drivers/net/ethernet/mellanox/mlx4/main.c

   1 /*
   2  * Copyright (c) 2004, 2005 Topspin Communications.  All rights reserved.
   3  * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved.
   4  * Copyright (c) 2005, 2006, 2007, 2008 Mellanox Technologies. All rights reserved.
   5  * Copyright (c) 2006, 2007 Cisco Systems, Inc. All rights reserved.
   6  *
   7  * This software is available to you under a choice of one of two
   8  * licenses.  You may choose to be licensed under the terms of the GNU
   9  * General Public License (GPL) Version 2, available from the file
  10  * COPYING in the main directory of this source tree, or the
  11  * OpenIB.org BSD license below:
  12  *
  13  *     Redistribution and use in source and binary forms, with or
  14  *     without modification, are permitted provided that the following
  15  *     conditions are met:
  16  *
  17  *      - Redistributions of source code must retain the above
  18  *        copyright notice, this list of conditions and the following
  19  *        disclaimer.
  20  *
  21  *      - Redistributions in binary form must reproduce the above
  22  *        copyright notice, this list of conditions and the following
  23  *        disclaimer in the documentation and/or other materials
  24  *        provided with the distribution.
  25  *
  26  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
  27  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  28  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
  29  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
  30  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
  31  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
  32  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  33  * SOFTWARE.
  34  */
  35
  36 #include <linux/module.h>
  37 #include <linux/init.h>
  38 #include <linux/errno.h>
  39 #include <linux/pci.h>
  40 #include <linux/dma-mapping.h>
  41 #include <linux/slab.h>
  42 #include <linux/io-mapping.h>
  43 #include <linux/delay.h>
  44 #include <linux/kmod.h>
  45
  46 #include <linux/mlx4/device.h>
  47 #include <linux/mlx4/doorbell.h>
  48
  49 #include "mlx4.h"
  50 #include "fw.h"
  51 #include "icm.h"
  52
  53 MODULE_AUTHOR("Roland Dreier");
  54 MODULE_DESCRIPTION("Mellanox ConnectX HCA low-level driver");
  55 MODULE_LICENSE("Dual BSD/GPL");
  56 MODULE_VERSION(DRV_VERSION);
  57
  58 struct workqueue_struct *mlx4_wq;
  59
  60 #ifdef CONFIG_MLX4_DEBUG
  61
  62 int mlx4_debug_level = 0;
  63 module_param_named(debug_level, mlx4_debug_level, int, 0644);
  64 MODULE_PARM_DESC(debug_level, "Enable debug tracing if > 0");
  65
  66 #endif /* CONFIG_MLX4_DEBUG */
  67
  68 #ifdef CONFIG_PCI_MSI
  69
  70 static int msi_x = 1;
  71 module_param(msi_x, int, 0444);
  72 MODULE_PARM_DESC(msi_x, "attempt to use MSI-X if nonzero");
  73
  74 #else /* CONFIG_PCI_MSI */
  75
  76 #define msi_x (0)
  77
  78 #endif /* CONFIG_PCI_MSI */
  79
  80 static uint8_t num_vfs[3] = {0, 0, 0};
  81 static int num_vfs_argc;
  82 module_param_array(num_vfs, byte , &num_vfs_argc, 0444);
  83 MODULE_PARM_DESC(num_vfs, "enable #num_vfs functions if num_vfs > 0\n"
  84                           "num_vfs=port1,port2,port1+2");
  85
  86 static uint8_t probe_vf[3] = {0, 0, 0};
  87 static int probe_vfs_argc;
  88 module_param_array(probe_vf, byte, &probe_vfs_argc, 0444);
  89 MODULE_PARM_DESC(probe_vf, "number of vfs to probe by pf driver (num_vfs > 0)\n"
  90                            "probe_vf=port1,port2,port1+2");
  91
  92 int mlx4_log_num_mgm_entry_size = MLX4_DEFAULT_MGM_LOG_ENTRY_SIZE;
  93 module_param_named(log_num_mgm_entry_size,
  94                         mlx4_log_num_mgm_entry_size, int, 0444);
  95 MODULE_PARM_DESC(log_num_mgm_entry_size, "log mgm size, that defines the num"
  96                                          " of qp per mcg, for example:"
  97                                          " 10 gives 248.range: 7 <="
  98                                          " log_num_mgm_entry_size <= 12."
  99                                          " To activate device managed"
 100                                          " flow steering when available, set to -1");
 101
 102 static bool enable_64b_cqe_eqe = true;
 103 module_param(enable_64b_cqe_eqe, bool, 0444);
 104 MODULE_PARM_DESC(enable_64b_cqe_eqe,
 105                  "Enable 64 byte CQEs/EQEs when the FW supports this (default: True)");
 106
 107 #define PF_CONTEXT_BEHAVIOUR_MASK       (MLX4_FUNC_CAP_64B_EQE_CQE | \
 108                                          MLX4_FUNC_CAP_EQE_CQE_STRIDE | \
 109                                          MLX4_FUNC_CAP_DMFS_A0_STATIC)
 110
 111 #define RESET_PERSIST_MASK_FLAGS        (MLX4_FLAG_SRIOV)
 112
 113 static char mlx4_version[] =
 114         DRV_NAME ": Mellanox ConnectX core driver v"
 115         DRV_VERSION " (" DRV_RELDATE ")\n";
 116
 117 static struct mlx4_profile default_profile = {
 118         .num_qp         = 1 << 18,
 119         .num_srq        = 1 << 16,
 120         .rdmarc_per_qp  = 1 << 4,
 121         .num_cq         = 1 << 16,
 122         .num_mcg        = 1 << 13,
 123         .num_mpt        = 1 << 19,
 124         .num_mtt        = 1 << 20, /* It is really num mtt segements */
 125 };
 126
 127 static struct mlx4_profile low_mem_profile = {
 128         .num_qp         = 1 << 17,
 129         .num_srq        = 1 << 6,
 130         .rdmarc_per_qp  = 1 << 4,
 131         .num_cq         = 1 << 8,
 132         .num_mcg        = 1 << 8,
 133         .num_mpt        = 1 << 9,
 134         .num_mtt        = 1 << 7,
 135 };
 136
 137 static int log_num_mac = 7;
 138 module_param_named(log_num_mac, log_num_mac, int, 0444);
 139 MODULE_PARM_DESC(log_num_mac, "Log2 max number of MACs per ETH port (1-7)");
 140
 141 static int log_num_vlan;
 142 module_param_named(log_num_vlan, log_num_vlan, int, 0444);
 143 MODULE_PARM_DESC(log_num_vlan, "Log2 max number of VLANs per ETH port (0-7)");
 144 /* Log2 max number of VLANs per ETH port (0-7) */
 145 #define MLX4_LOG_NUM_VLANS 7
 146 #define MLX4_MIN_LOG_NUM_VLANS 0
 147 #define MLX4_MIN_LOG_NUM_MAC 1
 148
 149 static bool use_prio;
 150 module_param_named(use_prio, use_prio, bool, 0444);
 151 MODULE_PARM_DESC(use_prio, "Enable steering by VLAN priority on ETH ports (deprecated)");
 152
 153 int log_mtts_per_seg = ilog2(MLX4_MTT_ENTRY_PER_SEG);
 154 module_param_named(log_mtts_per_seg, log_mtts_per_seg, int, 0444);
 155 MODULE_PARM_DESC(log_mtts_per_seg, "Log2 number of MTT entries per segment (1-7)");
 156
 157 static int port_type_array[2] = {MLX4_PORT_TYPE_NONE, MLX4_PORT_TYPE_NONE};
 158 static int arr_argc = 2;
 159 module_param_array(port_type_array, int, &arr_argc, 0444);
 160 MODULE_PARM_DESC(port_type_array, "Array of port types: HW_DEFAULT (0) is default "
 161                                 "1 for IB, 2 for Ethernet");
 162
 163 struct mlx4_port_config {
 164         struct list_head list;
 165         enum mlx4_port_type port_type[MLX4_MAX_PORTS + 1];
 166         struct pci_dev *pdev;
 167 };
 168
 169 static atomic_t pf_loading = ATOMIC_INIT(0);
 170
 171 int mlx4_check_port_params(struct mlx4_dev *dev,
 172                            enum mlx4_port_type *port_type)
 173 {
 174         int i;
 175
 176         if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_DPDP)) {
 177                 for (i = 0; i < dev->caps.num_ports - 1; i++) {
 178                         if (port_type[i] != port_type[i + 1]) {
 179                                 mlx4_err(dev, "Only same port types supported on this HCA, aborting\n");
 180                                 return -EINVAL;
 181                         }
 182                 }
 183         }
 184
 185         for (i = 0; i < dev->caps.num_ports; i++) {
 186                 if (!(port_type[i] & dev->caps.supported_type[i+1])) {
 187                         mlx4_err(dev, "Requested port type for port %d is not supported on this HCA\n",
 188                                  i + 1);
 189                         return -EINVAL;
 190                 }
 191         }
 192         return 0;
 193 }
 194
 195 static void mlx4_set_port_mask(struct mlx4_dev *dev)
 196 {
 197         int i;
 198
 199         for (i = 1; i <= dev->caps.num_ports; ++i)
 200                 dev->caps.port_mask[i] = dev->caps.port_type[i];
 201 }
 202
 203 enum {
 204         MLX4_QUERY_FUNC_NUM_SYS_EQS = 1 << 0,
 205 };
 206
 207 static int mlx4_query_func(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap)
 208 {
 209         int err = 0;
 210         struct mlx4_func func;
 211
 212         if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS) {
 213                 err = mlx4_QUERY_FUNC(dev, &func, 0);
 214                 if (err) {
 215                         mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting.\n");
 216                         return err;
 217                 }
 218                 dev_cap->max_eqs = func.max_eq;
 219                 dev_cap->reserved_eqs = func.rsvd_eqs;
 220                 dev_cap->reserved_uars = func.rsvd_uars;
 221                 err |= MLX4_QUERY_FUNC_NUM_SYS_EQS;
 222         }
 223         return err;
 224 }
 225
 226 static void mlx4_enable_cqe_eqe_stride(struct mlx4_dev *dev)
 227 {
 228         struct mlx4_caps *dev_cap = &dev->caps;
 229
 230         /* FW not supporting or cancelled by user */
 231         if (!(dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_EQE_STRIDE) ||
 232             !(dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_CQE_STRIDE))
 233                 return;
 234
 235         /* Must have 64B CQE_EQE enabled by FW to use bigger stride
 236          * When FW has NCSI it may decide not to report 64B CQE/EQEs
 237          */
 238         if (!(dev_cap->flags & MLX4_DEV_CAP_FLAG_64B_EQE) ||
 239             !(dev_cap->flags & MLX4_DEV_CAP_FLAG_64B_CQE)) {
 240                 dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_CQE_STRIDE;
 241                 dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_EQE_STRIDE;
 242                 return;
 243         }
 244
 245         if (cache_line_size() == 128 || cache_line_size() == 256) {
 246                 mlx4_dbg(dev, "Enabling CQE stride cacheLine supported\n");
 247                 /* Changing the real data inside CQE size to 32B */
 248                 dev_cap->flags &= ~MLX4_DEV_CAP_FLAG_64B_CQE;
 249                 dev_cap->flags &= ~MLX4_DEV_CAP_FLAG_64B_EQE;
 250
 251                 if (mlx4_is_master(dev))
 252                         dev_cap->function_caps |= MLX4_FUNC_CAP_EQE_CQE_STRIDE;
 253         } else {
 254                 mlx4_dbg(dev, "Disabling CQE stride cacheLine unsupported\n");
 255                 dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_CQE_STRIDE;
 256                 dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_EQE_STRIDE;
 257         }
 258 }
 259
 260 static int _mlx4_dev_port(struct mlx4_dev *dev, int port,
 261                           struct mlx4_port_cap *port_cap)
 262 {
 263         dev->caps.vl_cap[port]      = port_cap->max_vl;
 264         dev->caps.ib_mtu_cap[port]          = port_cap->ib_mtu;
 265         dev->phys_caps.gid_phys_table_len[port]  = port_cap->max_gids;
 266         dev->phys_caps.pkey_phys_table_len[port] = port_cap->max_pkeys;
 267         /* set gid and pkey table operating lengths by default
 268          * to non-sriov values
 269          */
 270         dev->caps.gid_table_len[port]  = port_cap->max_gids;
 271         dev->caps.pkey_table_len[port] = port_cap->max_pkeys;
 272         dev->caps.port_width_cap[port] = port_cap->max_port_width;
 273         dev->caps.eth_mtu_cap[port]    = port_cap->eth_mtu;
 274         dev->caps.def_mac[port]        = port_cap->def_mac;
 275         dev->caps.supported_type[port] = port_cap->supported_port_types;
 276         dev->caps.suggested_type[port] = port_cap->suggested_type;
 277         dev->caps.default_sense[port] = port_cap->default_sense;
 278         dev->caps.trans_type[port]          = port_cap->trans_type;
 279         dev->caps.vendor_oui[port]     = port_cap->vendor_oui;
 280         dev->caps.wavelength[port]     = port_cap->wavelength;
 281         dev->caps.trans_code[port]     = port_cap->trans_code;
 282
 283         return 0;
 284 }
 285
 286 static int mlx4_dev_port(struct mlx4_dev *dev, int port,
 287                          struct mlx4_port_cap *port_cap)
 288 {
 289         int err = 0;
 290
 291         err = mlx4_QUERY_PORT(dev, port, port_cap);
 292
 293         if (err)
 294                 mlx4_err(dev, "QUERY_PORT command failed.\n");
 295
 296         return err;
 297 }
 298
 299 #define MLX4_A0_STEERING_TABLE_SIZE     256
 300 static int mlx4_dev_cap(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap)
 301 {
 302         int err;
 303         int i;
 304
 305         err = mlx4_QUERY_DEV_CAP(dev, dev_cap);
 306         if (err) {
 307                 mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting\n");
 308                 return err;
 309         }
 310         mlx4_dev_cap_dump(dev, dev_cap);
 311
 312         if (dev_cap->min_page_sz > PAGE_SIZE) {
 313                 mlx4_err(dev, "HCA minimum page size of %d bigger than kernel PAGE_SIZE of %ld, aborting\n",
 314                          dev_cap->min_page_sz, PAGE_SIZE);
 315                 return -ENODEV;
 316         }
 317         if (dev_cap->num_ports > MLX4_MAX_PORTS) {
 318                 mlx4_err(dev, "HCA has %d ports, but we only support %d, aborting\n",
 319                          dev_cap->num_ports, MLX4_MAX_PORTS);
 320                 return -ENODEV;
 321         }
 322
 323         if (dev_cap->uar_size > pci_resource_len(dev->persist->pdev, 2)) {
 324                 mlx4_err(dev, "HCA reported UAR size of 0x%x bigger than PCI resource 2 size of 0x%llx, aborting\n",
 325                          dev_cap->uar_size,
 326                          (unsigned long long)
 327                          pci_resource_len(dev->persist->pdev, 2));
 328                 return -ENODEV;
 329         }
 330
 331         dev->caps.num_ports          = dev_cap->num_ports;
 332         dev->caps.num_sys_eqs = dev_cap->num_sys_eqs;
 333         dev->phys_caps.num_phys_eqs = dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS ?
 334                                       dev->caps.num_sys_eqs :
 335                                       MLX4_MAX_EQ_NUM;
 336         for (i = 1; i <= dev->caps.num_ports; ++i) {
 337                 err = _mlx4_dev_port(dev, i, dev_cap->port_cap + i);
 338                 if (err) {
 339                         mlx4_err(dev, "QUERY_PORT command failed, aborting\n");
 340                         return err;
 341                 }
 342         }
 343
 344         dev->caps.uar_page_size      = PAGE_SIZE;
 345         dev->caps.num_uars           = dev_cap->uar_size / PAGE_SIZE;
 346         dev->caps.local_ca_ack_delay = dev_cap->local_ca_ack_delay;
 347         dev->caps.bf_reg_size        = dev_cap->bf_reg_size;
 348         dev->caps.bf_regs_per_page   = dev_cap->bf_regs_per_page;
 349         dev->caps.max_sq_sg          = dev_cap->max_sq_sg;
 350         dev->caps.max_rq_sg          = dev_cap->max_rq_sg;
 351         dev->caps.max_wqes           = dev_cap->max_qp_sz;
 352         dev->caps.max_qp_init_rdma   = dev_cap->max_requester_per_qp;
 353         dev->caps.max_srq_wqes       = dev_cap->max_srq_sz;
 354         dev->caps.max_srq_sge        = dev_cap->max_rq_sg - 1;
 355         dev->caps.reserved_srqs      = dev_cap->reserved_srqs;
 356         dev->caps.max_sq_desc_sz     = dev_cap->max_sq_desc_sz;
 357         dev->caps.max_rq_desc_sz     = dev_cap->max_rq_desc_sz;
 358         /*
 359          * Subtract 1 from the limit because we need to allocate a
 360          * spare CQE so the HCA HW can tell the difference between an
 361          * empty CQ and a full CQ.
 362          */
 363         dev->caps.max_cqes           = dev_cap->max_cq_sz - 1;
 364         dev->caps.reserved_cqs       = dev_cap->reserved_cqs;
 365         dev->caps.reserved_eqs       = dev_cap->reserved_eqs;
 366         dev->caps.reserved_mtts      = dev_cap->reserved_mtts;
 367         dev->caps.reserved_mrws      = dev_cap->reserved_mrws;
 368
 369         /* The first 128 UARs are used for EQ doorbells */
 370         dev->caps.reserved_uars      = max_t(int, 128, dev_cap->reserved_uars);
 371         dev->caps.reserved_pds       = dev_cap->reserved_pds;
 372         dev->caps.reserved_xrcds     = (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) ?
 373                                         dev_cap->reserved_xrcds : 0;
 374         dev->caps.max_xrcds          = (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) ?
 375                                         dev_cap->max_xrcds : 0;
 376         dev->caps.mtt_entry_sz       = dev_cap->mtt_entry_sz;
 377
 378         dev->caps.max_msg_sz         = dev_cap->max_msg_sz;
 379         dev->caps.page_size_cap      = ~(u32) (dev_cap->min_page_sz - 1);
 380         dev->caps.flags              = dev_cap->flags;
 381         dev->caps.flags2             = dev_cap->flags2;
 382         dev->caps.bmme_flags         = dev_cap->bmme_flags;
 383         dev->caps.reserved_lkey      = dev_cap->reserved_lkey;
 384         dev->caps.stat_rate_support  = dev_cap->stat_rate_support;
 385         dev->caps.max_gso_sz         = dev_cap->max_gso_sz;
 386         dev->caps.max_rss_tbl_sz     = dev_cap->max_rss_tbl_sz;
 387
 388         /* Sense port always allowed on supported devices for ConnectX-1 and -2 */
 389         if (mlx4_priv(dev)->pci_dev_data & MLX4_PCI_DEV_FORCE_SENSE_PORT)
 390                 dev->caps.flags |= MLX4_DEV_CAP_FLAG_SENSE_SUPPORT;
 391         /* Don't do sense port on multifunction devices (for now at least) */
 392         if (mlx4_is_mfunc(dev))
 393                 dev->caps.flags &= ~MLX4_DEV_CAP_FLAG_SENSE_SUPPORT;
 394
 395         if (mlx4_low_memory_profile()) {
 396                 dev->caps.log_num_macs  = MLX4_MIN_LOG_NUM_MAC;
 397                 dev->caps.log_num_vlans = MLX4_MIN_LOG_NUM_VLANS;
 398         } else {
 399                 dev->caps.log_num_macs  = log_num_mac;
 400                 dev->caps.log_num_vlans = MLX4_LOG_NUM_VLANS;
 401         }
 402
 403         for (i = 1; i <= dev->caps.num_ports; ++i) {
 404                 dev->caps.port_type[i] = MLX4_PORT_TYPE_NONE;
 405                 if (dev->caps.supported_type[i]) {
 406                         /* if only ETH is supported - assign ETH */
 407                         if (dev->caps.supported_type[i] == MLX4_PORT_TYPE_ETH)
 408                                 dev->caps.port_type[i] = MLX4_PORT_TYPE_ETH;
 409                         /* if only IB is supported, assign IB */
 410                         else if (dev->caps.supported_type[i] ==
 411                                  MLX4_PORT_TYPE_IB)
 412                                 dev->caps.port_type[i] = MLX4_PORT_TYPE_IB;
 413                         else {
 414                                 /* if IB and ETH are supported, we set the port
 415                                  * type according to user selection of port type;
 416                                  * if user selected none, take the FW hint */
 417                                 if (port_type_array[i - 1] == MLX4_PORT_TYPE_NONE)
 418                                         dev->caps.port_type[i] = dev->caps.suggested_type[i] ?
 419                                                 MLX4_PORT_TYPE_ETH : MLX4_PORT_TYPE_IB;
 420                                 else
 421                                         dev->caps.port_type[i] = port_type_array[i - 1];
 422                         }
 423                 }
 424                 /*
 425                  * Link sensing is allowed on the port if 3 conditions are true:
 426                  * 1. Both protocols are supported on the port.
 427                  * 2. Different types are supported on the port
 428                  * 3. FW declared that it supports link sensing
 429                  */
 430                 mlx4_priv(dev)->sense.sense_allowed[i] =
 431                         ((dev->caps.supported_type[i] == MLX4_PORT_TYPE_AUTO) &&
 432                          (dev->caps.flags & MLX4_DEV_CAP_FLAG_DPDP) &&
 433                          (dev->caps.flags & MLX4_DEV_CAP_FLAG_SENSE_SUPPORT));
 434
 435                 /*
 436                  * If "default_sense" bit is set, we move the port to "AUTO" mode
 437                  * and perform sense_port FW command to try and set the correct
 438                  * port type from beginning
 439                  */
 440                 if (mlx4_priv(dev)->sense.sense_allowed[i] && dev->caps.default_sense[i]) {
 441                         enum mlx4_port_type sensed_port = MLX4_PORT_TYPE_NONE;
 442                         dev->caps.possible_type[i] = MLX4_PORT_TYPE_AUTO;
 443                         mlx4_SENSE_PORT(dev, i, &sensed_port);
 444                         if (sensed_port != MLX4_PORT_TYPE_NONE)
 445                                 dev->caps.port_type[i] = sensed_port;
 446                 } else {
 447                         dev->caps.possible_type[i] = dev->caps.port_type[i];
 448                 }
 449
 450                 if (dev->caps.log_num_macs > dev_cap->port_cap[i].log_max_macs) {
 451                         dev->caps.log_num_macs = dev_cap->port_cap[i].log_max_macs;
 452                         mlx4_warn(dev, "Requested number of MACs is too much for port %d, reducing to %d\n",
 453                                   i, 1 << dev->caps.log_num_macs);
 454                 }
 455                 if (dev->caps.log_num_vlans > dev_cap->port_cap[i].log_max_vlans) {
 456                         dev->caps.log_num_vlans = dev_cap->port_cap[i].log_max_vlans;
 457                         mlx4_warn(dev, "Requested number of VLANs is too much for port %d, reducing to %d\n",
 458                                   i, 1 << dev->caps.log_num_vlans);
 459                 }
 460         }
 461
 462         dev->caps.max_counters = 1 << ilog2(dev_cap->max_counters);
 463
 464         dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW] = dev_cap->reserved_qps;
 465         dev->caps.reserved_qps_cnt[MLX4_QP_REGION_ETH_ADDR] =
 466                 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_ADDR] =
 467                 (1 << dev->caps.log_num_macs) *
 468                 (1 << dev->caps.log_num_vlans) *
 469                 dev->caps.num_ports;
 470         dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_EXCH] = MLX4_NUM_FEXCH;
 471
 472         if (dev_cap->dmfs_high_rate_qpn_base > 0 &&
 473             dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_FS_EN)
 474                 dev->caps.dmfs_high_rate_qpn_base = dev_cap->dmfs_high_rate_qpn_base;
 475         else
 476                 dev->caps.dmfs_high_rate_qpn_base =
 477                         dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW];
 478
 479         if (dev_cap->dmfs_high_rate_qpn_range > 0 &&
 480             dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_FS_EN) {
 481                 dev->caps.dmfs_high_rate_qpn_range = dev_cap->dmfs_high_rate_qpn_range;
 482                 dev->caps.dmfs_high_steer_mode = MLX4_STEERING_DMFS_A0_DEFAULT;
 483                 dev->caps.flags2 |= MLX4_DEV_CAP_FLAG2_FS_A0;
 484         } else {
 485                 dev->caps.dmfs_high_steer_mode = MLX4_STEERING_DMFS_A0_NOT_SUPPORTED;
 486                 dev->caps.dmfs_high_rate_qpn_base =
 487                         dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW];
 488                 dev->caps.dmfs_high_rate_qpn_range = MLX4_A0_STEERING_TABLE_SIZE;
 489         }
 490
 491         dev->caps.reserved_qps_cnt[MLX4_QP_REGION_RSS_RAW_ETH] =
 492                 dev->caps.dmfs_high_rate_qpn_range;
 493
 494         dev->caps.reserved_qps = dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW] +
 495                 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_ETH_ADDR] +
 496                 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_ADDR] +
 497                 dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FC_EXCH];
 498
 499         dev->caps.sqp_demux = (mlx4_is_master(dev)) ? MLX4_MAX_NUM_SLAVES : 0;
 500
 501         if (!enable_64b_cqe_eqe && !mlx4_is_slave(dev)) {
 502                 if (dev_cap->flags &
 503                     (MLX4_DEV_CAP_FLAG_64B_CQE | MLX4_DEV_CAP_FLAG_64B_EQE)) {
 504                         mlx4_warn(dev, "64B EQEs/CQEs supported by the device but not enabled\n");
 505                         dev->caps.flags &= ~MLX4_DEV_CAP_FLAG_64B_CQE;
 506                         dev->caps.flags &= ~MLX4_DEV_CAP_FLAG_64B_EQE;
 507                 }
 508
 509                 if (dev_cap->flags2 &
 510                     (MLX4_DEV_CAP_FLAG2_CQE_STRIDE |
 511                      MLX4_DEV_CAP_FLAG2_EQE_STRIDE)) {
 512                         mlx4_warn(dev, "Disabling EQE/CQE stride per user request\n");
 513                         dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_CQE_STRIDE;
 514                         dev_cap->flags2 &= ~MLX4_DEV_CAP_FLAG2_EQE_STRIDE;
 515                 }
 516         }
 517
 518         if ((dev->caps.flags &
 519             (MLX4_DEV_CAP_FLAG_64B_CQE | MLX4_DEV_CAP_FLAG_64B_EQE)) &&
 520             mlx4_is_master(dev))
 521                 dev->caps.function_caps |= MLX4_FUNC_CAP_64B_EQE_CQE;
 522
 523         if (!mlx4_is_slave(dev)) {
 524                 mlx4_enable_cqe_eqe_stride(dev);
 525                 dev->caps.alloc_res_qp_mask =
 526                         (dev->caps.bf_reg_size ? MLX4_RESERVE_ETH_BF_QP : 0) |
 527                         MLX4_RESERVE_A0_QP;
 528         } else {
 529                 dev->caps.alloc_res_qp_mask = 0;
 530         }
 531
 532         return 0;
 533 }
 534
 535 static int mlx4_get_pcie_dev_link_caps(struct mlx4_dev *dev,
 536                                        enum pci_bus_speed *speed,
 537                                        enum pcie_link_width *width)
 538 {
 539         u32 lnkcap1, lnkcap2;
 540         int err1, err2;
 541
 542 #define  PCIE_MLW_CAP_SHIFT 4   /* start of MLW mask in link capabilities */
 543
 544         *speed = PCI_SPEED_UNKNOWN;
 545         *width = PCIE_LNK_WIDTH_UNKNOWN;
 546
 547         err1 = pcie_capability_read_dword(dev->persist->pdev, PCI_EXP_LNKCAP,
 548                                           &lnkcap1);
 549         err2 = pcie_capability_read_dword(dev->persist->pdev, PCI_EXP_LNKCAP2,
 550                                           &lnkcap2);
 551         if (!err2 && lnkcap2) { /* PCIe r3.0-compliant */
 552                 if (lnkcap2 & PCI_EXP_LNKCAP2_SLS_8_0GB)
 553                         *speed = PCIE_SPEED_8_0GT;
 554                 else if (lnkcap2 & PCI_EXP_LNKCAP2_SLS_5_0GB)
 555                         *speed = PCIE_SPEED_5_0GT;
 556                 else if (lnkcap2 & PCI_EXP_LNKCAP2_SLS_2_5GB)
 557                         *speed = PCIE_SPEED_2_5GT;
 558         }
 559         if (!err1) {
 560                 *width = (lnkcap1 & PCI_EXP_LNKCAP_MLW) >> PCIE_MLW_CAP_SHIFT;
 561                 if (!lnkcap2) { /* pre-r3.0 */
 562                         if (lnkcap1 & PCI_EXP_LNKCAP_SLS_5_0GB)
 563                                 *speed = PCIE_SPEED_5_0GT;
 564                         else if (lnkcap1 & PCI_EXP_LNKCAP_SLS_2_5GB)
 565                                 *speed = PCIE_SPEED_2_5GT;
 566                 }
 567         }
 568
 569         if (*speed == PCI_SPEED_UNKNOWN || *width == PCIE_LNK_WIDTH_UNKNOWN) {
 570                 return err1 ? err1 :
 571                         err2 ? err2 : -EINVAL;
 572         }
 573         return 0;
 574 }
 575
 576 static void mlx4_check_pcie_caps(struct mlx4_dev *dev)
 577 {
 578         enum pcie_link_width width, width_cap;
 579         enum pci_bus_speed speed, speed_cap;
 580         int err;
 581
 582 #define PCIE_SPEED_STR(speed) \
 583         (speed == PCIE_SPEED_8_0GT ? "8.0GT/s" : \
 584          speed == PCIE_SPEED_5_0GT ? "5.0GT/s" : \
 585          speed == PCIE_SPEED_2_5GT ? "2.5GT/s" : \
 586          "Unknown")
 587
 588         err = mlx4_get_pcie_dev_link_caps(dev, &speed_cap, &width_cap);
 589         if (err) {
 590                 mlx4_warn(dev,
 591                           "Unable to determine PCIe device BW capabilities\n");
 592                 return;
 593         }
 594
 595         err = pcie_get_minimum_link(dev->persist->pdev, &speed, &width);
 596         if (err || speed == PCI_SPEED_UNKNOWN ||
 597             width == PCIE_LNK_WIDTH_UNKNOWN) {
 598                 mlx4_warn(dev,
 599                           "Unable to determine PCI device chain minimum BW\n");
 600                 return;
 601         }
 602
 603         if (width != width_cap || speed != speed_cap)
 604                 mlx4_warn(dev,
 605                           "PCIe BW is different than device's capability\n");
 606
 607         mlx4_info(dev, "PCIe link speed is %s, device supports %s\n",
 608                   PCIE_SPEED_STR(speed), PCIE_SPEED_STR(speed_cap));
 609         mlx4_info(dev, "PCIe link width is x%d, device supports x%d\n",
 610                   width, width_cap);
 611         return;
 612 }
 613
 614 /*The function checks if there are live vf, return the num of them*/
 615 static int mlx4_how_many_lives_vf(struct mlx4_dev *dev)
 616 {
 617         struct mlx4_priv *priv = mlx4_priv(dev);
 618         struct mlx4_slave_state *s_state;
 619         int i;
 620         int ret = 0;
 621
 622         for (i = 1/*the ppf is 0*/; i < dev->num_slaves; ++i) {
 623                 s_state = &priv->mfunc.master.slave_state[i];
 624                 if (s_state->active && s_state->last_cmd !=
 625                     MLX4_COMM_CMD_RESET) {
 626                         mlx4_warn(dev, "%s: slave: %d is still active\n",
 627                                   __func__, i);
 628                         ret++;
 629                 }
 630         }
 631         return ret;
 632 }
 633
 634 int mlx4_get_parav_qkey(struct mlx4_dev *dev, u32 qpn, u32 *qkey)
 635 {
 636         u32 qk = MLX4_RESERVED_QKEY_BASE;
 637
 638         if (qpn >= dev->phys_caps.base_tunnel_sqpn + 8 * MLX4_MFUNC_MAX ||
 639             qpn < dev->phys_caps.base_proxy_sqpn)
 640                 return -EINVAL;
 641
 642         if (qpn >= dev->phys_caps.base_tunnel_sqpn)
 643                 /* tunnel qp */
 644                 qk += qpn - dev->phys_caps.base_tunnel_sqpn;
 645         else
 646                 qk += qpn - dev->phys_caps.base_proxy_sqpn;
 647         *qkey = qk;
 648         return 0;
 649 }
 650 EXPORT_SYMBOL(mlx4_get_parav_qkey);
 651
 652 void mlx4_sync_pkey_table(struct mlx4_dev *dev, int slave, int port, int i, int val)
 653 {
 654         struct mlx4_priv *priv = container_of(dev, struct mlx4_priv, dev);
 655
 656         if (!mlx4_is_master(dev))
 657                 return;
 658
 659         priv->virt2phys_pkey[slave][port - 1][i] = val;
 660 }
 661 EXPORT_SYMBOL(mlx4_sync_pkey_table);
 662
 663 void mlx4_put_slave_node_guid(struct mlx4_dev *dev, int slave, __be64 guid)
 664 {
 665         struct mlx4_priv *priv = container_of(dev, struct mlx4_priv, dev);
 666
 667         if (!mlx4_is_master(dev))
 668                 return;
 669
 670         priv->slave_node_guids[slave] = guid;
 671 }
 672 EXPORT_SYMBOL(mlx4_put_slave_node_guid);
 673
 674 __be64 mlx4_get_slave_node_guid(struct mlx4_dev *dev, int slave)
 675 {
 676         struct mlx4_priv *priv = container_of(dev, struct mlx4_priv, dev);
 677
 678         if (!mlx4_is_master(dev))
 679                 return 0;
 680
 681         return priv->slave_node_guids[slave];
 682 }
 683 EXPORT_SYMBOL(mlx4_get_slave_node_guid);
 684
 685 int mlx4_is_slave_active(struct mlx4_dev *dev, int slave)
 686 {
 687         struct mlx4_priv *priv = mlx4_priv(dev);
 688         struct mlx4_slave_state *s_slave;
 689
 690         if (!mlx4_is_master(dev))
 691                 return 0;
 692
 693         s_slave = &priv->mfunc.master.slave_state[slave];
 694         return !!s_slave->active;
 695 }
 696 EXPORT_SYMBOL(mlx4_is_slave_active);
 697
 698 static void slave_adjust_steering_mode(struct mlx4_dev *dev,
 699                                        struct mlx4_dev_cap *dev_cap,
 700                                        struct mlx4_init_hca_param *hca_param)
 701 {
 702         dev->caps.steering_mode = hca_param->steering_mode;
 703         if (dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED) {
 704                 dev->caps.num_qp_per_mgm = dev_cap->fs_max_num_qp_per_entry;
 705                 dev->caps.fs_log_max_ucast_qp_range_size =
 706                         dev_cap->fs_log_max_ucast_qp_range_size;
 707         } else
 708                 dev->caps.num_qp_per_mgm =
 709                         4 * ((1 << hca_param->log_mc_entry_sz)/16 - 2);
 710
 711         mlx4_dbg(dev, "Steering mode is: %s\n",
 712                  mlx4_steering_mode_str(dev->caps.steering_mode));
 713 }
 714
 715 static int mlx4_slave_cap(struct mlx4_dev *dev)
 716 {
 717         int                        err;
 718         u32                        page_size;
 719         struct mlx4_dev_cap        dev_cap;
 720         struct mlx4_func_cap       func_cap;
 721         struct mlx4_init_hca_param hca_param;
 722         u8                         i;
 723
 724         memset(&hca_param, 0, sizeof(hca_param));
 725         err = mlx4_QUERY_HCA(dev, &hca_param);
 726         if (err) {
 727                 mlx4_err(dev, "QUERY_HCA command failed, aborting\n");
 728                 return err;
 729         }
 730
 731         /* fail if the hca has an unknown global capability
 732          * at this time global_caps should be always zeroed
 733          */
 734         if (hca_param.global_caps) {
 735                 mlx4_err(dev, "Unknown hca global capabilities\n");
 736                 return -ENOSYS;
 737         }
 738
 739         mlx4_log_num_mgm_entry_size = hca_param.log_mc_entry_sz;
 740
 741         dev->caps.hca_core_clock = hca_param.hca_core_clock;
 742
 743         memset(&dev_cap, 0, sizeof(dev_cap));
 744         dev->caps.max_qp_dest_rdma = 1 << hca_param.log_rd_per_qp;
 745         err = mlx4_dev_cap(dev, &dev_cap);
 746         if (err) {
 747                 mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting\n");
 748                 return err;
 749         }
 750
 751         err = mlx4_QUERY_FW(dev);
 752         if (err)
 753                 mlx4_err(dev, "QUERY_FW command failed: could not get FW version\n");
 754
 755         page_size = ~dev->caps.page_size_cap + 1;
 756         mlx4_warn(dev, "HCA minimum page size:%d\n", page_size);
 757         if (page_size > PAGE_SIZE) {
 758                 mlx4_err(dev, "HCA minimum page size of %d bigger than kernel PAGE_SIZE of %ld, aborting\n",
 759                          page_size, PAGE_SIZE);
 760                 return -ENODEV;
 761         }
 762
 763         /* slave gets uar page size from QUERY_HCA fw command */
 764         dev->caps.uar_page_size = 1 << (hca_param.uar_page_sz + 12);
 765
 766         /* TODO: relax this assumption */
 767         if (dev->caps.uar_page_size != PAGE_SIZE) {
 768                 mlx4_err(dev, "UAR size:%d != kernel PAGE_SIZE of %ld\n",
 769                          dev->caps.uar_page_size, PAGE_SIZE);
 770                 return -ENODEV;
 771         }
 772
 773         memset(&func_cap, 0, sizeof(func_cap));
 774         err = mlx4_QUERY_FUNC_CAP(dev, 0, &func_cap);
 775         if (err) {
 776                 mlx4_err(dev, "QUERY_FUNC_CAP general command failed, aborting (%d)\n",
 777                          err);
 778                 return err;
 779         }
 780
 781         if ((func_cap.pf_context_behaviour | PF_CONTEXT_BEHAVIOUR_MASK) !=
 782             PF_CONTEXT_BEHAVIOUR_MASK) {
 783                 mlx4_err(dev, "Unknown pf context behaviour %x known flags %x\n",
 784                          func_cap.pf_context_behaviour, PF_CONTEXT_BEHAVIOUR_MASK);
 785                 return -ENOSYS;
 786         }
 787
 788         dev->caps.num_ports             = func_cap.num_ports;
 789         dev->quotas.qp                  = func_cap.qp_quota;
 790         dev->quotas.srq                 = func_cap.srq_quota;
 791         dev->quotas.cq                  = func_cap.cq_quota;
 792         dev->quotas.mpt                 = func_cap.mpt_quota;
 793         dev->quotas.mtt                 = func_cap.mtt_quota;
 794         dev->caps.num_qps               = 1 << hca_param.log_num_qps;
 795         dev->caps.num_srqs              = 1 << hca_param.log_num_srqs;
 796         dev->caps.num_cqs               = 1 << hca_param.log_num_cqs;
 797         dev->caps.num_mpts              = 1 << hca_param.log_mpt_sz;
 798         dev->caps.num_eqs               = func_cap.max_eq;
 799         dev->caps.reserved_eqs          = func_cap.reserved_eq;
 800         dev->caps.num_pds               = MLX4_NUM_PDS;
 801         dev->caps.num_mgms              = 0;
 802         dev->caps.num_amgms             = 0;
 803
 804         if (dev->caps.num_ports > MLX4_MAX_PORTS) {
 805                 mlx4_err(dev, "HCA has %d ports, but we only support %d, aborting\n",
 806                          dev->caps.num_ports, MLX4_MAX_PORTS);
 807                 return -ENODEV;
 808         }
 809
 810         dev->caps.qp0_qkey = kcalloc(dev->caps.num_ports, sizeof(u32), GFP_KERNEL);
 811         dev->caps.qp0_tunnel = kcalloc(dev->caps.num_ports, sizeof (u32), GFP_KERNEL);
 812         dev->caps.qp0_proxy = kcalloc(dev->caps.num_ports, sizeof (u32), GFP_KERNEL);
 813         dev->caps.qp1_tunnel = kcalloc(dev->caps.num_ports, sizeof (u32), GFP_KERNEL);
 814         dev->caps.qp1_proxy = kcalloc(dev->caps.num_ports, sizeof (u32), GFP_KERNEL);
 815
 816         if (!dev->caps.qp0_tunnel || !dev->caps.qp0_proxy ||
 817             !dev->caps.qp1_tunnel || !dev->caps.qp1_proxy ||
 818             !dev->caps.qp0_qkey) {
 819                 err = -ENOMEM;
 820                 goto err_mem;
 821         }
 822
 823         for (i = 1; i <= dev->caps.num_ports; ++i) {
 824                 err = mlx4_QUERY_FUNC_CAP(dev, i, &func_cap);
 825                 if (err) {
 826                         mlx4_err(dev, "QUERY_FUNC_CAP port command failed for port %d, aborting (%d)\n",
 827                                  i, err);
 828                         goto err_mem;
 829                 }
 830                 dev->caps.qp0_qkey[i - 1] = func_cap.qp0_qkey;
 831                 dev->caps.qp0_tunnel[i - 1] = func_cap.qp0_tunnel_qpn;
 832                 dev->caps.qp0_proxy[i - 1] = func_cap.qp0_proxy_qpn;
 833                 dev->caps.qp1_tunnel[i - 1] = func_cap.qp1_tunnel_qpn;
 834                 dev->caps.qp1_proxy[i - 1] = func_cap.qp1_proxy_qpn;
 835                 dev->caps.port_mask[i] = dev->caps.port_type[i];
 836                 dev->caps.phys_port_id[i] = func_cap.phys_port_id;
 837                 if (mlx4_get_slave_pkey_gid_tbl_len(dev, i,
 838                                                     &dev->caps.gid_table_len[i],
 839                                                     &dev->caps.pkey_table_len[i]))
 840                         goto err_mem;
 841         }
 842
 843         if (dev->caps.uar_page_size * (dev->caps.num_uars -
 844                                        dev->caps.reserved_uars) >
 845                                        pci_resource_len(dev->persist->pdev,
 846                                                         2)) {
 847                 mlx4_err(dev, "HCA reported UAR region size of 0x%x bigger than PCI resource 2 size of 0x%llx, aborting\n",
 848                          dev->caps.uar_page_size * dev->caps.num_uars,
 849                          (unsigned long long)
 850                          pci_resource_len(dev->persist->pdev, 2));
 851                 goto err_mem;
 852         }
 853
 854         if (hca_param.dev_cap_enabled & MLX4_DEV_CAP_64B_EQE_ENABLED) {
 855                 dev->caps.eqe_size   = 64;
 856                 dev->caps.eqe_factor = 1;
 857         } else {
 858                 dev->caps.eqe_size   = 32;
 859                 dev->caps.eqe_factor = 0;
 860         }
 861
 862         if (hca_param.dev_cap_enabled & MLX4_DEV_CAP_64B_CQE_ENABLED) {
 863                 dev->caps.cqe_size   = 64;
 864                 dev->caps.userspace_caps |= MLX4_USER_DEV_CAP_LARGE_CQE;
 865         } else {
 866                 dev->caps.cqe_size   = 32;
 867         }
 868
 869         if (hca_param.dev_cap_enabled & MLX4_DEV_CAP_EQE_STRIDE_ENABLED) {
 870                 dev->caps.eqe_size = hca_param.eqe_size;
 871                 dev->caps.eqe_factor = 0;
 872         }
 873
 874         if (hca_param.dev_cap_enabled & MLX4_DEV_CAP_CQE_STRIDE_ENABLED) {
 875                 dev->caps.cqe_size = hca_param.cqe_size;
 876                 /* User still need to know when CQE > 32B */
 877                 dev->caps.userspace_caps |= MLX4_USER_DEV_CAP_LARGE_CQE;
 878         }
 879
 880         dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_TS;
 881         mlx4_warn(dev, "Timestamping is not supported in slave mode\n");
 882
 883         slave_adjust_steering_mode(dev, &dev_cap, &hca_param);
 884
 885         if (func_cap.extra_flags & MLX4_QUERY_FUNC_FLAGS_BF_RES_QP &&
 886             dev->caps.bf_reg_size)
 887                 dev->caps.alloc_res_qp_mask |= MLX4_RESERVE_ETH_BF_QP;
 888
 889         if (func_cap.extra_flags & MLX4_QUERY_FUNC_FLAGS_A0_RES_QP)
 890                 dev->caps.alloc_res_qp_mask |= MLX4_RESERVE_A0_QP;
 891
 892         return 0;
 893
 894 err_mem:
 895         kfree(dev->caps.qp0_qkey);
 896         kfree(dev->caps.qp0_tunnel);
 897         kfree(dev->caps.qp0_proxy);
 898         kfree(dev->caps.qp1_tunnel);
 899         kfree(dev->caps.qp1_proxy);
 900         dev->caps.qp0_qkey = NULL;
 901         dev->caps.qp0_tunnel = NULL;
 902         dev->caps.qp0_proxy = NULL;
 903         dev->caps.qp1_tunnel = NULL;
 904         dev->caps.qp1_proxy = NULL;
 905
 906         return err;
 907 }
 908
 909 static void mlx4_request_modules(struct mlx4_dev *dev)
 910 {
 911         int port;
 912         int has_ib_port = false;
 913         int has_eth_port = false;
 914 #define EN_DRV_NAME     "mlx4_en"
 915 #define IB_DRV_NAME     "mlx4_ib"
 916
 917         for (port = 1; port <= dev->caps.num_ports; port++) {
 918                 if (dev->caps.port_type[port] == MLX4_PORT_TYPE_IB)
 919                         has_ib_port = true;
 920                 else if (dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH)
 921                         has_eth_port = true;
 922         }
 923
 924         if (has_eth_port)
 925                 request_module_nowait(EN_DRV_NAME);
 926         if (has_ib_port || (dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE))
 927                 request_module_nowait(IB_DRV_NAME);
 928 }
 929
 930 /*
 931  * Change the port configuration of the device.
 932  * Every user of this function must hold the port mutex.
 933  */
 934 int mlx4_change_port_types(struct mlx4_dev *dev,
 935                            enum mlx4_port_type *port_types)
 936 {
 937         int err = 0;
 938         int change = 0;
 939         int port;
 940
 941         for (port = 0; port <  dev->caps.num_ports; port++) {
 942                 /* Change the port type only if the new type is different
 943                  * from the current, and not set to Auto */
 944                 if (port_types[port] != dev->caps.port_type[port + 1])
 945                         change = 1;
 946         }
 947         if (change) {
 948                 mlx4_unregister_device(dev);
 949                 for (port = 1; port <= dev->caps.num_ports; port++) {
 950                         mlx4_CLOSE_PORT(dev, port);
 951                         dev->caps.port_type[port] = port_types[port - 1];
 952                         err = mlx4_SET_PORT(dev, port, -1);
 953                         if (err) {
 954                                 mlx4_err(dev, "Failed to set port %d, aborting\n",
 955                                          port);
 956                                 goto out;
 957                         }
 958                 }
 959                 mlx4_set_port_mask(dev);
 960                 err = mlx4_register_device(dev);
 961                 if (err) {
 962                         mlx4_err(dev, "Failed to register device\n");
 963                         goto out;
 964                 }
 965                 mlx4_request_modules(dev);
 966         }
 967
 968 out:
 969         return err;
 970 }
 971
 972 static ssize_t show_port_type(struct device *dev,
 973                               struct device_attribute *attr,
 974                               char *buf)
 975 {
 976         struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
 977                                                    port_attr);
 978         struct mlx4_dev *mdev = info->dev;
 979         char type[8];
 980
 981         sprintf(type, "%s",
 982                 (mdev->caps.port_type[info->port] == MLX4_PORT_TYPE_IB) ?
 983                 "ib" : "eth");
 984         if (mdev->caps.possible_type[info->port] == MLX4_PORT_TYPE_AUTO)
 985                 sprintf(buf, "auto (%s)\n", type);
 986         else
 987                 sprintf(buf, "%s\n", type);
 988
 989         return strlen(buf);
 990 }
 991
 992 static ssize_t set_port_type(struct device *dev,
 993                              struct device_attribute *attr,
 994                              const char *buf, size_t count)
 995 {
 996         struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
 997                                                    port_attr);
 998         struct mlx4_dev *mdev = info->dev;
 999         struct mlx4_priv *priv = mlx4_priv(mdev);
1000         enum mlx4_port_type types[MLX4_MAX_PORTS];
1001         enum mlx4_port_type new_types[MLX4_MAX_PORTS];
1002         static DEFINE_MUTEX(set_port_type_mutex);
1003         int i;
1004         int err = 0;
1005
1006         mutex_lock(&set_port_type_mutex);
1007
1008         if (!strcmp(buf, "ib\n"))
1009                 info->tmp_type = MLX4_PORT_TYPE_IB;
1010         else if (!strcmp(buf, "eth\n"))
1011                 info->tmp_type = MLX4_PORT_TYPE_ETH;
1012         else if (!strcmp(buf, "auto\n"))
1013                 info->tmp_type = MLX4_PORT_TYPE_AUTO;
1014         else {
1015                 mlx4_err(mdev, "%s is not supported port type\n", buf);
1016                 err = -EINVAL;
1017                 goto err_out;
1018         }
1019
1020         mlx4_stop_sense(mdev);
1021         mutex_lock(&priv->port_mutex);
1022         /* Possible type is always the one that was delivered */
1023         mdev->caps.possible_type[info->port] = info->tmp_type;
1024
1025         for (i = 0; i < mdev->caps.num_ports; i++) {
1026                 types[i] = priv->port[i+1].tmp_type ? priv->port[i+1].tmp_type :
1027                                         mdev->caps.possible_type[i+1];
1028                 if (types[i] == MLX4_PORT_TYPE_AUTO)
1029                         types[i] = mdev->caps.port_type[i+1];
1030         }
1031
1032         if (!(mdev->caps.flags & MLX4_DEV_CAP_FLAG_DPDP) &&
1033             !(mdev->caps.flags & MLX4_DEV_CAP_FLAG_SENSE_SUPPORT)) {
1034                 for (i = 1; i <= mdev->caps.num_ports; i++) {
1035                         if (mdev->caps.possible_type[i] == MLX4_PORT_TYPE_AUTO) {
1036                                 mdev->caps.possible_type[i] = mdev->caps.port_type[i];
1037                                 err = -EINVAL;
1038                         }
1039                 }
1040         }
1041         if (err) {
1042                 mlx4_err(mdev, "Auto sensing is not supported on this HCA. Set only 'eth' or 'ib' for both ports (should be the same)\n");
1043                 goto out;
1044         }
1045
1046         mlx4_do_sense_ports(mdev, new_types, types);
1047
1048         err = mlx4_check_port_params(mdev, new_types);
1049         if (err)
1050                 goto out;
1051
1052         /* We are about to apply the changes after the configuration
1053          * was verified, no need to remember the temporary types
1054          * any more */
1055         for (i = 0; i < mdev->caps.num_ports; i++)
1056                 priv->port[i + 1].tmp_type = 0;
1057
1058         err = mlx4_change_port_types(mdev, new_types);
1059
1060 out:
1061         mlx4_start_sense(mdev);
1062         mutex_unlock(&priv->port_mutex);
1063 err_out:
1064         mutex_unlock(&set_port_type_mutex);
1065
1066         return err ? err : count;
1067 }
1068
1069 enum ibta_mtu {
1070         IB_MTU_256  = 1,
1071         IB_MTU_512  = 2,
1072         IB_MTU_1024 = 3,
1073         IB_MTU_2048 = 4,
1074         IB_MTU_4096 = 5
1075 };
1076
1077 static inline int int_to_ibta_mtu(int mtu)
1078 {
1079         switch (mtu) {
1080         case 256:  return IB_MTU_256;
1081         case 512:  return IB_MTU_512;
1082         case 1024: return IB_MTU_1024;
1083         case 2048: return IB_MTU_2048;
1084         case 4096: return IB_MTU_4096;
1085         default: return -1;
1086         }
1087 }
1088
1089 static inline int ibta_mtu_to_int(enum ibta_mtu mtu)
1090 {
1091         switch (mtu) {
1092         case IB_MTU_256:  return  256;
1093         case IB_MTU_512:  return  512;
1094         case IB_MTU_1024: return 1024;
1095         case IB_MTU_2048: return 2048;
1096         case IB_MTU_4096: return 4096;
1097         default: return -1;
1098         }
1099 }
1100
1101 static ssize_t show_port_ib_mtu(struct device *dev,
1102                              struct device_attribute *attr,
1103                              char *buf)
1104 {
1105         struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
1106                                                    port_mtu_attr);
1107         struct mlx4_dev *mdev = info->dev;
1108
1109         if (mdev->caps.port_type[info->port] == MLX4_PORT_TYPE_ETH)
1110                 mlx4_warn(mdev, "port level mtu is only used for IB ports\n");
1111
1112         sprintf(buf, "%d\n",
1113                         ibta_mtu_to_int(mdev->caps.port_ib_mtu[info->port]));
1114         return strlen(buf);
1115 }
1116
1117 static ssize_t set_port_ib_mtu(struct device *dev,
1118                              struct device_attribute *attr,
1119                              const char *buf, size_t count)
1120 {
1121         struct mlx4_port_info *info = container_of(attr, struct mlx4_port_info,
1122                                                    port_mtu_attr);
1123         struct mlx4_dev *mdev = info->dev;
1124         struct mlx4_priv *priv = mlx4_priv(mdev);
1125         int err, port, mtu, ibta_mtu = -1;
1126
1127         if (mdev->caps.port_type[info->port] == MLX4_PORT_TYPE_ETH) {
1128                 mlx4_warn(mdev, "port level mtu is only used for IB ports\n");
1129                 return -EINVAL;
1130         }
1131
1132         err = kstrtoint(buf, 0, &mtu);
1133         if (!err)
1134                 ibta_mtu = int_to_ibta_mtu(mtu);
1135
1136         if (err || ibta_mtu < 0) {
1137                 mlx4_err(mdev, "%s is invalid IBTA mtu\n", buf);
1138                 return -EINVAL;
1139         }
1140
1141         mdev->caps.port_ib_mtu[info->port] = ibta_mtu;
1142
1143         mlx4_stop_sense(mdev);
1144         mutex_lock(&priv->port_mutex);
1145         mlx4_unregister_device(mdev);
1146         for (port = 1; port <= mdev->caps.num_ports; port++) {
1147                 mlx4_CLOSE_PORT(mdev, port);
1148                 err = mlx4_SET_PORT(mdev, port, -1);
1149                 if (err) {
1150                         mlx4_err(mdev, "Failed to set port %d, aborting\n",
1151                                  port);
1152                         goto err_set_port;
1153                 }
1154         }
1155         err = mlx4_register_device(mdev);
1156 err_set_port:
1157         mutex_unlock(&priv->port_mutex);
1158         mlx4_start_sense(mdev);
1159         return err ? err : count;
1160 }
1161
1162 static int mlx4_load_fw(struct mlx4_dev *dev)
1163 {
1164         struct mlx4_priv *priv = mlx4_priv(dev);
1165         int err;
1166
1167         priv->fw.fw_icm = mlx4_alloc_icm(dev, priv->fw.fw_pages,
1168                                          GFP_HIGHUSER | __GFP_NOWARN, 0);
1169         if (!priv->fw.fw_icm) {
1170                 mlx4_err(dev, "Couldn't allocate FW area, aborting\n");
1171                 return -ENOMEM;
1172         }
1173
1174         err = mlx4_MAP_FA(dev, priv->fw.fw_icm);
1175         if (err) {
1176                 mlx4_err(dev, "MAP_FA command failed, aborting\n");
1177                 goto err_free;
1178         }
1179
1180         err = mlx4_RUN_FW(dev);
1181         if (err) {
1182                 mlx4_err(dev, "RUN_FW command failed, aborting\n");
1183                 goto err_unmap_fa;
1184         }
1185
1186         return 0;
1187
1188 err_unmap_fa:
1189         mlx4_UNMAP_FA(dev);
1190
1191 err_free:
1192         mlx4_free_icm(dev, priv->fw.fw_icm, 0);
1193         return err;
1194 }
1195
1196 static int mlx4_init_cmpt_table(struct mlx4_dev *dev, u64 cmpt_base,
1197                                 int cmpt_entry_sz)
1198 {
1199         struct mlx4_priv *priv = mlx4_priv(dev);
1200         int err;
1201         int num_eqs;
1202
1203         err = mlx4_init_icm_table(dev, &priv->qp_table.cmpt_table,
1204                                   cmpt_base +
1205                                   ((u64) (MLX4_CMPT_TYPE_QP *
1206                                           cmpt_entry_sz) << MLX4_CMPT_SHIFT),
1207                                   cmpt_entry_sz, dev->caps.num_qps,
1208                                   dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1209                                   0, 0);
1210         if (err)
1211                 goto err;
1212
1213         err = mlx4_init_icm_table(dev, &priv->srq_table.cmpt_table,
1214                                   cmpt_base +
1215                                   ((u64) (MLX4_CMPT_TYPE_SRQ *
1216                                           cmpt_entry_sz) << MLX4_CMPT_SHIFT),
1217                                   cmpt_entry_sz, dev->caps.num_srqs,
1218                                   dev->caps.reserved_srqs, 0, 0);
1219         if (err)
1220                 goto err_qp;
1221
1222         err = mlx4_init_icm_table(dev, &priv->cq_table.cmpt_table,
1223                                   cmpt_base +
1224                                   ((u64) (MLX4_CMPT_TYPE_CQ *
1225                                           cmpt_entry_sz) << MLX4_CMPT_SHIFT),
1226                                   cmpt_entry_sz, dev->caps.num_cqs,
1227                                   dev->caps.reserved_cqs, 0, 0);
1228         if (err)
1229                 goto err_srq;
1230
1231         num_eqs = dev->phys_caps.num_phys_eqs;
1232         err = mlx4_init_icm_table(dev, &priv->eq_table.cmpt_table,
1233                                   cmpt_base +
1234                                   ((u64) (MLX4_CMPT_TYPE_EQ *
1235                                           cmpt_entry_sz) << MLX4_CMPT_SHIFT),
1236                                   cmpt_entry_sz, num_eqs, num_eqs, 0, 0);
1237         if (err)
1238                 goto err_cq;
1239
1240         return 0;
1241
1242 err_cq:
1243         mlx4_cleanup_icm_table(dev, &priv->cq_table.cmpt_table);
1244
1245 err_srq:
1246         mlx4_cleanup_icm_table(dev, &priv->srq_table.cmpt_table);
1247
1248 err_qp:
1249         mlx4_cleanup_icm_table(dev, &priv->qp_table.cmpt_table);
1250
1251 err:
1252         return err;
1253 }
1254
1255 static int mlx4_init_icm(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap,
1256                          struct mlx4_init_hca_param *init_hca, u64 icm_size)
1257 {
1258         struct mlx4_priv *priv = mlx4_priv(dev);
1259         u64 aux_pages;
1260         int num_eqs;
1261         int err;
1262
1263         err = mlx4_SET_ICM_SIZE(dev, icm_size, &aux_pages);
1264         if (err) {
1265                 mlx4_err(dev, "SET_ICM_SIZE command failed, aborting\n");
1266                 return err;
1267         }
1268
1269         mlx4_dbg(dev, "%lld KB of HCA context requires %lld KB aux memory\n",
1270                  (unsigned long long) icm_size >> 10,
1271                  (unsigned long long) aux_pages << 2);
1272
1273         priv->fw.aux_icm = mlx4_alloc_icm(dev, aux_pages,
1274                                           GFP_HIGHUSER | __GFP_NOWARN, 0);
1275         if (!priv->fw.aux_icm) {
1276                 mlx4_err(dev, "Couldn't allocate aux memory, aborting\n");
1277                 return -ENOMEM;
1278         }
1279
1280         err = mlx4_MAP_ICM_AUX(dev, priv->fw.aux_icm);
1281         if (err) {
1282                 mlx4_err(dev, "MAP_ICM_AUX command failed, aborting\n");
1283                 goto err_free_aux;
1284         }
1285
1286         err = mlx4_init_cmpt_table(dev, init_hca->cmpt_base, dev_cap->cmpt_entry_sz);
1287         if (err) {
1288                 mlx4_err(dev, "Failed to map cMPT context memory, aborting\n");
1289                 goto err_unmap_aux;
1290         }
1291
1292
1293         num_eqs = dev->phys_caps.num_phys_eqs;
1294         err = mlx4_init_icm_table(dev, &priv->eq_table.table,
1295                                   init_hca->eqc_base, dev_cap->eqc_entry_sz,
1296                                   num_eqs, num_eqs, 0, 0);
1297         if (err) {
1298                 mlx4_err(dev, "Failed to map EQ context memory, aborting\n");
1299                 goto err_unmap_cmpt;
1300         }
1301
1302         /*
1303          * Reserved MTT entries must be aligned up to a cacheline
1304          * boundary, since the FW will write to them, while the driver
1305          * writes to all other MTT entries. (The variable
1306          * dev->caps.mtt_entry_sz below is really the MTT segment
1307          * size, not the raw entry size)
1308          */
1309         dev->caps.reserved_mtts =
1310                 ALIGN(dev->caps.reserved_mtts * dev->caps.mtt_entry_sz,
1311                       dma_get_cache_alignment()) / dev->caps.mtt_entry_sz;
1312
1313         err = mlx4_init_icm_table(dev, &priv->mr_table.mtt_table,
1314                                   init_hca->mtt_base,
1315                                   dev->caps.mtt_entry_sz,
1316                                   dev->caps.num_mtts,
1317                                   dev->caps.reserved_mtts, 1, 0);
1318         if (err) {
1319                 mlx4_err(dev, "Failed to map MTT context memory, aborting\n");
1320                 goto err_unmap_eq;
1321         }
1322
1323         err = mlx4_init_icm_table(dev, &priv->mr_table.dmpt_table,
1324                                   init_hca->dmpt_base,
1325                                   dev_cap->dmpt_entry_sz,
1326                                   dev->caps.num_mpts,
1327                                   dev->caps.reserved_mrws, 1, 1);
1328         if (err) {
1329                 mlx4_err(dev, "Failed to map dMPT context memory, aborting\n");
1330                 goto err_unmap_mtt;
1331         }
1332
1333         err = mlx4_init_icm_table(dev, &priv->qp_table.qp_table,
1334                                   init_hca->qpc_base,
1335                                   dev_cap->qpc_entry_sz,
1336                                   dev->caps.num_qps,
1337                                   dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1338                                   0, 0);
1339         if (err) {
1340                 mlx4_err(dev, "Failed to map QP context memory, aborting\n");
1341                 goto err_unmap_dmpt;
1342         }
1343
1344         err = mlx4_init_icm_table(dev, &priv->qp_table.auxc_table,
1345                                   init_hca->auxc_base,
1346                                   dev_cap->aux_entry_sz,
1347                                   dev->caps.num_qps,
1348                                   dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1349                                   0, 0);
1350         if (err) {
1351                 mlx4_err(dev, "Failed to map AUXC context memory, aborting\n");
1352                 goto err_unmap_qp;
1353         }
1354
1355         err = mlx4_init_icm_table(dev, &priv->qp_table.altc_table,
1356                                   init_hca->altc_base,
1357                                   dev_cap->altc_entry_sz,
1358                                   dev->caps.num_qps,
1359                                   dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1360                                   0, 0);
1361         if (err) {
1362                 mlx4_err(dev, "Failed to map ALTC context memory, aborting\n");
1363                 goto err_unmap_auxc;
1364         }
1365
1366         err = mlx4_init_icm_table(dev, &priv->qp_table.rdmarc_table,
1367                                   init_hca->rdmarc_base,
1368                                   dev_cap->rdmarc_entry_sz << priv->qp_table.rdmarc_shift,
1369                                   dev->caps.num_qps,
1370                                   dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW],
1371                                   0, 0);
1372         if (err) {
1373                 mlx4_err(dev, "Failed to map RDMARC context memory, aborting\n");
1374                 goto err_unmap_altc;
1375         }
1376
1377         err = mlx4_init_icm_table(dev, &priv->cq_table.table,
1378                                   init_hca->cqc_base,
1379                                   dev_cap->cqc_entry_sz,
1380                                   dev->caps.num_cqs,
1381                                   dev->caps.reserved_cqs, 0, 0);
1382         if (err) {
1383                 mlx4_err(dev, "Failed to map CQ context memory, aborting\n");
1384                 goto err_unmap_rdmarc;
1385         }
1386
1387         err = mlx4_init_icm_table(dev, &priv->srq_table.table,
1388                                   init_hca->srqc_base,
1389                                   dev_cap->srq_entry_sz,
1390                                   dev->caps.num_srqs,
1391                                   dev->caps.reserved_srqs, 0, 0);
1392         if (err) {
1393                 mlx4_err(dev, "Failed to map SRQ context memory, aborting\n");
1394                 goto err_unmap_cq;
1395         }
1396
1397         /*
1398          * For flow steering device managed mode it is required to use
1399          * mlx4_init_icm_table. For B0 steering mode it's not strictly
1400          * required, but for simplicity just map the whole multicast
1401          * group table now.  The table isn't very big and it's a lot
1402          * easier than trying to track ref counts.
1403          */
1404         err = mlx4_init_icm_table(dev, &priv->mcg_table.table,
1405                                   init_hca->mc_base,
1406                                   mlx4_get_mgm_entry_size(dev),
1407                                   dev->caps.num_mgms + dev->caps.num_amgms,
1408                                   dev->caps.num_mgms + dev->caps.num_amgms,
1409                                   0, 0);
1410         if (err) {
1411                 mlx4_err(dev, "Failed to map MCG context memory, aborting\n");
1412                 goto err_unmap_srq;
1413         }
1414
1415         return 0;
1416
1417 err_unmap_srq:
1418         mlx4_cleanup_icm_table(dev, &priv->srq_table.table);
1419
1420 err_unmap_cq:
1421         mlx4_cleanup_icm_table(dev, &priv->cq_table.table);
1422
1423 err_unmap_rdmarc:
1424         mlx4_cleanup_icm_table(dev, &priv->qp_table.rdmarc_table);
1425
1426 err_unmap_altc:
1427         mlx4_cleanup_icm_table(dev, &priv->qp_table.altc_table);
1428
1429 err_unmap_auxc:
1430         mlx4_cleanup_icm_table(dev, &priv->qp_table.auxc_table);
1431
1432 err_unmap_qp:
1433         mlx4_cleanup_icm_table(dev, &priv->qp_table.qp_table);
1434
1435 err_unmap_dmpt:
1436         mlx4_cleanup_icm_table(dev, &priv->mr_table.dmpt_table);
1437
1438 err_unmap_mtt:
1439         mlx4_cleanup_icm_table(dev, &priv->mr_table.mtt_table);
1440
1441 err_unmap_eq:
1442         mlx4_cleanup_icm_table(dev, &priv->eq_table.table);
1443
1444 err_unmap_cmpt:
1445         mlx4_cleanup_icm_table(dev, &priv->eq_table.cmpt_table);
1446         mlx4_cleanup_icm_table(dev, &priv->cq_table.cmpt_table);
1447         mlx4_cleanup_icm_table(dev, &priv->srq_table.cmpt_table);
1448         mlx4_cleanup_icm_table(dev, &priv->qp_table.cmpt_table);
1449
1450 err_unmap_aux:
1451         mlx4_UNMAP_ICM_AUX(dev);
1452
1453 err_free_aux:
1454         mlx4_free_icm(dev, priv->fw.aux_icm, 0);
1455
1456         return err;
1457 }
1458
1459 static void mlx4_free_icms(struct mlx4_dev *dev)
1460 {
1461         struct mlx4_priv *priv = mlx4_priv(dev);
1462
1463         mlx4_cleanup_icm_table(dev, &priv->mcg_table.table);
1464         mlx4_cleanup_icm_table(dev, &priv->srq_table.table);
1465         mlx4_cleanup_icm_table(dev, &priv->cq_table.table);
1466         mlx4_cleanup_icm_table(dev, &priv->qp_table.rdmarc_table);
1467         mlx4_cleanup_icm_table(dev, &priv->qp_table.altc_table);
1468         mlx4_cleanup_icm_table(dev, &priv->qp_table.auxc_table);
1469         mlx4_cleanup_icm_table(dev, &priv->qp_table.qp_table);
1470         mlx4_cleanup_icm_table(dev, &priv->mr_table.dmpt_table);
1471         mlx4_cleanup_icm_table(dev, &priv->mr_table.mtt_table);
1472         mlx4_cleanup_icm_table(dev, &priv->eq_table.table);
1473         mlx4_cleanup_icm_table(dev, &priv->eq_table.cmpt_table);
1474         mlx4_cleanup_icm_table(dev, &priv->cq_table.cmpt_table);
1475         mlx4_cleanup_icm_table(dev, &priv->srq_table.cmpt_table);
1476         mlx4_cleanup_icm_table(dev, &priv->qp_table.cmpt_table);
1477
1478         mlx4_UNMAP_ICM_AUX(dev);
1479         mlx4_free_icm(dev, priv->fw.aux_icm, 0);
1480 }
1481
1482 static void mlx4_slave_exit(struct mlx4_dev *dev)
1483 {
1484         struct mlx4_priv *priv = mlx4_priv(dev);
1485
1486         mutex_lock(&priv->cmd.slave_cmd_mutex);
1487         if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_RESET, 0, MLX4_COMM_CMD_NA_OP,
1488                           MLX4_COMM_TIME))
1489                 mlx4_warn(dev, "Failed to close slave function\n");
1490         mutex_unlock(&priv->cmd.slave_cmd_mutex);
1491 }
1492
1493 static int map_bf_area(struct mlx4_dev *dev)
1494 {
1495         struct mlx4_priv *priv = mlx4_priv(dev);
1496         resource_size_t bf_start;
1497         resource_size_t bf_len;
1498         int err = 0;
1499
1500         if (!dev->caps.bf_reg_size)
1501                 return -ENXIO;
1502
1503         bf_start = pci_resource_start(dev->persist->pdev, 2) +
1504                         (dev->caps.num_uars << PAGE_SHIFT);
1505         bf_len = pci_resource_len(dev->persist->pdev, 2) -
1506                         (dev->caps.num_uars << PAGE_SHIFT);
1507         priv->bf_mapping = io_mapping_create_wc(bf_start, bf_len);
1508         if (!priv->bf_mapping)
1509                 err = -ENOMEM;
1510
1511         return err;
1512 }
1513
1514 static void unmap_bf_area(struct mlx4_dev *dev)
1515 {
1516         if (mlx4_priv(dev)->bf_mapping)
1517                 io_mapping_free(mlx4_priv(dev)->bf_mapping);
1518 }
1519
1520 cycle_t mlx4_read_clock(struct mlx4_dev *dev)
1521 {
1522         u32 clockhi, clocklo, clockhi1;
1523         cycle_t cycles;
1524         int i;
1525         struct mlx4_priv *priv = mlx4_priv(dev);
1526
1527         for (i = 0; i < 10; i++) {
1528                 clockhi = swab32(readl(priv->clock_mapping));
1529                 clocklo = swab32(readl(priv->clock_mapping + 4));
1530                 clockhi1 = swab32(readl(priv->clock_mapping));
1531                 if (clockhi == clockhi1)
1532                         break;
1533         }
1534
1535         cycles = (u64) clockhi << 32 | (u64) clocklo;
1536
1537         return cycles;
1538 }
1539 EXPORT_SYMBOL_GPL(mlx4_read_clock);
1540
1541
1542 static int map_internal_clock(struct mlx4_dev *dev)
1543 {
1544         struct mlx4_priv *priv = mlx4_priv(dev);
1545
1546         priv->clock_mapping =
1547                 ioremap(pci_resource_start(dev->persist->pdev,
1548                                            priv->fw.clock_bar) +
1549                         priv->fw.clock_offset, MLX4_CLOCK_SIZE);
1550
1551         if (!priv->clock_mapping)
1552                 return -ENOMEM;
1553
1554         return 0;
1555 }
1556
1557 static void unmap_internal_clock(struct mlx4_dev *dev)
1558 {
1559         struct mlx4_priv *priv = mlx4_priv(dev);
1560
1561         if (priv->clock_mapping)
1562                 iounmap(priv->clock_mapping);
1563 }
1564
1565 static void mlx4_close_hca(struct mlx4_dev *dev)
1566 {
1567         unmap_internal_clock(dev);
1568         unmap_bf_area(dev);
1569         if (mlx4_is_slave(dev))
1570                 mlx4_slave_exit(dev);
1571         else {
1572                 mlx4_CLOSE_HCA(dev, 0);
1573                 mlx4_free_icms(dev);
1574         }
1575 }
1576
1577 static void mlx4_close_fw(struct mlx4_dev *dev)
1578 {
1579         if (!mlx4_is_slave(dev)) {
1580                 mlx4_UNMAP_FA(dev);
1581                 mlx4_free_icm(dev, mlx4_priv(dev)->fw.fw_icm, 0);
1582         }
1583 }
1584
1585 static int mlx4_comm_check_offline(struct mlx4_dev *dev)
1586 {
1587 #define COMM_CHAN_OFFLINE_OFFSET 0x09
1588
1589         u32 comm_flags;
1590         u32 offline_bit;
1591         unsigned long end;
1592         struct mlx4_priv *priv = mlx4_priv(dev);
1593
1594         end = msecs_to_jiffies(MLX4_COMM_OFFLINE_TIME_OUT) + jiffies;
1595         while (time_before(jiffies, end)) {
1596                 comm_flags = swab32(readl((__iomem char *)priv->mfunc.comm +
1597                                           MLX4_COMM_CHAN_FLAGS));
1598                 offline_bit = (comm_flags &
1599                                (u32)(1 << COMM_CHAN_OFFLINE_OFFSET));
1600                 if (!offline_bit)
1601                         return 0;
1602                 /* There are cases as part of AER/Reset flow that PF needs
1603                  * around 100 msec to load. We therefore sleep for 100 msec
1604                  * to allow other tasks to make use of that CPU during this
1605                  * time interval.
1606                  */
1607                 msleep(100);
1608         }
1609         mlx4_err(dev, "Communication channel is offline.\n");
1610         return -EIO;
1611 }
1612
1613 static void mlx4_reset_vf_support(struct mlx4_dev *dev)
1614 {
1615 #define COMM_CHAN_RST_OFFSET 0x1e
1616
1617         struct mlx4_priv *priv = mlx4_priv(dev);
1618         u32 comm_rst;
1619         u32 comm_caps;
1620
1621         comm_caps = swab32(readl((__iomem char *)priv->mfunc.comm +
1622                                  MLX4_COMM_CHAN_CAPS));
1623         comm_rst = (comm_caps & (u32)(1 << COMM_CHAN_RST_OFFSET));
1624
1625         if (comm_rst)
1626                 dev->caps.vf_caps |= MLX4_VF_CAP_FLAG_RESET;
1627 }
1628
1629 static int mlx4_init_slave(struct mlx4_dev *dev)
1630 {
1631         struct mlx4_priv *priv = mlx4_priv(dev);
1632         u64 dma = (u64) priv->mfunc.vhcr_dma;
1633         int ret_from_reset = 0;
1634         u32 slave_read;
1635         u32 cmd_channel_ver;
1636
1637         if (atomic_read(&pf_loading)) {
1638                 mlx4_warn(dev, "PF is not ready - Deferring probe\n");
1639                 return -EPROBE_DEFER;
1640         }
1641
1642         mutex_lock(&priv->cmd.slave_cmd_mutex);
1643         priv->cmd.max_cmds = 1;
1644         if (mlx4_comm_check_offline(dev)) {
1645                 mlx4_err(dev, "PF is not responsive, skipping initialization\n");
1646                 goto err_offline;
1647         }
1648
1649         mlx4_reset_vf_support(dev);
1650         mlx4_warn(dev, "Sending reset\n");
1651         ret_from_reset = mlx4_comm_cmd(dev, MLX4_COMM_CMD_RESET, 0,
1652                                        MLX4_COMM_CMD_NA_OP, MLX4_COMM_TIME);
1653         /* if we are in the middle of flr the slave will try
1654          * NUM_OF_RESET_RETRIES times before leaving.*/
1655         if (ret_from_reset) {
1656                 if (MLX4_DELAY_RESET_SLAVE == ret_from_reset) {
1657                         mlx4_warn(dev, "slave is currently in the middle of FLR - Deferring probe\n");
1658                         mutex_unlock(&priv->cmd.slave_cmd_mutex);
1659                         return -EPROBE_DEFER;
1660                 } else
1661                         goto err;
1662         }
1663
1664         /* check the driver version - the slave I/F revision
1665          * must match the master's */
1666         slave_read = swab32(readl(&priv->mfunc.comm->slave_read));
1667         cmd_channel_ver = mlx4_comm_get_version();
1668
1669         if (MLX4_COMM_GET_IF_REV(cmd_channel_ver) !=
1670                 MLX4_COMM_GET_IF_REV(slave_read)) {
1671                 mlx4_err(dev, "slave driver version is not supported by the master\n");
1672                 goto err;
1673         }
1674
1675         mlx4_warn(dev, "Sending vhcr0\n");
1676         if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR0, dma >> 48,
1677                              MLX4_COMM_CMD_NA_OP, MLX4_COMM_TIME))
1678                 goto err;
1679         if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR1, dma >> 32,
1680                              MLX4_COMM_CMD_NA_OP, MLX4_COMM_TIME))
1681                 goto err;
1682         if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR2, dma >> 16,
1683                              MLX4_COMM_CMD_NA_OP, MLX4_COMM_TIME))
1684                 goto err;
1685         if (mlx4_comm_cmd(dev, MLX4_COMM_CMD_VHCR_EN, dma,
1686                           MLX4_COMM_CMD_NA_OP, MLX4_COMM_TIME))
1687                 goto err;
1688
1689         mutex_unlock(&priv->cmd.slave_cmd_mutex);
1690         return 0;
1691
1692 err:
1693         mlx4_comm_cmd(dev, MLX4_COMM_CMD_RESET, 0, MLX4_COMM_CMD_NA_OP, 0);
1694 err_offline:
1695         mutex_unlock(&priv->cmd.slave_cmd_mutex);
1696         return -EIO;
1697 }
1698
1699 static void mlx4_parav_master_pf_caps(struct mlx4_dev *dev)
1700 {
1701         int i;
1702
1703         for (i = 1; i <= dev->caps.num_ports; i++) {
1704                 if (dev->caps.port_type[i] == MLX4_PORT_TYPE_ETH)
1705                         dev->caps.gid_table_len[i] =
1706                                 mlx4_get_slave_num_gids(dev, 0, i);
1707                 else
1708                         dev->caps.gid_table_len[i] = 1;
1709                 dev->caps.pkey_table_len[i] =
1710                         dev->phys_caps.pkey_phys_table_len[i] - 1;
1711         }
1712 }
1713
1714 static int choose_log_fs_mgm_entry_size(int qp_per_entry)
1715 {
1716         int i = MLX4_MIN_MGM_LOG_ENTRY_SIZE;
1717
1718         for (i = MLX4_MIN_MGM_LOG_ENTRY_SIZE; i <= MLX4_MAX_MGM_LOG_ENTRY_SIZE;
1719               i++) {
1720                 if (qp_per_entry <= 4 * ((1 << i) / 16 - 2))
1721                         break;
1722         }
1723
1724         return (i <= MLX4_MAX_MGM_LOG_ENTRY_SIZE) ? i : -1;
1725 }
1726
1727 static const char *dmfs_high_rate_steering_mode_str(int dmfs_high_steer_mode)
1728 {
1729         switch (dmfs_high_steer_mode) {
1730         case MLX4_STEERING_DMFS_A0_DEFAULT:
1731                 return "default performance";
1732
1733         case MLX4_STEERING_DMFS_A0_DYNAMIC:
1734                 return "dynamic hybrid mode";
1735
1736         case MLX4_STEERING_DMFS_A0_STATIC:
1737                 return "performance optimized for limited rule configuration (static)";
1738
1739         case MLX4_STEERING_DMFS_A0_DISABLE:
1740                 return "disabled performance optimized steering";
1741
1742         case MLX4_STEERING_DMFS_A0_NOT_SUPPORTED:
1743                 return "performance optimized steering not supported";
1744
1745         default:
1746                 return "Unrecognized mode";
1747         }
1748 }
1749
1750 #define MLX4_DMFS_A0_STEERING                   (1UL << 2)
1751
1752 static void choose_steering_mode(struct mlx4_dev *dev,
1753                                  struct mlx4_dev_cap *dev_cap)
1754 {
1755         if (mlx4_log_num_mgm_entry_size <= 0) {
1756                 if ((-mlx4_log_num_mgm_entry_size) & MLX4_DMFS_A0_STEERING) {
1757                         if (dev->caps.dmfs_high_steer_mode ==
1758                             MLX4_STEERING_DMFS_A0_NOT_SUPPORTED)
1759                                 mlx4_err(dev, "DMFS high rate mode not supported\n");
1760                         else
1761                                 dev->caps.dmfs_high_steer_mode =
1762                                         MLX4_STEERING_DMFS_A0_STATIC;
1763                 }
1764         }
1765
1766         if (mlx4_log_num_mgm_entry_size <= 0 &&
1767             dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_FS_EN &&
1768             (!mlx4_is_mfunc(dev) ||
1769              (dev_cap->fs_max_num_qp_per_entry >=
1770              (dev->persist->num_vfs + 1))) &&
1771             choose_log_fs_mgm_entry_size(dev_cap->fs_max_num_qp_per_entry) >=
1772                 MLX4_MIN_MGM_LOG_ENTRY_SIZE) {
1773                 dev->oper_log_mgm_entry_size =
1774                         choose_log_fs_mgm_entry_size(dev_cap->fs_max_num_qp_per_entry);
1775                 dev->caps.steering_mode = MLX4_STEERING_MODE_DEVICE_MANAGED;
1776                 dev->caps.num_qp_per_mgm = dev_cap->fs_max_num_qp_per_entry;
1777                 dev->caps.fs_log_max_ucast_qp_range_size =
1778                         dev_cap->fs_log_max_ucast_qp_range_size;
1779         } else {
1780                 if (dev->caps.dmfs_high_steer_mode !=
1781                     MLX4_STEERING_DMFS_A0_NOT_SUPPORTED)
1782                         dev->caps.dmfs_high_steer_mode = MLX4_STEERING_DMFS_A0_DISABLE;
1783                 if (dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_UC_STEER &&
1784                     dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_MC_STEER)
1785                         dev->caps.steering_mode = MLX4_STEERING_MODE_B0;
1786                 else {
1787                         dev->caps.steering_mode = MLX4_STEERING_MODE_A0;
1788
1789                         if (dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_UC_STEER ||
1790                             dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_MC_STEER)
1791                                 mlx4_warn(dev, "Must have both UC_STEER and MC_STEER flags set to use B0 steering - falling back to A0 steering mode\n");
1792                 }
1793                 dev->oper_log_mgm_entry_size =
1794                         mlx4_log_num_mgm_entry_size > 0 ?
1795                         mlx4_log_num_mgm_entry_size :
1796                         MLX4_DEFAULT_MGM_LOG_ENTRY_SIZE;
1797                 dev->caps.num_qp_per_mgm = mlx4_get_qp_per_mgm(dev);
1798         }
1799         mlx4_dbg(dev, "Steering mode is: %s, oper_log_mgm_entry_size = %d, modparam log_num_mgm_entry_size = %d\n",
1800                  mlx4_steering_mode_str(dev->caps.steering_mode),
1801                  dev->oper_log_mgm_entry_size,
1802                  mlx4_log_num_mgm_entry_size);
1803 }
1804
1805 static void choose_tunnel_offload_mode(struct mlx4_dev *dev,
1806                                        struct mlx4_dev_cap *dev_cap)
1807 {
1808         if (dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED &&
1809             dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_VXLAN_OFFLOADS)
1810                 dev->caps.tunnel_offload_mode = MLX4_TUNNEL_OFFLOAD_MODE_VXLAN;
1811         else
1812                 dev->caps.tunnel_offload_mode = MLX4_TUNNEL_OFFLOAD_MODE_NONE;
1813
1814         mlx4_dbg(dev, "Tunneling offload mode is: %s\n",  (dev->caps.tunnel_offload_mode
1815                  == MLX4_TUNNEL_OFFLOAD_MODE_VXLAN) ? "vxlan" : "none");
1816 }
1817
1818 static int mlx4_validate_optimized_steering(struct mlx4_dev *dev)
1819 {
1820         int i;
1821         struct mlx4_port_cap port_cap;
1822
1823         if (dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_NOT_SUPPORTED)
1824                 return -EINVAL;
1825
1826         for (i = 1; i <= dev->caps.num_ports; i++) {
1827                 if (mlx4_dev_port(dev, i, &port_cap)) {
1828                         mlx4_err(dev,
1829                                  "QUERY_DEV_CAP command failed, can't veify DMFS high rate steering.\n");
1830                 } else if ((dev->caps.dmfs_high_steer_mode !=
1831                             MLX4_STEERING_DMFS_A0_DEFAULT) &&
1832                            (port_cap.dmfs_optimized_state ==
1833                             !!(dev->caps.dmfs_high_steer_mode ==
1834                             MLX4_STEERING_DMFS_A0_DISABLE))) {
1835                         mlx4_err(dev,
1836                                  "DMFS high rate steer mode differ, driver requested %s but %s in FW.\n",
1837                                  dmfs_high_rate_steering_mode_str(
1838                                         dev->caps.dmfs_high_steer_mode),
1839                                  (port_cap.dmfs_optimized_state ?
1840                                         "enabled" : "disabled"));
1841                 }
1842         }
1843
1844         return 0;
1845 }
1846
1847 static int mlx4_init_fw(struct mlx4_dev *dev)
1848 {
1849         struct mlx4_mod_stat_cfg   mlx4_cfg;
1850         int err = 0;
1851
1852         if (!mlx4_is_slave(dev)) {
1853                 err = mlx4_QUERY_FW(dev);
1854                 if (err) {
1855                         if (err == -EACCES)
1856                                 mlx4_info(dev, "non-primary physical function, skipping\n");
1857                         else
1858                                 mlx4_err(dev, "QUERY_FW command failed, aborting\n");
1859                         return err;
1860                 }
1861
1862                 err = mlx4_load_fw(dev);
1863                 if (err) {
1864                         mlx4_err(dev, "Failed to start FW, aborting\n");
1865                         return err;
1866                 }
1867
1868                 mlx4_cfg.log_pg_sz_m = 1;
1869                 mlx4_cfg.log_pg_sz = 0;
1870                 err = mlx4_MOD_STAT_CFG(dev, &mlx4_cfg);
1871                 if (err)
1872                         mlx4_warn(dev, "Failed to override log_pg_sz parameter\n");
1873         }
1874
1875         return err;
1876 }
1877
1878 static int mlx4_init_hca(struct mlx4_dev *dev)
1879 {
1880         struct mlx4_priv          *priv = mlx4_priv(dev);
1881         struct mlx4_adapter        adapter;
1882         struct mlx4_dev_cap        dev_cap;
1883         struct mlx4_profile        profile;
1884         struct mlx4_init_hca_param init_hca;
1885         u64 icm_size;
1886         struct mlx4_config_dev_params params;
1887         int err;
1888
1889         if (!mlx4_is_slave(dev)) {
1890                 err = mlx4_dev_cap(dev, &dev_cap);
1891                 if (err) {
1892                         mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting\n");
1893                         return err;
1894                 }
1895
1896                 choose_steering_mode(dev, &dev_cap);
1897                 choose_tunnel_offload_mode(dev, &dev_cap);
1898
1899                 if (dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC &&
1900                     mlx4_is_master(dev))
1901                         dev->caps.function_caps |= MLX4_FUNC_CAP_DMFS_A0_STATIC;
1902
1903                 err = mlx4_get_phys_port_id(dev);
1904                 if (err)
1905                         mlx4_err(dev, "Fail to get physical port id\n");
1906
1907                 if (mlx4_is_master(dev))
1908                         mlx4_parav_master_pf_caps(dev);
1909
1910                 if (mlx4_low_memory_profile()) {
1911                         mlx4_info(dev, "Running from within kdump kernel. Using low memory profile\n");
1912                         profile = low_mem_profile;
1913                 } else {
1914                         profile = default_profile;
1915                 }
1916                 if (dev->caps.steering_mode ==
1917                     MLX4_STEERING_MODE_DEVICE_MANAGED)
1918                         profile.num_mcg = MLX4_FS_NUM_MCG;
1919
1920                 icm_size = mlx4_make_profile(dev, &profile, &dev_cap,
1921                                              &init_hca);
1922                 if ((long long) icm_size < 0) {
1923                         err = icm_size;
1924                         return err;
1925                 }
1926
1927                 dev->caps.max_fmr_maps = (1 << (32 - ilog2(dev->caps.num_mpts))) - 1;
1928
1929                 init_hca.log_uar_sz = ilog2(dev->caps.num_uars);
1930                 init_hca.uar_page_sz = PAGE_SHIFT - 12;
1931                 init_hca.mw_enabled = 0;
1932                 if (dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW ||
1933                     dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN)
1934                         init_hca.mw_enabled = INIT_HCA_TPT_MW_ENABLE;
1935
1936                 err = mlx4_init_icm(dev, &dev_cap, &init_hca, icm_size);
1937                 if (err)
1938                         return err;
1939
1940                 err = mlx4_INIT_HCA(dev, &init_hca);
1941                 if (err) {
1942                         mlx4_err(dev, "INIT_HCA command failed, aborting\n");
1943                         goto err_free_icm;
1944                 }
1945
1946                 if (dev_cap.flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS) {
1947                         err = mlx4_query_func(dev, &dev_cap);
1948                         if (err < 0) {
1949                                 mlx4_err(dev, "QUERY_FUNC command failed, aborting.\n");
1950                                 goto err_close;
1951                         } else if (err & MLX4_QUERY_FUNC_NUM_SYS_EQS) {
1952                                 dev->caps.num_eqs = dev_cap.max_eqs;
1953                                 dev->caps.reserved_eqs = dev_cap.reserved_eqs;
1954                                 dev->caps.reserved_uars = dev_cap.reserved_uars;
1955                         }
1956                 }
1957
1958                 /*
1959                  * If TS is supported by FW
1960                  * read HCA frequency by QUERY_HCA command
1961                  */
1962                 if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_TS) {
1963                         memset(&init_hca, 0, sizeof(init_hca));
1964                         err = mlx4_QUERY_HCA(dev, &init_hca);
1965                         if (err) {
1966                                 mlx4_err(dev, "QUERY_HCA command failed, disable timestamp\n");
1967                                 dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_TS;
1968                         } else {
1969                                 dev->caps.hca_core_clock =
1970                                         init_hca.hca_core_clock;
1971                         }
1972
1973                         /* In case we got HCA frequency 0 - disable timestamping
1974                          * to avoid dividing by zero
1975                          */
1976                         if (!dev->caps.hca_core_clock) {
1977                                 dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_TS;
1978                                 mlx4_err(dev,
1979                                          "HCA frequency is 0 - timestamping is not supported\n");
1980                         } else if (map_internal_clock(dev)) {
1981                                 /*
1982                                  * Map internal clock,
1983                                  * in case of failure disable timestamping
1984                                  */
1985                                 dev->caps.flags2 &= ~MLX4_DEV_CAP_FLAG2_TS;
1986                                 mlx4_err(dev, "Failed to map internal clock. Timestamping is not supported\n");
1987                         }
1988                 }
1989
1990                 if (dev->caps.dmfs_high_steer_mode !=
1991                     MLX4_STEERING_DMFS_A0_NOT_SUPPORTED) {
1992                         if (mlx4_validate_optimized_steering(dev))
1993                                 mlx4_warn(dev, "Optimized steering validation failed\n");
1994
1995                         if (dev->caps.dmfs_high_steer_mode ==
1996                             MLX4_STEERING_DMFS_A0_DISABLE) {
1997                                 dev->caps.dmfs_high_rate_qpn_base =
1998                                         dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW];
1999                                 dev->caps.dmfs_high_rate_qpn_range =
2000                                         MLX4_A0_STEERING_TABLE_SIZE;
2001                         }
2002
2003                         mlx4_dbg(dev, "DMFS high rate steer mode is: %s\n",
2004                                  dmfs_high_rate_steering_mode_str(
2005                                         dev->caps.dmfs_high_steer_mode));
2006                 }
2007         } else {
2008                 err = mlx4_init_slave(dev);
2009                 if (err) {
2010                         if (err != -EPROBE_DEFER)
2011                                 mlx4_err(dev, "Failed to initialize slave\n");
2012                         return err;
2013                 }
2014
2015                 err = mlx4_slave_cap(dev);
2016                 if (err) {
2017                         mlx4_err(dev, "Failed to obtain slave caps\n");
2018                         goto err_close;
2019                 }
2020         }
2021
2022         if (map_bf_area(dev))
2023                 mlx4_dbg(dev, "Failed to map blue flame area\n");
2024
2025         /*Only the master set the ports, all the rest got it from it.*/
2026         if (!mlx4_is_slave(dev))
2027                 mlx4_set_port_mask(dev);
2028
2029         err = mlx4_QUERY_ADAPTER(dev, &adapter);
2030         if (err) {
2031                 mlx4_err(dev, "QUERY_ADAPTER command failed, aborting\n");
2032                 goto unmap_bf;
2033         }
2034
2035         /* Query CONFIG_DEV parameters */
2036         err = mlx4_config_dev_retrieval(dev, &params);
2037         if (err && err != -ENOTSUPP) {
2038                 mlx4_err(dev, "Failed to query CONFIG_DEV parameters\n");
2039         } else if (!err) {
2040                 dev->caps.rx_checksum_flags_port[1] = params.rx_csum_flags_port_1;
2041                 dev->caps.rx_checksum_flags_port[2] = params.rx_csum_flags_port_2;
2042         }
2043         priv->eq_table.inta_pin = adapter.inta_pin;
2044         memcpy(dev->board_id, adapter.board_id, sizeof dev->board_id);
2045
2046         return 0;
2047
2048 unmap_bf:
2049         unmap_internal_clock(dev);
2050         unmap_bf_area(dev);
2051
2052         if (mlx4_is_slave(dev)) {
2053                 kfree(dev->caps.qp0_qkey);
2054                 kfree(dev->caps.qp0_tunnel);
2055                 kfree(dev->caps.qp0_proxy);
2056                 kfree(dev->caps.qp1_tunnel);
2057                 kfree(dev->caps.qp1_proxy);
2058         }
2059
2060 err_close:
2061         if (mlx4_is_slave(dev))
2062                 mlx4_slave_exit(dev);
2063         else
2064                 mlx4_CLOSE_HCA(dev, 0);
2065
2066 err_free_icm:
2067         if (!mlx4_is_slave(dev))
2068                 mlx4_free_icms(dev);
2069
2070         return err;
2071 }
2072
2073 static int mlx4_init_counters_table(struct mlx4_dev *dev)
2074 {
2075         struct mlx4_priv *priv = mlx4_priv(dev);
2076         int nent;
2077
2078         if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_COUNTERS))
2079                 return -ENOENT;
2080
2081         nent = dev->caps.max_counters;
2082         return mlx4_bitmap_init(&priv->counters_bitmap, nent, nent - 1, 0, 0);
2083 }
2084
2085 static void mlx4_cleanup_counters_table(struct mlx4_dev *dev)
2086 {
2087         mlx4_bitmap_cleanup(&mlx4_priv(dev)->counters_bitmap);
2088 }
2089
2090 int __mlx4_counter_alloc(struct mlx4_dev *dev, u32 *idx)
2091 {
2092         struct mlx4_priv *priv = mlx4_priv(dev);
2093
2094         if (!(dev->caps.flags & MLX4_DEV_CAP_FLAG_COUNTERS))
2095                 return -ENOENT;
2096
2097         *idx = mlx4_bitmap_alloc(&priv->counters_bitmap);
2098         if (*idx == -1)
2099                 return -ENOMEM;
2100
2101         return 0;
2102 }
2103
2104 int mlx4_counter_alloc(struct mlx4_dev *dev, u32 *idx)
2105 {
2106         u64 out_param;
2107         int err;
2108
2109         if (mlx4_is_mfunc(dev)) {
2110                 err = mlx4_cmd_imm(dev, 0, &out_param, RES_COUNTER,
2111                                    RES_OP_RESERVE, MLX4_CMD_ALLOC_RES,
2112                                    MLX4_CMD_TIME_CLASS_A, MLX4_CMD_WRAPPED);
2113                 if (!err)
2114                         *idx = get_param_l(&out_param);
2115
2116                 return err;
2117         }
2118         return __mlx4_counter_alloc(dev, idx);
2119 }
2120 EXPORT_SYMBOL_GPL(mlx4_counter_alloc);
2121
2122 void __mlx4_counter_free(struct mlx4_dev *dev, u32 idx)
2123 {
2124         mlx4_bitmap_free(&mlx4_priv(dev)->counters_bitmap, idx, MLX4_USE_RR);
2125         return;
2126 }
2127
2128 void mlx4_counter_free(struct mlx4_dev *dev, u32 idx)
2129 {
2130         u64 in_param = 0;
2131
2132         if (mlx4_is_mfunc(dev)) {
2133                 set_param_l(&in_param, idx);
2134                 mlx4_cmd(dev, in_param, RES_COUNTER, RES_OP_RESERVE,
2135                          MLX4_CMD_FREE_RES, MLX4_CMD_TIME_CLASS_A,
2136                          MLX4_CMD_WRAPPED);
2137                 return;
2138         }
2139         __mlx4_counter_free(dev, idx);
2140 }
2141 EXPORT_SYMBOL_GPL(mlx4_counter_free);
2142
2143 static int mlx4_setup_hca(struct mlx4_dev *dev)
2144 {
2145         struct mlx4_priv *priv = mlx4_priv(dev);
2146         int err;
2147         int port;
2148         __be32 ib_port_default_caps;
2149
2150         err = mlx4_init_uar_table(dev);
2151         if (err) {
2152                 mlx4_err(dev, "Failed to initialize user access region table, aborting\n");
2153                  return err;
2154         }
2155
2156         err = mlx4_uar_alloc(dev, &priv->driver_uar);
2157         if (err) {
2158                 mlx4_err(dev, "Failed to allocate driver access region, aborting\n");
2159                 goto err_uar_table_free;
2160         }
2161
2162         priv->kar = ioremap((phys_addr_t) priv->driver_uar.pfn << PAGE_SHIFT, PAGE_SIZE);
2163         if (!priv->kar) {
2164                 mlx4_err(dev, "Couldn't map kernel access region, aborting\n");
2165                 err = -ENOMEM;
2166                 goto err_uar_free;
2167         }
2168
2169         err = mlx4_init_pd_table(dev);
2170         if (err) {
2171                 mlx4_err(dev, "Failed to initialize protection domain table, aborting\n");
2172                 goto err_kar_unmap;
2173         }
2174
2175         err = mlx4_init_xrcd_table(dev);
2176         if (err) {
2177                 mlx4_err(dev, "Failed to initialize reliable connection domain table, aborting\n");
2178                 goto err_pd_table_free;
2179         }
2180
2181         err = mlx4_init_mr_table(dev);
2182         if (err) {
2183                 mlx4_err(dev, "Failed to initialize memory region table, aborting\n");
2184                 goto err_xrcd_table_free;
2185         }
2186
2187         if (!mlx4_is_slave(dev)) {
2188                 err = mlx4_init_mcg_table(dev);
2189                 if (err) {
2190                         mlx4_err(dev, "Failed to initialize multicast group table, aborting\n");
2191                         goto err_mr_table_free;
2192                 }
2193                 err = mlx4_config_mad_demux(dev);
2194                 if (err) {
2195                         mlx4_err(dev, "Failed in config_mad_demux, aborting\n");
2196                         goto err_mcg_table_free;
2197                 }
2198         }
2199
2200         err = mlx4_init_eq_table(dev);
2201         if (err) {
2202                 mlx4_err(dev, "Failed to initialize event queue table, aborting\n");
2203                 goto err_mcg_table_free;
2204         }
2205
2206         err = mlx4_cmd_use_events(dev);
2207         if (err) {
2208                 mlx4_err(dev, "Failed to switch to event-driven firmware commands, aborting\n");
2209                 goto err_eq_table_free;
2210         }
2211
2212         err = mlx4_NOP(dev);
2213         if (err) {
2214                 if (dev->flags & MLX4_FLAG_MSI_X) {
2215                         mlx4_warn(dev, "NOP command failed to generate MSI-X interrupt IRQ %d)\n",
2216                                   priv->eq_table.eq[dev->caps.num_comp_vectors].irq);
2217                         mlx4_warn(dev, "Trying again without MSI-X\n");
2218                 } else {
2219                         mlx4_err(dev, "NOP command failed to generate interrupt (IRQ %d), aborting\n",
2220                                  priv->eq_table.eq[dev->caps.num_comp_vectors].irq);
2221                         mlx4_err(dev, "BIOS or ACPI interrupt routing problem?\n");
2222                 }
2223
2224                 goto err_cmd_poll;
2225         }
2226
2227         mlx4_dbg(dev, "NOP command IRQ test passed\n");
2228
2229         err = mlx4_init_cq_table(dev);
2230         if (err) {
2231                 mlx4_err(dev, "Failed to initialize completion queue table, aborting\n");
2232                 goto err_cmd_poll;
2233         }
2234
2235         err = mlx4_init_srq_table(dev);
2236         if (err) {
2237                 mlx4_err(dev, "Failed to initialize shared receive queue table, aborting\n");
2238                 goto err_cq_table_free;
2239         }
2240
2241         err = mlx4_init_qp_table(dev);
2242         if (err) {
2243                 mlx4_err(dev, "Failed to initialize queue pair table, aborting\n");
2244                 goto err_srq_table_free;
2245         }
2246
2247         err = mlx4_init_counters_table(dev);
2248         if (err && err != -ENOENT) {
2249                 mlx4_err(dev, "Failed to initialize counters table, aborting\n");
2250                 goto err_qp_table_free;
2251         }
2252
2253         if (!mlx4_is_slave(dev)) {
2254                 for (port = 1; port <= dev->caps.num_ports; port++) {
2255                         ib_port_default_caps = 0;
2256                         err = mlx4_get_port_ib_caps(dev, port,
2257                                                     &ib_port_default_caps);
2258                         if (err)
2259                                 mlx4_warn(dev, "failed to get port %d default ib capabilities (%d). Continuing with caps = 0\n",
2260                                           port, err);
2261                         dev->caps.ib_port_def_cap[port] = ib_port_default_caps;
2262
2263                         /* initialize per-slave default ib port capabilities */
2264                         if (mlx4_is_master(dev)) {
2265                                 int i;
2266                                 for (i = 0; i < dev->num_slaves; i++) {
2267                                         if (i == mlx4_master_func_num(dev))
2268                                                 continue;
2269                                         priv->mfunc.master.slave_state[i].ib_cap_mask[port] =
2270                                                 ib_port_default_caps;
2271                                 }
2272                         }
2273
2274                         if (mlx4_is_mfunc(dev))
2275                                 dev->caps.port_ib_mtu[port] = IB_MTU_2048;
2276                         else
2277                                 dev->caps.port_ib_mtu[port] = IB_MTU_4096;
2278
2279                         err = mlx4_SET_PORT(dev, port, mlx4_is_master(dev) ?
2280                                             dev->caps.pkey_table_len[port] : -1);
2281                         if (err) {
2282                                 mlx4_err(dev, "Failed to set port %d, aborting\n",
2283                                          port);
2284                                 goto err_counters_table_free;
2285                         }
2286                 }
2287         }
2288
2289         return 0;
2290
2291 err_counters_table_free:
2292         mlx4_cleanup_counters_table(dev);
2293
2294 err_qp_table_free:
2295         mlx4_cleanup_qp_table(dev);
2296
2297 err_srq_table_free:
2298         mlx4_cleanup_srq_table(dev);
2299
2300 err_cq_table_free:
2301         mlx4_cleanup_cq_table(dev);
2302
2303 err_cmd_poll:
2304         mlx4_cmd_use_polling(dev);
2305
2306 err_eq_table_free:
2307         mlx4_cleanup_eq_table(dev);
2308
2309 err_mcg_table_free:
2310         if (!mlx4_is_slave(dev))
2311                 mlx4_cleanup_mcg_table(dev);
2312
2313 err_mr_table_free:
2314         mlx4_cleanup_mr_table(dev);
2315
2316 err_xrcd_table_free:
2317         mlx4_cleanup_xrcd_table(dev);
2318
2319 err_pd_table_free:
2320         mlx4_cleanup_pd_table(dev);
2321
2322 err_kar_unmap:
2323         iounmap(priv->kar);
2324
2325 err_uar_free:
2326         mlx4_uar_free(dev, &priv->driver_uar);
2327
2328 err_uar_table_free:
2329         mlx4_cleanup_uar_table(dev);
2330         return err;
2331 }
2332
2333 static void mlx4_enable_msi_x(struct mlx4_dev *dev)
2334 {
2335         struct mlx4_priv *priv = mlx4_priv(dev);
2336         struct msix_entry *entries;
2337         int i;
2338
2339         if (msi_x) {
2340                 int nreq = dev->caps.num_ports * num_online_cpus() + MSIX_LEGACY_SZ;
2341
2342                 nreq = min_t(int, dev->caps.num_eqs - dev->caps.reserved_eqs,
2343                              nreq);
2344
2345                 entries = kcalloc(nreq, sizeof *entries, GFP_KERNEL);
2346                 if (!entries)
2347                         goto no_msi;
2348
2349                 for (i = 0; i < nreq; ++i)
2350                         entries[i].entry = i;
2351
2352                 nreq = pci_enable_msix_range(dev->persist->pdev, entries, 2,
2353                                              nreq);
2354
2355                 if (nreq < 0) {
2356                         kfree(entries);
2357                         goto no_msi;
2358                 } else if (nreq < MSIX_LEGACY_SZ +
2359                            dev->caps.num_ports * MIN_MSIX_P_PORT) {
2360                         /*Working in legacy mode , all EQ's shared*/
2361                         dev->caps.comp_pool           = 0;
2362                         dev->caps.num_comp_vectors = nreq - 1;
2363                 } else {
2364                         dev->caps.comp_pool           = nreq - MSIX_LEGACY_SZ;
2365                         dev->caps.num_comp_vectors = MSIX_LEGACY_SZ - 1;
2366                 }
2367                 for (i = 0; i < nreq; ++i)
2368                         priv->eq_table.eq[i].irq = entries[i].vector;
2369
2370                 dev->flags |= MLX4_FLAG_MSI_X;
2371
2372                 kfree(entries);
2373                 return;
2374         }
2375
2376 no_msi:
2377         dev->caps.num_comp_vectors = 1;
2378         dev->caps.comp_pool        = 0;
2379
2380         for (i = 0; i < 2; ++i)
2381                 priv->eq_table.eq[i].irq = dev->persist->pdev->irq;
2382 }
2383
2384 static int mlx4_init_port_info(struct mlx4_dev *dev, int port)
2385 {
2386         struct mlx4_port_info *info = &mlx4_priv(dev)->port[port];
2387         int err = 0;
2388
2389         info->dev = dev;
2390         info->port = port;
2391         if (!mlx4_is_slave(dev)) {
2392                 mlx4_init_mac_table(dev, &info->mac_table);
2393                 mlx4_init_vlan_table(dev, &info->vlan_table);
2394                 mlx4_init_roce_gid_table(dev, &info->gid_table);
2395                 info->base_qpn = mlx4_get_base_qpn(dev, port);
2396         }
2397
2398         sprintf(info->dev_name, "mlx4_port%d", port);
2399         info->port_attr.attr.name = info->dev_name;
2400         if (mlx4_is_mfunc(dev))
2401                 info->port_attr.attr.mode = S_IRUGO;
2402         else {
2403                 info->port_attr.attr.mode = S_IRUGO | S_IWUSR;
2404                 info->port_attr.store     = set_port_type;
2405         }
2406         info->port_attr.show      = show_port_type;
2407         sysfs_attr_init(&info->port_attr.attr);
2408
2409         err = device_create_file(&dev->persist->pdev->dev, &info->port_attr);
2410         if (err) {
2411                 mlx4_err(dev, "Failed to create file for port %d\n", port);
2412                 info->port = -1;
2413         }
2414
2415         sprintf(info->dev_mtu_name, "mlx4_port%d_mtu", port);
2416         info->port_mtu_attr.attr.name = info->dev_mtu_name;
2417         if (mlx4_is_mfunc(dev))
2418                 info->port_mtu_attr.attr.mode = S_IRUGO;
2419         else {
2420                 info->port_mtu_attr.attr.mode = S_IRUGO | S_IWUSR;
2421                 info->port_mtu_attr.store     = set_port_ib_mtu;
2422         }
2423         info->port_mtu_attr.show      = show_port_ib_mtu;
2424         sysfs_attr_init(&info->port_mtu_attr.attr);
2425
2426         err = device_create_file(&dev->persist->pdev->dev,
2427                                  &info->port_mtu_attr);
2428         if (err) {
2429                 mlx4_err(dev, "Failed to create mtu file for port %d\n", port);
2430                 device_remove_file(&info->dev->persist->pdev->dev,
2431                                    &info->port_attr);
2432                 info->port = -1;
2433         }
2434
2435         return err;
2436 }
2437
2438 static void mlx4_cleanup_port_info(struct mlx4_port_info *info)
2439 {
2440         if (info->port < 0)
2441                 return;
2442
2443         device_remove_file(&info->dev->persist->pdev->dev, &info->port_attr);
2444         device_remove_file(&info->dev->persist->pdev->dev,
2445                            &info->port_mtu_attr);
2446 }
2447
2448 static int mlx4_init_steering(struct mlx4_dev *dev)
2449 {
2450         struct mlx4_priv *priv = mlx4_priv(dev);
2451         int num_entries = dev->caps.num_ports;
2452         int i, j;
2453
2454         priv->steer = kzalloc(sizeof(struct mlx4_steer) * num_entries, GFP_KERNEL);
2455         if (!priv->steer)
2456                 return -ENOMEM;
2457
2458         for (i = 0; i < num_entries; i++)
2459                 for (j = 0; j < MLX4_NUM_STEERS; j++) {
2460                         INIT_LIST_HEAD(&priv->steer[i].promisc_qps[j]);
2461                         INIT_LIST_HEAD(&priv->steer[i].steer_entries[j]);
2462                 }
2463         return 0;
2464 }
2465
2466 static void mlx4_clear_steering(struct mlx4_dev *dev)
2467 {
2468         struct mlx4_priv *priv = mlx4_priv(dev);
2469         struct mlx4_steer_index *entry, *tmp_entry;
2470         struct mlx4_promisc_qp *pqp, *tmp_pqp;
2471         int num_entries = dev->caps.num_ports;
2472         int i, j;
2473
2474         for (i = 0; i < num_entries; i++) {
2475                 for (j = 0; j < MLX4_NUM_STEERS; j++) {
2476                         list_for_each_entry_safe(pqp, tmp_pqp,
2477                                                  &priv->steer[i].promisc_qps[j],
2478                                                  list) {
2479                                 list_del(&pqp->list);
2480                                 kfree(pqp);
2481                         }
2482                         list_for_each_entry_safe(entry, tmp_entry,
2483                                                  &priv->steer[i].steer_entries[j],
2484                                                  list) {
2485                                 list_del(&entry->list);
2486                                 list_for_each_entry_safe(pqp, tmp_pqp,
2487                                                          &entry->duplicates,
2488                                                          list) {
2489                                         list_del(&pqp->list);
2490                                         kfree(pqp);
2491                                 }
2492                                 kfree(entry);
2493                         }
2494                 }
2495         }
2496         kfree(priv->steer);
2497 }
2498
2499 static int extended_func_num(struct pci_dev *pdev)
2500 {
2501         return PCI_SLOT(pdev->devfn) * 8 + PCI_FUNC(pdev->devfn);
2502 }
2503
2504 #define MLX4_OWNER_BASE 0x8069c
2505 #define MLX4_OWNER_SIZE 4
2506
2507 static int mlx4_get_ownership(struct mlx4_dev *dev)
2508 {
2509         void __iomem *owner;
2510         u32 ret;
2511
2512         if (pci_channel_offline(dev->persist->pdev))
2513                 return -EIO;
2514
2515         owner = ioremap(pci_resource_start(dev->persist->pdev, 0) +
2516                         MLX4_OWNER_BASE,
2517                         MLX4_OWNER_SIZE);
2518         if (!owner) {
2519                 mlx4_err(dev, "Failed to obtain ownership bit\n");
2520                 return -ENOMEM;
2521         }
2522
2523         ret = readl(owner);
2524         iounmap(owner);
2525         return (int) !!ret;
2526 }
2527
2528 static void mlx4_free_ownership(struct mlx4_dev *dev)
2529 {
2530         void __iomem *owner;
2531
2532         if (pci_channel_offline(dev->persist->pdev))
2533                 return;
2534
2535         owner = ioremap(pci_resource_start(dev->persist->pdev, 0) +
2536                         MLX4_OWNER_BASE,
2537                         MLX4_OWNER_SIZE);
2538         if (!owner) {
2539                 mlx4_err(dev, "Failed to obtain ownership bit\n");
2540                 return;
2541         }
2542         writel(0, owner);
2543         msleep(1000);
2544         iounmap(owner);
2545 }
2546
2547 #define SRIOV_VALID_STATE(flags) (!!((flags) & MLX4_FLAG_SRIOV) ==\
2548                                   !!((flags) & MLX4_FLAG_MASTER))
2549
2550 static u64 mlx4_enable_sriov(struct mlx4_dev *dev, struct pci_dev *pdev,
2551                              u8 total_vfs, int existing_vfs, int reset_flow)
2552 {
2553         u64 dev_flags = dev->flags;
2554         int err = 0;
2555
2556         if (reset_flow) {
2557                 dev->dev_vfs = kcalloc(total_vfs, sizeof(*dev->dev_vfs),
2558                                        GFP_KERNEL);
2559                 if (!dev->dev_vfs)
2560                         goto free_mem;
2561                 return dev_flags;
2562         }
2563
2564         atomic_inc(&pf_loading);
2565         if (dev->flags &  MLX4_FLAG_SRIOV) {
2566                 if (existing_vfs != total_vfs) {
2567                         mlx4_err(dev, "SR-IOV was already enabled, but with num_vfs (%d) different than requested (%d)\n",
2568                                  existing_vfs, total_vfs);
2569                         total_vfs = existing_vfs;
2570                 }
2571         }
2572
2573         dev->dev_vfs = kzalloc(total_vfs * sizeof(*dev->dev_vfs), GFP_KERNEL);
2574         if (NULL == dev->dev_vfs) {
2575                 mlx4_err(dev, "Failed to allocate memory for VFs\n");
2576                 goto disable_sriov;
2577         }
2578
2579         if (!(dev->flags &  MLX4_FLAG_SRIOV)) {
2580                 mlx4_warn(dev, "Enabling SR-IOV with %d VFs\n", total_vfs);
2581                 err = pci_enable_sriov(pdev, total_vfs);
2582         }
2583         if (err) {
2584                 mlx4_err(dev, "Failed to enable SR-IOV, continuing without SR-IOV (err = %d)\n",
2585                          err);
2586                 goto disable_sriov;
2587         } else {
2588                 mlx4_warn(dev, "Running in master mode\n");
2589                 dev_flags |= MLX4_FLAG_SRIOV |
2590                         MLX4_FLAG_MASTER;
2591                 dev_flags &= ~MLX4_FLAG_SLAVE;
2592                 dev->persist->num_vfs = total_vfs;
2593         }
2594         return dev_flags;
2595
2596 disable_sriov:
2597         atomic_dec(&pf_loading);
2598 free_mem:
2599         dev->persist->num_vfs = 0;
2600         kfree(dev->dev_vfs);
2601         return dev_flags & ~MLX4_FLAG_MASTER;
2602 }
2603
2604 enum {
2605         MLX4_DEV_CAP_CHECK_NUM_VFS_ABOVE_64 = -1,
2606 };
2607
2608 static int mlx4_check_dev_cap(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap,
2609                               int *nvfs)
2610 {
2611         int requested_vfs = nvfs[0] + nvfs[1] + nvfs[2];
2612         /* Checking for 64 VFs as a limitation of CX2 */
2613         if (!(dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_80_VFS) &&
2614             requested_vfs >= 64) {
2615                 mlx4_err(dev, "Requested %d VFs, but FW does not support more than 64\n",
2616                          requested_vfs);
2617                 return MLX4_DEV_CAP_CHECK_NUM_VFS_ABOVE_64;
2618         }
2619         return 0;
2620 }
2621
2622 static int mlx4_load_one(struct pci_dev *pdev, int pci_dev_data,
2623                          int total_vfs, int *nvfs, struct mlx4_priv *priv,
2624                          int reset_flow)
2625 {
2626         struct mlx4_dev *dev;
2627         unsigned sum = 0;
2628         int err;
2629         int port;
2630         int i;
2631         struct mlx4_dev_cap *dev_cap = NULL;
2632         int existing_vfs = 0;
2633
2634         dev = &priv->dev;
2635
2636         INIT_LIST_HEAD(&priv->ctx_list);
2637         spin_lock_init(&priv->ctx_lock);
2638
2639         mutex_init(&priv->port_mutex);
2640
2641         INIT_LIST_HEAD(&priv->pgdir_list);
2642         mutex_init(&priv->pgdir_mutex);
2643
2644         INIT_LIST_HEAD(&priv->bf_list);
2645         mutex_init(&priv->bf_mutex);
2646
2647         dev->rev_id = pdev->revision;
2648         dev->numa_node = dev_to_node(&pdev->dev);
2649
2650         /* Detect if this device is a virtual function */
2651         if (pci_dev_data & MLX4_PCI_DEV_IS_VF) {
2652                 mlx4_warn(dev, "Detected virtual function - running in slave mode\n");
2653                 dev->flags |= MLX4_FLAG_SLAVE;
2654         } else {
2655                 /* We reset the device and enable SRIOV only for physical
2656                  * devices.  Try to claim ownership on the device;
2657                  * if already taken, skip -- do not allow multiple PFs */
2658                 err = mlx4_get_ownership(dev);
2659                 if (err) {
2660                         if (err < 0)
2661                                 return err;
2662                         else {
2663                                 mlx4_warn(dev, "Multiple PFs not yet supported - Skipping PF\n");
2664                                 return -EINVAL;
2665                         }
2666                 }
2667
2668                 atomic_set(&priv->opreq_count, 0);
2669                 INIT_WORK(&priv->opreq_task, mlx4_opreq_action);
2670
2671                 /*
2672                  * Now reset the HCA before we touch the PCI capabilities or
2673                  * attempt a firmware command, since a boot ROM may have left
2674                  * the HCA in an undefined state.
2675                  */
2676                 err = mlx4_reset(dev);
2677                 if (err) {
2678                         mlx4_err(dev, "Failed to reset HCA, aborting\n");
2679                         goto err_sriov;
2680                 }
2681
2682                 if (total_vfs) {
2683                         dev->flags = MLX4_FLAG_MASTER;
2684                         existing_vfs = pci_num_vf(pdev);
2685                         if (existing_vfs)
2686                                 dev->flags |= MLX4_FLAG_SRIOV;
2687                         dev->persist->num_vfs = total_vfs;
2688                 }
2689         }
2690
2691         /* on load remove any previous indication of internal error,
2692          * device is up.
2693          */
2694         dev->persist->state = MLX4_DEVICE_STATE_UP;
2695
2696 slave_start:
2697         err = mlx4_cmd_init(dev);
2698         if (err) {
2699                 mlx4_err(dev, "Failed to init command interface, aborting\n");
2700                 goto err_sriov;
2701         }
2702
2703         /* In slave functions, the communication channel must be initialized
2704          * before posting commands. Also, init num_slaves before calling
2705          * mlx4_init_hca */
2706         if (mlx4_is_mfunc(dev)) {
2707                 if (mlx4_is_master(dev)) {
2708                         dev->num_slaves = MLX4_MAX_NUM_SLAVES;
2709
2710                 } else {
2711                         dev->num_slaves = 0;
2712                         err = mlx4_multi_func_init(dev);
2713                         if (err) {
2714                                 mlx4_err(dev, "Failed to init slave mfunc interface, aborting\n");
2715                                 goto err_cmd;
2716                         }
2717                 }
2718         }
2719
2720         err = mlx4_init_fw(dev);
2721         if (err) {
2722                 mlx4_err(dev, "Failed to init fw, aborting.\n");
2723                 goto err_mfunc;
2724         }
2725
2726         if (mlx4_is_master(dev)) {
2727                 /* when we hit the goto slave_start below, dev_cap already initialized */
2728                 if (!dev_cap) {
2729                         dev_cap = kzalloc(sizeof(*dev_cap), GFP_KERNEL);
2730
2731                         if (!dev_cap) {
2732                                 err = -ENOMEM;
2733                                 goto err_fw;
2734                         }
2735
2736                         err = mlx4_QUERY_DEV_CAP(dev, dev_cap);
2737                         if (err) {
2738                                 mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting.\n");
2739                                 goto err_fw;
2740                         }
2741
2742                         if (mlx4_check_dev_cap(dev, dev_cap, nvfs))
2743                                 goto err_fw;
2744
2745                         if (!(dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS)) {
2746                                 u64 dev_flags = mlx4_enable_sriov(dev, pdev,
2747                                                                   total_vfs,
2748                                                                   existing_vfs,
2749                                                                   reset_flow);
2750
2751                                 mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_ALL);
2752                                 dev->flags = dev_flags;
2753                                 if (!SRIOV_VALID_STATE(dev->flags)) {
2754                                         mlx4_err(dev, "Invalid SRIOV state\n");
2755                                         goto err_sriov;
2756                                 }
2757                                 err = mlx4_reset(dev);
2758                                 if (err) {
2759                                         mlx4_err(dev, "Failed to reset HCA, aborting.\n");
2760                                         goto err_sriov;
2761                                 }
2762                                 goto slave_start;
2763                         }
2764                 } else {
2765                         /* Legacy mode FW requires SRIOV to be enabled before
2766                          * doing QUERY_DEV_CAP, since max_eq's value is different if
2767                          * SRIOV is enabled.
2768                          */
2769                         memset(dev_cap, 0, sizeof(*dev_cap));
2770                         err = mlx4_QUERY_DEV_CAP(dev, dev_cap);
2771                         if (err) {
2772                                 mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting.\n");
2773                                 goto err_fw;
2774                         }
2775
2776                         if (mlx4_check_dev_cap(dev, dev_cap, nvfs))
2777                                 goto err_fw;
2778                 }
2779         }
2780
2781         err = mlx4_init_hca(dev);
2782         if (err) {
2783                 if (err == -EACCES) {
2784                         /* Not primary Physical function
2785                          * Running in slave mode */
2786                         mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_ALL);
2787                         /* We're not a PF */
2788                         if (dev->flags & MLX4_FLAG_SRIOV) {
2789                                 if (!existing_vfs)
2790                                         pci_disable_sriov(pdev);
2791                                 if (mlx4_is_master(dev) && !reset_flow)
2792                                         atomic_dec(&pf_loading);
2793                                 dev->flags &= ~MLX4_FLAG_SRIOV;
2794                         }
2795                         if (!mlx4_is_slave(dev))
2796                                 mlx4_free_ownership(dev);
2797                         dev->flags |= MLX4_FLAG_SLAVE;
2798                         dev->flags &= ~MLX4_FLAG_MASTER;
2799                         goto slave_start;
2800                 } else
2801                         goto err_fw;
2802         }
2803
2804         if (mlx4_is_master(dev) && (dev_cap->flags2 & MLX4_DEV_CAP_FLAG2_SYS_EQS)) {
2805                 u64 dev_flags = mlx4_enable_sriov(dev, pdev, total_vfs,
2806                                                   existing_vfs, reset_flow);
2807
2808                 if ((dev->flags ^ dev_flags) & (MLX4_FLAG_MASTER | MLX4_FLAG_SLAVE)) {
2809                         mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_VHCR);
2810                         dev->flags = dev_flags;
2811                         err = mlx4_cmd_init(dev);
2812                         if (err) {
2813                                 /* Only VHCR is cleaned up, so could still
2814                                  * send FW commands
2815                                  */
2816                                 mlx4_err(dev, "Failed to init VHCR command interface, aborting\n");
2817                                 goto err_close;
2818                         }
2819                 } else {
2820                         dev->flags = dev_flags;
2821                 }
2822
2823                 if (!SRIOV_VALID_STATE(dev->flags)) {
2824                         mlx4_err(dev, "Invalid SRIOV state\n");
2825                         goto err_close;
2826                 }
2827         }
2828
2829         /* check if the device is functioning at its maximum possible speed.
2830          * No return code for this call, just warn the user in case of PCI
2831          * express device capabilities are under-satisfied by the bus.
2832          */
2833         if (!mlx4_is_slave(dev))
2834                 mlx4_check_pcie_caps(dev);
2835
2836         /* In master functions, the communication channel must be initialized
2837          * after obtaining its address from fw */
2838         if (mlx4_is_master(dev)) {
2839                 int ib_ports = 0;
2840
2841                 mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
2842                         ib_ports++;
2843
2844                 if (ib_ports &&
2845                     (num_vfs_argc > 1 || probe_vfs_argc > 1)) {
2846                         mlx4_err(dev,
2847                                  "Invalid syntax of num_vfs/probe_vfs with IB port - single port VFs syntax is only supported when all ports are configured as ethernet\n");
2848                         err = -EINVAL;
2849                         goto err_close;
2850                 }
2851                 if (dev->caps.num_ports < 2 &&
2852                     num_vfs_argc > 1) {
2853                         err = -EINVAL;
2854                         mlx4_err(dev,
2855                                  "Error: Trying to configure VFs on port 2, but HCA has only %d physical ports\n",
2856                                  dev->caps.num_ports);
2857                         goto err_close;
2858                 }
2859                 memcpy(dev->persist->nvfs, nvfs, sizeof(dev->persist->nvfs));
2860
2861                 for (i = 0;
2862                      i < sizeof(dev->persist->nvfs)/
2863                      sizeof(dev->persist->nvfs[0]); i++) {
2864                         unsigned j;
2865
2866                         for (j = 0; j < dev->persist->nvfs[i]; ++sum, ++j) {
2867                                 dev->dev_vfs[sum].min_port = i < 2 ? i + 1 : 1;
2868                                 dev->dev_vfs[sum].n_ports = i < 2 ? 1 :
2869                                         dev->caps.num_ports;
2870                         }
2871                 }
2872
2873                 /* In master functions, the communication channel
2874                  * must be initialized after obtaining its address from fw
2875                  */
2876                 err = mlx4_multi_func_init(dev);
2877                 if (err) {
2878                         mlx4_err(dev, "Failed to init master mfunc interface, aborting.\n");
2879                         goto err_close;
2880                 }
2881         }
2882
2883         err = mlx4_alloc_eq_table(dev);
2884         if (err)
2885                 goto err_master_mfunc;
2886
2887         priv->msix_ctl.pool_bm = 0;
2888         mutex_init(&priv->msix_ctl.pool_lock);
2889
2890         mlx4_enable_msi_x(dev);
2891         if ((mlx4_is_mfunc(dev)) &&
2892             !(dev->flags & MLX4_FLAG_MSI_X)) {
2893                 err = -ENOSYS;
2894                 mlx4_err(dev, "INTx is not supported in multi-function mode, aborting\n");
2895                 goto err_free_eq;
2896         }
2897
2898         if (!mlx4_is_slave(dev)) {
2899                 err = mlx4_init_steering(dev);
2900                 if (err)
2901                         goto err_disable_msix;
2902         }
2903
2904         err = mlx4_setup_hca(dev);
2905         if (err == -EBUSY && (dev->flags & MLX4_FLAG_MSI_X) &&
2906             !mlx4_is_mfunc(dev)) {
2907                 dev->flags &= ~MLX4_FLAG_MSI_X;
2908                 dev->caps.num_comp_vectors = 1;
2909                 dev->caps.comp_pool        = 0;
2910                 pci_disable_msix(pdev);
2911                 err = mlx4_setup_hca(dev);
2912         }
2913
2914         if (err)
2915                 goto err_steer;
2916
2917         mlx4_init_quotas(dev);
2918         /* When PF resources are ready arm its comm channel to enable
2919          * getting commands
2920          */
2921         if (mlx4_is_master(dev)) {
2922                 err = mlx4_ARM_COMM_CHANNEL(dev);
2923                 if (err) {
2924                         mlx4_err(dev, " Failed to arm comm channel eq: %x\n",
2925                                  err);
2926                         goto err_steer;
2927                 }
2928         }
2929
2930         for (port = 1; port <= dev->caps.num_ports; port++) {
2931                 err = mlx4_init_port_info(dev, port);
2932                 if (err)
2933                         goto err_port;
2934         }
2935
2936         err = mlx4_register_device(dev);
2937         if (err)
2938                 goto err_port;
2939
2940         mlx4_request_modules(dev);
2941
2942         mlx4_sense_init(dev);
2943         mlx4_start_sense(dev);
2944
2945         priv->removed = 0;
2946
2947         if (mlx4_is_master(dev) && dev->persist->num_vfs && !reset_flow)
2948                 atomic_dec(&pf_loading);
2949
2950         kfree(dev_cap);
2951         return 0;
2952
2953 err_port:
2954         for (--port; port >= 1; --port)
2955                 mlx4_cleanup_port_info(&priv->port[port]);
2956
2957         mlx4_cleanup_counters_table(dev);
2958         mlx4_cleanup_qp_table(dev);
2959         mlx4_cleanup_srq_table(dev);
2960         mlx4_cleanup_cq_table(dev);
2961         mlx4_cmd_use_polling(dev);
2962         mlx4_cleanup_eq_table(dev);
2963         mlx4_cleanup_mcg_table(dev);
2964         mlx4_cleanup_mr_table(dev);
2965         mlx4_cleanup_xrcd_table(dev);
2966         mlx4_cleanup_pd_table(dev);
2967         mlx4_cleanup_uar_table(dev);
2968
2969 err_steer:
2970         if (!mlx4_is_slave(dev))
2971                 mlx4_clear_steering(dev);
2972
2973 err_disable_msix:
2974         if (dev->flags & MLX4_FLAG_MSI_X)
2975                 pci_disable_msix(pdev);
2976
2977 err_free_eq:
2978         mlx4_free_eq_table(dev);
2979
2980 err_master_mfunc:
2981         if (mlx4_is_master(dev))
2982                 mlx4_multi_func_cleanup(dev);
2983
2984         if (mlx4_is_slave(dev)) {
2985                 kfree(dev->caps.qp0_qkey);
2986                 kfree(dev->caps.qp0_tunnel);
2987                 kfree(dev->caps.qp0_proxy);
2988                 kfree(dev->caps.qp1_tunnel);
2989                 kfree(dev->caps.qp1_proxy);
2990         }
2991
2992 err_close:
2993         mlx4_close_hca(dev);
2994
2995 err_fw:
2996         mlx4_close_fw(dev);
2997
2998 err_mfunc:
2999         if (mlx4_is_slave(dev))
3000                 mlx4_multi_func_cleanup(dev);
3001
3002 err_cmd:
3003         mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_ALL);
3004
3005 err_sriov:
3006         if (dev->flags & MLX4_FLAG_SRIOV && !existing_vfs) {
3007                 pci_disable_sriov(pdev);
3008                 dev->flags &= ~MLX4_FLAG_SRIOV;
3009         }
3010
3011         if (mlx4_is_master(dev) && dev->persist->num_vfs && !reset_flow)
3012                 atomic_dec(&pf_loading);
3013
3014         kfree(priv->dev.dev_vfs);
3015
3016         if (!mlx4_is_slave(dev))
3017                 mlx4_free_ownership(dev);
3018
3019         kfree(dev_cap);
3020         return err;
3021 }
3022
3023 static int __mlx4_init_one(struct pci_dev *pdev, int pci_dev_data,
3024                            struct mlx4_priv *priv)
3025 {
3026         int err;
3027         int nvfs[MLX4_MAX_PORTS + 1] = {0, 0, 0};
3028         int prb_vf[MLX4_MAX_PORTS + 1] = {0, 0, 0};
3029         const int param_map[MLX4_MAX_PORTS + 1][MLX4_MAX_PORTS + 1] = {
3030                 {2, 0, 0}, {0, 1, 2}, {0, 1, 2} };
3031         unsigned total_vfs = 0;
3032         unsigned int i;
3033
3034         pr_info(DRV_NAME ": Initializing %s\n", pci_name(pdev));
3035
3036         err = pci_enable_device(pdev);
3037         if (err) {
3038                 dev_err(&pdev->dev, "Cannot enable PCI device, aborting\n");
3039                 return err;
3040         }
3041
3042         /* Due to requirement that all VFs and the PF are *guaranteed* 2 MACS
3043          * per port, we must limit the number of VFs to 63 (since their are
3044          * 128 MACs)
3045          */
3046         for (i = 0; i < sizeof(nvfs)/sizeof(nvfs[0]) && i < num_vfs_argc;
3047              total_vfs += nvfs[param_map[num_vfs_argc - 1][i]], i++) {
3048                 nvfs[param_map[num_vfs_argc - 1][i]] = num_vfs[i];
3049                 if (nvfs[i] < 0) {
3050                         dev_err(&pdev->dev, "num_vfs module parameter cannot be negative\n");
3051                         err = -EINVAL;
3052                         goto err_disable_pdev;
3053                 }
3054         }
3055         for (i = 0; i < sizeof(prb_vf)/sizeof(prb_vf[0]) && i < probe_vfs_argc;
3056              i++) {
3057                 prb_vf[param_map[probe_vfs_argc - 1][i]] = probe_vf[i];
3058                 if (prb_vf[i] < 0 || prb_vf[i] > nvfs[i]) {
3059                         dev_err(&pdev->dev, "probe_vf module parameter cannot be negative or greater than num_vfs\n");
3060                         err = -EINVAL;
3061                         goto err_disable_pdev;
3062                 }
3063         }
3064         if (total_vfs >= MLX4_MAX_NUM_VF) {
3065                 dev_err(&pdev->dev,
3066                         "Requested more VF's (%d) than allowed (%d)\n",
3067                         total_vfs, MLX4_MAX_NUM_VF - 1);
3068                 err = -EINVAL;
3069                 goto err_disable_pdev;
3070         }
3071
3072         for (i = 0; i < MLX4_MAX_PORTS; i++) {
3073                 if (nvfs[i] + nvfs[2] >= MLX4_MAX_NUM_VF_P_PORT) {
3074                         dev_err(&pdev->dev,
3075                                 "Requested more VF's (%d) for port (%d) than allowed (%d)\n",
3076                                 nvfs[i] + nvfs[2], i + 1,
3077                                 MLX4_MAX_NUM_VF_P_PORT - 1);
3078                         err = -EINVAL;
3079                         goto err_disable_pdev;
3080                 }
3081         }
3082
3083         /* Check for BARs. */
3084         if (!(pci_dev_data & MLX4_PCI_DEV_IS_VF) &&
3085             !(pci_resource_flags(pdev, 0) & IORESOURCE_MEM)) {
3086                 dev_err(&pdev->dev, "Missing DCS, aborting (driver_data: 0x%x, pci_resource_flags(pdev, 0):0x%lx)\n",
3087                         pci_dev_data, pci_resource_flags(pdev, 0));
3088                 err = -ENODEV;
3089                 goto err_disable_pdev;
3090         }
3091         if (!(pci_resource_flags(pdev, 2) & IORESOURCE_MEM)) {
3092                 dev_err(&pdev->dev, "Missing UAR, aborting\n");
3093                 err = -ENODEV;
3094                 goto err_disable_pdev;
3095         }
3096
3097         err = pci_request_regions(pdev, DRV_NAME);
3098         if (err) {
3099                 dev_err(&pdev->dev, "Couldn't get PCI resources, aborting\n");
3100                 goto err_disable_pdev;
3101         }
3102
3103         pci_set_master(pdev);
3104
3105         err = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
3106         if (err) {
3107                 dev_warn(&pdev->dev, "Warning: couldn't set 64-bit PCI DMA mask\n");
3108                 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
3109                 if (err) {
3110                         dev_err(&pdev->dev, "Can't set PCI DMA mask, aborting\n");
3111                         goto err_release_regions;
3112                 }
3113         }
3114         err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
3115         if (err) {
3116                 dev_warn(&pdev->dev, "Warning: couldn't set 64-bit consistent PCI DMA mask\n");
3117                 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
3118                 if (err) {
3119                         dev_err(&pdev->dev, "Can't set consistent PCI DMA mask, aborting\n");
3120                         goto err_release_regions;
3121                 }
3122         }
3123
3124         /* Allow large DMA segments, up to the firmware limit of 1 GB */
3125         dma_set_max_seg_size(&pdev->dev, 1024 * 1024 * 1024);
3126         /* Detect if this device is a virtual function */
3127         if (pci_dev_data & MLX4_PCI_DEV_IS_VF) {
3128                 /* When acting as pf, we normally skip vfs unless explicitly
3129                  * requested to probe them.
3130                  */
3131                 if (total_vfs) {
3132                         unsigned vfs_offset = 0;
3133
3134                         for (i = 0; i < sizeof(nvfs)/sizeof(nvfs[0]) &&
3135                              vfs_offset + nvfs[i] < extended_func_num(pdev);
3136                              vfs_offset += nvfs[i], i++)
3137                                 ;
3138                         if (i == sizeof(nvfs)/sizeof(nvfs[0])) {
3139                                 err = -ENODEV;
3140                                 goto err_release_regions;
3141                         }
3142                         if ((extended_func_num(pdev) - vfs_offset)
3143                             > prb_vf[i]) {
3144                                 dev_warn(&pdev->dev, "Skipping virtual function:%d\n",
3145                                          extended_func_num(pdev));
3146                                 err = -ENODEV;
3147                                 goto err_release_regions;
3148                         }
3149                 }
3150         }
3151
3152         err = mlx4_catas_init(&priv->dev);
3153         if (err)
3154                 goto err_release_regions;
3155
3156         err = mlx4_load_one(pdev, pci_dev_data, total_vfs, nvfs, priv, 0);
3157         if (err)
3158                 goto err_catas;
3159
3160         return 0;
3161
3162 err_catas:
3163         mlx4_catas_end(&priv->dev);
3164
3165 err_release_regions:
3166         pci_release_regions(pdev);
3167
3168 err_disable_pdev:
3169         pci_disable_device(pdev);
3170         pci_set_drvdata(pdev, NULL);
3171         return err;
3172 }
3173
3174 static int mlx4_init_one(struct pci_dev *pdev, const struct pci_device_id *id)
3175 {
3176         struct mlx4_priv *priv;
3177         struct mlx4_dev *dev;
3178         int ret;
3179
3180         printk_once(KERN_INFO "%s", mlx4_version);
3181
3182         priv = kzalloc(sizeof(*priv), GFP_KERNEL);
3183         if (!priv)
3184                 return -ENOMEM;
3185
3186         dev       = &priv->dev;
3187         dev->persist = kzalloc(sizeof(*dev->persist), GFP_KERNEL);
3188         if (!dev->persist) {
3189                 kfree(priv);
3190                 return -ENOMEM;
3191         }
3192         dev->persist->pdev = pdev;
3193         dev->persist->dev = dev;
3194         pci_set_drvdata(pdev, dev->persist);
3195         priv->pci_dev_data = id->driver_data;
3196         mutex_init(&dev->persist->device_state_mutex);
3197         mutex_init(&dev->persist->interface_state_mutex);
3198
3199         ret =  __mlx4_init_one(pdev, id->driver_data, priv);
3200         if (ret) {
3201                 kfree(dev->persist);
3202                 kfree(priv);
3203         } else {
3204                 pci_save_state(pdev);
3205         }
3206
3207         return ret;
3208 }
3209
3210 static void mlx4_clean_dev(struct mlx4_dev *dev)
3211 {
3212         struct mlx4_dev_persistent *persist = dev->persist;
3213         struct mlx4_priv *priv = mlx4_priv(dev);
3214         unsigned long   flags = (dev->flags & RESET_PERSIST_MASK_FLAGS);
3215
3216         memset(priv, 0, sizeof(*priv));
3217         priv->dev.persist = persist;
3218         priv->dev.flags = flags;
3219 }
3220
3221 static void mlx4_unload_one(struct pci_dev *pdev)
3222 {
3223         struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
3224         struct mlx4_dev  *dev  = persist->dev;
3225         struct mlx4_priv *priv = mlx4_priv(dev);
3226         int               pci_dev_data;
3227         int p, i;
3228
3229         if (priv->removed)
3230                 return;
3231
3232         /* saving current ports type for further use */
3233         for (i = 0; i < dev->caps.num_ports; i++) {
3234                 dev->persist->curr_port_type[i] = dev->caps.port_type[i + 1];
3235                 dev->persist->curr_port_poss_type[i] = dev->caps.
3236                                                        possible_type[i + 1];
3237         }
3238
3239         pci_dev_data = priv->pci_dev_data;
3240
3241         mlx4_stop_sense(dev);
3242         mlx4_unregister_device(dev);
3243
3244         for (p = 1; p <= dev->caps.num_ports; p++) {
3245                 mlx4_cleanup_port_info(&priv->port[p]);
3246                 mlx4_CLOSE_PORT(dev, p);
3247         }
3248
3249         if (mlx4_is_master(dev))
3250                 mlx4_free_resource_tracker(dev,
3251                                            RES_TR_FREE_SLAVES_ONLY);
3252
3253         mlx4_cleanup_counters_table(dev);
3254         mlx4_cleanup_qp_table(dev);
3255         mlx4_cleanup_srq_table(dev);
3256         mlx4_cleanup_cq_table(dev);
3257         mlx4_cmd_use_polling(dev);
3258         mlx4_cleanup_eq_table(dev);
3259         mlx4_cleanup_mcg_table(dev);
3260         mlx4_cleanup_mr_table(dev);
3261         mlx4_cleanup_xrcd_table(dev);
3262         mlx4_cleanup_pd_table(dev);
3263
3264         if (mlx4_is_master(dev))
3265                 mlx4_free_resource_tracker(dev,
3266                                            RES_TR_FREE_STRUCTS_ONLY);
3267
3268         iounmap(priv->kar);
3269         mlx4_uar_free(dev, &priv->driver_uar);
3270         mlx4_cleanup_uar_table(dev);
3271         if (!mlx4_is_slave(dev))
3272                 mlx4_clear_steering(dev);
3273         mlx4_free_eq_table(dev);
3274         if (mlx4_is_master(dev))
3275                 mlx4_multi_func_cleanup(dev);
3276         mlx4_close_hca(dev);
3277         mlx4_close_fw(dev);
3278         if (mlx4_is_slave(dev))
3279                 mlx4_multi_func_cleanup(dev);
3280         mlx4_cmd_cleanup(dev, MLX4_CMD_CLEANUP_ALL);
3281
3282         if (dev->flags & MLX4_FLAG_MSI_X)
3283                 pci_disable_msix(pdev);
3284
3285         if (!mlx4_is_slave(dev))
3286                 mlx4_free_ownership(dev);
3287
3288         kfree(dev->caps.qp0_qkey);
3289         kfree(dev->caps.qp0_tunnel);
3290         kfree(dev->caps.qp0_proxy);
3291         kfree(dev->caps.qp1_tunnel);
3292         kfree(dev->caps.qp1_proxy);
3293         kfree(dev->dev_vfs);
3294
3295         mlx4_clean_dev(dev);
3296         priv->pci_dev_data = pci_dev_data;
3297         priv->removed = 1;
3298 }
3299
3300 static void mlx4_remove_one(struct pci_dev *pdev)
3301 {
3302         struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
3303         struct mlx4_dev  *dev  = persist->dev;
3304         struct mlx4_priv *priv = mlx4_priv(dev);
3305         int active_vfs = 0;
3306
3307         mutex_lock(&persist->interface_state_mutex);
3308         persist->interface_state |= MLX4_INTERFACE_STATE_DELETION;
3309         mutex_unlock(&persist->interface_state_mutex);
3310
3311         /* Disabling SR-IOV is not allowed while there are active vf's */
3312         if (mlx4_is_master(dev) && dev->flags & MLX4_FLAG_SRIOV) {
3313                 active_vfs = mlx4_how_many_lives_vf(dev);
3314                 if (active_vfs) {
3315                         pr_warn("Removing PF when there are active VF's !!\n");
3316                         pr_warn("Will not disable SR-IOV.\n");
3317                 }
3318         }
3319
3320         /* device marked to be under deletion running now without the lock
3321          * letting other tasks to be terminated
3322          */
3323         if (persist->interface_state & MLX4_INTERFACE_STATE_UP)
3324                 mlx4_unload_one(pdev);
3325         else
3326                 mlx4_info(dev, "%s: interface is down\n", __func__);
3327         mlx4_catas_end(dev);
3328         if (dev->flags & MLX4_FLAG_SRIOV && !active_vfs) {
3329                 mlx4_warn(dev, "Disabling SR-IOV\n");
3330                 pci_disable_sriov(pdev);
3331         }
3332
3333         pci_release_regions(pdev);
3334         pci_disable_device(pdev);
3335         kfree(dev->persist);
3336         kfree(priv);
3337         pci_set_drvdata(pdev, NULL);
3338 }
3339
3340 static int restore_current_port_types(struct mlx4_dev *dev,
3341                                       enum mlx4_port_type *types,
3342                                       enum mlx4_port_type *poss_types)
3343 {
3344         struct mlx4_priv *priv = mlx4_priv(dev);
3345         int err, i;
3346
3347         mlx4_stop_sense(dev);
3348
3349         mutex_lock(&priv->port_mutex);
3350         for (i = 0; i < dev->caps.num_ports; i++)
3351                 dev->caps.possible_type[i + 1] = poss_types[i];
3352         err = mlx4_change_port_types(dev, types);
3353         mlx4_start_sense(dev);
3354         mutex_unlock(&priv->port_mutex);
3355
3356         return err;
3357 }
3358
3359 int mlx4_restart_one(struct pci_dev *pdev)
3360 {
3361         struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
3362         struct mlx4_dev  *dev  = persist->dev;
3363         struct mlx4_priv *priv = mlx4_priv(dev);
3364         int nvfs[MLX4_MAX_PORTS + 1] = {0, 0, 0};
3365         int pci_dev_data, err, total_vfs;
3366
3367         pci_dev_data = priv->pci_dev_data;
3368         total_vfs = dev->persist->num_vfs;
3369         memcpy(nvfs, dev->persist->nvfs, sizeof(dev->persist->nvfs));
3370
3371         mlx4_unload_one(pdev);
3372         err = mlx4_load_one(pdev, pci_dev_data, total_vfs, nvfs, priv, 1);
3373         if (err) {
3374                 mlx4_err(dev, "%s: ERROR: mlx4_load_one failed, pci_name=%s, err=%d\n",
3375                          __func__, pci_name(pdev), err);
3376                 return err;
3377         }
3378
3379         err = restore_current_port_types(dev, dev->persist->curr_port_type,
3380                                          dev->persist->curr_port_poss_type);
3381         if (err)
3382                 mlx4_err(dev, "could not restore original port types (%d)\n",
3383                          err);
3384
3385         return err;
3386 }
3387
3388 static const struct pci_device_id mlx4_pci_table[] = {
3389         /* MT25408 "Hermon" SDR */
3390         { PCI_VDEVICE(MELLANOX, 0x6340), MLX4_PCI_DEV_FORCE_SENSE_PORT },
3391         /* MT25408 "Hermon" DDR */
3392         { PCI_VDEVICE(MELLANOX, 0x634a), MLX4_PCI_DEV_FORCE_SENSE_PORT },
3393         /* MT25408 "Hermon" QDR */
3394         { PCI_VDEVICE(MELLANOX, 0x6354), MLX4_PCI_DEV_FORCE_SENSE_PORT },
3395         /* MT25408 "Hermon" DDR PCIe gen2 */
3396         { PCI_VDEVICE(MELLANOX, 0x6732), MLX4_PCI_DEV_FORCE_SENSE_PORT },
3397         /* MT25408 "Hermon" QDR PCIe gen2 */
3398         { PCI_VDEVICE(MELLANOX, 0x673c), MLX4_PCI_DEV_FORCE_SENSE_PORT },
3399         /* MT25408 "Hermon" EN 10GigE */
3400         { PCI_VDEVICE(MELLANOX, 0x6368), MLX4_PCI_DEV_FORCE_SENSE_PORT },
3401         /* MT25408 "Hermon" EN 10GigE PCIe gen2 */
3402         { PCI_VDEVICE(MELLANOX, 0x6750), MLX4_PCI_DEV_FORCE_SENSE_PORT },
3403         /* MT25458 ConnectX EN 10GBASE-T 10GigE */
3404         { PCI_VDEVICE(MELLANOX, 0x6372), MLX4_PCI_DEV_FORCE_SENSE_PORT },
3405         /* MT25458 ConnectX EN 10GBASE-T+Gen2 10GigE */
3406         { PCI_VDEVICE(MELLANOX, 0x675a), MLX4_PCI_DEV_FORCE_SENSE_PORT },
3407         /* MT26468 ConnectX EN 10GigE PCIe gen2*/
3408         { PCI_VDEVICE(MELLANOX, 0x6764), MLX4_PCI_DEV_FORCE_SENSE_PORT },
3409         /* MT26438 ConnectX EN 40GigE PCIe gen2 5GT/s */
3410         { PCI_VDEVICE(MELLANOX, 0x6746), MLX4_PCI_DEV_FORCE_SENSE_PORT },
3411         /* MT26478 ConnectX2 40GigE PCIe gen2 */
3412         { PCI_VDEVICE(MELLANOX, 0x676e), MLX4_PCI_DEV_FORCE_SENSE_PORT },
3413         /* MT25400 Family [ConnectX-2 Virtual Function] */
3414         { PCI_VDEVICE(MELLANOX, 0x1002), MLX4_PCI_DEV_IS_VF },
3415         /* MT27500 Family [ConnectX-3] */
3416         { PCI_VDEVICE(MELLANOX, 0x1003), 0 },
3417         /* MT27500 Family [ConnectX-3 Virtual Function] */
3418         { PCI_VDEVICE(MELLANOX, 0x1004), MLX4_PCI_DEV_IS_VF },
3419         { PCI_VDEVICE(MELLANOX, 0x1005), 0 }, /* MT27510 Family */
3420         { PCI_VDEVICE(MELLANOX, 0x1006), 0 }, /* MT27511 Family */
3421         { PCI_VDEVICE(MELLANOX, 0x1007), 0 }, /* MT27520 Family */
3422         { PCI_VDEVICE(MELLANOX, 0x1008), 0 }, /* MT27521 Family */
3423         { PCI_VDEVICE(MELLANOX, 0x1009), 0 }, /* MT27530 Family */
3424         { PCI_VDEVICE(MELLANOX, 0x100a), 0 }, /* MT27531 Family */
3425         { PCI_VDEVICE(MELLANOX, 0x100b), 0 }, /* MT27540 Family */
3426         { PCI_VDEVICE(MELLANOX, 0x100c), 0 }, /* MT27541 Family */
3427         { PCI_VDEVICE(MELLANOX, 0x100d), 0 }, /* MT27550 Family */
3428         { PCI_VDEVICE(MELLANOX, 0x100e), 0 }, /* MT27551 Family */
3429         { PCI_VDEVICE(MELLANOX, 0x100f), 0 }, /* MT27560 Family */
3430         { PCI_VDEVICE(MELLANOX, 0x1010), 0 }, /* MT27561 Family */
3431         { 0, }
3432 };
3433
3434 MODULE_DEVICE_TABLE(pci, mlx4_pci_table);
3435
3436 static pci_ers_result_t mlx4_pci_err_detected(struct pci_dev *pdev,
3437                                               pci_channel_state_t state)
3438 {
3439         struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
3440
3441         mlx4_err(persist->dev, "mlx4_pci_err_detected was called\n");
3442         mlx4_enter_error_state(persist);
3443
3444         mutex_lock(&persist->interface_state_mutex);
3445         if (persist->interface_state & MLX4_INTERFACE_STATE_UP)
3446                 mlx4_unload_one(pdev);
3447
3448         mutex_unlock(&persist->interface_state_mutex);
3449         if (state == pci_channel_io_perm_failure)
3450                 return PCI_ERS_RESULT_DISCONNECT;
3451
3452         pci_disable_device(pdev);
3453         return PCI_ERS_RESULT_NEED_RESET;
3454 }
3455
3456 static pci_ers_result_t mlx4_pci_slot_reset(struct pci_dev *pdev)
3457 {
3458         struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
3459         struct mlx4_dev  *dev  = persist->dev;
3460         struct mlx4_priv *priv = mlx4_priv(dev);
3461         int               ret;
3462         int nvfs[MLX4_MAX_PORTS + 1] = {0, 0, 0};
3463         int total_vfs;
3464
3465         mlx4_err(dev, "mlx4_pci_slot_reset was called\n");
3466         ret = pci_enable_device(pdev);
3467         if (ret) {
3468                 mlx4_err(dev, "Can not re-enable device, ret=%d\n", ret);
3469                 return PCI_ERS_RESULT_DISCONNECT;
3470         }
3471
3472         pci_set_master(pdev);
3473         pci_restore_state(pdev);
3474         pci_save_state(pdev);
3475
3476         total_vfs = dev->persist->num_vfs;
3477         memcpy(nvfs, dev->persist->nvfs, sizeof(dev->persist->nvfs));
3478
3479         mutex_lock(&persist->interface_state_mutex);
3480         if (!(persist->interface_state & MLX4_INTERFACE_STATE_UP)) {
3481                 ret = mlx4_load_one(pdev, priv->pci_dev_data, total_vfs, nvfs,
3482                                     priv, 1);
3483                 if (ret) {
3484                         mlx4_err(dev, "%s: mlx4_load_one failed, ret=%d\n",
3485                                  __func__,  ret);
3486                         goto end;
3487                 }
3488
3489                 ret = restore_current_port_types(dev, dev->persist->
3490                                                  curr_port_type, dev->persist->
3491                                                  curr_port_poss_type);
3492                 if (ret)
3493                         mlx4_err(dev, "could not restore original port types (%d)\n", ret);
3494         }
3495 end:
3496         mutex_unlock(&persist->interface_state_mutex);
3497
3498         return ret ? PCI_ERS_RESULT_DISCONNECT : PCI_ERS_RESULT_RECOVERED;
3499 }
3500
3501 static void mlx4_shutdown(struct pci_dev *pdev)
3502 {
3503         struct mlx4_dev_persistent *persist = pci_get_drvdata(pdev);
3504
3505         mlx4_info(persist->dev, "mlx4_shutdown was called\n");
3506         mutex_lock(&persist->interface_state_mutex);
3507         if (persist->interface_state & MLX4_INTERFACE_STATE_UP)
3508                 mlx4_unload_one(pdev);
3509         mutex_unlock(&persist->interface_state_mutex);
3510 }
3511
3512 static const struct pci_error_handlers mlx4_err_handler = {
3513         .error_detected = mlx4_pci_err_detected,
3514         .slot_reset     = mlx4_pci_slot_reset,
3515 };
3516
3517 static struct pci_driver mlx4_driver = {
3518         .name           = DRV_NAME,
3519         .id_table       = mlx4_pci_table,
3520         .probe          = mlx4_init_one,
3521         .shutdown       = mlx4_shutdown,
3522         .remove         = mlx4_remove_one,
3523         .err_handler    = &mlx4_err_handler,
3524 };
3525
3526 static int __init mlx4_verify_params(void)
3527 {
3528         if ((log_num_mac < 0) || (log_num_mac > 7)) {
3529                 pr_warn("mlx4_core: bad num_mac: %d\n", log_num_mac);
3530                 return -1;
3531         }
3532
3533         if (log_num_vlan != 0)
3534                 pr_warn("mlx4_core: log_num_vlan - obsolete module param, using %d\n",
3535                         MLX4_LOG_NUM_VLANS);
3536
3537         if (use_prio != 0)
3538                 pr_warn("mlx4_core: use_prio - obsolete module param, ignored\n");
3539
3540         if ((log_mtts_per_seg < 1) || (log_mtts_per_seg > 7)) {
3541                 pr_warn("mlx4_core: bad log_mtts_per_seg: %d\n",
3542                         log_mtts_per_seg);
3543                 return -1;
3544         }
3545
3546         /* Check if module param for ports type has legal combination */
3547         if (port_type_array[0] == false && port_type_array[1] == true) {
3548                 pr_warn("Module parameter configuration ETH/IB is not supported. Switching to default configuration IB/IB\n");
3549                 port_type_array[0] = true;
3550         }
3551
3552         if (mlx4_log_num_mgm_entry_size < -7 ||
3553             (mlx4_log_num_mgm_entry_size > 0 &&
3554              (mlx4_log_num_mgm_entry_size < MLX4_MIN_MGM_LOG_ENTRY_SIZE ||
3555               mlx4_log_num_mgm_entry_size > MLX4_MAX_MGM_LOG_ENTRY_SIZE))) {
3556                 pr_warn("mlx4_core: mlx4_log_num_mgm_entry_size (%d) not in legal range (-7..0 or %d..%d)\n",
3557                         mlx4_log_num_mgm_entry_size,
3558                         MLX4_MIN_MGM_LOG_ENTRY_SIZE,
3559                         MLX4_MAX_MGM_LOG_ENTRY_SIZE);
3560                 return -1;
3561         }
3562
3563         return 0;
3564 }
3565
3566 static int __init mlx4_init(void)
3567 {
3568         int ret;
3569
3570         if (mlx4_verify_params())
3571                 return -EINVAL;
3572
3573
3574         mlx4_wq = create_singlethread_workqueue("mlx4");
3575         if (!mlx4_wq)
3576                 return -ENOMEM;
3577
3578         ret = pci_register_driver(&mlx4_driver);
3579         if (ret < 0)
3580                 destroy_workqueue(mlx4_wq);
3581         return ret < 0 ? ret : 0;
3582 }
3583
3584 static void __exit mlx4_cleanup(void)
3585 {
3586         pci_unregister_driver(&mlx4_driver);
3587         destroy_workqueue(mlx4_wq);
3588 }
3589
3590 module_init(mlx4_init);
3591 module_exit(mlx4_cleanup);