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 #include <rdma/rw.h>
  52
  53 #include "core_priv.h"
  54
  55 static const char * const ib_events[] = {
  56         [IB_EVENT_CQ_ERR]               = "CQ error",
  57         [IB_EVENT_QP_FATAL]             = "QP fatal error",
  58         [IB_EVENT_QP_REQ_ERR]           = "QP request error",
  59         [IB_EVENT_QP_ACCESS_ERR]        = "QP access error",
  60         [IB_EVENT_COMM_EST]             = "communication established",
  61         [IB_EVENT_SQ_DRAINED]           = "send queue drained",
  62         [IB_EVENT_PATH_MIG]             = "path migration successful",
  63         [IB_EVENT_PATH_MIG_ERR]         = "path migration error",
  64         [IB_EVENT_DEVICE_FATAL]         = "device fatal error",
  65         [IB_EVENT_PORT_ACTIVE]          = "port active",
  66         [IB_EVENT_PORT_ERR]             = "port error",
  67         [IB_EVENT_LID_CHANGE]           = "LID change",
  68         [IB_EVENT_PKEY_CHANGE]          = "P_key change",
  69         [IB_EVENT_SM_CHANGE]            = "SM change",
  70         [IB_EVENT_SRQ_ERR]              = "SRQ error",
  71         [IB_EVENT_SRQ_LIMIT_REACHED]    = "SRQ limit reached",
  72         [IB_EVENT_QP_LAST_WQE_REACHED]  = "last WQE reached",
  73         [IB_EVENT_CLIENT_REREGISTER]    = "client reregister",
  74         [IB_EVENT_GID_CHANGE]           = "GID changed",
  75 };
  76
  77 const char *__attribute_const__ ib_event_msg(enum ib_event_type event)
  78 {
  79         size_t index = event;
  80
  81         return (index < ARRAY_SIZE(ib_events) && ib_events[index]) ?
  82                         ib_events[index] : "unrecognized event";
  83 }
  84 EXPORT_SYMBOL(ib_event_msg);
  85
  86 static const char * const wc_statuses[] = {
  87         [IB_WC_SUCCESS]                 = "success",
  88         [IB_WC_LOC_LEN_ERR]             = "local length error",
  89         [IB_WC_LOC_QP_OP_ERR]           = "local QP operation error",
  90         [IB_WC_LOC_EEC_OP_ERR]          = "local EE context operation error",
  91         [IB_WC_LOC_PROT_ERR]            = "local protection error",
  92         [IB_WC_WR_FLUSH_ERR]            = "WR flushed",
  93         [IB_WC_MW_BIND_ERR]             = "memory management operation error",
  94         [IB_WC_BAD_RESP_ERR]            = "bad response error",
  95         [IB_WC_LOC_ACCESS_ERR]          = "local access error",
  96         [IB_WC_REM_INV_REQ_ERR]         = "invalid request error",
  97         [IB_WC_REM_ACCESS_ERR]          = "remote access error",
  98         [IB_WC_REM_OP_ERR]              = "remote operation error",
  99         [IB_WC_RETRY_EXC_ERR]           = "transport retry counter exceeded",
 100         [IB_WC_RNR_RETRY_EXC_ERR]       = "RNR retry counter exceeded",
 101         [IB_WC_LOC_RDD_VIOL_ERR]        = "local RDD violation error",
 102         [IB_WC_REM_INV_RD_REQ_ERR]      = "remote invalid RD request",
 103         [IB_WC_REM_ABORT_ERR]           = "operation aborted",
 104         [IB_WC_INV_EECN_ERR]            = "invalid EE context number",
 105         [IB_WC_INV_EEC_STATE_ERR]       = "invalid EE context state",
 106         [IB_WC_FATAL_ERR]               = "fatal error",
 107         [IB_WC_RESP_TIMEOUT_ERR]        = "response timeout error",
 108         [IB_WC_GENERAL_ERR]             = "general error",
 109 };
 110
 111 const char *__attribute_const__ ib_wc_status_msg(enum ib_wc_status status)
 112 {
 113         size_t index = status;
 114
 115         return (index < ARRAY_SIZE(wc_statuses) && wc_statuses[index]) ?
 116                         wc_statuses[index] : "unrecognized status";
 117 }
 118 EXPORT_SYMBOL(ib_wc_status_msg);
 119
 120 __attribute_const__ int ib_rate_to_mult(enum ib_rate rate)
 121 {
 122         switch (rate) {
 123         case IB_RATE_2_5_GBPS: return  1;
 124         case IB_RATE_5_GBPS:   return  2;
 125         case IB_RATE_10_GBPS:  return  4;
 126         case IB_RATE_20_GBPS:  return  8;
 127         case IB_RATE_30_GBPS:  return 12;
 128         case IB_RATE_40_GBPS:  return 16;
 129         case IB_RATE_60_GBPS:  return 24;
 130         case IB_RATE_80_GBPS:  return 32;
 131         case IB_RATE_120_GBPS: return 48;
 132         default:               return -1;
 133         }
 134 }
 135 EXPORT_SYMBOL(ib_rate_to_mult);
 136
 137 __attribute_const__ enum ib_rate mult_to_ib_rate(int mult)
 138 {
 139         switch (mult) {
 140         case 1:  return IB_RATE_2_5_GBPS;
 141         case 2:  return IB_RATE_5_GBPS;
 142         case 4:  return IB_RATE_10_GBPS;
 143         case 8:  return IB_RATE_20_GBPS;
 144         case 12: return IB_RATE_30_GBPS;
 145         case 16: return IB_RATE_40_GBPS;
 146         case 24: return IB_RATE_60_GBPS;
 147         case 32: return IB_RATE_80_GBPS;
 148         case 48: return IB_RATE_120_GBPS;
 149         default: return IB_RATE_PORT_CURRENT;
 150         }
 151 }
 152 EXPORT_SYMBOL(mult_to_ib_rate);
 153
 154 __attribute_const__ int ib_rate_to_mbps(enum ib_rate rate)
 155 {
 156         switch (rate) {
 157         case IB_RATE_2_5_GBPS: return 2500;
 158         case IB_RATE_5_GBPS:   return 5000;
 159         case IB_RATE_10_GBPS:  return 10000;
 160         case IB_RATE_20_GBPS:  return 20000;
 161         case IB_RATE_30_GBPS:  return 30000;
 162         case IB_RATE_40_GBPS:  return 40000;
 163         case IB_RATE_60_GBPS:  return 60000;
 164         case IB_RATE_80_GBPS:  return 80000;
 165         case IB_RATE_120_GBPS: return 120000;
 166         case IB_RATE_14_GBPS:  return 14062;
 167         case IB_RATE_56_GBPS:  return 56250;
 168         case IB_RATE_112_GBPS: return 112500;
 169         case IB_RATE_168_GBPS: return 168750;
 170         case IB_RATE_25_GBPS:  return 25781;
 171         case IB_RATE_100_GBPS: return 103125;
 172         case IB_RATE_200_GBPS: return 206250;
 173         case IB_RATE_300_GBPS: return 309375;
 174         default:               return -1;
 175         }
 176 }
 177 EXPORT_SYMBOL(ib_rate_to_mbps);
 178
 179 __attribute_const__ enum rdma_transport_type
 180 rdma_node_get_transport(enum rdma_node_type node_type)
 181 {
 182         switch (node_type) {
 183         case RDMA_NODE_IB_CA:
 184         case RDMA_NODE_IB_SWITCH:
 185         case RDMA_NODE_IB_ROUTER:
 186                 return RDMA_TRANSPORT_IB;
 187         case RDMA_NODE_RNIC:
 188                 return RDMA_TRANSPORT_IWARP;
 189         case RDMA_NODE_USNIC:
 190                 return RDMA_TRANSPORT_USNIC;
 191         case RDMA_NODE_USNIC_UDP:
 192                 return RDMA_TRANSPORT_USNIC_UDP;
 193         default:
 194                 BUG();
 195                 return 0;
 196         }
 197 }
 198 EXPORT_SYMBOL(rdma_node_get_transport);
 199
 200 enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device, u8 port_num)
 201 {
 202         if (device->get_link_layer)
 203                 return device->get_link_layer(device, port_num);
 204
 205         switch (rdma_node_get_transport(device->node_type)) {
 206         case RDMA_TRANSPORT_IB:
 207                 return IB_LINK_LAYER_INFINIBAND;
 208         case RDMA_TRANSPORT_IWARP:
 209         case RDMA_TRANSPORT_USNIC:
 210         case RDMA_TRANSPORT_USNIC_UDP:
 211                 return IB_LINK_LAYER_ETHERNET;
 212         default:
 213                 return IB_LINK_LAYER_UNSPECIFIED;
 214         }
 215 }
 216 EXPORT_SYMBOL(rdma_port_get_link_layer);
 217
 218 /* Protection domains */
 219
 220 /**
 221  * ib_alloc_pd - Allocates an unused protection domain.
 222  * @device: The device on which to allocate the protection domain.
 223  *
 224  * A protection domain object provides an association between QPs, shared
 225  * receive queues, address handles, memory regions, and memory windows.
 226  *
 227  * Every PD has a local_dma_lkey which can be used as the lkey value for local
 228  * memory operations.
 229  */
 230 struct ib_pd *ib_alloc_pd(struct ib_device *device)
 231 {
 232         struct ib_pd *pd;
 233
 234         pd = device->alloc_pd(device, NULL, NULL);
 235         if (IS_ERR(pd))
 236                 return pd;
 237
 238         pd->device = device;
 239         pd->uobject = NULL;
 240         pd->local_mr = NULL;
 241         atomic_set(&pd->usecnt, 0);
 242
 243         if (device->attrs.device_cap_flags & IB_DEVICE_LOCAL_DMA_LKEY)
 244                 pd->local_dma_lkey = device->local_dma_lkey;
 245         else {
 246                 struct ib_mr *mr;
 247
 248                 mr = ib_get_dma_mr(pd, IB_ACCESS_LOCAL_WRITE);
 249                 if (IS_ERR(mr)) {
 250                         ib_dealloc_pd(pd);
 251                         return (struct ib_pd *)mr;
 252                 }
 253
 254                 pd->local_mr = mr;
 255                 pd->local_dma_lkey = pd->local_mr->lkey;
 256         }
 257         return pd;
 258 }
 259 EXPORT_SYMBOL(ib_alloc_pd);
 260
 261 /**
 262  * ib_dealloc_pd - Deallocates a protection domain.
 263  * @pd: The protection domain to deallocate.
 264  *
 265  * It is an error to call this function while any resources in the pd still
 266  * exist.  The caller is responsible to synchronously destroy them and
 267  * guarantee no new allocations will happen.
 268  */
 269 void ib_dealloc_pd(struct ib_pd *pd)
 270 {
 271         int ret;
 272
 273         if (pd->local_mr) {
 274                 ret = ib_dereg_mr(pd->local_mr);
 275                 WARN_ON(ret);
 276                 pd->local_mr = NULL;
 277         }
 278
 279         /* uverbs manipulates usecnt with proper locking, while the kabi
 280            requires the caller to guarantee we can't race here. */
 281         WARN_ON(atomic_read(&pd->usecnt));
 282
 283         /* Making delalloc_pd a void return is a WIP, no driver should return
 284            an error here. */
 285         ret = pd->device->dealloc_pd(pd);
 286         WARN_ONCE(ret, "Infiniband HW driver failed dealloc_pd");
 287 }
 288 EXPORT_SYMBOL(ib_dealloc_pd);
 289
 290 /* Address handles */
 291
 292 struct ib_ah *ib_create_ah(struct ib_pd *pd, struct ib_ah_attr *ah_attr)
 293 {
 294         struct ib_ah *ah;
 295
 296         ah = pd->device->create_ah(pd, ah_attr);
 297
 298         if (!IS_ERR(ah)) {
 299                 ah->device  = pd->device;
 300                 ah->pd      = pd;
 301                 ah->uobject = NULL;
 302                 atomic_inc(&pd->usecnt);
 303         }
 304
 305         return ah;
 306 }
 307 EXPORT_SYMBOL(ib_create_ah);
 308
 309 static int ib_get_header_version(const union rdma_network_hdr *hdr)
 310 {
 311         const struct iphdr *ip4h = (struct iphdr *)&hdr->roce4grh;
 312         struct iphdr ip4h_checked;
 313         const struct ipv6hdr *ip6h = (struct ipv6hdr *)&hdr->ibgrh;
 314
 315         /* If it's IPv6, the version must be 6, otherwise, the first
 316          * 20 bytes (before the IPv4 header) are garbled.
 317          */
 318         if (ip6h->version != 6)
 319                 return (ip4h->version == 4) ? 4 : 0;
 320         /* version may be 6 or 4 because the first 20 bytes could be garbled */
 321
 322         /* RoCE v2 requires no options, thus header length
 323          * must be 5 words
 324          */
 325         if (ip4h->ihl != 5)
 326                 return 6;
 327
 328         /* Verify checksum.
 329          * We can't write on scattered buffers so we need to copy to
 330          * temp buffer.
 331          */
 332         memcpy(&ip4h_checked, ip4h, sizeof(ip4h_checked));
 333         ip4h_checked.check = 0;
 334         ip4h_checked.check = ip_fast_csum((u8 *)&ip4h_checked, 5);
 335         /* if IPv4 header checksum is OK, believe it */
 336         if (ip4h->check == ip4h_checked.check)
 337                 return 4;
 338         return 6;
 339 }
 340
 341 static enum rdma_network_type ib_get_net_type_by_grh(struct ib_device *device,
 342                                                      u8 port_num,
 343                                                      const struct ib_grh *grh)
 344 {
 345         int grh_version;
 346
 347         if (rdma_protocol_ib(device, port_num))
 348                 return RDMA_NETWORK_IB;
 349
 350         grh_version = ib_get_header_version((union rdma_network_hdr *)grh);
 351
 352         if (grh_version == 4)
 353                 return RDMA_NETWORK_IPV4;
 354
 355         if (grh->next_hdr == IPPROTO_UDP)
 356                 return RDMA_NETWORK_IPV6;
 357
 358         return RDMA_NETWORK_ROCE_V1;
 359 }
 360
 361 struct find_gid_index_context {
 362         u16 vlan_id;
 363         enum ib_gid_type gid_type;
 364 };
 365
 366 static bool find_gid_index(const union ib_gid *gid,
 367                            const struct ib_gid_attr *gid_attr,
 368                            void *context)
 369 {
 370         struct find_gid_index_context *ctx =
 371                 (struct find_gid_index_context *)context;
 372
 373         if (ctx->gid_type != gid_attr->gid_type)
 374                 return false;
 375
 376         if ((!!(ctx->vlan_id != 0xffff) == !is_vlan_dev(gid_attr->ndev)) ||
 377             (is_vlan_dev(gid_attr->ndev) &&
 378              vlan_dev_vlan_id(gid_attr->ndev) != ctx->vlan_id))
 379                 return false;
 380
 381         return true;
 382 }
 383
 384 static int get_sgid_index_from_eth(struct ib_device *device, u8 port_num,
 385                                    u16 vlan_id, const union ib_gid *sgid,
 386                                    enum ib_gid_type gid_type,
 387                                    u16 *gid_index)
 388 {
 389         struct find_gid_index_context context = {.vlan_id = vlan_id,
 390                                                  .gid_type = gid_type};
 391
 392         return ib_find_gid_by_filter(device, sgid, port_num, find_gid_index,
 393                                      &context, gid_index);
 394 }
 395
 396 static int get_gids_from_rdma_hdr(union rdma_network_hdr *hdr,
 397                                   enum rdma_network_type net_type,
 398                                   union ib_gid *sgid, union ib_gid *dgid)
 399 {
 400         struct sockaddr_in  src_in;
 401         struct sockaddr_in  dst_in;
 402         __be32 src_saddr, dst_saddr;
 403
 404         if (!sgid || !dgid)
 405                 return -EINVAL;
 406
 407         if (net_type == RDMA_NETWORK_IPV4) {
 408                 memcpy(&src_in.sin_addr.s_addr,
 409                        &hdr->roce4grh.saddr, 4);
 410                 memcpy(&dst_in.sin_addr.s_addr,
 411                        &hdr->roce4grh.daddr, 4);
 412                 src_saddr = src_in.sin_addr.s_addr;
 413                 dst_saddr = dst_in.sin_addr.s_addr;
 414                 ipv6_addr_set_v4mapped(src_saddr,
 415                                        (struct in6_addr *)sgid);
 416                 ipv6_addr_set_v4mapped(dst_saddr,
 417                                        (struct in6_addr *)dgid);
 418                 return 0;
 419         } else if (net_type == RDMA_NETWORK_IPV6 ||
 420                    net_type == RDMA_NETWORK_IB) {
 421                 *dgid = hdr->ibgrh.dgid;
 422                 *sgid = hdr->ibgrh.sgid;
 423                 return 0;
 424         } else {
 425                 return -EINVAL;
 426         }
 427 }
 428
 429 int ib_init_ah_from_wc(struct ib_device *device, u8 port_num,
 430                        const struct ib_wc *wc, const struct ib_grh *grh,
 431                        struct ib_ah_attr *ah_attr)
 432 {
 433         u32 flow_class;
 434         u16 gid_index;
 435         int ret;
 436         enum rdma_network_type net_type = RDMA_NETWORK_IB;
 437         enum ib_gid_type gid_type = IB_GID_TYPE_IB;
 438         int hoplimit = 0xff;
 439         union ib_gid dgid;
 440         union ib_gid sgid;
 441
 442         memset(ah_attr, 0, sizeof *ah_attr);
 443         if (rdma_cap_eth_ah(device, port_num)) {
 444                 if (wc->wc_flags & IB_WC_WITH_NETWORK_HDR_TYPE)
 445                         net_type = wc->network_hdr_type;
 446                 else
 447                         net_type = ib_get_net_type_by_grh(device, port_num, grh);
 448                 gid_type = ib_network_to_gid_type(net_type);
 449         }
 450         ret = get_gids_from_rdma_hdr((union rdma_network_hdr *)grh, net_type,
 451                                      &sgid, &dgid);
 452         if (ret)
 453                 return ret;
 454
 455         if (rdma_protocol_roce(device, port_num)) {
 456                 int if_index = 0;
 457                 u16 vlan_id = wc->wc_flags & IB_WC_WITH_VLAN ?
 458                                 wc->vlan_id : 0xffff;
 459                 struct net_device *idev;
 460                 struct net_device *resolved_dev;
 461
 462                 if (!(wc->wc_flags & IB_WC_GRH))
 463                         return -EPROTOTYPE;
 464
 465                 if (!device->get_netdev)
 466                         return -EOPNOTSUPP;
 467
 468                 idev = device->get_netdev(device, port_num);
 469                 if (!idev)
 470                         return -ENODEV;
 471
 472                 ret = rdma_addr_find_l2_eth_by_grh(&dgid, &sgid,
 473                                                    ah_attr->dmac,
 474                                                    wc->wc_flags & IB_WC_WITH_VLAN ?
 475                                                    NULL : &vlan_id,
 476                                                    &if_index, &hoplimit);
 477                 if (ret) {
 478                         dev_put(idev);
 479                         return ret;
 480                 }
 481
 482                 resolved_dev = dev_get_by_index(&init_net, if_index);
 483                 if (resolved_dev->flags & IFF_LOOPBACK) {
 484                         dev_put(resolved_dev);
 485                         resolved_dev = idev;
 486                         dev_hold(resolved_dev);
 487                 }
 488                 rcu_read_lock();
 489                 if (resolved_dev != idev && !rdma_is_upper_dev_rcu(idev,
 490                                                                    resolved_dev))
 491                         ret = -EHOSTUNREACH;
 492                 rcu_read_unlock();
 493                 dev_put(idev);
 494                 dev_put(resolved_dev);
 495                 if (ret)
 496                         return ret;
 497
 498                 ret = get_sgid_index_from_eth(device, port_num, vlan_id,
 499                                               &dgid, gid_type, &gid_index);
 500                 if (ret)
 501                         return ret;
 502         }
 503
 504         ah_attr->dlid = wc->slid;
 505         ah_attr->sl = wc->sl;
 506         ah_attr->src_path_bits = wc->dlid_path_bits;
 507         ah_attr->port_num = port_num;
 508
 509         if (wc->wc_flags & IB_WC_GRH) {
 510                 ah_attr->ah_flags = IB_AH_GRH;
 511                 ah_attr->grh.dgid = sgid;
 512
 513                 if (!rdma_cap_eth_ah(device, port_num)) {
 514                         if (dgid.global.interface_id != cpu_to_be64(IB_SA_WELL_KNOWN_GUID)) {
 515                                 ret = ib_find_cached_gid_by_port(device, &dgid,
 516                                                                  IB_GID_TYPE_IB,
 517                                                                  port_num, NULL,
 518                                                                  &gid_index);
 519                                 if (ret)
 520                                         return ret;
 521                         } else {
 522                                 gid_index = 0;
 523                         }
 524                 }
 525
 526                 ah_attr->grh.sgid_index = (u8) gid_index;
 527                 flow_class = be32_to_cpu(grh->version_tclass_flow);
 528                 ah_attr->grh.flow_label = flow_class & 0xFFFFF;
 529                 ah_attr->grh.hop_limit = hoplimit;
 530                 ah_attr->grh.traffic_class = (flow_class >> 20) & 0xFF;
 531         }
 532         return 0;
 533 }
 534 EXPORT_SYMBOL(ib_init_ah_from_wc);
 535
 536 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, const struct ib_wc *wc,
 537                                    const struct ib_grh *grh, u8 port_num)
 538 {
 539         struct ib_ah_attr ah_attr;
 540         int ret;
 541
 542         ret = ib_init_ah_from_wc(pd->device, port_num, wc, grh, &ah_attr);
 543         if (ret)
 544                 return ERR_PTR(ret);
 545
 546         return ib_create_ah(pd, &ah_attr);
 547 }
 548 EXPORT_SYMBOL(ib_create_ah_from_wc);
 549
 550 int ib_modify_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr)
 551 {
 552         return ah->device->modify_ah ?
 553                 ah->device->modify_ah(ah, ah_attr) :
 554                 -ENOSYS;
 555 }
 556 EXPORT_SYMBOL(ib_modify_ah);
 557
 558 int ib_query_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr)
 559 {
 560         return ah->device->query_ah ?
 561                 ah->device->query_ah(ah, ah_attr) :
 562                 -ENOSYS;
 563 }
 564 EXPORT_SYMBOL(ib_query_ah);
 565
 566 int ib_destroy_ah(struct ib_ah *ah)
 567 {
 568         struct ib_pd *pd;
 569         int ret;
 570
 571         pd = ah->pd;
 572         ret = ah->device->destroy_ah(ah);
 573         if (!ret)
 574                 atomic_dec(&pd->usecnt);
 575
 576         return ret;
 577 }
 578 EXPORT_SYMBOL(ib_destroy_ah);
 579
 580 /* Shared receive queues */
 581
 582 struct ib_srq *ib_create_srq(struct ib_pd *pd,
 583                              struct ib_srq_init_attr *srq_init_attr)
 584 {
 585         struct ib_srq *srq;
 586
 587         if (!pd->device->create_srq)
 588                 return ERR_PTR(-ENOSYS);
 589
 590         srq = pd->device->create_srq(pd, srq_init_attr, NULL);
 591
 592         if (!IS_ERR(srq)) {
 593                 srq->device        = pd->device;
 594                 srq->pd            = pd;
 595                 srq->uobject       = NULL;
 596                 srq->event_handler = srq_init_attr->event_handler;
 597                 srq->srq_context   = srq_init_attr->srq_context;
 598                 srq->srq_type      = srq_init_attr->srq_type;
 599                 if (srq->srq_type == IB_SRQT_XRC) {
 600                         srq->ext.xrc.xrcd = srq_init_attr->ext.xrc.xrcd;
 601                         srq->ext.xrc.cq   = srq_init_attr->ext.xrc.cq;
 602                         atomic_inc(&srq->ext.xrc.xrcd->usecnt);
 603                         atomic_inc(&srq->ext.xrc.cq->usecnt);
 604                 }
 605                 atomic_inc(&pd->usecnt);
 606                 atomic_set(&srq->usecnt, 0);
 607         }
 608
 609         return srq;
 610 }
 611 EXPORT_SYMBOL(ib_create_srq);
 612
 613 int ib_modify_srq(struct ib_srq *srq,
 614                   struct ib_srq_attr *srq_attr,
 615                   enum ib_srq_attr_mask srq_attr_mask)
 616 {
 617         return srq->device->modify_srq ?
 618                 srq->device->modify_srq(srq, srq_attr, srq_attr_mask, NULL) :
 619                 -ENOSYS;
 620 }
 621 EXPORT_SYMBOL(ib_modify_srq);
 622
 623 int ib_query_srq(struct ib_srq *srq,
 624                  struct ib_srq_attr *srq_attr)
 625 {
 626         return srq->device->query_srq ?
 627                 srq->device->query_srq(srq, srq_attr) : -ENOSYS;
 628 }
 629 EXPORT_SYMBOL(ib_query_srq);
 630
 631 int ib_destroy_srq(struct ib_srq *srq)
 632 {
 633         struct ib_pd *pd;
 634         enum ib_srq_type srq_type;
 635         struct ib_xrcd *uninitialized_var(xrcd);
 636         struct ib_cq *uninitialized_var(cq);
 637         int ret;
 638
 639         if (atomic_read(&srq->usecnt))
 640                 return -EBUSY;
 641
 642         pd = srq->pd;
 643         srq_type = srq->srq_type;
 644         if (srq_type == IB_SRQT_XRC) {
 645                 xrcd = srq->ext.xrc.xrcd;
 646                 cq = srq->ext.xrc.cq;
 647         }
 648
 649         ret = srq->device->destroy_srq(srq);
 650         if (!ret) {
 651                 atomic_dec(&pd->usecnt);
 652                 if (srq_type == IB_SRQT_XRC) {
 653                         atomic_dec(&xrcd->usecnt);
 654                         atomic_dec(&cq->usecnt);
 655                 }
 656         }
 657
 658         return ret;
 659 }
 660 EXPORT_SYMBOL(ib_destroy_srq);
 661
 662 /* Queue pairs */
 663
 664 static void __ib_shared_qp_event_handler(struct ib_event *event, void *context)
 665 {
 666         struct ib_qp *qp = context;
 667         unsigned long flags;
 668
 669         spin_lock_irqsave(&qp->device->event_handler_lock, flags);
 670         list_for_each_entry(event->element.qp, &qp->open_list, open_list)
 671                 if (event->element.qp->event_handler)
 672                         event->element.qp->event_handler(event, event->element.qp->qp_context);
 673         spin_unlock_irqrestore(&qp->device->event_handler_lock, flags);
 674 }
 675
 676 static void __ib_insert_xrcd_qp(struct ib_xrcd *xrcd, struct ib_qp *qp)
 677 {
 678         mutex_lock(&xrcd->tgt_qp_mutex);
 679         list_add(&qp->xrcd_list, &xrcd->tgt_qp_list);
 680         mutex_unlock(&xrcd->tgt_qp_mutex);
 681 }
 682
 683 static struct ib_qp *__ib_open_qp(struct ib_qp *real_qp,
 684                                   void (*event_handler)(struct ib_event *, void *),
 685                                   void *qp_context)
 686 {
 687         struct ib_qp *qp;
 688         unsigned long flags;
 689
 690         qp = kzalloc(sizeof *qp, GFP_KERNEL);
 691         if (!qp)
 692                 return ERR_PTR(-ENOMEM);
 693
 694         qp->real_qp = real_qp;
 695         atomic_inc(&real_qp->usecnt);
 696         qp->device = real_qp->device;
 697         qp->event_handler = event_handler;
 698         qp->qp_context = qp_context;
 699         qp->qp_num = real_qp->qp_num;
 700         qp->qp_type = real_qp->qp_type;
 701
 702         spin_lock_irqsave(&real_qp->device->event_handler_lock, flags);
 703         list_add(&qp->open_list, &real_qp->open_list);
 704         spin_unlock_irqrestore(&real_qp->device->event_handler_lock, flags);
 705
 706         return qp;
 707 }
 708
 709 struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
 710                          struct ib_qp_open_attr *qp_open_attr)
 711 {
 712         struct ib_qp *qp, *real_qp;
 713
 714         if (qp_open_attr->qp_type != IB_QPT_XRC_TGT)
 715                 return ERR_PTR(-EINVAL);
 716
 717         qp = ERR_PTR(-EINVAL);
 718         mutex_lock(&xrcd->tgt_qp_mutex);
 719         list_for_each_entry(real_qp, &xrcd->tgt_qp_list, xrcd_list) {
 720                 if (real_qp->qp_num == qp_open_attr->qp_num) {
 721                         qp = __ib_open_qp(real_qp, qp_open_attr->event_handler,
 722                                           qp_open_attr->qp_context);
 723                         break;
 724                 }
 725         }
 726         mutex_unlock(&xrcd->tgt_qp_mutex);
 727         return qp;
 728 }
 729 EXPORT_SYMBOL(ib_open_qp);
 730
 731 static struct ib_qp *ib_create_xrc_qp(struct ib_qp *qp,
 732                 struct ib_qp_init_attr *qp_init_attr)
 733 {
 734         struct ib_qp *real_qp = qp;
 735
 736         qp->event_handler = __ib_shared_qp_event_handler;
 737         qp->qp_context = qp;
 738         qp->pd = NULL;
 739         qp->send_cq = qp->recv_cq = NULL;
 740         qp->srq = NULL;
 741         qp->xrcd = qp_init_attr->xrcd;
 742         atomic_inc(&qp_init_attr->xrcd->usecnt);
 743         INIT_LIST_HEAD(&qp->open_list);
 744
 745         qp = __ib_open_qp(real_qp, qp_init_attr->event_handler,
 746                           qp_init_attr->qp_context);
 747         if (!IS_ERR(qp))
 748                 __ib_insert_xrcd_qp(qp_init_attr->xrcd, real_qp);
 749         else
 750                 real_qp->device->destroy_qp(real_qp);
 751         return qp;
 752 }
 753
 754 struct ib_qp *ib_create_qp(struct ib_pd *pd,
 755                            struct ib_qp_init_attr *qp_init_attr)
 756 {
 757         struct ib_device *device = pd ? pd->device : qp_init_attr->xrcd->device;
 758         struct ib_qp *qp;
 759         int ret;
 760
 761         if (qp_init_attr->rwq_ind_tbl &&
 762             (qp_init_attr->recv_cq ||
 763             qp_init_attr->srq || qp_init_attr->cap.max_recv_wr ||
 764             qp_init_attr->cap.max_recv_sge))
 765                 return ERR_PTR(-EINVAL);
 766
 767         /*
 768          * If the callers is using the RDMA API calculate the resources
 769          * needed for the RDMA READ/WRITE operations.
 770          *
 771          * Note that these callers need to pass in a port number.
 772          */
 773         if (qp_init_attr->cap.max_rdma_ctxs)
 774                 rdma_rw_init_qp(device, qp_init_attr);
 775
 776         qp = device->create_qp(pd, qp_init_attr, NULL);
 777         if (IS_ERR(qp))
 778                 return qp;
 779
 780         qp->device     = device;
 781         qp->real_qp    = qp;
 782         qp->uobject    = NULL;
 783         qp->qp_type    = qp_init_attr->qp_type;
 784         qp->rwq_ind_tbl = qp_init_attr->rwq_ind_tbl;
 785
 786         atomic_set(&qp->usecnt, 0);
 787         qp->mrs_used = 0;
 788         spin_lock_init(&qp->mr_lock);
 789         INIT_LIST_HEAD(&qp->rdma_mrs);
 790         INIT_LIST_HEAD(&qp->sig_mrs);
 791
 792         if (qp_init_attr->qp_type == IB_QPT_XRC_TGT)
 793                 return ib_create_xrc_qp(qp, qp_init_attr);
 794
 795         qp->event_handler = qp_init_attr->event_handler;
 796         qp->qp_context = qp_init_attr->qp_context;
 797         if (qp_init_attr->qp_type == IB_QPT_XRC_INI) {
 798                 qp->recv_cq = NULL;
 799                 qp->srq = NULL;
 800         } else {
 801                 qp->recv_cq = qp_init_attr->recv_cq;
 802                 if (qp_init_attr->recv_cq)
 803                         atomic_inc(&qp_init_attr->recv_cq->usecnt);
 804                 qp->srq = qp_init_attr->srq;
 805                 if (qp->srq)
 806                         atomic_inc(&qp_init_attr->srq->usecnt);
 807         }
 808
 809         qp->pd      = pd;
 810         qp->send_cq = qp_init_attr->send_cq;
 811         qp->xrcd    = NULL;
 812
 813         atomic_inc(&pd->usecnt);
 814         if (qp_init_attr->send_cq)
 815                 atomic_inc(&qp_init_attr->send_cq->usecnt);
 816         if (qp_init_attr->rwq_ind_tbl)
 817                 atomic_inc(&qp->rwq_ind_tbl->usecnt);
 818
 819         if (qp_init_attr->cap.max_rdma_ctxs) {
 820                 ret = rdma_rw_init_mrs(qp, qp_init_attr);
 821                 if (ret) {
 822                         pr_err("failed to init MR pool ret= %d\n", ret);
 823                         ib_destroy_qp(qp);
 824                         qp = ERR_PTR(ret);
 825                 }
 826         }
 827
 828         return qp;
 829 }
 830 EXPORT_SYMBOL(ib_create_qp);
 831
 832 static const struct {
 833         int                     valid;
 834         enum ib_qp_attr_mask    req_param[IB_QPT_MAX];
 835         enum ib_qp_attr_mask    opt_param[IB_QPT_MAX];
 836 } qp_state_table[IB_QPS_ERR + 1][IB_QPS_ERR + 1] = {
 837         [IB_QPS_RESET] = {
 838                 [IB_QPS_RESET] = { .valid = 1 },
 839                 [IB_QPS_INIT]  = {
 840                         .valid = 1,
 841                         .req_param = {
 842                                 [IB_QPT_UD]  = (IB_QP_PKEY_INDEX                |
 843                                                 IB_QP_PORT                      |
 844                                                 IB_QP_QKEY),
 845                                 [IB_QPT_RAW_PACKET] = IB_QP_PORT,
 846                                 [IB_QPT_UC]  = (IB_QP_PKEY_INDEX                |
 847                                                 IB_QP_PORT                      |
 848                                                 IB_QP_ACCESS_FLAGS),
 849                                 [IB_QPT_RC]  = (IB_QP_PKEY_INDEX                |
 850                                                 IB_QP_PORT                      |
 851                                                 IB_QP_ACCESS_FLAGS),
 852                                 [IB_QPT_XRC_INI] = (IB_QP_PKEY_INDEX            |
 853                                                 IB_QP_PORT                      |
 854                                                 IB_QP_ACCESS_FLAGS),
 855                                 [IB_QPT_XRC_TGT] = (IB_QP_PKEY_INDEX            |
 856                                                 IB_QP_PORT                      |
 857                                                 IB_QP_ACCESS_FLAGS),
 858                                 [IB_QPT_SMI] = (IB_QP_PKEY_INDEX                |
 859                                                 IB_QP_QKEY),
 860                                 [IB_QPT_GSI] = (IB_QP_PKEY_INDEX                |
 861                                                 IB_QP_QKEY),
 862                         }
 863                 },
 864         },
 865         [IB_QPS_INIT]  = {
 866                 [IB_QPS_RESET] = { .valid = 1 },
 867                 [IB_QPS_ERR] =   { .valid = 1 },
 868                 [IB_QPS_INIT]  = {
 869                         .valid = 1,
 870                         .opt_param = {
 871                                 [IB_QPT_UD]  = (IB_QP_PKEY_INDEX                |
 872                                                 IB_QP_PORT                      |
 873                                                 IB_QP_QKEY),
 874                                 [IB_QPT_UC]  = (IB_QP_PKEY_INDEX                |
 875                                                 IB_QP_PORT                      |
 876                                                 IB_QP_ACCESS_FLAGS),
 877                                 [IB_QPT_RC]  = (IB_QP_PKEY_INDEX                |
 878                                                 IB_QP_PORT                      |
 879                                                 IB_QP_ACCESS_FLAGS),
 880                                 [IB_QPT_XRC_INI] = (IB_QP_PKEY_INDEX            |
 881                                                 IB_QP_PORT                      |
 882                                                 IB_QP_ACCESS_FLAGS),
 883                                 [IB_QPT_XRC_TGT] = (IB_QP_PKEY_INDEX            |
 884                                                 IB_QP_PORT                      |
 885                                                 IB_QP_ACCESS_FLAGS),
 886                                 [IB_QPT_SMI] = (IB_QP_PKEY_INDEX                |
 887                                                 IB_QP_QKEY),
 888                                 [IB_QPT_GSI] = (IB_QP_PKEY_INDEX                |
 889                                                 IB_QP_QKEY),
 890                         }
 891                 },
 892                 [IB_QPS_RTR]   = {
 893                         .valid = 1,
 894                         .req_param = {
 895                                 [IB_QPT_UC]  = (IB_QP_AV                        |
 896                                                 IB_QP_PATH_MTU                  |
 897                                                 IB_QP_DEST_QPN                  |
 898                                                 IB_QP_RQ_PSN),
 899                                 [IB_QPT_RC]  = (IB_QP_AV                        |
 900                                                 IB_QP_PATH_MTU                  |
 901                                                 IB_QP_DEST_QPN                  |
 902                                                 IB_QP_RQ_PSN                    |
 903                                                 IB_QP_MAX_DEST_RD_ATOMIC        |
 904                                                 IB_QP_MIN_RNR_TIMER),
 905                                 [IB_QPT_XRC_INI] = (IB_QP_AV                    |
 906                                                 IB_QP_PATH_MTU                  |
 907                                                 IB_QP_DEST_QPN                  |
 908                                                 IB_QP_RQ_PSN),
 909                                 [IB_QPT_XRC_TGT] = (IB_QP_AV                    |
 910                                                 IB_QP_PATH_MTU                  |
 911                                                 IB_QP_DEST_QPN                  |
 912                                                 IB_QP_RQ_PSN                    |
 913                                                 IB_QP_MAX_DEST_RD_ATOMIC        |
 914                                                 IB_QP_MIN_RNR_TIMER),
 915                         },
 916                         .opt_param = {
 917                                  [IB_QPT_UD]  = (IB_QP_PKEY_INDEX               |
 918                                                  IB_QP_QKEY),
 919                                  [IB_QPT_UC]  = (IB_QP_ALT_PATH                 |
 920                                                  IB_QP_ACCESS_FLAGS             |
 921                                                  IB_QP_PKEY_INDEX),
 922                                  [IB_QPT_RC]  = (IB_QP_ALT_PATH                 |
 923                                                  IB_QP_ACCESS_FLAGS             |
 924                                                  IB_QP_PKEY_INDEX),
 925                                  [IB_QPT_XRC_INI] = (IB_QP_ALT_PATH             |
 926                                                  IB_QP_ACCESS_FLAGS             |
 927                                                  IB_QP_PKEY_INDEX),
 928                                  [IB_QPT_XRC_TGT] = (IB_QP_ALT_PATH             |
 929                                                  IB_QP_ACCESS_FLAGS             |
 930                                                  IB_QP_PKEY_INDEX),
 931                                  [IB_QPT_SMI] = (IB_QP_PKEY_INDEX               |
 932                                                  IB_QP_QKEY),
 933                                  [IB_QPT_GSI] = (IB_QP_PKEY_INDEX               |
 934                                                  IB_QP_QKEY),
 935                          },
 936                 },
 937         },
 938         [IB_QPS_RTR]   = {
 939                 [IB_QPS_RESET] = { .valid = 1 },
 940                 [IB_QPS_ERR] =   { .valid = 1 },
 941                 [IB_QPS_RTS]   = {
 942                         .valid = 1,
 943                         .req_param = {
 944                                 [IB_QPT_UD]  = IB_QP_SQ_PSN,
 945                                 [IB_QPT_UC]  = IB_QP_SQ_PSN,
 946                                 [IB_QPT_RC]  = (IB_QP_TIMEOUT                   |
 947                                                 IB_QP_RETRY_CNT                 |
 948                                                 IB_QP_RNR_RETRY                 |
 949                                                 IB_QP_SQ_PSN                    |
 950                                                 IB_QP_MAX_QP_RD_ATOMIC),
 951                                 [IB_QPT_XRC_INI] = (IB_QP_TIMEOUT               |
 952                                                 IB_QP_RETRY_CNT                 |
 953                                                 IB_QP_RNR_RETRY                 |
 954                                                 IB_QP_SQ_PSN                    |
 955                                                 IB_QP_MAX_QP_RD_ATOMIC),
 956                                 [IB_QPT_XRC_TGT] = (IB_QP_TIMEOUT               |
 957                                                 IB_QP_SQ_PSN),
 958                                 [IB_QPT_SMI] = IB_QP_SQ_PSN,
 959                                 [IB_QPT_GSI] = IB_QP_SQ_PSN,
 960                         },
 961                         .opt_param = {
 962                                  [IB_QPT_UD]  = (IB_QP_CUR_STATE                |
 963                                                  IB_QP_QKEY),
 964                                  [IB_QPT_UC]  = (IB_QP_CUR_STATE                |
 965                                                  IB_QP_ALT_PATH                 |
 966                                                  IB_QP_ACCESS_FLAGS             |
 967                                                  IB_QP_PATH_MIG_STATE),
 968                                  [IB_QPT_RC]  = (IB_QP_CUR_STATE                |
 969                                                  IB_QP_ALT_PATH                 |
 970                                                  IB_QP_ACCESS_FLAGS             |
 971                                                  IB_QP_MIN_RNR_TIMER            |
 972                                                  IB_QP_PATH_MIG_STATE),
 973                                  [IB_QPT_XRC_INI] = (IB_QP_CUR_STATE            |
 974                                                  IB_QP_ALT_PATH                 |
 975                                                  IB_QP_ACCESS_FLAGS             |
 976                                                  IB_QP_PATH_MIG_STATE),
 977                                  [IB_QPT_XRC_TGT] = (IB_QP_CUR_STATE            |
 978                                                  IB_QP_ALT_PATH                 |
 979                                                  IB_QP_ACCESS_FLAGS             |
 980                                                  IB_QP_MIN_RNR_TIMER            |
 981                                                  IB_QP_PATH_MIG_STATE),
 982                                  [IB_QPT_SMI] = (IB_QP_CUR_STATE                |
 983                                                  IB_QP_QKEY),
 984                                  [IB_QPT_GSI] = (IB_QP_CUR_STATE                |
 985                                                  IB_QP_QKEY),
 986                          }
 987                 }
 988         },
 989         [IB_QPS_RTS]   = {
 990                 [IB_QPS_RESET] = { .valid = 1 },
 991                 [IB_QPS_ERR] =   { .valid = 1 },
 992                 [IB_QPS_RTS]   = {
 993                         .valid = 1,
 994                         .opt_param = {
 995                                 [IB_QPT_UD]  = (IB_QP_CUR_STATE                 |
 996                                                 IB_QP_QKEY),
 997                                 [IB_QPT_UC]  = (IB_QP_CUR_STATE                 |
 998                                                 IB_QP_ACCESS_FLAGS              |
 999                                                 IB_QP_ALT_PATH                  |
1000                                                 IB_QP_PATH_MIG_STATE),
1001                                 [IB_QPT_RC]  = (IB_QP_CUR_STATE                 |
1002                                                 IB_QP_ACCESS_FLAGS              |
1003                                                 IB_QP_ALT_PATH                  |
1004                                                 IB_QP_PATH_MIG_STATE            |
1005                                                 IB_QP_MIN_RNR_TIMER),
1006                                 [IB_QPT_XRC_INI] = (IB_QP_CUR_STATE             |
1007                                                 IB_QP_ACCESS_FLAGS              |
1008                                                 IB_QP_ALT_PATH                  |
1009                                                 IB_QP_PATH_MIG_STATE),
1010                                 [IB_QPT_XRC_TGT] = (IB_QP_CUR_STATE             |
1011                                                 IB_QP_ACCESS_FLAGS              |
1012                                                 IB_QP_ALT_PATH                  |
1013                                                 IB_QP_PATH_MIG_STATE            |
1014                                                 IB_QP_MIN_RNR_TIMER),
1015                                 [IB_QPT_SMI] = (IB_QP_CUR_STATE                 |
1016                                                 IB_QP_QKEY),
1017                                 [IB_QPT_GSI] = (IB_QP_CUR_STATE                 |
1018                                                 IB_QP_QKEY),
1019                         }
1020                 },
1021                 [IB_QPS_SQD]   = {
1022                         .valid = 1,
1023                         .opt_param = {
1024                                 [IB_QPT_UD]  = IB_QP_EN_SQD_ASYNC_NOTIFY,
1025                                 [IB_QPT_UC]  = IB_QP_EN_SQD_ASYNC_NOTIFY,
1026                                 [IB_QPT_RC]  = IB_QP_EN_SQD_ASYNC_NOTIFY,
1027                                 [IB_QPT_XRC_INI] = IB_QP_EN_SQD_ASYNC_NOTIFY,
1028                                 [IB_QPT_XRC_TGT] = IB_QP_EN_SQD_ASYNC_NOTIFY, /* ??? */
1029                                 [IB_QPT_SMI] = IB_QP_EN_SQD_ASYNC_NOTIFY,
1030                                 [IB_QPT_GSI] = IB_QP_EN_SQD_ASYNC_NOTIFY
1031                         }
1032                 },
1033         },
1034         [IB_QPS_SQD]   = {
1035                 [IB_QPS_RESET] = { .valid = 1 },
1036                 [IB_QPS_ERR] =   { .valid = 1 },
1037                 [IB_QPS_RTS]   = {
1038                         .valid = 1,
1039                         .opt_param = {
1040                                 [IB_QPT_UD]  = (IB_QP_CUR_STATE                 |
1041                                                 IB_QP_QKEY),
1042                                 [IB_QPT_UC]  = (IB_QP_CUR_STATE                 |
1043                                                 IB_QP_ALT_PATH                  |
1044                                                 IB_QP_ACCESS_FLAGS              |
1045                                                 IB_QP_PATH_MIG_STATE),
1046                                 [IB_QPT_RC]  = (IB_QP_CUR_STATE                 |
1047                                                 IB_QP_ALT_PATH                  |
1048                                                 IB_QP_ACCESS_FLAGS              |
1049                                                 IB_QP_MIN_RNR_TIMER             |
1050                                                 IB_QP_PATH_MIG_STATE),
1051                                 [IB_QPT_XRC_INI] = (IB_QP_CUR_STATE             |
1052                                                 IB_QP_ALT_PATH                  |
1053                                                 IB_QP_ACCESS_FLAGS              |
1054                                                 IB_QP_PATH_MIG_STATE),
1055                                 [IB_QPT_XRC_TGT] = (IB_QP_CUR_STATE             |
1056                                                 IB_QP_ALT_PATH                  |
1057                                                 IB_QP_ACCESS_FLAGS              |
1058                                                 IB_QP_MIN_RNR_TIMER             |
1059                                                 IB_QP_PATH_MIG_STATE),
1060                                 [IB_QPT_SMI] = (IB_QP_CUR_STATE                 |
1061                                                 IB_QP_QKEY),
1062                                 [IB_QPT_GSI] = (IB_QP_CUR_STATE                 |
1063                                                 IB_QP_QKEY),
1064                         }
1065                 },
1066                 [IB_QPS_SQD]   = {
1067                         .valid = 1,
1068                         .opt_param = {
1069                                 [IB_QPT_UD]  = (IB_QP_PKEY_INDEX                |
1070                                                 IB_QP_QKEY),
1071                                 [IB_QPT_UC]  = (IB_QP_AV                        |
1072                                                 IB_QP_ALT_PATH                  |
1073                                                 IB_QP_ACCESS_FLAGS              |
1074                                                 IB_QP_PKEY_INDEX                |
1075                                                 IB_QP_PATH_MIG_STATE),
1076                                 [IB_QPT_RC]  = (IB_QP_PORT                      |
1077                                                 IB_QP_AV                        |
1078                                                 IB_QP_TIMEOUT                   |
1079                                                 IB_QP_RETRY_CNT                 |
1080                                                 IB_QP_RNR_RETRY                 |
1081                                                 IB_QP_MAX_QP_RD_ATOMIC          |
1082                                                 IB_QP_MAX_DEST_RD_ATOMIC        |
1083                                                 IB_QP_ALT_PATH                  |
1084                                                 IB_QP_ACCESS_FLAGS              |
1085                                                 IB_QP_PKEY_INDEX                |
1086                                                 IB_QP_MIN_RNR_TIMER             |
1087                                                 IB_QP_PATH_MIG_STATE),
1088                                 [IB_QPT_XRC_INI] = (IB_QP_PORT                  |
1089                                                 IB_QP_AV                        |
1090                                                 IB_QP_TIMEOUT                   |
1091                                                 IB_QP_RETRY_CNT                 |
1092                                                 IB_QP_RNR_RETRY                 |
1093                                                 IB_QP_MAX_QP_RD_ATOMIC          |
1094                                                 IB_QP_ALT_PATH                  |
1095                                                 IB_QP_ACCESS_FLAGS              |
1096                                                 IB_QP_PKEY_INDEX                |
1097                                                 IB_QP_PATH_MIG_STATE),
1098                                 [IB_QPT_XRC_TGT] = (IB_QP_PORT                  |
1099                                                 IB_QP_AV                        |
1100                                                 IB_QP_TIMEOUT                   |
1101                                                 IB_QP_MAX_DEST_RD_ATOMIC        |
1102                                                 IB_QP_ALT_PATH                  |
1103                                                 IB_QP_ACCESS_FLAGS              |
1104                                                 IB_QP_PKEY_INDEX                |
1105                                                 IB_QP_MIN_RNR_TIMER             |
1106                                                 IB_QP_PATH_MIG_STATE),
1107                                 [IB_QPT_SMI] = (IB_QP_PKEY_INDEX                |
1108                                                 IB_QP_QKEY),
1109                                 [IB_QPT_GSI] = (IB_QP_PKEY_INDEX                |
1110                                                 IB_QP_QKEY),
1111                         }
1112                 }
1113         },
1114         [IB_QPS_SQE]   = {
1115                 [IB_QPS_RESET] = { .valid = 1 },
1116                 [IB_QPS_ERR] =   { .valid = 1 },
1117                 [IB_QPS_RTS]   = {
1118                         .valid = 1,
1119                         .opt_param = {
1120                                 [IB_QPT_UD]  = (IB_QP_CUR_STATE                 |
1121                                                 IB_QP_QKEY),
1122                                 [IB_QPT_UC]  = (IB_QP_CUR_STATE                 |
1123                                                 IB_QP_ACCESS_FLAGS),
1124                                 [IB_QPT_SMI] = (IB_QP_CUR_STATE                 |
1125                                                 IB_QP_QKEY),
1126                                 [IB_QPT_GSI] = (IB_QP_CUR_STATE                 |
1127                                                 IB_QP_QKEY),
1128                         }
1129                 }
1130         },
1131         [IB_QPS_ERR] = {
1132                 [IB_QPS_RESET] = { .valid = 1 },
1133                 [IB_QPS_ERR] =   { .valid = 1 }
1134         }
1135 };
1136
1137 int ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
1138                        enum ib_qp_type type, enum ib_qp_attr_mask mask,
1139                        enum rdma_link_layer ll)
1140 {
1141         enum ib_qp_attr_mask req_param, opt_param;
1142
1143         if (cur_state  < 0 || cur_state  > IB_QPS_ERR ||
1144             next_state < 0 || next_state > IB_QPS_ERR)
1145                 return 0;
1146
1147         if (mask & IB_QP_CUR_STATE  &&
1148             cur_state != IB_QPS_RTR && cur_state != IB_QPS_RTS &&
1149             cur_state != IB_QPS_SQD && cur_state != IB_QPS_SQE)
1150                 return 0;
1151
1152         if (!qp_state_table[cur_state][next_state].valid)
1153                 return 0;
1154
1155         req_param = qp_state_table[cur_state][next_state].req_param[type];
1156         opt_param = qp_state_table[cur_state][next_state].opt_param[type];
1157
1158         if ((mask & req_param) != req_param)
1159                 return 0;
1160
1161         if (mask & ~(req_param | opt_param | IB_QP_STATE))
1162                 return 0;
1163
1164         return 1;
1165 }
1166 EXPORT_SYMBOL(ib_modify_qp_is_ok);
1167
1168 int ib_resolve_eth_dmac(struct ib_qp *qp,
1169                         struct ib_qp_attr *qp_attr, int *qp_attr_mask)
1170 {
1171         int           ret = 0;
1172
1173         if (*qp_attr_mask & IB_QP_AV) {
1174                 if (qp_attr->ah_attr.port_num < rdma_start_port(qp->device) ||
1175                     qp_attr->ah_attr.port_num > rdma_end_port(qp->device))
1176                         return -EINVAL;
1177
1178                 if (!rdma_cap_eth_ah(qp->device, qp_attr->ah_attr.port_num))
1179                         return 0;
1180
1181                 if (rdma_link_local_addr((struct in6_addr *)qp_attr->ah_attr.grh.dgid.raw)) {
1182                         rdma_get_ll_mac((struct in6_addr *)qp_attr->ah_attr.grh.dgid.raw,
1183                                         qp_attr->ah_attr.dmac);
1184                 } else {
1185                         union ib_gid            sgid;
1186                         struct ib_gid_attr      sgid_attr;
1187                         int                     ifindex;
1188                         int                     hop_limit;
1189
1190                         ret = ib_query_gid(qp->device,
1191                                            qp_attr->ah_attr.port_num,
1192                                            qp_attr->ah_attr.grh.sgid_index,
1193                                            &sgid, &sgid_attr);
1194
1195                         if (ret || !sgid_attr.ndev) {
1196                                 if (!ret)
1197                                         ret = -ENXIO;
1198                                 goto out;
1199                         }
1200
1201                         ifindex = sgid_attr.ndev->ifindex;
1202
1203                         ret = rdma_addr_find_l2_eth_by_grh(&sgid,
1204                                                            &qp_attr->ah_attr.grh.dgid,
1205                                                            qp_attr->ah_attr.dmac,
1206                                                            NULL, &ifindex, &hop_limit);
1207
1208                         dev_put(sgid_attr.ndev);
1209
1210                         qp_attr->ah_attr.grh.hop_limit = hop_limit;
1211                 }
1212         }
1213 out:
1214         return ret;
1215 }
1216 EXPORT_SYMBOL(ib_resolve_eth_dmac);
1217
1218
1219 int ib_modify_qp(struct ib_qp *qp,
1220                  struct ib_qp_attr *qp_attr,
1221                  int qp_attr_mask)
1222 {
1223         int ret;
1224
1225         ret = ib_resolve_eth_dmac(qp, qp_attr, &qp_attr_mask);
1226         if (ret)
1227                 return ret;
1228
1229         return qp->device->modify_qp(qp->real_qp, qp_attr, qp_attr_mask, NULL);
1230 }
1231 EXPORT_SYMBOL(ib_modify_qp);
1232
1233 int ib_query_qp(struct ib_qp *qp,
1234                 struct ib_qp_attr *qp_attr,
1235                 int qp_attr_mask,
1236                 struct ib_qp_init_attr *qp_init_attr)
1237 {
1238         return qp->device->query_qp ?
1239                 qp->device->query_qp(qp->real_qp, qp_attr, qp_attr_mask, qp_init_attr) :
1240                 -ENOSYS;
1241 }
1242 EXPORT_SYMBOL(ib_query_qp);
1243
1244 int ib_close_qp(struct ib_qp *qp)
1245 {
1246         struct ib_qp *real_qp;
1247         unsigned long flags;
1248
1249         real_qp = qp->real_qp;
1250         if (real_qp == qp)
1251                 return -EINVAL;
1252
1253         spin_lock_irqsave(&real_qp->device->event_handler_lock, flags);
1254         list_del(&qp->open_list);
1255         spin_unlock_irqrestore(&real_qp->device->event_handler_lock, flags);
1256
1257         atomic_dec(&real_qp->usecnt);
1258         kfree(qp);
1259
1260         return 0;
1261 }
1262 EXPORT_SYMBOL(ib_close_qp);
1263
1264 static int __ib_destroy_shared_qp(struct ib_qp *qp)
1265 {
1266         struct ib_xrcd *xrcd;
1267         struct ib_qp *real_qp;
1268         int ret;
1269
1270         real_qp = qp->real_qp;
1271         xrcd = real_qp->xrcd;
1272
1273         mutex_lock(&xrcd->tgt_qp_mutex);
1274         ib_close_qp(qp);
1275         if (atomic_read(&real_qp->usecnt) == 0)
1276                 list_del(&real_qp->xrcd_list);
1277         else
1278                 real_qp = NULL;
1279         mutex_unlock(&xrcd->tgt_qp_mutex);
1280
1281         if (real_qp) {
1282                 ret = ib_destroy_qp(real_qp);
1283                 if (!ret)
1284                         atomic_dec(&xrcd->usecnt);
1285                 else
1286                         __ib_insert_xrcd_qp(xrcd, real_qp);
1287         }
1288
1289         return 0;
1290 }
1291
1292 int ib_destroy_qp(struct ib_qp *qp)
1293 {
1294         struct ib_pd *pd;
1295         struct ib_cq *scq, *rcq;
1296         struct ib_srq *srq;
1297         struct ib_rwq_ind_table *ind_tbl;
1298         int ret;
1299
1300         WARN_ON_ONCE(qp->mrs_used > 0);
1301
1302         if (atomic_read(&qp->usecnt))
1303                 return -EBUSY;
1304
1305         if (qp->real_qp != qp)
1306                 return __ib_destroy_shared_qp(qp);
1307
1308         pd   = qp->pd;
1309         scq  = qp->send_cq;
1310         rcq  = qp->recv_cq;
1311         srq  = qp->srq;
1312         ind_tbl = qp->rwq_ind_tbl;
1313
1314         if (!qp->uobject)
1315                 rdma_rw_cleanup_mrs(qp);
1316
1317         ret = qp->device->destroy_qp(qp);
1318         if (!ret) {
1319                 if (pd)
1320                         atomic_dec(&pd->usecnt);
1321                 if (scq)
1322                         atomic_dec(&scq->usecnt);
1323                 if (rcq)
1324                         atomic_dec(&rcq->usecnt);
1325                 if (srq)
1326                         atomic_dec(&srq->usecnt);
1327                 if (ind_tbl)
1328                         atomic_dec(&ind_tbl->usecnt);
1329         }
1330
1331         return ret;
1332 }
1333 EXPORT_SYMBOL(ib_destroy_qp);
1334
1335 /* Completion queues */
1336
1337 struct ib_cq *ib_create_cq(struct ib_device *device,
1338                            ib_comp_handler comp_handler,
1339                            void (*event_handler)(struct ib_event *, void *),
1340                            void *cq_context,
1341                            const struct ib_cq_init_attr *cq_attr)
1342 {
1343         struct ib_cq *cq;
1344
1345         cq = device->create_cq(device, cq_attr, NULL, NULL);
1346
1347         if (!IS_ERR(cq)) {
1348                 cq->device        = device;
1349                 cq->uobject       = NULL;
1350                 cq->comp_handler  = comp_handler;
1351                 cq->event_handler = event_handler;
1352                 cq->cq_context    = cq_context;
1353                 atomic_set(&cq->usecnt, 0);
1354         }
1355
1356         return cq;
1357 }
1358 EXPORT_SYMBOL(ib_create_cq);
1359
1360 int ib_modify_cq(struct ib_cq *cq, u16 cq_count, u16 cq_period)
1361 {
1362         return cq->device->modify_cq ?
1363                 cq->device->modify_cq(cq, cq_count, cq_period) : -ENOSYS;
1364 }
1365 EXPORT_SYMBOL(ib_modify_cq);
1366
1367 int ib_destroy_cq(struct ib_cq *cq)
1368 {
1369         if (atomic_read(&cq->usecnt))
1370                 return -EBUSY;
1371
1372         return cq->device->destroy_cq(cq);
1373 }
1374 EXPORT_SYMBOL(ib_destroy_cq);
1375
1376 int ib_resize_cq(struct ib_cq *cq, int cqe)
1377 {
1378         return cq->device->resize_cq ?
1379                 cq->device->resize_cq(cq, cqe, NULL) : -ENOSYS;
1380 }
1381 EXPORT_SYMBOL(ib_resize_cq);
1382
1383 /* Memory regions */
1384
1385 struct ib_mr *ib_get_dma_mr(struct ib_pd *pd, int mr_access_flags)
1386 {
1387         struct ib_mr *mr;
1388         int err;
1389
1390         err = ib_check_mr_access(mr_access_flags);
1391         if (err)
1392                 return ERR_PTR(err);
1393
1394         mr = pd->device->get_dma_mr(pd, mr_access_flags);
1395
1396         if (!IS_ERR(mr)) {
1397                 mr->device  = pd->device;
1398                 mr->pd      = pd;
1399                 mr->uobject = NULL;
1400                 atomic_inc(&pd->usecnt);
1401                 mr->need_inval = false;
1402         }
1403
1404         return mr;
1405 }
1406 EXPORT_SYMBOL(ib_get_dma_mr);
1407
1408 int ib_dereg_mr(struct ib_mr *mr)
1409 {
1410         struct ib_pd *pd = mr->pd;
1411         int ret;
1412
1413         ret = mr->device->dereg_mr(mr);
1414         if (!ret)
1415                 atomic_dec(&pd->usecnt);
1416
1417         return ret;
1418 }
1419 EXPORT_SYMBOL(ib_dereg_mr);
1420
1421 /**
1422  * ib_alloc_mr() - Allocates a memory region
1423  * @pd:            protection domain associated with the region
1424  * @mr_type:       memory region type
1425  * @max_num_sg:    maximum sg entries available for registration.
1426  *
1427  * Notes:
1428  * Memory registeration page/sg lists must not exceed max_num_sg.
1429  * For mr_type IB_MR_TYPE_MEM_REG, the total length cannot exceed
1430  * max_num_sg * used_page_size.
1431  *
1432  */
1433 struct ib_mr *ib_alloc_mr(struct ib_pd *pd,
1434                           enum ib_mr_type mr_type,
1435                           u32 max_num_sg)
1436 {
1437         struct ib_mr *mr;
1438
1439         if (!pd->device->alloc_mr)
1440                 return ERR_PTR(-ENOSYS);
1441
1442         mr = pd->device->alloc_mr(pd, mr_type, max_num_sg);
1443         if (!IS_ERR(mr)) {
1444                 mr->device  = pd->device;
1445                 mr->pd      = pd;
1446                 mr->uobject = NULL;
1447                 atomic_inc(&pd->usecnt);
1448                 mr->need_inval = false;
1449         }
1450
1451         return mr;
1452 }
1453 EXPORT_SYMBOL(ib_alloc_mr);
1454
1455 /* "Fast" memory regions */
1456
1457 struct ib_fmr *ib_alloc_fmr(struct ib_pd *pd,
1458                             int mr_access_flags,
1459                             struct ib_fmr_attr *fmr_attr)
1460 {
1461         struct ib_fmr *fmr;
1462
1463         if (!pd->device->alloc_fmr)
1464                 return ERR_PTR(-ENOSYS);
1465
1466         fmr = pd->device->alloc_fmr(pd, mr_access_flags, fmr_attr);
1467         if (!IS_ERR(fmr)) {
1468                 fmr->device = pd->device;
1469                 fmr->pd     = pd;
1470                 atomic_inc(&pd->usecnt);
1471         }
1472
1473         return fmr;
1474 }
1475 EXPORT_SYMBOL(ib_alloc_fmr);
1476
1477 int ib_unmap_fmr(struct list_head *fmr_list)
1478 {
1479         struct ib_fmr *fmr;
1480
1481         if (list_empty(fmr_list))
1482                 return 0;
1483
1484         fmr = list_entry(fmr_list->next, struct ib_fmr, list);
1485         return fmr->device->unmap_fmr(fmr_list);
1486 }
1487 EXPORT_SYMBOL(ib_unmap_fmr);
1488
1489 int ib_dealloc_fmr(struct ib_fmr *fmr)
1490 {
1491         struct ib_pd *pd;
1492         int ret;
1493
1494         pd = fmr->pd;
1495         ret = fmr->device->dealloc_fmr(fmr);
1496         if (!ret)
1497                 atomic_dec(&pd->usecnt);
1498
1499         return ret;
1500 }
1501 EXPORT_SYMBOL(ib_dealloc_fmr);
1502
1503 /* Multicast groups */
1504
1505 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid)
1506 {
1507         int ret;
1508
1509         if (!qp->device->attach_mcast)
1510                 return -ENOSYS;
1511         if (gid->raw[0] != 0xff || qp->qp_type != IB_QPT_UD)
1512                 return -EINVAL;
1513
1514         ret = qp->device->attach_mcast(qp, gid, lid);
1515         if (!ret)
1516                 atomic_inc(&qp->usecnt);
1517         return ret;
1518 }
1519 EXPORT_SYMBOL(ib_attach_mcast);
1520
1521 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid)
1522 {
1523         int ret;
1524
1525         if (!qp->device->detach_mcast)
1526                 return -ENOSYS;
1527         if (gid->raw[0] != 0xff || qp->qp_type != IB_QPT_UD)
1528                 return -EINVAL;
1529
1530         ret = qp->device->detach_mcast(qp, gid, lid);
1531         if (!ret)
1532                 atomic_dec(&qp->usecnt);
1533         return ret;
1534 }
1535 EXPORT_SYMBOL(ib_detach_mcast);
1536
1537 struct ib_xrcd *ib_alloc_xrcd(struct ib_device *device)
1538 {
1539         struct ib_xrcd *xrcd;
1540
1541         if (!device->alloc_xrcd)
1542                 return ERR_PTR(-ENOSYS);
1543
1544         xrcd = device->alloc_xrcd(device, NULL, NULL);
1545         if (!IS_ERR(xrcd)) {
1546                 xrcd->device = device;
1547                 xrcd->inode = NULL;
1548                 atomic_set(&xrcd->usecnt, 0);
1549                 mutex_init(&xrcd->tgt_qp_mutex);
1550                 INIT_LIST_HEAD(&xrcd->tgt_qp_list);
1551         }
1552
1553         return xrcd;
1554 }
1555 EXPORT_SYMBOL(ib_alloc_xrcd);
1556
1557 int ib_dealloc_xrcd(struct ib_xrcd *xrcd)
1558 {
1559         struct ib_qp *qp;
1560         int ret;
1561
1562         if (atomic_read(&xrcd->usecnt))
1563                 return -EBUSY;
1564
1565         while (!list_empty(&xrcd->tgt_qp_list)) {
1566                 qp = list_entry(xrcd->tgt_qp_list.next, struct ib_qp, xrcd_list);
1567                 ret = ib_destroy_qp(qp);
1568                 if (ret)
1569                         return ret;
1570         }
1571
1572         return xrcd->device->dealloc_xrcd(xrcd);
1573 }
1574 EXPORT_SYMBOL(ib_dealloc_xrcd);
1575
1576 /**
1577  * ib_create_wq - Creates a WQ associated with the specified protection
1578  * domain.
1579  * @pd: The protection domain associated with the WQ.
1580  * @wq_init_attr: A list of initial attributes required to create the
1581  * WQ. If WQ creation succeeds, then the attributes are updated to
1582  * the actual capabilities of the created WQ.
1583  *
1584  * wq_init_attr->max_wr and wq_init_attr->max_sge determine
1585  * the requested size of the WQ, and set to the actual values allocated
1586  * on return.
1587  * If ib_create_wq() succeeds, then max_wr and max_sge will always be
1588  * at least as large as the requested values.
1589  */
1590 struct ib_wq *ib_create_wq(struct ib_pd *pd,
1591                            struct ib_wq_init_attr *wq_attr)
1592 {
1593         struct ib_wq *wq;
1594
1595         if (!pd->device->create_wq)
1596                 return ERR_PTR(-ENOSYS);
1597
1598         wq = pd->device->create_wq(pd, wq_attr, NULL);
1599         if (!IS_ERR(wq)) {
1600                 wq->event_handler = wq_attr->event_handler;
1601                 wq->wq_context = wq_attr->wq_context;
1602                 wq->wq_type = wq_attr->wq_type;
1603                 wq->cq = wq_attr->cq;
1604                 wq->device = pd->device;
1605                 wq->pd = pd;
1606                 wq->uobject = NULL;
1607                 atomic_inc(&pd->usecnt);
1608                 atomic_inc(&wq_attr->cq->usecnt);
1609                 atomic_set(&wq->usecnt, 0);
1610         }
1611         return wq;
1612 }
1613 EXPORT_SYMBOL(ib_create_wq);
1614
1615 /**
1616  * ib_destroy_wq - Destroys the specified WQ.
1617  * @wq: The WQ to destroy.
1618  */
1619 int ib_destroy_wq(struct ib_wq *wq)
1620 {
1621         int err;
1622         struct ib_cq *cq = wq->cq;
1623         struct ib_pd *pd = wq->pd;
1624
1625         if (atomic_read(&wq->usecnt))
1626                 return -EBUSY;
1627
1628         err = wq->device->destroy_wq(wq);
1629         if (!err) {
1630                 atomic_dec(&pd->usecnt);
1631                 atomic_dec(&cq->usecnt);
1632         }
1633         return err;
1634 }
1635 EXPORT_SYMBOL(ib_destroy_wq);
1636
1637 /**
1638  * ib_modify_wq - Modifies the specified WQ.
1639  * @wq: The WQ to modify.
1640  * @wq_attr: On input, specifies the WQ attributes to modify.
1641  * @wq_attr_mask: A bit-mask used to specify which attributes of the WQ
1642  *   are being modified.
1643  * On output, the current values of selected WQ attributes are returned.
1644  */
1645 int ib_modify_wq(struct ib_wq *wq, struct ib_wq_attr *wq_attr,
1646                  u32 wq_attr_mask)
1647 {
1648         int err;
1649
1650         if (!wq->device->modify_wq)
1651                 return -ENOSYS;
1652
1653         err = wq->device->modify_wq(wq, wq_attr, wq_attr_mask, NULL);
1654         return err;
1655 }
1656 EXPORT_SYMBOL(ib_modify_wq);
1657
1658 /*
1659  * ib_create_rwq_ind_table - Creates a RQ Indirection Table.
1660  * @device: The device on which to create the rwq indirection table.
1661  * @ib_rwq_ind_table_init_attr: A list of initial attributes required to
1662  * create the Indirection Table.
1663  *
1664  * Note: The life time of ib_rwq_ind_table_init_attr->ind_tbl is not less
1665  *      than the created ib_rwq_ind_table object and the caller is responsible
1666  *      for its memory allocation/free.
1667  */
1668 struct ib_rwq_ind_table *ib_create_rwq_ind_table(struct ib_device *device,
1669                                                  struct ib_rwq_ind_table_init_attr *init_attr)
1670 {
1671         struct ib_rwq_ind_table *rwq_ind_table;
1672         int i;
1673         u32 table_size;
1674
1675         if (!device->create_rwq_ind_table)
1676                 return ERR_PTR(-ENOSYS);
1677
1678         table_size = (1 << init_attr->log_ind_tbl_size);
1679         rwq_ind_table = device->create_rwq_ind_table(device,
1680                                 init_attr, NULL);
1681         if (IS_ERR(rwq_ind_table))
1682                 return rwq_ind_table;
1683
1684         rwq_ind_table->ind_tbl = init_attr->ind_tbl;
1685         rwq_ind_table->log_ind_tbl_size = init_attr->log_ind_tbl_size;
1686         rwq_ind_table->device = device;
1687         rwq_ind_table->uobject = NULL;
1688         atomic_set(&rwq_ind_table->usecnt, 0);
1689
1690         for (i = 0; i < table_size; i++)
1691                 atomic_inc(&rwq_ind_table->ind_tbl[i]->usecnt);
1692
1693         return rwq_ind_table;
1694 }
1695 EXPORT_SYMBOL(ib_create_rwq_ind_table);
1696
1697 /*
1698  * ib_destroy_rwq_ind_table - Destroys the specified Indirection Table.
1699  * @wq_ind_table: The Indirection Table to destroy.
1700 */
1701 int ib_destroy_rwq_ind_table(struct ib_rwq_ind_table *rwq_ind_table)
1702 {
1703         int err, i;
1704         u32 table_size = (1 << rwq_ind_table->log_ind_tbl_size);
1705         struct ib_wq **ind_tbl = rwq_ind_table->ind_tbl;
1706
1707         if (atomic_read(&rwq_ind_table->usecnt))
1708                 return -EBUSY;
1709
1710         err = rwq_ind_table->device->destroy_rwq_ind_table(rwq_ind_table);
1711         if (!err) {
1712                 for (i = 0; i < table_size; i++)
1713                         atomic_dec(&ind_tbl[i]->usecnt);
1714         }
1715
1716         return err;
1717 }
1718 EXPORT_SYMBOL(ib_destroy_rwq_ind_table);
1719
1720 struct ib_flow *ib_create_flow(struct ib_qp *qp,
1721                                struct ib_flow_attr *flow_attr,
1722                                int domain)
1723 {
1724         struct ib_flow *flow_id;
1725         if (!qp->device->create_flow)
1726                 return ERR_PTR(-ENOSYS);
1727
1728         flow_id = qp->device->create_flow(qp, flow_attr, domain);
1729         if (!IS_ERR(flow_id))
1730                 atomic_inc(&qp->usecnt);
1731         return flow_id;
1732 }
1733 EXPORT_SYMBOL(ib_create_flow);
1734
1735 int ib_destroy_flow(struct ib_flow *flow_id)
1736 {
1737         int err;
1738         struct ib_qp *qp = flow_id->qp;
1739
1740         err = qp->device->destroy_flow(flow_id);
1741         if (!err)
1742                 atomic_dec(&qp->usecnt);
1743         return err;
1744 }
1745 EXPORT_SYMBOL(ib_destroy_flow);
1746
1747 int ib_check_mr_status(struct ib_mr *mr, u32 check_mask,
1748                        struct ib_mr_status *mr_status)
1749 {
1750         return mr->device->check_mr_status ?
1751                 mr->device->check_mr_status(mr, check_mask, mr_status) : -ENOSYS;
1752 }
1753 EXPORT_SYMBOL(ib_check_mr_status);
1754
1755 int ib_set_vf_link_state(struct ib_device *device, int vf, u8 port,
1756                          int state)
1757 {
1758         if (!device->set_vf_link_state)
1759                 return -ENOSYS;
1760
1761         return device->set_vf_link_state(device, vf, port, state);
1762 }
1763 EXPORT_SYMBOL(ib_set_vf_link_state);
1764
1765 int ib_get_vf_config(struct ib_device *device, int vf, u8 port,
1766                      struct ifla_vf_info *info)
1767 {
1768         if (!device->get_vf_config)
1769                 return -ENOSYS;
1770
1771         return device->get_vf_config(device, vf, port, info);
1772 }
1773 EXPORT_SYMBOL(ib_get_vf_config);
1774
1775 int ib_get_vf_stats(struct ib_device *device, int vf, u8 port,
1776                     struct ifla_vf_stats *stats)
1777 {
1778         if (!device->get_vf_stats)
1779                 return -ENOSYS;
1780
1781         return device->get_vf_stats(device, vf, port, stats);
1782 }
1783 EXPORT_SYMBOL(ib_get_vf_stats);
1784
1785 int ib_set_vf_guid(struct ib_device *device, int vf, u8 port, u64 guid,
1786                    int type)
1787 {
1788         if (!device->set_vf_guid)
1789                 return -ENOSYS;
1790
1791         return device->set_vf_guid(device, vf, port, guid, type);
1792 }
1793 EXPORT_SYMBOL(ib_set_vf_guid);
1794
1795 /**
1796  * ib_map_mr_sg() - Map the largest prefix of a dma mapped SG list
1797  *     and set it the memory region.
1798  * @mr:            memory region
1799  * @sg:            dma mapped scatterlist
1800  * @sg_nents:      number of entries in sg
1801  * @sg_offset:     offset in bytes into sg
1802  * @page_size:     page vector desired page size
1803  *
1804  * Constraints:
1805  * - The first sg element is allowed to have an offset.
1806  * - Each sg element must be aligned to page_size (or physically
1807  *   contiguous to the previous element). In case an sg element has a
1808  *   non contiguous offset, the mapping prefix will not include it.
1809  * - The last sg element is allowed to have length less than page_size.
1810  * - If sg_nents total byte length exceeds the mr max_num_sge * page_size
1811  *   then only max_num_sg entries will be mapped.
1812  * - If the MR was allocated with type IB_MR_TYPE_SG_GAPS_REG, non of these
1813  *   constraints holds and the page_size argument is ignored.
1814  *
1815  * Returns the number of sg elements that were mapped to the memory region.
1816  *
1817  * After this completes successfully, the  memory region
1818  * is ready for registration.
1819  */
1820 int ib_map_mr_sg(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
1821                  unsigned int *sg_offset, unsigned int page_size)
1822 {
1823         if (unlikely(!mr->device->map_mr_sg))
1824                 return -ENOSYS;
1825
1826         mr->page_size = page_size;
1827
1828         return mr->device->map_mr_sg(mr, sg, sg_nents, sg_offset);
1829 }
1830 EXPORT_SYMBOL(ib_map_mr_sg);
1831
1832 /**
1833  * ib_sg_to_pages() - Convert the largest prefix of a sg list
1834  *     to a page vector
1835  * @mr:            memory region
1836  * @sgl:           dma mapped scatterlist
1837  * @sg_nents:      number of entries in sg
1838  * @sg_offset_p:   IN:  start offset in bytes into sg
1839  *                 OUT: offset in bytes for element n of the sg of the first
1840  *                      byte that has not been processed where n is the return
1841  *                      value of this function.
1842  * @set_page:      driver page assignment function pointer
1843  *
1844  * Core service helper for drivers to convert the largest
1845  * prefix of given sg list to a page vector. The sg list
1846  * prefix converted is the prefix that meet the requirements
1847  * of ib_map_mr_sg.
1848  *
1849  * Returns the number of sg elements that were assigned to
1850  * a page vector.
1851  */
1852 int ib_sg_to_pages(struct ib_mr *mr, struct scatterlist *sgl, int sg_nents,
1853                 unsigned int *sg_offset_p, int (*set_page)(struct ib_mr *, u64))
1854 {
1855         struct scatterlist *sg;
1856         u64 last_end_dma_addr = 0;
1857         unsigned int sg_offset = sg_offset_p ? *sg_offset_p : 0;
1858         unsigned int last_page_off = 0;
1859         u64 page_mask = ~((u64)mr->page_size - 1);
1860         int i, ret;
1861
1862         if (unlikely(sg_nents <= 0 || sg_offset > sg_dma_len(&sgl[0])))
1863                 return -EINVAL;
1864
1865         mr->iova = sg_dma_address(&sgl[0]) + sg_offset;
1866         mr->length = 0;
1867
1868         for_each_sg(sgl, sg, sg_nents, i) {
1869                 u64 dma_addr = sg_dma_address(sg) + sg_offset;
1870                 u64 prev_addr = dma_addr;
1871                 unsigned int dma_len = sg_dma_len(sg) - sg_offset;
1872                 u64 end_dma_addr = dma_addr + dma_len;
1873                 u64 page_addr = dma_addr & page_mask;
1874
1875                 /*
1876                  * For the second and later elements, check whether either the
1877                  * end of element i-1 or the start of element i is not aligned
1878                  * on a page boundary.
1879                  */
1880                 if (i && (last_page_off != 0 || page_addr != dma_addr)) {
1881                         /* Stop mapping if there is a gap. */
1882                         if (last_end_dma_addr != dma_addr)
1883                                 break;
1884
1885                         /*
1886                          * Coalesce this element with the last. If it is small
1887                          * enough just update mr->length. Otherwise start
1888                          * mapping from the next page.
1889                          */
1890                         goto next_page;
1891                 }
1892
1893                 do {
1894                         ret = set_page(mr, page_addr);
1895                         if (unlikely(ret < 0)) {
1896                                 sg_offset = prev_addr - sg_dma_address(sg);
1897                                 mr->length += prev_addr - dma_addr;
1898                                 if (sg_offset_p)
1899                                         *sg_offset_p = sg_offset;
1900                                 return i || sg_offset ? i : ret;
1901                         }
1902                         prev_addr = page_addr;
1903 next_page:
1904                         page_addr += mr->page_size;
1905                 } while (page_addr < end_dma_addr);
1906
1907                 mr->length += dma_len;
1908                 last_end_dma_addr = end_dma_addr;
1909                 last_page_off = end_dma_addr & ~page_mask;
1910
1911                 sg_offset = 0;
1912         }
1913
1914         if (sg_offset_p)
1915                 *sg_offset_p = 0;
1916         return i;
1917 }
1918 EXPORT_SYMBOL(ib_sg_to_pages);
1919
1920 struct ib_drain_cqe {
1921         struct ib_cqe cqe;
1922         struct completion done;
1923 };
1924
1925 static void ib_drain_qp_done(struct ib_cq *cq, struct ib_wc *wc)
1926 {
1927         struct ib_drain_cqe *cqe = container_of(wc->wr_cqe, struct ib_drain_cqe,
1928                                                 cqe);
1929
1930         complete(&cqe->done);
1931 }
1932
1933 /*
1934  * Post a WR and block until its completion is reaped for the SQ.
1935  */
1936 static void __ib_drain_sq(struct ib_qp *qp)
1937 {
1938         struct ib_qp_attr attr = { .qp_state = IB_QPS_ERR };
1939         struct ib_drain_cqe sdrain;
1940         struct ib_send_wr swr = {}, *bad_swr;
1941         int ret;
1942
1943         if (qp->send_cq->poll_ctx == IB_POLL_DIRECT) {
1944                 WARN_ONCE(qp->send_cq->poll_ctx == IB_POLL_DIRECT,
1945                           "IB_POLL_DIRECT poll_ctx not supported for drain\n");
1946                 return;
1947         }
1948
1949         swr.wr_cqe = &sdrain.cqe;
1950         sdrain.cqe.done = ib_drain_qp_done;
1951         init_completion(&sdrain.done);
1952
1953         ret = ib_modify_qp(qp, &attr, IB_QP_STATE);
1954         if (ret) {
1955                 WARN_ONCE(ret, "failed to drain send queue: %d\n", ret);
1956                 return;
1957         }
1958
1959         ret = ib_post_send(qp, &swr, &bad_swr);
1960         if (ret) {
1961                 WARN_ONCE(ret, "failed to drain send queue: %d\n", ret);
1962                 return;
1963         }
1964
1965         wait_for_completion(&sdrain.done);
1966 }
1967
1968 /*
1969  * Post a WR and block until its completion is reaped for the RQ.
1970  */
1971 static void __ib_drain_rq(struct ib_qp *qp)
1972 {
1973         struct ib_qp_attr attr = { .qp_state = IB_QPS_ERR };
1974         struct ib_drain_cqe rdrain;
1975         struct ib_recv_wr rwr = {}, *bad_rwr;
1976         int ret;
1977
1978         if (qp->recv_cq->poll_ctx == IB_POLL_DIRECT) {
1979                 WARN_ONCE(qp->recv_cq->poll_ctx == IB_POLL_DIRECT,
1980                           "IB_POLL_DIRECT poll_ctx not supported for drain\n");
1981                 return;
1982         }
1983
1984         rwr.wr_cqe = &rdrain.cqe;
1985         rdrain.cqe.done = ib_drain_qp_done;
1986         init_completion(&rdrain.done);
1987
1988         ret = ib_modify_qp(qp, &attr, IB_QP_STATE);
1989         if (ret) {
1990                 WARN_ONCE(ret, "failed to drain recv queue: %d\n", ret);
1991                 return;
1992         }
1993
1994         ret = ib_post_recv(qp, &rwr, &bad_rwr);
1995         if (ret) {
1996                 WARN_ONCE(ret, "failed to drain recv queue: %d\n", ret);
1997                 return;
1998         }
1999
2000         wait_for_completion(&rdrain.done);
2001 }
2002
2003 /**
2004  * ib_drain_sq() - Block until all SQ CQEs have been consumed by the
2005  *                 application.
2006  * @qp:            queue pair to drain
2007  *
2008  * If the device has a provider-specific drain function, then
2009  * call that.  Otherwise call the generic drain function
2010  * __ib_drain_sq().
2011  *
2012  * The caller must:
2013  *
2014  * ensure there is room in the CQ and SQ for the drain work request and
2015  * completion.
2016  *
2017  * allocate the CQ using ib_alloc_cq() and the CQ poll context cannot be
2018  * IB_POLL_DIRECT.
2019  *
2020  * ensure that there are no other contexts that are posting WRs concurrently.
2021  * Otherwise the drain is not guaranteed.
2022  */
2023 void ib_drain_sq(struct ib_qp *qp)
2024 {
2025         if (qp->device->drain_sq)
2026                 qp->device->drain_sq(qp);
2027         else
2028                 __ib_drain_sq(qp);
2029 }
2030 EXPORT_SYMBOL(ib_drain_sq);
2031
2032 /**
2033  * ib_drain_rq() - Block until all RQ CQEs have been consumed by the
2034  *                 application.
2035  * @qp:            queue pair to drain
2036  *
2037  * If the device has a provider-specific drain function, then
2038  * call that.  Otherwise call the generic drain function
2039  * __ib_drain_rq().
2040  *
2041  * The caller must:
2042  *
2043  * ensure there is room in the CQ and RQ for the drain work request and
2044  * completion.
2045  *
2046  * allocate the CQ using ib_alloc_cq() and the CQ poll context cannot be
2047  * IB_POLL_DIRECT.
2048  *
2049  * ensure that there are no other contexts that are posting WRs concurrently.
2050  * Otherwise the drain is not guaranteed.
2051  */
2052 void ib_drain_rq(struct ib_qp *qp)
2053 {
2054         if (qp->device->drain_rq)
2055                 qp->device->drain_rq(qp);
2056         else
2057                 __ib_drain_rq(qp);
2058 }
2059 EXPORT_SYMBOL(ib_drain_rq);
2060
2061 /**
2062  * ib_drain_qp() - Block until all CQEs have been consumed by the
2063  *                 application on both the RQ and SQ.
2064  * @qp:            queue pair to drain
2065  *
2066  * The caller must:
2067  *
2068  * ensure there is room in the CQ(s), SQ, and RQ for drain work requests
2069  * and completions.
2070  *
2071  * allocate the CQs using ib_alloc_cq() and the CQ poll context cannot be
2072  * IB_POLL_DIRECT.
2073  *
2074  * ensure that there are no other contexts that are posting WRs concurrently.
2075  * Otherwise the drain is not guaranteed.
2076  */
2077 void ib_drain_qp(struct ib_qp *qp)
2078 {
2079         ib_drain_sq(qp);
2080         if (!qp->srq)
2081                 ib_drain_rq(qp);
2082 }
2083 EXPORT_SYMBOL(ib_drain_qp);