  = pd->device;
1400                 mr->pd      = pd;
1401                 mr->uobject = NULL;
1402                 atomic_inc(&pd->usecnt);
1403                 mr->need_inval = false;
1404         }
1405
1406         return mr;
1407 }
1408 EXPORT_SYMBOL(ib_alloc_mr);
1409
1410 /* "Fast" memory regions */
1411
1412 struct ib_fmr *ib_alloc_fmr(struct ib_pd *pd,
1413                             int mr_access_flags,
1414                             struct ib_fmr_attr *fmr_attr)
1415 {
1416         struct ib_fmr *fmr;
1417
1418         if (!pd->device->alloc_fmr)
1419                 return ERR_PTR(-ENOSYS);
1420
1421         fmr = pd->device->alloc_fmr(pd, mr_access_flags, fmr_attr);
1422         if (!IS_ERR(fmr)) {
1423                 fmr->device = pd->device;
1424                 fmr->pd     = pd;
1425                 atomic_inc(&pd->usecnt);
1426         }
1427
1428         return fmr;
1429 }
1430 EXPORT_SYMBOL(ib_alloc_fmr);
1431
1432 int ib_unmap_fmr(struct list_head *fmr_list)
1433 {
1434         struct ib_fmr *fmr;
1435
1436         if (list_empty(fmr_list))
1437                 return 0;
1438
1439         fmr = list_entry(fmr_list->next, struct ib_fmr, list);
1440         return fmr->device->unmap_fmr(fmr_list);
1441 }
1442 EXPORT_SYMBOL(ib_unmap_fmr);
1443
1444 int ib_dealloc_fmr(struct ib_fmr *fmr)
1445 {
1446         struct ib_pd *pd;
1447         int ret;
1448
1449         pd = fmr->pd;
1450         ret = fmr->device->dealloc_fmr(fmr);
1451         if (!ret)
1452                 atomic_dec(&pd->usecnt);
1453
1454         return ret;
1455 }
1456 EXPORT_SYMBOL(ib_dealloc_fmr);
1457
1458 /* Multicast groups */
1459
1460 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid)
1461 {
1462         int ret;
1463
1464         if (!qp->device->attach_mcast)
1465                 return -ENOSYS;
1466         if (gid->raw[0] != 0xff || qp->qp_type != IB_QPT_UD)
1467                 return -EINVAL;
1468
1469         ret = qp->device->attach_mcast(qp, gid, lid);
1470         if (!ret)
1471                 atomic_inc(&qp->usecnt);
1472         return ret;
1473 }
1474 EXPORT_SYMBOL(ib_attach_mcast);
1475
1476 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid)
1477 {
1478         int ret;
1479
1480         if (!qp->device->detach_mcast)
1481                 return -ENOSYS;
1482         if (gid->raw[0] != 0xff || qp->qp_type != IB_QPT_UD)
1483                 return -EINVAL;
1484
1485         ret = qp->device->detach_mcast(qp, gid, lid);
1486         if (!ret)
1487                 atomic_dec(&qp->usecnt);
1488         return ret;
1489 }
1490 EXPORT_SYMBOL(ib_detach_mcast);
1491
1492 struct ib_xrcd *ib_alloc_xrcd(struct ib_device *device)
1493 {
1494         struct ib_xrcd *xrcd;
1495
1496         if (!device->alloc_xrcd)
1497                 return ERR_PTR(-ENOSYS);
1498
1499         xrcd = device->alloc_xrcd(device, NULL, NULL);
1500         if (!IS_ERR(xrcd)) {
1501                 xrcd->device = device;
1502                 xrcd->inode = NULL;
1503                 atomic_set(&xrcd->usecnt, 0);
1504                 mutex_init(&xrcd->tgt_qp_mutex);
1505                 INIT_LIST_HEAD(&xrcd->tgt_qp_list);
1506         }
1507
1508         return xrcd;
1509 }
1510 EXPORT_SYMBOL(ib_alloc_xrcd);
1511
1512 int ib_dealloc_xrcd(struct ib_xrcd *xrcd)
1513 {
1514         struct ib_qp *qp;
1515         int ret;
1516
1517         if (atomic_read(&xrcd->usecnt))
1518                 return -EBUSY;
1519
1520         while (!list_empty(&xrcd->tgt_qp_list)) {
1521                 qp = list_entry(xrcd->tgt_qp_list.next, struct ib_qp, xrcd_list);
1522                 ret = ib_destroy_qp(qp);
1523                 if (ret)
1524                         return ret;
1525         }
1526
1527         return xrcd->device->dealloc_xrcd(xrcd);
1528 }
1529 EXPORT_SYMBOL(ib_dealloc_xrcd);
1530
1531 struct ib_flow *ib_create_flow(struct ib_qp *qp,
1532                                struct ib_flow_attr *flow_attr,
1533                                int domain)
1534 {
1535         struct ib_flow *flow_id;
1536         if (!qp->device->create_flow)
1537                 return ERR_PTR(-ENOSYS);
1538
1539         flow_id = qp->device->create_flow(qp, flow_attr, domain);
1540         if (!IS_ERR(flow_id))
1541                 atomic_inc(&qp->usecnt);
1542         return flow_id;
1543 }
1544 EXPORT_SYMBOL(ib_create_flow);
1545
1546 int ib_destroy_flow(struct ib_flow *flow_id)
1547 {
1548         int err;
1549         struct ib_qp *qp = flow_id->qp;
1550
1551         err = qp->device->destroy_flow(flow_id);
1552         if (!err)
1553                 atomic_dec(&qp->usecnt);
1554         return err;
1555 }
1556 EXPORT_SYMBOL(ib_destroy_flow);
1557
1558 int ib_check_mr_status(struct ib_mr *mr, u32 check_mask,
1559                        struct ib_mr_status *mr_status)
1560 {
1561         return mr->device->check_mr_status ?
1562                 mr->device->check_mr_status(mr, check_mask, mr_status) : -ENOSYS;
1563 }
1564 EXPORT_SYMBOL(ib_check_mr_status);
1565
1566 int ib_set_vf_link_state(struct ib_device *device, int vf, u8 port,
1567                          int state)
1568 {
1569         if (!device->set_vf_link_state)
1570                 return -ENOSYS;
1571
1572         return device->set_vf_link_state(device, vf, port, state);
1573 }
1574 EXPORT_SYMBOL(ib_set_vf_link_state);
1575
1576 int ib_get_vf_config(struct ib_device *device, int vf, u8 port,
1577                      struct ifla_vf_info *info)
1578 {
1579         if (!device->get_vf_config)
1580                 return -ENOSYS;
1581
1582         return device->get_vf_config(device, vf, port, info);
1583 }
1584 EXPORT_SYMBOL(ib_get_vf_config);
1585
1586 int ib_get_vf_stats(struct ib_device *device, int vf, u8 port,
1587                     struct ifla_vf_stats *stats)
1588 {
1589         if (!device->get_vf_stats)
1590                 return -ENOSYS;
1591
1592         return device->get_vf_stats(device, vf, port, stats);
1593 }
1594 EXPORT_SYMBOL(ib_get_vf_stats);
1595
1596 int ib_set_vf_guid(struct ib_device *device, int vf, u8 port, u64 guid,
1597                    int type)
1598 {
1599         if (!device->set_vf_guid)
1600                 return -ENOSYS;
1601
1602         return device->set_vf_guid(device, vf, port, guid, type);
1603 }
1604 EXPORT_SYMBOL(ib_set_vf_guid);
1605
1606 /**
1607  * ib_map_mr_sg() - Map the largest prefix of a dma mapped SG list
1608  *     and set it the memory region.
1609  * @mr:            memory region
1610  * @sg:            dma mapped scatterlist
1611  * @sg_nents:      number of entries in sg
1612  * @sg_offset:     offset in bytes into sg
1613  * @page_size:     page vector desired page size
1614  *
1615  * Constraints:
1616  * - The first sg element is allowed to have an offset.
1617  * - Each sg element must be aligned to page_size (or physically
1618  *   contiguous to the previous element). In case an sg element has a
1619  *   non contiguous offset, the mapping prefix will not include it.
1620  * - The last sg element is allowed to have length less than page_size.
1621  * - If sg_nents total byte length exceeds the mr max_num_sge * page_size
1622  *   then only max_num_sg entries will be mapped.
1623  * - If the MR was allocated with type IB_MR_TYPE_SG_GAPS_REG, non of these
1624  *   constraints holds and the page_size argument is ignored.
1625  *
1626  * Returns the number of sg elements that were mapped to the memory region.
1627  *
1628  * After this completes successfully, the  memory region
1629  * is ready for registration.
1630  */
1631 int ib_map_mr_sg(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
1632                 unsigned int sg_offset, unsigned int page_size)
1633 {
1634         if (unlikely(!mr->device->map_mr_sg))
1635                 return -ENOSYS;
1636
1637         mr->page_size = page_size;
1638
1639         return mr->device->map_mr_sg(mr, sg, sg_nents, sg_offset);
1640 }
1641 EXPORT_SYMBOL(ib_map_mr_sg);
1642
1643 /**
1644  * ib_sg_to_pages() - Convert the largest prefix of a sg list
1645  *     to a page vector
1646  * @mr:            memory region
1647  * @sgl:           dma mapped scatterlist
1648  * @sg_nents:      number of entries in sg
1649  * @sg_offset:     offset in bytes into sg
1650  * @set_page:      driver page assignment function pointer
1651  *
1652  * Core service helper for drivers to convert the largest
1653  * prefix of given sg list to a page vector. The sg list
1654  * prefix converted is the prefix that meet the requirements
1655  * of ib_map_mr_sg.
1656  *
1657  * Returns the number of sg elements that were assigned to
1658  * a page vector.
1659  */
1660 int ib_sg_to_pages(struct ib_mr *mr, struct scatterlist *sgl, int sg_nents,
1661                 unsigned int sg_offset, int (*set_page)(struct ib_mr *, u64))
1662 {
1663         struct scatterlist *sg;
1664         u64 last_end_dma_addr = 0;
1665         unsigned int last_page_off = 0;
1666         u64 page_mask = ~((u64)mr->page_size - 1);
1667         int i, ret;
1668
1669         mr->iova = sg_dma_address(&sgl[0]) + sg_offset;
1670         mr->length = 0;
1671
1672         for_each_sg(sgl, sg, sg_nents, i) {
1673                 u64 dma_addr = sg_dma_address(sg) + sg_offset;
1674                 unsigned int dma_len = sg_dma_len(sg) - sg_offset;
1675                 u64 end_dma_addr = dma_addr + dma_len;
1676                 u64 page_addr = dma_addr & page_mask;
1677
1678                 /*
1679                  * For the second and later elements, check whether either the
1680                  * end of element i-1 or the start of element i is not aligned
1681                  * on a page boundary.
1682                  */
1683                 if (i && (last_page_off != 0 || page_addr != dma_addr)) {
1684                         /* Stop mapping if there is a gap. */
1685                         if (last_end_dma_addr != dma_addr)
1686                                 break;
1687
1688                         /*
1689                          * Coalesce this element with the last. If it is small
1690                          * enough just update mr->length. Otherwise start
1691                          * mapping from the next page.
1692                          */
1693                         goto next_page;
1694                 }
1695
1696                 do {
1697                         ret = set_page(mr, page_addr);
1698                         if (unlikely(ret < 0))
1699                                 return i ? : ret;
1700 next_page:
1701                         page_addr += mr->page_size;
1702                 } while (page_addr < end_dma_addr);
1703
1704                 mr->length += dma_len;
1705                 last_end_dma_addr = end_dma_addr;
1706                 last_page_off = end_dma_addr & ~page_mask;
1707
1708                 sg_offset = 0;
1709         }
1710
1711         return i;
1712 }
1713 EXPORT_SYMBOL(ib_sg_to_pages);
1714
1715 struct ib_drain_cqe {
1716         struct ib_cqe cqe;
1717         struct completion done;
1718 };
1719
1720 static void ib_drain_qp_done(struct ib_cq *cq, struct ib_wc *wc)
1721 {
1722         struct ib_drain_cqe *cqe = container_of(wc->wr_cqe, struct ib_drain_cqe,
1723                                                 cqe);
1724
1725         complete(&cqe->done);
1726 }
1727
1728 /*
1729  * Post a WR and block until its completion is reaped for the SQ.
1730  */
1731 static void __ib_drain_sq(struct ib_qp *qp)
1732 {
1733         struct ib_qp_attr attr = { .qp_state = IB_QPS_ERR };
1734         struct ib_drain_cqe sdrain;
1735         struct ib_send_wr swr = {}, *bad_swr;
1736         int ret;
1737
1738         if (qp->send_cq->poll_ctx == IB_POLL_DIRECT) {
1739                 WARN_ONCE(qp->send_cq->poll_ctx == IB_POLL_DIRECT,
1740                           "IB_POLL_DIRECT poll_ctx not supported for drain\n");
1741                 return;
1742         }
1743
1744         swr.wr_cqe = &sdrain.cqe;
1745         sdrain.cqe.done = ib_drain_qp_done;
1746         init_completion(&sdrain.done);
1747
1748         ret = ib_modify_qp(qp, &attr, IB_QP_STATE);
1749         if (ret) {
1750                 WARN_ONCE(ret, "failed to drain send queue: %d\n", ret);
1751                 return;
1752         }
1753
1754         ret = ib_post_send(qp, &swr, &bad_swr);
1755         if (ret) {
1756                 WARN_ONCE(ret, "failed to drain send queue: %d\n", ret);
1757                 return;
1758         }
1759
1760         wait_for_completion(&sdrain.done);
1761 }
1762
1763 /*
1764  * Post a WR and block until its completion is reaped for the RQ.
1765  */
1766 static void __ib_drain_rq(struct ib_qp *qp)
1767 {
1768         struct ib_qp_attr attr = { .qp_state = IB_QPS_ERR };
1769         struct ib_drain_cqe rdrain;
1770         struct ib_recv_wr rwr = {}, *bad_rwr;
1771         int ret;
1772
1773         if (qp->recv_cq->poll_ctx == IB_POLL_DIRECT) {
1774                 WARN_ONCE(qp->recv_cq->poll_ctx == IB_POLL_DIRECT,
1775                           "IB_POLL_DIRECT poll_ctx not supported for drain\n");
1776                 return;
1777         }
1778
1779         rwr.wr_cqe = &rdrain.cqe;
1780         rdrain.cqe.done = ib_drain_qp_done;
1781         init_completion(&rdrain.done);
1782
1783         ret = ib_modify_qp(qp, &attr, IB_QP_STATE);
1784         if (ret) {
1785                 WARN_ONCE(ret, "failed to drain recv queue: %d\n", ret);
1786                 return;
1787         }
1788
1789         ret = ib_post_recv(qp, &rwr, &bad_rwr);
1790         if (ret) {
1791                 WARN_ONCE(ret, "failed to drain recv queue: %d\n", ret);
1792                 return;
1793         }
1794
1795         wait_for_completion(&rdrain.done);
1796 }
1797
1798 /**
1799  * ib_drain_sq() - Block until all SQ CQEs have been consumed by the
1800  *                 application.
1801  * @qp:            queue pair to drain
1802  *
1803  * If the device has a provider-specific drain function, then
1804  * call that.  Otherwise call the generic drain function
1805  * __ib_drain_sq().
1806  *
1807  * The caller must:
1808  *
1809  * ensure there is room in the CQ and SQ for the drain work request and
1810  * completion.
1811  *
1812  * allocate the CQ using ib_alloc_cq() and the CQ poll context cannot be
1813  * IB_POLL_DIRECT.
1814  *
1815  * ensure that there are no other contexts that are posting WRs concurrently.
1816  * Otherwise the drain is not guaranteed.
1817  */
1818 void ib_drain_sq(struct ib_qp *qp)
1819 {
1820         if (qp->device->drain_sq)
1821                 qp->device->drain_sq(qp);
1822         else
1823                 __ib_drain_sq(qp);
1824 }
1825 EXPORT_SYMBOL(ib_drain_sq);
1826
1827 /**
1828  * ib_drain_rq() - Block until all RQ CQEs have been consumed by the
1829  *                 application.
1830  * @qp:            queue pair to drain
1831  *
1832  * If the device has a provider-specific drain function, then
1833  * call that.  Otherwise call the generic drain function
1834  * __ib_drain_rq().
1835  *
1836  * The caller must:
1837  *
1838  * ensure there is room in the CQ and RQ for the drain work request and
1839  * completion.
1840  *
1841  * allocate the CQ using ib_alloc_cq() and the CQ poll context cannot be
1842  * IB_POLL_DIRECT.
1843  *
1844  * ensure that there are no other contexts that are posting WRs concurrently.
1845  * Otherwise the drain is not guaranteed.
1846  */
1847 void ib_drain_rq(struct ib_qp *qp)
1848 {
1849         if (qp->device->drain_rq)
1850                 qp->device->drain_rq(qp);
1851         else
1852                 __ib_drain_rq(qp);
1853 }
1854 EXPORT_SYMBOL(ib_drain_rq);
1855
1856 /**
1857  * ib_drain_qp() - Block until all CQEs have been consumed by the
1858  *                 application on both the RQ and SQ.
1859  * @qp:            queue pair to drain
1860  *
1861  * The caller must:
1862  *
1863  * ensure there is room in the CQ(s), SQ, and RQ for drain work requests
1864  * and completions.
1865  *
1866  * allocate the CQs using ib_alloc_cq() and the CQ poll context cannot be
1867  * IB_POLL_DIRECT.
1868  *
1869  * ensure that there are no other contexts that are posting WRs concurrently.
1870  * Otherwise the drain is not guaranteed.
1871  */
1872 void ib_drain_qp(struct ib_qp *qp)
1873 {
1874         ib_drain_sq(qp);
1875         if (!qp->srq)
1876                 ib_drain_rq(qp);
1877 }
1878 EXPORT_SYMBOL(ib_drain_qp);