drivers/infiniband/hw/mlx4/main.c

   1 /*
   2  * Copyright (c) 2006, 2007 Cisco Systems, Inc. All rights reserved.
   3  * Copyright (c) 2007, 2008 Mellanox Technologies. All rights reserved.
   4  *
   5  * This software is available to you under a choice of one of two
   6  * licenses.  You may choose to be licensed under the terms of the GNU
   7  * General Public License (GPL) Version 2, available from the file
   8  * COPYING in the main directory of this source tree, or the
   9  * OpenIB.org BSD license below:
  10  *
  11  *     Redistribution and use in source and binary forms, with or
  12  *     without modification, are permitted provided that the following
  13  *     conditions are met:
  14  *
  15  *      - Redistributions of source code must retain the above
  16  *        copyright notice, this list of conditions and the following
  17  *        disclaimer.
  18  *
  19  *      - Redistributions in binary form must reproduce the above
  20  *        copyright notice, this list of conditions and the following
  21  *        disclaimer in the documentation and/or other materials
  22  *        provided with the distribution.
  23  *
  24  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
  25  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  26  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
  27  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
  28  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
  29  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
  30  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  31  * SOFTWARE.
  32  */
  33
  34 #include <linux/module.h>
  35 #include <linux/init.h>
  36 #include <linux/slab.h>
  37 #include <linux/errno.h>
  38 #include <linux/netdevice.h>
  39 #include <linux/inetdevice.h>
  40 #include <linux/rtnetlink.h>
  41 #include <linux/if_vlan.h>
  42 #include <linux/sched/mm.h>
  43 #include <linux/sched/task.h>
  44
  45 #include <net/ipv6.h>
  46 #include <net/addrconf.h>
  47 #include <net/devlink.h>
  48
  49 #include <rdma/ib_smi.h>
  50 #include <rdma/ib_user_verbs.h>
  51 #include <rdma/ib_addr.h>
  52 #include <rdma/ib_cache.h>
  53
  54 #include <net/bonding.h>
  55
  56 #include <linux/mlx4/driver.h>
  57 #include <linux/mlx4/cmd.h>
  58 #include <linux/mlx4/qp.h>
  59
  60 #include "mlx4_ib.h"
  61 #include <rdma/mlx4-abi.h>
  62
  63 #define DRV_NAME        MLX4_IB_DRV_NAME
  64 #define DRV_VERSION     "4.0-0"
  65
  66 #define MLX4_IB_FLOW_MAX_PRIO 0xFFF
  67 #define MLX4_IB_FLOW_QPN_MASK 0xFFFFFF
  68 #define MLX4_IB_CARD_REV_A0   0xA0
  69
  70 MODULE_AUTHOR("Roland Dreier");
  71 MODULE_DESCRIPTION("Mellanox ConnectX HCA InfiniBand driver");
  72 MODULE_LICENSE("Dual BSD/GPL");
  73
  74 int mlx4_ib_sm_guid_assign = 0;
  75 module_param_named(sm_guid_assign, mlx4_ib_sm_guid_assign, int, 0444);
  76 MODULE_PARM_DESC(sm_guid_assign, "Enable SM alias_GUID assignment if sm_guid_assign > 0 (Default: 0)");
  77
  78 static const char mlx4_ib_version[] =
  79         DRV_NAME ": Mellanox ConnectX InfiniBand driver v"
  80         DRV_VERSION "\n";
  81
  82 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init);
  83 static enum rdma_link_layer mlx4_ib_port_link_layer(struct ib_device *device,
  84                                                     u8 port_num);
  85
  86 static struct workqueue_struct *wq;
  87
  88 static void init_query_mad(struct ib_smp *mad)
  89 {
  90         mad->base_version  = 1;
  91         mad->mgmt_class    = IB_MGMT_CLASS_SUBN_LID_ROUTED;
  92         mad->class_version = 1;
  93         mad->method        = IB_MGMT_METHOD_GET;
  94 }
  95
  96 static int check_flow_steering_support(struct mlx4_dev *dev)
  97 {
  98         int eth_num_ports = 0;
  99         int ib_num_ports = 0;
 100
 101         int dmfs = dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED;
 102
 103         if (dmfs) {
 104                 int i;
 105                 mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_ETH)
 106                         eth_num_ports++;
 107                 mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
 108                         ib_num_ports++;
 109                 dmfs &= (!ib_num_ports ||
 110                          (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DMFS_IPOIB)) &&
 111                         (!eth_num_ports ||
 112                          (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_FS_EN));
 113                 if (ib_num_ports && mlx4_is_mfunc(dev)) {
 114                         pr_warn("Device managed flow steering is unavailable for IB port in multifunction env.\n");
 115                         dmfs = 0;
 116                 }
 117         }
 118         return dmfs;
 119 }
 120
 121 static int num_ib_ports(struct mlx4_dev *dev)
 122 {
 123         int ib_ports = 0;
 124         int i;
 125
 126         mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
 127                 ib_ports++;
 128
 129         return ib_ports;
 130 }
 131
 132 static struct net_device *mlx4_ib_get_netdev(struct ib_device *device, u8 port_num)
 133 {
 134         struct mlx4_ib_dev *ibdev = to_mdev(device);
 135         struct net_device *dev;
 136
 137         rcu_read_lock();
 138         dev = mlx4_get_protocol_dev(ibdev->dev, MLX4_PROT_ETH, port_num);
 139
 140         if (dev) {
 141                 if (mlx4_is_bonded(ibdev->dev)) {
 142                         struct net_device *upper = NULL;
 143
 144                         upper = netdev_master_upper_dev_get_rcu(dev);
 145                         if (upper) {
 146                                 struct net_device *active;
 147
 148                                 active = bond_option_active_slave_get_rcu(netdev_priv(upper));
 149                                 if (active)
 150                                         dev = active;
 151                         }
 152                 }
 153         }
 154         if (dev)
 155                 dev_hold(dev);
 156
 157         rcu_read_unlock();
 158         return dev;
 159 }
 160
 161 static int mlx4_ib_update_gids_v1(struct gid_entry *gids,
 162                                   struct mlx4_ib_dev *ibdev,
 163                                   u8 port_num)
 164 {
 165         struct mlx4_cmd_mailbox *mailbox;
 166         int err;
 167         struct mlx4_dev *dev = ibdev->dev;
 168         int i;
 169         union ib_gid *gid_tbl;
 170
 171         mailbox = mlx4_alloc_cmd_mailbox(dev);
 172         if (IS_ERR(mailbox))
 173                 return -ENOMEM;
 174
 175         gid_tbl = mailbox->buf;
 176
 177         for (i = 0; i < MLX4_MAX_PORT_GIDS; ++i)
 178                 memcpy(&gid_tbl[i], &gids[i].gid, sizeof(union ib_gid));
 179
 180         err = mlx4_cmd(dev, mailbox->dma,
 181                        MLX4_SET_PORT_GID_TABLE << 8 | port_num,
 182                        1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
 183                        MLX4_CMD_WRAPPED);
 184         if (mlx4_is_bonded(dev))
 185                 err += mlx4_cmd(dev, mailbox->dma,
 186                                 MLX4_SET_PORT_GID_TABLE << 8 | 2,
 187                                 1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
 188                                 MLX4_CMD_WRAPPED);
 189
 190         mlx4_free_cmd_mailbox(dev, mailbox);
 191         return err;
 192 }
 193
 194 static int mlx4_ib_update_gids_v1_v2(struct gid_entry *gids,
 195                                      struct mlx4_ib_dev *ibdev,
 196                                      u8 port_num)
 197 {
 198         struct mlx4_cmd_mailbox *mailbox;
 199         int err;
 200         struct mlx4_dev *dev = ibdev->dev;
 201         int i;
 202         struct {
 203                 union ib_gid    gid;
 204                 __be32          rsrvd1[2];
 205                 __be16          rsrvd2;
 206                 u8              type;
 207                 u8              version;
 208                 __be32          rsrvd3;
 209         } *gid_tbl;
 210
 211         mailbox = mlx4_alloc_cmd_mailbox(dev);
 212         if (IS_ERR(mailbox))
 213                 return -ENOMEM;
 214
 215         gid_tbl = mailbox->buf;
 216         for (i = 0; i < MLX4_MAX_PORT_GIDS; ++i) {
 217                 memcpy(&gid_tbl[i].gid, &gids[i].gid, sizeof(union ib_gid));
 218                 if (gids[i].gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP) {
 219                         gid_tbl[i].version = 2;
 220                         if (!ipv6_addr_v4mapped((struct in6_addr *)&gids[i].gid))
 221                                 gid_tbl[i].type = 1;
 222                 }
 223         }
 224
 225         err = mlx4_cmd(dev, mailbox->dma,
 226                        MLX4_SET_PORT_ROCE_ADDR << 8 | port_num,
 227                        1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
 228                        MLX4_CMD_WRAPPED);
 229         if (mlx4_is_bonded(dev))
 230                 err += mlx4_cmd(dev, mailbox->dma,
 231                                 MLX4_SET_PORT_ROCE_ADDR << 8 | 2,
 232                                 1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
 233                                 MLX4_CMD_WRAPPED);
 234
 235         mlx4_free_cmd_mailbox(dev, mailbox);
 236         return err;
 237 }
 238
 239 static int mlx4_ib_update_gids(struct gid_entry *gids,
 240                                struct mlx4_ib_dev *ibdev,
 241                                u8 port_num)
 242 {
 243         if (ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ROCE_V1_V2)
 244                 return mlx4_ib_update_gids_v1_v2(gids, ibdev, port_num);
 245
 246         return mlx4_ib_update_gids_v1(gids, ibdev, port_num);
 247 }
 248
 249 static int mlx4_ib_add_gid(const struct ib_gid_attr *attr, void **context)
 250 {
 251         struct mlx4_ib_dev *ibdev = to_mdev(attr->device);
 252         struct mlx4_ib_iboe *iboe = &ibdev->iboe;
 253         struct mlx4_port_gid_table   *port_gid_table;
 254         int free = -1, found = -1;
 255         int ret = 0;
 256         int hw_update = 0;
 257         int i;
 258         struct gid_entry *gids = NULL;
 259
 260         if (!rdma_cap_roce_gid_table(attr->device, attr->port_num))
 261                 return -EINVAL;
 262
 263         if (attr->port_num > MLX4_MAX_PORTS)
 264                 return -EINVAL;
 265
 266         if (!context)
 267                 return -EINVAL;
 268
 269         port_gid_table = &iboe->gids[attr->port_num - 1];
 270         spin_lock_bh(&iboe->lock);
 271         for (i = 0; i < MLX4_MAX_PORT_GIDS; ++i) {
 272                 if (!memcmp(&port_gid_table->gids[i].gid,
 273                             &attr->gid, sizeof(attr->gid)) &&
 274                     port_gid_table->gids[i].gid_type == attr->gid_type)  {
 275                         found = i;
 276                         break;
 277                 }
 278                 if (free < 0 && rdma_is_zero_gid(&port_gid_table->gids[i].gid))
 279                         free = i; /* HW has space */
 280         }
 281
 282         if (found < 0) {
 283                 if (free < 0) {
 284                         ret = -ENOSPC;
 285                 } else {
 286                         port_gid_table->gids[free].ctx = kmalloc(sizeof(*port_gid_table->gids[free].ctx), GFP_ATOMIC);
 287                         if (!port_gid_table->gids[free].ctx) {
 288                                 ret = -ENOMEM;
 289                         } else {
 290                                 *context = port_gid_table->gids[free].ctx;
 291                                 memcpy(&port_gid_table->gids[free].gid,
 292                                        &attr->gid, sizeof(attr->gid));
 293                                 port_gid_table->gids[free].gid_type = attr->gid_type;
 294                                 port_gid_table->gids[free].ctx->real_index = free;
 295                                 port_gid_table->gids[free].ctx->refcount = 1;
 296                                 hw_update = 1;
 297                         }
 298                 }
 299         } else {
 300                 struct gid_cache_context *ctx = port_gid_table->gids[found].ctx;
 301                 *context = ctx;
 302                 ctx->refcount++;
 303         }
 304         if (!ret && hw_update) {
 305                 gids = kmalloc_array(MLX4_MAX_PORT_GIDS, sizeof(*gids),
 306                                      GFP_ATOMIC);
 307                 if (!gids) {
 308                         ret = -ENOMEM;
 309                 } else {
 310                         for (i = 0; i < MLX4_MAX_PORT_GIDS; i++) {
 311                                 memcpy(&gids[i].gid, &port_gid_table->gids[i].gid, sizeof(union ib_gid));
 312                                 gids[i].gid_type = port_gid_table->gids[i].gid_type;
 313                         }
 314                 }
 315         }
 316         spin_unlock_bh(&iboe->lock);
 317
 318         if (!ret && hw_update) {
 319                 ret = mlx4_ib_update_gids(gids, ibdev, attr->port_num);
 320                 kfree(gids);
 321         }
 322
 323         return ret;
 324 }
 325
 326 static int mlx4_ib_del_gid(const struct ib_gid_attr *attr, void **context)
 327 {
 328         struct gid_cache_context *ctx = *context;
 329         struct mlx4_ib_dev *ibdev = to_mdev(attr->device);
 330         struct mlx4_ib_iboe *iboe = &ibdev->iboe;
 331         struct mlx4_port_gid_table   *port_gid_table;
 332         int ret = 0;
 333         int hw_update = 0;
 334         struct gid_entry *gids = NULL;
 335
 336         if (!rdma_cap_roce_gid_table(attr->device, attr->port_num))
 337                 return -EINVAL;
 338
 339         if (attr->port_num > MLX4_MAX_PORTS)
 340                 return -EINVAL;
 341
 342         port_gid_table = &iboe->gids[attr->port_num - 1];
 343         spin_lock_bh(&iboe->lock);
 344         if (ctx) {
 345                 ctx->refcount--;
 346                 if (!ctx->refcount) {
 347                         unsigned int real_index = ctx->real_index;
 348
 349                         memset(&port_gid_table->gids[real_index].gid, 0,
 350                                sizeof(port_gid_table->gids[real_index].gid));
 351                         kfree(port_gid_table->gids[real_index].ctx);
 352                         port_gid_table->gids[real_index].ctx = NULL;
 353                         hw_update = 1;
 354                 }
 355         }
 356         if (!ret && hw_update) {
 357                 int i;
 358
 359                 gids = kmalloc_array(MLX4_MAX_PORT_GIDS, sizeof(*gids),
 360                                      GFP_ATOMIC);
 361                 if (!gids) {
 362                         ret = -ENOMEM;
 363                 } else {
 364                         for (i = 0; i < MLX4_MAX_PORT_GIDS; i++) {
 365                                 memcpy(&gids[i].gid,
 366                                        &port_gid_table->gids[i].gid,
 367                                        sizeof(union ib_gid));
 368                                 gids[i].gid_type =
 369                                     port_gid_table->gids[i].gid_type;
 370                         }
 371                 }
 372         }
 373         spin_unlock_bh(&iboe->lock);
 374
 375         if (!ret && hw_update) {
 376                 ret = mlx4_ib_update_gids(gids, ibdev, attr->port_num);
 377                 kfree(gids);
 378         }
 379         return ret;
 380 }
 381
 382 int mlx4_ib_gid_index_to_real_index(struct mlx4_ib_dev *ibdev,
 383                                     const struct ib_gid_attr *attr)
 384 {
 385         struct mlx4_ib_iboe *iboe = &ibdev->iboe;
 386         struct gid_cache_context *ctx = NULL;
 387         struct mlx4_port_gid_table   *port_gid_table;
 388         int real_index = -EINVAL;
 389         int i;
 390         unsigned long flags;
 391         u8 port_num = attr->port_num;
 392
 393         if (port_num > MLX4_MAX_PORTS)
 394                 return -EINVAL;
 395
 396         if (mlx4_is_bonded(ibdev->dev))
 397                 port_num = 1;
 398
 399         if (!rdma_cap_roce_gid_table(&ibdev->ib_dev, port_num))
 400                 return attr->index;
 401
 402         spin_lock_irqsave(&iboe->lock, flags);
 403         port_gid_table = &iboe->gids[port_num - 1];
 404
 405         for (i = 0; i < MLX4_MAX_PORT_GIDS; ++i)
 406                 if (!memcmp(&port_gid_table->gids[i].gid,
 407                             &attr->gid, sizeof(attr->gid)) &&
 408                     attr->gid_type == port_gid_table->gids[i].gid_type) {
 409                         ctx = port_gid_table->gids[i].ctx;
 410                         break;
 411                 }
 412         if (ctx)
 413                 real_index = ctx->real_index;
 414         spin_unlock_irqrestore(&iboe->lock, flags);
 415         return real_index;
 416 }
 417
 418 #define field_avail(type, fld, sz) (offsetof(type, fld) + \
 419                                     sizeof(((type *)0)->fld) <= (sz))
 420
 421 static int mlx4_ib_query_device(struct ib_device *ibdev,
 422                                 struct ib_device_attr *props,
 423                                 struct ib_udata *uhw)
 424 {
 425         struct mlx4_ib_dev *dev = to_mdev(ibdev);
 426         struct ib_smp *in_mad  = NULL;
 427         struct ib_smp *out_mad = NULL;
 428         int err;
 429         int have_ib_ports;
 430         struct mlx4_uverbs_ex_query_device cmd;
 431         struct mlx4_uverbs_ex_query_device_resp resp = {.comp_mask = 0};
 432         struct mlx4_clock_params clock_params;
 433
 434         if (uhw->inlen) {
 435                 if (uhw->inlen < sizeof(cmd))
 436                         return -EINVAL;
 437
 438                 err = ib_copy_from_udata(&cmd, uhw, sizeof(cmd));
 439                 if (err)
 440                         return err;
 441
 442                 if (cmd.comp_mask)
 443                         return -EINVAL;
 444
 445                 if (cmd.reserved)
 446                         return -EINVAL;
 447         }
 448
 449         resp.response_length = offsetof(typeof(resp), response_length) +
 450                 sizeof(resp.response_length);
 451         in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
 452         out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
 453         err = -ENOMEM;
 454         if (!in_mad || !out_mad)
 455                 goto out;
 456
 457         init_query_mad(in_mad);
 458         in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
 459
 460         err = mlx4_MAD_IFC(to_mdev(ibdev), MLX4_MAD_IFC_IGNORE_KEYS,
 461                            1, NULL, NULL, in_mad, out_mad);
 462         if (err)
 463                 goto out;
 464
 465         memset(props, 0, sizeof *props);
 466
 467         have_ib_ports = num_ib_ports(dev->dev);
 468
 469         props->fw_ver = dev->dev->caps.fw_ver;
 470         props->device_cap_flags    = IB_DEVICE_CHANGE_PHY_PORT |
 471                 IB_DEVICE_PORT_ACTIVE_EVENT             |
 472                 IB_DEVICE_SYS_IMAGE_GUID                |
 473                 IB_DEVICE_RC_RNR_NAK_GEN                |
 474                 IB_DEVICE_BLOCK_MULTICAST_LOOPBACK;
 475         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_PKEY_CNTR)
 476                 props->device_cap_flags |= IB_DEVICE_BAD_PKEY_CNTR;
 477         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_QKEY_CNTR)
 478                 props->device_cap_flags |= IB_DEVICE_BAD_QKEY_CNTR;
 479         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_APM && have_ib_ports)
 480                 props->device_cap_flags |= IB_DEVICE_AUTO_PATH_MIG;
 481         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_UD_AV_PORT)
 482                 props->device_cap_flags |= IB_DEVICE_UD_AV_PORT_ENFORCE;
 483         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_IPOIB_CSUM)
 484                 props->device_cap_flags |= IB_DEVICE_UD_IP_CSUM;
 485         if (dev->dev->caps.max_gso_sz &&
 486             (dev->dev->rev_id != MLX4_IB_CARD_REV_A0) &&
 487             (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BLH))
 488                 props->device_cap_flags |= IB_DEVICE_UD_TSO;
 489         if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_RESERVED_LKEY)
 490                 props->device_cap_flags |= IB_DEVICE_LOCAL_DMA_LKEY;
 491         if ((dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_LOCAL_INV) &&
 492             (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_REMOTE_INV) &&
 493             (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_FAST_REG_WR))
 494                 props->device_cap_flags |= IB_DEVICE_MEM_MGT_EXTENSIONS;
 495         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC)
 496                 props->device_cap_flags |= IB_DEVICE_XRC;
 497         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW)
 498                 props->device_cap_flags |= IB_DEVICE_MEM_WINDOW;
 499         if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
 500                 if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_WIN_TYPE_2B)
 501                         props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2B;
 502                 else
 503                         props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2A;
 504         }
 505         if (dev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED)
 506                 props->device_cap_flags |= IB_DEVICE_MANAGED_FLOW_STEERING;
 507
 508         props->device_cap_flags |= IB_DEVICE_RAW_IP_CSUM;
 509
 510         props->vendor_id           = be32_to_cpup((__be32 *) (out_mad->data + 36)) &
 511                 0xffffff;
 512         props->vendor_part_id      = dev->dev->persist->pdev->device;
 513         props->hw_ver              = be32_to_cpup((__be32 *) (out_mad->data + 32));
 514         memcpy(&props->sys_image_guid, out_mad->data +  4, 8);
 515
 516         props->max_mr_size         = ~0ull;
 517         props->page_size_cap       = dev->dev->caps.page_size_cap;
 518         props->max_qp              = dev->dev->quotas.qp;
 519         props->max_qp_wr           = dev->dev->caps.max_wqes - MLX4_IB_SQ_MAX_SPARE;
 520         props->max_send_sge        = dev->dev->caps.max_sq_sg;
 521         props->max_recv_sge        = dev->dev->caps.max_rq_sg;
 522         props->max_sge_rd          = MLX4_MAX_SGE_RD;
 523         props->max_cq              = dev->dev->quotas.cq;
 524         props->max_cqe             = dev->dev->caps.max_cqes;
 525         props->max_mr              = dev->dev->quotas.mpt;
 526         props->max_pd              = dev->dev->caps.num_pds - dev->dev->caps.reserved_pds;
 527         props->max_qp_rd_atom      = dev->dev->caps.max_qp_dest_rdma;
 528         props->max_qp_init_rd_atom = dev->dev->caps.max_qp_init_rdma;
 529         props->max_res_rd_atom     = props->max_qp_rd_atom * props->max_qp;
 530         props->max_srq             = dev->dev->quotas.srq;
 531         props->max_srq_wr          = dev->dev->caps.max_srq_wqes - 1;
 532         props->max_srq_sge         = dev->dev->caps.max_srq_sge;
 533         props->max_fast_reg_page_list_len = MLX4_MAX_FAST_REG_PAGES;
 534         props->local_ca_ack_delay  = dev->dev->caps.local_ca_ack_delay;
 535         props->atomic_cap          = dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_ATOMIC ?
 536                 IB_ATOMIC_HCA : IB_ATOMIC_NONE;
 537         props->masked_atomic_cap   = props->atomic_cap;
 538         props->max_pkeys           = dev->dev->caps.pkey_table_len[1];
 539         props->max_mcast_grp       = dev->dev->caps.num_mgms + dev->dev->caps.num_amgms;
 540         props->max_mcast_qp_attach = dev->dev->caps.num_qp_per_mgm;
 541         props->max_total_mcast_qp_attach = props->max_mcast_qp_attach *
 542                                            props->max_mcast_grp;
 543         props->max_map_per_fmr = dev->dev->caps.max_fmr_maps;
 544         props->hca_core_clock = dev->dev->caps.hca_core_clock * 1000UL;
 545         props->timestamp_mask = 0xFFFFFFFFFFFFULL;
 546         props->max_ah = INT_MAX;
 547
 548         if (mlx4_ib_port_link_layer(ibdev, 1) == IB_LINK_LAYER_ETHERNET ||
 549             mlx4_ib_port_link_layer(ibdev, 2) == IB_LINK_LAYER_ETHERNET) {
 550                 if (dev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_RSS) {
 551                         props->rss_caps.max_rwq_indirection_tables =
 552                                 props->max_qp;
 553                         props->rss_caps.max_rwq_indirection_table_size =
 554                                 dev->dev->caps.max_rss_tbl_sz;
 555                         props->rss_caps.supported_qpts = 1 << IB_QPT_RAW_PACKET;
 556                         props->max_wq_type_rq = props->max_qp;
 557                 }
 558
 559                 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_FCS_KEEP)
 560                         props->raw_packet_caps |= IB_RAW_PACKET_CAP_SCATTER_FCS;
 561         }
 562
 563         props->cq_caps.max_cq_moderation_count = MLX4_MAX_CQ_COUNT;
 564         props->cq_caps.max_cq_moderation_period = MLX4_MAX_CQ_PERIOD;
 565
 566         if (!mlx4_is_slave(dev->dev))
 567                 err = mlx4_get_internal_clock_params(dev->dev, &clock_params);
 568
 569         if (uhw->outlen >= resp.response_length + sizeof(resp.hca_core_clock_offset)) {
 570                 resp.response_length += sizeof(resp.hca_core_clock_offset);
 571                 if (!err && !mlx4_is_slave(dev->dev)) {
 572                         resp.comp_mask |= MLX4_IB_QUERY_DEV_RESP_MASK_CORE_CLOCK_OFFSET;
 573                         resp.hca_core_clock_offset = clock_params.offset % PAGE_SIZE;
 574                 }
 575         }
 576
 577         if (uhw->outlen >= resp.response_length +
 578             sizeof(resp.max_inl_recv_sz)) {
 579                 resp.response_length += sizeof(resp.max_inl_recv_sz);
 580                 resp.max_inl_recv_sz  = dev->dev->caps.max_rq_sg *
 581                         sizeof(struct mlx4_wqe_data_seg);
 582         }
 583
 584         if (field_avail(typeof(resp), rss_caps, uhw->outlen)) {
 585                 if (props->rss_caps.supported_qpts) {
 586                         resp.rss_caps.rx_hash_function =
 587                                 MLX4_IB_RX_HASH_FUNC_TOEPLITZ;
 588
 589                         resp.rss_caps.rx_hash_fields_mask =
 590                                 MLX4_IB_RX_HASH_SRC_IPV4 |
 591                                 MLX4_IB_RX_HASH_DST_IPV4 |
 592                                 MLX4_IB_RX_HASH_SRC_IPV6 |
 593                                 MLX4_IB_RX_HASH_DST_IPV6 |
 594                                 MLX4_IB_RX_HASH_SRC_PORT_TCP |
 595                                 MLX4_IB_RX_HASH_DST_PORT_TCP |
 596                                 MLX4_IB_RX_HASH_SRC_PORT_UDP |
 597                                 MLX4_IB_RX_HASH_DST_PORT_UDP;
 598
 599                         if (dev->dev->caps.tunnel_offload_mode ==
 600                             MLX4_TUNNEL_OFFLOAD_MODE_VXLAN)
 601                                 resp.rss_caps.rx_hash_fields_mask |=
 602                                         MLX4_IB_RX_HASH_INNER;
 603                 }
 604                 resp.response_length = offsetof(typeof(resp), rss_caps) +
 605                                        sizeof(resp.rss_caps);
 606         }
 607
 608         if (field_avail(typeof(resp), tso_caps, uhw->outlen)) {
 609                 if (dev->dev->caps.max_gso_sz &&
 610                     ((mlx4_ib_port_link_layer(ibdev, 1) ==
 611                     IB_LINK_LAYER_ETHERNET) ||
 612                     (mlx4_ib_port_link_layer(ibdev, 2) ==
 613                     IB_LINK_LAYER_ETHERNET))) {
 614                         resp.tso_caps.max_tso = dev->dev->caps.max_gso_sz;
 615                         resp.tso_caps.supported_qpts |=
 616                                 1 << IB_QPT_RAW_PACKET;
 617                 }
 618                 resp.response_length = offsetof(typeof(resp), tso_caps) +
 619                                        sizeof(resp.tso_caps);
 620         }
 621
 622         if (uhw->outlen) {
 623                 err = ib_copy_to_udata(uhw, &resp, resp.response_length);
 624                 if (err)
 625                         goto out;
 626         }
 627 out:
 628         kfree(in_mad);
 629         kfree(out_mad);
 630
 631         return err;
 632 }
 633
 634 static enum rdma_link_layer
 635 mlx4_ib_port_link_layer(struct ib_device *device, u8 port_num)
 636 {
 637         struct mlx4_dev *dev = to_mdev(device)->dev;
 638
 639         return dev->caps.port_mask[port_num] == MLX4_PORT_TYPE_IB ?
 640                 IB_LINK_LAYER_INFINIBAND : IB_LINK_LAYER_ETHERNET;
 641 }
 642
 643 static int ib_link_query_port(struct ib_device *ibdev, u8 port,
 644                               struct ib_port_attr *props, int netw_view)
 645 {
 646         struct ib_smp *in_mad  = NULL;
 647         struct ib_smp *out_mad = NULL;
 648         int ext_active_speed;
 649         int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
 650         int err = -ENOMEM;
 651
 652         in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
 653         out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
 654         if (!in_mad || !out_mad)
 655                 goto out;
 656
 657         init_query_mad(in_mad);
 658         in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
 659         in_mad->attr_mod = cpu_to_be32(port);
 660
 661         if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
 662                 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
 663
 664         err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
 665                                 in_mad, out_mad);
 666         if (err)
 667                 goto out;
 668
 669
 670         props->lid              = be16_to_cpup((__be16 *) (out_mad->data + 16));
 671         props->lmc              = out_mad->data[34] & 0x7;
 672         props->sm_lid           = be16_to_cpup((__be16 *) (out_mad->data + 18));
 673         props->sm_sl            = out_mad->data[36] & 0xf;
 674         props->state            = out_mad->data[32] & 0xf;
 675         props->phys_state       = out_mad->data[33] >> 4;
 676         props->port_cap_flags   = be32_to_cpup((__be32 *) (out_mad->data + 20));
 677         if (netw_view)
 678                 props->gid_tbl_len = out_mad->data[50];
 679         else
 680                 props->gid_tbl_len = to_mdev(ibdev)->dev->caps.gid_table_len[port];
 681         props->max_msg_sz       = to_mdev(ibdev)->dev->caps.max_msg_sz;
 682         props->pkey_tbl_len     = to_mdev(ibdev)->dev->caps.pkey_table_len[port];
 683         props->bad_pkey_cntr    = be16_to_cpup((__be16 *) (out_mad->data + 46));
 684         props->qkey_viol_cntr   = be16_to_cpup((__be16 *) (out_mad->data + 48));
 685         props->active_width     = out_mad->data[31] & 0xf;
 686         props->active_speed     = out_mad->data[35] >> 4;
 687         props->max_mtu          = out_mad->data[41] & 0xf;
 688         props->active_mtu       = out_mad->data[36] >> 4;
 689         props->subnet_timeout   = out_mad->data[51] & 0x1f;
 690         props->max_vl_num       = out_mad->data[37] >> 4;
 691         props->init_type_reply  = out_mad->data[41] >> 4;
 692
 693         /* Check if extended speeds (EDR/FDR/...) are supported */
 694         if (props->port_cap_flags & IB_PORT_EXTENDED_SPEEDS_SUP) {
 695                 ext_active_speed = out_mad->data[62] >> 4;
 696
 697                 switch (ext_active_speed) {
 698                 case 1:
 699                         props->active_speed = IB_SPEED_FDR;
 700                         break;
 701                 case 2:
 702                         props->active_speed = IB_SPEED_EDR;
 703                         break;
 704                 }
 705         }
 706
 707         /* If reported active speed is QDR, check if is FDR-10 */
 708         if (props->active_speed == IB_SPEED_QDR) {
 709                 init_query_mad(in_mad);
 710                 in_mad->attr_id = MLX4_ATTR_EXTENDED_PORT_INFO;
 711                 in_mad->attr_mod = cpu_to_be32(port);
 712
 713                 err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port,
 714                                    NULL, NULL, in_mad, out_mad);
 715                 if (err)
 716                         goto out;
 717
 718                 /* Checking LinkSpeedActive for FDR-10 */
 719                 if (out_mad->data[15] & 0x1)
 720                         props->active_speed = IB_SPEED_FDR10;
 721         }
 722
 723         /* Avoid wrong speed value returned by FW if the IB link is down. */
 724         if (props->state == IB_PORT_DOWN)
 725                  props->active_speed = IB_SPEED_SDR;
 726
 727 out:
 728         kfree(in_mad);
 729         kfree(out_mad);
 730         return err;
 731 }
 732
 733 static u8 state_to_phys_state(enum ib_port_state state)
 734 {
 735         return state == IB_PORT_ACTIVE ? 5 : 3;
 736 }
 737
 738 static int eth_link_query_port(struct ib_device *ibdev, u8 port,
 739                                struct ib_port_attr *props)
 740 {
 741
 742         struct mlx4_ib_dev *mdev = to_mdev(ibdev);
 743         struct mlx4_ib_iboe *iboe = &mdev->iboe;
 744         struct net_device *ndev;
 745         enum ib_mtu tmp;
 746         struct mlx4_cmd_mailbox *mailbox;
 747         int err = 0;
 748         int is_bonded = mlx4_is_bonded(mdev->dev);
 749
 750         mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
 751         if (IS_ERR(mailbox))
 752                 return PTR_ERR(mailbox);
 753
 754         err = mlx4_cmd_box(mdev->dev, 0, mailbox->dma, port, 0,
 755                            MLX4_CMD_QUERY_PORT, MLX4_CMD_TIME_CLASS_B,
 756                            MLX4_CMD_WRAPPED);
 757         if (err)
 758                 goto out;
 759
 760         props->active_width     =  (((u8 *)mailbox->buf)[5] == 0x40) ||
 761                                    (((u8 *)mailbox->buf)[5] == 0x20 /*56Gb*/) ?
 762                                            IB_WIDTH_4X : IB_WIDTH_1X;
 763         props->active_speed     =  (((u8 *)mailbox->buf)[5] == 0x20 /*56Gb*/) ?
 764                                            IB_SPEED_FDR : IB_SPEED_QDR;
 765         props->port_cap_flags   = IB_PORT_CM_SUP;
 766         props->ip_gids = true;
 767         props->gid_tbl_len      = mdev->dev->caps.gid_table_len[port];
 768         props->max_msg_sz       = mdev->dev->caps.max_msg_sz;
 769         props->pkey_tbl_len     = 1;
 770         props->max_mtu          = IB_MTU_4096;
 771         props->max_vl_num       = 2;
 772         props->state            = IB_PORT_DOWN;
 773         props->phys_state       = state_to_phys_state(props->state);
 774         props->active_mtu       = IB_MTU_256;
 775         spin_lock_bh(&iboe->lock);
 776         ndev = iboe->netdevs[port - 1];
 777         if (ndev && is_bonded) {
 778                 rcu_read_lock(); /* required to get upper dev */
 779                 ndev = netdev_master_upper_dev_get_rcu(ndev);
 780                 rcu_read_unlock();
 781         }
 782         if (!ndev)
 783                 goto out_unlock;
 784
 785         tmp = iboe_get_mtu(ndev->mtu);
 786         props->active_mtu = tmp ? min(props->max_mtu, tmp) : IB_MTU_256;
 787
 788         props->state            = (netif_running(ndev) && netif_carrier_ok(ndev)) ?
 789                                         IB_PORT_ACTIVE : IB_PORT_DOWN;
 790         props->phys_state       = state_to_phys_state(props->state);
 791 out_unlock:
 792         spin_unlock_bh(&iboe->lock);
 793 out:
 794         mlx4_free_cmd_mailbox(mdev->dev, mailbox);
 795         return err;
 796 }
 797
 798 int __mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
 799                          struct ib_port_attr *props, int netw_view)
 800 {
 801         int err;
 802
 803         /* props being zeroed by the caller, avoid zeroing it here */
 804
 805         err = mlx4_ib_port_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND ?
 806                 ib_link_query_port(ibdev, port, props, netw_view) :
 807                                 eth_link_query_port(ibdev, port, props);
 808
 809         return err;
 810 }
 811
 812 static int mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
 813                               struct ib_port_attr *props)
 814 {
 815         /* returns host view */
 816         return __mlx4_ib_query_port(ibdev, port, props, 0);
 817 }
 818
 819 int __mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
 820                         union ib_gid *gid, int netw_view)
 821 {
 822         struct ib_smp *in_mad  = NULL;
 823         struct ib_smp *out_mad = NULL;
 824         int err = -ENOMEM;
 825         struct mlx4_ib_dev *dev = to_mdev(ibdev);
 826         int clear = 0;
 827         int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
 828
 829         in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
 830         out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
 831         if (!in_mad || !out_mad)
 832                 goto out;
 833
 834         init_query_mad(in_mad);
 835         in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
 836         in_mad->attr_mod = cpu_to_be32(port);
 837
 838         if (mlx4_is_mfunc(dev->dev) && netw_view)
 839                 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
 840
 841         err = mlx4_MAD_IFC(dev, mad_ifc_flags, port, NULL, NULL, in_mad, out_mad);
 842         if (err)
 843                 goto out;
 844
 845         memcpy(gid->raw, out_mad->data + 8, 8);
 846
 847         if (mlx4_is_mfunc(dev->dev) && !netw_view) {
 848                 if (index) {
 849                         /* For any index > 0, return the null guid */
 850                         err = 0;
 851                         clear = 1;
 852                         goto out;
 853                 }
 854         }
 855
 856         init_query_mad(in_mad);
 857         in_mad->attr_id  = IB_SMP_ATTR_GUID_INFO;
 858         in_mad->attr_mod = cpu_to_be32(index / 8);
 859
 860         err = mlx4_MAD_IFC(dev, mad_ifc_flags, port,
 861                            NULL, NULL, in_mad, out_mad);
 862         if (err)
 863                 goto out;
 864
 865         memcpy(gid->raw + 8, out_mad->data + (index % 8) * 8, 8);
 866
 867 out:
 868         if (clear)
 869                 memset(gid->raw + 8, 0, 8);
 870         kfree(in_mad);
 871         kfree(out_mad);
 872         return err;
 873 }
 874
 875 static int mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
 876                              union ib_gid *gid)
 877 {
 878         if (rdma_protocol_ib(ibdev, port))
 879                 return __mlx4_ib_query_gid(ibdev, port, index, gid, 0);
 880         return 0;
 881 }
 882
 883 static int mlx4_ib_query_sl2vl(struct ib_device *ibdev, u8 port, u64 *sl2vl_tbl)
 884 {
 885         union sl2vl_tbl_to_u64 sl2vl64;
 886         struct ib_smp *in_mad  = NULL;
 887         struct ib_smp *out_mad = NULL;
 888         int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
 889         int err = -ENOMEM;
 890         int jj;
 891
 892         if (mlx4_is_slave(to_mdev(ibdev)->dev)) {
 893                 *sl2vl_tbl = 0;
 894                 return 0;
 895         }
 896
 897         in_mad  = kzalloc(sizeof(*in_mad), GFP_KERNEL);
 898         out_mad = kmalloc(sizeof(*out_mad), GFP_KERNEL);
 899         if (!in_mad || !out_mad)
 900                 goto out;
 901
 902         init_query_mad(in_mad);
 903         in_mad->attr_id  = IB_SMP_ATTR_SL_TO_VL_TABLE;
 904         in_mad->attr_mod = 0;
 905
 906         if (mlx4_is_mfunc(to_mdev(ibdev)->dev))
 907                 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
 908
 909         err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
 910                            in_mad, out_mad);
 911         if (err)
 912                 goto out;
 913
 914         for (jj = 0; jj < 8; jj++)
 915                 sl2vl64.sl8[jj] = ((struct ib_smp *)out_mad)->data[jj];
 916         *sl2vl_tbl = sl2vl64.sl64;
 917
 918 out:
 919         kfree(in_mad);
 920         kfree(out_mad);
 921         return err;
 922 }
 923
 924 static void mlx4_init_sl2vl_tbl(struct mlx4_ib_dev *mdev)
 925 {
 926         u64 sl2vl;
 927         int i;
 928         int err;
 929
 930         for (i = 1; i <= mdev->dev->caps.num_ports; i++) {
 931                 if (mdev->dev->caps.port_type[i] == MLX4_PORT_TYPE_ETH)
 932                         continue;
 933                 err = mlx4_ib_query_sl2vl(&mdev->ib_dev, i, &sl2vl);
 934                 if (err) {
 935                         pr_err("Unable to get default sl to vl mapping for port %d.  Using all zeroes (%d)\n",
 936                                i, err);
 937                         sl2vl = 0;
 938                 }
 939                 atomic64_set(&mdev->sl2vl[i - 1], sl2vl);
 940         }
 941 }
 942
 943 int __mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index,
 944                          u16 *pkey, int netw_view)
 945 {
 946         struct ib_smp *in_mad  = NULL;
 947         struct ib_smp *out_mad = NULL;
 948         int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
 949         int err = -ENOMEM;
 950
 951         in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
 952         out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
 953         if (!in_mad || !out_mad)
 954                 goto out;
 955
 956         init_query_mad(in_mad);
 957         in_mad->attr_id  = IB_SMP_ATTR_PKEY_TABLE;
 958         in_mad->attr_mod = cpu_to_be32(index / 32);
 959
 960         if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
 961                 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
 962
 963         err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
 964                            in_mad, out_mad);
 965         if (err)
 966                 goto out;
 967
 968         *pkey = be16_to_cpu(((__be16 *) out_mad->data)[index % 32]);
 969
 970 out:
 971         kfree(in_mad);
 972         kfree(out_mad);
 973         return err;
 974 }
 975
 976 static int mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index, u16 *pkey)
 977 {
 978         return __mlx4_ib_query_pkey(ibdev, port, index, pkey, 0);
 979 }
 980
 981 static int mlx4_ib_modify_device(struct ib_device *ibdev, int mask,
 982                                  struct ib_device_modify *props)
 983 {
 984         struct mlx4_cmd_mailbox *mailbox;
 985         unsigned long flags;
 986
 987         if (mask & ~IB_DEVICE_MODIFY_NODE_DESC)
 988                 return -EOPNOTSUPP;
 989
 990         if (!(mask & IB_DEVICE_MODIFY_NODE_DESC))
 991                 return 0;
 992
 993         if (mlx4_is_slave(to_mdev(ibdev)->dev))
 994                 return -EOPNOTSUPP;
 995
 996         spin_lock_irqsave(&to_mdev(ibdev)->sm_lock, flags);
 997         memcpy(ibdev->node_desc, props->node_desc, IB_DEVICE_NODE_DESC_MAX);
 998         spin_unlock_irqrestore(&to_mdev(ibdev)->sm_lock, flags);
 999
1000         /*
1001          * If possible, pass node desc to FW, so it can generate
1002          * a 144 trap.  If cmd fails, just ignore.
1003          */
1004         mailbox = mlx4_alloc_cmd_mailbox(to_mdev(ibdev)->dev);
1005         if (IS_ERR(mailbox))
1006                 return 0;
1007
1008         memcpy(mailbox->buf, props->node_desc, IB_DEVICE_NODE_DESC_MAX);
1009         mlx4_cmd(to_mdev(ibdev)->dev, mailbox->dma, 1, 0,
1010                  MLX4_CMD_SET_NODE, MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
1011
1012         mlx4_free_cmd_mailbox(to_mdev(ibdev)->dev, mailbox);
1013
1014         return 0;
1015 }
1016
1017 static int mlx4_ib_SET_PORT(struct mlx4_ib_dev *dev, u8 port, int reset_qkey_viols,
1018                             u32 cap_mask)
1019 {
1020         struct mlx4_cmd_mailbox *mailbox;
1021         int err;
1022
1023         mailbox = mlx4_alloc_cmd_mailbox(dev->dev);
1024         if (IS_ERR(mailbox))
1025                 return PTR_ERR(mailbox);
1026
1027         if (dev->dev->flags & MLX4_FLAG_OLD_PORT_CMDS) {
1028                 *(u8 *) mailbox->buf         = !!reset_qkey_viols << 6;
1029                 ((__be32 *) mailbox->buf)[2] = cpu_to_be32(cap_mask);
1030         } else {
1031                 ((u8 *) mailbox->buf)[3]     = !!reset_qkey_viols;
1032                 ((__be32 *) mailbox->buf)[1] = cpu_to_be32(cap_mask);
1033         }
1034
1035         err = mlx4_cmd(dev->dev, mailbox->dma, port, MLX4_SET_PORT_IB_OPCODE,
1036                        MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
1037                        MLX4_CMD_WRAPPED);
1038
1039         mlx4_free_cmd_mailbox(dev->dev, mailbox);
1040         return err;
1041 }
1042
1043 static int mlx4_ib_modify_port(struct ib_device *ibdev, u8 port, int mask,
1044                                struct ib_port_modify *props)
1045 {
1046         struct mlx4_ib_dev *mdev = to_mdev(ibdev);
1047         u8 is_eth = mdev->dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH;
1048         struct ib_port_attr attr;
1049         u32 cap_mask;
1050         int err;
1051
1052         /* return OK if this is RoCE. CM calls ib_modify_port() regardless
1053          * of whether port link layer is ETH or IB. For ETH ports, qkey
1054          * violations and port capabilities are not meaningful.
1055          */
1056         if (is_eth)
1057                 return 0;
1058
1059         mutex_lock(&mdev->cap_mask_mutex);
1060
1061         err = ib_query_port(ibdev, port, &attr);
1062         if (err)
1063                 goto out;
1064
1065         cap_mask = (attr.port_cap_flags | props->set_port_cap_mask) &
1066                 ~props->clr_port_cap_mask;
1067
1068         err = mlx4_ib_SET_PORT(mdev, port,
1069                                !!(mask & IB_PORT_RESET_QKEY_CNTR),
1070                                cap_mask);
1071
1072 out:
1073         mutex_unlock(&to_mdev(ibdev)->cap_mask_mutex);
1074         return err;
1075 }
1076
1077 static struct ib_ucontext *mlx4_ib_alloc_ucontext(struct ib_device *ibdev,
1078                                                   struct ib_udata *udata)
1079 {
1080         struct mlx4_ib_dev *dev = to_mdev(ibdev);
1081         struct mlx4_ib_ucontext *context;
1082         struct mlx4_ib_alloc_ucontext_resp_v3 resp_v3;
1083         struct mlx4_ib_alloc_ucontext_resp resp;
1084         int err;
1085
1086         if (!dev->ib_active)
1087                 return ERR_PTR(-EAGAIN);
1088
1089         if (ibdev->uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION) {
1090                 resp_v3.qp_tab_size      = dev->dev->caps.num_qps;
1091                 resp_v3.bf_reg_size      = dev->dev->caps.bf_reg_size;
1092                 resp_v3.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
1093         } else {
1094                 resp.dev_caps         = dev->dev->caps.userspace_caps;
1095                 resp.qp_tab_size      = dev->dev->caps.num_qps;
1096                 resp.bf_reg_size      = dev->dev->caps.bf_reg_size;
1097                 resp.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
1098                 resp.cqe_size         = dev->dev->caps.cqe_size;
1099         }
1100
1101         context = kzalloc(sizeof(*context), GFP_KERNEL);
1102         if (!context)
1103                 return ERR_PTR(-ENOMEM);
1104
1105         err = mlx4_uar_alloc(to_mdev(ibdev)->dev, &context->uar);
1106         if (err) {
1107                 kfree(context);
1108                 return ERR_PTR(err);
1109         }
1110
1111         INIT_LIST_HEAD(&context->db_page_list);
1112         mutex_init(&context->db_page_mutex);
1113
1114         INIT_LIST_HEAD(&context->wqn_ranges_list);
1115         mutex_init(&context->wqn_ranges_mutex);
1116
1117         if (ibdev->uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION)
1118                 err = ib_copy_to_udata(udata, &resp_v3, sizeof(resp_v3));
1119         else
1120                 err = ib_copy_to_udata(udata, &resp, sizeof(resp));
1121
1122         if (err) {
1123                 mlx4_uar_free(to_mdev(ibdev)->dev, &context->uar);
1124                 kfree(context);
1125                 return ERR_PTR(-EFAULT);
1126         }
1127
1128         return &context->ibucontext;
1129 }
1130
1131 static int mlx4_ib_dealloc_ucontext(struct ib_ucontext *ibcontext)
1132 {
1133         struct mlx4_ib_ucontext *context = to_mucontext(ibcontext);
1134
1135         mlx4_uar_free(to_mdev(ibcontext->device)->dev, &context->uar);
1136         kfree(context);
1137
1138         return 0;
1139 }
1140
1141 static void  mlx4_ib_vma_open(struct vm_area_struct *area)
1142 {
1143         /* vma_open is called when a new VMA is created on top of our VMA.
1144          * This is done through either mremap flow or split_vma (usually due
1145          * to mlock, madvise, munmap, etc.). We do not support a clone of the
1146          * vma, as this VMA is strongly hardware related. Therefore we set the
1147          * vm_ops of the newly created/cloned VMA to NULL, to prevent it from
1148          * calling us again and trying to do incorrect actions. We assume that
1149          * the original vma size is exactly a single page that there will be no
1150          * "splitting" operations on.
1151          */
1152         area->vm_ops = NULL;
1153 }
1154
1155 static void  mlx4_ib_vma_close(struct vm_area_struct *area)
1156 {
1157         struct mlx4_ib_vma_private_data *mlx4_ib_vma_priv_data;
1158
1159         /* It's guaranteed that all VMAs opened on a FD are closed before the
1160          * file itself is closed, therefore no sync is needed with the regular
1161          * closing flow. (e.g. mlx4_ib_dealloc_ucontext) However need a sync
1162          * with accessing the vma as part of mlx4_ib_disassociate_ucontext.
1163          * The close operation is usually called under mm->mmap_sem except when
1164          * process is exiting.  The exiting case is handled explicitly as part
1165          * of mlx4_ib_disassociate_ucontext.
1166          */
1167         mlx4_ib_vma_priv_data = (struct mlx4_ib_vma_private_data *)
1168                                 area->vm_private_data;
1169
1170         /* set the vma context pointer to null in the mlx4_ib driver's private
1171          * data to protect against a race condition in mlx4_ib_dissassociate_ucontext().
1172          */
1173         mlx4_ib_vma_priv_data->vma = NULL;
1174 }
1175
1176 static const struct vm_operations_struct mlx4_ib_vm_ops = {
1177         .open = mlx4_ib_vma_open,
1178         .close = mlx4_ib_vma_close
1179 };
1180
1181 static void mlx4_ib_disassociate_ucontext(struct ib_ucontext *ibcontext)
1182 {
1183         int i;
1184         struct vm_area_struct *vma;
1185         struct mlx4_ib_ucontext *context = to_mucontext(ibcontext);
1186
1187         /* need to protect from a race on closing the vma as part of
1188          * mlx4_ib_vma_close().
1189          */
1190         for (i = 0; i < HW_BAR_COUNT; i++) {
1191                 vma = context->hw_bar_info[i].vma;
1192                 if (!vma)
1193                         continue;
1194
1195                 zap_vma_ptes(context->hw_bar_info[i].vma,
1196                              context->hw_bar_info[i].vma->vm_start, PAGE_SIZE);
1197
1198                 context->hw_bar_info[i].vma->vm_flags &=
1199                         ~(VM_SHARED | VM_MAYSHARE);
1200                 /* context going to be destroyed, should not access ops any more */
1201                 context->hw_bar_info[i].vma->vm_ops = NULL;
1202         }
1203 }
1204
1205 static void mlx4_ib_set_vma_data(struct vm_area_struct *vma,
1206                                  struct mlx4_ib_vma_private_data *vma_private_data)
1207 {
1208         vma_private_data->vma = vma;
1209         vma->vm_private_data = vma_private_data;
1210         vma->vm_ops =  &mlx4_ib_vm_ops;
1211 }
1212
1213 static int mlx4_ib_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
1214 {
1215         struct mlx4_ib_dev *dev = to_mdev(context->device);
1216         struct mlx4_ib_ucontext *mucontext = to_mucontext(context);
1217
1218         if (vma->vm_end - vma->vm_start != PAGE_SIZE)
1219                 return -EINVAL;
1220
1221         if (vma->vm_pgoff == 0) {
1222                 /* We prevent double mmaping on same context */
1223                 if (mucontext->hw_bar_info[HW_BAR_DB].vma)
1224                         return -EINVAL;
1225
1226                 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
1227
1228                 if (io_remap_pfn_range(vma, vma->vm_start,
1229                                        to_mucontext(context)->uar.pfn,
1230                                        PAGE_SIZE, vma->vm_page_prot))
1231                         return -EAGAIN;
1232
1233                 mlx4_ib_set_vma_data(vma, &mucontext->hw_bar_info[HW_BAR_DB]);
1234
1235         } else if (vma->vm_pgoff == 1 && dev->dev->caps.bf_reg_size != 0) {
1236                 /* We prevent double mmaping on same context */
1237                 if (mucontext->hw_bar_info[HW_BAR_BF].vma)
1238                         return -EINVAL;
1239
1240                 vma->vm_page_prot = pgprot_writecombine(vma->vm_page_prot);
1241
1242                 if (io_remap_pfn_range(vma, vma->vm_start,
1243                                        to_mucontext(context)->uar.pfn +
1244                                        dev->dev->caps.num_uars,
1245                                        PAGE_SIZE, vma->vm_page_prot))
1246                         return -EAGAIN;
1247
1248                 mlx4_ib_set_vma_data(vma, &mucontext->hw_bar_info[HW_BAR_BF]);
1249
1250         } else if (vma->vm_pgoff == 3) {
1251                 struct mlx4_clock_params params;
1252                 int ret;
1253
1254                 /* We prevent double mmaping on same context */
1255                 if (mucontext->hw_bar_info[HW_BAR_CLOCK].vma)
1256                         return -EINVAL;
1257
1258                 ret = mlx4_get_internal_clock_params(dev->dev, &params);
1259
1260                 if (ret)
1261                         return ret;
1262
1263                 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
1264                 if (io_remap_pfn_range(vma, vma->vm_start,
1265                                        (pci_resource_start(dev->dev->persist->pdev,
1266                                                            params.bar) +
1267                                         params.offset)
1268                                        >> PAGE_SHIFT,
1269                                        PAGE_SIZE, vma->vm_page_prot))
1270                         return -EAGAIN;
1271
1272                 mlx4_ib_set_vma_data(vma,
1273                                      &mucontext->hw_bar_info[HW_BAR_CLOCK]);
1274         } else {
1275                 return -EINVAL;
1276         }
1277
1278         return 0;
1279 }
1280
1281 static struct ib_pd *mlx4_ib_alloc_pd(struct ib_device *ibdev,
1282                                       struct ib_ucontext *context,
1283                                       struct ib_udata *udata)
1284 {
1285         struct mlx4_ib_pd *pd;
1286         int err;
1287
1288         pd = kzalloc(sizeof(*pd), GFP_KERNEL);
1289         if (!pd)
1290                 return ERR_PTR(-ENOMEM);
1291
1292         err = mlx4_pd_alloc(to_mdev(ibdev)->dev, &pd->pdn);
1293         if (err) {
1294                 kfree(pd);
1295                 return ERR_PTR(err);
1296         }
1297
1298         if (context)
1299                 if (ib_copy_to_udata(udata, &pd->pdn, sizeof (__u32))) {
1300                         mlx4_pd_free(to_mdev(ibdev)->dev, pd->pdn);
1301                         kfree(pd);
1302                         return ERR_PTR(-EFAULT);
1303                 }
1304         return &pd->ibpd;
1305 }
1306
1307 static int mlx4_ib_dealloc_pd(struct ib_pd *pd)
1308 {
1309         mlx4_pd_free(to_mdev(pd->device)->dev, to_mpd(pd)->pdn);
1310         kfree(pd);
1311
1312         return 0;
1313 }
1314
1315 static struct ib_xrcd *mlx4_ib_alloc_xrcd(struct ib_device *ibdev,
1316                                           struct ib_ucontext *context,
1317                                           struct ib_udata *udata)
1318 {
1319         struct mlx4_ib_xrcd *xrcd;
1320         struct ib_cq_init_attr cq_attr = {};
1321         int err;
1322
1323         if (!(to_mdev(ibdev)->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC))
1324                 return ERR_PTR(-ENOSYS);
1325
1326         xrcd = kmalloc(sizeof *xrcd, GFP_KERNEL);
1327         if (!xrcd)
1328                 return ERR_PTR(-ENOMEM);
1329
1330         err = mlx4_xrcd_alloc(to_mdev(ibdev)->dev, &xrcd->xrcdn);
1331         if (err)
1332                 goto err1;
1333
1334         xrcd->pd = ib_alloc_pd(ibdev, 0);
1335         if (IS_ERR(xrcd->pd)) {
1336                 err = PTR_ERR(xrcd->pd);
1337                 goto err2;
1338         }
1339
1340         cq_attr.cqe = 1;
1341         xrcd->cq = ib_create_cq(ibdev, NULL, NULL, xrcd, &cq_attr);
1342         if (IS_ERR(xrcd->cq)) {
1343                 err = PTR_ERR(xrcd->cq);
1344                 goto err3;
1345         }
1346
1347         return &xrcd->ibxrcd;
1348
1349 err3:
1350         ib_dealloc_pd(xrcd->pd);
1351 err2:
1352         mlx4_xrcd_free(to_mdev(ibdev)->dev, xrcd->xrcdn);
1353 err1:
1354         kfree(xrcd);
1355         return ERR_PTR(err);
1356 }
1357
1358 static int mlx4_ib_dealloc_xrcd(struct ib_xrcd *xrcd)
1359 {
1360         ib_destroy_cq(to_mxrcd(xrcd)->cq);
1361         ib_dealloc_pd(to_mxrcd(xrcd)->pd);
1362         mlx4_xrcd_free(to_mdev(xrcd->device)->dev, to_mxrcd(xrcd)->xrcdn);
1363         kfree(xrcd);
1364
1365         return 0;
1366 }
1367
1368 static int add_gid_entry(struct ib_qp *ibqp, union ib_gid *gid)
1369 {
1370         struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1371         struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1372         struct mlx4_ib_gid_entry *ge;
1373
1374         ge = kzalloc(sizeof *ge, GFP_KERNEL);
1375         if (!ge)
1376                 return -ENOMEM;
1377
1378         ge->gid = *gid;
1379         if (mlx4_ib_add_mc(mdev, mqp, gid)) {
1380                 ge->port = mqp->port;
1381                 ge->added = 1;
1382         }
1383
1384         mutex_lock(&mqp->mutex);
1385         list_add_tail(&ge->list, &mqp->gid_list);
1386         mutex_unlock(&mqp->mutex);
1387
1388         return 0;
1389 }
1390
1391 static void mlx4_ib_delete_counters_table(struct mlx4_ib_dev *ibdev,
1392                                           struct mlx4_ib_counters *ctr_table)
1393 {
1394         struct counter_index *counter, *tmp_count;
1395
1396         mutex_lock(&ctr_table->mutex);
1397         list_for_each_entry_safe(counter, tmp_count, &ctr_table->counters_list,
1398                                  list) {
1399                 if (counter->allocated)
1400                         mlx4_counter_free(ibdev->dev, counter->index);
1401                 list_del(&counter->list);
1402                 kfree(counter);
1403         }
1404         mutex_unlock(&ctr_table->mutex);
1405 }
1406
1407 int mlx4_ib_add_mc(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
1408                    union ib_gid *gid)
1409 {
1410         struct net_device *ndev;
1411         int ret = 0;
1412
1413         if (!mqp->port)
1414                 return 0;
1415
1416         spin_lock_bh(&mdev->iboe.lock);
1417         ndev = mdev->iboe.netdevs[mqp->port - 1];
1418         if (ndev)
1419                 dev_hold(ndev);
1420         spin_unlock_bh(&mdev->iboe.lock);
1421
1422         if (ndev) {
1423                 ret = 1;
1424                 dev_put(ndev);
1425         }
1426
1427         return ret;
1428 }
1429
1430 struct mlx4_ib_steering {
1431         struct list_head list;
1432         struct mlx4_flow_reg_id reg_id;
1433         union ib_gid gid;
1434 };
1435
1436 #define LAST_ETH_FIELD vlan_tag
1437 #define LAST_IB_FIELD sl
1438 #define LAST_IPV4_FIELD dst_ip
1439 #define LAST_TCP_UDP_FIELD src_port
1440
1441 /* Field is the last supported field */
1442 #define FIELDS_NOT_SUPPORTED(filter, field)\
1443         memchr_inv((void *)&filter.field  +\
1444                    sizeof(filter.field), 0,\
1445                    sizeof(filter) -\
1446                    offsetof(typeof(filter), field) -\
1447                    sizeof(filter.field))
1448
1449 static int parse_flow_attr(struct mlx4_dev *dev,
1450                            u32 qp_num,
1451                            union ib_flow_spec *ib_spec,
1452                            struct _rule_hw *mlx4_spec)
1453 {
1454         enum mlx4_net_trans_rule_id type;
1455
1456         switch (ib_spec->type) {
1457         case IB_FLOW_SPEC_ETH:
1458                 if (FIELDS_NOT_SUPPORTED(ib_spec->eth.mask, LAST_ETH_FIELD))
1459                         return -ENOTSUPP;
1460
1461                 type = MLX4_NET_TRANS_RULE_ID_ETH;
1462                 memcpy(mlx4_spec->eth.dst_mac, ib_spec->eth.val.dst_mac,
1463                        ETH_ALEN);
1464                 memcpy(mlx4_spec->eth.dst_mac_msk, ib_spec->eth.mask.dst_mac,
1465                        ETH_ALEN);
1466                 mlx4_spec->eth.vlan_tag = ib_spec->eth.val.vlan_tag;
1467                 mlx4_spec->eth.vlan_tag_msk = ib_spec->eth.mask.vlan_tag;
1468                 break;
1469         case IB_FLOW_SPEC_IB:
1470                 if (FIELDS_NOT_SUPPORTED(ib_spec->ib.mask, LAST_IB_FIELD))
1471                         return -ENOTSUPP;
1472
1473                 type = MLX4_NET_TRANS_RULE_ID_IB;
1474                 mlx4_spec->ib.l3_qpn =
1475                         cpu_to_be32(qp_num);
1476                 mlx4_spec->ib.qpn_mask =
1477                         cpu_to_be32(MLX4_IB_FLOW_QPN_MASK);
1478                 break;
1479
1480
1481         case IB_FLOW_SPEC_IPV4:
1482                 if (FIELDS_NOT_SUPPORTED(ib_spec->ipv4.mask, LAST_IPV4_FIELD))
1483                         return -ENOTSUPP;
1484
1485                 type = MLX4_NET_TRANS_RULE_ID_IPV4;
1486                 mlx4_spec->ipv4.src_ip = ib_spec->ipv4.val.src_ip;
1487                 mlx4_spec->ipv4.src_ip_msk = ib_spec->ipv4.mask.src_ip;
1488                 mlx4_spec->ipv4.dst_ip = ib_spec->ipv4.val.dst_ip;
1489                 mlx4_spec->ipv4.dst_ip_msk = ib_spec->ipv4.mask.dst_ip;
1490                 break;
1491
1492         case IB_FLOW_SPEC_TCP:
1493         case IB_FLOW_SPEC_UDP:
1494                 if (FIELDS_NOT_SUPPORTED(ib_spec->tcp_udp.mask, LAST_TCP_UDP_FIELD))
1495                         return -ENOTSUPP;
1496
1497                 type = ib_spec->type == IB_FLOW_SPEC_TCP ?
1498                                         MLX4_NET_TRANS_RULE_ID_TCP :
1499                                         MLX4_NET_TRANS_RULE_ID_UDP;
1500                 mlx4_spec->tcp_udp.dst_port = ib_spec->tcp_udp.val.dst_port;
1501                 mlx4_spec->tcp_udp.dst_port_msk = ib_spec->tcp_udp.mask.dst_port;
1502                 mlx4_spec->tcp_udp.src_port = ib_spec->tcp_udp.val.src_port;
1503                 mlx4_spec->tcp_udp.src_port_msk = ib_spec->tcp_udp.mask.src_port;
1504                 break;
1505
1506         default:
1507                 return -EINVAL;
1508         }
1509         if (mlx4_map_sw_to_hw_steering_id(dev, type) < 0 ||
1510             mlx4_hw_rule_sz(dev, type) < 0)
1511                 return -EINVAL;
1512         mlx4_spec->id = cpu_to_be16(mlx4_map_sw_to_hw_steering_id(dev, type));
1513         mlx4_spec->size = mlx4_hw_rule_sz(dev, type) >> 2;
1514         return mlx4_hw_rule_sz(dev, type);
1515 }
1516
1517 struct default_rules {
1518         __u32 mandatory_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
1519         __u32 mandatory_not_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
1520         __u32 rules_create_list[IB_FLOW_SPEC_SUPPORT_LAYERS];
1521         __u8  link_layer;
1522 };
1523 static const struct default_rules default_table[] = {
1524         {
1525                 .mandatory_fields = {IB_FLOW_SPEC_IPV4},
1526                 .mandatory_not_fields = {IB_FLOW_SPEC_ETH},
1527                 .rules_create_list = {IB_FLOW_SPEC_IB},
1528                 .link_layer = IB_LINK_LAYER_INFINIBAND
1529         }
1530 };
1531
1532 static int __mlx4_ib_default_rules_match(struct ib_qp *qp,
1533                                          struct ib_flow_attr *flow_attr)
1534 {
1535         int i, j, k;
1536         void *ib_flow;
1537         const struct default_rules *pdefault_rules = default_table;
1538         u8 link_layer = rdma_port_get_link_layer(qp->device, flow_attr->port);
1539
1540         for (i = 0; i < ARRAY_SIZE(default_table); i++, pdefault_rules++) {
1541                 __u32 field_types[IB_FLOW_SPEC_SUPPORT_LAYERS];
1542                 memset(&field_types, 0, sizeof(field_types));
1543
1544                 if (link_layer != pdefault_rules->link_layer)
1545                         continue;
1546
1547                 ib_flow = flow_attr + 1;
1548                 /* we assume the specs are sorted */
1549                 for (j = 0, k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS &&
1550                      j < flow_attr->num_of_specs; k++) {
1551                         union ib_flow_spec *current_flow =
1552                                 (union ib_flow_spec *)ib_flow;
1553
1554                         /* same layer but different type */
1555                         if (((current_flow->type & IB_FLOW_SPEC_LAYER_MASK) ==
1556                              (pdefault_rules->mandatory_fields[k] &
1557                               IB_FLOW_SPEC_LAYER_MASK)) &&
1558                             (current_flow->type !=
1559                              pdefault_rules->mandatory_fields[k]))
1560                                 goto out;
1561
1562                         /* same layer, try match next one */
1563                         if (current_flow->type ==
1564                             pdefault_rules->mandatory_fields[k]) {
1565                                 j++;
1566                                 ib_flow +=
1567                                         ((union ib_flow_spec *)ib_flow)->size;
1568                         }
1569                 }
1570
1571                 ib_flow = flow_attr + 1;
1572                 for (j = 0; j < flow_attr->num_of_specs;
1573                      j++, ib_flow += ((union ib_flow_spec *)ib_flow)->size)
1574                         for (k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS; k++)
1575                                 /* same layer and same type */
1576                                 if (((union ib_flow_spec *)ib_flow)->type ==
1577                                     pdefault_rules->mandatory_not_fields[k])
1578                                         goto out;
1579
1580                 return i;
1581         }
1582 out:
1583         return -1;
1584 }
1585
1586 static int __mlx4_ib_create_default_rules(
1587                 struct mlx4_ib_dev *mdev,
1588                 struct ib_qp *qp,
1589                 const struct default_rules *pdefault_rules,
1590                 struct _rule_hw *mlx4_spec) {
1591         int size = 0;
1592         int i;
1593
1594         for (i = 0; i < ARRAY_SIZE(pdefault_rules->rules_create_list); i++) {
1595                 int ret;
1596                 union ib_flow_spec ib_spec;
1597                 switch (pdefault_rules->rules_create_list[i]) {
1598                 case 0:
1599                         /* no rule */
1600                         continue;
1601                 case IB_FLOW_SPEC_IB:
1602                         ib_spec.type = IB_FLOW_SPEC_IB;
1603                         ib_spec.size = sizeof(struct ib_flow_spec_ib);
1604
1605                         break;
1606                 default:
1607                         /* invalid rule */
1608                         return -EINVAL;
1609                 }
1610                 /* We must put empty rule, qpn is being ignored */
1611                 ret = parse_flow_attr(mdev->dev, 0, &ib_spec,
1612                                       mlx4_spec);
1613                 if (ret < 0) {
1614                         pr_info("invalid parsing\n");
1615                         return -EINVAL;
1616                 }
1617
1618                 mlx4_spec = (void *)mlx4_spec + ret;
1619                 size += ret;
1620         }
1621         return size;
1622 }
1623
1624 static int __mlx4_ib_create_flow(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
1625                           int domain,
1626                           enum mlx4_net_trans_promisc_mode flow_type,
1627                           u64 *reg_id)
1628 {
1629         int ret, i;
1630         int size = 0;
1631         void *ib_flow;
1632         struct mlx4_ib_dev *mdev = to_mdev(qp->device);
1633         struct mlx4_cmd_mailbox *mailbox;
1634         struct mlx4_net_trans_rule_hw_ctrl *ctrl;
1635         int default_flow;
1636
1637         static const u16 __mlx4_domain[] = {
1638                 [IB_FLOW_DOMAIN_USER] = MLX4_DOMAIN_UVERBS,
1639                 [IB_FLOW_DOMAIN_ETHTOOL] = MLX4_DOMAIN_ETHTOOL,
1640                 [IB_FLOW_DOMAIN_RFS] = MLX4_DOMAIN_RFS,
1641                 [IB_FLOW_DOMAIN_NIC] = MLX4_DOMAIN_NIC,
1642         };
1643
1644         if (flow_attr->priority > MLX4_IB_FLOW_MAX_PRIO) {
1645                 pr_err("Invalid priority value %d\n", flow_attr->priority);
1646                 return -EINVAL;
1647         }
1648
1649         if (domain >= IB_FLOW_DOMAIN_NUM) {
1650                 pr_err("Invalid domain value %d\n", domain);
1651                 return -EINVAL;
1652         }
1653
1654         if (mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type) < 0)
1655                 return -EINVAL;
1656
1657         mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
1658         if (IS_ERR(mailbox))
1659                 return PTR_ERR(mailbox);
1660         ctrl = mailbox->buf;
1661
1662         ctrl->prio = cpu_to_be16(__mlx4_domain[domain] |
1663                                  flow_attr->priority);
1664         ctrl->type = mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type);
1665         ctrl->port = flow_attr->port;
1666         ctrl->qpn = cpu_to_be32(qp->qp_num);
1667
1668         ib_flow = flow_attr + 1;
1669         size += sizeof(struct mlx4_net_trans_rule_hw_ctrl);
1670         /* Add default flows */
1671         default_flow = __mlx4_ib_default_rules_match(qp, flow_attr);
1672         if (default_flow >= 0) {
1673                 ret = __mlx4_ib_create_default_rules(
1674                                 mdev, qp, default_table + default_flow,
1675                                 mailbox->buf + size);
1676                 if (ret < 0) {
1677                         mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1678                         return -EINVAL;
1679                 }
1680                 size += ret;
1681         }
1682         for (i = 0; i < flow_attr->num_of_specs; i++) {
1683                 ret = parse_flow_attr(mdev->dev, qp->qp_num, ib_flow,
1684                                       mailbox->buf + size);
1685                 if (ret < 0) {
1686                         mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1687                         return -EINVAL;
1688                 }
1689                 ib_flow += ((union ib_flow_spec *) ib_flow)->size;
1690                 size += ret;
1691         }
1692
1693         if (mlx4_is_master(mdev->dev) && flow_type == MLX4_FS_REGULAR &&
1694             flow_attr->num_of_specs == 1) {
1695                 struct _rule_hw *rule_header = (struct _rule_hw *)(ctrl + 1);
1696                 enum ib_flow_spec_type header_spec =
1697                         ((union ib_flow_spec *)(flow_attr + 1))->type;
1698
1699                 if (header_spec == IB_FLOW_SPEC_ETH)
1700                         mlx4_handle_eth_header_mcast_prio(ctrl, rule_header);
1701         }
1702
1703         ret = mlx4_cmd_imm(mdev->dev, mailbox->dma, reg_id, size >> 2, 0,
1704                            MLX4_QP_FLOW_STEERING_ATTACH, MLX4_CMD_TIME_CLASS_A,
1705                            MLX4_CMD_NATIVE);
1706         if (ret == -ENOMEM)
1707                 pr_err("mcg table is full. Fail to register network rule.\n");
1708         else if (ret == -ENXIO)
1709                 pr_err("Device managed flow steering is disabled. Fail to register network rule.\n");
1710         else if (ret)
1711                 pr_err("Invalid argument. Fail to register network rule.\n");
1712
1713         mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1714         return ret;
1715 }
1716
1717 static int __mlx4_ib_destroy_flow(struct mlx4_dev *dev, u64 reg_id)
1718 {
1719         int err;
1720         err = mlx4_cmd(dev, reg_id, 0, 0,
1721                        MLX4_QP_FLOW_STEERING_DETACH, MLX4_CMD_TIME_CLASS_A,
1722                        MLX4_CMD_NATIVE);
1723         if (err)
1724                 pr_err("Fail to detach network rule. registration id = 0x%llx\n",
1725                        reg_id);
1726         return err;
1727 }
1728
1729 static int mlx4_ib_tunnel_steer_add(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
1730                                     u64 *reg_id)
1731 {
1732         void *ib_flow;
1733         union ib_flow_spec *ib_spec;
1734         struct mlx4_dev *dev = to_mdev(qp->device)->dev;
1735         int err = 0;
1736
1737         if (dev->caps.tunnel_offload_mode != MLX4_TUNNEL_OFFLOAD_MODE_VXLAN ||
1738             dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC)
1739                 return 0; /* do nothing */
1740
1741         ib_flow = flow_attr + 1;
1742         ib_spec = (union ib_flow_spec *)ib_flow;
1743
1744         if (ib_spec->type !=  IB_FLOW_SPEC_ETH || flow_attr->num_of_specs != 1)
1745                 return 0; /* do nothing */
1746
1747         err = mlx4_tunnel_steer_add(to_mdev(qp->device)->dev, ib_spec->eth.val.dst_mac,
1748                                     flow_attr->port, qp->qp_num,
1749                                     MLX4_DOMAIN_UVERBS | (flow_attr->priority & 0xff),
1750                                     reg_id);
1751         return err;
1752 }
1753
1754 static int mlx4_ib_add_dont_trap_rule(struct mlx4_dev *dev,
1755                                       struct ib_flow_attr *flow_attr,
1756                                       enum mlx4_net_trans_promisc_mode *type)
1757 {
1758         int err = 0;
1759
1760         if (!(dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DMFS_UC_MC_SNIFFER) ||
1761             (dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC) ||
1762             (flow_attr->num_of_specs > 1) || (flow_attr->priority != 0)) {
1763                 return -EOPNOTSUPP;
1764         }
1765
1766         if (flow_attr->num_of_specs == 0) {
1767                 type[0] = MLX4_FS_MC_SNIFFER;
1768                 type[1] = MLX4_FS_UC_SNIFFER;
1769         } else {
1770                 union ib_flow_spec *ib_spec;
1771
1772                 ib_spec = (union ib_flow_spec *)(flow_attr + 1);
1773                 if (ib_spec->type !=  IB_FLOW_SPEC_ETH)
1774                         return -EINVAL;
1775
1776                 /* if all is zero than MC and UC */
1777                 if (is_zero_ether_addr(ib_spec->eth.mask.dst_mac)) {
1778                         type[0] = MLX4_FS_MC_SNIFFER;
1779                         type[1] = MLX4_FS_UC_SNIFFER;
1780                 } else {
1781                         u8 mac[ETH_ALEN] = {ib_spec->eth.mask.dst_mac[0] ^ 0x01,
1782                                             ib_spec->eth.mask.dst_mac[1],
1783                                             ib_spec->eth.mask.dst_mac[2],
1784                                             ib_spec->eth.mask.dst_mac[3],
1785                                             ib_spec->eth.mask.dst_mac[4],
1786                                             ib_spec->eth.mask.dst_mac[5]};
1787
1788                         /* Above xor was only on MC bit, non empty mask is valid
1789                          * only if this bit is set and rest are zero.
1790                          */
1791                         if (!is_zero_ether_addr(&mac[0]))
1792                                 return -EINVAL;
1793
1794                         if (is_multicast_ether_addr(ib_spec->eth.val.dst_mac))
1795                                 type[0] = MLX4_FS_MC_SNIFFER;
1796                         else
1797                                 type[0] = MLX4_FS_UC_SNIFFER;
1798                 }
1799         }
1800
1801         return err;
1802 }
1803
1804 static struct ib_flow *mlx4_ib_create_flow(struct ib_qp *qp,
1805                                     struct ib_flow_attr *flow_attr,
1806                                     int domain, struct ib_udata *udata)
1807 {
1808         int err = 0, i = 0, j = 0;
1809         struct mlx4_ib_flow *mflow;
1810         enum mlx4_net_trans_promisc_mode type[2];
1811         struct mlx4_dev *dev = (to_mdev(qp->device))->dev;
1812         int is_bonded = mlx4_is_bonded(dev);
1813
1814         if (flow_attr->port < 1 || flow_attr->port > qp->device->phys_port_cnt)
1815                 return ERR_PTR(-EINVAL);
1816
1817         if (flow_attr->flags & ~IB_FLOW_ATTR_FLAGS_DONT_TRAP)
1818                 return ERR_PTR(-EOPNOTSUPP);
1819
1820         if ((flow_attr->flags & IB_FLOW_ATTR_FLAGS_DONT_TRAP) &&
1821             (flow_attr->type != IB_FLOW_ATTR_NORMAL))
1822                 return ERR_PTR(-EOPNOTSUPP);
1823
1824         if (udata &&
1825             udata->inlen && !ib_is_udata_cleared(udata, 0, udata->inlen))
1826                 return ERR_PTR(-EOPNOTSUPP);
1827
1828         memset(type, 0, sizeof(type));
1829
1830         mflow = kzalloc(sizeof(*mflow), GFP_KERNEL);
1831         if (!mflow) {
1832                 err = -ENOMEM;
1833                 goto err_free;
1834         }
1835
1836         switch (flow_attr->type) {
1837         case IB_FLOW_ATTR_NORMAL:
1838                 /* If dont trap flag (continue match) is set, under specific
1839                  * condition traffic be replicated to given qp,
1840                  * without stealing it
1841                  */
1842                 if (unlikely(flow_attr->flags & IB_FLOW_ATTR_FLAGS_DONT_TRAP)) {
1843                         err = mlx4_ib_add_dont_trap_rule(dev,
1844                                                          flow_attr,
1845                                                          type);
1846                         if (err)
1847                                 goto err_free;
1848                 } else {
1849                         type[0] = MLX4_FS_REGULAR;
1850                 }
1851                 break;
1852
1853         case IB_FLOW_ATTR_ALL_DEFAULT:
1854                 type[0] = MLX4_FS_ALL_DEFAULT;
1855                 break;
1856
1857         case IB_FLOW_ATTR_MC_DEFAULT:
1858                 type[0] = MLX4_FS_MC_DEFAULT;
1859                 break;
1860
1861         case IB_FLOW_ATTR_SNIFFER:
1862                 type[0] = MLX4_FS_MIRROR_RX_PORT;
1863                 type[1] = MLX4_FS_MIRROR_SX_PORT;
1864                 break;
1865
1866         default:
1867                 err = -EINVAL;
1868                 goto err_free;
1869         }
1870
1871         while (i < ARRAY_SIZE(type) && type[i]) {
1872                 err = __mlx4_ib_create_flow(qp, flow_attr, domain, type[i],
1873                                             &mflow->reg_id[i].id);
1874                 if (err)
1875                         goto err_create_flow;
1876                 if (is_bonded) {
1877                         /* Application always sees one port so the mirror rule
1878                          * must be on port #2
1879                          */
1880                         flow_attr->port = 2;
1881                         err = __mlx4_ib_create_flow(qp, flow_attr,
1882                                                     domain, type[j],
1883                                                     &mflow->reg_id[j].mirror);
1884                         flow_attr->port = 1;
1885                         if (err)
1886                                 goto err_create_flow;
1887                         j++;
1888                 }
1889
1890                 i++;
1891         }
1892
1893         if (i < ARRAY_SIZE(type) && flow_attr->type == IB_FLOW_ATTR_NORMAL) {
1894                 err = mlx4_ib_tunnel_steer_add(qp, flow_attr,
1895                                                &mflow->reg_id[i].id);
1896                 if (err)
1897                         goto err_create_flow;
1898
1899                 if (is_bonded) {
1900                         flow_attr->port = 2;
1901                         err = mlx4_ib_tunnel_steer_add(qp, flow_attr,
1902                                                        &mflow->reg_id[j].mirror);
1903                         flow_attr->port = 1;
1904                         if (err)
1905                                 goto err_create_flow;
1906                         j++;
1907                 }
1908                 /* function to create mirror rule */
1909                 i++;
1910         }
1911
1912         return &mflow->ibflow;
1913
1914 err_create_flow:
1915         while (i) {
1916                 (void)__mlx4_ib_destroy_flow(to_mdev(qp->device)->dev,
1917                                              mflow->reg_id[i].id);
1918                 i--;
1919         }
1920
1921         while (j) {
1922                 (void)__mlx4_ib_destroy_flow(to_mdev(qp->device)->dev,
1923                                              mflow->reg_id[j].mirror);
1924                 j--;
1925         }
1926 err_free:
1927         kfree(mflow);
1928         return ERR_PTR(err);
1929 }
1930
1931 static int mlx4_ib_destroy_flow(struct ib_flow *flow_id)
1932 {
1933         int err, ret = 0;
1934         int i = 0;
1935         struct mlx4_ib_dev *mdev = to_mdev(flow_id->qp->device);
1936         struct mlx4_ib_flow *mflow = to_mflow(flow_id);
1937
1938         while (i < ARRAY_SIZE(mflow->reg_id) && mflow->reg_id[i].id) {
1939                 err = __mlx4_ib_destroy_flow(mdev->dev, mflow->reg_id[i].id);
1940                 if (err)
1941                         ret = err;
1942                 if (mflow->reg_id[i].mirror) {
1943                         err = __mlx4_ib_destroy_flow(mdev->dev,
1944                                                      mflow->reg_id[i].mirror);
1945                         if (err)
1946                                 ret = err;
1947                 }
1948                 i++;
1949         }
1950
1951         kfree(mflow);
1952         return ret;
1953 }
1954
1955 static int mlx4_ib_mcg_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
1956 {
1957         int err;
1958         struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1959         struct mlx4_dev *dev = mdev->dev;
1960         struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1961         struct mlx4_ib_steering *ib_steering = NULL;
1962         enum mlx4_protocol prot = MLX4_PROT_IB_IPV6;
1963         struct mlx4_flow_reg_id reg_id;
1964
1965         if (mdev->dev->caps.steering_mode ==
1966             MLX4_STEERING_MODE_DEVICE_MANAGED) {
1967                 ib_steering = kmalloc(sizeof(*ib_steering), GFP_KERNEL);
1968                 if (!ib_steering)
1969                         return -ENOMEM;
1970         }
1971
1972         err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw, mqp->port,
1973                                     !!(mqp->flags &
1974                                        MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
1975                                     prot, &reg_id.id);
1976         if (err) {
1977                 pr_err("multicast attach op failed, err %d\n", err);
1978                 goto err_malloc;
1979         }
1980
1981         reg_id.mirror = 0;
1982         if (mlx4_is_bonded(dev)) {
1983                 err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw,
1984                                             (mqp->port == 1) ? 2 : 1,
1985                                             !!(mqp->flags &
1986                                             MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
1987                                             prot, &reg_id.mirror);
1988                 if (err)
1989                         goto err_add;
1990         }
1991
1992         err = add_gid_entry(ibqp, gid);
1993         if (err)
1994                 goto err_add;
1995
1996         if (ib_steering) {
1997                 memcpy(ib_steering->gid.raw, gid->raw, 16);
1998                 ib_steering->reg_id = reg_id;
1999                 mutex_lock(&mqp->mutex);
2000                 list_add(&ib_steering->list, &mqp->steering_rules);
2001                 mutex_unlock(&mqp->mutex);
2002         }
2003         return 0;
2004
2005 err_add:
2006         mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
2007                               prot, reg_id.id);
2008         if (reg_id.mirror)
2009                 mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
2010                                       prot, reg_id.mirror);
2011 err_malloc:
2012         kfree(ib_steering);
2013
2014         return err;
2015 }
2016
2017 static struct mlx4_ib_gid_entry *find_gid_entry(struct mlx4_ib_qp *qp, u8 *raw)
2018 {
2019         struct mlx4_ib_gid_entry *ge;
2020         struct mlx4_ib_gid_entry *tmp;
2021         struct mlx4_ib_gid_entry *ret = NULL;
2022
2023         list_for_each_entry_safe(ge, tmp, &qp->gid_list, list) {
2024                 if (!memcmp(raw, ge->gid.raw, 16)) {
2025                         ret = ge;
2026                         break;
2027                 }
2028         }
2029
2030         return ret;
2031 }
2032
2033 static int mlx4_ib_mcg_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
2034 {
2035         int err;
2036         struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
2037         struct mlx4_dev *dev = mdev->dev;
2038         struct mlx4_ib_qp *mqp = to_mqp(ibqp);
2039         struct net_device *ndev;
2040         struct mlx4_ib_gid_entry *ge;
2041         struct mlx4_flow_reg_id reg_id = {0, 0};
2042         enum mlx4_protocol prot =  MLX4_PROT_IB_IPV6;
2043
2044         if (mdev->dev->caps.steering_mode ==
2045             MLX4_STEERING_MODE_DEVICE_MANAGED) {
2046                 struct mlx4_ib_steering *ib_steering;
2047
2048                 mutex_lock(&mqp->mutex);
2049                 list_for_each_entry(ib_steering, &mqp->steering_rules, list) {
2050                         if (!memcmp(ib_steering->gid.raw, gid->raw, 16)) {
2051                                 list_del(&ib_steering->list);
2052                                 break;
2053                         }
2054                 }
2055                 mutex_unlock(&mqp->mutex);
2056                 if (&ib_steering->list == &mqp->steering_rules) {
2057                         pr_err("Couldn't find reg_id for mgid. Steering rule is left attached\n");
2058                         return -EINVAL;
2059                 }
2060                 reg_id = ib_steering->reg_id;
2061                 kfree(ib_steering);
2062         }
2063
2064         err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
2065                                     prot, reg_id.id);
2066         if (err)
2067                 return err;
2068
2069         if (mlx4_is_bonded(dev)) {
2070                 err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
2071                                             prot, reg_id.mirror);
2072                 if (err)
2073                         return err;
2074         }
2075
2076         mutex_lock(&mqp->mutex);
2077         ge = find_gid_entry(mqp, gid->raw);
2078         if (ge) {
2079                 spin_lock_bh(&mdev->iboe.lock);
2080                 ndev = ge->added ? mdev->iboe.netdevs[ge->port - 1] : NULL;
2081                 if (ndev)
2082                         dev_hold(ndev);
2083                 spin_unlock_bh(&mdev->iboe.lock);
2084                 if (ndev)
2085                         dev_put(ndev);
2086                 list_del(&ge->list);
2087                 kfree(ge);
2088         } else
2089                 pr_warn("could not find mgid entry\n");
2090
2091         mutex_unlock(&mqp->mutex);
2092
2093         return 0;
2094 }
2095
2096 static int init_node_data(struct mlx4_ib_dev *dev)
2097 {
2098         struct ib_smp *in_mad  = NULL;
2099         struct ib_smp *out_mad = NULL;
2100         int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
2101         int err = -ENOMEM;
2102
2103         in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
2104         out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
2105         if (!in_mad || !out_mad)
2106                 goto out;
2107
2108         init_query_mad(in_mad);
2109         in_mad->attr_id = IB_SMP_ATTR_NODE_DESC;
2110         if (mlx4_is_master(dev->dev))
2111                 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
2112
2113         err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
2114         if (err)
2115                 goto out;
2116
2117         memcpy(dev->ib_dev.node_desc, out_mad->data, IB_DEVICE_NODE_DESC_MAX);
2118
2119         in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
2120
2121         err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
2122         if (err)
2123                 goto out;
2124
2125         dev->dev->rev_id = be32_to_cpup((__be32 *) (out_mad->data + 32));
2126         memcpy(&dev->ib_dev.node_guid, out_mad->data + 12, 8);
2127
2128 out:
2129         kfree(in_mad);
2130         kfree(out_mad);
2131         return err;
2132 }
2133
2134 static ssize_t show_hca(struct device *device, struct device_attribute *attr,
2135                         char *buf)
2136 {
2137         struct mlx4_ib_dev *dev =
2138                 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
2139         return sprintf(buf, "MT%d\n", dev->dev->persist->pdev->device);
2140 }
2141
2142 static ssize_t show_rev(struct device *device, struct device_attribute *attr,
2143                         char *buf)
2144 {
2145         struct mlx4_ib_dev *dev =
2146                 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
2147         return sprintf(buf, "%x\n", dev->dev->rev_id);
2148 }
2149
2150 static ssize_t show_board(struct device *device, struct device_attribute *attr,
2151                           char *buf)
2152 {
2153         struct mlx4_ib_dev *dev =
2154                 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
2155         return sprintf(buf, "%.*s\n", MLX4_BOARD_ID_LEN,
2156                        dev->dev->board_id);
2157 }
2158
2159 static DEVICE_ATTR(hw_rev,   S_IRUGO, show_rev,    NULL);
2160 static DEVICE_ATTR(hca_type, S_IRUGO, show_hca,    NULL);
2161 static DEVICE_ATTR(board_id, S_IRUGO, show_board,  NULL);
2162
2163 static struct device_attribute *mlx4_class_attributes[] = {
2164         &dev_attr_hw_rev,
2165         &dev_attr_hca_type,
2166         &dev_attr_board_id
2167 };
2168
2169 struct diag_counter {
2170         const char *name;
2171         u32 offset;
2172 };
2173
2174 #define DIAG_COUNTER(_name, _offset)                    \
2175         { .name = #_name, .offset = _offset }
2176
2177 static const struct diag_counter diag_basic[] = {
2178         DIAG_COUNTER(rq_num_lle, 0x00),
2179         DIAG_COUNTER(sq_num_lle, 0x04),
2180         DIAG_COUNTER(rq_num_lqpoe, 0x08),
2181         DIAG_COUNTER(sq_num_lqpoe, 0x0C),
2182         DIAG_COUNTER(rq_num_lpe, 0x18),
2183         DIAG_COUNTER(sq_num_lpe, 0x1C),
2184         DIAG_COUNTER(rq_num_wrfe, 0x20),
2185         DIAG_COUNTER(sq_num_wrfe, 0x24),
2186         DIAG_COUNTER(sq_num_mwbe, 0x2C),
2187         DIAG_COUNTER(sq_num_bre, 0x34),
2188         DIAG_COUNTER(sq_num_rire, 0x44),
2189         DIAG_COUNTER(rq_num_rire, 0x48),
2190         DIAG_COUNTER(sq_num_rae, 0x4C),
2191         DIAG_COUNTER(rq_num_rae, 0x50),
2192         DIAG_COUNTER(sq_num_roe, 0x54),
2193         DIAG_COUNTER(sq_num_tree, 0x5C),
2194         DIAG_COUNTER(sq_num_rree, 0x64),
2195         DIAG_COUNTER(rq_num_rnr, 0x68),
2196         DIAG_COUNTER(sq_num_rnr, 0x6C),
2197         DIAG_COUNTER(rq_num_oos, 0x100),
2198         DIAG_COUNTER(sq_num_oos, 0x104),
2199 };
2200
2201 static const struct diag_counter diag_ext[] = {
2202         DIAG_COUNTER(rq_num_dup, 0x130),
2203         DIAG_COUNTER(sq_num_to, 0x134),
2204 };
2205
2206 static const struct diag_counter diag_device_only[] = {
2207         DIAG_COUNTER(num_cqovf, 0x1A0),
2208         DIAG_COUNTER(rq_num_udsdprd, 0x118),
2209 };
2210
2211 static struct rdma_hw_stats *mlx4_ib_alloc_hw_stats(struct ib_device *ibdev,
2212                                                     u8 port_num)
2213 {
2214         struct mlx4_ib_dev *dev = to_mdev(ibdev);
2215         struct mlx4_ib_diag_counters *diag = dev->diag_counters;
2216
2217         if (!diag[!!port_num].name)
2218                 return NULL;
2219
2220         return rdma_alloc_hw_stats_struct(diag[!!port_num].name,
2221                                           diag[!!port_num].num_counters,
2222                                           RDMA_HW_STATS_DEFAULT_LIFESPAN);
2223 }
2224
2225 static int mlx4_ib_get_hw_stats(struct ib_device *ibdev,
2226                                 struct rdma_hw_stats *stats,
2227                                 u8 port, int index)
2228 {
2229         struct mlx4_ib_dev *dev = to_mdev(ibdev);
2230         struct mlx4_ib_diag_counters *diag = dev->diag_counters;
2231         u32 hw_value[ARRAY_SIZE(diag_device_only) +
2232                 ARRAY_SIZE(diag_ext) + ARRAY_SIZE(diag_basic)] = {};
2233         int ret;
2234         int i;
2235
2236         ret = mlx4_query_diag_counters(dev->dev,
2237                                        MLX4_OP_MOD_QUERY_TRANSPORT_CI_ERRORS,
2238                                        diag[!!port].offset, hw_value,
2239                                        diag[!!port].num_counters, port);
2240
2241         if (ret)
2242                 return ret;
2243
2244         for (i = 0; i < diag[!!port].num_counters; i++)
2245                 stats->value[i] = hw_value[i];
2246
2247         return diag[!!port].num_counters;
2248 }
2249
2250 static int __mlx4_ib_alloc_diag_counters(struct mlx4_ib_dev *ibdev,
2251                                          const char ***name,
2252                                          u32 **offset,
2253                                          u32 *num,
2254                                          bool port)
2255 {
2256         u32 num_counters;
2257
2258         num_counters = ARRAY_SIZE(diag_basic);
2259
2260         if (ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DIAG_PER_PORT)
2261                 num_counters += ARRAY_SIZE(diag_ext);
2262
2263         if (!port)
2264                 num_counters += ARRAY_SIZE(diag_device_only);
2265
2266         *name = kcalloc(num_counters, sizeof(**name), GFP_KERNEL);
2267         if (!*name)
2268                 return -ENOMEM;
2269
2270         *offset = kcalloc(num_counters, sizeof(**offset), GFP_KERNEL);
2271         if (!*offset)
2272                 goto err_name;
2273
2274         *num = num_counters;
2275
2276         return 0;
2277
2278 err_name:
2279         kfree(*name);
2280         return -ENOMEM;
2281 }
2282
2283 static void mlx4_ib_fill_diag_counters(struct mlx4_ib_dev *ibdev,
2284                                        const char **name,
2285                                        u32 *offset,
2286                                        bool port)
2287 {
2288         int i;
2289         int j;
2290
2291         for (i = 0, j = 0; i < ARRAY_SIZE(diag_basic); i++, j++) {
2292                 name[i] = diag_basic[i].name;
2293                 offset[i] = diag_basic[i].offset;
2294         }
2295
2296         if (ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DIAG_PER_PORT) {
2297                 for (i = 0; i < ARRAY_SIZE(diag_ext); i++, j++) {
2298                         name[j] = diag_ext[i].name;
2299                         offset[j] = diag_ext[i].offset;
2300                 }
2301         }
2302
2303         if (!port) {
2304                 for (i = 0; i < ARRAY_SIZE(diag_device_only); i++, j++) {
2305                         name[j] = diag_device_only[i].name;
2306                         offset[j] = diag_device_only[i].offset;
2307                 }
2308         }
2309 }
2310
2311 static int mlx4_ib_alloc_diag_counters(struct mlx4_ib_dev *ibdev)
2312 {
2313         struct mlx4_ib_diag_counters *diag = ibdev->diag_counters;
2314         int i;
2315         int ret;
2316         bool per_port = !!(ibdev->dev->caps.flags2 &
2317                 MLX4_DEV_CAP_FLAG2_DIAG_PER_PORT);
2318
2319         if (mlx4_is_slave(ibdev->dev))
2320                 return 0;
2321
2322         for (i = 0; i < MLX4_DIAG_COUNTERS_TYPES; i++) {
2323                 /* i == 1 means we are building port counters */
2324                 if (i && !per_port)
2325                         continue;
2326
2327                 ret = __mlx4_ib_alloc_diag_counters(ibdev, &diag[i].name,
2328                                                     &diag[i].offset,
2329                                                     &diag[i].num_counters, i);
2330                 if (ret)
2331                         goto err_alloc;
2332
2333                 mlx4_ib_fill_diag_counters(ibdev, diag[i].name,
2334                                            diag[i].offset, i);
2335         }
2336
2337         ibdev->ib_dev.get_hw_stats      = mlx4_ib_get_hw_stats;
2338         ibdev->ib_dev.alloc_hw_stats    = mlx4_ib_alloc_hw_stats;
2339
2340         return 0;
2341
2342 err_alloc:
2343         if (i) {
2344                 kfree(diag[i - 1].name);
2345                 kfree(diag[i - 1].offset);
2346         }
2347
2348         return ret;
2349 }
2350
2351 static void mlx4_ib_diag_cleanup(struct mlx4_ib_dev *ibdev)
2352 {
2353         int i;
2354
2355         for (i = 0; i < MLX4_DIAG_COUNTERS_TYPES; i++) {
2356                 kfree(ibdev->diag_counters[i].offset);
2357                 kfree(ibdev->diag_counters[i].name);
2358         }
2359 }
2360
2361 #define MLX4_IB_INVALID_MAC     ((u64)-1)
2362 static void mlx4_ib_update_qps(struct mlx4_ib_dev *ibdev,
2363                                struct net_device *dev,
2364                                int port)
2365 {
2366         u64 new_smac = 0;
2367         u64 release_mac = MLX4_IB_INVALID_MAC;
2368         struct mlx4_ib_qp *qp;
2369
2370         read_lock(&dev_base_lock);
2371         new_smac = mlx4_mac_to_u64(dev->dev_addr);
2372         read_unlock(&dev_base_lock);
2373
2374         atomic64_set(&ibdev->iboe.mac[port - 1], new_smac);
2375
2376         /* no need for update QP1 and mac registration in non-SRIOV */
2377         if (!mlx4_is_mfunc(ibdev->dev))
2378                 return;
2379
2380         mutex_lock(&ibdev->qp1_proxy_lock[port - 1]);
2381         qp = ibdev->qp1_proxy[port - 1];
2382         if (qp) {
2383                 int new_smac_index;
2384                 u64 old_smac;
2385                 struct mlx4_update_qp_params update_params;
2386
2387                 mutex_lock(&qp->mutex);
2388                 old_smac = qp->pri.smac;
2389                 if (new_smac == old_smac)
2390                         goto unlock;
2391
2392                 new_smac_index = mlx4_register_mac(ibdev->dev, port, new_smac);
2393
2394                 if (new_smac_index < 0)
2395                         goto unlock;
2396
2397                 update_params.smac_index = new_smac_index;
2398                 if (mlx4_update_qp(ibdev->dev, qp->mqp.qpn, MLX4_UPDATE_QP_SMAC,
2399                                    &update_params)) {
2400                         release_mac = new_smac;
2401                         goto unlock;
2402                 }
2403                 /* if old port was zero, no mac was yet registered for this QP */
2404                 if (qp->pri.smac_port)
2405                         release_mac = old_smac;
2406                 qp->pri.smac = new_smac;
2407                 qp->pri.smac_port = port;
2408                 qp->pri.smac_index = new_smac_index;
2409         }
2410
2411 unlock:
2412         if (release_mac != MLX4_IB_INVALID_MAC)
2413                 mlx4_unregister_mac(ibdev->dev, port, release_mac);
2414         if (qp)
2415                 mutex_unlock(&qp->mutex);
2416         mutex_unlock(&ibdev->qp1_proxy_lock[port - 1]);
2417 }
2418
2419 static void mlx4_ib_scan_netdevs(struct mlx4_ib_dev *ibdev,
2420                                  struct net_device *dev,
2421                                  unsigned long event)
2422
2423 {
2424         struct mlx4_ib_iboe *iboe;
2425         int update_qps_port = -1;
2426         int port;
2427
2428         ASSERT_RTNL();
2429
2430         iboe = &ibdev->iboe;
2431
2432         spin_lock_bh(&iboe->lock);
2433         mlx4_foreach_ib_transport_port(port, ibdev->dev) {
2434
2435                 iboe->netdevs[port - 1] =
2436                         mlx4_get_protocol_dev(ibdev->dev, MLX4_PROT_ETH, port);
2437
2438                 if (dev == iboe->netdevs[port - 1] &&
2439                     (event == NETDEV_CHANGEADDR || event == NETDEV_REGISTER ||
2440                      event == NETDEV_UP || event == NETDEV_CHANGE))
2441                         update_qps_port = port;
2442
2443         }
2444         spin_unlock_bh(&iboe->lock);
2445
2446         if (update_qps_port > 0)
2447                 mlx4_ib_update_qps(ibdev, dev, update_qps_port);
2448 }
2449
2450 static int mlx4_ib_netdev_event(struct notifier_block *this,
2451                                 unsigned long event, void *ptr)
2452 {
2453         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
2454         struct mlx4_ib_dev *ibdev;
2455
2456         if (!net_eq(dev_net(dev), &init_net))
2457                 return NOTIFY_DONE;
2458
2459         ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb);
2460         mlx4_ib_scan_netdevs(ibdev, dev, event);
2461
2462         return NOTIFY_DONE;
2463 }
2464
2465 static void init_pkeys(struct mlx4_ib_dev *ibdev)
2466 {
2467         int port;
2468         int slave;
2469         int i;
2470
2471         if (mlx4_is_master(ibdev->dev)) {
2472                 for (slave = 0; slave <= ibdev->dev->persist->num_vfs;
2473                      ++slave) {
2474                         for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
2475                                 for (i = 0;
2476                                      i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
2477                                      ++i) {
2478                                         ibdev->pkeys.virt2phys_pkey[slave][port - 1][i] =
2479                                         /* master has the identity virt2phys pkey mapping */
2480                                                 (slave == mlx4_master_func_num(ibdev->dev) || !i) ? i :
2481                                                         ibdev->dev->phys_caps.pkey_phys_table_len[port] - 1;
2482                                         mlx4_sync_pkey_table(ibdev->dev, slave, port, i,
2483                                                              ibdev->pkeys.virt2phys_pkey[slave][port - 1][i]);
2484                                 }
2485                         }
2486                 }
2487                 /* initialize pkey cache */
2488                 for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
2489                         for (i = 0;
2490                              i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
2491                              ++i)
2492                                 ibdev->pkeys.phys_pkey_cache[port-1][i] =
2493                                         (i) ? 0 : 0xFFFF;
2494                 }
2495         }
2496 }
2497
2498 static void mlx4_ib_alloc_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
2499 {
2500         int i, j, eq = 0, total_eqs = 0;
2501
2502         ibdev->eq_table = kcalloc(dev->caps.num_comp_vectors,
2503                                   sizeof(ibdev->eq_table[0]), GFP_KERNEL);
2504         if (!ibdev->eq_table)
2505                 return;
2506
2507         for (i = 1; i <= dev->caps.num_ports; i++) {
2508                 for (j = 0; j < mlx4_get_eqs_per_port(dev, i);
2509                      j++, total_eqs++) {
2510                         if (i > 1 &&  mlx4_is_eq_shared(dev, total_eqs))
2511                                 continue;
2512                         ibdev->eq_table[eq] = total_eqs;
2513                         if (!mlx4_assign_eq(dev, i,
2514                                             &ibdev->eq_table[eq]))
2515                                 eq++;
2516                         else
2517                                 ibdev->eq_table[eq] = -1;
2518                 }
2519         }
2520
2521         for (i = eq; i < dev->caps.num_comp_vectors;
2522              ibdev->eq_table[i++] = -1)
2523                 ;
2524
2525         /* Advertise the new number of EQs to clients */
2526         ibdev->ib_dev.num_comp_vectors = eq;
2527 }
2528
2529 static void mlx4_ib_free_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
2530 {
2531         int i;
2532         int total_eqs = ibdev->ib_dev.num_comp_vectors;
2533
2534         /* no eqs were allocated */
2535         if (!ibdev->eq_table)
2536                 return;
2537
2538         /* Reset the advertised EQ number */
2539         ibdev->ib_dev.num_comp_vectors = 0;
2540
2541         for (i = 0; i < total_eqs; i++)
2542                 mlx4_release_eq(dev, ibdev->eq_table[i]);
2543
2544         kfree(ibdev->eq_table);
2545         ibdev->eq_table = NULL;
2546 }
2547
2548 static int mlx4_port_immutable(struct ib_device *ibdev, u8 port_num,
2549                                struct ib_port_immutable *immutable)
2550 {
2551         struct ib_port_attr attr;
2552         struct mlx4_ib_dev *mdev = to_mdev(ibdev);
2553         int err;
2554
2555         if (mlx4_ib_port_link_layer(ibdev, port_num) == IB_LINK_LAYER_INFINIBAND) {
2556                 immutable->core_cap_flags = RDMA_CORE_PORT_IBA_IB;
2557                 immutable->max_mad_size = IB_MGMT_MAD_SIZE;
2558         } else {
2559                 if (mdev->dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE)
2560                         immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE;
2561                 if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ROCE_V1_V2)
2562                         immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE |
2563                                 RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP;
2564                 immutable->core_cap_flags |= RDMA_CORE_PORT_RAW_PACKET;
2565                 if (immutable->core_cap_flags & (RDMA_CORE_PORT_IBA_ROCE |
2566                     RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP))
2567                         immutable->max_mad_size = IB_MGMT_MAD_SIZE;
2568         }
2569
2570         err = ib_query_port(ibdev, port_num, &attr);
2571         if (err)
2572                 return err;
2573
2574         immutable->pkey_tbl_len = attr.pkey_tbl_len;
2575         immutable->gid_tbl_len = attr.gid_tbl_len;
2576
2577         return 0;
2578 }
2579
2580 static void get_fw_ver_str(struct ib_device *device, char *str)
2581 {
2582         struct mlx4_ib_dev *dev =
2583                 container_of(device, struct mlx4_ib_dev, ib_dev);
2584         snprintf(str, IB_FW_VERSION_NAME_MAX, "%d.%d.%d",
2585                  (int) (dev->dev->caps.fw_ver >> 32),
2586                  (int) (dev->dev->caps.fw_ver >> 16) & 0xffff,
2587                  (int) dev->dev->caps.fw_ver & 0xffff);
2588 }
2589
2590 static void *mlx4_ib_add(struct mlx4_dev *dev)
2591 {
2592         struct mlx4_ib_dev *ibdev;
2593         int num_ports = 0;
2594         int i, j;
2595         int err;
2596         struct mlx4_ib_iboe *iboe;
2597         int ib_num_ports = 0;
2598         int num_req_counters;
2599         int allocated;
2600         u32 counter_index;
2601         struct counter_index *new_counter_index = NULL;
2602
2603         pr_info_once("%s", mlx4_ib_version);
2604
2605         num_ports = 0;
2606         mlx4_foreach_ib_transport_port(i, dev)
2607                 num_ports++;
2608
2609         /* No point in registering a device with no ports... */
2610         if (num_ports == 0)
2611                 return NULL;
2612
2613         ibdev = (struct mlx4_ib_dev *) ib_alloc_device(sizeof *ibdev);
2614         if (!ibdev) {
2615                 dev_err(&dev->persist->pdev->dev,
2616                         "Device struct alloc failed\n");
2617                 return NULL;
2618         }
2619
2620         iboe = &ibdev->iboe;
2621
2622         if (mlx4_pd_alloc(dev, &ibdev->priv_pdn))
2623                 goto err_dealloc;
2624
2625         if (mlx4_uar_alloc(dev, &ibdev->priv_uar))
2626                 goto err_pd;
2627
2628         ibdev->uar_map = ioremap((phys_addr_t) ibdev->priv_uar.pfn << PAGE_SHIFT,
2629                                  PAGE_SIZE);
2630         if (!ibdev->uar_map)
2631                 goto err_uar;
2632         MLX4_INIT_DOORBELL_LOCK(&ibdev->uar_lock);
2633
2634         ibdev->dev = dev;
2635         ibdev->bond_next_port   = 0;
2636
2637         strlcpy(ibdev->ib_dev.name, "mlx4_%d", IB_DEVICE_NAME_MAX);
2638         ibdev->ib_dev.owner             = THIS_MODULE;
2639         ibdev->ib_dev.node_type         = RDMA_NODE_IB_CA;
2640         ibdev->ib_dev.local_dma_lkey    = dev->caps.reserved_lkey;
2641         ibdev->num_ports                = num_ports;
2642         ibdev->ib_dev.phys_port_cnt     = mlx4_is_bonded(dev) ?
2643                                                 1 : ibdev->num_ports;
2644         ibdev->ib_dev.num_comp_vectors  = dev->caps.num_comp_vectors;
2645         ibdev->ib_dev.dev.parent        = &dev->persist->pdev->dev;
2646         ibdev->ib_dev.get_netdev        = mlx4_ib_get_netdev;
2647         ibdev->ib_dev.add_gid           = mlx4_ib_add_gid;
2648         ibdev->ib_dev.del_gid           = mlx4_ib_del_gid;
2649
2650         if (dev->caps.userspace_caps)
2651                 ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_ABI_VERSION;
2652         else
2653                 ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION;
2654
2655         ibdev->ib_dev.uverbs_cmd_mask   =
2656                 (1ull << IB_USER_VERBS_CMD_GET_CONTEXT)         |
2657                 (1ull << IB_USER_VERBS_CMD_QUERY_DEVICE)        |
2658                 (1ull << IB_USER_VERBS_CMD_QUERY_PORT)          |
2659                 (1ull << IB_USER_VERBS_CMD_ALLOC_PD)            |
2660                 (1ull << IB_USER_VERBS_CMD_DEALLOC_PD)          |
2661                 (1ull << IB_USER_VERBS_CMD_REG_MR)              |
2662                 (1ull << IB_USER_VERBS_CMD_REREG_MR)            |
2663                 (1ull << IB_USER_VERBS_CMD_DEREG_MR)            |
2664                 (1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL) |
2665                 (1ull << IB_USER_VERBS_CMD_CREATE_CQ)           |
2666                 (1ull << IB_USER_VERBS_CMD_RESIZE_CQ)           |
2667                 (1ull << IB_USER_VERBS_CMD_DESTROY_CQ)          |
2668                 (1ull << IB_USER_VERBS_CMD_CREATE_QP)           |
2669                 (1ull << IB_USER_VERBS_CMD_MODIFY_QP)           |
2670                 (1ull << IB_USER_VERBS_CMD_QUERY_QP)            |
2671                 (1ull << IB_USER_VERBS_CMD_DESTROY_QP)          |
2672                 (1ull << IB_USER_VERBS_CMD_ATTACH_MCAST)        |
2673                 (1ull << IB_USER_VERBS_CMD_DETACH_MCAST)        |
2674                 (1ull << IB_USER_VERBS_CMD_CREATE_SRQ)          |
2675                 (1ull << IB_USER_VERBS_CMD_MODIFY_SRQ)          |
2676                 (1ull << IB_USER_VERBS_CMD_QUERY_SRQ)           |
2677                 (1ull << IB_USER_VERBS_CMD_DESTROY_SRQ)         |
2678                 (1ull << IB_USER_VERBS_CMD_CREATE_XSRQ)         |
2679                 (1ull << IB_USER_VERBS_CMD_OPEN_QP);
2680
2681         ibdev->ib_dev.query_device      = mlx4_ib_query_device;
2682         ibdev->ib_dev.query_port        = mlx4_ib_query_port;
2683         ibdev->ib_dev.get_link_layer    = mlx4_ib_port_link_layer;
2684         ibdev->ib_dev.query_gid         = mlx4_ib_query_gid;
2685         ibdev->ib_dev.query_pkey        = mlx4_ib_query_pkey;
2686         ibdev->ib_dev.modify_device     = mlx4_ib_modify_device;
2687         ibdev->ib_dev.modify_port       = mlx4_ib_modify_port;
2688         ibdev->ib_dev.alloc_ucontext    = mlx4_ib_alloc_ucontext;
2689         ibdev->ib_dev.dealloc_ucontext  = mlx4_ib_dealloc_ucontext;
2690         ibdev->ib_dev.mmap              = mlx4_ib_mmap;
2691         ibdev->ib_dev.alloc_pd          = mlx4_ib_alloc_pd;
2692         ibdev->ib_dev.dealloc_pd        = mlx4_ib_dealloc_pd;
2693         ibdev->ib_dev.create_ah         = mlx4_ib_create_ah;
2694         ibdev->ib_dev.query_ah          = mlx4_ib_query_ah;
2695         ibdev->ib_dev.destroy_ah        = mlx4_ib_destroy_ah;
2696         ibdev->ib_dev.create_srq        = mlx4_ib_create_srq;
2697         ibdev->ib_dev.modify_srq        = mlx4_ib_modify_srq;
2698         ibdev->ib_dev.query_srq         = mlx4_ib_query_srq;
2699         ibdev->ib_dev.destroy_srq       = mlx4_ib_destroy_srq;
2700         ibdev->ib_dev.post_srq_recv     = mlx4_ib_post_srq_recv;
2701         ibdev->ib_dev.create_qp         = mlx4_ib_create_qp;
2702         ibdev->ib_dev.modify_qp         = mlx4_ib_modify_qp;
2703         ibdev->ib_dev.query_qp          = mlx4_ib_query_qp;
2704         ibdev->ib_dev.destroy_qp        = mlx4_ib_destroy_qp;
2705         ibdev->ib_dev.drain_sq          = mlx4_ib_drain_sq;
2706         ibdev->ib_dev.drain_rq          = mlx4_ib_drain_rq;
2707         ibdev->ib_dev.post_send         = mlx4_ib_post_send;
2708         ibdev->ib_dev.post_recv         = mlx4_ib_post_recv;
2709         ibdev->ib_dev.create_cq         = mlx4_ib_create_cq;
2710         ibdev->ib_dev.modify_cq         = mlx4_ib_modify_cq;
2711         ibdev->ib_dev.resize_cq         = mlx4_ib_resize_cq;
2712         ibdev->ib_dev.destroy_cq        = mlx4_ib_destroy_cq;
2713         ibdev->ib_dev.poll_cq           = mlx4_ib_poll_cq;
2714         ibdev->ib_dev.req_notify_cq     = mlx4_ib_arm_cq;
2715         ibdev->ib_dev.get_dma_mr        = mlx4_ib_get_dma_mr;
2716         ibdev->ib_dev.reg_user_mr       = mlx4_ib_reg_user_mr;
2717         ibdev->ib_dev.rereg_user_mr     = mlx4_ib_rereg_user_mr;
2718         ibdev->ib_dev.dereg_mr          = mlx4_ib_dereg_mr;
2719         ibdev->ib_dev.alloc_mr          = mlx4_ib_alloc_mr;
2720         ibdev->ib_dev.map_mr_sg         = mlx4_ib_map_mr_sg;
2721         ibdev->ib_dev.attach_mcast      = mlx4_ib_mcg_attach;
2722         ibdev->ib_dev.detach_mcast      = mlx4_ib_mcg_detach;
2723         ibdev->ib_dev.process_mad       = mlx4_ib_process_mad;
2724         ibdev->ib_dev.get_port_immutable = mlx4_port_immutable;
2725         ibdev->ib_dev.get_dev_fw_str    = get_fw_ver_str;
2726         ibdev->ib_dev.disassociate_ucontext = mlx4_ib_disassociate_ucontext;
2727
2728         ibdev->ib_dev.uverbs_ex_cmd_mask |=
2729                 (1ull << IB_USER_VERBS_EX_CMD_MODIFY_CQ);
2730
2731         if ((dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_RSS) &&
2732             ((mlx4_ib_port_link_layer(&ibdev->ib_dev, 1) ==
2733             IB_LINK_LAYER_ETHERNET) ||
2734             (mlx4_ib_port_link_layer(&ibdev->ib_dev, 2) ==
2735             IB_LINK_LAYER_ETHERNET))) {
2736                 ibdev->ib_dev.create_wq         = mlx4_ib_create_wq;
2737                 ibdev->ib_dev.modify_wq         = mlx4_ib_modify_wq;
2738                 ibdev->ib_dev.destroy_wq        = mlx4_ib_destroy_wq;
2739                 ibdev->ib_dev.create_rwq_ind_table  =
2740                         mlx4_ib_create_rwq_ind_table;
2741                 ibdev->ib_dev.destroy_rwq_ind_table =
2742                         mlx4_ib_destroy_rwq_ind_table;
2743                 ibdev->ib_dev.uverbs_ex_cmd_mask |=
2744                         (1ull << IB_USER_VERBS_EX_CMD_CREATE_WQ)          |
2745                         (1ull << IB_USER_VERBS_EX_CMD_MODIFY_WQ)          |
2746                         (1ull << IB_USER_VERBS_EX_CMD_DESTROY_WQ)         |
2747                         (1ull << IB_USER_VERBS_EX_CMD_CREATE_RWQ_IND_TBL) |
2748                         (1ull << IB_USER_VERBS_EX_CMD_DESTROY_RWQ_IND_TBL);
2749         }
2750
2751         if (!mlx4_is_slave(ibdev->dev)) {
2752                 ibdev->ib_dev.alloc_fmr         = mlx4_ib_fmr_alloc;
2753                 ibdev->ib_dev.map_phys_fmr      = mlx4_ib_map_phys_fmr;
2754                 ibdev->ib_dev.unmap_fmr         = mlx4_ib_unmap_fmr;
2755                 ibdev->ib_dev.dealloc_fmr       = mlx4_ib_fmr_dealloc;
2756         }
2757
2758         if (dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW ||
2759             dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
2760                 ibdev->ib_dev.alloc_mw = mlx4_ib_alloc_mw;
2761                 ibdev->ib_dev.dealloc_mw = mlx4_ib_dealloc_mw;
2762
2763                 ibdev->ib_dev.uverbs_cmd_mask |=
2764                         (1ull << IB_USER_VERBS_CMD_ALLOC_MW) |
2765                         (1ull << IB_USER_VERBS_CMD_DEALLOC_MW);
2766         }
2767
2768         if (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) {
2769                 ibdev->ib_dev.alloc_xrcd = mlx4_ib_alloc_xrcd;
2770                 ibdev->ib_dev.dealloc_xrcd = mlx4_ib_dealloc_xrcd;
2771                 ibdev->ib_dev.uverbs_cmd_mask |=
2772                         (1ull << IB_USER_VERBS_CMD_OPEN_XRCD) |
2773                         (1ull << IB_USER_VERBS_CMD_CLOSE_XRCD);
2774         }
2775
2776         if (check_flow_steering_support(dev)) {
2777                 ibdev->steering_support = MLX4_STEERING_MODE_DEVICE_MANAGED;
2778                 ibdev->ib_dev.create_flow       = mlx4_ib_create_flow;
2779                 ibdev->ib_dev.destroy_flow      = mlx4_ib_destroy_flow;
2780
2781                 ibdev->ib_dev.uverbs_ex_cmd_mask        |=
2782                         (1ull << IB_USER_VERBS_EX_CMD_CREATE_FLOW) |
2783                         (1ull << IB_USER_VERBS_EX_CMD_DESTROY_FLOW);
2784         }
2785
2786         ibdev->ib_dev.uverbs_ex_cmd_mask |=
2787                 (1ull << IB_USER_VERBS_EX_CMD_QUERY_DEVICE) |
2788                 (1ull << IB_USER_VERBS_EX_CMD_CREATE_CQ) |
2789                 (1ull << IB_USER_VERBS_EX_CMD_CREATE_QP);
2790
2791         mlx4_ib_alloc_eqs(dev, ibdev);
2792
2793         spin_lock_init(&iboe->lock);
2794
2795         if (init_node_data(ibdev))
2796                 goto err_map;
2797         mlx4_init_sl2vl_tbl(ibdev);
2798
2799         for (i = 0; i < ibdev->num_ports; ++i) {
2800                 mutex_init(&ibdev->counters_table[i].mutex);
2801                 INIT_LIST_HEAD(&ibdev->counters_table[i].counters_list);
2802         }
2803
2804         num_req_counters = mlx4_is_bonded(dev) ? 1 : ibdev->num_ports;
2805         for (i = 0; i < num_req_counters; ++i) {
2806                 mutex_init(&ibdev->qp1_proxy_lock[i]);
2807                 allocated = 0;
2808                 if (mlx4_ib_port_link_layer(&ibdev->ib_dev, i + 1) ==
2809                                                 IB_LINK_LAYER_ETHERNET) {
2810                         err = mlx4_counter_alloc(ibdev->dev, &counter_index,
2811                                                  MLX4_RES_USAGE_DRIVER);
2812                         /* if failed to allocate a new counter, use default */
2813                         if (err)
2814                                 counter_index =
2815                                         mlx4_get_default_counter_index(dev,
2816                                                                        i + 1);
2817                         else
2818                                 allocated = 1;
2819                 } else { /* IB_LINK_LAYER_INFINIBAND use the default counter */
2820                         counter_index = mlx4_get_default_counter_index(dev,
2821                                                                        i + 1);
2822                 }
2823                 new_counter_index = kmalloc(sizeof(*new_counter_index),
2824                                             GFP_KERNEL);
2825                 if (!new_counter_index) {
2826                         if (allocated)
2827                                 mlx4_counter_free(ibdev->dev, counter_index);
2828                         goto err_counter;
2829                 }
2830                 new_counter_index->index = counter_index;
2831                 new_counter_index->allocated = allocated;
2832                 list_add_tail(&new_counter_index->list,
2833                               &ibdev->counters_table[i].counters_list);
2834                 ibdev->counters_table[i].default_counter = counter_index;
2835                 pr_info("counter index %d for port %d allocated %d\n",
2836                         counter_index, i + 1, allocated);
2837         }
2838         if (mlx4_is_bonded(dev))
2839                 for (i = 1; i < ibdev->num_ports ; ++i) {
2840                         new_counter_index =
2841                                         kmalloc(sizeof(struct counter_index),
2842                                                 GFP_KERNEL);
2843                         if (!new_counter_index)
2844                                 goto err_counter;
2845                         new_counter_index->index = counter_index;
2846                         new_counter_index->allocated = 0;
2847                         list_add_tail(&new_counter_index->list,
2848                                       &ibdev->counters_table[i].counters_list);
2849                         ibdev->counters_table[i].default_counter =
2850                                                                 counter_index;
2851                 }
2852
2853         mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
2854                 ib_num_ports++;
2855
2856         spin_lock_init(&ibdev->sm_lock);
2857         mutex_init(&ibdev->cap_mask_mutex);
2858         INIT_LIST_HEAD(&ibdev->qp_list);
2859         spin_lock_init(&ibdev->reset_flow_resource_lock);
2860
2861         if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED &&
2862             ib_num_ports) {
2863                 ibdev->steer_qpn_count = MLX4_IB_UC_MAX_NUM_QPS;
2864                 err = mlx4_qp_reserve_range(dev, ibdev->steer_qpn_count,
2865                                             MLX4_IB_UC_STEER_QPN_ALIGN,
2866                                             &ibdev->steer_qpn_base, 0,
2867                                             MLX4_RES_USAGE_DRIVER);
2868                 if (err)
2869                         goto err_counter;
2870
2871                 ibdev->ib_uc_qpns_bitmap =
2872                         kmalloc_array(BITS_TO_LONGS(ibdev->steer_qpn_count),
2873                                       sizeof(long),
2874                                       GFP_KERNEL);
2875                 if (!ibdev->ib_uc_qpns_bitmap)
2876                         goto err_steer_qp_release;
2877
2878                 if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DMFS_IPOIB) {
2879                         bitmap_zero(ibdev->ib_uc_qpns_bitmap,
2880                                     ibdev->steer_qpn_count);
2881                         err = mlx4_FLOW_STEERING_IB_UC_QP_RANGE(
2882                                         dev, ibdev->steer_qpn_base,
2883                                         ibdev->steer_qpn_base +
2884                                         ibdev->steer_qpn_count - 1);
2885                         if (err)
2886                                 goto err_steer_free_bitmap;
2887                 } else {
2888                         bitmap_fill(ibdev->ib_uc_qpns_bitmap,
2889                                     ibdev->steer_qpn_count);
2890                 }
2891         }
2892
2893         for (j = 1; j <= ibdev->dev->caps.num_ports; j++)
2894                 atomic64_set(&iboe->mac[j - 1], ibdev->dev->caps.def_mac[j]);
2895
2896         if (mlx4_ib_alloc_diag_counters(ibdev))
2897                 goto err_steer_free_bitmap;
2898
2899         ibdev->ib_dev.driver_id = RDMA_DRIVER_MLX4;
2900         if (ib_register_device(&ibdev->ib_dev, NULL))
2901                 goto err_diag_counters;
2902
2903         if (mlx4_ib_mad_init(ibdev))
2904                 goto err_reg;
2905
2906         if (mlx4_ib_init_sriov(ibdev))
2907                 goto err_mad;
2908
2909         if (!iboe->nb.notifier_call) {
2910                 iboe->nb.notifier_call = mlx4_ib_netdev_event;
2911                 err = register_netdevice_notifier(&iboe->nb);
2912                 if (err) {
2913                         iboe->nb.notifier_call = NULL;
2914                         goto err_notif;
2915                 }
2916         }
2917         if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ROCE_V1_V2) {
2918                 err = mlx4_config_roce_v2_port(dev, ROCE_V2_UDP_DPORT);
2919                 if (err)
2920                         goto err_notif;
2921         }
2922
2923         for (j = 0; j < ARRAY_SIZE(mlx4_class_attributes); ++j) {
2924                 if (device_create_file(&ibdev->ib_dev.dev,
2925                                        mlx4_class_attributes[j]))
2926                         goto err_notif;
2927         }
2928
2929         ibdev->ib_active = true;
2930         mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
2931                 devlink_port_type_ib_set(mlx4_get_devlink_port(dev, i),
2932                                          &ibdev->ib_dev);
2933
2934         if (mlx4_is_mfunc(ibdev->dev))
2935                 init_pkeys(ibdev);
2936
2937         /* create paravirt contexts for any VFs which are active */
2938         if (mlx4_is_master(ibdev->dev)) {
2939                 for (j = 0; j < MLX4_MFUNC_MAX; j++) {
2940                         if (j == mlx4_master_func_num(ibdev->dev))
2941                                 continue;
2942                         if (mlx4_is_slave_active(ibdev->dev, j))
2943                                 do_slave_init(ibdev, j, 1);
2944                 }
2945         }
2946         return ibdev;
2947
2948 err_notif:
2949         if (ibdev->iboe.nb.notifier_call) {
2950                 if (unregister_netdevice_notifier(&ibdev->iboe.nb))
2951                         pr_warn("failure unregistering notifier\n");
2952                 ibdev->iboe.nb.notifier_call = NULL;
2953         }
2954         flush_workqueue(wq);
2955
2956         mlx4_ib_close_sriov(ibdev);
2957
2958 err_mad:
2959         mlx4_ib_mad_cleanup(ibdev);
2960
2961 err_reg:
2962         ib_unregister_device(&ibdev->ib_dev);
2963
2964 err_diag_counters:
2965         mlx4_ib_diag_cleanup(ibdev);
2966
2967 err_steer_free_bitmap:
2968         kfree(ibdev->ib_uc_qpns_bitmap);
2969
2970 err_steer_qp_release:
2971         mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
2972                               ibdev->steer_qpn_count);
2973 err_counter:
2974         for (i = 0; i < ibdev->num_ports; ++i)
2975                 mlx4_ib_delete_counters_table(ibdev, &ibdev->counters_table[i]);
2976
2977 err_map:
2978         mlx4_ib_free_eqs(dev, ibdev);
2979         iounmap(ibdev->uar_map);
2980
2981 err_uar:
2982         mlx4_uar_free(dev, &ibdev->priv_uar);
2983
2984 err_pd:
2985         mlx4_pd_free(dev, ibdev->priv_pdn);
2986
2987 err_dealloc:
2988         ib_dealloc_device(&ibdev->ib_dev);
2989
2990         return NULL;
2991 }
2992
2993 int mlx4_ib_steer_qp_alloc(struct mlx4_ib_dev *dev, int count, int *qpn)
2994 {
2995         int offset;
2996
2997         WARN_ON(!dev->ib_uc_qpns_bitmap);
2998
2999         offset = bitmap_find_free_region(dev->ib_uc_qpns_bitmap,
3000                                          dev->steer_qpn_count,
3001                                          get_count_order(count));
3002         if (offset < 0)
3003                 return offset;
3004
3005         *qpn = dev->steer_qpn_base + offset;
3006         return 0;
3007 }
3008
3009 void mlx4_ib_steer_qp_free(struct mlx4_ib_dev *dev, u32 qpn, int count)
3010 {
3011         if (!qpn ||
3012             dev->steering_support != MLX4_STEERING_MODE_DEVICE_MANAGED)
3013                 return;
3014
3015         if (WARN(qpn < dev->steer_qpn_base, "qpn = %u, steer_qpn_base = %u\n",
3016                  qpn, dev->steer_qpn_base))
3017                 /* not supposed to be here */
3018                 return;
3019
3020         bitmap_release_region(dev->ib_uc_qpns_bitmap,
3021                               qpn - dev->steer_qpn_base,
3022                               get_count_order(count));
3023 }
3024
3025 int mlx4_ib_steer_qp_reg(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
3026                          int is_attach)
3027 {
3028         int err;
3029         size_t flow_size;
3030         struct ib_flow_attr *flow = NULL;
3031         struct ib_flow_spec_ib *ib_spec;
3032
3033         if (is_attach) {
3034                 flow_size = sizeof(struct ib_flow_attr) +
3035                             sizeof(struct ib_flow_spec_ib);
3036                 flow = kzalloc(flow_size, GFP_KERNEL);
3037                 if (!flow)
3038                         return -ENOMEM;
3039                 flow->port = mqp->port;
3040                 flow->num_of_specs = 1;
3041                 flow->size = flow_size;
3042                 ib_spec = (struct ib_flow_spec_ib *)(flow + 1);
3043                 ib_spec->type = IB_FLOW_SPEC_IB;
3044                 ib_spec->size = sizeof(struct ib_flow_spec_ib);
3045                 /* Add an empty rule for IB L2 */
3046                 memset(&ib_spec->mask, 0, sizeof(ib_spec->mask));
3047
3048                 err = __mlx4_ib_create_flow(&mqp->ibqp, flow,
3049                                             IB_FLOW_DOMAIN_NIC,
3050                                             MLX4_FS_REGULAR,
3051                                             &mqp->reg_id);
3052         } else {
3053                 err = __mlx4_ib_destroy_flow(mdev->dev, mqp->reg_id);
3054         }
3055         kfree(flow);
3056         return err;
3057 }
3058
3059 static void mlx4_ib_remove(struct mlx4_dev *dev, void *ibdev_ptr)
3060 {
3061         struct mlx4_ib_dev *ibdev = ibdev_ptr;
3062         int p;
3063         int i;
3064
3065         mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
3066                 devlink_port_type_clear(mlx4_get_devlink_port(dev, i));
3067         ibdev->ib_active = false;
3068         flush_workqueue(wq);
3069
3070         mlx4_ib_close_sriov(ibdev);
3071         mlx4_ib_mad_cleanup(ibdev);
3072         ib_unregister_device(&ibdev->ib_dev);
3073         mlx4_ib_diag_cleanup(ibdev);
3074         if (ibdev->iboe.nb.notifier_call) {
3075                 if (unregister_netdevice_notifier(&ibdev->iboe.nb))
3076                         pr_warn("failure unregistering notifier\n");
3077                 ibdev->iboe.nb.notifier_call = NULL;
3078         }
3079
3080         mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
3081                               ibdev->steer_qpn_count);
3082         kfree(ibdev->ib_uc_qpns_bitmap);
3083
3084         iounmap(ibdev->uar_map);
3085         for (p = 0; p < ibdev->num_ports; ++p)
3086                 mlx4_ib_delete_counters_table(ibdev, &ibdev->counters_table[p]);
3087
3088         mlx4_foreach_port(p, dev, MLX4_PORT_TYPE_IB)
3089                 mlx4_CLOSE_PORT(dev, p);
3090
3091         mlx4_ib_free_eqs(dev, ibdev);
3092
3093         mlx4_uar_free(dev, &ibdev->priv_uar);
3094         mlx4_pd_free(dev, ibdev->priv_pdn);
3095         ib_dealloc_device(&ibdev->ib_dev);
3096 }
3097
3098 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init)
3099 {
3100         struct mlx4_ib_demux_work **dm = NULL;
3101         struct mlx4_dev *dev = ibdev->dev;
3102         int i;
3103         unsigned long flags;
3104         struct mlx4_active_ports actv_ports;
3105         unsigned int ports;
3106         unsigned int first_port;
3107
3108         if (!mlx4_is_master(dev))
3109                 return;
3110
3111         actv_ports = mlx4_get_active_ports(dev, slave);
3112         ports = bitmap_weight(actv_ports.ports, dev->caps.num_ports);
3113         first_port = find_first_bit(actv_ports.ports, dev->caps.num_ports);
3114
3115         dm = kcalloc(ports, sizeof(*dm), GFP_ATOMIC);
3116         if (!dm)
3117                 return;
3118
3119         for (i = 0; i < ports; i++) {
3120                 dm[i] = kmalloc(sizeof (struct mlx4_ib_demux_work), GFP_ATOMIC);
3121                 if (!dm[i]) {
3122                         while (--i >= 0)
3123                                 kfree(dm[i]);
3124                         goto out;
3125                 }
3126                 INIT_WORK(&dm[i]->work, mlx4_ib_tunnels_update_work);
3127                 dm[i]->port = first_port + i + 1;
3128                 dm[i]->slave = slave;
3129                 dm[i]->do_init = do_init;
3130                 dm[i]->dev = ibdev;
3131         }
3132         /* initialize or tear down tunnel QPs for the slave */
3133         spin_lock_irqsave(&ibdev->sriov.going_down_lock, flags);
3134         if (!ibdev->sriov.is_going_down) {
3135                 for (i = 0; i < ports; i++)
3136                         queue_work(ibdev->sriov.demux[i].ud_wq, &dm[i]->work);
3137                 spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
3138         } else {
3139                 spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
3140                 for (i = 0; i < ports; i++)
3141                         kfree(dm[i]);
3142         }
3143 out:
3144         kfree(dm);
3145         return;
3146 }
3147
3148 static void mlx4_ib_handle_catas_error(struct mlx4_ib_dev *ibdev)
3149 {
3150         struct mlx4_ib_qp *mqp;
3151         unsigned long flags_qp;
3152         unsigned long flags_cq;
3153         struct mlx4_ib_cq *send_mcq, *recv_mcq;
3154         struct list_head    cq_notify_list;
3155         struct mlx4_cq *mcq;
3156         unsigned long flags;
3157
3158         pr_warn("mlx4_ib_handle_catas_error was started\n");
3159         INIT_LIST_HEAD(&cq_notify_list);
3160
3161         /* Go over qp list reside on that ibdev, sync with create/destroy qp.*/
3162         spin_lock_irqsave(&ibdev->reset_flow_resource_lock, flags);
3163
3164         list_for_each_entry(mqp, &ibdev->qp_list, qps_list) {
3165                 spin_lock_irqsave(&mqp->sq.lock, flags_qp);
3166                 if (mqp->sq.tail != mqp->sq.head) {
3167                         send_mcq = to_mcq(mqp->ibqp.send_cq);
3168                         spin_lock_irqsave(&send_mcq->lock, flags_cq);
3169                         if (send_mcq->mcq.comp &&
3170                             mqp->ibqp.send_cq->comp_handler) {
3171                                 if (!send_mcq->mcq.reset_notify_added) {
3172                                         send_mcq->mcq.reset_notify_added = 1;
3173                                         list_add_tail(&send_mcq->mcq.reset_notify,
3174                                                       &cq_notify_list);
3175                                 }
3176                         }
3177                         spin_unlock_irqrestore(&send_mcq->lock, flags_cq);
3178                 }
3179                 spin_unlock_irqrestore(&mqp->sq.lock, flags_qp);
3180                 /* Now, handle the QP's receive queue */
3181                 spin_lock_irqsave(&mqp->rq.lock, flags_qp);
3182                 /* no handling is needed for SRQ */
3183                 if (!mqp->ibqp.srq) {
3184                         if (mqp->rq.tail != mqp->rq.head) {
3185                                 recv_mcq = to_mcq(mqp->ibqp.recv_cq);
3186                                 spin_lock_irqsave(&recv_mcq->lock, flags_cq);
3187                                 if (recv_mcq->mcq.comp &&
3188                                     mqp->ibqp.recv_cq->comp_handler) {
3189                                         if (!recv_mcq->mcq.reset_notify_added) {
3190                                                 recv_mcq->mcq.reset_notify_added = 1;
3191                                                 list_add_tail(&recv_mcq->mcq.reset_notify,
3192                                                               &cq_notify_list);
3193                                         }
3194                                 }
3195                                 spin_unlock_irqrestore(&recv_mcq->lock,
3196                                                        flags_cq);
3197                         }
3198                 }
3199                 spin_unlock_irqrestore(&mqp->rq.lock, flags_qp);
3200         }
3201
3202         list_for_each_entry(mcq, &cq_notify_list, reset_notify) {
3203                 mcq->comp(mcq);
3204         }
3205         spin_unlock_irqrestore(&ibdev->reset_flow_resource_lock, flags);
3206         pr_warn("mlx4_ib_handle_catas_error ended\n");
3207 }
3208
3209 static void handle_bonded_port_state_event(struct work_struct *work)
3210 {
3211         struct ib_event_work *ew =
3212                 container_of(work, struct ib_event_work, work);
3213         struct mlx4_ib_dev *ibdev = ew->ib_dev;
3214         enum ib_port_state bonded_port_state = IB_PORT_NOP;
3215         int i;
3216         struct ib_event ibev;
3217
3218         kfree(ew);
3219         spin_lock_bh(&ibdev->iboe.lock);
3220         for (i = 0; i < MLX4_MAX_PORTS; ++i) {
3221                 struct net_device *curr_netdev = ibdev->iboe.netdevs[i];
3222                 enum ib_port_state curr_port_state;
3223
3224                 if (!curr_netdev)
3225                         continue;
3226
3227                 curr_port_state =
3228                         (netif_running(curr_netdev) &&
3229                          netif_carrier_ok(curr_netdev)) ?
3230                         IB_PORT_ACTIVE : IB_PORT_DOWN;
3231
3232                 bonded_port_state = (bonded_port_state != IB_PORT_ACTIVE) ?
3233                         curr_port_state : IB_PORT_ACTIVE;
3234         }
3235         spin_unlock_bh(&ibdev->iboe.lock);
3236
3237         ibev.device = &ibdev->ib_dev;
3238         ibev.element.port_num = 1;
3239         ibev.event = (bonded_port_state == IB_PORT_ACTIVE) ?
3240                 IB_EVENT_PORT_ACTIVE : IB_EVENT_PORT_ERR;
3241
3242         ib_dispatch_event(&ibev);
3243 }
3244
3245 void mlx4_ib_sl2vl_update(struct mlx4_ib_dev *mdev, int port)
3246 {
3247         u64 sl2vl;
3248         int err;
3249
3250         err = mlx4_ib_query_sl2vl(&mdev->ib_dev, port, &sl2vl);
3251         if (err) {
3252                 pr_err("Unable to get current sl to vl mapping for port %d.  Using all zeroes (%d)\n",
3253                        port, err);
3254                 sl2vl = 0;
3255         }
3256         atomic64_set(&mdev->sl2vl[port - 1], sl2vl);
3257 }
3258
3259 static void ib_sl2vl_update_work(struct work_struct *work)
3260 {
3261         struct ib_event_work *ew = container_of(work, struct ib_event_work, work);
3262         struct mlx4_ib_dev *mdev = ew->ib_dev;
3263         int port = ew->port;
3264
3265         mlx4_ib_sl2vl_update(mdev, port);
3266
3267         kfree(ew);
3268 }
3269
3270 void mlx4_sched_ib_sl2vl_update_work(struct mlx4_ib_dev *ibdev,
3271                                      int port)
3272 {
3273         struct ib_event_work *ew;
3274
3275         ew = kmalloc(sizeof(*ew), GFP_ATOMIC);
3276         if (ew) {
3277                 INIT_WORK(&ew->work, ib_sl2vl_update_work);
3278                 ew->port = port;
3279                 ew->ib_dev = ibdev;
3280                 queue_work(wq, &ew->work);
3281         }
3282 }
3283
3284 static void mlx4_ib_event(struct mlx4_dev *dev, void *ibdev_ptr,
3285                           enum mlx4_dev_event event, unsigned long param)
3286 {
3287         struct ib_event ibev;
3288         struct mlx4_ib_dev *ibdev = to_mdev((struct ib_device *) ibdev_ptr);
3289         struct mlx4_eqe *eqe = NULL;
3290         struct ib_event_work *ew;
3291         int p = 0;
3292
3293         if (mlx4_is_bonded(dev) &&
3294             ((event == MLX4_DEV_EVENT_PORT_UP) ||
3295             (event == MLX4_DEV_EVENT_PORT_DOWN))) {
3296                 ew = kmalloc(sizeof(*ew), GFP_ATOMIC);
3297                 if (!ew)
3298                         return;
3299                 INIT_WORK(&ew->work, handle_bonded_port_state_event);
3300                 ew->ib_dev = ibdev;
3301                 queue_work(wq, &ew->work);
3302                 return;
3303         }
3304
3305         if (event == MLX4_DEV_EVENT_PORT_MGMT_CHANGE)
3306                 eqe = (struct mlx4_eqe *)param;
3307         else
3308                 p = (int) param;
3309
3310         switch (event) {
3311         case MLX4_DEV_EVENT_PORT_UP:
3312                 if (p > ibdev->num_ports)
3313                         return;
3314                 if (!mlx4_is_slave(dev) &&
3315                     rdma_port_get_link_layer(&ibdev->ib_dev, p) ==
3316                         IB_LINK_LAYER_INFINIBAND) {
3317                         if (mlx4_is_master(dev))
3318                                 mlx4_ib_invalidate_all_guid_record(ibdev, p);
3319                         if (ibdev->dev->flags & MLX4_FLAG_SECURE_HOST &&
3320                             !(ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_SL_TO_VL_CHANGE_EVENT))
3321                                 mlx4_sched_ib_sl2vl_update_work(ibdev, p);
3322                 }
3323                 ibev.event = IB_EVENT_PORT_ACTIVE;
3324                 break;
3325
3326         case MLX4_DEV_EVENT_PORT_DOWN:
3327                 if (p > ibdev->num_ports)
3328                         return;
3329                 ibev.event = IB_EVENT_PORT_ERR;
3330                 break;
3331
3332         case MLX4_DEV_EVENT_CATASTROPHIC_ERROR:
3333                 ibdev->ib_active = false;
3334                 ibev.event = IB_EVENT_DEVICE_FATAL;
3335                 mlx4_ib_handle_catas_error(ibdev);
3336                 break;
3337
3338         case MLX4_DEV_EVENT_PORT_MGMT_CHANGE:
3339                 ew = kmalloc(sizeof *ew, GFP_ATOMIC);
3340                 if (!ew)
3341                         break;
3342
3343                 INIT_WORK(&ew->work, handle_port_mgmt_change_event);
3344                 memcpy(&ew->ib_eqe, eqe, sizeof *eqe);
3345                 ew->ib_dev = ibdev;
3346                 /* need to queue only for port owner, which uses GEN_EQE */
3347                 if (mlx4_is_master(dev))
3348                         queue_work(wq, &ew->work);
3349                 else
3350                         handle_port_mgmt_change_event(&ew->work);
3351                 return;
3352
3353         case MLX4_DEV_EVENT_SLAVE_INIT:
3354                 /* here, p is the slave id */
3355                 do_slave_init(ibdev, p, 1);
3356                 if (mlx4_is_master(dev)) {
3357                         int i;
3358
3359                         for (i = 1; i <= ibdev->num_ports; i++) {
3360                                 if (rdma_port_get_link_layer(&ibdev->ib_dev, i)
3361                                         == IB_LINK_LAYER_INFINIBAND)
3362                                         mlx4_ib_slave_alias_guid_event(ibdev,
3363                                                                        p, i,
3364                                                                        1);
3365                         }
3366                 }
3367                 return;
3368
3369         case MLX4_DEV_EVENT_SLAVE_SHUTDOWN:
3370                 if (mlx4_is_master(dev)) {
3371                         int i;
3372
3373                         for (i = 1; i <= ibdev->num_ports; i++) {
3374                                 if (rdma_port_get_link_layer(&ibdev->ib_dev, i)
3375                                         == IB_LINK_LAYER_INFINIBAND)
3376                                         mlx4_ib_slave_alias_guid_event(ibdev,
3377                                                                        p, i,
3378                                                                        0);
3379                         }
3380                 }
3381                 /* here, p is the slave id */
3382                 do_slave_init(ibdev, p, 0);
3383                 return;
3384
3385         default:
3386                 return;
3387         }
3388
3389         ibev.device           = ibdev_ptr;
3390         ibev.element.port_num = mlx4_is_bonded(ibdev->dev) ? 1 : (u8)p;
3391
3392         ib_dispatch_event(&ibev);
3393 }
3394
3395 static struct mlx4_interface mlx4_ib_interface = {
3396         .add            = mlx4_ib_add,
3397         .remove         = mlx4_ib_remove,
3398         .event          = mlx4_ib_event,
3399         .protocol       = MLX4_PROT_IB_IPV6,
3400         .flags          = MLX4_INTFF_BONDING
3401 };
3402
3403 static int __init mlx4_ib_init(void)
3404 {
3405         int err;
3406
3407         wq = alloc_ordered_workqueue("mlx4_ib", WQ_MEM_RECLAIM);
3408         if (!wq)
3409                 return -ENOMEM;
3410
3411         err = mlx4_ib_mcg_init();
3412         if (err)
3413                 goto clean_wq;
3414
3415         err = mlx4_register_interface(&mlx4_ib_interface);
3416         if (err)
3417                 goto clean_mcg;
3418
3419         return 0;
3420
3421 clean_mcg:
3422         mlx4_ib_mcg_destroy();
3423
3424 clean_wq:
3425         destroy_workqueue(wq);
3426         return err;
3427 }
3428
3429 static void __exit mlx4_ib_cleanup(void)
3430 {
3431         mlx4_unregister_interface(&mlx4_ib_interface);
3432         mlx4_ib_mcg_destroy();
3433         destroy_workqueue(wq);
3434 }
3435
3436 module_init(mlx4_ib_init);
3437 module_exit(mlx4_ib_cleanup);