drivers/net/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 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
  42 #include <linux/mlx4/device.h>
  43 #include <linux/mlx4/doorbell.h>
  44
  45 #include "mlx4.h"
  46 #include "fw.h"
  47 #include "icm.h"
  48
  49 MODULE_AUTHOR("Roland Dreier");
  50 MODULE_DESCRIPTION("Mellanox ConnectX HCA low-level driver");
  51 MODULE_LICENSE("Dual BSD/GPL");
  52 MODULE_VERSION(DRV_VERSION);
  53
  54 #ifdef CONFIG_MLX4_DEBUG
  55
  56 int mlx4_debug_level = 0;
  57 module_param_named(debug_level, mlx4_debug_level, int, 0644);
  58 MODULE_PARM_DESC(debug_level, "Enable debug tracing if > 0");
  59
  60 #endif /* CONFIG_MLX4_DEBUG */
  61
  62 #ifdef CONFIG_PCI_MSI
  63
  64 static int msi_x = 1;
  65 module_param(msi_x, int, 0444);
  66 MODULE_PARM_DESC(msi_x, "attempt to use MSI-X if nonzero");
  67
  68 #else /* CONFIG_PCI_MSI */
  69
  70 #define msi_x (0)
  71
  72 #endif /* CONFIG_PCI_MSI */
  73
  74 static char mlx4_version[] __devinitdata =
  75         DRV_NAME ": Mellanox ConnectX core driver v"
  76         DRV_VERSION " (" DRV_RELDATE ")\n";
  77
  78 static struct mlx4_profile default_profile = {
  79         .num_qp         = 1 << 16,
  80         .num_srq        = 1 << 16,
  81         .rdmarc_per_qp  = 1 << 4,
  82         .num_cq         = 1 << 16,
  83         .num_mcg        = 1 << 13,
  84         .num_mpt        = 1 << 17,
  85         .num_mtt        = 1 << 20,
  86 };
  87
  88 static int mlx4_dev_cap(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap)
  89 {
  90         int err;
  91         int i;
  92
  93         err = mlx4_QUERY_DEV_CAP(dev, dev_cap);
  94         if (err) {
  95                 mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting.\n");
  96                 return err;
  97         }
  98
  99         if (dev_cap->min_page_sz > PAGE_SIZE) {
 100                 mlx4_err(dev, "HCA minimum page size of %d bigger than "
 101                          "kernel PAGE_SIZE of %ld, aborting.\n",
 102                          dev_cap->min_page_sz, PAGE_SIZE);
 103                 return -ENODEV;
 104         }
 105         if (dev_cap->num_ports > MLX4_MAX_PORTS) {
 106                 mlx4_err(dev, "HCA has %d ports, but we only support %d, "
 107                          "aborting.\n",
 108                          dev_cap->num_ports, MLX4_MAX_PORTS);
 109                 return -ENODEV;
 110         }
 111
 112         if (dev_cap->uar_size > pci_resource_len(dev->pdev, 2)) {
 113                 mlx4_err(dev, "HCA reported UAR size of 0x%x bigger than "
 114                          "PCI resource 2 size of 0x%llx, aborting.\n",
 115                          dev_cap->uar_size,
 116                          (unsigned long long) pci_resource_len(dev->pdev, 2));
 117                 return -ENODEV;
 118         }
 119
 120         dev->caps.num_ports          = dev_cap->num_ports;
 121         for (i = 1; i <= dev->caps.num_ports; ++i) {
 122                 dev->caps.vl_cap[i]         = dev_cap->max_vl[i];
 123                 dev->caps.mtu_cap[i]        = dev_cap->max_mtu[i];
 124                 dev->caps.gid_table_len[i]  = dev_cap->max_gids[i];
 125                 dev->caps.pkey_table_len[i] = dev_cap->max_pkeys[i];
 126                 dev->caps.port_width_cap[i] = dev_cap->max_port_width[i];
 127         }
 128
 129         dev->caps.num_uars           = dev_cap->uar_size / PAGE_SIZE;
 130         dev->caps.local_ca_ack_delay = dev_cap->local_ca_ack_delay;
 131         dev->caps.bf_reg_size        = dev_cap->bf_reg_size;
 132         dev->caps.bf_regs_per_page   = dev_cap->bf_regs_per_page;
 133         dev->caps.max_sq_sg          = dev_cap->max_sq_sg;
 134         dev->caps.max_rq_sg          = dev_cap->max_rq_sg;
 135         dev->caps.max_wqes           = dev_cap->max_qp_sz;
 136         dev->caps.max_qp_init_rdma   = dev_cap->max_requester_per_qp;
 137         dev->caps.reserved_qps       = dev_cap->reserved_qps;
 138         dev->caps.max_srq_wqes       = dev_cap->max_srq_sz;
 139         dev->caps.max_srq_sge        = dev_cap->max_rq_sg - 1;
 140         dev->caps.reserved_srqs      = dev_cap->reserved_srqs;
 141         dev->caps.max_sq_desc_sz     = dev_cap->max_sq_desc_sz;
 142         dev->caps.max_rq_desc_sz     = dev_cap->max_rq_desc_sz;
 143         dev->caps.num_qp_per_mgm     = MLX4_QP_PER_MGM;
 144         /*
 145          * Subtract 1 from the limit because we need to allocate a
 146          * spare CQE so the HCA HW can tell the difference between an
 147          * empty CQ and a full CQ.
 148          */
 149         dev->caps.max_cqes           = dev_cap->max_cq_sz - 1;
 150         dev->caps.reserved_cqs       = dev_cap->reserved_cqs;
 151         dev->caps.reserved_eqs       = dev_cap->reserved_eqs;
 152         dev->caps.reserved_mtts      = DIV_ROUND_UP(dev_cap->reserved_mtts,
 153                                                     MLX4_MTT_ENTRY_PER_SEG);
 154         dev->caps.reserved_mrws      = dev_cap->reserved_mrws;
 155         dev->caps.reserved_uars      = dev_cap->reserved_uars;
 156         dev->caps.reserved_pds       = dev_cap->reserved_pds;
 157         dev->caps.mtt_entry_sz       = MLX4_MTT_ENTRY_PER_SEG * dev_cap->mtt_entry_sz;
 158         dev->caps.max_msg_sz         = dev_cap->max_msg_sz;
 159         dev->caps.page_size_cap      = ~(u32) (dev_cap->min_page_sz - 1);
 160         dev->caps.flags              = dev_cap->flags;
 161         dev->caps.stat_rate_support  = dev_cap->stat_rate_support;
 162
 163         return 0;
 164 }
 165
 166 static int mlx4_load_fw(struct mlx4_dev *dev)
 167 {
 168         struct mlx4_priv *priv = mlx4_priv(dev);
 169         int err;
 170
 171         priv->fw.fw_icm = mlx4_alloc_icm(dev, priv->fw.fw_pages,
 172                                          GFP_HIGHUSER | __GFP_NOWARN, 0);
 173         if (!priv->fw.fw_icm) {
 174                 mlx4_err(dev, "Couldn't allocate FW area, aborting.\n");
 175                 return -ENOMEM;
 176         }
 177
 178         err = mlx4_MAP_FA(dev, priv->fw.fw_icm);
 179         if (err) {
 180                 mlx4_err(dev, "MAP_FA command failed, aborting.\n");
 181                 goto err_free;
 182         }
 183
 184         err = mlx4_RUN_FW(dev);
 185         if (err) {
 186                 mlx4_err(dev, "RUN_FW command failed, aborting.\n");
 187                 goto err_unmap_fa;
 188         }
 189
 190         return 0;
 191
 192 err_unmap_fa:
 193         mlx4_UNMAP_FA(dev);
 194
 195 err_free:
 196         mlx4_free_icm(dev, priv->fw.fw_icm, 0);
 197         return err;
 198 }
 199
 200 static int mlx4_init_cmpt_table(struct mlx4_dev *dev, u64 cmpt_base,
 201                                 int cmpt_entry_sz)
 202 {
 203         struct mlx4_priv *priv = mlx4_priv(dev);
 204         int err;
 205
 206         err = mlx4_init_icm_table(dev, &priv->qp_table.cmpt_table,
 207                                   cmpt_base +
 208                                   ((u64) (MLX4_CMPT_TYPE_QP *
 209                                           cmpt_entry_sz) << MLX4_CMPT_SHIFT),
 210                                   cmpt_entry_sz, dev->caps.num_qps,
 211                                   dev->caps.reserved_qps, 0, 0);
 212         if (err)
 213                 goto err;
 214
 215         err = mlx4_init_icm_table(dev, &priv->srq_table.cmpt_table,
 216                                   cmpt_base +
 217                                   ((u64) (MLX4_CMPT_TYPE_SRQ *
 218                                           cmpt_entry_sz) << MLX4_CMPT_SHIFT),
 219                                   cmpt_entry_sz, dev->caps.num_srqs,
 220                                   dev->caps.reserved_srqs, 0, 0);
 221         if (err)
 222                 goto err_qp;
 223
 224         err = mlx4_init_icm_table(dev, &priv->cq_table.cmpt_table,
 225                                   cmpt_base +
 226                                   ((u64) (MLX4_CMPT_TYPE_CQ *
 227                                           cmpt_entry_sz) << MLX4_CMPT_SHIFT),
 228                                   cmpt_entry_sz, dev->caps.num_cqs,
 229                                   dev->caps.reserved_cqs, 0, 0);
 230         if (err)
 231                 goto err_srq;
 232
 233         err = mlx4_init_icm_table(dev, &priv->eq_table.cmpt_table,
 234                                   cmpt_base +
 235                                   ((u64) (MLX4_CMPT_TYPE_EQ *
 236                                           cmpt_entry_sz) << MLX4_CMPT_SHIFT),
 237                                   cmpt_entry_sz,
 238                                   roundup_pow_of_two(MLX4_NUM_EQ +
 239                                                      dev->caps.reserved_eqs),
 240                                   MLX4_NUM_EQ + dev->caps.reserved_eqs, 0, 0);
 241         if (err)
 242                 goto err_cq;
 243
 244         return 0;
 245
 246 err_cq:
 247         mlx4_cleanup_icm_table(dev, &priv->cq_table.cmpt_table);
 248
 249 err_srq:
 250         mlx4_cleanup_icm_table(dev, &priv->srq_table.cmpt_table);
 251
 252 err_qp:
 253         mlx4_cleanup_icm_table(dev, &priv->qp_table.cmpt_table);
 254
 255 err:
 256         return err;
 257 }
 258
 259 static int mlx4_init_icm(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap,
 260                          struct mlx4_init_hca_param *init_hca, u64 icm_size)
 261 {
 262         struct mlx4_priv *priv = mlx4_priv(dev);
 263         u64 aux_pages;
 264         int err;
 265
 266         err = mlx4_SET_ICM_SIZE(dev, icm_size, &aux_pages);
 267         if (err) {
 268                 mlx4_err(dev, "SET_ICM_SIZE command failed, aborting.\n");
 269                 return err;
 270         }
 271
 272         mlx4_dbg(dev, "%lld KB of HCA context requires %lld KB aux memory.\n",
 273                  (unsigned long long) icm_size >> 10,
 274                  (unsigned long long) aux_pages << 2);
 275
 276         priv->fw.aux_icm = mlx4_alloc_icm(dev, aux_pages,
 277                                           GFP_HIGHUSER | __GFP_NOWARN, 0);
 278         if (!priv->fw.aux_icm) {
 279                 mlx4_err(dev, "Couldn't allocate aux memory, aborting.\n");
 280                 return -ENOMEM;
 281         }
 282
 283         err = mlx4_MAP_ICM_AUX(dev, priv->fw.aux_icm);
 284         if (err) {
 285                 mlx4_err(dev, "MAP_ICM_AUX command failed, aborting.\n");
 286                 goto err_free_aux;
 287         }
 288
 289         err = mlx4_init_cmpt_table(dev, init_hca->cmpt_base, dev_cap->cmpt_entry_sz);
 290         if (err) {
 291                 mlx4_err(dev, "Failed to map cMPT context memory, aborting.\n");
 292                 goto err_unmap_aux;
 293         }
 294
 295         err = mlx4_map_eq_icm(dev, init_hca->eqc_base);
 296         if (err) {
 297                 mlx4_err(dev, "Failed to map EQ context memory, aborting.\n");
 298                 goto err_unmap_cmpt;
 299         }
 300
 301         /*
 302          * Reserved MTT entries must be aligned up to a cacheline
 303          * boundary, since the FW will write to them, while the driver
 304          * writes to all other MTT entries. (The variable
 305          * dev->caps.mtt_entry_sz below is really the MTT segment
 306          * size, not the raw entry size)
 307          */
 308         dev->caps.reserved_mtts =
 309                 ALIGN(dev->caps.reserved_mtts * dev->caps.mtt_entry_sz,
 310                       dma_get_cache_alignment()) / dev->caps.mtt_entry_sz;
 311
 312         err = mlx4_init_icm_table(dev, &priv->mr_table.mtt_table,
 313                                   init_hca->mtt_base,
 314                                   dev->caps.mtt_entry_sz,
 315                                   dev->caps.num_mtt_segs,
 316                                   dev->caps.reserved_mtts, 1, 0);
 317         if (err) {
 318                 mlx4_err(dev, "Failed to map MTT context memory, aborting.\n");
 319                 goto err_unmap_eq;
 320         }
 321
 322         err = mlx4_init_icm_table(dev, &priv->mr_table.dmpt_table,
 323                                   init_hca->dmpt_base,
 324                                   dev_cap->dmpt_entry_sz,
 325                                   dev->caps.num_mpts,
 326                                   dev->caps.reserved_mrws, 1, 1);
 327         if (err) {
 328                 mlx4_err(dev, "Failed to map dMPT context memory, aborting.\n");
 329                 goto err_unmap_mtt;
 330         }
 331
 332         err = mlx4_init_icm_table(dev, &priv->qp_table.qp_table,
 333                                   init_hca->qpc_base,
 334                                   dev_cap->qpc_entry_sz,
 335                                   dev->caps.num_qps,
 336                                   dev->caps.reserved_qps, 0, 0);
 337         if (err) {
 338                 mlx4_err(dev, "Failed to map QP context memory, aborting.\n");
 339                 goto err_unmap_dmpt;
 340         }
 341
 342         err = mlx4_init_icm_table(dev, &priv->qp_table.auxc_table,
 343                                   init_hca->auxc_base,
 344                                   dev_cap->aux_entry_sz,
 345                                   dev->caps.num_qps,
 346                                   dev->caps.reserved_qps, 0, 0);
 347         if (err) {
 348                 mlx4_err(dev, "Failed to map AUXC context memory, aborting.\n");
 349                 goto err_unmap_qp;
 350         }
 351
 352         err = mlx4_init_icm_table(dev, &priv->qp_table.altc_table,
 353                                   init_hca->altc_base,
 354                                   dev_cap->altc_entry_sz,
 355                                   dev->caps.num_qps,
 356                                   dev->caps.reserved_qps, 0, 0);
 357         if (err) {
 358                 mlx4_err(dev, "Failed to map ALTC context memory, aborting.\n");
 359                 goto err_unmap_auxc;
 360         }
 361
 362         err = mlx4_init_icm_table(dev, &priv->qp_table.rdmarc_table,
 363                                   init_hca->rdmarc_base,
 364                                   dev_cap->rdmarc_entry_sz << priv->qp_table.rdmarc_shift,
 365                                   dev->caps.num_qps,
 366                                   dev->caps.reserved_qps, 0, 0);
 367         if (err) {
 368                 mlx4_err(dev, "Failed to map RDMARC context memory, aborting\n");
 369                 goto err_unmap_altc;
 370         }
 371
 372         err = mlx4_init_icm_table(dev, &priv->cq_table.table,
 373                                   init_hca->cqc_base,
 374                                   dev_cap->cqc_entry_sz,
 375                                   dev->caps.num_cqs,
 376                                   dev->caps.reserved_cqs, 0, 0);
 377         if (err) {
 378                 mlx4_err(dev, "Failed to map CQ context memory, aborting.\n");
 379                 goto err_unmap_rdmarc;
 380         }
 381
 382         err = mlx4_init_icm_table(dev, &priv->srq_table.table,
 383                                   init_hca->srqc_base,
 384                                   dev_cap->srq_entry_sz,
 385                                   dev->caps.num_srqs,
 386                                   dev->caps.reserved_srqs, 0, 0);
 387         if (err) {
 388                 mlx4_err(dev, "Failed to map SRQ context memory, aborting.\n");
 389                 goto err_unmap_cq;
 390         }
 391
 392         /*
 393          * It's not strictly required, but for simplicity just map the
 394          * whole multicast group table now.  The table isn't very big
 395          * and it's a lot easier than trying to track ref counts.
 396          */
 397         err = mlx4_init_icm_table(dev, &priv->mcg_table.table,
 398                                   init_hca->mc_base, MLX4_MGM_ENTRY_SIZE,
 399                                   dev->caps.num_mgms + dev->caps.num_amgms,
 400                                   dev->caps.num_mgms + dev->caps.num_amgms,
 401                                   0, 0);
 402         if (err) {
 403                 mlx4_err(dev, "Failed to map MCG context memory, aborting.\n");
 404                 goto err_unmap_srq;
 405         }
 406
 407         return 0;
 408
 409 err_unmap_srq:
 410         mlx4_cleanup_icm_table(dev, &priv->srq_table.table);
 411
 412 err_unmap_cq:
 413         mlx4_cleanup_icm_table(dev, &priv->cq_table.table);
 414
 415 err_unmap_rdmarc:
 416         mlx4_cleanup_icm_table(dev, &priv->qp_table.rdmarc_table);
 417
 418 err_unmap_altc:
 419         mlx4_cleanup_icm_table(dev, &priv->qp_table.altc_table);
 420
 421 err_unmap_auxc:
 422         mlx4_cleanup_icm_table(dev, &priv->qp_table.auxc_table);
 423
 424 err_unmap_qp:
 425         mlx4_cleanup_icm_table(dev, &priv->qp_table.qp_table);
 426
 427 err_unmap_dmpt:
 428         mlx4_cleanup_icm_table(dev, &priv->mr_table.dmpt_table);
 429
 430 err_unmap_mtt:
 431         mlx4_cleanup_icm_table(dev, &priv->mr_table.mtt_table);
 432
 433 err_unmap_eq:
 434         mlx4_unmap_eq_icm(dev);
 435
 436 err_unmap_cmpt:
 437         mlx4_cleanup_icm_table(dev, &priv->eq_table.cmpt_table);
 438         mlx4_cleanup_icm_table(dev, &priv->cq_table.cmpt_table);
 439         mlx4_cleanup_icm_table(dev, &priv->srq_table.cmpt_table);
 440         mlx4_cleanup_icm_table(dev, &priv->qp_table.cmpt_table);
 441
 442 err_unmap_aux:
 443         mlx4_UNMAP_ICM_AUX(dev);
 444
 445 err_free_aux:
 446         mlx4_free_icm(dev, priv->fw.aux_icm, 0);
 447
 448         return err;
 449 }
 450
 451 static void mlx4_free_icms(struct mlx4_dev *dev)
 452 {
 453         struct mlx4_priv *priv = mlx4_priv(dev);
 454
 455         mlx4_cleanup_icm_table(dev, &priv->mcg_table.table);
 456         mlx4_cleanup_icm_table(dev, &priv->srq_table.table);
 457         mlx4_cleanup_icm_table(dev, &priv->cq_table.table);
 458         mlx4_cleanup_icm_table(dev, &priv->qp_table.rdmarc_table);
 459         mlx4_cleanup_icm_table(dev, &priv->qp_table.altc_table);
 460         mlx4_cleanup_icm_table(dev, &priv->qp_table.auxc_table);
 461         mlx4_cleanup_icm_table(dev, &priv->qp_table.qp_table);
 462         mlx4_cleanup_icm_table(dev, &priv->mr_table.dmpt_table);
 463         mlx4_cleanup_icm_table(dev, &priv->mr_table.mtt_table);
 464         mlx4_cleanup_icm_table(dev, &priv->eq_table.cmpt_table);
 465         mlx4_cleanup_icm_table(dev, &priv->cq_table.cmpt_table);
 466         mlx4_cleanup_icm_table(dev, &priv->srq_table.cmpt_table);
 467         mlx4_cleanup_icm_table(dev, &priv->qp_table.cmpt_table);
 468         mlx4_unmap_eq_icm(dev);
 469
 470         mlx4_UNMAP_ICM_AUX(dev);
 471         mlx4_free_icm(dev, priv->fw.aux_icm, 0);
 472 }
 473
 474 static void mlx4_close_hca(struct mlx4_dev *dev)
 475 {
 476         mlx4_CLOSE_HCA(dev, 0);
 477         mlx4_free_icms(dev);
 478         mlx4_UNMAP_FA(dev);
 479         mlx4_free_icm(dev, mlx4_priv(dev)->fw.fw_icm, 0);
 480 }
 481
 482 static int mlx4_init_hca(struct mlx4_dev *dev)
 483 {
 484         struct mlx4_priv          *priv = mlx4_priv(dev);
 485         struct mlx4_adapter        adapter;
 486         struct mlx4_dev_cap        dev_cap;
 487         struct mlx4_profile        profile;
 488         struct mlx4_init_hca_param init_hca;
 489         u64 icm_size;
 490         int err;
 491
 492         err = mlx4_QUERY_FW(dev);
 493         if (err) {
 494                 mlx4_err(dev, "QUERY_FW command failed, aborting.\n");
 495                 return err;
 496         }
 497
 498         err = mlx4_load_fw(dev);
 499         if (err) {
 500                 mlx4_err(dev, "Failed to start FW, aborting.\n");
 501                 return err;
 502         }
 503
 504         err = mlx4_dev_cap(dev, &dev_cap);
 505         if (err) {
 506                 mlx4_err(dev, "QUERY_DEV_CAP command failed, aborting.\n");
 507                 goto err_stop_fw;
 508         }
 509
 510         profile = default_profile;
 511
 512         icm_size = mlx4_make_profile(dev, &profile, &dev_cap, &init_hca);
 513         if ((long long) icm_size < 0) {
 514                 err = icm_size;
 515                 goto err_stop_fw;
 516         }
 517
 518         init_hca.log_uar_sz = ilog2(dev->caps.num_uars);
 519
 520         err = mlx4_init_icm(dev, &dev_cap, &init_hca, icm_size);
 521         if (err)
 522                 goto err_stop_fw;
 523
 524         err = mlx4_INIT_HCA(dev, &init_hca);
 525         if (err) {
 526                 mlx4_err(dev, "INIT_HCA command failed, aborting.\n");
 527                 goto err_free_icm;
 528         }
 529
 530         err = mlx4_QUERY_ADAPTER(dev, &adapter);
 531         if (err) {
 532                 mlx4_err(dev, "QUERY_ADAPTER command failed, aborting.\n");
 533                 goto err_close;
 534         }
 535
 536         priv->eq_table.inta_pin = adapter.inta_pin;
 537         memcpy(dev->board_id, adapter.board_id, sizeof dev->board_id);
 538
 539         return 0;
 540
 541 err_close:
 542         mlx4_close_hca(dev);
 543
 544 err_free_icm:
 545         mlx4_free_icms(dev);
 546
 547 err_stop_fw:
 548         mlx4_UNMAP_FA(dev);
 549         mlx4_free_icm(dev, priv->fw.fw_icm, 0);
 550
 551         return err;
 552 }
 553
 554 static int mlx4_setup_hca(struct mlx4_dev *dev)
 555 {
 556         struct mlx4_priv *priv = mlx4_priv(dev);
 557         int err;
 558
 559         err = mlx4_init_uar_table(dev);
 560         if (err) {
 561                 mlx4_err(dev, "Failed to initialize "
 562                          "user access region table, aborting.\n");
 563                 return err;
 564         }
 565
 566         err = mlx4_uar_alloc(dev, &priv->driver_uar);
 567         if (err) {
 568                 mlx4_err(dev, "Failed to allocate driver access region, "
 569                          "aborting.\n");
 570                 goto err_uar_table_free;
 571         }
 572
 573         priv->kar = ioremap(priv->driver_uar.pfn << PAGE_SHIFT, PAGE_SIZE);
 574         if (!priv->kar) {
 575                 mlx4_err(dev, "Couldn't map kernel access region, "
 576                          "aborting.\n");
 577                 err = -ENOMEM;
 578                 goto err_uar_free;
 579         }
 580
 581         err = mlx4_init_pd_table(dev);
 582         if (err) {
 583                 mlx4_err(dev, "Failed to initialize "
 584                          "protection domain table, aborting.\n");
 585                 goto err_kar_unmap;
 586         }
 587
 588         err = mlx4_init_mr_table(dev);
 589         if (err) {
 590                 mlx4_err(dev, "Failed to initialize "
 591                          "memory region table, aborting.\n");
 592                 goto err_pd_table_free;
 593         }
 594
 595         err = mlx4_init_eq_table(dev);
 596         if (err) {
 597                 mlx4_err(dev, "Failed to initialize "
 598                          "event queue table, aborting.\n");
 599                 goto err_mr_table_free;
 600         }
 601
 602         err = mlx4_cmd_use_events(dev);
 603         if (err) {
 604                 mlx4_err(dev, "Failed to switch to event-driven "
 605                          "firmware commands, aborting.\n");
 606                 goto err_eq_table_free;
 607         }
 608
 609         err = mlx4_NOP(dev);
 610         if (err) {
 611                 if (dev->flags & MLX4_FLAG_MSI_X) {
 612                         mlx4_warn(dev, "NOP command failed to generate MSI-X "
 613                                   "interrupt IRQ %d).\n",
 614                                   priv->eq_table.eq[MLX4_EQ_ASYNC].irq);
 615                         mlx4_warn(dev, "Trying again without MSI-X.\n");
 616                 } else {
 617                         mlx4_err(dev, "NOP command failed to generate interrupt "
 618                                  "(IRQ %d), aborting.\n",
 619                                  priv->eq_table.eq[MLX4_EQ_ASYNC].irq);
 620                         mlx4_err(dev, "BIOS or ACPI interrupt routing problem?\n");
 621                 }
 622
 623                 goto err_cmd_poll;
 624         }
 625
 626         mlx4_dbg(dev, "NOP command IRQ test passed\n");
 627
 628         err = mlx4_init_cq_table(dev);
 629         if (err) {
 630                 mlx4_err(dev, "Failed to initialize "
 631                          "completion queue table, aborting.\n");
 632                 goto err_cmd_poll;
 633         }
 634
 635         err = mlx4_init_srq_table(dev);
 636         if (err) {
 637                 mlx4_err(dev, "Failed to initialize "
 638                          "shared receive queue table, aborting.\n");
 639                 goto err_cq_table_free;
 640         }
 641
 642         err = mlx4_init_qp_table(dev);
 643         if (err) {
 644                 mlx4_err(dev, "Failed to initialize "
 645                          "queue pair table, aborting.\n");
 646                 goto err_srq_table_free;
 647         }
 648
 649         err = mlx4_init_mcg_table(dev);
 650         if (err) {
 651                 mlx4_err(dev, "Failed to initialize "
 652                          "multicast group table, aborting.\n");
 653                 goto err_qp_table_free;
 654         }
 655
 656         return 0;
 657
 658 err_qp_table_free:
 659         mlx4_cleanup_qp_table(dev);
 660
 661 err_srq_table_free:
 662         mlx4_cleanup_srq_table(dev);
 663
 664 err_cq_table_free:
 665         mlx4_cleanup_cq_table(dev);
 666
 667 err_cmd_poll:
 668         mlx4_cmd_use_polling(dev);
 669
 670 err_eq_table_free:
 671         mlx4_cleanup_eq_table(dev);
 672
 673 err_mr_table_free:
 674         mlx4_cleanup_mr_table(dev);
 675
 676 err_pd_table_free:
 677         mlx4_cleanup_pd_table(dev);
 678
 679 err_kar_unmap:
 680         iounmap(priv->kar);
 681
 682 err_uar_free:
 683         mlx4_uar_free(dev, &priv->driver_uar);
 684
 685 err_uar_table_free:
 686         mlx4_cleanup_uar_table(dev);
 687         return err;
 688 }
 689
 690 static void mlx4_enable_msi_x(struct mlx4_dev *dev)
 691 {
 692         struct mlx4_priv *priv = mlx4_priv(dev);
 693         struct msix_entry entries[MLX4_NUM_EQ];
 694         int err;
 695         int i;
 696
 697         if (msi_x) {
 698                 for (i = 0; i < MLX4_NUM_EQ; ++i)
 699                         entries[i].entry = i;
 700
 701                 err = pci_enable_msix(dev->pdev, entries, ARRAY_SIZE(entries));
 702                 if (err) {
 703                         if (err > 0)
 704                                 mlx4_info(dev, "Only %d MSI-X vectors available, "
 705                                           "not using MSI-X\n", err);
 706                         goto no_msi;
 707                 }
 708
 709                 for (i = 0; i < MLX4_NUM_EQ; ++i)
 710                         priv->eq_table.eq[i].irq = entries[i].vector;
 711
 712                 dev->flags |= MLX4_FLAG_MSI_X;
 713                 return;
 714         }
 715
 716 no_msi:
 717         for (i = 0; i < MLX4_NUM_EQ; ++i)
 718                 priv->eq_table.eq[i].irq = dev->pdev->irq;
 719 }
 720
 721 static int __mlx4_init_one(struct pci_dev *pdev, const struct pci_device_id *id)
 722 {
 723         struct mlx4_priv *priv;
 724         struct mlx4_dev *dev;
 725         int err;
 726
 727         printk(KERN_INFO PFX "Initializing %s\n",
 728                pci_name(pdev));
 729
 730         err = pci_enable_device(pdev);
 731         if (err) {
 732                 dev_err(&pdev->dev, "Cannot enable PCI device, "
 733                         "aborting.\n");
 734                 return err;
 735         }
 736
 737         /*
 738          * Check for BARs.  We expect 0: 1MB, 2: 8MB, 4: DDR (may not
 739          * be present)
 740          */
 741         if (!(pci_resource_flags(pdev, 0) & IORESOURCE_MEM) ||
 742             pci_resource_len(pdev, 0) != 1 << 20) {
 743                 dev_err(&pdev->dev, "Missing DCS, aborting.\n");
 744                 err = -ENODEV;
 745                 goto err_disable_pdev;
 746         }
 747         if (!(pci_resource_flags(pdev, 2) & IORESOURCE_MEM)) {
 748                 dev_err(&pdev->dev, "Missing UAR, aborting.\n");
 749                 err = -ENODEV;
 750                 goto err_disable_pdev;
 751         }
 752
 753         err = pci_request_region(pdev, 0, DRV_NAME);
 754         if (err) {
 755                 dev_err(&pdev->dev, "Cannot request control region, aborting.\n");
 756                 goto err_disable_pdev;
 757         }
 758
 759         err = pci_request_region(pdev, 2, DRV_NAME);
 760         if (err) {
 761                 dev_err(&pdev->dev, "Cannot request UAR region, aborting.\n");
 762                 goto err_release_bar0;
 763         }
 764
 765         pci_set_master(pdev);
 766
 767         err = pci_set_dma_mask(pdev, DMA_64BIT_MASK);
 768         if (err) {
 769                 dev_warn(&pdev->dev, "Warning: couldn't set 64-bit PCI DMA mask.\n");
 770                 err = pci_set_dma_mask(pdev, DMA_32BIT_MASK);
 771                 if (err) {
 772                         dev_err(&pdev->dev, "Can't set PCI DMA mask, aborting.\n");
 773                         goto err_release_bar2;
 774                 }
 775         }
 776         err = pci_set_consistent_dma_mask(pdev, DMA_64BIT_MASK);
 777         if (err) {
 778                 dev_warn(&pdev->dev, "Warning: couldn't set 64-bit "
 779                          "consistent PCI DMA mask.\n");
 780                 err = pci_set_consistent_dma_mask(pdev, DMA_32BIT_MASK);
 781                 if (err) {
 782                         dev_err(&pdev->dev, "Can't set consistent PCI DMA mask, "
 783                                 "aborting.\n");
 784                         goto err_release_bar2;
 785                 }
 786         }
 787
 788         priv = kzalloc(sizeof *priv, GFP_KERNEL);
 789         if (!priv) {
 790                 dev_err(&pdev->dev, "Device struct alloc failed, "
 791                         "aborting.\n");
 792                 err = -ENOMEM;
 793                 goto err_release_bar2;
 794         }
 795
 796         dev       = &priv->dev;
 797         dev->pdev = pdev;
 798         INIT_LIST_HEAD(&priv->ctx_list);
 799         spin_lock_init(&priv->ctx_lock);
 800
 801         /*
 802          * Now reset the HCA before we touch the PCI capabilities or
 803          * attempt a firmware command, since a boot ROM may have left
 804          * the HCA in an undefined state.
 805          */
 806         err = mlx4_reset(dev);
 807         if (err) {
 808                 mlx4_err(dev, "Failed to reset HCA, aborting.\n");
 809                 goto err_free_dev;
 810         }
 811
 812         if (mlx4_cmd_init(dev)) {
 813                 mlx4_err(dev, "Failed to init command interface, aborting.\n");
 814                 goto err_free_dev;
 815         }
 816
 817         err = mlx4_init_hca(dev);
 818         if (err)
 819                 goto err_cmd;
 820
 821         mlx4_enable_msi_x(dev);
 822
 823         err = mlx4_setup_hca(dev);
 824         if (err == -EBUSY && (dev->flags & MLX4_FLAG_MSI_X)) {
 825                 dev->flags &= ~MLX4_FLAG_MSI_X;
 826                 pci_disable_msix(pdev);
 827                 err = mlx4_setup_hca(dev);
 828         }
 829
 830         if (err)
 831                 goto err_close;
 832
 833         err = mlx4_register_device(dev);
 834         if (err)
 835                 goto err_cleanup;
 836
 837         pci_set_drvdata(pdev, dev);
 838
 839         return 0;
 840
 841 err_cleanup:
 842         mlx4_cleanup_mcg_table(dev);
 843         mlx4_cleanup_qp_table(dev);
 844         mlx4_cleanup_srq_table(dev);
 845         mlx4_cleanup_cq_table(dev);
 846         mlx4_cmd_use_polling(dev);
 847         mlx4_cleanup_eq_table(dev);
 848         mlx4_cleanup_mr_table(dev);
 849         mlx4_cleanup_pd_table(dev);
 850         mlx4_cleanup_uar_table(dev);
 851
 852 err_close:
 853         if (dev->flags & MLX4_FLAG_MSI_X)
 854                 pci_disable_msix(pdev);
 855
 856         mlx4_close_hca(dev);
 857
 858 err_cmd:
 859         mlx4_cmd_cleanup(dev);
 860
 861 err_free_dev:
 862         kfree(priv);
 863
 864 err_release_bar2:
 865         pci_release_region(pdev, 2);
 866
 867 err_release_bar0:
 868         pci_release_region(pdev, 0);
 869
 870 err_disable_pdev:
 871         pci_disable_device(pdev);
 872         pci_set_drvdata(pdev, NULL);
 873         return err;
 874 }
 875
 876 static int __devinit mlx4_init_one(struct pci_dev *pdev,
 877                                    const struct pci_device_id *id)
 878 {
 879         static int mlx4_version_printed;
 880
 881         if (!mlx4_version_printed) {
 882                 printk(KERN_INFO "%s", mlx4_version);
 883                 ++mlx4_version_printed;
 884         }
 885
 886         return __mlx4_init_one(pdev, id);
 887 }
 888
 889 static void mlx4_remove_one(struct pci_dev *pdev)
 890 {
 891         struct mlx4_dev  *dev  = pci_get_drvdata(pdev);
 892         struct mlx4_priv *priv = mlx4_priv(dev);
 893         int p;
 894
 895         if (dev) {
 896                 mlx4_unregister_device(dev);
 897
 898                 for (p = 1; p <= dev->caps.num_ports; ++p)
 899                         mlx4_CLOSE_PORT(dev, p);
 900
 901                 mlx4_cleanup_mcg_table(dev);
 902                 mlx4_cleanup_qp_table(dev);
 903                 mlx4_cleanup_srq_table(dev);
 904                 mlx4_cleanup_cq_table(dev);
 905                 mlx4_cmd_use_polling(dev);
 906                 mlx4_cleanup_eq_table(dev);
 907                 mlx4_cleanup_mr_table(dev);
 908                 mlx4_cleanup_pd_table(dev);
 909
 910                 iounmap(priv->kar);
 911                 mlx4_uar_free(dev, &priv->driver_uar);
 912                 mlx4_cleanup_uar_table(dev);
 913                 mlx4_close_hca(dev);
 914                 mlx4_cmd_cleanup(dev);
 915
 916                 if (dev->flags & MLX4_FLAG_MSI_X)
 917                         pci_disable_msix(pdev);
 918
 919                 kfree(priv);
 920                 pci_release_region(pdev, 2);
 921                 pci_release_region(pdev, 0);
 922                 pci_disable_device(pdev);
 923                 pci_set_drvdata(pdev, NULL);
 924         }
 925 }
 926
 927 int mlx4_restart_one(struct pci_dev *pdev)
 928 {
 929         mlx4_remove_one(pdev);
 930         return __mlx4_init_one(pdev, NULL);
 931 }
 932
 933 static struct pci_device_id mlx4_pci_table[] = {
 934         { PCI_VDEVICE(MELLANOX, 0x6340) }, /* MT25408 "Hermon" SDR */
 935         { PCI_VDEVICE(MELLANOX, 0x634a) }, /* MT25408 "Hermon" DDR */
 936         { PCI_VDEVICE(MELLANOX, 0x6354) }, /* MT25408 "Hermon" QDR */
 937         { PCI_VDEVICE(MELLANOX, 0x6732) }, /* MT25408 "Hermon" DDR PCIe gen2 */
 938         { PCI_VDEVICE(MELLANOX, 0x673c) }, /* MT25408 "Hermon" QDR PCIe gen2 */
 939         { 0, }
 940 };
 941
 942 MODULE_DEVICE_TABLE(pci, mlx4_pci_table);
 943
 944 static struct pci_driver mlx4_driver = {
 945         .name           = DRV_NAME,
 946         .id_table       = mlx4_pci_table,
 947         .probe          = mlx4_init_one,
 948         .remove         = __devexit_p(mlx4_remove_one)
 949 };
 950
 951 static int __init mlx4_init(void)
 952 {
 953         int ret;
 954
 955         ret = mlx4_catas_init();
 956         if (ret)
 957                 return ret;
 958
 959         ret = pci_register_driver(&mlx4_driver);
 960         return ret < 0 ? ret : 0;
 961 }
 962
 963 static void __exit mlx4_cleanup(void)
 964 {
 965         pci_unregister_driver(&mlx4_driver);
 966         mlx4_catas_cleanup();
 967 }
 968
 969 module_init(mlx4_init);
 970 module_exit(mlx4_cleanup);