linux/drivers/infiniband/core/sa_query.c
<<
>>
Prefs
   1/*
   2 * Copyright (c) 2004 Topspin Communications.  All rights reserved.
   3 * Copyright (c) 2005 Voltaire, Inc.  All rights reserved.
   4 * Copyright (c) 2006 Intel Corporation.  All rights reserved.
   5 *
   6 * This software is available to you under a choice of one of two
   7 * licenses.  You may choose to be licensed under the terms of the GNU
   8 * General Public License (GPL) Version 2, available from the file
   9 * COPYING in the main directory of this source tree, or the
  10 * OpenIB.org BSD license below:
  11 *
  12 *     Redistribution and use in source and binary forms, with or
  13 *     without modification, are permitted provided that the following
  14 *     conditions are met:
  15 *
  16 *      - Redistributions of source code must retain the above
  17 *        copyright notice, this list of conditions and the following
  18 *        disclaimer.
  19 *
  20 *      - Redistributions in binary form must reproduce the above
  21 *        copyright notice, this list of conditions and the following
  22 *        disclaimer in the documentation and/or other materials
  23 *        provided with the distribution.
  24 *
  25 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
  26 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  27 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
  28 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
  29 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
  30 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
  31 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  32 * SOFTWARE.
  33 */
  34
  35#include <linux/module.h>
  36#include <linux/init.h>
  37#include <linux/err.h>
  38#include <linux/random.h>
  39#include <linux/spinlock.h>
  40#include <linux/slab.h>
  41#include <linux/dma-mapping.h>
  42#include <linux/kref.h>
  43#include <linux/idr.h>
  44#include <linux/workqueue.h>
  45#include <uapi/linux/if_ether.h>
  46#include <rdma/ib_pack.h>
  47#include <rdma/ib_cache.h>
  48#include <rdma/rdma_netlink.h>
  49#include <net/netlink.h>
  50#include <uapi/rdma/ib_user_sa.h>
  51#include <rdma/ib_marshall.h>
  52#include <rdma/ib_addr.h>
  53#include <rdma/opa_addr.h>
  54#include "sa.h"
  55#include "core_priv.h"
  56
  57#define IB_SA_LOCAL_SVC_TIMEOUT_MIN             100
  58#define IB_SA_LOCAL_SVC_TIMEOUT_DEFAULT         2000
  59#define IB_SA_LOCAL_SVC_TIMEOUT_MAX             200000
  60#define IB_SA_CPI_MAX_RETRY_CNT                 3
  61#define IB_SA_CPI_RETRY_WAIT                    1000 /*msecs */
  62static int sa_local_svc_timeout_ms = IB_SA_LOCAL_SVC_TIMEOUT_DEFAULT;
  63
  64struct ib_sa_sm_ah {
  65        struct ib_ah        *ah;
  66        struct kref          ref;
  67        u16                  pkey_index;
  68        u8                   src_path_mask;
  69};
  70
  71enum rdma_class_port_info_type {
  72        RDMA_CLASS_PORT_INFO_IB,
  73        RDMA_CLASS_PORT_INFO_OPA
  74};
  75
  76struct rdma_class_port_info {
  77        enum rdma_class_port_info_type type;
  78        union {
  79                struct ib_class_port_info ib;
  80                struct opa_class_port_info opa;
  81        };
  82};
  83
  84struct ib_sa_classport_cache {
  85        bool valid;
  86        int retry_cnt;
  87        struct rdma_class_port_info data;
  88};
  89
  90struct ib_sa_port {
  91        struct ib_mad_agent *agent;
  92        struct ib_sa_sm_ah  *sm_ah;
  93        struct work_struct   update_task;
  94        struct ib_sa_classport_cache classport_info;
  95        struct delayed_work ib_cpi_work;
  96        spinlock_t                   classport_lock; /* protects class port info set */
  97        spinlock_t           ah_lock;
  98        u8                   port_num;
  99};
 100
 101struct ib_sa_device {
 102        int                     start_port, end_port;
 103        struct ib_event_handler event_handler;
 104        struct ib_sa_port port[0];
 105};
 106
 107struct ib_sa_query {
 108        void (*callback)(struct ib_sa_query *, int, struct ib_sa_mad *);
 109        void (*release)(struct ib_sa_query *);
 110        struct ib_sa_client    *client;
 111        struct ib_sa_port      *port;
 112        struct ib_mad_send_buf *mad_buf;
 113        struct ib_sa_sm_ah     *sm_ah;
 114        int                     id;
 115        u32                     flags;
 116        struct list_head        list; /* Local svc request list */
 117        u32                     seq; /* Local svc request sequence number */
 118        unsigned long           timeout; /* Local svc timeout */
 119        u8                      path_use; /* How will the pathrecord be used */
 120};
 121
 122#define IB_SA_ENABLE_LOCAL_SERVICE      0x00000001
 123#define IB_SA_CANCEL                    0x00000002
 124#define IB_SA_QUERY_OPA                 0x00000004
 125
 126struct ib_sa_service_query {
 127        void (*callback)(int, struct ib_sa_service_rec *, void *);
 128        void *context;
 129        struct ib_sa_query sa_query;
 130};
 131
 132struct ib_sa_path_query {
 133        void (*callback)(int, struct sa_path_rec *, void *);
 134        void *context;
 135        struct ib_sa_query sa_query;
 136        struct sa_path_rec *conv_pr;
 137};
 138
 139struct ib_sa_guidinfo_query {
 140        void (*callback)(int, struct ib_sa_guidinfo_rec *, void *);
 141        void *context;
 142        struct ib_sa_query sa_query;
 143};
 144
 145struct ib_sa_classport_info_query {
 146        void (*callback)(void *);
 147        void *context;
 148        struct ib_sa_query sa_query;
 149};
 150
 151struct ib_sa_mcmember_query {
 152        void (*callback)(int, struct ib_sa_mcmember_rec *, void *);
 153        void *context;
 154        struct ib_sa_query sa_query;
 155};
 156
 157static LIST_HEAD(ib_nl_request_list);
 158static DEFINE_SPINLOCK(ib_nl_request_lock);
 159static atomic_t ib_nl_sa_request_seq;
 160static struct workqueue_struct *ib_nl_wq;
 161static struct delayed_work ib_nl_timed_work;
 162static const struct nla_policy ib_nl_policy[LS_NLA_TYPE_MAX] = {
 163        [LS_NLA_TYPE_PATH_RECORD]       = {.type = NLA_BINARY,
 164                .len = sizeof(struct ib_path_rec_data)},
 165        [LS_NLA_TYPE_TIMEOUT]           = {.type = NLA_U32},
 166        [LS_NLA_TYPE_SERVICE_ID]        = {.type = NLA_U64},
 167        [LS_NLA_TYPE_DGID]              = {.type = NLA_BINARY,
 168                .len = sizeof(struct rdma_nla_ls_gid)},
 169        [LS_NLA_TYPE_SGID]              = {.type = NLA_BINARY,
 170                .len = sizeof(struct rdma_nla_ls_gid)},
 171        [LS_NLA_TYPE_TCLASS]            = {.type = NLA_U8},
 172        [LS_NLA_TYPE_PKEY]              = {.type = NLA_U16},
 173        [LS_NLA_TYPE_QOS_CLASS]         = {.type = NLA_U16},
 174};
 175
 176
 177static void ib_sa_add_one(struct ib_device *device);
 178static void ib_sa_remove_one(struct ib_device *device, void *client_data);
 179
 180static struct ib_client sa_client = {
 181        .name   = "sa",
 182        .add    = ib_sa_add_one,
 183        .remove = ib_sa_remove_one
 184};
 185
 186static DEFINE_SPINLOCK(idr_lock);
 187static DEFINE_IDR(query_idr);
 188
 189static DEFINE_SPINLOCK(tid_lock);
 190static u32 tid;
 191
 192#define PATH_REC_FIELD(field) \
 193        .struct_offset_bytes = offsetof(struct sa_path_rec, field),     \
 194        .struct_size_bytes   = sizeof((struct sa_path_rec *)0)->field,  \
 195        .field_name          = "sa_path_rec:" #field
 196
 197static const struct ib_field path_rec_table[] = {
 198        { PATH_REC_FIELD(service_id),
 199          .offset_words = 0,
 200          .offset_bits  = 0,
 201          .size_bits    = 64 },
 202        { PATH_REC_FIELD(dgid),
 203          .offset_words = 2,
 204          .offset_bits  = 0,
 205          .size_bits    = 128 },
 206        { PATH_REC_FIELD(sgid),
 207          .offset_words = 6,
 208          .offset_bits  = 0,
 209          .size_bits    = 128 },
 210        { PATH_REC_FIELD(ib.dlid),
 211          .offset_words = 10,
 212          .offset_bits  = 0,
 213          .size_bits    = 16 },
 214        { PATH_REC_FIELD(ib.slid),
 215          .offset_words = 10,
 216          .offset_bits  = 16,
 217          .size_bits    = 16 },
 218        { PATH_REC_FIELD(ib.raw_traffic),
 219          .offset_words = 11,
 220          .offset_bits  = 0,
 221          .size_bits    = 1 },
 222        { RESERVED,
 223          .offset_words = 11,
 224          .offset_bits  = 1,
 225          .size_bits    = 3 },
 226        { PATH_REC_FIELD(flow_label),
 227          .offset_words = 11,
 228          .offset_bits  = 4,
 229          .size_bits    = 20 },
 230        { PATH_REC_FIELD(hop_limit),
 231          .offset_words = 11,
 232          .offset_bits  = 24,
 233          .size_bits    = 8 },
 234        { PATH_REC_FIELD(traffic_class),
 235          .offset_words = 12,
 236          .offset_bits  = 0,
 237          .size_bits    = 8 },
 238        { PATH_REC_FIELD(reversible),
 239          .offset_words = 12,
 240          .offset_bits  = 8,
 241          .size_bits    = 1 },
 242        { PATH_REC_FIELD(numb_path),
 243          .offset_words = 12,
 244          .offset_bits  = 9,
 245          .size_bits    = 7 },
 246        { PATH_REC_FIELD(pkey),
 247          .offset_words = 12,
 248          .offset_bits  = 16,
 249          .size_bits    = 16 },
 250        { PATH_REC_FIELD(qos_class),
 251          .offset_words = 13,
 252          .offset_bits  = 0,
 253          .size_bits    = 12 },
 254        { PATH_REC_FIELD(sl),
 255          .offset_words = 13,
 256          .offset_bits  = 12,
 257          .size_bits    = 4 },
 258        { PATH_REC_FIELD(mtu_selector),
 259          .offset_words = 13,
 260          .offset_bits  = 16,
 261          .size_bits    = 2 },
 262        { PATH_REC_FIELD(mtu),
 263          .offset_words = 13,
 264          .offset_bits  = 18,
 265          .size_bits    = 6 },
 266        { PATH_REC_FIELD(rate_selector),
 267          .offset_words = 13,
 268          .offset_bits  = 24,
 269          .size_bits    = 2 },
 270        { PATH_REC_FIELD(rate),
 271          .offset_words = 13,
 272          .offset_bits  = 26,
 273          .size_bits    = 6 },
 274        { PATH_REC_FIELD(packet_life_time_selector),
 275          .offset_words = 14,
 276          .offset_bits  = 0,
 277          .size_bits    = 2 },
 278        { PATH_REC_FIELD(packet_life_time),
 279          .offset_words = 14,
 280          .offset_bits  = 2,
 281          .size_bits    = 6 },
 282        { PATH_REC_FIELD(preference),
 283          .offset_words = 14,
 284          .offset_bits  = 8,
 285          .size_bits    = 8 },
 286        { RESERVED,
 287          .offset_words = 14,
 288          .offset_bits  = 16,
 289          .size_bits    = 48 },
 290};
 291
 292#define OPA_PATH_REC_FIELD(field) \
 293        .struct_offset_bytes = \
 294                offsetof(struct sa_path_rec, field), \
 295        .struct_size_bytes   = \
 296                sizeof((struct sa_path_rec *)0)->field, \
 297        .field_name          = "sa_path_rec:" #field
 298
 299static const struct ib_field opa_path_rec_table[] = {
 300        { OPA_PATH_REC_FIELD(service_id),
 301          .offset_words = 0,
 302          .offset_bits  = 0,
 303          .size_bits    = 64 },
 304        { OPA_PATH_REC_FIELD(dgid),
 305          .offset_words = 2,
 306          .offset_bits  = 0,
 307          .size_bits    = 128 },
 308        { OPA_PATH_REC_FIELD(sgid),
 309          .offset_words = 6,
 310          .offset_bits  = 0,
 311          .size_bits    = 128 },
 312        { OPA_PATH_REC_FIELD(opa.dlid),
 313          .offset_words = 10,
 314          .offset_bits  = 0,
 315          .size_bits    = 32 },
 316        { OPA_PATH_REC_FIELD(opa.slid),
 317          .offset_words = 11,
 318          .offset_bits  = 0,
 319          .size_bits    = 32 },
 320        { OPA_PATH_REC_FIELD(opa.raw_traffic),
 321          .offset_words = 12,
 322          .offset_bits  = 0,
 323          .size_bits    = 1 },
 324        { RESERVED,
 325          .offset_words = 12,
 326          .offset_bits  = 1,
 327          .size_bits    = 3 },
 328        { OPA_PATH_REC_FIELD(flow_label),
 329          .offset_words = 12,
 330          .offset_bits  = 4,
 331          .size_bits    = 20 },
 332        { OPA_PATH_REC_FIELD(hop_limit),
 333          .offset_words = 12,
 334          .offset_bits  = 24,
 335          .size_bits    = 8 },
 336        { OPA_PATH_REC_FIELD(traffic_class),
 337          .offset_words = 13,
 338          .offset_bits  = 0,
 339          .size_bits    = 8 },
 340        { OPA_PATH_REC_FIELD(reversible),
 341          .offset_words = 13,
 342          .offset_bits  = 8,
 343          .size_bits    = 1 },
 344        { OPA_PATH_REC_FIELD(numb_path),
 345          .offset_words = 13,
 346          .offset_bits  = 9,
 347          .size_bits    = 7 },
 348        { OPA_PATH_REC_FIELD(pkey),
 349          .offset_words = 13,
 350          .offset_bits  = 16,
 351          .size_bits    = 16 },
 352        { OPA_PATH_REC_FIELD(opa.l2_8B),
 353          .offset_words = 14,
 354          .offset_bits  = 0,
 355          .size_bits    = 1 },
 356        { OPA_PATH_REC_FIELD(opa.l2_10B),
 357          .offset_words = 14,
 358          .offset_bits  = 1,
 359          .size_bits    = 1 },
 360        { OPA_PATH_REC_FIELD(opa.l2_9B),
 361          .offset_words = 14,
 362          .offset_bits  = 2,
 363          .size_bits    = 1 },
 364        { OPA_PATH_REC_FIELD(opa.l2_16B),
 365          .offset_words = 14,
 366          .offset_bits  = 3,
 367          .size_bits    = 1 },
 368        { RESERVED,
 369          .offset_words = 14,
 370          .offset_bits  = 4,
 371          .size_bits    = 2 },
 372        { OPA_PATH_REC_FIELD(opa.qos_type),
 373          .offset_words = 14,
 374          .offset_bits  = 6,
 375          .size_bits    = 2 },
 376        { OPA_PATH_REC_FIELD(opa.qos_priority),
 377          .offset_words = 14,
 378          .offset_bits  = 8,
 379          .size_bits    = 8 },
 380        { RESERVED,
 381          .offset_words = 14,
 382          .offset_bits  = 16,
 383          .size_bits    = 3 },
 384        { OPA_PATH_REC_FIELD(sl),
 385          .offset_words = 14,
 386          .offset_bits  = 19,
 387          .size_bits    = 5 },
 388        { RESERVED,
 389          .offset_words = 14,
 390          .offset_bits  = 24,
 391          .size_bits    = 8 },
 392        { OPA_PATH_REC_FIELD(mtu_selector),
 393          .offset_words = 15,
 394          .offset_bits  = 0,
 395          .size_bits    = 2 },
 396        { OPA_PATH_REC_FIELD(mtu),
 397          .offset_words = 15,
 398          .offset_bits  = 2,
 399          .size_bits    = 6 },
 400        { OPA_PATH_REC_FIELD(rate_selector),
 401          .offset_words = 15,
 402          .offset_bits  = 8,
 403          .size_bits    = 2 },
 404        { OPA_PATH_REC_FIELD(rate),
 405          .offset_words = 15,
 406          .offset_bits  = 10,
 407          .size_bits    = 6 },
 408        { OPA_PATH_REC_FIELD(packet_life_time_selector),
 409          .offset_words = 15,
 410          .offset_bits  = 16,
 411          .size_bits    = 2 },
 412        { OPA_PATH_REC_FIELD(packet_life_time),
 413          .offset_words = 15,
 414          .offset_bits  = 18,
 415          .size_bits    = 6 },
 416        { OPA_PATH_REC_FIELD(preference),
 417          .offset_words = 15,
 418          .offset_bits  = 24,
 419          .size_bits    = 8 },
 420};
 421
 422#define MCMEMBER_REC_FIELD(field) \
 423        .struct_offset_bytes = offsetof(struct ib_sa_mcmember_rec, field),      \
 424        .struct_size_bytes   = sizeof ((struct ib_sa_mcmember_rec *) 0)->field, \
 425        .field_name          = "sa_mcmember_rec:" #field
 426
 427static const struct ib_field mcmember_rec_table[] = {
 428        { MCMEMBER_REC_FIELD(mgid),
 429          .offset_words = 0,
 430          .offset_bits  = 0,
 431          .size_bits    = 128 },
 432        { MCMEMBER_REC_FIELD(port_gid),
 433          .offset_words = 4,
 434          .offset_bits  = 0,
 435          .size_bits    = 128 },
 436        { MCMEMBER_REC_FIELD(qkey),
 437          .offset_words = 8,
 438          .offset_bits  = 0,
 439          .size_bits    = 32 },
 440        { MCMEMBER_REC_FIELD(mlid),
 441          .offset_words = 9,
 442          .offset_bits  = 0,
 443          .size_bits    = 16 },
 444        { MCMEMBER_REC_FIELD(mtu_selector),
 445          .offset_words = 9,
 446          .offset_bits  = 16,
 447          .size_bits    = 2 },
 448        { MCMEMBER_REC_FIELD(mtu),
 449          .offset_words = 9,
 450          .offset_bits  = 18,
 451          .size_bits    = 6 },
 452        { MCMEMBER_REC_FIELD(traffic_class),
 453          .offset_words = 9,
 454          .offset_bits  = 24,
 455          .size_bits    = 8 },
 456        { MCMEMBER_REC_FIELD(pkey),
 457          .offset_words = 10,
 458          .offset_bits  = 0,
 459          .size_bits    = 16 },
 460        { MCMEMBER_REC_FIELD(rate_selector),
 461          .offset_words = 10,
 462          .offset_bits  = 16,
 463          .size_bits    = 2 },
 464        { MCMEMBER_REC_FIELD(rate),
 465          .offset_words = 10,
 466          .offset_bits  = 18,
 467          .size_bits    = 6 },
 468        { MCMEMBER_REC_FIELD(packet_life_time_selector),
 469          .offset_words = 10,
 470          .offset_bits  = 24,
 471          .size_bits    = 2 },
 472        { MCMEMBER_REC_FIELD(packet_life_time),
 473          .offset_words = 10,
 474          .offset_bits  = 26,
 475          .size_bits    = 6 },
 476        { MCMEMBER_REC_FIELD(sl),
 477          .offset_words = 11,
 478          .offset_bits  = 0,
 479          .size_bits    = 4 },
 480        { MCMEMBER_REC_FIELD(flow_label),
 481          .offset_words = 11,
 482          .offset_bits  = 4,
 483          .size_bits    = 20 },
 484        { MCMEMBER_REC_FIELD(hop_limit),
 485          .offset_words = 11,
 486          .offset_bits  = 24,
 487          .size_bits    = 8 },
 488        { MCMEMBER_REC_FIELD(scope),
 489          .offset_words = 12,
 490          .offset_bits  = 0,
 491          .size_bits    = 4 },
 492        { MCMEMBER_REC_FIELD(join_state),
 493          .offset_words = 12,
 494          .offset_bits  = 4,
 495          .size_bits    = 4 },
 496        { MCMEMBER_REC_FIELD(proxy_join),
 497          .offset_words = 12,
 498          .offset_bits  = 8,
 499          .size_bits    = 1 },
 500        { RESERVED,
 501          .offset_words = 12,
 502          .offset_bits  = 9,
 503          .size_bits    = 23 },
 504};
 505
 506#define SERVICE_REC_FIELD(field) \
 507        .struct_offset_bytes = offsetof(struct ib_sa_service_rec, field),       \
 508        .struct_size_bytes   = sizeof ((struct ib_sa_service_rec *) 0)->field,  \
 509        .field_name          = "sa_service_rec:" #field
 510
 511static const struct ib_field service_rec_table[] = {
 512        { SERVICE_REC_FIELD(id),
 513          .offset_words = 0,
 514          .offset_bits  = 0,
 515          .size_bits    = 64 },
 516        { SERVICE_REC_FIELD(gid),
 517          .offset_words = 2,
 518          .offset_bits  = 0,
 519          .size_bits    = 128 },
 520        { SERVICE_REC_FIELD(pkey),
 521          .offset_words = 6,
 522          .offset_bits  = 0,
 523          .size_bits    = 16 },
 524        { SERVICE_REC_FIELD(lease),
 525          .offset_words = 7,
 526          .offset_bits  = 0,
 527          .size_bits    = 32 },
 528        { SERVICE_REC_FIELD(key),
 529          .offset_words = 8,
 530          .offset_bits  = 0,
 531          .size_bits    = 128 },
 532        { SERVICE_REC_FIELD(name),
 533          .offset_words = 12,
 534          .offset_bits  = 0,
 535          .size_bits    = 64*8 },
 536        { SERVICE_REC_FIELD(data8),
 537          .offset_words = 28,
 538          .offset_bits  = 0,
 539          .size_bits    = 16*8 },
 540        { SERVICE_REC_FIELD(data16),
 541          .offset_words = 32,
 542          .offset_bits  = 0,
 543          .size_bits    = 8*16 },
 544        { SERVICE_REC_FIELD(data32),
 545          .offset_words = 36,
 546          .offset_bits  = 0,
 547          .size_bits    = 4*32 },
 548        { SERVICE_REC_FIELD(data64),
 549          .offset_words = 40,
 550          .offset_bits  = 0,
 551          .size_bits    = 2*64 },
 552};
 553
 554#define CLASSPORTINFO_REC_FIELD(field) \
 555        .struct_offset_bytes = offsetof(struct ib_class_port_info, field),      \
 556        .struct_size_bytes   = sizeof((struct ib_class_port_info *)0)->field,   \
 557        .field_name          = "ib_class_port_info:" #field
 558
 559static const struct ib_field ib_classport_info_rec_table[] = {
 560        { CLASSPORTINFO_REC_FIELD(base_version),
 561          .offset_words = 0,
 562          .offset_bits  = 0,
 563          .size_bits    = 8 },
 564        { CLASSPORTINFO_REC_FIELD(class_version),
 565          .offset_words = 0,
 566          .offset_bits  = 8,
 567          .size_bits    = 8 },
 568        { CLASSPORTINFO_REC_FIELD(capability_mask),
 569          .offset_words = 0,
 570          .offset_bits  = 16,
 571          .size_bits    = 16 },
 572        { CLASSPORTINFO_REC_FIELD(cap_mask2_resp_time),
 573          .offset_words = 1,
 574          .offset_bits  = 0,
 575          .size_bits    = 32 },
 576        { CLASSPORTINFO_REC_FIELD(redirect_gid),
 577          .offset_words = 2,
 578          .offset_bits  = 0,
 579          .size_bits    = 128 },
 580        { CLASSPORTINFO_REC_FIELD(redirect_tcslfl),
 581          .offset_words = 6,
 582          .offset_bits  = 0,
 583          .size_bits    = 32 },
 584        { CLASSPORTINFO_REC_FIELD(redirect_lid),
 585          .offset_words = 7,
 586          .offset_bits  = 0,
 587          .size_bits    = 16 },
 588        { CLASSPORTINFO_REC_FIELD(redirect_pkey),
 589          .offset_words = 7,
 590          .offset_bits  = 16,
 591          .size_bits    = 16 },
 592
 593        { CLASSPORTINFO_REC_FIELD(redirect_qp),
 594          .offset_words = 8,
 595          .offset_bits  = 0,
 596          .size_bits    = 32 },
 597        { CLASSPORTINFO_REC_FIELD(redirect_qkey),
 598          .offset_words = 9,
 599          .offset_bits  = 0,
 600          .size_bits    = 32 },
 601
 602        { CLASSPORTINFO_REC_FIELD(trap_gid),
 603          .offset_words = 10,
 604          .offset_bits  = 0,
 605          .size_bits    = 128 },
 606        { CLASSPORTINFO_REC_FIELD(trap_tcslfl),
 607          .offset_words = 14,
 608          .offset_bits  = 0,
 609          .size_bits    = 32 },
 610
 611        { CLASSPORTINFO_REC_FIELD(trap_lid),
 612          .offset_words = 15,
 613          .offset_bits  = 0,
 614          .size_bits    = 16 },
 615        { CLASSPORTINFO_REC_FIELD(trap_pkey),
 616          .offset_words = 15,
 617          .offset_bits  = 16,
 618          .size_bits    = 16 },
 619
 620        { CLASSPORTINFO_REC_FIELD(trap_hlqp),
 621          .offset_words = 16,
 622          .offset_bits  = 0,
 623          .size_bits    = 32 },
 624        { CLASSPORTINFO_REC_FIELD(trap_qkey),
 625          .offset_words = 17,
 626          .offset_bits  = 0,
 627          .size_bits    = 32 },
 628};
 629
 630#define OPA_CLASSPORTINFO_REC_FIELD(field) \
 631        .struct_offset_bytes =\
 632                offsetof(struct opa_class_port_info, field),    \
 633        .struct_size_bytes   = \
 634                sizeof((struct opa_class_port_info *)0)->field, \
 635        .field_name          = "opa_class_port_info:" #field
 636
 637static const struct ib_field opa_classport_info_rec_table[] = {
 638        { OPA_CLASSPORTINFO_REC_FIELD(base_version),
 639          .offset_words = 0,
 640          .offset_bits  = 0,
 641          .size_bits    = 8 },
 642        { OPA_CLASSPORTINFO_REC_FIELD(class_version),
 643          .offset_words = 0,
 644          .offset_bits  = 8,
 645          .size_bits    = 8 },
 646        { OPA_CLASSPORTINFO_REC_FIELD(cap_mask),
 647          .offset_words = 0,
 648          .offset_bits  = 16,
 649          .size_bits    = 16 },
 650        { OPA_CLASSPORTINFO_REC_FIELD(cap_mask2_resp_time),
 651          .offset_words = 1,
 652          .offset_bits  = 0,
 653          .size_bits    = 32 },
 654        { OPA_CLASSPORTINFO_REC_FIELD(redirect_gid),
 655          .offset_words = 2,
 656          .offset_bits  = 0,
 657          .size_bits    = 128 },
 658        { OPA_CLASSPORTINFO_REC_FIELD(redirect_tc_fl),
 659          .offset_words = 6,
 660          .offset_bits  = 0,
 661          .size_bits    = 32 },
 662        { OPA_CLASSPORTINFO_REC_FIELD(redirect_lid),
 663          .offset_words = 7,
 664          .offset_bits  = 0,
 665          .size_bits    = 32 },
 666        { OPA_CLASSPORTINFO_REC_FIELD(redirect_sl_qp),
 667          .offset_words = 8,
 668          .offset_bits  = 0,
 669          .size_bits    = 32 },
 670        { OPA_CLASSPORTINFO_REC_FIELD(redirect_qkey),
 671          .offset_words = 9,
 672          .offset_bits  = 0,
 673          .size_bits    = 32 },
 674        { OPA_CLASSPORTINFO_REC_FIELD(trap_gid),
 675          .offset_words = 10,
 676          .offset_bits  = 0,
 677          .size_bits    = 128 },
 678        { OPA_CLASSPORTINFO_REC_FIELD(trap_tc_fl),
 679          .offset_words = 14,
 680          .offset_bits  = 0,
 681          .size_bits    = 32 },
 682        { OPA_CLASSPORTINFO_REC_FIELD(trap_lid),
 683          .offset_words = 15,
 684          .offset_bits  = 0,
 685          .size_bits    = 32 },
 686        { OPA_CLASSPORTINFO_REC_FIELD(trap_hl_qp),
 687          .offset_words = 16,
 688          .offset_bits  = 0,
 689          .size_bits    = 32 },
 690        { OPA_CLASSPORTINFO_REC_FIELD(trap_qkey),
 691          .offset_words = 17,
 692          .offset_bits  = 0,
 693          .size_bits    = 32 },
 694        { OPA_CLASSPORTINFO_REC_FIELD(trap_pkey),
 695          .offset_words = 18,
 696          .offset_bits  = 0,
 697          .size_bits    = 16 },
 698        { OPA_CLASSPORTINFO_REC_FIELD(redirect_pkey),
 699          .offset_words = 18,
 700          .offset_bits  = 16,
 701          .size_bits    = 16 },
 702        { OPA_CLASSPORTINFO_REC_FIELD(trap_sl_rsvd),
 703          .offset_words = 19,
 704          .offset_bits  = 0,
 705          .size_bits    = 8 },
 706        { RESERVED,
 707          .offset_words = 19,
 708          .offset_bits  = 8,
 709          .size_bits    = 24 },
 710};
 711
 712#define GUIDINFO_REC_FIELD(field) \
 713        .struct_offset_bytes = offsetof(struct ib_sa_guidinfo_rec, field),      \
 714        .struct_size_bytes   = sizeof((struct ib_sa_guidinfo_rec *) 0)->field,  \
 715        .field_name          = "sa_guidinfo_rec:" #field
 716
 717static const struct ib_field guidinfo_rec_table[] = {
 718        { GUIDINFO_REC_FIELD(lid),
 719          .offset_words = 0,
 720          .offset_bits  = 0,
 721          .size_bits    = 16 },
 722        { GUIDINFO_REC_FIELD(block_num),
 723          .offset_words = 0,
 724          .offset_bits  = 16,
 725          .size_bits    = 8 },
 726        { GUIDINFO_REC_FIELD(res1),
 727          .offset_words = 0,
 728          .offset_bits  = 24,
 729          .size_bits    = 8 },
 730        { GUIDINFO_REC_FIELD(res2),
 731          .offset_words = 1,
 732          .offset_bits  = 0,
 733          .size_bits    = 32 },
 734        { GUIDINFO_REC_FIELD(guid_info_list),
 735          .offset_words = 2,
 736          .offset_bits  = 0,
 737          .size_bits    = 512 },
 738};
 739
 740static inline void ib_sa_disable_local_svc(struct ib_sa_query *query)
 741{
 742        query->flags &= ~IB_SA_ENABLE_LOCAL_SERVICE;
 743}
 744
 745static inline int ib_sa_query_cancelled(struct ib_sa_query *query)
 746{
 747        return (query->flags & IB_SA_CANCEL);
 748}
 749
 750static void ib_nl_set_path_rec_attrs(struct sk_buff *skb,
 751                                     struct ib_sa_query *query)
 752{
 753        struct sa_path_rec *sa_rec = query->mad_buf->context[1];
 754        struct ib_sa_mad *mad = query->mad_buf->mad;
 755        ib_sa_comp_mask comp_mask = mad->sa_hdr.comp_mask;
 756        u16 val16;
 757        u64 val64;
 758        struct rdma_ls_resolve_header *header;
 759
 760        query->mad_buf->context[1] = NULL;
 761
 762        /* Construct the family header first */
 763        header = skb_put(skb, NLMSG_ALIGN(sizeof(*header)));
 764        memcpy(header->device_name, query->port->agent->device->name,
 765               LS_DEVICE_NAME_MAX);
 766        header->port_num = query->port->port_num;
 767
 768        if ((comp_mask & IB_SA_PATH_REC_REVERSIBLE) &&
 769            sa_rec->reversible != 0)
 770                query->path_use = LS_RESOLVE_PATH_USE_GMP;
 771        else
 772                query->path_use = LS_RESOLVE_PATH_USE_UNIDIRECTIONAL;
 773        header->path_use = query->path_use;
 774
 775        /* Now build the attributes */
 776        if (comp_mask & IB_SA_PATH_REC_SERVICE_ID) {
 777                val64 = be64_to_cpu(sa_rec->service_id);
 778                nla_put(skb, RDMA_NLA_F_MANDATORY | LS_NLA_TYPE_SERVICE_ID,
 779                        sizeof(val64), &val64);
 780        }
 781        if (comp_mask & IB_SA_PATH_REC_DGID)
 782                nla_put(skb, RDMA_NLA_F_MANDATORY | LS_NLA_TYPE_DGID,
 783                        sizeof(sa_rec->dgid), &sa_rec->dgid);
 784        if (comp_mask & IB_SA_PATH_REC_SGID)
 785                nla_put(skb, RDMA_NLA_F_MANDATORY | LS_NLA_TYPE_SGID,
 786                        sizeof(sa_rec->sgid), &sa_rec->sgid);
 787        if (comp_mask & IB_SA_PATH_REC_TRAFFIC_CLASS)
 788                nla_put(skb, RDMA_NLA_F_MANDATORY | LS_NLA_TYPE_TCLASS,
 789                        sizeof(sa_rec->traffic_class), &sa_rec->traffic_class);
 790
 791        if (comp_mask & IB_SA_PATH_REC_PKEY) {
 792                val16 = be16_to_cpu(sa_rec->pkey);
 793                nla_put(skb, RDMA_NLA_F_MANDATORY | LS_NLA_TYPE_PKEY,
 794                        sizeof(val16), &val16);
 795        }
 796        if (comp_mask & IB_SA_PATH_REC_QOS_CLASS) {
 797                val16 = be16_to_cpu(sa_rec->qos_class);
 798                nla_put(skb, RDMA_NLA_F_MANDATORY | LS_NLA_TYPE_QOS_CLASS,
 799                        sizeof(val16), &val16);
 800        }
 801}
 802
 803static int ib_nl_get_path_rec_attrs_len(ib_sa_comp_mask comp_mask)
 804{
 805        int len = 0;
 806
 807        if (comp_mask & IB_SA_PATH_REC_SERVICE_ID)
 808                len += nla_total_size(sizeof(u64));
 809        if (comp_mask & IB_SA_PATH_REC_DGID)
 810                len += nla_total_size(sizeof(struct rdma_nla_ls_gid));
 811        if (comp_mask & IB_SA_PATH_REC_SGID)
 812                len += nla_total_size(sizeof(struct rdma_nla_ls_gid));
 813        if (comp_mask & IB_SA_PATH_REC_TRAFFIC_CLASS)
 814                len += nla_total_size(sizeof(u8));
 815        if (comp_mask & IB_SA_PATH_REC_PKEY)
 816                len += nla_total_size(sizeof(u16));
 817        if (comp_mask & IB_SA_PATH_REC_QOS_CLASS)
 818                len += nla_total_size(sizeof(u16));
 819
 820        /*
 821         * Make sure that at least some of the required comp_mask bits are
 822         * set.
 823         */
 824        if (WARN_ON(len == 0))
 825                return len;
 826
 827        /* Add the family header */
 828        len += NLMSG_ALIGN(sizeof(struct rdma_ls_resolve_header));
 829
 830        return len;
 831}
 832
 833static int ib_nl_send_msg(struct ib_sa_query *query, gfp_t gfp_mask)
 834{
 835        struct sk_buff *skb = NULL;
 836        struct nlmsghdr *nlh;
 837        void *data;
 838        int ret = 0;
 839        struct ib_sa_mad *mad;
 840        int len;
 841
 842        mad = query->mad_buf->mad;
 843        len = ib_nl_get_path_rec_attrs_len(mad->sa_hdr.comp_mask);
 844        if (len <= 0)
 845                return -EMSGSIZE;
 846
 847        skb = nlmsg_new(len, gfp_mask);
 848        if (!skb)
 849                return -ENOMEM;
 850
 851        /* Put nlmsg header only for now */
 852        data = ibnl_put_msg(skb, &nlh, query->seq, 0, RDMA_NL_LS,
 853                            RDMA_NL_LS_OP_RESOLVE, NLM_F_REQUEST);
 854        if (!data) {
 855                nlmsg_free(skb);
 856                return -EMSGSIZE;
 857        }
 858
 859        /* Add attributes */
 860        ib_nl_set_path_rec_attrs(skb, query);
 861
 862        /* Repair the nlmsg header length */
 863        nlmsg_end(skb, nlh);
 864
 865        ret = rdma_nl_multicast(skb, RDMA_NL_GROUP_LS, gfp_mask);
 866        if (!ret)
 867                ret = len;
 868        else
 869                ret = 0;
 870
 871        return ret;
 872}
 873
 874static int ib_nl_make_request(struct ib_sa_query *query, gfp_t gfp_mask)
 875{
 876        unsigned long flags;
 877        unsigned long delay;
 878        int ret;
 879
 880        INIT_LIST_HEAD(&query->list);
 881        query->seq = (u32)atomic_inc_return(&ib_nl_sa_request_seq);
 882
 883        /* Put the request on the list first.*/
 884        spin_lock_irqsave(&ib_nl_request_lock, flags);
 885        delay = msecs_to_jiffies(sa_local_svc_timeout_ms);
 886        query->timeout = delay + jiffies;
 887        list_add_tail(&query->list, &ib_nl_request_list);
 888        /* Start the timeout if this is the only request */
 889        if (ib_nl_request_list.next == &query->list)
 890                queue_delayed_work(ib_nl_wq, &ib_nl_timed_work, delay);
 891        spin_unlock_irqrestore(&ib_nl_request_lock, flags);
 892
 893        ret = ib_nl_send_msg(query, gfp_mask);
 894        if (ret <= 0) {
 895                ret = -EIO;
 896                /* Remove the request */
 897                spin_lock_irqsave(&ib_nl_request_lock, flags);
 898                list_del(&query->list);
 899                spin_unlock_irqrestore(&ib_nl_request_lock, flags);
 900        } else {
 901                ret = 0;
 902        }
 903
 904        return ret;
 905}
 906
 907static int ib_nl_cancel_request(struct ib_sa_query *query)
 908{
 909        unsigned long flags;
 910        struct ib_sa_query *wait_query;
 911        int found = 0;
 912
 913        spin_lock_irqsave(&ib_nl_request_lock, flags);
 914        list_for_each_entry(wait_query, &ib_nl_request_list, list) {
 915                /* Let the timeout to take care of the callback */
 916                if (query == wait_query) {
 917                        query->flags |= IB_SA_CANCEL;
 918                        query->timeout = jiffies;
 919                        list_move(&query->list, &ib_nl_request_list);
 920                        found = 1;
 921                        mod_delayed_work(ib_nl_wq, &ib_nl_timed_work, 1);
 922                        break;
 923                }
 924        }
 925        spin_unlock_irqrestore(&ib_nl_request_lock, flags);
 926
 927        return found;
 928}
 929
 930static void send_handler(struct ib_mad_agent *agent,
 931                         struct ib_mad_send_wc *mad_send_wc);
 932
 933static void ib_nl_process_good_resolve_rsp(struct ib_sa_query *query,
 934                                           const struct nlmsghdr *nlh)
 935{
 936        struct ib_mad_send_wc mad_send_wc;
 937        struct ib_sa_mad *mad = NULL;
 938        const struct nlattr *head, *curr;
 939        struct ib_path_rec_data  *rec;
 940        int len, rem;
 941        u32 mask = 0;
 942        int status = -EIO;
 943
 944        if (query->callback) {
 945                head = (const struct nlattr *) nlmsg_data(nlh);
 946                len = nlmsg_len(nlh);
 947                switch (query->path_use) {
 948                case LS_RESOLVE_PATH_USE_UNIDIRECTIONAL:
 949                        mask = IB_PATH_PRIMARY | IB_PATH_OUTBOUND;
 950                        break;
 951
 952                case LS_RESOLVE_PATH_USE_ALL:
 953                case LS_RESOLVE_PATH_USE_GMP:
 954                default:
 955                        mask = IB_PATH_PRIMARY | IB_PATH_GMP |
 956                                IB_PATH_BIDIRECTIONAL;
 957                        break;
 958                }
 959                nla_for_each_attr(curr, head, len, rem) {
 960                        if (curr->nla_type == LS_NLA_TYPE_PATH_RECORD) {
 961                                rec = nla_data(curr);
 962                                /*
 963                                 * Get the first one. In the future, we may
 964                                 * need to get up to 6 pathrecords.
 965                                 */
 966                                if ((rec->flags & mask) == mask) {
 967                                        mad = query->mad_buf->mad;
 968                                        mad->mad_hdr.method |=
 969                                                IB_MGMT_METHOD_RESP;
 970                                        memcpy(mad->data, rec->path_rec,
 971                                               sizeof(rec->path_rec));
 972                                        status = 0;
 973                                        break;
 974                                }
 975                        }
 976                }
 977                query->callback(query, status, mad);
 978        }
 979
 980        mad_send_wc.send_buf = query->mad_buf;
 981        mad_send_wc.status = IB_WC_SUCCESS;
 982        send_handler(query->mad_buf->mad_agent, &mad_send_wc);
 983}
 984
 985static void ib_nl_request_timeout(struct work_struct *work)
 986{
 987        unsigned long flags;
 988        struct ib_sa_query *query;
 989        unsigned long delay;
 990        struct ib_mad_send_wc mad_send_wc;
 991        int ret;
 992
 993        spin_lock_irqsave(&ib_nl_request_lock, flags);
 994        while (!list_empty(&ib_nl_request_list)) {
 995                query = list_entry(ib_nl_request_list.next,
 996                                   struct ib_sa_query, list);
 997
 998                if (time_after(query->timeout, jiffies)) {
 999                        delay = query->timeout - jiffies;
1000                        if ((long)delay <= 0)
1001                                delay = 1;
1002                        queue_delayed_work(ib_nl_wq, &ib_nl_timed_work, delay);
1003                        break;
1004                }
1005
1006                list_del(&query->list);
1007                ib_sa_disable_local_svc(query);
1008                /* Hold the lock to protect against query cancellation */
1009                if (ib_sa_query_cancelled(query))
1010                        ret = -1;
1011                else
1012                        ret = ib_post_send_mad(query->mad_buf, NULL);
1013                if (ret) {
1014                        mad_send_wc.send_buf = query->mad_buf;
1015                        mad_send_wc.status = IB_WC_WR_FLUSH_ERR;
1016                        spin_unlock_irqrestore(&ib_nl_request_lock, flags);
1017                        send_handler(query->port->agent, &mad_send_wc);
1018                        spin_lock_irqsave(&ib_nl_request_lock, flags);
1019                }
1020        }
1021        spin_unlock_irqrestore(&ib_nl_request_lock, flags);
1022}
1023
1024int ib_nl_handle_set_timeout(struct sk_buff *skb,
1025                             struct nlmsghdr *nlh,
1026                             struct netlink_ext_ack *extack)
1027{
1028        int timeout, delta, abs_delta;
1029        const struct nlattr *attr;
1030        unsigned long flags;
1031        struct ib_sa_query *query;
1032        long delay = 0;
1033        struct nlattr *tb[LS_NLA_TYPE_MAX];
1034        int ret;
1035
1036        if (!(nlh->nlmsg_flags & NLM_F_REQUEST) ||
1037            !(NETLINK_CB(skb).sk))
1038                return -EPERM;
1039
1040        ret = nla_parse(tb, LS_NLA_TYPE_MAX - 1, nlmsg_data(nlh),
1041                        nlmsg_len(nlh), ib_nl_policy, NULL);
1042        attr = (const struct nlattr *)tb[LS_NLA_TYPE_TIMEOUT];
1043        if (ret || !attr)
1044                goto settimeout_out;
1045
1046        timeout = *(int *) nla_data(attr);
1047        if (timeout < IB_SA_LOCAL_SVC_TIMEOUT_MIN)
1048                timeout = IB_SA_LOCAL_SVC_TIMEOUT_MIN;
1049        if (timeout > IB_SA_LOCAL_SVC_TIMEOUT_MAX)
1050                timeout = IB_SA_LOCAL_SVC_TIMEOUT_MAX;
1051
1052        delta = timeout - sa_local_svc_timeout_ms;
1053        if (delta < 0)
1054                abs_delta = -delta;
1055        else
1056                abs_delta = delta;
1057
1058        if (delta != 0) {
1059                spin_lock_irqsave(&ib_nl_request_lock, flags);
1060                sa_local_svc_timeout_ms = timeout;
1061                list_for_each_entry(query, &ib_nl_request_list, list) {
1062                        if (delta < 0 && abs_delta > query->timeout)
1063                                query->timeout = 0;
1064                        else
1065                                query->timeout += delta;
1066
1067                        /* Get the new delay from the first entry */
1068                        if (!delay) {
1069                                delay = query->timeout - jiffies;
1070                                if (delay <= 0)
1071                                        delay = 1;
1072                        }
1073                }
1074                if (delay)
1075                        mod_delayed_work(ib_nl_wq, &ib_nl_timed_work,
1076                                         (unsigned long)delay);
1077                spin_unlock_irqrestore(&ib_nl_request_lock, flags);
1078        }
1079
1080settimeout_out:
1081        return skb->len;
1082}
1083
1084static inline int ib_nl_is_good_resolve_resp(const struct nlmsghdr *nlh)
1085{
1086        struct nlattr *tb[LS_NLA_TYPE_MAX];
1087        int ret;
1088
1089        if (nlh->nlmsg_flags & RDMA_NL_LS_F_ERR)
1090                return 0;
1091
1092        ret = nla_parse(tb, LS_NLA_TYPE_MAX - 1, nlmsg_data(nlh),
1093                        nlmsg_len(nlh), ib_nl_policy, NULL);
1094        if (ret)
1095                return 0;
1096
1097        return 1;
1098}
1099
1100int ib_nl_handle_resolve_resp(struct sk_buff *skb,
1101                              struct nlmsghdr *nlh,
1102                              struct netlink_ext_ack *extack)
1103{
1104        unsigned long flags;
1105        struct ib_sa_query *query;
1106        struct ib_mad_send_buf *send_buf;
1107        struct ib_mad_send_wc mad_send_wc;
1108        int found = 0;
1109        int ret;
1110
1111        if ((nlh->nlmsg_flags & NLM_F_REQUEST) ||
1112            !(NETLINK_CB(skb).sk))
1113                return -EPERM;
1114
1115        spin_lock_irqsave(&ib_nl_request_lock, flags);
1116        list_for_each_entry(query, &ib_nl_request_list, list) {
1117                /*
1118                 * If the query is cancelled, let the timeout routine
1119                 * take care of it.
1120                 */
1121                if (nlh->nlmsg_seq == query->seq) {
1122                        found = !ib_sa_query_cancelled(query);
1123                        if (found)
1124                                list_del(&query->list);
1125                        break;
1126                }
1127        }
1128
1129        if (!found) {
1130                spin_unlock_irqrestore(&ib_nl_request_lock, flags);
1131                goto resp_out;
1132        }
1133
1134        send_buf = query->mad_buf;
1135
1136        if (!ib_nl_is_good_resolve_resp(nlh)) {
1137                /* if the result is a failure, send out the packet via IB */
1138                ib_sa_disable_local_svc(query);
1139                ret = ib_post_send_mad(query->mad_buf, NULL);
1140                spin_unlock_irqrestore(&ib_nl_request_lock, flags);
1141                if (ret) {
1142                        mad_send_wc.send_buf = send_buf;
1143                        mad_send_wc.status = IB_WC_GENERAL_ERR;
1144                        send_handler(query->port->agent, &mad_send_wc);
1145                }
1146        } else {
1147                spin_unlock_irqrestore(&ib_nl_request_lock, flags);
1148                ib_nl_process_good_resolve_rsp(query, nlh);
1149        }
1150
1151resp_out:
1152        return skb->len;
1153}
1154
1155static void free_sm_ah(struct kref *kref)
1156{
1157        struct ib_sa_sm_ah *sm_ah = container_of(kref, struct ib_sa_sm_ah, ref);
1158
1159        rdma_destroy_ah(sm_ah->ah);
1160        kfree(sm_ah);
1161}
1162
1163void ib_sa_register_client(struct ib_sa_client *client)
1164{
1165        atomic_set(&client->users, 1);
1166        init_completion(&client->comp);
1167}
1168EXPORT_SYMBOL(ib_sa_register_client);
1169
1170void ib_sa_unregister_client(struct ib_sa_client *client)
1171{
1172        ib_sa_client_put(client);
1173        wait_for_completion(&client->comp);
1174}
1175EXPORT_SYMBOL(ib_sa_unregister_client);
1176
1177/**
1178 * ib_sa_cancel_query - try to cancel an SA query
1179 * @id:ID of query to cancel
1180 * @query:query pointer to cancel
1181 *
1182 * Try to cancel an SA query.  If the id and query don't match up or
1183 * the query has already completed, nothing is done.  Otherwise the
1184 * query is canceled and will complete with a status of -EINTR.
1185 */
1186void ib_sa_cancel_query(int id, struct ib_sa_query *query)
1187{
1188        unsigned long flags;
1189        struct ib_mad_agent *agent;
1190        struct ib_mad_send_buf *mad_buf;
1191
1192        spin_lock_irqsave(&idr_lock, flags);
1193        if (idr_find(&query_idr, id) != query) {
1194                spin_unlock_irqrestore(&idr_lock, flags);
1195                return;
1196        }
1197        agent = query->port->agent;
1198        mad_buf = query->mad_buf;
1199        spin_unlock_irqrestore(&idr_lock, flags);
1200
1201        /*
1202         * If the query is still on the netlink request list, schedule
1203         * it to be cancelled by the timeout routine. Otherwise, it has been
1204         * sent to the MAD layer and has to be cancelled from there.
1205         */
1206        if (!ib_nl_cancel_request(query))
1207                ib_cancel_mad(agent, mad_buf);
1208}
1209EXPORT_SYMBOL(ib_sa_cancel_query);
1210
1211static u8 get_src_path_mask(struct ib_device *device, u8 port_num)
1212{
1213        struct ib_sa_device *sa_dev;
1214        struct ib_sa_port   *port;
1215        unsigned long flags;
1216        u8 src_path_mask;
1217
1218        sa_dev = ib_get_client_data(device, &sa_client);
1219        if (!sa_dev)
1220                return 0x7f;
1221
1222        port  = &sa_dev->port[port_num - sa_dev->start_port];
1223        spin_lock_irqsave(&port->ah_lock, flags);
1224        src_path_mask = port->sm_ah ? port->sm_ah->src_path_mask : 0x7f;
1225        spin_unlock_irqrestore(&port->ah_lock, flags);
1226
1227        return src_path_mask;
1228}
1229
1230int ib_init_ah_attr_from_path(struct ib_device *device, u8 port_num,
1231                              struct sa_path_rec *rec,
1232                              struct rdma_ah_attr *ah_attr)
1233{
1234        int ret;
1235        u16 gid_index;
1236        int use_roce;
1237        struct net_device *ndev = NULL;
1238
1239        memset(ah_attr, 0, sizeof *ah_attr);
1240        ah_attr->type = rdma_ah_find_type(device, port_num);
1241
1242        rdma_ah_set_dlid(ah_attr, be32_to_cpu(sa_path_get_dlid(rec)));
1243
1244        if ((ah_attr->type == RDMA_AH_ATTR_TYPE_OPA) &&
1245            (rdma_ah_get_dlid(ah_attr) == be16_to_cpu(IB_LID_PERMISSIVE)))
1246                rdma_ah_set_make_grd(ah_attr, true);
1247
1248        rdma_ah_set_sl(ah_attr, rec->sl);
1249        rdma_ah_set_path_bits(ah_attr, be32_to_cpu(sa_path_get_slid(rec)) &
1250                              get_src_path_mask(device, port_num));
1251        rdma_ah_set_port_num(ah_attr, port_num);
1252        rdma_ah_set_static_rate(ah_attr, rec->rate);
1253        use_roce = rdma_cap_eth_ah(device, port_num);
1254
1255        if (use_roce) {
1256                struct net_device *idev;
1257                struct net_device *resolved_dev;
1258                struct rdma_dev_addr dev_addr = {
1259                        .bound_dev_if = ((sa_path_get_ifindex(rec) >= 0) ?
1260                                         sa_path_get_ifindex(rec) : 0),
1261                        .net = sa_path_get_ndev(rec) ?
1262                                sa_path_get_ndev(rec) :
1263                                &init_net
1264                };
1265                union {
1266                        struct sockaddr     _sockaddr;
1267                        struct sockaddr_in  _sockaddr_in;
1268                        struct sockaddr_in6 _sockaddr_in6;
1269                } sgid_addr, dgid_addr;
1270
1271                if (!device->get_netdev)
1272                        return -EOPNOTSUPP;
1273
1274                rdma_gid2ip(&sgid_addr._sockaddr, &rec->sgid);
1275                rdma_gid2ip(&dgid_addr._sockaddr, &rec->dgid);
1276
1277                /* validate the route */
1278                ret = rdma_resolve_ip_route(&sgid_addr._sockaddr,
1279                                            &dgid_addr._sockaddr, &dev_addr);
1280                if (ret)
1281                        return ret;
1282
1283                if ((dev_addr.network == RDMA_NETWORK_IPV4 ||
1284                     dev_addr.network == RDMA_NETWORK_IPV6) &&
1285                    rec->rec_type != SA_PATH_REC_TYPE_ROCE_V2)
1286                        return -EINVAL;
1287
1288                idev = device->get_netdev(device, port_num);
1289                if (!idev)
1290                        return -ENODEV;
1291
1292                resolved_dev = dev_get_by_index(dev_addr.net,
1293                                                dev_addr.bound_dev_if);
1294                if (!resolved_dev) {
1295                        dev_put(idev);
1296                        return -ENODEV;
1297                }
1298                ndev = ib_get_ndev_from_path(rec);
1299                rcu_read_lock();
1300                if ((ndev && ndev != resolved_dev) ||
1301                    (resolved_dev != idev &&
1302                     !rdma_is_upper_dev_rcu(idev, resolved_dev)))
1303                        ret = -EHOSTUNREACH;
1304                rcu_read_unlock();
1305                dev_put(idev);
1306                dev_put(resolved_dev);
1307                if (ret) {
1308                        if (ndev)
1309                                dev_put(ndev);
1310                        return ret;
1311                }
1312        }
1313
1314        if (rec->hop_limit > 0 || use_roce) {
1315                enum ib_gid_type type = sa_conv_pathrec_to_gid_type(rec);
1316
1317                ret = ib_find_cached_gid_by_port(device, &rec->sgid, type,
1318                                                 port_num, ndev, &gid_index);
1319                if (ret) {
1320                        if (ndev)
1321                                dev_put(ndev);
1322                        return ret;
1323                }
1324
1325                rdma_ah_set_grh(ah_attr, &rec->dgid,
1326                                be32_to_cpu(rec->flow_label),
1327                                gid_index, rec->hop_limit,
1328                                rec->traffic_class);
1329                if (ndev)
1330                        dev_put(ndev);
1331        }
1332
1333        if (use_roce) {
1334                u8 *dmac = sa_path_get_dmac(rec);
1335
1336                if (!dmac)
1337                        return -EINVAL;
1338                memcpy(ah_attr->roce.dmac, dmac, ETH_ALEN);
1339        }
1340
1341        return 0;
1342}
1343EXPORT_SYMBOL(ib_init_ah_attr_from_path);
1344
1345static int alloc_mad(struct ib_sa_query *query, gfp_t gfp_mask)
1346{
1347        struct rdma_ah_attr ah_attr;
1348        unsigned long flags;
1349
1350        spin_lock_irqsave(&query->port->ah_lock, flags);
1351        if (!query->port->sm_ah) {
1352                spin_unlock_irqrestore(&query->port->ah_lock, flags);
1353                return -EAGAIN;
1354        }
1355        kref_get(&query->port->sm_ah->ref);
1356        query->sm_ah = query->port->sm_ah;
1357        spin_unlock_irqrestore(&query->port->ah_lock, flags);
1358
1359        /*
1360         * Always check if sm_ah has valid dlid assigned,
1361         * before querying for class port info
1362         */
1363        if ((rdma_query_ah(query->sm_ah->ah, &ah_attr) < 0) ||
1364            !rdma_is_valid_unicast_lid(&ah_attr)) {
1365                kref_put(&query->sm_ah->ref, free_sm_ah);
1366                return -EAGAIN;
1367        }
1368        query->mad_buf = ib_create_send_mad(query->port->agent, 1,
1369                                            query->sm_ah->pkey_index,
1370                                            0, IB_MGMT_SA_HDR, IB_MGMT_SA_DATA,
1371                                            gfp_mask,
1372                                            ((query->flags & IB_SA_QUERY_OPA) ?
1373                                             OPA_MGMT_BASE_VERSION :
1374                                             IB_MGMT_BASE_VERSION));
1375        if (IS_ERR(query->mad_buf)) {
1376                kref_put(&query->sm_ah->ref, free_sm_ah);
1377                return -ENOMEM;
1378        }
1379
1380        query->mad_buf->ah = query->sm_ah->ah;
1381
1382        return 0;
1383}
1384
1385static void free_mad(struct ib_sa_query *query)
1386{
1387        ib_free_send_mad(query->mad_buf);
1388        kref_put(&query->sm_ah->ref, free_sm_ah);
1389}
1390
1391static void init_mad(struct ib_sa_query *query, struct ib_mad_agent *agent)
1392{
1393        struct ib_sa_mad *mad = query->mad_buf->mad;
1394        unsigned long flags;
1395
1396        memset(mad, 0, sizeof *mad);
1397
1398        if (query->flags & IB_SA_QUERY_OPA) {
1399                mad->mad_hdr.base_version  = OPA_MGMT_BASE_VERSION;
1400                mad->mad_hdr.class_version = OPA_SA_CLASS_VERSION;
1401        } else {
1402                mad->mad_hdr.base_version  = IB_MGMT_BASE_VERSION;
1403                mad->mad_hdr.class_version = IB_SA_CLASS_VERSION;
1404        }
1405        mad->mad_hdr.mgmt_class    = IB_MGMT_CLASS_SUBN_ADM;
1406        spin_lock_irqsave(&tid_lock, flags);
1407        mad->mad_hdr.tid           =
1408                cpu_to_be64(((u64) agent->hi_tid) << 32 | tid++);
1409        spin_unlock_irqrestore(&tid_lock, flags);
1410}
1411
1412static int send_mad(struct ib_sa_query *query, int timeout_ms, gfp_t gfp_mask)
1413{
1414        bool preload = gfpflags_allow_blocking(gfp_mask);
1415        unsigned long flags;
1416        int ret, id;
1417
1418        if (preload)
1419                idr_preload(gfp_mask);
1420        spin_lock_irqsave(&idr_lock, flags);
1421
1422        id = idr_alloc(&query_idr, query, 0, 0, GFP_NOWAIT);
1423
1424        spin_unlock_irqrestore(&idr_lock, flags);
1425        if (preload)
1426                idr_preload_end();
1427        if (id < 0)
1428                return id;
1429
1430        query->mad_buf->timeout_ms  = timeout_ms;
1431        query->mad_buf->context[0] = query;
1432        query->id = id;
1433
1434        if ((query->flags & IB_SA_ENABLE_LOCAL_SERVICE) &&
1435            (!(query->flags & IB_SA_QUERY_OPA))) {
1436                if (!rdma_nl_chk_listeners(RDMA_NL_GROUP_LS)) {
1437                        if (!ib_nl_make_request(query, gfp_mask))
1438                                return id;
1439                }
1440                ib_sa_disable_local_svc(query);
1441        }
1442
1443        ret = ib_post_send_mad(query->mad_buf, NULL);
1444        if (ret) {
1445                spin_lock_irqsave(&idr_lock, flags);
1446                idr_remove(&query_idr, id);
1447                spin_unlock_irqrestore(&idr_lock, flags);
1448        }
1449
1450        /*
1451         * It's not safe to dereference query any more, because the
1452         * send may already have completed and freed the query in
1453         * another context.
1454         */
1455        return ret ? ret : id;
1456}
1457
1458void ib_sa_unpack_path(void *attribute, struct sa_path_rec *rec)
1459{
1460        ib_unpack(path_rec_table, ARRAY_SIZE(path_rec_table), attribute, rec);
1461}
1462EXPORT_SYMBOL(ib_sa_unpack_path);
1463
1464void ib_sa_pack_path(struct sa_path_rec *rec, void *attribute)
1465{
1466        ib_pack(path_rec_table, ARRAY_SIZE(path_rec_table), rec, attribute);
1467}
1468EXPORT_SYMBOL(ib_sa_pack_path);
1469
1470static bool ib_sa_opa_pathrecord_support(struct ib_sa_client *client,
1471                                         struct ib_device *device,
1472                                         u8 port_num)
1473{
1474        struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
1475        struct ib_sa_port *port;
1476        unsigned long flags;
1477        bool ret = false;
1478
1479        if (!sa_dev)
1480                return ret;
1481
1482        port = &sa_dev->port[port_num - sa_dev->start_port];
1483        spin_lock_irqsave(&port->classport_lock, flags);
1484        if (!port->classport_info.valid)
1485                goto ret;
1486
1487        if (port->classport_info.data.type == RDMA_CLASS_PORT_INFO_OPA)
1488                ret = opa_get_cpi_capmask2(&port->classport_info.data.opa) &
1489                        OPA_CLASS_PORT_INFO_PR_SUPPORT;
1490ret:
1491        spin_unlock_irqrestore(&port->classport_lock, flags);
1492        return ret;
1493}
1494
1495enum opa_pr_supported {
1496        PR_NOT_SUPPORTED,
1497        PR_OPA_SUPPORTED,
1498        PR_IB_SUPPORTED
1499};
1500
1501/**
1502 * Check if current PR query can be an OPA query.
1503 * Retuns PR_NOT_SUPPORTED if a path record query is not
1504 * possible, PR_OPA_SUPPORTED if an OPA path record query
1505 * is possible and PR_IB_SUPPORTED if an IB path record
1506 * query is possible.
1507 */
1508static int opa_pr_query_possible(struct ib_sa_client *client,
1509                                 struct ib_device *device,
1510                                 u8 port_num,
1511                                 struct sa_path_rec *rec)
1512{
1513        struct ib_port_attr port_attr;
1514
1515        if (ib_query_port(device, port_num, &port_attr))
1516                return PR_NOT_SUPPORTED;
1517
1518        if (ib_sa_opa_pathrecord_support(client, device, port_num))
1519                return PR_OPA_SUPPORTED;
1520
1521        if (port_attr.lid >= be16_to_cpu(IB_MULTICAST_LID_BASE))
1522                return PR_NOT_SUPPORTED;
1523        else
1524                return PR_IB_SUPPORTED;
1525}
1526
1527static void ib_sa_path_rec_callback(struct ib_sa_query *sa_query,
1528                                    int status,
1529                                    struct ib_sa_mad *mad)
1530{
1531        struct ib_sa_path_query *query =
1532                container_of(sa_query, struct ib_sa_path_query, sa_query);
1533
1534        if (mad) {
1535                struct sa_path_rec rec;
1536
1537                if (sa_query->flags & IB_SA_QUERY_OPA) {
1538                        ib_unpack(opa_path_rec_table,
1539                                  ARRAY_SIZE(opa_path_rec_table),
1540                                  mad->data, &rec);
1541                        rec.rec_type = SA_PATH_REC_TYPE_OPA;
1542                        query->callback(status, &rec, query->context);
1543                } else {
1544                        ib_unpack(path_rec_table,
1545                                  ARRAY_SIZE(path_rec_table),
1546                                  mad->data, &rec);
1547                        rec.rec_type = SA_PATH_REC_TYPE_IB;
1548                        sa_path_set_ndev(&rec, NULL);
1549                        sa_path_set_ifindex(&rec, 0);
1550                        sa_path_set_dmac_zero(&rec);
1551
1552                        if (query->conv_pr) {
1553                                struct sa_path_rec opa;
1554
1555                                memset(&opa, 0, sizeof(struct sa_path_rec));
1556                                sa_convert_path_ib_to_opa(&opa, &rec);
1557                                query->callback(status, &opa, query->context);
1558                        } else {
1559                                query->callback(status, &rec, query->context);
1560                        }
1561                }
1562        } else
1563                query->callback(status, NULL, query->context);
1564}
1565
1566static void ib_sa_path_rec_release(struct ib_sa_query *sa_query)
1567{
1568        struct ib_sa_path_query *query =
1569                container_of(sa_query, struct ib_sa_path_query, sa_query);
1570
1571        kfree(query->conv_pr);
1572        kfree(query);
1573}
1574
1575/**
1576 * ib_sa_path_rec_get - Start a Path get query
1577 * @client:SA client
1578 * @device:device to send query on
1579 * @port_num: port number to send query on
1580 * @rec:Path Record to send in query
1581 * @comp_mask:component mask to send in query
1582 * @timeout_ms:time to wait for response
1583 * @gfp_mask:GFP mask to use for internal allocations
1584 * @callback:function called when query completes, times out or is
1585 * canceled
1586 * @context:opaque user context passed to callback
1587 * @sa_query:query context, used to cancel query
1588 *
1589 * Send a Path Record Get query to the SA to look up a path.  The
1590 * callback function will be called when the query completes (or
1591 * fails); status is 0 for a successful response, -EINTR if the query
1592 * is canceled, -ETIMEDOUT is the query timed out, or -EIO if an error
1593 * occurred sending the query.  The resp parameter of the callback is
1594 * only valid if status is 0.
1595 *
1596 * If the return value of ib_sa_path_rec_get() is negative, it is an
1597 * error code.  Otherwise it is a query ID that can be used to cancel
1598 * the query.
1599 */
1600int ib_sa_path_rec_get(struct ib_sa_client *client,
1601                       struct ib_device *device, u8 port_num,
1602                       struct sa_path_rec *rec,
1603                       ib_sa_comp_mask comp_mask,
1604                       int timeout_ms, gfp_t gfp_mask,
1605                       void (*callback)(int status,
1606                                        struct sa_path_rec *resp,
1607                                        void *context),
1608                       void *context,
1609                       struct ib_sa_query **sa_query)
1610{
1611        struct ib_sa_path_query *query;
1612        struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
1613        struct ib_sa_port   *port;
1614        struct ib_mad_agent *agent;
1615        struct ib_sa_mad *mad;
1616        enum opa_pr_supported status;
1617        int ret;
1618
1619        if (!sa_dev)
1620                return -ENODEV;
1621
1622        if ((rec->rec_type != SA_PATH_REC_TYPE_IB) &&
1623            (rec->rec_type != SA_PATH_REC_TYPE_OPA))
1624                return -EINVAL;
1625
1626        port  = &sa_dev->port[port_num - sa_dev->start_port];
1627        agent = port->agent;
1628
1629        query = kzalloc(sizeof(*query), gfp_mask);
1630        if (!query)
1631                return -ENOMEM;
1632
1633        query->sa_query.port     = port;
1634        if (rec->rec_type == SA_PATH_REC_TYPE_OPA) {
1635                status = opa_pr_query_possible(client, device, port_num, rec);
1636                if (status == PR_NOT_SUPPORTED) {
1637                        ret = -EINVAL;
1638                        goto err1;
1639                } else if (status == PR_OPA_SUPPORTED) {
1640                        query->sa_query.flags |= IB_SA_QUERY_OPA;
1641                } else {
1642                        query->conv_pr =
1643                                kmalloc(sizeof(*query->conv_pr), gfp_mask);
1644                        if (!query->conv_pr) {
1645                                ret = -ENOMEM;
1646                                goto err1;
1647                        }
1648                }
1649        }
1650
1651        ret = alloc_mad(&query->sa_query, gfp_mask);
1652        if (ret)
1653                goto err2;
1654
1655        ib_sa_client_get(client);
1656        query->sa_query.client = client;
1657        query->callback        = callback;
1658        query->context         = context;
1659
1660        mad = query->sa_query.mad_buf->mad;
1661        init_mad(&query->sa_query, agent);
1662
1663        query->sa_query.callback = callback ? ib_sa_path_rec_callback : NULL;
1664        query->sa_query.release  = ib_sa_path_rec_release;
1665        mad->mad_hdr.method      = IB_MGMT_METHOD_GET;
1666        mad->mad_hdr.attr_id     = cpu_to_be16(IB_SA_ATTR_PATH_REC);
1667        mad->sa_hdr.comp_mask    = comp_mask;
1668
1669        if (query->sa_query.flags & IB_SA_QUERY_OPA) {
1670                ib_pack(opa_path_rec_table, ARRAY_SIZE(opa_path_rec_table),
1671                        rec, mad->data);
1672        } else if (query->conv_pr) {
1673                sa_convert_path_opa_to_ib(query->conv_pr, rec);
1674                ib_pack(path_rec_table, ARRAY_SIZE(path_rec_table),
1675                        query->conv_pr, mad->data);
1676        } else {
1677                ib_pack(path_rec_table, ARRAY_SIZE(path_rec_table),
1678                        rec, mad->data);
1679        }
1680
1681        *sa_query = &query->sa_query;
1682
1683        query->sa_query.flags |= IB_SA_ENABLE_LOCAL_SERVICE;
1684        query->sa_query.mad_buf->context[1] = (query->conv_pr) ?
1685                                                query->conv_pr : rec;
1686
1687        ret = send_mad(&query->sa_query, timeout_ms, gfp_mask);
1688        if (ret < 0)
1689                goto err3;
1690
1691        return ret;
1692
1693err3:
1694        *sa_query = NULL;
1695        ib_sa_client_put(query->sa_query.client);
1696        free_mad(&query->sa_query);
1697err2:
1698        kfree(query->conv_pr);
1699err1:
1700        kfree(query);
1701        return ret;
1702}
1703EXPORT_SYMBOL(ib_sa_path_rec_get);
1704
1705static void ib_sa_service_rec_callback(struct ib_sa_query *sa_query,
1706                                    int status,
1707                                    struct ib_sa_mad *mad)
1708{
1709        struct ib_sa_service_query *query =
1710                container_of(sa_query, struct ib_sa_service_query, sa_query);
1711
1712        if (mad) {
1713                struct ib_sa_service_rec rec;
1714
1715                ib_unpack(service_rec_table, ARRAY_SIZE(service_rec_table),
1716                          mad->data, &rec);
1717                query->callback(status, &rec, query->context);
1718        } else
1719                query->callback(status, NULL, query->context);
1720}
1721
1722static void ib_sa_service_rec_release(struct ib_sa_query *sa_query)
1723{
1724        kfree(container_of(sa_query, struct ib_sa_service_query, sa_query));
1725}
1726
1727/**
1728 * ib_sa_service_rec_query - Start Service Record operation
1729 * @client:SA client
1730 * @device:device to send request on
1731 * @port_num: port number to send request on
1732 * @method:SA method - should be get, set, or delete
1733 * @rec:Service Record to send in request
1734 * @comp_mask:component mask to send in request
1735 * @timeout_ms:time to wait for response
1736 * @gfp_mask:GFP mask to use for internal allocations
1737 * @callback:function called when request completes, times out or is
1738 * canceled
1739 * @context:opaque user context passed to callback
1740 * @sa_query:request context, used to cancel request
1741 *
1742 * Send a Service Record set/get/delete to the SA to register,
1743 * unregister or query a service record.
1744 * The callback function will be called when the request completes (or
1745 * fails); status is 0 for a successful response, -EINTR if the query
1746 * is canceled, -ETIMEDOUT is the query timed out, or -EIO if an error
1747 * occurred sending the query.  The resp parameter of the callback is
1748 * only valid if status is 0.
1749 *
1750 * If the return value of ib_sa_service_rec_query() is negative, it is an
1751 * error code.  Otherwise it is a request ID that can be used to cancel
1752 * the query.
1753 */
1754int ib_sa_service_rec_query(struct ib_sa_client *client,
1755                            struct ib_device *device, u8 port_num, u8 method,
1756                            struct ib_sa_service_rec *rec,
1757                            ib_sa_comp_mask comp_mask,
1758                            int timeout_ms, gfp_t gfp_mask,
1759                            void (*callback)(int status,
1760                                             struct ib_sa_service_rec *resp,
1761                                             void *context),
1762                            void *context,
1763                            struct ib_sa_query **sa_query)
1764{
1765        struct ib_sa_service_query *query;
1766        struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
1767        struct ib_sa_port   *port;
1768        struct ib_mad_agent *agent;
1769        struct ib_sa_mad *mad;
1770        int ret;
1771
1772        if (!sa_dev)
1773                return -ENODEV;
1774
1775        port  = &sa_dev->port[port_num - sa_dev->start_port];
1776        agent = port->agent;
1777
1778        if (method != IB_MGMT_METHOD_GET &&
1779            method != IB_MGMT_METHOD_SET &&
1780            method != IB_SA_METHOD_DELETE)
1781                return -EINVAL;
1782
1783        query = kzalloc(sizeof(*query), gfp_mask);
1784        if (!query)
1785                return -ENOMEM;
1786
1787        query->sa_query.port     = port;
1788        ret = alloc_mad(&query->sa_query, gfp_mask);
1789        if (ret)
1790                goto err1;
1791
1792        ib_sa_client_get(client);
1793        query->sa_query.client = client;
1794        query->callback        = callback;
1795        query->context         = context;
1796
1797        mad = query->sa_query.mad_buf->mad;
1798        init_mad(&query->sa_query, agent);
1799
1800        query->sa_query.callback = callback ? ib_sa_service_rec_callback : NULL;
1801        query->sa_query.release  = ib_sa_service_rec_release;
1802        mad->mad_hdr.method      = method;
1803        mad->mad_hdr.attr_id     = cpu_to_be16(IB_SA_ATTR_SERVICE_REC);
1804        mad->sa_hdr.comp_mask    = comp_mask;
1805
1806        ib_pack(service_rec_table, ARRAY_SIZE(service_rec_table),
1807                rec, mad->data);
1808
1809        *sa_query = &query->sa_query;
1810
1811        ret = send_mad(&query->sa_query, timeout_ms, gfp_mask);
1812        if (ret < 0)
1813                goto err2;
1814
1815        return ret;
1816
1817err2:
1818        *sa_query = NULL;
1819        ib_sa_client_put(query->sa_query.client);
1820        free_mad(&query->sa_query);
1821
1822err1:
1823        kfree(query);
1824        return ret;
1825}
1826EXPORT_SYMBOL(ib_sa_service_rec_query);
1827
1828static void ib_sa_mcmember_rec_callback(struct ib_sa_query *sa_query,
1829                                        int status,
1830                                        struct ib_sa_mad *mad)
1831{
1832        struct ib_sa_mcmember_query *query =
1833                container_of(sa_query, struct ib_sa_mcmember_query, sa_query);
1834
1835        if (mad) {
1836                struct ib_sa_mcmember_rec rec;
1837
1838                ib_unpack(mcmember_rec_table, ARRAY_SIZE(mcmember_rec_table),
1839                          mad->data, &rec);
1840                query->callback(status, &rec, query->context);
1841        } else
1842                query->callback(status, NULL, query->context);
1843}
1844
1845static void ib_sa_mcmember_rec_release(struct ib_sa_query *sa_query)
1846{
1847        kfree(container_of(sa_query, struct ib_sa_mcmember_query, sa_query));
1848}
1849
1850int ib_sa_mcmember_rec_query(struct ib_sa_client *client,
1851                             struct ib_device *device, u8 port_num,
1852                             u8 method,
1853                             struct ib_sa_mcmember_rec *rec,
1854                             ib_sa_comp_mask comp_mask,
1855                             int timeout_ms, gfp_t gfp_mask,
1856                             void (*callback)(int status,
1857                                              struct ib_sa_mcmember_rec *resp,
1858                                              void *context),
1859                             void *context,
1860                             struct ib_sa_query **sa_query)
1861{
1862        struct ib_sa_mcmember_query *query;
1863        struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
1864        struct ib_sa_port   *port;
1865        struct ib_mad_agent *agent;
1866        struct ib_sa_mad *mad;
1867        int ret;
1868
1869        if (!sa_dev)
1870                return -ENODEV;
1871
1872        port  = &sa_dev->port[port_num - sa_dev->start_port];
1873        agent = port->agent;
1874
1875        query = kzalloc(sizeof(*query), gfp_mask);
1876        if (!query)
1877                return -ENOMEM;
1878
1879        query->sa_query.port     = port;
1880        ret = alloc_mad(&query->sa_query, gfp_mask);
1881        if (ret)
1882                goto err1;
1883
1884        ib_sa_client_get(client);
1885        query->sa_query.client = client;
1886        query->callback        = callback;
1887        query->context         = context;
1888
1889        mad = query->sa_query.mad_buf->mad;
1890        init_mad(&query->sa_query, agent);
1891
1892        query->sa_query.callback = callback ? ib_sa_mcmember_rec_callback : NULL;
1893        query->sa_query.release  = ib_sa_mcmember_rec_release;
1894        mad->mad_hdr.method      = method;
1895        mad->mad_hdr.attr_id     = cpu_to_be16(IB_SA_ATTR_MC_MEMBER_REC);
1896        mad->sa_hdr.comp_mask    = comp_mask;
1897
1898        ib_pack(mcmember_rec_table, ARRAY_SIZE(mcmember_rec_table),
1899                rec, mad->data);
1900
1901        *sa_query = &query->sa_query;
1902
1903        ret = send_mad(&query->sa_query, timeout_ms, gfp_mask);
1904        if (ret < 0)
1905                goto err2;
1906
1907        return ret;
1908
1909err2:
1910        *sa_query = NULL;
1911        ib_sa_client_put(query->sa_query.client);
1912        free_mad(&query->sa_query);
1913
1914err1:
1915        kfree(query);
1916        return ret;
1917}
1918
1919/* Support GuidInfoRecord */
1920static void ib_sa_guidinfo_rec_callback(struct ib_sa_query *sa_query,
1921                                        int status,
1922                                        struct ib_sa_mad *mad)
1923{
1924        struct ib_sa_guidinfo_query *query =
1925                container_of(sa_query, struct ib_sa_guidinfo_query, sa_query);
1926
1927        if (mad) {
1928                struct ib_sa_guidinfo_rec rec;
1929
1930                ib_unpack(guidinfo_rec_table, ARRAY_SIZE(guidinfo_rec_table),
1931                          mad->data, &rec);
1932                query->callback(status, &rec, query->context);
1933        } else
1934                query->callback(status, NULL, query->context);
1935}
1936
1937static void ib_sa_guidinfo_rec_release(struct ib_sa_query *sa_query)
1938{
1939        kfree(container_of(sa_query, struct ib_sa_guidinfo_query, sa_query));
1940}
1941
1942int ib_sa_guid_info_rec_query(struct ib_sa_client *client,
1943                              struct ib_device *device, u8 port_num,
1944                              struct ib_sa_guidinfo_rec *rec,
1945                              ib_sa_comp_mask comp_mask, u8 method,
1946                              int timeout_ms, gfp_t gfp_mask,
1947                              void (*callback)(int status,
1948                                               struct ib_sa_guidinfo_rec *resp,
1949                                               void *context),
1950                              void *context,
1951                              struct ib_sa_query **sa_query)
1952{
1953        struct ib_sa_guidinfo_query *query;
1954        struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
1955        struct ib_sa_port *port;
1956        struct ib_mad_agent *agent;
1957        struct ib_sa_mad *mad;
1958        int ret;
1959
1960        if (!sa_dev)
1961                return -ENODEV;
1962
1963        if (method != IB_MGMT_METHOD_GET &&
1964            method != IB_MGMT_METHOD_SET &&
1965            method != IB_SA_METHOD_DELETE) {
1966                return -EINVAL;
1967        }
1968
1969        port  = &sa_dev->port[port_num - sa_dev->start_port];
1970        agent = port->agent;
1971
1972        query = kzalloc(sizeof(*query), gfp_mask);
1973        if (!query)
1974                return -ENOMEM;
1975
1976        query->sa_query.port = port;
1977        ret = alloc_mad(&query->sa_query, gfp_mask);
1978        if (ret)
1979                goto err1;
1980
1981        ib_sa_client_get(client);
1982        query->sa_query.client = client;
1983        query->callback        = callback;
1984        query->context         = context;
1985
1986        mad = query->sa_query.mad_buf->mad;
1987        init_mad(&query->sa_query, agent);
1988
1989        query->sa_query.callback = callback ? ib_sa_guidinfo_rec_callback : NULL;
1990        query->sa_query.release  = ib_sa_guidinfo_rec_release;
1991
1992        mad->mad_hdr.method      = method;
1993        mad->mad_hdr.attr_id     = cpu_to_be16(IB_SA_ATTR_GUID_INFO_REC);
1994        mad->sa_hdr.comp_mask    = comp_mask;
1995
1996        ib_pack(guidinfo_rec_table, ARRAY_SIZE(guidinfo_rec_table), rec,
1997                mad->data);
1998
1999        *sa_query = &query->sa_query;
2000
2001        ret = send_mad(&query->sa_query, timeout_ms, gfp_mask);
2002        if (ret < 0)
2003                goto err2;
2004
2005        return ret;
2006
2007err2:
2008        *sa_query = NULL;
2009        ib_sa_client_put(query->sa_query.client);
2010        free_mad(&query->sa_query);
2011
2012err1:
2013        kfree(query);
2014        return ret;
2015}
2016EXPORT_SYMBOL(ib_sa_guid_info_rec_query);
2017
2018bool ib_sa_sendonly_fullmem_support(struct ib_sa_client *client,
2019                                    struct ib_device *device,
2020                                    u8 port_num)
2021{
2022        struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
2023        struct ib_sa_port *port;
2024        bool ret = false;
2025        unsigned long flags;
2026
2027        if (!sa_dev)
2028                return ret;
2029
2030        port  = &sa_dev->port[port_num - sa_dev->start_port];
2031
2032        spin_lock_irqsave(&port->classport_lock, flags);
2033        if ((port->classport_info.valid) &&
2034            (port->classport_info.data.type == RDMA_CLASS_PORT_INFO_IB))
2035                ret = ib_get_cpi_capmask2(&port->classport_info.data.ib)
2036                        & IB_SA_CAP_MASK2_SENDONLY_FULL_MEM_SUPPORT;
2037        spin_unlock_irqrestore(&port->classport_lock, flags);
2038        return ret;
2039}
2040EXPORT_SYMBOL(ib_sa_sendonly_fullmem_support);
2041
2042struct ib_classport_info_context {
2043        struct completion       done;
2044        struct ib_sa_query      *sa_query;
2045};
2046
2047static void ib_classportinfo_cb(void *context)
2048{
2049        struct ib_classport_info_context *cb_ctx = context;
2050
2051        complete(&cb_ctx->done);
2052}
2053
2054static void ib_sa_classport_info_rec_callback(struct ib_sa_query *sa_query,
2055                                              int status,
2056                                              struct ib_sa_mad *mad)
2057{
2058        unsigned long flags;
2059        struct ib_sa_classport_info_query *query =
2060                container_of(sa_query, struct ib_sa_classport_info_query, sa_query);
2061        struct ib_sa_classport_cache *info = &sa_query->port->classport_info;
2062
2063        if (mad) {
2064                if (sa_query->flags & IB_SA_QUERY_OPA) {
2065                        struct opa_class_port_info rec;
2066
2067                        ib_unpack(opa_classport_info_rec_table,
2068                                  ARRAY_SIZE(opa_classport_info_rec_table),
2069                                  mad->data, &rec);
2070
2071                        spin_lock_irqsave(&sa_query->port->classport_lock,
2072                                          flags);
2073                        if (!status && !info->valid) {
2074                                memcpy(&info->data.opa, &rec,
2075                                       sizeof(info->data.opa));
2076
2077                                info->valid = true;
2078                                info->data.type = RDMA_CLASS_PORT_INFO_OPA;
2079                        }
2080                        spin_unlock_irqrestore(&sa_query->port->classport_lock,
2081                                               flags);
2082
2083                } else {
2084                        struct ib_class_port_info rec;
2085
2086                        ib_unpack(ib_classport_info_rec_table,
2087                                  ARRAY_SIZE(ib_classport_info_rec_table),
2088                                  mad->data, &rec);
2089
2090                        spin_lock_irqsave(&sa_query->port->classport_lock,
2091                                          flags);
2092                        if (!status && !info->valid) {
2093                                memcpy(&info->data.ib, &rec,
2094                                       sizeof(info->data.ib));
2095
2096                                info->valid = true;
2097                                info->data.type = RDMA_CLASS_PORT_INFO_IB;
2098                        }
2099                        spin_unlock_irqrestore(&sa_query->port->classport_lock,
2100                                               flags);
2101                }
2102        }
2103        query->callback(query->context);
2104}
2105
2106static void ib_sa_classport_info_rec_release(struct ib_sa_query *sa_query)
2107{
2108        kfree(container_of(sa_query, struct ib_sa_classport_info_query,
2109                           sa_query));
2110}
2111
2112static int ib_sa_classport_info_rec_query(struct ib_sa_port *port,
2113                                          int timeout_ms,
2114                                          void (*callback)(void *context),
2115                                          void *context,
2116                                          struct ib_sa_query **sa_query)
2117{
2118        struct ib_mad_agent *agent;
2119        struct ib_sa_classport_info_query *query;
2120        struct ib_sa_mad *mad;
2121        gfp_t gfp_mask = GFP_KERNEL;
2122        int ret;
2123
2124        agent = port->agent;
2125
2126        query = kzalloc(sizeof(*query), gfp_mask);
2127        if (!query)
2128                return -ENOMEM;
2129
2130        query->sa_query.port = port;
2131        query->sa_query.flags |= rdma_cap_opa_ah(port->agent->device,
2132                                                 port->port_num) ?
2133                                 IB_SA_QUERY_OPA : 0;
2134        ret = alloc_mad(&query->sa_query, gfp_mask);
2135        if (ret)
2136                goto err_free;
2137
2138        query->callback = callback;
2139        query->context = context;
2140
2141        mad = query->sa_query.mad_buf->mad;
2142        init_mad(&query->sa_query, agent);
2143
2144        query->sa_query.callback = ib_sa_classport_info_rec_callback;
2145        query->sa_query.release  = ib_sa_classport_info_rec_release;
2146        mad->mad_hdr.method      = IB_MGMT_METHOD_GET;
2147        mad->mad_hdr.attr_id     = cpu_to_be16(IB_SA_ATTR_CLASS_PORTINFO);
2148        mad->sa_hdr.comp_mask    = 0;
2149        *sa_query = &query->sa_query;
2150
2151        ret = send_mad(&query->sa_query, timeout_ms, gfp_mask);
2152        if (ret < 0)
2153                goto err_free_mad;
2154
2155        return ret;
2156
2157err_free_mad:
2158        *sa_query = NULL;
2159        free_mad(&query->sa_query);
2160
2161err_free:
2162        kfree(query);
2163        return ret;
2164}
2165
2166static void update_ib_cpi(struct work_struct *work)
2167{
2168        struct ib_sa_port *port =
2169                container_of(work, struct ib_sa_port, ib_cpi_work.work);
2170        struct ib_classport_info_context *cb_context;
2171        unsigned long flags;
2172        int ret;
2173
2174        /* If the classport info is valid, nothing
2175         * to do here.
2176         */
2177        spin_lock_irqsave(&port->classport_lock, flags);
2178        if (port->classport_info.valid) {
2179                spin_unlock_irqrestore(&port->classport_lock, flags);
2180                return;
2181        }
2182        spin_unlock_irqrestore(&port->classport_lock, flags);
2183
2184        cb_context = kmalloc(sizeof(*cb_context), GFP_KERNEL);
2185        if (!cb_context)
2186                goto err_nomem;
2187
2188        init_completion(&cb_context->done);
2189
2190        ret = ib_sa_classport_info_rec_query(port, 3000,
2191                                             ib_classportinfo_cb, cb_context,
2192                                             &cb_context->sa_query);
2193        if (ret < 0)
2194                goto free_cb_err;
2195        wait_for_completion(&cb_context->done);
2196free_cb_err:
2197        kfree(cb_context);
2198        spin_lock_irqsave(&port->classport_lock, flags);
2199
2200        /* If the classport info is still not valid, the query should have
2201         * failed for some reason. Retry issuing the query
2202         */
2203        if (!port->classport_info.valid) {
2204                port->classport_info.retry_cnt++;
2205                if (port->classport_info.retry_cnt <=
2206                    IB_SA_CPI_MAX_RETRY_CNT) {
2207                        unsigned long delay =
2208                                msecs_to_jiffies(IB_SA_CPI_RETRY_WAIT);
2209
2210                        queue_delayed_work(ib_wq, &port->ib_cpi_work, delay);
2211                }
2212        }
2213        spin_unlock_irqrestore(&port->classport_lock, flags);
2214
2215err_nomem:
2216        return;
2217}
2218
2219static void send_handler(struct ib_mad_agent *agent,
2220                         struct ib_mad_send_wc *mad_send_wc)
2221{
2222        struct ib_sa_query *query = mad_send_wc->send_buf->context[0];
2223        unsigned long flags;
2224
2225        if (query->callback)
2226                switch (mad_send_wc->status) {
2227                case IB_WC_SUCCESS:
2228                        /* No callback -- already got recv */
2229                        break;
2230                case IB_WC_RESP_TIMEOUT_ERR:
2231                        query->callback(query, -ETIMEDOUT, NULL);
2232                        break;
2233                case IB_WC_WR_FLUSH_ERR:
2234                        query->callback(query, -EINTR, NULL);
2235                        break;
2236                default:
2237                        query->callback(query, -EIO, NULL);
2238                        break;
2239                }
2240
2241        spin_lock_irqsave(&idr_lock, flags);
2242        idr_remove(&query_idr, query->id);
2243        spin_unlock_irqrestore(&idr_lock, flags);
2244
2245        free_mad(query);
2246        if (query->client)
2247                ib_sa_client_put(query->client);
2248        query->release(query);
2249}
2250
2251static void recv_handler(struct ib_mad_agent *mad_agent,
2252                         struct ib_mad_send_buf *send_buf,
2253                         struct ib_mad_recv_wc *mad_recv_wc)
2254{
2255        struct ib_sa_query *query;
2256
2257        if (!send_buf)
2258                return;
2259
2260        query = send_buf->context[0];
2261        if (query->callback) {
2262                if (mad_recv_wc->wc->status == IB_WC_SUCCESS)
2263                        query->callback(query,
2264                                        mad_recv_wc->recv_buf.mad->mad_hdr.status ?
2265                                        -EINVAL : 0,
2266                                        (struct ib_sa_mad *) mad_recv_wc->recv_buf.mad);
2267                else
2268                        query->callback(query, -EIO, NULL);
2269        }
2270
2271        ib_free_recv_mad(mad_recv_wc);
2272}
2273
2274static void update_sm_ah(struct work_struct *work)
2275{
2276        struct ib_sa_port *port =
2277                container_of(work, struct ib_sa_port, update_task);
2278        struct ib_sa_sm_ah *new_ah;
2279        struct ib_port_attr port_attr;
2280        struct rdma_ah_attr   ah_attr;
2281
2282        if (ib_query_port(port->agent->device, port->port_num, &port_attr)) {
2283                pr_warn("Couldn't query port\n");
2284                return;
2285        }
2286
2287        new_ah = kmalloc(sizeof(*new_ah), GFP_KERNEL);
2288        if (!new_ah)
2289                return;
2290
2291        kref_init(&new_ah->ref);
2292        new_ah->src_path_mask = (1 << port_attr.lmc) - 1;
2293
2294        new_ah->pkey_index = 0;
2295        if (ib_find_pkey(port->agent->device, port->port_num,
2296                         IB_DEFAULT_PKEY_FULL, &new_ah->pkey_index))
2297                pr_err("Couldn't find index for default PKey\n");
2298
2299        memset(&ah_attr, 0, sizeof(ah_attr));
2300        ah_attr.type = rdma_ah_find_type(port->agent->device,
2301                                         port->port_num);
2302        rdma_ah_set_dlid(&ah_attr, port_attr.sm_lid);
2303        rdma_ah_set_sl(&ah_attr, port_attr.sm_sl);
2304        rdma_ah_set_port_num(&ah_attr, port->port_num);
2305        if (port_attr.grh_required) {
2306                if (ah_attr.type == RDMA_AH_ATTR_TYPE_OPA) {
2307                        rdma_ah_set_make_grd(&ah_attr, true);
2308                } else {
2309                        rdma_ah_set_ah_flags(&ah_attr, IB_AH_GRH);
2310                        rdma_ah_set_subnet_prefix(&ah_attr,
2311                                                  cpu_to_be64(port_attr.subnet_prefix));
2312                        rdma_ah_set_interface_id(&ah_attr,
2313                                                 cpu_to_be64(IB_SA_WELL_KNOWN_GUID));
2314                }
2315        }
2316
2317        new_ah->ah = rdma_create_ah(port->agent->qp->pd, &ah_attr);
2318        if (IS_ERR(new_ah->ah)) {
2319                pr_warn("Couldn't create new SM AH\n");
2320                kfree(new_ah);
2321                return;
2322        }
2323
2324        spin_lock_irq(&port->ah_lock);
2325        if (port->sm_ah)
2326                kref_put(&port->sm_ah->ref, free_sm_ah);
2327        port->sm_ah = new_ah;
2328        spin_unlock_irq(&port->ah_lock);
2329}
2330
2331static void ib_sa_event(struct ib_event_handler *handler,
2332                        struct ib_event *event)
2333{
2334        if (event->event == IB_EVENT_PORT_ERR    ||
2335            event->event == IB_EVENT_PORT_ACTIVE ||
2336            event->event == IB_EVENT_LID_CHANGE  ||
2337            event->event == IB_EVENT_PKEY_CHANGE ||
2338            event->event == IB_EVENT_SM_CHANGE   ||
2339            event->event == IB_EVENT_CLIENT_REREGISTER) {
2340                unsigned long flags;
2341                struct ib_sa_device *sa_dev =
2342                        container_of(handler, typeof(*sa_dev), event_handler);
2343                u8 port_num = event->element.port_num - sa_dev->start_port;
2344                struct ib_sa_port *port = &sa_dev->port[port_num];
2345
2346                if (!rdma_cap_ib_sa(handler->device, port->port_num))
2347                        return;
2348
2349                spin_lock_irqsave(&port->ah_lock, flags);
2350                if (port->sm_ah)
2351                        kref_put(&port->sm_ah->ref, free_sm_ah);
2352                port->sm_ah = NULL;
2353                spin_unlock_irqrestore(&port->ah_lock, flags);
2354
2355                if (event->event == IB_EVENT_SM_CHANGE ||
2356                    event->event == IB_EVENT_CLIENT_REREGISTER ||
2357                    event->event == IB_EVENT_LID_CHANGE ||
2358                    event->event == IB_EVENT_PORT_ACTIVE) {
2359                        unsigned long delay =
2360                                msecs_to_jiffies(IB_SA_CPI_RETRY_WAIT);
2361
2362                        spin_lock_irqsave(&port->classport_lock, flags);
2363                        port->classport_info.valid = false;
2364                        port->classport_info.retry_cnt = 0;
2365                        spin_unlock_irqrestore(&port->classport_lock, flags);
2366                        queue_delayed_work(ib_wq,
2367                                           &port->ib_cpi_work, delay);
2368                }
2369                queue_work(ib_wq, &sa_dev->port[port_num].update_task);
2370        }
2371}
2372
2373static void ib_sa_add_one(struct ib_device *device)
2374{
2375        struct ib_sa_device *sa_dev;
2376        int s, e, i;
2377        int count = 0;
2378
2379        s = rdma_start_port(device);
2380        e = rdma_end_port(device);
2381
2382        sa_dev = kzalloc(sizeof *sa_dev +
2383                         (e - s + 1) * sizeof (struct ib_sa_port),
2384                         GFP_KERNEL);
2385        if (!sa_dev)
2386                return;
2387
2388        sa_dev->start_port = s;
2389        sa_dev->end_port   = e;
2390
2391        for (i = 0; i <= e - s; ++i) {
2392                spin_lock_init(&sa_dev->port[i].ah_lock);
2393                if (!rdma_cap_ib_sa(device, i + 1))
2394                        continue;
2395
2396                sa_dev->port[i].sm_ah    = NULL;
2397                sa_dev->port[i].port_num = i + s;
2398
2399                spin_lock_init(&sa_dev->port[i].classport_lock);
2400                sa_dev->port[i].classport_info.valid = false;
2401
2402                sa_dev->port[i].agent =
2403                        ib_register_mad_agent(device, i + s, IB_QPT_GSI,
2404                                              NULL, 0, send_handler,
2405                                              recv_handler, sa_dev, 0);
2406                if (IS_ERR(sa_dev->port[i].agent))
2407                        goto err;
2408
2409                INIT_WORK(&sa_dev->port[i].update_task, update_sm_ah);
2410                INIT_DELAYED_WORK(&sa_dev->port[i].ib_cpi_work,
2411                                  update_ib_cpi);
2412
2413                count++;
2414        }
2415
2416        if (!count)
2417                goto free;
2418
2419        ib_set_client_data(device, &sa_client, sa_dev);
2420
2421        /*
2422         * We register our event handler after everything is set up,
2423         * and then update our cached info after the event handler is
2424         * registered to avoid any problems if a port changes state
2425         * during our initialization.
2426         */
2427
2428        INIT_IB_EVENT_HANDLER(&sa_dev->event_handler, device, ib_sa_event);
2429        ib_register_event_handler(&sa_dev->event_handler);
2430
2431        for (i = 0; i <= e - s; ++i) {
2432                if (rdma_cap_ib_sa(device, i + 1))
2433                        update_sm_ah(&sa_dev->port[i].update_task);
2434        }
2435
2436        return;
2437
2438err:
2439        while (--i >= 0) {
2440                if (rdma_cap_ib_sa(device, i + 1))
2441                        ib_unregister_mad_agent(sa_dev->port[i].agent);
2442        }
2443free:
2444        kfree(sa_dev);
2445        return;
2446}
2447
2448static void ib_sa_remove_one(struct ib_device *device, void *client_data)
2449{
2450        struct ib_sa_device *sa_dev = client_data;
2451        int i;
2452
2453        if (!sa_dev)
2454                return;
2455
2456        ib_unregister_event_handler(&sa_dev->event_handler);
2457        flush_workqueue(ib_wq);
2458
2459        for (i = 0; i <= sa_dev->end_port - sa_dev->start_port; ++i) {
2460                if (rdma_cap_ib_sa(device, i + 1)) {
2461                        cancel_delayed_work_sync(&sa_dev->port[i].ib_cpi_work);
2462                        ib_unregister_mad_agent(sa_dev->port[i].agent);
2463                        if (sa_dev->port[i].sm_ah)
2464                                kref_put(&sa_dev->port[i].sm_ah->ref, free_sm_ah);
2465                }
2466
2467        }
2468
2469        kfree(sa_dev);
2470}
2471
2472int ib_sa_init(void)
2473{
2474        int ret;
2475
2476        get_random_bytes(&tid, sizeof tid);
2477
2478        atomic_set(&ib_nl_sa_request_seq, 0);
2479
2480        ret = ib_register_client(&sa_client);
2481        if (ret) {
2482                pr_err("Couldn't register ib_sa client\n");
2483                goto err1;
2484        }
2485
2486        ret = mcast_init();
2487        if (ret) {
2488                pr_err("Couldn't initialize multicast handling\n");
2489                goto err2;
2490        }
2491
2492        ib_nl_wq = alloc_ordered_workqueue("ib_nl_sa_wq", WQ_MEM_RECLAIM);
2493        if (!ib_nl_wq) {
2494                ret = -ENOMEM;
2495                goto err3;
2496        }
2497
2498        INIT_DELAYED_WORK(&ib_nl_timed_work, ib_nl_request_timeout);
2499
2500        return 0;
2501
2502err3:
2503        mcast_cleanup();
2504err2:
2505        ib_unregister_client(&sa_client);
2506err1:
2507        return ret;
2508}
2509
2510void ib_sa_cleanup(void)
2511{
2512        cancel_delayed_work(&ib_nl_timed_work);
2513        flush_workqueue(ib_nl_wq);
2514        destroy_workqueue(ib_nl_wq);
2515        mcast_cleanup();
2516        ib_unregister_client(&sa_client);
2517        idr_destroy(&query_idr);
2518}
2519