drivers/infiniband/hw/mlx4/main.c

   1 /*
   2  * Copyright (c) 2006, 2007 Cisco Systems, Inc. All rights reserved.
   3  * Copyright (c) 2007, 2008 Mellanox Technologies. All rights reserved.
   4  *
   5  * This software is available to you under a choice of one of two
   6  * licenses.  You may choose to be licensed under the terms of the GNU
   7  * General Public License (GPL) Version 2, available from the file
   8  * COPYING in the main directory of this source tree, or the
   9  * OpenIB.org BSD license below:
  10  *
  11  *     Redistribution and use in source and binary forms, with or
  12  *     without modification, are permitted provided that the following
  13  *     conditions are met:
  14  *
  15  *      - Redistributions of source code must retain the above
  16  *        copyright notice, this list of conditions and the following
  17  *        disclaimer.
  18  *
  19  *      - Redistributions in binary form must reproduce the above
  20  *        copyright notice, this list of conditions and the following
  21  *        disclaimer in the documentation and/or other materials
  22  *        provided with the distribution.
  23  *
  24  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
  25  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  26  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
  27  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
  28  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
  29  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
  30  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  31  * SOFTWARE.
  32  */
  33
  34 #include <linux/module.h>
  35 #include <linux/init.h>
  36 #include <linux/slab.h>
  37 #include <linux/errno.h>
  38 #include <linux/netdevice.h>
  39 #include <linux/inetdevice.h>
  40 #include <linux/rtnetlink.h>
  41 #include <linux/if_vlan.h>
  42 #include <linux/sched/mm.h>
  43 #include <linux/sched/task.h>
  44
  45 #include <net/ipv6.h>
  46 #include <net/addrconf.h>
  47 #include <net/devlink.h>
  48
  49 #include <rdma/ib_smi.h>
  50 #include <rdma/ib_user_verbs.h>
  51 #include <rdma/ib_addr.h>
  52 #include <rdma/ib_cache.h>
  53
  54 #include <net/bonding.h>
  55
  56 #include <linux/mlx4/driver.h>
  57 #include <linux/mlx4/cmd.h>
  58 #include <linux/mlx4/qp.h>
  59
  60 #include "mlx4_ib.h"
  61 #include <rdma/mlx4-abi.h>
  62
  63 #define DRV_NAME        MLX4_IB_DRV_NAME
  64 #define DRV_VERSION     "4.0-0"
  65
  66 #define MLX4_IB_FLOW_MAX_PRIO 0xFFF
  67 #define MLX4_IB_FLOW_QPN_MASK 0xFFFFFF
  68 #define MLX4_IB_CARD_REV_A0   0xA0
  69
  70 MODULE_AUTHOR("Roland Dreier");
  71 MODULE_DESCRIPTION("Mellanox ConnectX HCA InfiniBand driver");
  72 MODULE_LICENSE("Dual BSD/GPL");
  73
  74 int mlx4_ib_sm_guid_assign = 0;
  75 module_param_named(sm_guid_assign, mlx4_ib_sm_guid_assign, int, 0444);
  76 MODULE_PARM_DESC(sm_guid_assign, "Enable SM alias_GUID assignment if sm_guid_assign > 0 (Default: 0)");
  77
  78 static const char mlx4_ib_version[] =
  79         DRV_NAME ": Mellanox ConnectX InfiniBand driver v"
  80         DRV_VERSION "\n";
  81
  82 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init);
  83 static enum rdma_link_layer mlx4_ib_port_link_layer(struct ib_device *device,
  84                                                     u8 port_num);
  85
  86 static struct workqueue_struct *wq;
  87
  88 static void init_query_mad(struct ib_smp *mad)
  89 {
  90         mad->base_version  = 1;
  91         mad->mgmt_class    = IB_MGMT_CLASS_SUBN_LID_ROUTED;
  92         mad->class_version = 1;
  93         mad->method        = IB_MGMT_METHOD_GET;
  94 }
  95
  96 static int check_flow_steering_support(struct mlx4_dev *dev)
  97 {
  98         int eth_num_ports = 0;
  99         int ib_num_ports = 0;
 100
 101         int dmfs = dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED;
 102
 103         if (dmfs) {
 104                 int i;
 105                 mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_ETH)
 106                         eth_num_ports++;
 107                 mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
 108                         ib_num_ports++;
 109                 dmfs &= (!ib_num_ports ||
 110                          (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DMFS_IPOIB)) &&
 111                         (!eth_num_ports ||
 112                          (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_FS_EN));
 113                 if (ib_num_ports && mlx4_is_mfunc(dev)) {
 114                         pr_warn("Device managed flow steering is unavailable for IB port in multifunction env.\n");
 115                         dmfs = 0;
 116                 }
 117         }
 118         return dmfs;
 119 }
 120
 121 static int num_ib_ports(struct mlx4_dev *dev)
 122 {
 123         int ib_ports = 0;
 124         int i;
 125
 126         mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
 127                 ib_ports++;
 128
 129         return ib_ports;
 130 }
 131
 132 static struct net_device *mlx4_ib_get_netdev(struct ib_device *device, u8 port_num)
 133 {
 134         struct mlx4_ib_dev *ibdev = to_mdev(device);
 135         struct net_device *dev;
 136
 137         rcu_read_lock();
 138         dev = mlx4_get_protocol_dev(ibdev->dev, MLX4_PROT_ETH, port_num);
 139
 140         if (dev) {
 141                 if (mlx4_is_bonded(ibdev->dev)) {
 142                         struct net_device *upper = NULL;
 143
 144                         upper = netdev_master_upper_dev_get_rcu(dev);
 145                         if (upper) {
 146                                 struct net_device *active;
 147
 148                                 active = bond_option_active_slave_get_rcu(netdev_priv(upper));
 149                                 if (active)
 150                                         dev = active;
 151                         }
 152                 }
 153         }
 154         if (dev)
 155                 dev_hold(dev);
 156
 157         rcu_read_unlock();
 158         return dev;
 159 }
 160
 161 static int mlx4_ib_update_gids_v1(struct gid_entry *gids,
 162                                   struct mlx4_ib_dev *ibdev,
 163                                   u8 port_num)
 164 {
 165         struct mlx4_cmd_mailbox *mailbox;
 166         int err;
 167         struct mlx4_dev *dev = ibdev->dev;
 168         int i;
 169         union ib_gid *gid_tbl;
 170
 171         mailbox = mlx4_alloc_cmd_mailbox(dev);
 172         if (IS_ERR(mailbox))
 173                 return -ENOMEM;
 174
 175         gid_tbl = mailbox->buf;
 176
 177         for (i = 0; i < MLX4_MAX_PORT_GIDS; ++i)
 178                 memcpy(&gid_tbl[i], &gids[i].gid, sizeof(union ib_gid));
 179
 180         err = mlx4_cmd(dev, mailbox->dma,
 181                        MLX4_SET_PORT_GID_TABLE << 8 | port_num,
 182                        1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
 183                        MLX4_CMD_WRAPPED);
 184         if (mlx4_is_bonded(dev))
 185                 err += mlx4_cmd(dev, mailbox->dma,
 186                                 MLX4_SET_PORT_GID_TABLE << 8 | 2,
 187                                 1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
 188                                 MLX4_CMD_WRAPPED);
 189
 190         mlx4_free_cmd_mailbox(dev, mailbox);
 191         return err;
 192 }
 193
 194 static int mlx4_ib_update_gids_v1_v2(struct gid_entry *gids,
 195                                      struct mlx4_ib_dev *ibdev,
 196                                      u8 port_num)
 197 {
 198         struct mlx4_cmd_mailbox *mailbox;
 199         int err;
 200         struct mlx4_dev *dev = ibdev->dev;
 201         int i;
 202         struct {
 203                 union ib_gid    gid;
 204                 __be32          rsrvd1[2];
 205                 __be16          rsrvd2;
 206                 u8              type;
 207                 u8              version;
 208                 __be32          rsrvd3;
 209         } *gid_tbl;
 210
 211         mailbox = mlx4_alloc_cmd_mailbox(dev);
 212         if (IS_ERR(mailbox))
 213                 return -ENOMEM;
 214
 215         gid_tbl = mailbox->buf;
 216         for (i = 0; i < MLX4_MAX_PORT_GIDS; ++i) {
 217                 memcpy(&gid_tbl[i].gid, &gids[i].gid, sizeof(union ib_gid));
 218                 if (gids[i].gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP) {
 219                         gid_tbl[i].version = 2;
 220                         if (!ipv6_addr_v4mapped((struct in6_addr *)&gids[i].gid))
 221                                 gid_tbl[i].type = 1;
 222                 }
 223         }
 224
 225         err = mlx4_cmd(dev, mailbox->dma,
 226                        MLX4_SET_PORT_ROCE_ADDR << 8 | port_num,
 227                        1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
 228                        MLX4_CMD_WRAPPED);
 229         if (mlx4_is_bonded(dev))
 230                 err += mlx4_cmd(dev, mailbox->dma,
 231                                 MLX4_SET_PORT_ROCE_ADDR << 8 | 2,
 232                                 1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
 233                                 MLX4_CMD_WRAPPED);
 234
 235         mlx4_free_cmd_mailbox(dev, mailbox);
 236         return err;
 237 }
 238
 239 static int mlx4_ib_update_gids(struct gid_entry *gids,
 240                                struct mlx4_ib_dev *ibdev,
 241                                u8 port_num)
 242 {
 243         if (ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ROCE_V1_V2)
 244                 return mlx4_ib_update_gids_v1_v2(gids, ibdev, port_num);
 245
 246         return mlx4_ib_update_gids_v1(gids, ibdev, port_num);
 247 }
 248
 249 static int mlx4_ib_add_gid(struct ib_device *device,
 250                            u8 port_num,
 251                            unsigned int index,
 252                            const union ib_gid *gid,
 253                            const struct ib_gid_attr *attr,
 254                            void **context)
 255 {
 256         struct mlx4_ib_dev *ibdev = to_mdev(device);
 257         struct mlx4_ib_iboe *iboe = &ibdev->iboe;
 258         struct mlx4_port_gid_table   *port_gid_table;
 259         int free = -1, found = -1;
 260         int ret = 0;
 261         int hw_update = 0;
 262         int i;
 263         struct gid_entry *gids = NULL;
 264
 265         if (!rdma_cap_roce_gid_table(device, port_num))
 266                 return -EINVAL;
 267
 268         if (port_num > MLX4_MAX_PORTS)
 269                 return -EINVAL;
 270
 271         if (!context)
 272                 return -EINVAL;
 273
 274         port_gid_table = &iboe->gids[port_num - 1];
 275         spin_lock_bh(&iboe->lock);
 276         for (i = 0; i < MLX4_MAX_PORT_GIDS; ++i) {
 277                 if (!memcmp(&port_gid_table->gids[i].gid, gid, sizeof(*gid)) &&
 278                     (port_gid_table->gids[i].gid_type == attr->gid_type))  {
 279                         found = i;
 280                         break;
 281                 }
 282                 if (free < 0 && !memcmp(&port_gid_table->gids[i].gid, &zgid, sizeof(*gid)))
 283                         free = i; /* HW has space */
 284         }
 285
 286         if (found < 0) {
 287                 if (free < 0) {
 288                         ret = -ENOSPC;
 289                 } else {
 290                         port_gid_table->gids[free].ctx = kmalloc(sizeof(*port_gid_table->gids[free].ctx), GFP_ATOMIC);
 291                         if (!port_gid_table->gids[free].ctx) {
 292                                 ret = -ENOMEM;
 293                         } else {
 294                                 *context = port_gid_table->gids[free].ctx;
 295                                 memcpy(&port_gid_table->gids[free].gid, gid, sizeof(*gid));
 296                                 port_gid_table->gids[free].gid_type = attr->gid_type;
 297                                 port_gid_table->gids[free].ctx->real_index = free;
 298                                 port_gid_table->gids[free].ctx->refcount = 1;
 299                                 hw_update = 1;
 300                         }
 301                 }
 302         } else {
 303                 struct gid_cache_context *ctx = port_gid_table->gids[found].ctx;
 304                 *context = ctx;
 305                 ctx->refcount++;
 306         }
 307         if (!ret && hw_update) {
 308                 gids = kmalloc(sizeof(*gids) * MLX4_MAX_PORT_GIDS, GFP_ATOMIC);
 309                 if (!gids) {
 310                         ret = -ENOMEM;
 311                 } else {
 312                         for (i = 0; i < MLX4_MAX_PORT_GIDS; i++) {
 313                                 memcpy(&gids[i].gid, &port_gid_table->gids[i].gid, sizeof(union ib_gid));
 314                                 gids[i].gid_type = port_gid_table->gids[i].gid_type;
 315                         }
 316                 }
 317         }
 318         spin_unlock_bh(&iboe->lock);
 319
 320         if (!ret && hw_update) {
 321                 ret = mlx4_ib_update_gids(gids, ibdev, port_num);
 322                 kfree(gids);
 323         }
 324
 325         return ret;
 326 }
 327
 328 static int mlx4_ib_del_gid(struct ib_device *device,
 329                            u8 port_num,
 330                            unsigned int index,
 331                            void **context)
 332 {
 333         struct gid_cache_context *ctx = *context;
 334         struct mlx4_ib_dev *ibdev = to_mdev(device);
 335         struct mlx4_ib_iboe *iboe = &ibdev->iboe;
 336         struct mlx4_port_gid_table   *port_gid_table;
 337         int ret = 0;
 338         int hw_update = 0;
 339         struct gid_entry *gids = NULL;
 340
 341         if (!rdma_cap_roce_gid_table(device, port_num))
 342                 return -EINVAL;
 343
 344         if (port_num > MLX4_MAX_PORTS)
 345                 return -EINVAL;
 346
 347         port_gid_table = &iboe->gids[port_num - 1];
 348         spin_lock_bh(&iboe->lock);
 349         if (ctx) {
 350                 ctx->refcount--;
 351                 if (!ctx->refcount) {
 352                         unsigned int real_index = ctx->real_index;
 353
 354                         memcpy(&port_gid_table->gids[real_index].gid, &zgid, sizeof(zgid));
 355                         kfree(port_gid_table->gids[real_index].ctx);
 356                         port_gid_table->gids[real_index].ctx = NULL;
 357                         hw_update = 1;
 358                 }
 359         }
 360         if (!ret && hw_update) {
 361                 int i;
 362
 363                 gids = kmalloc(sizeof(*gids) * MLX4_MAX_PORT_GIDS, GFP_ATOMIC);
 364                 if (!gids) {
 365                         ret = -ENOMEM;
 366                 } else {
 367                         for (i = 0; i < MLX4_MAX_PORT_GIDS; i++) {
 368                                 memcpy(&gids[i].gid,
 369                                        &port_gid_table->gids[i].gid,
 370                                        sizeof(union ib_gid));
 371                                 gids[i].gid_type =
 372                                     port_gid_table->gids[i].gid_type;
 373                         }
 374                 }
 375         }
 376         spin_unlock_bh(&iboe->lock);
 377
 378         if (!ret && hw_update) {
 379                 ret = mlx4_ib_update_gids(gids, ibdev, port_num);
 380                 kfree(gids);
 381         }
 382         return ret;
 383 }
 384
 385 int mlx4_ib_gid_index_to_real_index(struct mlx4_ib_dev *ibdev,
 386                                     u8 port_num, int index)
 387 {
 388         struct mlx4_ib_iboe *iboe = &ibdev->iboe;
 389         struct gid_cache_context *ctx = NULL;
 390         union ib_gid gid;
 391         struct mlx4_port_gid_table   *port_gid_table;
 392         int real_index = -EINVAL;
 393         int i;
 394         int ret;
 395         unsigned long flags;
 396         struct ib_gid_attr attr;
 397
 398         if (port_num > MLX4_MAX_PORTS)
 399                 return -EINVAL;
 400
 401         if (mlx4_is_bonded(ibdev->dev))
 402                 port_num = 1;
 403
 404         if (!rdma_cap_roce_gid_table(&ibdev->ib_dev, port_num))
 405                 return index;
 406
 407         ret = ib_get_cached_gid(&ibdev->ib_dev, port_num, index, &gid, &attr);
 408         if (ret)
 409                 return ret;
 410
 411         if (attr.ndev)
 412                 dev_put(attr.ndev);
 413
 414         if (!memcmp(&gid, &zgid, sizeof(gid)))
 415                 return -EINVAL;
 416
 417         spin_lock_irqsave(&iboe->lock, flags);
 418         port_gid_table = &iboe->gids[port_num - 1];
 419
 420         for (i = 0; i < MLX4_MAX_PORT_GIDS; ++i)
 421                 if (!memcmp(&port_gid_table->gids[i].gid, &gid, sizeof(gid)) &&
 422                     attr.gid_type == port_gid_table->gids[i].gid_type) {
 423                         ctx = port_gid_table->gids[i].ctx;
 424                         break;
 425                 }
 426         if (ctx)
 427                 real_index = ctx->real_index;
 428         spin_unlock_irqrestore(&iboe->lock, flags);
 429         return real_index;
 430 }
 431
 432 static int mlx4_ib_query_device(struct ib_device *ibdev,
 433                                 struct ib_device_attr *props,
 434                                 struct ib_udata *uhw)
 435 {
 436         struct mlx4_ib_dev *dev = to_mdev(ibdev);
 437         struct ib_smp *in_mad  = NULL;
 438         struct ib_smp *out_mad = NULL;
 439         int err;
 440         int have_ib_ports;
 441         struct mlx4_uverbs_ex_query_device cmd;
 442         struct mlx4_uverbs_ex_query_device_resp resp = {.comp_mask = 0};
 443         struct mlx4_clock_params clock_params;
 444
 445         if (uhw->inlen) {
 446                 if (uhw->inlen < sizeof(cmd))
 447                         return -EINVAL;
 448
 449                 err = ib_copy_from_udata(&cmd, uhw, sizeof(cmd));
 450                 if (err)
 451                         return err;
 452
 453                 if (cmd.comp_mask)
 454                         return -EINVAL;
 455
 456                 if (cmd.reserved)
 457                         return -EINVAL;
 458         }
 459
 460         resp.response_length = offsetof(typeof(resp), response_length) +
 461                 sizeof(resp.response_length);
 462         in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
 463         out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
 464         err = -ENOMEM;
 465         if (!in_mad || !out_mad)
 466                 goto out;
 467
 468         init_query_mad(in_mad);
 469         in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
 470
 471         err = mlx4_MAD_IFC(to_mdev(ibdev), MLX4_MAD_IFC_IGNORE_KEYS,
 472                            1, NULL, NULL, in_mad, out_mad);
 473         if (err)
 474                 goto out;
 475
 476         memset(props, 0, sizeof *props);
 477
 478         have_ib_ports = num_ib_ports(dev->dev);
 479
 480         props->fw_ver = dev->dev->caps.fw_ver;
 481         props->device_cap_flags    = IB_DEVICE_CHANGE_PHY_PORT |
 482                 IB_DEVICE_PORT_ACTIVE_EVENT             |
 483                 IB_DEVICE_SYS_IMAGE_GUID                |
 484                 IB_DEVICE_RC_RNR_NAK_GEN                |
 485                 IB_DEVICE_BLOCK_MULTICAST_LOOPBACK;
 486         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_PKEY_CNTR)
 487                 props->device_cap_flags |= IB_DEVICE_BAD_PKEY_CNTR;
 488         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_QKEY_CNTR)
 489                 props->device_cap_flags |= IB_DEVICE_BAD_QKEY_CNTR;
 490         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_APM && have_ib_ports)
 491                 props->device_cap_flags |= IB_DEVICE_AUTO_PATH_MIG;
 492         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_UD_AV_PORT)
 493                 props->device_cap_flags |= IB_DEVICE_UD_AV_PORT_ENFORCE;
 494         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_IPOIB_CSUM)
 495                 props->device_cap_flags |= IB_DEVICE_UD_IP_CSUM;
 496         if (dev->dev->caps.max_gso_sz &&
 497             (dev->dev->rev_id != MLX4_IB_CARD_REV_A0) &&
 498             (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BLH))
 499                 props->device_cap_flags |= IB_DEVICE_UD_TSO;
 500         if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_RESERVED_LKEY)
 501                 props->device_cap_flags |= IB_DEVICE_LOCAL_DMA_LKEY;
 502         if ((dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_LOCAL_INV) &&
 503             (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_REMOTE_INV) &&
 504             (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_FAST_REG_WR))
 505                 props->device_cap_flags |= IB_DEVICE_MEM_MGT_EXTENSIONS;
 506         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC)
 507                 props->device_cap_flags |= IB_DEVICE_XRC;
 508         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW)
 509                 props->device_cap_flags |= IB_DEVICE_MEM_WINDOW;
 510         if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
 511                 if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_WIN_TYPE_2B)
 512                         props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2B;
 513                 else
 514                         props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2A;
 515         }
 516         if (dev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED)
 517                 props->device_cap_flags |= IB_DEVICE_MANAGED_FLOW_STEERING;
 518
 519         props->device_cap_flags |= IB_DEVICE_RAW_IP_CSUM;
 520
 521         props->vendor_id           = be32_to_cpup((__be32 *) (out_mad->data + 36)) &
 522                 0xffffff;
 523         props->vendor_part_id      = dev->dev->persist->pdev->device;
 524         props->hw_ver              = be32_to_cpup((__be32 *) (out_mad->data + 32));
 525         memcpy(&props->sys_image_guid, out_mad->data +  4, 8);
 526
 527         props->max_mr_size         = ~0ull;
 528         props->page_size_cap       = dev->dev->caps.page_size_cap;
 529         props->max_qp              = dev->dev->quotas.qp;
 530         props->max_qp_wr           = dev->dev->caps.max_wqes - MLX4_IB_SQ_MAX_SPARE;
 531         props->max_sge             = min(dev->dev->caps.max_sq_sg,
 532                                          dev->dev->caps.max_rq_sg);
 533         props->max_sge_rd          = MLX4_MAX_SGE_RD;
 534         props->max_cq              = dev->dev->quotas.cq;
 535         props->max_cqe             = dev->dev->caps.max_cqes;
 536         props->max_mr              = dev->dev->quotas.mpt;
 537         props->max_pd              = dev->dev->caps.num_pds - dev->dev->caps.reserved_pds;
 538         props->max_qp_rd_atom      = dev->dev->caps.max_qp_dest_rdma;
 539         props->max_qp_init_rd_atom = dev->dev->caps.max_qp_init_rdma;
 540         props->max_res_rd_atom     = props->max_qp_rd_atom * props->max_qp;
 541         props->max_srq             = dev->dev->quotas.srq;
 542         props->max_srq_wr          = dev->dev->caps.max_srq_wqes - 1;
 543         props->max_srq_sge         = dev->dev->caps.max_srq_sge;
 544         props->max_fast_reg_page_list_len = MLX4_MAX_FAST_REG_PAGES;
 545         props->local_ca_ack_delay  = dev->dev->caps.local_ca_ack_delay;
 546         props->atomic_cap          = dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_ATOMIC ?
 547                 IB_ATOMIC_HCA : IB_ATOMIC_NONE;
 548         props->masked_atomic_cap   = props->atomic_cap;
 549         props->max_pkeys           = dev->dev->caps.pkey_table_len[1];
 550         props->max_mcast_grp       = dev->dev->caps.num_mgms + dev->dev->caps.num_amgms;
 551         props->max_mcast_qp_attach = dev->dev->caps.num_qp_per_mgm;
 552         props->max_total_mcast_qp_attach = props->max_mcast_qp_attach *
 553                                            props->max_mcast_grp;
 554         props->max_map_per_fmr = dev->dev->caps.max_fmr_maps;
 555         props->hca_core_clock = dev->dev->caps.hca_core_clock * 1000UL;
 556         props->timestamp_mask = 0xFFFFFFFFFFFFULL;
 557         props->max_ah = INT_MAX;
 558
 559         if ((dev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_RSS) &&
 560             (mlx4_ib_port_link_layer(ibdev, 1) == IB_LINK_LAYER_ETHERNET ||
 561              mlx4_ib_port_link_layer(ibdev, 2) == IB_LINK_LAYER_ETHERNET)) {
 562                 props->rss_caps.max_rwq_indirection_tables = props->max_qp;
 563                 props->rss_caps.max_rwq_indirection_table_size =
 564                         dev->dev->caps.max_rss_tbl_sz;
 565                 props->rss_caps.supported_qpts = 1 << IB_QPT_RAW_PACKET;
 566                 props->max_wq_type_rq = props->max_qp;
 567         }
 568
 569         if (!mlx4_is_slave(dev->dev))
 570                 err = mlx4_get_internal_clock_params(dev->dev, &clock_params);
 571
 572         if (uhw->outlen >= resp.response_length + sizeof(resp.hca_core_clock_offset)) {
 573                 resp.response_length += sizeof(resp.hca_core_clock_offset);
 574                 if (!err && !mlx4_is_slave(dev->dev)) {
 575                         resp.comp_mask |= QUERY_DEVICE_RESP_MASK_TIMESTAMP;
 576                         resp.hca_core_clock_offset = clock_params.offset % PAGE_SIZE;
 577                 }
 578         }
 579
 580         if (uhw->outlen >= resp.response_length +
 581             sizeof(resp.max_inl_recv_sz)) {
 582                 resp.response_length += sizeof(resp.max_inl_recv_sz);
 583                 resp.max_inl_recv_sz  = dev->dev->caps.max_rq_sg *
 584                         sizeof(struct mlx4_wqe_data_seg);
 585         }
 586
 587         if (uhw->outlen) {
 588                 err = ib_copy_to_udata(uhw, &resp, resp.response_length);
 589                 if (err)
 590                         goto out;
 591         }
 592 out:
 593         kfree(in_mad);
 594         kfree(out_mad);
 595
 596         return err;
 597 }
 598
 599 static enum rdma_link_layer
 600 mlx4_ib_port_link_layer(struct ib_device *device, u8 port_num)
 601 {
 602         struct mlx4_dev *dev = to_mdev(device)->dev;
 603
 604         return dev->caps.port_mask[port_num] == MLX4_PORT_TYPE_IB ?
 605                 IB_LINK_LAYER_INFINIBAND : IB_LINK_LAYER_ETHERNET;
 606 }
 607
 608 static int ib_link_query_port(struct ib_device *ibdev, u8 port,
 609                               struct ib_port_attr *props, int netw_view)
 610 {
 611         struct ib_smp *in_mad  = NULL;
 612         struct ib_smp *out_mad = NULL;
 613         int ext_active_speed;
 614         int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
 615         int err = -ENOMEM;
 616
 617         in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
 618         out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
 619         if (!in_mad || !out_mad)
 620                 goto out;
 621
 622         init_query_mad(in_mad);
 623         in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
 624         in_mad->attr_mod = cpu_to_be32(port);
 625
 626         if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
 627                 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
 628
 629         err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
 630                                 in_mad, out_mad);
 631         if (err)
 632                 goto out;
 633
 634
 635         props->lid              = be16_to_cpup((__be16 *) (out_mad->data + 16));
 636         props->lmc              = out_mad->data[34] & 0x7;
 637         props->sm_lid           = be16_to_cpup((__be16 *) (out_mad->data + 18));
 638         props->sm_sl            = out_mad->data[36] & 0xf;
 639         props->state            = out_mad->data[32] & 0xf;
 640         props->phys_state       = out_mad->data[33] >> 4;
 641         props->port_cap_flags   = be32_to_cpup((__be32 *) (out_mad->data + 20));
 642         if (netw_view)
 643                 props->gid_tbl_len = out_mad->data[50];
 644         else
 645                 props->gid_tbl_len = to_mdev(ibdev)->dev->caps.gid_table_len[port];
 646         props->max_msg_sz       = to_mdev(ibdev)->dev->caps.max_msg_sz;
 647         props->pkey_tbl_len     = to_mdev(ibdev)->dev->caps.pkey_table_len[port];
 648         props->bad_pkey_cntr    = be16_to_cpup((__be16 *) (out_mad->data + 46));
 649         props->qkey_viol_cntr   = be16_to_cpup((__be16 *) (out_mad->data + 48));
 650         props->active_width     = out_mad->data[31] & 0xf;
 651         props->active_speed     = out_mad->data[35] >> 4;
 652         props->max_mtu          = out_mad->data[41] & 0xf;
 653         props->active_mtu       = out_mad->data[36] >> 4;
 654         props->subnet_timeout   = out_mad->data[51] & 0x1f;
 655         props->max_vl_num       = out_mad->data[37] >> 4;
 656         props->init_type_reply  = out_mad->data[41] >> 4;
 657
 658         /* Check if extended speeds (EDR/FDR/...) are supported */
 659         if (props->port_cap_flags & IB_PORT_EXTENDED_SPEEDS_SUP) {
 660                 ext_active_speed = out_mad->data[62] >> 4;
 661
 662                 switch (ext_active_speed) {
 663                 case 1:
 664                         props->active_speed = IB_SPEED_FDR;
 665                         break;
 666                 case 2:
 667                         props->active_speed = IB_SPEED_EDR;
 668                         break;
 669                 }
 670         }
 671
 672         /* If reported active speed is QDR, check if is FDR-10 */
 673         if (props->active_speed == IB_SPEED_QDR) {
 674                 init_query_mad(in_mad);
 675                 in_mad->attr_id = MLX4_ATTR_EXTENDED_PORT_INFO;
 676                 in_mad->attr_mod = cpu_to_be32(port);
 677
 678                 err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port,
 679                                    NULL, NULL, in_mad, out_mad);
 680                 if (err)
 681                         goto out;
 682
 683                 /* Checking LinkSpeedActive for FDR-10 */
 684                 if (out_mad->data[15] & 0x1)
 685                         props->active_speed = IB_SPEED_FDR10;
 686         }
 687
 688         /* Avoid wrong speed value returned by FW if the IB link is down. */
 689         if (props->state == IB_PORT_DOWN)
 690                  props->active_speed = IB_SPEED_SDR;
 691
 692 out:
 693         kfree(in_mad);
 694         kfree(out_mad);
 695         return err;
 696 }
 697
 698 static u8 state_to_phys_state(enum ib_port_state state)
 699 {
 700         return state == IB_PORT_ACTIVE ? 5 : 3;
 701 }
 702
 703 static int eth_link_query_port(struct ib_device *ibdev, u8 port,
 704                                struct ib_port_attr *props)
 705 {
 706
 707         struct mlx4_ib_dev *mdev = to_mdev(ibdev);
 708         struct mlx4_ib_iboe *iboe = &mdev->iboe;
 709         struct net_device *ndev;
 710         enum ib_mtu tmp;
 711         struct mlx4_cmd_mailbox *mailbox;
 712         int err = 0;
 713         int is_bonded = mlx4_is_bonded(mdev->dev);
 714
 715         mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
 716         if (IS_ERR(mailbox))
 717                 return PTR_ERR(mailbox);
 718
 719         err = mlx4_cmd_box(mdev->dev, 0, mailbox->dma, port, 0,
 720                            MLX4_CMD_QUERY_PORT, MLX4_CMD_TIME_CLASS_B,
 721                            MLX4_CMD_WRAPPED);
 722         if (err)
 723                 goto out;
 724
 725         props->active_width     =  (((u8 *)mailbox->buf)[5] == 0x40) ||
 726                                    (((u8 *)mailbox->buf)[5] == 0x20 /*56Gb*/) ?
 727                                            IB_WIDTH_4X : IB_WIDTH_1X;
 728         props->active_speed     =  (((u8 *)mailbox->buf)[5] == 0x20 /*56Gb*/) ?
 729                                            IB_SPEED_FDR : IB_SPEED_QDR;
 730         props->port_cap_flags   = IB_PORT_CM_SUP | IB_PORT_IP_BASED_GIDS;
 731         props->gid_tbl_len      = mdev->dev->caps.gid_table_len[port];
 732         props->max_msg_sz       = mdev->dev->caps.max_msg_sz;
 733         props->pkey_tbl_len     = 1;
 734         props->max_mtu          = IB_MTU_4096;
 735         props->max_vl_num       = 2;
 736         props->state            = IB_PORT_DOWN;
 737         props->phys_state       = state_to_phys_state(props->state);
 738         props->active_mtu       = IB_MTU_256;
 739         spin_lock_bh(&iboe->lock);
 740         ndev = iboe->netdevs[port - 1];
 741         if (ndev && is_bonded) {
 742                 rcu_read_lock(); /* required to get upper dev */
 743                 ndev = netdev_master_upper_dev_get_rcu(ndev);
 744                 rcu_read_unlock();
 745         }
 746         if (!ndev)
 747                 goto out_unlock;
 748
 749         tmp = iboe_get_mtu(ndev->mtu);
 750         props->active_mtu = tmp ? min(props->max_mtu, tmp) : IB_MTU_256;
 751
 752         props->state            = (netif_running(ndev) && netif_carrier_ok(ndev)) ?
 753                                         IB_PORT_ACTIVE : IB_PORT_DOWN;
 754         props->phys_state       = state_to_phys_state(props->state);
 755 out_unlock:
 756         spin_unlock_bh(&iboe->lock);
 757 out:
 758         mlx4_free_cmd_mailbox(mdev->dev, mailbox);
 759         return err;
 760 }
 761
 762 int __mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
 763                          struct ib_port_attr *props, int netw_view)
 764 {
 765         int err;
 766
 767         /* props being zeroed by the caller, avoid zeroing it here */
 768
 769         err = mlx4_ib_port_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND ?
 770                 ib_link_query_port(ibdev, port, props, netw_view) :
 771                                 eth_link_query_port(ibdev, port, props);
 772
 773         return err;
 774 }
 775
 776 static int mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
 777                               struct ib_port_attr *props)
 778 {
 779         /* returns host view */
 780         return __mlx4_ib_query_port(ibdev, port, props, 0);
 781 }
 782
 783 int __mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
 784                         union ib_gid *gid, int netw_view)
 785 {
 786         struct ib_smp *in_mad  = NULL;
 787         struct ib_smp *out_mad = NULL;
 788         int err = -ENOMEM;
 789         struct mlx4_ib_dev *dev = to_mdev(ibdev);
 790         int clear = 0;
 791         int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
 792
 793         in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
 794         out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
 795         if (!in_mad || !out_mad)
 796                 goto out;
 797
 798         init_query_mad(in_mad);
 799         in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
 800         in_mad->attr_mod = cpu_to_be32(port);
 801
 802         if (mlx4_is_mfunc(dev->dev) && netw_view)
 803                 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
 804
 805         err = mlx4_MAD_IFC(dev, mad_ifc_flags, port, NULL, NULL, in_mad, out_mad);
 806         if (err)
 807                 goto out;
 808
 809         memcpy(gid->raw, out_mad->data + 8, 8);
 810
 811         if (mlx4_is_mfunc(dev->dev) && !netw_view) {
 812                 if (index) {
 813                         /* For any index > 0, return the null guid */
 814                         err = 0;
 815                         clear = 1;
 816                         goto out;
 817                 }
 818         }
 819
 820         init_query_mad(in_mad);
 821         in_mad->attr_id  = IB_SMP_ATTR_GUID_INFO;
 822         in_mad->attr_mod = cpu_to_be32(index / 8);
 823
 824         err = mlx4_MAD_IFC(dev, mad_ifc_flags, port,
 825                            NULL, NULL, in_mad, out_mad);
 826         if (err)
 827                 goto out;
 828
 829         memcpy(gid->raw + 8, out_mad->data + (index % 8) * 8, 8);
 830
 831 out:
 832         if (clear)
 833                 memset(gid->raw + 8, 0, 8);
 834         kfree(in_mad);
 835         kfree(out_mad);
 836         return err;
 837 }
 838
 839 static int mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
 840                              union ib_gid *gid)
 841 {
 842         int ret;
 843
 844         if (rdma_protocol_ib(ibdev, port))
 845                 return __mlx4_ib_query_gid(ibdev, port, index, gid, 0);
 846
 847         if (!rdma_protocol_roce(ibdev, port))
 848                 return -ENODEV;
 849
 850         if (!rdma_cap_roce_gid_table(ibdev, port))
 851                 return -ENODEV;
 852
 853         ret = ib_get_cached_gid(ibdev, port, index, gid, NULL);
 854         if (ret == -EAGAIN) {
 855                 memcpy(gid, &zgid, sizeof(*gid));
 856                 return 0;
 857         }
 858
 859         return ret;
 860 }
 861
 862 static int mlx4_ib_query_sl2vl(struct ib_device *ibdev, u8 port, u64 *sl2vl_tbl)
 863 {
 864         union sl2vl_tbl_to_u64 sl2vl64;
 865         struct ib_smp *in_mad  = NULL;
 866         struct ib_smp *out_mad = NULL;
 867         int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
 868         int err = -ENOMEM;
 869         int jj;
 870
 871         if (mlx4_is_slave(to_mdev(ibdev)->dev)) {
 872                 *sl2vl_tbl = 0;
 873                 return 0;
 874         }
 875
 876         in_mad  = kzalloc(sizeof(*in_mad), GFP_KERNEL);
 877         out_mad = kmalloc(sizeof(*out_mad), GFP_KERNEL);
 878         if (!in_mad || !out_mad)
 879                 goto out;
 880
 881         init_query_mad(in_mad);
 882         in_mad->attr_id  = IB_SMP_ATTR_SL_TO_VL_TABLE;
 883         in_mad->attr_mod = 0;
 884
 885         if (mlx4_is_mfunc(to_mdev(ibdev)->dev))
 886                 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
 887
 888         err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
 889                            in_mad, out_mad);
 890         if (err)
 891                 goto out;
 892
 893         for (jj = 0; jj < 8; jj++)
 894                 sl2vl64.sl8[jj] = ((struct ib_smp *)out_mad)->data[jj];
 895         *sl2vl_tbl = sl2vl64.sl64;
 896
 897 out:
 898         kfree(in_mad);
 899         kfree(out_mad);
 900         return err;
 901 }
 902
 903 static void mlx4_init_sl2vl_tbl(struct mlx4_ib_dev *mdev)
 904 {
 905         u64 sl2vl;
 906         int i;
 907         int err;
 908
 909         for (i = 1; i <= mdev->dev->caps.num_ports; i++) {
 910                 if (mdev->dev->caps.port_type[i] == MLX4_PORT_TYPE_ETH)
 911                         continue;
 912                 err = mlx4_ib_query_sl2vl(&mdev->ib_dev, i, &sl2vl);
 913                 if (err) {
 914                         pr_err("Unable to get default sl to vl mapping for port %d.  Using all zeroes (%d)\n",
 915                                i, err);
 916                         sl2vl = 0;
 917                 }
 918                 atomic64_set(&mdev->sl2vl[i - 1], sl2vl);
 919         }
 920 }
 921
 922 int __mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index,
 923                          u16 *pkey, int netw_view)
 924 {
 925         struct ib_smp *in_mad  = NULL;
 926         struct ib_smp *out_mad = NULL;
 927         int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
 928         int err = -ENOMEM;
 929
 930         in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
 931         out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
 932         if (!in_mad || !out_mad)
 933                 goto out;
 934
 935         init_query_mad(in_mad);
 936         in_mad->attr_id  = IB_SMP_ATTR_PKEY_TABLE;
 937         in_mad->attr_mod = cpu_to_be32(index / 32);
 938
 939         if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
 940                 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
 941
 942         err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
 943                            in_mad, out_mad);
 944         if (err)
 945                 goto out;
 946
 947         *pkey = be16_to_cpu(((__be16 *) out_mad->data)[index % 32]);
 948
 949 out:
 950         kfree(in_mad);
 951         kfree(out_mad);
 952         return err;
 953 }
 954
 955 static int mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index, u16 *pkey)
 956 {
 957         return __mlx4_ib_query_pkey(ibdev, port, index, pkey, 0);
 958 }
 959
 960 static int mlx4_ib_modify_device(struct ib_device *ibdev, int mask,
 961                                  struct ib_device_modify *props)
 962 {
 963         struct mlx4_cmd_mailbox *mailbox;
 964         unsigned long flags;
 965
 966         if (mask & ~IB_DEVICE_MODIFY_NODE_DESC)
 967                 return -EOPNOTSUPP;
 968
 969         if (!(mask & IB_DEVICE_MODIFY_NODE_DESC))
 970                 return 0;
 971
 972         if (mlx4_is_slave(to_mdev(ibdev)->dev))
 973                 return -EOPNOTSUPP;
 974
 975         spin_lock_irqsave(&to_mdev(ibdev)->sm_lock, flags);
 976         memcpy(ibdev->node_desc, props->node_desc, IB_DEVICE_NODE_DESC_MAX);
 977         spin_unlock_irqrestore(&to_mdev(ibdev)->sm_lock, flags);
 978
 979         /*
 980          * If possible, pass node desc to FW, so it can generate
 981          * a 144 trap.  If cmd fails, just ignore.
 982          */
 983         mailbox = mlx4_alloc_cmd_mailbox(to_mdev(ibdev)->dev);
 984         if (IS_ERR(mailbox))
 985                 return 0;
 986
 987         memcpy(mailbox->buf, props->node_desc, IB_DEVICE_NODE_DESC_MAX);
 988         mlx4_cmd(to_mdev(ibdev)->dev, mailbox->dma, 1, 0,
 989                  MLX4_CMD_SET_NODE, MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
 990
 991         mlx4_free_cmd_mailbox(to_mdev(ibdev)->dev, mailbox);
 992
 993         return 0;
 994 }
 995
 996 static int mlx4_ib_SET_PORT(struct mlx4_ib_dev *dev, u8 port, int reset_qkey_viols,
 997                             u32 cap_mask)
 998 {
 999         struct mlx4_cmd_mailbox *mailbox;
1000         int err;
1001
1002         mailbox = mlx4_alloc_cmd_mailbox(dev->dev);
1003         if (IS_ERR(mailbox))
1004                 return PTR_ERR(mailbox);
1005
1006         if (dev->dev->flags & MLX4_FLAG_OLD_PORT_CMDS) {
1007                 *(u8 *) mailbox->buf         = !!reset_qkey_viols << 6;
1008                 ((__be32 *) mailbox->buf)[2] = cpu_to_be32(cap_mask);
1009         } else {
1010                 ((u8 *) mailbox->buf)[3]     = !!reset_qkey_viols;
1011                 ((__be32 *) mailbox->buf)[1] = cpu_to_be32(cap_mask);
1012         }
1013
1014         err = mlx4_cmd(dev->dev, mailbox->dma, port, MLX4_SET_PORT_IB_OPCODE,
1015                        MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
1016                        MLX4_CMD_WRAPPED);
1017
1018         mlx4_free_cmd_mailbox(dev->dev, mailbox);
1019         return err;
1020 }
1021
1022 static int mlx4_ib_modify_port(struct ib_device *ibdev, u8 port, int mask,
1023                                struct ib_port_modify *props)
1024 {
1025         struct mlx4_ib_dev *mdev = to_mdev(ibdev);
1026         u8 is_eth = mdev->dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH;
1027         struct ib_port_attr attr;
1028         u32 cap_mask;
1029         int err;
1030
1031         /* return OK if this is RoCE. CM calls ib_modify_port() regardless
1032          * of whether port link layer is ETH or IB. For ETH ports, qkey
1033          * violations and port capabilities are not meaningful.
1034          */
1035         if (is_eth)
1036                 return 0;
1037
1038         mutex_lock(&mdev->cap_mask_mutex);
1039
1040         err = ib_query_port(ibdev, port, &attr);
1041         if (err)
1042                 goto out;
1043
1044         cap_mask = (attr.port_cap_flags | props->set_port_cap_mask) &
1045                 ~props->clr_port_cap_mask;
1046
1047         err = mlx4_ib_SET_PORT(mdev, port,
1048                                !!(mask & IB_PORT_RESET_QKEY_CNTR),
1049                                cap_mask);
1050
1051 out:
1052         mutex_unlock(&to_mdev(ibdev)->cap_mask_mutex);
1053         return err;
1054 }
1055
1056 static struct ib_ucontext *mlx4_ib_alloc_ucontext(struct ib_device *ibdev,
1057                                                   struct ib_udata *udata)
1058 {
1059         struct mlx4_ib_dev *dev = to_mdev(ibdev);
1060         struct mlx4_ib_ucontext *context;
1061         struct mlx4_ib_alloc_ucontext_resp_v3 resp_v3;
1062         struct mlx4_ib_alloc_ucontext_resp resp;
1063         int err;
1064
1065         if (!dev->ib_active)
1066                 return ERR_PTR(-EAGAIN);
1067
1068         if (ibdev->uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION) {
1069                 resp_v3.qp_tab_size      = dev->dev->caps.num_qps;
1070                 resp_v3.bf_reg_size      = dev->dev->caps.bf_reg_size;
1071                 resp_v3.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
1072         } else {
1073                 resp.dev_caps         = dev->dev->caps.userspace_caps;
1074                 resp.qp_tab_size      = dev->dev->caps.num_qps;
1075                 resp.bf_reg_size      = dev->dev->caps.bf_reg_size;
1076                 resp.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
1077                 resp.cqe_size         = dev->dev->caps.cqe_size;
1078         }
1079
1080         context = kzalloc(sizeof(*context), GFP_KERNEL);
1081         if (!context)
1082                 return ERR_PTR(-ENOMEM);
1083
1084         err = mlx4_uar_alloc(to_mdev(ibdev)->dev, &context->uar);
1085         if (err) {
1086                 kfree(context);
1087                 return ERR_PTR(err);
1088         }
1089
1090         INIT_LIST_HEAD(&context->db_page_list);
1091         mutex_init(&context->db_page_mutex);
1092
1093         INIT_LIST_HEAD(&context->wqn_ranges_list);
1094         mutex_init(&context->wqn_ranges_mutex);
1095
1096         if (ibdev->uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION)
1097                 err = ib_copy_to_udata(udata, &resp_v3, sizeof(resp_v3));
1098         else
1099                 err = ib_copy_to_udata(udata, &resp, sizeof(resp));
1100
1101         if (err) {
1102                 mlx4_uar_free(to_mdev(ibdev)->dev, &context->uar);
1103                 kfree(context);
1104                 return ERR_PTR(-EFAULT);
1105         }
1106
1107         return &context->ibucontext;
1108 }
1109
1110 static int mlx4_ib_dealloc_ucontext(struct ib_ucontext *ibcontext)
1111 {
1112         struct mlx4_ib_ucontext *context = to_mucontext(ibcontext);
1113
1114         mlx4_uar_free(to_mdev(ibcontext->device)->dev, &context->uar);
1115         kfree(context);
1116
1117         return 0;
1118 }
1119
1120 static void  mlx4_ib_vma_open(struct vm_area_struct *area)
1121 {
1122         /* vma_open is called when a new VMA is created on top of our VMA.
1123          * This is done through either mremap flow or split_vma (usually due
1124          * to mlock, madvise, munmap, etc.). We do not support a clone of the
1125          * vma, as this VMA is strongly hardware related. Therefore we set the
1126          * vm_ops of the newly created/cloned VMA to NULL, to prevent it from
1127          * calling us again and trying to do incorrect actions. We assume that
1128          * the original vma size is exactly a single page that there will be no
1129          * "splitting" operations on.
1130          */
1131         area->vm_ops = NULL;
1132 }
1133
1134 static void  mlx4_ib_vma_close(struct vm_area_struct *area)
1135 {
1136         struct mlx4_ib_vma_private_data *mlx4_ib_vma_priv_data;
1137
1138         /* It's guaranteed that all VMAs opened on a FD are closed before the
1139          * file itself is closed, therefore no sync is needed with the regular
1140          * closing flow. (e.g. mlx4_ib_dealloc_ucontext) However need a sync
1141          * with accessing the vma as part of mlx4_ib_disassociate_ucontext.
1142          * The close operation is usually called under mm->mmap_sem except when
1143          * process is exiting.  The exiting case is handled explicitly as part
1144          * of mlx4_ib_disassociate_ucontext.
1145          */
1146         mlx4_ib_vma_priv_data = (struct mlx4_ib_vma_private_data *)
1147                                 area->vm_private_data;
1148
1149         /* set the vma context pointer to null in the mlx4_ib driver's private
1150          * data to protect against a race condition in mlx4_ib_dissassociate_ucontext().
1151          */
1152         mlx4_ib_vma_priv_data->vma = NULL;
1153 }
1154
1155 static const struct vm_operations_struct mlx4_ib_vm_ops = {
1156         .open = mlx4_ib_vma_open,
1157         .close = mlx4_ib_vma_close
1158 };
1159
1160 static void mlx4_ib_disassociate_ucontext(struct ib_ucontext *ibcontext)
1161 {
1162         int i;
1163         int ret = 0;
1164         struct vm_area_struct *vma;
1165         struct mlx4_ib_ucontext *context = to_mucontext(ibcontext);
1166         struct task_struct *owning_process  = NULL;
1167         struct mm_struct   *owning_mm       = NULL;
1168
1169         owning_process = get_pid_task(ibcontext->tgid, PIDTYPE_PID);
1170         if (!owning_process)
1171                 return;
1172
1173         owning_mm = get_task_mm(owning_process);
1174         if (!owning_mm) {
1175                 pr_info("no mm, disassociate ucontext is pending task termination\n");
1176                 while (1) {
1177                         /* make sure that task is dead before returning, it may
1178                          * prevent a rare case of module down in parallel to a
1179                          * call to mlx4_ib_vma_close.
1180                          */
1181                         put_task_struct(owning_process);
1182                         usleep_range(1000, 2000);
1183                         owning_process = get_pid_task(ibcontext->tgid,
1184                                                       PIDTYPE_PID);
1185                         if (!owning_process ||
1186                             owning_process->state == TASK_DEAD) {
1187                                 pr_info("disassociate ucontext done, task was terminated\n");
1188                                 /* in case task was dead need to release the task struct */
1189                                 if (owning_process)
1190                                         put_task_struct(owning_process);
1191                                 return;
1192                         }
1193                 }
1194         }
1195
1196         /* need to protect from a race on closing the vma as part of
1197          * mlx4_ib_vma_close().
1198          */
1199         down_write(&owning_mm->mmap_sem);
1200         if (!mmget_still_valid(owning_mm))
1201                 goto skip_mm;
1202         for (i = 0; i < HW_BAR_COUNT; i++) {
1203                 vma = context->hw_bar_info[i].vma;
1204                 if (!vma)
1205                         continue;
1206
1207                 ret = zap_vma_ptes(context->hw_bar_info[i].vma,
1208                                    context->hw_bar_info[i].vma->vm_start,
1209                                    PAGE_SIZE);
1210                 if (ret) {
1211                         pr_err("Error: zap_vma_ptes failed for index=%d, ret=%d\n", i, ret);
1212                         BUG_ON(1);
1213                 }
1214
1215                 context->hw_bar_info[i].vma->vm_flags &=
1216                         ~(VM_SHARED | VM_MAYSHARE);
1217                 /* context going to be destroyed, should not access ops any more */
1218                 context->hw_bar_info[i].vma->vm_ops = NULL;
1219         }
1220 skip_mm:
1221         up_write(&owning_mm->mmap_sem);
1222         mmput(owning_mm);
1223         put_task_struct(owning_process);
1224 }
1225
1226 static void mlx4_ib_set_vma_data(struct vm_area_struct *vma,
1227                                  struct mlx4_ib_vma_private_data *vma_private_data)
1228 {
1229         vma_private_data->vma = vma;
1230         vma->vm_private_data = vma_private_data;
1231         vma->vm_ops =  &mlx4_ib_vm_ops;
1232 }
1233
1234 static int mlx4_ib_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
1235 {
1236         struct mlx4_ib_dev *dev = to_mdev(context->device);
1237         struct mlx4_ib_ucontext *mucontext = to_mucontext(context);
1238
1239         if (vma->vm_end - vma->vm_start != PAGE_SIZE)
1240                 return -EINVAL;
1241
1242         if (vma->vm_pgoff == 0) {
1243                 /* We prevent double mmaping on same context */
1244                 if (mucontext->hw_bar_info[HW_BAR_DB].vma)
1245                         return -EINVAL;
1246
1247                 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
1248
1249                 if (io_remap_pfn_range(vma, vma->vm_start,
1250                                        to_mucontext(context)->uar.pfn,
1251                                        PAGE_SIZE, vma->vm_page_prot))
1252                         return -EAGAIN;
1253
1254                 mlx4_ib_set_vma_data(vma, &mucontext->hw_bar_info[HW_BAR_DB]);
1255
1256         } else if (vma->vm_pgoff == 1 && dev->dev->caps.bf_reg_size != 0) {
1257                 /* We prevent double mmaping on same context */
1258                 if (mucontext->hw_bar_info[HW_BAR_BF].vma)
1259                         return -EINVAL;
1260
1261                 vma->vm_page_prot = pgprot_writecombine(vma->vm_page_prot);
1262
1263                 if (io_remap_pfn_range(vma, vma->vm_start,
1264                                        to_mucontext(context)->uar.pfn +
1265                                        dev->dev->caps.num_uars,
1266                                        PAGE_SIZE, vma->vm_page_prot))
1267                         return -EAGAIN;
1268
1269                 mlx4_ib_set_vma_data(vma, &mucontext->hw_bar_info[HW_BAR_BF]);
1270
1271         } else if (vma->vm_pgoff == 3) {
1272                 struct mlx4_clock_params params;
1273                 int ret;
1274
1275                 /* We prevent double mmaping on same context */
1276                 if (mucontext->hw_bar_info[HW_BAR_CLOCK].vma)
1277                         return -EINVAL;
1278
1279                 ret = mlx4_get_internal_clock_params(dev->dev, &params);
1280
1281                 if (ret)
1282                         return ret;
1283
1284                 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
1285                 if (io_remap_pfn_range(vma, vma->vm_start,
1286                                        (pci_resource_start(dev->dev->persist->pdev,
1287                                                            params.bar) +
1288                                         params.offset)
1289                                        >> PAGE_SHIFT,
1290                                        PAGE_SIZE, vma->vm_page_prot))
1291                         return -EAGAIN;
1292
1293                 mlx4_ib_set_vma_data(vma,
1294                                      &mucontext->hw_bar_info[HW_BAR_CLOCK]);
1295         } else {
1296                 return -EINVAL;
1297         }
1298
1299         return 0;
1300 }
1301
1302 static struct ib_pd *mlx4_ib_alloc_pd(struct ib_device *ibdev,
1303                                       struct ib_ucontext *context,
1304                                       struct ib_udata *udata)
1305 {
1306         struct mlx4_ib_pd *pd;
1307         int err;
1308
1309         pd = kmalloc(sizeof *pd, GFP_KERNEL);
1310         if (!pd)
1311                 return ERR_PTR(-ENOMEM);
1312
1313         err = mlx4_pd_alloc(to_mdev(ibdev)->dev, &pd->pdn);
1314         if (err) {
1315                 kfree(pd);
1316                 return ERR_PTR(err);
1317         }
1318
1319         if (context)
1320                 if (ib_copy_to_udata(udata, &pd->pdn, sizeof (__u32))) {
1321                         mlx4_pd_free(to_mdev(ibdev)->dev, pd->pdn);
1322                         kfree(pd);
1323                         return ERR_PTR(-EFAULT);
1324                 }
1325
1326         return &pd->ibpd;
1327 }
1328
1329 static int mlx4_ib_dealloc_pd(struct ib_pd *pd)
1330 {
1331         mlx4_pd_free(to_mdev(pd->device)->dev, to_mpd(pd)->pdn);
1332         kfree(pd);
1333
1334         return 0;
1335 }
1336
1337 static struct ib_xrcd *mlx4_ib_alloc_xrcd(struct ib_device *ibdev,
1338                                           struct ib_ucontext *context,
1339                                           struct ib_udata *udata)
1340 {
1341         struct mlx4_ib_xrcd *xrcd;
1342         struct ib_cq_init_attr cq_attr = {};
1343         int err;
1344
1345         if (!(to_mdev(ibdev)->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC))
1346                 return ERR_PTR(-ENOSYS);
1347
1348         xrcd = kmalloc(sizeof *xrcd, GFP_KERNEL);
1349         if (!xrcd)
1350                 return ERR_PTR(-ENOMEM);
1351
1352         err = mlx4_xrcd_alloc(to_mdev(ibdev)->dev, &xrcd->xrcdn);
1353         if (err)
1354                 goto err1;
1355
1356         xrcd->pd = ib_alloc_pd(ibdev, 0);
1357         if (IS_ERR(xrcd->pd)) {
1358                 err = PTR_ERR(xrcd->pd);
1359                 goto err2;
1360         }
1361
1362         cq_attr.cqe = 1;
1363         xrcd->cq = ib_create_cq(ibdev, NULL, NULL, xrcd, &cq_attr);
1364         if (IS_ERR(xrcd->cq)) {
1365                 err = PTR_ERR(xrcd->cq);
1366                 goto err3;
1367         }
1368
1369         return &xrcd->ibxrcd;
1370
1371 err3:
1372         ib_dealloc_pd(xrcd->pd);
1373 err2:
1374         mlx4_xrcd_free(to_mdev(ibdev)->dev, xrcd->xrcdn);
1375 err1:
1376         kfree(xrcd);
1377         return ERR_PTR(err);
1378 }
1379
1380 static int mlx4_ib_dealloc_xrcd(struct ib_xrcd *xrcd)
1381 {
1382         ib_destroy_cq(to_mxrcd(xrcd)->cq);
1383         ib_dealloc_pd(to_mxrcd(xrcd)->pd);
1384         mlx4_xrcd_free(to_mdev(xrcd->device)->dev, to_mxrcd(xrcd)->xrcdn);
1385         kfree(xrcd);
1386
1387         return 0;
1388 }
1389
1390 static int add_gid_entry(struct ib_qp *ibqp, union ib_gid *gid)
1391 {
1392         struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1393         struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1394         struct mlx4_ib_gid_entry *ge;
1395
1396         ge = kzalloc(sizeof *ge, GFP_KERNEL);
1397         if (!ge)
1398                 return -ENOMEM;
1399
1400         ge->gid = *gid;
1401         if (mlx4_ib_add_mc(mdev, mqp, gid)) {
1402                 ge->port = mqp->port;
1403                 ge->added = 1;
1404         }
1405
1406         mutex_lock(&mqp->mutex);
1407         list_add_tail(&ge->list, &mqp->gid_list);
1408         mutex_unlock(&mqp->mutex);
1409
1410         return 0;
1411 }
1412
1413 static void mlx4_ib_delete_counters_table(struct mlx4_ib_dev *ibdev,
1414                                           struct mlx4_ib_counters *ctr_table)
1415 {
1416         struct counter_index *counter, *tmp_count;
1417
1418         mutex_lock(&ctr_table->mutex);
1419         list_for_each_entry_safe(counter, tmp_count, &ctr_table->counters_list,
1420                                  list) {
1421                 if (counter->allocated)
1422                         mlx4_counter_free(ibdev->dev, counter->index);
1423                 list_del(&counter->list);
1424                 kfree(counter);
1425         }
1426         mutex_unlock(&ctr_table->mutex);
1427 }
1428
1429 int mlx4_ib_add_mc(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
1430                    union ib_gid *gid)
1431 {
1432         struct net_device *ndev;
1433         int ret = 0;
1434
1435         if (!mqp->port)
1436                 return 0;
1437
1438         spin_lock_bh(&mdev->iboe.lock);
1439         ndev = mdev->iboe.netdevs[mqp->port - 1];
1440         if (ndev)
1441                 dev_hold(ndev);
1442         spin_unlock_bh(&mdev->iboe.lock);
1443
1444         if (ndev) {
1445                 ret = 1;
1446                 dev_put(ndev);
1447         }
1448
1449         return ret;
1450 }
1451
1452 struct mlx4_ib_steering {
1453         struct list_head list;
1454         struct mlx4_flow_reg_id reg_id;
1455         union ib_gid gid;
1456 };
1457
1458 #define LAST_ETH_FIELD vlan_tag
1459 #define LAST_IB_FIELD sl
1460 #define LAST_IPV4_FIELD dst_ip
1461 #define LAST_TCP_UDP_FIELD src_port
1462
1463 /* Field is the last supported field */
1464 #define FIELDS_NOT_SUPPORTED(filter, field)\
1465         memchr_inv((void *)&filter.field  +\
1466                    sizeof(filter.field), 0,\
1467                    sizeof(filter) -\
1468                    offsetof(typeof(filter), field) -\
1469                    sizeof(filter.field))
1470
1471 static int parse_flow_attr(struct mlx4_dev *dev,
1472                            u32 qp_num,
1473                            union ib_flow_spec *ib_spec,
1474                            struct _rule_hw *mlx4_spec)
1475 {
1476         enum mlx4_net_trans_rule_id type;
1477
1478         switch (ib_spec->type) {
1479         case IB_FLOW_SPEC_ETH:
1480                 if (FIELDS_NOT_SUPPORTED(ib_spec->eth.mask, LAST_ETH_FIELD))
1481                         return -ENOTSUPP;
1482
1483                 type = MLX4_NET_TRANS_RULE_ID_ETH;
1484                 memcpy(mlx4_spec->eth.dst_mac, ib_spec->eth.val.dst_mac,
1485                        ETH_ALEN);
1486                 memcpy(mlx4_spec->eth.dst_mac_msk, ib_spec->eth.mask.dst_mac,
1487                        ETH_ALEN);
1488                 mlx4_spec->eth.vlan_tag = ib_spec->eth.val.vlan_tag;
1489                 mlx4_spec->eth.vlan_tag_msk = ib_spec->eth.mask.vlan_tag;
1490                 break;
1491         case IB_FLOW_SPEC_IB:
1492                 if (FIELDS_NOT_SUPPORTED(ib_spec->ib.mask, LAST_IB_FIELD))
1493                         return -ENOTSUPP;
1494
1495                 type = MLX4_NET_TRANS_RULE_ID_IB;
1496                 mlx4_spec->ib.l3_qpn =
1497                         cpu_to_be32(qp_num);
1498                 mlx4_spec->ib.qpn_mask =
1499                         cpu_to_be32(MLX4_IB_FLOW_QPN_MASK);
1500                 break;
1501
1502
1503         case IB_FLOW_SPEC_IPV4:
1504                 if (FIELDS_NOT_SUPPORTED(ib_spec->ipv4.mask, LAST_IPV4_FIELD))
1505                         return -ENOTSUPP;
1506
1507                 type = MLX4_NET_TRANS_RULE_ID_IPV4;
1508                 mlx4_spec->ipv4.src_ip = ib_spec->ipv4.val.src_ip;
1509                 mlx4_spec->ipv4.src_ip_msk = ib_spec->ipv4.mask.src_ip;
1510                 mlx4_spec->ipv4.dst_ip = ib_spec->ipv4.val.dst_ip;
1511                 mlx4_spec->ipv4.dst_ip_msk = ib_spec->ipv4.mask.dst_ip;
1512                 break;
1513
1514         case IB_FLOW_SPEC_TCP:
1515         case IB_FLOW_SPEC_UDP:
1516                 if (FIELDS_NOT_SUPPORTED(ib_spec->tcp_udp.mask, LAST_TCP_UDP_FIELD))
1517                         return -ENOTSUPP;
1518
1519                 type = ib_spec->type == IB_FLOW_SPEC_TCP ?
1520                                         MLX4_NET_TRANS_RULE_ID_TCP :
1521                                         MLX4_NET_TRANS_RULE_ID_UDP;
1522                 mlx4_spec->tcp_udp.dst_port = ib_spec->tcp_udp.val.dst_port;
1523                 mlx4_spec->tcp_udp.dst_port_msk = ib_spec->tcp_udp.mask.dst_port;
1524                 mlx4_spec->tcp_udp.src_port = ib_spec->tcp_udp.val.src_port;
1525                 mlx4_spec->tcp_udp.src_port_msk = ib_spec->tcp_udp.mask.src_port;
1526                 break;
1527
1528         default:
1529                 return -EINVAL;
1530         }
1531         if (mlx4_map_sw_to_hw_steering_id(dev, type) < 0 ||
1532             mlx4_hw_rule_sz(dev, type) < 0)
1533                 return -EINVAL;
1534         mlx4_spec->id = cpu_to_be16(mlx4_map_sw_to_hw_steering_id(dev, type));
1535         mlx4_spec->size = mlx4_hw_rule_sz(dev, type) >> 2;
1536         return mlx4_hw_rule_sz(dev, type);
1537 }
1538
1539 struct default_rules {
1540         __u32 mandatory_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
1541         __u32 mandatory_not_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
1542         __u32 rules_create_list[IB_FLOW_SPEC_SUPPORT_LAYERS];
1543         __u8  link_layer;
1544 };
1545 static const struct default_rules default_table[] = {
1546         {
1547                 .mandatory_fields = {IB_FLOW_SPEC_IPV4},
1548                 .mandatory_not_fields = {IB_FLOW_SPEC_ETH},
1549                 .rules_create_list = {IB_FLOW_SPEC_IB},
1550                 .link_layer = IB_LINK_LAYER_INFINIBAND
1551         }
1552 };
1553
1554 static int __mlx4_ib_default_rules_match(struct ib_qp *qp,
1555                                          struct ib_flow_attr *flow_attr)
1556 {
1557         int i, j, k;
1558         void *ib_flow;
1559         const struct default_rules *pdefault_rules = default_table;
1560         u8 link_layer = rdma_port_get_link_layer(qp->device, flow_attr->port);
1561
1562         for (i = 0; i < ARRAY_SIZE(default_table); i++, pdefault_rules++) {
1563                 __u32 field_types[IB_FLOW_SPEC_SUPPORT_LAYERS];
1564                 memset(&field_types, 0, sizeof(field_types));
1565
1566                 if (link_layer != pdefault_rules->link_layer)
1567                         continue;
1568
1569                 ib_flow = flow_attr + 1;
1570                 /* we assume the specs are sorted */
1571                 for (j = 0, k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS &&
1572                      j < flow_attr->num_of_specs; k++) {
1573                         union ib_flow_spec *current_flow =
1574                                 (union ib_flow_spec *)ib_flow;
1575
1576                         /* same layer but different type */
1577                         if (((current_flow->type & IB_FLOW_SPEC_LAYER_MASK) ==
1578                              (pdefault_rules->mandatory_fields[k] &
1579                               IB_FLOW_SPEC_LAYER_MASK)) &&
1580                             (current_flow->type !=
1581                              pdefault_rules->mandatory_fields[k]))
1582                                 goto out;
1583
1584                         /* same layer, try match next one */
1585                         if (current_flow->type ==
1586                             pdefault_rules->mandatory_fields[k]) {
1587                                 j++;
1588                                 ib_flow +=
1589                                         ((union ib_flow_spec *)ib_flow)->size;
1590                         }
1591                 }
1592
1593                 ib_flow = flow_attr + 1;
1594                 for (j = 0; j < flow_attr->num_of_specs;
1595                      j++, ib_flow += ((union ib_flow_spec *)ib_flow)->size)
1596                         for (k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS; k++)
1597                                 /* same layer and same type */
1598                                 if (((union ib_flow_spec *)ib_flow)->type ==
1599                                     pdefault_rules->mandatory_not_fields[k])
1600                                         goto out;
1601
1602                 return i;
1603         }
1604 out:
1605         return -1;
1606 }
1607
1608 static int __mlx4_ib_create_default_rules(
1609                 struct mlx4_ib_dev *mdev,
1610                 struct ib_qp *qp,
1611                 const struct default_rules *pdefault_rules,
1612                 struct _rule_hw *mlx4_spec) {
1613         int size = 0;
1614         int i;
1615
1616         for (i = 0; i < ARRAY_SIZE(pdefault_rules->rules_create_list); i++) {
1617                 union ib_flow_spec ib_spec = {};
1618                 int ret;
1619
1620                 switch (pdefault_rules->rules_create_list[i]) {
1621                 case 0:
1622                         /* no rule */
1623                         continue;
1624                 case IB_FLOW_SPEC_IB:
1625                         ib_spec.type = IB_FLOW_SPEC_IB;
1626                         ib_spec.size = sizeof(struct ib_flow_spec_ib);
1627
1628                         break;
1629                 default:
1630                         /* invalid rule */
1631                         return -EINVAL;
1632                 }
1633                 /* We must put empty rule, qpn is being ignored */
1634                 ret = parse_flow_attr(mdev->dev, 0, &ib_spec,
1635                                       mlx4_spec);
1636                 if (ret < 0) {
1637                         pr_info("invalid parsing\n");
1638                         return -EINVAL;
1639                 }
1640
1641                 mlx4_spec = (void *)mlx4_spec + ret;
1642                 size += ret;
1643         }
1644         return size;
1645 }
1646
1647 static int __mlx4_ib_create_flow(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
1648                           int domain,
1649                           enum mlx4_net_trans_promisc_mode flow_type,
1650                           u64 *reg_id)
1651 {
1652         int ret, i;
1653         int size = 0;
1654         void *ib_flow;
1655         struct mlx4_ib_dev *mdev = to_mdev(qp->device);
1656         struct mlx4_cmd_mailbox *mailbox;
1657         struct mlx4_net_trans_rule_hw_ctrl *ctrl;
1658         int default_flow;
1659
1660         static const u16 __mlx4_domain[] = {
1661                 [IB_FLOW_DOMAIN_USER] = MLX4_DOMAIN_UVERBS,
1662                 [IB_FLOW_DOMAIN_ETHTOOL] = MLX4_DOMAIN_ETHTOOL,
1663                 [IB_FLOW_DOMAIN_RFS] = MLX4_DOMAIN_RFS,
1664                 [IB_FLOW_DOMAIN_NIC] = MLX4_DOMAIN_NIC,
1665         };
1666
1667         if (flow_attr->priority > MLX4_IB_FLOW_MAX_PRIO) {
1668                 pr_err("Invalid priority value %d\n", flow_attr->priority);
1669                 return -EINVAL;
1670         }
1671
1672         if (domain >= IB_FLOW_DOMAIN_NUM) {
1673                 pr_err("Invalid domain value %d\n", domain);
1674                 return -EINVAL;
1675         }
1676
1677         if (mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type) < 0)
1678                 return -EINVAL;
1679
1680         mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
1681         if (IS_ERR(mailbox))
1682                 return PTR_ERR(mailbox);
1683         ctrl = mailbox->buf;
1684
1685         ctrl->prio = cpu_to_be16(__mlx4_domain[domain] |
1686                                  flow_attr->priority);
1687         ctrl->type = mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type);
1688         ctrl->port = flow_attr->port;
1689         ctrl->qpn = cpu_to_be32(qp->qp_num);
1690
1691         ib_flow = flow_attr + 1;
1692         size += sizeof(struct mlx4_net_trans_rule_hw_ctrl);
1693         /* Add default flows */
1694         default_flow = __mlx4_ib_default_rules_match(qp, flow_attr);
1695         if (default_flow >= 0) {
1696                 ret = __mlx4_ib_create_default_rules(
1697                                 mdev, qp, default_table + default_flow,
1698                                 mailbox->buf + size);
1699                 if (ret < 0) {
1700                         mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1701                         return -EINVAL;
1702                 }
1703                 size += ret;
1704         }
1705         for (i = 0; i < flow_attr->num_of_specs; i++) {
1706                 ret = parse_flow_attr(mdev->dev, qp->qp_num, ib_flow,
1707                                       mailbox->buf + size);
1708                 if (ret < 0) {
1709                         mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1710                         return -EINVAL;
1711                 }
1712                 ib_flow += ((union ib_flow_spec *) ib_flow)->size;
1713                 size += ret;
1714         }
1715
1716         if (mlx4_is_master(mdev->dev) && flow_type == MLX4_FS_REGULAR &&
1717             flow_attr->num_of_specs == 1) {
1718                 struct _rule_hw *rule_header = (struct _rule_hw *)(ctrl + 1);
1719                 enum ib_flow_spec_type header_spec =
1720                         ((union ib_flow_spec *)(flow_attr + 1))->type;
1721
1722                 if (header_spec == IB_FLOW_SPEC_ETH)
1723                         mlx4_handle_eth_header_mcast_prio(ctrl, rule_header);
1724         }
1725
1726         ret = mlx4_cmd_imm(mdev->dev, mailbox->dma, reg_id, size >> 2, 0,
1727                            MLX4_QP_FLOW_STEERING_ATTACH, MLX4_CMD_TIME_CLASS_A,
1728                            MLX4_CMD_NATIVE);
1729         if (ret == -ENOMEM)
1730                 pr_err("mcg table is full. Fail to register network rule.\n");
1731         else if (ret == -ENXIO)
1732                 pr_err("Device managed flow steering is disabled. Fail to register network rule.\n");
1733         else if (ret)
1734                 pr_err("Invalid argument. Fail to register network rule.\n");
1735
1736         mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1737         return ret;
1738 }
1739
1740 static int __mlx4_ib_destroy_flow(struct mlx4_dev *dev, u64 reg_id)
1741 {
1742         int err;
1743         err = mlx4_cmd(dev, reg_id, 0, 0,
1744                        MLX4_QP_FLOW_STEERING_DETACH, MLX4_CMD_TIME_CLASS_A,
1745                        MLX4_CMD_NATIVE);
1746         if (err)
1747                 pr_err("Fail to detach network rule. registration id = 0x%llx\n",
1748                        reg_id);
1749         return err;
1750 }
1751
1752 static int mlx4_ib_tunnel_steer_add(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
1753                                     u64 *reg_id)
1754 {
1755         void *ib_flow;
1756         union ib_flow_spec *ib_spec;
1757         struct mlx4_dev *dev = to_mdev(qp->device)->dev;
1758         int err = 0;
1759
1760         if (dev->caps.tunnel_offload_mode != MLX4_TUNNEL_OFFLOAD_MODE_VXLAN ||
1761             dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC)
1762                 return 0; /* do nothing */
1763
1764         ib_flow = flow_attr + 1;
1765         ib_spec = (union ib_flow_spec *)ib_flow;
1766
1767         if (ib_spec->type !=  IB_FLOW_SPEC_ETH || flow_attr->num_of_specs != 1)
1768                 return 0; /* do nothing */
1769
1770         err = mlx4_tunnel_steer_add(to_mdev(qp->device)->dev, ib_spec->eth.val.dst_mac,
1771                                     flow_attr->port, qp->qp_num,
1772                                     MLX4_DOMAIN_UVERBS | (flow_attr->priority & 0xff),
1773                                     reg_id);
1774         return err;
1775 }
1776
1777 static int mlx4_ib_add_dont_trap_rule(struct mlx4_dev *dev,
1778                                       struct ib_flow_attr *flow_attr,
1779                                       enum mlx4_net_trans_promisc_mode *type)
1780 {
1781         int err = 0;
1782
1783         if (!(dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DMFS_UC_MC_SNIFFER) ||
1784             (dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC) ||
1785             (flow_attr->num_of_specs > 1) || (flow_attr->priority != 0)) {
1786                 return -EOPNOTSUPP;
1787         }
1788
1789         if (flow_attr->num_of_specs == 0) {
1790                 type[0] = MLX4_FS_MC_SNIFFER;
1791                 type[1] = MLX4_FS_UC_SNIFFER;
1792         } else {
1793                 union ib_flow_spec *ib_spec;
1794
1795                 ib_spec = (union ib_flow_spec *)(flow_attr + 1);
1796                 if (ib_spec->type !=  IB_FLOW_SPEC_ETH)
1797                         return -EINVAL;
1798
1799                 /* if all is zero than MC and UC */
1800                 if (is_zero_ether_addr(ib_spec->eth.mask.dst_mac)) {
1801                         type[0] = MLX4_FS_MC_SNIFFER;
1802                         type[1] = MLX4_FS_UC_SNIFFER;
1803                 } else {
1804                         u8 mac[ETH_ALEN] = {ib_spec->eth.mask.dst_mac[0] ^ 0x01,
1805                                             ib_spec->eth.mask.dst_mac[1],
1806                                             ib_spec->eth.mask.dst_mac[2],
1807                                             ib_spec->eth.mask.dst_mac[3],
1808                                             ib_spec->eth.mask.dst_mac[4],
1809                                             ib_spec->eth.mask.dst_mac[5]};
1810
1811                         /* Above xor was only on MC bit, non empty mask is valid
1812                          * only if this bit is set and rest are zero.
1813                          */
1814                         if (!is_zero_ether_addr(&mac[0]))
1815                                 return -EINVAL;
1816
1817                         if (is_multicast_ether_addr(ib_spec->eth.val.dst_mac))
1818                                 type[0] = MLX4_FS_MC_SNIFFER;
1819                         else
1820                                 type[0] = MLX4_FS_UC_SNIFFER;
1821                 }
1822         }
1823
1824         return err;
1825 }
1826
1827 static struct ib_flow *mlx4_ib_create_flow(struct ib_qp *qp,
1828                                     struct ib_flow_attr *flow_attr,
1829                                     int domain)
1830 {
1831         int err = 0, i = 0, j = 0;
1832         struct mlx4_ib_flow *mflow;
1833         enum mlx4_net_trans_promisc_mode type[2];
1834         struct mlx4_dev *dev = (to_mdev(qp->device))->dev;
1835         int is_bonded = mlx4_is_bonded(dev);
1836
1837         if (flow_attr->port < 1 || flow_attr->port > qp->device->phys_port_cnt)
1838                 return ERR_PTR(-EINVAL);
1839
1840         if ((flow_attr->flags & IB_FLOW_ATTR_FLAGS_DONT_TRAP) &&
1841             (flow_attr->type != IB_FLOW_ATTR_NORMAL))
1842                 return ERR_PTR(-EOPNOTSUPP);
1843
1844         memset(type, 0, sizeof(type));
1845
1846         mflow = kzalloc(sizeof(*mflow), GFP_KERNEL);
1847         if (!mflow) {
1848                 err = -ENOMEM;
1849                 goto err_free;
1850         }
1851
1852         switch (flow_attr->type) {
1853         case IB_FLOW_ATTR_NORMAL:
1854                 /* If dont trap flag (continue match) is set, under specific
1855                  * condition traffic be replicated to given qp,
1856                  * without stealing it
1857                  */
1858                 if (unlikely(flow_attr->flags & IB_FLOW_ATTR_FLAGS_DONT_TRAP)) {
1859                         err = mlx4_ib_add_dont_trap_rule(dev,
1860                                                          flow_attr,
1861                                                          type);
1862                         if (err)
1863                                 goto err_free;
1864                 } else {
1865                         type[0] = MLX4_FS_REGULAR;
1866                 }
1867                 break;
1868
1869         case IB_FLOW_ATTR_ALL_DEFAULT:
1870                 type[0] = MLX4_FS_ALL_DEFAULT;
1871                 break;
1872
1873         case IB_FLOW_ATTR_MC_DEFAULT:
1874                 type[0] = MLX4_FS_MC_DEFAULT;
1875                 break;
1876
1877         case IB_FLOW_ATTR_SNIFFER:
1878                 type[0] = MLX4_FS_MIRROR_RX_PORT;
1879                 type[1] = MLX4_FS_MIRROR_SX_PORT;
1880                 break;
1881
1882         default:
1883                 err = -EINVAL;
1884                 goto err_free;
1885         }
1886
1887         while (i < ARRAY_SIZE(type) && type[i]) {
1888                 err = __mlx4_ib_create_flow(qp, flow_attr, domain, type[i],
1889                                             &mflow->reg_id[i].id);
1890                 if (err)
1891                         goto err_create_flow;
1892                 if (is_bonded) {
1893                         /* Application always sees one port so the mirror rule
1894                          * must be on port #2
1895                          */
1896                         flow_attr->port = 2;
1897                         err = __mlx4_ib_create_flow(qp, flow_attr,
1898                                                     domain, type[j],
1899                                                     &mflow->reg_id[j].mirror);
1900                         flow_attr->port = 1;
1901                         if (err)
1902                                 goto err_create_flow;
1903                         j++;
1904                 }
1905
1906                 i++;
1907         }
1908
1909         if (i < ARRAY_SIZE(type) && flow_attr->type == IB_FLOW_ATTR_NORMAL) {
1910                 err = mlx4_ib_tunnel_steer_add(qp, flow_attr,
1911                                                &mflow->reg_id[i].id);
1912                 if (err)
1913                         goto err_create_flow;
1914
1915                 if (is_bonded) {
1916                         flow_attr->port = 2;
1917                         err = mlx4_ib_tunnel_steer_add(qp, flow_attr,
1918                                                        &mflow->reg_id[j].mirror);
1919                         flow_attr->port = 1;
1920                         if (err)
1921                                 goto err_create_flow;
1922                         j++;
1923                 }
1924                 /* function to create mirror rule */
1925                 i++;
1926         }
1927
1928         return &mflow->ibflow;
1929
1930 err_create_flow:
1931         while (i) {
1932                 (void)__mlx4_ib_destroy_flow(to_mdev(qp->device)->dev,
1933                                              mflow->reg_id[i].id);
1934                 i--;
1935         }
1936
1937         while (j) {
1938                 (void)__mlx4_ib_destroy_flow(to_mdev(qp->device)->dev,
1939                                              mflow->reg_id[j].mirror);
1940                 j--;
1941         }
1942 err_free:
1943         kfree(mflow);
1944         return ERR_PTR(err);
1945 }
1946
1947 static int mlx4_ib_destroy_flow(struct ib_flow *flow_id)
1948 {
1949         int err, ret = 0;
1950         int i = 0;
1951         struct mlx4_ib_dev *mdev = to_mdev(flow_id->qp->device);
1952         struct mlx4_ib_flow *mflow = to_mflow(flow_id);
1953
1954         while (i < ARRAY_SIZE(mflow->reg_id) && mflow->reg_id[i].id) {
1955                 err = __mlx4_ib_destroy_flow(mdev->dev, mflow->reg_id[i].id);
1956                 if (err)
1957                         ret = err;
1958                 if (mflow->reg_id[i].mirror) {
1959                         err = __mlx4_ib_destroy_flow(mdev->dev,
1960                                                      mflow->reg_id[i].mirror);
1961                         if (err)
1962                                 ret = err;
1963                 }
1964                 i++;
1965         }
1966
1967         kfree(mflow);
1968         return ret;
1969 }
1970
1971 static int mlx4_ib_mcg_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
1972 {
1973         int err;
1974         struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1975         struct mlx4_dev *dev = mdev->dev;
1976         struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1977         struct mlx4_ib_steering *ib_steering = NULL;
1978         enum mlx4_protocol prot = MLX4_PROT_IB_IPV6;
1979         struct mlx4_flow_reg_id reg_id;
1980
1981         if (mdev->dev->caps.steering_mode ==
1982             MLX4_STEERING_MODE_DEVICE_MANAGED) {
1983                 ib_steering = kmalloc(sizeof(*ib_steering), GFP_KERNEL);
1984                 if (!ib_steering)
1985                         return -ENOMEM;
1986         }
1987
1988         err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw, mqp->port,
1989                                     !!(mqp->flags &
1990                                        MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
1991                                     prot, &reg_id.id);
1992         if (err) {
1993                 pr_err("multicast attach op failed, err %d\n", err);
1994                 goto err_malloc;
1995         }
1996
1997         reg_id.mirror = 0;
1998         if (mlx4_is_bonded(dev)) {
1999                 err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw,
2000                                             (mqp->port == 1) ? 2 : 1,
2001                                             !!(mqp->flags &
2002                                             MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
2003                                             prot, &reg_id.mirror);
2004                 if (err)
2005                         goto err_add;
2006         }
2007
2008         err = add_gid_entry(ibqp, gid);
2009         if (err)
2010                 goto err_add;
2011
2012         if (ib_steering) {
2013                 memcpy(ib_steering->gid.raw, gid->raw, 16);
2014                 ib_steering->reg_id = reg_id;
2015                 mutex_lock(&mqp->mutex);
2016                 list_add(&ib_steering->list, &mqp->steering_rules);
2017                 mutex_unlock(&mqp->mutex);
2018         }
2019         return 0;
2020
2021 err_add:
2022         mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
2023                               prot, reg_id.id);
2024         if (reg_id.mirror)
2025                 mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
2026                                       prot, reg_id.mirror);
2027 err_malloc:
2028         kfree(ib_steering);
2029
2030         return err;
2031 }
2032
2033 static struct mlx4_ib_gid_entry *find_gid_entry(struct mlx4_ib_qp *qp, u8 *raw)
2034 {
2035         struct mlx4_ib_gid_entry *ge;
2036         struct mlx4_ib_gid_entry *tmp;
2037         struct mlx4_ib_gid_entry *ret = NULL;
2038
2039         list_for_each_entry_safe(ge, tmp, &qp->gid_list, list) {
2040                 if (!memcmp(raw, ge->gid.raw, 16)) {
2041                         ret = ge;
2042                         break;
2043                 }
2044         }
2045
2046         return ret;
2047 }
2048
2049 static int mlx4_ib_mcg_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
2050 {
2051         int err;
2052         struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
2053         struct mlx4_dev *dev = mdev->dev;
2054         struct mlx4_ib_qp *mqp = to_mqp(ibqp);
2055         struct net_device *ndev;
2056         struct mlx4_ib_gid_entry *ge;
2057         struct mlx4_flow_reg_id reg_id = {0, 0};
2058         enum mlx4_protocol prot =  MLX4_PROT_IB_IPV6;
2059
2060         if (mdev->dev->caps.steering_mode ==
2061             MLX4_STEERING_MODE_DEVICE_MANAGED) {
2062                 struct mlx4_ib_steering *ib_steering;
2063
2064                 mutex_lock(&mqp->mutex);
2065                 list_for_each_entry(ib_steering, &mqp->steering_rules, list) {
2066                         if (!memcmp(ib_steering->gid.raw, gid->raw, 16)) {
2067                                 list_del(&ib_steering->list);
2068                                 break;
2069                         }
2070                 }
2071                 mutex_unlock(&mqp->mutex);
2072                 if (&ib_steering->list == &mqp->steering_rules) {
2073                         pr_err("Couldn't find reg_id for mgid. Steering rule is left attached\n");
2074                         return -EINVAL;
2075                 }
2076                 reg_id = ib_steering->reg_id;
2077                 kfree(ib_steering);
2078         }
2079
2080         err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
2081                                     prot, reg_id.id);
2082         if (err)
2083                 return err;
2084
2085         if (mlx4_is_bonded(dev)) {
2086                 err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
2087                                             prot, reg_id.mirror);
2088                 if (err)
2089                         return err;
2090         }
2091
2092         mutex_lock(&mqp->mutex);
2093         ge = find_gid_entry(mqp, gid->raw);
2094         if (ge) {
2095                 spin_lock_bh(&mdev->iboe.lock);
2096                 ndev = ge->added ? mdev->iboe.netdevs[ge->port - 1] : NULL;
2097                 if (ndev)
2098                         dev_hold(ndev);
2099                 spin_unlock_bh(&mdev->iboe.lock);
2100                 if (ndev)
2101                         dev_put(ndev);
2102                 list_del(&ge->list);
2103                 kfree(ge);
2104         } else
2105                 pr_warn("could not find mgid entry\n");
2106
2107         mutex_unlock(&mqp->mutex);
2108
2109         return 0;
2110 }
2111
2112 static int init_node_data(struct mlx4_ib_dev *dev)
2113 {
2114         struct ib_smp *in_mad  = NULL;
2115         struct ib_smp *out_mad = NULL;
2116         int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
2117         int err = -ENOMEM;
2118
2119         in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
2120         out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
2121         if (!in_mad || !out_mad)
2122                 goto out;
2123
2124         init_query_mad(in_mad);
2125         in_mad->attr_id = IB_SMP_ATTR_NODE_DESC;
2126         if (mlx4_is_master(dev->dev))
2127                 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
2128
2129         err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
2130         if (err)
2131                 goto out;
2132
2133         memcpy(dev->ib_dev.node_desc, out_mad->data, IB_DEVICE_NODE_DESC_MAX);
2134
2135         in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
2136
2137         err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
2138         if (err)
2139                 goto out;
2140
2141         dev->dev->rev_id = be32_to_cpup((__be32 *) (out_mad->data + 32));
2142         memcpy(&dev->ib_dev.node_guid, out_mad->data + 12, 8);
2143
2144 out:
2145         kfree(in_mad);
2146         kfree(out_mad);
2147         return err;
2148 }
2149
2150 static ssize_t show_hca(struct device *device, struct device_attribute *attr,
2151                         char *buf)
2152 {
2153         struct mlx4_ib_dev *dev =
2154                 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
2155         return sprintf(buf, "MT%d\n", dev->dev->persist->pdev->device);
2156 }
2157
2158 static ssize_t show_rev(struct device *device, struct device_attribute *attr,
2159                         char *buf)
2160 {
2161         struct mlx4_ib_dev *dev =
2162                 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
2163         return sprintf(buf, "%x\n", dev->dev->rev_id);
2164 }
2165
2166 static ssize_t show_board(struct device *device, struct device_attribute *attr,
2167                           char *buf)
2168 {
2169         struct mlx4_ib_dev *dev =
2170                 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
2171         return sprintf(buf, "%.*s\n", MLX4_BOARD_ID_LEN,
2172                        dev->dev->board_id);
2173 }
2174
2175 static DEVICE_ATTR(hw_rev,   S_IRUGO, show_rev,    NULL);
2176 static DEVICE_ATTR(hca_type, S_IRUGO, show_hca,    NULL);
2177 static DEVICE_ATTR(board_id, S_IRUGO, show_board,  NULL);
2178
2179 static struct device_attribute *mlx4_class_attributes[] = {
2180         &dev_attr_hw_rev,
2181         &dev_attr_hca_type,
2182         &dev_attr_board_id
2183 };
2184
2185 struct diag_counter {
2186         const char *name;
2187         u32 offset;
2188 };
2189
2190 #define DIAG_COUNTER(_name, _offset)                    \
2191         { .name = #_name, .offset = _offset }
2192
2193 static const struct diag_counter diag_basic[] = {
2194         DIAG_COUNTER(rq_num_lle, 0x00),
2195         DIAG_COUNTER(sq_num_lle, 0x04),
2196         DIAG_COUNTER(rq_num_lqpoe, 0x08),
2197         DIAG_COUNTER(sq_num_lqpoe, 0x0C),
2198         DIAG_COUNTER(rq_num_lpe, 0x18),
2199         DIAG_COUNTER(sq_num_lpe, 0x1C),
2200         DIAG_COUNTER(rq_num_wrfe, 0x20),
2201         DIAG_COUNTER(sq_num_wrfe, 0x24),
2202         DIAG_COUNTER(sq_num_mwbe, 0x2C),
2203         DIAG_COUNTER(sq_num_bre, 0x34),
2204         DIAG_COUNTER(sq_num_rire, 0x44),
2205         DIAG_COUNTER(rq_num_rire, 0x48),
2206         DIAG_COUNTER(sq_num_rae, 0x4C),
2207         DIAG_COUNTER(rq_num_rae, 0x50),
2208         DIAG_COUNTER(sq_num_roe, 0x54),
2209         DIAG_COUNTER(sq_num_tree, 0x5C),
2210         DIAG_COUNTER(sq_num_rree, 0x64),
2211         DIAG_COUNTER(rq_num_rnr, 0x68),
2212         DIAG_COUNTER(sq_num_rnr, 0x6C),
2213         DIAG_COUNTER(rq_num_oos, 0x100),
2214         DIAG_COUNTER(sq_num_oos, 0x104),
2215 };
2216
2217 static const struct diag_counter diag_ext[] = {
2218         DIAG_COUNTER(rq_num_dup, 0x130),
2219         DIAG_COUNTER(sq_num_to, 0x134),
2220 };
2221
2222 static const struct diag_counter diag_device_only[] = {
2223         DIAG_COUNTER(num_cqovf, 0x1A0),
2224         DIAG_COUNTER(rq_num_udsdprd, 0x118),
2225 };
2226
2227 static struct rdma_hw_stats *mlx4_ib_alloc_hw_stats(struct ib_device *ibdev,
2228                                                     u8 port_num)
2229 {
2230         struct mlx4_ib_dev *dev = to_mdev(ibdev);
2231         struct mlx4_ib_diag_counters *diag = dev->diag_counters;
2232
2233         if (!diag[!!port_num].name)
2234                 return NULL;
2235
2236         return rdma_alloc_hw_stats_struct(diag[!!port_num].name,
2237                                           diag[!!port_num].num_counters,
2238                                           RDMA_HW_STATS_DEFAULT_LIFESPAN);
2239 }
2240
2241 static int mlx4_ib_get_hw_stats(struct ib_device *ibdev,
2242                                 struct rdma_hw_stats *stats,
2243                                 u8 port, int index)
2244 {
2245         struct mlx4_ib_dev *dev = to_mdev(ibdev);
2246         struct mlx4_ib_diag_counters *diag = dev->diag_counters;
2247         u32 hw_value[ARRAY_SIZE(diag_device_only) +
2248                 ARRAY_SIZE(diag_ext) + ARRAY_SIZE(diag_basic)] = {};
2249         int ret;
2250         int i;
2251
2252         ret = mlx4_query_diag_counters(dev->dev,
2253                                        MLX4_OP_MOD_QUERY_TRANSPORT_CI_ERRORS,
2254                                        diag[!!port].offset, hw_value,
2255                                        diag[!!port].num_counters, port);
2256
2257         if (ret)
2258                 return ret;
2259
2260         for (i = 0; i < diag[!!port].num_counters; i++)
2261                 stats->value[i] = hw_value[i];
2262
2263         return diag[!!port].num_counters;
2264 }
2265
2266 static int __mlx4_ib_alloc_diag_counters(struct mlx4_ib_dev *ibdev,
2267                                          const char ***name,
2268                                          u32 **offset,
2269                                          u32 *num,
2270                                          bool port)
2271 {
2272         u32 num_counters;
2273
2274         num_counters = ARRAY_SIZE(diag_basic);
2275
2276         if (ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DIAG_PER_PORT)
2277                 num_counters += ARRAY_SIZE(diag_ext);
2278
2279         if (!port)
2280                 num_counters += ARRAY_SIZE(diag_device_only);
2281
2282         *name = kcalloc(num_counters, sizeof(**name), GFP_KERNEL);
2283         if (!*name)
2284                 return -ENOMEM;
2285
2286         *offset = kcalloc(num_counters, sizeof(**offset), GFP_KERNEL);
2287         if (!*offset)
2288                 goto err_name;
2289
2290         *num = num_counters;
2291
2292         return 0;
2293
2294 err_name:
2295         kfree(*name);
2296         return -ENOMEM;
2297 }
2298
2299 static void mlx4_ib_fill_diag_counters(struct mlx4_ib_dev *ibdev,
2300                                        const char **name,
2301                                        u32 *offset,
2302                                        bool port)
2303 {
2304         int i;
2305         int j;
2306
2307         for (i = 0, j = 0; i < ARRAY_SIZE(diag_basic); i++, j++) {
2308                 name[i] = diag_basic[i].name;
2309                 offset[i] = diag_basic[i].offset;
2310         }
2311
2312         if (ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DIAG_PER_PORT) {
2313                 for (i = 0; i < ARRAY_SIZE(diag_ext); i++, j++) {
2314                         name[j] = diag_ext[i].name;
2315                         offset[j] = diag_ext[i].offset;
2316                 }
2317         }
2318
2319         if (!port) {
2320                 for (i = 0; i < ARRAY_SIZE(diag_device_only); i++, j++) {
2321                         name[j] = diag_device_only[i].name;
2322                         offset[j] = diag_device_only[i].offset;
2323                 }
2324         }
2325 }
2326
2327 static int mlx4_ib_alloc_diag_counters(struct mlx4_ib_dev *ibdev)
2328 {
2329         struct mlx4_ib_diag_counters *diag = ibdev->diag_counters;
2330         int i;
2331         int ret;
2332         bool per_port = !!(ibdev->dev->caps.flags2 &
2333                 MLX4_DEV_CAP_FLAG2_DIAG_PER_PORT);
2334
2335         if (mlx4_is_slave(ibdev->dev))
2336                 return 0;
2337
2338         for (i = 0; i < MLX4_DIAG_COUNTERS_TYPES; i++) {
2339                 /* i == 1 means we are building port counters */
2340                 if (i && !per_port)
2341                         continue;
2342
2343                 ret = __mlx4_ib_alloc_diag_counters(ibdev, &diag[i].name,
2344                                                     &diag[i].offset,
2345                                                     &diag[i].num_counters, i);
2346                 if (ret)
2347                         goto err_alloc;
2348
2349                 mlx4_ib_fill_diag_counters(ibdev, diag[i].name,
2350                                            diag[i].offset, i);
2351         }
2352
2353         ibdev->ib_dev.get_hw_stats      = mlx4_ib_get_hw_stats;
2354         ibdev->ib_dev.alloc_hw_stats    = mlx4_ib_alloc_hw_stats;
2355
2356         return 0;
2357
2358 err_alloc:
2359         if (i) {
2360                 kfree(diag[i - 1].name);
2361                 kfree(diag[i - 1].offset);
2362         }
2363
2364         return ret;
2365 }
2366
2367 static void mlx4_ib_diag_cleanup(struct mlx4_ib_dev *ibdev)
2368 {
2369         int i;
2370
2371         for (i = 0; i < MLX4_DIAG_COUNTERS_TYPES; i++) {
2372                 kfree(ibdev->diag_counters[i].offset);
2373                 kfree(ibdev->diag_counters[i].name);
2374         }
2375 }
2376
2377 #define MLX4_IB_INVALID_MAC     ((u64)-1)
2378 static void mlx4_ib_update_qps(struct mlx4_ib_dev *ibdev,
2379                                struct net_device *dev,
2380                                int port)
2381 {
2382         u64 new_smac = 0;
2383         u64 release_mac = MLX4_IB_INVALID_MAC;
2384         struct mlx4_ib_qp *qp;
2385
2386         read_lock(&dev_base_lock);
2387         new_smac = mlx4_mac_to_u64(dev->dev_addr);
2388         read_unlock(&dev_base_lock);
2389
2390         atomic64_set(&ibdev->iboe.mac[port - 1], new_smac);
2391
2392         /* no need for update QP1 and mac registration in non-SRIOV */
2393         if (!mlx4_is_mfunc(ibdev->dev))
2394                 return;
2395
2396         mutex_lock(&ibdev->qp1_proxy_lock[port - 1]);
2397         qp = ibdev->qp1_proxy[port - 1];
2398         if (qp) {
2399                 int new_smac_index;
2400                 u64 old_smac;
2401                 struct mlx4_update_qp_params update_params;
2402
2403                 mutex_lock(&qp->mutex);
2404                 old_smac = qp->pri.smac;
2405                 if (new_smac == old_smac)
2406                         goto unlock;
2407
2408                 new_smac_index = mlx4_register_mac(ibdev->dev, port, new_smac);
2409
2410                 if (new_smac_index < 0)
2411                         goto unlock;
2412
2413                 update_params.smac_index = new_smac_index;
2414                 if (mlx4_update_qp(ibdev->dev, qp->mqp.qpn, MLX4_UPDATE_QP_SMAC,
2415                                    &update_params)) {
2416                         release_mac = new_smac;
2417                         goto unlock;
2418                 }
2419                 /* if old port was zero, no mac was yet registered for this QP */
2420                 if (qp->pri.smac_port)
2421                         release_mac = old_smac;
2422                 qp->pri.smac = new_smac;
2423                 qp->pri.smac_port = port;
2424                 qp->pri.smac_index = new_smac_index;
2425         }
2426
2427 unlock:
2428         if (release_mac != MLX4_IB_INVALID_MAC)
2429                 mlx4_unregister_mac(ibdev->dev, port, release_mac);
2430         if (qp)
2431                 mutex_unlock(&qp->mutex);
2432         mutex_unlock(&ibdev->qp1_proxy_lock[port - 1]);
2433 }
2434
2435 static void mlx4_ib_scan_netdevs(struct mlx4_ib_dev *ibdev,
2436                                  struct net_device *dev,
2437                                  unsigned long event)
2438
2439 {
2440         struct mlx4_ib_iboe *iboe;
2441         int update_qps_port = -1;
2442         int port;
2443
2444         ASSERT_RTNL();
2445
2446         iboe = &ibdev->iboe;
2447
2448         spin_lock_bh(&iboe->lock);
2449         mlx4_foreach_ib_transport_port(port, ibdev->dev) {
2450
2451                 iboe->netdevs[port - 1] =
2452                         mlx4_get_protocol_dev(ibdev->dev, MLX4_PROT_ETH, port);
2453
2454                 if (dev == iboe->netdevs[port - 1] &&
2455                     (event == NETDEV_CHANGEADDR || event == NETDEV_REGISTER ||
2456                      event == NETDEV_UP || event == NETDEV_CHANGE))
2457                         update_qps_port = port;
2458
2459         }
2460         spin_unlock_bh(&iboe->lock);
2461
2462         if (update_qps_port > 0)
2463                 mlx4_ib_update_qps(ibdev, dev, update_qps_port);
2464 }
2465
2466 static int mlx4_ib_netdev_event(struct notifier_block *this,
2467                                 unsigned long event, void *ptr)
2468 {
2469         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
2470         struct mlx4_ib_dev *ibdev;
2471
2472         if (!net_eq(dev_net(dev), &init_net))
2473                 return NOTIFY_DONE;
2474
2475         ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb);
2476         mlx4_ib_scan_netdevs(ibdev, dev, event);
2477
2478         return NOTIFY_DONE;
2479 }
2480
2481 static void init_pkeys(struct mlx4_ib_dev *ibdev)
2482 {
2483         int port;
2484         int slave;
2485         int i;
2486
2487         if (mlx4_is_master(ibdev->dev)) {
2488                 for (slave = 0; slave <= ibdev->dev->persist->num_vfs;
2489                      ++slave) {
2490                         for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
2491                                 for (i = 0;
2492                                      i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
2493                                      ++i) {
2494                                         ibdev->pkeys.virt2phys_pkey[slave][port - 1][i] =
2495                                         /* master has the identity virt2phys pkey mapping */
2496                                                 (slave == mlx4_master_func_num(ibdev->dev) || !i) ? i :
2497                                                         ibdev->dev->phys_caps.pkey_phys_table_len[port] - 1;
2498                                         mlx4_sync_pkey_table(ibdev->dev, slave, port, i,
2499                                                              ibdev->pkeys.virt2phys_pkey[slave][port - 1][i]);
2500                                 }
2501                         }
2502                 }
2503                 /* initialize pkey cache */
2504                 for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
2505                         for (i = 0;
2506                              i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
2507                              ++i)
2508                                 ibdev->pkeys.phys_pkey_cache[port-1][i] =
2509                                         (i) ? 0 : 0xFFFF;
2510                 }
2511         }
2512 }
2513
2514 static void mlx4_ib_alloc_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
2515 {
2516         int i, j, eq = 0, total_eqs = 0;
2517
2518         ibdev->eq_table = kcalloc(dev->caps.num_comp_vectors,
2519                                   sizeof(ibdev->eq_table[0]), GFP_KERNEL);
2520         if (!ibdev->eq_table)
2521                 return;
2522
2523         for (i = 1; i <= dev->caps.num_ports; i++) {
2524                 for (j = 0; j < mlx4_get_eqs_per_port(dev, i);
2525                      j++, total_eqs++) {
2526                         if (i > 1 &&  mlx4_is_eq_shared(dev, total_eqs))
2527                                 continue;
2528                         ibdev->eq_table[eq] = total_eqs;
2529                         if (!mlx4_assign_eq(dev, i,
2530                                             &ibdev->eq_table[eq]))
2531                                 eq++;
2532                         else
2533                                 ibdev->eq_table[eq] = -1;
2534                 }
2535         }
2536
2537         for (i = eq; i < dev->caps.num_comp_vectors;
2538              ibdev->eq_table[i++] = -1)
2539                 ;
2540
2541         /* Advertise the new number of EQs to clients */
2542         ibdev->ib_dev.num_comp_vectors = eq;
2543 }
2544
2545 static void mlx4_ib_free_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
2546 {
2547         int i;
2548         int total_eqs = ibdev->ib_dev.num_comp_vectors;
2549
2550         /* no eqs were allocated */
2551         if (!ibdev->eq_table)
2552                 return;
2553
2554         /* Reset the advertised EQ number */
2555         ibdev->ib_dev.num_comp_vectors = 0;
2556
2557         for (i = 0; i < total_eqs; i++)
2558                 mlx4_release_eq(dev, ibdev->eq_table[i]);
2559
2560         kfree(ibdev->eq_table);
2561         ibdev->eq_table = NULL;
2562 }
2563
2564 static int mlx4_port_immutable(struct ib_device *ibdev, u8 port_num,
2565                                struct ib_port_immutable *immutable)
2566 {
2567         struct ib_port_attr attr;
2568         struct mlx4_ib_dev *mdev = to_mdev(ibdev);
2569         int err;
2570
2571         if (mlx4_ib_port_link_layer(ibdev, port_num) == IB_LINK_LAYER_INFINIBAND) {
2572                 immutable->core_cap_flags = RDMA_CORE_PORT_IBA_IB;
2573                 immutable->max_mad_size = IB_MGMT_MAD_SIZE;
2574         } else {
2575                 if (mdev->dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE)
2576                         immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE;
2577                 if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ROCE_V1_V2)
2578                         immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE |
2579                                 RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP;
2580                 immutable->core_cap_flags |= RDMA_CORE_PORT_RAW_PACKET;
2581                 if (immutable->core_cap_flags & (RDMA_CORE_PORT_IBA_ROCE |
2582                     RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP))
2583                         immutable->max_mad_size = IB_MGMT_MAD_SIZE;
2584         }
2585
2586         err = ib_query_port(ibdev, port_num, &attr);
2587         if (err)
2588                 return err;
2589
2590         immutable->pkey_tbl_len = attr.pkey_tbl_len;
2591         immutable->gid_tbl_len = attr.gid_tbl_len;
2592
2593         return 0;
2594 }
2595
2596 static void get_fw_ver_str(struct ib_device *device, char *str)
2597 {
2598         struct mlx4_ib_dev *dev =
2599                 container_of(device, struct mlx4_ib_dev, ib_dev);
2600         snprintf(str, IB_FW_VERSION_NAME_MAX, "%d.%d.%d",
2601                  (int) (dev->dev->caps.fw_ver >> 32),
2602                  (int) (dev->dev->caps.fw_ver >> 16) & 0xffff,
2603                  (int) dev->dev->caps.fw_ver & 0xffff);
2604 }
2605
2606 static void *mlx4_ib_add(struct mlx4_dev *dev)
2607 {
2608         struct mlx4_ib_dev *ibdev;
2609         int num_ports = 0;
2610         int i, j;
2611         int err;
2612         struct mlx4_ib_iboe *iboe;
2613         int ib_num_ports = 0;
2614         int num_req_counters;
2615         int allocated;
2616         u32 counter_index;
2617         struct counter_index *new_counter_index = NULL;
2618
2619         pr_info_once("%s", mlx4_ib_version);
2620
2621         num_ports = 0;
2622         mlx4_foreach_ib_transport_port(i, dev)
2623                 num_ports++;
2624
2625         /* No point in registering a device with no ports... */
2626         if (num_ports == 0)
2627                 return NULL;
2628
2629         ibdev = (struct mlx4_ib_dev *) ib_alloc_device(sizeof *ibdev);
2630         if (!ibdev) {
2631                 dev_err(&dev->persist->pdev->dev,
2632                         "Device struct alloc failed\n");
2633                 return NULL;
2634         }
2635
2636         iboe = &ibdev->iboe;
2637
2638         if (mlx4_pd_alloc(dev, &ibdev->priv_pdn))
2639                 goto err_dealloc;
2640
2641         if (mlx4_uar_alloc(dev, &ibdev->priv_uar))
2642                 goto err_pd;
2643
2644         ibdev->uar_map = ioremap((phys_addr_t) ibdev->priv_uar.pfn << PAGE_SHIFT,
2645                                  PAGE_SIZE);
2646         if (!ibdev->uar_map)
2647                 goto err_uar;
2648         MLX4_INIT_DOORBELL_LOCK(&ibdev->uar_lock);
2649
2650         ibdev->dev = dev;
2651         ibdev->bond_next_port   = 0;
2652
2653         strlcpy(ibdev->ib_dev.name, "mlx4_%d", IB_DEVICE_NAME_MAX);
2654         ibdev->ib_dev.owner             = THIS_MODULE;
2655         ibdev->ib_dev.node_type         = RDMA_NODE_IB_CA;
2656         ibdev->ib_dev.local_dma_lkey    = dev->caps.reserved_lkey;
2657         ibdev->num_ports                = num_ports;
2658         ibdev->ib_dev.phys_port_cnt     = mlx4_is_bonded(dev) ?
2659                                                 1 : ibdev->num_ports;
2660         ibdev->ib_dev.num_comp_vectors  = dev->caps.num_comp_vectors;
2661         ibdev->ib_dev.dev.parent        = &dev->persist->pdev->dev;
2662         ibdev->ib_dev.get_netdev        = mlx4_ib_get_netdev;
2663         ibdev->ib_dev.add_gid           = mlx4_ib_add_gid;
2664         ibdev->ib_dev.del_gid           = mlx4_ib_del_gid;
2665
2666         if (dev->caps.userspace_caps)
2667                 ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_ABI_VERSION;
2668         else
2669                 ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION;
2670
2671         ibdev->ib_dev.uverbs_cmd_mask   =
2672                 (1ull << IB_USER_VERBS_CMD_GET_CONTEXT)         |
2673                 (1ull << IB_USER_VERBS_CMD_QUERY_DEVICE)        |
2674                 (1ull << IB_USER_VERBS_CMD_QUERY_PORT)          |
2675                 (1ull << IB_USER_VERBS_CMD_ALLOC_PD)            |
2676                 (1ull << IB_USER_VERBS_CMD_DEALLOC_PD)          |
2677                 (1ull << IB_USER_VERBS_CMD_REG_MR)              |
2678                 (1ull << IB_USER_VERBS_CMD_REREG_MR)            |
2679                 (1ull << IB_USER_VERBS_CMD_DEREG_MR)            |
2680                 (1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL) |
2681                 (1ull << IB_USER_VERBS_CMD_CREATE_CQ)           |
2682                 (1ull << IB_USER_VERBS_CMD_RESIZE_CQ)           |
2683                 (1ull << IB_USER_VERBS_CMD_DESTROY_CQ)          |
2684                 (1ull << IB_USER_VERBS_CMD_CREATE_QP)           |
2685                 (1ull << IB_USER_VERBS_CMD_MODIFY_QP)           |
2686                 (1ull << IB_USER_VERBS_CMD_QUERY_QP)            |
2687                 (1ull << IB_USER_VERBS_CMD_DESTROY_QP)          |
2688                 (1ull << IB_USER_VERBS_CMD_ATTACH_MCAST)        |
2689                 (1ull << IB_USER_VERBS_CMD_DETACH_MCAST)        |
2690                 (1ull << IB_USER_VERBS_CMD_CREATE_SRQ)          |
2691                 (1ull << IB_USER_VERBS_CMD_MODIFY_SRQ)          |
2692                 (1ull << IB_USER_VERBS_CMD_QUERY_SRQ)           |
2693                 (1ull << IB_USER_VERBS_CMD_DESTROY_SRQ)         |
2694                 (1ull << IB_USER_VERBS_CMD_CREATE_XSRQ)         |
2695                 (1ull << IB_USER_VERBS_CMD_OPEN_QP);
2696
2697         ibdev->ib_dev.query_device      = mlx4_ib_query_device;
2698         ibdev->ib_dev.query_port        = mlx4_ib_query_port;
2699         ibdev->ib_dev.get_link_layer    = mlx4_ib_port_link_layer;
2700         ibdev->ib_dev.query_gid         = mlx4_ib_query_gid;
2701         ibdev->ib_dev.query_pkey        = mlx4_ib_query_pkey;
2702         ibdev->ib_dev.modify_device     = mlx4_ib_modify_device;
2703         ibdev->ib_dev.modify_port       = mlx4_ib_modify_port;
2704         ibdev->ib_dev.alloc_ucontext    = mlx4_ib_alloc_ucontext;
2705         ibdev->ib_dev.dealloc_ucontext  = mlx4_ib_dealloc_ucontext;
2706         ibdev->ib_dev.mmap              = mlx4_ib_mmap;
2707         ibdev->ib_dev.alloc_pd          = mlx4_ib_alloc_pd;
2708         ibdev->ib_dev.dealloc_pd        = mlx4_ib_dealloc_pd;
2709         ibdev->ib_dev.create_ah         = mlx4_ib_create_ah;
2710         ibdev->ib_dev.query_ah          = mlx4_ib_query_ah;
2711         ibdev->ib_dev.destroy_ah        = mlx4_ib_destroy_ah;
2712         ibdev->ib_dev.create_srq        = mlx4_ib_create_srq;
2713         ibdev->ib_dev.modify_srq        = mlx4_ib_modify_srq;
2714         ibdev->ib_dev.query_srq         = mlx4_ib_query_srq;
2715         ibdev->ib_dev.destroy_srq       = mlx4_ib_destroy_srq;
2716         ibdev->ib_dev.post_srq_recv     = mlx4_ib_post_srq_recv;
2717         ibdev->ib_dev.create_qp         = mlx4_ib_create_qp;
2718         ibdev->ib_dev.modify_qp         = mlx4_ib_modify_qp;
2719         ibdev->ib_dev.query_qp          = mlx4_ib_query_qp;
2720         ibdev->ib_dev.destroy_qp        = mlx4_ib_destroy_qp;
2721         ibdev->ib_dev.post_send         = mlx4_ib_post_send;
2722         ibdev->ib_dev.post_recv         = mlx4_ib_post_recv;
2723         ibdev->ib_dev.create_cq         = mlx4_ib_create_cq;
2724         ibdev->ib_dev.modify_cq         = mlx4_ib_modify_cq;
2725         ibdev->ib_dev.resize_cq         = mlx4_ib_resize_cq;
2726         ibdev->ib_dev.destroy_cq        = mlx4_ib_destroy_cq;
2727         ibdev->ib_dev.poll_cq           = mlx4_ib_poll_cq;
2728         ibdev->ib_dev.req_notify_cq     = mlx4_ib_arm_cq;
2729         ibdev->ib_dev.get_dma_mr        = mlx4_ib_get_dma_mr;
2730         ibdev->ib_dev.reg_user_mr       = mlx4_ib_reg_user_mr;
2731         ibdev->ib_dev.rereg_user_mr     = mlx4_ib_rereg_user_mr;
2732         ibdev->ib_dev.dereg_mr          = mlx4_ib_dereg_mr;
2733         ibdev->ib_dev.alloc_mr          = mlx4_ib_alloc_mr;
2734         ibdev->ib_dev.map_mr_sg         = mlx4_ib_map_mr_sg;
2735         ibdev->ib_dev.attach_mcast      = mlx4_ib_mcg_attach;
2736         ibdev->ib_dev.detach_mcast      = mlx4_ib_mcg_detach;
2737         ibdev->ib_dev.process_mad       = mlx4_ib_process_mad;
2738         ibdev->ib_dev.get_port_immutable = mlx4_port_immutable;
2739         ibdev->ib_dev.get_dev_fw_str    = get_fw_ver_str;
2740         ibdev->ib_dev.disassociate_ucontext = mlx4_ib_disassociate_ucontext;
2741
2742         if ((dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_RSS) &&
2743             ((mlx4_ib_port_link_layer(&ibdev->ib_dev, 1) ==
2744             IB_LINK_LAYER_ETHERNET) ||
2745             (mlx4_ib_port_link_layer(&ibdev->ib_dev, 2) ==
2746             IB_LINK_LAYER_ETHERNET))) {
2747                 ibdev->ib_dev.create_wq         = mlx4_ib_create_wq;
2748                 ibdev->ib_dev.modify_wq         = mlx4_ib_modify_wq;
2749                 ibdev->ib_dev.destroy_wq        = mlx4_ib_destroy_wq;
2750                 ibdev->ib_dev.create_rwq_ind_table  =
2751                         mlx4_ib_create_rwq_ind_table;
2752                 ibdev->ib_dev.destroy_rwq_ind_table =
2753                         mlx4_ib_destroy_rwq_ind_table;
2754                 ibdev->ib_dev.uverbs_ex_cmd_mask |=
2755                         (1ull << IB_USER_VERBS_EX_CMD_CREATE_WQ)          |
2756                         (1ull << IB_USER_VERBS_EX_CMD_MODIFY_WQ)          |
2757                         (1ull << IB_USER_VERBS_EX_CMD_DESTROY_WQ)         |
2758                         (1ull << IB_USER_VERBS_EX_CMD_CREATE_RWQ_IND_TBL) |
2759                         (1ull << IB_USER_VERBS_EX_CMD_DESTROY_RWQ_IND_TBL);
2760         }
2761
2762         if (!mlx4_is_slave(ibdev->dev)) {
2763                 ibdev->ib_dev.alloc_fmr         = mlx4_ib_fmr_alloc;
2764                 ibdev->ib_dev.map_phys_fmr      = mlx4_ib_map_phys_fmr;
2765                 ibdev->ib_dev.unmap_fmr         = mlx4_ib_unmap_fmr;
2766                 ibdev->ib_dev.dealloc_fmr       = mlx4_ib_fmr_dealloc;
2767         }
2768
2769         if (dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW ||
2770             dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
2771                 ibdev->ib_dev.alloc_mw = mlx4_ib_alloc_mw;
2772                 ibdev->ib_dev.dealloc_mw = mlx4_ib_dealloc_mw;
2773
2774                 ibdev->ib_dev.uverbs_cmd_mask |=
2775                         (1ull << IB_USER_VERBS_CMD_ALLOC_MW) |
2776                         (1ull << IB_USER_VERBS_CMD_DEALLOC_MW);
2777         }
2778
2779         if (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) {
2780                 ibdev->ib_dev.alloc_xrcd = mlx4_ib_alloc_xrcd;
2781                 ibdev->ib_dev.dealloc_xrcd = mlx4_ib_dealloc_xrcd;
2782                 ibdev->ib_dev.uverbs_cmd_mask |=
2783                         (1ull << IB_USER_VERBS_CMD_OPEN_XRCD) |
2784                         (1ull << IB_USER_VERBS_CMD_CLOSE_XRCD);
2785         }
2786
2787         if (check_flow_steering_support(dev)) {
2788                 ibdev->steering_support = MLX4_STEERING_MODE_DEVICE_MANAGED;
2789                 ibdev->ib_dev.create_flow       = mlx4_ib_create_flow;
2790                 ibdev->ib_dev.destroy_flow      = mlx4_ib_destroy_flow;
2791
2792                 ibdev->ib_dev.uverbs_ex_cmd_mask        |=
2793                         (1ull << IB_USER_VERBS_EX_CMD_CREATE_FLOW) |
2794                         (1ull << IB_USER_VERBS_EX_CMD_DESTROY_FLOW);
2795         }
2796
2797         ibdev->ib_dev.uverbs_ex_cmd_mask |=
2798                 (1ull << IB_USER_VERBS_EX_CMD_QUERY_DEVICE) |
2799                 (1ull << IB_USER_VERBS_EX_CMD_CREATE_CQ) |
2800                 (1ull << IB_USER_VERBS_EX_CMD_CREATE_QP);
2801
2802         mlx4_ib_alloc_eqs(dev, ibdev);
2803
2804         spin_lock_init(&iboe->lock);
2805
2806         if (init_node_data(ibdev))
2807                 goto err_map;
2808         mlx4_init_sl2vl_tbl(ibdev);
2809
2810         for (i = 0; i < ibdev->num_ports; ++i) {
2811                 mutex_init(&ibdev->counters_table[i].mutex);
2812                 INIT_LIST_HEAD(&ibdev->counters_table[i].counters_list);
2813         }
2814
2815         num_req_counters = mlx4_is_bonded(dev) ? 1 : ibdev->num_ports;
2816         for (i = 0; i < num_req_counters; ++i) {
2817                 mutex_init(&ibdev->qp1_proxy_lock[i]);
2818                 allocated = 0;
2819                 if (mlx4_ib_port_link_layer(&ibdev->ib_dev, i + 1) ==
2820                                                 IB_LINK_LAYER_ETHERNET) {
2821                         err = mlx4_counter_alloc(ibdev->dev, &counter_index,
2822                                                  MLX4_RES_USAGE_DRIVER);
2823                         /* if failed to allocate a new counter, use default */
2824                         if (err)
2825                                 counter_index =
2826                                         mlx4_get_default_counter_index(dev,
2827                                                                        i + 1);
2828                         else
2829                                 allocated = 1;
2830                 } else { /* IB_LINK_LAYER_INFINIBAND use the default counter */
2831                         counter_index = mlx4_get_default_counter_index(dev,
2832                                                                        i + 1);
2833                 }
2834                 new_counter_index = kmalloc(sizeof(*new_counter_index),
2835                                             GFP_KERNEL);
2836                 if (!new_counter_index) {
2837                         if (allocated)
2838                                 mlx4_counter_free(ibdev->dev, counter_index);
2839                         goto err_counter;
2840                 }
2841                 new_counter_index->index = counter_index;
2842                 new_counter_index->allocated = allocated;
2843                 list_add_tail(&new_counter_index->list,
2844                               &ibdev->counters_table[i].counters_list);
2845                 ibdev->counters_table[i].default_counter = counter_index;
2846                 pr_info("counter index %d for port %d allocated %d\n",
2847                         counter_index, i + 1, allocated);
2848         }
2849         if (mlx4_is_bonded(dev))
2850                 for (i = 1; i < ibdev->num_ports ; ++i) {
2851                         new_counter_index =
2852                                         kmalloc(sizeof(struct counter_index),
2853                                                 GFP_KERNEL);
2854                         if (!new_counter_index)
2855                                 goto err_counter;
2856                         new_counter_index->index = counter_index;
2857                         new_counter_index->allocated = 0;
2858                         list_add_tail(&new_counter_index->list,
2859                                       &ibdev->counters_table[i].counters_list);
2860                         ibdev->counters_table[i].default_counter =
2861                                                                 counter_index;
2862                 }
2863
2864         mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
2865                 ib_num_ports++;
2866
2867         spin_lock_init(&ibdev->sm_lock);
2868         mutex_init(&ibdev->cap_mask_mutex);
2869         INIT_LIST_HEAD(&ibdev->qp_list);
2870         spin_lock_init(&ibdev->reset_flow_resource_lock);
2871
2872         if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED &&
2873             ib_num_ports) {
2874                 ibdev->steer_qpn_count = MLX4_IB_UC_MAX_NUM_QPS;
2875                 err = mlx4_qp_reserve_range(dev, ibdev->steer_qpn_count,
2876                                             MLX4_IB_UC_STEER_QPN_ALIGN,
2877                                             &ibdev->steer_qpn_base, 0,
2878                                             MLX4_RES_USAGE_DRIVER);
2879                 if (err)
2880                         goto err_counter;
2881
2882                 ibdev->ib_uc_qpns_bitmap =
2883                         kmalloc(BITS_TO_LONGS(ibdev->steer_qpn_count) *
2884                                 sizeof(long),
2885                                 GFP_KERNEL);
2886                 if (!ibdev->ib_uc_qpns_bitmap)
2887                         goto err_steer_qp_release;
2888
2889                 if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DMFS_IPOIB) {
2890                         bitmap_zero(ibdev->ib_uc_qpns_bitmap,
2891                                     ibdev->steer_qpn_count);
2892                         err = mlx4_FLOW_STEERING_IB_UC_QP_RANGE(
2893                                         dev, ibdev->steer_qpn_base,
2894                                         ibdev->steer_qpn_base +
2895                                         ibdev->steer_qpn_count - 1);
2896                         if (err)
2897                                 goto err_steer_free_bitmap;
2898                 } else {
2899                         bitmap_fill(ibdev->ib_uc_qpns_bitmap,
2900                                     ibdev->steer_qpn_count);
2901                 }
2902         }
2903
2904         for (j = 1; j <= ibdev->dev->caps.num_ports; j++)
2905                 atomic64_set(&iboe->mac[j - 1], ibdev->dev->caps.def_mac[j]);
2906
2907         if (mlx4_ib_alloc_diag_counters(ibdev))
2908                 goto err_steer_free_bitmap;
2909
2910         if (ib_register_device(&ibdev->ib_dev, NULL))
2911                 goto err_diag_counters;
2912
2913         if (mlx4_ib_mad_init(ibdev))
2914                 goto err_reg;
2915
2916         if (mlx4_ib_init_sriov(ibdev))
2917                 goto err_mad;
2918
2919         if (!iboe->nb.notifier_call) {
2920                 iboe->nb.notifier_call = mlx4_ib_netdev_event;
2921                 err = register_netdevice_notifier(&iboe->nb);
2922                 if (err) {
2923                         iboe->nb.notifier_call = NULL;
2924                         goto err_notif;
2925                 }
2926         }
2927         if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ROCE_V1_V2) {
2928                 err = mlx4_config_roce_v2_port(dev, ROCE_V2_UDP_DPORT);
2929                 if (err)
2930                         goto err_notif;
2931         }
2932
2933         for (j = 0; j < ARRAY_SIZE(mlx4_class_attributes); ++j) {
2934                 if (device_create_file(&ibdev->ib_dev.dev,
2935                                        mlx4_class_attributes[j]))
2936                         goto err_notif;
2937         }
2938
2939         ibdev->ib_active = true;
2940         mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
2941                 devlink_port_type_ib_set(mlx4_get_devlink_port(dev, i),
2942                                          &ibdev->ib_dev);
2943
2944         if (mlx4_is_mfunc(ibdev->dev))
2945                 init_pkeys(ibdev);
2946
2947         /* create paravirt contexts for any VFs which are active */
2948         if (mlx4_is_master(ibdev->dev)) {
2949                 for (j = 0; j < MLX4_MFUNC_MAX; j++) {
2950                         if (j == mlx4_master_func_num(ibdev->dev))
2951                                 continue;
2952                         if (mlx4_is_slave_active(ibdev->dev, j))
2953                                 do_slave_init(ibdev, j, 1);
2954                 }
2955         }
2956         return ibdev;
2957
2958 err_notif:
2959         if (ibdev->iboe.nb.notifier_call) {
2960                 if (unregister_netdevice_notifier(&ibdev->iboe.nb))
2961                         pr_warn("failure unregistering notifier\n");
2962                 ibdev->iboe.nb.notifier_call = NULL;
2963         }
2964         flush_workqueue(wq);
2965
2966         mlx4_ib_close_sriov(ibdev);
2967
2968 err_mad:
2969         mlx4_ib_mad_cleanup(ibdev);
2970
2971 err_reg:
2972         ib_unregister_device(&ibdev->ib_dev);
2973
2974 err_diag_counters:
2975         mlx4_ib_diag_cleanup(ibdev);
2976
2977 err_steer_free_bitmap:
2978         kfree(ibdev->ib_uc_qpns_bitmap);
2979
2980 err_steer_qp_release:
2981         mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
2982                               ibdev->steer_qpn_count);
2983 err_counter:
2984         for (i = 0; i < ibdev->num_ports; ++i)
2985                 mlx4_ib_delete_counters_table(ibdev, &ibdev->counters_table[i]);
2986
2987 err_map:
2988         mlx4_ib_free_eqs(dev, ibdev);
2989         iounmap(ibdev->uar_map);
2990
2991 err_uar:
2992         mlx4_uar_free(dev, &ibdev->priv_uar);
2993
2994 err_pd:
2995         mlx4_pd_free(dev, ibdev->priv_pdn);
2996
2997 err_dealloc:
2998         ib_dealloc_device(&ibdev->ib_dev);
2999
3000         return NULL;
3001 }
3002
3003 int mlx4_ib_steer_qp_alloc(struct mlx4_ib_dev *dev, int count, int *qpn)
3004 {
3005         int offset;
3006
3007         WARN_ON(!dev->ib_uc_qpns_bitmap);
3008
3009         offset = bitmap_find_free_region(dev->ib_uc_qpns_bitmap,
3010                                          dev->steer_qpn_count,
3011                                          get_count_order(count));
3012         if (offset < 0)
3013                 return offset;
3014
3015         *qpn = dev->steer_qpn_base + offset;
3016         return 0;
3017 }
3018
3019 void mlx4_ib_steer_qp_free(struct mlx4_ib_dev *dev, u32 qpn, int count)
3020 {
3021         if (!qpn ||
3022             dev->steering_support != MLX4_STEERING_MODE_DEVICE_MANAGED)
3023                 return;
3024
3025         BUG_ON(qpn < dev->steer_qpn_base);
3026
3027         bitmap_release_region(dev->ib_uc_qpns_bitmap,
3028                               qpn - dev->steer_qpn_base,
3029                               get_count_order(count));
3030 }
3031
3032 int mlx4_ib_steer_qp_reg(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
3033                          int is_attach)
3034 {
3035         int err;
3036         size_t flow_size;
3037         struct ib_flow_attr *flow = NULL;
3038         struct ib_flow_spec_ib *ib_spec;
3039
3040         if (is_attach) {
3041                 flow_size = sizeof(struct ib_flow_attr) +
3042                             sizeof(struct ib_flow_spec_ib);
3043                 flow = kzalloc(flow_size, GFP_KERNEL);
3044                 if (!flow)
3045                         return -ENOMEM;
3046                 flow->port = mqp->port;
3047                 flow->num_of_specs = 1;
3048                 flow->size = flow_size;
3049                 ib_spec = (struct ib_flow_spec_ib *)(flow + 1);
3050                 ib_spec->type = IB_FLOW_SPEC_IB;
3051                 ib_spec->size = sizeof(struct ib_flow_spec_ib);
3052                 /* Add an empty rule for IB L2 */
3053                 memset(&ib_spec->mask, 0, sizeof(ib_spec->mask));
3054
3055                 err = __mlx4_ib_create_flow(&mqp->ibqp, flow,
3056                                             IB_FLOW_DOMAIN_NIC,
3057                                             MLX4_FS_REGULAR,
3058                                             &mqp->reg_id);
3059         } else {
3060                 err = __mlx4_ib_destroy_flow(mdev->dev, mqp->reg_id);
3061         }
3062         kfree(flow);
3063         return err;
3064 }
3065
3066 static void mlx4_ib_remove(struct mlx4_dev *dev, void *ibdev_ptr)
3067 {
3068         struct mlx4_ib_dev *ibdev = ibdev_ptr;
3069         int p;
3070         int i;
3071
3072         mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
3073                 devlink_port_type_clear(mlx4_get_devlink_port(dev, i));
3074         ibdev->ib_active = false;
3075         flush_workqueue(wq);
3076
3077         if (ibdev->iboe.nb.notifier_call) {
3078                 if (unregister_netdevice_notifier(&ibdev->iboe.nb))
3079                         pr_warn("failure unregistering notifier\n");
3080                 ibdev->iboe.nb.notifier_call = NULL;
3081         }
3082
3083         mlx4_ib_close_sriov(ibdev);
3084         mlx4_ib_mad_cleanup(ibdev);
3085         ib_unregister_device(&ibdev->ib_dev);
3086         mlx4_ib_diag_cleanup(ibdev);
3087
3088         mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
3089                               ibdev->steer_qpn_count);
3090         kfree(ibdev->ib_uc_qpns_bitmap);
3091
3092         iounmap(ibdev->uar_map);
3093         for (p = 0; p < ibdev->num_ports; ++p)
3094                 mlx4_ib_delete_counters_table(ibdev, &ibdev->counters_table[p]);
3095
3096         mlx4_foreach_port(p, dev, MLX4_PORT_TYPE_IB)
3097                 mlx4_CLOSE_PORT(dev, p);
3098
3099         mlx4_ib_free_eqs(dev, ibdev);
3100
3101         mlx4_uar_free(dev, &ibdev->priv_uar);
3102         mlx4_pd_free(dev, ibdev->priv_pdn);
3103         ib_dealloc_device(&ibdev->ib_dev);
3104 }
3105
3106 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init)
3107 {
3108         struct mlx4_ib_demux_work **dm = NULL;
3109         struct mlx4_dev *dev = ibdev->dev;
3110         int i;
3111         unsigned long flags;
3112         struct mlx4_active_ports actv_ports;
3113         unsigned int ports;
3114         unsigned int first_port;
3115
3116         if (!mlx4_is_master(dev))
3117                 return;
3118
3119         actv_ports = mlx4_get_active_ports(dev, slave);
3120         ports = bitmap_weight(actv_ports.ports, dev->caps.num_ports);
3121         first_port = find_first_bit(actv_ports.ports, dev->caps.num_ports);
3122
3123         dm = kcalloc(ports, sizeof(*dm), GFP_ATOMIC);
3124         if (!dm)
3125                 return;
3126
3127         for (i = 0; i < ports; i++) {
3128                 dm[i] = kmalloc(sizeof (struct mlx4_ib_demux_work), GFP_ATOMIC);
3129                 if (!dm[i]) {
3130                         while (--i >= 0)
3131                                 kfree(dm[i]);
3132                         goto out;
3133                 }
3134                 INIT_WORK(&dm[i]->work, mlx4_ib_tunnels_update_work);
3135                 dm[i]->port = first_port + i + 1;
3136                 dm[i]->slave = slave;
3137                 dm[i]->do_init = do_init;
3138                 dm[i]->dev = ibdev;
3139         }
3140         /* initialize or tear down tunnel QPs for the slave */
3141         spin_lock_irqsave(&ibdev->sriov.going_down_lock, flags);
3142         if (!ibdev->sriov.is_going_down) {
3143                 for (i = 0; i < ports; i++)
3144                         queue_work(ibdev->sriov.demux[i].ud_wq, &dm[i]->work);
3145                 spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
3146         } else {
3147                 spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
3148                 for (i = 0; i < ports; i++)
3149                         kfree(dm[i]);
3150         }
3151 out:
3152         kfree(dm);
3153         return;
3154 }
3155
3156 static void mlx4_ib_handle_catas_error(struct mlx4_ib_dev *ibdev)
3157 {
3158         struct mlx4_ib_qp *mqp;
3159         unsigned long flags_qp;
3160         unsigned long flags_cq;
3161         struct mlx4_ib_cq *send_mcq, *recv_mcq;
3162         struct list_head    cq_notify_list;
3163         struct mlx4_cq *mcq;
3164         unsigned long flags;
3165
3166         pr_warn("mlx4_ib_handle_catas_error was started\n");
3167         INIT_LIST_HEAD(&cq_notify_list);
3168
3169         /* Go over qp list reside on that ibdev, sync with create/destroy qp.*/
3170         spin_lock_irqsave(&ibdev->reset_flow_resource_lock, flags);
3171
3172         list_for_each_entry(mqp, &ibdev->qp_list, qps_list) {
3173                 spin_lock_irqsave(&mqp->sq.lock, flags_qp);
3174                 if (mqp->sq.tail != mqp->sq.head) {
3175                         send_mcq = to_mcq(mqp->ibqp.send_cq);
3176                         spin_lock_irqsave(&send_mcq->lock, flags_cq);
3177                         if (send_mcq->mcq.comp &&
3178                             mqp->ibqp.send_cq->comp_handler) {
3179                                 if (!send_mcq->mcq.reset_notify_added) {
3180                                         send_mcq->mcq.reset_notify_added = 1;
3181                                         list_add_tail(&send_mcq->mcq.reset_notify,
3182                                                       &cq_notify_list);
3183                                 }
3184                         }
3185                         spin_unlock_irqrestore(&send_mcq->lock, flags_cq);
3186                 }
3187                 spin_unlock_irqrestore(&mqp->sq.lock, flags_qp);
3188                 /* Now, handle the QP's receive queue */
3189                 spin_lock_irqsave(&mqp->rq.lock, flags_qp);
3190                 /* no handling is needed for SRQ */
3191                 if (!mqp->ibqp.srq) {
3192                         if (mqp->rq.tail != mqp->rq.head) {
3193                                 recv_mcq = to_mcq(mqp->ibqp.recv_cq);
3194                                 spin_lock_irqsave(&recv_mcq->lock, flags_cq);
3195                                 if (recv_mcq->mcq.comp &&
3196                                     mqp->ibqp.recv_cq->comp_handler) {
3197                                         if (!recv_mcq->mcq.reset_notify_added) {
3198                                                 recv_mcq->mcq.reset_notify_added = 1;
3199                                                 list_add_tail(&recv_mcq->mcq.reset_notify,
3200                                                               &cq_notify_list);
3201                                         }
3202                                 }
3203                                 spin_unlock_irqrestore(&recv_mcq->lock,
3204                                                        flags_cq);
3205                         }
3206                 }
3207                 spin_unlock_irqrestore(&mqp->rq.lock, flags_qp);
3208         }
3209
3210         list_for_each_entry(mcq, &cq_notify_list, reset_notify) {
3211                 mcq->comp(mcq);
3212         }
3213         spin_unlock_irqrestore(&ibdev->reset_flow_resource_lock, flags);
3214         pr_warn("mlx4_ib_handle_catas_error ended\n");
3215 }
3216
3217 static void handle_bonded_port_state_event(struct work_struct *work)
3218 {
3219         struct ib_event_work *ew =
3220                 container_of(work, struct ib_event_work, work);
3221         struct mlx4_ib_dev *ibdev = ew->ib_dev;
3222         enum ib_port_state bonded_port_state = IB_PORT_NOP;
3223         int i;
3224         struct ib_event ibev;
3225
3226         kfree(ew);
3227         spin_lock_bh(&ibdev->iboe.lock);
3228         for (i = 0; i < MLX4_MAX_PORTS; ++i) {
3229                 struct net_device *curr_netdev = ibdev->iboe.netdevs[i];
3230                 enum ib_port_state curr_port_state;
3231
3232                 if (!curr_netdev)
3233                         continue;
3234
3235                 curr_port_state =
3236                         (netif_running(curr_netdev) &&
3237                          netif_carrier_ok(curr_netdev)) ?
3238                         IB_PORT_ACTIVE : IB_PORT_DOWN;
3239
3240                 bonded_port_state = (bonded_port_state != IB_PORT_ACTIVE) ?
3241                         curr_port_state : IB_PORT_ACTIVE;
3242         }
3243         spin_unlock_bh(&ibdev->iboe.lock);
3244
3245         ibev.device = &ibdev->ib_dev;
3246         ibev.element.port_num = 1;
3247         ibev.event = (bonded_port_state == IB_PORT_ACTIVE) ?
3248                 IB_EVENT_PORT_ACTIVE : IB_EVENT_PORT_ERR;
3249
3250         ib_dispatch_event(&ibev);
3251 }
3252
3253 void mlx4_ib_sl2vl_update(struct mlx4_ib_dev *mdev, int port)
3254 {
3255         u64 sl2vl;
3256         int err;
3257
3258         err = mlx4_ib_query_sl2vl(&mdev->ib_dev, port, &sl2vl);
3259         if (err) {
3260                 pr_err("Unable to get current sl to vl mapping for port %d.  Using all zeroes (%d)\n",
3261                        port, err);
3262                 sl2vl = 0;
3263         }
3264         atomic64_set(&mdev->sl2vl[port - 1], sl2vl);
3265 }
3266
3267 static void ib_sl2vl_update_work(struct work_struct *work)
3268 {
3269         struct ib_event_work *ew = container_of(work, struct ib_event_work, work);
3270         struct mlx4_ib_dev *mdev = ew->ib_dev;
3271         int port = ew->port;
3272
3273         mlx4_ib_sl2vl_update(mdev, port);
3274
3275         kfree(ew);
3276 }
3277
3278 void mlx4_sched_ib_sl2vl_update_work(struct mlx4_ib_dev *ibdev,
3279                                      int port)
3280 {
3281         struct ib_event_work *ew;
3282
3283         ew = kmalloc(sizeof(*ew), GFP_ATOMIC);
3284         if (ew) {
3285                 INIT_WORK(&ew->work, ib_sl2vl_update_work);
3286                 ew->port = port;
3287                 ew->ib_dev = ibdev;
3288                 queue_work(wq, &ew->work);
3289         }
3290 }
3291
3292 static void mlx4_ib_event(struct mlx4_dev *dev, void *ibdev_ptr,
3293                           enum mlx4_dev_event event, unsigned long param)
3294 {
3295         struct ib_event ibev;
3296         struct mlx4_ib_dev *ibdev = to_mdev((struct ib_device *) ibdev_ptr);
3297         struct mlx4_eqe *eqe = NULL;
3298         struct ib_event_work *ew;
3299         int p = 0;
3300
3301         if (mlx4_is_bonded(dev) &&
3302             ((event == MLX4_DEV_EVENT_PORT_UP) ||
3303             (event == MLX4_DEV_EVENT_PORT_DOWN))) {
3304                 ew = kmalloc(sizeof(*ew), GFP_ATOMIC);
3305                 if (!ew)
3306                         return;
3307                 INIT_WORK(&ew->work, handle_bonded_port_state_event);
3308                 ew->ib_dev = ibdev;
3309                 queue_work(wq, &ew->work);
3310                 return;
3311         }
3312
3313         if (event == MLX4_DEV_EVENT_PORT_MGMT_CHANGE)
3314                 eqe = (struct mlx4_eqe *)param;
3315         else
3316                 p = (int) param;
3317
3318         switch (event) {
3319         case MLX4_DEV_EVENT_PORT_UP:
3320                 if (p > ibdev->num_ports)
3321                         return;
3322                 if (!mlx4_is_slave(dev) &&
3323                     rdma_port_get_link_layer(&ibdev->ib_dev, p) ==
3324                         IB_LINK_LAYER_INFINIBAND) {
3325                         if (mlx4_is_master(dev))
3326                                 mlx4_ib_invalidate_all_guid_record(ibdev, p);
3327                         if (ibdev->dev->flags & MLX4_FLAG_SECURE_HOST &&
3328                             !(ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_SL_TO_VL_CHANGE_EVENT))
3329                                 mlx4_sched_ib_sl2vl_update_work(ibdev, p);
3330                 }
3331                 ibev.event = IB_EVENT_PORT_ACTIVE;
3332                 break;
3333
3334         case MLX4_DEV_EVENT_PORT_DOWN:
3335                 if (p > ibdev->num_ports)
3336                         return;
3337                 ibev.event = IB_EVENT_PORT_ERR;
3338                 break;
3339
3340         case MLX4_DEV_EVENT_CATASTROPHIC_ERROR:
3341                 ibdev->ib_active = false;
3342                 ibev.event = IB_EVENT_DEVICE_FATAL;
3343                 mlx4_ib_handle_catas_error(ibdev);
3344                 break;
3345
3346         case MLX4_DEV_EVENT_PORT_MGMT_CHANGE:
3347                 ew = kmalloc(sizeof *ew, GFP_ATOMIC);
3348                 if (!ew)
3349                         break;
3350
3351                 INIT_WORK(&ew->work, handle_port_mgmt_change_event);
3352                 memcpy(&ew->ib_eqe, eqe, sizeof *eqe);
3353                 ew->ib_dev = ibdev;
3354                 /* need to queue only for port owner, which uses GEN_EQE */
3355                 if (mlx4_is_master(dev))
3356                         queue_work(wq, &ew->work);
3357                 else
3358                         handle_port_mgmt_change_event(&ew->work);
3359                 return;
3360
3361         case MLX4_DEV_EVENT_SLAVE_INIT:
3362                 /* here, p is the slave id */
3363                 do_slave_init(ibdev, p, 1);
3364                 if (mlx4_is_master(dev)) {
3365                         int i;
3366
3367                         for (i = 1; i <= ibdev->num_ports; i++) {
3368                                 if (rdma_port_get_link_layer(&ibdev->ib_dev, i)
3369                                         == IB_LINK_LAYER_INFINIBAND)
3370                                         mlx4_ib_slave_alias_guid_event(ibdev,
3371                                                                        p, i,
3372                                                                        1);
3373                         }
3374                 }
3375                 return;
3376
3377         case MLX4_DEV_EVENT_SLAVE_SHUTDOWN:
3378                 if (mlx4_is_master(dev)) {
3379                         int i;
3380
3381                         for (i = 1; i <= ibdev->num_ports; i++) {
3382                                 if (rdma_port_get_link_layer(&ibdev->ib_dev, i)
3383                                         == IB_LINK_LAYER_INFINIBAND)
3384                                         mlx4_ib_slave_alias_guid_event(ibdev,
3385                                                                        p, i,
3386                                                                        0);
3387                         }
3388                 }
3389                 /* here, p is the slave id */
3390                 do_slave_init(ibdev, p, 0);
3391                 return;
3392
3393         default:
3394                 return;
3395         }
3396
3397         ibev.device           = ibdev_ptr;
3398         ibev.element.port_num = mlx4_is_bonded(ibdev->dev) ? 1 : (u8)p;
3399
3400         ib_dispatch_event(&ibev);
3401 }
3402
3403 static struct mlx4_interface mlx4_ib_interface = {
3404         .add            = mlx4_ib_add,
3405         .remove         = mlx4_ib_remove,
3406         .event          = mlx4_ib_event,
3407         .protocol       = MLX4_PROT_IB_IPV6,
3408         .flags          = MLX4_INTFF_BONDING
3409 };
3410
3411 static int __init mlx4_ib_init(void)
3412 {
3413         int err;
3414
3415         wq = alloc_ordered_workqueue("mlx4_ib", WQ_MEM_RECLAIM);
3416         if (!wq)
3417                 return -ENOMEM;
3418
3419         err = mlx4_ib_mcg_init();
3420         if (err)
3421                 goto clean_wq;
3422
3423         err = mlx4_register_interface(&mlx4_ib_interface);
3424         if (err)
3425                 goto clean_mcg;
3426
3427         return 0;
3428
3429 clean_mcg:
3430         mlx4_ib_mcg_destroy();
3431
3432 clean_wq:
3433         destroy_workqueue(wq);
3434         return err;
3435 }
3436
3437 static void __exit mlx4_ib_cleanup(void)
3438 {
3439         mlx4_unregister_interface(&mlx4_ib_interface);
3440         mlx4_ib_mcg_destroy();
3441         destroy_workqueue(wq);
3442 }
3443
3444 module_init(mlx4_ib_init);
3445 module_exit(mlx4_ib_cleanup);