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         /*
 762          * If the callers is using the RDMA API calculate the resources
 763          * needed for the RDMA READ/WRITE operations.
 764          *
 765          * Note that these callers need to pass in a port number.
 766          */
 767         if (qp_init_attr->cap.max_rdma_ctxs)
 768                 rdma_rw_init_qp(device, qp_init_attr);
 769
 770         qp = device->create_qp(pd, qp_init_attr, NULL);
 771         if (IS_ERR(qp))
 772                 return qp;
 773
 774         qp->device     = device;
 775         qp->real_qp    = qp;
 776         qp->uobject    = NULL;
 777         qp->qp_type    = qp_init_attr->qp_type;
 778
 779         atomic_set(&qp->usecnt, 0);
 780         qp->mrs_used = 0;
 781         spin_lock_init(&qp->mr_lock);
 782         INIT_LIST_HEAD(&qp->rdma_mrs);
 783         INIT_LIST_HEAD(&qp->sig_mrs);
 784
 785         if (qp_init_attr->qp_type == IB_QPT_XRC_TGT)
 786                 return ib_create_xrc_qp(qp, qp_init_attr);
 787
 788         qp->event_handler = qp_init_attr->event_handler;
 789         qp->qp_context = qp_init_attr->qp_context;
 790         if (qp_init_attr->qp_type == IB_QPT_XRC_INI) {
 791                 qp->recv_cq = NULL;
 792                 qp->srq = NULL;
 793         } else {
 794                 qp->recv_cq = qp_init_attr->recv_cq;
 795                 atomic_inc(&qp_init_attr->recv_cq->usecnt);
 796                 qp->srq = qp_init_attr->srq;
 797                 if (qp->srq)
 798                         atomic_inc(&qp_init_attr->srq->usecnt);
 799         }
 800
 801         qp->pd      = pd;
 802         qp->send_cq = qp_init_attr->send_cq;
 803         qp->xrcd    = NULL;
 804
 805         atomic_inc(&pd->usecnt);
 806         atomic_inc(&qp_init_attr->send_cq->usecnt);
 807
 808         if (qp_init_attr->cap.max_rdma_ctxs) {
 809                 ret = rdma_rw_init_mrs(qp, qp_init_attr);
 810                 if (ret) {
 811                         pr_err("failed to init MR pool ret= %d\n", ret);
 812                         ib_destroy_qp(qp);
 813                         qp = ERR_PTR(ret);
 814                 }
 815         }
 816
 817         return qp;
 818 }
 819 EXPORT_SYMBOL(ib_create_qp);
 820
 821 static const struct {
 822         int                     valid;
 823         enum ib_qp_attr_mask    req_param[IB_QPT_MAX];
 824         enum ib_qp_attr_mask    opt_param[IB_QPT_MAX];
 825 } qp_state_table[IB_QPS_ERR + 1][IB_QPS_ERR + 1] = {
 826         [IB_QPS_RESET] = {
 827                 [IB_QPS_RESET] = { .valid = 1 },
 828                 [IB_QPS_INIT]  = {
 829                         .valid = 1,
 830                         .req_param = {
 831                                 [IB_QPT_UD]  = (IB_QP_PKEY_INDEX                |
 832                                                 IB_QP_PORT                      |
 833                                                 IB_QP_QKEY),
 834                                 [IB_QPT_RAW_PACKET] = IB_QP_PORT,
 835                                 [IB_QPT_UC]  = (IB_QP_PKEY_INDEX                |
 836                                                 IB_QP_PORT                      |
 837                                                 IB_QP_ACCESS_FLAGS),
 838                                 [IB_QPT_RC]  = (IB_QP_PKEY_INDEX                |
 839                                                 IB_QP_PORT                      |
 840                                                 IB_QP_ACCESS_FLAGS),
 841                                 [IB_QPT_XRC_INI] = (IB_QP_PKEY_INDEX            |
 842                                                 IB_QP_PORT                      |
 843                                                 IB_QP_ACCESS_FLAGS),
 844                                 [IB_QPT_XRC_TGT] = (IB_QP_PKEY_INDEX            |
 845                                                 IB_QP_PORT                      |
 846                                                 IB_QP_ACCESS_FLAGS),
 847                                 [IB_QPT_SMI] = (IB_QP_PKEY_INDEX                |
 848                                                 IB_QP_QKEY),
 849                                 [IB_QPT_GSI] = (IB_QP_PKEY_INDEX                |
 850                                                 IB_QP_QKEY),
 851                         }
 852                 },
 853         },
 854         [IB_QPS_INIT]  = {
 855                 [IB_QPS_RESET] = { .valid = 1 },
 856                 [IB_QPS_ERR] =   { .valid = 1 },
 857                 [IB_QPS_INIT]  = {
 858                         .valid = 1,
 859                         .opt_param = {
 860                                 [IB_QPT_UD]  = (IB_QP_PKEY_INDEX                |
 861                                                 IB_QP_PORT                      |
 862                                                 IB_QP_QKEY),
 863                                 [IB_QPT_UC]  = (IB_QP_PKEY_INDEX                |
 864                                                 IB_QP_PORT                      |
 865                                                 IB_QP_ACCESS_FLAGS),
 866                                 [IB_QPT_RC]  = (IB_QP_PKEY_INDEX                |
 867                                                 IB_QP_PORT                      |
 868                                                 IB_QP_ACCESS_FLAGS),
 869                                 [IB_QPT_XRC_INI] = (IB_QP_PKEY_INDEX            |
 870                                                 IB_QP_PORT                      |
 871                                                 IB_QP_ACCESS_FLAGS),
 872                                 [IB_QPT_XRC_TGT] = (IB_QP_PKEY_INDEX            |
 873                                                 IB_QP_PORT                      |
 874                                                 IB_QP_ACCESS_FLAGS),
 875                                 [IB_QPT_SMI] = (IB_QP_PKEY_INDEX                |
 876                                                 IB_QP_QKEY),
 877                                 [IB_QPT_GSI] = (IB_QP_PKEY_INDEX                |
 878                                                 IB_QP_QKEY),
 879                         }
 880                 },
 881                 [IB_QPS_RTR]   = {
 882                         .valid = 1,
 883                         .req_param = {
 884                                 [IB_QPT_UC]  = (IB_QP_AV                        |
 885                                                 IB_QP_PATH_MTU                  |
 886                                                 IB_QP_DEST_QPN                  |
 887                                                 IB_QP_RQ_PSN),
 888                                 [IB_QPT_RC]  = (IB_QP_AV                        |
 889                                                 IB_QP_PATH_MTU                  |
 890                                                 IB_QP_DEST_QPN                  |
 891                                                 IB_QP_RQ_PSN                    |
 892                                                 IB_QP_MAX_DEST_RD_ATOMIC        |
 893                                                 IB_QP_MIN_RNR_TIMER),
 894                                 [IB_QPT_XRC_INI] = (IB_QP_AV                    |
 895                                                 IB_QP_PATH_MTU                  |
 896                                                 IB_QP_DEST_QPN                  |
 897                                                 IB_QP_RQ_PSN),
 898                                 [IB_QPT_XRC_TGT] = (IB_QP_AV                    |
 899                                                 IB_QP_PATH_MTU                  |
 900                                                 IB_QP_DEST_QPN                  |
 901                                                 IB_QP_RQ_PSN                    |
 902                                                 IB_QP_MAX_DEST_RD_ATOMIC        |
 903                                                 IB_QP_MIN_RNR_TIMER),
 904                         },
 905                         .opt_param = {
 906                                  [IB_QPT_UD]  = (IB_QP_PKEY_INDEX               |
 907                                                  IB_QP_QKEY),
 908                                  [IB_QPT_UC]  = (IB_QP_ALT_PATH                 |
 909                                                  IB_QP_ACCESS_FLAGS             |
 910                                                  IB_QP_PKEY_INDEX),
 911                                  [IB_QPT_RC]  = (IB_QP_ALT_PATH                 |
 912                                                  IB_QP_ACCESS_FLAGS             |
 913                                                  IB_QP_PKEY_INDEX),
 914                                  [IB_QPT_XRC_INI] = (IB_QP_ALT_PATH             |
 915                                                  IB_QP_ACCESS_FLAGS             |
 916                                                  IB_QP_PKEY_INDEX),
 917                                  [IB_QPT_XRC_TGT] = (IB_QP_ALT_PATH             |
 918                                                  IB_QP_ACCESS_FLAGS             |
 919                                                  IB_QP_PKEY_INDEX),
 920                                  [IB_QPT_SMI] = (IB_QP_PKEY_INDEX               |
 921                                                  IB_QP_QKEY),
 922                                  [IB_QPT_GSI] = (IB_QP_PKEY_INDEX               |
 923                                                  IB_QP_QKEY),
 924                          },
 925                 },
 926         },
 927         [IB_QPS_RTR]   = {
 928                 [IB_QPS_RESET] = { .valid = 1 },
 929                 [IB_QPS_ERR] =   { .valid = 1 },
 930                 [IB_QPS_RTS]   = {
 931                         .valid = 1,
 932                         .req_param = {
 933                                 [IB_QPT_UD]  = IB_QP_SQ_PSN,
 934                                 [IB_QPT_UC]  = IB_QP_SQ_PSN,
 935                                 [IB_QPT_RC]  = (IB_QP_TIMEOUT                   |
 936                                                 IB_QP_RETRY_CNT                 |
 937                                                 IB_QP_RNR_RETRY                 |
 938                                                 IB_QP_SQ_PSN                    |
 939                                                 IB_QP_MAX_QP_RD_ATOMIC),
 940                                 [IB_QPT_XRC_INI] = (IB_QP_TIMEOUT               |
 941                                                 IB_QP_RETRY_CNT                 |
 942                                                 IB_QP_RNR_RETRY                 |
 943                                                 IB_QP_SQ_PSN                    |
 944                                                 IB_QP_MAX_QP_RD_ATOMIC),
 945                                 [IB_QPT_XRC_TGT] = (IB_QP_TIMEOUT               |
 946                                                 IB_QP_SQ_PSN),
 947                                 [IB_QPT_SMI] = IB_QP_SQ_PSN,
 948                                 [IB_QPT_GSI] = IB_QP_SQ_PSN,
 949                         },
 950                         .opt_param = {
 951                                  [IB_QPT_UD]  = (IB_QP_CUR_STATE                |
 952                                                  IB_QP_QKEY),
 953                                  [IB_QPT_UC]  = (IB_QP_CUR_STATE                |
 954                                                  IB_QP_ALT_PATH                 |
 955                                                  IB_QP_ACCESS_FLAGS             |
 956                                                  IB_QP_PATH_MIG_STATE),
 957                                  [IB_QPT_RC]  = (IB_QP_CUR_STATE                |
 958                                                  IB_QP_ALT_PATH                 |
 959                                                  IB_QP_ACCESS_FLAGS             |
 960                                                  IB_QP_MIN_RNR_TIMER            |
 961                                                  IB_QP_PATH_MIG_STATE),
 962                                  [IB_QPT_XRC_INI] = (IB_QP_CUR_STATE            |
 963                                                  IB_QP_ALT_PATH                 |
 964                                                  IB_QP_ACCESS_FLAGS             |
 965                                                  IB_QP_PATH_MIG_STATE),
 966                                  [IB_QPT_XRC_TGT] = (IB_QP_CUR_STATE            |
 967                                                  IB_QP_ALT_PATH                 |
 968                                                  IB_QP_ACCESS_FLAGS             |
 969                                                  IB_QP_MIN_RNR_TIMER            |
 970                                                  IB_QP_PATH_MIG_STATE),
 971                                  [IB_QPT_SMI] = (IB_QP_CUR_STATE                |
 972                                                  IB_QP_QKEY),
 973                                  [IB_QPT_GSI] = (IB_QP_CUR_STATE                |
 974                                                  IB_QP_QKEY),
 975                          }
 976                 }
 977         },
 978         [IB_QPS_RTS]   = {
 979                 [IB_QPS_RESET] = { .valid = 1 },
 980                 [IB_QPS_ERR] =   { .valid = 1 },
 981                 [IB_QPS_RTS]   = {
 982                         .valid = 1,
 983                         .opt_param = {
 984                                 [IB_QPT_UD]  = (IB_QP_CUR_STATE                 |
 985                                                 IB_QP_QKEY),
 986                                 [IB_QPT_UC]  = (IB_QP_CUR_STATE                 |
 987                                                 IB_QP_ACCESS_FLAGS              |
 988                                                 IB_QP_ALT_PATH                  |
 989                                                 IB_QP_PATH_MIG_STATE),
 990                                 [IB_QPT_RC]  = (IB_QP_CUR_STATE                 |
 991                                                 IB_QP_ACCESS_FLAGS              |
 992                                                 IB_QP_ALT_PATH                  |
 993                                                 IB_QP_PATH_MIG_STATE            |
 994                                                 IB_QP_MIN_RNR_TIMER),
 995                                 [IB_QPT_XRC_INI] = (IB_QP_CUR_STATE             |
 996                                                 IB_QP_ACCESS_FLAGS              |
 997                                                 IB_QP_ALT_PATH                  |
 998                                                 IB_QP_PATH_MIG_STATE),
 999                                 [IB_QPT_XRC_TGT] = (IB_QP_CUR_STATE             |
1000                                                 IB_QP_ACCESS_FLAGS              |
1001                                                 IB_QP_ALT_PATH                  |
1002                                                 IB_QP_PATH_MIG_STATE            |
1003                                                 IB_QP_MIN_RNR_TIMER),
1004                                 [IB_QPT_SMI] = (IB_QP_CUR_STATE                 |
1005                                                 IB_QP_QKEY),
1006                                 [IB_QPT_GSI] = (IB_QP_CUR_STATE                 |
1007                                                 IB_QP_QKEY),
1008                         }
1009                 },
1010                 [IB_QPS_SQD]   = {
1011                         .valid = 1,
1012                         .opt_param = {
1013                                 [IB_QPT_UD]  = IB_QP_EN_SQD_ASYNC_NOTIFY,
1014                                 [IB_QPT_UC]  = IB_QP_EN_SQD_ASYNC_NOTIFY,
1015                                 [IB_QPT_RC]  = IB_QP_EN_SQD_ASYNC_NOTIFY,
1016                                 [IB_QPT_XRC_INI] = IB_QP_EN_SQD_ASYNC_NOTIFY,
1017                                 [IB_QPT_XRC_TGT] = IB_QP_EN_SQD_ASYNC_NOTIFY, /* ??? */
1018                                 [IB_QPT_SMI] = IB_QP_EN_SQD_ASYNC_NOTIFY,
1019                                 [IB_QPT_GSI] = IB_QP_EN_SQD_ASYNC_NOTIFY
1020                         }
1021                 },
1022         },
1023         [IB_QPS_SQD]   = {
1024                 [IB_QPS_RESET] = { .valid = 1 },
1025                 [IB_QPS_ERR] =   { .valid = 1 },
1026                 [IB_QPS_RTS]   = {
1027                         .valid = 1,
1028                         .opt_param = {
1029                                 [IB_QPT_UD]  = (IB_QP_CUR_STATE                 |
1030                                                 IB_QP_QKEY),
1031                                 [IB_QPT_UC]  = (IB_QP_CUR_STATE                 |
1032                                                 IB_QP_ALT_PATH                  |
1033                                                 IB_QP_ACCESS_FLAGS              |
1034                                                 IB_QP_PATH_MIG_STATE),
1035                                 [IB_QPT_RC]  = (IB_QP_CUR_STATE                 |
1036                                                 IB_QP_ALT_PATH                  |
1037                                                 IB_QP_ACCESS_FLAGS              |
1038                                                 IB_QP_MIN_RNR_TIMER             |
1039                                                 IB_QP_PATH_MIG_STATE),
1040                                 [IB_QPT_XRC_INI] = (IB_QP_CUR_STATE             |
1041                                                 IB_QP_ALT_PATH                  |
1042                                                 IB_QP_ACCESS_FLAGS              |
1043                                                 IB_QP_PATH_MIG_STATE),
1044                                 [IB_QPT_XRC_TGT] = (IB_QP_CUR_STATE             |
1045                                                 IB_QP_ALT_PATH                  |
1046                                                 IB_QP_ACCESS_FLAGS              |
1047                                                 IB_QP_MIN_RNR_TIMER             |
1048                                                 IB_QP_PATH_MIG_STATE),
1049                                 [IB_QPT_SMI] = (IB_QP_CUR_STATE                 |
1050                                                 IB_QP_QKEY),
1051                                 [IB_QPT_GSI] = (IB_QP_CUR_STATE                 |
1052                                                 IB_QP_QKEY),
1053                         }
1054                 },
1055                 [IB_QPS_SQD]   = {
1056                         .valid = 1,
1057                         .opt_param = {
1058                                 [IB_QPT_UD]  = (IB_QP_PKEY_INDEX                |
1059                                                 IB_QP_QKEY),
1060                                 [IB_QPT_UC]  = (IB_QP_AV                        |
1061                                                 IB_QP_ALT_PATH                  |
1062                                                 IB_QP_ACCESS_FLAGS              |
1063                                                 IB_QP_PKEY_INDEX                |
1064                                                 IB_QP_PATH_MIG_STATE),
1065                                 [IB_QPT_RC]  = (IB_QP_PORT                      |
1066                                                 IB_QP_AV                        |
1067                                                 IB_QP_TIMEOUT                   |
1068                                                 IB_QP_RETRY_CNT                 |
1069                                                 IB_QP_RNR_RETRY                 |
1070                                                 IB_QP_MAX_QP_RD_ATOMIC          |
1071                                                 IB_QP_MAX_DEST_RD_ATOMIC        |
1072                                                 IB_QP_ALT_PATH                  |
1073                                                 IB_QP_ACCESS_FLAGS              |
1074                                                 IB_QP_PKEY_INDEX                |
1075                                                 IB_QP_MIN_RNR_TIMER             |
1076                                                 IB_QP_PATH_MIG_STATE),
1077                                 [IB_QPT_XRC_INI] = (IB_QP_PORT                  |
1078                                                 IB_QP_AV                        |
1079                                                 IB_QP_TIMEOUT                   |
1080                                                 IB_QP_RETRY_CNT                 |
1081                                                 IB_QP_RNR_RETRY                 |
1082                                                 IB_QP_MAX_QP_RD_ATOMIC          |
1083                                                 IB_QP_ALT_PATH                  |
1084                                                 IB_QP_ACCESS_FLAGS              |
1085                                                 IB_QP_PKEY_INDEX                |
1086                                                 IB_QP_PATH_MIG_STATE),
1087                                 [IB_QPT_XRC_TGT] = (IB_QP_PORT                  |
1088                                                 IB_QP_AV                        |
1089                                                 IB_QP_TIMEOUT                   |
1090                                                 IB_QP_MAX_DEST_RD_ATOMIC        |
1091                                                 IB_QP_ALT_PATH                  |
1092                                                 IB_QP_ACCESS_FLAGS              |
1093                                                 IB_QP_PKEY_INDEX                |
1094                                                 IB_QP_MIN_RNR_TIMER             |
1095                                                 IB_QP_PATH_MIG_STATE),
1096                                 [IB_QPT_SMI] = (IB_QP_PKEY_INDEX                |
1097                                                 IB_QP_QKEY),
1098                                 [IB_QPT_GSI] = (IB_QP_PKEY_INDEX                |
1099                                                 IB_QP_QKEY),
1100                         }
1101                 }
1102         },
1103         [IB_QPS_SQE]   = {
1104                 [IB_QPS_RESET] = { .valid = 1 },
1105                 [IB_QPS_ERR] =   { .valid = 1 },
1106                 [IB_QPS_RTS]   = {
1107                         .valid = 1,
1108                         .opt_param = {
1109                                 [IB_QPT_UD]  = (IB_QP_CUR_STATE                 |
1110                                                 IB_QP_QKEY),
1111                                 [IB_QPT_UC]  = (IB_QP_CUR_STATE                 |
1112                                                 IB_QP_ACCESS_FLAGS),
1113                                 [IB_QPT_SMI] = (IB_QP_CUR_STATE                 |
1114                                                 IB_QP_QKEY),
1115                                 [IB_QPT_GSI] = (IB_QP_CUR_STATE                 |
1116                                                 IB_QP_QKEY),
1117                         }
1118                 }
1119         },
1120         [IB_QPS_ERR] = {
1121                 [IB_QPS_RESET] = { .valid = 1 },
1122                 [IB_QPS_ERR] =   { .valid = 1 }
1123         }
1124 };
1125
1126 int ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
1127                        enum ib_qp_type type, enum ib_qp_attr_mask mask,
1128                        enum rdma_link_layer ll)
1129 {
1130         enum ib_qp_attr_mask req_param, opt_param;
1131
1132         if (cur_state  < 0 || cur_state  > IB_QPS_ERR ||
1133             next_state < 0 || next_state > IB_QPS_ERR)
1134                 return 0;
1135
1136         if (mask & IB_QP_CUR_STATE  &&
1137             cur_state != IB_QPS_RTR && cur_state != IB_QPS_RTS &&
1138             cur_state != IB_QPS_SQD && cur_state != IB_QPS_SQE)
1139                 return 0;
1140
1141         if (!qp_state_table[cur_state][next_state].valid)
1142                 return 0;
1143
1144         req_param = qp_state_table[cur_state][next_state].req_param[type];
1145         opt_param = qp_state_table[cur_state][next_state].opt_param[type];
1146
1147         if ((mask & req_param) != req_param)
1148                 return 0;
1149
1150         if (mask & ~(req_param | opt_param | IB_QP_STATE))
1151                 return 0;
1152
1153         return 1;
1154 }
1155 EXPORT_SYMBOL(ib_modify_qp_is_ok);
1156
1157 int ib_resolve_eth_dmac(struct ib_qp *qp,
1158                         struct ib_qp_attr *qp_attr, int *qp_attr_mask)
1159 {
1160         int           ret = 0;
1161
1162         if (*qp_attr_mask & IB_QP_AV) {
1163                 if (qp_attr->ah_attr.port_num < rdma_start_port(qp->device) ||
1164                     qp_attr->ah_attr.port_num > rdma_end_port(qp->device))
1165                         return -EINVAL;
1166
1167                 if (!rdma_cap_eth_ah(qp->device, qp_attr->ah_attr.port_num))
1168                         return 0;
1169
1170                 if (rdma_link_local_addr((struct in6_addr *)qp_attr->ah_attr.grh.dgid.raw)) {
1171                         rdma_get_ll_mac((struct in6_addr *)qp_attr->ah_attr.grh.dgid.raw,
1172                                         qp_attr->ah_attr.dmac);
1173                 } else {
1174                         union ib_gid            sgid;
1175                         struct ib_gid_attr      sgid_attr;
1176                         int                     ifindex;
1177                         int                     hop_limit;
1178
1179                         ret = ib_query_gid(qp->device,
1180                                            qp_attr->ah_attr.port_num,
1181                                            qp_attr->ah_attr.grh.sgid_index,
1182                                            &sgid, &sgid_attr);
1183
1184                         if (ret || !sgid_attr.ndev) {
1185                                 if (!ret)
1186                                         ret = -ENXIO;
1187                                 goto out;
1188                         }
1189
1190                         ifindex = sgid_attr.ndev->ifindex;
1191
1192                         ret = rdma_addr_find_l2_eth_by_grh(&sgid,
1193                                                            &qp_attr->ah_attr.grh.dgid,
1194                                                            qp_attr->ah_attr.dmac,
1195                                                            NULL, &ifindex, &hop_limit);
1196
1197                         dev_put(sgid_attr.ndev);
1198
1199                         qp_attr->ah_attr.grh.hop_limit = hop_limit;
1200                 }
1201         }
1202 out:
1203         return ret;
1204 }
1205 EXPORT_SYMBOL(ib_resolve_eth_dmac);
1206
1207
1208 int ib_modify_qp(struct ib_qp *qp,
1209                  struct ib_qp_attr *qp_attr,
1210                  int qp_attr_mask)
1211 {
1212         int ret;
1213
1214         ret = ib_resolve_eth_dmac(qp, qp_attr, &qp_attr_mask);
1215         if (ret)
1216                 return ret;
1217
1218         return qp->device->modify_qp(qp->real_qp, qp_attr, qp_attr_mask, NULL);
1219 }
1220 EXPORT_SYMBOL(ib_modify_qp);
1221
1222 int ib_query_qp(struct ib_qp *qp,
1223                 struct ib_qp_attr *qp_attr,
1224                 int qp_attr_mask,
1225                 struct ib_qp_init_attr *qp_init_attr)
1226 {
1227         return qp->device->query_qp ?
1228                 qp->device->query_qp(qp->real_qp, qp_attr, qp_attr_mask, qp_init_attr) :
1229                 -ENOSYS;
1230 }
1231 EXPORT_SYMBOL(ib_query_qp);
1232
1233 int ib_close_qp(struct ib_qp *qp)
1234 {
1235         struct ib_qp *real_qp;
1236         unsigned long flags;
1237
1238         real_qp = qp->real_qp;
1239         if (real_qp == qp)
1240                 return -EINVAL;
1241
1242         spin_lock_irqsave(&real_qp->device->event_handler_lock, flags);
1243         list_del(&qp->open_list);
1244         spin_unlock_irqrestore(&real_qp->device->event_handler_lock, flags);
1245
1246         atomic_dec(&real_qp->usecnt);
1247         kfree(qp);
1248
1249         return 0;
1250 }
1251 EXPORT_SYMBOL(ib_close_qp);
1252
1253 static int __ib_destroy_shared_qp(struct ib_qp *qp)
1254 {
1255         struct ib_xrcd *xrcd;
1256         struct ib_qp *real_qp;
1257         int ret;
1258
1259         real_qp = qp->real_qp;
1260         xrcd = real_qp->xrcd;
1261
1262         mutex_lock(&xrcd->tgt_qp_mutex);
1263         ib_close_qp(qp);
1264         if (atomic_read(&real_qp->usecnt) == 0)
1265                 list_del(&real_qp->xrcd_list);
1266         else
1267                 real_qp = NULL;
1268         mutex_unlock(&xrcd->tgt_qp_mutex);
1269
1270         if (real_qp) {
1271                 ret = ib_destroy_qp(real_qp);
1272                 if (!ret)
1273                         atomic_dec(&xrcd->usecnt);
1274                 else
1275                         __ib_insert_xrcd_qp(xrcd, real_qp);
1276         }
1277
1278         return 0;
1279 }
1280
1281 int ib_destroy_qp(struct ib_qp *qp)
1282 {
1283         struct ib_pd *pd;
1284         struct ib_cq *scq, *rcq;
1285         struct ib_srq *srq;
1286         int ret;
1287
1288         WARN_ON_ONCE(qp->mrs_used > 0);
1289
1290         if (atomic_read(&qp->usecnt))
1291                 return -EBUSY;
1292
1293         if (qp->real_qp != qp)
1294                 return __ib_destroy_shared_qp(qp);
1295
1296         pd   = qp->pd;
1297         scq  = qp->send_cq;
1298         rcq  = qp->recv_cq;
1299         srq  = qp->srq;
1300
1301         if (!qp->uobject)
1302                 rdma_rw_cleanup_mrs(qp);
1303
1304         ret = qp->device->destroy_qp(qp);
1305         if (!ret) {
1306                 if (pd)
1307                         atomic_dec(&pd->usecnt);
1308                 if (scq)
1309                         atomic_dec(&scq->usecnt);
1310                 if (rcq)
1311                         atomic_dec(&rcq->usecnt);
1312                 if (srq)
1313                         atomic_dec(&srq->usecnt);
1314         }
1315
1316         return ret;
1317 }
1318 EXPORT_SYMBOL(ib_destroy_qp);
1319
1320 /* Completion queues */
1321
1322 struct ib_cq *ib_create_cq(struct ib_device *device,
1323                            ib_comp_handler comp_handler,
1324                            void (*event_handler)(struct ib_event *, void *),
1325                            void *cq_context,
1326                            const struct ib_cq_init_attr *cq_attr)
1327 {
1328         struct ib_cq *cq;
1329
1330         cq = device->create_cq(device, cq_attr, NULL, NULL);
1331
1332         if (!IS_ERR(cq)) {
1333                 cq->device        = device;
1334                 cq->uobject       = NULL;
1335                 cq->comp_handler  = comp_handler;
1336                 cq->event_handler = event_handler;
1337                 cq->cq_context    = cq_context;
1338                 atomic_set(&cq->usecnt, 0);
1339         }
1340
1341         return cq;
1342 }
1343 EXPORT_SYMBOL(ib_create_cq);
1344
1345 int ib_modify_cq(struct ib_cq *cq, u16 cq_count, u16 cq_period)
1346 {
1347         return cq->device->modify_cq ?
1348                 cq->device->modify_cq(cq, cq_count, cq_period) : -ENOSYS;
1349 }
1350 EXPORT_SYMBOL(ib_modify_cq);
1351
1352 int ib_destroy_cq(struct ib_cq *cq)
1353 {
1354         if (atomic_read(&cq->usecnt))
1355                 return -EBUSY;
1356
1357         return cq->device->destroy_cq(cq);
1358 }
1359 EXPORT_SYMBOL(ib_destroy_cq);
1360
1361 int ib_resize_cq(struct ib_cq *cq, int cqe)
1362 {
1363         return cq->device->resize_cq ?
1364                 cq->device->resize_cq(cq, cqe, NULL) : -ENOSYS;
1365 }
1366 EXPORT_SYMBOL(ib_resize_cq);
1367
1368 /* Memory regions */
1369
1370 struct ib_mr *ib_get_dma_mr(struct ib_pd *pd, int mr_access_flags)
1371 {
1372         struct ib_mr *mr;
1373         int err;
1374
1375         err = ib_check_mr_access(mr_access_flags);
1376         if (err)
1377                 return ERR_PTR(err);
1378
1379         mr = pd->device->get_dma_mr(pd, mr_access_flags);
1380
1381         if (!IS_ERR(mr)) {
1382                 mr->device  = pd->device;
1383                 mr->pd      = pd;
1384                 mr->uobject = NULL;
1385                 atomic_inc(&pd->usecnt);
1386                 mr->need_inval = false;
1387         }
1388
1389         return mr;
1390 }
1391 EXPORT_SYMBOL(ib_get_dma_mr);
1392
1393 int ib_dereg_mr(struct ib_mr *mr)
1394 {
1395         struct ib_pd *pd = mr->pd;
1396         int ret;
1397
1398         ret = mr->device->dereg_mr(mr);
1399         if (!ret)
1400                 atomic_dec(&pd->usecnt);
1401
1402         return ret;
1403 }
1404 EXPORT_SYMBOL(ib_dereg_mr);
1405
1406 /**
1407  * ib_alloc_mr() - Allocates a memory region
1408  * @pd:            protection domain associated with the region
1409  * @mr_type:       memory region type
1410  * @max_num_sg:    maximum sg entries available for registration.
1411  *
1412  * Notes:
1413  * Memory registeration page/sg lists must not exceed max_num_sg.
1414  * For mr_type IB_MR_TYPE_MEM_REG, the total length cannot exceed
1415  * max_num_sg * used_page_size.
1416  *
1417  */
1418 struct ib_mr *ib_alloc_mr(struct ib_pd *pd,
1419                           enum ib_mr_type mr_type,
1420                           u32 max_num_sg)
1421 {
1422         struct ib_mr *mr;
1423
1424         if (!pd->device->alloc_mr)
1425                 return ERR_PTR(-ENOSYS);
1426
1427         mr = pd->device->alloc_mr(pd, mr_type, max_num_sg);
1428         if (!IS_ERR(mr)) {
1429                 mr->device  = pd->device;
1430                 mr->pd      = pd;
1431                 mr->uobject = NULL;
1432                 atomic_inc(&pd->usecnt);
1433                 mr->need_inval = false;
1434         }
1435
1436         return mr;
1437 }
1438 EXPORT_SYMBOL(ib_alloc_mr);
1439
1440 /* "Fast" memory regions */
1441
1442 struct ib_fmr *ib_alloc_fmr(struct ib_pd *pd,
1443                             int mr_access_flags,
1444                             struct ib_fmr_attr *fmr_attr)
1445 {
1446         struct ib_fmr *fmr;
1447
1448         if (!pd->device->alloc_fmr)
1449                 return ERR_PTR(-ENOSYS);
1450
1451         fmr = pd->device->alloc_fmr(pd, mr_access_flags, fmr_attr);
1452         if (!IS_ERR(fmr)) {
1453                 fmr->device = pd->device;
1454                 fmr->pd     = pd;
1455                 atomic_inc(&pd->usecnt);
1456         }
1457
1458         return fmr;
1459 }
1460 EXPORT_SYMBOL(ib_alloc_fmr);
1461
1462 int ib_unmap_fmr(struct list_head *fmr_list)
1463 {
1464         struct ib_fmr *fmr;
1465
1466         if (list_empty(fmr_list))
1467                 return 0;
1468
1469         fmr = list_entry(fmr_list->next, struct ib_fmr, list);
1470         return fmr->device->unmap_fmr(fmr_list);
1471 }
1472 EXPORT_SYMBOL(ib_unmap_fmr);
1473
1474 int ib_dealloc_fmr(struct ib_fmr *fmr)
1475 {
1476         struct ib_pd *pd;
1477         int ret;
1478
1479         pd = fmr->pd;
1480         ret = fmr->device->dealloc_fmr(fmr);
1481         if (!ret)
1482                 atomic_dec(&pd->usecnt);
1483
1484         return ret;
1485 }
1486 EXPORT_SYMBOL(ib_dealloc_fmr);
1487
1488 /* Multicast groups */
1489
1490 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid)
1491 {
1492         int ret;
1493
1494         if (!qp->device->attach_mcast)
1495                 return -ENOSYS;
1496         if (gid->raw[0] != 0xff || qp->qp_type != IB_QPT_UD)
1497                 return -EINVAL;
1498
1499         ret = qp->device->attach_mcast(qp, gid, lid);
1500         if (!ret)
1501                 atomic_inc(&qp->usecnt);
1502         return ret;
1503 }
1504 EXPORT_SYMBOL(ib_attach_mcast);
1505
1506 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid)
1507 {
1508         int ret;
1509
1510         if (!qp->device->detach_mcast)
1511                 return -ENOSYS;
1512         if (gid->raw[0] != 0xff || qp->qp_type != IB_QPT_UD)
1513                 return -EINVAL;
1514
1515         ret = qp->device->detach_mcast(qp, gid, lid);
1516         if (!ret)
1517                 atomic_dec(&qp->usecnt);
1518         return ret;
1519 }
1520 EXPORT_SYMBOL(ib_detach_mcast);
1521
1522 struct ib_xrcd *ib_alloc_xrcd(struct ib_device *device)
1523 {
1524         struct ib_xrcd *xrcd;
1525
1526         if (!device->alloc_xrcd)
1527                 return ERR_PTR(-ENOSYS);
1528
1529         xrcd = device->alloc_xrcd(device, NULL, NULL);
1530         if (!IS_ERR(xrcd)) {
1531                 xrcd->device = device;
1532                 xrcd->inode = NULL;
1533                 atomic_set(&xrcd->usecnt, 0);
1534                 mutex_init(&xrcd->tgt_qp_mutex);
1535                 INIT_LIST_HEAD(&xrcd->tgt_qp_list);
1536         }
1537
1538         return xrcd;
1539 }
1540 EXPORT_SYMBOL(ib_alloc_xrcd);
1541
1542 int ib_dealloc_xrcd(struct ib_xrcd *xrcd)
1543 {
1544         struct ib_qp *qp;
1545         int ret;
1546
1547         if (atomic_read(&xrcd->usecnt))
1548                 return -EBUSY;
1549
1550         while (!list_empty(&xrcd->tgt_qp_list)) {
1551                 qp = list_entry(xrcd->tgt_qp_list.next, struct ib_qp, xrcd_list);
1552                 ret = ib_destroy_qp(qp);
1553                 if (ret)
1554                         return ret;
1555         }
1556
1557         return xrcd->device->dealloc_xrcd(xrcd);
1558 }
1559 EXPORT_SYMBOL(ib_dealloc_xrcd);
1560
1561 struct ib_flow *ib_create_flow(struct ib_qp *qp,
1562                                struct ib_flow_attr *flow_attr,
1563                                int domain)
1564 {
1565         struct ib_flow *flow_id;
1566         if (!qp->device->create_flow)
1567                 return ERR_PTR(-ENOSYS);
1568
1569         flow_id = qp->device->create_flow(qp, flow_attr, domain);
1570         if (!IS_ERR(flow_id))
1571                 atomic_inc(&qp->usecnt);
1572         return flow_id;
1573 }
1574 EXPORT_SYMBOL(ib_create_flow);
1575
1576 int ib_destroy_flow(struct ib_flow *flow_id)
1577 {
1578         int err;
1579         struct ib_qp *qp = flow_id->qp;
1580
1581         err = qp->device->destroy_flow(flow_id);
1582         if (!err)
1583                 atomic_dec(&qp->usecnt);
1584         return err;
1585 }
1586 EXPORT_SYMBOL(ib_destroy_flow);
1587
1588 int ib_check_mr_status(struct ib_mr *mr, u32 check_mask,
1589                        struct ib_mr_status *mr_status)
1590 {
1591         return mr->device->check_mr_status ?
1592                 mr->device->check_mr_status(mr, check_mask, mr_status) : -ENOSYS;
1593 }
1594 EXPORT_SYMBOL(ib_check_mr_status);
1595
1596 int ib_set_vf_link_state(struct ib_device *device, int vf, u8 port,
1597                          int state)
1598 {
1599         if (!device->set_vf_link_state)
1600                 return -ENOSYS;
1601
1602         return device->set_vf_link_state(device, vf, port, state);
1603 }
1604 EXPORT_SYMBOL(ib_set_vf_link_state);
1605
1606 int ib_get_vf_config(struct ib_device *device, int vf, u8 port,
1607                      struct ifla_vf_info *info)
1608 {
1609         if (!device->get_vf_config)
1610                 return -ENOSYS;
1611
1612         return device->get_vf_config(device, vf, port, info);
1613 }
1614 EXPORT_SYMBOL(ib_get_vf_config);
1615
1616 int ib_get_vf_stats(struct ib_device *device, int vf, u8 port,
1617                     struct ifla_vf_stats *stats)
1618 {
1619         if (!device->get_vf_stats)
1620                 return -ENOSYS;
1621
1622         return device->get_vf_stats(device, vf, port, stats);
1623 }
1624 EXPORT_SYMBOL(ib_get_vf_stats);
1625
1626 int ib_set_vf_guid(struct ib_device *device, int vf, u8 port, u64 guid,
1627                    int type)
1628 {
1629         if (!device->set_vf_guid)
1630                 return -ENOSYS;
1631
1632         return device->set_vf_guid(device, vf, port, guid, type);
1633 }
1634 EXPORT_SYMBOL(ib_set_vf_guid);
1635
1636 /**
1637  * ib_map_mr_sg() - Map the largest prefix of a dma mapped SG list
1638  *     and set it the memory region.
1639  * @mr:            memory region
1640  * @sg:            dma mapped scatterlist
1641  * @sg_nents:      number of entries in sg
1642  * @sg_offset:     offset in bytes into sg
1643  * @page_size:     page vector desired page size
1644  *
1645  * Constraints:
1646  * - The first sg element is allowed to have an offset.
1647  * - Each sg element must be aligned to page_size (or physically
1648  *   contiguous to the previous element). In case an sg element has a
1649  *   non contiguous offset, the mapping prefix will not include it.
1650  * - The last sg element is allowed to have length less than page_size.
1651  * - If sg_nents total byte length exceeds the mr max_num_sge * page_size
1652  *   then only max_num_sg entries will be mapped.
1653  * - If the MR was allocated with type IB_MR_TYPE_SG_GAPS_REG, non of these
1654  *   constraints holds and the page_size argument is ignored.
1655  *
1656  * Returns the number of sg elements that were mapped to the memory region.
1657  *
1658  * After this completes successfully, the  memory region
1659  * is ready for registration.
1660  */
1661 int ib_map_mr_sg(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
1662                  unsigned int *sg_offset, unsigned int page_size)
1663 {
1664         if (unlikely(!mr->device->map_mr_sg))
1665                 return -ENOSYS;
1666
1667         mr->page_size = page_size;
1668
1669         return mr->device->map_mr_sg(mr, sg, sg_nents, sg_offset);
1670 }
1671 EXPORT_SYMBOL(ib_map_mr_sg);
1672
1673 /**
1674  * ib_sg_to_pages() - Convert the largest prefix of a sg list
1675  *     to a page vector
1676  * @mr:            memory region
1677  * @sgl:           dma mapped scatterlist
1678  * @sg_nents:      number of entries in sg
1679  * @sg_offset_p:   IN:  start offset in bytes into sg
1680  *                 OUT: offset in bytes for element n of the sg of the first
1681  *                      byte that has not been processed where n is the return
1682  *                      value of this function.
1683  * @set_page:      driver page assignment function pointer
1684  *
1685  * Core service helper for drivers to convert the largest
1686  * prefix of given sg list to a page vector. The sg list
1687  * prefix converted is the prefix that meet the requirements
1688  * of ib_map_mr_sg.
1689  *
1690  * Returns the number of sg elements that were assigned to
1691  * a page vector.
1692  */
1693 int ib_sg_to_pages(struct ib_mr *mr, struct scatterlist *sgl, int sg_nents,
1694                 unsigned int *sg_offset_p, int (*set_page)(struct ib_mr *, u64))
1695 {
1696         struct scatterlist *sg;
1697         u64 last_end_dma_addr = 0;
1698         unsigned int sg_offset = sg_offset_p ? *sg_offset_p : 0;
1699         unsigned int last_page_off = 0;
1700         u64 page_mask = ~((u64)mr->page_size - 1);
1701         int i, ret;
1702
1703         if (unlikely(sg_nents <= 0 || sg_offset > sg_dma_len(&sgl[0])))
1704                 return -EINVAL;
1705
1706         mr->iova = sg_dma_address(&sgl[0]) + sg_offset;
1707         mr->length = 0;
1708
1709         for_each_sg(sgl, sg, sg_nents, i) {
1710                 u64 dma_addr = sg_dma_address(sg) + sg_offset;
1711                 u64 prev_addr = dma_addr;
1712                 unsigned int dma_len = sg_dma_len(sg) - sg_offset;
1713                 u64 end_dma_addr = dma_addr + dma_len;
1714                 u64 page_addr = dma_addr & page_mask;
1715
1716                 /*
1717                  * For the second and later elements, check whether either the
1718                  * end of element i-1 or the start of element i is not aligned
1719                  * on a page boundary.
1720                  */
1721                 if (i && (last_page_off != 0 || page_addr != dma_addr)) {
1722                         /* Stop mapping if there is a gap. */
1723                         if (last_end_dma_addr != dma_addr)
1724                                 break;
1725
1726                         /*
1727                          * Coalesce this element with the last. If it is small
1728                          * enough just update mr->length. Otherwise start
1729                          * mapping from the next page.
1730                          */
1731                         goto next_page;
1732                 }
1733
1734                 do {
1735                         ret = set_page(mr, page_addr);
1736                         if (unlikely(ret < 0)) {
1737                                 sg_offset = prev_addr - sg_dma_address(sg);
1738                                 mr->length += prev_addr - dma_addr;
1739                                 if (sg_offset_p)
1740                                         *sg_offset_p = sg_offset;
1741                                 return i || sg_offset ? i : ret;
1742                         }
1743                         prev_addr = page_addr;
1744 next_page:
1745                         page_addr += mr->page_size;
1746                 } while (page_addr < end_dma_addr);
1747
1748                 mr->length += dma_len;
1749                 last_end_dma_addr = end_dma_addr;
1750                 last_page_off = end_dma_addr & ~page_mask;
1751
1752                 sg_offset = 0;
1753         }
1754
1755         if (sg_offset_p)
1756                 *sg_offset_p = 0;
1757         return i;
1758 }
1759 EXPORT_SYMBOL(ib_sg_to_pages);
1760
1761 struct ib_drain_cqe {
1762         struct ib_cqe cqe;
1763         struct completion done;
1764 };
1765
1766 static void ib_drain_qp_done(struct ib_cq *cq, struct ib_wc *wc)
1767 {
1768         struct ib_drain_cqe *cqe = container_of(wc->wr_cqe, struct ib_drain_cqe,
1769                                                 cqe);
1770
1771         complete(&cqe->done);
1772 }
1773
1774 /*
1775  * Post a WR and block until its completion is reaped for the SQ.
1776  */
1777 static void __ib_drain_sq(struct ib_qp *qp)
1778 {
1779         struct ib_qp_attr attr = { .qp_state = IB_QPS_ERR };
1780         struct ib_drain_cqe sdrain;
1781         struct ib_send_wr swr = {}, *bad_swr;
1782         int ret;
1783
1784         if (qp->send_cq->poll_ctx == IB_POLL_DIRECT) {
1785                 WARN_ONCE(qp->send_cq->poll_ctx == IB_POLL_DIRECT,
1786                           "IB_POLL_DIRECT poll_ctx not supported for drain\n");
1787                 return;
1788         }
1789
1790         swr.wr_cqe = &sdrain.cqe;
1791         sdrain.cqe.done = ib_drain_qp_done;
1792         init_completion(&sdrain.done);
1793
1794         ret = ib_modify_qp(qp, &attr, IB_QP_STATE);
1795         if (ret) {
1796                 WARN_ONCE(ret, "failed to drain send queue: %d\n", ret);
1797                 return;
1798         }
1799
1800         ret = ib_post_send(qp, &swr, &bad_swr);
1801         if (ret) {
1802                 WARN_ONCE(ret, "failed to drain send queue: %d\n", ret);
1803                 return;
1804         }
1805
1806         wait_for_completion(&sdrain.done);
1807 }
1808
1809 /*
1810  * Post a WR and block until its completion is reaped for the RQ.
1811  */
1812 static void __ib_drain_rq(struct ib_qp *qp)
1813 {
1814         struct ib_qp_attr attr = { .qp_state = IB_QPS_ERR };
1815         struct ib_drain_cqe rdrain;
1816         struct ib_recv_wr rwr = {}, *bad_rwr;
1817         int ret;
1818
1819         if (qp->recv_cq->poll_ctx == IB_POLL_DIRECT) {
1820                 WARN_ONCE(qp->recv_cq->poll_ctx == IB_POLL_DIRECT,
1821                           "IB_POLL_DIRECT poll_ctx not supported for drain\n");
1822                 return;
1823         }
1824
1825         rwr.wr_cqe = &rdrain.cqe;
1826         rdrain.cqe.done = ib_drain_qp_done;
1827         init_completion(&rdrain.done);
1828
1829         ret = ib_modify_qp(qp, &attr, IB_QP_STATE);
1830         if (ret) {
1831                 WARN_ONCE(ret, "failed to drain recv queue: %d\n", ret);
1832                 return;
1833         }
1834
1835         ret = ib_post_recv(qp, &rwr, &bad_rwr);
1836         if (ret) {
1837                 WARN_ONCE(ret, "failed to drain recv queue: %d\n", ret);
1838                 return;
1839         }
1840
1841         wait_for_completion(&rdrain.done);
1842 }
1843
1844 /**
1845  * ib_drain_sq() - Block until all SQ CQEs have been consumed by the
1846  *                 application.
1847  * @qp:            queue pair to drain
1848  *
1849  * If the device has a provider-specific drain function, then
1850  * call that.  Otherwise call the generic drain function
1851  * __ib_drain_sq().
1852  *
1853  * The caller must:
1854  *
1855  * ensure there is room in the CQ and SQ for the drain work request and
1856  * completion.
1857  *
1858  * allocate the CQ using ib_alloc_cq() and the CQ poll context cannot be
1859  * IB_POLL_DIRECT.
1860  *
1861  * ensure that there are no other contexts that are posting WRs concurrently.
1862  * Otherwise the drain is not guaranteed.
1863  */
1864 void ib_drain_sq(struct ib_qp *qp)
1865 {
1866         if (qp->device->drain_sq)
1867                 qp->device->drain_sq(qp);
1868         else
1869                 __ib_drain_sq(qp);
1870 }
1871 EXPORT_SYMBOL(ib_drain_sq);
1872
1873 /**
1874  * ib_drain_rq() - Block until all RQ CQEs have been consumed by the
1875  *                 application.
1876  * @qp:            queue pair to drain
1877  *
1878  * If the device has a provider-specific drain function, then
1879  * call that.  Otherwise call the generic drain function
1880  * __ib_drain_rq().
1881  *
1882  * The caller must:
1883  *
1884  * ensure there is room in the CQ and RQ for the drain work request and
1885  * completion.
1886  *
1887  * allocate the CQ using ib_alloc_cq() and the CQ poll context cannot be
1888  * IB_POLL_DIRECT.
1889  *
1890  * ensure that there are no other contexts that are posting WRs concurrently.
1891  * Otherwise the drain is not guaranteed.
1892  */
1893 void ib_drain_rq(struct ib_qp *qp)
1894 {
1895         if (qp->device->drain_rq)
1896                 qp->device->drain_rq(qp);
1897         else
1898                 __ib_drain_rq(qp);
1899 }
1900 EXPORT_SYMBOL(ib_drain_rq);
1901
1902 /**
1903  * ib_drain_qp() - Block until all CQEs have been consumed by the
1904  *                 application on both the RQ and SQ.
1905  * @qp:            queue pair to drain
1906  *
1907  * The caller must:
1908  *
1909  * ensure there is room in the CQ(s), SQ, and RQ for drain work requests
1910  * and completions.
1911  *
1912  * allocate the CQs using ib_alloc_cq() and the CQ poll context cannot be
1913  * IB_POLL_DIRECT.
1914  *
1915  * ensure that there are no other contexts that are posting WRs concurrently.
1916  * Otherwise the drain is not guaranteed.
1917  */
1918 void ib_drain_qp(struct ib_qp *qp)
1919 {
1920         ib_drain_sq(qp);
1921         if (!qp->srq)
1922                 ib_drain_rq(qp);
1923 }
1924 EXPORT_SYMBOL(ib_drain_qp);