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 #define MLX4_IB_CARD_REV_A0   0xA0
  63
  64 MODULE_AUTHOR("Roland Dreier");
  65 MODULE_DESCRIPTION("Mellanox ConnectX HCA InfiniBand driver");
  66 MODULE_LICENSE("Dual BSD/GPL");
  67 MODULE_VERSION(DRV_VERSION);
  68
  69 int mlx4_ib_sm_guid_assign = 1;
  70 module_param_named(sm_guid_assign, mlx4_ib_sm_guid_assign, int, 0444);
  71 MODULE_PARM_DESC(sm_guid_assign, "Enable SM alias_GUID assignment if sm_guid_assign > 0 (Default: 1)");
  72
  73 static const char mlx4_ib_version[] =
  74         DRV_NAME ": Mellanox ConnectX InfiniBand driver v"
  75         DRV_VERSION " (" DRV_RELDATE ")\n";
  76
  77 struct update_gid_work {
  78         struct work_struct      work;
  79         union ib_gid            gids[128];
  80         struct mlx4_ib_dev     *dev;
  81         int                     port;
  82 };
  83
  84 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init);
  85
  86 static struct workqueue_struct *wq;
  87
  88 static void init_query_mad(struct ib_smp *mad)
  89 {
  90         mad->base_version  = 1;
  91         mad->mgmt_class    = IB_MGMT_CLASS_SUBN_LID_ROUTED;
  92         mad->class_version = 1;
  93         mad->method        = IB_MGMT_METHOD_GET;
  94 }
  95
  96 static union ib_gid zgid;
  97
  98 static int check_flow_steering_support(struct mlx4_dev *dev)
  99 {
 100         int eth_num_ports = 0;
 101         int ib_num_ports = 0;
 102
 103         int dmfs = dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED;
 104
 105         if (dmfs) {
 106                 int i;
 107                 mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_ETH)
 108                         eth_num_ports++;
 109                 mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
 110                         ib_num_ports++;
 111                 dmfs &= (!ib_num_ports ||
 112                          (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DMFS_IPOIB)) &&
 113                         (!eth_num_ports ||
 114                          (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_FS_EN));
 115                 if (ib_num_ports && mlx4_is_mfunc(dev)) {
 116                         pr_warn("Device managed flow steering is unavailable for IB port in multifunction env.\n");
 117                         dmfs = 0;
 118                 }
 119         }
 120         return dmfs;
 121 }
 122
 123 static int mlx4_ib_query_device(struct ib_device *ibdev,
 124                                 struct ib_device_attr *props)
 125 {
 126         struct mlx4_ib_dev *dev = to_mdev(ibdev);
 127         struct ib_smp *in_mad  = NULL;
 128         struct ib_smp *out_mad = NULL;
 129         int err = -ENOMEM;
 130
 131         in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
 132         out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
 133         if (!in_mad || !out_mad)
 134                 goto out;
 135
 136         init_query_mad(in_mad);
 137         in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
 138
 139         err = mlx4_MAD_IFC(to_mdev(ibdev), MLX4_MAD_IFC_IGNORE_KEYS,
 140                            1, NULL, NULL, in_mad, out_mad);
 141         if (err)
 142                 goto out;
 143
 144         memset(props, 0, sizeof *props);
 145
 146         props->fw_ver = dev->dev->caps.fw_ver;
 147         props->device_cap_flags    = IB_DEVICE_CHANGE_PHY_PORT |
 148                 IB_DEVICE_PORT_ACTIVE_EVENT             |
 149                 IB_DEVICE_SYS_IMAGE_GUID                |
 150                 IB_DEVICE_RC_RNR_NAK_GEN                |
 151                 IB_DEVICE_BLOCK_MULTICAST_LOOPBACK;
 152         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_PKEY_CNTR)
 153                 props->device_cap_flags |= IB_DEVICE_BAD_PKEY_CNTR;
 154         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_QKEY_CNTR)
 155                 props->device_cap_flags |= IB_DEVICE_BAD_QKEY_CNTR;
 156         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_APM)
 157                 props->device_cap_flags |= IB_DEVICE_AUTO_PATH_MIG;
 158         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_UD_AV_PORT)
 159                 props->device_cap_flags |= IB_DEVICE_UD_AV_PORT_ENFORCE;
 160         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_IPOIB_CSUM)
 161                 props->device_cap_flags |= IB_DEVICE_UD_IP_CSUM;
 162         if (dev->dev->caps.max_gso_sz &&
 163             (dev->dev->rev_id != MLX4_IB_CARD_REV_A0) &&
 164             (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BLH))
 165                 props->device_cap_flags |= IB_DEVICE_UD_TSO;
 166         if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_RESERVED_LKEY)
 167                 props->device_cap_flags |= IB_DEVICE_LOCAL_DMA_LKEY;
 168         if ((dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_LOCAL_INV) &&
 169             (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_REMOTE_INV) &&
 170             (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_FAST_REG_WR))
 171                 props->device_cap_flags |= IB_DEVICE_MEM_MGT_EXTENSIONS;
 172         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC)
 173                 props->device_cap_flags |= IB_DEVICE_XRC;
 174         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW)
 175                 props->device_cap_flags |= IB_DEVICE_MEM_WINDOW;
 176         if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
 177                 if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_WIN_TYPE_2B)
 178                         props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2B;
 179                 else
 180                         props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2A;
 181         if (dev->steering_support ==  MLX4_STEERING_MODE_DEVICE_MANAGED)
 182                 props->device_cap_flags |= IB_DEVICE_MANAGED_FLOW_STEERING;
 183         }
 184
 185         props->vendor_id           = be32_to_cpup((__be32 *) (out_mad->data + 36)) &
 186                 0xffffff;
 187         props->vendor_part_id      = dev->dev->pdev->device;
 188         props->hw_ver              = be32_to_cpup((__be32 *) (out_mad->data + 32));
 189         memcpy(&props->sys_image_guid, out_mad->data +  4, 8);
 190
 191         props->max_mr_size         = ~0ull;
 192         props->page_size_cap       = dev->dev->caps.page_size_cap;
 193         props->max_qp              = dev->dev->quotas.qp;
 194         props->max_qp_wr           = dev->dev->caps.max_wqes - MLX4_IB_SQ_MAX_SPARE;
 195         props->max_sge             = min(dev->dev->caps.max_sq_sg,
 196                                          dev->dev->caps.max_rq_sg);
 197         props->max_cq              = dev->dev->quotas.cq;
 198         props->max_cqe             = dev->dev->caps.max_cqes;
 199         props->max_mr              = dev->dev->quotas.mpt;
 200         props->max_pd              = dev->dev->caps.num_pds - dev->dev->caps.reserved_pds;
 201         props->max_qp_rd_atom      = dev->dev->caps.max_qp_dest_rdma;
 202         props->max_qp_init_rd_atom = dev->dev->caps.max_qp_init_rdma;
 203         props->max_res_rd_atom     = props->max_qp_rd_atom * props->max_qp;
 204         props->max_srq             = dev->dev->quotas.srq;
 205         props->max_srq_wr          = dev->dev->caps.max_srq_wqes - 1;
 206         props->max_srq_sge         = dev->dev->caps.max_srq_sge;
 207         props->max_fast_reg_page_list_len = MLX4_MAX_FAST_REG_PAGES;
 208         props->local_ca_ack_delay  = dev->dev->caps.local_ca_ack_delay;
 209         props->atomic_cap          = dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_ATOMIC ?
 210                 IB_ATOMIC_HCA : IB_ATOMIC_NONE;
 211         props->masked_atomic_cap   = props->atomic_cap;
 212         props->max_pkeys           = dev->dev->caps.pkey_table_len[1];
 213         props->max_mcast_grp       = dev->dev->caps.num_mgms + dev->dev->caps.num_amgms;
 214         props->max_mcast_qp_attach = dev->dev->caps.num_qp_per_mgm;
 215         props->max_total_mcast_qp_attach = props->max_mcast_qp_attach *
 216                                            props->max_mcast_grp;
 217         props->max_map_per_fmr = dev->dev->caps.max_fmr_maps;
 218
 219 out:
 220         kfree(in_mad);
 221         kfree(out_mad);
 222
 223         return err;
 224 }
 225
 226 static enum rdma_link_layer
 227 mlx4_ib_port_link_layer(struct ib_device *device, u8 port_num)
 228 {
 229         struct mlx4_dev *dev = to_mdev(device)->dev;
 230
 231         return dev->caps.port_mask[port_num] == MLX4_PORT_TYPE_IB ?
 232                 IB_LINK_LAYER_INFINIBAND : IB_LINK_LAYER_ETHERNET;
 233 }
 234
 235 static int ib_link_query_port(struct ib_device *ibdev, u8 port,
 236                               struct ib_port_attr *props, int netw_view)
 237 {
 238         struct ib_smp *in_mad  = NULL;
 239         struct ib_smp *out_mad = NULL;
 240         int ext_active_speed;
 241         int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
 242         int err = -ENOMEM;
 243
 244         in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
 245         out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
 246         if (!in_mad || !out_mad)
 247                 goto out;
 248
 249         init_query_mad(in_mad);
 250         in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
 251         in_mad->attr_mod = cpu_to_be32(port);
 252
 253         if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
 254                 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
 255
 256         err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
 257                                 in_mad, out_mad);
 258         if (err)
 259                 goto out;
 260
 261
 262         props->lid              = be16_to_cpup((__be16 *) (out_mad->data + 16));
 263         props->lmc              = out_mad->data[34] & 0x7;
 264         props->sm_lid           = be16_to_cpup((__be16 *) (out_mad->data + 18));
 265         props->sm_sl            = out_mad->data[36] & 0xf;
 266         props->state            = out_mad->data[32] & 0xf;
 267         props->phys_state       = out_mad->data[33] >> 4;
 268         props->port_cap_flags   = be32_to_cpup((__be32 *) (out_mad->data + 20));
 269         if (netw_view)
 270                 props->gid_tbl_len = out_mad->data[50];
 271         else
 272                 props->gid_tbl_len = to_mdev(ibdev)->dev->caps.gid_table_len[port];
 273         props->max_msg_sz       = to_mdev(ibdev)->dev->caps.max_msg_sz;
 274         props->pkey_tbl_len     = to_mdev(ibdev)->dev->caps.pkey_table_len[port];
 275         props->bad_pkey_cntr    = be16_to_cpup((__be16 *) (out_mad->data + 46));
 276         props->qkey_viol_cntr   = be16_to_cpup((__be16 *) (out_mad->data + 48));
 277         props->active_width     = out_mad->data[31] & 0xf;
 278         props->active_speed     = out_mad->data[35] >> 4;
 279         props->max_mtu          = out_mad->data[41] & 0xf;
 280         props->active_mtu       = out_mad->data[36] >> 4;
 281         props->subnet_timeout   = out_mad->data[51] & 0x1f;
 282         props->max_vl_num       = out_mad->data[37] >> 4;
 283         props->init_type_reply  = out_mad->data[41] >> 4;
 284
 285         /* Check if extended speeds (EDR/FDR/...) are supported */
 286         if (props->port_cap_flags & IB_PORT_EXTENDED_SPEEDS_SUP) {
 287                 ext_active_speed = out_mad->data[62] >> 4;
 288
 289                 switch (ext_active_speed) {
 290                 case 1:
 291                         props->active_speed = IB_SPEED_FDR;
 292                         break;
 293                 case 2:
 294                         props->active_speed = IB_SPEED_EDR;
 295                         break;
 296                 }
 297         }
 298
 299         /* If reported active speed is QDR, check if is FDR-10 */
 300         if (props->active_speed == IB_SPEED_QDR) {
 301                 init_query_mad(in_mad);
 302                 in_mad->attr_id = MLX4_ATTR_EXTENDED_PORT_INFO;
 303                 in_mad->attr_mod = cpu_to_be32(port);
 304
 305                 err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port,
 306                                    NULL, NULL, in_mad, out_mad);
 307                 if (err)
 308                         goto out;
 309
 310                 /* Checking LinkSpeedActive for FDR-10 */
 311                 if (out_mad->data[15] & 0x1)
 312                         props->active_speed = IB_SPEED_FDR10;
 313         }
 314
 315         /* Avoid wrong speed value returned by FW if the IB link is down. */
 316         if (props->state == IB_PORT_DOWN)
 317                  props->active_speed = IB_SPEED_SDR;
 318
 319 out:
 320         kfree(in_mad);
 321         kfree(out_mad);
 322         return err;
 323 }
 324
 325 static u8 state_to_phys_state(enum ib_port_state state)
 326 {
 327         return state == IB_PORT_ACTIVE ? 5 : 3;
 328 }
 329
 330 static int eth_link_query_port(struct ib_device *ibdev, u8 port,
 331                                struct ib_port_attr *props, int netw_view)
 332 {
 333
 334         struct mlx4_ib_dev *mdev = to_mdev(ibdev);
 335         struct mlx4_ib_iboe *iboe = &mdev->iboe;
 336         struct net_device *ndev;
 337         enum ib_mtu tmp;
 338         struct mlx4_cmd_mailbox *mailbox;
 339         int err = 0;
 340
 341         mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
 342         if (IS_ERR(mailbox))
 343                 return PTR_ERR(mailbox);
 344
 345         err = mlx4_cmd_box(mdev->dev, 0, mailbox->dma, port, 0,
 346                            MLX4_CMD_QUERY_PORT, MLX4_CMD_TIME_CLASS_B,
 347                            MLX4_CMD_WRAPPED);
 348         if (err)
 349                 goto out;
 350
 351         props->active_width     =  (((u8 *)mailbox->buf)[5] == 0x40) ?
 352                                                 IB_WIDTH_4X : IB_WIDTH_1X;
 353         props->active_speed     = 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 < ARRAY_SIZE(default_table); i++, pdefault_rules++) {
 917                 __u32 field_types[IB_FLOW_SPEC_SUPPORT_LAYERS];
 918                 memset(&field_types, 0, sizeof(field_types));
 919
 920                 if (link_layer != pdefault_rules->link_layer)
 921                         continue;
 922
 923                 ib_flow = flow_attr + 1;
 924                 /* we assume the specs are sorted */
 925                 for (j = 0, k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS &&
 926                      j < flow_attr->num_of_specs; k++) {
 927                         union ib_flow_spec *current_flow =
 928                                 (union ib_flow_spec *)ib_flow;
 929
 930                         /* same layer but different type */
 931                         if (((current_flow->type & IB_FLOW_SPEC_LAYER_MASK) ==
 932                              (pdefault_rules->mandatory_fields[k] &
 933                               IB_FLOW_SPEC_LAYER_MASK)) &&
 934                             (current_flow->type !=
 935                              pdefault_rules->mandatory_fields[k]))
 936                                 goto out;
 937
 938                         /* same layer, try match next one */
 939                         if (current_flow->type ==
 940                             pdefault_rules->mandatory_fields[k]) {
 941                                 j++;
 942                                 ib_flow +=
 943                                         ((union ib_flow_spec *)ib_flow)->size;
 944                         }
 945                 }
 946
 947                 ib_flow = flow_attr + 1;
 948                 for (j = 0; j < flow_attr->num_of_specs;
 949                      j++, ib_flow += ((union ib_flow_spec *)ib_flow)->size)
 950                         for (k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS; k++)
 951                                 /* same layer and same type */
 952                                 if (((union ib_flow_spec *)ib_flow)->type ==
 953                                     pdefault_rules->mandatory_not_fields[k])
 954                                         goto out;
 955
 956                 return i;
 957         }
 958 out:
 959         return -1;
 960 }
 961
 962 static int __mlx4_ib_create_default_rules(
 963                 struct mlx4_ib_dev *mdev,
 964                 struct ib_qp *qp,
 965                 const struct default_rules *pdefault_rules,
 966                 struct _rule_hw *mlx4_spec) {
 967         int size = 0;
 968         int i;
 969
 970         for (i = 0; i < ARRAY_SIZE(pdefault_rules->rules_create_list); i++) {
 971                 int ret;
 972                 union ib_flow_spec ib_spec;
 973                 switch (pdefault_rules->rules_create_list[i]) {
 974                 case 0:
 975                         /* no rule */
 976                         continue;
 977                 case IB_FLOW_SPEC_IB:
 978                         ib_spec.type = IB_FLOW_SPEC_IB;
 979                         ib_spec.size = sizeof(struct ib_flow_spec_ib);
 980
 981                         break;
 982                 default:
 983                         /* invalid rule */
 984                         return -EINVAL;
 985                 }
 986                 /* We must put empty rule, qpn is being ignored */
 987                 ret = parse_flow_attr(mdev->dev, 0, &ib_spec,
 988                                       mlx4_spec);
 989                 if (ret < 0) {
 990                         pr_info("invalid parsing\n");
 991                         return -EINVAL;
 992                 }
 993
 994                 mlx4_spec = (void *)mlx4_spec + ret;
 995                 size += ret;
 996         }
 997         return size;
 998 }
 999
1000 static int __mlx4_ib_create_flow(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
1001                           int domain,
1002                           enum mlx4_net_trans_promisc_mode flow_type,
1003                           u64 *reg_id)
1004 {
1005         int ret, i;
1006         int size = 0;
1007         void *ib_flow;
1008         struct mlx4_ib_dev *mdev = to_mdev(qp->device);
1009         struct mlx4_cmd_mailbox *mailbox;
1010         struct mlx4_net_trans_rule_hw_ctrl *ctrl;
1011         int default_flow;
1012
1013         static const u16 __mlx4_domain[] = {
1014                 [IB_FLOW_DOMAIN_USER] = MLX4_DOMAIN_UVERBS,
1015                 [IB_FLOW_DOMAIN_ETHTOOL] = MLX4_DOMAIN_ETHTOOL,
1016                 [IB_FLOW_DOMAIN_RFS] = MLX4_DOMAIN_RFS,
1017                 [IB_FLOW_DOMAIN_NIC] = MLX4_DOMAIN_NIC,
1018         };
1019
1020         if (flow_attr->priority > MLX4_IB_FLOW_MAX_PRIO) {
1021                 pr_err("Invalid priority value %d\n", flow_attr->priority);
1022                 return -EINVAL;
1023         }
1024
1025         if (domain >= IB_FLOW_DOMAIN_NUM) {
1026                 pr_err("Invalid domain value %d\n", domain);
1027                 return -EINVAL;
1028         }
1029
1030         if (mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type) < 0)
1031                 return -EINVAL;
1032
1033         mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
1034         if (IS_ERR(mailbox))
1035                 return PTR_ERR(mailbox);
1036         ctrl = mailbox->buf;
1037
1038         ctrl->prio = cpu_to_be16(__mlx4_domain[domain] |
1039                                  flow_attr->priority);
1040         ctrl->type = mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type);
1041         ctrl->port = flow_attr->port;
1042         ctrl->qpn = cpu_to_be32(qp->qp_num);
1043
1044         ib_flow = flow_attr + 1;
1045         size += sizeof(struct mlx4_net_trans_rule_hw_ctrl);
1046         /* Add default flows */
1047         default_flow = __mlx4_ib_default_rules_match(qp, flow_attr);
1048         if (default_flow >= 0) {
1049                 ret = __mlx4_ib_create_default_rules(
1050                                 mdev, qp, default_table + default_flow,
1051                                 mailbox->buf + size);
1052                 if (ret < 0) {
1053                         mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1054                         return -EINVAL;
1055                 }
1056                 size += ret;
1057         }
1058         for (i = 0; i < flow_attr->num_of_specs; i++) {
1059                 ret = parse_flow_attr(mdev->dev, qp->qp_num, ib_flow,
1060                                       mailbox->buf + size);
1061                 if (ret < 0) {
1062                         mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1063                         return -EINVAL;
1064                 }
1065                 ib_flow += ((union ib_flow_spec *) ib_flow)->size;
1066                 size += ret;
1067         }
1068
1069         ret = mlx4_cmd_imm(mdev->dev, mailbox->dma, reg_id, size >> 2, 0,
1070                            MLX4_QP_FLOW_STEERING_ATTACH, MLX4_CMD_TIME_CLASS_A,
1071                            MLX4_CMD_NATIVE);
1072         if (ret == -ENOMEM)
1073                 pr_err("mcg table is full. Fail to register network rule.\n");
1074         else if (ret == -ENXIO)
1075                 pr_err("Device managed flow steering is disabled. Fail to register network rule.\n");
1076         else if (ret)
1077                 pr_err("Invalid argumant. Fail to register network rule.\n");
1078
1079         mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1080         return ret;
1081 }
1082
1083 static int __mlx4_ib_destroy_flow(struct mlx4_dev *dev, u64 reg_id)
1084 {
1085         int err;
1086         err = mlx4_cmd(dev, reg_id, 0, 0,
1087                        MLX4_QP_FLOW_STEERING_DETACH, MLX4_CMD_TIME_CLASS_A,
1088                        MLX4_CMD_NATIVE);
1089         if (err)
1090                 pr_err("Fail to detach network rule. registration id = 0x%llx\n",
1091                        reg_id);
1092         return err;
1093 }
1094
1095 static struct ib_flow *mlx4_ib_create_flow(struct ib_qp *qp,
1096                                     struct ib_flow_attr *flow_attr,
1097                                     int domain)
1098 {
1099         int err = 0, i = 0;
1100         struct mlx4_ib_flow *mflow;
1101         enum mlx4_net_trans_promisc_mode type[2];
1102
1103         memset(type, 0, sizeof(type));
1104
1105         mflow = kzalloc(sizeof(*mflow), GFP_KERNEL);
1106         if (!mflow) {
1107                 err = -ENOMEM;
1108                 goto err_free;
1109         }
1110
1111         switch (flow_attr->type) {
1112         case IB_FLOW_ATTR_NORMAL:
1113                 type[0] = MLX4_FS_REGULAR;
1114                 break;
1115
1116         case IB_FLOW_ATTR_ALL_DEFAULT:
1117                 type[0] = MLX4_FS_ALL_DEFAULT;
1118                 break;
1119
1120         case IB_FLOW_ATTR_MC_DEFAULT:
1121                 type[0] = MLX4_FS_MC_DEFAULT;
1122                 break;
1123
1124         case IB_FLOW_ATTR_SNIFFER:
1125                 type[0] = MLX4_FS_UC_SNIFFER;
1126                 type[1] = MLX4_FS_MC_SNIFFER;
1127                 break;
1128
1129         default:
1130                 err = -EINVAL;
1131                 goto err_free;
1132         }
1133
1134         while (i < ARRAY_SIZE(type) && type[i]) {
1135                 err = __mlx4_ib_create_flow(qp, flow_attr, domain, type[i],
1136                                             &mflow->reg_id[i]);
1137                 if (err)
1138                         goto err_free;
1139                 i++;
1140         }
1141
1142         return &mflow->ibflow;
1143
1144 err_free:
1145         kfree(mflow);
1146         return ERR_PTR(err);
1147 }
1148
1149 static int mlx4_ib_destroy_flow(struct ib_flow *flow_id)
1150 {
1151         int err, ret = 0;
1152         int i = 0;
1153         struct mlx4_ib_dev *mdev = to_mdev(flow_id->qp->device);
1154         struct mlx4_ib_flow *mflow = to_mflow(flow_id);
1155
1156         while (i < ARRAY_SIZE(mflow->reg_id) && mflow->reg_id[i]) {
1157                 err = __mlx4_ib_destroy_flow(mdev->dev, mflow->reg_id[i]);
1158                 if (err)
1159                         ret = err;
1160                 i++;
1161         }
1162
1163         kfree(mflow);
1164         return ret;
1165 }
1166
1167 static int mlx4_ib_mcg_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
1168 {
1169         int err;
1170         struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1171         struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1172         u64 reg_id;
1173         struct mlx4_ib_steering *ib_steering = NULL;
1174         enum mlx4_protocol prot = (gid->raw[1] == 0x0e) ?
1175                 MLX4_PROT_IB_IPV4 : MLX4_PROT_IB_IPV6;
1176
1177         if (mdev->dev->caps.steering_mode ==
1178             MLX4_STEERING_MODE_DEVICE_MANAGED) {
1179                 ib_steering = kmalloc(sizeof(*ib_steering), GFP_KERNEL);
1180                 if (!ib_steering)
1181                         return -ENOMEM;
1182         }
1183
1184         err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw, mqp->port,
1185                                     !!(mqp->flags &
1186                                        MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
1187                                     prot, &reg_id);
1188         if (err)
1189                 goto err_malloc;
1190
1191         err = add_gid_entry(ibqp, gid);
1192         if (err)
1193                 goto err_add;
1194
1195         if (ib_steering) {
1196                 memcpy(ib_steering->gid.raw, gid->raw, 16);
1197                 ib_steering->reg_id = reg_id;
1198                 mutex_lock(&mqp->mutex);
1199                 list_add(&ib_steering->list, &mqp->steering_rules);
1200                 mutex_unlock(&mqp->mutex);
1201         }
1202         return 0;
1203
1204 err_add:
1205         mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1206                               prot, reg_id);
1207 err_malloc:
1208         kfree(ib_steering);
1209
1210         return err;
1211 }
1212
1213 static struct mlx4_ib_gid_entry *find_gid_entry(struct mlx4_ib_qp *qp, u8 *raw)
1214 {
1215         struct mlx4_ib_gid_entry *ge;
1216         struct mlx4_ib_gid_entry *tmp;
1217         struct mlx4_ib_gid_entry *ret = NULL;
1218
1219         list_for_each_entry_safe(ge, tmp, &qp->gid_list, list) {
1220                 if (!memcmp(raw, ge->gid.raw, 16)) {
1221                         ret = ge;
1222                         break;
1223                 }
1224         }
1225
1226         return ret;
1227 }
1228
1229 static int mlx4_ib_mcg_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
1230 {
1231         int err;
1232         struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1233         struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1234         struct net_device *ndev;
1235         struct mlx4_ib_gid_entry *ge;
1236         u64 reg_id = 0;
1237         enum mlx4_protocol prot = (gid->raw[1] == 0x0e) ?
1238                 MLX4_PROT_IB_IPV4 : MLX4_PROT_IB_IPV6;
1239
1240         if (mdev->dev->caps.steering_mode ==
1241             MLX4_STEERING_MODE_DEVICE_MANAGED) {
1242                 struct mlx4_ib_steering *ib_steering;
1243
1244                 mutex_lock(&mqp->mutex);
1245                 list_for_each_entry(ib_steering, &mqp->steering_rules, list) {
1246                         if (!memcmp(ib_steering->gid.raw, gid->raw, 16)) {
1247                                 list_del(&ib_steering->list);
1248                                 break;
1249                         }
1250                 }
1251                 mutex_unlock(&mqp->mutex);
1252                 if (&ib_steering->list == &mqp->steering_rules) {
1253                         pr_err("Couldn't find reg_id for mgid. Steering rule is left attached\n");
1254                         return -EINVAL;
1255                 }
1256                 reg_id = ib_steering->reg_id;
1257                 kfree(ib_steering);
1258         }
1259
1260         err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1261                                     prot, reg_id);
1262         if (err)
1263                 return err;
1264
1265         mutex_lock(&mqp->mutex);
1266         ge = find_gid_entry(mqp, gid->raw);
1267         if (ge) {
1268                 spin_lock(&mdev->iboe.lock);
1269                 ndev = ge->added ? mdev->iboe.netdevs[ge->port - 1] : NULL;
1270                 if (ndev)
1271                         dev_hold(ndev);
1272                 spin_unlock(&mdev->iboe.lock);
1273                 if (ndev)
1274                         dev_put(ndev);
1275                 list_del(&ge->list);
1276                 kfree(ge);
1277         } else
1278                 pr_warn("could not find mgid entry\n");
1279
1280         mutex_unlock(&mqp->mutex);
1281
1282         return 0;
1283 }
1284
1285 static int init_node_data(struct mlx4_ib_dev *dev)
1286 {
1287         struct ib_smp *in_mad  = NULL;
1288         struct ib_smp *out_mad = NULL;
1289         int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
1290         int err = -ENOMEM;
1291
1292         in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
1293         out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
1294         if (!in_mad || !out_mad)
1295                 goto out;
1296
1297         init_query_mad(in_mad);
1298         in_mad->attr_id = IB_SMP_ATTR_NODE_DESC;
1299         if (mlx4_is_master(dev->dev))
1300                 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
1301
1302         err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
1303         if (err)
1304                 goto out;
1305
1306         memcpy(dev->ib_dev.node_desc, out_mad->data, 64);
1307
1308         in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
1309
1310         err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
1311         if (err)
1312                 goto out;
1313
1314         dev->dev->rev_id = be32_to_cpup((__be32 *) (out_mad->data + 32));
1315         memcpy(&dev->ib_dev.node_guid, out_mad->data + 12, 8);
1316
1317 out:
1318         kfree(in_mad);
1319         kfree(out_mad);
1320         return err;
1321 }
1322
1323 static ssize_t show_hca(struct device *device, struct device_attribute *attr,
1324                         char *buf)
1325 {
1326         struct mlx4_ib_dev *dev =
1327                 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1328         return sprintf(buf, "MT%d\n", dev->dev->pdev->device);
1329 }
1330
1331 static ssize_t show_fw_ver(struct device *device, struct device_attribute *attr,
1332                            char *buf)
1333 {
1334         struct mlx4_ib_dev *dev =
1335                 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1336         return sprintf(buf, "%d.%d.%d\n", (int) (dev->dev->caps.fw_ver >> 32),
1337                        (int) (dev->dev->caps.fw_ver >> 16) & 0xffff,
1338                        (int) dev->dev->caps.fw_ver & 0xffff);
1339 }
1340
1341 static ssize_t show_rev(struct device *device, struct device_attribute *attr,
1342                         char *buf)
1343 {
1344         struct mlx4_ib_dev *dev =
1345                 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1346         return sprintf(buf, "%x\n", dev->dev->rev_id);
1347 }
1348
1349 static ssize_t show_board(struct device *device, struct device_attribute *attr,
1350                           char *buf)
1351 {
1352         struct mlx4_ib_dev *dev =
1353                 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1354         return sprintf(buf, "%.*s\n", MLX4_BOARD_ID_LEN,
1355                        dev->dev->board_id);
1356 }
1357
1358 static DEVICE_ATTR(hw_rev,   S_IRUGO, show_rev,    NULL);
1359 static DEVICE_ATTR(fw_ver,   S_IRUGO, show_fw_ver, NULL);
1360 static DEVICE_ATTR(hca_type, S_IRUGO, show_hca,    NULL);
1361 static DEVICE_ATTR(board_id, S_IRUGO, show_board,  NULL);
1362
1363 static struct device_attribute *mlx4_class_attributes[] = {
1364         &dev_attr_hw_rev,
1365         &dev_attr_fw_ver,
1366         &dev_attr_hca_type,
1367         &dev_attr_board_id
1368 };
1369
1370 static void mlx4_addrconf_ifid_eui48(u8 *eui, u16 vlan_id,
1371                                      struct net_device *dev)
1372 {
1373         memcpy(eui, dev->dev_addr, 3);
1374         memcpy(eui + 5, dev->dev_addr + 3, 3);
1375         if (vlan_id < 0x1000) {
1376                 eui[3] = vlan_id >> 8;
1377                 eui[4] = vlan_id & 0xff;
1378         } else {
1379                 eui[3] = 0xff;
1380                 eui[4] = 0xfe;
1381         }
1382         eui[0] ^= 2;
1383 }
1384
1385 static void update_gids_task(struct work_struct *work)
1386 {
1387         struct update_gid_work *gw = container_of(work, struct update_gid_work, work);
1388         struct mlx4_cmd_mailbox *mailbox;
1389         union ib_gid *gids;
1390         int err;
1391         struct mlx4_dev *dev = gw->dev->dev;
1392
1393         mailbox = mlx4_alloc_cmd_mailbox(dev);
1394         if (IS_ERR(mailbox)) {
1395                 pr_warn("update gid table failed %ld\n", PTR_ERR(mailbox));
1396                 return;
1397         }
1398
1399         gids = mailbox->buf;
1400         memcpy(gids, gw->gids, sizeof gw->gids);
1401
1402         err = mlx4_cmd(dev, mailbox->dma, MLX4_SET_PORT_GID_TABLE << 8 | gw->port,
1403                        1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
1404                        MLX4_CMD_WRAPPED);
1405         if (err)
1406                 pr_warn("set port command failed\n");
1407         else
1408                 mlx4_ib_dispatch_event(gw->dev, gw->port, IB_EVENT_GID_CHANGE);
1409
1410         mlx4_free_cmd_mailbox(dev, mailbox);
1411         kfree(gw);
1412 }
1413
1414 static void reset_gids_task(struct work_struct *work)
1415 {
1416         struct update_gid_work *gw =
1417                         container_of(work, struct update_gid_work, work);
1418         struct mlx4_cmd_mailbox *mailbox;
1419         union ib_gid *gids;
1420         int err;
1421         struct mlx4_dev *dev = gw->dev->dev;
1422
1423         mailbox = mlx4_alloc_cmd_mailbox(dev);
1424         if (IS_ERR(mailbox)) {
1425                 pr_warn("reset gid table failed\n");
1426                 goto free;
1427         }
1428
1429         gids = mailbox->buf;
1430         memcpy(gids, gw->gids, sizeof(gw->gids));
1431
1432         if (mlx4_ib_port_link_layer(&gw->dev->ib_dev, gw->port) ==
1433                                     IB_LINK_LAYER_ETHERNET) {
1434                 err = mlx4_cmd(dev, mailbox->dma,
1435                                MLX4_SET_PORT_GID_TABLE << 8 | gw->port,
1436                                1, MLX4_CMD_SET_PORT,
1437                                MLX4_CMD_TIME_CLASS_B,
1438                                MLX4_CMD_WRAPPED);
1439                 if (err)
1440                         pr_warn(KERN_WARNING
1441                                 "set port %d command failed\n", gw->port);
1442         }
1443
1444         mlx4_free_cmd_mailbox(dev, mailbox);
1445 free:
1446         kfree(gw);
1447 }
1448
1449 static int update_gid_table(struct mlx4_ib_dev *dev, int port,
1450                             union ib_gid *gid, int clear,
1451                             int default_gid)
1452 {
1453         struct update_gid_work *work;
1454         int i;
1455         int need_update = 0;
1456         int free = -1;
1457         int found = -1;
1458         int max_gids;
1459
1460         if (default_gid) {
1461                 free = 0;
1462         } else {
1463                 max_gids = dev->dev->caps.gid_table_len[port];
1464                 for (i = 1; i < max_gids; ++i) {
1465                         if (!memcmp(&dev->iboe.gid_table[port - 1][i], gid,
1466                                     sizeof(*gid)))
1467                                 found = i;
1468
1469                         if (clear) {
1470                                 if (found >= 0) {
1471                                         need_update = 1;
1472                                         dev->iboe.gid_table[port - 1][found] =
1473                                                 zgid;
1474                                         break;
1475                                 }
1476                         } else {
1477                                 if (found >= 0)
1478                                         break;
1479
1480                                 if (free < 0 &&
1481                                     !memcmp(&dev->iboe.gid_table[port - 1][i],
1482                                             &zgid, sizeof(*gid)))
1483                                         free = i;
1484                         }
1485                 }
1486         }
1487
1488         if (found == -1 && !clear && free >= 0) {
1489                 dev->iboe.gid_table[port - 1][free] = *gid;
1490                 need_update = 1;
1491         }
1492
1493         if (!need_update)
1494                 return 0;
1495
1496         work = kzalloc(sizeof(*work), GFP_ATOMIC);
1497         if (!work)
1498                 return -ENOMEM;
1499
1500         memcpy(work->gids, dev->iboe.gid_table[port - 1], sizeof(work->gids));
1501         INIT_WORK(&work->work, update_gids_task);
1502         work->port = port;
1503         work->dev = dev;
1504         queue_work(wq, &work->work);
1505
1506         return 0;
1507 }
1508
1509 static void mlx4_make_default_gid(struct  net_device *dev, union ib_gid *gid)
1510 {
1511         gid->global.subnet_prefix = cpu_to_be64(0xfe80000000000000LL);
1512         mlx4_addrconf_ifid_eui48(&gid->raw[8], 0xffff, dev);
1513 }
1514
1515
1516 static int reset_gid_table(struct mlx4_ib_dev *dev, u8 port)
1517 {
1518         struct update_gid_work *work;
1519
1520         work = kzalloc(sizeof(*work), GFP_ATOMIC);
1521         if (!work)
1522                 return -ENOMEM;
1523
1524         memset(dev->iboe.gid_table[port - 1], 0, sizeof(work->gids));
1525         memset(work->gids, 0, sizeof(work->gids));
1526         INIT_WORK(&work->work, reset_gids_task);
1527         work->dev = dev;
1528         work->port = port;
1529         queue_work(wq, &work->work);
1530         return 0;
1531 }
1532
1533 static int mlx4_ib_addr_event(int event, struct net_device *event_netdev,
1534                               struct mlx4_ib_dev *ibdev, union ib_gid *gid)
1535 {
1536         struct mlx4_ib_iboe *iboe;
1537         int port = 0;
1538         struct net_device *real_dev = rdma_vlan_dev_real_dev(event_netdev) ?
1539                                 rdma_vlan_dev_real_dev(event_netdev) :
1540                                 event_netdev;
1541         union ib_gid default_gid;
1542
1543         mlx4_make_default_gid(real_dev, &default_gid);
1544
1545         if (!memcmp(gid, &default_gid, sizeof(*gid)))
1546                 return 0;
1547
1548         if (event != NETDEV_DOWN && event != NETDEV_UP)
1549                 return 0;
1550
1551         if ((real_dev != event_netdev) &&
1552             (event == NETDEV_DOWN) &&
1553             rdma_link_local_addr((struct in6_addr *)gid))
1554                 return 0;
1555
1556         iboe = &ibdev->iboe;
1557         spin_lock(&iboe->lock);
1558
1559         for (port = 1; port <= ibdev->dev->caps.num_ports; ++port)
1560                 if ((netif_is_bond_master(real_dev) &&
1561                      (real_dev == iboe->masters[port - 1])) ||
1562                      (!netif_is_bond_master(real_dev) &&
1563                      (real_dev == iboe->netdevs[port - 1])))
1564                         update_gid_table(ibdev, port, gid,
1565                                          event == NETDEV_DOWN, 0);
1566
1567         spin_unlock(&iboe->lock);
1568         return 0;
1569
1570 }
1571
1572 static u8 mlx4_ib_get_dev_port(struct net_device *dev,
1573                                struct mlx4_ib_dev *ibdev)
1574 {
1575         u8 port = 0;
1576         struct mlx4_ib_iboe *iboe;
1577         struct net_device *real_dev = rdma_vlan_dev_real_dev(dev) ?
1578                                 rdma_vlan_dev_real_dev(dev) : dev;
1579
1580         iboe = &ibdev->iboe;
1581
1582         for (port = 1; port <= ibdev->dev->caps.num_ports; ++port)
1583                 if ((netif_is_bond_master(real_dev) &&
1584                      (real_dev == iboe->masters[port - 1])) ||
1585                      (!netif_is_bond_master(real_dev) &&
1586                      (real_dev == iboe->netdevs[port - 1])))
1587                         break;
1588
1589         if ((port == 0) || (port > ibdev->dev->caps.num_ports))
1590                 return 0;
1591         else
1592                 return port;
1593 }
1594
1595 static int mlx4_ib_inet_event(struct notifier_block *this, unsigned long event,
1596                                 void *ptr)
1597 {
1598         struct mlx4_ib_dev *ibdev;
1599         struct in_ifaddr *ifa = ptr;
1600         union ib_gid gid;
1601         struct net_device *event_netdev = ifa->ifa_dev->dev;
1602
1603         ipv6_addr_set_v4mapped(ifa->ifa_address, (struct in6_addr *)&gid);
1604
1605         ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb_inet);
1606
1607         mlx4_ib_addr_event(event, event_netdev, ibdev, &gid);
1608         return NOTIFY_DONE;
1609 }
1610
1611 #if IS_ENABLED(CONFIG_IPV6)
1612 static int mlx4_ib_inet6_event(struct notifier_block *this, unsigned long event,
1613                                 void *ptr)
1614 {
1615         struct mlx4_ib_dev *ibdev;
1616         struct inet6_ifaddr *ifa = ptr;
1617         union  ib_gid *gid = (union ib_gid *)&ifa->addr;
1618         struct net_device *event_netdev = ifa->idev->dev;
1619
1620         ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb_inet6);
1621
1622         mlx4_ib_addr_event(event, event_netdev, ibdev, gid);
1623         return NOTIFY_DONE;
1624 }
1625 #endif
1626
1627 #define MLX4_IB_INVALID_MAC     ((u64)-1)
1628 static void mlx4_ib_update_qps(struct mlx4_ib_dev *ibdev,
1629                                struct net_device *dev,
1630                                int port)
1631 {
1632         u64 new_smac = 0;
1633         u64 release_mac = MLX4_IB_INVALID_MAC;
1634         struct mlx4_ib_qp *qp;
1635
1636         read_lock(&dev_base_lock);
1637         new_smac = mlx4_mac_to_u64(dev->dev_addr);
1638         read_unlock(&dev_base_lock);
1639
1640         mutex_lock(&ibdev->qp1_proxy_lock[port - 1]);
1641         qp = ibdev->qp1_proxy[port - 1];
1642         if (qp) {
1643                 int new_smac_index;
1644                 u64 old_smac = qp->pri.smac;
1645                 struct mlx4_update_qp_params update_params;
1646
1647                 if (new_smac == old_smac)
1648                         goto unlock;
1649
1650                 new_smac_index = mlx4_register_mac(ibdev->dev, port, new_smac);
1651
1652                 if (new_smac_index < 0)
1653                         goto unlock;
1654
1655                 update_params.smac_index = new_smac_index;
1656                 if (mlx4_update_qp(ibdev->dev, &qp->mqp, MLX4_UPDATE_QP_SMAC,
1657                                    &update_params)) {
1658                         release_mac = new_smac;
1659                         goto unlock;
1660                 }
1661
1662                 qp->pri.smac = new_smac;
1663                 qp->pri.smac_index = new_smac_index;
1664
1665                 release_mac = old_smac;
1666         }
1667
1668 unlock:
1669         mutex_unlock(&ibdev->qp1_proxy_lock[port - 1]);
1670         if (release_mac != MLX4_IB_INVALID_MAC)
1671                 mlx4_unregister_mac(ibdev->dev, port, release_mac);
1672 }
1673
1674 static void mlx4_ib_get_dev_addr(struct net_device *dev,
1675                                  struct mlx4_ib_dev *ibdev, u8 port)
1676 {
1677         struct in_device *in_dev;
1678 #if IS_ENABLED(CONFIG_IPV6)
1679         struct inet6_dev *in6_dev;
1680         union ib_gid  *pgid;
1681         struct inet6_ifaddr *ifp;
1682         union ib_gid default_gid;
1683 #endif
1684         union ib_gid gid;
1685
1686
1687         if ((port == 0) || (port > ibdev->dev->caps.num_ports))
1688                 return;
1689
1690         /* IPv4 gids */
1691         in_dev = in_dev_get(dev);
1692         if (in_dev) {
1693                 for_ifa(in_dev) {
1694                         /*ifa->ifa_address;*/
1695                         ipv6_addr_set_v4mapped(ifa->ifa_address,
1696                                                (struct in6_addr *)&gid);
1697                         update_gid_table(ibdev, port, &gid, 0, 0);
1698                 }
1699                 endfor_ifa(in_dev);
1700                 in_dev_put(in_dev);
1701         }
1702 #if IS_ENABLED(CONFIG_IPV6)
1703         mlx4_make_default_gid(dev, &default_gid);
1704         /* IPv6 gids */
1705         in6_dev = in6_dev_get(dev);
1706         if (in6_dev) {
1707                 read_lock_bh(&in6_dev->lock);
1708                 list_for_each_entry(ifp, &in6_dev->addr_list, if_list) {
1709                         pgid = (union ib_gid *)&ifp->addr;
1710                         if (!memcmp(pgid, &default_gid, sizeof(*pgid)))
1711                                 continue;
1712                         update_gid_table(ibdev, port, pgid, 0, 0);
1713                 }
1714                 read_unlock_bh(&in6_dev->lock);
1715                 in6_dev_put(in6_dev);
1716         }
1717 #endif
1718 }
1719
1720 static void mlx4_ib_set_default_gid(struct mlx4_ib_dev *ibdev,
1721                                  struct  net_device *dev, u8 port)
1722 {
1723         union ib_gid gid;
1724         mlx4_make_default_gid(dev, &gid);
1725         update_gid_table(ibdev, port, &gid, 0, 1);
1726 }
1727
1728 static int mlx4_ib_init_gid_table(struct mlx4_ib_dev *ibdev)
1729 {
1730         struct  net_device *dev;
1731         struct mlx4_ib_iboe *iboe = &ibdev->iboe;
1732         int i;
1733         int err = 0;
1734
1735         for (i = 1; i <= ibdev->num_ports; ++i) {
1736                 if (rdma_port_get_link_layer(&ibdev->ib_dev, i) ==
1737                     IB_LINK_LAYER_ETHERNET) {
1738                         err = reset_gid_table(ibdev, i);
1739                         if (err)
1740                                 goto out;
1741                 }
1742         }
1743
1744         read_lock(&dev_base_lock);
1745         spin_lock(&iboe->lock);
1746
1747         for_each_netdev(&init_net, dev) {
1748                 u8 port = mlx4_ib_get_dev_port(dev, ibdev);
1749                 /* port will be non-zero only for ETH ports */
1750                 if (port) {
1751                         mlx4_ib_set_default_gid(ibdev, dev, port);
1752                         mlx4_ib_get_dev_addr(dev, ibdev, port);
1753                 }
1754         }
1755
1756         spin_unlock(&iboe->lock);
1757         read_unlock(&dev_base_lock);
1758 out:
1759         return err;
1760 }
1761
1762 static void mlx4_ib_scan_netdevs(struct mlx4_ib_dev *ibdev,
1763                                  struct net_device *dev,
1764                                  unsigned long event)
1765
1766 {
1767         struct mlx4_ib_iboe *iboe;
1768         int update_qps_port = -1;
1769         int port;
1770
1771         iboe = &ibdev->iboe;
1772
1773         spin_lock(&iboe->lock);
1774         mlx4_foreach_ib_transport_port(port, ibdev->dev) {
1775                 enum ib_port_state      port_state = IB_PORT_NOP;
1776                 struct net_device *old_master = iboe->masters[port - 1];
1777                 struct net_device *curr_netdev;
1778                 struct net_device *curr_master;
1779
1780                 iboe->netdevs[port - 1] =
1781                         mlx4_get_protocol_dev(ibdev->dev, MLX4_PROT_ETH, port);
1782                 if (iboe->netdevs[port - 1])
1783                         mlx4_ib_set_default_gid(ibdev,
1784                                                 iboe->netdevs[port - 1], port);
1785                 curr_netdev = iboe->netdevs[port - 1];
1786
1787                 if (iboe->netdevs[port - 1] &&
1788                     netif_is_bond_slave(iboe->netdevs[port - 1])) {
1789                         iboe->masters[port - 1] = netdev_master_upper_dev_get(
1790                                 iboe->netdevs[port - 1]);
1791                 } else {
1792                         iboe->masters[port - 1] = NULL;
1793                 }
1794                 curr_master = iboe->masters[port - 1];
1795
1796                 if (dev == iboe->netdevs[port - 1] &&
1797                     (event == NETDEV_CHANGEADDR || event == NETDEV_REGISTER ||
1798                      event == NETDEV_UP || event == NETDEV_CHANGE))
1799                         update_qps_port = port;
1800
1801                 if (curr_netdev) {
1802                         port_state = (netif_running(curr_netdev) && netif_carrier_ok(curr_netdev)) ?
1803                                                 IB_PORT_ACTIVE : IB_PORT_DOWN;
1804                         mlx4_ib_set_default_gid(ibdev, curr_netdev, port);
1805                         if (curr_master) {
1806                                 /* if using bonding/team and a slave port is down, we
1807                                  * don't want the bond IP based gids in the table since
1808                                  * flows that select port by gid may get the down port.
1809                                 */
1810                                 if (port_state == IB_PORT_DOWN) {
1811                                         reset_gid_table(ibdev, port);
1812                                         mlx4_ib_set_default_gid(ibdev,
1813                                                                 curr_netdev,
1814                                                                 port);
1815                                 } else {
1816                                         /* gids from the upper dev (bond/team)
1817                                          * should appear in port's gid table
1818                                         */
1819                                         mlx4_ib_get_dev_addr(curr_master,
1820                                                              ibdev, port);
1821                                 }
1822                         }
1823                         /* if bonding is used it is possible that we add it to
1824                          * masters only after IP address is assigned to the
1825                          * net bonding interface.
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_master, ibdev, port);
1832                         }
1833
1834                         if (!curr_master && (old_master != curr_master)) {
1835                                 reset_gid_table(ibdev, port);
1836                                 mlx4_ib_set_default_gid(ibdev,
1837                                                         curr_netdev, port);
1838                                 mlx4_ib_get_dev_addr(curr_netdev, ibdev, port);
1839                         }
1840                 } else {
1841                         reset_gid_table(ibdev, port);
1842                 }
1843         }
1844
1845         spin_unlock(&iboe->lock);
1846
1847         if (update_qps_port > 0)
1848                 mlx4_ib_update_qps(ibdev, dev, update_qps_port);
1849 }
1850
1851 static int mlx4_ib_netdev_event(struct notifier_block *this,
1852                                 unsigned long event, void *ptr)
1853 {
1854         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1855         struct mlx4_ib_dev *ibdev;
1856
1857         if (!net_eq(dev_net(dev), &init_net))
1858                 return NOTIFY_DONE;
1859
1860         ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb);
1861         mlx4_ib_scan_netdevs(ibdev, dev, event);
1862
1863         return NOTIFY_DONE;
1864 }
1865
1866 static void init_pkeys(struct mlx4_ib_dev *ibdev)
1867 {
1868         int port;
1869         int slave;
1870         int i;
1871
1872         if (mlx4_is_master(ibdev->dev)) {
1873                 for (slave = 0; slave <= ibdev->dev->num_vfs; ++slave) {
1874                         for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
1875                                 for (i = 0;
1876                                      i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
1877                                      ++i) {
1878                                         ibdev->pkeys.virt2phys_pkey[slave][port - 1][i] =
1879                                         /* master has the identity virt2phys pkey mapping */
1880                                                 (slave == mlx4_master_func_num(ibdev->dev) || !i) ? i :
1881                                                         ibdev->dev->phys_caps.pkey_phys_table_len[port] - 1;
1882                                         mlx4_sync_pkey_table(ibdev->dev, slave, port, i,
1883                                                              ibdev->pkeys.virt2phys_pkey[slave][port - 1][i]);
1884                                 }
1885                         }
1886                 }
1887                 /* initialize pkey cache */
1888                 for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
1889                         for (i = 0;
1890                              i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
1891                              ++i)
1892                                 ibdev->pkeys.phys_pkey_cache[port-1][i] =
1893                                         (i) ? 0 : 0xFFFF;
1894                 }
1895         }
1896 }
1897
1898 static void mlx4_ib_alloc_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
1899 {
1900         char name[80];
1901         int eq_per_port = 0;
1902         int added_eqs = 0;
1903         int total_eqs = 0;
1904         int i, j, eq;
1905
1906         /* Legacy mode or comp_pool is not large enough */
1907         if (dev->caps.comp_pool == 0 ||
1908             dev->caps.num_ports > dev->caps.comp_pool)
1909                 return;
1910
1911         eq_per_port = rounddown_pow_of_two(dev->caps.comp_pool/
1912                                         dev->caps.num_ports);
1913
1914         /* Init eq table */
1915         added_eqs = 0;
1916         mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
1917                 added_eqs += eq_per_port;
1918
1919         total_eqs = dev->caps.num_comp_vectors + added_eqs;
1920
1921         ibdev->eq_table = kzalloc(total_eqs * sizeof(int), GFP_KERNEL);
1922         if (!ibdev->eq_table)
1923                 return;
1924
1925         ibdev->eq_added = added_eqs;
1926
1927         eq = 0;
1928         mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB) {
1929                 for (j = 0; j < eq_per_port; j++) {
1930                         snprintf(name, sizeof(name), "mlx4-ib-%d-%d@%s",
1931                                  i, j, dev->pdev->bus->name);
1932                         /* Set IRQ for specific name (per ring) */
1933                         if (mlx4_assign_eq(dev, name, NULL,
1934                                            &ibdev->eq_table[eq])) {
1935                                 /* Use legacy (same as mlx4_en driver) */
1936                                 pr_warn("Can't allocate EQ %d; reverting to legacy\n", eq);
1937                                 ibdev->eq_table[eq] =
1938                                         (eq % dev->caps.num_comp_vectors);
1939                         }
1940                         eq++;
1941                 }
1942         }
1943
1944         /* Fill the reset of the vector with legacy EQ */
1945         for (i = 0, eq = added_eqs; i < dev->caps.num_comp_vectors; i++)
1946                 ibdev->eq_table[eq++] = i;
1947
1948         /* Advertise the new number of EQs to clients */
1949         ibdev->ib_dev.num_comp_vectors = total_eqs;
1950 }
1951
1952 static void mlx4_ib_free_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
1953 {
1954         int i;
1955
1956         /* no additional eqs were added */
1957         if (!ibdev->eq_table)
1958                 return;
1959
1960         /* Reset the advertised EQ number */
1961         ibdev->ib_dev.num_comp_vectors = dev->caps.num_comp_vectors;
1962
1963         /* Free only the added eqs */
1964         for (i = 0; i < ibdev->eq_added; i++) {
1965                 /* Don't free legacy eqs if used */
1966                 if (ibdev->eq_table[i] <= dev->caps.num_comp_vectors)
1967                         continue;
1968                 mlx4_release_eq(dev, ibdev->eq_table[i]);
1969         }
1970
1971         kfree(ibdev->eq_table);
1972 }
1973
1974 static void *mlx4_ib_add(struct mlx4_dev *dev)
1975 {
1976         struct mlx4_ib_dev *ibdev;
1977         int num_ports = 0;
1978         int i, j;
1979         int err;
1980         struct mlx4_ib_iboe *iboe;
1981         int ib_num_ports = 0;
1982
1983         pr_info_once("%s", mlx4_ib_version);
1984
1985         num_ports = 0;
1986         mlx4_foreach_ib_transport_port(i, dev)
1987                 num_ports++;
1988
1989         /* No point in registering a device with no ports... */
1990         if (num_ports == 0)
1991                 return NULL;
1992
1993         ibdev = (struct mlx4_ib_dev *) ib_alloc_device(sizeof *ibdev);
1994         if (!ibdev) {
1995                 dev_err(&dev->pdev->dev, "Device struct alloc failed\n");
1996                 return NULL;
1997         }
1998
1999         iboe = &ibdev->iboe;
2000
2001         if (mlx4_pd_alloc(dev, &ibdev->priv_pdn))
2002                 goto err_dealloc;
2003
2004         if (mlx4_uar_alloc(dev, &ibdev->priv_uar))
2005                 goto err_pd;
2006
2007         ibdev->uar_map = ioremap((phys_addr_t) ibdev->priv_uar.pfn << PAGE_SHIFT,
2008                                  PAGE_SIZE);
2009         if (!ibdev->uar_map)
2010                 goto err_uar;
2011         MLX4_INIT_DOORBELL_LOCK(&ibdev->uar_lock);
2012
2013         ibdev->dev = dev;
2014
2015         strlcpy(ibdev->ib_dev.name, "mlx4_%d", IB_DEVICE_NAME_MAX);
2016         ibdev->ib_dev.owner             = THIS_MODULE;
2017         ibdev->ib_dev.node_type         = RDMA_NODE_IB_CA;
2018         ibdev->ib_dev.local_dma_lkey    = dev->caps.reserved_lkey;
2019         ibdev->num_ports                = num_ports;
2020         ibdev->ib_dev.phys_port_cnt     = ibdev->num_ports;
2021         ibdev->ib_dev.num_comp_vectors  = dev->caps.num_comp_vectors;
2022         ibdev->ib_dev.dma_device        = &dev->pdev->dev;
2023
2024         if (dev->caps.userspace_caps)
2025                 ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_ABI_VERSION;
2026         else
2027                 ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION;
2028
2029         ibdev->ib_dev.uverbs_cmd_mask   =
2030                 (1ull << IB_USER_VERBS_CMD_GET_CONTEXT)         |
2031                 (1ull << IB_USER_VERBS_CMD_QUERY_DEVICE)        |
2032                 (1ull << IB_USER_VERBS_CMD_QUERY_PORT)          |
2033                 (1ull << IB_USER_VERBS_CMD_ALLOC_PD)            |
2034                 (1ull << IB_USER_VERBS_CMD_DEALLOC_PD)          |
2035                 (1ull << IB_USER_VERBS_CMD_REG_MR)              |
2036                 (1ull << IB_USER_VERBS_CMD_REREG_MR)            |
2037                 (1ull << IB_USER_VERBS_CMD_DEREG_MR)            |
2038                 (1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL) |
2039                 (1ull << IB_USER_VERBS_CMD_CREATE_CQ)           |
2040                 (1ull << IB_USER_VERBS_CMD_RESIZE_CQ)           |
2041                 (1ull << IB_USER_VERBS_CMD_DESTROY_CQ)          |
2042                 (1ull << IB_USER_VERBS_CMD_CREATE_QP)           |
2043                 (1ull << IB_USER_VERBS_CMD_MODIFY_QP)           |
2044                 (1ull << IB_USER_VERBS_CMD_QUERY_QP)            |
2045                 (1ull << IB_USER_VERBS_CMD_DESTROY_QP)          |
2046                 (1ull << IB_USER_VERBS_CMD_ATTACH_MCAST)        |
2047                 (1ull << IB_USER_VERBS_CMD_DETACH_MCAST)        |
2048                 (1ull << IB_USER_VERBS_CMD_CREATE_SRQ)          |
2049                 (1ull << IB_USER_VERBS_CMD_MODIFY_SRQ)          |
2050                 (1ull << IB_USER_VERBS_CMD_QUERY_SRQ)           |
2051                 (1ull << IB_USER_VERBS_CMD_DESTROY_SRQ)         |
2052                 (1ull << IB_USER_VERBS_CMD_CREATE_XSRQ)         |
2053                 (1ull << IB_USER_VERBS_CMD_OPEN_QP);
2054
2055         ibdev->ib_dev.query_device      = mlx4_ib_query_device;
2056         ibdev->ib_dev.query_port        = mlx4_ib_query_port;
2057         ibdev->ib_dev.get_link_layer    = mlx4_ib_port_link_layer;
2058         ibdev->ib_dev.query_gid         = mlx4_ib_query_gid;
2059         ibdev->ib_dev.query_pkey        = mlx4_ib_query_pkey;
2060         ibdev->ib_dev.modify_device     = mlx4_ib_modify_device;
2061         ibdev->ib_dev.modify_port       = mlx4_ib_modify_port;
2062         ibdev->ib_dev.alloc_ucontext    = mlx4_ib_alloc_ucontext;
2063         ibdev->ib_dev.dealloc_ucontext  = mlx4_ib_dealloc_ucontext;
2064         ibdev->ib_dev.mmap              = mlx4_ib_mmap;
2065         ibdev->ib_dev.alloc_pd          = mlx4_ib_alloc_pd;
2066         ibdev->ib_dev.dealloc_pd        = mlx4_ib_dealloc_pd;
2067         ibdev->ib_dev.create_ah         = mlx4_ib_create_ah;
2068         ibdev->ib_dev.query_ah          = mlx4_ib_query_ah;
2069         ibdev->ib_dev.destroy_ah        = mlx4_ib_destroy_ah;
2070         ibdev->ib_dev.create_srq        = mlx4_ib_create_srq;
2071         ibdev->ib_dev.modify_srq        = mlx4_ib_modify_srq;
2072         ibdev->ib_dev.query_srq         = mlx4_ib_query_srq;
2073         ibdev->ib_dev.destroy_srq       = mlx4_ib_destroy_srq;
2074         ibdev->ib_dev.post_srq_recv     = mlx4_ib_post_srq_recv;
2075         ibdev->ib_dev.create_qp         = mlx4_ib_create_qp;
2076         ibdev->ib_dev.modify_qp         = mlx4_ib_modify_qp;
2077         ibdev->ib_dev.query_qp          = mlx4_ib_query_qp;
2078         ibdev->ib_dev.destroy_qp        = mlx4_ib_destroy_qp;
2079         ibdev->ib_dev.post_send         = mlx4_ib_post_send;
2080         ibdev->ib_dev.post_recv         = mlx4_ib_post_recv;
2081         ibdev->ib_dev.create_cq         = mlx4_ib_create_cq;
2082         ibdev->ib_dev.modify_cq         = mlx4_ib_modify_cq;
2083         ibdev->ib_dev.resize_cq         = mlx4_ib_resize_cq;
2084         ibdev->ib_dev.destroy_cq        = mlx4_ib_destroy_cq;
2085         ibdev->ib_dev.poll_cq           = mlx4_ib_poll_cq;
2086         ibdev->ib_dev.req_notify_cq     = mlx4_ib_arm_cq;
2087         ibdev->ib_dev.get_dma_mr        = mlx4_ib_get_dma_mr;
2088         ibdev->ib_dev.reg_user_mr       = mlx4_ib_reg_user_mr;
2089         ibdev->ib_dev.rereg_user_mr     = mlx4_ib_rereg_user_mr;
2090         ibdev->ib_dev.dereg_mr          = mlx4_ib_dereg_mr;
2091         ibdev->ib_dev.alloc_fast_reg_mr = mlx4_ib_alloc_fast_reg_mr;
2092         ibdev->ib_dev.alloc_fast_reg_page_list = mlx4_ib_alloc_fast_reg_page_list;
2093         ibdev->ib_dev.free_fast_reg_page_list  = mlx4_ib_free_fast_reg_page_list;
2094         ibdev->ib_dev.attach_mcast      = mlx4_ib_mcg_attach;
2095         ibdev->ib_dev.detach_mcast      = mlx4_ib_mcg_detach;
2096         ibdev->ib_dev.process_mad       = mlx4_ib_process_mad;
2097
2098         if (!mlx4_is_slave(ibdev->dev)) {
2099                 ibdev->ib_dev.alloc_fmr         = mlx4_ib_fmr_alloc;
2100                 ibdev->ib_dev.map_phys_fmr      = mlx4_ib_map_phys_fmr;
2101                 ibdev->ib_dev.unmap_fmr         = mlx4_ib_unmap_fmr;
2102                 ibdev->ib_dev.dealloc_fmr       = mlx4_ib_fmr_dealloc;
2103         }
2104
2105         if (dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW ||
2106             dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
2107                 ibdev->ib_dev.alloc_mw = mlx4_ib_alloc_mw;
2108                 ibdev->ib_dev.bind_mw = mlx4_ib_bind_mw;
2109                 ibdev->ib_dev.dealloc_mw = mlx4_ib_dealloc_mw;
2110
2111                 ibdev->ib_dev.uverbs_cmd_mask |=
2112                         (1ull << IB_USER_VERBS_CMD_ALLOC_MW) |
2113                         (1ull << IB_USER_VERBS_CMD_DEALLOC_MW);
2114         }
2115
2116         if (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) {
2117                 ibdev->ib_dev.alloc_xrcd = mlx4_ib_alloc_xrcd;
2118                 ibdev->ib_dev.dealloc_xrcd = mlx4_ib_dealloc_xrcd;
2119                 ibdev->ib_dev.uverbs_cmd_mask |=
2120                         (1ull << IB_USER_VERBS_CMD_OPEN_XRCD) |
2121                         (1ull << IB_USER_VERBS_CMD_CLOSE_XRCD);
2122         }
2123
2124         if (check_flow_steering_support(dev)) {
2125                 ibdev->steering_support = MLX4_STEERING_MODE_DEVICE_MANAGED;
2126                 ibdev->ib_dev.create_flow       = mlx4_ib_create_flow;
2127                 ibdev->ib_dev.destroy_flow      = mlx4_ib_destroy_flow;
2128
2129                 ibdev->ib_dev.uverbs_ex_cmd_mask        |=
2130                         (1ull << IB_USER_VERBS_EX_CMD_CREATE_FLOW) |
2131                         (1ull << IB_USER_VERBS_EX_CMD_DESTROY_FLOW);
2132         }
2133
2134         mlx4_ib_alloc_eqs(dev, ibdev);
2135
2136         spin_lock_init(&iboe->lock);
2137
2138         if (init_node_data(ibdev))
2139                 goto err_map;
2140
2141         for (i = 0; i < ibdev->num_ports; ++i) {
2142                 mutex_init(&ibdev->qp1_proxy_lock[i]);
2143                 if (mlx4_ib_port_link_layer(&ibdev->ib_dev, i + 1) ==
2144                                                 IB_LINK_LAYER_ETHERNET) {
2145                         err = mlx4_counter_alloc(ibdev->dev, &ibdev->counters[i]);
2146                         if (err)
2147                                 ibdev->counters[i] = -1;
2148                 } else {
2149                         ibdev->counters[i] = -1;
2150                 }
2151         }
2152
2153         mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
2154                 ib_num_ports++;
2155
2156         spin_lock_init(&ibdev->sm_lock);
2157         mutex_init(&ibdev->cap_mask_mutex);
2158
2159         if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED &&
2160             ib_num_ports) {
2161                 ibdev->steer_qpn_count = MLX4_IB_UC_MAX_NUM_QPS;
2162                 err = mlx4_qp_reserve_range(dev, ibdev->steer_qpn_count,
2163                                             MLX4_IB_UC_STEER_QPN_ALIGN,
2164                                             &ibdev->steer_qpn_base);
2165                 if (err)
2166                         goto err_counter;
2167
2168                 ibdev->ib_uc_qpns_bitmap =
2169                         kmalloc(BITS_TO_LONGS(ibdev->steer_qpn_count) *
2170                                 sizeof(long),
2171                                 GFP_KERNEL);
2172                 if (!ibdev->ib_uc_qpns_bitmap) {
2173                         dev_err(&dev->pdev->dev, "bit map alloc failed\n");
2174                         goto err_steer_qp_release;
2175                 }
2176
2177                 bitmap_zero(ibdev->ib_uc_qpns_bitmap, ibdev->steer_qpn_count);
2178
2179                 err = mlx4_FLOW_STEERING_IB_UC_QP_RANGE(
2180                                 dev, ibdev->steer_qpn_base,
2181                                 ibdev->steer_qpn_base +
2182                                 ibdev->steer_qpn_count - 1);
2183                 if (err)
2184                         goto err_steer_free_bitmap;
2185         }
2186
2187         if (ib_register_device(&ibdev->ib_dev, NULL))
2188                 goto err_steer_free_bitmap;
2189
2190         if (mlx4_ib_mad_init(ibdev))
2191                 goto err_reg;
2192
2193         if (mlx4_ib_init_sriov(ibdev))
2194                 goto err_mad;
2195
2196         if (dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE) {
2197                 if (!iboe->nb.notifier_call) {
2198                         iboe->nb.notifier_call = mlx4_ib_netdev_event;
2199                         err = register_netdevice_notifier(&iboe->nb);
2200                         if (err) {
2201                                 iboe->nb.notifier_call = NULL;
2202                                 goto err_notif;
2203                         }
2204                 }
2205                 if (!iboe->nb_inet.notifier_call) {
2206                         iboe->nb_inet.notifier_call = mlx4_ib_inet_event;
2207                         err = register_inetaddr_notifier(&iboe->nb_inet);
2208                         if (err) {
2209                                 iboe->nb_inet.notifier_call = NULL;
2210                                 goto err_notif;
2211                         }
2212                 }
2213 #if IS_ENABLED(CONFIG_IPV6)
2214                 if (!iboe->nb_inet6.notifier_call) {
2215                         iboe->nb_inet6.notifier_call = mlx4_ib_inet6_event;
2216                         err = register_inet6addr_notifier(&iboe->nb_inet6);
2217                         if (err) {
2218                                 iboe->nb_inet6.notifier_call = NULL;
2219                                 goto err_notif;
2220                         }
2221                 }
2222 #endif
2223                 if (mlx4_ib_init_gid_table(ibdev))
2224                         goto err_notif;
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         if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED)
2282                 mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
2283                                       ibdev->steer_qpn_count);
2284 err_counter:
2285         for (; i; --i)
2286                 if (ibdev->counters[i - 1] != -1)
2287                         mlx4_counter_free(ibdev->dev, ibdev->counters[i - 1]);
2288
2289 err_map:
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         mlx4_ib_close_sriov(ibdev);
2373         mlx4_ib_mad_cleanup(ibdev);
2374         ib_unregister_device(&ibdev->ib_dev);
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         if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED) {
2382                 mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
2383                                       ibdev->steer_qpn_count);
2384                 kfree(ibdev->ib_uc_qpns_bitmap);
2385         }
2386
2387         if (ibdev->iboe.nb_inet.notifier_call) {
2388                 if (unregister_inetaddr_notifier(&ibdev->iboe.nb_inet))
2389                         pr_warn("failure unregistering notifier\n");
2390                 ibdev->iboe.nb_inet.notifier_call = NULL;
2391         }
2392 #if IS_ENABLED(CONFIG_IPV6)
2393         if (ibdev->iboe.nb_inet6.notifier_call) {
2394                 if (unregister_inet6addr_notifier(&ibdev->iboe.nb_inet6))
2395                         pr_warn("failure unregistering notifier\n");
2396                 ibdev->iboe.nb_inet6.notifier_call = NULL;
2397         }
2398 #endif
2399
2400         iounmap(ibdev->uar_map);
2401         for (p = 0; p < ibdev->num_ports; ++p)
2402                 if (ibdev->counters[p] != -1)
2403                         mlx4_counter_free(ibdev->dev, ibdev->counters[p]);
2404         mlx4_foreach_port(p, dev, MLX4_PORT_TYPE_IB)
2405                 mlx4_CLOSE_PORT(dev, p);
2406
2407         mlx4_ib_free_eqs(dev, ibdev);
2408
2409         mlx4_uar_free(dev, &ibdev->priv_uar);
2410         mlx4_pd_free(dev, ibdev->priv_pdn);
2411         ib_dealloc_device(&ibdev->ib_dev);
2412 }
2413
2414 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init)
2415 {
2416         struct mlx4_ib_demux_work **dm = NULL;
2417         struct mlx4_dev *dev = ibdev->dev;
2418         int i;
2419         unsigned long flags;
2420         struct mlx4_active_ports actv_ports;
2421         unsigned int ports;
2422         unsigned int first_port;
2423
2424         if (!mlx4_is_master(dev))
2425                 return;
2426
2427         actv_ports = mlx4_get_active_ports(dev, slave);
2428         ports = bitmap_weight(actv_ports.ports, dev->caps.num_ports);
2429         first_port = find_first_bit(actv_ports.ports, dev->caps.num_ports);
2430
2431         dm = kcalloc(ports, sizeof(*dm), GFP_ATOMIC);
2432         if (!dm) {
2433                 pr_err("failed to allocate memory for tunneling qp update\n");
2434                 goto out;
2435         }
2436
2437         for (i = 0; i < ports; i++) {
2438                 dm[i] = kmalloc(sizeof (struct mlx4_ib_demux_work), GFP_ATOMIC);
2439                 if (!dm[i]) {
2440                         pr_err("failed to allocate memory for tunneling qp update work struct\n");
2441                         for (i = 0; i < dev->caps.num_ports; i++) {
2442                                 if (dm[i])
2443                                         kfree(dm[i]);
2444                         }
2445                         goto out;
2446                 }
2447         }
2448         /* initialize or tear down tunnel QPs for the slave */
2449         for (i = 0; i < ports; i++) {
2450                 INIT_WORK(&dm[i]->work, mlx4_ib_tunnels_update_work);
2451                 dm[i]->port = first_port + i + 1;
2452                 dm[i]->slave = slave;
2453                 dm[i]->do_init = do_init;
2454                 dm[i]->dev = ibdev;
2455                 spin_lock_irqsave(&ibdev->sriov.going_down_lock, flags);
2456                 if (!ibdev->sriov.is_going_down)
2457                         queue_work(ibdev->sriov.demux[i].ud_wq, &dm[i]->work);
2458                 spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
2459         }
2460 out:
2461         kfree(dm);
2462         return;
2463 }
2464
2465 static void mlx4_ib_event(struct mlx4_dev *dev, void *ibdev_ptr,
2466                           enum mlx4_dev_event event, unsigned long param)
2467 {
2468         struct ib_event ibev;
2469         struct mlx4_ib_dev *ibdev = to_mdev((struct ib_device *) ibdev_ptr);
2470         struct mlx4_eqe *eqe = NULL;
2471         struct ib_event_work *ew;
2472         int p = 0;
2473
2474         if (event == MLX4_DEV_EVENT_PORT_MGMT_CHANGE)
2475                 eqe = (struct mlx4_eqe *)param;
2476         else
2477                 p = (int) param;
2478
2479         switch (event) {
2480         case MLX4_DEV_EVENT_PORT_UP:
2481                 if (p > ibdev->num_ports)
2482                         return;
2483                 if (mlx4_is_master(dev) &&
2484                     rdma_port_get_link_layer(&ibdev->ib_dev, p) ==
2485                         IB_LINK_LAYER_INFINIBAND) {
2486                         mlx4_ib_invalidate_all_guid_record(ibdev, p);
2487                 }
2488                 ibev.event = IB_EVENT_PORT_ACTIVE;
2489                 break;
2490
2491         case MLX4_DEV_EVENT_PORT_DOWN:
2492                 if (p > ibdev->num_ports)
2493                         return;
2494                 ibev.event = IB_EVENT_PORT_ERR;
2495                 break;
2496
2497         case MLX4_DEV_EVENT_CATASTROPHIC_ERROR:
2498                 ibdev->ib_active = false;
2499                 ibev.event = IB_EVENT_DEVICE_FATAL;
2500                 break;
2501
2502         case MLX4_DEV_EVENT_PORT_MGMT_CHANGE:
2503                 ew = kmalloc(sizeof *ew, GFP_ATOMIC);
2504                 if (!ew) {
2505                         pr_err("failed to allocate memory for events work\n");
2506                         break;
2507                 }
2508
2509                 INIT_WORK(&ew->work, handle_port_mgmt_change_event);
2510                 memcpy(&ew->ib_eqe, eqe, sizeof *eqe);
2511                 ew->ib_dev = ibdev;
2512                 /* need to queue only for port owner, which uses GEN_EQE */
2513                 if (mlx4_is_master(dev))
2514                         queue_work(wq, &ew->work);
2515                 else
2516                         handle_port_mgmt_change_event(&ew->work);
2517                 return;
2518
2519         case MLX4_DEV_EVENT_SLAVE_INIT:
2520                 /* here, p is the slave id */
2521                 do_slave_init(ibdev, p, 1);
2522                 return;
2523
2524         case MLX4_DEV_EVENT_SLAVE_SHUTDOWN:
2525                 /* here, p is the slave id */
2526                 do_slave_init(ibdev, p, 0);
2527                 return;
2528
2529         default:
2530                 return;
2531         }
2532
2533         ibev.device           = ibdev_ptr;
2534         ibev.element.port_num = (u8) p;
2535
2536         ib_dispatch_event(&ibev);
2537 }
2538
2539 static struct mlx4_interface mlx4_ib_interface = {
2540         .add            = mlx4_ib_add,
2541         .remove         = mlx4_ib_remove,
2542         .event          = mlx4_ib_event,
2543         .protocol       = MLX4_PROT_IB_IPV6
2544 };
2545
2546 static int __init mlx4_ib_init(void)
2547 {
2548         int err;
2549
2550         wq = create_singlethread_workqueue("mlx4_ib");
2551         if (!wq)
2552                 return -ENOMEM;
2553
2554         err = mlx4_ib_mcg_init();
2555         if (err)
2556                 goto clean_wq;
2557
2558         err = mlx4_register_interface(&mlx4_ib_interface);
2559         if (err)
2560                 goto clean_mcg;
2561
2562         return 0;
2563
2564 clean_mcg:
2565         mlx4_ib_mcg_destroy();
2566
2567 clean_wq:
2568         destroy_workqueue(wq);
2569         return err;
2570 }
2571
2572 static void __exit mlx4_ib_cleanup(void)
2573 {
2574         mlx4_unregister_interface(&mlx4_ib_interface);
2575         mlx4_ib_mcg_destroy();
2576         destroy_workqueue(wq);
2577 }
2578
2579 module_init(mlx4_ib_init);
2580 module_exit(mlx4_ib_cleanup);