sys/ofed/drivers/infiniband/core/ib_verbs.c

   1 /*-
   2  * SPDX-License-Identifier: BSD-2-Clause OR GPL-2.0
   3  *
   4  * Copyright (c) 2004 Mellanox Technologies Ltd.  All rights reserved.
   5  * Copyright (c) 2004 Infinicon Corporation.  All rights reserved.
   6  * Copyright (c) 2004 Intel Corporation.  All rights reserved.
   7  * Copyright (c) 2004 Topspin Corporation.  All rights reserved.
   8  * Copyright (c) 2004 Voltaire Corporation.  All rights reserved.
   9  * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved.
  10  * Copyright (c) 2005, 2006 Cisco Systems.  All rights reserved.
  11  *
  12  * This software is available to you under a choice of one of two
  13  * licenses.  You may choose to be licensed under the terms of the GNU
  14  * General Public License (GPL) Version 2, available from the file
  15  * COPYING in the main directory of this source tree, or the
  16  * OpenIB.org BSD license below:
  17  *
  18  *     Redistribution and use in source and binary forms, with or
  19  *     without modification, are permitted provided that the following
  20  *     conditions are met:
  21  *
  22  *      - Redistributions of source code must retain the above
  23  *        copyright notice, this list of conditions and the following
  24  *        disclaimer.
  25  *
  26  *      - Redistributions in binary form must reproduce the above
  27  *        copyright notice, this list of conditions and the following
  28  *        disclaimer in the documentation and/or other materials
  29  *        provided with the distribution.
  30  *
  31  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
  32  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  33  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
  34  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
  35  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
  36  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
  37  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  38  * SOFTWARE.
  39  */
  40
  41 #include <sys/cdefs.h>
  42 __FBSDID("$FreeBSD$");
  43
  44 #include <linux/errno.h>
  45 #include <linux/err.h>
  46 #include <linux/string.h>
  47 #include <linux/slab.h>
  48 #include <linux/in.h>
  49 #include <linux/in6.h>
  50
  51 #include <rdma/ib_verbs.h>
  52 #include <rdma/ib_cache.h>
  53 #include <rdma/ib_addr.h>
  54
  55 #include <netinet/ip.h>
  56 #include <netinet/ip6.h>
  57
  58 #include <machine/in_cksum.h>
  59
  60 #include "core_priv.h"
  61
  62 static const char * const ib_events[] = {
  63         [IB_EVENT_CQ_ERR]               = "CQ error",
  64         [IB_EVENT_QP_FATAL]             = "QP fatal error",
  65         [IB_EVENT_QP_REQ_ERR]           = "QP request error",
  66         [IB_EVENT_QP_ACCESS_ERR]        = "QP access error",
  67         [IB_EVENT_COMM_EST]             = "communication established",
  68         [IB_EVENT_SQ_DRAINED]           = "send queue drained",
  69         [IB_EVENT_PATH_MIG]             = "path migration successful",
  70         [IB_EVENT_PATH_MIG_ERR]         = "path migration error",
  71         [IB_EVENT_DEVICE_FATAL]         = "device fatal error",
  72         [IB_EVENT_PORT_ACTIVE]          = "port active",
  73         [IB_EVENT_PORT_ERR]             = "port error",
  74         [IB_EVENT_LID_CHANGE]           = "LID change",
  75         [IB_EVENT_PKEY_CHANGE]          = "P_key change",
  76         [IB_EVENT_SM_CHANGE]            = "SM change",
  77         [IB_EVENT_SRQ_ERR]              = "SRQ error",
  78         [IB_EVENT_SRQ_LIMIT_REACHED]    = "SRQ limit reached",
  79         [IB_EVENT_QP_LAST_WQE_REACHED]  = "last WQE reached",
  80         [IB_EVENT_CLIENT_REREGISTER]    = "client reregister",
  81         [IB_EVENT_GID_CHANGE]           = "GID changed",
  82 };
  83
  84 const char *__attribute_const__ ib_event_msg(enum ib_event_type event)
  85 {
  86         size_t index = event;
  87
  88         return (index < ARRAY_SIZE(ib_events) && ib_events[index]) ?
  89                         ib_events[index] : "unrecognized event";
  90 }
  91 EXPORT_SYMBOL(ib_event_msg);
  92
  93 static const char * const wc_statuses[] = {
  94         [IB_WC_SUCCESS]                 = "success",
  95         [IB_WC_LOC_LEN_ERR]             = "local length error",
  96         [IB_WC_LOC_QP_OP_ERR]           = "local QP operation error",
  97         [IB_WC_LOC_EEC_OP_ERR]          = "local EE context operation error",
  98         [IB_WC_LOC_PROT_ERR]            = "local protection error",
  99         [IB_WC_WR_FLUSH_ERR]            = "WR flushed",
 100         [IB_WC_MW_BIND_ERR]             = "memory management operation error",
 101         [IB_WC_BAD_RESP_ERR]            = "bad response error",
 102         [IB_WC_LOC_ACCESS_ERR]          = "local access error",
 103         [IB_WC_REM_INV_REQ_ERR]         = "invalid request error",
 104         [IB_WC_REM_ACCESS_ERR]          = "remote access error",
 105         [IB_WC_REM_OP_ERR]              = "remote operation error",
 106         [IB_WC_RETRY_EXC_ERR]           = "transport retry counter exceeded",
 107         [IB_WC_RNR_RETRY_EXC_ERR]       = "RNR retry counter exceeded",
 108         [IB_WC_LOC_RDD_VIOL_ERR]        = "local RDD violation error",
 109         [IB_WC_REM_INV_RD_REQ_ERR]      = "remote invalid RD request",
 110         [IB_WC_REM_ABORT_ERR]           = "operation aborted",
 111         [IB_WC_INV_EECN_ERR]            = "invalid EE context number",
 112         [IB_WC_INV_EEC_STATE_ERR]       = "invalid EE context state",
 113         [IB_WC_FATAL_ERR]               = "fatal error",
 114         [IB_WC_RESP_TIMEOUT_ERR]        = "response timeout error",
 115         [IB_WC_GENERAL_ERR]             = "general error",
 116 };
 117
 118 const char *__attribute_const__ ib_wc_status_msg(enum ib_wc_status status)
 119 {
 120         size_t index = status;
 121
 122         return (index < ARRAY_SIZE(wc_statuses) && wc_statuses[index]) ?
 123                         wc_statuses[index] : "unrecognized status";
 124 }
 125 EXPORT_SYMBOL(ib_wc_status_msg);
 126
 127 __attribute_const__ int ib_rate_to_mult(enum ib_rate rate)
 128 {
 129         switch (rate) {
 130         case IB_RATE_2_5_GBPS: return  1;
 131         case IB_RATE_5_GBPS:   return  2;
 132         case IB_RATE_10_GBPS:  return  4;
 133         case IB_RATE_20_GBPS:  return  8;
 134         case IB_RATE_30_GBPS:  return 12;
 135         case IB_RATE_40_GBPS:  return 16;
 136         case IB_RATE_60_GBPS:  return 24;
 137         case IB_RATE_80_GBPS:  return 32;
 138         case IB_RATE_120_GBPS: return 48;
 139         case IB_RATE_28_GBPS:  return  11;
 140         case IB_RATE_50_GBPS:  return  20;
 141         case IB_RATE_400_GBPS: return 160;
 142         case IB_RATE_600_GBPS: return 240;
 143         default:               return -1;
 144         }
 145 }
 146 EXPORT_SYMBOL(ib_rate_to_mult);
 147
 148 __attribute_const__ enum ib_rate mult_to_ib_rate(int mult)
 149 {
 150         switch (mult) {
 151         case 1:  return IB_RATE_2_5_GBPS;
 152         case 2:  return IB_RATE_5_GBPS;
 153         case 4:  return IB_RATE_10_GBPS;
 154         case 8:  return IB_RATE_20_GBPS;
 155         case 12: return IB_RATE_30_GBPS;
 156         case 16: return IB_RATE_40_GBPS;
 157         case 24: return IB_RATE_60_GBPS;
 158         case 32: return IB_RATE_80_GBPS;
 159         case 48: return IB_RATE_120_GBPS;
 160         case 6:   return IB_RATE_14_GBPS;
 161         case 22:  return IB_RATE_56_GBPS;
 162         case 45:  return IB_RATE_112_GBPS;
 163         case 67:  return IB_RATE_168_GBPS;
 164         case 10:  return IB_RATE_25_GBPS;
 165         case 40:  return IB_RATE_100_GBPS;
 166         case 80:  return IB_RATE_200_GBPS;
 167         case 120: return IB_RATE_300_GBPS;
 168         case 11:  return IB_RATE_28_GBPS;
 169         case 20:  return IB_RATE_50_GBPS;
 170         case 160: return IB_RATE_400_GBPS;
 171         case 240: return IB_RATE_600_GBPS;
 172         default: return IB_RATE_PORT_CURRENT;
 173         }
 174 }
 175 EXPORT_SYMBOL(mult_to_ib_rate);
 176
 177 __attribute_const__ int ib_rate_to_mbps(enum ib_rate rate)
 178 {
 179         switch (rate) {
 180         case IB_RATE_2_5_GBPS: return 2500;
 181         case IB_RATE_5_GBPS:   return 5000;
 182         case IB_RATE_10_GBPS:  return 10000;
 183         case IB_RATE_20_GBPS:  return 20000;
 184         case IB_RATE_30_GBPS:  return 30000;
 185         case IB_RATE_40_GBPS:  return 40000;
 186         case IB_RATE_60_GBPS:  return 60000;
 187         case IB_RATE_80_GBPS:  return 80000;
 188         case IB_RATE_120_GBPS: return 120000;
 189         case IB_RATE_14_GBPS:  return 14062;
 190         case IB_RATE_56_GBPS:  return 56250;
 191         case IB_RATE_112_GBPS: return 112500;
 192         case IB_RATE_168_GBPS: return 168750;
 193         case IB_RATE_25_GBPS:  return 25781;
 194         case IB_RATE_100_GBPS: return 103125;
 195         case IB_RATE_200_GBPS: return 206250;
 196         case IB_RATE_300_GBPS: return 309375;
 197         case IB_RATE_28_GBPS:  return 28125;
 198         case IB_RATE_50_GBPS:  return 53125;
 199         case IB_RATE_400_GBPS: return 425000;
 200         case IB_RATE_600_GBPS: return 637500;
 201         default:               return -1;
 202         }
 203 }
 204 EXPORT_SYMBOL(ib_rate_to_mbps);
 205
 206 __attribute_const__ enum rdma_transport_type
 207 rdma_node_get_transport(enum rdma_node_type node_type)
 208 {
 209         switch (node_type) {
 210         case RDMA_NODE_IB_CA:
 211         case RDMA_NODE_IB_SWITCH:
 212         case RDMA_NODE_IB_ROUTER:
 213                 return RDMA_TRANSPORT_IB;
 214         case RDMA_NODE_RNIC:
 215                 return RDMA_TRANSPORT_IWARP;
 216         case RDMA_NODE_USNIC:
 217                 return RDMA_TRANSPORT_USNIC;
 218         case RDMA_NODE_USNIC_UDP:
 219                 return RDMA_TRANSPORT_USNIC_UDP;
 220         default:
 221                 BUG();
 222                 return 0;
 223         }
 224 }
 225 EXPORT_SYMBOL(rdma_node_get_transport);
 226
 227 enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device, u8 port_num)
 228 {
 229         if (device->get_link_layer)
 230                 return device->get_link_layer(device, port_num);
 231
 232         switch (rdma_node_get_transport(device->node_type)) {
 233         case RDMA_TRANSPORT_IB:
 234                 return IB_LINK_LAYER_INFINIBAND;
 235         case RDMA_TRANSPORT_IWARP:
 236         case RDMA_TRANSPORT_USNIC:
 237         case RDMA_TRANSPORT_USNIC_UDP:
 238                 return IB_LINK_LAYER_ETHERNET;
 239         default:
 240                 return IB_LINK_LAYER_UNSPECIFIED;
 241         }
 242 }
 243 EXPORT_SYMBOL(rdma_port_get_link_layer);
 244
 245 /* Protection domains */
 246
 247 /**
 248  * ib_alloc_pd - Allocates an unused protection domain.
 249  * @device: The device on which to allocate the protection domain.
 250  *
 251  * A protection domain object provides an association between QPs, shared
 252  * receive queues, address handles, memory regions, and memory windows.
 253  *
 254  * Every PD has a local_dma_lkey which can be used as the lkey value for local
 255  * memory operations.
 256  */
 257 struct ib_pd *__ib_alloc_pd(struct ib_device *device, unsigned int flags,
 258                 const char *caller)
 259 {
 260         struct ib_pd *pd;
 261         int mr_access_flags = 0;
 262
 263         pd = device->alloc_pd(device, NULL, NULL);
 264         if (IS_ERR(pd))
 265                 return pd;
 266
 267         pd->device = device;
 268         pd->uobject = NULL;
 269         pd->__internal_mr = NULL;
 270         atomic_set(&pd->usecnt, 0);
 271         pd->flags = flags;
 272
 273         if (device->attrs.device_cap_flags & IB_DEVICE_LOCAL_DMA_LKEY)
 274                 pd->local_dma_lkey = device->local_dma_lkey;
 275         else
 276                 mr_access_flags |= IB_ACCESS_LOCAL_WRITE;
 277
 278         if (flags & IB_PD_UNSAFE_GLOBAL_RKEY) {
 279                 pr_warn("%s: enabling unsafe global rkey\n", caller);
 280                 mr_access_flags |= IB_ACCESS_REMOTE_READ | IB_ACCESS_REMOTE_WRITE;
 281         }
 282
 283         if (mr_access_flags) {
 284                 struct ib_mr *mr;
 285
 286                 mr = pd->device->get_dma_mr(pd, mr_access_flags);
 287                 if (IS_ERR(mr)) {
 288                         ib_dealloc_pd(pd);
 289                         return ERR_CAST(mr);
 290                 }
 291
 292                 mr->device      = pd->device;
 293                 mr->pd          = pd;
 294                 mr->uobject     = NULL;
 295                 mr->need_inval  = false;
 296
 297                 pd->__internal_mr = mr;
 298
 299                 if (!(device->attrs.device_cap_flags & IB_DEVICE_LOCAL_DMA_LKEY))
 300                         pd->local_dma_lkey = pd->__internal_mr->lkey;
 301
 302                 if (flags & IB_PD_UNSAFE_GLOBAL_RKEY)
 303                         pd->unsafe_global_rkey = pd->__internal_mr->rkey;
 304         }
 305
 306         return pd;
 307 }
 308 EXPORT_SYMBOL(__ib_alloc_pd);
 309
 310 /**
 311  * ib_dealloc_pd - Deallocates a protection domain.
 312  * @pd: The protection domain to deallocate.
 313  *
 314  * It is an error to call this function while any resources in the pd still
 315  * exist.  The caller is responsible to synchronously destroy them and
 316  * guarantee no new allocations will happen.
 317  */
 318 void ib_dealloc_pd(struct ib_pd *pd)
 319 {
 320         int ret;
 321
 322         if (pd->__internal_mr) {
 323                 ret = pd->device->dereg_mr(pd->__internal_mr);
 324                 WARN_ON(ret);
 325                 pd->__internal_mr = NULL;
 326         }
 327
 328         /* uverbs manipulates usecnt with proper locking, while the kabi
 329            requires the caller to guarantee we can't race here. */
 330         WARN_ON(atomic_read(&pd->usecnt));
 331
 332         /* Making delalloc_pd a void return is a WIP, no driver should return
 333            an error here. */
 334         ret = pd->device->dealloc_pd(pd);
 335         WARN_ONCE(ret, "Infiniband HW driver failed dealloc_pd");
 336 }
 337 EXPORT_SYMBOL(ib_dealloc_pd);
 338
 339 /* Address handles */
 340
 341 struct ib_ah *ib_create_ah(struct ib_pd *pd, struct ib_ah_attr *ah_attr)
 342 {
 343         struct ib_ah *ah;
 344
 345         ah = pd->device->create_ah(pd, ah_attr, NULL);
 346
 347         if (!IS_ERR(ah)) {
 348                 ah->device  = pd->device;
 349                 ah->pd      = pd;
 350                 ah->uobject = NULL;
 351                 atomic_inc(&pd->usecnt);
 352         }
 353
 354         return ah;
 355 }
 356 EXPORT_SYMBOL(ib_create_ah);
 357
 358 static int ib_get_header_version(const union rdma_network_hdr *hdr)
 359 {
 360         const struct ip *ip4h = (const struct ip *)&hdr->roce4grh;
 361         struct ip ip4h_checked;
 362         const struct ip6_hdr *ip6h = (const struct ip6_hdr *)&hdr->ibgrh;
 363
 364         /* If it's IPv6, the version must be 6, otherwise, the first
 365          * 20 bytes (before the IPv4 header) are garbled.
 366          */
 367         if ((ip6h->ip6_vfc & IPV6_VERSION_MASK) != IPV6_VERSION)
 368                 return (ip4h->ip_v == 4) ? 4 : 0;
 369         /* version may be 6 or 4 because the first 20 bytes could be garbled */
 370
 371         /* RoCE v2 requires no options, thus header length
 372          * must be 5 words
 373          */
 374         if (ip4h->ip_hl != 5)
 375                 return 6;
 376
 377         /* Verify checksum.
 378          * We can't write on scattered buffers so we need to copy to
 379          * temp buffer.
 380          */
 381         memcpy(&ip4h_checked, ip4h, sizeof(ip4h_checked));
 382         ip4h_checked.ip_sum = 0;
 383 #if defined(INET) || defined(INET6)
 384         ip4h_checked.ip_sum = in_cksum_hdr(&ip4h_checked);
 385 #endif
 386         /* if IPv4 header checksum is OK, believe it */
 387         if (ip4h->ip_sum == ip4h_checked.ip_sum)
 388                 return 4;
 389         return 6;
 390 }
 391
 392 static enum rdma_network_type ib_get_net_type_by_grh(struct ib_device *device,
 393                                                      u8 port_num,
 394                                                      const struct ib_grh *grh)
 395 {
 396         int grh_version;
 397
 398         if (rdma_protocol_ib(device, port_num))
 399                 return RDMA_NETWORK_IB;
 400
 401         grh_version = ib_get_header_version((const union rdma_network_hdr *)grh);
 402
 403         if (grh_version == 4)
 404                 return RDMA_NETWORK_IPV4;
 405
 406         if (grh->next_hdr == IPPROTO_UDP)
 407                 return RDMA_NETWORK_IPV6;
 408
 409         return RDMA_NETWORK_ROCE_V1;
 410 }
 411
 412 struct find_gid_index_context {
 413         u16 vlan_id;
 414         enum ib_gid_type gid_type;
 415 };
 416
 417
 418 /*
 419  * This function will return true only if a inspected GID index
 420  * matches the request based on the GID type and VLAN configuration
 421  */
 422 static bool find_gid_index(const union ib_gid *gid,
 423                            const struct ib_gid_attr *gid_attr,
 424                            void *context)
 425 {
 426         u16 vlan_diff;
 427         struct find_gid_index_context *ctx =
 428                 (struct find_gid_index_context *)context;
 429
 430         if (ctx->gid_type != gid_attr->gid_type)
 431                 return false;
 432
 433         /*
 434          * The following will verify:
 435          * 1. VLAN ID matching for VLAN tagged requests.
 436          * 2. prio-tagged/untagged to prio-tagged/untagged matching.
 437          *
 438          * This XOR is valid, since 0x0 < vlan_id < 0x0FFF.
 439          */
 440         vlan_diff = rdma_vlan_dev_vlan_id(gid_attr->ndev) ^ ctx->vlan_id;
 441
 442         return (vlan_diff == 0x0000 || vlan_diff == 0xFFFF);
 443 }
 444
 445 static int get_sgid_index_from_eth(struct ib_device *device, u8 port_num,
 446                                    u16 vlan_id, const union ib_gid *sgid,
 447                                    enum ib_gid_type gid_type,
 448                                    u16 *gid_index)
 449 {
 450         struct find_gid_index_context context = {.vlan_id = vlan_id,
 451                                                  .gid_type = gid_type};
 452
 453         return ib_find_gid_by_filter(device, sgid, port_num, find_gid_index,
 454                                      &context, gid_index);
 455 }
 456
 457 static int get_gids_from_rdma_hdr(const union rdma_network_hdr *hdr,
 458                                   enum rdma_network_type net_type,
 459                                   union ib_gid *sgid, union ib_gid *dgid)
 460 {
 461         struct sockaddr_in  src_in;
 462         struct sockaddr_in  dst_in;
 463         __be32 src_saddr, dst_saddr;
 464
 465         if (!sgid || !dgid)
 466                 return -EINVAL;
 467
 468         if (net_type == RDMA_NETWORK_IPV4) {
 469                 memcpy(&src_in.sin_addr.s_addr,
 470                        &hdr->roce4grh.ip_src, 4);
 471                 memcpy(&dst_in.sin_addr.s_addr,
 472                        &hdr->roce4grh.ip_dst, 4);
 473                 src_saddr = src_in.sin_addr.s_addr;
 474                 dst_saddr = dst_in.sin_addr.s_addr;
 475                 ipv6_addr_set_v4mapped(src_saddr,
 476                                        (struct in6_addr *)sgid);
 477                 ipv6_addr_set_v4mapped(dst_saddr,
 478                                        (struct in6_addr *)dgid);
 479                 return 0;
 480         } else if (net_type == RDMA_NETWORK_IPV6 ||
 481                    net_type == RDMA_NETWORK_IB) {
 482                 *dgid = hdr->ibgrh.dgid;
 483                 *sgid = hdr->ibgrh.sgid;
 484                 return 0;
 485         } else {
 486                 return -EINVAL;
 487         }
 488 }
 489
 490 int ib_init_ah_from_wc(struct ib_device *device, u8 port_num,
 491                        const struct ib_wc *wc, const struct ib_grh *grh,
 492                        struct ib_ah_attr *ah_attr)
 493 {
 494         u32 flow_class;
 495         u16 gid_index = 0;
 496         int ret;
 497         enum rdma_network_type net_type = RDMA_NETWORK_IB;
 498         enum ib_gid_type gid_type = IB_GID_TYPE_IB;
 499         int hoplimit = 0xff;
 500         union ib_gid dgid;
 501         union ib_gid sgid;
 502
 503         memset(ah_attr, 0, sizeof *ah_attr);
 504         if (rdma_cap_eth_ah(device, port_num)) {
 505                 if (wc->wc_flags & IB_WC_WITH_NETWORK_HDR_TYPE)
 506                         net_type = wc->network_hdr_type;
 507                 else
 508                         net_type = ib_get_net_type_by_grh(device, port_num, grh);
 509                 gid_type = ib_network_to_gid_type(net_type);
 510         }
 511         ret = get_gids_from_rdma_hdr((const union rdma_network_hdr *)grh, net_type,
 512                                      &sgid, &dgid);
 513         if (ret)
 514                 return ret;
 515
 516         if (rdma_protocol_roce(device, port_num)) {
 517                 struct ib_gid_attr dgid_attr;
 518                 const u16 vlan_id = (wc->wc_flags & IB_WC_WITH_VLAN) ?
 519                                 wc->vlan_id : 0xffff;
 520
 521                 if (!(wc->wc_flags & IB_WC_GRH))
 522                         return -EPROTOTYPE;
 523
 524                 ret = get_sgid_index_from_eth(device, port_num, vlan_id,
 525                                               &dgid, gid_type, &gid_index);
 526                 if (ret)
 527                         return ret;
 528
 529                 ret = ib_get_cached_gid(device, port_num, gid_index, &dgid, &dgid_attr);
 530                 if (ret)
 531                         return ret;
 532
 533                 if (dgid_attr.ndev == NULL)
 534                         return -ENODEV;
 535
 536                 ret = rdma_addr_find_l2_eth_by_grh(&dgid, &sgid, ah_attr->dmac,
 537                     dgid_attr.ndev, &hoplimit);
 538
 539                 dev_put(dgid_attr.ndev);
 540                 if (ret)
 541                         return ret;
 542         }
 543
 544         ah_attr->dlid = wc->slid;
 545         ah_attr->sl = wc->sl;
 546         ah_attr->src_path_bits = wc->dlid_path_bits;
 547         ah_attr->port_num = port_num;
 548
 549         if (wc->wc_flags & IB_WC_GRH) {
 550                 ah_attr->ah_flags = IB_AH_GRH;
 551                 ah_attr->grh.dgid = sgid;
 552
 553                 if (!rdma_cap_eth_ah(device, port_num)) {
 554                         if (dgid.global.interface_id != cpu_to_be64(IB_SA_WELL_KNOWN_GUID)) {
 555                                 ret = ib_find_cached_gid_by_port(device, &dgid,
 556                                                                  IB_GID_TYPE_IB,
 557                                                                  port_num, NULL,
 558                                                                  &gid_index);
 559                                 if (ret)
 560                                         return ret;
 561                         }
 562                 }
 563
 564                 ah_attr->grh.sgid_index = (u8) gid_index;
 565                 flow_class = be32_to_cpu(grh->version_tclass_flow);
 566                 ah_attr->grh.flow_label = flow_class & 0xFFFFF;
 567                 ah_attr->grh.hop_limit = hoplimit;
 568                 ah_attr->grh.traffic_class = (flow_class >> 20) & 0xFF;
 569         }
 570         return 0;
 571 }
 572 EXPORT_SYMBOL(ib_init_ah_from_wc);
 573
 574 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, const struct ib_wc *wc,
 575                                    const struct ib_grh *grh, u8 port_num)
 576 {
 577         struct ib_ah_attr ah_attr;
 578         int ret;
 579
 580         ret = ib_init_ah_from_wc(pd->device, port_num, wc, grh, &ah_attr);
 581         if (ret)
 582                 return ERR_PTR(ret);
 583
 584         return ib_create_ah(pd, &ah_attr);
 585 }
 586 EXPORT_SYMBOL(ib_create_ah_from_wc);
 587
 588 int ib_modify_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr)
 589 {
 590         return ah->device->modify_ah ?
 591                 ah->device->modify_ah(ah, ah_attr) :
 592                 -ENOSYS;
 593 }
 594 EXPORT_SYMBOL(ib_modify_ah);
 595
 596 int ib_query_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr)
 597 {
 598         return ah->device->query_ah ?
 599                 ah->device->query_ah(ah, ah_attr) :
 600                 -ENOSYS;
 601 }
 602 EXPORT_SYMBOL(ib_query_ah);
 603
 604 int ib_destroy_ah(struct ib_ah *ah)
 605 {
 606         struct ib_pd *pd;
 607         int ret;
 608
 609         pd = ah->pd;
 610         ret = ah->device->destroy_ah(ah);
 611         if (!ret)
 612                 atomic_dec(&pd->usecnt);
 613
 614         return ret;
 615 }
 616 EXPORT_SYMBOL(ib_destroy_ah);
 617
 618 /* Shared receive queues */
 619
 620 struct ib_srq *ib_create_srq(struct ib_pd *pd,
 621                              struct ib_srq_init_attr *srq_init_attr)
 622 {
 623         struct ib_srq *srq;
 624
 625         if (!pd->device->create_srq)
 626                 return ERR_PTR(-ENOSYS);
 627
 628         srq = pd->device->create_srq(pd, srq_init_attr, NULL);
 629
 630         if (!IS_ERR(srq)) {
 631                 srq->device        = pd->device;
 632                 srq->pd            = pd;
 633                 srq->uobject       = NULL;
 634                 srq->event_handler = srq_init_attr->event_handler;
 635                 srq->srq_context   = srq_init_attr->srq_context;
 636                 srq->srq_type      = srq_init_attr->srq_type;
 637                 if (srq->srq_type == IB_SRQT_XRC) {
 638                         srq->ext.xrc.xrcd = srq_init_attr->ext.xrc.xrcd;
 639                         srq->ext.xrc.cq   = srq_init_attr->ext.xrc.cq;
 640                         atomic_inc(&srq->ext.xrc.xrcd->usecnt);
 641                         atomic_inc(&srq->ext.xrc.cq->usecnt);
 642                 }
 643                 atomic_inc(&pd->usecnt);
 644                 atomic_set(&srq->usecnt, 0);
 645         }
 646
 647         return srq;
 648 }
 649 EXPORT_SYMBOL(ib_create_srq);
 650
 651 int ib_modify_srq(struct ib_srq *srq,
 652                   struct ib_srq_attr *srq_attr,
 653                   enum ib_srq_attr_mask srq_attr_mask)
 654 {
 655         return srq->device->modify_srq ?
 656                 srq->device->modify_srq(srq, srq_attr, srq_attr_mask, NULL) :
 657                 -ENOSYS;
 658 }
 659 EXPORT_SYMBOL(ib_modify_srq);
 660
 661 int ib_query_srq(struct ib_srq *srq,
 662                  struct ib_srq_attr *srq_attr)
 663 {
 664         return srq->device->query_srq ?
 665                 srq->device->query_srq(srq, srq_attr) : -ENOSYS;
 666 }
 667 EXPORT_SYMBOL(ib_query_srq);
 668
 669 int ib_destroy_srq(struct ib_srq *srq)
 670 {
 671         struct ib_pd *pd;
 672         enum ib_srq_type srq_type;
 673         struct ib_xrcd *uninitialized_var(xrcd);
 674         struct ib_cq *uninitialized_var(cq);
 675         int ret;
 676
 677         if (atomic_read(&srq->usecnt))
 678                 return -EBUSY;
 679
 680         pd = srq->pd;
 681         srq_type = srq->srq_type;
 682         if (srq_type == IB_SRQT_XRC) {
 683                 xrcd = srq->ext.xrc.xrcd;
 684                 cq = srq->ext.xrc.cq;
 685         }
 686
 687         ret = srq->device->destroy_srq(srq);
 688         if (!ret) {
 689                 atomic_dec(&pd->usecnt);
 690                 if (srq_type == IB_SRQT_XRC) {
 691                         atomic_dec(&xrcd->usecnt);
 692                         atomic_dec(&cq->usecnt);
 693                 }
 694         }
 695
 696         return ret;
 697 }
 698 EXPORT_SYMBOL(ib_destroy_srq);
 699
 700 /* Queue pairs */
 701
 702 static void __ib_shared_qp_event_handler(struct ib_event *event, void *context)
 703 {
 704         struct ib_qp *qp = context;
 705         unsigned long flags;
 706
 707         spin_lock_irqsave(&qp->device->event_handler_lock, flags);
 708         list_for_each_entry(event->element.qp, &qp->open_list, open_list)
 709                 if (event->element.qp->event_handler)
 710                         event->element.qp->event_handler(event, event->element.qp->qp_context);
 711         spin_unlock_irqrestore(&qp->device->event_handler_lock, flags);
 712 }
 713
 714 static void __ib_insert_xrcd_qp(struct ib_xrcd *xrcd, struct ib_qp *qp)
 715 {
 716         mutex_lock(&xrcd->tgt_qp_mutex);
 717         list_add(&qp->xrcd_list, &xrcd->tgt_qp_list);
 718         mutex_unlock(&xrcd->tgt_qp_mutex);
 719 }
 720
 721 static struct ib_qp *__ib_open_qp(struct ib_qp *real_qp,
 722                                   void (*event_handler)(struct ib_event *, void *),
 723                                   void *qp_context)
 724 {
 725         struct ib_qp *qp;
 726         unsigned long flags;
 727
 728         qp = kzalloc(sizeof *qp, GFP_KERNEL);
 729         if (!qp)
 730                 return ERR_PTR(-ENOMEM);
 731
 732         qp->real_qp = real_qp;
 733         atomic_inc(&real_qp->usecnt);
 734         qp->device = real_qp->device;
 735         qp->event_handler = event_handler;
 736         qp->qp_context = qp_context;
 737         qp->qp_num = real_qp->qp_num;
 738         qp->qp_type = real_qp->qp_type;
 739
 740         spin_lock_irqsave(&real_qp->device->event_handler_lock, flags);
 741         list_add(&qp->open_list, &real_qp->open_list);
 742         spin_unlock_irqrestore(&real_qp->device->event_handler_lock, flags);
 743
 744         return qp;
 745 }
 746
 747 struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
 748                          struct ib_qp_open_attr *qp_open_attr)
 749 {
 750         struct ib_qp *qp, *real_qp;
 751
 752         if (qp_open_attr->qp_type != IB_QPT_XRC_TGT)
 753                 return ERR_PTR(-EINVAL);
 754
 755         qp = ERR_PTR(-EINVAL);
 756         mutex_lock(&xrcd->tgt_qp_mutex);
 757         list_for_each_entry(real_qp, &xrcd->tgt_qp_list, xrcd_list) {
 758                 if (real_qp->qp_num == qp_open_attr->qp_num) {
 759                         qp = __ib_open_qp(real_qp, qp_open_attr->event_handler,
 760                                           qp_open_attr->qp_context);
 761                         break;
 762                 }
 763         }
 764         mutex_unlock(&xrcd->tgt_qp_mutex);
 765         return qp;
 766 }
 767 EXPORT_SYMBOL(ib_open_qp);
 768
 769 static struct ib_qp *ib_create_xrc_qp(struct ib_qp *qp,
 770                 struct ib_qp_init_attr *qp_init_attr)
 771 {
 772         struct ib_qp *real_qp = qp;
 773
 774         qp->event_handler = __ib_shared_qp_event_handler;
 775         qp->qp_context = qp;
 776         qp->pd = NULL;
 777         qp->send_cq = qp->recv_cq = NULL;
 778         qp->srq = NULL;
 779         qp->xrcd = qp_init_attr->xrcd;
 780         atomic_inc(&qp_init_attr->xrcd->usecnt);
 781         INIT_LIST_HEAD(&qp->open_list);
 782
 783         qp = __ib_open_qp(real_qp, qp_init_attr->event_handler,
 784                           qp_init_attr->qp_context);
 785         if (!IS_ERR(qp))
 786                 __ib_insert_xrcd_qp(qp_init_attr->xrcd, real_qp);
 787         else
 788                 real_qp->device->destroy_qp(real_qp);
 789         return qp;
 790 }
 791
 792 struct ib_qp *ib_create_qp(struct ib_pd *pd,
 793                            struct ib_qp_init_attr *qp_init_attr)
 794 {
 795         struct ib_device *device = pd ? pd->device : qp_init_attr->xrcd->device;
 796         struct ib_qp *qp;
 797
 798         if (qp_init_attr->rwq_ind_tbl &&
 799             (qp_init_attr->recv_cq ||
 800             qp_init_attr->srq || qp_init_attr->cap.max_recv_wr ||
 801             qp_init_attr->cap.max_recv_sge))
 802                 return ERR_PTR(-EINVAL);
 803
 804         qp = device->create_qp(pd, qp_init_attr, NULL);
 805         if (IS_ERR(qp))
 806                 return qp;
 807
 808         qp->device     = device;
 809         qp->real_qp    = qp;
 810         qp->uobject    = NULL;
 811         qp->qp_type    = qp_init_attr->qp_type;
 812         qp->rwq_ind_tbl = qp_init_attr->rwq_ind_tbl;
 813
 814         atomic_set(&qp->usecnt, 0);
 815         spin_lock_init(&qp->mr_lock);
 816
 817         if (qp_init_attr->qp_type == IB_QPT_XRC_TGT)
 818                 return ib_create_xrc_qp(qp, qp_init_attr);
 819
 820         qp->event_handler = qp_init_attr->event_handler;
 821         qp->qp_context = qp_init_attr->qp_context;
 822         if (qp_init_attr->qp_type == IB_QPT_XRC_INI) {
 823                 qp->recv_cq = NULL;
 824                 qp->srq = NULL;
 825         } else {
 826                 qp->recv_cq = qp_init_attr->recv_cq;
 827                 if (qp_init_attr->recv_cq)
 828                         atomic_inc(&qp_init_attr->recv_cq->usecnt);
 829                 qp->srq = qp_init_attr->srq;
 830                 if (qp->srq)
 831                         atomic_inc(&qp_init_attr->srq->usecnt);
 832         }
 833
 834         qp->pd      = pd;
 835         qp->send_cq = qp_init_attr->send_cq;
 836         qp->xrcd    = NULL;
 837
 838         atomic_inc(&pd->usecnt);
 839         if (qp_init_attr->send_cq)
 840                 atomic_inc(&qp_init_attr->send_cq->usecnt);
 841         if (qp_init_attr->rwq_ind_tbl)
 842                 atomic_inc(&qp->rwq_ind_tbl->usecnt);
 843
 844         /*
 845          * Note: all hw drivers guarantee that max_send_sge is lower than
 846          * the device RDMA WRITE SGE limit but not all hw drivers ensure that
 847          * max_send_sge <= max_sge_rd.
 848          */
 849         qp->max_write_sge = qp_init_attr->cap.max_send_sge;
 850         qp->max_read_sge = min_t(u32, qp_init_attr->cap.max_send_sge,
 851                                  device->attrs.max_sge_rd);
 852
 853         return qp;
 854 }
 855 EXPORT_SYMBOL(ib_create_qp);
 856
 857 static const struct {
 858         int                     valid;
 859         enum ib_qp_attr_mask    req_param[IB_QPT_MAX];
 860         enum ib_qp_attr_mask    opt_param[IB_QPT_MAX];
 861 } qp_state_table[IB_QPS_ERR + 1][IB_QPS_ERR + 1] = {
 862         [IB_QPS_RESET] = {
 863                 [IB_QPS_RESET] = { .valid = 1 },
 864                 [IB_QPS_INIT]  = {
 865                         .valid = 1,
 866                         .req_param = {
 867                                 [IB_QPT_UD]  = (IB_QP_PKEY_INDEX                |
 868                                                 IB_QP_PORT                      |
 869                                                 IB_QP_QKEY),
 870                                 [IB_QPT_RAW_PACKET] = IB_QP_PORT,
 871                                 [IB_QPT_UC]  = (IB_QP_PKEY_INDEX                |
 872                                                 IB_QP_PORT                      |
 873                                                 IB_QP_ACCESS_FLAGS),
 874                                 [IB_QPT_RC]  = (IB_QP_PKEY_INDEX                |
 875                                                 IB_QP_PORT                      |
 876                                                 IB_QP_ACCESS_FLAGS),
 877                                 [IB_QPT_XRC_INI] = (IB_QP_PKEY_INDEX            |
 878                                                 IB_QP_PORT                      |
 879                                                 IB_QP_ACCESS_FLAGS),
 880                                 [IB_QPT_XRC_TGT] = (IB_QP_PKEY_INDEX            |
 881                                                 IB_QP_PORT                      |
 882                                                 IB_QP_ACCESS_FLAGS),
 883                                 [IB_QPT_SMI] = (IB_QP_PKEY_INDEX                |
 884                                                 IB_QP_QKEY),
 885                                 [IB_QPT_GSI] = (IB_QP_PKEY_INDEX                |
 886                                                 IB_QP_QKEY),
 887                         }
 888                 },
 889         },
 890         [IB_QPS_INIT]  = {
 891                 [IB_QPS_RESET] = { .valid = 1 },
 892                 [IB_QPS_ERR] =   { .valid = 1 },
 893                 [IB_QPS_INIT]  = {
 894                         .valid = 1,
 895                         .opt_param = {
 896                                 [IB_QPT_UD]  = (IB_QP_PKEY_INDEX                |
 897                                                 IB_QP_PORT                      |
 898                                                 IB_QP_QKEY),
 899                                 [IB_QPT_UC]  = (IB_QP_PKEY_INDEX                |
 900                                                 IB_QP_PORT                      |
 901                                                 IB_QP_ACCESS_FLAGS),
 902                                 [IB_QPT_RC]  = (IB_QP_PKEY_INDEX                |
 903                                                 IB_QP_PORT                      |
 904                                                 IB_QP_ACCESS_FLAGS),
 905                                 [IB_QPT_XRC_INI] = (IB_QP_PKEY_INDEX            |
 906                                                 IB_QP_PORT                      |
 907                                                 IB_QP_ACCESS_FLAGS),
 908                                 [IB_QPT_XRC_TGT] = (IB_QP_PKEY_INDEX            |
 909                                                 IB_QP_PORT                      |
 910                                                 IB_QP_ACCESS_FLAGS),
 911                                 [IB_QPT_SMI] = (IB_QP_PKEY_INDEX                |
 912                                                 IB_QP_QKEY),
 913                                 [IB_QPT_GSI] = (IB_QP_PKEY_INDEX                |
 914                                                 IB_QP_QKEY),
 915                         }
 916                 },
 917                 [IB_QPS_RTR]   = {
 918                         .valid = 1,
 919                         .req_param = {
 920                                 [IB_QPT_UC]  = (IB_QP_AV                        |
 921                                                 IB_QP_PATH_MTU                  |
 922                                                 IB_QP_DEST_QPN                  |
 923                                                 IB_QP_RQ_PSN),
 924                                 [IB_QPT_RC]  = (IB_QP_AV                        |
 925                                                 IB_QP_PATH_MTU                  |
 926                                                 IB_QP_DEST_QPN                  |
 927                                                 IB_QP_RQ_PSN                    |
 928                                                 IB_QP_MAX_DEST_RD_ATOMIC        |
 929                                                 IB_QP_MIN_RNR_TIMER),
 930                                 [IB_QPT_XRC_INI] = (IB_QP_AV                    |
 931                                                 IB_QP_PATH_MTU                  |
 932                                                 IB_QP_DEST_QPN                  |
 933                                                 IB_QP_RQ_PSN),
 934                                 [IB_QPT_XRC_TGT] = (IB_QP_AV                    |
 935                                                 IB_QP_PATH_MTU                  |
 936                                                 IB_QP_DEST_QPN                  |
 937                                                 IB_QP_RQ_PSN                    |
 938                                                 IB_QP_MAX_DEST_RD_ATOMIC        |
 939                                                 IB_QP_MIN_RNR_TIMER),
 940                         },
 941                         .opt_param = {
 942                                  [IB_QPT_UD]  = (IB_QP_PKEY_INDEX               |
 943                                                  IB_QP_QKEY),
 944                                  [IB_QPT_UC]  = (IB_QP_ALT_PATH                 |
 945                                                  IB_QP_ACCESS_FLAGS             |
 946                                                  IB_QP_PKEY_INDEX),
 947                                  [IB_QPT_RC]  = (IB_QP_ALT_PATH                 |
 948                                                  IB_QP_ACCESS_FLAGS             |
 949                                                  IB_QP_PKEY_INDEX),
 950                                  [IB_QPT_XRC_INI] = (IB_QP_ALT_PATH             |
 951                                                  IB_QP_ACCESS_FLAGS             |
 952                                                  IB_QP_PKEY_INDEX),
 953                                  [IB_QPT_XRC_TGT] = (IB_QP_ALT_PATH             |
 954                                                  IB_QP_ACCESS_FLAGS             |
 955                                                  IB_QP_PKEY_INDEX),
 956                                  [IB_QPT_SMI] = (IB_QP_PKEY_INDEX               |
 957                                                  IB_QP_QKEY),
 958                                  [IB_QPT_GSI] = (IB_QP_PKEY_INDEX               |
 959                                                  IB_QP_QKEY),
 960                          },
 961                 },
 962         },
 963         [IB_QPS_RTR]   = {
 964                 [IB_QPS_RESET] = { .valid = 1 },
 965                 [IB_QPS_ERR] =   { .valid = 1 },
 966                 [IB_QPS_RTS]   = {
 967                         .valid = 1,
 968                         .req_param = {
 969                                 [IB_QPT_UD]  = IB_QP_SQ_PSN,
 970                                 [IB_QPT_UC]  = IB_QP_SQ_PSN,
 971                                 [IB_QPT_RC]  = (IB_QP_TIMEOUT                   |
 972                                                 IB_QP_RETRY_CNT                 |
 973                                                 IB_QP_RNR_RETRY                 |
 974                                                 IB_QP_SQ_PSN                    |
 975                                                 IB_QP_MAX_QP_RD_ATOMIC),
 976                                 [IB_QPT_XRC_INI] = (IB_QP_TIMEOUT               |
 977                                                 IB_QP_RETRY_CNT                 |
 978                                                 IB_QP_RNR_RETRY                 |
 979                                                 IB_QP_SQ_PSN                    |
 980                                                 IB_QP_MAX_QP_RD_ATOMIC),
 981                                 [IB_QPT_XRC_TGT] = (IB_QP_TIMEOUT               |
 982                                                 IB_QP_SQ_PSN),
 983                                 [IB_QPT_SMI] = IB_QP_SQ_PSN,
 984                                 [IB_QPT_GSI] = IB_QP_SQ_PSN,
 985                         },
 986                         .opt_param = {
 987                                  [IB_QPT_UD]  = (IB_QP_CUR_STATE                |
 988                                                  IB_QP_QKEY),
 989                                  [IB_QPT_UC]  = (IB_QP_CUR_STATE                |
 990                                                  IB_QP_ALT_PATH                 |
 991                                                  IB_QP_ACCESS_FLAGS             |
 992                                                  IB_QP_PATH_MIG_STATE),
 993                                  [IB_QPT_RC]  = (IB_QP_CUR_STATE                |
 994                                                  IB_QP_ALT_PATH                 |
 995                                                  IB_QP_ACCESS_FLAGS             |
 996                                                  IB_QP_MIN_RNR_TIMER            |
 997                                                  IB_QP_PATH_MIG_STATE),
 998                                  [IB_QPT_XRC_INI] = (IB_QP_CUR_STATE            |
 999                                                  IB_QP_ALT_PATH                 |
1000                                                  IB_QP_ACCESS_FLAGS             |
1001                                                  IB_QP_PATH_MIG_STATE),
1002                                  [IB_QPT_XRC_TGT] = (IB_QP_CUR_STATE            |
1003                                                  IB_QP_ALT_PATH                 |
1004                                                  IB_QP_ACCESS_FLAGS             |
1005                                                  IB_QP_MIN_RNR_TIMER            |
1006                                                  IB_QP_PATH_MIG_STATE),
1007                                  [IB_QPT_SMI] = (IB_QP_CUR_STATE                |
1008                                                  IB_QP_QKEY),
1009                                  [IB_QPT_GSI] = (IB_QP_CUR_STATE                |
1010                                                  IB_QP_QKEY),
1011                          }
1012                 }
1013         },
1014         [IB_QPS_RTS]   = {
1015                 [IB_QPS_RESET] = { .valid = 1 },
1016                 [IB_QPS_ERR] =   { .valid = 1 },
1017                 [IB_QPS_RTS]   = {
1018                         .valid = 1,
1019                         .opt_param = {
1020                                 [IB_QPT_UD]  = (IB_QP_CUR_STATE                 |
1021                                                 IB_QP_QKEY),
1022                                 [IB_QPT_UC]  = (IB_QP_CUR_STATE                 |
1023                                                 IB_QP_ACCESS_FLAGS              |
1024                                                 IB_QP_ALT_PATH                  |
1025                                                 IB_QP_PATH_MIG_STATE),
1026                                 [IB_QPT_RC]  = (IB_QP_CUR_STATE                 |
1027                                                 IB_QP_ACCESS_FLAGS              |
1028                                                 IB_QP_ALT_PATH                  |
1029                                                 IB_QP_PATH_MIG_STATE            |
1030                                                 IB_QP_MIN_RNR_TIMER),
1031                                 [IB_QPT_XRC_INI] = (IB_QP_CUR_STATE             |
1032                                                 IB_QP_ACCESS_FLAGS              |
1033                                                 IB_QP_ALT_PATH                  |
1034                                                 IB_QP_PATH_MIG_STATE),
1035                                 [IB_QPT_XRC_TGT] = (IB_QP_CUR_STATE             |
1036                                                 IB_QP_ACCESS_FLAGS              |
1037                                                 IB_QP_ALT_PATH                  |
1038                                                 IB_QP_PATH_MIG_STATE            |
1039                                                 IB_QP_MIN_RNR_TIMER),
1040                                 [IB_QPT_SMI] = (IB_QP_CUR_STATE                 |
1041                                                 IB_QP_QKEY),
1042                                 [IB_QPT_GSI] = (IB_QP_CUR_STATE                 |
1043                                                 IB_QP_QKEY),
1044                         }
1045                 },
1046                 [IB_QPS_SQD]   = {
1047                         .valid = 1,
1048                         .opt_param = {
1049                                 [IB_QPT_UD]  = IB_QP_EN_SQD_ASYNC_NOTIFY,
1050                                 [IB_QPT_UC]  = IB_QP_EN_SQD_ASYNC_NOTIFY,
1051                                 [IB_QPT_RC]  = IB_QP_EN_SQD_ASYNC_NOTIFY,
1052                                 [IB_QPT_XRC_INI] = IB_QP_EN_SQD_ASYNC_NOTIFY,
1053                                 [IB_QPT_XRC_TGT] = IB_QP_EN_SQD_ASYNC_NOTIFY, /* ??? */
1054                                 [IB_QPT_SMI] = IB_QP_EN_SQD_ASYNC_NOTIFY,
1055                                 [IB_QPT_GSI] = IB_QP_EN_SQD_ASYNC_NOTIFY
1056                         }
1057                 },
1058         },
1059         [IB_QPS_SQD]   = {
1060                 [IB_QPS_RESET] = { .valid = 1 },
1061                 [IB_QPS_ERR] =   { .valid = 1 },
1062                 [IB_QPS_RTS]   = {
1063                         .valid = 1,
1064                         .opt_param = {
1065                                 [IB_QPT_UD]  = (IB_QP_CUR_STATE                 |
1066                                                 IB_QP_QKEY),
1067                                 [IB_QPT_UC]  = (IB_QP_CUR_STATE                 |
1068                                                 IB_QP_ALT_PATH                  |
1069                                                 IB_QP_ACCESS_FLAGS              |
1070                                                 IB_QP_PATH_MIG_STATE),
1071                                 [IB_QPT_RC]  = (IB_QP_CUR_STATE                 |
1072                                                 IB_QP_ALT_PATH                  |
1073                                                 IB_QP_ACCESS_FLAGS              |
1074                                                 IB_QP_MIN_RNR_TIMER             |
1075                                                 IB_QP_PATH_MIG_STATE),
1076                                 [IB_QPT_XRC_INI] = (IB_QP_CUR_STATE             |
1077                                                 IB_QP_ALT_PATH                  |
1078                                                 IB_QP_ACCESS_FLAGS              |
1079                                                 IB_QP_PATH_MIG_STATE),
1080                                 [IB_QPT_XRC_TGT] = (IB_QP_CUR_STATE             |
1081                                                 IB_QP_ALT_PATH                  |
1082                                                 IB_QP_ACCESS_FLAGS              |
1083                                                 IB_QP_MIN_RNR_TIMER             |
1084                                                 IB_QP_PATH_MIG_STATE),
1085                                 [IB_QPT_SMI] = (IB_QP_CUR_STATE                 |
1086                                                 IB_QP_QKEY),
1087                                 [IB_QPT_GSI] = (IB_QP_CUR_STATE                 |
1088                                                 IB_QP_QKEY),
1089                         }
1090                 },
1091                 [IB_QPS_SQD]   = {
1092                         .valid = 1,
1093                         .opt_param = {
1094                                 [IB_QPT_UD]  = (IB_QP_PKEY_INDEX                |
1095                                                 IB_QP_QKEY),
1096                                 [IB_QPT_UC]  = (IB_QP_AV                        |
1097                                                 IB_QP_ALT_PATH                  |
1098                                                 IB_QP_ACCESS_FLAGS              |
1099                                                 IB_QP_PKEY_INDEX                |
1100                                                 IB_QP_PATH_MIG_STATE),
1101                                 [IB_QPT_RC]  = (IB_QP_PORT                      |
1102                                                 IB_QP_AV                        |
1103                                                 IB_QP_TIMEOUT                   |
1104                                                 IB_QP_RETRY_CNT                 |
1105                                                 IB_QP_RNR_RETRY                 |
1106                                                 IB_QP_MAX_QP_RD_ATOMIC          |
1107                                                 IB_QP_MAX_DEST_RD_ATOMIC        |
1108                                                 IB_QP_ALT_PATH                  |
1109                                                 IB_QP_ACCESS_FLAGS              |
1110                                                 IB_QP_PKEY_INDEX                |
1111                                                 IB_QP_MIN_RNR_TIMER             |
1112                                                 IB_QP_PATH_MIG_STATE),
1113                                 [IB_QPT_XRC_INI] = (IB_QP_PORT                  |
1114                                                 IB_QP_AV                        |
1115                                                 IB_QP_TIMEOUT                   |
1116                                                 IB_QP_RETRY_CNT                 |
1117                                                 IB_QP_RNR_RETRY                 |
1118                                                 IB_QP_MAX_QP_RD_ATOMIC          |
1119                                                 IB_QP_ALT_PATH                  |
1120                                                 IB_QP_ACCESS_FLAGS              |
1121                                                 IB_QP_PKEY_INDEX                |
1122                                                 IB_QP_PATH_MIG_STATE),
1123                                 [IB_QPT_XRC_TGT] = (IB_QP_PORT                  |
1124                                                 IB_QP_AV                        |
1125                                                 IB_QP_TIMEOUT                   |
1126                                                 IB_QP_MAX_DEST_RD_ATOMIC        |
1127                                                 IB_QP_ALT_PATH                  |
1128                                                 IB_QP_ACCESS_FLAGS              |
1129                                                 IB_QP_PKEY_INDEX                |
1130                                                 IB_QP_MIN_RNR_TIMER             |
1131                                                 IB_QP_PATH_MIG_STATE),
1132                                 [IB_QPT_SMI] = (IB_QP_PKEY_INDEX                |
1133                                                 IB_QP_QKEY),
1134                                 [IB_QPT_GSI] = (IB_QP_PKEY_INDEX                |
1135                                                 IB_QP_QKEY),
1136                         }
1137                 }
1138         },
1139         [IB_QPS_SQE]   = {
1140                 [IB_QPS_RESET] = { .valid = 1 },
1141                 [IB_QPS_ERR] =   { .valid = 1 },
1142                 [IB_QPS_RTS]   = {
1143                         .valid = 1,
1144                         .opt_param = {
1145                                 [IB_QPT_UD]  = (IB_QP_CUR_STATE                 |
1146                                                 IB_QP_QKEY),
1147                                 [IB_QPT_UC]  = (IB_QP_CUR_STATE                 |
1148                                                 IB_QP_ACCESS_FLAGS),
1149                                 [IB_QPT_SMI] = (IB_QP_CUR_STATE                 |
1150                                                 IB_QP_QKEY),
1151                                 [IB_QPT_GSI] = (IB_QP_CUR_STATE                 |
1152                                                 IB_QP_QKEY),
1153                         }
1154                 }
1155         },
1156         [IB_QPS_ERR] = {
1157                 [IB_QPS_RESET] = { .valid = 1 },
1158                 [IB_QPS_ERR] =   { .valid = 1 }
1159         }
1160 };
1161
1162 int ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
1163                        enum ib_qp_type type, enum ib_qp_attr_mask mask,
1164                        enum rdma_link_layer ll)
1165 {
1166         enum ib_qp_attr_mask req_param, opt_param;
1167
1168         if (cur_state  < 0 || cur_state  > IB_QPS_ERR ||
1169             next_state < 0 || next_state > IB_QPS_ERR)
1170                 return 0;
1171
1172         if (mask & IB_QP_CUR_STATE  &&
1173             cur_state != IB_QPS_RTR && cur_state != IB_QPS_RTS &&
1174             cur_state != IB_QPS_SQD && cur_state != IB_QPS_SQE)
1175                 return 0;
1176
1177         if (!qp_state_table[cur_state][next_state].valid)
1178                 return 0;
1179
1180         req_param = qp_state_table[cur_state][next_state].req_param[type];
1181         opt_param = qp_state_table[cur_state][next_state].opt_param[type];
1182
1183         if ((mask & req_param) != req_param)
1184                 return 0;
1185
1186         if (mask & ~(req_param | opt_param | IB_QP_STATE))
1187                 return 0;
1188
1189         return 1;
1190 }
1191 EXPORT_SYMBOL(ib_modify_qp_is_ok);
1192
1193 int ib_resolve_eth_dmac(struct ib_device *device,
1194                         struct ib_ah_attr *ah_attr)
1195 {
1196         struct ib_gid_attr sgid_attr;
1197         union ib_gid sgid;
1198         int hop_limit;
1199         int ret;
1200
1201         if (ah_attr->port_num < rdma_start_port(device) ||
1202             ah_attr->port_num > rdma_end_port(device))
1203                 return -EINVAL;
1204
1205         if (!rdma_cap_eth_ah(device, ah_attr->port_num))
1206                 return 0;
1207
1208         if (rdma_is_multicast_addr((struct in6_addr *)ah_attr->grh.dgid.raw)) {
1209                 if (ipv6_addr_v4mapped((struct in6_addr *)ah_attr->grh.dgid.raw)) {
1210                         __be32 addr = 0;
1211
1212                         memcpy(&addr, ah_attr->grh.dgid.raw + 12, 4);
1213                         ip_eth_mc_map(addr, (char *)ah_attr->dmac);
1214                 } else {
1215                         ipv6_eth_mc_map((struct in6_addr *)ah_attr->grh.dgid.raw,
1216                                         (char *)ah_attr->dmac);
1217                 }
1218                 return 0;
1219         }
1220
1221         ret = ib_query_gid(device,
1222                            ah_attr->port_num,
1223                            ah_attr->grh.sgid_index,
1224                            &sgid, &sgid_attr);
1225         if (ret != 0)
1226                 return (ret);
1227         if (!sgid_attr.ndev)
1228                 return -ENXIO;
1229
1230         ret = rdma_addr_find_l2_eth_by_grh(&sgid,
1231                                            &ah_attr->grh.dgid,
1232                                            ah_attr->dmac,
1233                                            sgid_attr.ndev, &hop_limit);
1234         dev_put(sgid_attr.ndev);
1235
1236         ah_attr->grh.hop_limit = hop_limit;
1237         return ret;
1238 }
1239 EXPORT_SYMBOL(ib_resolve_eth_dmac);
1240
1241
1242 int ib_modify_qp(struct ib_qp *qp,
1243                  struct ib_qp_attr *qp_attr,
1244                  int qp_attr_mask)
1245 {
1246         if (qp_attr_mask & IB_QP_AV) {
1247                 int ret;
1248
1249                 ret = ib_resolve_eth_dmac(qp->device, &qp_attr->ah_attr);
1250                 if (ret)
1251                         return ret;
1252         }
1253
1254         return qp->device->modify_qp(qp->real_qp, qp_attr, qp_attr_mask, NULL);
1255 }
1256 EXPORT_SYMBOL(ib_modify_qp);
1257
1258 int ib_query_qp(struct ib_qp *qp,
1259                 struct ib_qp_attr *qp_attr,
1260                 int qp_attr_mask,
1261                 struct ib_qp_init_attr *qp_init_attr)
1262 {
1263         return qp->device->query_qp ?
1264                 qp->device->query_qp(qp->real_qp, qp_attr, qp_attr_mask, qp_init_attr) :
1265                 -ENOSYS;
1266 }
1267 EXPORT_SYMBOL(ib_query_qp);
1268
1269 int ib_close_qp(struct ib_qp *qp)
1270 {
1271         struct ib_qp *real_qp;
1272         unsigned long flags;
1273
1274         real_qp = qp->real_qp;
1275         if (real_qp == qp)
1276                 return -EINVAL;
1277
1278         spin_lock_irqsave(&real_qp->device->event_handler_lock, flags);
1279         list_del(&qp->open_list);
1280         spin_unlock_irqrestore(&real_qp->device->event_handler_lock, flags);
1281
1282         atomic_dec(&real_qp->usecnt);
1283         kfree(qp);
1284
1285         return 0;
1286 }
1287 EXPORT_SYMBOL(ib_close_qp);
1288
1289 static int __ib_destroy_shared_qp(struct ib_qp *qp)
1290 {
1291         struct ib_xrcd *xrcd;
1292         struct ib_qp *real_qp;
1293         int ret;
1294
1295         real_qp = qp->real_qp;
1296         xrcd = real_qp->xrcd;
1297
1298         mutex_lock(&xrcd->tgt_qp_mutex);
1299         ib_close_qp(qp);
1300         if (atomic_read(&real_qp->usecnt) == 0)
1301                 list_del(&real_qp->xrcd_list);
1302         else
1303                 real_qp = NULL;
1304         mutex_unlock(&xrcd->tgt_qp_mutex);
1305
1306         if (real_qp) {
1307                 ret = ib_destroy_qp(real_qp);
1308                 if (!ret)
1309                         atomic_dec(&xrcd->usecnt);
1310                 else
1311                         __ib_insert_xrcd_qp(xrcd, real_qp);
1312         }
1313
1314         return 0;
1315 }
1316
1317 int ib_destroy_qp(struct ib_qp *qp)
1318 {
1319         struct ib_pd *pd;
1320         struct ib_cq *scq, *rcq;
1321         struct ib_srq *srq;
1322         struct ib_rwq_ind_table *ind_tbl;
1323         int ret;
1324
1325         if (atomic_read(&qp->usecnt))
1326                 return -EBUSY;
1327
1328         if (qp->real_qp != qp)
1329                 return __ib_destroy_shared_qp(qp);
1330
1331         pd   = qp->pd;
1332         scq  = qp->send_cq;
1333         rcq  = qp->recv_cq;
1334         srq  = qp->srq;
1335         ind_tbl = qp->rwq_ind_tbl;
1336
1337         ret = qp->device->destroy_qp(qp);
1338         if (!ret) {
1339                 if (pd)
1340                         atomic_dec(&pd->usecnt);
1341                 if (scq)
1342                         atomic_dec(&scq->usecnt);
1343                 if (rcq)
1344                         atomic_dec(&rcq->usecnt);
1345                 if (srq)
1346                         atomic_dec(&srq->usecnt);
1347                 if (ind_tbl)
1348                         atomic_dec(&ind_tbl->usecnt);
1349         }
1350
1351         return ret;
1352 }
1353 EXPORT_SYMBOL(ib_destroy_qp);
1354
1355 /* Completion queues */
1356
1357 struct ib_cq *ib_create_cq(struct ib_device *device,
1358                            ib_comp_handler comp_handler,
1359                            void (*event_handler)(struct ib_event *, void *),
1360                            void *cq_context,
1361                            const struct ib_cq_init_attr *cq_attr)
1362 {
1363         struct ib_cq *cq;
1364
1365         cq = device->create_cq(device, cq_attr, NULL, NULL);
1366
1367         if (!IS_ERR(cq)) {
1368                 cq->device        = device;
1369                 cq->uobject       = NULL;
1370                 cq->comp_handler  = comp_handler;
1371                 cq->event_handler = event_handler;
1372                 cq->cq_context    = cq_context;
1373                 atomic_set(&cq->usecnt, 0);
1374         }
1375
1376         return cq;
1377 }
1378 EXPORT_SYMBOL(ib_create_cq);
1379
1380 int ib_modify_cq(struct ib_cq *cq, u16 cq_count, u16 cq_period)
1381 {
1382         return cq->device->modify_cq ?
1383                 cq->device->modify_cq(cq, cq_count, cq_period) : -ENOSYS;
1384 }
1385 EXPORT_SYMBOL(ib_modify_cq);
1386
1387 int ib_destroy_cq(struct ib_cq *cq)
1388 {
1389         if (atomic_read(&cq->usecnt))
1390                 return -EBUSY;
1391
1392         return cq->device->destroy_cq(cq);
1393 }
1394 EXPORT_SYMBOL(ib_destroy_cq);
1395
1396 int ib_resize_cq(struct ib_cq *cq, int cqe)
1397 {
1398         return cq->device->resize_cq ?
1399                 cq->device->resize_cq(cq, cqe, NULL) : -ENOSYS;
1400 }
1401 EXPORT_SYMBOL(ib_resize_cq);
1402
1403 /* Memory regions */
1404
1405 int ib_dereg_mr(struct ib_mr *mr)
1406 {
1407         struct ib_pd *pd = mr->pd;
1408         int ret;
1409
1410         ret = mr->device->dereg_mr(mr);
1411         if (!ret)
1412                 atomic_dec(&pd->usecnt);
1413
1414         return ret;
1415 }
1416 EXPORT_SYMBOL(ib_dereg_mr);
1417
1418 /**
1419  * ib_alloc_mr() - Allocates a memory region
1420  * @pd:            protection domain associated with the region
1421  * @mr_type:       memory region type
1422  * @max_num_sg:    maximum sg entries available for registration.
1423  *
1424  * Notes:
1425  * Memory registeration page/sg lists must not exceed max_num_sg.
1426  * For mr_type IB_MR_TYPE_MEM_REG, the total length cannot exceed
1427  * max_num_sg * used_page_size.
1428  *
1429  */
1430 struct ib_mr *ib_alloc_mr(struct ib_pd *pd,
1431                           enum ib_mr_type mr_type,
1432                           u32 max_num_sg)
1433 {
1434         struct ib_mr *mr;
1435
1436         if (!pd->device->alloc_mr)
1437                 return ERR_PTR(-ENOSYS);
1438
1439         mr = pd->device->alloc_mr(pd, mr_type, max_num_sg);
1440         if (!IS_ERR(mr)) {
1441                 mr->device  = pd->device;
1442                 mr->pd      = pd;
1443                 mr->uobject = NULL;
1444                 atomic_inc(&pd->usecnt);
1445                 mr->need_inval = false;
1446         }
1447
1448         return mr;
1449 }
1450 EXPORT_SYMBOL(ib_alloc_mr);
1451
1452 /* "Fast" memory regions */
1453
1454 struct ib_fmr *ib_alloc_fmr(struct ib_pd *pd,
1455                             int mr_access_flags,
1456                             struct ib_fmr_attr *fmr_attr)
1457 {
1458         struct ib_fmr *fmr;
1459
1460         if (!pd->device->alloc_fmr)
1461                 return ERR_PTR(-ENOSYS);
1462
1463         fmr = pd->device->alloc_fmr(pd, mr_access_flags, fmr_attr);
1464         if (!IS_ERR(fmr)) {
1465                 fmr->device = pd->device;
1466                 fmr->pd     = pd;
1467                 atomic_inc(&pd->usecnt);
1468         }
1469
1470         return fmr;
1471 }
1472 EXPORT_SYMBOL(ib_alloc_fmr);
1473
1474 int ib_unmap_fmr(struct list_head *fmr_list)
1475 {
1476         struct ib_fmr *fmr;
1477
1478         if (list_empty(fmr_list))
1479                 return 0;
1480
1481         fmr = list_entry(fmr_list->next, struct ib_fmr, list);
1482         return fmr->device->unmap_fmr(fmr_list);
1483 }
1484 EXPORT_SYMBOL(ib_unmap_fmr);
1485
1486 int ib_dealloc_fmr(struct ib_fmr *fmr)
1487 {
1488         struct ib_pd *pd;
1489         int ret;
1490
1491         pd = fmr->pd;
1492         ret = fmr->device->dealloc_fmr(fmr);
1493         if (!ret)
1494                 atomic_dec(&pd->usecnt);
1495
1496         return ret;
1497 }
1498 EXPORT_SYMBOL(ib_dealloc_fmr);
1499
1500 /* Multicast groups */
1501
1502 static bool is_valid_mcast_lid(struct ib_qp *qp, u16 lid)
1503 {
1504         struct ib_qp_init_attr init_attr = {};
1505         struct ib_qp_attr attr = {};
1506         int num_eth_ports = 0;
1507         int port;
1508
1509         /* If QP state >= init, it is assigned to a port and we can check this
1510          * port only.
1511          */
1512         if (!ib_query_qp(qp, &attr, IB_QP_STATE | IB_QP_PORT, &init_attr)) {
1513                 if (attr.qp_state >= IB_QPS_INIT) {
1514                         if (rdma_port_get_link_layer(qp->device, attr.port_num) !=
1515                             IB_LINK_LAYER_INFINIBAND)
1516                                 return true;
1517                         goto lid_check;
1518                 }
1519         }
1520
1521         /* Can't get a quick answer, iterate over all ports */
1522         for (port = 0; port < qp->device->phys_port_cnt; port++)
1523                 if (rdma_port_get_link_layer(qp->device, port) !=
1524                     IB_LINK_LAYER_INFINIBAND)
1525                         num_eth_ports++;
1526
1527         /* If we have at lease one Ethernet port, RoCE annex declares that
1528          * multicast LID should be ignored. We can't tell at this step if the
1529          * QP belongs to an IB or Ethernet port.
1530          */
1531         if (num_eth_ports)
1532                 return true;
1533
1534         /* If all the ports are IB, we can check according to IB spec. */
1535 lid_check:
1536         return !(lid < be16_to_cpu(IB_MULTICAST_LID_BASE) ||
1537                  lid == be16_to_cpu(IB_LID_PERMISSIVE));
1538 }
1539
1540 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid)
1541 {
1542         int ret;
1543
1544         if (!qp->device->attach_mcast)
1545                 return -ENOSYS;
1546
1547         if (!rdma_is_multicast_addr((struct in6_addr *)gid->raw) ||
1548             qp->qp_type != IB_QPT_UD || !is_valid_mcast_lid(qp, lid))
1549                 return -EINVAL;
1550
1551         ret = qp->device->attach_mcast(qp, gid, lid);
1552         if (!ret)
1553                 atomic_inc(&qp->usecnt);
1554         return ret;
1555 }
1556 EXPORT_SYMBOL(ib_attach_mcast);
1557
1558 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid)
1559 {
1560         int ret;
1561
1562         if (!qp->device->detach_mcast)
1563                 return -ENOSYS;
1564
1565         if (!rdma_is_multicast_addr((struct in6_addr *)gid->raw) ||
1566             qp->qp_type != IB_QPT_UD || !is_valid_mcast_lid(qp, lid))
1567                 return -EINVAL;
1568
1569         ret = qp->device->detach_mcast(qp, gid, lid);
1570         if (!ret)
1571                 atomic_dec(&qp->usecnt);
1572         return ret;
1573 }
1574 EXPORT_SYMBOL(ib_detach_mcast);
1575
1576 struct ib_xrcd *ib_alloc_xrcd(struct ib_device *device)
1577 {
1578         struct ib_xrcd *xrcd;
1579
1580         if (!device->alloc_xrcd)
1581                 return ERR_PTR(-ENOSYS);
1582
1583         xrcd = device->alloc_xrcd(device, NULL, NULL);
1584         if (!IS_ERR(xrcd)) {
1585                 xrcd->device = device;
1586                 xrcd->inode = NULL;
1587                 atomic_set(&xrcd->usecnt, 0);
1588                 mutex_init(&xrcd->tgt_qp_mutex);
1589                 INIT_LIST_HEAD(&xrcd->tgt_qp_list);
1590         }
1591
1592         return xrcd;
1593 }
1594 EXPORT_SYMBOL(ib_alloc_xrcd);
1595
1596 int ib_dealloc_xrcd(struct ib_xrcd *xrcd)
1597 {
1598         struct ib_qp *qp;
1599         int ret;
1600
1601         if (atomic_read(&xrcd->usecnt))
1602                 return -EBUSY;
1603
1604         while (!list_empty(&xrcd->tgt_qp_list)) {
1605                 qp = list_entry(xrcd->tgt_qp_list.next, struct ib_qp, xrcd_list);
1606                 ret = ib_destroy_qp(qp);
1607                 if (ret)
1608                         return ret;
1609         }
1610
1611         return xrcd->device->dealloc_xrcd(xrcd);
1612 }
1613 EXPORT_SYMBOL(ib_dealloc_xrcd);
1614
1615 /**
1616  * ib_create_wq - Creates a WQ associated with the specified protection
1617  * domain.
1618  * @pd: The protection domain associated with the WQ.
1619  * @wq_init_attr: A list of initial attributes required to create the
1620  * WQ. If WQ creation succeeds, then the attributes are updated to
1621  * the actual capabilities of the created WQ.
1622  *
1623  * wq_init_attr->max_wr and wq_init_attr->max_sge determine
1624  * the requested size of the WQ, and set to the actual values allocated
1625  * on return.
1626  * If ib_create_wq() succeeds, then max_wr and max_sge will always be
1627  * at least as large as the requested values.
1628  */
1629 struct ib_wq *ib_create_wq(struct ib_pd *pd,
1630                            struct ib_wq_init_attr *wq_attr)
1631 {
1632         struct ib_wq *wq;
1633
1634         if (!pd->device->create_wq)
1635                 return ERR_PTR(-ENOSYS);
1636
1637         wq = pd->device->create_wq(pd, wq_attr, NULL);
1638         if (!IS_ERR(wq)) {
1639                 wq->event_handler = wq_attr->event_handler;
1640                 wq->wq_context = wq_attr->wq_context;
1641                 wq->wq_type = wq_attr->wq_type;
1642                 wq->cq = wq_attr->cq;
1643                 wq->device = pd->device;
1644                 wq->pd = pd;
1645                 wq->uobject = NULL;
1646                 atomic_inc(&pd->usecnt);
1647                 atomic_inc(&wq_attr->cq->usecnt);
1648                 atomic_set(&wq->usecnt, 0);
1649         }
1650         return wq;
1651 }
1652 EXPORT_SYMBOL(ib_create_wq);
1653
1654 /**
1655  * ib_destroy_wq - Destroys the specified WQ.
1656  * @wq: The WQ to destroy.
1657  */
1658 int ib_destroy_wq(struct ib_wq *wq)
1659 {
1660         int err;
1661         struct ib_cq *cq = wq->cq;
1662         struct ib_pd *pd = wq->pd;
1663
1664         if (atomic_read(&wq->usecnt))
1665                 return -EBUSY;
1666
1667         err = wq->device->destroy_wq(wq);
1668         if (!err) {
1669                 atomic_dec(&pd->usecnt);
1670                 atomic_dec(&cq->usecnt);
1671         }
1672         return err;
1673 }
1674 EXPORT_SYMBOL(ib_destroy_wq);
1675
1676 /**
1677  * ib_modify_wq - Modifies the specified WQ.
1678  * @wq: The WQ to modify.
1679  * @wq_attr: On input, specifies the WQ attributes to modify.
1680  * @wq_attr_mask: A bit-mask used to specify which attributes of the WQ
1681  *   are being modified.
1682  * On output, the current values of selected WQ attributes are returned.
1683  */
1684 int ib_modify_wq(struct ib_wq *wq, struct ib_wq_attr *wq_attr,
1685                  u32 wq_attr_mask)
1686 {
1687         int err;
1688
1689         if (!wq->device->modify_wq)
1690                 return -ENOSYS;
1691
1692         err = wq->device->modify_wq(wq, wq_attr, wq_attr_mask, NULL);
1693         return err;
1694 }
1695 EXPORT_SYMBOL(ib_modify_wq);
1696
1697 /*
1698  * ib_create_rwq_ind_table - Creates a RQ Indirection Table.
1699  * @device: The device on which to create the rwq indirection table.
1700  * @ib_rwq_ind_table_init_attr: A list of initial attributes required to
1701  * create the Indirection Table.
1702  *
1703  * Note: The life time of ib_rwq_ind_table_init_attr->ind_tbl is not less
1704  *      than the created ib_rwq_ind_table object and the caller is responsible
1705  *      for its memory allocation/free.
1706  */
1707 struct ib_rwq_ind_table *ib_create_rwq_ind_table(struct ib_device *device,
1708                                                  struct ib_rwq_ind_table_init_attr *init_attr)
1709 {
1710         struct ib_rwq_ind_table *rwq_ind_table;
1711         int i;
1712         u32 table_size;
1713
1714         if (!device->create_rwq_ind_table)
1715                 return ERR_PTR(-ENOSYS);
1716
1717         table_size = (1 << init_attr->log_ind_tbl_size);
1718         rwq_ind_table = device->create_rwq_ind_table(device,
1719                                 init_attr, NULL);
1720         if (IS_ERR(rwq_ind_table))
1721                 return rwq_ind_table;
1722
1723         rwq_ind_table->ind_tbl = init_attr->ind_tbl;
1724         rwq_ind_table->log_ind_tbl_size = init_attr->log_ind_tbl_size;
1725         rwq_ind_table->device = device;
1726         rwq_ind_table->uobject = NULL;
1727         atomic_set(&rwq_ind_table->usecnt, 0);
1728
1729         for (i = 0; i < table_size; i++)
1730                 atomic_inc(&rwq_ind_table->ind_tbl[i]->usecnt);
1731
1732         return rwq_ind_table;
1733 }
1734 EXPORT_SYMBOL(ib_create_rwq_ind_table);
1735
1736 /*
1737  * ib_destroy_rwq_ind_table - Destroys the specified Indirection Table.
1738  * @wq_ind_table: The Indirection Table to destroy.
1739 */
1740 int ib_destroy_rwq_ind_table(struct ib_rwq_ind_table *rwq_ind_table)
1741 {
1742         int err, i;
1743         u32 table_size = (1 << rwq_ind_table->log_ind_tbl_size);
1744         struct ib_wq **ind_tbl = rwq_ind_table->ind_tbl;
1745
1746         if (atomic_read(&rwq_ind_table->usecnt))
1747                 return -EBUSY;
1748
1749         err = rwq_ind_table->device->destroy_rwq_ind_table(rwq_ind_table);
1750         if (!err) {
1751                 for (i = 0; i < table_size; i++)
1752                         atomic_dec(&ind_tbl[i]->usecnt);
1753         }
1754
1755         return err;
1756 }
1757 EXPORT_SYMBOL(ib_destroy_rwq_ind_table);
1758
1759 struct ib_flow *ib_create_flow(struct ib_qp *qp,
1760                                struct ib_flow_attr *flow_attr,
1761                                int domain)
1762 {
1763         struct ib_flow *flow_id;
1764         if (!qp->device->create_flow)
1765                 return ERR_PTR(-ENOSYS);
1766
1767         flow_id = qp->device->create_flow(qp, flow_attr, domain);
1768         if (!IS_ERR(flow_id))
1769                 atomic_inc(&qp->usecnt);
1770         return flow_id;
1771 }
1772 EXPORT_SYMBOL(ib_create_flow);
1773
1774 int ib_destroy_flow(struct ib_flow *flow_id)
1775 {
1776         int err;
1777         struct ib_qp *qp = flow_id->qp;
1778
1779         err = qp->device->destroy_flow(flow_id);
1780         if (!err)
1781                 atomic_dec(&qp->usecnt);
1782         return err;
1783 }
1784 EXPORT_SYMBOL(ib_destroy_flow);
1785
1786 int ib_check_mr_status(struct ib_mr *mr, u32 check_mask,
1787                        struct ib_mr_status *mr_status)
1788 {
1789         return mr->device->check_mr_status ?
1790                 mr->device->check_mr_status(mr, check_mask, mr_status) : -ENOSYS;
1791 }
1792 EXPORT_SYMBOL(ib_check_mr_status);
1793
1794 int ib_set_vf_link_state(struct ib_device *device, int vf, u8 port,
1795                          int state)
1796 {
1797         if (!device->set_vf_link_state)
1798                 return -ENOSYS;
1799
1800         return device->set_vf_link_state(device, vf, port, state);
1801 }
1802 EXPORT_SYMBOL(ib_set_vf_link_state);
1803
1804 int ib_get_vf_config(struct ib_device *device, int vf, u8 port,
1805                      struct ifla_vf_info *info)
1806 {
1807         if (!device->get_vf_config)
1808                 return -ENOSYS;
1809
1810         return device->get_vf_config(device, vf, port, info);
1811 }
1812 EXPORT_SYMBOL(ib_get_vf_config);
1813
1814 int ib_get_vf_stats(struct ib_device *device, int vf, u8 port,
1815                     struct ifla_vf_stats *stats)
1816 {
1817         if (!device->get_vf_stats)
1818                 return -ENOSYS;
1819
1820         return device->get_vf_stats(device, vf, port, stats);
1821 }
1822 EXPORT_SYMBOL(ib_get_vf_stats);
1823
1824 int ib_set_vf_guid(struct ib_device *device, int vf, u8 port, u64 guid,
1825                    int type)
1826 {
1827         if (!device->set_vf_guid)
1828                 return -ENOSYS;
1829
1830         return device->set_vf_guid(device, vf, port, guid, type);
1831 }
1832 EXPORT_SYMBOL(ib_set_vf_guid);
1833
1834 /**
1835  * ib_map_mr_sg() - Map the largest prefix of a dma mapped SG list
1836  *     and set it the memory region.
1837  * @mr:            memory region
1838  * @sg:            dma mapped scatterlist
1839  * @sg_nents:      number of entries in sg
1840  * @sg_offset:     offset in bytes into sg
1841  * @page_size:     page vector desired page size
1842  *
1843  * Constraints:
1844  * - The first sg element is allowed to have an offset.
1845  * - Each sg element must either be aligned to page_size or virtually
1846  *   contiguous to the previous element. In case an sg element has a
1847  *   non-contiguous offset, the mapping prefix will not include it.
1848  * - The last sg element is allowed to have length less than page_size.
1849  * - If sg_nents total byte length exceeds the mr max_num_sge * page_size
1850  *   then only max_num_sg entries will be mapped.
1851  * - If the MR was allocated with type IB_MR_TYPE_SG_GAPS, none of these
1852  *   constraints holds and the page_size argument is ignored.
1853  *
1854  * Returns the number of sg elements that were mapped to the memory region.
1855  *
1856  * After this completes successfully, the  memory region
1857  * is ready for registration.
1858  */
1859 int ib_map_mr_sg(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
1860                  unsigned int *sg_offset, unsigned int page_size)
1861 {
1862         if (unlikely(!mr->device->map_mr_sg))
1863                 return -ENOSYS;
1864
1865         mr->page_size = page_size;
1866
1867         return mr->device->map_mr_sg(mr, sg, sg_nents, sg_offset);
1868 }
1869 EXPORT_SYMBOL(ib_map_mr_sg);
1870
1871 /**
1872  * ib_sg_to_pages() - Convert the largest prefix of a sg list
1873  *     to a page vector
1874  * @mr:            memory region
1875  * @sgl:           dma mapped scatterlist
1876  * @sg_nents:      number of entries in sg
1877  * @sg_offset_p:   IN:  start offset in bytes into sg
1878  *                 OUT: offset in bytes for element n of the sg of the first
1879  *                      byte that has not been processed where n is the return
1880  *                      value of this function.
1881  * @set_page:      driver page assignment function pointer
1882  *
1883  * Core service helper for drivers to convert the largest
1884  * prefix of given sg list to a page vector. The sg list
1885  * prefix converted is the prefix that meet the requirements
1886  * of ib_map_mr_sg.
1887  *
1888  * Returns the number of sg elements that were assigned to
1889  * a page vector.
1890  */
1891 int ib_sg_to_pages(struct ib_mr *mr, struct scatterlist *sgl, int sg_nents,
1892                 unsigned int *sg_offset_p, int (*set_page)(struct ib_mr *, u64))
1893 {
1894         struct scatterlist *sg;
1895         u64 last_end_dma_addr = 0;
1896         unsigned int sg_offset = sg_offset_p ? *sg_offset_p : 0;
1897         unsigned int last_page_off = 0;
1898         u64 page_mask = ~((u64)mr->page_size - 1);
1899         int i, ret;
1900
1901         if (unlikely(sg_nents <= 0 || sg_offset > sg_dma_len(&sgl[0])))
1902                 return -EINVAL;
1903
1904         mr->iova = sg_dma_address(&sgl[0]) + sg_offset;
1905         mr->length = 0;
1906
1907         for_each_sg(sgl, sg, sg_nents, i) {
1908                 u64 dma_addr = sg_dma_address(sg) + sg_offset;
1909                 u64 prev_addr = dma_addr;
1910                 unsigned int dma_len = sg_dma_len(sg) - sg_offset;
1911                 u64 end_dma_addr = dma_addr + dma_len;
1912                 u64 page_addr = dma_addr & page_mask;
1913
1914                 /*
1915                  * For the second and later elements, check whether either the
1916                  * end of element i-1 or the start of element i is not aligned
1917                  * on a page boundary.
1918                  */
1919                 if (i && (last_page_off != 0 || page_addr != dma_addr)) {
1920                         /* Stop mapping if there is a gap. */
1921                         if (last_end_dma_addr != dma_addr)
1922                                 break;
1923
1924                         /*
1925                          * Coalesce this element with the last. If it is small
1926                          * enough just update mr->length. Otherwise start
1927                          * mapping from the next page.
1928                          */
1929                         goto next_page;
1930                 }
1931
1932                 do {
1933                         ret = set_page(mr, page_addr);
1934                         if (unlikely(ret < 0)) {
1935                                 sg_offset = prev_addr - sg_dma_address(sg);
1936                                 mr->length += prev_addr - dma_addr;
1937                                 if (sg_offset_p)
1938                                         *sg_offset_p = sg_offset;
1939                                 return i || sg_offset ? i : ret;
1940                         }
1941                         prev_addr = page_addr;
1942 next_page:
1943                         page_addr += mr->page_size;
1944                 } while (page_addr < end_dma_addr);
1945
1946                 mr->length += dma_len;
1947                 last_end_dma_addr = end_dma_addr;
1948                 last_page_off = end_dma_addr & ~page_mask;
1949
1950                 sg_offset = 0;
1951         }
1952
1953         if (sg_offset_p)
1954                 *sg_offset_p = 0;
1955         return i;
1956 }
1957 EXPORT_SYMBOL(ib_sg_to_pages);
1958
1959 struct ib_drain_cqe {
1960         struct ib_cqe cqe;
1961         struct completion done;
1962 };
1963
1964 static void ib_drain_qp_done(struct ib_cq *cq, struct ib_wc *wc)
1965 {
1966         struct ib_drain_cqe *cqe = container_of(wc->wr_cqe, struct ib_drain_cqe,
1967                                                 cqe);
1968
1969         complete(&cqe->done);
1970 }
1971
1972 /*
1973  * Post a WR and block until its completion is reaped for the SQ.
1974  */
1975 static void __ib_drain_sq(struct ib_qp *qp)
1976 {
1977         struct ib_qp_attr attr = { .qp_state = IB_QPS_ERR };
1978         struct ib_drain_cqe sdrain;
1979         struct ib_send_wr *bad_swr;
1980         struct ib_rdma_wr swr = {
1981                 .wr = {
1982                         .opcode = IB_WR_RDMA_WRITE,
1983                         .wr_cqe = &sdrain.cqe,
1984                 },
1985         };
1986         int ret;
1987
1988         if (qp->send_cq->poll_ctx == IB_POLL_DIRECT) {
1989                 WARN_ONCE(qp->send_cq->poll_ctx == IB_POLL_DIRECT,
1990                           "IB_POLL_DIRECT poll_ctx not supported for drain\n");
1991                 return;
1992         }
1993
1994         sdrain.cqe.done = ib_drain_qp_done;
1995         init_completion(&sdrain.done);
1996
1997         ret = ib_modify_qp(qp, &attr, IB_QP_STATE);
1998         if (ret) {
1999                 WARN_ONCE(ret, "failed to drain send queue: %d\n", ret);
2000                 return;
2001         }
2002
2003         ret = ib_post_send(qp, &swr.wr, &bad_swr);
2004         if (ret) {
2005                 WARN_ONCE(ret, "failed to drain send queue: %d\n", ret);
2006                 return;
2007         }
2008
2009         wait_for_completion(&sdrain.done);
2010 }
2011
2012 /*
2013  * Post a WR and block until its completion is reaped for the RQ.
2014  */
2015 static void __ib_drain_rq(struct ib_qp *qp)
2016 {
2017         struct ib_qp_attr attr = { .qp_state = IB_QPS_ERR };
2018         struct ib_drain_cqe rdrain;
2019         struct ib_recv_wr rwr = {}, *bad_rwr;
2020         int ret;
2021
2022         if (qp->recv_cq->poll_ctx == IB_POLL_DIRECT) {
2023                 WARN_ONCE(qp->recv_cq->poll_ctx == IB_POLL_DIRECT,
2024                           "IB_POLL_DIRECT poll_ctx not supported for drain\n");
2025                 return;
2026         }
2027
2028         rwr.wr_cqe = &rdrain.cqe;
2029         rdrain.cqe.done = ib_drain_qp_done;
2030         init_completion(&rdrain.done);
2031
2032         ret = ib_modify_qp(qp, &attr, IB_QP_STATE);
2033         if (ret) {
2034                 WARN_ONCE(ret, "failed to drain recv queue: %d\n", ret);
2035                 return;
2036         }
2037
2038         ret = ib_post_recv(qp, &rwr, &bad_rwr);
2039         if (ret) {
2040                 WARN_ONCE(ret, "failed to drain recv queue: %d\n", ret);
2041                 return;
2042         }
2043
2044         wait_for_completion(&rdrain.done);
2045 }
2046
2047 /**
2048  * ib_drain_sq() - Block until all SQ CQEs have been consumed by the
2049  *                 application.
2050  * @qp:            queue pair to drain
2051  *
2052  * If the device has a provider-specific drain function, then
2053  * call that.  Otherwise call the generic drain function
2054  * __ib_drain_sq().
2055  *
2056  * The caller must:
2057  *
2058  * ensure there is room in the CQ and SQ for the drain work request and
2059  * completion.
2060  *
2061  * allocate the CQ using ib_alloc_cq() and the CQ poll context cannot be
2062  * IB_POLL_DIRECT.
2063  *
2064  * ensure that there are no other contexts that are posting WRs concurrently.
2065  * Otherwise the drain is not guaranteed.
2066  */
2067 void ib_drain_sq(struct ib_qp *qp)
2068 {
2069         if (qp->device->drain_sq)
2070                 qp->device->drain_sq(qp);
2071         else
2072                 __ib_drain_sq(qp);
2073 }
2074 EXPORT_SYMBOL(ib_drain_sq);
2075
2076 /**
2077  * ib_drain_rq() - Block until all RQ CQEs have been consumed by the
2078  *                 application.
2079  * @qp:            queue pair to drain
2080  *
2081  * If the device has a provider-specific drain function, then
2082  * call that.  Otherwise call the generic drain function
2083  * __ib_drain_rq().
2084  *
2085  * The caller must:
2086  *
2087  * ensure there is room in the CQ and RQ for the drain work request and
2088  * completion.
2089  *
2090  * allocate the CQ using ib_alloc_cq() and the CQ poll context cannot be
2091  * IB_POLL_DIRECT.
2092  *
2093  * ensure that there are no other contexts that are posting WRs concurrently.
2094  * Otherwise the drain is not guaranteed.
2095  */
2096 void ib_drain_rq(struct ib_qp *qp)
2097 {
2098         if (qp->device->drain_rq)
2099                 qp->device->drain_rq(qp);
2100         else
2101                 __ib_drain_rq(qp);
2102 }
2103 EXPORT_SYMBOL(ib_drain_rq);
2104
2105 /**
2106  * ib_drain_qp() - Block until all CQEs have been consumed by the
2107  *                 application on both the RQ and SQ.
2108  * @qp:            queue pair to drain
2109  *
2110  * The caller must:
2111  *
2112  * ensure there is room in the CQ(s), SQ, and RQ for drain work requests
2113  * and completions.
2114  *
2115  * allocate the CQs using ib_alloc_cq() and the CQ poll context cannot be
2116  * IB_POLL_DIRECT.
2117  *
2118  * ensure that there are no other contexts that are posting WRs concurrently.
2119  * Otherwise the drain is not guaranteed.
2120  */
2121 void ib_drain_qp(struct ib_qp *qp)
2122 {
2123         ib_drain_sq(qp);
2124         if (!qp->srq)
2125                 ib_drain_rq(qp);
2126 }
2127 EXPORT_SYMBOL(ib_drain_qp);