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 void free_gid_entry(struct gid_entry *entry)
 250 {
 251         memset(&entry->gid, 0, sizeof(entry->gid));
 252         kfree(entry->ctx);
 253         entry->ctx = NULL;
 254 }
 255
 256 static int mlx4_ib_add_gid(const struct ib_gid_attr *attr, void **context)
 257 {
 258         struct mlx4_ib_dev *ibdev = to_mdev(attr->device);
 259         struct mlx4_ib_iboe *iboe = &ibdev->iboe;
 260         struct mlx4_port_gid_table   *port_gid_table;
 261         int free = -1, found = -1;
 262         int ret = 0;
 263         int hw_update = 0;
 264         int i;
 265         struct gid_entry *gids = NULL;
 266         u16 vlan_id = 0xffff;
 267         u8 mac[ETH_ALEN];
 268
 269         if (!rdma_cap_roce_gid_table(attr->device, attr->port_num))
 270                 return -EINVAL;
 271
 272         if (attr->port_num > MLX4_MAX_PORTS)
 273                 return -EINVAL;
 274
 275         if (!context)
 276                 return -EINVAL;
 277
 278         ret = rdma_read_gid_l2_fields(attr, &vlan_id, &mac[0]);
 279         if (ret)
 280                 return ret;
 281         port_gid_table = &iboe->gids[attr->port_num - 1];
 282         spin_lock_bh(&iboe->lock);
 283         for (i = 0; i < MLX4_MAX_PORT_GIDS; ++i) {
 284                 if (!memcmp(&port_gid_table->gids[i].gid,
 285                             &attr->gid, sizeof(attr->gid)) &&
 286                     port_gid_table->gids[i].gid_type == attr->gid_type &&
 287                     port_gid_table->gids[i].vlan_id == vlan_id)  {
 288                         found = i;
 289                         break;
 290                 }
 291                 if (free < 0 && rdma_is_zero_gid(&port_gid_table->gids[i].gid))
 292                         free = i; /* HW has space */
 293         }
 294
 295         if (found < 0) {
 296                 if (free < 0) {
 297                         ret = -ENOSPC;
 298                 } else {
 299                         port_gid_table->gids[free].ctx = kmalloc(sizeof(*port_gid_table->gids[free].ctx), GFP_ATOMIC);
 300                         if (!port_gid_table->gids[free].ctx) {
 301                                 ret = -ENOMEM;
 302                         } else {
 303                                 *context = port_gid_table->gids[free].ctx;
 304                                 memcpy(&port_gid_table->gids[free].gid,
 305                                        &attr->gid, sizeof(attr->gid));
 306                                 port_gid_table->gids[free].gid_type = attr->gid_type;
 307                                 port_gid_table->gids[free].vlan_id = vlan_id;
 308                                 port_gid_table->gids[free].ctx->real_index = free;
 309                                 port_gid_table->gids[free].ctx->refcount = 1;
 310                                 hw_update = 1;
 311                         }
 312                 }
 313         } else {
 314                 struct gid_cache_context *ctx = port_gid_table->gids[found].ctx;
 315                 *context = ctx;
 316                 ctx->refcount++;
 317         }
 318         if (!ret && hw_update) {
 319                 gids = kmalloc_array(MLX4_MAX_PORT_GIDS, sizeof(*gids),
 320                                      GFP_ATOMIC);
 321                 if (!gids) {
 322                         ret = -ENOMEM;
 323                         *context = NULL;
 324                         free_gid_entry(&port_gid_table->gids[free]);
 325                 } else {
 326                         for (i = 0; i < MLX4_MAX_PORT_GIDS; i++) {
 327                                 memcpy(&gids[i].gid, &port_gid_table->gids[i].gid, sizeof(union ib_gid));
 328                                 gids[i].gid_type = port_gid_table->gids[i].gid_type;
 329                         }
 330                 }
 331         }
 332         spin_unlock_bh(&iboe->lock);
 333
 334         if (!ret && hw_update) {
 335                 ret = mlx4_ib_update_gids(gids, ibdev, attr->port_num);
 336                 if (ret) {
 337                         spin_lock_bh(&iboe->lock);
 338                         *context = NULL;
 339                         free_gid_entry(&port_gid_table->gids[free]);
 340                         spin_unlock_bh(&iboe->lock);
 341                 }
 342                 kfree(gids);
 343         }
 344
 345         return ret;
 346 }
 347
 348 static int mlx4_ib_del_gid(const struct ib_gid_attr *attr, void **context)
 349 {
 350         struct gid_cache_context *ctx = *context;
 351         struct mlx4_ib_dev *ibdev = to_mdev(attr->device);
 352         struct mlx4_ib_iboe *iboe = &ibdev->iboe;
 353         struct mlx4_port_gid_table   *port_gid_table;
 354         int ret = 0;
 355         int hw_update = 0;
 356         struct gid_entry *gids = NULL;
 357
 358         if (!rdma_cap_roce_gid_table(attr->device, attr->port_num))
 359                 return -EINVAL;
 360
 361         if (attr->port_num > MLX4_MAX_PORTS)
 362                 return -EINVAL;
 363
 364         port_gid_table = &iboe->gids[attr->port_num - 1];
 365         spin_lock_bh(&iboe->lock);
 366         if (ctx) {
 367                 ctx->refcount--;
 368                 if (!ctx->refcount) {
 369                         unsigned int real_index = ctx->real_index;
 370
 371                         free_gid_entry(&port_gid_table->gids[real_index]);
 372                         hw_update = 1;
 373                 }
 374         }
 375         if (!ret && hw_update) {
 376                 int i;
 377
 378                 gids = kmalloc_array(MLX4_MAX_PORT_GIDS, sizeof(*gids),
 379                                      GFP_ATOMIC);
 380                 if (!gids) {
 381                         ret = -ENOMEM;
 382                 } else {
 383                         for (i = 0; i < MLX4_MAX_PORT_GIDS; i++) {
 384                                 memcpy(&gids[i].gid,
 385                                        &port_gid_table->gids[i].gid,
 386                                        sizeof(union ib_gid));
 387                                 gids[i].gid_type =
 388                                     port_gid_table->gids[i].gid_type;
 389                         }
 390                 }
 391         }
 392         spin_unlock_bh(&iboe->lock);
 393
 394         if (!ret && hw_update) {
 395                 ret = mlx4_ib_update_gids(gids, ibdev, attr->port_num);
 396                 kfree(gids);
 397         }
 398         return ret;
 399 }
 400
 401 int mlx4_ib_gid_index_to_real_index(struct mlx4_ib_dev *ibdev,
 402                                     const struct ib_gid_attr *attr)
 403 {
 404         struct mlx4_ib_iboe *iboe = &ibdev->iboe;
 405         struct gid_cache_context *ctx = NULL;
 406         struct mlx4_port_gid_table   *port_gid_table;
 407         int real_index = -EINVAL;
 408         int i;
 409         unsigned long flags;
 410         u8 port_num = attr->port_num;
 411
 412         if (port_num > MLX4_MAX_PORTS)
 413                 return -EINVAL;
 414
 415         if (mlx4_is_bonded(ibdev->dev))
 416                 port_num = 1;
 417
 418         if (!rdma_cap_roce_gid_table(&ibdev->ib_dev, port_num))
 419                 return attr->index;
 420
 421         spin_lock_irqsave(&iboe->lock, flags);
 422         port_gid_table = &iboe->gids[port_num - 1];
 423
 424         for (i = 0; i < MLX4_MAX_PORT_GIDS; ++i)
 425                 if (!memcmp(&port_gid_table->gids[i].gid,
 426                             &attr->gid, sizeof(attr->gid)) &&
 427                     attr->gid_type == port_gid_table->gids[i].gid_type) {
 428                         ctx = port_gid_table->gids[i].ctx;
 429                         break;
 430                 }
 431         if (ctx)
 432                 real_index = ctx->real_index;
 433         spin_unlock_irqrestore(&iboe->lock, flags);
 434         return real_index;
 435 }
 436
 437 static int mlx4_ib_query_device(struct ib_device *ibdev,
 438                                 struct ib_device_attr *props,
 439                                 struct ib_udata *uhw)
 440 {
 441         struct mlx4_ib_dev *dev = to_mdev(ibdev);
 442         struct ib_smp *in_mad  = NULL;
 443         struct ib_smp *out_mad = NULL;
 444         int err;
 445         int have_ib_ports;
 446         struct mlx4_uverbs_ex_query_device cmd;
 447         struct mlx4_uverbs_ex_query_device_resp resp = {};
 448         struct mlx4_clock_params clock_params;
 449
 450         if (uhw->inlen) {
 451                 if (uhw->inlen < sizeof(cmd))
 452                         return -EINVAL;
 453
 454                 err = ib_copy_from_udata(&cmd, uhw, sizeof(cmd));
 455                 if (err)
 456                         return err;
 457
 458                 if (cmd.comp_mask)
 459                         return -EINVAL;
 460
 461                 if (cmd.reserved)
 462                         return -EINVAL;
 463         }
 464
 465         resp.response_length = offsetof(typeof(resp), response_length) +
 466                 sizeof(resp.response_length);
 467         in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
 468         out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
 469         err = -ENOMEM;
 470         if (!in_mad || !out_mad)
 471                 goto out;
 472
 473         init_query_mad(in_mad);
 474         in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
 475
 476         err = mlx4_MAD_IFC(to_mdev(ibdev), MLX4_MAD_IFC_IGNORE_KEYS,
 477                            1, NULL, NULL, in_mad, out_mad);
 478         if (err)
 479                 goto out;
 480
 481         memset(props, 0, sizeof *props);
 482
 483         have_ib_ports = num_ib_ports(dev->dev);
 484
 485         props->fw_ver = dev->dev->caps.fw_ver;
 486         props->device_cap_flags    = IB_DEVICE_CHANGE_PHY_PORT |
 487                 IB_DEVICE_PORT_ACTIVE_EVENT             |
 488                 IB_DEVICE_SYS_IMAGE_GUID                |
 489                 IB_DEVICE_RC_RNR_NAK_GEN                |
 490                 IB_DEVICE_BLOCK_MULTICAST_LOOPBACK;
 491         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_PKEY_CNTR)
 492                 props->device_cap_flags |= IB_DEVICE_BAD_PKEY_CNTR;
 493         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_QKEY_CNTR)
 494                 props->device_cap_flags |= IB_DEVICE_BAD_QKEY_CNTR;
 495         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_APM && have_ib_ports)
 496                 props->device_cap_flags |= IB_DEVICE_AUTO_PATH_MIG;
 497         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_UD_AV_PORT)
 498                 props->device_cap_flags |= IB_DEVICE_UD_AV_PORT_ENFORCE;
 499         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_IPOIB_CSUM)
 500                 props->device_cap_flags |= IB_DEVICE_UD_IP_CSUM;
 501         if (dev->dev->caps.max_gso_sz &&
 502             (dev->dev->rev_id != MLX4_IB_CARD_REV_A0) &&
 503             (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BLH))
 504                 props->device_cap_flags |= IB_DEVICE_UD_TSO;
 505         if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_RESERVED_LKEY)
 506                 props->device_cap_flags |= IB_DEVICE_LOCAL_DMA_LKEY;
 507         if ((dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_LOCAL_INV) &&
 508             (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_REMOTE_INV) &&
 509             (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_FAST_REG_WR))
 510                 props->device_cap_flags |= IB_DEVICE_MEM_MGT_EXTENSIONS;
 511         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC)
 512                 props->device_cap_flags |= IB_DEVICE_XRC;
 513         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW)
 514                 props->device_cap_flags |= IB_DEVICE_MEM_WINDOW;
 515         if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
 516                 if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_WIN_TYPE_2B)
 517                         props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2B;
 518                 else
 519                         props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2A;
 520         }
 521         if (dev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED)
 522                 props->device_cap_flags |= IB_DEVICE_MANAGED_FLOW_STEERING;
 523
 524         props->device_cap_flags |= IB_DEVICE_RAW_IP_CSUM;
 525
 526         props->vendor_id           = be32_to_cpup((__be32 *) (out_mad->data + 36)) &
 527                 0xffffff;
 528         props->vendor_part_id      = dev->dev->persist->pdev->device;
 529         props->hw_ver              = be32_to_cpup((__be32 *) (out_mad->data + 32));
 530         memcpy(&props->sys_image_guid, out_mad->data +  4, 8);
 531
 532         props->max_mr_size         = ~0ull;
 533         props->page_size_cap       = dev->dev->caps.page_size_cap;
 534         props->max_qp              = dev->dev->quotas.qp;
 535         props->max_qp_wr           = dev->dev->caps.max_wqes - MLX4_IB_SQ_MAX_SPARE;
 536         props->max_send_sge =
 537                 min(dev->dev->caps.max_sq_sg, dev->dev->caps.max_rq_sg);
 538         props->max_recv_sge =
 539                 min(dev->dev->caps.max_sq_sg, dev->dev->caps.max_rq_sg);
 540         props->max_sge_rd = MLX4_MAX_SGE_RD;
 541         props->max_cq              = dev->dev->quotas.cq;
 542         props->max_cqe             = dev->dev->caps.max_cqes;
 543         props->max_mr              = dev->dev->quotas.mpt;
 544         props->max_pd              = dev->dev->caps.num_pds - dev->dev->caps.reserved_pds;
 545         props->max_qp_rd_atom      = dev->dev->caps.max_qp_dest_rdma;
 546         props->max_qp_init_rd_atom = dev->dev->caps.max_qp_init_rdma;
 547         props->max_res_rd_atom     = props->max_qp_rd_atom * props->max_qp;
 548         props->max_srq             = dev->dev->quotas.srq;
 549         props->max_srq_wr          = dev->dev->caps.max_srq_wqes - 1;
 550         props->max_srq_sge         = dev->dev->caps.max_srq_sge;
 551         props->max_fast_reg_page_list_len = MLX4_MAX_FAST_REG_PAGES;
 552         props->local_ca_ack_delay  = dev->dev->caps.local_ca_ack_delay;
 553         props->atomic_cap          = dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_ATOMIC ?
 554                 IB_ATOMIC_HCA : IB_ATOMIC_NONE;
 555         props->masked_atomic_cap   = props->atomic_cap;
 556         props->max_pkeys           = dev->dev->caps.pkey_table_len[1];
 557         props->max_mcast_grp       = dev->dev->caps.num_mgms + dev->dev->caps.num_amgms;
 558         props->max_mcast_qp_attach = dev->dev->caps.num_qp_per_mgm;
 559         props->max_total_mcast_qp_attach = props->max_mcast_qp_attach *
 560                                            props->max_mcast_grp;
 561         props->hca_core_clock = dev->dev->caps.hca_core_clock * 1000UL;
 562         props->timestamp_mask = 0xFFFFFFFFFFFFULL;
 563         props->max_ah = INT_MAX;
 564
 565         if (mlx4_ib_port_link_layer(ibdev, 1) == IB_LINK_LAYER_ETHERNET ||
 566             mlx4_ib_port_link_layer(ibdev, 2) == IB_LINK_LAYER_ETHERNET) {
 567                 if (dev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_RSS) {
 568                         props->rss_caps.max_rwq_indirection_tables =
 569                                 props->max_qp;
 570                         props->rss_caps.max_rwq_indirection_table_size =
 571                                 dev->dev->caps.max_rss_tbl_sz;
 572                         props->rss_caps.supported_qpts = 1 << IB_QPT_RAW_PACKET;
 573                         props->max_wq_type_rq = props->max_qp;
 574                 }
 575
 576                 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_FCS_KEEP)
 577                         props->raw_packet_caps |= IB_RAW_PACKET_CAP_SCATTER_FCS;
 578         }
 579
 580         props->cq_caps.max_cq_moderation_count = MLX4_MAX_CQ_COUNT;
 581         props->cq_caps.max_cq_moderation_period = MLX4_MAX_CQ_PERIOD;
 582
 583         if (uhw->outlen >= resp.response_length + sizeof(resp.hca_core_clock_offset)) {
 584                 resp.response_length += sizeof(resp.hca_core_clock_offset);
 585                 if (!mlx4_get_internal_clock_params(dev->dev, &clock_params)) {
 586                         resp.comp_mask |= MLX4_IB_QUERY_DEV_RESP_MASK_CORE_CLOCK_OFFSET;
 587                         resp.hca_core_clock_offset = clock_params.offset % PAGE_SIZE;
 588                 }
 589         }
 590
 591         if (uhw->outlen >= resp.response_length +
 592             sizeof(resp.max_inl_recv_sz)) {
 593                 resp.response_length += sizeof(resp.max_inl_recv_sz);
 594                 resp.max_inl_recv_sz  = dev->dev->caps.max_rq_sg *
 595                         sizeof(struct mlx4_wqe_data_seg);
 596         }
 597
 598         if (offsetofend(typeof(resp), rss_caps) <= uhw->outlen) {
 599                 if (props->rss_caps.supported_qpts) {
 600                         resp.rss_caps.rx_hash_function =
 601                                 MLX4_IB_RX_HASH_FUNC_TOEPLITZ;
 602
 603                         resp.rss_caps.rx_hash_fields_mask =
 604                                 MLX4_IB_RX_HASH_SRC_IPV4 |
 605                                 MLX4_IB_RX_HASH_DST_IPV4 |
 606                                 MLX4_IB_RX_HASH_SRC_IPV6 |
 607                                 MLX4_IB_RX_HASH_DST_IPV6 |
 608                                 MLX4_IB_RX_HASH_SRC_PORT_TCP |
 609                                 MLX4_IB_RX_HASH_DST_PORT_TCP |
 610                                 MLX4_IB_RX_HASH_SRC_PORT_UDP |
 611                                 MLX4_IB_RX_HASH_DST_PORT_UDP;
 612
 613                         if (dev->dev->caps.tunnel_offload_mode ==
 614                             MLX4_TUNNEL_OFFLOAD_MODE_VXLAN)
 615                                 resp.rss_caps.rx_hash_fields_mask |=
 616                                         MLX4_IB_RX_HASH_INNER;
 617                 }
 618                 resp.response_length = offsetof(typeof(resp), rss_caps) +
 619                                        sizeof(resp.rss_caps);
 620         }
 621
 622         if (offsetofend(typeof(resp), tso_caps) <= uhw->outlen) {
 623                 if (dev->dev->caps.max_gso_sz &&
 624                     ((mlx4_ib_port_link_layer(ibdev, 1) ==
 625                     IB_LINK_LAYER_ETHERNET) ||
 626                     (mlx4_ib_port_link_layer(ibdev, 2) ==
 627                     IB_LINK_LAYER_ETHERNET))) {
 628                         resp.tso_caps.max_tso = dev->dev->caps.max_gso_sz;
 629                         resp.tso_caps.supported_qpts |=
 630                                 1 << IB_QPT_RAW_PACKET;
 631                 }
 632                 resp.response_length = offsetof(typeof(resp), tso_caps) +
 633                                        sizeof(resp.tso_caps);
 634         }
 635
 636         if (uhw->outlen) {
 637                 err = ib_copy_to_udata(uhw, &resp, resp.response_length);
 638                 if (err)
 639                         goto out;
 640         }
 641 out:
 642         kfree(in_mad);
 643         kfree(out_mad);
 644
 645         return err;
 646 }
 647
 648 static enum rdma_link_layer
 649 mlx4_ib_port_link_layer(struct ib_device *device, u8 port_num)
 650 {
 651         struct mlx4_dev *dev = to_mdev(device)->dev;
 652
 653         return dev->caps.port_mask[port_num] == MLX4_PORT_TYPE_IB ?
 654                 IB_LINK_LAYER_INFINIBAND : IB_LINK_LAYER_ETHERNET;
 655 }
 656
 657 static int ib_link_query_port(struct ib_device *ibdev, u8 port,
 658                               struct ib_port_attr *props, int netw_view)
 659 {
 660         struct ib_smp *in_mad  = NULL;
 661         struct ib_smp *out_mad = NULL;
 662         int ext_active_speed;
 663         int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
 664         int err = -ENOMEM;
 665
 666         in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
 667         out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
 668         if (!in_mad || !out_mad)
 669                 goto out;
 670
 671         init_query_mad(in_mad);
 672         in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
 673         in_mad->attr_mod = cpu_to_be32(port);
 674
 675         if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
 676                 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
 677
 678         err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
 679                                 in_mad, out_mad);
 680         if (err)
 681                 goto out;
 682
 683
 684         props->lid              = be16_to_cpup((__be16 *) (out_mad->data + 16));
 685         props->lmc              = out_mad->data[34] & 0x7;
 686         props->sm_lid           = be16_to_cpup((__be16 *) (out_mad->data + 18));
 687         props->sm_sl            = out_mad->data[36] & 0xf;
 688         props->state            = out_mad->data[32] & 0xf;
 689         props->phys_state       = out_mad->data[33] >> 4;
 690         props->port_cap_flags   = be32_to_cpup((__be32 *) (out_mad->data + 20));
 691         if (netw_view)
 692                 props->gid_tbl_len = out_mad->data[50];
 693         else
 694                 props->gid_tbl_len = to_mdev(ibdev)->dev->caps.gid_table_len[port];
 695         props->max_msg_sz       = to_mdev(ibdev)->dev->caps.max_msg_sz;
 696         props->pkey_tbl_len     = to_mdev(ibdev)->dev->caps.pkey_table_len[port];
 697         props->bad_pkey_cntr    = be16_to_cpup((__be16 *) (out_mad->data + 46));
 698         props->qkey_viol_cntr   = be16_to_cpup((__be16 *) (out_mad->data + 48));
 699         props->active_width     = out_mad->data[31] & 0xf;
 700         props->active_speed     = out_mad->data[35] >> 4;
 701         props->max_mtu          = out_mad->data[41] & 0xf;
 702         props->active_mtu       = out_mad->data[36] >> 4;
 703         props->subnet_timeout   = out_mad->data[51] & 0x1f;
 704         props->max_vl_num       = out_mad->data[37] >> 4;
 705         props->init_type_reply  = out_mad->data[41] >> 4;
 706
 707         /* Check if extended speeds (EDR/FDR/...) are supported */
 708         if (props->port_cap_flags & IB_PORT_EXTENDED_SPEEDS_SUP) {
 709                 ext_active_speed = out_mad->data[62] >> 4;
 710
 711                 switch (ext_active_speed) {
 712                 case 1:
 713                         props->active_speed = IB_SPEED_FDR;
 714                         break;
 715                 case 2:
 716                         props->active_speed = IB_SPEED_EDR;
 717                         break;
 718                 }
 719         }
 720
 721         /* If reported active speed is QDR, check if is FDR-10 */
 722         if (props->active_speed == IB_SPEED_QDR) {
 723                 init_query_mad(in_mad);
 724                 in_mad->attr_id = MLX4_ATTR_EXTENDED_PORT_INFO;
 725                 in_mad->attr_mod = cpu_to_be32(port);
 726
 727                 err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port,
 728                                    NULL, NULL, in_mad, out_mad);
 729                 if (err)
 730                         goto out;
 731
 732                 /* Checking LinkSpeedActive for FDR-10 */
 733                 if (out_mad->data[15] & 0x1)
 734                         props->active_speed = IB_SPEED_FDR10;
 735         }
 736
 737         /* Avoid wrong speed value returned by FW if the IB link is down. */
 738         if (props->state == IB_PORT_DOWN)
 739                  props->active_speed = IB_SPEED_SDR;
 740
 741 out:
 742         kfree(in_mad);
 743         kfree(out_mad);
 744         return err;
 745 }
 746
 747 static u8 state_to_phys_state(enum ib_port_state state)
 748 {
 749         return state == IB_PORT_ACTIVE ?
 750                 IB_PORT_PHYS_STATE_LINK_UP : IB_PORT_PHYS_STATE_DISABLED;
 751 }
 752
 753 static int eth_link_query_port(struct ib_device *ibdev, u8 port,
 754                                struct ib_port_attr *props)
 755 {
 756
 757         struct mlx4_ib_dev *mdev = to_mdev(ibdev);
 758         struct mlx4_ib_iboe *iboe = &mdev->iboe;
 759         struct net_device *ndev;
 760         enum ib_mtu tmp;
 761         struct mlx4_cmd_mailbox *mailbox;
 762         int err = 0;
 763         int is_bonded = mlx4_is_bonded(mdev->dev);
 764
 765         mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
 766         if (IS_ERR(mailbox))
 767                 return PTR_ERR(mailbox);
 768
 769         err = mlx4_cmd_box(mdev->dev, 0, mailbox->dma, port, 0,
 770                            MLX4_CMD_QUERY_PORT, MLX4_CMD_TIME_CLASS_B,
 771                            MLX4_CMD_WRAPPED);
 772         if (err)
 773                 goto out;
 774
 775         props->active_width     =  (((u8 *)mailbox->buf)[5] == 0x40) ||
 776                                    (((u8 *)mailbox->buf)[5] == 0x20 /*56Gb*/) ?
 777                                            IB_WIDTH_4X : IB_WIDTH_1X;
 778         props->active_speed     =  (((u8 *)mailbox->buf)[5] == 0x20 /*56Gb*/) ?
 779                                            IB_SPEED_FDR : IB_SPEED_QDR;
 780         props->port_cap_flags   = IB_PORT_CM_SUP;
 781         props->ip_gids = true;
 782         props->gid_tbl_len      = mdev->dev->caps.gid_table_len[port];
 783         props->max_msg_sz       = mdev->dev->caps.max_msg_sz;
 784         if (mdev->dev->caps.pkey_table_len[port])
 785                 props->pkey_tbl_len = 1;
 786         props->max_mtu          = IB_MTU_4096;
 787         props->max_vl_num       = 2;
 788         props->state            = IB_PORT_DOWN;
 789         props->phys_state       = state_to_phys_state(props->state);
 790         props->active_mtu       = IB_MTU_256;
 791         spin_lock_bh(&iboe->lock);
 792         ndev = iboe->netdevs[port - 1];
 793         if (ndev && is_bonded) {
 794                 rcu_read_lock(); /* required to get upper dev */
 795                 ndev = netdev_master_upper_dev_get_rcu(ndev);
 796                 rcu_read_unlock();
 797         }
 798         if (!ndev)
 799                 goto out_unlock;
 800
 801         tmp = iboe_get_mtu(ndev->mtu);
 802         props->active_mtu = tmp ? min(props->max_mtu, tmp) : IB_MTU_256;
 803
 804         props->state            = (netif_running(ndev) && netif_carrier_ok(ndev)) ?
 805                                         IB_PORT_ACTIVE : IB_PORT_DOWN;
 806         props->phys_state       = state_to_phys_state(props->state);
 807 out_unlock:
 808         spin_unlock_bh(&iboe->lock);
 809 out:
 810         mlx4_free_cmd_mailbox(mdev->dev, mailbox);
 811         return err;
 812 }
 813
 814 int __mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
 815                          struct ib_port_attr *props, int netw_view)
 816 {
 817         int err;
 818
 819         /* props being zeroed by the caller, avoid zeroing it here */
 820
 821         err = mlx4_ib_port_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND ?
 822                 ib_link_query_port(ibdev, port, props, netw_view) :
 823                                 eth_link_query_port(ibdev, port, props);
 824
 825         return err;
 826 }
 827
 828 static int mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
 829                               struct ib_port_attr *props)
 830 {
 831         /* returns host view */
 832         return __mlx4_ib_query_port(ibdev, port, props, 0);
 833 }
 834
 835 int __mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
 836                         union ib_gid *gid, int netw_view)
 837 {
 838         struct ib_smp *in_mad  = NULL;
 839         struct ib_smp *out_mad = NULL;
 840         int err = -ENOMEM;
 841         struct mlx4_ib_dev *dev = to_mdev(ibdev);
 842         int clear = 0;
 843         int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
 844
 845         in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
 846         out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
 847         if (!in_mad || !out_mad)
 848                 goto out;
 849
 850         init_query_mad(in_mad);
 851         in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
 852         in_mad->attr_mod = cpu_to_be32(port);
 853
 854         if (mlx4_is_mfunc(dev->dev) && netw_view)
 855                 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
 856
 857         err = mlx4_MAD_IFC(dev, mad_ifc_flags, port, NULL, NULL, in_mad, out_mad);
 858         if (err)
 859                 goto out;
 860
 861         memcpy(gid->raw, out_mad->data + 8, 8);
 862
 863         if (mlx4_is_mfunc(dev->dev) && !netw_view) {
 864                 if (index) {
 865                         /* For any index > 0, return the null guid */
 866                         err = 0;
 867                         clear = 1;
 868                         goto out;
 869                 }
 870         }
 871
 872         init_query_mad(in_mad);
 873         in_mad->attr_id  = IB_SMP_ATTR_GUID_INFO;
 874         in_mad->attr_mod = cpu_to_be32(index / 8);
 875
 876         err = mlx4_MAD_IFC(dev, mad_ifc_flags, port,
 877                            NULL, NULL, in_mad, out_mad);
 878         if (err)
 879                 goto out;
 880
 881         memcpy(gid->raw + 8, out_mad->data + (index % 8) * 8, 8);
 882
 883 out:
 884         if (clear)
 885                 memset(gid->raw + 8, 0, 8);
 886         kfree(in_mad);
 887         kfree(out_mad);
 888         return err;
 889 }
 890
 891 static int mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
 892                              union ib_gid *gid)
 893 {
 894         if (rdma_protocol_ib(ibdev, port))
 895                 return __mlx4_ib_query_gid(ibdev, port, index, gid, 0);
 896         return 0;
 897 }
 898
 899 static int mlx4_ib_query_sl2vl(struct ib_device *ibdev, u8 port, u64 *sl2vl_tbl)
 900 {
 901         union sl2vl_tbl_to_u64 sl2vl64;
 902         struct ib_smp *in_mad  = NULL;
 903         struct ib_smp *out_mad = NULL;
 904         int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
 905         int err = -ENOMEM;
 906         int jj;
 907
 908         if (mlx4_is_slave(to_mdev(ibdev)->dev)) {
 909                 *sl2vl_tbl = 0;
 910                 return 0;
 911         }
 912
 913         in_mad  = kzalloc(sizeof(*in_mad), GFP_KERNEL);
 914         out_mad = kmalloc(sizeof(*out_mad), GFP_KERNEL);
 915         if (!in_mad || !out_mad)
 916                 goto out;
 917
 918         init_query_mad(in_mad);
 919         in_mad->attr_id  = IB_SMP_ATTR_SL_TO_VL_TABLE;
 920         in_mad->attr_mod = 0;
 921
 922         if (mlx4_is_mfunc(to_mdev(ibdev)->dev))
 923                 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
 924
 925         err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
 926                            in_mad, out_mad);
 927         if (err)
 928                 goto out;
 929
 930         for (jj = 0; jj < 8; jj++)
 931                 sl2vl64.sl8[jj] = ((struct ib_smp *)out_mad)->data[jj];
 932         *sl2vl_tbl = sl2vl64.sl64;
 933
 934 out:
 935         kfree(in_mad);
 936         kfree(out_mad);
 937         return err;
 938 }
 939
 940 static void mlx4_init_sl2vl_tbl(struct mlx4_ib_dev *mdev)
 941 {
 942         u64 sl2vl;
 943         int i;
 944         int err;
 945
 946         for (i = 1; i <= mdev->dev->caps.num_ports; i++) {
 947                 if (mdev->dev->caps.port_type[i] == MLX4_PORT_TYPE_ETH)
 948                         continue;
 949                 err = mlx4_ib_query_sl2vl(&mdev->ib_dev, i, &sl2vl);
 950                 if (err) {
 951                         pr_err("Unable to get default sl to vl mapping for port %d.  Using all zeroes (%d)\n",
 952                                i, err);
 953                         sl2vl = 0;
 954                 }
 955                 atomic64_set(&mdev->sl2vl[i - 1], sl2vl);
 956         }
 957 }
 958
 959 int __mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index,
 960                          u16 *pkey, int netw_view)
 961 {
 962         struct ib_smp *in_mad  = NULL;
 963         struct ib_smp *out_mad = NULL;
 964         int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
 965         int err = -ENOMEM;
 966
 967         in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
 968         out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
 969         if (!in_mad || !out_mad)
 970                 goto out;
 971
 972         init_query_mad(in_mad);
 973         in_mad->attr_id  = IB_SMP_ATTR_PKEY_TABLE;
 974         in_mad->attr_mod = cpu_to_be32(index / 32);
 975
 976         if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
 977                 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
 978
 979         err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
 980                            in_mad, out_mad);
 981         if (err)
 982                 goto out;
 983
 984         *pkey = be16_to_cpu(((__be16 *) out_mad->data)[index % 32]);
 985
 986 out:
 987         kfree(in_mad);
 988         kfree(out_mad);
 989         return err;
 990 }
 991
 992 static int mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index, u16 *pkey)
 993 {
 994         return __mlx4_ib_query_pkey(ibdev, port, index, pkey, 0);
 995 }
 996
 997 static int mlx4_ib_modify_device(struct ib_device *ibdev, int mask,
 998                                  struct ib_device_modify *props)
 999 {
1000         struct mlx4_cmd_mailbox *mailbox;
1001         unsigned long flags;
1002
1003         if (mask & ~IB_DEVICE_MODIFY_NODE_DESC)
1004                 return -EOPNOTSUPP;
1005
1006         if (!(mask & IB_DEVICE_MODIFY_NODE_DESC))
1007                 return 0;
1008
1009         if (mlx4_is_slave(to_mdev(ibdev)->dev))
1010                 return -EOPNOTSUPP;
1011
1012         spin_lock_irqsave(&to_mdev(ibdev)->sm_lock, flags);
1013         memcpy(ibdev->node_desc, props->node_desc, IB_DEVICE_NODE_DESC_MAX);
1014         spin_unlock_irqrestore(&to_mdev(ibdev)->sm_lock, flags);
1015
1016         /*
1017          * If possible, pass node desc to FW, so it can generate
1018          * a 144 trap.  If cmd fails, just ignore.
1019          */
1020         mailbox = mlx4_alloc_cmd_mailbox(to_mdev(ibdev)->dev);
1021         if (IS_ERR(mailbox))
1022                 return 0;
1023
1024         memcpy(mailbox->buf, props->node_desc, IB_DEVICE_NODE_DESC_MAX);
1025         mlx4_cmd(to_mdev(ibdev)->dev, mailbox->dma, 1, 0,
1026                  MLX4_CMD_SET_NODE, MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
1027
1028         mlx4_free_cmd_mailbox(to_mdev(ibdev)->dev, mailbox);
1029
1030         return 0;
1031 }
1032
1033 static int mlx4_ib_SET_PORT(struct mlx4_ib_dev *dev, u8 port, int reset_qkey_viols,
1034                             u32 cap_mask)
1035 {
1036         struct mlx4_cmd_mailbox *mailbox;
1037         int err;
1038
1039         mailbox = mlx4_alloc_cmd_mailbox(dev->dev);
1040         if (IS_ERR(mailbox))
1041                 return PTR_ERR(mailbox);
1042
1043         if (dev->dev->flags & MLX4_FLAG_OLD_PORT_CMDS) {
1044                 *(u8 *) mailbox->buf         = !!reset_qkey_viols << 6;
1045                 ((__be32 *) mailbox->buf)[2] = cpu_to_be32(cap_mask);
1046         } else {
1047                 ((u8 *) mailbox->buf)[3]     = !!reset_qkey_viols;
1048                 ((__be32 *) mailbox->buf)[1] = cpu_to_be32(cap_mask);
1049         }
1050
1051         err = mlx4_cmd(dev->dev, mailbox->dma, port, MLX4_SET_PORT_IB_OPCODE,
1052                        MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
1053                        MLX4_CMD_WRAPPED);
1054
1055         mlx4_free_cmd_mailbox(dev->dev, mailbox);
1056         return err;
1057 }
1058
1059 static int mlx4_ib_modify_port(struct ib_device *ibdev, u8 port, int mask,
1060                                struct ib_port_modify *props)
1061 {
1062         struct mlx4_ib_dev *mdev = to_mdev(ibdev);
1063         u8 is_eth = mdev->dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH;
1064         struct ib_port_attr attr;
1065         u32 cap_mask;
1066         int err;
1067
1068         /* return OK if this is RoCE. CM calls ib_modify_port() regardless
1069          * of whether port link layer is ETH or IB. For ETH ports, qkey
1070          * violations and port capabilities are not meaningful.
1071          */
1072         if (is_eth)
1073                 return 0;
1074
1075         mutex_lock(&mdev->cap_mask_mutex);
1076
1077         err = ib_query_port(ibdev, port, &attr);
1078         if (err)
1079                 goto out;
1080
1081         cap_mask = (attr.port_cap_flags | props->set_port_cap_mask) &
1082                 ~props->clr_port_cap_mask;
1083
1084         err = mlx4_ib_SET_PORT(mdev, port,
1085                                !!(mask & IB_PORT_RESET_QKEY_CNTR),
1086                                cap_mask);
1087
1088 out:
1089         mutex_unlock(&to_mdev(ibdev)->cap_mask_mutex);
1090         return err;
1091 }
1092
1093 static int mlx4_ib_alloc_ucontext(struct ib_ucontext *uctx,
1094                                   struct ib_udata *udata)
1095 {
1096         struct ib_device *ibdev = uctx->device;
1097         struct mlx4_ib_dev *dev = to_mdev(ibdev);
1098         struct mlx4_ib_ucontext *context = to_mucontext(uctx);
1099         struct mlx4_ib_alloc_ucontext_resp_v3 resp_v3;
1100         struct mlx4_ib_alloc_ucontext_resp resp;
1101         int err;
1102
1103         if (!dev->ib_active)
1104                 return -EAGAIN;
1105
1106         if (ibdev->ops.uverbs_abi_ver ==
1107             MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION) {
1108                 resp_v3.qp_tab_size      = dev->dev->caps.num_qps;
1109                 resp_v3.bf_reg_size      = dev->dev->caps.bf_reg_size;
1110                 resp_v3.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
1111         } else {
1112                 resp.dev_caps         = dev->dev->caps.userspace_caps;
1113                 resp.qp_tab_size      = dev->dev->caps.num_qps;
1114                 resp.bf_reg_size      = dev->dev->caps.bf_reg_size;
1115                 resp.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
1116                 resp.cqe_size         = dev->dev->caps.cqe_size;
1117         }
1118
1119         err = mlx4_uar_alloc(to_mdev(ibdev)->dev, &context->uar);
1120         if (err)
1121                 return err;
1122
1123         INIT_LIST_HEAD(&context->db_page_list);
1124         mutex_init(&context->db_page_mutex);
1125
1126         INIT_LIST_HEAD(&context->wqn_ranges_list);
1127         mutex_init(&context->wqn_ranges_mutex);
1128
1129         if (ibdev->ops.uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION)
1130                 err = ib_copy_to_udata(udata, &resp_v3, sizeof(resp_v3));
1131         else
1132                 err = ib_copy_to_udata(udata, &resp, sizeof(resp));
1133
1134         if (err) {
1135                 mlx4_uar_free(to_mdev(ibdev)->dev, &context->uar);
1136                 return -EFAULT;
1137         }
1138
1139         return err;
1140 }
1141
1142 static void mlx4_ib_dealloc_ucontext(struct ib_ucontext *ibcontext)
1143 {
1144         struct mlx4_ib_ucontext *context = to_mucontext(ibcontext);
1145
1146         mlx4_uar_free(to_mdev(ibcontext->device)->dev, &context->uar);
1147 }
1148
1149 static void mlx4_ib_disassociate_ucontext(struct ib_ucontext *ibcontext)
1150 {
1151 }
1152
1153 static int mlx4_ib_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
1154 {
1155         struct mlx4_ib_dev *dev = to_mdev(context->device);
1156
1157         switch (vma->vm_pgoff) {
1158         case 0:
1159                 return rdma_user_mmap_io(context, vma,
1160                                          to_mucontext(context)->uar.pfn,
1161                                          PAGE_SIZE,
1162                                          pgprot_noncached(vma->vm_page_prot),
1163                                          NULL);
1164
1165         case 1:
1166                 if (dev->dev->caps.bf_reg_size == 0)
1167                         return -EINVAL;
1168                 return rdma_user_mmap_io(
1169                         context, vma,
1170                         to_mucontext(context)->uar.pfn +
1171                                 dev->dev->caps.num_uars,
1172                         PAGE_SIZE, pgprot_writecombine(vma->vm_page_prot),
1173                         NULL);
1174
1175         case 3: {
1176                 struct mlx4_clock_params params;
1177                 int ret;
1178
1179                 ret = mlx4_get_internal_clock_params(dev->dev, &params);
1180                 if (ret)
1181                         return ret;
1182
1183                 return rdma_user_mmap_io(
1184                         context, vma,
1185                         (pci_resource_start(dev->dev->persist->pdev,
1186                                             params.bar) +
1187                          params.offset) >>
1188                                 PAGE_SHIFT,
1189                         PAGE_SIZE, pgprot_noncached(vma->vm_page_prot),
1190                         NULL);
1191         }
1192
1193         default:
1194                 return -EINVAL;
1195         }
1196 }
1197
1198 static int mlx4_ib_alloc_pd(struct ib_pd *ibpd, struct ib_udata *udata)
1199 {
1200         struct mlx4_ib_pd *pd = to_mpd(ibpd);
1201         struct ib_device *ibdev = ibpd->device;
1202         int err;
1203
1204         err = mlx4_pd_alloc(to_mdev(ibdev)->dev, &pd->pdn);
1205         if (err)
1206                 return err;
1207
1208         if (udata && ib_copy_to_udata(udata, &pd->pdn, sizeof(__u32))) {
1209                 mlx4_pd_free(to_mdev(ibdev)->dev, pd->pdn);
1210                 return -EFAULT;
1211         }
1212         return 0;
1213 }
1214
1215 static int mlx4_ib_dealloc_pd(struct ib_pd *pd, struct ib_udata *udata)
1216 {
1217         mlx4_pd_free(to_mdev(pd->device)->dev, to_mpd(pd)->pdn);
1218         return 0;
1219 }
1220
1221 static int mlx4_ib_alloc_xrcd(struct ib_xrcd *ibxrcd, struct ib_udata *udata)
1222 {
1223         struct mlx4_ib_dev *dev = to_mdev(ibxrcd->device);
1224         struct mlx4_ib_xrcd *xrcd = to_mxrcd(ibxrcd);
1225         struct ib_cq_init_attr cq_attr = {};
1226         int err;
1227
1228         if (!(dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC))
1229                 return -EOPNOTSUPP;
1230
1231         err = mlx4_xrcd_alloc(dev->dev, &xrcd->xrcdn);
1232         if (err)
1233                 return err;
1234
1235         xrcd->pd = ib_alloc_pd(ibxrcd->device, 0);
1236         if (IS_ERR(xrcd->pd)) {
1237                 err = PTR_ERR(xrcd->pd);
1238                 goto err2;
1239         }
1240
1241         cq_attr.cqe = 1;
1242         xrcd->cq = ib_create_cq(ibxrcd->device, NULL, NULL, xrcd, &cq_attr);
1243         if (IS_ERR(xrcd->cq)) {
1244                 err = PTR_ERR(xrcd->cq);
1245                 goto err3;
1246         }
1247
1248         return 0;
1249
1250 err3:
1251         ib_dealloc_pd(xrcd->pd);
1252 err2:
1253         mlx4_xrcd_free(dev->dev, xrcd->xrcdn);
1254         return err;
1255 }
1256
1257 static int mlx4_ib_dealloc_xrcd(struct ib_xrcd *xrcd, struct ib_udata *udata)
1258 {
1259         ib_destroy_cq(to_mxrcd(xrcd)->cq);
1260         ib_dealloc_pd(to_mxrcd(xrcd)->pd);
1261         mlx4_xrcd_free(to_mdev(xrcd->device)->dev, to_mxrcd(xrcd)->xrcdn);
1262         return 0;
1263 }
1264
1265 static int add_gid_entry(struct ib_qp *ibqp, union ib_gid *gid)
1266 {
1267         struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1268         struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1269         struct mlx4_ib_gid_entry *ge;
1270
1271         ge = kzalloc(sizeof *ge, GFP_KERNEL);
1272         if (!ge)
1273                 return -ENOMEM;
1274
1275         ge->gid = *gid;
1276         if (mlx4_ib_add_mc(mdev, mqp, gid)) {
1277                 ge->port = mqp->port;
1278                 ge->added = 1;
1279         }
1280
1281         mutex_lock(&mqp->mutex);
1282         list_add_tail(&ge->list, &mqp->gid_list);
1283         mutex_unlock(&mqp->mutex);
1284
1285         return 0;
1286 }
1287
1288 static void mlx4_ib_delete_counters_table(struct mlx4_ib_dev *ibdev,
1289                                           struct mlx4_ib_counters *ctr_table)
1290 {
1291         struct counter_index *counter, *tmp_count;
1292
1293         mutex_lock(&ctr_table->mutex);
1294         list_for_each_entry_safe(counter, tmp_count, &ctr_table->counters_list,
1295                                  list) {
1296                 if (counter->allocated)
1297                         mlx4_counter_free(ibdev->dev, counter->index);
1298                 list_del(&counter->list);
1299                 kfree(counter);
1300         }
1301         mutex_unlock(&ctr_table->mutex);
1302 }
1303
1304 int mlx4_ib_add_mc(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
1305                    union ib_gid *gid)
1306 {
1307         struct net_device *ndev;
1308         int ret = 0;
1309
1310         if (!mqp->port)
1311                 return 0;
1312
1313         spin_lock_bh(&mdev->iboe.lock);
1314         ndev = mdev->iboe.netdevs[mqp->port - 1];
1315         if (ndev)
1316                 dev_hold(ndev);
1317         spin_unlock_bh(&mdev->iboe.lock);
1318
1319         if (ndev) {
1320                 ret = 1;
1321                 dev_put(ndev);
1322         }
1323
1324         return ret;
1325 }
1326
1327 struct mlx4_ib_steering {
1328         struct list_head list;
1329         struct mlx4_flow_reg_id reg_id;
1330         union ib_gid gid;
1331 };
1332
1333 #define LAST_ETH_FIELD vlan_tag
1334 #define LAST_IB_FIELD sl
1335 #define LAST_IPV4_FIELD dst_ip
1336 #define LAST_TCP_UDP_FIELD src_port
1337
1338 /* Field is the last supported field */
1339 #define FIELDS_NOT_SUPPORTED(filter, field)\
1340         memchr_inv((void *)&filter.field  +\
1341                    sizeof(filter.field), 0,\
1342                    sizeof(filter) -\
1343                    offsetof(typeof(filter), field) -\
1344                    sizeof(filter.field))
1345
1346 static int parse_flow_attr(struct mlx4_dev *dev,
1347                            u32 qp_num,
1348                            union ib_flow_spec *ib_spec,
1349                            struct _rule_hw *mlx4_spec)
1350 {
1351         enum mlx4_net_trans_rule_id type;
1352
1353         switch (ib_spec->type) {
1354         case IB_FLOW_SPEC_ETH:
1355                 if (FIELDS_NOT_SUPPORTED(ib_spec->eth.mask, LAST_ETH_FIELD))
1356                         return -ENOTSUPP;
1357
1358                 type = MLX4_NET_TRANS_RULE_ID_ETH;
1359                 memcpy(mlx4_spec->eth.dst_mac, ib_spec->eth.val.dst_mac,
1360                        ETH_ALEN);
1361                 memcpy(mlx4_spec->eth.dst_mac_msk, ib_spec->eth.mask.dst_mac,
1362                        ETH_ALEN);
1363                 mlx4_spec->eth.vlan_tag = ib_spec->eth.val.vlan_tag;
1364                 mlx4_spec->eth.vlan_tag_msk = ib_spec->eth.mask.vlan_tag;
1365                 break;
1366         case IB_FLOW_SPEC_IB:
1367                 if (FIELDS_NOT_SUPPORTED(ib_spec->ib.mask, LAST_IB_FIELD))
1368                         return -ENOTSUPP;
1369
1370                 type = MLX4_NET_TRANS_RULE_ID_IB;
1371                 mlx4_spec->ib.l3_qpn =
1372                         cpu_to_be32(qp_num);
1373                 mlx4_spec->ib.qpn_mask =
1374                         cpu_to_be32(MLX4_IB_FLOW_QPN_MASK);
1375                 break;
1376
1377
1378         case IB_FLOW_SPEC_IPV4:
1379                 if (FIELDS_NOT_SUPPORTED(ib_spec->ipv4.mask, LAST_IPV4_FIELD))
1380                         return -ENOTSUPP;
1381
1382                 type = MLX4_NET_TRANS_RULE_ID_IPV4;
1383                 mlx4_spec->ipv4.src_ip = ib_spec->ipv4.val.src_ip;
1384                 mlx4_spec->ipv4.src_ip_msk = ib_spec->ipv4.mask.src_ip;
1385                 mlx4_spec->ipv4.dst_ip = ib_spec->ipv4.val.dst_ip;
1386                 mlx4_spec->ipv4.dst_ip_msk = ib_spec->ipv4.mask.dst_ip;
1387                 break;
1388
1389         case IB_FLOW_SPEC_TCP:
1390         case IB_FLOW_SPEC_UDP:
1391                 if (FIELDS_NOT_SUPPORTED(ib_spec->tcp_udp.mask, LAST_TCP_UDP_FIELD))
1392                         return -ENOTSUPP;
1393
1394                 type = ib_spec->type == IB_FLOW_SPEC_TCP ?
1395                                         MLX4_NET_TRANS_RULE_ID_TCP :
1396                                         MLX4_NET_TRANS_RULE_ID_UDP;
1397                 mlx4_spec->tcp_udp.dst_port = ib_spec->tcp_udp.val.dst_port;
1398                 mlx4_spec->tcp_udp.dst_port_msk = ib_spec->tcp_udp.mask.dst_port;
1399                 mlx4_spec->tcp_udp.src_port = ib_spec->tcp_udp.val.src_port;
1400                 mlx4_spec->tcp_udp.src_port_msk = ib_spec->tcp_udp.mask.src_port;
1401                 break;
1402
1403         default:
1404                 return -EINVAL;
1405         }
1406         if (mlx4_map_sw_to_hw_steering_id(dev, type) < 0 ||
1407             mlx4_hw_rule_sz(dev, type) < 0)
1408                 return -EINVAL;
1409         mlx4_spec->id = cpu_to_be16(mlx4_map_sw_to_hw_steering_id(dev, type));
1410         mlx4_spec->size = mlx4_hw_rule_sz(dev, type) >> 2;
1411         return mlx4_hw_rule_sz(dev, type);
1412 }
1413
1414 struct default_rules {
1415         __u32 mandatory_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
1416         __u32 mandatory_not_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
1417         __u32 rules_create_list[IB_FLOW_SPEC_SUPPORT_LAYERS];
1418         __u8  link_layer;
1419 };
1420 static const struct default_rules default_table[] = {
1421         {
1422                 .mandatory_fields = {IB_FLOW_SPEC_IPV4},
1423                 .mandatory_not_fields = {IB_FLOW_SPEC_ETH},
1424                 .rules_create_list = {IB_FLOW_SPEC_IB},
1425                 .link_layer = IB_LINK_LAYER_INFINIBAND
1426         }
1427 };
1428
1429 static int __mlx4_ib_default_rules_match(struct ib_qp *qp,
1430                                          struct ib_flow_attr *flow_attr)
1431 {
1432         int i, j, k;
1433         void *ib_flow;
1434         const struct default_rules *pdefault_rules = default_table;
1435         u8 link_layer = rdma_port_get_link_layer(qp->device, flow_attr->port);
1436
1437         for (i = 0; i < ARRAY_SIZE(default_table); i++, pdefault_rules++) {
1438                 __u32 field_types[IB_FLOW_SPEC_SUPPORT_LAYERS];
1439                 memset(&field_types, 0, sizeof(field_types));
1440
1441                 if (link_layer != pdefault_rules->link_layer)
1442                         continue;
1443
1444                 ib_flow = flow_attr + 1;
1445                 /* we assume the specs are sorted */
1446                 for (j = 0, k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS &&
1447                      j < flow_attr->num_of_specs; k++) {
1448                         union ib_flow_spec *current_flow =
1449                                 (union ib_flow_spec *)ib_flow;
1450
1451                         /* same layer but different type */
1452                         if (((current_flow->type & IB_FLOW_SPEC_LAYER_MASK) ==
1453                              (pdefault_rules->mandatory_fields[k] &
1454                               IB_FLOW_SPEC_LAYER_MASK)) &&
1455                             (current_flow->type !=
1456                              pdefault_rules->mandatory_fields[k]))
1457                                 goto out;
1458
1459                         /* same layer, try match next one */
1460                         if (current_flow->type ==
1461                             pdefault_rules->mandatory_fields[k]) {
1462                                 j++;
1463                                 ib_flow +=
1464                                         ((union ib_flow_spec *)ib_flow)->size;
1465                         }
1466                 }
1467
1468                 ib_flow = flow_attr + 1;
1469                 for (j = 0; j < flow_attr->num_of_specs;
1470                      j++, ib_flow += ((union ib_flow_spec *)ib_flow)->size)
1471                         for (k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS; k++)
1472                                 /* same layer and same type */
1473                                 if (((union ib_flow_spec *)ib_flow)->type ==
1474                                     pdefault_rules->mandatory_not_fields[k])
1475                                         goto out;
1476
1477                 return i;
1478         }
1479 out:
1480         return -1;
1481 }
1482
1483 static int __mlx4_ib_create_default_rules(
1484                 struct mlx4_ib_dev *mdev,
1485                 struct ib_qp *qp,
1486                 const struct default_rules *pdefault_rules,
1487                 struct _rule_hw *mlx4_spec) {
1488         int size = 0;
1489         int i;
1490
1491         for (i = 0; i < ARRAY_SIZE(pdefault_rules->rules_create_list); i++) {
1492                 union ib_flow_spec ib_spec = {};
1493                 int ret;
1494
1495                 switch (pdefault_rules->rules_create_list[i]) {
1496                 case 0:
1497                         /* no rule */
1498                         continue;
1499                 case IB_FLOW_SPEC_IB:
1500                         ib_spec.type = IB_FLOW_SPEC_IB;
1501                         ib_spec.size = sizeof(struct ib_flow_spec_ib);
1502
1503                         break;
1504                 default:
1505                         /* invalid rule */
1506                         return -EINVAL;
1507                 }
1508                 /* We must put empty rule, qpn is being ignored */
1509                 ret = parse_flow_attr(mdev->dev, 0, &ib_spec,
1510                                       mlx4_spec);
1511                 if (ret < 0) {
1512                         pr_info("invalid parsing\n");
1513                         return -EINVAL;
1514                 }
1515
1516                 mlx4_spec = (void *)mlx4_spec + ret;
1517                 size += ret;
1518         }
1519         return size;
1520 }
1521
1522 static int __mlx4_ib_create_flow(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
1523                           int domain,
1524                           enum mlx4_net_trans_promisc_mode flow_type,
1525                           u64 *reg_id)
1526 {
1527         int ret, i;
1528         int size = 0;
1529         void *ib_flow;
1530         struct mlx4_ib_dev *mdev = to_mdev(qp->device);
1531         struct mlx4_cmd_mailbox *mailbox;
1532         struct mlx4_net_trans_rule_hw_ctrl *ctrl;
1533         int default_flow;
1534
1535         if (flow_attr->priority > MLX4_IB_FLOW_MAX_PRIO) {
1536                 pr_err("Invalid priority value %d\n", flow_attr->priority);
1537                 return -EINVAL;
1538         }
1539
1540         if (mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type) < 0)
1541                 return -EINVAL;
1542
1543         mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
1544         if (IS_ERR(mailbox))
1545                 return PTR_ERR(mailbox);
1546         ctrl = mailbox->buf;
1547
1548         ctrl->prio = cpu_to_be16(domain | flow_attr->priority);
1549         ctrl->type = mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type);
1550         ctrl->port = flow_attr->port;
1551         ctrl->qpn = cpu_to_be32(qp->qp_num);
1552
1553         ib_flow = flow_attr + 1;
1554         size += sizeof(struct mlx4_net_trans_rule_hw_ctrl);
1555         /* Add default flows */
1556         default_flow = __mlx4_ib_default_rules_match(qp, flow_attr);
1557         if (default_flow >= 0) {
1558                 ret = __mlx4_ib_create_default_rules(
1559                                 mdev, qp, default_table + default_flow,
1560                                 mailbox->buf + size);
1561                 if (ret < 0) {
1562                         mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1563                         return -EINVAL;
1564                 }
1565                 size += ret;
1566         }
1567         for (i = 0; i < flow_attr->num_of_specs; i++) {
1568                 ret = parse_flow_attr(mdev->dev, qp->qp_num, ib_flow,
1569                                       mailbox->buf + size);
1570                 if (ret < 0) {
1571                         mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1572                         return -EINVAL;
1573                 }
1574                 ib_flow += ((union ib_flow_spec *) ib_flow)->size;
1575                 size += ret;
1576         }
1577
1578         if (mlx4_is_master(mdev->dev) && flow_type == MLX4_FS_REGULAR &&
1579             flow_attr->num_of_specs == 1) {
1580                 struct _rule_hw *rule_header = (struct _rule_hw *)(ctrl + 1);
1581                 enum ib_flow_spec_type header_spec =
1582                         ((union ib_flow_spec *)(flow_attr + 1))->type;
1583
1584                 if (header_spec == IB_FLOW_SPEC_ETH)
1585                         mlx4_handle_eth_header_mcast_prio(ctrl, rule_header);
1586         }
1587
1588         ret = mlx4_cmd_imm(mdev->dev, mailbox->dma, reg_id, size >> 2, 0,
1589                            MLX4_QP_FLOW_STEERING_ATTACH, MLX4_CMD_TIME_CLASS_A,
1590                            MLX4_CMD_NATIVE);
1591         if (ret == -ENOMEM)
1592                 pr_err("mcg table is full. Fail to register network rule.\n");
1593         else if (ret == -ENXIO)
1594                 pr_err("Device managed flow steering is disabled. Fail to register network rule.\n");
1595         else if (ret)
1596                 pr_err("Invalid argument. Fail to register network rule.\n");
1597
1598         mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1599         return ret;
1600 }
1601
1602 static int __mlx4_ib_destroy_flow(struct mlx4_dev *dev, u64 reg_id)
1603 {
1604         int err;
1605         err = mlx4_cmd(dev, reg_id, 0, 0,
1606                        MLX4_QP_FLOW_STEERING_DETACH, MLX4_CMD_TIME_CLASS_A,
1607                        MLX4_CMD_NATIVE);
1608         if (err)
1609                 pr_err("Fail to detach network rule. registration id = 0x%llx\n",
1610                        reg_id);
1611         return err;
1612 }
1613
1614 static int mlx4_ib_tunnel_steer_add(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
1615                                     u64 *reg_id)
1616 {
1617         void *ib_flow;
1618         union ib_flow_spec *ib_spec;
1619         struct mlx4_dev *dev = to_mdev(qp->device)->dev;
1620         int err = 0;
1621
1622         if (dev->caps.tunnel_offload_mode != MLX4_TUNNEL_OFFLOAD_MODE_VXLAN ||
1623             dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC)
1624                 return 0; /* do nothing */
1625
1626         ib_flow = flow_attr + 1;
1627         ib_spec = (union ib_flow_spec *)ib_flow;
1628
1629         if (ib_spec->type !=  IB_FLOW_SPEC_ETH || flow_attr->num_of_specs != 1)
1630                 return 0; /* do nothing */
1631
1632         err = mlx4_tunnel_steer_add(to_mdev(qp->device)->dev, ib_spec->eth.val.dst_mac,
1633                                     flow_attr->port, qp->qp_num,
1634                                     MLX4_DOMAIN_UVERBS | (flow_attr->priority & 0xff),
1635                                     reg_id);
1636         return err;
1637 }
1638
1639 static int mlx4_ib_add_dont_trap_rule(struct mlx4_dev *dev,
1640                                       struct ib_flow_attr *flow_attr,
1641                                       enum mlx4_net_trans_promisc_mode *type)
1642 {
1643         int err = 0;
1644
1645         if (!(dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DMFS_UC_MC_SNIFFER) ||
1646             (dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC) ||
1647             (flow_attr->num_of_specs > 1) || (flow_attr->priority != 0)) {
1648                 return -EOPNOTSUPP;
1649         }
1650
1651         if (flow_attr->num_of_specs == 0) {
1652                 type[0] = MLX4_FS_MC_SNIFFER;
1653                 type[1] = MLX4_FS_UC_SNIFFER;
1654         } else {
1655                 union ib_flow_spec *ib_spec;
1656
1657                 ib_spec = (union ib_flow_spec *)(flow_attr + 1);
1658                 if (ib_spec->type !=  IB_FLOW_SPEC_ETH)
1659                         return -EINVAL;
1660
1661                 /* if all is zero than MC and UC */
1662                 if (is_zero_ether_addr(ib_spec->eth.mask.dst_mac)) {
1663                         type[0] = MLX4_FS_MC_SNIFFER;
1664                         type[1] = MLX4_FS_UC_SNIFFER;
1665                 } else {
1666                         u8 mac[ETH_ALEN] = {ib_spec->eth.mask.dst_mac[0] ^ 0x01,
1667                                             ib_spec->eth.mask.dst_mac[1],
1668                                             ib_spec->eth.mask.dst_mac[2],
1669                                             ib_spec->eth.mask.dst_mac[3],
1670                                             ib_spec->eth.mask.dst_mac[4],
1671                                             ib_spec->eth.mask.dst_mac[5]};
1672
1673                         /* Above xor was only on MC bit, non empty mask is valid
1674                          * only if this bit is set and rest are zero.
1675                          */
1676                         if (!is_zero_ether_addr(&mac[0]))
1677                                 return -EINVAL;
1678
1679                         if (is_multicast_ether_addr(ib_spec->eth.val.dst_mac))
1680                                 type[0] = MLX4_FS_MC_SNIFFER;
1681                         else
1682                                 type[0] = MLX4_FS_UC_SNIFFER;
1683                 }
1684         }
1685
1686         return err;
1687 }
1688
1689 static struct ib_flow *mlx4_ib_create_flow(struct ib_qp *qp,
1690                                            struct ib_flow_attr *flow_attr,
1691                                            struct ib_udata *udata)
1692 {
1693         int err = 0, i = 0, j = 0;
1694         struct mlx4_ib_flow *mflow;
1695         enum mlx4_net_trans_promisc_mode type[2];
1696         struct mlx4_dev *dev = (to_mdev(qp->device))->dev;
1697         int is_bonded = mlx4_is_bonded(dev);
1698
1699         if (flow_attr->port < 1 || flow_attr->port > qp->device->phys_port_cnt)
1700                 return ERR_PTR(-EINVAL);
1701
1702         if (flow_attr->flags & ~IB_FLOW_ATTR_FLAGS_DONT_TRAP)
1703                 return ERR_PTR(-EOPNOTSUPP);
1704
1705         if ((flow_attr->flags & IB_FLOW_ATTR_FLAGS_DONT_TRAP) &&
1706             (flow_attr->type != IB_FLOW_ATTR_NORMAL))
1707                 return ERR_PTR(-EOPNOTSUPP);
1708
1709         if (udata &&
1710             udata->inlen && !ib_is_udata_cleared(udata, 0, udata->inlen))
1711                 return ERR_PTR(-EOPNOTSUPP);
1712
1713         memset(type, 0, sizeof(type));
1714
1715         mflow = kzalloc(sizeof(*mflow), GFP_KERNEL);
1716         if (!mflow) {
1717                 err = -ENOMEM;
1718                 goto err_free;
1719         }
1720
1721         switch (flow_attr->type) {
1722         case IB_FLOW_ATTR_NORMAL:
1723                 /* If dont trap flag (continue match) is set, under specific
1724                  * condition traffic be replicated to given qp,
1725                  * without stealing it
1726                  */
1727                 if (unlikely(flow_attr->flags & IB_FLOW_ATTR_FLAGS_DONT_TRAP)) {
1728                         err = mlx4_ib_add_dont_trap_rule(dev,
1729                                                          flow_attr,
1730                                                          type);
1731                         if (err)
1732                                 goto err_free;
1733                 } else {
1734                         type[0] = MLX4_FS_REGULAR;
1735                 }
1736                 break;
1737
1738         case IB_FLOW_ATTR_ALL_DEFAULT:
1739                 type[0] = MLX4_FS_ALL_DEFAULT;
1740                 break;
1741
1742         case IB_FLOW_ATTR_MC_DEFAULT:
1743                 type[0] = MLX4_FS_MC_DEFAULT;
1744                 break;
1745
1746         case IB_FLOW_ATTR_SNIFFER:
1747                 type[0] = MLX4_FS_MIRROR_RX_PORT;
1748                 type[1] = MLX4_FS_MIRROR_SX_PORT;
1749                 break;
1750
1751         default:
1752                 err = -EINVAL;
1753                 goto err_free;
1754         }
1755
1756         while (i < ARRAY_SIZE(type) && type[i]) {
1757                 err = __mlx4_ib_create_flow(qp, flow_attr, MLX4_DOMAIN_UVERBS,
1758                                             type[i], &mflow->reg_id[i].id);
1759                 if (err)
1760                         goto err_create_flow;
1761                 if (is_bonded) {
1762                         /* Application always sees one port so the mirror rule
1763                          * must be on port #2
1764                          */
1765                         flow_attr->port = 2;
1766                         err = __mlx4_ib_create_flow(qp, flow_attr,
1767                                                     MLX4_DOMAIN_UVERBS, type[j],
1768                                                     &mflow->reg_id[j].mirror);
1769                         flow_attr->port = 1;
1770                         if (err)
1771                                 goto err_create_flow;
1772                         j++;
1773                 }
1774
1775                 i++;
1776         }
1777
1778         if (i < ARRAY_SIZE(type) && flow_attr->type == IB_FLOW_ATTR_NORMAL) {
1779                 err = mlx4_ib_tunnel_steer_add(qp, flow_attr,
1780                                                &mflow->reg_id[i].id);
1781                 if (err)
1782                         goto err_create_flow;
1783
1784                 if (is_bonded) {
1785                         flow_attr->port = 2;
1786                         err = mlx4_ib_tunnel_steer_add(qp, flow_attr,
1787                                                        &mflow->reg_id[j].mirror);
1788                         flow_attr->port = 1;
1789                         if (err)
1790                                 goto err_create_flow;
1791                         j++;
1792                 }
1793                 /* function to create mirror rule */
1794                 i++;
1795         }
1796
1797         return &mflow->ibflow;
1798
1799 err_create_flow:
1800         while (i) {
1801                 (void)__mlx4_ib_destroy_flow(to_mdev(qp->device)->dev,
1802                                              mflow->reg_id[i].id);
1803                 i--;
1804         }
1805
1806         while (j) {
1807                 (void)__mlx4_ib_destroy_flow(to_mdev(qp->device)->dev,
1808                                              mflow->reg_id[j].mirror);
1809                 j--;
1810         }
1811 err_free:
1812         kfree(mflow);
1813         return ERR_PTR(err);
1814 }
1815
1816 static int mlx4_ib_destroy_flow(struct ib_flow *flow_id)
1817 {
1818         int err, ret = 0;
1819         int i = 0;
1820         struct mlx4_ib_dev *mdev = to_mdev(flow_id->qp->device);
1821         struct mlx4_ib_flow *mflow = to_mflow(flow_id);
1822
1823         while (i < ARRAY_SIZE(mflow->reg_id) && mflow->reg_id[i].id) {
1824                 err = __mlx4_ib_destroy_flow(mdev->dev, mflow->reg_id[i].id);
1825                 if (err)
1826                         ret = err;
1827                 if (mflow->reg_id[i].mirror) {
1828                         err = __mlx4_ib_destroy_flow(mdev->dev,
1829                                                      mflow->reg_id[i].mirror);
1830                         if (err)
1831                                 ret = err;
1832                 }
1833                 i++;
1834         }
1835
1836         kfree(mflow);
1837         return ret;
1838 }
1839
1840 static int mlx4_ib_mcg_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
1841 {
1842         int err;
1843         struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1844         struct mlx4_dev *dev = mdev->dev;
1845         struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1846         struct mlx4_ib_steering *ib_steering = NULL;
1847         enum mlx4_protocol prot = MLX4_PROT_IB_IPV6;
1848         struct mlx4_flow_reg_id reg_id;
1849
1850         if (mdev->dev->caps.steering_mode ==
1851             MLX4_STEERING_MODE_DEVICE_MANAGED) {
1852                 ib_steering = kmalloc(sizeof(*ib_steering), GFP_KERNEL);
1853                 if (!ib_steering)
1854                         return -ENOMEM;
1855         }
1856
1857         err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw, mqp->port,
1858                                     !!(mqp->flags &
1859                                        MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
1860                                     prot, &reg_id.id);
1861         if (err) {
1862                 pr_err("multicast attach op failed, err %d\n", err);
1863                 goto err_malloc;
1864         }
1865
1866         reg_id.mirror = 0;
1867         if (mlx4_is_bonded(dev)) {
1868                 err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw,
1869                                             (mqp->port == 1) ? 2 : 1,
1870                                             !!(mqp->flags &
1871                                             MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
1872                                             prot, &reg_id.mirror);
1873                 if (err)
1874                         goto err_add;
1875         }
1876
1877         err = add_gid_entry(ibqp, gid);
1878         if (err)
1879                 goto err_add;
1880
1881         if (ib_steering) {
1882                 memcpy(ib_steering->gid.raw, gid->raw, 16);
1883                 ib_steering->reg_id = reg_id;
1884                 mutex_lock(&mqp->mutex);
1885                 list_add(&ib_steering->list, &mqp->steering_rules);
1886                 mutex_unlock(&mqp->mutex);
1887         }
1888         return 0;
1889
1890 err_add:
1891         mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1892                               prot, reg_id.id);
1893         if (reg_id.mirror)
1894                 mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1895                                       prot, reg_id.mirror);
1896 err_malloc:
1897         kfree(ib_steering);
1898
1899         return err;
1900 }
1901
1902 static struct mlx4_ib_gid_entry *find_gid_entry(struct mlx4_ib_qp *qp, u8 *raw)
1903 {
1904         struct mlx4_ib_gid_entry *ge;
1905         struct mlx4_ib_gid_entry *tmp;
1906         struct mlx4_ib_gid_entry *ret = NULL;
1907
1908         list_for_each_entry_safe(ge, tmp, &qp->gid_list, list) {
1909                 if (!memcmp(raw, ge->gid.raw, 16)) {
1910                         ret = ge;
1911                         break;
1912                 }
1913         }
1914
1915         return ret;
1916 }
1917
1918 static int mlx4_ib_mcg_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
1919 {
1920         int err;
1921         struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1922         struct mlx4_dev *dev = mdev->dev;
1923         struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1924         struct net_device *ndev;
1925         struct mlx4_ib_gid_entry *ge;
1926         struct mlx4_flow_reg_id reg_id = {0, 0};
1927         enum mlx4_protocol prot =  MLX4_PROT_IB_IPV6;
1928
1929         if (mdev->dev->caps.steering_mode ==
1930             MLX4_STEERING_MODE_DEVICE_MANAGED) {
1931                 struct mlx4_ib_steering *ib_steering;
1932
1933                 mutex_lock(&mqp->mutex);
1934                 list_for_each_entry(ib_steering, &mqp->steering_rules, list) {
1935                         if (!memcmp(ib_steering->gid.raw, gid->raw, 16)) {
1936                                 list_del(&ib_steering->list);
1937                                 break;
1938                         }
1939                 }
1940                 mutex_unlock(&mqp->mutex);
1941                 if (&ib_steering->list == &mqp->steering_rules) {
1942                         pr_err("Couldn't find reg_id for mgid. Steering rule is left attached\n");
1943                         return -EINVAL;
1944                 }
1945                 reg_id = ib_steering->reg_id;
1946                 kfree(ib_steering);
1947         }
1948
1949         err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1950                                     prot, reg_id.id);
1951         if (err)
1952                 return err;
1953
1954         if (mlx4_is_bonded(dev)) {
1955                 err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1956                                             prot, reg_id.mirror);
1957                 if (err)
1958                         return err;
1959         }
1960
1961         mutex_lock(&mqp->mutex);
1962         ge = find_gid_entry(mqp, gid->raw);
1963         if (ge) {
1964                 spin_lock_bh(&mdev->iboe.lock);
1965                 ndev = ge->added ? mdev->iboe.netdevs[ge->port - 1] : NULL;
1966                 if (ndev)
1967                         dev_hold(ndev);
1968                 spin_unlock_bh(&mdev->iboe.lock);
1969                 if (ndev)
1970                         dev_put(ndev);
1971                 list_del(&ge->list);
1972                 kfree(ge);
1973         } else
1974                 pr_warn("could not find mgid entry\n");
1975
1976         mutex_unlock(&mqp->mutex);
1977
1978         return 0;
1979 }
1980
1981 static int init_node_data(struct mlx4_ib_dev *dev)
1982 {
1983         struct ib_smp *in_mad  = NULL;
1984         struct ib_smp *out_mad = NULL;
1985         int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
1986         int err = -ENOMEM;
1987
1988         in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
1989         out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
1990         if (!in_mad || !out_mad)
1991                 goto out;
1992
1993         init_query_mad(in_mad);
1994         in_mad->attr_id = IB_SMP_ATTR_NODE_DESC;
1995         if (mlx4_is_master(dev->dev))
1996                 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
1997
1998         err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
1999         if (err)
2000                 goto out;
2001
2002         memcpy(dev->ib_dev.node_desc, out_mad->data, IB_DEVICE_NODE_DESC_MAX);
2003
2004         in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
2005
2006         err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
2007         if (err)
2008                 goto out;
2009
2010         dev->dev->rev_id = be32_to_cpup((__be32 *) (out_mad->data + 32));
2011         memcpy(&dev->ib_dev.node_guid, out_mad->data + 12, 8);
2012
2013 out:
2014         kfree(in_mad);
2015         kfree(out_mad);
2016         return err;
2017 }
2018
2019 static ssize_t hca_type_show(struct device *device,
2020                              struct device_attribute *attr, char *buf)
2021 {
2022         struct mlx4_ib_dev *dev =
2023                 rdma_device_to_drv_device(device, struct mlx4_ib_dev, ib_dev);
2024         return sprintf(buf, "MT%d\n", dev->dev->persist->pdev->device);
2025 }
2026 static DEVICE_ATTR_RO(hca_type);
2027
2028 static ssize_t hw_rev_show(struct device *device,
2029                            struct device_attribute *attr, char *buf)
2030 {
2031         struct mlx4_ib_dev *dev =
2032                 rdma_device_to_drv_device(device, struct mlx4_ib_dev, ib_dev);
2033         return sprintf(buf, "%x\n", dev->dev->rev_id);
2034 }
2035 static DEVICE_ATTR_RO(hw_rev);
2036
2037 static ssize_t board_id_show(struct device *device,
2038                              struct device_attribute *attr, char *buf)
2039 {
2040         struct mlx4_ib_dev *dev =
2041                 rdma_device_to_drv_device(device, struct mlx4_ib_dev, ib_dev);
2042
2043         return sprintf(buf, "%.*s\n", MLX4_BOARD_ID_LEN,
2044                        dev->dev->board_id);
2045 }
2046 static DEVICE_ATTR_RO(board_id);
2047
2048 static struct attribute *mlx4_class_attributes[] = {
2049         &dev_attr_hw_rev.attr,
2050         &dev_attr_hca_type.attr,
2051         &dev_attr_board_id.attr,
2052         NULL
2053 };
2054
2055 static const struct attribute_group mlx4_attr_group = {
2056         .attrs = mlx4_class_attributes,
2057 };
2058
2059 struct diag_counter {
2060         const char *name;
2061         u32 offset;
2062 };
2063
2064 #define DIAG_COUNTER(_name, _offset)                    \
2065         { .name = #_name, .offset = _offset }
2066
2067 static const struct diag_counter diag_basic[] = {
2068         DIAG_COUNTER(rq_num_lle, 0x00),
2069         DIAG_COUNTER(sq_num_lle, 0x04),
2070         DIAG_COUNTER(rq_num_lqpoe, 0x08),
2071         DIAG_COUNTER(sq_num_lqpoe, 0x0C),
2072         DIAG_COUNTER(rq_num_lpe, 0x18),
2073         DIAG_COUNTER(sq_num_lpe, 0x1C),
2074         DIAG_COUNTER(rq_num_wrfe, 0x20),
2075         DIAG_COUNTER(sq_num_wrfe, 0x24),
2076         DIAG_COUNTER(sq_num_mwbe, 0x2C),
2077         DIAG_COUNTER(sq_num_bre, 0x34),
2078         DIAG_COUNTER(sq_num_rire, 0x44),
2079         DIAG_COUNTER(rq_num_rire, 0x48),
2080         DIAG_COUNTER(sq_num_rae, 0x4C),
2081         DIAG_COUNTER(rq_num_rae, 0x50),
2082         DIAG_COUNTER(sq_num_roe, 0x54),
2083         DIAG_COUNTER(sq_num_tree, 0x5C),
2084         DIAG_COUNTER(sq_num_rree, 0x64),
2085         DIAG_COUNTER(rq_num_rnr, 0x68),
2086         DIAG_COUNTER(sq_num_rnr, 0x6C),
2087         DIAG_COUNTER(rq_num_oos, 0x100),
2088         DIAG_COUNTER(sq_num_oos, 0x104),
2089 };
2090
2091 static const struct diag_counter diag_ext[] = {
2092         DIAG_COUNTER(rq_num_dup, 0x130),
2093         DIAG_COUNTER(sq_num_to, 0x134),
2094 };
2095
2096 static const struct diag_counter diag_device_only[] = {
2097         DIAG_COUNTER(num_cqovf, 0x1A0),
2098         DIAG_COUNTER(rq_num_udsdprd, 0x118),
2099 };
2100
2101 static struct rdma_hw_stats *mlx4_ib_alloc_hw_stats(struct ib_device *ibdev,
2102                                                     u8 port_num)
2103 {
2104         struct mlx4_ib_dev *dev = to_mdev(ibdev);
2105         struct mlx4_ib_diag_counters *diag = dev->diag_counters;
2106
2107         if (!diag[!!port_num].name)
2108                 return NULL;
2109
2110         return rdma_alloc_hw_stats_struct(diag[!!port_num].name,
2111                                           diag[!!port_num].num_counters,
2112                                           RDMA_HW_STATS_DEFAULT_LIFESPAN);
2113 }
2114
2115 static int mlx4_ib_get_hw_stats(struct ib_device *ibdev,
2116                                 struct rdma_hw_stats *stats,
2117                                 u8 port, int index)
2118 {
2119         struct mlx4_ib_dev *dev = to_mdev(ibdev);
2120         struct mlx4_ib_diag_counters *diag = dev->diag_counters;
2121         u32 hw_value[ARRAY_SIZE(diag_device_only) +
2122                 ARRAY_SIZE(diag_ext) + ARRAY_SIZE(diag_basic)] = {};
2123         int ret;
2124         int i;
2125
2126         ret = mlx4_query_diag_counters(dev->dev,
2127                                        MLX4_OP_MOD_QUERY_TRANSPORT_CI_ERRORS,
2128                                        diag[!!port].offset, hw_value,
2129                                        diag[!!port].num_counters, port);
2130
2131         if (ret)
2132                 return ret;
2133
2134         for (i = 0; i < diag[!!port].num_counters; i++)
2135                 stats->value[i] = hw_value[i];
2136
2137         return diag[!!port].num_counters;
2138 }
2139
2140 static int __mlx4_ib_alloc_diag_counters(struct mlx4_ib_dev *ibdev,
2141                                          const char ***name,
2142                                          u32 **offset,
2143                                          u32 *num,
2144                                          bool port)
2145 {
2146         u32 num_counters;
2147
2148         num_counters = ARRAY_SIZE(diag_basic);
2149
2150         if (ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DIAG_PER_PORT)
2151                 num_counters += ARRAY_SIZE(diag_ext);
2152
2153         if (!port)
2154                 num_counters += ARRAY_SIZE(diag_device_only);
2155
2156         *name = kcalloc(num_counters, sizeof(**name), GFP_KERNEL);
2157         if (!*name)
2158                 return -ENOMEM;
2159
2160         *offset = kcalloc(num_counters, sizeof(**offset), GFP_KERNEL);
2161         if (!*offset)
2162                 goto err_name;
2163
2164         *num = num_counters;
2165
2166         return 0;
2167
2168 err_name:
2169         kfree(*name);
2170         return -ENOMEM;
2171 }
2172
2173 static void mlx4_ib_fill_diag_counters(struct mlx4_ib_dev *ibdev,
2174                                        const char **name,
2175                                        u32 *offset,
2176                                        bool port)
2177 {
2178         int i;
2179         int j;
2180
2181         for (i = 0, j = 0; i < ARRAY_SIZE(diag_basic); i++, j++) {
2182                 name[i] = diag_basic[i].name;
2183                 offset[i] = diag_basic[i].offset;
2184         }
2185
2186         if (ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DIAG_PER_PORT) {
2187                 for (i = 0; i < ARRAY_SIZE(diag_ext); i++, j++) {
2188                         name[j] = diag_ext[i].name;
2189                         offset[j] = diag_ext[i].offset;
2190                 }
2191         }
2192
2193         if (!port) {
2194                 for (i = 0; i < ARRAY_SIZE(diag_device_only); i++, j++) {
2195                         name[j] = diag_device_only[i].name;
2196                         offset[j] = diag_device_only[i].offset;
2197                 }
2198         }
2199 }
2200
2201 static const struct ib_device_ops mlx4_ib_hw_stats_ops = {
2202         .alloc_hw_stats = mlx4_ib_alloc_hw_stats,
2203         .get_hw_stats = mlx4_ib_get_hw_stats,
2204 };
2205
2206 static int mlx4_ib_alloc_diag_counters(struct mlx4_ib_dev *ibdev)
2207 {
2208         struct mlx4_ib_diag_counters *diag = ibdev->diag_counters;
2209         int i;
2210         int ret;
2211         bool per_port = !!(ibdev->dev->caps.flags2 &
2212                 MLX4_DEV_CAP_FLAG2_DIAG_PER_PORT);
2213
2214         if (mlx4_is_slave(ibdev->dev))
2215                 return 0;
2216
2217         for (i = 0; i < MLX4_DIAG_COUNTERS_TYPES; i++) {
2218                 /* i == 1 means we are building port counters */
2219                 if (i && !per_port)
2220                         continue;
2221
2222                 ret = __mlx4_ib_alloc_diag_counters(ibdev, &diag[i].name,
2223                                                     &diag[i].offset,
2224                                                     &diag[i].num_counters, i);
2225                 if (ret)
2226                         goto err_alloc;
2227
2228                 mlx4_ib_fill_diag_counters(ibdev, diag[i].name,
2229                                            diag[i].offset, i);
2230         }
2231
2232         ib_set_device_ops(&ibdev->ib_dev, &mlx4_ib_hw_stats_ops);
2233
2234         return 0;
2235
2236 err_alloc:
2237         if (i) {
2238                 kfree(diag[i - 1].name);
2239                 kfree(diag[i - 1].offset);
2240         }
2241
2242         return ret;
2243 }
2244
2245 static void mlx4_ib_diag_cleanup(struct mlx4_ib_dev *ibdev)
2246 {
2247         int i;
2248
2249         for (i = 0; i < MLX4_DIAG_COUNTERS_TYPES; i++) {
2250                 kfree(ibdev->diag_counters[i].offset);
2251                 kfree(ibdev->diag_counters[i].name);
2252         }
2253 }
2254
2255 #define MLX4_IB_INVALID_MAC     ((u64)-1)
2256 static void mlx4_ib_update_qps(struct mlx4_ib_dev *ibdev,
2257                                struct net_device *dev,
2258                                int port)
2259 {
2260         u64 new_smac = 0;
2261         u64 release_mac = MLX4_IB_INVALID_MAC;
2262         struct mlx4_ib_qp *qp;
2263
2264         read_lock(&dev_base_lock);
2265         new_smac = mlx4_mac_to_u64(dev->dev_addr);
2266         read_unlock(&dev_base_lock);
2267
2268         atomic64_set(&ibdev->iboe.mac[port - 1], new_smac);
2269
2270         /* no need for update QP1 and mac registration in non-SRIOV */
2271         if (!mlx4_is_mfunc(ibdev->dev))
2272                 return;
2273
2274         mutex_lock(&ibdev->qp1_proxy_lock[port - 1]);
2275         qp = ibdev->qp1_proxy[port - 1];
2276         if (qp) {
2277                 int new_smac_index;
2278                 u64 old_smac;
2279                 struct mlx4_update_qp_params update_params;
2280
2281                 mutex_lock(&qp->mutex);
2282                 old_smac = qp->pri.smac;
2283                 if (new_smac == old_smac)
2284                         goto unlock;
2285
2286                 new_smac_index = mlx4_register_mac(ibdev->dev, port, new_smac);
2287
2288                 if (new_smac_index < 0)
2289                         goto unlock;
2290
2291                 update_params.smac_index = new_smac_index;
2292                 if (mlx4_update_qp(ibdev->dev, qp->mqp.qpn, MLX4_UPDATE_QP_SMAC,
2293                                    &update_params)) {
2294                         release_mac = new_smac;
2295                         goto unlock;
2296                 }
2297                 /* if old port was zero, no mac was yet registered for this QP */
2298                 if (qp->pri.smac_port)
2299                         release_mac = old_smac;
2300                 qp->pri.smac = new_smac;
2301                 qp->pri.smac_port = port;
2302                 qp->pri.smac_index = new_smac_index;
2303         }
2304
2305 unlock:
2306         if (release_mac != MLX4_IB_INVALID_MAC)
2307                 mlx4_unregister_mac(ibdev->dev, port, release_mac);
2308         if (qp)
2309                 mutex_unlock(&qp->mutex);
2310         mutex_unlock(&ibdev->qp1_proxy_lock[port - 1]);
2311 }
2312
2313 static void mlx4_ib_scan_netdevs(struct mlx4_ib_dev *ibdev,
2314                                  struct net_device *dev,
2315                                  unsigned long event)
2316
2317 {
2318         struct mlx4_ib_iboe *iboe;
2319         int update_qps_port = -1;
2320         int port;
2321
2322         ASSERT_RTNL();
2323
2324         iboe = &ibdev->iboe;
2325
2326         spin_lock_bh(&iboe->lock);
2327         mlx4_foreach_ib_transport_port(port, ibdev->dev) {
2328
2329                 iboe->netdevs[port - 1] =
2330                         mlx4_get_protocol_dev(ibdev->dev, MLX4_PROT_ETH, port);
2331
2332                 if (dev == iboe->netdevs[port - 1] &&
2333                     (event == NETDEV_CHANGEADDR || event == NETDEV_REGISTER ||
2334                      event == NETDEV_UP || event == NETDEV_CHANGE))
2335                         update_qps_port = port;
2336
2337                 if (dev == iboe->netdevs[port - 1] &&
2338                     (event == NETDEV_UP || event == NETDEV_DOWN)) {
2339                         enum ib_port_state port_state;
2340                         struct ib_event ibev = { };
2341
2342                         if (ib_get_cached_port_state(&ibdev->ib_dev, port,
2343                                                      &port_state))
2344                                 continue;
2345
2346                         if (event == NETDEV_UP &&
2347                             (port_state != IB_PORT_ACTIVE ||
2348                              iboe->last_port_state[port - 1] != IB_PORT_DOWN))
2349                                 continue;
2350                         if (event == NETDEV_DOWN &&
2351                             (port_state != IB_PORT_DOWN ||
2352                              iboe->last_port_state[port - 1] != IB_PORT_ACTIVE))
2353                                 continue;
2354                         iboe->last_port_state[port - 1] = port_state;
2355
2356                         ibev.device = &ibdev->ib_dev;
2357                         ibev.element.port_num = port;
2358                         ibev.event = event == NETDEV_UP ? IB_EVENT_PORT_ACTIVE :
2359                                                           IB_EVENT_PORT_ERR;
2360                         ib_dispatch_event(&ibev);
2361                 }
2362
2363         }
2364         spin_unlock_bh(&iboe->lock);
2365
2366         if (update_qps_port > 0)
2367                 mlx4_ib_update_qps(ibdev, dev, update_qps_port);
2368 }
2369
2370 static int mlx4_ib_netdev_event(struct notifier_block *this,
2371                                 unsigned long event, void *ptr)
2372 {
2373         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
2374         struct mlx4_ib_dev *ibdev;
2375
2376         if (!net_eq(dev_net(dev), &init_net))
2377                 return NOTIFY_DONE;
2378
2379         ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb);
2380         mlx4_ib_scan_netdevs(ibdev, dev, event);
2381
2382         return NOTIFY_DONE;
2383 }
2384
2385 static void init_pkeys(struct mlx4_ib_dev *ibdev)
2386 {
2387         int port;
2388         int slave;
2389         int i;
2390
2391         if (mlx4_is_master(ibdev->dev)) {
2392                 for (slave = 0; slave <= ibdev->dev->persist->num_vfs;
2393                      ++slave) {
2394                         for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
2395                                 for (i = 0;
2396                                      i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
2397                                      ++i) {
2398                                         ibdev->pkeys.virt2phys_pkey[slave][port - 1][i] =
2399                                         /* master has the identity virt2phys pkey mapping */
2400                                                 (slave == mlx4_master_func_num(ibdev->dev) || !i) ? i :
2401                                                         ibdev->dev->phys_caps.pkey_phys_table_len[port] - 1;
2402                                         mlx4_sync_pkey_table(ibdev->dev, slave, port, i,
2403                                                              ibdev->pkeys.virt2phys_pkey[slave][port - 1][i]);
2404                                 }
2405                         }
2406                 }
2407                 /* initialize pkey cache */
2408                 for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
2409                         for (i = 0;
2410                              i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
2411                              ++i)
2412                                 ibdev->pkeys.phys_pkey_cache[port-1][i] =
2413                                         (i) ? 0 : 0xFFFF;
2414                 }
2415         }
2416 }
2417
2418 static void mlx4_ib_alloc_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
2419 {
2420         int i, j, eq = 0, total_eqs = 0;
2421
2422         ibdev->eq_table = kcalloc(dev->caps.num_comp_vectors,
2423                                   sizeof(ibdev->eq_table[0]), GFP_KERNEL);
2424         if (!ibdev->eq_table)
2425                 return;
2426
2427         for (i = 1; i <= dev->caps.num_ports; i++) {
2428                 for (j = 0; j < mlx4_get_eqs_per_port(dev, i);
2429                      j++, total_eqs++) {
2430                         if (i > 1 &&  mlx4_is_eq_shared(dev, total_eqs))
2431                                 continue;
2432                         ibdev->eq_table[eq] = total_eqs;
2433                         if (!mlx4_assign_eq(dev, i,
2434                                             &ibdev->eq_table[eq]))
2435                                 eq++;
2436                         else
2437                                 ibdev->eq_table[eq] = -1;
2438                 }
2439         }
2440
2441         for (i = eq; i < dev->caps.num_comp_vectors;
2442              ibdev->eq_table[i++] = -1)
2443                 ;
2444
2445         /* Advertise the new number of EQs to clients */
2446         ibdev->ib_dev.num_comp_vectors = eq;
2447 }
2448
2449 static void mlx4_ib_free_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
2450 {
2451         int i;
2452         int total_eqs = ibdev->ib_dev.num_comp_vectors;
2453
2454         /* no eqs were allocated */
2455         if (!ibdev->eq_table)
2456                 return;
2457
2458         /* Reset the advertised EQ number */
2459         ibdev->ib_dev.num_comp_vectors = 0;
2460
2461         for (i = 0; i < total_eqs; i++)
2462                 mlx4_release_eq(dev, ibdev->eq_table[i]);
2463
2464         kfree(ibdev->eq_table);
2465         ibdev->eq_table = NULL;
2466 }
2467
2468 static int mlx4_port_immutable(struct ib_device *ibdev, u8 port_num,
2469                                struct ib_port_immutable *immutable)
2470 {
2471         struct ib_port_attr attr;
2472         struct mlx4_ib_dev *mdev = to_mdev(ibdev);
2473         int err;
2474
2475         if (mlx4_ib_port_link_layer(ibdev, port_num) == IB_LINK_LAYER_INFINIBAND) {
2476                 immutable->core_cap_flags = RDMA_CORE_PORT_IBA_IB;
2477                 immutable->max_mad_size = IB_MGMT_MAD_SIZE;
2478         } else {
2479                 if (mdev->dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE)
2480                         immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE;
2481                 if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ROCE_V1_V2)
2482                         immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE |
2483                                 RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP;
2484                 immutable->core_cap_flags |= RDMA_CORE_PORT_RAW_PACKET;
2485                 if (immutable->core_cap_flags & (RDMA_CORE_PORT_IBA_ROCE |
2486                     RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP))
2487                         immutable->max_mad_size = IB_MGMT_MAD_SIZE;
2488         }
2489
2490         err = ib_query_port(ibdev, port_num, &attr);
2491         if (err)
2492                 return err;
2493
2494         immutable->pkey_tbl_len = attr.pkey_tbl_len;
2495         immutable->gid_tbl_len = attr.gid_tbl_len;
2496
2497         return 0;
2498 }
2499
2500 static void get_fw_ver_str(struct ib_device *device, char *str)
2501 {
2502         struct mlx4_ib_dev *dev =
2503                 container_of(device, struct mlx4_ib_dev, ib_dev);
2504         snprintf(str, IB_FW_VERSION_NAME_MAX, "%d.%d.%d",
2505                  (int) (dev->dev->caps.fw_ver >> 32),
2506                  (int) (dev->dev->caps.fw_ver >> 16) & 0xffff,
2507                  (int) dev->dev->caps.fw_ver & 0xffff);
2508 }
2509
2510 static const struct ib_device_ops mlx4_ib_dev_ops = {
2511         .owner = THIS_MODULE,
2512         .driver_id = RDMA_DRIVER_MLX4,
2513         .uverbs_abi_ver = MLX4_IB_UVERBS_ABI_VERSION,
2514
2515         .add_gid = mlx4_ib_add_gid,
2516         .alloc_mr = mlx4_ib_alloc_mr,
2517         .alloc_pd = mlx4_ib_alloc_pd,
2518         .alloc_ucontext = mlx4_ib_alloc_ucontext,
2519         .attach_mcast = mlx4_ib_mcg_attach,
2520         .create_ah = mlx4_ib_create_ah,
2521         .create_cq = mlx4_ib_create_cq,
2522         .create_qp = mlx4_ib_create_qp,
2523         .create_srq = mlx4_ib_create_srq,
2524         .dealloc_pd = mlx4_ib_dealloc_pd,
2525         .dealloc_ucontext = mlx4_ib_dealloc_ucontext,
2526         .del_gid = mlx4_ib_del_gid,
2527         .dereg_mr = mlx4_ib_dereg_mr,
2528         .destroy_ah = mlx4_ib_destroy_ah,
2529         .destroy_cq = mlx4_ib_destroy_cq,
2530         .destroy_qp = mlx4_ib_destroy_qp,
2531         .destroy_srq = mlx4_ib_destroy_srq,
2532         .detach_mcast = mlx4_ib_mcg_detach,
2533         .disassociate_ucontext = mlx4_ib_disassociate_ucontext,
2534         .drain_rq = mlx4_ib_drain_rq,
2535         .drain_sq = mlx4_ib_drain_sq,
2536         .get_dev_fw_str = get_fw_ver_str,
2537         .get_dma_mr = mlx4_ib_get_dma_mr,
2538         .get_link_layer = mlx4_ib_port_link_layer,
2539         .get_netdev = mlx4_ib_get_netdev,
2540         .get_port_immutable = mlx4_port_immutable,
2541         .map_mr_sg = mlx4_ib_map_mr_sg,
2542         .mmap = mlx4_ib_mmap,
2543         .modify_cq = mlx4_ib_modify_cq,
2544         .modify_device = mlx4_ib_modify_device,
2545         .modify_port = mlx4_ib_modify_port,
2546         .modify_qp = mlx4_ib_modify_qp,
2547         .modify_srq = mlx4_ib_modify_srq,
2548         .poll_cq = mlx4_ib_poll_cq,
2549         .post_recv = mlx4_ib_post_recv,
2550         .post_send = mlx4_ib_post_send,
2551         .post_srq_recv = mlx4_ib_post_srq_recv,
2552         .process_mad = mlx4_ib_process_mad,
2553         .query_ah = mlx4_ib_query_ah,
2554         .query_device = mlx4_ib_query_device,
2555         .query_gid = mlx4_ib_query_gid,
2556         .query_pkey = mlx4_ib_query_pkey,
2557         .query_port = mlx4_ib_query_port,
2558         .query_qp = mlx4_ib_query_qp,
2559         .query_srq = mlx4_ib_query_srq,
2560         .reg_user_mr = mlx4_ib_reg_user_mr,
2561         .req_notify_cq = mlx4_ib_arm_cq,
2562         .rereg_user_mr = mlx4_ib_rereg_user_mr,
2563         .resize_cq = mlx4_ib_resize_cq,
2564
2565         INIT_RDMA_OBJ_SIZE(ib_ah, mlx4_ib_ah, ibah),
2566         INIT_RDMA_OBJ_SIZE(ib_cq, mlx4_ib_cq, ibcq),
2567         INIT_RDMA_OBJ_SIZE(ib_pd, mlx4_ib_pd, ibpd),
2568         INIT_RDMA_OBJ_SIZE(ib_srq, mlx4_ib_srq, ibsrq),
2569         INIT_RDMA_OBJ_SIZE(ib_ucontext, mlx4_ib_ucontext, ibucontext),
2570 };
2571
2572 static const struct ib_device_ops mlx4_ib_dev_wq_ops = {
2573         .create_rwq_ind_table = mlx4_ib_create_rwq_ind_table,
2574         .create_wq = mlx4_ib_create_wq,
2575         .destroy_rwq_ind_table = mlx4_ib_destroy_rwq_ind_table,
2576         .destroy_wq = mlx4_ib_destroy_wq,
2577         .modify_wq = mlx4_ib_modify_wq,
2578
2579         INIT_RDMA_OBJ_SIZE(ib_rwq_ind_table, mlx4_ib_rwq_ind_table,
2580                            ib_rwq_ind_tbl),
2581 };
2582
2583 static const struct ib_device_ops mlx4_ib_dev_mw_ops = {
2584         .alloc_mw = mlx4_ib_alloc_mw,
2585         .dealloc_mw = mlx4_ib_dealloc_mw,
2586
2587         INIT_RDMA_OBJ_SIZE(ib_mw, mlx4_ib_mw, ibmw),
2588 };
2589
2590 static const struct ib_device_ops mlx4_ib_dev_xrc_ops = {
2591         .alloc_xrcd = mlx4_ib_alloc_xrcd,
2592         .dealloc_xrcd = mlx4_ib_dealloc_xrcd,
2593
2594         INIT_RDMA_OBJ_SIZE(ib_xrcd, mlx4_ib_xrcd, ibxrcd),
2595 };
2596
2597 static const struct ib_device_ops mlx4_ib_dev_fs_ops = {
2598         .create_flow = mlx4_ib_create_flow,
2599         .destroy_flow = mlx4_ib_destroy_flow,
2600 };
2601
2602 static void *mlx4_ib_add(struct mlx4_dev *dev)
2603 {
2604         struct mlx4_ib_dev *ibdev;
2605         int num_ports = 0;
2606         int i, j;
2607         int err;
2608         struct mlx4_ib_iboe *iboe;
2609         int ib_num_ports = 0;
2610         int num_req_counters;
2611         int allocated;
2612         u32 counter_index;
2613         struct counter_index *new_counter_index = NULL;
2614
2615         pr_info_once("%s", mlx4_ib_version);
2616
2617         num_ports = 0;
2618         mlx4_foreach_ib_transport_port(i, dev)
2619                 num_ports++;
2620
2621         /* No point in registering a device with no ports... */
2622         if (num_ports == 0)
2623                 return NULL;
2624
2625         ibdev = ib_alloc_device(mlx4_ib_dev, ib_dev);
2626         if (!ibdev) {
2627                 dev_err(&dev->persist->pdev->dev,
2628                         "Device struct alloc failed\n");
2629                 return NULL;
2630         }
2631
2632         iboe = &ibdev->iboe;
2633
2634         if (mlx4_pd_alloc(dev, &ibdev->priv_pdn))
2635                 goto err_dealloc;
2636
2637         if (mlx4_uar_alloc(dev, &ibdev->priv_uar))
2638                 goto err_pd;
2639
2640         ibdev->uar_map = ioremap((phys_addr_t) ibdev->priv_uar.pfn << PAGE_SHIFT,
2641                                  PAGE_SIZE);
2642         if (!ibdev->uar_map)
2643                 goto err_uar;
2644         MLX4_INIT_DOORBELL_LOCK(&ibdev->uar_lock);
2645
2646         ibdev->dev = dev;
2647         ibdev->bond_next_port   = 0;
2648
2649         ibdev->ib_dev.node_type         = RDMA_NODE_IB_CA;
2650         ibdev->ib_dev.local_dma_lkey    = dev->caps.reserved_lkey;
2651         ibdev->num_ports                = num_ports;
2652         ibdev->ib_dev.phys_port_cnt     = mlx4_is_bonded(dev) ?
2653                                                 1 : ibdev->num_ports;
2654         ibdev->ib_dev.num_comp_vectors  = dev->caps.num_comp_vectors;
2655         ibdev->ib_dev.dev.parent        = &dev->persist->pdev->dev;
2656
2657         ibdev->ib_dev.uverbs_cmd_mask   =
2658                 (1ull << IB_USER_VERBS_CMD_GET_CONTEXT)         |
2659                 (1ull << IB_USER_VERBS_CMD_QUERY_DEVICE)        |
2660                 (1ull << IB_USER_VERBS_CMD_QUERY_PORT)          |
2661                 (1ull << IB_USER_VERBS_CMD_ALLOC_PD)            |
2662                 (1ull << IB_USER_VERBS_CMD_DEALLOC_PD)          |
2663                 (1ull << IB_USER_VERBS_CMD_REG_MR)              |
2664                 (1ull << IB_USER_VERBS_CMD_REREG_MR)            |
2665                 (1ull << IB_USER_VERBS_CMD_DEREG_MR)            |
2666                 (1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL) |
2667                 (1ull << IB_USER_VERBS_CMD_CREATE_CQ)           |
2668                 (1ull << IB_USER_VERBS_CMD_RESIZE_CQ)           |
2669                 (1ull << IB_USER_VERBS_CMD_DESTROY_CQ)          |
2670                 (1ull << IB_USER_VERBS_CMD_CREATE_QP)           |
2671                 (1ull << IB_USER_VERBS_CMD_MODIFY_QP)           |
2672                 (1ull << IB_USER_VERBS_CMD_QUERY_QP)            |
2673                 (1ull << IB_USER_VERBS_CMD_DESTROY_QP)          |
2674                 (1ull << IB_USER_VERBS_CMD_ATTACH_MCAST)        |
2675                 (1ull << IB_USER_VERBS_CMD_DETACH_MCAST)        |
2676                 (1ull << IB_USER_VERBS_CMD_CREATE_SRQ)          |
2677                 (1ull << IB_USER_VERBS_CMD_MODIFY_SRQ)          |
2678                 (1ull << IB_USER_VERBS_CMD_QUERY_SRQ)           |
2679                 (1ull << IB_USER_VERBS_CMD_DESTROY_SRQ)         |
2680                 (1ull << IB_USER_VERBS_CMD_CREATE_XSRQ)         |
2681                 (1ull << IB_USER_VERBS_CMD_OPEN_QP);
2682
2683         ib_set_device_ops(&ibdev->ib_dev, &mlx4_ib_dev_ops);
2684         ibdev->ib_dev.uverbs_ex_cmd_mask |=
2685                 (1ull << IB_USER_VERBS_EX_CMD_CREATE_CQ) |
2686                 (1ull << IB_USER_VERBS_EX_CMD_CREATE_QP);
2687
2688         if ((dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_RSS) &&
2689             ((mlx4_ib_port_link_layer(&ibdev->ib_dev, 1) ==
2690             IB_LINK_LAYER_ETHERNET) ||
2691             (mlx4_ib_port_link_layer(&ibdev->ib_dev, 2) ==
2692             IB_LINK_LAYER_ETHERNET)))
2693                 ib_set_device_ops(&ibdev->ib_dev, &mlx4_ib_dev_wq_ops);
2694
2695         if (dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW ||
2696             dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
2697                 ibdev->ib_dev.uverbs_cmd_mask |=
2698                         (1ull << IB_USER_VERBS_CMD_ALLOC_MW) |
2699                         (1ull << IB_USER_VERBS_CMD_DEALLOC_MW);
2700                 ib_set_device_ops(&ibdev->ib_dev, &mlx4_ib_dev_mw_ops);
2701         }
2702
2703         if (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) {
2704                 ibdev->ib_dev.uverbs_cmd_mask |=
2705                         (1ull << IB_USER_VERBS_CMD_OPEN_XRCD) |
2706                         (1ull << IB_USER_VERBS_CMD_CLOSE_XRCD);
2707                 ib_set_device_ops(&ibdev->ib_dev, &mlx4_ib_dev_xrc_ops);
2708         }
2709
2710         if (check_flow_steering_support(dev)) {
2711                 ibdev->steering_support = MLX4_STEERING_MODE_DEVICE_MANAGED;
2712                 ib_set_device_ops(&ibdev->ib_dev, &mlx4_ib_dev_fs_ops);
2713         }
2714
2715         if (!dev->caps.userspace_caps)
2716                 ibdev->ib_dev.ops.uverbs_abi_ver =
2717                         MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION;
2718
2719         mlx4_ib_alloc_eqs(dev, ibdev);
2720
2721         spin_lock_init(&iboe->lock);
2722
2723         if (init_node_data(ibdev))
2724                 goto err_map;
2725         mlx4_init_sl2vl_tbl(ibdev);
2726
2727         for (i = 0; i < ibdev->num_ports; ++i) {
2728                 mutex_init(&ibdev->counters_table[i].mutex);
2729                 INIT_LIST_HEAD(&ibdev->counters_table[i].counters_list);
2730                 iboe->last_port_state[i] = IB_PORT_DOWN;
2731         }
2732
2733         num_req_counters = mlx4_is_bonded(dev) ? 1 : ibdev->num_ports;
2734         for (i = 0; i < num_req_counters; ++i) {
2735                 mutex_init(&ibdev->qp1_proxy_lock[i]);
2736                 allocated = 0;
2737                 if (mlx4_ib_port_link_layer(&ibdev->ib_dev, i + 1) ==
2738                                                 IB_LINK_LAYER_ETHERNET) {
2739                         err = mlx4_counter_alloc(ibdev->dev, &counter_index,
2740                                                  MLX4_RES_USAGE_DRIVER);
2741                         /* if failed to allocate a new counter, use default */
2742                         if (err)
2743                                 counter_index =
2744                                         mlx4_get_default_counter_index(dev,
2745                                                                        i + 1);
2746                         else
2747                                 allocated = 1;
2748                 } else { /* IB_LINK_LAYER_INFINIBAND use the default counter */
2749                         counter_index = mlx4_get_default_counter_index(dev,
2750                                                                        i + 1);
2751                 }
2752                 new_counter_index = kmalloc(sizeof(*new_counter_index),
2753                                             GFP_KERNEL);
2754                 if (!new_counter_index) {
2755                         if (allocated)
2756                                 mlx4_counter_free(ibdev->dev, counter_index);
2757                         goto err_counter;
2758                 }
2759                 new_counter_index->index = counter_index;
2760                 new_counter_index->allocated = allocated;
2761                 list_add_tail(&new_counter_index->list,
2762                               &ibdev->counters_table[i].counters_list);
2763                 ibdev->counters_table[i].default_counter = counter_index;
2764                 pr_info("counter index %d for port %d allocated %d\n",
2765                         counter_index, i + 1, allocated);
2766         }
2767         if (mlx4_is_bonded(dev))
2768                 for (i = 1; i < ibdev->num_ports ; ++i) {
2769                         new_counter_index =
2770                                         kmalloc(sizeof(struct counter_index),
2771                                                 GFP_KERNEL);
2772                         if (!new_counter_index)
2773                                 goto err_counter;
2774                         new_counter_index->index = counter_index;
2775                         new_counter_index->allocated = 0;
2776                         list_add_tail(&new_counter_index->list,
2777                                       &ibdev->counters_table[i].counters_list);
2778                         ibdev->counters_table[i].default_counter =
2779                                                                 counter_index;
2780                 }
2781
2782         mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
2783                 ib_num_ports++;
2784
2785         spin_lock_init(&ibdev->sm_lock);
2786         mutex_init(&ibdev->cap_mask_mutex);
2787         INIT_LIST_HEAD(&ibdev->qp_list);
2788         spin_lock_init(&ibdev->reset_flow_resource_lock);
2789
2790         if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED &&
2791             ib_num_ports) {
2792                 ibdev->steer_qpn_count = MLX4_IB_UC_MAX_NUM_QPS;
2793                 err = mlx4_qp_reserve_range(dev, ibdev->steer_qpn_count,
2794                                             MLX4_IB_UC_STEER_QPN_ALIGN,
2795                                             &ibdev->steer_qpn_base, 0,
2796                                             MLX4_RES_USAGE_DRIVER);
2797                 if (err)
2798                         goto err_counter;
2799
2800                 ibdev->ib_uc_qpns_bitmap =
2801                         kmalloc_array(BITS_TO_LONGS(ibdev->steer_qpn_count),
2802                                       sizeof(long),
2803                                       GFP_KERNEL);
2804                 if (!ibdev->ib_uc_qpns_bitmap)
2805                         goto err_steer_qp_release;
2806
2807                 if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DMFS_IPOIB) {
2808                         bitmap_zero(ibdev->ib_uc_qpns_bitmap,
2809                                     ibdev->steer_qpn_count);
2810                         err = mlx4_FLOW_STEERING_IB_UC_QP_RANGE(
2811                                         dev, ibdev->steer_qpn_base,
2812                                         ibdev->steer_qpn_base +
2813                                         ibdev->steer_qpn_count - 1);
2814                         if (err)
2815                                 goto err_steer_free_bitmap;
2816                 } else {
2817                         bitmap_fill(ibdev->ib_uc_qpns_bitmap,
2818                                     ibdev->steer_qpn_count);
2819                 }
2820         }
2821
2822         for (j = 1; j <= ibdev->dev->caps.num_ports; j++)
2823                 atomic64_set(&iboe->mac[j - 1], ibdev->dev->caps.def_mac[j]);
2824
2825         if (mlx4_ib_alloc_diag_counters(ibdev))
2826                 goto err_steer_free_bitmap;
2827
2828         rdma_set_device_sysfs_group(&ibdev->ib_dev, &mlx4_attr_group);
2829         if (ib_register_device(&ibdev->ib_dev, "mlx4_%d",
2830                                &dev->persist->pdev->dev))
2831                 goto err_diag_counters;
2832
2833         if (mlx4_ib_mad_init(ibdev))
2834                 goto err_reg;
2835
2836         if (mlx4_ib_init_sriov(ibdev))
2837                 goto err_mad;
2838
2839         if (!iboe->nb.notifier_call) {
2840                 iboe->nb.notifier_call = mlx4_ib_netdev_event;
2841                 err = register_netdevice_notifier(&iboe->nb);
2842                 if (err) {
2843                         iboe->nb.notifier_call = NULL;
2844                         goto err_notif;
2845                 }
2846         }
2847         if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ROCE_V1_V2) {
2848                 err = mlx4_config_roce_v2_port(dev, ROCE_V2_UDP_DPORT);
2849                 if (err)
2850                         goto err_notif;
2851         }
2852
2853         ibdev->ib_active = true;
2854         mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
2855                 devlink_port_type_ib_set(mlx4_get_devlink_port(dev, i),
2856                                          &ibdev->ib_dev);
2857
2858         if (mlx4_is_mfunc(ibdev->dev))
2859                 init_pkeys(ibdev);
2860
2861         /* create paravirt contexts for any VFs which are active */
2862         if (mlx4_is_master(ibdev->dev)) {
2863                 for (j = 0; j < MLX4_MFUNC_MAX; j++) {
2864                         if (j == mlx4_master_func_num(ibdev->dev))
2865                                 continue;
2866                         if (mlx4_is_slave_active(ibdev->dev, j))
2867                                 do_slave_init(ibdev, j, 1);
2868                 }
2869         }
2870         return ibdev;
2871
2872 err_notif:
2873         if (ibdev->iboe.nb.notifier_call) {
2874                 if (unregister_netdevice_notifier(&ibdev->iboe.nb))
2875                         pr_warn("failure unregistering notifier\n");
2876                 ibdev->iboe.nb.notifier_call = NULL;
2877         }
2878         flush_workqueue(wq);
2879
2880         mlx4_ib_close_sriov(ibdev);
2881
2882 err_mad:
2883         mlx4_ib_mad_cleanup(ibdev);
2884
2885 err_reg:
2886         ib_unregister_device(&ibdev->ib_dev);
2887
2888 err_diag_counters:
2889         mlx4_ib_diag_cleanup(ibdev);
2890
2891 err_steer_free_bitmap:
2892         kfree(ibdev->ib_uc_qpns_bitmap);
2893
2894 err_steer_qp_release:
2895         mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
2896                               ibdev->steer_qpn_count);
2897 err_counter:
2898         for (i = 0; i < ibdev->num_ports; ++i)
2899                 mlx4_ib_delete_counters_table(ibdev, &ibdev->counters_table[i]);
2900
2901 err_map:
2902         mlx4_ib_free_eqs(dev, ibdev);
2903         iounmap(ibdev->uar_map);
2904
2905 err_uar:
2906         mlx4_uar_free(dev, &ibdev->priv_uar);
2907
2908 err_pd:
2909         mlx4_pd_free(dev, ibdev->priv_pdn);
2910
2911 err_dealloc:
2912         ib_dealloc_device(&ibdev->ib_dev);
2913
2914         return NULL;
2915 }
2916
2917 int mlx4_ib_steer_qp_alloc(struct mlx4_ib_dev *dev, int count, int *qpn)
2918 {
2919         int offset;
2920
2921         WARN_ON(!dev->ib_uc_qpns_bitmap);
2922
2923         offset = bitmap_find_free_region(dev->ib_uc_qpns_bitmap,
2924                                          dev->steer_qpn_count,
2925                                          get_count_order(count));
2926         if (offset < 0)
2927                 return offset;
2928
2929         *qpn = dev->steer_qpn_base + offset;
2930         return 0;
2931 }
2932
2933 void mlx4_ib_steer_qp_free(struct mlx4_ib_dev *dev, u32 qpn, int count)
2934 {
2935         if (!qpn ||
2936             dev->steering_support != MLX4_STEERING_MODE_DEVICE_MANAGED)
2937                 return;
2938
2939         if (WARN(qpn < dev->steer_qpn_base, "qpn = %u, steer_qpn_base = %u\n",
2940                  qpn, dev->steer_qpn_base))
2941                 /* not supposed to be here */
2942                 return;
2943
2944         bitmap_release_region(dev->ib_uc_qpns_bitmap,
2945                               qpn - dev->steer_qpn_base,
2946                               get_count_order(count));
2947 }
2948
2949 int mlx4_ib_steer_qp_reg(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
2950                          int is_attach)
2951 {
2952         int err;
2953         size_t flow_size;
2954         struct ib_flow_attr *flow = NULL;
2955         struct ib_flow_spec_ib *ib_spec;
2956
2957         if (is_attach) {
2958                 flow_size = sizeof(struct ib_flow_attr) +
2959                             sizeof(struct ib_flow_spec_ib);
2960                 flow = kzalloc(flow_size, GFP_KERNEL);
2961                 if (!flow)
2962                         return -ENOMEM;
2963                 flow->port = mqp->port;
2964                 flow->num_of_specs = 1;
2965                 flow->size = flow_size;
2966                 ib_spec = (struct ib_flow_spec_ib *)(flow + 1);
2967                 ib_spec->type = IB_FLOW_SPEC_IB;
2968                 ib_spec->size = sizeof(struct ib_flow_spec_ib);
2969                 /* Add an empty rule for IB L2 */
2970                 memset(&ib_spec->mask, 0, sizeof(ib_spec->mask));
2971
2972                 err = __mlx4_ib_create_flow(&mqp->ibqp, flow, MLX4_DOMAIN_NIC,
2973                                             MLX4_FS_REGULAR, &mqp->reg_id);
2974         } else {
2975                 err = __mlx4_ib_destroy_flow(mdev->dev, mqp->reg_id);
2976         }
2977         kfree(flow);
2978         return err;
2979 }
2980
2981 static void mlx4_ib_remove(struct mlx4_dev *dev, void *ibdev_ptr)
2982 {
2983         struct mlx4_ib_dev *ibdev = ibdev_ptr;
2984         int p;
2985         int i;
2986
2987         mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
2988                 devlink_port_type_clear(mlx4_get_devlink_port(dev, i));
2989         ibdev->ib_active = false;
2990         flush_workqueue(wq);
2991
2992         if (ibdev->iboe.nb.notifier_call) {
2993                 if (unregister_netdevice_notifier(&ibdev->iboe.nb))
2994                         pr_warn("failure unregistering notifier\n");
2995                 ibdev->iboe.nb.notifier_call = NULL;
2996         }
2997
2998         mlx4_ib_close_sriov(ibdev);
2999         mlx4_ib_mad_cleanup(ibdev);
3000         ib_unregister_device(&ibdev->ib_dev);
3001         mlx4_ib_diag_cleanup(ibdev);
3002
3003         mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
3004                               ibdev->steer_qpn_count);
3005         kfree(ibdev->ib_uc_qpns_bitmap);
3006
3007         iounmap(ibdev->uar_map);
3008         for (p = 0; p < ibdev->num_ports; ++p)
3009                 mlx4_ib_delete_counters_table(ibdev, &ibdev->counters_table[p]);
3010
3011         mlx4_foreach_port(p, dev, MLX4_PORT_TYPE_IB)
3012                 mlx4_CLOSE_PORT(dev, p);
3013
3014         mlx4_ib_free_eqs(dev, ibdev);
3015
3016         mlx4_uar_free(dev, &ibdev->priv_uar);
3017         mlx4_pd_free(dev, ibdev->priv_pdn);
3018         ib_dealloc_device(&ibdev->ib_dev);
3019 }
3020
3021 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init)
3022 {
3023         struct mlx4_ib_demux_work **dm = NULL;
3024         struct mlx4_dev *dev = ibdev->dev;
3025         int i;
3026         unsigned long flags;
3027         struct mlx4_active_ports actv_ports;
3028         unsigned int ports;
3029         unsigned int first_port;
3030
3031         if (!mlx4_is_master(dev))
3032                 return;
3033
3034         actv_ports = mlx4_get_active_ports(dev, slave);
3035         ports = bitmap_weight(actv_ports.ports, dev->caps.num_ports);
3036         first_port = find_first_bit(actv_ports.ports, dev->caps.num_ports);
3037
3038         dm = kcalloc(ports, sizeof(*dm), GFP_ATOMIC);
3039         if (!dm)
3040                 return;
3041
3042         for (i = 0; i < ports; i++) {
3043                 dm[i] = kmalloc(sizeof (struct mlx4_ib_demux_work), GFP_ATOMIC);
3044                 if (!dm[i]) {
3045                         while (--i >= 0)
3046                                 kfree(dm[i]);
3047                         goto out;
3048                 }
3049                 INIT_WORK(&dm[i]->work, mlx4_ib_tunnels_update_work);
3050                 dm[i]->port = first_port + i + 1;
3051                 dm[i]->slave = slave;
3052                 dm[i]->do_init = do_init;
3053                 dm[i]->dev = ibdev;
3054         }
3055         /* initialize or tear down tunnel QPs for the slave */
3056         spin_lock_irqsave(&ibdev->sriov.going_down_lock, flags);
3057         if (!ibdev->sriov.is_going_down) {
3058                 for (i = 0; i < ports; i++)
3059                         queue_work(ibdev->sriov.demux[i].ud_wq, &dm[i]->work);
3060                 spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
3061         } else {
3062                 spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
3063                 for (i = 0; i < ports; i++)
3064                         kfree(dm[i]);
3065         }
3066 out:
3067         kfree(dm);
3068         return;
3069 }
3070
3071 static void mlx4_ib_handle_catas_error(struct mlx4_ib_dev *ibdev)
3072 {
3073         struct mlx4_ib_qp *mqp;
3074         unsigned long flags_qp;
3075         unsigned long flags_cq;
3076         struct mlx4_ib_cq *send_mcq, *recv_mcq;
3077         struct list_head    cq_notify_list;
3078         struct mlx4_cq *mcq;
3079         unsigned long flags;
3080
3081         pr_warn("mlx4_ib_handle_catas_error was started\n");
3082         INIT_LIST_HEAD(&cq_notify_list);
3083
3084         /* Go over qp list reside on that ibdev, sync with create/destroy qp.*/
3085         spin_lock_irqsave(&ibdev->reset_flow_resource_lock, flags);
3086
3087         list_for_each_entry(mqp, &ibdev->qp_list, qps_list) {
3088                 spin_lock_irqsave(&mqp->sq.lock, flags_qp);
3089                 if (mqp->sq.tail != mqp->sq.head) {
3090                         send_mcq = to_mcq(mqp->ibqp.send_cq);
3091                         spin_lock_irqsave(&send_mcq->lock, flags_cq);
3092                         if (send_mcq->mcq.comp &&
3093                             mqp->ibqp.send_cq->comp_handler) {
3094                                 if (!send_mcq->mcq.reset_notify_added) {
3095                                         send_mcq->mcq.reset_notify_added = 1;
3096                                         list_add_tail(&send_mcq->mcq.reset_notify,
3097                                                       &cq_notify_list);
3098                                 }
3099                         }
3100                         spin_unlock_irqrestore(&send_mcq->lock, flags_cq);
3101                 }
3102                 spin_unlock_irqrestore(&mqp->sq.lock, flags_qp);
3103                 /* Now, handle the QP's receive queue */
3104                 spin_lock_irqsave(&mqp->rq.lock, flags_qp);
3105                 /* no handling is needed for SRQ */
3106                 if (!mqp->ibqp.srq) {
3107                         if (mqp->rq.tail != mqp->rq.head) {
3108                                 recv_mcq = to_mcq(mqp->ibqp.recv_cq);
3109                                 spin_lock_irqsave(&recv_mcq->lock, flags_cq);
3110                                 if (recv_mcq->mcq.comp &&
3111                                     mqp->ibqp.recv_cq->comp_handler) {
3112                                         if (!recv_mcq->mcq.reset_notify_added) {
3113                                                 recv_mcq->mcq.reset_notify_added = 1;
3114                                                 list_add_tail(&recv_mcq->mcq.reset_notify,
3115                                                               &cq_notify_list);
3116                                         }
3117                                 }
3118                                 spin_unlock_irqrestore(&recv_mcq->lock,
3119                                                        flags_cq);
3120                         }
3121                 }
3122                 spin_unlock_irqrestore(&mqp->rq.lock, flags_qp);
3123         }
3124
3125         list_for_each_entry(mcq, &cq_notify_list, reset_notify) {
3126                 mcq->comp(mcq);
3127         }
3128         spin_unlock_irqrestore(&ibdev->reset_flow_resource_lock, flags);
3129         pr_warn("mlx4_ib_handle_catas_error ended\n");
3130 }
3131
3132 static void handle_bonded_port_state_event(struct work_struct *work)
3133 {
3134         struct ib_event_work *ew =
3135                 container_of(work, struct ib_event_work, work);
3136         struct mlx4_ib_dev *ibdev = ew->ib_dev;
3137         enum ib_port_state bonded_port_state = IB_PORT_NOP;
3138         int i;
3139         struct ib_event ibev;
3140
3141         kfree(ew);
3142         spin_lock_bh(&ibdev->iboe.lock);
3143         for (i = 0; i < MLX4_MAX_PORTS; ++i) {
3144                 struct net_device *curr_netdev = ibdev->iboe.netdevs[i];
3145                 enum ib_port_state curr_port_state;
3146
3147                 if (!curr_netdev)
3148                         continue;
3149
3150                 curr_port_state =
3151                         (netif_running(curr_netdev) &&
3152                          netif_carrier_ok(curr_netdev)) ?
3153                         IB_PORT_ACTIVE : IB_PORT_DOWN;
3154
3155                 bonded_port_state = (bonded_port_state != IB_PORT_ACTIVE) ?
3156                         curr_port_state : IB_PORT_ACTIVE;
3157         }
3158         spin_unlock_bh(&ibdev->iboe.lock);
3159
3160         ibev.device = &ibdev->ib_dev;
3161         ibev.element.port_num = 1;
3162         ibev.event = (bonded_port_state == IB_PORT_ACTIVE) ?
3163                 IB_EVENT_PORT_ACTIVE : IB_EVENT_PORT_ERR;
3164
3165         ib_dispatch_event(&ibev);
3166 }
3167
3168 void mlx4_ib_sl2vl_update(struct mlx4_ib_dev *mdev, int port)
3169 {
3170         u64 sl2vl;
3171         int err;
3172
3173         err = mlx4_ib_query_sl2vl(&mdev->ib_dev, port, &sl2vl);
3174         if (err) {
3175                 pr_err("Unable to get current sl to vl mapping for port %d.  Using all zeroes (%d)\n",
3176                        port, err);
3177                 sl2vl = 0;
3178         }
3179         atomic64_set(&mdev->sl2vl[port - 1], sl2vl);
3180 }
3181
3182 static void ib_sl2vl_update_work(struct work_struct *work)
3183 {
3184         struct ib_event_work *ew = container_of(work, struct ib_event_work, work);
3185         struct mlx4_ib_dev *mdev = ew->ib_dev;
3186         int port = ew->port;
3187
3188         mlx4_ib_sl2vl_update(mdev, port);
3189
3190         kfree(ew);
3191 }
3192
3193 void mlx4_sched_ib_sl2vl_update_work(struct mlx4_ib_dev *ibdev,
3194                                      int port)
3195 {
3196         struct ib_event_work *ew;
3197
3198         ew = kmalloc(sizeof(*ew), GFP_ATOMIC);
3199         if (ew) {
3200                 INIT_WORK(&ew->work, ib_sl2vl_update_work);
3201                 ew->port = port;
3202                 ew->ib_dev = ibdev;
3203                 queue_work(wq, &ew->work);
3204         }
3205 }
3206
3207 static void mlx4_ib_event(struct mlx4_dev *dev, void *ibdev_ptr,
3208                           enum mlx4_dev_event event, unsigned long param)
3209 {
3210         struct ib_event ibev;
3211         struct mlx4_ib_dev *ibdev = to_mdev((struct ib_device *) ibdev_ptr);
3212         struct mlx4_eqe *eqe = NULL;
3213         struct ib_event_work *ew;
3214         int p = 0;
3215
3216         if (mlx4_is_bonded(dev) &&
3217             ((event == MLX4_DEV_EVENT_PORT_UP) ||
3218             (event == MLX4_DEV_EVENT_PORT_DOWN))) {
3219                 ew = kmalloc(sizeof(*ew), GFP_ATOMIC);
3220                 if (!ew)
3221                         return;
3222                 INIT_WORK(&ew->work, handle_bonded_port_state_event);
3223                 ew->ib_dev = ibdev;
3224                 queue_work(wq, &ew->work);
3225                 return;
3226         }
3227
3228         if (event == MLX4_DEV_EVENT_PORT_MGMT_CHANGE)
3229                 eqe = (struct mlx4_eqe *)param;
3230         else
3231                 p = (int) param;
3232
3233         switch (event) {
3234         case MLX4_DEV_EVENT_PORT_UP:
3235                 if (p > ibdev->num_ports)
3236                         return;
3237                 if (!mlx4_is_slave(dev) &&
3238                     rdma_port_get_link_layer(&ibdev->ib_dev, p) ==
3239                         IB_LINK_LAYER_INFINIBAND) {
3240                         if (mlx4_is_master(dev))
3241                                 mlx4_ib_invalidate_all_guid_record(ibdev, p);
3242                         if (ibdev->dev->flags & MLX4_FLAG_SECURE_HOST &&
3243                             !(ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_SL_TO_VL_CHANGE_EVENT))
3244                                 mlx4_sched_ib_sl2vl_update_work(ibdev, p);
3245                 }
3246                 ibev.event = IB_EVENT_PORT_ACTIVE;
3247                 break;
3248
3249         case MLX4_DEV_EVENT_PORT_DOWN:
3250                 if (p > ibdev->num_ports)
3251                         return;
3252                 ibev.event = IB_EVENT_PORT_ERR;
3253                 break;
3254
3255         case MLX4_DEV_EVENT_CATASTROPHIC_ERROR:
3256                 ibdev->ib_active = false;
3257                 ibev.event = IB_EVENT_DEVICE_FATAL;
3258                 mlx4_ib_handle_catas_error(ibdev);
3259                 break;
3260
3261         case MLX4_DEV_EVENT_PORT_MGMT_CHANGE:
3262                 ew = kmalloc(sizeof *ew, GFP_ATOMIC);
3263                 if (!ew)
3264                         return;
3265
3266                 INIT_WORK(&ew->work, handle_port_mgmt_change_event);
3267                 memcpy(&ew->ib_eqe, eqe, sizeof *eqe);
3268                 ew->ib_dev = ibdev;
3269                 /* need to queue only for port owner, which uses GEN_EQE */
3270                 if (mlx4_is_master(dev))
3271                         queue_work(wq, &ew->work);
3272                 else
3273                         handle_port_mgmt_change_event(&ew->work);
3274                 return;
3275
3276         case MLX4_DEV_EVENT_SLAVE_INIT:
3277                 /* here, p is the slave id */
3278                 do_slave_init(ibdev, p, 1);
3279                 if (mlx4_is_master(dev)) {
3280                         int i;
3281
3282                         for (i = 1; i <= ibdev->num_ports; i++) {
3283                                 if (rdma_port_get_link_layer(&ibdev->ib_dev, i)
3284                                         == IB_LINK_LAYER_INFINIBAND)
3285                                         mlx4_ib_slave_alias_guid_event(ibdev,
3286                                                                        p, i,
3287                                                                        1);
3288                         }
3289                 }
3290                 return;
3291
3292         case MLX4_DEV_EVENT_SLAVE_SHUTDOWN:
3293                 if (mlx4_is_master(dev)) {
3294                         int i;
3295
3296                         for (i = 1; i <= ibdev->num_ports; i++) {
3297                                 if (rdma_port_get_link_layer(&ibdev->ib_dev, i)
3298                                         == IB_LINK_LAYER_INFINIBAND)
3299                                         mlx4_ib_slave_alias_guid_event(ibdev,
3300                                                                        p, i,
3301                                                                        0);
3302                         }
3303                 }
3304                 /* here, p is the slave id */
3305                 do_slave_init(ibdev, p, 0);
3306                 return;
3307
3308         default:
3309                 return;
3310         }
3311
3312         ibev.device           = ibdev_ptr;
3313         ibev.element.port_num = mlx4_is_bonded(ibdev->dev) ? 1 : (u8)p;
3314
3315         ib_dispatch_event(&ibev);
3316 }
3317
3318 static struct mlx4_interface mlx4_ib_interface = {
3319         .add            = mlx4_ib_add,
3320         .remove         = mlx4_ib_remove,
3321         .event          = mlx4_ib_event,
3322         .protocol       = MLX4_PROT_IB_IPV6,
3323         .flags          = MLX4_INTFF_BONDING
3324 };
3325
3326 static int __init mlx4_ib_init(void)
3327 {
3328         int err;
3329
3330         wq = alloc_ordered_workqueue("mlx4_ib", WQ_MEM_RECLAIM);
3331         if (!wq)
3332                 return -ENOMEM;
3333
3334         err = mlx4_ib_mcg_init();
3335         if (err)
3336                 goto clean_wq;
3337
3338         err = mlx4_register_interface(&mlx4_ib_interface);
3339         if (err)
3340                 goto clean_mcg;
3341
3342         return 0;
3343
3344 clean_mcg:
3345         mlx4_ib_mcg_destroy();
3346
3347 clean_wq:
3348         destroy_workqueue(wq);
3349         return err;
3350 }
3351
3352 static void __exit mlx4_ib_cleanup(void)
3353 {
3354         mlx4_unregister_interface(&mlx4_ib_interface);
3355         mlx4_ib_mcg_destroy();
3356         destroy_workqueue(wq);
3357 }
3358
3359 module_init(mlx4_ib_init);
3360 module_exit(mlx4_ib_cleanup);