drivers/infiniband/hw/vmw_pvrdma/pvrdma_main.c

   1 /*
   2  * Copyright (c) 2012-2016 VMware, Inc.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of EITHER the GNU General Public License
   6  * version 2 as published by the Free Software Foundation or the BSD
   7  * 2-Clause License. This program is distributed in the hope that it
   8  * will be useful, but WITHOUT ANY WARRANTY; WITHOUT EVEN THE IMPLIED
   9  * WARRANTY OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE.
  10  * See the GNU General Public License version 2 for more details at
  11  * http://www.gnu.org/licenses/old-licenses/gpl-2.0.en.html.
  12  *
  13  * You should have received a copy of the GNU General Public License
  14  * along with this program available in the file COPYING in the main
  15  * directory of this source tree.
  16  *
  17  * The BSD 2-Clause License
  18  *
  19  *     Redistribution and use in source and binary forms, with or
  20  *     without modification, are permitted provided that the following
  21  *     conditions are met:
  22  *
  23  *      - Redistributions of source code must retain the above
  24  *        copyright notice, this list of conditions and the following
  25  *        disclaimer.
  26  *
  27  *      - Redistributions in binary form must reproduce the above
  28  *        copyright notice, this list of conditions and the following
  29  *        disclaimer in the documentation and/or other materials
  30  *        provided with the distribution.
  31  *
  32  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  33  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  34  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
  35  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
  36  * COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT,
  37  * INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
  38  * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
  39  * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  40  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
  41  * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
  42  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
  43  * OF THE POSSIBILITY OF SUCH DAMAGE.
  44  */
  45
  46 #include <linux/errno.h>
  47 #include <linux/inetdevice.h>
  48 #include <linux/init.h>
  49 #include <linux/module.h>
  50 #include <linux/slab.h>
  51 #include <rdma/ib_addr.h>
  52 #include <rdma/ib_smi.h>
  53 #include <rdma/ib_user_verbs.h>
  54 #include <net/addrconf.h>
  55
  56 #include "pvrdma.h"
  57
  58 #define DRV_NAME        "vmw_pvrdma"
  59 #define DRV_VERSION     "1.0.1.0-k"
  60
  61 static DEFINE_MUTEX(pvrdma_device_list_lock);
  62 static LIST_HEAD(pvrdma_device_list);
  63 static struct workqueue_struct *event_wq;
  64
  65 static int pvrdma_add_gid(const struct ib_gid_attr *attr, void **context);
  66 static int pvrdma_del_gid(const struct ib_gid_attr *attr, void **context);
  67
  68 static ssize_t hca_type_show(struct device *device,
  69                              struct device_attribute *attr, char *buf)
  70 {
  71         return sprintf(buf, "VMW_PVRDMA-%s\n", DRV_VERSION);
  72 }
  73 static DEVICE_ATTR_RO(hca_type);
  74
  75 static ssize_t hw_rev_show(struct device *device,
  76                            struct device_attribute *attr, char *buf)
  77 {
  78         return sprintf(buf, "%d\n", PVRDMA_REV_ID);
  79 }
  80 static DEVICE_ATTR_RO(hw_rev);
  81
  82 static ssize_t board_id_show(struct device *device,
  83                              struct device_attribute *attr, char *buf)
  84 {
  85         return sprintf(buf, "%d\n", PVRDMA_BOARD_ID);
  86 }
  87 static DEVICE_ATTR_RO(board_id);
  88
  89 static struct attribute *pvrdma_class_attributes[] = {
  90         &dev_attr_hw_rev.attr,
  91         &dev_attr_hca_type.attr,
  92         &dev_attr_board_id.attr,
  93         NULL,
  94 };
  95
  96 static const struct attribute_group pvrdma_attr_group = {
  97         .attrs = pvrdma_class_attributes,
  98 };
  99
 100 static void pvrdma_get_fw_ver_str(struct ib_device *device, char *str)
 101 {
 102         struct pvrdma_dev *dev =
 103                 container_of(device, struct pvrdma_dev, ib_dev);
 104         snprintf(str, IB_FW_VERSION_NAME_MAX, "%d.%d.%d\n",
 105                  (int) (dev->dsr->caps.fw_ver >> 32),
 106                  (int) (dev->dsr->caps.fw_ver >> 16) & 0xffff,
 107                  (int) dev->dsr->caps.fw_ver & 0xffff);
 108 }
 109
 110 static int pvrdma_init_device(struct pvrdma_dev *dev)
 111 {
 112         /*  Initialize some device related stuff */
 113         spin_lock_init(&dev->cmd_lock);
 114         sema_init(&dev->cmd_sema, 1);
 115         atomic_set(&dev->num_qps, 0);
 116         atomic_set(&dev->num_srqs, 0);
 117         atomic_set(&dev->num_cqs, 0);
 118         atomic_set(&dev->num_pds, 0);
 119         atomic_set(&dev->num_ahs, 0);
 120
 121         return 0;
 122 }
 123
 124 static int pvrdma_port_immutable(struct ib_device *ibdev, u8 port_num,
 125                                  struct ib_port_immutable *immutable)
 126 {
 127         struct pvrdma_dev *dev = to_vdev(ibdev);
 128         struct ib_port_attr attr;
 129         int err;
 130
 131         if (dev->dsr->caps.gid_types == PVRDMA_GID_TYPE_FLAG_ROCE_V1)
 132                 immutable->core_cap_flags |= RDMA_CORE_PORT_IBA_ROCE;
 133         else if (dev->dsr->caps.gid_types == PVRDMA_GID_TYPE_FLAG_ROCE_V2)
 134                 immutable->core_cap_flags |= RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP;
 135
 136         err = ib_query_port(ibdev, port_num, &attr);
 137         if (err)
 138                 return err;
 139
 140         immutable->pkey_tbl_len = attr.pkey_tbl_len;
 141         immutable->gid_tbl_len = attr.gid_tbl_len;
 142         immutable->max_mad_size = IB_MGMT_MAD_SIZE;
 143         return 0;
 144 }
 145
 146 static struct net_device *pvrdma_get_netdev(struct ib_device *ibdev,
 147                                             u8 port_num)
 148 {
 149         struct net_device *netdev;
 150         struct pvrdma_dev *dev = to_vdev(ibdev);
 151
 152         if (port_num != 1)
 153                 return NULL;
 154
 155         rcu_read_lock();
 156         netdev = dev->netdev;
 157         if (netdev)
 158                 dev_hold(netdev);
 159         rcu_read_unlock();
 160
 161         return netdev;
 162 }
 163
 164 static const struct ib_device_ops pvrdma_dev_ops = {
 165         .add_gid = pvrdma_add_gid,
 166         .alloc_mr = pvrdma_alloc_mr,
 167         .alloc_pd = pvrdma_alloc_pd,
 168         .alloc_ucontext = pvrdma_alloc_ucontext,
 169         .create_ah = pvrdma_create_ah,
 170         .create_cq = pvrdma_create_cq,
 171         .create_qp = pvrdma_create_qp,
 172         .dealloc_pd = pvrdma_dealloc_pd,
 173         .dealloc_ucontext = pvrdma_dealloc_ucontext,
 174         .del_gid = pvrdma_del_gid,
 175         .dereg_mr = pvrdma_dereg_mr,
 176         .destroy_ah = pvrdma_destroy_ah,
 177         .destroy_cq = pvrdma_destroy_cq,
 178         .destroy_qp = pvrdma_destroy_qp,
 179         .get_dev_fw_str = pvrdma_get_fw_ver_str,
 180         .get_dma_mr = pvrdma_get_dma_mr,
 181         .get_link_layer = pvrdma_port_link_layer,
 182         .get_netdev = pvrdma_get_netdev,
 183         .get_port_immutable = pvrdma_port_immutable,
 184         .map_mr_sg = pvrdma_map_mr_sg,
 185         .mmap = pvrdma_mmap,
 186         .modify_port = pvrdma_modify_port,
 187         .modify_qp = pvrdma_modify_qp,
 188         .poll_cq = pvrdma_poll_cq,
 189         .post_recv = pvrdma_post_recv,
 190         .post_send = pvrdma_post_send,
 191         .query_device = pvrdma_query_device,
 192         .query_gid = pvrdma_query_gid,
 193         .query_pkey = pvrdma_query_pkey,
 194         .query_port = pvrdma_query_port,
 195         .query_qp = pvrdma_query_qp,
 196         .reg_user_mr = pvrdma_reg_user_mr,
 197         .req_notify_cq = pvrdma_req_notify_cq,
 198 };
 199
 200 static const struct ib_device_ops pvrdma_dev_srq_ops = {
 201         .create_srq = pvrdma_create_srq,
 202         .destroy_srq = pvrdma_destroy_srq,
 203         .modify_srq = pvrdma_modify_srq,
 204         .query_srq = pvrdma_query_srq,
 205 };
 206
 207 static int pvrdma_register_device(struct pvrdma_dev *dev)
 208 {
 209         int ret = -1;
 210
 211         dev->ib_dev.node_guid = dev->dsr->caps.node_guid;
 212         dev->sys_image_guid = dev->dsr->caps.sys_image_guid;
 213         dev->flags = 0;
 214         dev->ib_dev.owner = THIS_MODULE;
 215         dev->ib_dev.num_comp_vectors = 1;
 216         dev->ib_dev.dev.parent = &dev->pdev->dev;
 217         dev->ib_dev.uverbs_abi_ver = PVRDMA_UVERBS_ABI_VERSION;
 218         dev->ib_dev.uverbs_cmd_mask =
 219                 (1ull << IB_USER_VERBS_CMD_GET_CONTEXT)         |
 220                 (1ull << IB_USER_VERBS_CMD_QUERY_DEVICE)        |
 221                 (1ull << IB_USER_VERBS_CMD_QUERY_PORT)          |
 222                 (1ull << IB_USER_VERBS_CMD_ALLOC_PD)            |
 223                 (1ull << IB_USER_VERBS_CMD_DEALLOC_PD)          |
 224                 (1ull << IB_USER_VERBS_CMD_REG_MR)              |
 225                 (1ull << IB_USER_VERBS_CMD_DEREG_MR)            |
 226                 (1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL) |
 227                 (1ull << IB_USER_VERBS_CMD_CREATE_CQ)           |
 228                 (1ull << IB_USER_VERBS_CMD_POLL_CQ)             |
 229                 (1ull << IB_USER_VERBS_CMD_REQ_NOTIFY_CQ)       |
 230                 (1ull << IB_USER_VERBS_CMD_DESTROY_CQ)          |
 231                 (1ull << IB_USER_VERBS_CMD_CREATE_QP)           |
 232                 (1ull << IB_USER_VERBS_CMD_MODIFY_QP)           |
 233                 (1ull << IB_USER_VERBS_CMD_QUERY_QP)            |
 234                 (1ull << IB_USER_VERBS_CMD_DESTROY_QP)          |
 235                 (1ull << IB_USER_VERBS_CMD_POST_SEND)           |
 236                 (1ull << IB_USER_VERBS_CMD_POST_RECV)           |
 237                 (1ull << IB_USER_VERBS_CMD_CREATE_AH)           |
 238                 (1ull << IB_USER_VERBS_CMD_DESTROY_AH);
 239
 240         dev->ib_dev.node_type = RDMA_NODE_IB_CA;
 241         dev->ib_dev.phys_port_cnt = dev->dsr->caps.phys_port_cnt;
 242
 243         ib_set_device_ops(&dev->ib_dev, &pvrdma_dev_ops);
 244
 245         mutex_init(&dev->port_mutex);
 246         spin_lock_init(&dev->desc_lock);
 247
 248         dev->cq_tbl = kcalloc(dev->dsr->caps.max_cq, sizeof(struct pvrdma_cq *),
 249                               GFP_KERNEL);
 250         if (!dev->cq_tbl)
 251                 return ret;
 252         spin_lock_init(&dev->cq_tbl_lock);
 253
 254         dev->qp_tbl = kcalloc(dev->dsr->caps.max_qp, sizeof(struct pvrdma_qp *),
 255                               GFP_KERNEL);
 256         if (!dev->qp_tbl)
 257                 goto err_cq_free;
 258         spin_lock_init(&dev->qp_tbl_lock);
 259
 260         /* Check if SRQ is supported by backend */
 261         if (dev->dsr->caps.max_srq) {
 262                 dev->ib_dev.uverbs_cmd_mask |=
 263                         (1ull << IB_USER_VERBS_CMD_CREATE_SRQ)  |
 264                         (1ull << IB_USER_VERBS_CMD_MODIFY_SRQ)  |
 265                         (1ull << IB_USER_VERBS_CMD_QUERY_SRQ)   |
 266                         (1ull << IB_USER_VERBS_CMD_DESTROY_SRQ) |
 267                         (1ull << IB_USER_VERBS_CMD_POST_SRQ_RECV);
 268
 269                 ib_set_device_ops(&dev->ib_dev, &pvrdma_dev_srq_ops);
 270
 271                 dev->srq_tbl = kcalloc(dev->dsr->caps.max_srq,
 272                                        sizeof(struct pvrdma_srq *),
 273                                        GFP_KERNEL);
 274                 if (!dev->srq_tbl)
 275                         goto err_qp_free;
 276         }
 277         dev->ib_dev.driver_id = RDMA_DRIVER_VMW_PVRDMA;
 278         spin_lock_init(&dev->srq_tbl_lock);
 279         rdma_set_device_sysfs_group(&dev->ib_dev, &pvrdma_attr_group);
 280
 281         ret = ib_register_device(&dev->ib_dev, "vmw_pvrdma%d", NULL);
 282         if (ret)
 283                 goto err_srq_free;
 284
 285         dev->ib_active = true;
 286
 287         return 0;
 288
 289 err_srq_free:
 290         kfree(dev->srq_tbl);
 291 err_qp_free:
 292         kfree(dev->qp_tbl);
 293 err_cq_free:
 294         kfree(dev->cq_tbl);
 295
 296         return ret;
 297 }
 298
 299 static irqreturn_t pvrdma_intr0_handler(int irq, void *dev_id)
 300 {
 301         u32 icr = PVRDMA_INTR_CAUSE_RESPONSE;
 302         struct pvrdma_dev *dev = dev_id;
 303
 304         dev_dbg(&dev->pdev->dev, "interrupt 0 (response) handler\n");
 305
 306         if (!dev->pdev->msix_enabled) {
 307                 /* Legacy intr */
 308                 icr = pvrdma_read_reg(dev, PVRDMA_REG_ICR);
 309                 if (icr == 0)
 310                         return IRQ_NONE;
 311         }
 312
 313         if (icr == PVRDMA_INTR_CAUSE_RESPONSE)
 314                 complete(&dev->cmd_done);
 315
 316         return IRQ_HANDLED;
 317 }
 318
 319 static void pvrdma_qp_event(struct pvrdma_dev *dev, u32 qpn, int type)
 320 {
 321         struct pvrdma_qp *qp;
 322         unsigned long flags;
 323
 324         spin_lock_irqsave(&dev->qp_tbl_lock, flags);
 325         qp = dev->qp_tbl[qpn % dev->dsr->caps.max_qp];
 326         if (qp)
 327                 refcount_inc(&qp->refcnt);
 328         spin_unlock_irqrestore(&dev->qp_tbl_lock, flags);
 329
 330         if (qp && qp->ibqp.event_handler) {
 331                 struct ib_qp *ibqp = &qp->ibqp;
 332                 struct ib_event e;
 333
 334                 e.device = ibqp->device;
 335                 e.element.qp = ibqp;
 336                 e.event = type; /* 1:1 mapping for now. */
 337                 ibqp->event_handler(&e, ibqp->qp_context);
 338         }
 339         if (qp) {
 340                 if (refcount_dec_and_test(&qp->refcnt))
 341                         complete(&qp->free);
 342         }
 343 }
 344
 345 static void pvrdma_cq_event(struct pvrdma_dev *dev, u32 cqn, int type)
 346 {
 347         struct pvrdma_cq *cq;
 348         unsigned long flags;
 349
 350         spin_lock_irqsave(&dev->cq_tbl_lock, flags);
 351         cq = dev->cq_tbl[cqn % dev->dsr->caps.max_cq];
 352         if (cq)
 353                 refcount_inc(&cq->refcnt);
 354         spin_unlock_irqrestore(&dev->cq_tbl_lock, flags);
 355
 356         if (cq && cq->ibcq.event_handler) {
 357                 struct ib_cq *ibcq = &cq->ibcq;
 358                 struct ib_event e;
 359
 360                 e.device = ibcq->device;
 361                 e.element.cq = ibcq;
 362                 e.event = type; /* 1:1 mapping for now. */
 363                 ibcq->event_handler(&e, ibcq->cq_context);
 364         }
 365         if (cq) {
 366                 if (refcount_dec_and_test(&cq->refcnt))
 367                         complete(&cq->free);
 368         }
 369 }
 370
 371 static void pvrdma_srq_event(struct pvrdma_dev *dev, u32 srqn, int type)
 372 {
 373         struct pvrdma_srq *srq;
 374         unsigned long flags;
 375
 376         spin_lock_irqsave(&dev->srq_tbl_lock, flags);
 377         if (dev->srq_tbl)
 378                 srq = dev->srq_tbl[srqn % dev->dsr->caps.max_srq];
 379         else
 380                 srq = NULL;
 381         if (srq)
 382                 refcount_inc(&srq->refcnt);
 383         spin_unlock_irqrestore(&dev->srq_tbl_lock, flags);
 384
 385         if (srq && srq->ibsrq.event_handler) {
 386                 struct ib_srq *ibsrq = &srq->ibsrq;
 387                 struct ib_event e;
 388
 389                 e.device = ibsrq->device;
 390                 e.element.srq = ibsrq;
 391                 e.event = type; /* 1:1 mapping for now. */
 392                 ibsrq->event_handler(&e, ibsrq->srq_context);
 393         }
 394         if (srq) {
 395                 if (refcount_dec_and_test(&srq->refcnt))
 396                         complete(&srq->free);
 397         }
 398 }
 399
 400 static void pvrdma_dispatch_event(struct pvrdma_dev *dev, int port,
 401                                   enum ib_event_type event)
 402 {
 403         struct ib_event ib_event;
 404
 405         memset(&ib_event, 0, sizeof(ib_event));
 406         ib_event.device = &dev->ib_dev;
 407         ib_event.element.port_num = port;
 408         ib_event.event = event;
 409         ib_dispatch_event(&ib_event);
 410 }
 411
 412 static void pvrdma_dev_event(struct pvrdma_dev *dev, u8 port, int type)
 413 {
 414         if (port < 1 || port > dev->dsr->caps.phys_port_cnt) {
 415                 dev_warn(&dev->pdev->dev, "event on port %d\n", port);
 416                 return;
 417         }
 418
 419         pvrdma_dispatch_event(dev, port, type);
 420 }
 421
 422 static inline struct pvrdma_eqe *get_eqe(struct pvrdma_dev *dev, unsigned int i)
 423 {
 424         return (struct pvrdma_eqe *)pvrdma_page_dir_get_ptr(
 425                                         &dev->async_pdir,
 426                                         PAGE_SIZE +
 427                                         sizeof(struct pvrdma_eqe) * i);
 428 }
 429
 430 static irqreturn_t pvrdma_intr1_handler(int irq, void *dev_id)
 431 {
 432         struct pvrdma_dev *dev = dev_id;
 433         struct pvrdma_ring *ring = &dev->async_ring_state->rx;
 434         int ring_slots = (dev->dsr->async_ring_pages.num_pages - 1) *
 435                          PAGE_SIZE / sizeof(struct pvrdma_eqe);
 436         unsigned int head;
 437
 438         dev_dbg(&dev->pdev->dev, "interrupt 1 (async event) handler\n");
 439
 440         /*
 441          * Don't process events until the IB device is registered. Otherwise
 442          * we'll try to ib_dispatch_event() on an invalid device.
 443          */
 444         if (!dev->ib_active)
 445                 return IRQ_HANDLED;
 446
 447         while (pvrdma_idx_ring_has_data(ring, ring_slots, &head) > 0) {
 448                 struct pvrdma_eqe *eqe;
 449
 450                 eqe = get_eqe(dev, head);
 451
 452                 switch (eqe->type) {
 453                 case PVRDMA_EVENT_QP_FATAL:
 454                 case PVRDMA_EVENT_QP_REQ_ERR:
 455                 case PVRDMA_EVENT_QP_ACCESS_ERR:
 456                 case PVRDMA_EVENT_COMM_EST:
 457                 case PVRDMA_EVENT_SQ_DRAINED:
 458                 case PVRDMA_EVENT_PATH_MIG:
 459                 case PVRDMA_EVENT_PATH_MIG_ERR:
 460                 case PVRDMA_EVENT_QP_LAST_WQE_REACHED:
 461                         pvrdma_qp_event(dev, eqe->info, eqe->type);
 462                         break;
 463
 464                 case PVRDMA_EVENT_CQ_ERR:
 465                         pvrdma_cq_event(dev, eqe->info, eqe->type);
 466                         break;
 467
 468                 case PVRDMA_EVENT_SRQ_ERR:
 469                 case PVRDMA_EVENT_SRQ_LIMIT_REACHED:
 470                         pvrdma_srq_event(dev, eqe->info, eqe->type);
 471                         break;
 472
 473                 case PVRDMA_EVENT_PORT_ACTIVE:
 474                 case PVRDMA_EVENT_PORT_ERR:
 475                 case PVRDMA_EVENT_LID_CHANGE:
 476                 case PVRDMA_EVENT_PKEY_CHANGE:
 477                 case PVRDMA_EVENT_SM_CHANGE:
 478                 case PVRDMA_EVENT_CLIENT_REREGISTER:
 479                 case PVRDMA_EVENT_GID_CHANGE:
 480                         pvrdma_dev_event(dev, eqe->info, eqe->type);
 481                         break;
 482
 483                 case PVRDMA_EVENT_DEVICE_FATAL:
 484                         pvrdma_dev_event(dev, 1, eqe->type);
 485                         break;
 486
 487                 default:
 488                         break;
 489                 }
 490
 491                 pvrdma_idx_ring_inc(&ring->cons_head, ring_slots);
 492         }
 493
 494         return IRQ_HANDLED;
 495 }
 496
 497 static inline struct pvrdma_cqne *get_cqne(struct pvrdma_dev *dev,
 498                                            unsigned int i)
 499 {
 500         return (struct pvrdma_cqne *)pvrdma_page_dir_get_ptr(
 501                                         &dev->cq_pdir,
 502                                         PAGE_SIZE +
 503                                         sizeof(struct pvrdma_cqne) * i);
 504 }
 505
 506 static irqreturn_t pvrdma_intrx_handler(int irq, void *dev_id)
 507 {
 508         struct pvrdma_dev *dev = dev_id;
 509         struct pvrdma_ring *ring = &dev->cq_ring_state->rx;
 510         int ring_slots = (dev->dsr->cq_ring_pages.num_pages - 1) * PAGE_SIZE /
 511                          sizeof(struct pvrdma_cqne);
 512         unsigned int head;
 513         unsigned long flags;
 514
 515         dev_dbg(&dev->pdev->dev, "interrupt x (completion) handler\n");
 516
 517         while (pvrdma_idx_ring_has_data(ring, ring_slots, &head) > 0) {
 518                 struct pvrdma_cqne *cqne;
 519                 struct pvrdma_cq *cq;
 520
 521                 cqne = get_cqne(dev, head);
 522                 spin_lock_irqsave(&dev->cq_tbl_lock, flags);
 523                 cq = dev->cq_tbl[cqne->info % dev->dsr->caps.max_cq];
 524                 if (cq)
 525                         refcount_inc(&cq->refcnt);
 526                 spin_unlock_irqrestore(&dev->cq_tbl_lock, flags);
 527
 528                 if (cq && cq->ibcq.comp_handler)
 529                         cq->ibcq.comp_handler(&cq->ibcq, cq->ibcq.cq_context);
 530                 if (cq) {
 531                         if (refcount_dec_and_test(&cq->refcnt))
 532                                 complete(&cq->free);
 533                 }
 534                 pvrdma_idx_ring_inc(&ring->cons_head, ring_slots);
 535         }
 536
 537         return IRQ_HANDLED;
 538 }
 539
 540 static void pvrdma_free_irq(struct pvrdma_dev *dev)
 541 {
 542         int i;
 543
 544         dev_dbg(&dev->pdev->dev, "freeing interrupts\n");
 545         for (i = 0; i < dev->nr_vectors; i++)
 546                 free_irq(pci_irq_vector(dev->pdev, i), dev);
 547 }
 548
 549 static void pvrdma_enable_intrs(struct pvrdma_dev *dev)
 550 {
 551         dev_dbg(&dev->pdev->dev, "enable interrupts\n");
 552         pvrdma_write_reg(dev, PVRDMA_REG_IMR, 0);
 553 }
 554
 555 static void pvrdma_disable_intrs(struct pvrdma_dev *dev)
 556 {
 557         dev_dbg(&dev->pdev->dev, "disable interrupts\n");
 558         pvrdma_write_reg(dev, PVRDMA_REG_IMR, ~0);
 559 }
 560
 561 static int pvrdma_alloc_intrs(struct pvrdma_dev *dev)
 562 {
 563         struct pci_dev *pdev = dev->pdev;
 564         int ret = 0, i;
 565
 566         ret = pci_alloc_irq_vectors(pdev, 1, PVRDMA_MAX_INTERRUPTS,
 567                         PCI_IRQ_MSIX);
 568         if (ret < 0) {
 569                 ret = pci_alloc_irq_vectors(pdev, 1, 1,
 570                                 PCI_IRQ_MSI | PCI_IRQ_LEGACY);
 571                 if (ret < 0)
 572                         return ret;
 573         }
 574         dev->nr_vectors = ret;
 575
 576         ret = request_irq(pci_irq_vector(dev->pdev, 0), pvrdma_intr0_handler,
 577                         pdev->msix_enabled ? 0 : IRQF_SHARED, DRV_NAME, dev);
 578         if (ret) {
 579                 dev_err(&dev->pdev->dev,
 580                         "failed to request interrupt 0\n");
 581                 goto out_free_vectors;
 582         }
 583
 584         for (i = 1; i < dev->nr_vectors; i++) {
 585                 ret = request_irq(pci_irq_vector(dev->pdev, i),
 586                                 i == 1 ? pvrdma_intr1_handler :
 587                                          pvrdma_intrx_handler,
 588                                 0, DRV_NAME, dev);
 589                 if (ret) {
 590                         dev_err(&dev->pdev->dev,
 591                                 "failed to request interrupt %d\n", i);
 592                         goto free_irqs;
 593                 }
 594         }
 595
 596         return 0;
 597
 598 free_irqs:
 599         while (--i >= 0)
 600                 free_irq(pci_irq_vector(dev->pdev, i), dev);
 601 out_free_vectors:
 602         pci_free_irq_vectors(pdev);
 603         return ret;
 604 }
 605
 606 static void pvrdma_free_slots(struct pvrdma_dev *dev)
 607 {
 608         struct pci_dev *pdev = dev->pdev;
 609
 610         if (dev->resp_slot)
 611                 dma_free_coherent(&pdev->dev, PAGE_SIZE, dev->resp_slot,
 612                                   dev->dsr->resp_slot_dma);
 613         if (dev->cmd_slot)
 614                 dma_free_coherent(&pdev->dev, PAGE_SIZE, dev->cmd_slot,
 615                                   dev->dsr->cmd_slot_dma);
 616 }
 617
 618 static int pvrdma_add_gid_at_index(struct pvrdma_dev *dev,
 619                                    const union ib_gid *gid,
 620                                    u8 gid_type,
 621                                    int index)
 622 {
 623         int ret;
 624         union pvrdma_cmd_req req;
 625         struct pvrdma_cmd_create_bind *cmd_bind = &req.create_bind;
 626
 627         if (!dev->sgid_tbl) {
 628                 dev_warn(&dev->pdev->dev, "sgid table not initialized\n");
 629                 return -EINVAL;
 630         }
 631
 632         memset(cmd_bind, 0, sizeof(*cmd_bind));
 633         cmd_bind->hdr.cmd = PVRDMA_CMD_CREATE_BIND;
 634         memcpy(cmd_bind->new_gid, gid->raw, 16);
 635         cmd_bind->mtu = ib_mtu_enum_to_int(IB_MTU_1024);
 636         cmd_bind->vlan = 0xfff;
 637         cmd_bind->index = index;
 638         cmd_bind->gid_type = gid_type;
 639
 640         ret = pvrdma_cmd_post(dev, &req, NULL, 0);
 641         if (ret < 0) {
 642                 dev_warn(&dev->pdev->dev,
 643                          "could not create binding, error: %d\n", ret);
 644                 return -EFAULT;
 645         }
 646         memcpy(&dev->sgid_tbl[index], gid, sizeof(*gid));
 647         return 0;
 648 }
 649
 650 static int pvrdma_add_gid(const struct ib_gid_attr *attr, void **context)
 651 {
 652         struct pvrdma_dev *dev = to_vdev(attr->device);
 653
 654         return pvrdma_add_gid_at_index(dev, &attr->gid,
 655                                        ib_gid_type_to_pvrdma(attr->gid_type),
 656                                        attr->index);
 657 }
 658
 659 static int pvrdma_del_gid_at_index(struct pvrdma_dev *dev, int index)
 660 {
 661         int ret;
 662         union pvrdma_cmd_req req;
 663         struct pvrdma_cmd_destroy_bind *cmd_dest = &req.destroy_bind;
 664
 665         /* Update sgid table. */
 666         if (!dev->sgid_tbl) {
 667                 dev_warn(&dev->pdev->dev, "sgid table not initialized\n");
 668                 return -EINVAL;
 669         }
 670
 671         memset(cmd_dest, 0, sizeof(*cmd_dest));
 672         cmd_dest->hdr.cmd = PVRDMA_CMD_DESTROY_BIND;
 673         memcpy(cmd_dest->dest_gid, &dev->sgid_tbl[index], 16);
 674         cmd_dest->index = index;
 675
 676         ret = pvrdma_cmd_post(dev, &req, NULL, 0);
 677         if (ret < 0) {
 678                 dev_warn(&dev->pdev->dev,
 679                          "could not destroy binding, error: %d\n", ret);
 680                 return ret;
 681         }
 682         memset(&dev->sgid_tbl[index], 0, 16);
 683         return 0;
 684 }
 685
 686 static int pvrdma_del_gid(const struct ib_gid_attr *attr, void **context)
 687 {
 688         struct pvrdma_dev *dev = to_vdev(attr->device);
 689
 690         dev_dbg(&dev->pdev->dev, "removing gid at index %u from %s",
 691                 attr->index, dev->netdev->name);
 692
 693         return pvrdma_del_gid_at_index(dev, attr->index);
 694 }
 695
 696 static void pvrdma_netdevice_event_handle(struct pvrdma_dev *dev,
 697                                           struct net_device *ndev,
 698                                           unsigned long event)
 699 {
 700         struct pci_dev *pdev_net;
 701         unsigned int slot;
 702
 703         switch (event) {
 704         case NETDEV_REBOOT:
 705         case NETDEV_DOWN:
 706                 pvrdma_dispatch_event(dev, 1, IB_EVENT_PORT_ERR);
 707                 break;
 708         case NETDEV_UP:
 709                 pvrdma_write_reg(dev, PVRDMA_REG_CTL,
 710                                  PVRDMA_DEVICE_CTL_UNQUIESCE);
 711
 712                 mb();
 713
 714                 if (pvrdma_read_reg(dev, PVRDMA_REG_ERR))
 715                         dev_err(&dev->pdev->dev,
 716                                 "failed to activate device during link up\n");
 717                 else
 718                         pvrdma_dispatch_event(dev, 1, IB_EVENT_PORT_ACTIVE);
 719                 break;
 720         case NETDEV_UNREGISTER:
 721                 dev_put(dev->netdev);
 722                 dev->netdev = NULL;
 723                 break;
 724         case NETDEV_REGISTER:
 725                 /* vmxnet3 will have same bus, slot. But func will be 0 */
 726                 slot = PCI_SLOT(dev->pdev->devfn);
 727                 pdev_net = pci_get_slot(dev->pdev->bus,
 728                                         PCI_DEVFN(slot, 0));
 729                 if ((dev->netdev == NULL) &&
 730                     (pci_get_drvdata(pdev_net) == ndev)) {
 731                         /* this is our netdev */
 732                         dev->netdev = ndev;
 733                         dev_hold(ndev);
 734                 }
 735                 pci_dev_put(pdev_net);
 736                 break;
 737
 738         default:
 739                 dev_dbg(&dev->pdev->dev, "ignore netdevice event %ld on %s\n",
 740                         event, dev_name(&dev->ib_dev.dev));
 741                 break;
 742         }
 743 }
 744
 745 static void pvrdma_netdevice_event_work(struct work_struct *work)
 746 {
 747         struct pvrdma_netdevice_work *netdev_work;
 748         struct pvrdma_dev *dev;
 749
 750         netdev_work = container_of(work, struct pvrdma_netdevice_work, work);
 751
 752         mutex_lock(&pvrdma_device_list_lock);
 753         list_for_each_entry(dev, &pvrdma_device_list, device_link) {
 754                 if ((netdev_work->event == NETDEV_REGISTER) ||
 755                     (dev->netdev == netdev_work->event_netdev)) {
 756                         pvrdma_netdevice_event_handle(dev,
 757                                                       netdev_work->event_netdev,
 758                                                       netdev_work->event);
 759                         break;
 760                 }
 761         }
 762         mutex_unlock(&pvrdma_device_list_lock);
 763
 764         kfree(netdev_work);
 765 }
 766
 767 static int pvrdma_netdevice_event(struct notifier_block *this,
 768                                   unsigned long event, void *ptr)
 769 {
 770         struct net_device *event_netdev = netdev_notifier_info_to_dev(ptr);
 771         struct pvrdma_netdevice_work *netdev_work;
 772
 773         netdev_work = kmalloc(sizeof(*netdev_work), GFP_ATOMIC);
 774         if (!netdev_work)
 775                 return NOTIFY_BAD;
 776
 777         INIT_WORK(&netdev_work->work, pvrdma_netdevice_event_work);
 778         netdev_work->event_netdev = event_netdev;
 779         netdev_work->event = event;
 780         queue_work(event_wq, &netdev_work->work);
 781
 782         return NOTIFY_DONE;
 783 }
 784
 785 static int pvrdma_pci_probe(struct pci_dev *pdev,
 786                             const struct pci_device_id *id)
 787 {
 788         struct pci_dev *pdev_net;
 789         struct pvrdma_dev *dev;
 790         int ret;
 791         unsigned long start;
 792         unsigned long len;
 793         dma_addr_t slot_dma = 0;
 794
 795         dev_dbg(&pdev->dev, "initializing driver %s\n", pci_name(pdev));
 796
 797         /* Allocate zero-out device */
 798         dev = (struct pvrdma_dev *)ib_alloc_device(sizeof(*dev));
 799         if (!dev) {
 800                 dev_err(&pdev->dev, "failed to allocate IB device\n");
 801                 return -ENOMEM;
 802         }
 803
 804         mutex_lock(&pvrdma_device_list_lock);
 805         list_add(&dev->device_link, &pvrdma_device_list);
 806         mutex_unlock(&pvrdma_device_list_lock);
 807
 808         ret = pvrdma_init_device(dev);
 809         if (ret)
 810                 goto err_free_device;
 811
 812         dev->pdev = pdev;
 813         pci_set_drvdata(pdev, dev);
 814
 815         ret = pci_enable_device(pdev);
 816         if (ret) {
 817                 dev_err(&pdev->dev, "cannot enable PCI device\n");
 818                 goto err_free_device;
 819         }
 820
 821         dev_dbg(&pdev->dev, "PCI resource flags BAR0 %#lx\n",
 822                 pci_resource_flags(pdev, 0));
 823         dev_dbg(&pdev->dev, "PCI resource len %#llx\n",
 824                 (unsigned long long)pci_resource_len(pdev, 0));
 825         dev_dbg(&pdev->dev, "PCI resource start %#llx\n",
 826                 (unsigned long long)pci_resource_start(pdev, 0));
 827         dev_dbg(&pdev->dev, "PCI resource flags BAR1 %#lx\n",
 828                 pci_resource_flags(pdev, 1));
 829         dev_dbg(&pdev->dev, "PCI resource len %#llx\n",
 830                 (unsigned long long)pci_resource_len(pdev, 1));
 831         dev_dbg(&pdev->dev, "PCI resource start %#llx\n",
 832                 (unsigned long long)pci_resource_start(pdev, 1));
 833
 834         if (!(pci_resource_flags(pdev, 0) & IORESOURCE_MEM) ||
 835             !(pci_resource_flags(pdev, 1) & IORESOURCE_MEM)) {
 836                 dev_err(&pdev->dev, "PCI BAR region not MMIO\n");
 837                 ret = -ENOMEM;
 838                 goto err_free_device;
 839         }
 840
 841         ret = pci_request_regions(pdev, DRV_NAME);
 842         if (ret) {
 843                 dev_err(&pdev->dev, "cannot request PCI resources\n");
 844                 goto err_disable_pdev;
 845         }
 846
 847         /* Enable 64-Bit DMA */
 848         if (pci_set_dma_mask(pdev, DMA_BIT_MASK(64)) == 0) {
 849                 ret = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
 850                 if (ret != 0) {
 851                         dev_err(&pdev->dev,
 852                                 "pci_set_consistent_dma_mask failed\n");
 853                         goto err_free_resource;
 854                 }
 855         } else {
 856                 ret = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
 857                 if (ret != 0) {
 858                         dev_err(&pdev->dev,
 859                                 "pci_set_dma_mask failed\n");
 860                         goto err_free_resource;
 861                 }
 862         }
 863
 864         pci_set_master(pdev);
 865
 866         /* Map register space */
 867         start = pci_resource_start(dev->pdev, PVRDMA_PCI_RESOURCE_REG);
 868         len = pci_resource_len(dev->pdev, PVRDMA_PCI_RESOURCE_REG);
 869         dev->regs = ioremap(start, len);
 870         if (!dev->regs) {
 871                 dev_err(&pdev->dev, "register mapping failed\n");
 872                 ret = -ENOMEM;
 873                 goto err_free_resource;
 874         }
 875
 876         /* Setup per-device UAR. */
 877         dev->driver_uar.index = 0;
 878         dev->driver_uar.pfn =
 879                 pci_resource_start(dev->pdev, PVRDMA_PCI_RESOURCE_UAR) >>
 880                 PAGE_SHIFT;
 881         dev->driver_uar.map =
 882                 ioremap(dev->driver_uar.pfn << PAGE_SHIFT, PAGE_SIZE);
 883         if (!dev->driver_uar.map) {
 884                 dev_err(&pdev->dev, "failed to remap UAR pages\n");
 885                 ret = -ENOMEM;
 886                 goto err_unmap_regs;
 887         }
 888
 889         dev->dsr_version = pvrdma_read_reg(dev, PVRDMA_REG_VERSION);
 890         dev_info(&pdev->dev, "device version %d, driver version %d\n",
 891                  dev->dsr_version, PVRDMA_VERSION);
 892
 893         dev->dsr = dma_alloc_coherent(&pdev->dev, sizeof(*dev->dsr),
 894                                       &dev->dsrbase, GFP_KERNEL);
 895         if (!dev->dsr) {
 896                 dev_err(&pdev->dev, "failed to allocate shared region\n");
 897                 ret = -ENOMEM;
 898                 goto err_uar_unmap;
 899         }
 900
 901         /* Setup the shared region */
 902         dev->dsr->driver_version = PVRDMA_VERSION;
 903         dev->dsr->gos_info.gos_bits = sizeof(void *) == 4 ?
 904                 PVRDMA_GOS_BITS_32 :
 905                 PVRDMA_GOS_BITS_64;
 906         dev->dsr->gos_info.gos_type = PVRDMA_GOS_TYPE_LINUX;
 907         dev->dsr->gos_info.gos_ver = 1;
 908         dev->dsr->uar_pfn = dev->driver_uar.pfn;
 909
 910         /* Command slot. */
 911         dev->cmd_slot = dma_alloc_coherent(&pdev->dev, PAGE_SIZE,
 912                                            &slot_dma, GFP_KERNEL);
 913         if (!dev->cmd_slot) {
 914                 ret = -ENOMEM;
 915                 goto err_free_dsr;
 916         }
 917
 918         dev->dsr->cmd_slot_dma = (u64)slot_dma;
 919
 920         /* Response slot. */
 921         dev->resp_slot = dma_alloc_coherent(&pdev->dev, PAGE_SIZE,
 922                                             &slot_dma, GFP_KERNEL);
 923         if (!dev->resp_slot) {
 924                 ret = -ENOMEM;
 925                 goto err_free_slots;
 926         }
 927
 928         dev->dsr->resp_slot_dma = (u64)slot_dma;
 929
 930         /* Async event ring */
 931         dev->dsr->async_ring_pages.num_pages = PVRDMA_NUM_RING_PAGES;
 932         ret = pvrdma_page_dir_init(dev, &dev->async_pdir,
 933                                    dev->dsr->async_ring_pages.num_pages, true);
 934         if (ret)
 935                 goto err_free_slots;
 936         dev->async_ring_state = dev->async_pdir.pages[0];
 937         dev->dsr->async_ring_pages.pdir_dma = dev->async_pdir.dir_dma;
 938
 939         /* CQ notification ring */
 940         dev->dsr->cq_ring_pages.num_pages = PVRDMA_NUM_RING_PAGES;
 941         ret = pvrdma_page_dir_init(dev, &dev->cq_pdir,
 942                                    dev->dsr->cq_ring_pages.num_pages, true);
 943         if (ret)
 944                 goto err_free_async_ring;
 945         dev->cq_ring_state = dev->cq_pdir.pages[0];
 946         dev->dsr->cq_ring_pages.pdir_dma = dev->cq_pdir.dir_dma;
 947
 948         /*
 949          * Write the PA of the shared region to the device. The writes must be
 950          * ordered such that the high bits are written last. When the writes
 951          * complete, the device will have filled out the capabilities.
 952          */
 953
 954         pvrdma_write_reg(dev, PVRDMA_REG_DSRLOW, (u32)dev->dsrbase);
 955         pvrdma_write_reg(dev, PVRDMA_REG_DSRHIGH,
 956                          (u32)((u64)(dev->dsrbase) >> 32));
 957
 958         /* Make sure the write is complete before reading status. */
 959         mb();
 960
 961         /* The driver supports RoCE V1 and V2. */
 962         if (!PVRDMA_SUPPORTED(dev)) {
 963                 dev_err(&pdev->dev, "driver needs RoCE v1 or v2 support\n");
 964                 ret = -EFAULT;
 965                 goto err_free_cq_ring;
 966         }
 967
 968         /* Paired vmxnet3 will have same bus, slot. But func will be 0 */
 969         pdev_net = pci_get_slot(pdev->bus, PCI_DEVFN(PCI_SLOT(pdev->devfn), 0));
 970         if (!pdev_net) {
 971                 dev_err(&pdev->dev, "failed to find paired net device\n");
 972                 ret = -ENODEV;
 973                 goto err_free_cq_ring;
 974         }
 975
 976         if (pdev_net->vendor != PCI_VENDOR_ID_VMWARE ||
 977             pdev_net->device != PCI_DEVICE_ID_VMWARE_VMXNET3) {
 978                 dev_err(&pdev->dev, "failed to find paired vmxnet3 device\n");
 979                 pci_dev_put(pdev_net);
 980                 ret = -ENODEV;
 981                 goto err_free_cq_ring;
 982         }
 983
 984         dev->netdev = pci_get_drvdata(pdev_net);
 985         pci_dev_put(pdev_net);
 986         if (!dev->netdev) {
 987                 dev_err(&pdev->dev, "failed to get vmxnet3 device\n");
 988                 ret = -ENODEV;
 989                 goto err_free_cq_ring;
 990         }
 991         dev_hold(dev->netdev);
 992
 993         dev_info(&pdev->dev, "paired device to %s\n", dev->netdev->name);
 994
 995         /* Interrupt setup */
 996         ret = pvrdma_alloc_intrs(dev);
 997         if (ret) {
 998                 dev_err(&pdev->dev, "failed to allocate interrupts\n");
 999                 ret = -ENOMEM;
1000                 goto err_free_cq_ring;
1001         }
1002
1003         /* Allocate UAR table. */
1004         ret = pvrdma_uar_table_init(dev);
1005         if (ret) {
1006                 dev_err(&pdev->dev, "failed to allocate UAR table\n");
1007                 ret = -ENOMEM;
1008                 goto err_free_intrs;
1009         }
1010
1011         /* Allocate GID table */
1012         dev->sgid_tbl = kcalloc(dev->dsr->caps.gid_tbl_len,
1013                                 sizeof(union ib_gid), GFP_KERNEL);
1014         if (!dev->sgid_tbl) {
1015                 ret = -ENOMEM;
1016                 goto err_free_uar_table;
1017         }
1018         dev_dbg(&pdev->dev, "gid table len %d\n", dev->dsr->caps.gid_tbl_len);
1019
1020         pvrdma_enable_intrs(dev);
1021
1022         /* Activate pvrdma device */
1023         pvrdma_write_reg(dev, PVRDMA_REG_CTL, PVRDMA_DEVICE_CTL_ACTIVATE);
1024
1025         /* Make sure the write is complete before reading status. */
1026         mb();
1027
1028         /* Check if device was successfully activated */
1029         ret = pvrdma_read_reg(dev, PVRDMA_REG_ERR);
1030         if (ret != 0) {
1031                 dev_err(&pdev->dev, "failed to activate device\n");
1032                 ret = -EFAULT;
1033                 goto err_disable_intr;
1034         }
1035
1036         /* Register IB device */
1037         ret = pvrdma_register_device(dev);
1038         if (ret) {
1039                 dev_err(&pdev->dev, "failed to register IB device\n");
1040                 goto err_disable_intr;
1041         }
1042
1043         dev->nb_netdev.notifier_call = pvrdma_netdevice_event;
1044         ret = register_netdevice_notifier(&dev->nb_netdev);
1045         if (ret) {
1046                 dev_err(&pdev->dev, "failed to register netdevice events\n");
1047                 goto err_unreg_ibdev;
1048         }
1049
1050         dev_info(&pdev->dev, "attached to device\n");
1051         return 0;
1052
1053 err_unreg_ibdev:
1054         ib_unregister_device(&dev->ib_dev);
1055 err_disable_intr:
1056         pvrdma_disable_intrs(dev);
1057         kfree(dev->sgid_tbl);
1058 err_free_uar_table:
1059         pvrdma_uar_table_cleanup(dev);
1060 err_free_intrs:
1061         pvrdma_free_irq(dev);
1062         pci_free_irq_vectors(pdev);
1063 err_free_cq_ring:
1064         if (dev->netdev) {
1065                 dev_put(dev->netdev);
1066                 dev->netdev = NULL;
1067         }
1068         pvrdma_page_dir_cleanup(dev, &dev->cq_pdir);
1069 err_free_async_ring:
1070         pvrdma_page_dir_cleanup(dev, &dev->async_pdir);
1071 err_free_slots:
1072         pvrdma_free_slots(dev);
1073 err_free_dsr:
1074         dma_free_coherent(&pdev->dev, sizeof(*dev->dsr), dev->dsr,
1075                           dev->dsrbase);
1076 err_uar_unmap:
1077         iounmap(dev->driver_uar.map);
1078 err_unmap_regs:
1079         iounmap(dev->regs);
1080 err_free_resource:
1081         pci_release_regions(pdev);
1082 err_disable_pdev:
1083         pci_disable_device(pdev);
1084         pci_set_drvdata(pdev, NULL);
1085 err_free_device:
1086         mutex_lock(&pvrdma_device_list_lock);
1087         list_del(&dev->device_link);
1088         mutex_unlock(&pvrdma_device_list_lock);
1089         ib_dealloc_device(&dev->ib_dev);
1090         return ret;
1091 }
1092
1093 static void pvrdma_pci_remove(struct pci_dev *pdev)
1094 {
1095         struct pvrdma_dev *dev = pci_get_drvdata(pdev);
1096
1097         if (!dev)
1098                 return;
1099
1100         dev_info(&pdev->dev, "detaching from device\n");
1101
1102         unregister_netdevice_notifier(&dev->nb_netdev);
1103         dev->nb_netdev.notifier_call = NULL;
1104
1105         flush_workqueue(event_wq);
1106
1107         if (dev->netdev) {
1108                 dev_put(dev->netdev);
1109                 dev->netdev = NULL;
1110         }
1111
1112         /* Unregister ib device */
1113         ib_unregister_device(&dev->ib_dev);
1114
1115         mutex_lock(&pvrdma_device_list_lock);
1116         list_del(&dev->device_link);
1117         mutex_unlock(&pvrdma_device_list_lock);
1118
1119         pvrdma_disable_intrs(dev);
1120         pvrdma_free_irq(dev);
1121         pci_free_irq_vectors(pdev);
1122
1123         /* Deactivate pvrdma device */
1124         pvrdma_write_reg(dev, PVRDMA_REG_CTL, PVRDMA_DEVICE_CTL_RESET);
1125         pvrdma_page_dir_cleanup(dev, &dev->cq_pdir);
1126         pvrdma_page_dir_cleanup(dev, &dev->async_pdir);
1127         pvrdma_free_slots(dev);
1128
1129         iounmap(dev->regs);
1130         kfree(dev->sgid_tbl);
1131         kfree(dev->cq_tbl);
1132         kfree(dev->srq_tbl);
1133         kfree(dev->qp_tbl);
1134         pvrdma_uar_table_cleanup(dev);
1135         iounmap(dev->driver_uar.map);
1136
1137         ib_dealloc_device(&dev->ib_dev);
1138
1139         /* Free pci resources */
1140         pci_release_regions(pdev);
1141         pci_disable_device(pdev);
1142         pci_set_drvdata(pdev, NULL);
1143 }
1144
1145 static const struct pci_device_id pvrdma_pci_table[] = {
1146         { PCI_DEVICE(PCI_VENDOR_ID_VMWARE, PCI_DEVICE_ID_VMWARE_PVRDMA), },
1147         { 0 },
1148 };
1149
1150 MODULE_DEVICE_TABLE(pci, pvrdma_pci_table);
1151
1152 static struct pci_driver pvrdma_driver = {
1153         .name           = DRV_NAME,
1154         .id_table       = pvrdma_pci_table,
1155         .probe          = pvrdma_pci_probe,
1156         .remove         = pvrdma_pci_remove,
1157 };
1158
1159 static int __init pvrdma_init(void)
1160 {
1161         int err;
1162
1163         event_wq = alloc_ordered_workqueue("pvrdma_event_wq", WQ_MEM_RECLAIM);
1164         if (!event_wq)
1165                 return -ENOMEM;
1166
1167         err = pci_register_driver(&pvrdma_driver);
1168         if (err)
1169                 destroy_workqueue(event_wq);
1170
1171         return err;
1172 }
1173
1174 static void __exit pvrdma_cleanup(void)
1175 {
1176         pci_unregister_driver(&pvrdma_driver);
1177
1178         destroy_workqueue(event_wq);
1179 }
1180
1181 module_init(pvrdma_init);
1182 module_exit(pvrdma_cleanup);
1183
1184 MODULE_AUTHOR("VMware, Inc");
1185 MODULE_DESCRIPTION("VMware Paravirtual RDMA driver");
1186 MODULE_LICENSE("Dual BSD/GPL");