drivers/infiniband/hw/mlx4/main.c

   1 /*
   2  * Copyright (c) 2006, 2007 Cisco Systems, Inc. All rights reserved.
   3  * Copyright (c) 2007, 2008 Mellanox Technologies. All rights reserved.
   4  *
   5  * This software is available to you under a choice of one of two
   6  * licenses.  You may choose to be licensed under the terms of the GNU
   7  * General Public License (GPL) Version 2, available from the file
   8  * COPYING in the main directory of this source tree, or the
   9  * OpenIB.org BSD license below:
  10  *
  11  *     Redistribution and use in source and binary forms, with or
  12  *     without modification, are permitted provided that the following
  13  *     conditions are met:
  14  *
  15  *      - Redistributions of source code must retain the above
  16  *        copyright notice, this list of conditions and the following
  17  *        disclaimer.
  18  *
  19  *      - Redistributions in binary form must reproduce the above
  20  *        copyright notice, this list of conditions and the following
  21  *        disclaimer in the documentation and/or other materials
  22  *        provided with the distribution.
  23  *
  24  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
  25  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  26  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
  27  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
  28  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
  29  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
  30  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  31  * SOFTWARE.
  32  */
  33
  34 #include <linux/module.h>
  35 #include <linux/init.h>
  36 #include <linux/slab.h>
  37 #include <linux/errno.h>
  38 #include <linux/netdevice.h>
  39 #include <linux/inetdevice.h>
  40 #include <linux/rtnetlink.h>
  41 #include <linux/if_vlan.h>
  42 #include <linux/sched/mm.h>
  43 #include <linux/sched/task.h>
  44
  45 #include <net/ipv6.h>
  46 #include <net/addrconf.h>
  47 #include <net/devlink.h>
  48
  49 #include <rdma/ib_smi.h>
  50 #include <rdma/ib_user_verbs.h>
  51 #include <rdma/ib_addr.h>
  52 #include <rdma/ib_cache.h>
  53
  54 #include <net/bonding.h>
  55
  56 #include <linux/mlx4/driver.h>
  57 #include <linux/mlx4/cmd.h>
  58 #include <linux/mlx4/qp.h>
  59
  60 #include "mlx4_ib.h"
  61 #include <rdma/mlx4-abi.h>
  62
  63 #define DRV_NAME        MLX4_IB_DRV_NAME
  64 #define DRV_VERSION     "4.0-0"
  65
  66 #define MLX4_IB_FLOW_MAX_PRIO 0xFFF
  67 #define MLX4_IB_FLOW_QPN_MASK 0xFFFFFF
  68 #define MLX4_IB_CARD_REV_A0   0xA0
  69
  70 MODULE_AUTHOR("Roland Dreier");
  71 MODULE_DESCRIPTION("Mellanox ConnectX HCA InfiniBand driver");
  72 MODULE_LICENSE("Dual BSD/GPL");
  73
  74 int mlx4_ib_sm_guid_assign = 0;
  75 module_param_named(sm_guid_assign, mlx4_ib_sm_guid_assign, int, 0444);
  76 MODULE_PARM_DESC(sm_guid_assign, "Enable SM alias_GUID assignment if sm_guid_assign > 0 (Default: 0)");
  77
  78 static const char mlx4_ib_version[] =
  79         DRV_NAME ": Mellanox ConnectX InfiniBand driver v"
  80         DRV_VERSION "\n";
  81
  82 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init);
  83 static enum rdma_link_layer mlx4_ib_port_link_layer(struct ib_device *device,
  84                                                     u8 port_num);
  85
  86 static struct workqueue_struct *wq;
  87
  88 static void init_query_mad(struct ib_smp *mad)
  89 {
  90         mad->base_version  = 1;
  91         mad->mgmt_class    = IB_MGMT_CLASS_SUBN_LID_ROUTED;
  92         mad->class_version = 1;
  93         mad->method        = IB_MGMT_METHOD_GET;
  94 }
  95
  96 static int check_flow_steering_support(struct mlx4_dev *dev)
  97 {
  98         int eth_num_ports = 0;
  99         int ib_num_ports = 0;
 100
 101         int dmfs = dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED;
 102
 103         if (dmfs) {
 104                 int i;
 105                 mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_ETH)
 106                         eth_num_ports++;
 107                 mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
 108                         ib_num_ports++;
 109                 dmfs &= (!ib_num_ports ||
 110                          (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DMFS_IPOIB)) &&
 111                         (!eth_num_ports ||
 112                          (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_FS_EN));
 113                 if (ib_num_ports && mlx4_is_mfunc(dev)) {
 114                         pr_warn("Device managed flow steering is unavailable for IB port in multifunction env.\n");
 115                         dmfs = 0;
 116                 }
 117         }
 118         return dmfs;
 119 }
 120
 121 static int num_ib_ports(struct mlx4_dev *dev)
 122 {
 123         int ib_ports = 0;
 124         int i;
 125
 126         mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
 127                 ib_ports++;
 128
 129         return ib_ports;
 130 }
 131
 132 static struct net_device *mlx4_ib_get_netdev(struct ib_device *device, u8 port_num)
 133 {
 134         struct mlx4_ib_dev *ibdev = to_mdev(device);
 135         struct net_device *dev;
 136
 137         rcu_read_lock();
 138         dev = mlx4_get_protocol_dev(ibdev->dev, MLX4_PROT_ETH, port_num);
 139
 140         if (dev) {
 141                 if (mlx4_is_bonded(ibdev->dev)) {
 142                         struct net_device *upper = NULL;
 143
 144                         upper = netdev_master_upper_dev_get_rcu(dev);
 145                         if (upper) {
 146                                 struct net_device *active;
 147
 148                                 active = bond_option_active_slave_get_rcu(netdev_priv(upper));
 149                                 if (active)
 150                                         dev = active;
 151                         }
 152                 }
 153         }
 154         if (dev)
 155                 dev_hold(dev);
 156
 157         rcu_read_unlock();
 158         return dev;
 159 }
 160
 161 static int mlx4_ib_update_gids_v1(struct gid_entry *gids,
 162                                   struct mlx4_ib_dev *ibdev,
 163                                   u8 port_num)
 164 {
 165         struct mlx4_cmd_mailbox *mailbox;
 166         int err;
 167         struct mlx4_dev *dev = ibdev->dev;
 168         int i;
 169         union ib_gid *gid_tbl;
 170
 171         mailbox = mlx4_alloc_cmd_mailbox(dev);
 172         if (IS_ERR(mailbox))
 173                 return -ENOMEM;
 174
 175         gid_tbl = mailbox->buf;
 176
 177         for (i = 0; i < MLX4_MAX_PORT_GIDS; ++i)
 178                 memcpy(&gid_tbl[i], &gids[i].gid, sizeof(union ib_gid));
 179
 180         err = mlx4_cmd(dev, mailbox->dma,
 181                        MLX4_SET_PORT_GID_TABLE << 8 | port_num,
 182                        1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
 183                        MLX4_CMD_WRAPPED);
 184         if (mlx4_is_bonded(dev))
 185                 err += mlx4_cmd(dev, mailbox->dma,
 186                                 MLX4_SET_PORT_GID_TABLE << 8 | 2,
 187                                 1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
 188                                 MLX4_CMD_WRAPPED);
 189
 190         mlx4_free_cmd_mailbox(dev, mailbox);
 191         return err;
 192 }
 193
 194 static int mlx4_ib_update_gids_v1_v2(struct gid_entry *gids,
 195                                      struct mlx4_ib_dev *ibdev,
 196                                      u8 port_num)
 197 {
 198         struct mlx4_cmd_mailbox *mailbox;
 199         int err;
 200         struct mlx4_dev *dev = ibdev->dev;
 201         int i;
 202         struct {
 203                 union ib_gid    gid;
 204                 __be32          rsrvd1[2];
 205                 __be16          rsrvd2;
 206                 u8              type;
 207                 u8              version;
 208                 __be32          rsrvd3;
 209         } *gid_tbl;
 210
 211         mailbox = mlx4_alloc_cmd_mailbox(dev);
 212         if (IS_ERR(mailbox))
 213                 return -ENOMEM;
 214
 215         gid_tbl = mailbox->buf;
 216         for (i = 0; i < MLX4_MAX_PORT_GIDS; ++i) {
 217                 memcpy(&gid_tbl[i].gid, &gids[i].gid, sizeof(union ib_gid));
 218                 if (gids[i].gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP) {
 219                         gid_tbl[i].version = 2;
 220                         if (!ipv6_addr_v4mapped((struct in6_addr *)&gids[i].gid))
 221                                 gid_tbl[i].type = 1;
 222                 }
 223         }
 224
 225         err = mlx4_cmd(dev, mailbox->dma,
 226                        MLX4_SET_PORT_ROCE_ADDR << 8 | port_num,
 227                        1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
 228                        MLX4_CMD_WRAPPED);
 229         if (mlx4_is_bonded(dev))
 230                 err += mlx4_cmd(dev, mailbox->dma,
 231                                 MLX4_SET_PORT_ROCE_ADDR << 8 | 2,
 232                                 1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
 233                                 MLX4_CMD_WRAPPED);
 234
 235         mlx4_free_cmd_mailbox(dev, mailbox);
 236         return err;
 237 }
 238
 239 static int mlx4_ib_update_gids(struct gid_entry *gids,
 240                                struct mlx4_ib_dev *ibdev,
 241                                u8 port_num)
 242 {
 243         if (ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ROCE_V1_V2)
 244                 return mlx4_ib_update_gids_v1_v2(gids, ibdev, port_num);
 245
 246         return mlx4_ib_update_gids_v1(gids, ibdev, port_num);
 247 }
 248
 249 static void free_gid_entry(struct gid_entry *entry)
 250 {
 251         memset(&entry->gid, 0, sizeof(entry->gid));
 252         kfree(entry->ctx);
 253         entry->ctx = NULL;
 254 }
 255
 256 static int mlx4_ib_add_gid(const struct ib_gid_attr *attr, void **context)
 257 {
 258         struct mlx4_ib_dev *ibdev = to_mdev(attr->device);
 259         struct mlx4_ib_iboe *iboe = &ibdev->iboe;
 260         struct mlx4_port_gid_table   *port_gid_table;
 261         int free = -1, found = -1;
 262         int ret = 0;
 263         int hw_update = 0;
 264         int i;
 265         struct gid_entry *gids = NULL;
 266         u16 vlan_id = 0xffff;
 267         u8 mac[ETH_ALEN];
 268
 269         if (!rdma_cap_roce_gid_table(attr->device, attr->port_num))
 270                 return -EINVAL;
 271
 272         if (attr->port_num > MLX4_MAX_PORTS)
 273                 return -EINVAL;
 274
 275         if (!context)
 276                 return -EINVAL;
 277
 278         ret = rdma_read_gid_l2_fields(attr, &vlan_id, &mac[0]);
 279         if (ret)
 280                 return ret;
 281         port_gid_table = &iboe->gids[attr->port_num - 1];
 282         spin_lock_bh(&iboe->lock);
 283         for (i = 0; i < MLX4_MAX_PORT_GIDS; ++i) {
 284                 if (!memcmp(&port_gid_table->gids[i].gid,
 285                             &attr->gid, sizeof(attr->gid)) &&
 286                     port_gid_table->gids[i].gid_type == attr->gid_type &&
 287                     port_gid_table->gids[i].vlan_id == vlan_id)  {
 288                         found = i;
 289                         break;
 290                 }
 291                 if (free < 0 && rdma_is_zero_gid(&port_gid_table->gids[i].gid))
 292                         free = i; /* HW has space */
 293         }
 294
 295         if (found < 0) {
 296                 if (free < 0) {
 297                         ret = -ENOSPC;
 298                 } else {
 299                         port_gid_table->gids[free].ctx = kmalloc(sizeof(*port_gid_table->gids[free].ctx), GFP_ATOMIC);
 300                         if (!port_gid_table->gids[free].ctx) {
 301                                 ret = -ENOMEM;
 302                         } else {
 303                                 *context = port_gid_table->gids[free].ctx;
 304                                 memcpy(&port_gid_table->gids[free].gid,
 305                                        &attr->gid, sizeof(attr->gid));
 306                                 port_gid_table->gids[free].gid_type = attr->gid_type;
 307                                 port_gid_table->gids[free].vlan_id = vlan_id;
 308                                 port_gid_table->gids[free].ctx->real_index = free;
 309                                 port_gid_table->gids[free].ctx->refcount = 1;
 310                                 hw_update = 1;
 311                         }
 312                 }
 313         } else {
 314                 struct gid_cache_context *ctx = port_gid_table->gids[found].ctx;
 315                 *context = ctx;
 316                 ctx->refcount++;
 317         }
 318         if (!ret && hw_update) {
 319                 gids = kmalloc_array(MLX4_MAX_PORT_GIDS, sizeof(*gids),
 320                                      GFP_ATOMIC);
 321                 if (!gids) {
 322                         ret = -ENOMEM;
 323                         *context = NULL;
 324                         free_gid_entry(&port_gid_table->gids[free]);
 325                 } else {
 326                         for (i = 0; i < MLX4_MAX_PORT_GIDS; i++) {
 327                                 memcpy(&gids[i].gid, &port_gid_table->gids[i].gid, sizeof(union ib_gid));
 328                                 gids[i].gid_type = port_gid_table->gids[i].gid_type;
 329                         }
 330                 }
 331         }
 332         spin_unlock_bh(&iboe->lock);
 333
 334         if (!ret && hw_update) {
 335                 ret = mlx4_ib_update_gids(gids, ibdev, attr->port_num);
 336                 if (ret) {
 337                         spin_lock_bh(&iboe->lock);
 338                         *context = NULL;
 339                         free_gid_entry(&port_gid_table->gids[free]);
 340                         spin_unlock_bh(&iboe->lock);
 341                 }
 342                 kfree(gids);
 343         }
 344
 345         return ret;
 346 }
 347
 348 static int mlx4_ib_del_gid(const struct ib_gid_attr *attr, void **context)
 349 {
 350         struct gid_cache_context *ctx = *context;
 351         struct mlx4_ib_dev *ibdev = to_mdev(attr->device);
 352         struct mlx4_ib_iboe *iboe = &ibdev->iboe;
 353         struct mlx4_port_gid_table   *port_gid_table;
 354         int ret = 0;
 355         int hw_update = 0;
 356         struct gid_entry *gids = NULL;
 357
 358         if (!rdma_cap_roce_gid_table(attr->device, attr->port_num))
 359                 return -EINVAL;
 360
 361         if (attr->port_num > MLX4_MAX_PORTS)
 362                 return -EINVAL;
 363
 364         port_gid_table = &iboe->gids[attr->port_num - 1];
 365         spin_lock_bh(&iboe->lock);
 366         if (ctx) {
 367                 ctx->refcount--;
 368                 if (!ctx->refcount) {
 369                         unsigned int real_index = ctx->real_index;
 370
 371                         free_gid_entry(&port_gid_table->gids[real_index]);
 372                         hw_update = 1;
 373                 }
 374         }
 375         if (!ret && hw_update) {
 376                 int i;
 377
 378                 gids = kmalloc_array(MLX4_MAX_PORT_GIDS, sizeof(*gids),
 379                                      GFP_ATOMIC);
 380                 if (!gids) {
 381                         ret = -ENOMEM;
 382                 } else {
 383                         for (i = 0; i < MLX4_MAX_PORT_GIDS; i++) {
 384                                 memcpy(&gids[i].gid,
 385                                        &port_gid_table->gids[i].gid,
 386                                        sizeof(union ib_gid));
 387                                 gids[i].gid_type =
 388                                     port_gid_table->gids[i].gid_type;
 389                         }
 390                 }
 391         }
 392         spin_unlock_bh(&iboe->lock);
 393
 394         if (!ret && hw_update) {
 395                 ret = mlx4_ib_update_gids(gids, ibdev, attr->port_num);
 396                 kfree(gids);
 397         }
 398         return ret;
 399 }
 400
 401 int mlx4_ib_gid_index_to_real_index(struct mlx4_ib_dev *ibdev,
 402                                     const struct ib_gid_attr *attr)
 403 {
 404         struct mlx4_ib_iboe *iboe = &ibdev->iboe;
 405         struct gid_cache_context *ctx = NULL;
 406         struct mlx4_port_gid_table   *port_gid_table;
 407         int real_index = -EINVAL;
 408         int i;
 409         unsigned long flags;
 410         u8 port_num = attr->port_num;
 411
 412         if (port_num > MLX4_MAX_PORTS)
 413                 return -EINVAL;
 414
 415         if (mlx4_is_bonded(ibdev->dev))
 416                 port_num = 1;
 417
 418         if (!rdma_cap_roce_gid_table(&ibdev->ib_dev, port_num))
 419                 return attr->index;
 420
 421         spin_lock_irqsave(&iboe->lock, flags);
 422         port_gid_table = &iboe->gids[port_num - 1];
 423
 424         for (i = 0; i < MLX4_MAX_PORT_GIDS; ++i)
 425                 if (!memcmp(&port_gid_table->gids[i].gid,
 426                             &attr->gid, sizeof(attr->gid)) &&
 427                     attr->gid_type == port_gid_table->gids[i].gid_type) {
 428                         ctx = port_gid_table->gids[i].ctx;
 429                         break;
 430                 }
 431         if (ctx)
 432                 real_index = ctx->real_index;
 433         spin_unlock_irqrestore(&iboe->lock, flags);
 434         return real_index;
 435 }
 436
 437 static int mlx4_ib_query_device(struct ib_device *ibdev,
 438                                 struct ib_device_attr *props,
 439                                 struct ib_udata *uhw)
 440 {
 441         struct mlx4_ib_dev *dev = to_mdev(ibdev);
 442         struct ib_smp *in_mad  = NULL;
 443         struct ib_smp *out_mad = NULL;
 444         int err;
 445         int have_ib_ports;
 446         struct mlx4_uverbs_ex_query_device cmd;
 447         struct mlx4_uverbs_ex_query_device_resp resp = {};
 448         struct mlx4_clock_params clock_params;
 449
 450         if (uhw->inlen) {
 451                 if (uhw->inlen < sizeof(cmd))
 452                         return -EINVAL;
 453
 454                 err = ib_copy_from_udata(&cmd, uhw, sizeof(cmd));
 455                 if (err)
 456                         return err;
 457
 458                 if (cmd.comp_mask)
 459                         return -EINVAL;
 460
 461                 if (cmd.reserved)
 462                         return -EINVAL;
 463         }
 464
 465         resp.response_length = offsetof(typeof(resp), response_length) +
 466                 sizeof(resp.response_length);
 467         in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
 468         out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
 469         err = -ENOMEM;
 470         if (!in_mad || !out_mad)
 471                 goto out;
 472
 473         init_query_mad(in_mad);
 474         in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
 475
 476         err = mlx4_MAD_IFC(to_mdev(ibdev), MLX4_MAD_IFC_IGNORE_KEYS,
 477                            1, NULL, NULL, in_mad, out_mad);
 478         if (err)
 479                 goto out;
 480
 481         memset(props, 0, sizeof *props);
 482
 483         have_ib_ports = num_ib_ports(dev->dev);
 484
 485         props->fw_ver = dev->dev->caps.fw_ver;
 486         props->device_cap_flags    = IB_DEVICE_CHANGE_PHY_PORT |
 487                 IB_DEVICE_PORT_ACTIVE_EVENT             |
 488                 IB_DEVICE_SYS_IMAGE_GUID                |
 489                 IB_DEVICE_RC_RNR_NAK_GEN                |
 490                 IB_DEVICE_BLOCK_MULTICAST_LOOPBACK;
 491         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_PKEY_CNTR)
 492                 props->device_cap_flags |= IB_DEVICE_BAD_PKEY_CNTR;
 493         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_QKEY_CNTR)
 494                 props->device_cap_flags |= IB_DEVICE_BAD_QKEY_CNTR;
 495         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_APM && have_ib_ports)
 496                 props->device_cap_flags |= IB_DEVICE_AUTO_PATH_MIG;
 497         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_UD_AV_PORT)
 498                 props->device_cap_flags |= IB_DEVICE_UD_AV_PORT_ENFORCE;
 499         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_IPOIB_CSUM)
 500                 props->device_cap_flags |= IB_DEVICE_UD_IP_CSUM;
 501         if (dev->dev->caps.max_gso_sz &&
 502             (dev->dev->rev_id != MLX4_IB_CARD_REV_A0) &&
 503             (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BLH))
 504                 props->device_cap_flags |= IB_DEVICE_UD_TSO;
 505         if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_RESERVED_LKEY)
 506                 props->device_cap_flags |= IB_DEVICE_LOCAL_DMA_LKEY;
 507         if ((dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_LOCAL_INV) &&
 508             (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_REMOTE_INV) &&
 509             (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_FAST_REG_WR))
 510                 props->device_cap_flags |= IB_DEVICE_MEM_MGT_EXTENSIONS;
 511         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC)
 512                 props->device_cap_flags |= IB_DEVICE_XRC;
 513         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW)
 514                 props->device_cap_flags |= IB_DEVICE_MEM_WINDOW;
 515         if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
 516                 if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_WIN_TYPE_2B)
 517                         props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2B;
 518                 else
 519                         props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2A;
 520         }
 521         if (dev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED)
 522                 props->device_cap_flags |= IB_DEVICE_MANAGED_FLOW_STEERING;
 523
 524         props->device_cap_flags |= IB_DEVICE_RAW_IP_CSUM;
 525
 526         props->vendor_id           = be32_to_cpup((__be32 *) (out_mad->data + 36)) &
 527                 0xffffff;
 528         props->vendor_part_id      = dev->dev->persist->pdev->device;
 529         props->hw_ver              = be32_to_cpup((__be32 *) (out_mad->data + 32));
 530         memcpy(&props->sys_image_guid, out_mad->data +  4, 8);
 531
 532         props->max_mr_size         = ~0ull;
 533         props->page_size_cap       = dev->dev->caps.page_size_cap;
 534         props->max_qp              = dev->dev->quotas.qp;
 535         props->max_qp_wr           = dev->dev->caps.max_wqes - MLX4_IB_SQ_MAX_SPARE;
 536         props->max_send_sge =
 537                 min(dev->dev->caps.max_sq_sg, dev->dev->caps.max_rq_sg);
 538         props->max_recv_sge =
 539                 min(dev->dev->caps.max_sq_sg, dev->dev->caps.max_rq_sg);
 540         props->max_sge_rd = MLX4_MAX_SGE_RD;
 541         props->max_cq              = dev->dev->quotas.cq;
 542         props->max_cqe             = dev->dev->caps.max_cqes;
 543         props->max_mr              = dev->dev->quotas.mpt;
 544         props->max_pd              = dev->dev->caps.num_pds - dev->dev->caps.reserved_pds;
 545         props->max_qp_rd_atom      = dev->dev->caps.max_qp_dest_rdma;
 546         props->max_qp_init_rd_atom = dev->dev->caps.max_qp_init_rdma;
 547         props->max_res_rd_atom     = props->max_qp_rd_atom * props->max_qp;
 548         props->max_srq             = dev->dev->quotas.srq;
 549         props->max_srq_wr          = dev->dev->caps.max_srq_wqes - 1;
 550         props->max_srq_sge         = dev->dev->caps.max_srq_sge;
 551         props->max_fast_reg_page_list_len = MLX4_MAX_FAST_REG_PAGES;
 552         props->local_ca_ack_delay  = dev->dev->caps.local_ca_ack_delay;
 553         props->atomic_cap          = dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_ATOMIC ?
 554                 IB_ATOMIC_HCA : IB_ATOMIC_NONE;
 555         props->masked_atomic_cap   = props->atomic_cap;
 556         props->max_pkeys           = dev->dev->caps.pkey_table_len[1];
 557         props->max_mcast_grp       = dev->dev->caps.num_mgms + dev->dev->caps.num_amgms;
 558         props->max_mcast_qp_attach = dev->dev->caps.num_qp_per_mgm;
 559         props->max_total_mcast_qp_attach = props->max_mcast_qp_attach *
 560                                            props->max_mcast_grp;
 561         props->max_map_per_fmr = dev->dev->caps.max_fmr_maps;
 562         props->hca_core_clock = dev->dev->caps.hca_core_clock * 1000UL;
 563         props->timestamp_mask = 0xFFFFFFFFFFFFULL;
 564         props->max_ah = INT_MAX;
 565
 566         if (mlx4_ib_port_link_layer(ibdev, 1) == IB_LINK_LAYER_ETHERNET ||
 567             mlx4_ib_port_link_layer(ibdev, 2) == IB_LINK_LAYER_ETHERNET) {
 568                 if (dev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_RSS) {
 569                         props->rss_caps.max_rwq_indirection_tables =
 570                                 props->max_qp;
 571                         props->rss_caps.max_rwq_indirection_table_size =
 572                                 dev->dev->caps.max_rss_tbl_sz;
 573                         props->rss_caps.supported_qpts = 1 << IB_QPT_RAW_PACKET;
 574                         props->max_wq_type_rq = props->max_qp;
 575                 }
 576
 577                 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_FCS_KEEP)
 578                         props->raw_packet_caps |= IB_RAW_PACKET_CAP_SCATTER_FCS;
 579         }
 580
 581         props->cq_caps.max_cq_moderation_count = MLX4_MAX_CQ_COUNT;
 582         props->cq_caps.max_cq_moderation_period = MLX4_MAX_CQ_PERIOD;
 583
 584         if (!mlx4_is_slave(dev->dev))
 585                 err = mlx4_get_internal_clock_params(dev->dev, &clock_params);
 586
 587         if (uhw->outlen >= resp.response_length + sizeof(resp.hca_core_clock_offset)) {
 588                 resp.response_length += sizeof(resp.hca_core_clock_offset);
 589                 if (!err && !mlx4_is_slave(dev->dev)) {
 590                         resp.comp_mask |= MLX4_IB_QUERY_DEV_RESP_MASK_CORE_CLOCK_OFFSET;
 591                         resp.hca_core_clock_offset = clock_params.offset % PAGE_SIZE;
 592                 }
 593         }
 594
 595         if (uhw->outlen >= resp.response_length +
 596             sizeof(resp.max_inl_recv_sz)) {
 597                 resp.response_length += sizeof(resp.max_inl_recv_sz);
 598                 resp.max_inl_recv_sz  = dev->dev->caps.max_rq_sg *
 599                         sizeof(struct mlx4_wqe_data_seg);
 600         }
 601
 602         if (offsetofend(typeof(resp), rss_caps) <= uhw->outlen) {
 603                 if (props->rss_caps.supported_qpts) {
 604                         resp.rss_caps.rx_hash_function =
 605                                 MLX4_IB_RX_HASH_FUNC_TOEPLITZ;
 606
 607                         resp.rss_caps.rx_hash_fields_mask =
 608                                 MLX4_IB_RX_HASH_SRC_IPV4 |
 609                                 MLX4_IB_RX_HASH_DST_IPV4 |
 610                                 MLX4_IB_RX_HASH_SRC_IPV6 |
 611                                 MLX4_IB_RX_HASH_DST_IPV6 |
 612                                 MLX4_IB_RX_HASH_SRC_PORT_TCP |
 613                                 MLX4_IB_RX_HASH_DST_PORT_TCP |
 614                                 MLX4_IB_RX_HASH_SRC_PORT_UDP |
 615                                 MLX4_IB_RX_HASH_DST_PORT_UDP;
 616
 617                         if (dev->dev->caps.tunnel_offload_mode ==
 618                             MLX4_TUNNEL_OFFLOAD_MODE_VXLAN)
 619                                 resp.rss_caps.rx_hash_fields_mask |=
 620                                         MLX4_IB_RX_HASH_INNER;
 621                 }
 622                 resp.response_length = offsetof(typeof(resp), rss_caps) +
 623                                        sizeof(resp.rss_caps);
 624         }
 625
 626         if (offsetofend(typeof(resp), tso_caps) <= uhw->outlen) {
 627                 if (dev->dev->caps.max_gso_sz &&
 628                     ((mlx4_ib_port_link_layer(ibdev, 1) ==
 629                     IB_LINK_LAYER_ETHERNET) ||
 630                     (mlx4_ib_port_link_layer(ibdev, 2) ==
 631                     IB_LINK_LAYER_ETHERNET))) {
 632                         resp.tso_caps.max_tso = dev->dev->caps.max_gso_sz;
 633                         resp.tso_caps.supported_qpts |=
 634                                 1 << IB_QPT_RAW_PACKET;
 635                 }
 636                 resp.response_length = offsetof(typeof(resp), tso_caps) +
 637                                        sizeof(resp.tso_caps);
 638         }
 639
 640         if (uhw->outlen) {
 641                 err = ib_copy_to_udata(uhw, &resp, resp.response_length);
 642                 if (err)
 643                         goto out;
 644         }
 645 out:
 646         kfree(in_mad);
 647         kfree(out_mad);
 648
 649         return err;
 650 }
 651
 652 static enum rdma_link_layer
 653 mlx4_ib_port_link_layer(struct ib_device *device, u8 port_num)
 654 {
 655         struct mlx4_dev *dev = to_mdev(device)->dev;
 656
 657         return dev->caps.port_mask[port_num] == MLX4_PORT_TYPE_IB ?
 658                 IB_LINK_LAYER_INFINIBAND : IB_LINK_LAYER_ETHERNET;
 659 }
 660
 661 static int ib_link_query_port(struct ib_device *ibdev, u8 port,
 662                               struct ib_port_attr *props, int netw_view)
 663 {
 664         struct ib_smp *in_mad  = NULL;
 665         struct ib_smp *out_mad = NULL;
 666         int ext_active_speed;
 667         int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
 668         int err = -ENOMEM;
 669
 670         in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
 671         out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
 672         if (!in_mad || !out_mad)
 673                 goto out;
 674
 675         init_query_mad(in_mad);
 676         in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
 677         in_mad->attr_mod = cpu_to_be32(port);
 678
 679         if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
 680                 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
 681
 682         err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
 683                                 in_mad, out_mad);
 684         if (err)
 685                 goto out;
 686
 687
 688         props->lid              = be16_to_cpup((__be16 *) (out_mad->data + 16));
 689         props->lmc              = out_mad->data[34] & 0x7;
 690         props->sm_lid           = be16_to_cpup((__be16 *) (out_mad->data + 18));
 691         props->sm_sl            = out_mad->data[36] & 0xf;
 692         props->state            = out_mad->data[32] & 0xf;
 693         props->phys_state       = out_mad->data[33] >> 4;
 694         props->port_cap_flags   = be32_to_cpup((__be32 *) (out_mad->data + 20));
 695         if (netw_view)
 696                 props->gid_tbl_len = out_mad->data[50];
 697         else
 698                 props->gid_tbl_len = to_mdev(ibdev)->dev->caps.gid_table_len[port];
 699         props->max_msg_sz       = to_mdev(ibdev)->dev->caps.max_msg_sz;
 700         props->pkey_tbl_len     = to_mdev(ibdev)->dev->caps.pkey_table_len[port];
 701         props->bad_pkey_cntr    = be16_to_cpup((__be16 *) (out_mad->data + 46));
 702         props->qkey_viol_cntr   = be16_to_cpup((__be16 *) (out_mad->data + 48));
 703         props->active_width     = out_mad->data[31] & 0xf;
 704         props->active_speed     = out_mad->data[35] >> 4;
 705         props->max_mtu          = out_mad->data[41] & 0xf;
 706         props->active_mtu       = out_mad->data[36] >> 4;
 707         props->subnet_timeout   = out_mad->data[51] & 0x1f;
 708         props->max_vl_num       = out_mad->data[37] >> 4;
 709         props->init_type_reply  = out_mad->data[41] >> 4;
 710
 711         /* Check if extended speeds (EDR/FDR/...) are supported */
 712         if (props->port_cap_flags & IB_PORT_EXTENDED_SPEEDS_SUP) {
 713                 ext_active_speed = out_mad->data[62] >> 4;
 714
 715                 switch (ext_active_speed) {
 716                 case 1:
 717                         props->active_speed = IB_SPEED_FDR;
 718                         break;
 719                 case 2:
 720                         props->active_speed = IB_SPEED_EDR;
 721                         break;
 722                 }
 723         }
 724
 725         /* If reported active speed is QDR, check if is FDR-10 */
 726         if (props->active_speed == IB_SPEED_QDR) {
 727                 init_query_mad(in_mad);
 728                 in_mad->attr_id = MLX4_ATTR_EXTENDED_PORT_INFO;
 729                 in_mad->attr_mod = cpu_to_be32(port);
 730
 731                 err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port,
 732                                    NULL, NULL, in_mad, out_mad);
 733                 if (err)
 734                         goto out;
 735
 736                 /* Checking LinkSpeedActive for FDR-10 */
 737                 if (out_mad->data[15] & 0x1)
 738                         props->active_speed = IB_SPEED_FDR10;
 739         }
 740
 741         /* Avoid wrong speed value returned by FW if the IB link is down. */
 742         if (props->state == IB_PORT_DOWN)
 743                  props->active_speed = IB_SPEED_SDR;
 744
 745 out:
 746         kfree(in_mad);
 747         kfree(out_mad);
 748         return err;
 749 }
 750
 751 static u8 state_to_phys_state(enum ib_port_state state)
 752 {
 753         return state == IB_PORT_ACTIVE ?
 754                 IB_PORT_PHYS_STATE_LINK_UP : IB_PORT_PHYS_STATE_DISABLED;
 755 }
 756
 757 static int eth_link_query_port(struct ib_device *ibdev, u8 port,
 758                                struct ib_port_attr *props)
 759 {
 760
 761         struct mlx4_ib_dev *mdev = to_mdev(ibdev);
 762         struct mlx4_ib_iboe *iboe = &mdev->iboe;
 763         struct net_device *ndev;
 764         enum ib_mtu tmp;
 765         struct mlx4_cmd_mailbox *mailbox;
 766         int err = 0;
 767         int is_bonded = mlx4_is_bonded(mdev->dev);
 768
 769         mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
 770         if (IS_ERR(mailbox))
 771                 return PTR_ERR(mailbox);
 772
 773         err = mlx4_cmd_box(mdev->dev, 0, mailbox->dma, port, 0,
 774                            MLX4_CMD_QUERY_PORT, MLX4_CMD_TIME_CLASS_B,
 775                            MLX4_CMD_WRAPPED);
 776         if (err)
 777                 goto out;
 778
 779         props->active_width     =  (((u8 *)mailbox->buf)[5] == 0x40) ||
 780                                    (((u8 *)mailbox->buf)[5] == 0x20 /*56Gb*/) ?
 781                                            IB_WIDTH_4X : IB_WIDTH_1X;
 782         props->active_speed     =  (((u8 *)mailbox->buf)[5] == 0x20 /*56Gb*/) ?
 783                                            IB_SPEED_FDR : IB_SPEED_QDR;
 784         props->port_cap_flags   = IB_PORT_CM_SUP;
 785         props->ip_gids = true;
 786         props->gid_tbl_len      = mdev->dev->caps.gid_table_len[port];
 787         props->max_msg_sz       = mdev->dev->caps.max_msg_sz;
 788         props->pkey_tbl_len     = 1;
 789         props->max_mtu          = IB_MTU_4096;
 790         props->max_vl_num       = 2;
 791         props->state            = IB_PORT_DOWN;
 792         props->phys_state       = state_to_phys_state(props->state);
 793         props->active_mtu       = IB_MTU_256;
 794         spin_lock_bh(&iboe->lock);
 795         ndev = iboe->netdevs[port - 1];
 796         if (ndev && is_bonded) {
 797                 rcu_read_lock(); /* required to get upper dev */
 798                 ndev = netdev_master_upper_dev_get_rcu(ndev);
 799                 rcu_read_unlock();
 800         }
 801         if (!ndev)
 802                 goto out_unlock;
 803
 804         tmp = iboe_get_mtu(ndev->mtu);
 805         props->active_mtu = tmp ? min(props->max_mtu, tmp) : IB_MTU_256;
 806
 807         props->state            = (netif_running(ndev) && netif_carrier_ok(ndev)) ?
 808                                         IB_PORT_ACTIVE : IB_PORT_DOWN;
 809         props->phys_state       = state_to_phys_state(props->state);
 810 out_unlock:
 811         spin_unlock_bh(&iboe->lock);
 812 out:
 813         mlx4_free_cmd_mailbox(mdev->dev, mailbox);
 814         return err;
 815 }
 816
 817 int __mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
 818                          struct ib_port_attr *props, int netw_view)
 819 {
 820         int err;
 821
 822         /* props being zeroed by the caller, avoid zeroing it here */
 823
 824         err = mlx4_ib_port_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND ?
 825                 ib_link_query_port(ibdev, port, props, netw_view) :
 826                                 eth_link_query_port(ibdev, port, props);
 827
 828         return err;
 829 }
 830
 831 static int mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
 832                               struct ib_port_attr *props)
 833 {
 834         /* returns host view */
 835         return __mlx4_ib_query_port(ibdev, port, props, 0);
 836 }
 837
 838 int __mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
 839                         union ib_gid *gid, int netw_view)
 840 {
 841         struct ib_smp *in_mad  = NULL;
 842         struct ib_smp *out_mad = NULL;
 843         int err = -ENOMEM;
 844         struct mlx4_ib_dev *dev = to_mdev(ibdev);
 845         int clear = 0;
 846         int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
 847
 848         in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
 849         out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
 850         if (!in_mad || !out_mad)
 851                 goto out;
 852
 853         init_query_mad(in_mad);
 854         in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
 855         in_mad->attr_mod = cpu_to_be32(port);
 856
 857         if (mlx4_is_mfunc(dev->dev) && netw_view)
 858                 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
 859
 860         err = mlx4_MAD_IFC(dev, mad_ifc_flags, port, NULL, NULL, in_mad, out_mad);
 861         if (err)
 862                 goto out;
 863
 864         memcpy(gid->raw, out_mad->data + 8, 8);
 865
 866         if (mlx4_is_mfunc(dev->dev) && !netw_view) {
 867                 if (index) {
 868                         /* For any index > 0, return the null guid */
 869                         err = 0;
 870                         clear = 1;
 871                         goto out;
 872                 }
 873         }
 874
 875         init_query_mad(in_mad);
 876         in_mad->attr_id  = IB_SMP_ATTR_GUID_INFO;
 877         in_mad->attr_mod = cpu_to_be32(index / 8);
 878
 879         err = mlx4_MAD_IFC(dev, mad_ifc_flags, port,
 880                            NULL, NULL, in_mad, out_mad);
 881         if (err)
 882                 goto out;
 883
 884         memcpy(gid->raw + 8, out_mad->data + (index % 8) * 8, 8);
 885
 886 out:
 887         if (clear)
 888                 memset(gid->raw + 8, 0, 8);
 889         kfree(in_mad);
 890         kfree(out_mad);
 891         return err;
 892 }
 893
 894 static int mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
 895                              union ib_gid *gid)
 896 {
 897         if (rdma_protocol_ib(ibdev, port))
 898                 return __mlx4_ib_query_gid(ibdev, port, index, gid, 0);
 899         return 0;
 900 }
 901
 902 static int mlx4_ib_query_sl2vl(struct ib_device *ibdev, u8 port, u64 *sl2vl_tbl)
 903 {
 904         union sl2vl_tbl_to_u64 sl2vl64;
 905         struct ib_smp *in_mad  = NULL;
 906         struct ib_smp *out_mad = NULL;
 907         int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
 908         int err = -ENOMEM;
 909         int jj;
 910
 911         if (mlx4_is_slave(to_mdev(ibdev)->dev)) {
 912                 *sl2vl_tbl = 0;
 913                 return 0;
 914         }
 915
 916         in_mad  = kzalloc(sizeof(*in_mad), GFP_KERNEL);
 917         out_mad = kmalloc(sizeof(*out_mad), GFP_KERNEL);
 918         if (!in_mad || !out_mad)
 919                 goto out;
 920
 921         init_query_mad(in_mad);
 922         in_mad->attr_id  = IB_SMP_ATTR_SL_TO_VL_TABLE;
 923         in_mad->attr_mod = 0;
 924
 925         if (mlx4_is_mfunc(to_mdev(ibdev)->dev))
 926                 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
 927
 928         err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
 929                            in_mad, out_mad);
 930         if (err)
 931                 goto out;
 932
 933         for (jj = 0; jj < 8; jj++)
 934                 sl2vl64.sl8[jj] = ((struct ib_smp *)out_mad)->data[jj];
 935         *sl2vl_tbl = sl2vl64.sl64;
 936
 937 out:
 938         kfree(in_mad);
 939         kfree(out_mad);
 940         return err;
 941 }
 942
 943 static void mlx4_init_sl2vl_tbl(struct mlx4_ib_dev *mdev)
 944 {
 945         u64 sl2vl;
 946         int i;
 947         int err;
 948
 949         for (i = 1; i <= mdev->dev->caps.num_ports; i++) {
 950                 if (mdev->dev->caps.port_type[i] == MLX4_PORT_TYPE_ETH)
 951                         continue;
 952                 err = mlx4_ib_query_sl2vl(&mdev->ib_dev, i, &sl2vl);
 953                 if (err) {
 954                         pr_err("Unable to get default sl to vl mapping for port %d.  Using all zeroes (%d)\n",
 955                                i, err);
 956                         sl2vl = 0;
 957                 }
 958                 atomic64_set(&mdev->sl2vl[i - 1], sl2vl);
 959         }
 960 }
 961
 962 int __mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index,
 963                          u16 *pkey, int netw_view)
 964 {
 965         struct ib_smp *in_mad  = NULL;
 966         struct ib_smp *out_mad = NULL;
 967         int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
 968         int err = -ENOMEM;
 969
 970         in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
 971         out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
 972         if (!in_mad || !out_mad)
 973                 goto out;
 974
 975         init_query_mad(in_mad);
 976         in_mad->attr_id  = IB_SMP_ATTR_PKEY_TABLE;
 977         in_mad->attr_mod = cpu_to_be32(index / 32);
 978
 979         if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
 980                 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
 981
 982         err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
 983                            in_mad, out_mad);
 984         if (err)
 985                 goto out;
 986
 987         *pkey = be16_to_cpu(((__be16 *) out_mad->data)[index % 32]);
 988
 989 out:
 990         kfree(in_mad);
 991         kfree(out_mad);
 992         return err;
 993 }
 994
 995 static int mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index, u16 *pkey)
 996 {
 997         return __mlx4_ib_query_pkey(ibdev, port, index, pkey, 0);
 998 }
 999
1000 static int mlx4_ib_modify_device(struct ib_device *ibdev, int mask,
1001                                  struct ib_device_modify *props)
1002 {
1003         struct mlx4_cmd_mailbox *mailbox;
1004         unsigned long flags;
1005
1006         if (mask & ~IB_DEVICE_MODIFY_NODE_DESC)
1007                 return -EOPNOTSUPP;
1008
1009         if (!(mask & IB_DEVICE_MODIFY_NODE_DESC))
1010                 return 0;
1011
1012         if (mlx4_is_slave(to_mdev(ibdev)->dev))
1013                 return -EOPNOTSUPP;
1014
1015         spin_lock_irqsave(&to_mdev(ibdev)->sm_lock, flags);
1016         memcpy(ibdev->node_desc, props->node_desc, IB_DEVICE_NODE_DESC_MAX);
1017         spin_unlock_irqrestore(&to_mdev(ibdev)->sm_lock, flags);
1018
1019         /*
1020          * If possible, pass node desc to FW, so it can generate
1021          * a 144 trap.  If cmd fails, just ignore.
1022          */
1023         mailbox = mlx4_alloc_cmd_mailbox(to_mdev(ibdev)->dev);
1024         if (IS_ERR(mailbox))
1025                 return 0;
1026
1027         memcpy(mailbox->buf, props->node_desc, IB_DEVICE_NODE_DESC_MAX);
1028         mlx4_cmd(to_mdev(ibdev)->dev, mailbox->dma, 1, 0,
1029                  MLX4_CMD_SET_NODE, MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
1030
1031         mlx4_free_cmd_mailbox(to_mdev(ibdev)->dev, mailbox);
1032
1033         return 0;
1034 }
1035
1036 static int mlx4_ib_SET_PORT(struct mlx4_ib_dev *dev, u8 port, int reset_qkey_viols,
1037                             u32 cap_mask)
1038 {
1039         struct mlx4_cmd_mailbox *mailbox;
1040         int err;
1041
1042         mailbox = mlx4_alloc_cmd_mailbox(dev->dev);
1043         if (IS_ERR(mailbox))
1044                 return PTR_ERR(mailbox);
1045
1046         if (dev->dev->flags & MLX4_FLAG_OLD_PORT_CMDS) {
1047                 *(u8 *) mailbox->buf         = !!reset_qkey_viols << 6;
1048                 ((__be32 *) mailbox->buf)[2] = cpu_to_be32(cap_mask);
1049         } else {
1050                 ((u8 *) mailbox->buf)[3]     = !!reset_qkey_viols;
1051                 ((__be32 *) mailbox->buf)[1] = cpu_to_be32(cap_mask);
1052         }
1053
1054         err = mlx4_cmd(dev->dev, mailbox->dma, port, MLX4_SET_PORT_IB_OPCODE,
1055                        MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
1056                        MLX4_CMD_WRAPPED);
1057
1058         mlx4_free_cmd_mailbox(dev->dev, mailbox);
1059         return err;
1060 }
1061
1062 static int mlx4_ib_modify_port(struct ib_device *ibdev, u8 port, int mask,
1063                                struct ib_port_modify *props)
1064 {
1065         struct mlx4_ib_dev *mdev = to_mdev(ibdev);
1066         u8 is_eth = mdev->dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH;
1067         struct ib_port_attr attr;
1068         u32 cap_mask;
1069         int err;
1070
1071         /* return OK if this is RoCE. CM calls ib_modify_port() regardless
1072          * of whether port link layer is ETH or IB. For ETH ports, qkey
1073          * violations and port capabilities are not meaningful.
1074          */
1075         if (is_eth)
1076                 return 0;
1077
1078         mutex_lock(&mdev->cap_mask_mutex);
1079
1080         err = ib_query_port(ibdev, port, &attr);
1081         if (err)
1082                 goto out;
1083
1084         cap_mask = (attr.port_cap_flags | props->set_port_cap_mask) &
1085                 ~props->clr_port_cap_mask;
1086
1087         err = mlx4_ib_SET_PORT(mdev, port,
1088                                !!(mask & IB_PORT_RESET_QKEY_CNTR),
1089                                cap_mask);
1090
1091 out:
1092         mutex_unlock(&to_mdev(ibdev)->cap_mask_mutex);
1093         return err;
1094 }
1095
1096 static int mlx4_ib_alloc_ucontext(struct ib_ucontext *uctx,
1097                                   struct ib_udata *udata)
1098 {
1099         struct ib_device *ibdev = uctx->device;
1100         struct mlx4_ib_dev *dev = to_mdev(ibdev);
1101         struct mlx4_ib_ucontext *context = to_mucontext(uctx);
1102         struct mlx4_ib_alloc_ucontext_resp_v3 resp_v3;
1103         struct mlx4_ib_alloc_ucontext_resp resp;
1104         int err;
1105
1106         if (!dev->ib_active)
1107                 return -EAGAIN;
1108
1109         if (ibdev->ops.uverbs_abi_ver ==
1110             MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION) {
1111                 resp_v3.qp_tab_size      = dev->dev->caps.num_qps;
1112                 resp_v3.bf_reg_size      = dev->dev->caps.bf_reg_size;
1113                 resp_v3.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
1114         } else {
1115                 resp.dev_caps         = dev->dev->caps.userspace_caps;
1116                 resp.qp_tab_size      = dev->dev->caps.num_qps;
1117                 resp.bf_reg_size      = dev->dev->caps.bf_reg_size;
1118                 resp.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
1119                 resp.cqe_size         = dev->dev->caps.cqe_size;
1120         }
1121
1122         err = mlx4_uar_alloc(to_mdev(ibdev)->dev, &context->uar);
1123         if (err)
1124                 return err;
1125
1126         INIT_LIST_HEAD(&context->db_page_list);
1127         mutex_init(&context->db_page_mutex);
1128
1129         INIT_LIST_HEAD(&context->wqn_ranges_list);
1130         mutex_init(&context->wqn_ranges_mutex);
1131
1132         if (ibdev->ops.uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION)
1133                 err = ib_copy_to_udata(udata, &resp_v3, sizeof(resp_v3));
1134         else
1135                 err = ib_copy_to_udata(udata, &resp, sizeof(resp));
1136
1137         if (err) {
1138                 mlx4_uar_free(to_mdev(ibdev)->dev, &context->uar);
1139                 return -EFAULT;
1140         }
1141
1142         return err;
1143 }
1144
1145 static void mlx4_ib_dealloc_ucontext(struct ib_ucontext *ibcontext)
1146 {
1147         struct mlx4_ib_ucontext *context = to_mucontext(ibcontext);
1148
1149         mlx4_uar_free(to_mdev(ibcontext->device)->dev, &context->uar);
1150 }
1151
1152 static void mlx4_ib_disassociate_ucontext(struct ib_ucontext *ibcontext)
1153 {
1154 }
1155
1156 static int mlx4_ib_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
1157 {
1158         struct mlx4_ib_dev *dev = to_mdev(context->device);
1159
1160         switch (vma->vm_pgoff) {
1161         case 0:
1162                 return rdma_user_mmap_io(context, vma,
1163                                          to_mucontext(context)->uar.pfn,
1164                                          PAGE_SIZE,
1165                                          pgprot_noncached(vma->vm_page_prot),
1166                                          NULL);
1167
1168         case 1:
1169                 if (dev->dev->caps.bf_reg_size == 0)
1170                         return -EINVAL;
1171                 return rdma_user_mmap_io(
1172                         context, vma,
1173                         to_mucontext(context)->uar.pfn +
1174                                 dev->dev->caps.num_uars,
1175                         PAGE_SIZE, pgprot_writecombine(vma->vm_page_prot),
1176                         NULL);
1177
1178         case 3: {
1179                 struct mlx4_clock_params params;
1180                 int ret;
1181
1182                 ret = mlx4_get_internal_clock_params(dev->dev, &params);
1183                 if (ret)
1184                         return ret;
1185
1186                 return rdma_user_mmap_io(
1187                         context, vma,
1188                         (pci_resource_start(dev->dev->persist->pdev,
1189                                             params.bar) +
1190                          params.offset) >>
1191                                 PAGE_SHIFT,
1192                         PAGE_SIZE, pgprot_noncached(vma->vm_page_prot),
1193                         NULL);
1194         }
1195
1196         default:
1197                 return -EINVAL;
1198         }
1199 }
1200
1201 static int mlx4_ib_alloc_pd(struct ib_pd *ibpd, struct ib_udata *udata)
1202 {
1203         struct mlx4_ib_pd *pd = to_mpd(ibpd);
1204         struct ib_device *ibdev = ibpd->device;
1205         int err;
1206
1207         err = mlx4_pd_alloc(to_mdev(ibdev)->dev, &pd->pdn);
1208         if (err)
1209                 return err;
1210
1211         if (udata && ib_copy_to_udata(udata, &pd->pdn, sizeof(__u32))) {
1212                 mlx4_pd_free(to_mdev(ibdev)->dev, pd->pdn);
1213                 return -EFAULT;
1214         }
1215         return 0;
1216 }
1217
1218 static void mlx4_ib_dealloc_pd(struct ib_pd *pd, struct ib_udata *udata)
1219 {
1220         mlx4_pd_free(to_mdev(pd->device)->dev, to_mpd(pd)->pdn);
1221 }
1222
1223 static struct ib_xrcd *mlx4_ib_alloc_xrcd(struct ib_device *ibdev,
1224                                           struct ib_udata *udata)
1225 {
1226         struct mlx4_ib_xrcd *xrcd;
1227         struct ib_cq_init_attr cq_attr = {};
1228         int err;
1229
1230         if (!(to_mdev(ibdev)->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC))
1231                 return ERR_PTR(-ENOSYS);
1232
1233         xrcd = kmalloc(sizeof *xrcd, GFP_KERNEL);
1234         if (!xrcd)
1235                 return ERR_PTR(-ENOMEM);
1236
1237         err = mlx4_xrcd_alloc(to_mdev(ibdev)->dev, &xrcd->xrcdn);
1238         if (err)
1239                 goto err1;
1240
1241         xrcd->pd = ib_alloc_pd(ibdev, 0);
1242         if (IS_ERR(xrcd->pd)) {
1243                 err = PTR_ERR(xrcd->pd);
1244                 goto err2;
1245         }
1246
1247         cq_attr.cqe = 1;
1248         xrcd->cq = ib_create_cq(ibdev, NULL, NULL, xrcd, &cq_attr);
1249         if (IS_ERR(xrcd->cq)) {
1250                 err = PTR_ERR(xrcd->cq);
1251                 goto err3;
1252         }
1253
1254         return &xrcd->ibxrcd;
1255
1256 err3:
1257         ib_dealloc_pd(xrcd->pd);
1258 err2:
1259         mlx4_xrcd_free(to_mdev(ibdev)->dev, xrcd->xrcdn);
1260 err1:
1261         kfree(xrcd);
1262         return ERR_PTR(err);
1263 }
1264
1265 static int mlx4_ib_dealloc_xrcd(struct ib_xrcd *xrcd, struct ib_udata *udata)
1266 {
1267         ib_destroy_cq(to_mxrcd(xrcd)->cq);
1268         ib_dealloc_pd(to_mxrcd(xrcd)->pd);
1269         mlx4_xrcd_free(to_mdev(xrcd->device)->dev, to_mxrcd(xrcd)->xrcdn);
1270         kfree(xrcd);
1271
1272         return 0;
1273 }
1274
1275 static int add_gid_entry(struct ib_qp *ibqp, union ib_gid *gid)
1276 {
1277         struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1278         struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1279         struct mlx4_ib_gid_entry *ge;
1280
1281         ge = kzalloc(sizeof *ge, GFP_KERNEL);
1282         if (!ge)
1283                 return -ENOMEM;
1284
1285         ge->gid = *gid;
1286         if (mlx4_ib_add_mc(mdev, mqp, gid)) {
1287                 ge->port = mqp->port;
1288                 ge->added = 1;
1289         }
1290
1291         mutex_lock(&mqp->mutex);
1292         list_add_tail(&ge->list, &mqp->gid_list);
1293         mutex_unlock(&mqp->mutex);
1294
1295         return 0;
1296 }
1297
1298 static void mlx4_ib_delete_counters_table(struct mlx4_ib_dev *ibdev,
1299                                           struct mlx4_ib_counters *ctr_table)
1300 {
1301         struct counter_index *counter, *tmp_count;
1302
1303         mutex_lock(&ctr_table->mutex);
1304         list_for_each_entry_safe(counter, tmp_count, &ctr_table->counters_list,
1305                                  list) {
1306                 if (counter->allocated)
1307                         mlx4_counter_free(ibdev->dev, counter->index);
1308                 list_del(&counter->list);
1309                 kfree(counter);
1310         }
1311         mutex_unlock(&ctr_table->mutex);
1312 }
1313
1314 int mlx4_ib_add_mc(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
1315                    union ib_gid *gid)
1316 {
1317         struct net_device *ndev;
1318         int ret = 0;
1319
1320         if (!mqp->port)
1321                 return 0;
1322
1323         spin_lock_bh(&mdev->iboe.lock);
1324         ndev = mdev->iboe.netdevs[mqp->port - 1];
1325         if (ndev)
1326                 dev_hold(ndev);
1327         spin_unlock_bh(&mdev->iboe.lock);
1328
1329         if (ndev) {
1330                 ret = 1;
1331                 dev_put(ndev);
1332         }
1333
1334         return ret;
1335 }
1336
1337 struct mlx4_ib_steering {
1338         struct list_head list;
1339         struct mlx4_flow_reg_id reg_id;
1340         union ib_gid gid;
1341 };
1342
1343 #define LAST_ETH_FIELD vlan_tag
1344 #define LAST_IB_FIELD sl
1345 #define LAST_IPV4_FIELD dst_ip
1346 #define LAST_TCP_UDP_FIELD src_port
1347
1348 /* Field is the last supported field */
1349 #define FIELDS_NOT_SUPPORTED(filter, field)\
1350         memchr_inv((void *)&filter.field  +\
1351                    sizeof(filter.field), 0,\
1352                    sizeof(filter) -\
1353                    offsetof(typeof(filter), field) -\
1354                    sizeof(filter.field))
1355
1356 static int parse_flow_attr(struct mlx4_dev *dev,
1357                            u32 qp_num,
1358                            union ib_flow_spec *ib_spec,
1359                            struct _rule_hw *mlx4_spec)
1360 {
1361         enum mlx4_net_trans_rule_id type;
1362
1363         switch (ib_spec->type) {
1364         case IB_FLOW_SPEC_ETH:
1365                 if (FIELDS_NOT_SUPPORTED(ib_spec->eth.mask, LAST_ETH_FIELD))
1366                         return -ENOTSUPP;
1367
1368                 type = MLX4_NET_TRANS_RULE_ID_ETH;
1369                 memcpy(mlx4_spec->eth.dst_mac, ib_spec->eth.val.dst_mac,
1370                        ETH_ALEN);
1371                 memcpy(mlx4_spec->eth.dst_mac_msk, ib_spec->eth.mask.dst_mac,
1372                        ETH_ALEN);
1373                 mlx4_spec->eth.vlan_tag = ib_spec->eth.val.vlan_tag;
1374                 mlx4_spec->eth.vlan_tag_msk = ib_spec->eth.mask.vlan_tag;
1375                 break;
1376         case IB_FLOW_SPEC_IB:
1377                 if (FIELDS_NOT_SUPPORTED(ib_spec->ib.mask, LAST_IB_FIELD))
1378                         return -ENOTSUPP;
1379
1380                 type = MLX4_NET_TRANS_RULE_ID_IB;
1381                 mlx4_spec->ib.l3_qpn =
1382                         cpu_to_be32(qp_num);
1383                 mlx4_spec->ib.qpn_mask =
1384                         cpu_to_be32(MLX4_IB_FLOW_QPN_MASK);
1385                 break;
1386
1387
1388         case IB_FLOW_SPEC_IPV4:
1389                 if (FIELDS_NOT_SUPPORTED(ib_spec->ipv4.mask, LAST_IPV4_FIELD))
1390                         return -ENOTSUPP;
1391
1392                 type = MLX4_NET_TRANS_RULE_ID_IPV4;
1393                 mlx4_spec->ipv4.src_ip = ib_spec->ipv4.val.src_ip;
1394                 mlx4_spec->ipv4.src_ip_msk = ib_spec->ipv4.mask.src_ip;
1395                 mlx4_spec->ipv4.dst_ip = ib_spec->ipv4.val.dst_ip;
1396                 mlx4_spec->ipv4.dst_ip_msk = ib_spec->ipv4.mask.dst_ip;
1397                 break;
1398
1399         case IB_FLOW_SPEC_TCP:
1400         case IB_FLOW_SPEC_UDP:
1401                 if (FIELDS_NOT_SUPPORTED(ib_spec->tcp_udp.mask, LAST_TCP_UDP_FIELD))
1402                         return -ENOTSUPP;
1403
1404                 type = ib_spec->type == IB_FLOW_SPEC_TCP ?
1405                                         MLX4_NET_TRANS_RULE_ID_TCP :
1406                                         MLX4_NET_TRANS_RULE_ID_UDP;
1407                 mlx4_spec->tcp_udp.dst_port = ib_spec->tcp_udp.val.dst_port;
1408                 mlx4_spec->tcp_udp.dst_port_msk = ib_spec->tcp_udp.mask.dst_port;
1409                 mlx4_spec->tcp_udp.src_port = ib_spec->tcp_udp.val.src_port;
1410                 mlx4_spec->tcp_udp.src_port_msk = ib_spec->tcp_udp.mask.src_port;
1411                 break;
1412
1413         default:
1414                 return -EINVAL;
1415         }
1416         if (mlx4_map_sw_to_hw_steering_id(dev, type) < 0 ||
1417             mlx4_hw_rule_sz(dev, type) < 0)
1418                 return -EINVAL;
1419         mlx4_spec->id = cpu_to_be16(mlx4_map_sw_to_hw_steering_id(dev, type));
1420         mlx4_spec->size = mlx4_hw_rule_sz(dev, type) >> 2;
1421         return mlx4_hw_rule_sz(dev, type);
1422 }
1423
1424 struct default_rules {
1425         __u32 mandatory_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
1426         __u32 mandatory_not_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
1427         __u32 rules_create_list[IB_FLOW_SPEC_SUPPORT_LAYERS];
1428         __u8  link_layer;
1429 };
1430 static const struct default_rules default_table[] = {
1431         {
1432                 .mandatory_fields = {IB_FLOW_SPEC_IPV4},
1433                 .mandatory_not_fields = {IB_FLOW_SPEC_ETH},
1434                 .rules_create_list = {IB_FLOW_SPEC_IB},
1435                 .link_layer = IB_LINK_LAYER_INFINIBAND
1436         }
1437 };
1438
1439 static int __mlx4_ib_default_rules_match(struct ib_qp *qp,
1440                                          struct ib_flow_attr *flow_attr)
1441 {
1442         int i, j, k;
1443         void *ib_flow;
1444         const struct default_rules *pdefault_rules = default_table;
1445         u8 link_layer = rdma_port_get_link_layer(qp->device, flow_attr->port);
1446
1447         for (i = 0; i < ARRAY_SIZE(default_table); i++, pdefault_rules++) {
1448                 __u32 field_types[IB_FLOW_SPEC_SUPPORT_LAYERS];
1449                 memset(&field_types, 0, sizeof(field_types));
1450
1451                 if (link_layer != pdefault_rules->link_layer)
1452                         continue;
1453
1454                 ib_flow = flow_attr + 1;
1455                 /* we assume the specs are sorted */
1456                 for (j = 0, k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS &&
1457                      j < flow_attr->num_of_specs; k++) {
1458                         union ib_flow_spec *current_flow =
1459                                 (union ib_flow_spec *)ib_flow;
1460
1461                         /* same layer but different type */
1462                         if (((current_flow->type & IB_FLOW_SPEC_LAYER_MASK) ==
1463                              (pdefault_rules->mandatory_fields[k] &
1464                               IB_FLOW_SPEC_LAYER_MASK)) &&
1465                             (current_flow->type !=
1466                              pdefault_rules->mandatory_fields[k]))
1467                                 goto out;
1468
1469                         /* same layer, try match next one */
1470                         if (current_flow->type ==
1471                             pdefault_rules->mandatory_fields[k]) {
1472                                 j++;
1473                                 ib_flow +=
1474                                         ((union ib_flow_spec *)ib_flow)->size;
1475                         }
1476                 }
1477
1478                 ib_flow = flow_attr + 1;
1479                 for (j = 0; j < flow_attr->num_of_specs;
1480                      j++, ib_flow += ((union ib_flow_spec *)ib_flow)->size)
1481                         for (k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS; k++)
1482                                 /* same layer and same type */
1483                                 if (((union ib_flow_spec *)ib_flow)->type ==
1484                                     pdefault_rules->mandatory_not_fields[k])
1485                                         goto out;
1486
1487                 return i;
1488         }
1489 out:
1490         return -1;
1491 }
1492
1493 static int __mlx4_ib_create_default_rules(
1494                 struct mlx4_ib_dev *mdev,
1495                 struct ib_qp *qp,
1496                 const struct default_rules *pdefault_rules,
1497                 struct _rule_hw *mlx4_spec) {
1498         int size = 0;
1499         int i;
1500
1501         for (i = 0; i < ARRAY_SIZE(pdefault_rules->rules_create_list); i++) {
1502                 union ib_flow_spec ib_spec = {};
1503                 int ret;
1504
1505                 switch (pdefault_rules->rules_create_list[i]) {
1506                 case 0:
1507                         /* no rule */
1508                         continue;
1509                 case IB_FLOW_SPEC_IB:
1510                         ib_spec.type = IB_FLOW_SPEC_IB;
1511                         ib_spec.size = sizeof(struct ib_flow_spec_ib);
1512
1513                         break;
1514                 default:
1515                         /* invalid rule */
1516                         return -EINVAL;
1517                 }
1518                 /* We must put empty rule, qpn is being ignored */
1519                 ret = parse_flow_attr(mdev->dev, 0, &ib_spec,
1520                                       mlx4_spec);
1521                 if (ret < 0) {
1522                         pr_info("invalid parsing\n");
1523                         return -EINVAL;
1524                 }
1525
1526                 mlx4_spec = (void *)mlx4_spec + ret;
1527                 size += ret;
1528         }
1529         return size;
1530 }
1531
1532 static int __mlx4_ib_create_flow(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
1533                           int domain,
1534                           enum mlx4_net_trans_promisc_mode flow_type,
1535                           u64 *reg_id)
1536 {
1537         int ret, i;
1538         int size = 0;
1539         void *ib_flow;
1540         struct mlx4_ib_dev *mdev = to_mdev(qp->device);
1541         struct mlx4_cmd_mailbox *mailbox;
1542         struct mlx4_net_trans_rule_hw_ctrl *ctrl;
1543         int default_flow;
1544
1545         static const u16 __mlx4_domain[] = {
1546                 [IB_FLOW_DOMAIN_USER] = MLX4_DOMAIN_UVERBS,
1547                 [IB_FLOW_DOMAIN_ETHTOOL] = MLX4_DOMAIN_ETHTOOL,
1548                 [IB_FLOW_DOMAIN_RFS] = MLX4_DOMAIN_RFS,
1549                 [IB_FLOW_DOMAIN_NIC] = MLX4_DOMAIN_NIC,
1550         };
1551
1552         if (flow_attr->priority > MLX4_IB_FLOW_MAX_PRIO) {
1553                 pr_err("Invalid priority value %d\n", flow_attr->priority);
1554                 return -EINVAL;
1555         }
1556
1557         if (domain >= IB_FLOW_DOMAIN_NUM) {
1558                 pr_err("Invalid domain value %d\n", domain);
1559                 return -EINVAL;
1560         }
1561
1562         if (mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type) < 0)
1563                 return -EINVAL;
1564
1565         mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
1566         if (IS_ERR(mailbox))
1567                 return PTR_ERR(mailbox);
1568         ctrl = mailbox->buf;
1569
1570         ctrl->prio = cpu_to_be16(__mlx4_domain[domain] |
1571                                  flow_attr->priority);
1572         ctrl->type = mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type);
1573         ctrl->port = flow_attr->port;
1574         ctrl->qpn = cpu_to_be32(qp->qp_num);
1575
1576         ib_flow = flow_attr + 1;
1577         size += sizeof(struct mlx4_net_trans_rule_hw_ctrl);
1578         /* Add default flows */
1579         default_flow = __mlx4_ib_default_rules_match(qp, flow_attr);
1580         if (default_flow >= 0) {
1581                 ret = __mlx4_ib_create_default_rules(
1582                                 mdev, qp, default_table + default_flow,
1583                                 mailbox->buf + size);
1584                 if (ret < 0) {
1585                         mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1586                         return -EINVAL;
1587                 }
1588                 size += ret;
1589         }
1590         for (i = 0; i < flow_attr->num_of_specs; i++) {
1591                 ret = parse_flow_attr(mdev->dev, qp->qp_num, ib_flow,
1592                                       mailbox->buf + size);
1593                 if (ret < 0) {
1594                         mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1595                         return -EINVAL;
1596                 }
1597                 ib_flow += ((union ib_flow_spec *) ib_flow)->size;
1598                 size += ret;
1599         }
1600
1601         if (mlx4_is_master(mdev->dev) && flow_type == MLX4_FS_REGULAR &&
1602             flow_attr->num_of_specs == 1) {
1603                 struct _rule_hw *rule_header = (struct _rule_hw *)(ctrl + 1);
1604                 enum ib_flow_spec_type header_spec =
1605                         ((union ib_flow_spec *)(flow_attr + 1))->type;
1606
1607                 if (header_spec == IB_FLOW_SPEC_ETH)
1608                         mlx4_handle_eth_header_mcast_prio(ctrl, rule_header);
1609         }
1610
1611         ret = mlx4_cmd_imm(mdev->dev, mailbox->dma, reg_id, size >> 2, 0,
1612                            MLX4_QP_FLOW_STEERING_ATTACH, MLX4_CMD_TIME_CLASS_A,
1613                            MLX4_CMD_NATIVE);
1614         if (ret == -ENOMEM)
1615                 pr_err("mcg table is full. Fail to register network rule.\n");
1616         else if (ret == -ENXIO)
1617                 pr_err("Device managed flow steering is disabled. Fail to register network rule.\n");
1618         else if (ret)
1619                 pr_err("Invalid argument. Fail to register network rule.\n");
1620
1621         mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1622         return ret;
1623 }
1624
1625 static int __mlx4_ib_destroy_flow(struct mlx4_dev *dev, u64 reg_id)
1626 {
1627         int err;
1628         err = mlx4_cmd(dev, reg_id, 0, 0,
1629                        MLX4_QP_FLOW_STEERING_DETACH, MLX4_CMD_TIME_CLASS_A,
1630                        MLX4_CMD_NATIVE);
1631         if (err)
1632                 pr_err("Fail to detach network rule. registration id = 0x%llx\n",
1633                        reg_id);
1634         return err;
1635 }
1636
1637 static int mlx4_ib_tunnel_steer_add(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
1638                                     u64 *reg_id)
1639 {
1640         void *ib_flow;
1641         union ib_flow_spec *ib_spec;
1642         struct mlx4_dev *dev = to_mdev(qp->device)->dev;
1643         int err = 0;
1644
1645         if (dev->caps.tunnel_offload_mode != MLX4_TUNNEL_OFFLOAD_MODE_VXLAN ||
1646             dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC)
1647                 return 0; /* do nothing */
1648
1649         ib_flow = flow_attr + 1;
1650         ib_spec = (union ib_flow_spec *)ib_flow;
1651
1652         if (ib_spec->type !=  IB_FLOW_SPEC_ETH || flow_attr->num_of_specs != 1)
1653                 return 0; /* do nothing */
1654
1655         err = mlx4_tunnel_steer_add(to_mdev(qp->device)->dev, ib_spec->eth.val.dst_mac,
1656                                     flow_attr->port, qp->qp_num,
1657                                     MLX4_DOMAIN_UVERBS | (flow_attr->priority & 0xff),
1658                                     reg_id);
1659         return err;
1660 }
1661
1662 static int mlx4_ib_add_dont_trap_rule(struct mlx4_dev *dev,
1663                                       struct ib_flow_attr *flow_attr,
1664                                       enum mlx4_net_trans_promisc_mode *type)
1665 {
1666         int err = 0;
1667
1668         if (!(dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DMFS_UC_MC_SNIFFER) ||
1669             (dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC) ||
1670             (flow_attr->num_of_specs > 1) || (flow_attr->priority != 0)) {
1671                 return -EOPNOTSUPP;
1672         }
1673
1674         if (flow_attr->num_of_specs == 0) {
1675                 type[0] = MLX4_FS_MC_SNIFFER;
1676                 type[1] = MLX4_FS_UC_SNIFFER;
1677         } else {
1678                 union ib_flow_spec *ib_spec;
1679
1680                 ib_spec = (union ib_flow_spec *)(flow_attr + 1);
1681                 if (ib_spec->type !=  IB_FLOW_SPEC_ETH)
1682                         return -EINVAL;
1683
1684                 /* if all is zero than MC and UC */
1685                 if (is_zero_ether_addr(ib_spec->eth.mask.dst_mac)) {
1686                         type[0] = MLX4_FS_MC_SNIFFER;
1687                         type[1] = MLX4_FS_UC_SNIFFER;
1688                 } else {
1689                         u8 mac[ETH_ALEN] = {ib_spec->eth.mask.dst_mac[0] ^ 0x01,
1690                                             ib_spec->eth.mask.dst_mac[1],
1691                                             ib_spec->eth.mask.dst_mac[2],
1692                                             ib_spec->eth.mask.dst_mac[3],
1693                                             ib_spec->eth.mask.dst_mac[4],
1694                                             ib_spec->eth.mask.dst_mac[5]};
1695
1696                         /* Above xor was only on MC bit, non empty mask is valid
1697                          * only if this bit is set and rest are zero.
1698                          */
1699                         if (!is_zero_ether_addr(&mac[0]))
1700                                 return -EINVAL;
1701
1702                         if (is_multicast_ether_addr(ib_spec->eth.val.dst_mac))
1703                                 type[0] = MLX4_FS_MC_SNIFFER;
1704                         else
1705                                 type[0] = MLX4_FS_UC_SNIFFER;
1706                 }
1707         }
1708
1709         return err;
1710 }
1711
1712 static struct ib_flow *mlx4_ib_create_flow(struct ib_qp *qp,
1713                                     struct ib_flow_attr *flow_attr,
1714                                     int domain, struct ib_udata *udata)
1715 {
1716         int err = 0, i = 0, j = 0;
1717         struct mlx4_ib_flow *mflow;
1718         enum mlx4_net_trans_promisc_mode type[2];
1719         struct mlx4_dev *dev = (to_mdev(qp->device))->dev;
1720         int is_bonded = mlx4_is_bonded(dev);
1721
1722         if (flow_attr->port < 1 || flow_attr->port > qp->device->phys_port_cnt)
1723                 return ERR_PTR(-EINVAL);
1724
1725         if (flow_attr->flags & ~IB_FLOW_ATTR_FLAGS_DONT_TRAP)
1726                 return ERR_PTR(-EOPNOTSUPP);
1727
1728         if ((flow_attr->flags & IB_FLOW_ATTR_FLAGS_DONT_TRAP) &&
1729             (flow_attr->type != IB_FLOW_ATTR_NORMAL))
1730                 return ERR_PTR(-EOPNOTSUPP);
1731
1732         if (udata &&
1733             udata->inlen && !ib_is_udata_cleared(udata, 0, udata->inlen))
1734                 return ERR_PTR(-EOPNOTSUPP);
1735
1736         memset(type, 0, sizeof(type));
1737
1738         mflow = kzalloc(sizeof(*mflow), GFP_KERNEL);
1739         if (!mflow) {
1740                 err = -ENOMEM;
1741                 goto err_free;
1742         }
1743
1744         switch (flow_attr->type) {
1745         case IB_FLOW_ATTR_NORMAL:
1746                 /* If dont trap flag (continue match) is set, under specific
1747                  * condition traffic be replicated to given qp,
1748                  * without stealing it
1749                  */
1750                 if (unlikely(flow_attr->flags & IB_FLOW_ATTR_FLAGS_DONT_TRAP)) {
1751                         err = mlx4_ib_add_dont_trap_rule(dev,
1752                                                          flow_attr,
1753                                                          type);
1754                         if (err)
1755                                 goto err_free;
1756                 } else {
1757                         type[0] = MLX4_FS_REGULAR;
1758                 }
1759                 break;
1760
1761         case IB_FLOW_ATTR_ALL_DEFAULT:
1762                 type[0] = MLX4_FS_ALL_DEFAULT;
1763                 break;
1764
1765         case IB_FLOW_ATTR_MC_DEFAULT:
1766                 type[0] = MLX4_FS_MC_DEFAULT;
1767                 break;
1768
1769         case IB_FLOW_ATTR_SNIFFER:
1770                 type[0] = MLX4_FS_MIRROR_RX_PORT;
1771                 type[1] = MLX4_FS_MIRROR_SX_PORT;
1772                 break;
1773
1774         default:
1775                 err = -EINVAL;
1776                 goto err_free;
1777         }
1778
1779         while (i < ARRAY_SIZE(type) && type[i]) {
1780                 err = __mlx4_ib_create_flow(qp, flow_attr, domain, type[i],
1781                                             &mflow->reg_id[i].id);
1782                 if (err)
1783                         goto err_create_flow;
1784                 if (is_bonded) {
1785                         /* Application always sees one port so the mirror rule
1786                          * must be on port #2
1787                          */
1788                         flow_attr->port = 2;
1789                         err = __mlx4_ib_create_flow(qp, flow_attr,
1790                                                     domain, type[j],
1791                                                     &mflow->reg_id[j].mirror);
1792                         flow_attr->port = 1;
1793                         if (err)
1794                                 goto err_create_flow;
1795                         j++;
1796                 }
1797
1798                 i++;
1799         }
1800
1801         if (i < ARRAY_SIZE(type) && flow_attr->type == IB_FLOW_ATTR_NORMAL) {
1802                 err = mlx4_ib_tunnel_steer_add(qp, flow_attr,
1803                                                &mflow->reg_id[i].id);
1804                 if (err)
1805                         goto err_create_flow;
1806
1807                 if (is_bonded) {
1808                         flow_attr->port = 2;
1809                         err = mlx4_ib_tunnel_steer_add(qp, flow_attr,
1810                                                        &mflow->reg_id[j].mirror);
1811                         flow_attr->port = 1;
1812                         if (err)
1813                                 goto err_create_flow;
1814                         j++;
1815                 }
1816                 /* function to create mirror rule */
1817                 i++;
1818         }
1819
1820         return &mflow->ibflow;
1821
1822 err_create_flow:
1823         while (i) {
1824                 (void)__mlx4_ib_destroy_flow(to_mdev(qp->device)->dev,
1825                                              mflow->reg_id[i].id);
1826                 i--;
1827         }
1828
1829         while (j) {
1830                 (void)__mlx4_ib_destroy_flow(to_mdev(qp->device)->dev,
1831                                              mflow->reg_id[j].mirror);
1832                 j--;
1833         }
1834 err_free:
1835         kfree(mflow);
1836         return ERR_PTR(err);
1837 }
1838
1839 static int mlx4_ib_destroy_flow(struct ib_flow *flow_id)
1840 {
1841         int err, ret = 0;
1842         int i = 0;
1843         struct mlx4_ib_dev *mdev = to_mdev(flow_id->qp->device);
1844         struct mlx4_ib_flow *mflow = to_mflow(flow_id);
1845
1846         while (i < ARRAY_SIZE(mflow->reg_id) && mflow->reg_id[i].id) {
1847                 err = __mlx4_ib_destroy_flow(mdev->dev, mflow->reg_id[i].id);
1848                 if (err)
1849                         ret = err;
1850                 if (mflow->reg_id[i].mirror) {
1851                         err = __mlx4_ib_destroy_flow(mdev->dev,
1852                                                      mflow->reg_id[i].mirror);
1853                         if (err)
1854                                 ret = err;
1855                 }
1856                 i++;
1857         }
1858
1859         kfree(mflow);
1860         return ret;
1861 }
1862
1863 static int mlx4_ib_mcg_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
1864 {
1865         int err;
1866         struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1867         struct mlx4_dev *dev = mdev->dev;
1868         struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1869         struct mlx4_ib_steering *ib_steering = NULL;
1870         enum mlx4_protocol prot = MLX4_PROT_IB_IPV6;
1871         struct mlx4_flow_reg_id reg_id;
1872
1873         if (mdev->dev->caps.steering_mode ==
1874             MLX4_STEERING_MODE_DEVICE_MANAGED) {
1875                 ib_steering = kmalloc(sizeof(*ib_steering), GFP_KERNEL);
1876                 if (!ib_steering)
1877                         return -ENOMEM;
1878         }
1879
1880         err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw, mqp->port,
1881                                     !!(mqp->flags &
1882                                        MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
1883                                     prot, &reg_id.id);
1884         if (err) {
1885                 pr_err("multicast attach op failed, err %d\n", err);
1886                 goto err_malloc;
1887         }
1888
1889         reg_id.mirror = 0;
1890         if (mlx4_is_bonded(dev)) {
1891                 err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw,
1892                                             (mqp->port == 1) ? 2 : 1,
1893                                             !!(mqp->flags &
1894                                             MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
1895                                             prot, &reg_id.mirror);
1896                 if (err)
1897                         goto err_add;
1898         }
1899
1900         err = add_gid_entry(ibqp, gid);
1901         if (err)
1902                 goto err_add;
1903
1904         if (ib_steering) {
1905                 memcpy(ib_steering->gid.raw, gid->raw, 16);
1906                 ib_steering->reg_id = reg_id;
1907                 mutex_lock(&mqp->mutex);
1908                 list_add(&ib_steering->list, &mqp->steering_rules);
1909                 mutex_unlock(&mqp->mutex);
1910         }
1911         return 0;
1912
1913 err_add:
1914         mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1915                               prot, reg_id.id);
1916         if (reg_id.mirror)
1917                 mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1918                                       prot, reg_id.mirror);
1919 err_malloc:
1920         kfree(ib_steering);
1921
1922         return err;
1923 }
1924
1925 static struct mlx4_ib_gid_entry *find_gid_entry(struct mlx4_ib_qp *qp, u8 *raw)
1926 {
1927         struct mlx4_ib_gid_entry *ge;
1928         struct mlx4_ib_gid_entry *tmp;
1929         struct mlx4_ib_gid_entry *ret = NULL;
1930
1931         list_for_each_entry_safe(ge, tmp, &qp->gid_list, list) {
1932                 if (!memcmp(raw, ge->gid.raw, 16)) {
1933                         ret = ge;
1934                         break;
1935                 }
1936         }
1937
1938         return ret;
1939 }
1940
1941 static int mlx4_ib_mcg_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
1942 {
1943         int err;
1944         struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1945         struct mlx4_dev *dev = mdev->dev;
1946         struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1947         struct net_device *ndev;
1948         struct mlx4_ib_gid_entry *ge;
1949         struct mlx4_flow_reg_id reg_id = {0, 0};
1950         enum mlx4_protocol prot =  MLX4_PROT_IB_IPV6;
1951
1952         if (mdev->dev->caps.steering_mode ==
1953             MLX4_STEERING_MODE_DEVICE_MANAGED) {
1954                 struct mlx4_ib_steering *ib_steering;
1955
1956                 mutex_lock(&mqp->mutex);
1957                 list_for_each_entry(ib_steering, &mqp->steering_rules, list) {
1958                         if (!memcmp(ib_steering->gid.raw, gid->raw, 16)) {
1959                                 list_del(&ib_steering->list);
1960                                 break;
1961                         }
1962                 }
1963                 mutex_unlock(&mqp->mutex);
1964                 if (&ib_steering->list == &mqp->steering_rules) {
1965                         pr_err("Couldn't find reg_id for mgid. Steering rule is left attached\n");
1966                         return -EINVAL;
1967                 }
1968                 reg_id = ib_steering->reg_id;
1969                 kfree(ib_steering);
1970         }
1971
1972         err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1973                                     prot, reg_id.id);
1974         if (err)
1975                 return err;
1976
1977         if (mlx4_is_bonded(dev)) {
1978                 err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1979                                             prot, reg_id.mirror);
1980                 if (err)
1981                         return err;
1982         }
1983
1984         mutex_lock(&mqp->mutex);
1985         ge = find_gid_entry(mqp, gid->raw);
1986         if (ge) {
1987                 spin_lock_bh(&mdev->iboe.lock);
1988                 ndev = ge->added ? mdev->iboe.netdevs[ge->port - 1] : NULL;
1989                 if (ndev)
1990                         dev_hold(ndev);
1991                 spin_unlock_bh(&mdev->iboe.lock);
1992                 if (ndev)
1993                         dev_put(ndev);
1994                 list_del(&ge->list);
1995                 kfree(ge);
1996         } else
1997                 pr_warn("could not find mgid entry\n");
1998
1999         mutex_unlock(&mqp->mutex);
2000
2001         return 0;
2002 }
2003
2004 static int init_node_data(struct mlx4_ib_dev *dev)
2005 {
2006         struct ib_smp *in_mad  = NULL;
2007         struct ib_smp *out_mad = NULL;
2008         int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
2009         int err = -ENOMEM;
2010
2011         in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
2012         out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
2013         if (!in_mad || !out_mad)
2014                 goto out;
2015
2016         init_query_mad(in_mad);
2017         in_mad->attr_id = IB_SMP_ATTR_NODE_DESC;
2018         if (mlx4_is_master(dev->dev))
2019                 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
2020
2021         err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
2022         if (err)
2023                 goto out;
2024
2025         memcpy(dev->ib_dev.node_desc, out_mad->data, IB_DEVICE_NODE_DESC_MAX);
2026
2027         in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
2028
2029         err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
2030         if (err)
2031                 goto out;
2032
2033         dev->dev->rev_id = be32_to_cpup((__be32 *) (out_mad->data + 32));
2034         memcpy(&dev->ib_dev.node_guid, out_mad->data + 12, 8);
2035
2036 out:
2037         kfree(in_mad);
2038         kfree(out_mad);
2039         return err;
2040 }
2041
2042 static ssize_t hca_type_show(struct device *device,
2043                              struct device_attribute *attr, char *buf)
2044 {
2045         struct mlx4_ib_dev *dev =
2046                 rdma_device_to_drv_device(device, struct mlx4_ib_dev, ib_dev);
2047         return sprintf(buf, "MT%d\n", dev->dev->persist->pdev->device);
2048 }
2049 static DEVICE_ATTR_RO(hca_type);
2050
2051 static ssize_t hw_rev_show(struct device *device,
2052                            struct device_attribute *attr, char *buf)
2053 {
2054         struct mlx4_ib_dev *dev =
2055                 rdma_device_to_drv_device(device, struct mlx4_ib_dev, ib_dev);
2056         return sprintf(buf, "%x\n", dev->dev->rev_id);
2057 }
2058 static DEVICE_ATTR_RO(hw_rev);
2059
2060 static ssize_t board_id_show(struct device *device,
2061                              struct device_attribute *attr, char *buf)
2062 {
2063         struct mlx4_ib_dev *dev =
2064                 rdma_device_to_drv_device(device, struct mlx4_ib_dev, ib_dev);
2065
2066         return sprintf(buf, "%.*s\n", MLX4_BOARD_ID_LEN,
2067                        dev->dev->board_id);
2068 }
2069 static DEVICE_ATTR_RO(board_id);
2070
2071 static struct attribute *mlx4_class_attributes[] = {
2072         &dev_attr_hw_rev.attr,
2073         &dev_attr_hca_type.attr,
2074         &dev_attr_board_id.attr,
2075         NULL
2076 };
2077
2078 static const struct attribute_group mlx4_attr_group = {
2079         .attrs = mlx4_class_attributes,
2080 };
2081
2082 struct diag_counter {
2083         const char *name;
2084         u32 offset;
2085 };
2086
2087 #define DIAG_COUNTER(_name, _offset)                    \
2088         { .name = #_name, .offset = _offset }
2089
2090 static const struct diag_counter diag_basic[] = {
2091         DIAG_COUNTER(rq_num_lle, 0x00),
2092         DIAG_COUNTER(sq_num_lle, 0x04),
2093         DIAG_COUNTER(rq_num_lqpoe, 0x08),
2094         DIAG_COUNTER(sq_num_lqpoe, 0x0C),
2095         DIAG_COUNTER(rq_num_lpe, 0x18),
2096         DIAG_COUNTER(sq_num_lpe, 0x1C),
2097         DIAG_COUNTER(rq_num_wrfe, 0x20),
2098         DIAG_COUNTER(sq_num_wrfe, 0x24),
2099         DIAG_COUNTER(sq_num_mwbe, 0x2C),
2100         DIAG_COUNTER(sq_num_bre, 0x34),
2101         DIAG_COUNTER(sq_num_rire, 0x44),
2102         DIAG_COUNTER(rq_num_rire, 0x48),
2103         DIAG_COUNTER(sq_num_rae, 0x4C),
2104         DIAG_COUNTER(rq_num_rae, 0x50),
2105         DIAG_COUNTER(sq_num_roe, 0x54),
2106         DIAG_COUNTER(sq_num_tree, 0x5C),
2107         DIAG_COUNTER(sq_num_rree, 0x64),
2108         DIAG_COUNTER(rq_num_rnr, 0x68),
2109         DIAG_COUNTER(sq_num_rnr, 0x6C),
2110         DIAG_COUNTER(rq_num_oos, 0x100),
2111         DIAG_COUNTER(sq_num_oos, 0x104),
2112 };
2113
2114 static const struct diag_counter diag_ext[] = {
2115         DIAG_COUNTER(rq_num_dup, 0x130),
2116         DIAG_COUNTER(sq_num_to, 0x134),
2117 };
2118
2119 static const struct diag_counter diag_device_only[] = {
2120         DIAG_COUNTER(num_cqovf, 0x1A0),
2121         DIAG_COUNTER(rq_num_udsdprd, 0x118),
2122 };
2123
2124 static struct rdma_hw_stats *mlx4_ib_alloc_hw_stats(struct ib_device *ibdev,
2125                                                     u8 port_num)
2126 {
2127         struct mlx4_ib_dev *dev = to_mdev(ibdev);
2128         struct mlx4_ib_diag_counters *diag = dev->diag_counters;
2129
2130         if (!diag[!!port_num].name)
2131                 return NULL;
2132
2133         return rdma_alloc_hw_stats_struct(diag[!!port_num].name,
2134                                           diag[!!port_num].num_counters,
2135                                           RDMA_HW_STATS_DEFAULT_LIFESPAN);
2136 }
2137
2138 static int mlx4_ib_get_hw_stats(struct ib_device *ibdev,
2139                                 struct rdma_hw_stats *stats,
2140                                 u8 port, int index)
2141 {
2142         struct mlx4_ib_dev *dev = to_mdev(ibdev);
2143         struct mlx4_ib_diag_counters *diag = dev->diag_counters;
2144         u32 hw_value[ARRAY_SIZE(diag_device_only) +
2145                 ARRAY_SIZE(diag_ext) + ARRAY_SIZE(diag_basic)] = {};
2146         int ret;
2147         int i;
2148
2149         ret = mlx4_query_diag_counters(dev->dev,
2150                                        MLX4_OP_MOD_QUERY_TRANSPORT_CI_ERRORS,
2151                                        diag[!!port].offset, hw_value,
2152                                        diag[!!port].num_counters, port);
2153
2154         if (ret)
2155                 return ret;
2156
2157         for (i = 0; i < diag[!!port].num_counters; i++)
2158                 stats->value[i] = hw_value[i];
2159
2160         return diag[!!port].num_counters;
2161 }
2162
2163 static int __mlx4_ib_alloc_diag_counters(struct mlx4_ib_dev *ibdev,
2164                                          const char ***name,
2165                                          u32 **offset,
2166                                          u32 *num,
2167                                          bool port)
2168 {
2169         u32 num_counters;
2170
2171         num_counters = ARRAY_SIZE(diag_basic);
2172
2173         if (ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DIAG_PER_PORT)
2174                 num_counters += ARRAY_SIZE(diag_ext);
2175
2176         if (!port)
2177                 num_counters += ARRAY_SIZE(diag_device_only);
2178
2179         *name = kcalloc(num_counters, sizeof(**name), GFP_KERNEL);
2180         if (!*name)
2181                 return -ENOMEM;
2182
2183         *offset = kcalloc(num_counters, sizeof(**offset), GFP_KERNEL);
2184         if (!*offset)
2185                 goto err_name;
2186
2187         *num = num_counters;
2188
2189         return 0;
2190
2191 err_name:
2192         kfree(*name);
2193         return -ENOMEM;
2194 }
2195
2196 static void mlx4_ib_fill_diag_counters(struct mlx4_ib_dev *ibdev,
2197                                        const char **name,
2198                                        u32 *offset,
2199                                        bool port)
2200 {
2201         int i;
2202         int j;
2203
2204         for (i = 0, j = 0; i < ARRAY_SIZE(diag_basic); i++, j++) {
2205                 name[i] = diag_basic[i].name;
2206                 offset[i] = diag_basic[i].offset;
2207         }
2208
2209         if (ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DIAG_PER_PORT) {
2210                 for (i = 0; i < ARRAY_SIZE(diag_ext); i++, j++) {
2211                         name[j] = diag_ext[i].name;
2212                         offset[j] = diag_ext[i].offset;
2213                 }
2214         }
2215
2216         if (!port) {
2217                 for (i = 0; i < ARRAY_SIZE(diag_device_only); i++, j++) {
2218                         name[j] = diag_device_only[i].name;
2219                         offset[j] = diag_device_only[i].offset;
2220                 }
2221         }
2222 }
2223
2224 static const struct ib_device_ops mlx4_ib_hw_stats_ops = {
2225         .alloc_hw_stats = mlx4_ib_alloc_hw_stats,
2226         .get_hw_stats = mlx4_ib_get_hw_stats,
2227 };
2228
2229 static int mlx4_ib_alloc_diag_counters(struct mlx4_ib_dev *ibdev)
2230 {
2231         struct mlx4_ib_diag_counters *diag = ibdev->diag_counters;
2232         int i;
2233         int ret;
2234         bool per_port = !!(ibdev->dev->caps.flags2 &
2235                 MLX4_DEV_CAP_FLAG2_DIAG_PER_PORT);
2236
2237         if (mlx4_is_slave(ibdev->dev))
2238                 return 0;
2239
2240         for (i = 0; i < MLX4_DIAG_COUNTERS_TYPES; i++) {
2241                 /* i == 1 means we are building port counters */
2242                 if (i && !per_port)
2243                         continue;
2244
2245                 ret = __mlx4_ib_alloc_diag_counters(ibdev, &diag[i].name,
2246                                                     &diag[i].offset,
2247                                                     &diag[i].num_counters, i);
2248                 if (ret)
2249                         goto err_alloc;
2250
2251                 mlx4_ib_fill_diag_counters(ibdev, diag[i].name,
2252                                            diag[i].offset, i);
2253         }
2254
2255         ib_set_device_ops(&ibdev->ib_dev, &mlx4_ib_hw_stats_ops);
2256
2257         return 0;
2258
2259 err_alloc:
2260         if (i) {
2261                 kfree(diag[i - 1].name);
2262                 kfree(diag[i - 1].offset);
2263         }
2264
2265         return ret;
2266 }
2267
2268 static void mlx4_ib_diag_cleanup(struct mlx4_ib_dev *ibdev)
2269 {
2270         int i;
2271
2272         for (i = 0; i < MLX4_DIAG_COUNTERS_TYPES; i++) {
2273                 kfree(ibdev->diag_counters[i].offset);
2274                 kfree(ibdev->diag_counters[i].name);
2275         }
2276 }
2277
2278 #define MLX4_IB_INVALID_MAC     ((u64)-1)
2279 static void mlx4_ib_update_qps(struct mlx4_ib_dev *ibdev,
2280                                struct net_device *dev,
2281                                int port)
2282 {
2283         u64 new_smac = 0;
2284         u64 release_mac = MLX4_IB_INVALID_MAC;
2285         struct mlx4_ib_qp *qp;
2286
2287         read_lock(&dev_base_lock);
2288         new_smac = mlx4_mac_to_u64(dev->dev_addr);
2289         read_unlock(&dev_base_lock);
2290
2291         atomic64_set(&ibdev->iboe.mac[port - 1], new_smac);
2292
2293         /* no need for update QP1 and mac registration in non-SRIOV */
2294         if (!mlx4_is_mfunc(ibdev->dev))
2295                 return;
2296
2297         mutex_lock(&ibdev->qp1_proxy_lock[port - 1]);
2298         qp = ibdev->qp1_proxy[port - 1];
2299         if (qp) {
2300                 int new_smac_index;
2301                 u64 old_smac;
2302                 struct mlx4_update_qp_params update_params;
2303
2304                 mutex_lock(&qp->mutex);
2305                 old_smac = qp->pri.smac;
2306                 if (new_smac == old_smac)
2307                         goto unlock;
2308
2309                 new_smac_index = mlx4_register_mac(ibdev->dev, port, new_smac);
2310
2311                 if (new_smac_index < 0)
2312                         goto unlock;
2313
2314                 update_params.smac_index = new_smac_index;
2315                 if (mlx4_update_qp(ibdev->dev, qp->mqp.qpn, MLX4_UPDATE_QP_SMAC,
2316                                    &update_params)) {
2317                         release_mac = new_smac;
2318                         goto unlock;
2319                 }
2320                 /* if old port was zero, no mac was yet registered for this QP */
2321                 if (qp->pri.smac_port)
2322                         release_mac = old_smac;
2323                 qp->pri.smac = new_smac;
2324                 qp->pri.smac_port = port;
2325                 qp->pri.smac_index = new_smac_index;
2326         }
2327
2328 unlock:
2329         if (release_mac != MLX4_IB_INVALID_MAC)
2330                 mlx4_unregister_mac(ibdev->dev, port, release_mac);
2331         if (qp)
2332                 mutex_unlock(&qp->mutex);
2333         mutex_unlock(&ibdev->qp1_proxy_lock[port - 1]);
2334 }
2335
2336 static void mlx4_ib_scan_netdevs(struct mlx4_ib_dev *ibdev,
2337                                  struct net_device *dev,
2338                                  unsigned long event)
2339
2340 {
2341         struct mlx4_ib_iboe *iboe;
2342         int update_qps_port = -1;
2343         int port;
2344
2345         ASSERT_RTNL();
2346
2347         iboe = &ibdev->iboe;
2348
2349         spin_lock_bh(&iboe->lock);
2350         mlx4_foreach_ib_transport_port(port, ibdev->dev) {
2351
2352                 iboe->netdevs[port - 1] =
2353                         mlx4_get_protocol_dev(ibdev->dev, MLX4_PROT_ETH, port);
2354
2355                 if (dev == iboe->netdevs[port - 1] &&
2356                     (event == NETDEV_CHANGEADDR || event == NETDEV_REGISTER ||
2357                      event == NETDEV_UP || event == NETDEV_CHANGE))
2358                         update_qps_port = port;
2359
2360                 if (dev == iboe->netdevs[port - 1] &&
2361                     (event == NETDEV_UP || event == NETDEV_DOWN)) {
2362                         enum ib_port_state port_state;
2363                         struct ib_event ibev = { };
2364
2365                         if (ib_get_cached_port_state(&ibdev->ib_dev, port,
2366                                                      &port_state))
2367                                 continue;
2368
2369                         if (event == NETDEV_UP &&
2370                             (port_state != IB_PORT_ACTIVE ||
2371                              iboe->last_port_state[port - 1] != IB_PORT_DOWN))
2372                                 continue;
2373                         if (event == NETDEV_DOWN &&
2374                             (port_state != IB_PORT_DOWN ||
2375                              iboe->last_port_state[port - 1] != IB_PORT_ACTIVE))
2376                                 continue;
2377                         iboe->last_port_state[port - 1] = port_state;
2378
2379                         ibev.device = &ibdev->ib_dev;
2380                         ibev.element.port_num = port;
2381                         ibev.event = event == NETDEV_UP ? IB_EVENT_PORT_ACTIVE :
2382                                                           IB_EVENT_PORT_ERR;
2383                         ib_dispatch_event(&ibev);
2384                 }
2385
2386         }
2387         spin_unlock_bh(&iboe->lock);
2388
2389         if (update_qps_port > 0)
2390                 mlx4_ib_update_qps(ibdev, dev, update_qps_port);
2391 }
2392
2393 static int mlx4_ib_netdev_event(struct notifier_block *this,
2394                                 unsigned long event, void *ptr)
2395 {
2396         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
2397         struct mlx4_ib_dev *ibdev;
2398
2399         if (!net_eq(dev_net(dev), &init_net))
2400                 return NOTIFY_DONE;
2401
2402         ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb);
2403         mlx4_ib_scan_netdevs(ibdev, dev, event);
2404
2405         return NOTIFY_DONE;
2406 }
2407
2408 static void init_pkeys(struct mlx4_ib_dev *ibdev)
2409 {
2410         int port;
2411         int slave;
2412         int i;
2413
2414         if (mlx4_is_master(ibdev->dev)) {
2415                 for (slave = 0; slave <= ibdev->dev->persist->num_vfs;
2416                      ++slave) {
2417                         for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
2418                                 for (i = 0;
2419                                      i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
2420                                      ++i) {
2421                                         ibdev->pkeys.virt2phys_pkey[slave][port - 1][i] =
2422                                         /* master has the identity virt2phys pkey mapping */
2423                                                 (slave == mlx4_master_func_num(ibdev->dev) || !i) ? i :
2424                                                         ibdev->dev->phys_caps.pkey_phys_table_len[port] - 1;
2425                                         mlx4_sync_pkey_table(ibdev->dev, slave, port, i,
2426                                                              ibdev->pkeys.virt2phys_pkey[slave][port - 1][i]);
2427                                 }
2428                         }
2429                 }
2430                 /* initialize pkey cache */
2431                 for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
2432                         for (i = 0;
2433                              i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
2434                              ++i)
2435                                 ibdev->pkeys.phys_pkey_cache[port-1][i] =
2436                                         (i) ? 0 : 0xFFFF;
2437                 }
2438         }
2439 }
2440
2441 static void mlx4_ib_alloc_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
2442 {
2443         int i, j, eq = 0, total_eqs = 0;
2444
2445         ibdev->eq_table = kcalloc(dev->caps.num_comp_vectors,
2446                                   sizeof(ibdev->eq_table[0]), GFP_KERNEL);
2447         if (!ibdev->eq_table)
2448                 return;
2449
2450         for (i = 1; i <= dev->caps.num_ports; i++) {
2451                 for (j = 0; j < mlx4_get_eqs_per_port(dev, i);
2452                      j++, total_eqs++) {
2453                         if (i > 1 &&  mlx4_is_eq_shared(dev, total_eqs))
2454                                 continue;
2455                         ibdev->eq_table[eq] = total_eqs;
2456                         if (!mlx4_assign_eq(dev, i,
2457                                             &ibdev->eq_table[eq]))
2458                                 eq++;
2459                         else
2460                                 ibdev->eq_table[eq] = -1;
2461                 }
2462         }
2463
2464         for (i = eq; i < dev->caps.num_comp_vectors;
2465              ibdev->eq_table[i++] = -1)
2466                 ;
2467
2468         /* Advertise the new number of EQs to clients */
2469         ibdev->ib_dev.num_comp_vectors = eq;
2470 }
2471
2472 static void mlx4_ib_free_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
2473 {
2474         int i;
2475         int total_eqs = ibdev->ib_dev.num_comp_vectors;
2476
2477         /* no eqs were allocated */
2478         if (!ibdev->eq_table)
2479                 return;
2480
2481         /* Reset the advertised EQ number */
2482         ibdev->ib_dev.num_comp_vectors = 0;
2483
2484         for (i = 0; i < total_eqs; i++)
2485                 mlx4_release_eq(dev, ibdev->eq_table[i]);
2486
2487         kfree(ibdev->eq_table);
2488         ibdev->eq_table = NULL;
2489 }
2490
2491 static int mlx4_port_immutable(struct ib_device *ibdev, u8 port_num,
2492                                struct ib_port_immutable *immutable)
2493 {
2494         struct ib_port_attr attr;
2495         struct mlx4_ib_dev *mdev = to_mdev(ibdev);
2496         int err;
2497
2498         if (mlx4_ib_port_link_layer(ibdev, port_num) == IB_LINK_LAYER_INFINIBAND) {
2499                 immutable->core_cap_flags = RDMA_CORE_PORT_IBA_IB;
2500                 immutable->max_mad_size = IB_MGMT_MAD_SIZE;
2501         } else {
2502                 if (mdev->dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE)
2503                         immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE;
2504                 if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ROCE_V1_V2)
2505                         immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE |
2506                                 RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP;
2507                 immutable->core_cap_flags |= RDMA_CORE_PORT_RAW_PACKET;
2508                 if (immutable->core_cap_flags & (RDMA_CORE_PORT_IBA_ROCE |
2509                     RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP))
2510                         immutable->max_mad_size = IB_MGMT_MAD_SIZE;
2511         }
2512
2513         err = ib_query_port(ibdev, port_num, &attr);
2514         if (err)
2515                 return err;
2516
2517         immutable->pkey_tbl_len = attr.pkey_tbl_len;
2518         immutable->gid_tbl_len = attr.gid_tbl_len;
2519
2520         return 0;
2521 }
2522
2523 static void get_fw_ver_str(struct ib_device *device, char *str)
2524 {
2525         struct mlx4_ib_dev *dev =
2526                 container_of(device, struct mlx4_ib_dev, ib_dev);
2527         snprintf(str, IB_FW_VERSION_NAME_MAX, "%d.%d.%d",
2528                  (int) (dev->dev->caps.fw_ver >> 32),
2529                  (int) (dev->dev->caps.fw_ver >> 16) & 0xffff,
2530                  (int) dev->dev->caps.fw_ver & 0xffff);
2531 }
2532
2533 static const struct ib_device_ops mlx4_ib_dev_ops = {
2534         .owner = THIS_MODULE,
2535         .driver_id = RDMA_DRIVER_MLX4,
2536         .uverbs_abi_ver = MLX4_IB_UVERBS_ABI_VERSION,
2537
2538         .add_gid = mlx4_ib_add_gid,
2539         .alloc_mr = mlx4_ib_alloc_mr,
2540         .alloc_pd = mlx4_ib_alloc_pd,
2541         .alloc_ucontext = mlx4_ib_alloc_ucontext,
2542         .attach_mcast = mlx4_ib_mcg_attach,
2543         .create_ah = mlx4_ib_create_ah,
2544         .create_cq = mlx4_ib_create_cq,
2545         .create_qp = mlx4_ib_create_qp,
2546         .create_srq = mlx4_ib_create_srq,
2547         .dealloc_pd = mlx4_ib_dealloc_pd,
2548         .dealloc_ucontext = mlx4_ib_dealloc_ucontext,
2549         .del_gid = mlx4_ib_del_gid,
2550         .dereg_mr = mlx4_ib_dereg_mr,
2551         .destroy_ah = mlx4_ib_destroy_ah,
2552         .destroy_cq = mlx4_ib_destroy_cq,
2553         .destroy_qp = mlx4_ib_destroy_qp,
2554         .destroy_srq = mlx4_ib_destroy_srq,
2555         .detach_mcast = mlx4_ib_mcg_detach,
2556         .disassociate_ucontext = mlx4_ib_disassociate_ucontext,
2557         .drain_rq = mlx4_ib_drain_rq,
2558         .drain_sq = mlx4_ib_drain_sq,
2559         .get_dev_fw_str = get_fw_ver_str,
2560         .get_dma_mr = mlx4_ib_get_dma_mr,
2561         .get_link_layer = mlx4_ib_port_link_layer,
2562         .get_netdev = mlx4_ib_get_netdev,
2563         .get_port_immutable = mlx4_port_immutable,
2564         .map_mr_sg = mlx4_ib_map_mr_sg,
2565         .mmap = mlx4_ib_mmap,
2566         .modify_cq = mlx4_ib_modify_cq,
2567         .modify_device = mlx4_ib_modify_device,
2568         .modify_port = mlx4_ib_modify_port,
2569         .modify_qp = mlx4_ib_modify_qp,
2570         .modify_srq = mlx4_ib_modify_srq,
2571         .poll_cq = mlx4_ib_poll_cq,
2572         .post_recv = mlx4_ib_post_recv,
2573         .post_send = mlx4_ib_post_send,
2574         .post_srq_recv = mlx4_ib_post_srq_recv,
2575         .process_mad = mlx4_ib_process_mad,
2576         .query_ah = mlx4_ib_query_ah,
2577         .query_device = mlx4_ib_query_device,
2578         .query_gid = mlx4_ib_query_gid,
2579         .query_pkey = mlx4_ib_query_pkey,
2580         .query_port = mlx4_ib_query_port,
2581         .query_qp = mlx4_ib_query_qp,
2582         .query_srq = mlx4_ib_query_srq,
2583         .reg_user_mr = mlx4_ib_reg_user_mr,
2584         .req_notify_cq = mlx4_ib_arm_cq,
2585         .rereg_user_mr = mlx4_ib_rereg_user_mr,
2586         .resize_cq = mlx4_ib_resize_cq,
2587
2588         INIT_RDMA_OBJ_SIZE(ib_ah, mlx4_ib_ah, ibah),
2589         INIT_RDMA_OBJ_SIZE(ib_cq, mlx4_ib_cq, ibcq),
2590         INIT_RDMA_OBJ_SIZE(ib_pd, mlx4_ib_pd, ibpd),
2591         INIT_RDMA_OBJ_SIZE(ib_srq, mlx4_ib_srq, ibsrq),
2592         INIT_RDMA_OBJ_SIZE(ib_ucontext, mlx4_ib_ucontext, ibucontext),
2593 };
2594
2595 static const struct ib_device_ops mlx4_ib_dev_wq_ops = {
2596         .create_rwq_ind_table = mlx4_ib_create_rwq_ind_table,
2597         .create_wq = mlx4_ib_create_wq,
2598         .destroy_rwq_ind_table = mlx4_ib_destroy_rwq_ind_table,
2599         .destroy_wq = mlx4_ib_destroy_wq,
2600         .modify_wq = mlx4_ib_modify_wq,
2601 };
2602
2603 static const struct ib_device_ops mlx4_ib_dev_fmr_ops = {
2604         .alloc_fmr = mlx4_ib_fmr_alloc,
2605         .dealloc_fmr = mlx4_ib_fmr_dealloc,
2606         .map_phys_fmr = mlx4_ib_map_phys_fmr,
2607         .unmap_fmr = mlx4_ib_unmap_fmr,
2608 };
2609
2610 static const struct ib_device_ops mlx4_ib_dev_mw_ops = {
2611         .alloc_mw = mlx4_ib_alloc_mw,
2612         .dealloc_mw = mlx4_ib_dealloc_mw,
2613 };
2614
2615 static const struct ib_device_ops mlx4_ib_dev_xrc_ops = {
2616         .alloc_xrcd = mlx4_ib_alloc_xrcd,
2617         .dealloc_xrcd = mlx4_ib_dealloc_xrcd,
2618 };
2619
2620 static const struct ib_device_ops mlx4_ib_dev_fs_ops = {
2621         .create_flow = mlx4_ib_create_flow,
2622         .destroy_flow = mlx4_ib_destroy_flow,
2623 };
2624
2625 static void *mlx4_ib_add(struct mlx4_dev *dev)
2626 {
2627         struct mlx4_ib_dev *ibdev;
2628         int num_ports = 0;
2629         int i, j;
2630         int err;
2631         struct mlx4_ib_iboe *iboe;
2632         int ib_num_ports = 0;
2633         int num_req_counters;
2634         int allocated;
2635         u32 counter_index;
2636         struct counter_index *new_counter_index = NULL;
2637
2638         pr_info_once("%s", mlx4_ib_version);
2639
2640         num_ports = 0;
2641         mlx4_foreach_ib_transport_port(i, dev)
2642                 num_ports++;
2643
2644         /* No point in registering a device with no ports... */
2645         if (num_ports == 0)
2646                 return NULL;
2647
2648         ibdev = ib_alloc_device(mlx4_ib_dev, ib_dev);
2649         if (!ibdev) {
2650                 dev_err(&dev->persist->pdev->dev,
2651                         "Device struct alloc failed\n");
2652                 return NULL;
2653         }
2654
2655         iboe = &ibdev->iboe;
2656
2657         if (mlx4_pd_alloc(dev, &ibdev->priv_pdn))
2658                 goto err_dealloc;
2659
2660         if (mlx4_uar_alloc(dev, &ibdev->priv_uar))
2661                 goto err_pd;
2662
2663         ibdev->uar_map = ioremap((phys_addr_t) ibdev->priv_uar.pfn << PAGE_SHIFT,
2664                                  PAGE_SIZE);
2665         if (!ibdev->uar_map)
2666                 goto err_uar;
2667         MLX4_INIT_DOORBELL_LOCK(&ibdev->uar_lock);
2668
2669         ibdev->dev = dev;
2670         ibdev->bond_next_port   = 0;
2671
2672         ibdev->ib_dev.node_type         = RDMA_NODE_IB_CA;
2673         ibdev->ib_dev.local_dma_lkey    = dev->caps.reserved_lkey;
2674         ibdev->num_ports                = num_ports;
2675         ibdev->ib_dev.phys_port_cnt     = mlx4_is_bonded(dev) ?
2676                                                 1 : ibdev->num_ports;
2677         ibdev->ib_dev.num_comp_vectors  = dev->caps.num_comp_vectors;
2678         ibdev->ib_dev.dev.parent        = &dev->persist->pdev->dev;
2679
2680         ibdev->ib_dev.uverbs_cmd_mask   =
2681                 (1ull << IB_USER_VERBS_CMD_GET_CONTEXT)         |
2682                 (1ull << IB_USER_VERBS_CMD_QUERY_DEVICE)        |
2683                 (1ull << IB_USER_VERBS_CMD_QUERY_PORT)          |
2684                 (1ull << IB_USER_VERBS_CMD_ALLOC_PD)            |
2685                 (1ull << IB_USER_VERBS_CMD_DEALLOC_PD)          |
2686                 (1ull << IB_USER_VERBS_CMD_REG_MR)              |
2687                 (1ull << IB_USER_VERBS_CMD_REREG_MR)            |
2688                 (1ull << IB_USER_VERBS_CMD_DEREG_MR)            |
2689                 (1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL) |
2690                 (1ull << IB_USER_VERBS_CMD_CREATE_CQ)           |
2691                 (1ull << IB_USER_VERBS_CMD_RESIZE_CQ)           |
2692                 (1ull << IB_USER_VERBS_CMD_DESTROY_CQ)          |
2693                 (1ull << IB_USER_VERBS_CMD_CREATE_QP)           |
2694                 (1ull << IB_USER_VERBS_CMD_MODIFY_QP)           |
2695                 (1ull << IB_USER_VERBS_CMD_QUERY_QP)            |
2696                 (1ull << IB_USER_VERBS_CMD_DESTROY_QP)          |
2697                 (1ull << IB_USER_VERBS_CMD_ATTACH_MCAST)        |
2698                 (1ull << IB_USER_VERBS_CMD_DETACH_MCAST)        |
2699                 (1ull << IB_USER_VERBS_CMD_CREATE_SRQ)          |
2700                 (1ull << IB_USER_VERBS_CMD_MODIFY_SRQ)          |
2701                 (1ull << IB_USER_VERBS_CMD_QUERY_SRQ)           |
2702                 (1ull << IB_USER_VERBS_CMD_DESTROY_SRQ)         |
2703                 (1ull << IB_USER_VERBS_CMD_CREATE_XSRQ)         |
2704                 (1ull << IB_USER_VERBS_CMD_OPEN_QP);
2705
2706         ib_set_device_ops(&ibdev->ib_dev, &mlx4_ib_dev_ops);
2707         ibdev->ib_dev.uverbs_ex_cmd_mask |=
2708                 (1ull << IB_USER_VERBS_EX_CMD_MODIFY_CQ) |
2709                 (1ull << IB_USER_VERBS_EX_CMD_QUERY_DEVICE) |
2710                 (1ull << IB_USER_VERBS_EX_CMD_CREATE_CQ) |
2711                 (1ull << IB_USER_VERBS_EX_CMD_CREATE_QP);
2712
2713         if ((dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_RSS) &&
2714             ((mlx4_ib_port_link_layer(&ibdev->ib_dev, 1) ==
2715             IB_LINK_LAYER_ETHERNET) ||
2716             (mlx4_ib_port_link_layer(&ibdev->ib_dev, 2) ==
2717             IB_LINK_LAYER_ETHERNET))) {
2718                 ibdev->ib_dev.uverbs_ex_cmd_mask |=
2719                         (1ull << IB_USER_VERBS_EX_CMD_CREATE_WQ)          |
2720                         (1ull << IB_USER_VERBS_EX_CMD_MODIFY_WQ)          |
2721                         (1ull << IB_USER_VERBS_EX_CMD_DESTROY_WQ)         |
2722                         (1ull << IB_USER_VERBS_EX_CMD_CREATE_RWQ_IND_TBL) |
2723                         (1ull << IB_USER_VERBS_EX_CMD_DESTROY_RWQ_IND_TBL);
2724                 ib_set_device_ops(&ibdev->ib_dev, &mlx4_ib_dev_wq_ops);
2725         }
2726
2727         if (!mlx4_is_slave(ibdev->dev))
2728                 ib_set_device_ops(&ibdev->ib_dev, &mlx4_ib_dev_fmr_ops);
2729
2730         if (dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW ||
2731             dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
2732                 ibdev->ib_dev.uverbs_cmd_mask |=
2733                         (1ull << IB_USER_VERBS_CMD_ALLOC_MW) |
2734                         (1ull << IB_USER_VERBS_CMD_DEALLOC_MW);
2735                 ib_set_device_ops(&ibdev->ib_dev, &mlx4_ib_dev_mw_ops);
2736         }
2737
2738         if (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) {
2739                 ibdev->ib_dev.uverbs_cmd_mask |=
2740                         (1ull << IB_USER_VERBS_CMD_OPEN_XRCD) |
2741                         (1ull << IB_USER_VERBS_CMD_CLOSE_XRCD);
2742                 ib_set_device_ops(&ibdev->ib_dev, &mlx4_ib_dev_xrc_ops);
2743         }
2744
2745         if (check_flow_steering_support(dev)) {
2746                 ibdev->steering_support = MLX4_STEERING_MODE_DEVICE_MANAGED;
2747                 ibdev->ib_dev.uverbs_ex_cmd_mask        |=
2748                         (1ull << IB_USER_VERBS_EX_CMD_CREATE_FLOW) |
2749                         (1ull << IB_USER_VERBS_EX_CMD_DESTROY_FLOW);
2750                 ib_set_device_ops(&ibdev->ib_dev, &mlx4_ib_dev_fs_ops);
2751         }
2752
2753         if (!dev->caps.userspace_caps)
2754                 ibdev->ib_dev.ops.uverbs_abi_ver =
2755                         MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION;
2756
2757         mlx4_ib_alloc_eqs(dev, ibdev);
2758
2759         spin_lock_init(&iboe->lock);
2760
2761         if (init_node_data(ibdev))
2762                 goto err_map;
2763         mlx4_init_sl2vl_tbl(ibdev);
2764
2765         for (i = 0; i < ibdev->num_ports; ++i) {
2766                 mutex_init(&ibdev->counters_table[i].mutex);
2767                 INIT_LIST_HEAD(&ibdev->counters_table[i].counters_list);
2768                 iboe->last_port_state[i] = IB_PORT_DOWN;
2769         }
2770
2771         num_req_counters = mlx4_is_bonded(dev) ? 1 : ibdev->num_ports;
2772         for (i = 0; i < num_req_counters; ++i) {
2773                 mutex_init(&ibdev->qp1_proxy_lock[i]);
2774                 allocated = 0;
2775                 if (mlx4_ib_port_link_layer(&ibdev->ib_dev, i + 1) ==
2776                                                 IB_LINK_LAYER_ETHERNET) {
2777                         err = mlx4_counter_alloc(ibdev->dev, &counter_index,
2778                                                  MLX4_RES_USAGE_DRIVER);
2779                         /* if failed to allocate a new counter, use default */
2780                         if (err)
2781                                 counter_index =
2782                                         mlx4_get_default_counter_index(dev,
2783                                                                        i + 1);
2784                         else
2785                                 allocated = 1;
2786                 } else { /* IB_LINK_LAYER_INFINIBAND use the default counter */
2787                         counter_index = mlx4_get_default_counter_index(dev,
2788                                                                        i + 1);
2789                 }
2790                 new_counter_index = kmalloc(sizeof(*new_counter_index),
2791                                             GFP_KERNEL);
2792                 if (!new_counter_index) {
2793                         if (allocated)
2794                                 mlx4_counter_free(ibdev->dev, counter_index);
2795                         goto err_counter;
2796                 }
2797                 new_counter_index->index = counter_index;
2798                 new_counter_index->allocated = allocated;
2799                 list_add_tail(&new_counter_index->list,
2800                               &ibdev->counters_table[i].counters_list);
2801                 ibdev->counters_table[i].default_counter = counter_index;
2802                 pr_info("counter index %d for port %d allocated %d\n",
2803                         counter_index, i + 1, allocated);
2804         }
2805         if (mlx4_is_bonded(dev))
2806                 for (i = 1; i < ibdev->num_ports ; ++i) {
2807                         new_counter_index =
2808                                         kmalloc(sizeof(struct counter_index),
2809                                                 GFP_KERNEL);
2810                         if (!new_counter_index)
2811                                 goto err_counter;
2812                         new_counter_index->index = counter_index;
2813                         new_counter_index->allocated = 0;
2814                         list_add_tail(&new_counter_index->list,
2815                                       &ibdev->counters_table[i].counters_list);
2816                         ibdev->counters_table[i].default_counter =
2817                                                                 counter_index;
2818                 }
2819
2820         mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
2821                 ib_num_ports++;
2822
2823         spin_lock_init(&ibdev->sm_lock);
2824         mutex_init(&ibdev->cap_mask_mutex);
2825         INIT_LIST_HEAD(&ibdev->qp_list);
2826         spin_lock_init(&ibdev->reset_flow_resource_lock);
2827
2828         if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED &&
2829             ib_num_ports) {
2830                 ibdev->steer_qpn_count = MLX4_IB_UC_MAX_NUM_QPS;
2831                 err = mlx4_qp_reserve_range(dev, ibdev->steer_qpn_count,
2832                                             MLX4_IB_UC_STEER_QPN_ALIGN,
2833                                             &ibdev->steer_qpn_base, 0,
2834                                             MLX4_RES_USAGE_DRIVER);
2835                 if (err)
2836                         goto err_counter;
2837
2838                 ibdev->ib_uc_qpns_bitmap =
2839                         kmalloc_array(BITS_TO_LONGS(ibdev->steer_qpn_count),
2840                                       sizeof(long),
2841                                       GFP_KERNEL);
2842                 if (!ibdev->ib_uc_qpns_bitmap)
2843                         goto err_steer_qp_release;
2844
2845                 if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DMFS_IPOIB) {
2846                         bitmap_zero(ibdev->ib_uc_qpns_bitmap,
2847                                     ibdev->steer_qpn_count);
2848                         err = mlx4_FLOW_STEERING_IB_UC_QP_RANGE(
2849                                         dev, ibdev->steer_qpn_base,
2850                                         ibdev->steer_qpn_base +
2851                                         ibdev->steer_qpn_count - 1);
2852                         if (err)
2853                                 goto err_steer_free_bitmap;
2854                 } else {
2855                         bitmap_fill(ibdev->ib_uc_qpns_bitmap,
2856                                     ibdev->steer_qpn_count);
2857                 }
2858         }
2859
2860         for (j = 1; j <= ibdev->dev->caps.num_ports; j++)
2861                 atomic64_set(&iboe->mac[j - 1], ibdev->dev->caps.def_mac[j]);
2862
2863         if (mlx4_ib_alloc_diag_counters(ibdev))
2864                 goto err_steer_free_bitmap;
2865
2866         rdma_set_device_sysfs_group(&ibdev->ib_dev, &mlx4_attr_group);
2867         if (ib_register_device(&ibdev->ib_dev, "mlx4_%d"))
2868                 goto err_diag_counters;
2869
2870         if (mlx4_ib_mad_init(ibdev))
2871                 goto err_reg;
2872
2873         if (mlx4_ib_init_sriov(ibdev))
2874                 goto err_mad;
2875
2876         if (!iboe->nb.notifier_call) {
2877                 iboe->nb.notifier_call = mlx4_ib_netdev_event;
2878                 err = register_netdevice_notifier(&iboe->nb);
2879                 if (err) {
2880                         iboe->nb.notifier_call = NULL;
2881                         goto err_notif;
2882                 }
2883         }
2884         if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ROCE_V1_V2) {
2885                 err = mlx4_config_roce_v2_port(dev, ROCE_V2_UDP_DPORT);
2886                 if (err)
2887                         goto err_notif;
2888         }
2889
2890         ibdev->ib_active = true;
2891         mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
2892                 devlink_port_type_ib_set(mlx4_get_devlink_port(dev, i),
2893                                          &ibdev->ib_dev);
2894
2895         if (mlx4_is_mfunc(ibdev->dev))
2896                 init_pkeys(ibdev);
2897
2898         /* create paravirt contexts for any VFs which are active */
2899         if (mlx4_is_master(ibdev->dev)) {
2900                 for (j = 0; j < MLX4_MFUNC_MAX; j++) {
2901                         if (j == mlx4_master_func_num(ibdev->dev))
2902                                 continue;
2903                         if (mlx4_is_slave_active(ibdev->dev, j))
2904                                 do_slave_init(ibdev, j, 1);
2905                 }
2906         }
2907         return ibdev;
2908
2909 err_notif:
2910         if (ibdev->iboe.nb.notifier_call) {
2911                 if (unregister_netdevice_notifier(&ibdev->iboe.nb))
2912                         pr_warn("failure unregistering notifier\n");
2913                 ibdev->iboe.nb.notifier_call = NULL;
2914         }
2915         flush_workqueue(wq);
2916
2917         mlx4_ib_close_sriov(ibdev);
2918
2919 err_mad:
2920         mlx4_ib_mad_cleanup(ibdev);
2921
2922 err_reg:
2923         ib_unregister_device(&ibdev->ib_dev);
2924
2925 err_diag_counters:
2926         mlx4_ib_diag_cleanup(ibdev);
2927
2928 err_steer_free_bitmap:
2929         kfree(ibdev->ib_uc_qpns_bitmap);
2930
2931 err_steer_qp_release:
2932         mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
2933                               ibdev->steer_qpn_count);
2934 err_counter:
2935         for (i = 0; i < ibdev->num_ports; ++i)
2936                 mlx4_ib_delete_counters_table(ibdev, &ibdev->counters_table[i]);
2937
2938 err_map:
2939         mlx4_ib_free_eqs(dev, ibdev);
2940         iounmap(ibdev->uar_map);
2941
2942 err_uar:
2943         mlx4_uar_free(dev, &ibdev->priv_uar);
2944
2945 err_pd:
2946         mlx4_pd_free(dev, ibdev->priv_pdn);
2947
2948 err_dealloc:
2949         ib_dealloc_device(&ibdev->ib_dev);
2950
2951         return NULL;
2952 }
2953
2954 int mlx4_ib_steer_qp_alloc(struct mlx4_ib_dev *dev, int count, int *qpn)
2955 {
2956         int offset;
2957
2958         WARN_ON(!dev->ib_uc_qpns_bitmap);
2959
2960         offset = bitmap_find_free_region(dev->ib_uc_qpns_bitmap,
2961                                          dev->steer_qpn_count,
2962                                          get_count_order(count));
2963         if (offset < 0)
2964                 return offset;
2965
2966         *qpn = dev->steer_qpn_base + offset;
2967         return 0;
2968 }
2969
2970 void mlx4_ib_steer_qp_free(struct mlx4_ib_dev *dev, u32 qpn, int count)
2971 {
2972         if (!qpn ||
2973             dev->steering_support != MLX4_STEERING_MODE_DEVICE_MANAGED)
2974                 return;
2975
2976         if (WARN(qpn < dev->steer_qpn_base, "qpn = %u, steer_qpn_base = %u\n",
2977                  qpn, dev->steer_qpn_base))
2978                 /* not supposed to be here */
2979                 return;
2980
2981         bitmap_release_region(dev->ib_uc_qpns_bitmap,
2982                               qpn - dev->steer_qpn_base,
2983                               get_count_order(count));
2984 }
2985
2986 int mlx4_ib_steer_qp_reg(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
2987                          int is_attach)
2988 {
2989         int err;
2990         size_t flow_size;
2991         struct ib_flow_attr *flow = NULL;
2992         struct ib_flow_spec_ib *ib_spec;
2993
2994         if (is_attach) {
2995                 flow_size = sizeof(struct ib_flow_attr) +
2996                             sizeof(struct ib_flow_spec_ib);
2997                 flow = kzalloc(flow_size, GFP_KERNEL);
2998                 if (!flow)
2999                         return -ENOMEM;
3000                 flow->port = mqp->port;
3001                 flow->num_of_specs = 1;
3002                 flow->size = flow_size;
3003                 ib_spec = (struct ib_flow_spec_ib *)(flow + 1);
3004                 ib_spec->type = IB_FLOW_SPEC_IB;
3005                 ib_spec->size = sizeof(struct ib_flow_spec_ib);
3006                 /* Add an empty rule for IB L2 */
3007                 memset(&ib_spec->mask, 0, sizeof(ib_spec->mask));
3008
3009                 err = __mlx4_ib_create_flow(&mqp->ibqp, flow,
3010                                             IB_FLOW_DOMAIN_NIC,
3011                                             MLX4_FS_REGULAR,
3012                                             &mqp->reg_id);
3013         } else {
3014                 err = __mlx4_ib_destroy_flow(mdev->dev, mqp->reg_id);
3015         }
3016         kfree(flow);
3017         return err;
3018 }
3019
3020 static void mlx4_ib_remove(struct mlx4_dev *dev, void *ibdev_ptr)
3021 {
3022         struct mlx4_ib_dev *ibdev = ibdev_ptr;
3023         int p;
3024         int i;
3025
3026         mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
3027                 devlink_port_type_clear(mlx4_get_devlink_port(dev, i));
3028         ibdev->ib_active = false;
3029         flush_workqueue(wq);
3030
3031         if (ibdev->iboe.nb.notifier_call) {
3032                 if (unregister_netdevice_notifier(&ibdev->iboe.nb))
3033                         pr_warn("failure unregistering notifier\n");
3034                 ibdev->iboe.nb.notifier_call = NULL;
3035         }
3036
3037         mlx4_ib_close_sriov(ibdev);
3038         mlx4_ib_mad_cleanup(ibdev);
3039         ib_unregister_device(&ibdev->ib_dev);
3040         mlx4_ib_diag_cleanup(ibdev);
3041
3042         mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
3043                               ibdev->steer_qpn_count);
3044         kfree(ibdev->ib_uc_qpns_bitmap);
3045
3046         iounmap(ibdev->uar_map);
3047         for (p = 0; p < ibdev->num_ports; ++p)
3048                 mlx4_ib_delete_counters_table(ibdev, &ibdev->counters_table[p]);
3049
3050         mlx4_foreach_port(p, dev, MLX4_PORT_TYPE_IB)
3051                 mlx4_CLOSE_PORT(dev, p);
3052
3053         mlx4_ib_free_eqs(dev, ibdev);
3054
3055         mlx4_uar_free(dev, &ibdev->priv_uar);
3056         mlx4_pd_free(dev, ibdev->priv_pdn);
3057         ib_dealloc_device(&ibdev->ib_dev);
3058 }
3059
3060 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init)
3061 {
3062         struct mlx4_ib_demux_work **dm = NULL;
3063         struct mlx4_dev *dev = ibdev->dev;
3064         int i;
3065         unsigned long flags;
3066         struct mlx4_active_ports actv_ports;
3067         unsigned int ports;
3068         unsigned int first_port;
3069
3070         if (!mlx4_is_master(dev))
3071                 return;
3072
3073         actv_ports = mlx4_get_active_ports(dev, slave);
3074         ports = bitmap_weight(actv_ports.ports, dev->caps.num_ports);
3075         first_port = find_first_bit(actv_ports.ports, dev->caps.num_ports);
3076
3077         dm = kcalloc(ports, sizeof(*dm), GFP_ATOMIC);
3078         if (!dm)
3079                 return;
3080
3081         for (i = 0; i < ports; i++) {
3082                 dm[i] = kmalloc(sizeof (struct mlx4_ib_demux_work), GFP_ATOMIC);
3083                 if (!dm[i]) {
3084                         while (--i >= 0)
3085                                 kfree(dm[i]);
3086                         goto out;
3087                 }
3088                 INIT_WORK(&dm[i]->work, mlx4_ib_tunnels_update_work);
3089                 dm[i]->port = first_port + i + 1;
3090                 dm[i]->slave = slave;
3091                 dm[i]->do_init = do_init;
3092                 dm[i]->dev = ibdev;
3093         }
3094         /* initialize or tear down tunnel QPs for the slave */
3095         spin_lock_irqsave(&ibdev->sriov.going_down_lock, flags);
3096         if (!ibdev->sriov.is_going_down) {
3097                 for (i = 0; i < ports; i++)
3098                         queue_work(ibdev->sriov.demux[i].ud_wq, &dm[i]->work);
3099                 spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
3100         } else {
3101                 spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
3102                 for (i = 0; i < ports; i++)
3103                         kfree(dm[i]);
3104         }
3105 out:
3106         kfree(dm);
3107         return;
3108 }
3109
3110 static void mlx4_ib_handle_catas_error(struct mlx4_ib_dev *ibdev)
3111 {
3112         struct mlx4_ib_qp *mqp;
3113         unsigned long flags_qp;
3114         unsigned long flags_cq;
3115         struct mlx4_ib_cq *send_mcq, *recv_mcq;
3116         struct list_head    cq_notify_list;
3117         struct mlx4_cq *mcq;
3118         unsigned long flags;
3119
3120         pr_warn("mlx4_ib_handle_catas_error was started\n");
3121         INIT_LIST_HEAD(&cq_notify_list);
3122
3123         /* Go over qp list reside on that ibdev, sync with create/destroy qp.*/
3124         spin_lock_irqsave(&ibdev->reset_flow_resource_lock, flags);
3125
3126         list_for_each_entry(mqp, &ibdev->qp_list, qps_list) {
3127                 spin_lock_irqsave(&mqp->sq.lock, flags_qp);
3128                 if (mqp->sq.tail != mqp->sq.head) {
3129                         send_mcq = to_mcq(mqp->ibqp.send_cq);
3130                         spin_lock_irqsave(&send_mcq->lock, flags_cq);
3131                         if (send_mcq->mcq.comp &&
3132                             mqp->ibqp.send_cq->comp_handler) {
3133                                 if (!send_mcq->mcq.reset_notify_added) {
3134                                         send_mcq->mcq.reset_notify_added = 1;
3135                                         list_add_tail(&send_mcq->mcq.reset_notify,
3136                                                       &cq_notify_list);
3137                                 }
3138                         }
3139                         spin_unlock_irqrestore(&send_mcq->lock, flags_cq);
3140                 }
3141                 spin_unlock_irqrestore(&mqp->sq.lock, flags_qp);
3142                 /* Now, handle the QP's receive queue */
3143                 spin_lock_irqsave(&mqp->rq.lock, flags_qp);
3144                 /* no handling is needed for SRQ */
3145                 if (!mqp->ibqp.srq) {
3146                         if (mqp->rq.tail != mqp->rq.head) {
3147                                 recv_mcq = to_mcq(mqp->ibqp.recv_cq);
3148                                 spin_lock_irqsave(&recv_mcq->lock, flags_cq);
3149                                 if (recv_mcq->mcq.comp &&
3150                                     mqp->ibqp.recv_cq->comp_handler) {
3151                                         if (!recv_mcq->mcq.reset_notify_added) {
3152                                                 recv_mcq->mcq.reset_notify_added = 1;
3153                                                 list_add_tail(&recv_mcq->mcq.reset_notify,
3154                                                               &cq_notify_list);
3155                                         }
3156                                 }
3157                                 spin_unlock_irqrestore(&recv_mcq->lock,
3158                                                        flags_cq);
3159                         }
3160                 }
3161                 spin_unlock_irqrestore(&mqp->rq.lock, flags_qp);
3162         }
3163
3164         list_for_each_entry(mcq, &cq_notify_list, reset_notify) {
3165                 mcq->comp(mcq);
3166         }
3167         spin_unlock_irqrestore(&ibdev->reset_flow_resource_lock, flags);
3168         pr_warn("mlx4_ib_handle_catas_error ended\n");
3169 }
3170
3171 static void handle_bonded_port_state_event(struct work_struct *work)
3172 {
3173         struct ib_event_work *ew =
3174                 container_of(work, struct ib_event_work, work);
3175         struct mlx4_ib_dev *ibdev = ew->ib_dev;
3176         enum ib_port_state bonded_port_state = IB_PORT_NOP;
3177         int i;
3178         struct ib_event ibev;
3179
3180         kfree(ew);
3181         spin_lock_bh(&ibdev->iboe.lock);
3182         for (i = 0; i < MLX4_MAX_PORTS; ++i) {
3183                 struct net_device *curr_netdev = ibdev->iboe.netdevs[i];
3184                 enum ib_port_state curr_port_state;
3185
3186                 if (!curr_netdev)
3187                         continue;
3188
3189                 curr_port_state =
3190                         (netif_running(curr_netdev) &&
3191                          netif_carrier_ok(curr_netdev)) ?
3192                         IB_PORT_ACTIVE : IB_PORT_DOWN;
3193
3194                 bonded_port_state = (bonded_port_state != IB_PORT_ACTIVE) ?
3195                         curr_port_state : IB_PORT_ACTIVE;
3196         }
3197         spin_unlock_bh(&ibdev->iboe.lock);
3198
3199         ibev.device = &ibdev->ib_dev;
3200         ibev.element.port_num = 1;
3201         ibev.event = (bonded_port_state == IB_PORT_ACTIVE) ?
3202                 IB_EVENT_PORT_ACTIVE : IB_EVENT_PORT_ERR;
3203
3204         ib_dispatch_event(&ibev);
3205 }
3206
3207 void mlx4_ib_sl2vl_update(struct mlx4_ib_dev *mdev, int port)
3208 {
3209         u64 sl2vl;
3210         int err;
3211
3212         err = mlx4_ib_query_sl2vl(&mdev->ib_dev, port, &sl2vl);
3213         if (err) {
3214                 pr_err("Unable to get current sl to vl mapping for port %d.  Using all zeroes (%d)\n",
3215                        port, err);
3216                 sl2vl = 0;
3217         }
3218         atomic64_set(&mdev->sl2vl[port - 1], sl2vl);
3219 }
3220
3221 static void ib_sl2vl_update_work(struct work_struct *work)
3222 {
3223         struct ib_event_work *ew = container_of(work, struct ib_event_work, work);
3224         struct mlx4_ib_dev *mdev = ew->ib_dev;
3225         int port = ew->port;
3226
3227         mlx4_ib_sl2vl_update(mdev, port);
3228
3229         kfree(ew);
3230 }
3231
3232 void mlx4_sched_ib_sl2vl_update_work(struct mlx4_ib_dev *ibdev,
3233                                      int port)
3234 {
3235         struct ib_event_work *ew;
3236
3237         ew = kmalloc(sizeof(*ew), GFP_ATOMIC);
3238         if (ew) {
3239                 INIT_WORK(&ew->work, ib_sl2vl_update_work);
3240                 ew->port = port;
3241                 ew->ib_dev = ibdev;
3242                 queue_work(wq, &ew->work);
3243         }
3244 }
3245
3246 static void mlx4_ib_event(struct mlx4_dev *dev, void *ibdev_ptr,
3247                           enum mlx4_dev_event event, unsigned long param)
3248 {
3249         struct ib_event ibev;
3250         struct mlx4_ib_dev *ibdev = to_mdev((struct ib_device *) ibdev_ptr);
3251         struct mlx4_eqe *eqe = NULL;
3252         struct ib_event_work *ew;
3253         int p = 0;
3254
3255         if (mlx4_is_bonded(dev) &&
3256             ((event == MLX4_DEV_EVENT_PORT_UP) ||
3257             (event == MLX4_DEV_EVENT_PORT_DOWN))) {
3258                 ew = kmalloc(sizeof(*ew), GFP_ATOMIC);
3259                 if (!ew)
3260                         return;
3261                 INIT_WORK(&ew->work, handle_bonded_port_state_event);
3262                 ew->ib_dev = ibdev;
3263                 queue_work(wq, &ew->work);
3264                 return;
3265         }
3266
3267         if (event == MLX4_DEV_EVENT_PORT_MGMT_CHANGE)
3268                 eqe = (struct mlx4_eqe *)param;
3269         else
3270                 p = (int) param;
3271
3272         switch (event) {
3273         case MLX4_DEV_EVENT_PORT_UP:
3274                 if (p > ibdev->num_ports)
3275                         return;
3276                 if (!mlx4_is_slave(dev) &&
3277                     rdma_port_get_link_layer(&ibdev->ib_dev, p) ==
3278                         IB_LINK_LAYER_INFINIBAND) {
3279                         if (mlx4_is_master(dev))
3280                                 mlx4_ib_invalidate_all_guid_record(ibdev, p);
3281                         if (ibdev->dev->flags & MLX4_FLAG_SECURE_HOST &&
3282                             !(ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_SL_TO_VL_CHANGE_EVENT))
3283                                 mlx4_sched_ib_sl2vl_update_work(ibdev, p);
3284                 }
3285                 ibev.event = IB_EVENT_PORT_ACTIVE;
3286                 break;
3287
3288         case MLX4_DEV_EVENT_PORT_DOWN:
3289                 if (p > ibdev->num_ports)
3290                         return;
3291                 ibev.event = IB_EVENT_PORT_ERR;
3292                 break;
3293
3294         case MLX4_DEV_EVENT_CATASTROPHIC_ERROR:
3295                 ibdev->ib_active = false;
3296                 ibev.event = IB_EVENT_DEVICE_FATAL;
3297                 mlx4_ib_handle_catas_error(ibdev);
3298                 break;
3299
3300         case MLX4_DEV_EVENT_PORT_MGMT_CHANGE:
3301                 ew = kmalloc(sizeof *ew, GFP_ATOMIC);
3302                 if (!ew)
3303                         break;
3304
3305                 INIT_WORK(&ew->work, handle_port_mgmt_change_event);
3306                 memcpy(&ew->ib_eqe, eqe, sizeof *eqe);
3307                 ew->ib_dev = ibdev;
3308                 /* need to queue only for port owner, which uses GEN_EQE */
3309                 if (mlx4_is_master(dev))
3310                         queue_work(wq, &ew->work);
3311                 else
3312                         handle_port_mgmt_change_event(&ew->work);
3313                 return;
3314
3315         case MLX4_DEV_EVENT_SLAVE_INIT:
3316                 /* here, p is the slave id */
3317                 do_slave_init(ibdev, p, 1);
3318                 if (mlx4_is_master(dev)) {
3319                         int i;
3320
3321                         for (i = 1; i <= ibdev->num_ports; i++) {
3322                                 if (rdma_port_get_link_layer(&ibdev->ib_dev, i)
3323                                         == IB_LINK_LAYER_INFINIBAND)
3324                                         mlx4_ib_slave_alias_guid_event(ibdev,
3325                                                                        p, i,
3326                                                                        1);
3327                         }
3328                 }
3329                 return;
3330
3331         case MLX4_DEV_EVENT_SLAVE_SHUTDOWN:
3332                 if (mlx4_is_master(dev)) {
3333                         int i;
3334
3335                         for (i = 1; i <= ibdev->num_ports; i++) {
3336                                 if (rdma_port_get_link_layer(&ibdev->ib_dev, i)
3337                                         == IB_LINK_LAYER_INFINIBAND)
3338                                         mlx4_ib_slave_alias_guid_event(ibdev,
3339                                                                        p, i,
3340                                                                        0);
3341                         }
3342                 }
3343                 /* here, p is the slave id */
3344                 do_slave_init(ibdev, p, 0);
3345                 return;
3346
3347         default:
3348                 return;
3349         }
3350
3351         ibev.device           = ibdev_ptr;
3352         ibev.element.port_num = mlx4_is_bonded(ibdev->dev) ? 1 : (u8)p;
3353
3354         ib_dispatch_event(&ibev);
3355 }
3356
3357 static struct mlx4_interface mlx4_ib_interface = {
3358         .add            = mlx4_ib_add,
3359         .remove         = mlx4_ib_remove,
3360         .event          = mlx4_ib_event,
3361         .protocol       = MLX4_PROT_IB_IPV6,
3362         .flags          = MLX4_INTFF_BONDING
3363 };
3364
3365 static int __init mlx4_ib_init(void)
3366 {
3367         int err;
3368
3369         wq = alloc_ordered_workqueue("mlx4_ib", WQ_MEM_RECLAIM);
3370         if (!wq)
3371                 return -ENOMEM;
3372
3373         err = mlx4_ib_mcg_init();
3374         if (err)
3375                 goto clean_wq;
3376
3377         err = mlx4_register_interface(&mlx4_ib_interface);
3378         if (err)
3379                 goto clean_mcg;
3380
3381         return 0;
3382
3383 clean_mcg:
3384         mlx4_ib_mcg_destroy();
3385
3386 clean_wq:
3387         destroy_workqueue(wq);
3388         return err;
3389 }
3390
3391 static void __exit mlx4_ib_cleanup(void)
3392 {
3393         mlx4_unregister_interface(&mlx4_ib_interface);
3394         mlx4_ib_mcg_destroy();
3395         destroy_workqueue(wq);
3396 }
3397
3398 module_init(mlx4_ib_init);
3399 module_exit(mlx4_ib_cleanup);