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                                                     u32 port_num);
  85
  86 static struct workqueue_struct *wq;
  87
  88 static int check_flow_steering_support(struct mlx4_dev *dev)
  89 {
  90         int eth_num_ports = 0;
  91         int ib_num_ports = 0;
  92
  93         int dmfs = dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED;
  94
  95         if (dmfs) {
  96                 int i;
  97                 mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_ETH)
  98                         eth_num_ports++;
  99                 mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
 100                         ib_num_ports++;
 101                 dmfs &= (!ib_num_ports ||
 102                          (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DMFS_IPOIB)) &&
 103                         (!eth_num_ports ||
 104                          (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_FS_EN));
 105                 if (ib_num_ports && mlx4_is_mfunc(dev)) {
 106                         pr_warn("Device managed flow steering is unavailable for IB port in multifunction env.\n");
 107                         dmfs = 0;
 108                 }
 109         }
 110         return dmfs;
 111 }
 112
 113 static int num_ib_ports(struct mlx4_dev *dev)
 114 {
 115         int ib_ports = 0;
 116         int i;
 117
 118         mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
 119                 ib_ports++;
 120
 121         return ib_ports;
 122 }
 123
 124 static struct net_device *mlx4_ib_get_netdev(struct ib_device *device,
 125                                              u32 port_num)
 126 {
 127         struct mlx4_ib_dev *ibdev = to_mdev(device);
 128         struct net_device *dev;
 129
 130         rcu_read_lock();
 131         dev = mlx4_get_protocol_dev(ibdev->dev, MLX4_PROT_ETH, port_num);
 132
 133         if (dev) {
 134                 if (mlx4_is_bonded(ibdev->dev)) {
 135                         struct net_device *upper;
 136
 137                         upper = netdev_master_upper_dev_get_rcu(dev);
 138                         if (upper) {
 139                                 struct net_device *active;
 140
 141                                 active = bond_option_active_slave_get_rcu(netdev_priv(upper));
 142                                 if (active)
 143                                         dev = active;
 144                         }
 145                 }
 146         }
 147         dev_hold(dev);
 148
 149         rcu_read_unlock();
 150         return dev;
 151 }
 152
 153 static int mlx4_ib_update_gids_v1(struct gid_entry *gids,
 154                                   struct mlx4_ib_dev *ibdev,
 155                                   u32 port_num)
 156 {
 157         struct mlx4_cmd_mailbox *mailbox;
 158         int err;
 159         struct mlx4_dev *dev = ibdev->dev;
 160         int i;
 161         union ib_gid *gid_tbl;
 162
 163         mailbox = mlx4_alloc_cmd_mailbox(dev);
 164         if (IS_ERR(mailbox))
 165                 return -ENOMEM;
 166
 167         gid_tbl = mailbox->buf;
 168
 169         for (i = 0; i < MLX4_MAX_PORT_GIDS; ++i)
 170                 memcpy(&gid_tbl[i], &gids[i].gid, sizeof(union ib_gid));
 171
 172         err = mlx4_cmd(dev, mailbox->dma,
 173                        MLX4_SET_PORT_GID_TABLE << 8 | port_num,
 174                        1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
 175                        MLX4_CMD_WRAPPED);
 176         if (mlx4_is_bonded(dev))
 177                 err += mlx4_cmd(dev, mailbox->dma,
 178                                 MLX4_SET_PORT_GID_TABLE << 8 | 2,
 179                                 1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
 180                                 MLX4_CMD_WRAPPED);
 181
 182         mlx4_free_cmd_mailbox(dev, mailbox);
 183         return err;
 184 }
 185
 186 static int mlx4_ib_update_gids_v1_v2(struct gid_entry *gids,
 187                                      struct mlx4_ib_dev *ibdev,
 188                                      u32 port_num)
 189 {
 190         struct mlx4_cmd_mailbox *mailbox;
 191         int err;
 192         struct mlx4_dev *dev = ibdev->dev;
 193         int i;
 194         struct {
 195                 union ib_gid    gid;
 196                 __be32          rsrvd1[2];
 197                 __be16          rsrvd2;
 198                 u8              type;
 199                 u8              version;
 200                 __be32          rsrvd3;
 201         } *gid_tbl;
 202
 203         mailbox = mlx4_alloc_cmd_mailbox(dev);
 204         if (IS_ERR(mailbox))
 205                 return -ENOMEM;
 206
 207         gid_tbl = mailbox->buf;
 208         for (i = 0; i < MLX4_MAX_PORT_GIDS; ++i) {
 209                 memcpy(&gid_tbl[i].gid, &gids[i].gid, sizeof(union ib_gid));
 210                 if (gids[i].gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP) {
 211                         gid_tbl[i].version = 2;
 212                         if (!ipv6_addr_v4mapped((struct in6_addr *)&gids[i].gid))
 213                                 gid_tbl[i].type = 1;
 214                 }
 215         }
 216
 217         err = mlx4_cmd(dev, mailbox->dma,
 218                        MLX4_SET_PORT_ROCE_ADDR << 8 | port_num,
 219                        1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
 220                        MLX4_CMD_WRAPPED);
 221         if (mlx4_is_bonded(dev))
 222                 err += mlx4_cmd(dev, mailbox->dma,
 223                                 MLX4_SET_PORT_ROCE_ADDR << 8 | 2,
 224                                 1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
 225                                 MLX4_CMD_WRAPPED);
 226
 227         mlx4_free_cmd_mailbox(dev, mailbox);
 228         return err;
 229 }
 230
 231 static int mlx4_ib_update_gids(struct gid_entry *gids,
 232                                struct mlx4_ib_dev *ibdev,
 233                                u32 port_num)
 234 {
 235         if (ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ROCE_V1_V2)
 236                 return mlx4_ib_update_gids_v1_v2(gids, ibdev, port_num);
 237
 238         return mlx4_ib_update_gids_v1(gids, ibdev, port_num);
 239 }
 240
 241 static void free_gid_entry(struct gid_entry *entry)
 242 {
 243         memset(&entry->gid, 0, sizeof(entry->gid));
 244         kfree(entry->ctx);
 245         entry->ctx = NULL;
 246 }
 247
 248 static int mlx4_ib_add_gid(const struct ib_gid_attr *attr, void **context)
 249 {
 250         struct mlx4_ib_dev *ibdev = to_mdev(attr->device);
 251         struct mlx4_ib_iboe *iboe = &ibdev->iboe;
 252         struct mlx4_port_gid_table   *port_gid_table;
 253         int free = -1, found = -1;
 254         int ret = 0;
 255         int hw_update = 0;
 256         int i;
 257         struct gid_entry *gids;
 258         u16 vlan_id = 0xffff;
 259         u8 mac[ETH_ALEN];
 260
 261         if (!rdma_cap_roce_gid_table(attr->device, attr->port_num))
 262                 return -EINVAL;
 263
 264         if (attr->port_num > MLX4_MAX_PORTS)
 265                 return -EINVAL;
 266
 267         if (!context)
 268                 return -EINVAL;
 269
 270         ret = rdma_read_gid_l2_fields(attr, &vlan_id, &mac[0]);
 271         if (ret)
 272                 return ret;
 273         port_gid_table = &iboe->gids[attr->port_num - 1];
 274         spin_lock_bh(&iboe->lock);
 275         for (i = 0; i < MLX4_MAX_PORT_GIDS; ++i) {
 276                 if (!memcmp(&port_gid_table->gids[i].gid,
 277                             &attr->gid, sizeof(attr->gid)) &&
 278                     port_gid_table->gids[i].gid_type == attr->gid_type &&
 279                     port_gid_table->gids[i].vlan_id == vlan_id)  {
 280                         found = i;
 281                         break;
 282                 }
 283                 if (free < 0 && rdma_is_zero_gid(&port_gid_table->gids[i].gid))
 284                         free = i; /* HW has space */
 285         }
 286
 287         if (found < 0) {
 288                 if (free < 0) {
 289                         ret = -ENOSPC;
 290                 } else {
 291                         port_gid_table->gids[free].ctx = kmalloc(sizeof(*port_gid_table->gids[free].ctx), GFP_ATOMIC);
 292                         if (!port_gid_table->gids[free].ctx) {
 293                                 ret = -ENOMEM;
 294                         } else {
 295                                 *context = port_gid_table->gids[free].ctx;
 296                                 port_gid_table->gids[free].gid = attr->gid;
 297                                 port_gid_table->gids[free].gid_type = attr->gid_type;
 298                                 port_gid_table->gids[free].vlan_id = vlan_id;
 299                                 port_gid_table->gids[free].ctx->real_index = free;
 300                                 port_gid_table->gids[free].ctx->refcount = 1;
 301                                 hw_update = 1;
 302                         }
 303                 }
 304         } else {
 305                 struct gid_cache_context *ctx = port_gid_table->gids[found].ctx;
 306                 *context = ctx;
 307                 ctx->refcount++;
 308         }
 309         if (!ret && hw_update) {
 310                 gids = kmalloc_array(MLX4_MAX_PORT_GIDS, sizeof(*gids),
 311                                      GFP_ATOMIC);
 312                 if (!gids) {
 313                         ret = -ENOMEM;
 314                         *context = NULL;
 315                         free_gid_entry(&port_gid_table->gids[free]);
 316                 } else {
 317                         for (i = 0; i < MLX4_MAX_PORT_GIDS; i++) {
 318                                 memcpy(&gids[i].gid, &port_gid_table->gids[i].gid, sizeof(union ib_gid));
 319                                 gids[i].gid_type = port_gid_table->gids[i].gid_type;
 320                         }
 321                 }
 322         }
 323         spin_unlock_bh(&iboe->lock);
 324
 325         if (!ret && hw_update) {
 326                 ret = mlx4_ib_update_gids(gids, ibdev, attr->port_num);
 327                 if (ret) {
 328                         spin_lock_bh(&iboe->lock);
 329                         *context = NULL;
 330                         free_gid_entry(&port_gid_table->gids[free]);
 331                         spin_unlock_bh(&iboe->lock);
 332                 }
 333                 kfree(gids);
 334         }
 335
 336         return ret;
 337 }
 338
 339 static int mlx4_ib_del_gid(const struct ib_gid_attr *attr, void **context)
 340 {
 341         struct gid_cache_context *ctx = *context;
 342         struct mlx4_ib_dev *ibdev = to_mdev(attr->device);
 343         struct mlx4_ib_iboe *iboe = &ibdev->iboe;
 344         struct mlx4_port_gid_table   *port_gid_table;
 345         int ret = 0;
 346         int hw_update = 0;
 347         struct gid_entry *gids;
 348
 349         if (!rdma_cap_roce_gid_table(attr->device, attr->port_num))
 350                 return -EINVAL;
 351
 352         if (attr->port_num > MLX4_MAX_PORTS)
 353                 return -EINVAL;
 354
 355         port_gid_table = &iboe->gids[attr->port_num - 1];
 356         spin_lock_bh(&iboe->lock);
 357         if (ctx) {
 358                 ctx->refcount--;
 359                 if (!ctx->refcount) {
 360                         unsigned int real_index = ctx->real_index;
 361
 362                         free_gid_entry(&port_gid_table->gids[real_index]);
 363                         hw_update = 1;
 364                 }
 365         }
 366         if (!ret && hw_update) {
 367                 int i;
 368
 369                 gids = kmalloc_array(MLX4_MAX_PORT_GIDS, sizeof(*gids),
 370                                      GFP_ATOMIC);
 371                 if (!gids) {
 372                         ret = -ENOMEM;
 373                 } else {
 374                         for (i = 0; i < MLX4_MAX_PORT_GIDS; i++) {
 375                                 memcpy(&gids[i].gid,
 376                                        &port_gid_table->gids[i].gid,
 377                                        sizeof(union ib_gid));
 378                                 gids[i].gid_type =
 379                                     port_gid_table->gids[i].gid_type;
 380                         }
 381                 }
 382         }
 383         spin_unlock_bh(&iboe->lock);
 384
 385         if (!ret && hw_update) {
 386                 ret = mlx4_ib_update_gids(gids, ibdev, attr->port_num);
 387                 kfree(gids);
 388         }
 389         return ret;
 390 }
 391
 392 int mlx4_ib_gid_index_to_real_index(struct mlx4_ib_dev *ibdev,
 393                                     const struct ib_gid_attr *attr)
 394 {
 395         struct mlx4_ib_iboe *iboe = &ibdev->iboe;
 396         struct gid_cache_context *ctx = NULL;
 397         struct mlx4_port_gid_table   *port_gid_table;
 398         int real_index = -EINVAL;
 399         int i;
 400         unsigned long flags;
 401         u32 port_num = attr->port_num;
 402
 403         if (port_num > MLX4_MAX_PORTS)
 404                 return -EINVAL;
 405
 406         if (mlx4_is_bonded(ibdev->dev))
 407                 port_num = 1;
 408
 409         if (!rdma_cap_roce_gid_table(&ibdev->ib_dev, port_num))
 410                 return attr->index;
 411
 412         spin_lock_irqsave(&iboe->lock, flags);
 413         port_gid_table = &iboe->gids[port_num - 1];
 414
 415         for (i = 0; i < MLX4_MAX_PORT_GIDS; ++i)
 416                 if (!memcmp(&port_gid_table->gids[i].gid,
 417                             &attr->gid, sizeof(attr->gid)) &&
 418                     attr->gid_type == port_gid_table->gids[i].gid_type) {
 419                         ctx = port_gid_table->gids[i].ctx;
 420                         break;
 421                 }
 422         if (ctx)
 423                 real_index = ctx->real_index;
 424         spin_unlock_irqrestore(&iboe->lock, flags);
 425         return real_index;
 426 }
 427
 428 static int mlx4_ib_query_device(struct ib_device *ibdev,
 429                                 struct ib_device_attr *props,
 430                                 struct ib_udata *uhw)
 431 {
 432         struct mlx4_ib_dev *dev = to_mdev(ibdev);
 433         struct ib_smp *in_mad;
 434         struct ib_smp *out_mad;
 435         int err;
 436         int have_ib_ports;
 437         struct mlx4_uverbs_ex_query_device cmd;
 438         struct mlx4_uverbs_ex_query_device_resp resp = {};
 439         struct mlx4_clock_params clock_params;
 440
 441         if (uhw->inlen) {
 442                 if (uhw->inlen < sizeof(cmd))
 443                         return -EINVAL;
 444
 445                 err = ib_copy_from_udata(&cmd, uhw, sizeof(cmd));
 446                 if (err)
 447                         return err;
 448
 449                 if (cmd.comp_mask)
 450                         return -EINVAL;
 451
 452                 if (cmd.reserved)
 453                         return -EINVAL;
 454         }
 455
 456         resp.response_length = offsetof(typeof(resp), response_length) +
 457                 sizeof(resp.response_length);
 458         in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
 459         out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
 460         err = -ENOMEM;
 461         if (!in_mad || !out_mad)
 462                 goto out;
 463
 464         ib_init_query_mad(in_mad);
 465         in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
 466
 467         err = mlx4_MAD_IFC(to_mdev(ibdev), MLX4_MAD_IFC_IGNORE_KEYS,
 468                            1, NULL, NULL, in_mad, out_mad);
 469         if (err)
 470                 goto out;
 471
 472         memset(props, 0, sizeof *props);
 473
 474         have_ib_ports = num_ib_ports(dev->dev);
 475
 476         props->fw_ver = dev->dev->caps.fw_ver;
 477         props->device_cap_flags    = IB_DEVICE_CHANGE_PHY_PORT |
 478                 IB_DEVICE_PORT_ACTIVE_EVENT             |
 479                 IB_DEVICE_SYS_IMAGE_GUID                |
 480                 IB_DEVICE_RC_RNR_NAK_GEN;
 481         props->kernel_cap_flags = IBK_BLOCK_MULTICAST_LOOPBACK;
 482         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_PKEY_CNTR)
 483                 props->device_cap_flags |= IB_DEVICE_BAD_PKEY_CNTR;
 484         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_QKEY_CNTR)
 485                 props->device_cap_flags |= IB_DEVICE_BAD_QKEY_CNTR;
 486         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_APM && have_ib_ports)
 487                 props->device_cap_flags |= IB_DEVICE_AUTO_PATH_MIG;
 488         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_UD_AV_PORT)
 489                 props->device_cap_flags |= IB_DEVICE_UD_AV_PORT_ENFORCE;
 490         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_IPOIB_CSUM)
 491                 props->device_cap_flags |= IB_DEVICE_UD_IP_CSUM;
 492         if (dev->dev->caps.max_gso_sz &&
 493             (dev->dev->rev_id != MLX4_IB_CARD_REV_A0) &&
 494             (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BLH))
 495                 props->kernel_cap_flags |= IBK_UD_TSO;
 496         if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_RESERVED_LKEY)
 497                 props->kernel_cap_flags |= IBK_LOCAL_DMA_LKEY;
 498         if ((dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_LOCAL_INV) &&
 499             (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_REMOTE_INV) &&
 500             (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_FAST_REG_WR))
 501                 props->device_cap_flags |= IB_DEVICE_MEM_MGT_EXTENSIONS;
 502         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC)
 503                 props->device_cap_flags |= IB_DEVICE_XRC;
 504         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW)
 505                 props->device_cap_flags |= IB_DEVICE_MEM_WINDOW;
 506         if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
 507                 if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_WIN_TYPE_2B)
 508                         props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2B;
 509                 else
 510                         props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2A;
 511         }
 512         if (dev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED)
 513                 props->device_cap_flags |= IB_DEVICE_MANAGED_FLOW_STEERING;
 514
 515         props->device_cap_flags |= IB_DEVICE_RAW_IP_CSUM;
 516
 517         props->vendor_id           = be32_to_cpup((__be32 *) (out_mad->data + 36)) &
 518                 0xffffff;
 519         props->vendor_part_id      = dev->dev->persist->pdev->device;
 520         props->hw_ver              = be32_to_cpup((__be32 *) (out_mad->data + 32));
 521         memcpy(&props->sys_image_guid, out_mad->data +  4, 8);
 522
 523         props->max_mr_size         = ~0ull;
 524         props->page_size_cap       = dev->dev->caps.page_size_cap;
 525         props->max_qp              = dev->dev->quotas.qp;
 526         props->max_qp_wr           = dev->dev->caps.max_wqes - MLX4_IB_SQ_MAX_SPARE;
 527         props->max_send_sge =
 528                 min(dev->dev->caps.max_sq_sg, dev->dev->caps.max_rq_sg);
 529         props->max_recv_sge =
 530                 min(dev->dev->caps.max_sq_sg, dev->dev->caps.max_rq_sg);
 531         props->max_sge_rd = MLX4_MAX_SGE_RD;
 532         props->max_cq              = dev->dev->quotas.cq;
 533         props->max_cqe             = dev->dev->caps.max_cqes;
 534         props->max_mr              = dev->dev->quotas.mpt;
 535         props->max_pd              = dev->dev->caps.num_pds - dev->dev->caps.reserved_pds;
 536         props->max_qp_rd_atom      = dev->dev->caps.max_qp_dest_rdma;
 537         props->max_qp_init_rd_atom = dev->dev->caps.max_qp_init_rdma;
 538         props->max_res_rd_atom     = props->max_qp_rd_atom * props->max_qp;
 539         props->max_srq             = dev->dev->quotas.srq;
 540         props->max_srq_wr          = dev->dev->caps.max_srq_wqes - 1;
 541         props->max_srq_sge         = dev->dev->caps.max_srq_sge;
 542         props->max_fast_reg_page_list_len = MLX4_MAX_FAST_REG_PAGES;
 543         props->local_ca_ack_delay  = dev->dev->caps.local_ca_ack_delay;
 544         props->atomic_cap          = dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_ATOMIC ?
 545                 IB_ATOMIC_HCA : IB_ATOMIC_NONE;
 546         props->masked_atomic_cap   = props->atomic_cap;
 547         props->max_pkeys           = dev->dev->caps.pkey_table_len[1];
 548         props->max_mcast_grp       = dev->dev->caps.num_mgms + dev->dev->caps.num_amgms;
 549         props->max_mcast_qp_attach = dev->dev->caps.num_qp_per_mgm;
 550         props->max_total_mcast_qp_attach = props->max_mcast_qp_attach *
 551                                            props->max_mcast_grp;
 552         props->hca_core_clock = dev->dev->caps.hca_core_clock * 1000UL;
 553         props->timestamp_mask = 0xFFFFFFFFFFFFULL;
 554         props->max_ah = INT_MAX;
 555
 556         if (mlx4_ib_port_link_layer(ibdev, 1) == IB_LINK_LAYER_ETHERNET ||
 557             mlx4_ib_port_link_layer(ibdev, 2) == IB_LINK_LAYER_ETHERNET) {
 558                 if (dev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_RSS) {
 559                         props->rss_caps.max_rwq_indirection_tables =
 560                                 props->max_qp;
 561                         props->rss_caps.max_rwq_indirection_table_size =
 562                                 dev->dev->caps.max_rss_tbl_sz;
 563                         props->rss_caps.supported_qpts = 1 << IB_QPT_RAW_PACKET;
 564                         props->max_wq_type_rq = props->max_qp;
 565                 }
 566
 567                 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_FCS_KEEP)
 568                         props->raw_packet_caps |= IB_RAW_PACKET_CAP_SCATTER_FCS;
 569         }
 570
 571         props->cq_caps.max_cq_moderation_count = MLX4_MAX_CQ_COUNT;
 572         props->cq_caps.max_cq_moderation_period = MLX4_MAX_CQ_PERIOD;
 573
 574         if (uhw->outlen >= resp.response_length + sizeof(resp.hca_core_clock_offset)) {
 575                 resp.response_length += sizeof(resp.hca_core_clock_offset);
 576                 if (!mlx4_get_internal_clock_params(dev->dev, &clock_params)) {
 577                         resp.comp_mask |= MLX4_IB_QUERY_DEV_RESP_MASK_CORE_CLOCK_OFFSET;
 578                         resp.hca_core_clock_offset = clock_params.offset % PAGE_SIZE;
 579                 }
 580         }
 581
 582         if (uhw->outlen >= resp.response_length +
 583             sizeof(resp.max_inl_recv_sz)) {
 584                 resp.response_length += sizeof(resp.max_inl_recv_sz);
 585                 resp.max_inl_recv_sz  = dev->dev->caps.max_rq_sg *
 586                         sizeof(struct mlx4_wqe_data_seg);
 587         }
 588
 589         if (offsetofend(typeof(resp), rss_caps) <= uhw->outlen) {
 590                 if (props->rss_caps.supported_qpts) {
 591                         resp.rss_caps.rx_hash_function =
 592                                 MLX4_IB_RX_HASH_FUNC_TOEPLITZ;
 593
 594                         resp.rss_caps.rx_hash_fields_mask =
 595                                 MLX4_IB_RX_HASH_SRC_IPV4 |
 596                                 MLX4_IB_RX_HASH_DST_IPV4 |
 597                                 MLX4_IB_RX_HASH_SRC_IPV6 |
 598                                 MLX4_IB_RX_HASH_DST_IPV6 |
 599                                 MLX4_IB_RX_HASH_SRC_PORT_TCP |
 600                                 MLX4_IB_RX_HASH_DST_PORT_TCP |
 601                                 MLX4_IB_RX_HASH_SRC_PORT_UDP |
 602                                 MLX4_IB_RX_HASH_DST_PORT_UDP;
 603
 604                         if (dev->dev->caps.tunnel_offload_mode ==
 605                             MLX4_TUNNEL_OFFLOAD_MODE_VXLAN)
 606                                 resp.rss_caps.rx_hash_fields_mask |=
 607                                         MLX4_IB_RX_HASH_INNER;
 608                 }
 609                 resp.response_length = offsetof(typeof(resp), rss_caps) +
 610                                        sizeof(resp.rss_caps);
 611         }
 612
 613         if (offsetofend(typeof(resp), tso_caps) <= uhw->outlen) {
 614                 if (dev->dev->caps.max_gso_sz &&
 615                     ((mlx4_ib_port_link_layer(ibdev, 1) ==
 616                     IB_LINK_LAYER_ETHERNET) ||
 617                     (mlx4_ib_port_link_layer(ibdev, 2) ==
 618                     IB_LINK_LAYER_ETHERNET))) {
 619                         resp.tso_caps.max_tso = dev->dev->caps.max_gso_sz;
 620                         resp.tso_caps.supported_qpts |=
 621                                 1 << IB_QPT_RAW_PACKET;
 622                 }
 623                 resp.response_length = offsetof(typeof(resp), tso_caps) +
 624                                        sizeof(resp.tso_caps);
 625         }
 626
 627         if (uhw->outlen) {
 628                 err = ib_copy_to_udata(uhw, &resp, resp.response_length);
 629                 if (err)
 630                         goto out;
 631         }
 632 out:
 633         kfree(in_mad);
 634         kfree(out_mad);
 635
 636         return err;
 637 }
 638
 639 static enum rdma_link_layer
 640 mlx4_ib_port_link_layer(struct ib_device *device, u32 port_num)
 641 {
 642         struct mlx4_dev *dev = to_mdev(device)->dev;
 643
 644         return dev->caps.port_mask[port_num] == MLX4_PORT_TYPE_IB ?
 645                 IB_LINK_LAYER_INFINIBAND : IB_LINK_LAYER_ETHERNET;
 646 }
 647
 648 static int ib_link_query_port(struct ib_device *ibdev, u32 port,
 649                               struct ib_port_attr *props, int netw_view)
 650 {
 651         struct ib_smp *in_mad;
 652         struct ib_smp *out_mad;
 653         int ext_active_speed;
 654         int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
 655         int err = -ENOMEM;
 656
 657         in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
 658         out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
 659         if (!in_mad || !out_mad)
 660                 goto out;
 661
 662         ib_init_query_mad(in_mad);
 663         in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
 664         in_mad->attr_mod = cpu_to_be32(port);
 665
 666         if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
 667                 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
 668
 669         err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
 670                                 in_mad, out_mad);
 671         if (err)
 672                 goto out;
 673
 674
 675         props->lid              = be16_to_cpup((__be16 *) (out_mad->data + 16));
 676         props->lmc              = out_mad->data[34] & 0x7;
 677         props->sm_lid           = be16_to_cpup((__be16 *) (out_mad->data + 18));
 678         props->sm_sl            = out_mad->data[36] & 0xf;
 679         props->state            = out_mad->data[32] & 0xf;
 680         props->phys_state       = out_mad->data[33] >> 4;
 681         props->port_cap_flags   = be32_to_cpup((__be32 *) (out_mad->data + 20));
 682         if (netw_view)
 683                 props->gid_tbl_len = out_mad->data[50];
 684         else
 685                 props->gid_tbl_len = to_mdev(ibdev)->dev->caps.gid_table_len[port];
 686         props->max_msg_sz       = to_mdev(ibdev)->dev->caps.max_msg_sz;
 687         props->pkey_tbl_len     = to_mdev(ibdev)->dev->caps.pkey_table_len[port];
 688         props->bad_pkey_cntr    = be16_to_cpup((__be16 *) (out_mad->data + 46));
 689         props->qkey_viol_cntr   = be16_to_cpup((__be16 *) (out_mad->data + 48));
 690         props->active_width     = out_mad->data[31] & 0xf;
 691         props->active_speed     = out_mad->data[35] >> 4;
 692         props->max_mtu          = out_mad->data[41] & 0xf;
 693         props->active_mtu       = out_mad->data[36] >> 4;
 694         props->subnet_timeout   = out_mad->data[51] & 0x1f;
 695         props->max_vl_num       = out_mad->data[37] >> 4;
 696         props->init_type_reply  = out_mad->data[41] >> 4;
 697
 698         /* Check if extended speeds (EDR/FDR/...) are supported */
 699         if (props->port_cap_flags & IB_PORT_EXTENDED_SPEEDS_SUP) {
 700                 ext_active_speed = out_mad->data[62] >> 4;
 701
 702                 switch (ext_active_speed) {
 703                 case 1:
 704                         props->active_speed = IB_SPEED_FDR;
 705                         break;
 706                 case 2:
 707                         props->active_speed = IB_SPEED_EDR;
 708                         break;
 709                 }
 710         }
 711
 712         /* If reported active speed is QDR, check if is FDR-10 */
 713         if (props->active_speed == IB_SPEED_QDR) {
 714                 ib_init_query_mad(in_mad);
 715                 in_mad->attr_id = MLX4_ATTR_EXTENDED_PORT_INFO;
 716                 in_mad->attr_mod = cpu_to_be32(port);
 717
 718                 err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port,
 719                                    NULL, NULL, in_mad, out_mad);
 720                 if (err)
 721                         goto out;
 722
 723                 /* Checking LinkSpeedActive for FDR-10 */
 724                 if (out_mad->data[15] & 0x1)
 725                         props->active_speed = IB_SPEED_FDR10;
 726         }
 727
 728         /* Avoid wrong speed value returned by FW if the IB link is down. */
 729         if (props->state == IB_PORT_DOWN)
 730                  props->active_speed = IB_SPEED_SDR;
 731
 732 out:
 733         kfree(in_mad);
 734         kfree(out_mad);
 735         return err;
 736 }
 737
 738 static u8 state_to_phys_state(enum ib_port_state state)
 739 {
 740         return state == IB_PORT_ACTIVE ?
 741                 IB_PORT_PHYS_STATE_LINK_UP : IB_PORT_PHYS_STATE_DISABLED;
 742 }
 743
 744 static int eth_link_query_port(struct ib_device *ibdev, u32 port,
 745                                struct ib_port_attr *props)
 746 {
 747
 748         struct mlx4_ib_dev *mdev = to_mdev(ibdev);
 749         struct mlx4_ib_iboe *iboe = &mdev->iboe;
 750         struct net_device *ndev;
 751         enum ib_mtu tmp;
 752         struct mlx4_cmd_mailbox *mailbox;
 753         int err = 0;
 754         int is_bonded = mlx4_is_bonded(mdev->dev);
 755
 756         mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
 757         if (IS_ERR(mailbox))
 758                 return PTR_ERR(mailbox);
 759
 760         err = mlx4_cmd_box(mdev->dev, 0, mailbox->dma, port, 0,
 761                            MLX4_CMD_QUERY_PORT, MLX4_CMD_TIME_CLASS_B,
 762                            MLX4_CMD_WRAPPED);
 763         if (err)
 764                 goto out;
 765
 766         props->active_width     =  (((u8 *)mailbox->buf)[5] == 0x40) ||
 767                                    (((u8 *)mailbox->buf)[5] == 0x20 /*56Gb*/) ?
 768                                            IB_WIDTH_4X : IB_WIDTH_1X;
 769         props->active_speed     =  (((u8 *)mailbox->buf)[5] == 0x20 /*56Gb*/) ?
 770                                            IB_SPEED_FDR : IB_SPEED_QDR;
 771         props->port_cap_flags   = IB_PORT_CM_SUP;
 772         props->ip_gids = true;
 773         props->gid_tbl_len      = mdev->dev->caps.gid_table_len[port];
 774         props->max_msg_sz       = mdev->dev->caps.max_msg_sz;
 775         if (mdev->dev->caps.pkey_table_len[port])
 776                 props->pkey_tbl_len = 1;
 777         props->max_mtu          = IB_MTU_4096;
 778         props->max_vl_num       = 2;
 779         props->state            = IB_PORT_DOWN;
 780         props->phys_state       = state_to_phys_state(props->state);
 781         props->active_mtu       = IB_MTU_256;
 782         spin_lock_bh(&iboe->lock);
 783         ndev = iboe->netdevs[port - 1];
 784         if (ndev && is_bonded) {
 785                 rcu_read_lock(); /* required to get upper dev */
 786                 ndev = netdev_master_upper_dev_get_rcu(ndev);
 787                 rcu_read_unlock();
 788         }
 789         if (!ndev)
 790                 goto out_unlock;
 791
 792         tmp = iboe_get_mtu(ndev->mtu);
 793         props->active_mtu = tmp ? min(props->max_mtu, tmp) : IB_MTU_256;
 794
 795         props->state            = (netif_running(ndev) && netif_carrier_ok(ndev)) ?
 796                                         IB_PORT_ACTIVE : IB_PORT_DOWN;
 797         props->phys_state       = state_to_phys_state(props->state);
 798 out_unlock:
 799         spin_unlock_bh(&iboe->lock);
 800 out:
 801         mlx4_free_cmd_mailbox(mdev->dev, mailbox);
 802         return err;
 803 }
 804
 805 int __mlx4_ib_query_port(struct ib_device *ibdev, u32 port,
 806                          struct ib_port_attr *props, int netw_view)
 807 {
 808         int err;
 809
 810         /* props being zeroed by the caller, avoid zeroing it here */
 811
 812         err = mlx4_ib_port_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND ?
 813                 ib_link_query_port(ibdev, port, props, netw_view) :
 814                                 eth_link_query_port(ibdev, port, props);
 815
 816         return err;
 817 }
 818
 819 static int mlx4_ib_query_port(struct ib_device *ibdev, u32 port,
 820                               struct ib_port_attr *props)
 821 {
 822         /* returns host view */
 823         return __mlx4_ib_query_port(ibdev, port, props, 0);
 824 }
 825
 826 int __mlx4_ib_query_gid(struct ib_device *ibdev, u32 port, int index,
 827                         union ib_gid *gid, int netw_view)
 828 {
 829         struct ib_smp *in_mad;
 830         struct ib_smp *out_mad;
 831         int err = -ENOMEM;
 832         struct mlx4_ib_dev *dev = to_mdev(ibdev);
 833         int clear = 0;
 834         int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
 835
 836         in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
 837         out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
 838         if (!in_mad || !out_mad)
 839                 goto out;
 840
 841         ib_init_query_mad(in_mad);
 842         in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
 843         in_mad->attr_mod = cpu_to_be32(port);
 844
 845         if (mlx4_is_mfunc(dev->dev) && netw_view)
 846                 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
 847
 848         err = mlx4_MAD_IFC(dev, mad_ifc_flags, port, NULL, NULL, in_mad, out_mad);
 849         if (err)
 850                 goto out;
 851
 852         memcpy(gid->raw, out_mad->data + 8, 8);
 853
 854         if (mlx4_is_mfunc(dev->dev) && !netw_view) {
 855                 if (index) {
 856                         /* For any index > 0, return the null guid */
 857                         err = 0;
 858                         clear = 1;
 859                         goto out;
 860                 }
 861         }
 862
 863         ib_init_query_mad(in_mad);
 864         in_mad->attr_id  = IB_SMP_ATTR_GUID_INFO;
 865         in_mad->attr_mod = cpu_to_be32(index / 8);
 866
 867         err = mlx4_MAD_IFC(dev, mad_ifc_flags, port,
 868                            NULL, NULL, in_mad, out_mad);
 869         if (err)
 870                 goto out;
 871
 872         memcpy(gid->raw + 8, out_mad->data + (index % 8) * 8, 8);
 873
 874 out:
 875         if (clear)
 876                 memset(gid->raw + 8, 0, 8);
 877         kfree(in_mad);
 878         kfree(out_mad);
 879         return err;
 880 }
 881
 882 static int mlx4_ib_query_gid(struct ib_device *ibdev, u32 port, int index,
 883                              union ib_gid *gid)
 884 {
 885         if (rdma_protocol_ib(ibdev, port))
 886                 return __mlx4_ib_query_gid(ibdev, port, index, gid, 0);
 887         return 0;
 888 }
 889
 890 static int mlx4_ib_query_sl2vl(struct ib_device *ibdev, u32 port,
 891                                u64 *sl2vl_tbl)
 892 {
 893         union sl2vl_tbl_to_u64 sl2vl64;
 894         struct ib_smp *in_mad;
 895         struct ib_smp *out_mad;
 896         int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
 897         int err = -ENOMEM;
 898         int jj;
 899
 900         if (mlx4_is_slave(to_mdev(ibdev)->dev)) {
 901                 *sl2vl_tbl = 0;
 902                 return 0;
 903         }
 904
 905         in_mad  = kzalloc(sizeof(*in_mad), GFP_KERNEL);
 906         out_mad = kmalloc(sizeof(*out_mad), GFP_KERNEL);
 907         if (!in_mad || !out_mad)
 908                 goto out;
 909
 910         ib_init_query_mad(in_mad);
 911         in_mad->attr_id  = IB_SMP_ATTR_SL_TO_VL_TABLE;
 912         in_mad->attr_mod = 0;
 913
 914         if (mlx4_is_mfunc(to_mdev(ibdev)->dev))
 915                 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
 916
 917         err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
 918                            in_mad, out_mad);
 919         if (err)
 920                 goto out;
 921
 922         for (jj = 0; jj < 8; jj++)
 923                 sl2vl64.sl8[jj] = ((struct ib_smp *)out_mad)->data[jj];
 924         *sl2vl_tbl = sl2vl64.sl64;
 925
 926 out:
 927         kfree(in_mad);
 928         kfree(out_mad);
 929         return err;
 930 }
 931
 932 static void mlx4_init_sl2vl_tbl(struct mlx4_ib_dev *mdev)
 933 {
 934         u64 sl2vl;
 935         int i;
 936         int err;
 937
 938         for (i = 1; i <= mdev->dev->caps.num_ports; i++) {
 939                 if (mdev->dev->caps.port_type[i] == MLX4_PORT_TYPE_ETH)
 940                         continue;
 941                 err = mlx4_ib_query_sl2vl(&mdev->ib_dev, i, &sl2vl);
 942                 if (err) {
 943                         pr_err("Unable to get default sl to vl mapping for port %d.  Using all zeroes (%d)\n",
 944                                i, err);
 945                         sl2vl = 0;
 946                 }
 947                 atomic64_set(&mdev->sl2vl[i - 1], sl2vl);
 948         }
 949 }
 950
 951 int __mlx4_ib_query_pkey(struct ib_device *ibdev, u32 port, u16 index,
 952                          u16 *pkey, int netw_view)
 953 {
 954         struct ib_smp *in_mad;
 955         struct ib_smp *out_mad;
 956         int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
 957         int err = -ENOMEM;
 958
 959         in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
 960         out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
 961         if (!in_mad || !out_mad)
 962                 goto out;
 963
 964         ib_init_query_mad(in_mad);
 965         in_mad->attr_id  = IB_SMP_ATTR_PKEY_TABLE;
 966         in_mad->attr_mod = cpu_to_be32(index / 32);
 967
 968         if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
 969                 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
 970
 971         err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
 972                            in_mad, out_mad);
 973         if (err)
 974                 goto out;
 975
 976         *pkey = be16_to_cpu(((__be16 *) out_mad->data)[index % 32]);
 977
 978 out:
 979         kfree(in_mad);
 980         kfree(out_mad);
 981         return err;
 982 }
 983
 984 static int mlx4_ib_query_pkey(struct ib_device *ibdev, u32 port, u16 index,
 985                               u16 *pkey)
 986 {
 987         return __mlx4_ib_query_pkey(ibdev, port, index, pkey, 0);
 988 }
 989
 990 static int mlx4_ib_modify_device(struct ib_device *ibdev, int mask,
 991                                  struct ib_device_modify *props)
 992 {
 993         struct mlx4_cmd_mailbox *mailbox;
 994         unsigned long flags;
 995
 996         if (mask & ~IB_DEVICE_MODIFY_NODE_DESC)
 997                 return -EOPNOTSUPP;
 998
 999         if (!(mask & IB_DEVICE_MODIFY_NODE_DESC))
1000                 return 0;
1001
1002         if (mlx4_is_slave(to_mdev(ibdev)->dev))
1003                 return -EOPNOTSUPP;
1004
1005         spin_lock_irqsave(&to_mdev(ibdev)->sm_lock, flags);
1006         memcpy(ibdev->node_desc, props->node_desc, IB_DEVICE_NODE_DESC_MAX);
1007         spin_unlock_irqrestore(&to_mdev(ibdev)->sm_lock, flags);
1008
1009         /*
1010          * If possible, pass node desc to FW, so it can generate
1011          * a 144 trap.  If cmd fails, just ignore.
1012          */
1013         mailbox = mlx4_alloc_cmd_mailbox(to_mdev(ibdev)->dev);
1014         if (IS_ERR(mailbox))
1015                 return 0;
1016
1017         memcpy(mailbox->buf, props->node_desc, IB_DEVICE_NODE_DESC_MAX);
1018         mlx4_cmd(to_mdev(ibdev)->dev, mailbox->dma, 1, 0,
1019                  MLX4_CMD_SET_NODE, MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
1020
1021         mlx4_free_cmd_mailbox(to_mdev(ibdev)->dev, mailbox);
1022
1023         return 0;
1024 }
1025
1026 static int mlx4_ib_SET_PORT(struct mlx4_ib_dev *dev, u32 port,
1027                             int reset_qkey_viols, u32 cap_mask)
1028 {
1029         struct mlx4_cmd_mailbox *mailbox;
1030         int err;
1031
1032         mailbox = mlx4_alloc_cmd_mailbox(dev->dev);
1033         if (IS_ERR(mailbox))
1034                 return PTR_ERR(mailbox);
1035
1036         if (dev->dev->flags & MLX4_FLAG_OLD_PORT_CMDS) {
1037                 *(u8 *) mailbox->buf         = !!reset_qkey_viols << 6;
1038                 ((__be32 *) mailbox->buf)[2] = cpu_to_be32(cap_mask);
1039         } else {
1040                 ((u8 *) mailbox->buf)[3]     = !!reset_qkey_viols;
1041                 ((__be32 *) mailbox->buf)[1] = cpu_to_be32(cap_mask);
1042         }
1043
1044         err = mlx4_cmd(dev->dev, mailbox->dma, port, MLX4_SET_PORT_IB_OPCODE,
1045                        MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
1046                        MLX4_CMD_WRAPPED);
1047
1048         mlx4_free_cmd_mailbox(dev->dev, mailbox);
1049         return err;
1050 }
1051
1052 static int mlx4_ib_modify_port(struct ib_device *ibdev, u32 port, int mask,
1053                                struct ib_port_modify *props)
1054 {
1055         struct mlx4_ib_dev *mdev = to_mdev(ibdev);
1056         u8 is_eth = mdev->dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH;
1057         struct ib_port_attr attr;
1058         u32 cap_mask;
1059         int err;
1060
1061         /* return OK if this is RoCE. CM calls ib_modify_port() regardless
1062          * of whether port link layer is ETH or IB. For ETH ports, qkey
1063          * violations and port capabilities are not meaningful.
1064          */
1065         if (is_eth)
1066                 return 0;
1067
1068         mutex_lock(&mdev->cap_mask_mutex);
1069
1070         err = ib_query_port(ibdev, port, &attr);
1071         if (err)
1072                 goto out;
1073
1074         cap_mask = (attr.port_cap_flags | props->set_port_cap_mask) &
1075                 ~props->clr_port_cap_mask;
1076
1077         err = mlx4_ib_SET_PORT(mdev, port,
1078                                !!(mask & IB_PORT_RESET_QKEY_CNTR),
1079                                cap_mask);
1080
1081 out:
1082         mutex_unlock(&to_mdev(ibdev)->cap_mask_mutex);
1083         return err;
1084 }
1085
1086 static int mlx4_ib_alloc_ucontext(struct ib_ucontext *uctx,
1087                                   struct ib_udata *udata)
1088 {
1089         struct ib_device *ibdev = uctx->device;
1090         struct mlx4_ib_dev *dev = to_mdev(ibdev);
1091         struct mlx4_ib_ucontext *context = to_mucontext(uctx);
1092         struct mlx4_ib_alloc_ucontext_resp_v3 resp_v3;
1093         struct mlx4_ib_alloc_ucontext_resp resp;
1094         int err;
1095
1096         if (!dev->ib_active)
1097                 return -EAGAIN;
1098
1099         if (ibdev->ops.uverbs_abi_ver ==
1100             MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION) {
1101                 resp_v3.qp_tab_size      = dev->dev->caps.num_qps;
1102                 resp_v3.bf_reg_size      = dev->dev->caps.bf_reg_size;
1103                 resp_v3.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
1104         } else {
1105                 resp.dev_caps         = dev->dev->caps.userspace_caps;
1106                 resp.qp_tab_size      = dev->dev->caps.num_qps;
1107                 resp.bf_reg_size      = dev->dev->caps.bf_reg_size;
1108                 resp.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
1109                 resp.cqe_size         = dev->dev->caps.cqe_size;
1110         }
1111
1112         err = mlx4_uar_alloc(to_mdev(ibdev)->dev, &context->uar);
1113         if (err)
1114                 return err;
1115
1116         INIT_LIST_HEAD(&context->db_page_list);
1117         mutex_init(&context->db_page_mutex);
1118
1119         INIT_LIST_HEAD(&context->wqn_ranges_list);
1120         mutex_init(&context->wqn_ranges_mutex);
1121
1122         if (ibdev->ops.uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION)
1123                 err = ib_copy_to_udata(udata, &resp_v3, sizeof(resp_v3));
1124         else
1125                 err = ib_copy_to_udata(udata, &resp, sizeof(resp));
1126
1127         if (err) {
1128                 mlx4_uar_free(to_mdev(ibdev)->dev, &context->uar);
1129                 return -EFAULT;
1130         }
1131
1132         return err;
1133 }
1134
1135 static void mlx4_ib_dealloc_ucontext(struct ib_ucontext *ibcontext)
1136 {
1137         struct mlx4_ib_ucontext *context = to_mucontext(ibcontext);
1138
1139         mlx4_uar_free(to_mdev(ibcontext->device)->dev, &context->uar);
1140 }
1141
1142 static void mlx4_ib_disassociate_ucontext(struct ib_ucontext *ibcontext)
1143 {
1144 }
1145
1146 static int mlx4_ib_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
1147 {
1148         struct mlx4_ib_dev *dev = to_mdev(context->device);
1149
1150         switch (vma->vm_pgoff) {
1151         case 0:
1152                 return rdma_user_mmap_io(context, vma,
1153                                          to_mucontext(context)->uar.pfn,
1154                                          PAGE_SIZE,
1155                                          pgprot_noncached(vma->vm_page_prot),
1156                                          NULL);
1157
1158         case 1:
1159                 if (dev->dev->caps.bf_reg_size == 0)
1160                         return -EINVAL;
1161                 return rdma_user_mmap_io(
1162                         context, vma,
1163                         to_mucontext(context)->uar.pfn +
1164                                 dev->dev->caps.num_uars,
1165                         PAGE_SIZE, pgprot_writecombine(vma->vm_page_prot),
1166                         NULL);
1167
1168         case 3: {
1169                 struct mlx4_clock_params params;
1170                 int ret;
1171
1172                 ret = mlx4_get_internal_clock_params(dev->dev, &params);
1173                 if (ret)
1174                         return ret;
1175
1176                 return rdma_user_mmap_io(
1177                         context, vma,
1178                         (pci_resource_start(dev->dev->persist->pdev,
1179                                             params.bar) +
1180                          params.offset) >>
1181                                 PAGE_SHIFT,
1182                         PAGE_SIZE, pgprot_noncached(vma->vm_page_prot),
1183                         NULL);
1184         }
1185
1186         default:
1187                 return -EINVAL;
1188         }
1189 }
1190
1191 static int mlx4_ib_alloc_pd(struct ib_pd *ibpd, struct ib_udata *udata)
1192 {
1193         struct mlx4_ib_pd *pd = to_mpd(ibpd);
1194         struct ib_device *ibdev = ibpd->device;
1195         int err;
1196
1197         err = mlx4_pd_alloc(to_mdev(ibdev)->dev, &pd->pdn);
1198         if (err)
1199                 return err;
1200
1201         if (udata && ib_copy_to_udata(udata, &pd->pdn, sizeof(__u32))) {
1202                 mlx4_pd_free(to_mdev(ibdev)->dev, pd->pdn);
1203                 return -EFAULT;
1204         }
1205         return 0;
1206 }
1207
1208 static int mlx4_ib_dealloc_pd(struct ib_pd *pd, struct ib_udata *udata)
1209 {
1210         mlx4_pd_free(to_mdev(pd->device)->dev, to_mpd(pd)->pdn);
1211         return 0;
1212 }
1213
1214 static int mlx4_ib_alloc_xrcd(struct ib_xrcd *ibxrcd, struct ib_udata *udata)
1215 {
1216         struct mlx4_ib_dev *dev = to_mdev(ibxrcd->device);
1217         struct mlx4_ib_xrcd *xrcd = to_mxrcd(ibxrcd);
1218         struct ib_cq_init_attr cq_attr = {};
1219         int err;
1220
1221         if (!(dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC))
1222                 return -EOPNOTSUPP;
1223
1224         err = mlx4_xrcd_alloc(dev->dev, &xrcd->xrcdn);
1225         if (err)
1226                 return err;
1227
1228         xrcd->pd = ib_alloc_pd(ibxrcd->device, 0);
1229         if (IS_ERR(xrcd->pd)) {
1230                 err = PTR_ERR(xrcd->pd);
1231                 goto err2;
1232         }
1233
1234         cq_attr.cqe = 1;
1235         xrcd->cq = ib_create_cq(ibxrcd->device, NULL, NULL, xrcd, &cq_attr);
1236         if (IS_ERR(xrcd->cq)) {
1237                 err = PTR_ERR(xrcd->cq);
1238                 goto err3;
1239         }
1240
1241         return 0;
1242
1243 err3:
1244         ib_dealloc_pd(xrcd->pd);
1245 err2:
1246         mlx4_xrcd_free(dev->dev, xrcd->xrcdn);
1247         return err;
1248 }
1249
1250 static int mlx4_ib_dealloc_xrcd(struct ib_xrcd *xrcd, struct ib_udata *udata)
1251 {
1252         ib_destroy_cq(to_mxrcd(xrcd)->cq);
1253         ib_dealloc_pd(to_mxrcd(xrcd)->pd);
1254         mlx4_xrcd_free(to_mdev(xrcd->device)->dev, to_mxrcd(xrcd)->xrcdn);
1255         return 0;
1256 }
1257
1258 static int add_gid_entry(struct ib_qp *ibqp, union ib_gid *gid)
1259 {
1260         struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1261         struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1262         struct mlx4_ib_gid_entry *ge;
1263
1264         ge = kzalloc(sizeof *ge, GFP_KERNEL);
1265         if (!ge)
1266                 return -ENOMEM;
1267
1268         ge->gid = *gid;
1269         if (mlx4_ib_add_mc(mdev, mqp, gid)) {
1270                 ge->port = mqp->port;
1271                 ge->added = 1;
1272         }
1273
1274         mutex_lock(&mqp->mutex);
1275         list_add_tail(&ge->list, &mqp->gid_list);
1276         mutex_unlock(&mqp->mutex);
1277
1278         return 0;
1279 }
1280
1281 static void mlx4_ib_delete_counters_table(struct mlx4_ib_dev *ibdev,
1282                                           struct mlx4_ib_counters *ctr_table)
1283 {
1284         struct counter_index *counter, *tmp_count;
1285
1286         mutex_lock(&ctr_table->mutex);
1287         list_for_each_entry_safe(counter, tmp_count, &ctr_table->counters_list,
1288                                  list) {
1289                 if (counter->allocated)
1290                         mlx4_counter_free(ibdev->dev, counter->index);
1291                 list_del(&counter->list);
1292                 kfree(counter);
1293         }
1294         mutex_unlock(&ctr_table->mutex);
1295 }
1296
1297 int mlx4_ib_add_mc(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
1298                    union ib_gid *gid)
1299 {
1300         struct net_device *ndev;
1301         int ret = 0;
1302
1303         if (!mqp->port)
1304                 return 0;
1305
1306         spin_lock_bh(&mdev->iboe.lock);
1307         ndev = mdev->iboe.netdevs[mqp->port - 1];
1308         dev_hold(ndev);
1309         spin_unlock_bh(&mdev->iboe.lock);
1310
1311         if (ndev) {
1312                 ret = 1;
1313                 dev_put(ndev);
1314         }
1315
1316         return ret;
1317 }
1318
1319 struct mlx4_ib_steering {
1320         struct list_head list;
1321         struct mlx4_flow_reg_id reg_id;
1322         union ib_gid gid;
1323 };
1324
1325 #define LAST_ETH_FIELD vlan_tag
1326 #define LAST_IB_FIELD sl
1327 #define LAST_IPV4_FIELD dst_ip
1328 #define LAST_TCP_UDP_FIELD src_port
1329
1330 /* Field is the last supported field */
1331 #define FIELDS_NOT_SUPPORTED(filter, field)\
1332         memchr_inv((void *)&filter.field  +\
1333                    sizeof(filter.field), 0,\
1334                    sizeof(filter) -\
1335                    offsetof(typeof(filter), field) -\
1336                    sizeof(filter.field))
1337
1338 static int parse_flow_attr(struct mlx4_dev *dev,
1339                            u32 qp_num,
1340                            union ib_flow_spec *ib_spec,
1341                            struct _rule_hw *mlx4_spec)
1342 {
1343         enum mlx4_net_trans_rule_id type;
1344
1345         switch (ib_spec->type) {
1346         case IB_FLOW_SPEC_ETH:
1347                 if (FIELDS_NOT_SUPPORTED(ib_spec->eth.mask, LAST_ETH_FIELD))
1348                         return -ENOTSUPP;
1349
1350                 type = MLX4_NET_TRANS_RULE_ID_ETH;
1351                 memcpy(mlx4_spec->eth.dst_mac, ib_spec->eth.val.dst_mac,
1352                        ETH_ALEN);
1353                 memcpy(mlx4_spec->eth.dst_mac_msk, ib_spec->eth.mask.dst_mac,
1354                        ETH_ALEN);
1355                 mlx4_spec->eth.vlan_tag = ib_spec->eth.val.vlan_tag;
1356                 mlx4_spec->eth.vlan_tag_msk = ib_spec->eth.mask.vlan_tag;
1357                 break;
1358         case IB_FLOW_SPEC_IB:
1359                 if (FIELDS_NOT_SUPPORTED(ib_spec->ib.mask, LAST_IB_FIELD))
1360                         return -ENOTSUPP;
1361
1362                 type = MLX4_NET_TRANS_RULE_ID_IB;
1363                 mlx4_spec->ib.l3_qpn =
1364                         cpu_to_be32(qp_num);
1365                 mlx4_spec->ib.qpn_mask =
1366                         cpu_to_be32(MLX4_IB_FLOW_QPN_MASK);
1367                 break;
1368
1369
1370         case IB_FLOW_SPEC_IPV4:
1371                 if (FIELDS_NOT_SUPPORTED(ib_spec->ipv4.mask, LAST_IPV4_FIELD))
1372                         return -ENOTSUPP;
1373
1374                 type = MLX4_NET_TRANS_RULE_ID_IPV4;
1375                 mlx4_spec->ipv4.src_ip = ib_spec->ipv4.val.src_ip;
1376                 mlx4_spec->ipv4.src_ip_msk = ib_spec->ipv4.mask.src_ip;
1377                 mlx4_spec->ipv4.dst_ip = ib_spec->ipv4.val.dst_ip;
1378                 mlx4_spec->ipv4.dst_ip_msk = ib_spec->ipv4.mask.dst_ip;
1379                 break;
1380
1381         case IB_FLOW_SPEC_TCP:
1382         case IB_FLOW_SPEC_UDP:
1383                 if (FIELDS_NOT_SUPPORTED(ib_spec->tcp_udp.mask, LAST_TCP_UDP_FIELD))
1384                         return -ENOTSUPP;
1385
1386                 type = ib_spec->type == IB_FLOW_SPEC_TCP ?
1387                                         MLX4_NET_TRANS_RULE_ID_TCP :
1388                                         MLX4_NET_TRANS_RULE_ID_UDP;
1389                 mlx4_spec->tcp_udp.dst_port = ib_spec->tcp_udp.val.dst_port;
1390                 mlx4_spec->tcp_udp.dst_port_msk = ib_spec->tcp_udp.mask.dst_port;
1391                 mlx4_spec->tcp_udp.src_port = ib_spec->tcp_udp.val.src_port;
1392                 mlx4_spec->tcp_udp.src_port_msk = ib_spec->tcp_udp.mask.src_port;
1393                 break;
1394
1395         default:
1396                 return -EINVAL;
1397         }
1398         if (mlx4_map_sw_to_hw_steering_id(dev, type) < 0 ||
1399             mlx4_hw_rule_sz(dev, type) < 0)
1400                 return -EINVAL;
1401         mlx4_spec->id = cpu_to_be16(mlx4_map_sw_to_hw_steering_id(dev, type));
1402         mlx4_spec->size = mlx4_hw_rule_sz(dev, type) >> 2;
1403         return mlx4_hw_rule_sz(dev, type);
1404 }
1405
1406 struct default_rules {
1407         __u32 mandatory_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
1408         __u32 mandatory_not_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
1409         __u32 rules_create_list[IB_FLOW_SPEC_SUPPORT_LAYERS];
1410         __u8  link_layer;
1411 };
1412 static const struct default_rules default_table[] = {
1413         {
1414                 .mandatory_fields = {IB_FLOW_SPEC_IPV4},
1415                 .mandatory_not_fields = {IB_FLOW_SPEC_ETH},
1416                 .rules_create_list = {IB_FLOW_SPEC_IB},
1417                 .link_layer = IB_LINK_LAYER_INFINIBAND
1418         }
1419 };
1420
1421 static int __mlx4_ib_default_rules_match(struct ib_qp *qp,
1422                                          struct ib_flow_attr *flow_attr)
1423 {
1424         int i, j, k;
1425         void *ib_flow;
1426         const struct default_rules *pdefault_rules = default_table;
1427         u8 link_layer = rdma_port_get_link_layer(qp->device, flow_attr->port);
1428
1429         for (i = 0; i < ARRAY_SIZE(default_table); i++, pdefault_rules++) {
1430                 __u32 field_types[IB_FLOW_SPEC_SUPPORT_LAYERS];
1431                 memset(&field_types, 0, sizeof(field_types));
1432
1433                 if (link_layer != pdefault_rules->link_layer)
1434                         continue;
1435
1436                 ib_flow = flow_attr + 1;
1437                 /* we assume the specs are sorted */
1438                 for (j = 0, k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS &&
1439                      j < flow_attr->num_of_specs; k++) {
1440                         union ib_flow_spec *current_flow =
1441                                 (union ib_flow_spec *)ib_flow;
1442
1443                         /* same layer but different type */
1444                         if (((current_flow->type & IB_FLOW_SPEC_LAYER_MASK) ==
1445                              (pdefault_rules->mandatory_fields[k] &
1446                               IB_FLOW_SPEC_LAYER_MASK)) &&
1447                             (current_flow->type !=
1448                              pdefault_rules->mandatory_fields[k]))
1449                                 goto out;
1450
1451                         /* same layer, try match next one */
1452                         if (current_flow->type ==
1453                             pdefault_rules->mandatory_fields[k]) {
1454                                 j++;
1455                                 ib_flow +=
1456                                         ((union ib_flow_spec *)ib_flow)->size;
1457                         }
1458                 }
1459
1460                 ib_flow = flow_attr + 1;
1461                 for (j = 0; j < flow_attr->num_of_specs;
1462                      j++, ib_flow += ((union ib_flow_spec *)ib_flow)->size)
1463                         for (k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS; k++)
1464                                 /* same layer and same type */
1465                                 if (((union ib_flow_spec *)ib_flow)->type ==
1466                                     pdefault_rules->mandatory_not_fields[k])
1467                                         goto out;
1468
1469                 return i;
1470         }
1471 out:
1472         return -1;
1473 }
1474
1475 static int __mlx4_ib_create_default_rules(
1476                 struct mlx4_ib_dev *mdev,
1477                 struct ib_qp *qp,
1478                 const struct default_rules *pdefault_rules,
1479                 struct _rule_hw *mlx4_spec) {
1480         int size = 0;
1481         int i;
1482
1483         for (i = 0; i < ARRAY_SIZE(pdefault_rules->rules_create_list); i++) {
1484                 union ib_flow_spec ib_spec = {};
1485                 int ret;
1486
1487                 switch (pdefault_rules->rules_create_list[i]) {
1488                 case 0:
1489                         /* no rule */
1490                         continue;
1491                 case IB_FLOW_SPEC_IB:
1492                         ib_spec.type = IB_FLOW_SPEC_IB;
1493                         ib_spec.size = sizeof(struct ib_flow_spec_ib);
1494
1495                         break;
1496                 default:
1497                         /* invalid rule */
1498                         return -EINVAL;
1499                 }
1500                 /* We must put empty rule, qpn is being ignored */
1501                 ret = parse_flow_attr(mdev->dev, 0, &ib_spec,
1502                                       mlx4_spec);
1503                 if (ret < 0) {
1504                         pr_info("invalid parsing\n");
1505                         return -EINVAL;
1506                 }
1507
1508                 mlx4_spec = (void *)mlx4_spec + ret;
1509                 size += ret;
1510         }
1511         return size;
1512 }
1513
1514 static int __mlx4_ib_create_flow(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
1515                           int domain,
1516                           enum mlx4_net_trans_promisc_mode flow_type,
1517                           u64 *reg_id)
1518 {
1519         int ret, i;
1520         int size = 0;
1521         void *ib_flow;
1522         struct mlx4_ib_dev *mdev = to_mdev(qp->device);
1523         struct mlx4_cmd_mailbox *mailbox;
1524         struct mlx4_net_trans_rule_hw_ctrl *ctrl;
1525         int default_flow;
1526
1527         if (flow_attr->priority > MLX4_IB_FLOW_MAX_PRIO) {
1528                 pr_err("Invalid priority value %d\n", flow_attr->priority);
1529                 return -EINVAL;
1530         }
1531
1532         if (mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type) < 0)
1533                 return -EINVAL;
1534
1535         mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
1536         if (IS_ERR(mailbox))
1537                 return PTR_ERR(mailbox);
1538         ctrl = mailbox->buf;
1539
1540         ctrl->prio = cpu_to_be16(domain | flow_attr->priority);
1541         ctrl->type = mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type);
1542         ctrl->port = flow_attr->port;
1543         ctrl->qpn = cpu_to_be32(qp->qp_num);
1544
1545         ib_flow = flow_attr + 1;
1546         size += sizeof(struct mlx4_net_trans_rule_hw_ctrl);
1547         /* Add default flows */
1548         default_flow = __mlx4_ib_default_rules_match(qp, flow_attr);
1549         if (default_flow >= 0) {
1550                 ret = __mlx4_ib_create_default_rules(
1551                                 mdev, qp, default_table + default_flow,
1552                                 mailbox->buf + size);
1553                 if (ret < 0) {
1554                         mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1555                         return -EINVAL;
1556                 }
1557                 size += ret;
1558         }
1559         for (i = 0; i < flow_attr->num_of_specs; i++) {
1560                 ret = parse_flow_attr(mdev->dev, qp->qp_num, ib_flow,
1561                                       mailbox->buf + size);
1562                 if (ret < 0) {
1563                         mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1564                         return -EINVAL;
1565                 }
1566                 ib_flow += ((union ib_flow_spec *) ib_flow)->size;
1567                 size += ret;
1568         }
1569
1570         if (mlx4_is_master(mdev->dev) && flow_type == MLX4_FS_REGULAR &&
1571             flow_attr->num_of_specs == 1) {
1572                 struct _rule_hw *rule_header = (struct _rule_hw *)(ctrl + 1);
1573                 enum ib_flow_spec_type header_spec =
1574                         ((union ib_flow_spec *)(flow_attr + 1))->type;
1575
1576                 if (header_spec == IB_FLOW_SPEC_ETH)
1577                         mlx4_handle_eth_header_mcast_prio(ctrl, rule_header);
1578         }
1579
1580         ret = mlx4_cmd_imm(mdev->dev, mailbox->dma, reg_id, size >> 2, 0,
1581                            MLX4_QP_FLOW_STEERING_ATTACH, MLX4_CMD_TIME_CLASS_A,
1582                            MLX4_CMD_NATIVE);
1583         if (ret == -ENOMEM)
1584                 pr_err("mcg table is full. Fail to register network rule.\n");
1585         else if (ret == -ENXIO)
1586                 pr_err("Device managed flow steering is disabled. Fail to register network rule.\n");
1587         else if (ret)
1588                 pr_err("Invalid argument. Fail to register network rule.\n");
1589
1590         mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1591         return ret;
1592 }
1593
1594 static int __mlx4_ib_destroy_flow(struct mlx4_dev *dev, u64 reg_id)
1595 {
1596         int err;
1597         err = mlx4_cmd(dev, reg_id, 0, 0,
1598                        MLX4_QP_FLOW_STEERING_DETACH, MLX4_CMD_TIME_CLASS_A,
1599                        MLX4_CMD_NATIVE);
1600         if (err)
1601                 pr_err("Fail to detach network rule. registration id = 0x%llx\n",
1602                        reg_id);
1603         return err;
1604 }
1605
1606 static int mlx4_ib_tunnel_steer_add(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
1607                                     u64 *reg_id)
1608 {
1609         void *ib_flow;
1610         union ib_flow_spec *ib_spec;
1611         struct mlx4_dev *dev = to_mdev(qp->device)->dev;
1612         int err = 0;
1613
1614         if (dev->caps.tunnel_offload_mode != MLX4_TUNNEL_OFFLOAD_MODE_VXLAN ||
1615             dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC)
1616                 return 0; /* do nothing */
1617
1618         ib_flow = flow_attr + 1;
1619         ib_spec = (union ib_flow_spec *)ib_flow;
1620
1621         if (ib_spec->type !=  IB_FLOW_SPEC_ETH || flow_attr->num_of_specs != 1)
1622                 return 0; /* do nothing */
1623
1624         err = mlx4_tunnel_steer_add(to_mdev(qp->device)->dev, ib_spec->eth.val.dst_mac,
1625                                     flow_attr->port, qp->qp_num,
1626                                     MLX4_DOMAIN_UVERBS | (flow_attr->priority & 0xff),
1627                                     reg_id);
1628         return err;
1629 }
1630
1631 static int mlx4_ib_add_dont_trap_rule(struct mlx4_dev *dev,
1632                                       struct ib_flow_attr *flow_attr,
1633                                       enum mlx4_net_trans_promisc_mode *type)
1634 {
1635         int err = 0;
1636
1637         if (!(dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DMFS_UC_MC_SNIFFER) ||
1638             (dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC) ||
1639             (flow_attr->num_of_specs > 1) || (flow_attr->priority != 0)) {
1640                 return -EOPNOTSUPP;
1641         }
1642
1643         if (flow_attr->num_of_specs == 0) {
1644                 type[0] = MLX4_FS_MC_SNIFFER;
1645                 type[1] = MLX4_FS_UC_SNIFFER;
1646         } else {
1647                 union ib_flow_spec *ib_spec;
1648
1649                 ib_spec = (union ib_flow_spec *)(flow_attr + 1);
1650                 if (ib_spec->type !=  IB_FLOW_SPEC_ETH)
1651                         return -EINVAL;
1652
1653                 /* if all is zero than MC and UC */
1654                 if (is_zero_ether_addr(ib_spec->eth.mask.dst_mac)) {
1655                         type[0] = MLX4_FS_MC_SNIFFER;
1656                         type[1] = MLX4_FS_UC_SNIFFER;
1657                 } else {
1658                         u8 mac[ETH_ALEN] = {ib_spec->eth.mask.dst_mac[0] ^ 0x01,
1659                                             ib_spec->eth.mask.dst_mac[1],
1660                                             ib_spec->eth.mask.dst_mac[2],
1661                                             ib_spec->eth.mask.dst_mac[3],
1662                                             ib_spec->eth.mask.dst_mac[4],
1663                                             ib_spec->eth.mask.dst_mac[5]};
1664
1665                         /* Above xor was only on MC bit, non empty mask is valid
1666                          * only if this bit is set and rest are zero.
1667                          */
1668                         if (!is_zero_ether_addr(&mac[0]))
1669                                 return -EINVAL;
1670
1671                         if (is_multicast_ether_addr(ib_spec->eth.val.dst_mac))
1672                                 type[0] = MLX4_FS_MC_SNIFFER;
1673                         else
1674                                 type[0] = MLX4_FS_UC_SNIFFER;
1675                 }
1676         }
1677
1678         return err;
1679 }
1680
1681 static struct ib_flow *mlx4_ib_create_flow(struct ib_qp *qp,
1682                                            struct ib_flow_attr *flow_attr,
1683                                            struct ib_udata *udata)
1684 {
1685         int err = 0, i = 0, j = 0;
1686         struct mlx4_ib_flow *mflow;
1687         enum mlx4_net_trans_promisc_mode type[2];
1688         struct mlx4_dev *dev = (to_mdev(qp->device))->dev;
1689         int is_bonded = mlx4_is_bonded(dev);
1690
1691         if (flow_attr->flags & ~IB_FLOW_ATTR_FLAGS_DONT_TRAP)
1692                 return ERR_PTR(-EOPNOTSUPP);
1693
1694         if ((flow_attr->flags & IB_FLOW_ATTR_FLAGS_DONT_TRAP) &&
1695             (flow_attr->type != IB_FLOW_ATTR_NORMAL))
1696                 return ERR_PTR(-EOPNOTSUPP);
1697
1698         if (udata &&
1699             udata->inlen && !ib_is_udata_cleared(udata, 0, udata->inlen))
1700                 return ERR_PTR(-EOPNOTSUPP);
1701
1702         memset(type, 0, sizeof(type));
1703
1704         mflow = kzalloc(sizeof(*mflow), GFP_KERNEL);
1705         if (!mflow) {
1706                 err = -ENOMEM;
1707                 goto err_free;
1708         }
1709
1710         switch (flow_attr->type) {
1711         case IB_FLOW_ATTR_NORMAL:
1712                 /* If dont trap flag (continue match) is set, under specific
1713                  * condition traffic be replicated to given qp,
1714                  * without stealing it
1715                  */
1716                 if (unlikely(flow_attr->flags & IB_FLOW_ATTR_FLAGS_DONT_TRAP)) {
1717                         err = mlx4_ib_add_dont_trap_rule(dev,
1718                                                          flow_attr,
1719                                                          type);
1720                         if (err)
1721                                 goto err_free;
1722                 } else {
1723                         type[0] = MLX4_FS_REGULAR;
1724                 }
1725                 break;
1726
1727         case IB_FLOW_ATTR_ALL_DEFAULT:
1728                 type[0] = MLX4_FS_ALL_DEFAULT;
1729                 break;
1730
1731         case IB_FLOW_ATTR_MC_DEFAULT:
1732                 type[0] = MLX4_FS_MC_DEFAULT;
1733                 break;
1734
1735         case IB_FLOW_ATTR_SNIFFER:
1736                 type[0] = MLX4_FS_MIRROR_RX_PORT;
1737                 type[1] = MLX4_FS_MIRROR_SX_PORT;
1738                 break;
1739
1740         default:
1741                 err = -EINVAL;
1742                 goto err_free;
1743         }
1744
1745         while (i < ARRAY_SIZE(type) && type[i]) {
1746                 err = __mlx4_ib_create_flow(qp, flow_attr, MLX4_DOMAIN_UVERBS,
1747                                             type[i], &mflow->reg_id[i].id);
1748                 if (err)
1749                         goto err_create_flow;
1750                 if (is_bonded) {
1751                         /* Application always sees one port so the mirror rule
1752                          * must be on port #2
1753                          */
1754                         flow_attr->port = 2;
1755                         err = __mlx4_ib_create_flow(qp, flow_attr,
1756                                                     MLX4_DOMAIN_UVERBS, type[j],
1757                                                     &mflow->reg_id[j].mirror);
1758                         flow_attr->port = 1;
1759                         if (err)
1760                                 goto err_create_flow;
1761                         j++;
1762                 }
1763
1764                 i++;
1765         }
1766
1767         if (i < ARRAY_SIZE(type) && flow_attr->type == IB_FLOW_ATTR_NORMAL) {
1768                 err = mlx4_ib_tunnel_steer_add(qp, flow_attr,
1769                                                &mflow->reg_id[i].id);
1770                 if (err)
1771                         goto err_create_flow;
1772
1773                 if (is_bonded) {
1774                         flow_attr->port = 2;
1775                         err = mlx4_ib_tunnel_steer_add(qp, flow_attr,
1776                                                        &mflow->reg_id[j].mirror);
1777                         flow_attr->port = 1;
1778                         if (err)
1779                                 goto err_create_flow;
1780                         j++;
1781                 }
1782                 /* function to create mirror rule */
1783                 i++;
1784         }
1785
1786         return &mflow->ibflow;
1787
1788 err_create_flow:
1789         while (i) {
1790                 (void)__mlx4_ib_destroy_flow(to_mdev(qp->device)->dev,
1791                                              mflow->reg_id[i].id);
1792                 i--;
1793         }
1794
1795         while (j) {
1796                 (void)__mlx4_ib_destroy_flow(to_mdev(qp->device)->dev,
1797                                              mflow->reg_id[j].mirror);
1798                 j--;
1799         }
1800 err_free:
1801         kfree(mflow);
1802         return ERR_PTR(err);
1803 }
1804
1805 static int mlx4_ib_destroy_flow(struct ib_flow *flow_id)
1806 {
1807         int err, ret = 0;
1808         int i = 0;
1809         struct mlx4_ib_dev *mdev = to_mdev(flow_id->qp->device);
1810         struct mlx4_ib_flow *mflow = to_mflow(flow_id);
1811
1812         while (i < ARRAY_SIZE(mflow->reg_id) && mflow->reg_id[i].id) {
1813                 err = __mlx4_ib_destroy_flow(mdev->dev, mflow->reg_id[i].id);
1814                 if (err)
1815                         ret = err;
1816                 if (mflow->reg_id[i].mirror) {
1817                         err = __mlx4_ib_destroy_flow(mdev->dev,
1818                                                      mflow->reg_id[i].mirror);
1819                         if (err)
1820                                 ret = err;
1821                 }
1822                 i++;
1823         }
1824
1825         kfree(mflow);
1826         return ret;
1827 }
1828
1829 static int mlx4_ib_mcg_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
1830 {
1831         int err;
1832         struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1833         struct mlx4_dev *dev = mdev->dev;
1834         struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1835         struct mlx4_ib_steering *ib_steering = NULL;
1836         enum mlx4_protocol prot = MLX4_PROT_IB_IPV6;
1837         struct mlx4_flow_reg_id reg_id;
1838
1839         if (mdev->dev->caps.steering_mode ==
1840             MLX4_STEERING_MODE_DEVICE_MANAGED) {
1841                 ib_steering = kmalloc(sizeof(*ib_steering), GFP_KERNEL);
1842                 if (!ib_steering)
1843                         return -ENOMEM;
1844         }
1845
1846         err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw, mqp->port,
1847                                     !!(mqp->flags &
1848                                        MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
1849                                     prot, &reg_id.id);
1850         if (err) {
1851                 pr_err("multicast attach op failed, err %d\n", err);
1852                 goto err_malloc;
1853         }
1854
1855         reg_id.mirror = 0;
1856         if (mlx4_is_bonded(dev)) {
1857                 err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw,
1858                                             (mqp->port == 1) ? 2 : 1,
1859                                             !!(mqp->flags &
1860                                             MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
1861                                             prot, &reg_id.mirror);
1862                 if (err)
1863                         goto err_add;
1864         }
1865
1866         err = add_gid_entry(ibqp, gid);
1867         if (err)
1868                 goto err_add;
1869
1870         if (ib_steering) {
1871                 memcpy(ib_steering->gid.raw, gid->raw, 16);
1872                 ib_steering->reg_id = reg_id;
1873                 mutex_lock(&mqp->mutex);
1874                 list_add(&ib_steering->list, &mqp->steering_rules);
1875                 mutex_unlock(&mqp->mutex);
1876         }
1877         return 0;
1878
1879 err_add:
1880         mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1881                               prot, reg_id.id);
1882         if (reg_id.mirror)
1883                 mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1884                                       prot, reg_id.mirror);
1885 err_malloc:
1886         kfree(ib_steering);
1887
1888         return err;
1889 }
1890
1891 static struct mlx4_ib_gid_entry *find_gid_entry(struct mlx4_ib_qp *qp, u8 *raw)
1892 {
1893         struct mlx4_ib_gid_entry *ge;
1894         struct mlx4_ib_gid_entry *tmp;
1895         struct mlx4_ib_gid_entry *ret = NULL;
1896
1897         list_for_each_entry_safe(ge, tmp, &qp->gid_list, list) {
1898                 if (!memcmp(raw, ge->gid.raw, 16)) {
1899                         ret = ge;
1900                         break;
1901                 }
1902         }
1903
1904         return ret;
1905 }
1906
1907 static int mlx4_ib_mcg_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
1908 {
1909         int err;
1910         struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1911         struct mlx4_dev *dev = mdev->dev;
1912         struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1913         struct net_device *ndev;
1914         struct mlx4_ib_gid_entry *ge;
1915         struct mlx4_flow_reg_id reg_id = {0, 0};
1916         enum mlx4_protocol prot =  MLX4_PROT_IB_IPV6;
1917
1918         if (mdev->dev->caps.steering_mode ==
1919             MLX4_STEERING_MODE_DEVICE_MANAGED) {
1920                 struct mlx4_ib_steering *ib_steering;
1921
1922                 mutex_lock(&mqp->mutex);
1923                 list_for_each_entry(ib_steering, &mqp->steering_rules, list) {
1924                         if (!memcmp(ib_steering->gid.raw, gid->raw, 16)) {
1925                                 list_del(&ib_steering->list);
1926                                 break;
1927                         }
1928                 }
1929                 mutex_unlock(&mqp->mutex);
1930                 if (&ib_steering->list == &mqp->steering_rules) {
1931                         pr_err("Couldn't find reg_id for mgid. Steering rule is left attached\n");
1932                         return -EINVAL;
1933                 }
1934                 reg_id = ib_steering->reg_id;
1935                 kfree(ib_steering);
1936         }
1937
1938         err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1939                                     prot, reg_id.id);
1940         if (err)
1941                 return err;
1942
1943         if (mlx4_is_bonded(dev)) {
1944                 err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1945                                             prot, reg_id.mirror);
1946                 if (err)
1947                         return err;
1948         }
1949
1950         mutex_lock(&mqp->mutex);
1951         ge = find_gid_entry(mqp, gid->raw);
1952         if (ge) {
1953                 spin_lock_bh(&mdev->iboe.lock);
1954                 ndev = ge->added ? mdev->iboe.netdevs[ge->port - 1] : NULL;
1955                 dev_hold(ndev);
1956                 spin_unlock_bh(&mdev->iboe.lock);
1957                 dev_put(ndev);
1958                 list_del(&ge->list);
1959                 kfree(ge);
1960         } else
1961                 pr_warn("could not find mgid entry\n");
1962
1963         mutex_unlock(&mqp->mutex);
1964
1965         return 0;
1966 }
1967
1968 static int init_node_data(struct mlx4_ib_dev *dev)
1969 {
1970         struct ib_smp *in_mad;
1971         struct ib_smp *out_mad;
1972         int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
1973         int err = -ENOMEM;
1974
1975         in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
1976         out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
1977         if (!in_mad || !out_mad)
1978                 goto out;
1979
1980         ib_init_query_mad(in_mad);
1981         in_mad->attr_id = IB_SMP_ATTR_NODE_DESC;
1982         if (mlx4_is_master(dev->dev))
1983                 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
1984
1985         err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
1986         if (err)
1987                 goto out;
1988
1989         memcpy(dev->ib_dev.node_desc, out_mad->data, IB_DEVICE_NODE_DESC_MAX);
1990
1991         in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
1992
1993         err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
1994         if (err)
1995                 goto out;
1996
1997         dev->dev->rev_id = be32_to_cpup((__be32 *) (out_mad->data + 32));
1998         memcpy(&dev->ib_dev.node_guid, out_mad->data + 12, 8);
1999
2000 out:
2001         kfree(in_mad);
2002         kfree(out_mad);
2003         return err;
2004 }
2005
2006 static ssize_t hca_type_show(struct device *device,
2007                              struct device_attribute *attr, char *buf)
2008 {
2009         struct mlx4_ib_dev *dev =
2010                 rdma_device_to_drv_device(device, struct mlx4_ib_dev, ib_dev);
2011
2012         return sysfs_emit(buf, "MT%d\n", dev->dev->persist->pdev->device);
2013 }
2014 static DEVICE_ATTR_RO(hca_type);
2015
2016 static ssize_t hw_rev_show(struct device *device,
2017                            struct device_attribute *attr, char *buf)
2018 {
2019         struct mlx4_ib_dev *dev =
2020                 rdma_device_to_drv_device(device, struct mlx4_ib_dev, ib_dev);
2021
2022         return sysfs_emit(buf, "%x\n", dev->dev->rev_id);
2023 }
2024 static DEVICE_ATTR_RO(hw_rev);
2025
2026 static ssize_t board_id_show(struct device *device,
2027                              struct device_attribute *attr, char *buf)
2028 {
2029         struct mlx4_ib_dev *dev =
2030                 rdma_device_to_drv_device(device, struct mlx4_ib_dev, ib_dev);
2031
2032         return sysfs_emit(buf, "%.*s\n", MLX4_BOARD_ID_LEN, dev->dev->board_id);
2033 }
2034 static DEVICE_ATTR_RO(board_id);
2035
2036 static struct attribute *mlx4_class_attributes[] = {
2037         &dev_attr_hw_rev.attr,
2038         &dev_attr_hca_type.attr,
2039         &dev_attr_board_id.attr,
2040         NULL
2041 };
2042
2043 static const struct attribute_group mlx4_attr_group = {
2044         .attrs = mlx4_class_attributes,
2045 };
2046
2047 struct diag_counter {
2048         const char *name;
2049         u32 offset;
2050 };
2051
2052 #define DIAG_COUNTER(_name, _offset)                    \
2053         { .name = #_name, .offset = _offset }
2054
2055 static const struct diag_counter diag_basic[] = {
2056         DIAG_COUNTER(rq_num_lle, 0x00),
2057         DIAG_COUNTER(sq_num_lle, 0x04),
2058         DIAG_COUNTER(rq_num_lqpoe, 0x08),
2059         DIAG_COUNTER(sq_num_lqpoe, 0x0C),
2060         DIAG_COUNTER(rq_num_lpe, 0x18),
2061         DIAG_COUNTER(sq_num_lpe, 0x1C),
2062         DIAG_COUNTER(rq_num_wrfe, 0x20),
2063         DIAG_COUNTER(sq_num_wrfe, 0x24),
2064         DIAG_COUNTER(sq_num_mwbe, 0x2C),
2065         DIAG_COUNTER(sq_num_bre, 0x34),
2066         DIAG_COUNTER(sq_num_rire, 0x44),
2067         DIAG_COUNTER(rq_num_rire, 0x48),
2068         DIAG_COUNTER(sq_num_rae, 0x4C),
2069         DIAG_COUNTER(rq_num_rae, 0x50),
2070         DIAG_COUNTER(sq_num_roe, 0x54),
2071         DIAG_COUNTER(sq_num_tree, 0x5C),
2072         DIAG_COUNTER(sq_num_rree, 0x64),
2073         DIAG_COUNTER(rq_num_rnr, 0x68),
2074         DIAG_COUNTER(sq_num_rnr, 0x6C),
2075         DIAG_COUNTER(rq_num_oos, 0x100),
2076         DIAG_COUNTER(sq_num_oos, 0x104),
2077 };
2078
2079 static const struct diag_counter diag_ext[] = {
2080         DIAG_COUNTER(rq_num_dup, 0x130),
2081         DIAG_COUNTER(sq_num_to, 0x134),
2082 };
2083
2084 static const struct diag_counter diag_device_only[] = {
2085         DIAG_COUNTER(num_cqovf, 0x1A0),
2086         DIAG_COUNTER(rq_num_udsdprd, 0x118),
2087 };
2088
2089 static struct rdma_hw_stats *
2090 mlx4_ib_alloc_hw_device_stats(struct ib_device *ibdev)
2091 {
2092         struct mlx4_ib_dev *dev = to_mdev(ibdev);
2093         struct mlx4_ib_diag_counters *diag = dev->diag_counters;
2094
2095         if (!diag[0].descs)
2096                 return NULL;
2097
2098         return rdma_alloc_hw_stats_struct(diag[0].descs, diag[0].num_counters,
2099                                           RDMA_HW_STATS_DEFAULT_LIFESPAN);
2100 }
2101
2102 static struct rdma_hw_stats *
2103 mlx4_ib_alloc_hw_port_stats(struct ib_device *ibdev, u32 port_num)
2104 {
2105         struct mlx4_ib_dev *dev = to_mdev(ibdev);
2106         struct mlx4_ib_diag_counters *diag = dev->diag_counters;
2107
2108         if (!diag[1].descs)
2109                 return NULL;
2110
2111         return rdma_alloc_hw_stats_struct(diag[1].descs, diag[1].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                                 u32 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                                          struct rdma_stat_desc **pdescs,
2142                                          u32 **offset, u32 *num, bool port)
2143 {
2144         u32 num_counters;
2145
2146         num_counters = ARRAY_SIZE(diag_basic);
2147
2148         if (ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DIAG_PER_PORT)
2149                 num_counters += ARRAY_SIZE(diag_ext);
2150
2151         if (!port)
2152                 num_counters += ARRAY_SIZE(diag_device_only);
2153
2154         *pdescs = kcalloc(num_counters, sizeof(struct rdma_stat_desc),
2155                           GFP_KERNEL);
2156         if (!*pdescs)
2157                 return -ENOMEM;
2158
2159         *offset = kcalloc(num_counters, sizeof(**offset), GFP_KERNEL);
2160         if (!*offset)
2161                 goto err;
2162
2163         *num = num_counters;
2164
2165         return 0;
2166
2167 err:
2168         kfree(*pdescs);
2169         return -ENOMEM;
2170 }
2171
2172 static void mlx4_ib_fill_diag_counters(struct mlx4_ib_dev *ibdev,
2173                                        struct rdma_stat_desc *descs,
2174                                        u32 *offset, bool port)
2175 {
2176         int i;
2177         int j;
2178
2179         for (i = 0, j = 0; i < ARRAY_SIZE(diag_basic); i++, j++) {
2180                 descs[i].name = diag_basic[i].name;
2181                 offset[i] = diag_basic[i].offset;
2182         }
2183
2184         if (ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DIAG_PER_PORT) {
2185                 for (i = 0; i < ARRAY_SIZE(diag_ext); i++, j++) {
2186                         descs[j].name = diag_ext[i].name;
2187                         offset[j] = diag_ext[i].offset;
2188                 }
2189         }
2190
2191         if (!port) {
2192                 for (i = 0; i < ARRAY_SIZE(diag_device_only); i++, j++) {
2193                         descs[j].name = diag_device_only[i].name;
2194                         offset[j] = diag_device_only[i].offset;
2195                 }
2196         }
2197 }
2198
2199 static const struct ib_device_ops mlx4_ib_hw_stats_ops = {
2200         .alloc_hw_device_stats = mlx4_ib_alloc_hw_device_stats,
2201         .alloc_hw_port_stats = mlx4_ib_alloc_hw_port_stats,
2202         .get_hw_stats = mlx4_ib_get_hw_stats,
2203 };
2204
2205 static const struct ib_device_ops mlx4_ib_hw_stats_ops1 = {
2206         .alloc_hw_device_stats = mlx4_ib_alloc_hw_device_stats,
2207         .get_hw_stats = mlx4_ib_get_hw_stats,
2208 };
2209
2210 static int mlx4_ib_alloc_diag_counters(struct mlx4_ib_dev *ibdev)
2211 {
2212         struct mlx4_ib_diag_counters *diag = ibdev->diag_counters;
2213         int i;
2214         int ret;
2215         bool per_port = !!(ibdev->dev->caps.flags2 &
2216                 MLX4_DEV_CAP_FLAG2_DIAG_PER_PORT);
2217
2218         if (mlx4_is_slave(ibdev->dev))
2219                 return 0;
2220
2221         for (i = 0; i < MLX4_DIAG_COUNTERS_TYPES; i++) {
2222                 /*
2223                  * i == 1 means we are building port counters, set a different
2224                  * stats ops without port stats callback.
2225                  */
2226                 if (i && !per_port) {
2227                         ib_set_device_ops(&ibdev->ib_dev,
2228                                           &mlx4_ib_hw_stats_ops1);
2229
2230                         return 0;
2231                 }
2232
2233                 ret = __mlx4_ib_alloc_diag_counters(ibdev, &diag[i].descs,
2234                                                     &diag[i].offset,
2235                                                     &diag[i].num_counters, i);
2236                 if (ret)
2237                         goto err_alloc;
2238
2239                 mlx4_ib_fill_diag_counters(ibdev, diag[i].descs,
2240                                            diag[i].offset, i);
2241         }
2242
2243         ib_set_device_ops(&ibdev->ib_dev, &mlx4_ib_hw_stats_ops);
2244
2245         return 0;
2246
2247 err_alloc:
2248         if (i) {
2249                 kfree(diag[i - 1].descs);
2250                 kfree(diag[i - 1].offset);
2251         }
2252
2253         return ret;
2254 }
2255
2256 static void mlx4_ib_diag_cleanup(struct mlx4_ib_dev *ibdev)
2257 {
2258         int i;
2259
2260         for (i = 0; i < MLX4_DIAG_COUNTERS_TYPES; i++) {
2261                 kfree(ibdev->diag_counters[i].offset);
2262                 kfree(ibdev->diag_counters[i].descs);
2263         }
2264 }
2265
2266 #define MLX4_IB_INVALID_MAC     ((u64)-1)
2267 static void mlx4_ib_update_qps(struct mlx4_ib_dev *ibdev,
2268                                struct net_device *dev,
2269                                int port)
2270 {
2271         u64 new_smac = 0;
2272         u64 release_mac = MLX4_IB_INVALID_MAC;
2273         struct mlx4_ib_qp *qp;
2274
2275         new_smac = ether_addr_to_u64(dev->dev_addr);
2276         atomic64_set(&ibdev->iboe.mac[port - 1], new_smac);
2277
2278         /* no need for update QP1 and mac registration in non-SRIOV */
2279         if (!mlx4_is_mfunc(ibdev->dev))
2280                 return;
2281
2282         mutex_lock(&ibdev->qp1_proxy_lock[port - 1]);
2283         qp = ibdev->qp1_proxy[port - 1];
2284         if (qp) {
2285                 int new_smac_index;
2286                 u64 old_smac;
2287                 struct mlx4_update_qp_params update_params;
2288
2289                 mutex_lock(&qp->mutex);
2290                 old_smac = qp->pri.smac;
2291                 if (new_smac == old_smac)
2292                         goto unlock;
2293
2294                 new_smac_index = mlx4_register_mac(ibdev->dev, port, new_smac);
2295
2296                 if (new_smac_index < 0)
2297                         goto unlock;
2298
2299                 update_params.smac_index = new_smac_index;
2300                 if (mlx4_update_qp(ibdev->dev, qp->mqp.qpn, MLX4_UPDATE_QP_SMAC,
2301                                    &update_params)) {
2302                         release_mac = new_smac;
2303                         goto unlock;
2304                 }
2305                 /* if old port was zero, no mac was yet registered for this QP */
2306                 if (qp->pri.smac_port)
2307                         release_mac = old_smac;
2308                 qp->pri.smac = new_smac;
2309                 qp->pri.smac_port = port;
2310                 qp->pri.smac_index = new_smac_index;
2311         }
2312
2313 unlock:
2314         if (release_mac != MLX4_IB_INVALID_MAC)
2315                 mlx4_unregister_mac(ibdev->dev, port, release_mac);
2316         if (qp)
2317                 mutex_unlock(&qp->mutex);
2318         mutex_unlock(&ibdev->qp1_proxy_lock[port - 1]);
2319 }
2320
2321 static void mlx4_ib_scan_netdevs(struct mlx4_ib_dev *ibdev,
2322                                  struct net_device *dev,
2323                                  unsigned long event)
2324
2325 {
2326         struct mlx4_ib_iboe *iboe;
2327         int update_qps_port = -1;
2328         int port;
2329
2330         ASSERT_RTNL();
2331
2332         iboe = &ibdev->iboe;
2333
2334         spin_lock_bh(&iboe->lock);
2335         mlx4_foreach_ib_transport_port(port, ibdev->dev) {
2336
2337                 iboe->netdevs[port - 1] =
2338                         mlx4_get_protocol_dev(ibdev->dev, MLX4_PROT_ETH, port);
2339
2340                 if (dev == iboe->netdevs[port - 1] &&
2341                     (event == NETDEV_CHANGEADDR || event == NETDEV_REGISTER ||
2342                      event == NETDEV_UP || event == NETDEV_CHANGE))
2343                         update_qps_port = port;
2344
2345                 if (dev == iboe->netdevs[port - 1] &&
2346                     (event == NETDEV_UP || event == NETDEV_DOWN)) {
2347                         enum ib_port_state port_state;
2348                         struct ib_event ibev = { };
2349
2350                         if (ib_get_cached_port_state(&ibdev->ib_dev, port,
2351                                                      &port_state))
2352                                 continue;
2353
2354                         if (event == NETDEV_UP &&
2355                             (port_state != IB_PORT_ACTIVE ||
2356                              iboe->last_port_state[port - 1] != IB_PORT_DOWN))
2357                                 continue;
2358                         if (event == NETDEV_DOWN &&
2359                             (port_state != IB_PORT_DOWN ||
2360                              iboe->last_port_state[port - 1] != IB_PORT_ACTIVE))
2361                                 continue;
2362                         iboe->last_port_state[port - 1] = port_state;
2363
2364                         ibev.device = &ibdev->ib_dev;
2365                         ibev.element.port_num = port;
2366                         ibev.event = event == NETDEV_UP ? IB_EVENT_PORT_ACTIVE :
2367                                                           IB_EVENT_PORT_ERR;
2368                         ib_dispatch_event(&ibev);
2369                 }
2370
2371         }
2372         spin_unlock_bh(&iboe->lock);
2373
2374         if (update_qps_port > 0)
2375                 mlx4_ib_update_qps(ibdev, dev, update_qps_port);
2376 }
2377
2378 static int mlx4_ib_netdev_event(struct notifier_block *this,
2379                                 unsigned long event, void *ptr)
2380 {
2381         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
2382         struct mlx4_ib_dev *ibdev;
2383
2384         if (!net_eq(dev_net(dev), &init_net))
2385                 return NOTIFY_DONE;
2386
2387         ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb);
2388         mlx4_ib_scan_netdevs(ibdev, dev, event);
2389
2390         return NOTIFY_DONE;
2391 }
2392
2393 static void init_pkeys(struct mlx4_ib_dev *ibdev)
2394 {
2395         int port;
2396         int slave;
2397         int i;
2398
2399         if (mlx4_is_master(ibdev->dev)) {
2400                 for (slave = 0; slave <= ibdev->dev->persist->num_vfs;
2401                      ++slave) {
2402                         for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
2403                                 for (i = 0;
2404                                      i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
2405                                      ++i) {
2406                                         ibdev->pkeys.virt2phys_pkey[slave][port - 1][i] =
2407                                         /* master has the identity virt2phys pkey mapping */
2408                                                 (slave == mlx4_master_func_num(ibdev->dev) || !i) ? i :
2409                                                         ibdev->dev->phys_caps.pkey_phys_table_len[port] - 1;
2410                                         mlx4_sync_pkey_table(ibdev->dev, slave, port, i,
2411                                                              ibdev->pkeys.virt2phys_pkey[slave][port - 1][i]);
2412                                 }
2413                         }
2414                 }
2415                 /* initialize pkey cache */
2416                 for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
2417                         for (i = 0;
2418                              i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
2419                              ++i)
2420                                 ibdev->pkeys.phys_pkey_cache[port-1][i] =
2421                                         (i) ? 0 : 0xFFFF;
2422                 }
2423         }
2424 }
2425
2426 static void mlx4_ib_alloc_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
2427 {
2428         int i, j, eq = 0, total_eqs = 0;
2429
2430         ibdev->eq_table = kcalloc(dev->caps.num_comp_vectors,
2431                                   sizeof(ibdev->eq_table[0]), GFP_KERNEL);
2432         if (!ibdev->eq_table)
2433                 return;
2434
2435         for (i = 1; i <= dev->caps.num_ports; i++) {
2436                 for (j = 0; j < mlx4_get_eqs_per_port(dev, i);
2437                      j++, total_eqs++) {
2438                         if (i > 1 &&  mlx4_is_eq_shared(dev, total_eqs))
2439                                 continue;
2440                         ibdev->eq_table[eq] = total_eqs;
2441                         if (!mlx4_assign_eq(dev, i,
2442                                             &ibdev->eq_table[eq]))
2443                                 eq++;
2444                         else
2445                                 ibdev->eq_table[eq] = -1;
2446                 }
2447         }
2448
2449         for (i = eq; i < dev->caps.num_comp_vectors;
2450              ibdev->eq_table[i++] = -1)
2451                 ;
2452
2453         /* Advertise the new number of EQs to clients */
2454         ibdev->ib_dev.num_comp_vectors = eq;
2455 }
2456
2457 static void mlx4_ib_free_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
2458 {
2459         int i;
2460         int total_eqs = ibdev->ib_dev.num_comp_vectors;
2461
2462         /* no eqs were allocated */
2463         if (!ibdev->eq_table)
2464                 return;
2465
2466         /* Reset the advertised EQ number */
2467         ibdev->ib_dev.num_comp_vectors = 0;
2468
2469         for (i = 0; i < total_eqs; i++)
2470                 mlx4_release_eq(dev, ibdev->eq_table[i]);
2471
2472         kfree(ibdev->eq_table);
2473         ibdev->eq_table = NULL;
2474 }
2475
2476 static int mlx4_port_immutable(struct ib_device *ibdev, u32 port_num,
2477                                struct ib_port_immutable *immutable)
2478 {
2479         struct ib_port_attr attr;
2480         struct mlx4_ib_dev *mdev = to_mdev(ibdev);
2481         int err;
2482
2483         if (mlx4_ib_port_link_layer(ibdev, port_num) == IB_LINK_LAYER_INFINIBAND) {
2484                 immutable->core_cap_flags = RDMA_CORE_PORT_IBA_IB;
2485                 immutable->max_mad_size = IB_MGMT_MAD_SIZE;
2486         } else {
2487                 if (mdev->dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE)
2488                         immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE;
2489                 if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ROCE_V1_V2)
2490                         immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE |
2491                                 RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP;
2492                 immutable->core_cap_flags |= RDMA_CORE_PORT_RAW_PACKET;
2493                 if (immutable->core_cap_flags & (RDMA_CORE_PORT_IBA_ROCE |
2494                     RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP))
2495                         immutable->max_mad_size = IB_MGMT_MAD_SIZE;
2496         }
2497
2498         err = ib_query_port(ibdev, port_num, &attr);
2499         if (err)
2500                 return err;
2501
2502         immutable->pkey_tbl_len = attr.pkey_tbl_len;
2503         immutable->gid_tbl_len = attr.gid_tbl_len;
2504
2505         return 0;
2506 }
2507
2508 static void get_fw_ver_str(struct ib_device *device, char *str)
2509 {
2510         struct mlx4_ib_dev *dev =
2511                 container_of(device, struct mlx4_ib_dev, ib_dev);
2512         snprintf(str, IB_FW_VERSION_NAME_MAX, "%d.%d.%d",
2513                  (int) (dev->dev->caps.fw_ver >> 32),
2514                  (int) (dev->dev->caps.fw_ver >> 16) & 0xffff,
2515                  (int) dev->dev->caps.fw_ver & 0xffff);
2516 }
2517
2518 static const struct ib_device_ops mlx4_ib_dev_ops = {
2519         .owner = THIS_MODULE,
2520         .driver_id = RDMA_DRIVER_MLX4,
2521         .uverbs_abi_ver = MLX4_IB_UVERBS_ABI_VERSION,
2522
2523         .add_gid = mlx4_ib_add_gid,
2524         .alloc_mr = mlx4_ib_alloc_mr,
2525         .alloc_pd = mlx4_ib_alloc_pd,
2526         .alloc_ucontext = mlx4_ib_alloc_ucontext,
2527         .attach_mcast = mlx4_ib_mcg_attach,
2528         .create_ah = mlx4_ib_create_ah,
2529         .create_cq = mlx4_ib_create_cq,
2530         .create_qp = mlx4_ib_create_qp,
2531         .create_srq = mlx4_ib_create_srq,
2532         .dealloc_pd = mlx4_ib_dealloc_pd,
2533         .dealloc_ucontext = mlx4_ib_dealloc_ucontext,
2534         .del_gid = mlx4_ib_del_gid,
2535         .dereg_mr = mlx4_ib_dereg_mr,
2536         .destroy_ah = mlx4_ib_destroy_ah,
2537         .destroy_cq = mlx4_ib_destroy_cq,
2538         .destroy_qp = mlx4_ib_destroy_qp,
2539         .destroy_srq = mlx4_ib_destroy_srq,
2540         .detach_mcast = mlx4_ib_mcg_detach,
2541         .device_group = &mlx4_attr_group,
2542         .disassociate_ucontext = mlx4_ib_disassociate_ucontext,
2543         .drain_rq = mlx4_ib_drain_rq,
2544         .drain_sq = mlx4_ib_drain_sq,
2545         .get_dev_fw_str = get_fw_ver_str,
2546         .get_dma_mr = mlx4_ib_get_dma_mr,
2547         .get_link_layer = mlx4_ib_port_link_layer,
2548         .get_netdev = mlx4_ib_get_netdev,
2549         .get_port_immutable = mlx4_port_immutable,
2550         .map_mr_sg = mlx4_ib_map_mr_sg,
2551         .mmap = mlx4_ib_mmap,
2552         .modify_cq = mlx4_ib_modify_cq,
2553         .modify_device = mlx4_ib_modify_device,
2554         .modify_port = mlx4_ib_modify_port,
2555         .modify_qp = mlx4_ib_modify_qp,
2556         .modify_srq = mlx4_ib_modify_srq,
2557         .poll_cq = mlx4_ib_poll_cq,
2558         .post_recv = mlx4_ib_post_recv,
2559         .post_send = mlx4_ib_post_send,
2560         .post_srq_recv = mlx4_ib_post_srq_recv,
2561         .process_mad = mlx4_ib_process_mad,
2562         .query_ah = mlx4_ib_query_ah,
2563         .query_device = mlx4_ib_query_device,
2564         .query_gid = mlx4_ib_query_gid,
2565         .query_pkey = mlx4_ib_query_pkey,
2566         .query_port = mlx4_ib_query_port,
2567         .query_qp = mlx4_ib_query_qp,
2568         .query_srq = mlx4_ib_query_srq,
2569         .reg_user_mr = mlx4_ib_reg_user_mr,
2570         .req_notify_cq = mlx4_ib_arm_cq,
2571         .rereg_user_mr = mlx4_ib_rereg_user_mr,
2572         .resize_cq = mlx4_ib_resize_cq,
2573
2574         INIT_RDMA_OBJ_SIZE(ib_ah, mlx4_ib_ah, ibah),
2575         INIT_RDMA_OBJ_SIZE(ib_cq, mlx4_ib_cq, ibcq),
2576         INIT_RDMA_OBJ_SIZE(ib_pd, mlx4_ib_pd, ibpd),
2577         INIT_RDMA_OBJ_SIZE(ib_qp, mlx4_ib_qp, ibqp),
2578         INIT_RDMA_OBJ_SIZE(ib_srq, mlx4_ib_srq, ibsrq),
2579         INIT_RDMA_OBJ_SIZE(ib_ucontext, mlx4_ib_ucontext, ibucontext),
2580 };
2581
2582 static const struct ib_device_ops mlx4_ib_dev_wq_ops = {
2583         .create_rwq_ind_table = mlx4_ib_create_rwq_ind_table,
2584         .create_wq = mlx4_ib_create_wq,
2585         .destroy_rwq_ind_table = mlx4_ib_destroy_rwq_ind_table,
2586         .destroy_wq = mlx4_ib_destroy_wq,
2587         .modify_wq = mlx4_ib_modify_wq,
2588
2589         INIT_RDMA_OBJ_SIZE(ib_rwq_ind_table, mlx4_ib_rwq_ind_table,
2590                            ib_rwq_ind_tbl),
2591 };
2592
2593 static const struct ib_device_ops mlx4_ib_dev_mw_ops = {
2594         .alloc_mw = mlx4_ib_alloc_mw,
2595         .dealloc_mw = mlx4_ib_dealloc_mw,
2596
2597         INIT_RDMA_OBJ_SIZE(ib_mw, mlx4_ib_mw, ibmw),
2598 };
2599
2600 static const struct ib_device_ops mlx4_ib_dev_xrc_ops = {
2601         .alloc_xrcd = mlx4_ib_alloc_xrcd,
2602         .dealloc_xrcd = mlx4_ib_dealloc_xrcd,
2603
2604         INIT_RDMA_OBJ_SIZE(ib_xrcd, mlx4_ib_xrcd, ibxrcd),
2605 };
2606
2607 static const struct ib_device_ops mlx4_ib_dev_fs_ops = {
2608         .create_flow = mlx4_ib_create_flow,
2609         .destroy_flow = mlx4_ib_destroy_flow,
2610 };
2611
2612 static void *mlx4_ib_add(struct mlx4_dev *dev)
2613 {
2614         struct mlx4_ib_dev *ibdev;
2615         int num_ports = 0;
2616         int i, j;
2617         int err;
2618         struct mlx4_ib_iboe *iboe;
2619         int ib_num_ports = 0;
2620         int num_req_counters;
2621         int allocated;
2622         u32 counter_index;
2623         struct counter_index *new_counter_index;
2624
2625         pr_info_once("%s", mlx4_ib_version);
2626
2627         num_ports = 0;
2628         mlx4_foreach_ib_transport_port(i, dev)
2629                 num_ports++;
2630
2631         /* No point in registering a device with no ports... */
2632         if (num_ports == 0)
2633                 return NULL;
2634
2635         ibdev = ib_alloc_device(mlx4_ib_dev, ib_dev);
2636         if (!ibdev) {
2637                 dev_err(&dev->persist->pdev->dev,
2638                         "Device struct alloc failed\n");
2639                 return NULL;
2640         }
2641
2642         iboe = &ibdev->iboe;
2643
2644         if (mlx4_pd_alloc(dev, &ibdev->priv_pdn))
2645                 goto err_dealloc;
2646
2647         if (mlx4_uar_alloc(dev, &ibdev->priv_uar))
2648                 goto err_pd;
2649
2650         ibdev->uar_map = ioremap((phys_addr_t) ibdev->priv_uar.pfn << PAGE_SHIFT,
2651                                  PAGE_SIZE);
2652         if (!ibdev->uar_map)
2653                 goto err_uar;
2654         MLX4_INIT_DOORBELL_LOCK(&ibdev->uar_lock);
2655
2656         ibdev->dev = dev;
2657         ibdev->bond_next_port   = 0;
2658
2659         ibdev->ib_dev.node_type         = RDMA_NODE_IB_CA;
2660         ibdev->ib_dev.local_dma_lkey    = dev->caps.reserved_lkey;
2661         ibdev->num_ports                = num_ports;
2662         ibdev->ib_dev.phys_port_cnt     = mlx4_is_bonded(dev) ?
2663                                                 1 : ibdev->num_ports;
2664         ibdev->ib_dev.num_comp_vectors  = dev->caps.num_comp_vectors;
2665         ibdev->ib_dev.dev.parent        = &dev->persist->pdev->dev;
2666
2667         ib_set_device_ops(&ibdev->ib_dev, &mlx4_ib_dev_ops);
2668
2669         if ((dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_RSS) &&
2670             ((mlx4_ib_port_link_layer(&ibdev->ib_dev, 1) ==
2671             IB_LINK_LAYER_ETHERNET) ||
2672             (mlx4_ib_port_link_layer(&ibdev->ib_dev, 2) ==
2673             IB_LINK_LAYER_ETHERNET)))
2674                 ib_set_device_ops(&ibdev->ib_dev, &mlx4_ib_dev_wq_ops);
2675
2676         if (dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW ||
2677             dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN)
2678                 ib_set_device_ops(&ibdev->ib_dev, &mlx4_ib_dev_mw_ops);
2679
2680         if (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) {
2681                 ib_set_device_ops(&ibdev->ib_dev, &mlx4_ib_dev_xrc_ops);
2682         }
2683
2684         if (check_flow_steering_support(dev)) {
2685                 ibdev->steering_support = MLX4_STEERING_MODE_DEVICE_MANAGED;
2686                 ib_set_device_ops(&ibdev->ib_dev, &mlx4_ib_dev_fs_ops);
2687         }
2688
2689         if (!dev->caps.userspace_caps)
2690                 ibdev->ib_dev.ops.uverbs_abi_ver =
2691                         MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION;
2692
2693         mlx4_ib_alloc_eqs(dev, ibdev);
2694
2695         spin_lock_init(&iboe->lock);
2696
2697         if (init_node_data(ibdev))
2698                 goto err_map;
2699         mlx4_init_sl2vl_tbl(ibdev);
2700
2701         for (i = 0; i < ibdev->num_ports; ++i) {
2702                 mutex_init(&ibdev->counters_table[i].mutex);
2703                 INIT_LIST_HEAD(&ibdev->counters_table[i].counters_list);
2704                 iboe->last_port_state[i] = IB_PORT_DOWN;
2705         }
2706
2707         num_req_counters = mlx4_is_bonded(dev) ? 1 : ibdev->num_ports;
2708         for (i = 0; i < num_req_counters; ++i) {
2709                 mutex_init(&ibdev->qp1_proxy_lock[i]);
2710                 allocated = 0;
2711                 if (mlx4_ib_port_link_layer(&ibdev->ib_dev, i + 1) ==
2712                                                 IB_LINK_LAYER_ETHERNET) {
2713                         err = mlx4_counter_alloc(ibdev->dev, &counter_index,
2714                                                  MLX4_RES_USAGE_DRIVER);
2715                         /* if failed to allocate a new counter, use default */
2716                         if (err)
2717                                 counter_index =
2718                                         mlx4_get_default_counter_index(dev,
2719                                                                        i + 1);
2720                         else
2721                                 allocated = 1;
2722                 } else { /* IB_LINK_LAYER_INFINIBAND use the default counter */
2723                         counter_index = mlx4_get_default_counter_index(dev,
2724                                                                        i + 1);
2725                 }
2726                 new_counter_index = kmalloc(sizeof(*new_counter_index),
2727                                             GFP_KERNEL);
2728                 if (!new_counter_index) {
2729                         if (allocated)
2730                                 mlx4_counter_free(ibdev->dev, counter_index);
2731                         goto err_counter;
2732                 }
2733                 new_counter_index->index = counter_index;
2734                 new_counter_index->allocated = allocated;
2735                 list_add_tail(&new_counter_index->list,
2736                               &ibdev->counters_table[i].counters_list);
2737                 ibdev->counters_table[i].default_counter = counter_index;
2738                 pr_info("counter index %d for port %d allocated %d\n",
2739                         counter_index, i + 1, allocated);
2740         }
2741         if (mlx4_is_bonded(dev))
2742                 for (i = 1; i < ibdev->num_ports ; ++i) {
2743                         new_counter_index =
2744                                         kmalloc(sizeof(struct counter_index),
2745                                                 GFP_KERNEL);
2746                         if (!new_counter_index)
2747                                 goto err_counter;
2748                         new_counter_index->index = counter_index;
2749                         new_counter_index->allocated = 0;
2750                         list_add_tail(&new_counter_index->list,
2751                                       &ibdev->counters_table[i].counters_list);
2752                         ibdev->counters_table[i].default_counter =
2753                                                                 counter_index;
2754                 }
2755
2756         mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
2757                 ib_num_ports++;
2758
2759         spin_lock_init(&ibdev->sm_lock);
2760         mutex_init(&ibdev->cap_mask_mutex);
2761         INIT_LIST_HEAD(&ibdev->qp_list);
2762         spin_lock_init(&ibdev->reset_flow_resource_lock);
2763
2764         if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED &&
2765             ib_num_ports) {
2766                 ibdev->steer_qpn_count = MLX4_IB_UC_MAX_NUM_QPS;
2767                 err = mlx4_qp_reserve_range(dev, ibdev->steer_qpn_count,
2768                                             MLX4_IB_UC_STEER_QPN_ALIGN,
2769                                             &ibdev->steer_qpn_base, 0,
2770                                             MLX4_RES_USAGE_DRIVER);
2771                 if (err)
2772                         goto err_counter;
2773
2774                 ibdev->ib_uc_qpns_bitmap = bitmap_alloc(ibdev->steer_qpn_count,
2775                                                         GFP_KERNEL);
2776                 if (!ibdev->ib_uc_qpns_bitmap)
2777                         goto err_steer_qp_release;
2778
2779                 if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DMFS_IPOIB) {
2780                         bitmap_zero(ibdev->ib_uc_qpns_bitmap,
2781                                     ibdev->steer_qpn_count);
2782                         err = mlx4_FLOW_STEERING_IB_UC_QP_RANGE(
2783                                         dev, ibdev->steer_qpn_base,
2784                                         ibdev->steer_qpn_base +
2785                                         ibdev->steer_qpn_count - 1);
2786                         if (err)
2787                                 goto err_steer_free_bitmap;
2788                 } else {
2789                         bitmap_fill(ibdev->ib_uc_qpns_bitmap,
2790                                     ibdev->steer_qpn_count);
2791                 }
2792         }
2793
2794         for (j = 1; j <= ibdev->dev->caps.num_ports; j++)
2795                 atomic64_set(&iboe->mac[j - 1], ibdev->dev->caps.def_mac[j]);
2796
2797         if (mlx4_ib_alloc_diag_counters(ibdev))
2798                 goto err_steer_free_bitmap;
2799
2800         if (ib_register_device(&ibdev->ib_dev, "mlx4_%d",
2801                                &dev->persist->pdev->dev))
2802                 goto err_diag_counters;
2803
2804         if (mlx4_ib_mad_init(ibdev))
2805                 goto err_reg;
2806
2807         if (mlx4_ib_init_sriov(ibdev))
2808                 goto err_mad;
2809
2810         if (!iboe->nb.notifier_call) {
2811                 iboe->nb.notifier_call = mlx4_ib_netdev_event;
2812                 err = register_netdevice_notifier(&iboe->nb);
2813                 if (err) {
2814                         iboe->nb.notifier_call = NULL;
2815                         goto err_notif;
2816                 }
2817         }
2818         if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ROCE_V1_V2) {
2819                 err = mlx4_config_roce_v2_port(dev, ROCE_V2_UDP_DPORT);
2820                 if (err)
2821                         goto err_notif;
2822         }
2823
2824         ibdev->ib_active = true;
2825         mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
2826                 devlink_port_type_ib_set(mlx4_get_devlink_port(dev, i),
2827                                          &ibdev->ib_dev);
2828
2829         if (mlx4_is_mfunc(ibdev->dev))
2830                 init_pkeys(ibdev);
2831
2832         /* create paravirt contexts for any VFs which are active */
2833         if (mlx4_is_master(ibdev->dev)) {
2834                 for (j = 0; j < MLX4_MFUNC_MAX; j++) {
2835                         if (j == mlx4_master_func_num(ibdev->dev))
2836                                 continue;
2837                         if (mlx4_is_slave_active(ibdev->dev, j))
2838                                 do_slave_init(ibdev, j, 1);
2839                 }
2840         }
2841         return ibdev;
2842
2843 err_notif:
2844         if (ibdev->iboe.nb.notifier_call) {
2845                 if (unregister_netdevice_notifier(&ibdev->iboe.nb))
2846                         pr_warn("failure unregistering notifier\n");
2847                 ibdev->iboe.nb.notifier_call = NULL;
2848         }
2849         flush_workqueue(wq);
2850
2851         mlx4_ib_close_sriov(ibdev);
2852
2853 err_mad:
2854         mlx4_ib_mad_cleanup(ibdev);
2855
2856 err_reg:
2857         ib_unregister_device(&ibdev->ib_dev);
2858
2859 err_diag_counters:
2860         mlx4_ib_diag_cleanup(ibdev);
2861
2862 err_steer_free_bitmap:
2863         bitmap_free(ibdev->ib_uc_qpns_bitmap);
2864
2865 err_steer_qp_release:
2866         mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
2867                               ibdev->steer_qpn_count);
2868 err_counter:
2869         for (i = 0; i < ibdev->num_ports; ++i)
2870                 mlx4_ib_delete_counters_table(ibdev, &ibdev->counters_table[i]);
2871
2872 err_map:
2873         mlx4_ib_free_eqs(dev, ibdev);
2874         iounmap(ibdev->uar_map);
2875
2876 err_uar:
2877         mlx4_uar_free(dev, &ibdev->priv_uar);
2878
2879 err_pd:
2880         mlx4_pd_free(dev, ibdev->priv_pdn);
2881
2882 err_dealloc:
2883         ib_dealloc_device(&ibdev->ib_dev);
2884
2885         return NULL;
2886 }
2887
2888 int mlx4_ib_steer_qp_alloc(struct mlx4_ib_dev *dev, int count, int *qpn)
2889 {
2890         int offset;
2891
2892         WARN_ON(!dev->ib_uc_qpns_bitmap);
2893
2894         offset = bitmap_find_free_region(dev->ib_uc_qpns_bitmap,
2895                                          dev->steer_qpn_count,
2896                                          get_count_order(count));
2897         if (offset < 0)
2898                 return offset;
2899
2900         *qpn = dev->steer_qpn_base + offset;
2901         return 0;
2902 }
2903
2904 void mlx4_ib_steer_qp_free(struct mlx4_ib_dev *dev, u32 qpn, int count)
2905 {
2906         if (!qpn ||
2907             dev->steering_support != MLX4_STEERING_MODE_DEVICE_MANAGED)
2908                 return;
2909
2910         if (WARN(qpn < dev->steer_qpn_base, "qpn = %u, steer_qpn_base = %u\n",
2911                  qpn, dev->steer_qpn_base))
2912                 /* not supposed to be here */
2913                 return;
2914
2915         bitmap_release_region(dev->ib_uc_qpns_bitmap,
2916                               qpn - dev->steer_qpn_base,
2917                               get_count_order(count));
2918 }
2919
2920 int mlx4_ib_steer_qp_reg(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
2921                          int is_attach)
2922 {
2923         int err;
2924         size_t flow_size;
2925         struct ib_flow_attr *flow;
2926         struct ib_flow_spec_ib *ib_spec;
2927
2928         if (is_attach) {
2929                 flow_size = sizeof(struct ib_flow_attr) +
2930                             sizeof(struct ib_flow_spec_ib);
2931                 flow = kzalloc(flow_size, GFP_KERNEL);
2932                 if (!flow)
2933                         return -ENOMEM;
2934                 flow->port = mqp->port;
2935                 flow->num_of_specs = 1;
2936                 flow->size = flow_size;
2937                 ib_spec = (struct ib_flow_spec_ib *)(flow + 1);
2938                 ib_spec->type = IB_FLOW_SPEC_IB;
2939                 ib_spec->size = sizeof(struct ib_flow_spec_ib);
2940                 /* Add an empty rule for IB L2 */
2941                 memset(&ib_spec->mask, 0, sizeof(ib_spec->mask));
2942
2943                 err = __mlx4_ib_create_flow(&mqp->ibqp, flow, MLX4_DOMAIN_NIC,
2944                                             MLX4_FS_REGULAR, &mqp->reg_id);
2945                 kfree(flow);
2946                 return err;
2947         }
2948
2949         return __mlx4_ib_destroy_flow(mdev->dev, mqp->reg_id);
2950 }
2951
2952 static void mlx4_ib_remove(struct mlx4_dev *dev, void *ibdev_ptr)
2953 {
2954         struct mlx4_ib_dev *ibdev = ibdev_ptr;
2955         int p;
2956         int i;
2957
2958         mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
2959                 devlink_port_type_clear(mlx4_get_devlink_port(dev, i));
2960         ibdev->ib_active = false;
2961         flush_workqueue(wq);
2962
2963         if (ibdev->iboe.nb.notifier_call) {
2964                 if (unregister_netdevice_notifier(&ibdev->iboe.nb))
2965                         pr_warn("failure unregistering notifier\n");
2966                 ibdev->iboe.nb.notifier_call = NULL;
2967         }
2968
2969         mlx4_ib_close_sriov(ibdev);
2970         mlx4_ib_mad_cleanup(ibdev);
2971         ib_unregister_device(&ibdev->ib_dev);
2972         mlx4_ib_diag_cleanup(ibdev);
2973
2974         mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
2975                               ibdev->steer_qpn_count);
2976         bitmap_free(ibdev->ib_uc_qpns_bitmap);
2977
2978         iounmap(ibdev->uar_map);
2979         for (p = 0; p < ibdev->num_ports; ++p)
2980                 mlx4_ib_delete_counters_table(ibdev, &ibdev->counters_table[p]);
2981
2982         mlx4_foreach_port(p, dev, MLX4_PORT_TYPE_IB)
2983                 mlx4_CLOSE_PORT(dev, p);
2984
2985         mlx4_ib_free_eqs(dev, ibdev);
2986
2987         mlx4_uar_free(dev, &ibdev->priv_uar);
2988         mlx4_pd_free(dev, ibdev->priv_pdn);
2989         ib_dealloc_device(&ibdev->ib_dev);
2990 }
2991
2992 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init)
2993 {
2994         struct mlx4_ib_demux_work **dm;
2995         struct mlx4_dev *dev = ibdev->dev;
2996         int i;
2997         unsigned long flags;
2998         struct mlx4_active_ports actv_ports;
2999         unsigned int ports;
3000         unsigned int first_port;
3001
3002         if (!mlx4_is_master(dev))
3003                 return;
3004
3005         actv_ports = mlx4_get_active_ports(dev, slave);
3006         ports = bitmap_weight(actv_ports.ports, dev->caps.num_ports);
3007         first_port = find_first_bit(actv_ports.ports, dev->caps.num_ports);
3008
3009         dm = kcalloc(ports, sizeof(*dm), GFP_ATOMIC);
3010         if (!dm)
3011                 return;
3012
3013         for (i = 0; i < ports; i++) {
3014                 dm[i] = kmalloc(sizeof (struct mlx4_ib_demux_work), GFP_ATOMIC);
3015                 if (!dm[i]) {
3016                         while (--i >= 0)
3017                                 kfree(dm[i]);
3018                         goto out;
3019                 }
3020                 INIT_WORK(&dm[i]->work, mlx4_ib_tunnels_update_work);
3021                 dm[i]->port = first_port + i + 1;
3022                 dm[i]->slave = slave;
3023                 dm[i]->do_init = do_init;
3024                 dm[i]->dev = ibdev;
3025         }
3026         /* initialize or tear down tunnel QPs for the slave */
3027         spin_lock_irqsave(&ibdev->sriov.going_down_lock, flags);
3028         if (!ibdev->sriov.is_going_down) {
3029                 for (i = 0; i < ports; i++)
3030                         queue_work(ibdev->sriov.demux[i].ud_wq, &dm[i]->work);
3031                 spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
3032         } else {
3033                 spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
3034                 for (i = 0; i < ports; i++)
3035                         kfree(dm[i]);
3036         }
3037 out:
3038         kfree(dm);
3039         return;
3040 }
3041
3042 static void mlx4_ib_handle_catas_error(struct mlx4_ib_dev *ibdev)
3043 {
3044         struct mlx4_ib_qp *mqp;
3045         unsigned long flags_qp;
3046         unsigned long flags_cq;
3047         struct mlx4_ib_cq *send_mcq, *recv_mcq;
3048         struct list_head    cq_notify_list;
3049         struct mlx4_cq *mcq;
3050         unsigned long flags;
3051
3052         pr_warn("mlx4_ib_handle_catas_error was started\n");
3053         INIT_LIST_HEAD(&cq_notify_list);
3054
3055         /* Go over qp list reside on that ibdev, sync with create/destroy qp.*/
3056         spin_lock_irqsave(&ibdev->reset_flow_resource_lock, flags);
3057
3058         list_for_each_entry(mqp, &ibdev->qp_list, qps_list) {
3059                 spin_lock_irqsave(&mqp->sq.lock, flags_qp);
3060                 if (mqp->sq.tail != mqp->sq.head) {
3061                         send_mcq = to_mcq(mqp->ibqp.send_cq);
3062                         spin_lock_irqsave(&send_mcq->lock, flags_cq);
3063                         if (send_mcq->mcq.comp &&
3064                             mqp->ibqp.send_cq->comp_handler) {
3065                                 if (!send_mcq->mcq.reset_notify_added) {
3066                                         send_mcq->mcq.reset_notify_added = 1;
3067                                         list_add_tail(&send_mcq->mcq.reset_notify,
3068                                                       &cq_notify_list);
3069                                 }
3070                         }
3071                         spin_unlock_irqrestore(&send_mcq->lock, flags_cq);
3072                 }
3073                 spin_unlock_irqrestore(&mqp->sq.lock, flags_qp);
3074                 /* Now, handle the QP's receive queue */
3075                 spin_lock_irqsave(&mqp->rq.lock, flags_qp);
3076                 /* no handling is needed for SRQ */
3077                 if (!mqp->ibqp.srq) {
3078                         if (mqp->rq.tail != mqp->rq.head) {
3079                                 recv_mcq = to_mcq(mqp->ibqp.recv_cq);
3080                                 spin_lock_irqsave(&recv_mcq->lock, flags_cq);
3081                                 if (recv_mcq->mcq.comp &&
3082                                     mqp->ibqp.recv_cq->comp_handler) {
3083                                         if (!recv_mcq->mcq.reset_notify_added) {
3084                                                 recv_mcq->mcq.reset_notify_added = 1;
3085                                                 list_add_tail(&recv_mcq->mcq.reset_notify,
3086                                                               &cq_notify_list);
3087                                         }
3088                                 }
3089                                 spin_unlock_irqrestore(&recv_mcq->lock,
3090                                                        flags_cq);
3091                         }
3092                 }
3093                 spin_unlock_irqrestore(&mqp->rq.lock, flags_qp);
3094         }
3095
3096         list_for_each_entry(mcq, &cq_notify_list, reset_notify) {
3097                 mcq->comp(mcq);
3098         }
3099         spin_unlock_irqrestore(&ibdev->reset_flow_resource_lock, flags);
3100         pr_warn("mlx4_ib_handle_catas_error ended\n");
3101 }
3102
3103 static void handle_bonded_port_state_event(struct work_struct *work)
3104 {
3105         struct ib_event_work *ew =
3106                 container_of(work, struct ib_event_work, work);
3107         struct mlx4_ib_dev *ibdev = ew->ib_dev;
3108         enum ib_port_state bonded_port_state = IB_PORT_NOP;
3109         int i;
3110         struct ib_event ibev;
3111
3112         kfree(ew);
3113         spin_lock_bh(&ibdev->iboe.lock);
3114         for (i = 0; i < MLX4_MAX_PORTS; ++i) {
3115                 struct net_device *curr_netdev = ibdev->iboe.netdevs[i];
3116                 enum ib_port_state curr_port_state;
3117
3118                 if (!curr_netdev)
3119                         continue;
3120
3121                 curr_port_state =
3122                         (netif_running(curr_netdev) &&
3123                          netif_carrier_ok(curr_netdev)) ?
3124                         IB_PORT_ACTIVE : IB_PORT_DOWN;
3125
3126                 bonded_port_state = (bonded_port_state != IB_PORT_ACTIVE) ?
3127                         curr_port_state : IB_PORT_ACTIVE;
3128         }
3129         spin_unlock_bh(&ibdev->iboe.lock);
3130
3131         ibev.device = &ibdev->ib_dev;
3132         ibev.element.port_num = 1;
3133         ibev.event = (bonded_port_state == IB_PORT_ACTIVE) ?
3134                 IB_EVENT_PORT_ACTIVE : IB_EVENT_PORT_ERR;
3135
3136         ib_dispatch_event(&ibev);
3137 }
3138
3139 void mlx4_ib_sl2vl_update(struct mlx4_ib_dev *mdev, int port)
3140 {
3141         u64 sl2vl;
3142         int err;
3143
3144         err = mlx4_ib_query_sl2vl(&mdev->ib_dev, port, &sl2vl);
3145         if (err) {
3146                 pr_err("Unable to get current sl to vl mapping for port %d.  Using all zeroes (%d)\n",
3147                        port, err);
3148                 sl2vl = 0;
3149         }
3150         atomic64_set(&mdev->sl2vl[port - 1], sl2vl);
3151 }
3152
3153 static void ib_sl2vl_update_work(struct work_struct *work)
3154 {
3155         struct ib_event_work *ew = container_of(work, struct ib_event_work, work);
3156         struct mlx4_ib_dev *mdev = ew->ib_dev;
3157         int port = ew->port;
3158
3159         mlx4_ib_sl2vl_update(mdev, port);
3160
3161         kfree(ew);
3162 }
3163
3164 void mlx4_sched_ib_sl2vl_update_work(struct mlx4_ib_dev *ibdev,
3165                                      int port)
3166 {
3167         struct ib_event_work *ew;
3168
3169         ew = kmalloc(sizeof(*ew), GFP_ATOMIC);
3170         if (ew) {
3171                 INIT_WORK(&ew->work, ib_sl2vl_update_work);
3172                 ew->port = port;
3173                 ew->ib_dev = ibdev;
3174                 queue_work(wq, &ew->work);
3175         }
3176 }
3177
3178 static void mlx4_ib_event(struct mlx4_dev *dev, void *ibdev_ptr,
3179                           enum mlx4_dev_event event, unsigned long param)
3180 {
3181         struct ib_event ibev;
3182         struct mlx4_ib_dev *ibdev = to_mdev((struct ib_device *) ibdev_ptr);
3183         struct mlx4_eqe *eqe = NULL;
3184         struct ib_event_work *ew;
3185         int p = 0;
3186
3187         if (mlx4_is_bonded(dev) &&
3188             ((event == MLX4_DEV_EVENT_PORT_UP) ||
3189             (event == MLX4_DEV_EVENT_PORT_DOWN))) {
3190                 ew = kmalloc(sizeof(*ew), GFP_ATOMIC);
3191                 if (!ew)
3192                         return;
3193                 INIT_WORK(&ew->work, handle_bonded_port_state_event);
3194                 ew->ib_dev = ibdev;
3195                 queue_work(wq, &ew->work);
3196                 return;
3197         }
3198
3199         if (event == MLX4_DEV_EVENT_PORT_MGMT_CHANGE)
3200                 eqe = (struct mlx4_eqe *)param;
3201         else
3202                 p = (int) param;
3203
3204         switch (event) {
3205         case MLX4_DEV_EVENT_PORT_UP:
3206                 if (p > ibdev->num_ports)
3207                         return;
3208                 if (!mlx4_is_slave(dev) &&
3209                     rdma_port_get_link_layer(&ibdev->ib_dev, p) ==
3210                         IB_LINK_LAYER_INFINIBAND) {
3211                         if (mlx4_is_master(dev))
3212                                 mlx4_ib_invalidate_all_guid_record(ibdev, p);
3213                         if (ibdev->dev->flags & MLX4_FLAG_SECURE_HOST &&
3214                             !(ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_SL_TO_VL_CHANGE_EVENT))
3215                                 mlx4_sched_ib_sl2vl_update_work(ibdev, p);
3216                 }
3217                 ibev.event = IB_EVENT_PORT_ACTIVE;
3218                 break;
3219
3220         case MLX4_DEV_EVENT_PORT_DOWN:
3221                 if (p > ibdev->num_ports)
3222                         return;
3223                 ibev.event = IB_EVENT_PORT_ERR;
3224                 break;
3225
3226         case MLX4_DEV_EVENT_CATASTROPHIC_ERROR:
3227                 ibdev->ib_active = false;
3228                 ibev.event = IB_EVENT_DEVICE_FATAL;
3229                 mlx4_ib_handle_catas_error(ibdev);
3230                 break;
3231
3232         case MLX4_DEV_EVENT_PORT_MGMT_CHANGE:
3233                 ew = kmalloc(sizeof *ew, GFP_ATOMIC);
3234                 if (!ew)
3235                         return;
3236
3237                 INIT_WORK(&ew->work, handle_port_mgmt_change_event);
3238                 memcpy(&ew->ib_eqe, eqe, sizeof *eqe);
3239                 ew->ib_dev = ibdev;
3240                 /* need to queue only for port owner, which uses GEN_EQE */
3241                 if (mlx4_is_master(dev))
3242                         queue_work(wq, &ew->work);
3243                 else
3244                         handle_port_mgmt_change_event(&ew->work);
3245                 return;
3246
3247         case MLX4_DEV_EVENT_SLAVE_INIT:
3248                 /* here, p is the slave id */
3249                 do_slave_init(ibdev, p, 1);
3250                 if (mlx4_is_master(dev)) {
3251                         int i;
3252
3253                         for (i = 1; i <= ibdev->num_ports; i++) {
3254                                 if (rdma_port_get_link_layer(&ibdev->ib_dev, i)
3255                                         == IB_LINK_LAYER_INFINIBAND)
3256                                         mlx4_ib_slave_alias_guid_event(ibdev,
3257                                                                        p, i,
3258                                                                        1);
3259                         }
3260                 }
3261                 return;
3262
3263         case MLX4_DEV_EVENT_SLAVE_SHUTDOWN:
3264                 if (mlx4_is_master(dev)) {
3265                         int i;
3266
3267                         for (i = 1; i <= ibdev->num_ports; i++) {
3268                                 if (rdma_port_get_link_layer(&ibdev->ib_dev, i)
3269                                         == IB_LINK_LAYER_INFINIBAND)
3270                                         mlx4_ib_slave_alias_guid_event(ibdev,
3271                                                                        p, i,
3272                                                                        0);
3273                         }
3274                 }
3275                 /* here, p is the slave id */
3276                 do_slave_init(ibdev, p, 0);
3277                 return;
3278
3279         default:
3280                 return;
3281         }
3282
3283         ibev.device           = ibdev_ptr;
3284         ibev.element.port_num = mlx4_is_bonded(ibdev->dev) ? 1 : (u8)p;
3285
3286         ib_dispatch_event(&ibev);
3287 }
3288
3289 static struct mlx4_interface mlx4_ib_interface = {
3290         .add            = mlx4_ib_add,
3291         .remove         = mlx4_ib_remove,
3292         .event          = mlx4_ib_event,
3293         .protocol       = MLX4_PROT_IB_IPV6,
3294         .flags          = MLX4_INTFF_BONDING
3295 };
3296
3297 static int __init mlx4_ib_init(void)
3298 {
3299         int err;
3300
3301         wq = alloc_ordered_workqueue("mlx4_ib", WQ_MEM_RECLAIM);
3302         if (!wq)
3303                 return -ENOMEM;
3304
3305         err = mlx4_ib_qp_event_init();
3306         if (err)
3307                 goto clean_qp_event;
3308
3309         err = mlx4_ib_cm_init();
3310         if (err)
3311                 goto clean_wq;
3312
3313         err = mlx4_ib_mcg_init();
3314         if (err)
3315                 goto clean_cm;
3316
3317         err = mlx4_register_interface(&mlx4_ib_interface);
3318         if (err)
3319                 goto clean_mcg;
3320
3321         return 0;
3322
3323 clean_mcg:
3324         mlx4_ib_mcg_destroy();
3325
3326 clean_cm:
3327         mlx4_ib_cm_destroy();
3328
3329 clean_wq:
3330         mlx4_ib_qp_event_cleanup();
3331
3332 clean_qp_event:
3333         destroy_workqueue(wq);
3334         return err;
3335 }
3336
3337 static void __exit mlx4_ib_cleanup(void)
3338 {
3339         mlx4_unregister_interface(&mlx4_ib_interface);
3340         mlx4_ib_mcg_destroy();
3341         mlx4_ib_cm_destroy();
3342         mlx4_ib_qp_event_cleanup();
3343         destroy_workqueue(wq);
3344 }
3345
3346 module_init(mlx4_ib_init);
3347 module_exit(mlx4_ib_cleanup);