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 <net/ipv6.h>
  43 #include <net/addrconf.h>
  44
  45 #include <rdma/ib_smi.h>
  46 #include <rdma/ib_user_verbs.h>
  47 #include <rdma/ib_addr.h>
  48
  49 #include <linux/mlx4/driver.h>
  50 #include <linux/mlx4/cmd.h>
  51 #include <linux/mlx4/qp.h>
  52
  53 #include "mlx4_ib.h"
  54 #include "user.h"
  55
  56 #define DRV_NAME        MLX4_IB_DRV_NAME
  57 #define DRV_VERSION     "2.2-1"
  58 #define DRV_RELDATE     "Feb 2014"
  59
  60 #define MLX4_IB_FLOW_MAX_PRIO 0xFFF
  61 #define MLX4_IB_FLOW_QPN_MASK 0xFFFFFF
  62
  63 MODULE_AUTHOR("Roland Dreier");
  64 MODULE_DESCRIPTION("Mellanox ConnectX HCA InfiniBand driver");
  65 MODULE_LICENSE("Dual BSD/GPL");
  66 MODULE_VERSION(DRV_VERSION);
  67
  68 int mlx4_ib_sm_guid_assign = 1;
  69 module_param_named(sm_guid_assign, mlx4_ib_sm_guid_assign, int, 0444);
  70 MODULE_PARM_DESC(sm_guid_assign, "Enable SM alias_GUID assignment if sm_guid_assign > 0 (Default: 1)");
  71
  72 static const char mlx4_ib_version[] =
  73         DRV_NAME ": Mellanox ConnectX InfiniBand driver v"
  74         DRV_VERSION " (" DRV_RELDATE ")\n";
  75
  76 struct update_gid_work {
  77         struct work_struct      work;
  78         union ib_gid            gids[128];
  79         struct mlx4_ib_dev     *dev;
  80         int                     port;
  81 };
  82
  83 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init);
  84
  85 static struct workqueue_struct *wq;
  86
  87 static void init_query_mad(struct ib_smp *mad)
  88 {
  89         mad->base_version  = 1;
  90         mad->mgmt_class    = IB_MGMT_CLASS_SUBN_LID_ROUTED;
  91         mad->class_version = 1;
  92         mad->method        = IB_MGMT_METHOD_GET;
  93 }
  94
  95 static union ib_gid zgid;
  96
  97 static int check_flow_steering_support(struct mlx4_dev *dev)
  98 {
  99         int eth_num_ports = 0;
 100         int ib_num_ports = 0;
 101
 102         int dmfs = dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED;
 103
 104         if (dmfs) {
 105                 int i;
 106                 mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_ETH)
 107                         eth_num_ports++;
 108                 mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
 109                         ib_num_ports++;
 110                 dmfs &= (!ib_num_ports ||
 111                          (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DMFS_IPOIB)) &&
 112                         (!eth_num_ports ||
 113                          (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_FS_EN));
 114                 if (ib_num_ports && mlx4_is_mfunc(dev)) {
 115                         pr_warn("Device managed flow steering is unavailable for IB port in multifunction env.\n");
 116                         dmfs = 0;
 117                 }
 118         }
 119         return dmfs;
 120 }
 121
 122 static int mlx4_ib_query_device(struct ib_device *ibdev,
 123                                 struct ib_device_attr *props)
 124 {
 125         struct mlx4_ib_dev *dev = to_mdev(ibdev);
 126         struct ib_smp *in_mad  = NULL;
 127         struct ib_smp *out_mad = NULL;
 128         int err = -ENOMEM;
 129
 130         in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
 131         out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
 132         if (!in_mad || !out_mad)
 133                 goto out;
 134
 135         init_query_mad(in_mad);
 136         in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
 137
 138         err = mlx4_MAD_IFC(to_mdev(ibdev), MLX4_MAD_IFC_IGNORE_KEYS,
 139                            1, NULL, NULL, in_mad, out_mad);
 140         if (err)
 141                 goto out;
 142
 143         memset(props, 0, sizeof *props);
 144
 145         props->fw_ver = dev->dev->caps.fw_ver;
 146         props->device_cap_flags    = IB_DEVICE_CHANGE_PHY_PORT |
 147                 IB_DEVICE_PORT_ACTIVE_EVENT             |
 148                 IB_DEVICE_SYS_IMAGE_GUID                |
 149                 IB_DEVICE_RC_RNR_NAK_GEN                |
 150                 IB_DEVICE_BLOCK_MULTICAST_LOOPBACK;
 151         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_PKEY_CNTR)
 152                 props->device_cap_flags |= IB_DEVICE_BAD_PKEY_CNTR;
 153         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_QKEY_CNTR)
 154                 props->device_cap_flags |= IB_DEVICE_BAD_QKEY_CNTR;
 155         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_APM)
 156                 props->device_cap_flags |= IB_DEVICE_AUTO_PATH_MIG;
 157         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_UD_AV_PORT)
 158                 props->device_cap_flags |= IB_DEVICE_UD_AV_PORT_ENFORCE;
 159         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_IPOIB_CSUM)
 160                 props->device_cap_flags |= IB_DEVICE_UD_IP_CSUM;
 161         if (dev->dev->caps.max_gso_sz && dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BLH)
 162                 props->device_cap_flags |= IB_DEVICE_UD_TSO;
 163         if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_RESERVED_LKEY)
 164                 props->device_cap_flags |= IB_DEVICE_LOCAL_DMA_LKEY;
 165         if ((dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_LOCAL_INV) &&
 166             (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_REMOTE_INV) &&
 167             (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_FAST_REG_WR))
 168                 props->device_cap_flags |= IB_DEVICE_MEM_MGT_EXTENSIONS;
 169         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC)
 170                 props->device_cap_flags |= IB_DEVICE_XRC;
 171         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW)
 172                 props->device_cap_flags |= IB_DEVICE_MEM_WINDOW;
 173         if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
 174                 if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_WIN_TYPE_2B)
 175                         props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2B;
 176                 else
 177                         props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2A;
 178         if (dev->steering_support ==  MLX4_STEERING_MODE_DEVICE_MANAGED)
 179                 props->device_cap_flags |= IB_DEVICE_MANAGED_FLOW_STEERING;
 180         }
 181
 182         props->vendor_id           = be32_to_cpup((__be32 *) (out_mad->data + 36)) &
 183                 0xffffff;
 184         props->vendor_part_id      = dev->dev->pdev->device;
 185         props->hw_ver              = be32_to_cpup((__be32 *) (out_mad->data + 32));
 186         memcpy(&props->sys_image_guid, out_mad->data +  4, 8);
 187
 188         props->max_mr_size         = ~0ull;
 189         props->page_size_cap       = dev->dev->caps.page_size_cap;
 190         props->max_qp              = dev->dev->quotas.qp;
 191         props->max_qp_wr           = dev->dev->caps.max_wqes - MLX4_IB_SQ_MAX_SPARE;
 192         props->max_sge             = min(dev->dev->caps.max_sq_sg,
 193                                          dev->dev->caps.max_rq_sg);
 194         props->max_cq              = dev->dev->quotas.cq;
 195         props->max_cqe             = dev->dev->caps.max_cqes;
 196         props->max_mr              = dev->dev->quotas.mpt;
 197         props->max_pd              = dev->dev->caps.num_pds - dev->dev->caps.reserved_pds;
 198         props->max_qp_rd_atom      = dev->dev->caps.max_qp_dest_rdma;
 199         props->max_qp_init_rd_atom = dev->dev->caps.max_qp_init_rdma;
 200         props->max_res_rd_atom     = props->max_qp_rd_atom * props->max_qp;
 201         props->max_srq             = dev->dev->quotas.srq;
 202         props->max_srq_wr          = dev->dev->caps.max_srq_wqes - 1;
 203         props->max_srq_sge         = dev->dev->caps.max_srq_sge;
 204         props->max_fast_reg_page_list_len = MLX4_MAX_FAST_REG_PAGES;
 205         props->local_ca_ack_delay  = dev->dev->caps.local_ca_ack_delay;
 206         props->atomic_cap          = dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_ATOMIC ?
 207                 IB_ATOMIC_HCA : IB_ATOMIC_NONE;
 208         props->masked_atomic_cap   = props->atomic_cap;
 209         props->max_pkeys           = dev->dev->caps.pkey_table_len[1];
 210         props->max_mcast_grp       = dev->dev->caps.num_mgms + dev->dev->caps.num_amgms;
 211         props->max_mcast_qp_attach = dev->dev->caps.num_qp_per_mgm;
 212         props->max_total_mcast_qp_attach = props->max_mcast_qp_attach *
 213                                            props->max_mcast_grp;
 214         props->max_map_per_fmr = dev->dev->caps.max_fmr_maps;
 215         props->max_ah = INT_MAX;
 216
 217 out:
 218         kfree(in_mad);
 219         kfree(out_mad);
 220
 221         return err;
 222 }
 223
 224 static enum rdma_link_layer
 225 mlx4_ib_port_link_layer(struct ib_device *device, u8 port_num)
 226 {
 227         struct mlx4_dev *dev = to_mdev(device)->dev;
 228
 229         return dev->caps.port_mask[port_num] == MLX4_PORT_TYPE_IB ?
 230                 IB_LINK_LAYER_INFINIBAND : IB_LINK_LAYER_ETHERNET;
 231 }
 232
 233 static int ib_link_query_port(struct ib_device *ibdev, u8 port,
 234                               struct ib_port_attr *props, int netw_view)
 235 {
 236         struct ib_smp *in_mad  = NULL;
 237         struct ib_smp *out_mad = NULL;
 238         int ext_active_speed;
 239         int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
 240         int err = -ENOMEM;
 241
 242         in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
 243         out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
 244         if (!in_mad || !out_mad)
 245                 goto out;
 246
 247         init_query_mad(in_mad);
 248         in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
 249         in_mad->attr_mod = cpu_to_be32(port);
 250
 251         if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
 252                 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
 253
 254         err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
 255                                 in_mad, out_mad);
 256         if (err)
 257                 goto out;
 258
 259
 260         props->lid              = be16_to_cpup((__be16 *) (out_mad->data + 16));
 261         props->lmc              = out_mad->data[34] & 0x7;
 262         props->sm_lid           = be16_to_cpup((__be16 *) (out_mad->data + 18));
 263         props->sm_sl            = out_mad->data[36] & 0xf;
 264         props->state            = out_mad->data[32] & 0xf;
 265         props->phys_state       = out_mad->data[33] >> 4;
 266         props->port_cap_flags   = be32_to_cpup((__be32 *) (out_mad->data + 20));
 267         if (netw_view)
 268                 props->gid_tbl_len = out_mad->data[50];
 269         else
 270                 props->gid_tbl_len = to_mdev(ibdev)->dev->caps.gid_table_len[port];
 271         props->max_msg_sz       = to_mdev(ibdev)->dev->caps.max_msg_sz;
 272         props->pkey_tbl_len     = to_mdev(ibdev)->dev->caps.pkey_table_len[port];
 273         props->bad_pkey_cntr    = be16_to_cpup((__be16 *) (out_mad->data + 46));
 274         props->qkey_viol_cntr   = be16_to_cpup((__be16 *) (out_mad->data + 48));
 275         props->active_width     = out_mad->data[31] & 0xf;
 276         props->active_speed     = out_mad->data[35] >> 4;
 277         props->max_mtu          = out_mad->data[41] & 0xf;
 278         props->active_mtu       = out_mad->data[36] >> 4;
 279         props->subnet_timeout   = out_mad->data[51] & 0x1f;
 280         props->max_vl_num       = out_mad->data[37] >> 4;
 281         props->init_type_reply  = out_mad->data[41] >> 4;
 282
 283         /* Check if extended speeds (EDR/FDR/...) are supported */
 284         if (props->port_cap_flags & IB_PORT_EXTENDED_SPEEDS_SUP) {
 285                 ext_active_speed = out_mad->data[62] >> 4;
 286
 287                 switch (ext_active_speed) {
 288                 case 1:
 289                         props->active_speed = IB_SPEED_FDR;
 290                         break;
 291                 case 2:
 292                         props->active_speed = IB_SPEED_EDR;
 293                         break;
 294                 }
 295         }
 296
 297         /* If reported active speed is QDR, check if is FDR-10 */
 298         if (props->active_speed == IB_SPEED_QDR) {
 299                 init_query_mad(in_mad);
 300                 in_mad->attr_id = MLX4_ATTR_EXTENDED_PORT_INFO;
 301                 in_mad->attr_mod = cpu_to_be32(port);
 302
 303                 err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port,
 304                                    NULL, NULL, in_mad, out_mad);
 305                 if (err)
 306                         goto out;
 307
 308                 /* Checking LinkSpeedActive for FDR-10 */
 309                 if (out_mad->data[15] & 0x1)
 310                         props->active_speed = IB_SPEED_FDR10;
 311         }
 312
 313         /* Avoid wrong speed value returned by FW if the IB link is down. */
 314         if (props->state == IB_PORT_DOWN)
 315                  props->active_speed = IB_SPEED_SDR;
 316
 317 out:
 318         kfree(in_mad);
 319         kfree(out_mad);
 320         return err;
 321 }
 322
 323 static u8 state_to_phys_state(enum ib_port_state state)
 324 {
 325         return state == IB_PORT_ACTIVE ? 5 : 3;
 326 }
 327
 328 static int eth_link_query_port(struct ib_device *ibdev, u8 port,
 329                                struct ib_port_attr *props, int netw_view)
 330 {
 331
 332         struct mlx4_ib_dev *mdev = to_mdev(ibdev);
 333         struct mlx4_ib_iboe *iboe = &mdev->iboe;
 334         struct net_device *ndev;
 335         enum ib_mtu tmp;
 336         struct mlx4_cmd_mailbox *mailbox;
 337         int err = 0;
 338
 339         mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
 340         if (IS_ERR(mailbox))
 341                 return PTR_ERR(mailbox);
 342
 343         err = mlx4_cmd_box(mdev->dev, 0, mailbox->dma, port, 0,
 344                            MLX4_CMD_QUERY_PORT, MLX4_CMD_TIME_CLASS_B,
 345                            MLX4_CMD_WRAPPED);
 346         if (err)
 347                 goto out;
 348
 349         props->active_width     =  (((u8 *)mailbox->buf)[5] == 0x40) ||
 350                                    (((u8 *)mailbox->buf)[5] == 0x20 /*56Gb*/) ?
 351                                            IB_WIDTH_4X : IB_WIDTH_1X;
 352         props->active_speed     =  (((u8 *)mailbox->buf)[5] == 0x20 /*56Gb*/) ?
 353                                            IB_SPEED_FDR : IB_SPEED_QDR;
 354         props->port_cap_flags   = IB_PORT_CM_SUP | IB_PORT_IP_BASED_GIDS;
 355         props->gid_tbl_len      = mdev->dev->caps.gid_table_len[port];
 356         props->max_msg_sz       = mdev->dev->caps.max_msg_sz;
 357         props->pkey_tbl_len     = 1;
 358         props->max_mtu          = IB_MTU_4096;
 359         props->max_vl_num       = 2;
 360         props->state            = IB_PORT_DOWN;
 361         props->phys_state       = state_to_phys_state(props->state);
 362         props->active_mtu       = IB_MTU_256;
 363         spin_lock(&iboe->lock);
 364         ndev = iboe->netdevs[port - 1];
 365         if (!ndev)
 366                 goto out_unlock;
 367
 368         tmp = iboe_get_mtu(ndev->mtu);
 369         props->active_mtu = tmp ? min(props->max_mtu, tmp) : IB_MTU_256;
 370
 371         props->state            = (netif_running(ndev) && netif_carrier_ok(ndev)) ?
 372                                         IB_PORT_ACTIVE : IB_PORT_DOWN;
 373         props->phys_state       = state_to_phys_state(props->state);
 374 out_unlock:
 375         spin_unlock(&iboe->lock);
 376 out:
 377         mlx4_free_cmd_mailbox(mdev->dev, mailbox);
 378         return err;
 379 }
 380
 381 int __mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
 382                          struct ib_port_attr *props, int netw_view)
 383 {
 384         int err;
 385
 386         memset(props, 0, sizeof *props);
 387
 388         err = mlx4_ib_port_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND ?
 389                 ib_link_query_port(ibdev, port, props, netw_view) :
 390                                 eth_link_query_port(ibdev, port, props, netw_view);
 391
 392         return err;
 393 }
 394
 395 static int mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
 396                               struct ib_port_attr *props)
 397 {
 398         /* returns host view */
 399         return __mlx4_ib_query_port(ibdev, port, props, 0);
 400 }
 401
 402 int __mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
 403                         union ib_gid *gid, int netw_view)
 404 {
 405         struct ib_smp *in_mad  = NULL;
 406         struct ib_smp *out_mad = NULL;
 407         int err = -ENOMEM;
 408         struct mlx4_ib_dev *dev = to_mdev(ibdev);
 409         int clear = 0;
 410         int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
 411
 412         in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
 413         out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
 414         if (!in_mad || !out_mad)
 415                 goto out;
 416
 417         init_query_mad(in_mad);
 418         in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
 419         in_mad->attr_mod = cpu_to_be32(port);
 420
 421         if (mlx4_is_mfunc(dev->dev) && netw_view)
 422                 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
 423
 424         err = mlx4_MAD_IFC(dev, mad_ifc_flags, port, NULL, NULL, in_mad, out_mad);
 425         if (err)
 426                 goto out;
 427
 428         memcpy(gid->raw, out_mad->data + 8, 8);
 429
 430         if (mlx4_is_mfunc(dev->dev) && !netw_view) {
 431                 if (index) {
 432                         /* For any index > 0, return the null guid */
 433                         err = 0;
 434                         clear = 1;
 435                         goto out;
 436                 }
 437         }
 438
 439         init_query_mad(in_mad);
 440         in_mad->attr_id  = IB_SMP_ATTR_GUID_INFO;
 441         in_mad->attr_mod = cpu_to_be32(index / 8);
 442
 443         err = mlx4_MAD_IFC(dev, mad_ifc_flags, port,
 444                            NULL, NULL, in_mad, out_mad);
 445         if (err)
 446                 goto out;
 447
 448         memcpy(gid->raw + 8, out_mad->data + (index % 8) * 8, 8);
 449
 450 out:
 451         if (clear)
 452                 memset(gid->raw + 8, 0, 8);
 453         kfree(in_mad);
 454         kfree(out_mad);
 455         return err;
 456 }
 457
 458 static int iboe_query_gid(struct ib_device *ibdev, u8 port, int index,
 459                           union ib_gid *gid)
 460 {
 461         struct mlx4_ib_dev *dev = to_mdev(ibdev);
 462
 463         *gid = dev->iboe.gid_table[port - 1][index];
 464
 465         return 0;
 466 }
 467
 468 static int mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
 469                              union ib_gid *gid)
 470 {
 471         if (rdma_port_get_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND)
 472                 return __mlx4_ib_query_gid(ibdev, port, index, gid, 0);
 473         else
 474                 return iboe_query_gid(ibdev, port, index, gid);
 475 }
 476
 477 int __mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index,
 478                          u16 *pkey, int netw_view)
 479 {
 480         struct ib_smp *in_mad  = NULL;
 481         struct ib_smp *out_mad = NULL;
 482         int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
 483         int err = -ENOMEM;
 484
 485         in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
 486         out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
 487         if (!in_mad || !out_mad)
 488                 goto out;
 489
 490         init_query_mad(in_mad);
 491         in_mad->attr_id  = IB_SMP_ATTR_PKEY_TABLE;
 492         in_mad->attr_mod = cpu_to_be32(index / 32);
 493
 494         if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
 495                 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
 496
 497         err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
 498                            in_mad, out_mad);
 499         if (err)
 500                 goto out;
 501
 502         *pkey = be16_to_cpu(((__be16 *) out_mad->data)[index % 32]);
 503
 504 out:
 505         kfree(in_mad);
 506         kfree(out_mad);
 507         return err;
 508 }
 509
 510 static int mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index, u16 *pkey)
 511 {
 512         return __mlx4_ib_query_pkey(ibdev, port, index, pkey, 0);
 513 }
 514
 515 static int mlx4_ib_modify_device(struct ib_device *ibdev, int mask,
 516                                  struct ib_device_modify *props)
 517 {
 518         struct mlx4_cmd_mailbox *mailbox;
 519         unsigned long flags;
 520
 521         if (mask & ~IB_DEVICE_MODIFY_NODE_DESC)
 522                 return -EOPNOTSUPP;
 523
 524         if (!(mask & IB_DEVICE_MODIFY_NODE_DESC))
 525                 return 0;
 526
 527         if (mlx4_is_slave(to_mdev(ibdev)->dev))
 528                 return -EOPNOTSUPP;
 529
 530         spin_lock_irqsave(&to_mdev(ibdev)->sm_lock, flags);
 531         memcpy(ibdev->node_desc, props->node_desc, 64);
 532         spin_unlock_irqrestore(&to_mdev(ibdev)->sm_lock, flags);
 533
 534         /*
 535          * If possible, pass node desc to FW, so it can generate
 536          * a 144 trap.  If cmd fails, just ignore.
 537          */
 538         mailbox = mlx4_alloc_cmd_mailbox(to_mdev(ibdev)->dev);
 539         if (IS_ERR(mailbox))
 540                 return 0;
 541
 542         memcpy(mailbox->buf, props->node_desc, 64);
 543         mlx4_cmd(to_mdev(ibdev)->dev, mailbox->dma, 1, 0,
 544                  MLX4_CMD_SET_NODE, MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
 545
 546         mlx4_free_cmd_mailbox(to_mdev(ibdev)->dev, mailbox);
 547
 548         return 0;
 549 }
 550
 551 static int mlx4_ib_SET_PORT(struct mlx4_ib_dev *dev, u8 port, int reset_qkey_viols,
 552                             u32 cap_mask)
 553 {
 554         struct mlx4_cmd_mailbox *mailbox;
 555         int err;
 556
 557         mailbox = mlx4_alloc_cmd_mailbox(dev->dev);
 558         if (IS_ERR(mailbox))
 559                 return PTR_ERR(mailbox);
 560
 561         if (dev->dev->flags & MLX4_FLAG_OLD_PORT_CMDS) {
 562                 *(u8 *) mailbox->buf         = !!reset_qkey_viols << 6;
 563                 ((__be32 *) mailbox->buf)[2] = cpu_to_be32(cap_mask);
 564         } else {
 565                 ((u8 *) mailbox->buf)[3]     = !!reset_qkey_viols;
 566                 ((__be32 *) mailbox->buf)[1] = cpu_to_be32(cap_mask);
 567         }
 568
 569         err = mlx4_cmd(dev->dev, mailbox->dma, port, 0, MLX4_CMD_SET_PORT,
 570                        MLX4_CMD_TIME_CLASS_B, MLX4_CMD_WRAPPED);
 571
 572         mlx4_free_cmd_mailbox(dev->dev, mailbox);
 573         return err;
 574 }
 575
 576 static int mlx4_ib_modify_port(struct ib_device *ibdev, u8 port, int mask,
 577                                struct ib_port_modify *props)
 578 {
 579         struct mlx4_ib_dev *mdev = to_mdev(ibdev);
 580         u8 is_eth = mdev->dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH;
 581         struct ib_port_attr attr;
 582         u32 cap_mask;
 583         int err;
 584
 585         /* return OK if this is RoCE. CM calls ib_modify_port() regardless
 586          * of whether port link layer is ETH or IB. For ETH ports, qkey
 587          * violations and port capabilities are not meaningful.
 588          */
 589         if (is_eth)
 590                 return 0;
 591
 592         mutex_lock(&mdev->cap_mask_mutex);
 593
 594         err = mlx4_ib_query_port(ibdev, port, &attr);
 595         if (err)
 596                 goto out;
 597
 598         cap_mask = (attr.port_cap_flags | props->set_port_cap_mask) &
 599                 ~props->clr_port_cap_mask;
 600
 601         err = mlx4_ib_SET_PORT(mdev, port,
 602                                !!(mask & IB_PORT_RESET_QKEY_CNTR),
 603                                cap_mask);
 604
 605 out:
 606         mutex_unlock(&to_mdev(ibdev)->cap_mask_mutex);
 607         return err;
 608 }
 609
 610 static struct ib_ucontext *mlx4_ib_alloc_ucontext(struct ib_device *ibdev,
 611                                                   struct ib_udata *udata)
 612 {
 613         struct mlx4_ib_dev *dev = to_mdev(ibdev);
 614         struct mlx4_ib_ucontext *context;
 615         struct mlx4_ib_alloc_ucontext_resp_v3 resp_v3;
 616         struct mlx4_ib_alloc_ucontext_resp resp;
 617         int err;
 618
 619         if (!dev->ib_active)
 620                 return ERR_PTR(-EAGAIN);
 621
 622         if (ibdev->uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION) {
 623                 resp_v3.qp_tab_size      = dev->dev->caps.num_qps;
 624                 resp_v3.bf_reg_size      = dev->dev->caps.bf_reg_size;
 625                 resp_v3.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
 626         } else {
 627                 resp.dev_caps         = dev->dev->caps.userspace_caps;
 628                 resp.qp_tab_size      = dev->dev->caps.num_qps;
 629                 resp.bf_reg_size      = dev->dev->caps.bf_reg_size;
 630                 resp.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
 631                 resp.cqe_size         = dev->dev->caps.cqe_size;
 632         }
 633
 634         context = kmalloc(sizeof *context, GFP_KERNEL);
 635         if (!context)
 636                 return ERR_PTR(-ENOMEM);
 637
 638         err = mlx4_uar_alloc(to_mdev(ibdev)->dev, &context->uar);
 639         if (err) {
 640                 kfree(context);
 641                 return ERR_PTR(err);
 642         }
 643
 644         INIT_LIST_HEAD(&context->db_page_list);
 645         mutex_init(&context->db_page_mutex);
 646
 647         if (ibdev->uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION)
 648                 err = ib_copy_to_udata(udata, &resp_v3, sizeof(resp_v3));
 649         else
 650                 err = ib_copy_to_udata(udata, &resp, sizeof(resp));
 651
 652         if (err) {
 653                 mlx4_uar_free(to_mdev(ibdev)->dev, &context->uar);
 654                 kfree(context);
 655                 return ERR_PTR(-EFAULT);
 656         }
 657
 658         return &context->ibucontext;
 659 }
 660
 661 static int mlx4_ib_dealloc_ucontext(struct ib_ucontext *ibcontext)
 662 {
 663         struct mlx4_ib_ucontext *context = to_mucontext(ibcontext);
 664
 665         mlx4_uar_free(to_mdev(ibcontext->device)->dev, &context->uar);
 666         kfree(context);
 667
 668         return 0;
 669 }
 670
 671 static int mlx4_ib_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
 672 {
 673         struct mlx4_ib_dev *dev = to_mdev(context->device);
 674
 675         if (vma->vm_end - vma->vm_start != PAGE_SIZE)
 676                 return -EINVAL;
 677
 678         if (vma->vm_pgoff == 0) {
 679                 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
 680
 681                 if (io_remap_pfn_range(vma, vma->vm_start,
 682                                        to_mucontext(context)->uar.pfn,
 683                                        PAGE_SIZE, vma->vm_page_prot))
 684                         return -EAGAIN;
 685         } else if (vma->vm_pgoff == 1 && dev->dev->caps.bf_reg_size != 0) {
 686                 vma->vm_page_prot = pgprot_writecombine(vma->vm_page_prot);
 687
 688                 if (io_remap_pfn_range(vma, vma->vm_start,
 689                                        to_mucontext(context)->uar.pfn +
 690                                        dev->dev->caps.num_uars,
 691                                        PAGE_SIZE, vma->vm_page_prot))
 692                         return -EAGAIN;
 693         } else
 694                 return -EINVAL;
 695
 696         return 0;
 697 }
 698
 699 static struct ib_pd *mlx4_ib_alloc_pd(struct ib_device *ibdev,
 700                                       struct ib_ucontext *context,
 701                                       struct ib_udata *udata)
 702 {
 703         struct mlx4_ib_pd *pd;
 704         int err;
 705
 706         pd = kmalloc(sizeof *pd, GFP_KERNEL);
 707         if (!pd)
 708                 return ERR_PTR(-ENOMEM);
 709
 710         err = mlx4_pd_alloc(to_mdev(ibdev)->dev, &pd->pdn);
 711         if (err) {
 712                 kfree(pd);
 713                 return ERR_PTR(err);
 714         }
 715
 716         if (context)
 717                 if (ib_copy_to_udata(udata, &pd->pdn, sizeof (__u32))) {
 718                         mlx4_pd_free(to_mdev(ibdev)->dev, pd->pdn);
 719                         kfree(pd);
 720                         return ERR_PTR(-EFAULT);
 721                 }
 722
 723         return &pd->ibpd;
 724 }
 725
 726 static int mlx4_ib_dealloc_pd(struct ib_pd *pd)
 727 {
 728         mlx4_pd_free(to_mdev(pd->device)->dev, to_mpd(pd)->pdn);
 729         kfree(pd);
 730
 731         return 0;
 732 }
 733
 734 static struct ib_xrcd *mlx4_ib_alloc_xrcd(struct ib_device *ibdev,
 735                                           struct ib_ucontext *context,
 736                                           struct ib_udata *udata)
 737 {
 738         struct mlx4_ib_xrcd *xrcd;
 739         int err;
 740
 741         if (!(to_mdev(ibdev)->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC))
 742                 return ERR_PTR(-ENOSYS);
 743
 744         xrcd = kmalloc(sizeof *xrcd, GFP_KERNEL);
 745         if (!xrcd)
 746                 return ERR_PTR(-ENOMEM);
 747
 748         err = mlx4_xrcd_alloc(to_mdev(ibdev)->dev, &xrcd->xrcdn);
 749         if (err)
 750                 goto err1;
 751
 752         xrcd->pd = ib_alloc_pd(ibdev);
 753         if (IS_ERR(xrcd->pd)) {
 754                 err = PTR_ERR(xrcd->pd);
 755                 goto err2;
 756         }
 757
 758         xrcd->cq = ib_create_cq(ibdev, NULL, NULL, xrcd, 1, 0);
 759         if (IS_ERR(xrcd->cq)) {
 760                 err = PTR_ERR(xrcd->cq);
 761                 goto err3;
 762         }
 763
 764         return &xrcd->ibxrcd;
 765
 766 err3:
 767         ib_dealloc_pd(xrcd->pd);
 768 err2:
 769         mlx4_xrcd_free(to_mdev(ibdev)->dev, xrcd->xrcdn);
 770 err1:
 771         kfree(xrcd);
 772         return ERR_PTR(err);
 773 }
 774
 775 static int mlx4_ib_dealloc_xrcd(struct ib_xrcd *xrcd)
 776 {
 777         ib_destroy_cq(to_mxrcd(xrcd)->cq);
 778         ib_dealloc_pd(to_mxrcd(xrcd)->pd);
 779         mlx4_xrcd_free(to_mdev(xrcd->device)->dev, to_mxrcd(xrcd)->xrcdn);
 780         kfree(xrcd);
 781
 782         return 0;
 783 }
 784
 785 static int add_gid_entry(struct ib_qp *ibqp, union ib_gid *gid)
 786 {
 787         struct mlx4_ib_qp *mqp = to_mqp(ibqp);
 788         struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
 789         struct mlx4_ib_gid_entry *ge;
 790
 791         ge = kzalloc(sizeof *ge, GFP_KERNEL);
 792         if (!ge)
 793                 return -ENOMEM;
 794
 795         ge->gid = *gid;
 796         if (mlx4_ib_add_mc(mdev, mqp, gid)) {
 797                 ge->port = mqp->port;
 798                 ge->added = 1;
 799         }
 800
 801         mutex_lock(&mqp->mutex);
 802         list_add_tail(&ge->list, &mqp->gid_list);
 803         mutex_unlock(&mqp->mutex);
 804
 805         return 0;
 806 }
 807
 808 int mlx4_ib_add_mc(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
 809                    union ib_gid *gid)
 810 {
 811         struct net_device *ndev;
 812         int ret = 0;
 813
 814         if (!mqp->port)
 815                 return 0;
 816
 817         spin_lock(&mdev->iboe.lock);
 818         ndev = mdev->iboe.netdevs[mqp->port - 1];
 819         if (ndev)
 820                 dev_hold(ndev);
 821         spin_unlock(&mdev->iboe.lock);
 822
 823         if (ndev) {
 824                 ret = 1;
 825                 dev_put(ndev);
 826         }
 827
 828         return ret;
 829 }
 830
 831 struct mlx4_ib_steering {
 832         struct list_head list;
 833         u64 reg_id;
 834         union ib_gid gid;
 835 };
 836
 837 static int parse_flow_attr(struct mlx4_dev *dev,
 838                            u32 qp_num,
 839                            union ib_flow_spec *ib_spec,
 840                            struct _rule_hw *mlx4_spec)
 841 {
 842         enum mlx4_net_trans_rule_id type;
 843
 844         switch (ib_spec->type) {
 845         case IB_FLOW_SPEC_ETH:
 846                 type = MLX4_NET_TRANS_RULE_ID_ETH;
 847                 memcpy(mlx4_spec->eth.dst_mac, ib_spec->eth.val.dst_mac,
 848                        ETH_ALEN);
 849                 memcpy(mlx4_spec->eth.dst_mac_msk, ib_spec->eth.mask.dst_mac,
 850                        ETH_ALEN);
 851                 mlx4_spec->eth.vlan_tag = ib_spec->eth.val.vlan_tag;
 852                 mlx4_spec->eth.vlan_tag_msk = ib_spec->eth.mask.vlan_tag;
 853                 break;
 854         case IB_FLOW_SPEC_IB:
 855                 type = MLX4_NET_TRANS_RULE_ID_IB;
 856                 mlx4_spec->ib.l3_qpn =
 857                         cpu_to_be32(qp_num);
 858                 mlx4_spec->ib.qpn_mask =
 859                         cpu_to_be32(MLX4_IB_FLOW_QPN_MASK);
 860                 break;
 861
 862
 863         case IB_FLOW_SPEC_IPV4:
 864                 type = MLX4_NET_TRANS_RULE_ID_IPV4;
 865                 mlx4_spec->ipv4.src_ip = ib_spec->ipv4.val.src_ip;
 866                 mlx4_spec->ipv4.src_ip_msk = ib_spec->ipv4.mask.src_ip;
 867                 mlx4_spec->ipv4.dst_ip = ib_spec->ipv4.val.dst_ip;
 868                 mlx4_spec->ipv4.dst_ip_msk = ib_spec->ipv4.mask.dst_ip;
 869                 break;
 870
 871         case IB_FLOW_SPEC_TCP:
 872         case IB_FLOW_SPEC_UDP:
 873                 type = ib_spec->type == IB_FLOW_SPEC_TCP ?
 874                                         MLX4_NET_TRANS_RULE_ID_TCP :
 875                                         MLX4_NET_TRANS_RULE_ID_UDP;
 876                 mlx4_spec->tcp_udp.dst_port = ib_spec->tcp_udp.val.dst_port;
 877                 mlx4_spec->tcp_udp.dst_port_msk = ib_spec->tcp_udp.mask.dst_port;
 878                 mlx4_spec->tcp_udp.src_port = ib_spec->tcp_udp.val.src_port;
 879                 mlx4_spec->tcp_udp.src_port_msk = ib_spec->tcp_udp.mask.src_port;
 880                 break;
 881
 882         default:
 883                 return -EINVAL;
 884         }
 885         if (mlx4_map_sw_to_hw_steering_id(dev, type) < 0 ||
 886             mlx4_hw_rule_sz(dev, type) < 0)
 887                 return -EINVAL;
 888         mlx4_spec->id = cpu_to_be16(mlx4_map_sw_to_hw_steering_id(dev, type));
 889         mlx4_spec->size = mlx4_hw_rule_sz(dev, type) >> 2;
 890         return mlx4_hw_rule_sz(dev, type);
 891 }
 892
 893 struct default_rules {
 894         __u32 mandatory_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
 895         __u32 mandatory_not_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
 896         __u32 rules_create_list[IB_FLOW_SPEC_SUPPORT_LAYERS];
 897         __u8  link_layer;
 898 };
 899 static const struct default_rules default_table[] = {
 900         {
 901                 .mandatory_fields = {IB_FLOW_SPEC_IPV4},
 902                 .mandatory_not_fields = {IB_FLOW_SPEC_ETH},
 903                 .rules_create_list = {IB_FLOW_SPEC_IB},
 904                 .link_layer = IB_LINK_LAYER_INFINIBAND
 905         }
 906 };
 907
 908 static int __mlx4_ib_default_rules_match(struct ib_qp *qp,
 909                                          struct ib_flow_attr *flow_attr)
 910 {
 911         int i, j, k;
 912         void *ib_flow;
 913         const struct default_rules *pdefault_rules = default_table;
 914         u8 link_layer = rdma_port_get_link_layer(qp->device, flow_attr->port);
 915
 916         for (i = 0; i < sizeof(default_table)/sizeof(default_table[0]); i++,
 917              pdefault_rules++) {
 918                 __u32 field_types[IB_FLOW_SPEC_SUPPORT_LAYERS];
 919                 memset(&field_types, 0, sizeof(field_types));
 920
 921                 if (link_layer != pdefault_rules->link_layer)
 922                         continue;
 923
 924                 ib_flow = flow_attr + 1;
 925                 /* we assume the specs are sorted */
 926                 for (j = 0, k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS &&
 927                      j < flow_attr->num_of_specs; k++) {
 928                         union ib_flow_spec *current_flow =
 929                                 (union ib_flow_spec *)ib_flow;
 930
 931                         /* same layer but different type */
 932                         if (((current_flow->type & IB_FLOW_SPEC_LAYER_MASK) ==
 933                              (pdefault_rules->mandatory_fields[k] &
 934                               IB_FLOW_SPEC_LAYER_MASK)) &&
 935                             (current_flow->type !=
 936                              pdefault_rules->mandatory_fields[k]))
 937                                 goto out;
 938
 939                         /* same layer, try match next one */
 940                         if (current_flow->type ==
 941                             pdefault_rules->mandatory_fields[k]) {
 942                                 j++;
 943                                 ib_flow +=
 944                                         ((union ib_flow_spec *)ib_flow)->size;
 945                         }
 946                 }
 947
 948                 ib_flow = flow_attr + 1;
 949                 for (j = 0; j < flow_attr->num_of_specs;
 950                      j++, ib_flow += ((union ib_flow_spec *)ib_flow)->size)
 951                         for (k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS; k++)
 952                                 /* same layer and same type */
 953                                 if (((union ib_flow_spec *)ib_flow)->type ==
 954                                     pdefault_rules->mandatory_not_fields[k])
 955                                         goto out;
 956
 957                 return i;
 958         }
 959 out:
 960         return -1;
 961 }
 962
 963 static int __mlx4_ib_create_default_rules(
 964                 struct mlx4_ib_dev *mdev,
 965                 struct ib_qp *qp,
 966                 const struct default_rules *pdefault_rules,
 967                 struct _rule_hw *mlx4_spec) {
 968         int size = 0;
 969         int i;
 970
 971         for (i = 0; i < sizeof(pdefault_rules->rules_create_list)/
 972                         sizeof(pdefault_rules->rules_create_list[0]); i++) {
 973                 int ret;
 974                 union ib_flow_spec ib_spec;
 975                 switch (pdefault_rules->rules_create_list[i]) {
 976                 case 0:
 977                         /* no rule */
 978                         continue;
 979                 case IB_FLOW_SPEC_IB:
 980                         ib_spec.type = IB_FLOW_SPEC_IB;
 981                         ib_spec.size = sizeof(struct ib_flow_spec_ib);
 982
 983                         break;
 984                 default:
 985                         /* invalid rule */
 986                         return -EINVAL;
 987                 }
 988                 /* We must put empty rule, qpn is being ignored */
 989                 ret = parse_flow_attr(mdev->dev, 0, &ib_spec,
 990                                       mlx4_spec);
 991                 if (ret < 0) {
 992                         pr_info("invalid parsing\n");
 993                         return -EINVAL;
 994                 }
 995
 996                 mlx4_spec = (void *)mlx4_spec + ret;
 997                 size += ret;
 998         }
 999         return size;
1000 }
1001
1002 static int __mlx4_ib_create_flow(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
1003                           int domain,
1004                           enum mlx4_net_trans_promisc_mode flow_type,
1005                           u64 *reg_id)
1006 {
1007         int ret, i;
1008         int size = 0;
1009         void *ib_flow;
1010         struct mlx4_ib_dev *mdev = to_mdev(qp->device);
1011         struct mlx4_cmd_mailbox *mailbox;
1012         struct mlx4_net_trans_rule_hw_ctrl *ctrl;
1013         int default_flow;
1014
1015         static const u16 __mlx4_domain[] = {
1016                 [IB_FLOW_DOMAIN_USER] = MLX4_DOMAIN_UVERBS,
1017                 [IB_FLOW_DOMAIN_ETHTOOL] = MLX4_DOMAIN_ETHTOOL,
1018                 [IB_FLOW_DOMAIN_RFS] = MLX4_DOMAIN_RFS,
1019                 [IB_FLOW_DOMAIN_NIC] = MLX4_DOMAIN_NIC,
1020         };
1021
1022         if (flow_attr->port < 1 || flow_attr->port > qp->device->phys_port_cnt)
1023                 return -EINVAL;
1024
1025         if (flow_attr->priority > MLX4_IB_FLOW_MAX_PRIO) {
1026                 pr_err("Invalid priority value %d\n", flow_attr->priority);
1027                 return -EINVAL;
1028         }
1029
1030         if (domain >= IB_FLOW_DOMAIN_NUM) {
1031                 pr_err("Invalid domain value %d\n", domain);
1032                 return -EINVAL;
1033         }
1034
1035         if (mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type) < 0)
1036                 return -EINVAL;
1037
1038         mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
1039         if (IS_ERR(mailbox))
1040                 return PTR_ERR(mailbox);
1041         ctrl = mailbox->buf;
1042
1043         ctrl->prio = cpu_to_be16(__mlx4_domain[domain] |
1044                                  flow_attr->priority);
1045         ctrl->type = mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type);
1046         ctrl->port = flow_attr->port;
1047         ctrl->qpn = cpu_to_be32(qp->qp_num);
1048
1049         ib_flow = flow_attr + 1;
1050         size += sizeof(struct mlx4_net_trans_rule_hw_ctrl);
1051         /* Add default flows */
1052         default_flow = __mlx4_ib_default_rules_match(qp, flow_attr);
1053         if (default_flow >= 0) {
1054                 ret = __mlx4_ib_create_default_rules(
1055                                 mdev, qp, default_table + default_flow,
1056                                 mailbox->buf + size);
1057                 if (ret < 0) {
1058                         mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1059                         return -EINVAL;
1060                 }
1061                 size += ret;
1062         }
1063         for (i = 0; i < flow_attr->num_of_specs; i++) {
1064                 ret = parse_flow_attr(mdev->dev, qp->qp_num, ib_flow,
1065                                       mailbox->buf + size);
1066                 if (ret < 0) {
1067                         mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1068                         return -EINVAL;
1069                 }
1070                 ib_flow += ((union ib_flow_spec *) ib_flow)->size;
1071                 size += ret;
1072         }
1073
1074         ret = mlx4_cmd_imm(mdev->dev, mailbox->dma, reg_id, size >> 2, 0,
1075                            MLX4_QP_FLOW_STEERING_ATTACH, MLX4_CMD_TIME_CLASS_A,
1076                            MLX4_CMD_NATIVE);
1077         if (ret == -ENOMEM)
1078                 pr_err("mcg table is full. Fail to register network rule.\n");
1079         else if (ret == -ENXIO)
1080                 pr_err("Device managed flow steering is disabled. Fail to register network rule.\n");
1081         else if (ret)
1082                 pr_err("Invalid argumant. Fail to register network rule.\n");
1083
1084         mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1085         return ret;
1086 }
1087
1088 static int __mlx4_ib_destroy_flow(struct mlx4_dev *dev, u64 reg_id)
1089 {
1090         int err;
1091         err = mlx4_cmd(dev, reg_id, 0, 0,
1092                        MLX4_QP_FLOW_STEERING_DETACH, MLX4_CMD_TIME_CLASS_A,
1093                        MLX4_CMD_NATIVE);
1094         if (err)
1095                 pr_err("Fail to detach network rule. registration id = 0x%llx\n",
1096                        reg_id);
1097         return err;
1098 }
1099
1100 static struct ib_flow *mlx4_ib_create_flow(struct ib_qp *qp,
1101                                     struct ib_flow_attr *flow_attr,
1102                                     int domain)
1103 {
1104         int err = 0, i = 0;
1105         struct mlx4_ib_flow *mflow;
1106         enum mlx4_net_trans_promisc_mode type[2];
1107
1108         memset(type, 0, sizeof(type));
1109
1110         mflow = kzalloc(sizeof(*mflow), GFP_KERNEL);
1111         if (!mflow) {
1112                 err = -ENOMEM;
1113                 goto err_free;
1114         }
1115
1116         switch (flow_attr->type) {
1117         case IB_FLOW_ATTR_NORMAL:
1118                 type[0] = MLX4_FS_REGULAR;
1119                 break;
1120
1121         case IB_FLOW_ATTR_ALL_DEFAULT:
1122                 type[0] = MLX4_FS_ALL_DEFAULT;
1123                 break;
1124
1125         case IB_FLOW_ATTR_MC_DEFAULT:
1126                 type[0] = MLX4_FS_MC_DEFAULT;
1127                 break;
1128
1129         case IB_FLOW_ATTR_SNIFFER:
1130                 type[0] = MLX4_FS_UC_SNIFFER;
1131                 type[1] = MLX4_FS_MC_SNIFFER;
1132                 break;
1133
1134         default:
1135                 err = -EINVAL;
1136                 goto err_free;
1137         }
1138
1139         while (i < ARRAY_SIZE(type) && type[i]) {
1140                 err = __mlx4_ib_create_flow(qp, flow_attr, domain, type[i],
1141                                             &mflow->reg_id[i]);
1142                 if (err)
1143                         goto err_free;
1144                 i++;
1145         }
1146
1147         return &mflow->ibflow;
1148
1149 err_free:
1150         kfree(mflow);
1151         return ERR_PTR(err);
1152 }
1153
1154 static int mlx4_ib_destroy_flow(struct ib_flow *flow_id)
1155 {
1156         int err, ret = 0;
1157         int i = 0;
1158         struct mlx4_ib_dev *mdev = to_mdev(flow_id->qp->device);
1159         struct mlx4_ib_flow *mflow = to_mflow(flow_id);
1160
1161         while (i < ARRAY_SIZE(mflow->reg_id) && mflow->reg_id[i]) {
1162                 err = __mlx4_ib_destroy_flow(mdev->dev, mflow->reg_id[i]);
1163                 if (err)
1164                         ret = err;
1165                 i++;
1166         }
1167
1168         kfree(mflow);
1169         return ret;
1170 }
1171
1172 static int mlx4_ib_mcg_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
1173 {
1174         int err;
1175         struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1176         struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1177         u64 reg_id;
1178         struct mlx4_ib_steering *ib_steering = NULL;
1179         enum mlx4_protocol prot = MLX4_PROT_IB_IPV6;
1180
1181         if (mdev->dev->caps.steering_mode ==
1182             MLX4_STEERING_MODE_DEVICE_MANAGED) {
1183                 ib_steering = kmalloc(sizeof(*ib_steering), GFP_KERNEL);
1184                 if (!ib_steering)
1185                         return -ENOMEM;
1186         }
1187
1188         err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw, mqp->port,
1189                                     !!(mqp->flags &
1190                                        MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
1191                                     prot, &reg_id);
1192         if (err) {
1193                 pr_err("multicast attach op failed, err %d\n", err);
1194                 goto err_malloc;
1195         }
1196
1197         err = add_gid_entry(ibqp, gid);
1198         if (err)
1199                 goto err_add;
1200
1201         if (ib_steering) {
1202                 memcpy(ib_steering->gid.raw, gid->raw, 16);
1203                 ib_steering->reg_id = reg_id;
1204                 mutex_lock(&mqp->mutex);
1205                 list_add(&ib_steering->list, &mqp->steering_rules);
1206                 mutex_unlock(&mqp->mutex);
1207         }
1208         return 0;
1209
1210 err_add:
1211         mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1212                               prot, reg_id);
1213 err_malloc:
1214         kfree(ib_steering);
1215
1216         return err;
1217 }
1218
1219 static struct mlx4_ib_gid_entry *find_gid_entry(struct mlx4_ib_qp *qp, u8 *raw)
1220 {
1221         struct mlx4_ib_gid_entry *ge;
1222         struct mlx4_ib_gid_entry *tmp;
1223         struct mlx4_ib_gid_entry *ret = NULL;
1224
1225         list_for_each_entry_safe(ge, tmp, &qp->gid_list, list) {
1226                 if (!memcmp(raw, ge->gid.raw, 16)) {
1227                         ret = ge;
1228                         break;
1229                 }
1230         }
1231
1232         return ret;
1233 }
1234
1235 static int mlx4_ib_mcg_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
1236 {
1237         int err;
1238         struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1239         struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1240         struct net_device *ndev;
1241         struct mlx4_ib_gid_entry *ge;
1242         u64 reg_id = 0;
1243         enum mlx4_protocol prot =  MLX4_PROT_IB_IPV6;
1244
1245         if (mdev->dev->caps.steering_mode ==
1246             MLX4_STEERING_MODE_DEVICE_MANAGED) {
1247                 struct mlx4_ib_steering *ib_steering;
1248
1249                 mutex_lock(&mqp->mutex);
1250                 list_for_each_entry(ib_steering, &mqp->steering_rules, list) {
1251                         if (!memcmp(ib_steering->gid.raw, gid->raw, 16)) {
1252                                 list_del(&ib_steering->list);
1253                                 break;
1254                         }
1255                 }
1256                 mutex_unlock(&mqp->mutex);
1257                 if (&ib_steering->list == &mqp->steering_rules) {
1258                         pr_err("Couldn't find reg_id for mgid. Steering rule is left attached\n");
1259                         return -EINVAL;
1260                 }
1261                 reg_id = ib_steering->reg_id;
1262                 kfree(ib_steering);
1263         }
1264
1265         err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1266                                     prot, reg_id);
1267         if (err)
1268                 return err;
1269
1270         mutex_lock(&mqp->mutex);
1271         ge = find_gid_entry(mqp, gid->raw);
1272         if (ge) {
1273                 spin_lock(&mdev->iboe.lock);
1274                 ndev = ge->added ? mdev->iboe.netdevs[ge->port - 1] : NULL;
1275                 if (ndev)
1276                         dev_hold(ndev);
1277                 spin_unlock(&mdev->iboe.lock);
1278                 if (ndev)
1279                         dev_put(ndev);
1280                 list_del(&ge->list);
1281                 kfree(ge);
1282         } else
1283                 pr_warn("could not find mgid entry\n");
1284
1285         mutex_unlock(&mqp->mutex);
1286
1287         return 0;
1288 }
1289
1290 static int init_node_data(struct mlx4_ib_dev *dev)
1291 {
1292         struct ib_smp *in_mad  = NULL;
1293         struct ib_smp *out_mad = NULL;
1294         int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
1295         int err = -ENOMEM;
1296
1297         in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
1298         out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
1299         if (!in_mad || !out_mad)
1300                 goto out;
1301
1302         init_query_mad(in_mad);
1303         in_mad->attr_id = IB_SMP_ATTR_NODE_DESC;
1304         if (mlx4_is_master(dev->dev))
1305                 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
1306
1307         err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
1308         if (err)
1309                 goto out;
1310
1311         memcpy(dev->ib_dev.node_desc, out_mad->data, 64);
1312
1313         in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
1314
1315         err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
1316         if (err)
1317                 goto out;
1318
1319         dev->dev->rev_id = be32_to_cpup((__be32 *) (out_mad->data + 32));
1320         memcpy(&dev->ib_dev.node_guid, out_mad->data + 12, 8);
1321
1322 out:
1323         kfree(in_mad);
1324         kfree(out_mad);
1325         return err;
1326 }
1327
1328 static ssize_t show_hca(struct device *device, struct device_attribute *attr,
1329                         char *buf)
1330 {
1331         struct mlx4_ib_dev *dev =
1332                 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1333         return sprintf(buf, "MT%d\n", dev->dev->pdev->device);
1334 }
1335
1336 static ssize_t show_fw_ver(struct device *device, struct device_attribute *attr,
1337                            char *buf)
1338 {
1339         struct mlx4_ib_dev *dev =
1340                 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1341         return sprintf(buf, "%d.%d.%d\n", (int) (dev->dev->caps.fw_ver >> 32),
1342                        (int) (dev->dev->caps.fw_ver >> 16) & 0xffff,
1343                        (int) dev->dev->caps.fw_ver & 0xffff);
1344 }
1345
1346 static ssize_t show_rev(struct device *device, struct device_attribute *attr,
1347                         char *buf)
1348 {
1349         struct mlx4_ib_dev *dev =
1350                 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1351         return sprintf(buf, "%x\n", dev->dev->rev_id);
1352 }
1353
1354 static ssize_t show_board(struct device *device, struct device_attribute *attr,
1355                           char *buf)
1356 {
1357         struct mlx4_ib_dev *dev =
1358                 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1359         return sprintf(buf, "%.*s\n", MLX4_BOARD_ID_LEN,
1360                        dev->dev->board_id);
1361 }
1362
1363 static DEVICE_ATTR(hw_rev,   S_IRUGO, show_rev,    NULL);
1364 static DEVICE_ATTR(fw_ver,   S_IRUGO, show_fw_ver, NULL);
1365 static DEVICE_ATTR(hca_type, S_IRUGO, show_hca,    NULL);
1366 static DEVICE_ATTR(board_id, S_IRUGO, show_board,  NULL);
1367
1368 static struct device_attribute *mlx4_class_attributes[] = {
1369         &dev_attr_hw_rev,
1370         &dev_attr_fw_ver,
1371         &dev_attr_hca_type,
1372         &dev_attr_board_id
1373 };
1374
1375 static void mlx4_addrconf_ifid_eui48(u8 *eui, u16 vlan_id,
1376                                      struct net_device *dev)
1377 {
1378         memcpy(eui, dev->dev_addr, 3);
1379         memcpy(eui + 5, dev->dev_addr + 3, 3);
1380         if (vlan_id < 0x1000) {
1381                 eui[3] = vlan_id >> 8;
1382                 eui[4] = vlan_id & 0xff;
1383         } else {
1384                 eui[3] = 0xff;
1385                 eui[4] = 0xfe;
1386         }
1387         eui[0] ^= 2;
1388 }
1389
1390 static void update_gids_task(struct work_struct *work)
1391 {
1392         struct update_gid_work *gw = container_of(work, struct update_gid_work, work);
1393         struct mlx4_cmd_mailbox *mailbox;
1394         union ib_gid *gids;
1395         int err;
1396         struct mlx4_dev *dev = gw->dev->dev;
1397
1398         if (!gw->dev->ib_active)
1399                 return;
1400
1401         mailbox = mlx4_alloc_cmd_mailbox(dev);
1402         if (IS_ERR(mailbox)) {
1403                 pr_warn("update gid table failed %ld\n", PTR_ERR(mailbox));
1404                 return;
1405         }
1406
1407         gids = mailbox->buf;
1408         memcpy(gids, gw->gids, sizeof gw->gids);
1409
1410         err = mlx4_cmd(dev, mailbox->dma, MLX4_SET_PORT_GID_TABLE << 8 | gw->port,
1411                        1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
1412                        MLX4_CMD_WRAPPED);
1413         if (err)
1414                 pr_warn("set port command failed\n");
1415         else
1416                 mlx4_ib_dispatch_event(gw->dev, gw->port, IB_EVENT_GID_CHANGE);
1417
1418         mlx4_free_cmd_mailbox(dev, mailbox);
1419         kfree(gw);
1420 }
1421
1422 static void reset_gids_task(struct work_struct *work)
1423 {
1424         struct update_gid_work *gw =
1425                         container_of(work, struct update_gid_work, work);
1426         struct mlx4_cmd_mailbox *mailbox;
1427         union ib_gid *gids;
1428         int err;
1429         struct mlx4_dev *dev = gw->dev->dev;
1430
1431         if (!gw->dev->ib_active)
1432                 return;
1433
1434         mailbox = mlx4_alloc_cmd_mailbox(dev);
1435         if (IS_ERR(mailbox)) {
1436                 pr_warn("reset gid table failed\n");
1437                 goto free;
1438         }
1439
1440         gids = mailbox->buf;
1441         memcpy(gids, gw->gids, sizeof(gw->gids));
1442
1443         if (mlx4_ib_port_link_layer(&gw->dev->ib_dev, gw->port) ==
1444                                     IB_LINK_LAYER_ETHERNET) {
1445                 err = mlx4_cmd(dev, mailbox->dma,
1446                                MLX4_SET_PORT_GID_TABLE << 8 | gw->port,
1447                                1, MLX4_CMD_SET_PORT,
1448                                MLX4_CMD_TIME_CLASS_B,
1449                                MLX4_CMD_WRAPPED);
1450                 if (err)
1451                         pr_warn(KERN_WARNING
1452                                 "set port %d command failed\n", gw->port);
1453         }
1454
1455         mlx4_free_cmd_mailbox(dev, mailbox);
1456 free:
1457         kfree(gw);
1458 }
1459
1460 static int update_gid_table(struct mlx4_ib_dev *dev, int port,
1461                             union ib_gid *gid, int clear,
1462                             int default_gid)
1463 {
1464         struct update_gid_work *work;
1465         int i;
1466         int need_update = 0;
1467         int free = -1;
1468         int found = -1;
1469         int max_gids;
1470
1471         if (default_gid) {
1472                 free = 0;
1473         } else {
1474                 max_gids = dev->dev->caps.gid_table_len[port];
1475                 for (i = 1; i < max_gids; ++i) {
1476                         if (!memcmp(&dev->iboe.gid_table[port - 1][i], gid,
1477                                     sizeof(*gid)))
1478                                 found = i;
1479
1480                         if (clear) {
1481                                 if (found >= 0) {
1482                                         need_update = 1;
1483                                         dev->iboe.gid_table[port - 1][found] =
1484                                                 zgid;
1485                                         break;
1486                                 }
1487                         } else {
1488                                 if (found >= 0)
1489                                         break;
1490
1491                                 if (free < 0 &&
1492                                     !memcmp(&dev->iboe.gid_table[port - 1][i],
1493                                             &zgid, sizeof(*gid)))
1494                                         free = i;
1495                         }
1496                 }
1497         }
1498
1499         if (found == -1 && !clear && free >= 0) {
1500                 dev->iboe.gid_table[port - 1][free] = *gid;
1501                 need_update = 1;
1502         }
1503
1504         if (!need_update)
1505                 return 0;
1506
1507         work = kzalloc(sizeof(*work), GFP_ATOMIC);
1508         if (!work)
1509                 return -ENOMEM;
1510
1511         memcpy(work->gids, dev->iboe.gid_table[port - 1], sizeof(work->gids));
1512         INIT_WORK(&work->work, update_gids_task);
1513         work->port = port;
1514         work->dev = dev;
1515         queue_work(wq, &work->work);
1516
1517         return 0;
1518 }
1519
1520 static void mlx4_make_default_gid(struct  net_device *dev, union ib_gid *gid)
1521 {
1522         gid->global.subnet_prefix = cpu_to_be64(0xfe80000000000000LL);
1523         mlx4_addrconf_ifid_eui48(&gid->raw[8], 0xffff, dev);
1524 }
1525
1526
1527 static int reset_gid_table(struct mlx4_ib_dev *dev, u8 port)
1528 {
1529         struct update_gid_work *work;
1530
1531         work = kzalloc(sizeof(*work), GFP_ATOMIC);
1532         if (!work)
1533                 return -ENOMEM;
1534
1535         memset(dev->iboe.gid_table[port - 1], 0, sizeof(work->gids));
1536         memset(work->gids, 0, sizeof(work->gids));
1537         INIT_WORK(&work->work, reset_gids_task);
1538         work->dev = dev;
1539         work->port = port;
1540         queue_work(wq, &work->work);
1541         return 0;
1542 }
1543
1544 static int mlx4_ib_addr_event(int event, struct net_device *event_netdev,
1545                               struct mlx4_ib_dev *ibdev, union ib_gid *gid)
1546 {
1547         struct mlx4_ib_iboe *iboe;
1548         int port = 0;
1549         struct net_device *real_dev = rdma_vlan_dev_real_dev(event_netdev) ?
1550                                 rdma_vlan_dev_real_dev(event_netdev) :
1551                                 event_netdev;
1552         union ib_gid default_gid;
1553
1554         mlx4_make_default_gid(real_dev, &default_gid);
1555
1556         if (!memcmp(gid, &default_gid, sizeof(*gid)))
1557                 return 0;
1558
1559         if (event != NETDEV_DOWN && event != NETDEV_UP)
1560                 return 0;
1561
1562         if ((real_dev != event_netdev) &&
1563             (event == NETDEV_DOWN) &&
1564             rdma_link_local_addr((struct in6_addr *)gid))
1565                 return 0;
1566
1567         iboe = &ibdev->iboe;
1568         spin_lock(&iboe->lock);
1569
1570         for (port = 1; port <= ibdev->dev->caps.num_ports; ++port)
1571                 if ((netif_is_bond_master(real_dev) &&
1572                      (real_dev == iboe->masters[port - 1])) ||
1573                      (!netif_is_bond_master(real_dev) &&
1574                      (real_dev == iboe->netdevs[port - 1])))
1575                         update_gid_table(ibdev, port, gid,
1576                                          event == NETDEV_DOWN, 0);
1577
1578         spin_unlock(&iboe->lock);
1579         return 0;
1580
1581 }
1582
1583 static u8 mlx4_ib_get_dev_port(struct net_device *dev,
1584                                struct mlx4_ib_dev *ibdev)
1585 {
1586         u8 port = 0;
1587         struct mlx4_ib_iboe *iboe;
1588         struct net_device *real_dev = rdma_vlan_dev_real_dev(dev) ?
1589                                 rdma_vlan_dev_real_dev(dev) : dev;
1590
1591         iboe = &ibdev->iboe;
1592
1593         for (port = 1; port <= ibdev->dev->caps.num_ports; ++port)
1594                 if ((netif_is_bond_master(real_dev) &&
1595                      (real_dev == iboe->masters[port - 1])) ||
1596                      (!netif_is_bond_master(real_dev) &&
1597                      (real_dev == iboe->netdevs[port - 1])))
1598                         break;
1599
1600         if ((port == 0) || (port > ibdev->dev->caps.num_ports))
1601                 return 0;
1602         else
1603                 return port;
1604 }
1605
1606 static int mlx4_ib_inet_event(struct notifier_block *this, unsigned long event,
1607                                 void *ptr)
1608 {
1609         struct mlx4_ib_dev *ibdev;
1610         struct in_ifaddr *ifa = ptr;
1611         union ib_gid gid;
1612         struct net_device *event_netdev = ifa->ifa_dev->dev;
1613
1614         ipv6_addr_set_v4mapped(ifa->ifa_address, (struct in6_addr *)&gid);
1615
1616         ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb_inet);
1617
1618         mlx4_ib_addr_event(event, event_netdev, ibdev, &gid);
1619         return NOTIFY_DONE;
1620 }
1621
1622 #if IS_ENABLED(CONFIG_IPV6)
1623 static int mlx4_ib_inet6_event(struct notifier_block *this, unsigned long event,
1624                                 void *ptr)
1625 {
1626         struct mlx4_ib_dev *ibdev;
1627         struct inet6_ifaddr *ifa = ptr;
1628         union  ib_gid *gid = (union ib_gid *)&ifa->addr;
1629         struct net_device *event_netdev = ifa->idev->dev;
1630
1631         ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb_inet6);
1632
1633         mlx4_ib_addr_event(event, event_netdev, ibdev, gid);
1634         return NOTIFY_DONE;
1635 }
1636 #endif
1637
1638 #define MLX4_IB_INVALID_MAC     ((u64)-1)
1639 static void mlx4_ib_update_qps(struct mlx4_ib_dev *ibdev,
1640                                struct net_device *dev,
1641                                int port)
1642 {
1643         u64 new_smac = 0;
1644         u64 release_mac = MLX4_IB_INVALID_MAC;
1645         struct mlx4_ib_qp *qp;
1646
1647         read_lock(&dev_base_lock);
1648         new_smac = mlx4_mac_to_u64(dev->dev_addr);
1649         read_unlock(&dev_base_lock);
1650
1651         mutex_lock(&ibdev->qp1_proxy_lock[port - 1]);
1652         qp = ibdev->qp1_proxy[port - 1];
1653         if (qp) {
1654                 int new_smac_index;
1655                 u64 old_smac = qp->pri.smac;
1656                 struct mlx4_update_qp_params update_params;
1657
1658                 if (new_smac == old_smac)
1659                         goto unlock;
1660
1661                 new_smac_index = mlx4_register_mac(ibdev->dev, port, new_smac);
1662
1663                 if (new_smac_index < 0)
1664                         goto unlock;
1665
1666                 update_params.smac_index = new_smac_index;
1667                 if (mlx4_update_qp(ibdev->dev, &qp->mqp, MLX4_UPDATE_QP_SMAC,
1668                                    &update_params)) {
1669                         release_mac = new_smac;
1670                         goto unlock;
1671                 }
1672
1673                 qp->pri.smac = new_smac;
1674                 qp->pri.smac_index = new_smac_index;
1675
1676                 release_mac = old_smac;
1677         }
1678
1679 unlock:
1680         mutex_unlock(&ibdev->qp1_proxy_lock[port - 1]);
1681         if (release_mac != MLX4_IB_INVALID_MAC)
1682                 mlx4_unregister_mac(ibdev->dev, port, release_mac);
1683 }
1684
1685 static void mlx4_ib_get_dev_addr(struct net_device *dev,
1686                                  struct mlx4_ib_dev *ibdev, u8 port)
1687 {
1688         struct in_device *in_dev;
1689 #if IS_ENABLED(CONFIG_IPV6)
1690         struct inet6_dev *in6_dev;
1691         union ib_gid  *pgid;
1692         struct inet6_ifaddr *ifp;
1693         union ib_gid default_gid;
1694 #endif
1695         union ib_gid gid;
1696
1697
1698         if ((port == 0) || (port > ibdev->dev->caps.num_ports))
1699                 return;
1700
1701         /* IPv4 gids */
1702         in_dev = in_dev_get(dev);
1703         if (in_dev) {
1704                 for_ifa(in_dev) {
1705                         /*ifa->ifa_address;*/
1706                         ipv6_addr_set_v4mapped(ifa->ifa_address,
1707                                                (struct in6_addr *)&gid);
1708                         update_gid_table(ibdev, port, &gid, 0, 0);
1709                 }
1710                 endfor_ifa(in_dev);
1711                 in_dev_put(in_dev);
1712         }
1713 #if IS_ENABLED(CONFIG_IPV6)
1714         mlx4_make_default_gid(dev, &default_gid);
1715         /* IPv6 gids */
1716         in6_dev = in6_dev_get(dev);
1717         if (in6_dev) {
1718                 read_lock_bh(&in6_dev->lock);
1719                 list_for_each_entry(ifp, &in6_dev->addr_list, if_list) {
1720                         pgid = (union ib_gid *)&ifp->addr;
1721                         if (!memcmp(pgid, &default_gid, sizeof(*pgid)))
1722                                 continue;
1723                         update_gid_table(ibdev, port, pgid, 0, 0);
1724                 }
1725                 read_unlock_bh(&in6_dev->lock);
1726                 in6_dev_put(in6_dev);
1727         }
1728 #endif
1729 }
1730
1731 static void mlx4_ib_set_default_gid(struct mlx4_ib_dev *ibdev,
1732                                  struct  net_device *dev, u8 port)
1733 {
1734         union ib_gid gid;
1735         mlx4_make_default_gid(dev, &gid);
1736         update_gid_table(ibdev, port, &gid, 0, 1);
1737 }
1738
1739 static int mlx4_ib_init_gid_table(struct mlx4_ib_dev *ibdev)
1740 {
1741         struct  net_device *dev;
1742         struct mlx4_ib_iboe *iboe = &ibdev->iboe;
1743         int i;
1744
1745         for (i = 1; i <= ibdev->num_ports; ++i)
1746                 if (reset_gid_table(ibdev, i))
1747                         return -1;
1748
1749         read_lock(&dev_base_lock);
1750         spin_lock(&iboe->lock);
1751
1752         for_each_netdev(&init_net, dev) {
1753                 u8 port = mlx4_ib_get_dev_port(dev, ibdev);
1754                 if (port)
1755                         mlx4_ib_get_dev_addr(dev, ibdev, port);
1756         }
1757
1758         spin_unlock(&iboe->lock);
1759         read_unlock(&dev_base_lock);
1760
1761         return 0;
1762 }
1763
1764 static void mlx4_ib_scan_netdevs(struct mlx4_ib_dev *ibdev,
1765                                  struct net_device *dev,
1766                                  unsigned long event)
1767
1768 {
1769         struct mlx4_ib_iboe *iboe;
1770         int update_qps_port = -1;
1771         int port;
1772
1773         iboe = &ibdev->iboe;
1774
1775         spin_lock(&iboe->lock);
1776         mlx4_foreach_ib_transport_port(port, ibdev->dev) {
1777                 enum ib_port_state      port_state = IB_PORT_NOP;
1778                 struct net_device *old_master = iboe->masters[port - 1];
1779                 struct net_device *curr_netdev;
1780                 struct net_device *curr_master;
1781
1782                 iboe->netdevs[port - 1] =
1783                         mlx4_get_protocol_dev(ibdev->dev, MLX4_PROT_ETH, port);
1784                 if (iboe->netdevs[port - 1])
1785                         mlx4_ib_set_default_gid(ibdev,
1786                                                 iboe->netdevs[port - 1], port);
1787                 curr_netdev = iboe->netdevs[port - 1];
1788
1789                 if (iboe->netdevs[port - 1] &&
1790                     netif_is_bond_slave(iboe->netdevs[port - 1])) {
1791                         iboe->masters[port - 1] = netdev_master_upper_dev_get(
1792                                 iboe->netdevs[port - 1]);
1793                 } else {
1794                         iboe->masters[port - 1] = NULL;
1795                 }
1796                 curr_master = iboe->masters[port - 1];
1797
1798                 if (dev == iboe->netdevs[port - 1] &&
1799                     (event == NETDEV_CHANGEADDR || event == NETDEV_REGISTER ||
1800                      event == NETDEV_UP || event == NETDEV_CHANGE))
1801                         update_qps_port = port;
1802
1803                 if (curr_netdev) {
1804                         port_state = (netif_running(curr_netdev) && netif_carrier_ok(curr_netdev)) ?
1805                                                 IB_PORT_ACTIVE : IB_PORT_DOWN;
1806                         mlx4_ib_set_default_gid(ibdev, curr_netdev, port);
1807                         /* if using bonding/team and a slave port is down, we
1808                          * don't the bond IP based gids in the table since
1809                          * flows that select port by gid may get the down port.
1810                          */
1811                         if (curr_master && (port_state == IB_PORT_DOWN)) {
1812                                 reset_gid_table(ibdev, port);
1813                                 mlx4_ib_set_default_gid(ibdev,
1814                                                         curr_netdev, port);
1815                         }
1816                         /* if bonding is used it is possible that we add it to
1817                          * masters only after IP address is assigned to the
1818                          * net bonding interface.
1819                         */
1820                         if (curr_master && (old_master != curr_master)) {
1821                                 reset_gid_table(ibdev, port);
1822                                 mlx4_ib_set_default_gid(ibdev,
1823                                                         curr_netdev, port);
1824                                 mlx4_ib_get_dev_addr(curr_master, ibdev, port);
1825                         }
1826
1827                         if (!curr_master && (old_master != curr_master)) {
1828                                 reset_gid_table(ibdev, port);
1829                                 mlx4_ib_set_default_gid(ibdev,
1830                                                         curr_netdev, port);
1831                                 mlx4_ib_get_dev_addr(curr_netdev, ibdev, port);
1832                         }
1833                 } else {
1834                         reset_gid_table(ibdev, port);
1835                 }
1836         }
1837
1838         spin_unlock(&iboe->lock);
1839
1840         if (update_qps_port > 0)
1841                 mlx4_ib_update_qps(ibdev, dev, update_qps_port);
1842 }
1843
1844 static int mlx4_ib_netdev_event(struct notifier_block *this,
1845                                 unsigned long event, void *ptr)
1846 {
1847         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1848         struct mlx4_ib_dev *ibdev;
1849
1850         if (!net_eq(dev_net(dev), &init_net))
1851                 return NOTIFY_DONE;
1852
1853         ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb);
1854         mlx4_ib_scan_netdevs(ibdev, dev, event);
1855
1856         return NOTIFY_DONE;
1857 }
1858
1859 static void init_pkeys(struct mlx4_ib_dev *ibdev)
1860 {
1861         int port;
1862         int slave;
1863         int i;
1864
1865         if (mlx4_is_master(ibdev->dev)) {
1866                 for (slave = 0; slave <= ibdev->dev->num_vfs; ++slave) {
1867                         for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
1868                                 for (i = 0;
1869                                      i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
1870                                      ++i) {
1871                                         ibdev->pkeys.virt2phys_pkey[slave][port - 1][i] =
1872                                         /* master has the identity virt2phys pkey mapping */
1873                                                 (slave == mlx4_master_func_num(ibdev->dev) || !i) ? i :
1874                                                         ibdev->dev->phys_caps.pkey_phys_table_len[port] - 1;
1875                                         mlx4_sync_pkey_table(ibdev->dev, slave, port, i,
1876                                                              ibdev->pkeys.virt2phys_pkey[slave][port - 1][i]);
1877                                 }
1878                         }
1879                 }
1880                 /* initialize pkey cache */
1881                 for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
1882                         for (i = 0;
1883                              i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
1884                              ++i)
1885                                 ibdev->pkeys.phys_pkey_cache[port-1][i] =
1886                                         (i) ? 0 : 0xFFFF;
1887                 }
1888         }
1889 }
1890
1891 static void mlx4_ib_alloc_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
1892 {
1893         char name[80];
1894         int eq_per_port = 0;
1895         int added_eqs = 0;
1896         int total_eqs = 0;
1897         int i, j, eq;
1898
1899         /* Legacy mode or comp_pool is not large enough */
1900         if (dev->caps.comp_pool == 0 ||
1901             dev->caps.num_ports > dev->caps.comp_pool)
1902                 return;
1903
1904         eq_per_port = rounddown_pow_of_two(dev->caps.comp_pool/
1905                                         dev->caps.num_ports);
1906
1907         /* Init eq table */
1908         added_eqs = 0;
1909         mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
1910                 added_eqs += eq_per_port;
1911
1912         total_eqs = dev->caps.num_comp_vectors + added_eqs;
1913
1914         ibdev->eq_table = kzalloc(total_eqs * sizeof(int), GFP_KERNEL);
1915         if (!ibdev->eq_table)
1916                 return;
1917
1918         ibdev->eq_added = added_eqs;
1919
1920         eq = 0;
1921         mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB) {
1922                 for (j = 0; j < eq_per_port; j++) {
1923                         snprintf(name, sizeof(name), "mlx4-ib-%d-%d@%s",
1924                                  i, j, dev->pdev->bus->name);
1925                         /* Set IRQ for specific name (per ring) */
1926                         if (mlx4_assign_eq(dev, name, NULL,
1927                                            &ibdev->eq_table[eq])) {
1928                                 /* Use legacy (same as mlx4_en driver) */
1929                                 pr_warn("Can't allocate EQ %d; reverting to legacy\n", eq);
1930                                 ibdev->eq_table[eq] =
1931                                         (eq % dev->caps.num_comp_vectors);
1932                         }
1933                         eq++;
1934                 }
1935         }
1936
1937         /* Fill the reset of the vector with legacy EQ */
1938         for (i = 0, eq = added_eqs; i < dev->caps.num_comp_vectors; i++)
1939                 ibdev->eq_table[eq++] = i;
1940
1941         /* Advertise the new number of EQs to clients */
1942         ibdev->ib_dev.num_comp_vectors = total_eqs;
1943 }
1944
1945 static void mlx4_ib_free_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
1946 {
1947         int i;
1948
1949         /* no additional eqs were added */
1950         if (!ibdev->eq_table)
1951                 return;
1952
1953         /* Reset the advertised EQ number */
1954         ibdev->ib_dev.num_comp_vectors = dev->caps.num_comp_vectors;
1955
1956         /* Free only the added eqs */
1957         for (i = 0; i < ibdev->eq_added; i++) {
1958                 /* Don't free legacy eqs if used */
1959                 if (ibdev->eq_table[i] <= dev->caps.num_comp_vectors)
1960                         continue;
1961                 mlx4_release_eq(dev, ibdev->eq_table[i]);
1962         }
1963
1964         kfree(ibdev->eq_table);
1965 }
1966
1967 static void *mlx4_ib_add(struct mlx4_dev *dev)
1968 {
1969         struct mlx4_ib_dev *ibdev;
1970         int num_ports = 0;
1971         int i, j;
1972         int err;
1973         struct mlx4_ib_iboe *iboe;
1974         int ib_num_ports = 0;
1975
1976         pr_info_once("%s", mlx4_ib_version);
1977
1978         num_ports = 0;
1979         mlx4_foreach_ib_transport_port(i, dev)
1980                 num_ports++;
1981
1982         /* No point in registering a device with no ports... */
1983         if (num_ports == 0)
1984                 return NULL;
1985
1986         ibdev = (struct mlx4_ib_dev *) ib_alloc_device(sizeof *ibdev);
1987         if (!ibdev) {
1988                 dev_err(&dev->pdev->dev, "Device struct alloc failed\n");
1989                 return NULL;
1990         }
1991
1992         iboe = &ibdev->iboe;
1993
1994         if (mlx4_pd_alloc(dev, &ibdev->priv_pdn))
1995                 goto err_dealloc;
1996
1997         if (mlx4_uar_alloc(dev, &ibdev->priv_uar))
1998                 goto err_pd;
1999
2000         ibdev->uar_map = ioremap((phys_addr_t) ibdev->priv_uar.pfn << PAGE_SHIFT,
2001                                  PAGE_SIZE);
2002         if (!ibdev->uar_map)
2003                 goto err_uar;
2004         MLX4_INIT_DOORBELL_LOCK(&ibdev->uar_lock);
2005
2006         ibdev->dev = dev;
2007
2008         strlcpy(ibdev->ib_dev.name, "mlx4_%d", IB_DEVICE_NAME_MAX);
2009         ibdev->ib_dev.owner             = THIS_MODULE;
2010         ibdev->ib_dev.node_type         = RDMA_NODE_IB_CA;
2011         ibdev->ib_dev.local_dma_lkey    = dev->caps.reserved_lkey;
2012         ibdev->num_ports                = num_ports;
2013         ibdev->ib_dev.phys_port_cnt     = ibdev->num_ports;
2014         ibdev->ib_dev.num_comp_vectors  = dev->caps.num_comp_vectors;
2015         ibdev->ib_dev.dma_device        = &dev->pdev->dev;
2016
2017         if (dev->caps.userspace_caps)
2018                 ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_ABI_VERSION;
2019         else
2020                 ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION;
2021
2022         ibdev->ib_dev.uverbs_cmd_mask   =
2023                 (1ull << IB_USER_VERBS_CMD_GET_CONTEXT)         |
2024                 (1ull << IB_USER_VERBS_CMD_QUERY_DEVICE)        |
2025                 (1ull << IB_USER_VERBS_CMD_QUERY_PORT)          |
2026                 (1ull << IB_USER_VERBS_CMD_ALLOC_PD)            |
2027                 (1ull << IB_USER_VERBS_CMD_DEALLOC_PD)          |
2028                 (1ull << IB_USER_VERBS_CMD_REG_MR)              |
2029                 (1ull << IB_USER_VERBS_CMD_DEREG_MR)            |
2030                 (1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL) |
2031                 (1ull << IB_USER_VERBS_CMD_CREATE_CQ)           |
2032                 (1ull << IB_USER_VERBS_CMD_RESIZE_CQ)           |
2033                 (1ull << IB_USER_VERBS_CMD_DESTROY_CQ)          |
2034                 (1ull << IB_USER_VERBS_CMD_CREATE_QP)           |
2035                 (1ull << IB_USER_VERBS_CMD_MODIFY_QP)           |
2036                 (1ull << IB_USER_VERBS_CMD_QUERY_QP)            |
2037                 (1ull << IB_USER_VERBS_CMD_DESTROY_QP)          |
2038                 (1ull << IB_USER_VERBS_CMD_ATTACH_MCAST)        |
2039                 (1ull << IB_USER_VERBS_CMD_DETACH_MCAST)        |
2040                 (1ull << IB_USER_VERBS_CMD_CREATE_SRQ)          |
2041                 (1ull << IB_USER_VERBS_CMD_MODIFY_SRQ)          |
2042                 (1ull << IB_USER_VERBS_CMD_QUERY_SRQ)           |
2043                 (1ull << IB_USER_VERBS_CMD_DESTROY_SRQ)         |
2044                 (1ull << IB_USER_VERBS_CMD_CREATE_XSRQ)         |
2045                 (1ull << IB_USER_VERBS_CMD_OPEN_QP);
2046
2047         ibdev->ib_dev.query_device      = mlx4_ib_query_device;
2048         ibdev->ib_dev.query_port        = mlx4_ib_query_port;
2049         ibdev->ib_dev.get_link_layer    = mlx4_ib_port_link_layer;
2050         ibdev->ib_dev.query_gid         = mlx4_ib_query_gid;
2051         ibdev->ib_dev.query_pkey        = mlx4_ib_query_pkey;
2052         ibdev->ib_dev.modify_device     = mlx4_ib_modify_device;
2053         ibdev->ib_dev.modify_port       = mlx4_ib_modify_port;
2054         ibdev->ib_dev.alloc_ucontext    = mlx4_ib_alloc_ucontext;
2055         ibdev->ib_dev.dealloc_ucontext  = mlx4_ib_dealloc_ucontext;
2056         ibdev->ib_dev.mmap              = mlx4_ib_mmap;
2057         ibdev->ib_dev.alloc_pd          = mlx4_ib_alloc_pd;
2058         ibdev->ib_dev.dealloc_pd        = mlx4_ib_dealloc_pd;
2059         ibdev->ib_dev.create_ah         = mlx4_ib_create_ah;
2060         ibdev->ib_dev.query_ah          = mlx4_ib_query_ah;
2061         ibdev->ib_dev.destroy_ah        = mlx4_ib_destroy_ah;
2062         ibdev->ib_dev.create_srq        = mlx4_ib_create_srq;
2063         ibdev->ib_dev.modify_srq        = mlx4_ib_modify_srq;
2064         ibdev->ib_dev.query_srq         = mlx4_ib_query_srq;
2065         ibdev->ib_dev.destroy_srq       = mlx4_ib_destroy_srq;
2066         ibdev->ib_dev.post_srq_recv     = mlx4_ib_post_srq_recv;
2067         ibdev->ib_dev.create_qp         = mlx4_ib_create_qp;
2068         ibdev->ib_dev.modify_qp         = mlx4_ib_modify_qp;
2069         ibdev->ib_dev.query_qp          = mlx4_ib_query_qp;
2070         ibdev->ib_dev.destroy_qp        = mlx4_ib_destroy_qp;
2071         ibdev->ib_dev.post_send         = mlx4_ib_post_send;
2072         ibdev->ib_dev.post_recv         = mlx4_ib_post_recv;
2073         ibdev->ib_dev.create_cq         = mlx4_ib_create_cq;
2074         ibdev->ib_dev.modify_cq         = mlx4_ib_modify_cq;
2075         ibdev->ib_dev.resize_cq         = mlx4_ib_resize_cq;
2076         ibdev->ib_dev.destroy_cq        = mlx4_ib_destroy_cq;
2077         ibdev->ib_dev.poll_cq           = mlx4_ib_poll_cq;
2078         ibdev->ib_dev.req_notify_cq     = mlx4_ib_arm_cq;
2079         ibdev->ib_dev.get_dma_mr        = mlx4_ib_get_dma_mr;
2080         ibdev->ib_dev.reg_user_mr       = mlx4_ib_reg_user_mr;
2081         ibdev->ib_dev.dereg_mr          = mlx4_ib_dereg_mr;
2082         ibdev->ib_dev.alloc_fast_reg_mr = mlx4_ib_alloc_fast_reg_mr;
2083         ibdev->ib_dev.alloc_fast_reg_page_list = mlx4_ib_alloc_fast_reg_page_list;
2084         ibdev->ib_dev.free_fast_reg_page_list  = mlx4_ib_free_fast_reg_page_list;
2085         ibdev->ib_dev.attach_mcast      = mlx4_ib_mcg_attach;
2086         ibdev->ib_dev.detach_mcast      = mlx4_ib_mcg_detach;
2087         ibdev->ib_dev.process_mad       = mlx4_ib_process_mad;
2088
2089         if (!mlx4_is_slave(ibdev->dev)) {
2090                 ibdev->ib_dev.alloc_fmr         = mlx4_ib_fmr_alloc;
2091                 ibdev->ib_dev.map_phys_fmr      = mlx4_ib_map_phys_fmr;
2092                 ibdev->ib_dev.unmap_fmr         = mlx4_ib_unmap_fmr;
2093                 ibdev->ib_dev.dealloc_fmr       = mlx4_ib_fmr_dealloc;
2094         }
2095
2096         if (dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW ||
2097             dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
2098                 ibdev->ib_dev.alloc_mw = mlx4_ib_alloc_mw;
2099                 ibdev->ib_dev.bind_mw = mlx4_ib_bind_mw;
2100                 ibdev->ib_dev.dealloc_mw = mlx4_ib_dealloc_mw;
2101
2102                 ibdev->ib_dev.uverbs_cmd_mask |=
2103                         (1ull << IB_USER_VERBS_CMD_ALLOC_MW) |
2104                         (1ull << IB_USER_VERBS_CMD_DEALLOC_MW);
2105         }
2106
2107         if (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) {
2108                 ibdev->ib_dev.alloc_xrcd = mlx4_ib_alloc_xrcd;
2109                 ibdev->ib_dev.dealloc_xrcd = mlx4_ib_dealloc_xrcd;
2110                 ibdev->ib_dev.uverbs_cmd_mask |=
2111                         (1ull << IB_USER_VERBS_CMD_OPEN_XRCD) |
2112                         (1ull << IB_USER_VERBS_CMD_CLOSE_XRCD);
2113         }
2114
2115         if (check_flow_steering_support(dev)) {
2116                 ibdev->steering_support = MLX4_STEERING_MODE_DEVICE_MANAGED;
2117                 ibdev->ib_dev.create_flow       = mlx4_ib_create_flow;
2118                 ibdev->ib_dev.destroy_flow      = mlx4_ib_destroy_flow;
2119
2120                 ibdev->ib_dev.uverbs_ex_cmd_mask        |=
2121                         (1ull << IB_USER_VERBS_EX_CMD_CREATE_FLOW) |
2122                         (1ull << IB_USER_VERBS_EX_CMD_DESTROY_FLOW);
2123         }
2124
2125         mlx4_ib_alloc_eqs(dev, ibdev);
2126
2127         spin_lock_init(&iboe->lock);
2128
2129         if (init_node_data(ibdev))
2130                 goto err_map;
2131
2132         for (i = 0; i < ibdev->num_ports; ++i) {
2133                 mutex_init(&ibdev->qp1_proxy_lock[i]);
2134                 if (mlx4_ib_port_link_layer(&ibdev->ib_dev, i + 1) ==
2135                                                 IB_LINK_LAYER_ETHERNET) {
2136                         err = mlx4_counter_alloc(ibdev->dev, &ibdev->counters[i]);
2137                         if (err)
2138                                 ibdev->counters[i] = -1;
2139                 } else {
2140                         ibdev->counters[i] = -1;
2141                 }
2142         }
2143
2144         mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
2145                 ib_num_ports++;
2146
2147         spin_lock_init(&ibdev->sm_lock);
2148         mutex_init(&ibdev->cap_mask_mutex);
2149
2150         if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED &&
2151             ib_num_ports) {
2152                 ibdev->steer_qpn_count = MLX4_IB_UC_MAX_NUM_QPS;
2153                 err = mlx4_qp_reserve_range(dev, ibdev->steer_qpn_count,
2154                                             MLX4_IB_UC_STEER_QPN_ALIGN,
2155                                             &ibdev->steer_qpn_base);
2156                 if (err)
2157                         goto err_counter;
2158
2159                 ibdev->ib_uc_qpns_bitmap =
2160                         kmalloc(BITS_TO_LONGS(ibdev->steer_qpn_count) *
2161                                 sizeof(long),
2162                                 GFP_KERNEL);
2163                 if (!ibdev->ib_uc_qpns_bitmap) {
2164                         dev_err(&dev->pdev->dev, "bit map alloc failed\n");
2165                         goto err_steer_qp_release;
2166                 }
2167
2168                 if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DMFS_IPOIB) {
2169                         bitmap_zero(ibdev->ib_uc_qpns_bitmap,
2170                                     ibdev->steer_qpn_count);
2171                         err = mlx4_FLOW_STEERING_IB_UC_QP_RANGE(
2172                                         dev, ibdev->steer_qpn_base,
2173                                         ibdev->steer_qpn_base +
2174                                         ibdev->steer_qpn_count - 1);
2175                         if (err)
2176                                 goto err_steer_free_bitmap;
2177                 } else {
2178                         bitmap_fill(ibdev->ib_uc_qpns_bitmap,
2179                                     ibdev->steer_qpn_count);
2180                 }
2181         }
2182
2183         if (ib_register_device(&ibdev->ib_dev, NULL))
2184                 goto err_steer_free_bitmap;
2185
2186         if (mlx4_ib_mad_init(ibdev))
2187                 goto err_reg;
2188
2189         if (mlx4_ib_init_sriov(ibdev))
2190                 goto err_mad;
2191
2192         if (dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE) {
2193                 if (!iboe->nb.notifier_call) {
2194                         iboe->nb.notifier_call = mlx4_ib_netdev_event;
2195                         err = register_netdevice_notifier(&iboe->nb);
2196                         if (err) {
2197                                 iboe->nb.notifier_call = NULL;
2198                                 goto err_notif;
2199                         }
2200                 }
2201                 if (!iboe->nb_inet.notifier_call) {
2202                         iboe->nb_inet.notifier_call = mlx4_ib_inet_event;
2203                         err = register_inetaddr_notifier(&iboe->nb_inet);
2204                         if (err) {
2205                                 iboe->nb_inet.notifier_call = NULL;
2206                                 goto err_notif;
2207                         }
2208                 }
2209 #if IS_ENABLED(CONFIG_IPV6)
2210                 if (!iboe->nb_inet6.notifier_call) {
2211                         iboe->nb_inet6.notifier_call = mlx4_ib_inet6_event;
2212                         err = register_inet6addr_notifier(&iboe->nb_inet6);
2213                         if (err) {
2214                                 iboe->nb_inet6.notifier_call = NULL;
2215                                 goto err_notif;
2216                         }
2217                 }
2218 #endif
2219                 for (i = 1 ; i <= ibdev->num_ports ; ++i)
2220                         reset_gid_table(ibdev, i);
2221                 rtnl_lock();
2222                 mlx4_ib_scan_netdevs(ibdev, NULL, 0);
2223                 rtnl_unlock();
2224                 mlx4_ib_init_gid_table(ibdev);
2225         }
2226
2227         for (j = 0; j < ARRAY_SIZE(mlx4_class_attributes); ++j) {
2228                 if (device_create_file(&ibdev->ib_dev.dev,
2229                                        mlx4_class_attributes[j]))
2230                         goto err_notif;
2231         }
2232
2233         ibdev->ib_active = true;
2234
2235         if (mlx4_is_mfunc(ibdev->dev))
2236                 init_pkeys(ibdev);
2237
2238         /* create paravirt contexts for any VFs which are active */
2239         if (mlx4_is_master(ibdev->dev)) {
2240                 for (j = 0; j < MLX4_MFUNC_MAX; j++) {
2241                         if (j == mlx4_master_func_num(ibdev->dev))
2242                                 continue;
2243                         if (mlx4_is_slave_active(ibdev->dev, j))
2244                                 do_slave_init(ibdev, j, 1);
2245                 }
2246         }
2247         return ibdev;
2248
2249 err_notif:
2250         if (ibdev->iboe.nb.notifier_call) {
2251                 if (unregister_netdevice_notifier(&ibdev->iboe.nb))
2252                         pr_warn("failure unregistering notifier\n");
2253                 ibdev->iboe.nb.notifier_call = NULL;
2254         }
2255         if (ibdev->iboe.nb_inet.notifier_call) {
2256                 if (unregister_inetaddr_notifier(&ibdev->iboe.nb_inet))
2257                         pr_warn("failure unregistering notifier\n");
2258                 ibdev->iboe.nb_inet.notifier_call = NULL;
2259         }
2260 #if IS_ENABLED(CONFIG_IPV6)
2261         if (ibdev->iboe.nb_inet6.notifier_call) {
2262                 if (unregister_inet6addr_notifier(&ibdev->iboe.nb_inet6))
2263                         pr_warn("failure unregistering notifier\n");
2264                 ibdev->iboe.nb_inet6.notifier_call = NULL;
2265         }
2266 #endif
2267         flush_workqueue(wq);
2268
2269         mlx4_ib_close_sriov(ibdev);
2270
2271 err_mad:
2272         mlx4_ib_mad_cleanup(ibdev);
2273
2274 err_reg:
2275         ib_unregister_device(&ibdev->ib_dev);
2276
2277 err_steer_free_bitmap:
2278         kfree(ibdev->ib_uc_qpns_bitmap);
2279
2280 err_steer_qp_release:
2281         mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
2282                               ibdev->steer_qpn_count);
2283 err_counter:
2284         for (; i; --i)
2285                 if (ibdev->counters[i - 1] != -1)
2286                         mlx4_counter_free(ibdev->dev, ibdev->counters[i - 1]);
2287
2288 err_map:
2289         mlx4_ib_free_eqs(dev, ibdev);
2290         iounmap(ibdev->uar_map);
2291
2292 err_uar:
2293         mlx4_uar_free(dev, &ibdev->priv_uar);
2294
2295 err_pd:
2296         mlx4_pd_free(dev, ibdev->priv_pdn);
2297
2298 err_dealloc:
2299         ib_dealloc_device(&ibdev->ib_dev);
2300
2301         return NULL;
2302 }
2303
2304 int mlx4_ib_steer_qp_alloc(struct mlx4_ib_dev *dev, int count, int *qpn)
2305 {
2306         int offset;
2307
2308         WARN_ON(!dev->ib_uc_qpns_bitmap);
2309
2310         offset = bitmap_find_free_region(dev->ib_uc_qpns_bitmap,
2311                                          dev->steer_qpn_count,
2312                                          get_count_order(count));
2313         if (offset < 0)
2314                 return offset;
2315
2316         *qpn = dev->steer_qpn_base + offset;
2317         return 0;
2318 }
2319
2320 void mlx4_ib_steer_qp_free(struct mlx4_ib_dev *dev, u32 qpn, int count)
2321 {
2322         if (!qpn ||
2323             dev->steering_support != MLX4_STEERING_MODE_DEVICE_MANAGED)
2324                 return;
2325
2326         BUG_ON(qpn < dev->steer_qpn_base);
2327
2328         bitmap_release_region(dev->ib_uc_qpns_bitmap,
2329                               qpn - dev->steer_qpn_base,
2330                               get_count_order(count));
2331 }
2332
2333 int mlx4_ib_steer_qp_reg(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
2334                          int is_attach)
2335 {
2336         int err;
2337         size_t flow_size;
2338         struct ib_flow_attr *flow = NULL;
2339         struct ib_flow_spec_ib *ib_spec;
2340
2341         if (is_attach) {
2342                 flow_size = sizeof(struct ib_flow_attr) +
2343                             sizeof(struct ib_flow_spec_ib);
2344                 flow = kzalloc(flow_size, GFP_KERNEL);
2345                 if (!flow)
2346                         return -ENOMEM;
2347                 flow->port = mqp->port;
2348                 flow->num_of_specs = 1;
2349                 flow->size = flow_size;
2350                 ib_spec = (struct ib_flow_spec_ib *)(flow + 1);
2351                 ib_spec->type = IB_FLOW_SPEC_IB;
2352                 ib_spec->size = sizeof(struct ib_flow_spec_ib);
2353                 /* Add an empty rule for IB L2 */
2354                 memset(&ib_spec->mask, 0, sizeof(ib_spec->mask));
2355
2356                 err = __mlx4_ib_create_flow(&mqp->ibqp, flow,
2357                                             IB_FLOW_DOMAIN_NIC,
2358                                             MLX4_FS_REGULAR,
2359                                             &mqp->reg_id);
2360         } else {
2361                 err = __mlx4_ib_destroy_flow(mdev->dev, mqp->reg_id);
2362         }
2363         kfree(flow);
2364         return err;
2365 }
2366
2367 static void mlx4_ib_remove(struct mlx4_dev *dev, void *ibdev_ptr)
2368 {
2369         struct mlx4_ib_dev *ibdev = ibdev_ptr;
2370         int p;
2371
2372         ibdev->ib_active = false;
2373         flush_workqueue(wq);
2374
2375         if (ibdev->iboe.nb.notifier_call) {
2376                 if (unregister_netdevice_notifier(&ibdev->iboe.nb))
2377                         pr_warn("failure unregistering notifier\n");
2378                 ibdev->iboe.nb.notifier_call = NULL;
2379         }
2380
2381         mlx4_ib_close_sriov(ibdev);
2382         mlx4_ib_mad_cleanup(ibdev);
2383         ib_unregister_device(&ibdev->ib_dev);
2384
2385         mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
2386                               ibdev->steer_qpn_count);
2387         kfree(ibdev->ib_uc_qpns_bitmap);
2388
2389         if (ibdev->iboe.nb_inet.notifier_call) {
2390                 if (unregister_inetaddr_notifier(&ibdev->iboe.nb_inet))
2391                         pr_warn("failure unregistering notifier\n");
2392                 ibdev->iboe.nb_inet.notifier_call = NULL;
2393         }
2394 #if IS_ENABLED(CONFIG_IPV6)
2395         if (ibdev->iboe.nb_inet6.notifier_call) {
2396                 if (unregister_inet6addr_notifier(&ibdev->iboe.nb_inet6))
2397                         pr_warn("failure unregistering notifier\n");
2398                 ibdev->iboe.nb_inet6.notifier_call = NULL;
2399         }
2400 #endif
2401
2402         iounmap(ibdev->uar_map);
2403         for (p = 0; p < ibdev->num_ports; ++p)
2404                 if (ibdev->counters[p] != -1)
2405                         mlx4_counter_free(ibdev->dev, ibdev->counters[p]);
2406         mlx4_foreach_port(p, dev, MLX4_PORT_TYPE_IB)
2407                 mlx4_CLOSE_PORT(dev, p);
2408
2409         mlx4_ib_free_eqs(dev, ibdev);
2410
2411         mlx4_uar_free(dev, &ibdev->priv_uar);
2412         mlx4_pd_free(dev, ibdev->priv_pdn);
2413         ib_dealloc_device(&ibdev->ib_dev);
2414 }
2415
2416 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init)
2417 {
2418         struct mlx4_ib_demux_work **dm = NULL;
2419         struct mlx4_dev *dev = ibdev->dev;
2420         int i;
2421         unsigned long flags;
2422         struct mlx4_active_ports actv_ports;
2423         unsigned int ports;
2424         unsigned int first_port;
2425
2426         if (!mlx4_is_master(dev))
2427                 return;
2428
2429         actv_ports = mlx4_get_active_ports(dev, slave);
2430         ports = bitmap_weight(actv_ports.ports, dev->caps.num_ports);
2431         first_port = find_first_bit(actv_ports.ports, dev->caps.num_ports);
2432
2433         dm = kcalloc(ports, sizeof(*dm), GFP_ATOMIC);
2434         if (!dm) {
2435                 pr_err("failed to allocate memory for tunneling qp update\n");
2436                 goto out;
2437         }
2438
2439         for (i = 0; i < ports; i++) {
2440                 dm[i] = kmalloc(sizeof (struct mlx4_ib_demux_work), GFP_ATOMIC);
2441                 if (!dm[i]) {
2442                         pr_err("failed to allocate memory for tunneling qp update work struct\n");
2443                         for (i = 0; i < dev->caps.num_ports; i++) {
2444                                 if (dm[i])
2445                                         kfree(dm[i]);
2446                         }
2447                         goto out;
2448                 }
2449         }
2450         /* initialize or tear down tunnel QPs for the slave */
2451         for (i = 0; i < ports; i++) {
2452                 INIT_WORK(&dm[i]->work, mlx4_ib_tunnels_update_work);
2453                 dm[i]->port = first_port + i + 1;
2454                 dm[i]->slave = slave;
2455                 dm[i]->do_init = do_init;
2456                 dm[i]->dev = ibdev;
2457                 spin_lock_irqsave(&ibdev->sriov.going_down_lock, flags);
2458                 if (!ibdev->sriov.is_going_down)
2459                         queue_work(ibdev->sriov.demux[i].ud_wq, &dm[i]->work);
2460                 spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
2461         }
2462 out:
2463         kfree(dm);
2464         return;
2465 }
2466
2467 static void mlx4_ib_event(struct mlx4_dev *dev, void *ibdev_ptr,
2468                           enum mlx4_dev_event event, unsigned long param)
2469 {
2470         struct ib_event ibev;
2471         struct mlx4_ib_dev *ibdev = to_mdev((struct ib_device *) ibdev_ptr);
2472         struct mlx4_eqe *eqe = NULL;
2473         struct ib_event_work *ew;
2474         int p = 0;
2475
2476         if (event == MLX4_DEV_EVENT_PORT_MGMT_CHANGE)
2477                 eqe = (struct mlx4_eqe *)param;
2478         else
2479                 p = (int) param;
2480
2481         switch (event) {
2482         case MLX4_DEV_EVENT_PORT_UP:
2483                 if (p > ibdev->num_ports)
2484                         return;
2485                 if (mlx4_is_master(dev) &&
2486                     rdma_port_get_link_layer(&ibdev->ib_dev, p) ==
2487                         IB_LINK_LAYER_INFINIBAND) {
2488                         mlx4_ib_invalidate_all_guid_record(ibdev, p);
2489                 }
2490                 ibev.event = IB_EVENT_PORT_ACTIVE;
2491                 break;
2492
2493         case MLX4_DEV_EVENT_PORT_DOWN:
2494                 if (p > ibdev->num_ports)
2495                         return;
2496                 ibev.event = IB_EVENT_PORT_ERR;
2497                 break;
2498
2499         case MLX4_DEV_EVENT_CATASTROPHIC_ERROR:
2500                 ibdev->ib_active = false;
2501                 ibev.event = IB_EVENT_DEVICE_FATAL;
2502                 break;
2503
2504         case MLX4_DEV_EVENT_PORT_MGMT_CHANGE:
2505                 ew = kmalloc(sizeof *ew, GFP_ATOMIC);
2506                 if (!ew) {
2507                         pr_err("failed to allocate memory for events work\n");
2508                         break;
2509                 }
2510
2511                 INIT_WORK(&ew->work, handle_port_mgmt_change_event);
2512                 memcpy(&ew->ib_eqe, eqe, sizeof *eqe);
2513                 ew->ib_dev = ibdev;
2514                 /* need to queue only for port owner, which uses GEN_EQE */
2515                 if (mlx4_is_master(dev))
2516                         queue_work(wq, &ew->work);
2517                 else
2518                         handle_port_mgmt_change_event(&ew->work);
2519                 return;
2520
2521         case MLX4_DEV_EVENT_SLAVE_INIT:
2522                 /* here, p is the slave id */
2523                 do_slave_init(ibdev, p, 1);
2524                 return;
2525
2526         case MLX4_DEV_EVENT_SLAVE_SHUTDOWN:
2527                 /* here, p is the slave id */
2528                 do_slave_init(ibdev, p, 0);
2529                 return;
2530
2531         default:
2532                 return;
2533         }
2534
2535         ibev.device           = ibdev_ptr;
2536         ibev.element.port_num = (u8) p;
2537
2538         ib_dispatch_event(&ibev);
2539 }
2540
2541 static struct mlx4_interface mlx4_ib_interface = {
2542         .add            = mlx4_ib_add,
2543         .remove         = mlx4_ib_remove,
2544         .event          = mlx4_ib_event,
2545         .protocol       = MLX4_PROT_IB_IPV6
2546 };
2547
2548 static int __init mlx4_ib_init(void)
2549 {
2550         int err;
2551
2552         wq = create_singlethread_workqueue("mlx4_ib");
2553         if (!wq)
2554                 return -ENOMEM;
2555
2556         err = mlx4_ib_mcg_init();
2557         if (err)
2558                 goto clean_wq;
2559
2560         err = mlx4_register_interface(&mlx4_ib_interface);
2561         if (err)
2562                 goto clean_mcg;
2563
2564         return 0;
2565
2566 clean_mcg:
2567         mlx4_ib_mcg_destroy();
2568
2569 clean_wq:
2570         destroy_workqueue(wq);
2571         return err;
2572 }
2573
2574 static void __exit mlx4_ib_cleanup(void)
2575 {
2576         mlx4_unregister_interface(&mlx4_ib_interface);
2577         mlx4_ib_mcg_destroy();
2578         destroy_workqueue(wq);
2579 }
2580
2581 module_init(mlx4_ib_init);
2582 module_exit(mlx4_ib_cleanup);