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 const struct ib_device_ops pvrdma_dev_ops = {
 147         .owner = THIS_MODULE,
 148         .driver_id = RDMA_DRIVER_VMW_PVRDMA,
 149         .uverbs_abi_ver = PVRDMA_UVERBS_ABI_VERSION,
 150
 151         .add_gid = pvrdma_add_gid,
 152         .alloc_mr = pvrdma_alloc_mr,
 153         .alloc_pd = pvrdma_alloc_pd,
 154         .alloc_ucontext = pvrdma_alloc_ucontext,
 155         .create_ah = pvrdma_create_ah,
 156         .create_cq = pvrdma_create_cq,
 157         .create_qp = pvrdma_create_qp,
 158         .dealloc_pd = pvrdma_dealloc_pd,
 159         .dealloc_ucontext = pvrdma_dealloc_ucontext,
 160         .del_gid = pvrdma_del_gid,
 161         .dereg_mr = pvrdma_dereg_mr,
 162         .destroy_ah = pvrdma_destroy_ah,
 163         .destroy_cq = pvrdma_destroy_cq,
 164         .destroy_qp = pvrdma_destroy_qp,
 165         .get_dev_fw_str = pvrdma_get_fw_ver_str,
 166         .get_dma_mr = pvrdma_get_dma_mr,
 167         .get_link_layer = pvrdma_port_link_layer,
 168         .get_port_immutable = pvrdma_port_immutable,
 169         .map_mr_sg = pvrdma_map_mr_sg,
 170         .mmap = pvrdma_mmap,
 171         .modify_port = pvrdma_modify_port,
 172         .modify_qp = pvrdma_modify_qp,
 173         .poll_cq = pvrdma_poll_cq,
 174         .post_recv = pvrdma_post_recv,
 175         .post_send = pvrdma_post_send,
 176         .query_device = pvrdma_query_device,
 177         .query_gid = pvrdma_query_gid,
 178         .query_pkey = pvrdma_query_pkey,
 179         .query_port = pvrdma_query_port,
 180         .query_qp = pvrdma_query_qp,
 181         .reg_user_mr = pvrdma_reg_user_mr,
 182         .req_notify_cq = pvrdma_req_notify_cq,
 183
 184         INIT_RDMA_OBJ_SIZE(ib_ah, pvrdma_ah, ibah),
 185         INIT_RDMA_OBJ_SIZE(ib_cq, pvrdma_cq, ibcq),
 186         INIT_RDMA_OBJ_SIZE(ib_pd, pvrdma_pd, ibpd),
 187         INIT_RDMA_OBJ_SIZE(ib_ucontext, pvrdma_ucontext, ibucontext),
 188 };
 189
 190 static const struct ib_device_ops pvrdma_dev_srq_ops = {
 191         .create_srq = pvrdma_create_srq,
 192         .destroy_srq = pvrdma_destroy_srq,
 193         .modify_srq = pvrdma_modify_srq,
 194         .query_srq = pvrdma_query_srq,
 195
 196         INIT_RDMA_OBJ_SIZE(ib_srq, pvrdma_srq, ibsrq),
 197 };
 198
 199 static int pvrdma_register_device(struct pvrdma_dev *dev)
 200 {
 201         int ret = -1;
 202
 203         dev->ib_dev.node_guid = dev->dsr->caps.node_guid;
 204         dev->sys_image_guid = dev->dsr->caps.sys_image_guid;
 205         dev->flags = 0;
 206         dev->ib_dev.num_comp_vectors = 1;
 207         dev->ib_dev.dev.parent = &dev->pdev->dev;
 208         dev->ib_dev.uverbs_cmd_mask =
 209                 (1ull << IB_USER_VERBS_CMD_GET_CONTEXT)         |
 210                 (1ull << IB_USER_VERBS_CMD_QUERY_DEVICE)        |
 211                 (1ull << IB_USER_VERBS_CMD_QUERY_PORT)          |
 212                 (1ull << IB_USER_VERBS_CMD_ALLOC_PD)            |
 213                 (1ull << IB_USER_VERBS_CMD_DEALLOC_PD)          |
 214                 (1ull << IB_USER_VERBS_CMD_REG_MR)              |
 215                 (1ull << IB_USER_VERBS_CMD_DEREG_MR)            |
 216                 (1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL) |
 217                 (1ull << IB_USER_VERBS_CMD_CREATE_CQ)           |
 218                 (1ull << IB_USER_VERBS_CMD_POLL_CQ)             |
 219                 (1ull << IB_USER_VERBS_CMD_REQ_NOTIFY_CQ)       |
 220                 (1ull << IB_USER_VERBS_CMD_DESTROY_CQ)          |
 221                 (1ull << IB_USER_VERBS_CMD_CREATE_QP)           |
 222                 (1ull << IB_USER_VERBS_CMD_MODIFY_QP)           |
 223                 (1ull << IB_USER_VERBS_CMD_QUERY_QP)            |
 224                 (1ull << IB_USER_VERBS_CMD_DESTROY_QP)          |
 225                 (1ull << IB_USER_VERBS_CMD_POST_SEND)           |
 226                 (1ull << IB_USER_VERBS_CMD_POST_RECV)           |
 227                 (1ull << IB_USER_VERBS_CMD_CREATE_AH)           |
 228                 (1ull << IB_USER_VERBS_CMD_DESTROY_AH);
 229
 230         dev->ib_dev.node_type = RDMA_NODE_IB_CA;
 231         dev->ib_dev.phys_port_cnt = dev->dsr->caps.phys_port_cnt;
 232
 233         ib_set_device_ops(&dev->ib_dev, &pvrdma_dev_ops);
 234
 235         mutex_init(&dev->port_mutex);
 236         spin_lock_init(&dev->desc_lock);
 237
 238         dev->cq_tbl = kcalloc(dev->dsr->caps.max_cq, sizeof(struct pvrdma_cq *),
 239                               GFP_KERNEL);
 240         if (!dev->cq_tbl)
 241                 return ret;
 242         spin_lock_init(&dev->cq_tbl_lock);
 243
 244         dev->qp_tbl = kcalloc(dev->dsr->caps.max_qp, sizeof(struct pvrdma_qp *),
 245                               GFP_KERNEL);
 246         if (!dev->qp_tbl)
 247                 goto err_cq_free;
 248         spin_lock_init(&dev->qp_tbl_lock);
 249
 250         /* Check if SRQ is supported by backend */
 251         if (dev->dsr->caps.max_srq) {
 252                 dev->ib_dev.uverbs_cmd_mask |=
 253                         (1ull << IB_USER_VERBS_CMD_CREATE_SRQ)  |
 254                         (1ull << IB_USER_VERBS_CMD_MODIFY_SRQ)  |
 255                         (1ull << IB_USER_VERBS_CMD_QUERY_SRQ)   |
 256                         (1ull << IB_USER_VERBS_CMD_DESTROY_SRQ) |
 257                         (1ull << IB_USER_VERBS_CMD_POST_SRQ_RECV);
 258
 259                 ib_set_device_ops(&dev->ib_dev, &pvrdma_dev_srq_ops);
 260
 261                 dev->srq_tbl = kcalloc(dev->dsr->caps.max_srq,
 262                                        sizeof(struct pvrdma_srq *),
 263                                        GFP_KERNEL);
 264                 if (!dev->srq_tbl)
 265                         goto err_qp_free;
 266         }
 267         ret = ib_device_set_netdev(&dev->ib_dev, dev->netdev, 1);
 268         if (ret)
 269                 return ret;
 270         spin_lock_init(&dev->srq_tbl_lock);
 271         rdma_set_device_sysfs_group(&dev->ib_dev, &pvrdma_attr_group);
 272
 273         ret = ib_register_device(&dev->ib_dev, "vmw_pvrdma%d");
 274         if (ret)
 275                 goto err_srq_free;
 276
 277         dev->ib_active = true;
 278
 279         return 0;
 280
 281 err_srq_free:
 282         kfree(dev->srq_tbl);
 283 err_qp_free:
 284         kfree(dev->qp_tbl);
 285 err_cq_free:
 286         kfree(dev->cq_tbl);
 287
 288         return ret;
 289 }
 290
 291 static irqreturn_t pvrdma_intr0_handler(int irq, void *dev_id)
 292 {
 293         u32 icr = PVRDMA_INTR_CAUSE_RESPONSE;
 294         struct pvrdma_dev *dev = dev_id;
 295
 296         dev_dbg(&dev->pdev->dev, "interrupt 0 (response) handler\n");
 297
 298         if (!dev->pdev->msix_enabled) {
 299                 /* Legacy intr */
 300                 icr = pvrdma_read_reg(dev, PVRDMA_REG_ICR);
 301                 if (icr == 0)
 302                         return IRQ_NONE;
 303         }
 304
 305         if (icr == PVRDMA_INTR_CAUSE_RESPONSE)
 306                 complete(&dev->cmd_done);
 307
 308         return IRQ_HANDLED;
 309 }
 310
 311 static void pvrdma_qp_event(struct pvrdma_dev *dev, u32 qpn, int type)
 312 {
 313         struct pvrdma_qp *qp;
 314         unsigned long flags;
 315
 316         spin_lock_irqsave(&dev->qp_tbl_lock, flags);
 317         qp = dev->qp_tbl[qpn % dev->dsr->caps.max_qp];
 318         if (qp)
 319                 refcount_inc(&qp->refcnt);
 320         spin_unlock_irqrestore(&dev->qp_tbl_lock, flags);
 321
 322         if (qp && qp->ibqp.event_handler) {
 323                 struct ib_qp *ibqp = &qp->ibqp;
 324                 struct ib_event e;
 325
 326                 e.device = ibqp->device;
 327                 e.element.qp = ibqp;
 328                 e.event = type; /* 1:1 mapping for now. */
 329                 ibqp->event_handler(&e, ibqp->qp_context);
 330         }
 331         if (qp) {
 332                 if (refcount_dec_and_test(&qp->refcnt))
 333                         complete(&qp->free);
 334         }
 335 }
 336
 337 static void pvrdma_cq_event(struct pvrdma_dev *dev, u32 cqn, int type)
 338 {
 339         struct pvrdma_cq *cq;
 340         unsigned long flags;
 341
 342         spin_lock_irqsave(&dev->cq_tbl_lock, flags);
 343         cq = dev->cq_tbl[cqn % dev->dsr->caps.max_cq];
 344         if (cq)
 345                 refcount_inc(&cq->refcnt);
 346         spin_unlock_irqrestore(&dev->cq_tbl_lock, flags);
 347
 348         if (cq && cq->ibcq.event_handler) {
 349                 struct ib_cq *ibcq = &cq->ibcq;
 350                 struct ib_event e;
 351
 352                 e.device = ibcq->device;
 353                 e.element.cq = ibcq;
 354                 e.event = type; /* 1:1 mapping for now. */
 355                 ibcq->event_handler(&e, ibcq->cq_context);
 356         }
 357         if (cq) {
 358                 if (refcount_dec_and_test(&cq->refcnt))
 359                         complete(&cq->free);
 360         }
 361 }
 362
 363 static void pvrdma_srq_event(struct pvrdma_dev *dev, u32 srqn, int type)
 364 {
 365         struct pvrdma_srq *srq;
 366         unsigned long flags;
 367
 368         spin_lock_irqsave(&dev->srq_tbl_lock, flags);
 369         if (dev->srq_tbl)
 370                 srq = dev->srq_tbl[srqn % dev->dsr->caps.max_srq];
 371         else
 372                 srq = NULL;
 373         if (srq)
 374                 refcount_inc(&srq->refcnt);
 375         spin_unlock_irqrestore(&dev->srq_tbl_lock, flags);
 376
 377         if (srq && srq->ibsrq.event_handler) {
 378                 struct ib_srq *ibsrq = &srq->ibsrq;
 379                 struct ib_event e;
 380
 381                 e.device = ibsrq->device;
 382                 e.element.srq = ibsrq;
 383                 e.event = type; /* 1:1 mapping for now. */
 384                 ibsrq->event_handler(&e, ibsrq->srq_context);
 385         }
 386         if (srq) {
 387                 if (refcount_dec_and_test(&srq->refcnt))
 388                         complete(&srq->free);
 389         }
 390 }
 391
 392 static void pvrdma_dispatch_event(struct pvrdma_dev *dev, int port,
 393                                   enum ib_event_type event)
 394 {
 395         struct ib_event ib_event;
 396
 397         memset(&ib_event, 0, sizeof(ib_event));
 398         ib_event.device = &dev->ib_dev;
 399         ib_event.element.port_num = port;
 400         ib_event.event = event;
 401         ib_dispatch_event(&ib_event);
 402 }
 403
 404 static void pvrdma_dev_event(struct pvrdma_dev *dev, u8 port, int type)
 405 {
 406         if (port < 1 || port > dev->dsr->caps.phys_port_cnt) {
 407                 dev_warn(&dev->pdev->dev, "event on port %d\n", port);
 408                 return;
 409         }
 410
 411         pvrdma_dispatch_event(dev, port, type);
 412 }
 413
 414 static inline struct pvrdma_eqe *get_eqe(struct pvrdma_dev *dev, unsigned int i)
 415 {
 416         return (struct pvrdma_eqe *)pvrdma_page_dir_get_ptr(
 417                                         &dev->async_pdir,
 418                                         PAGE_SIZE +
 419                                         sizeof(struct pvrdma_eqe) * i);
 420 }
 421
 422 static irqreturn_t pvrdma_intr1_handler(int irq, void *dev_id)
 423 {
 424         struct pvrdma_dev *dev = dev_id;
 425         struct pvrdma_ring *ring = &dev->async_ring_state->rx;
 426         int ring_slots = (dev->dsr->async_ring_pages.num_pages - 1) *
 427                          PAGE_SIZE / sizeof(struct pvrdma_eqe);
 428         unsigned int head;
 429
 430         dev_dbg(&dev->pdev->dev, "interrupt 1 (async event) handler\n");
 431
 432         /*
 433          * Don't process events until the IB device is registered. Otherwise
 434          * we'll try to ib_dispatch_event() on an invalid device.
 435          */
 436         if (!dev->ib_active)
 437                 return IRQ_HANDLED;
 438
 439         while (pvrdma_idx_ring_has_data(ring, ring_slots, &head) > 0) {
 440                 struct pvrdma_eqe *eqe;
 441
 442                 eqe = get_eqe(dev, head);
 443
 444                 switch (eqe->type) {
 445                 case PVRDMA_EVENT_QP_FATAL:
 446                 case PVRDMA_EVENT_QP_REQ_ERR:
 447                 case PVRDMA_EVENT_QP_ACCESS_ERR:
 448                 case PVRDMA_EVENT_COMM_EST:
 449                 case PVRDMA_EVENT_SQ_DRAINED:
 450                 case PVRDMA_EVENT_PATH_MIG:
 451                 case PVRDMA_EVENT_PATH_MIG_ERR:
 452                 case PVRDMA_EVENT_QP_LAST_WQE_REACHED:
 453                         pvrdma_qp_event(dev, eqe->info, eqe->type);
 454                         break;
 455
 456                 case PVRDMA_EVENT_CQ_ERR:
 457                         pvrdma_cq_event(dev, eqe->info, eqe->type);
 458                         break;
 459
 460                 case PVRDMA_EVENT_SRQ_ERR:
 461                 case PVRDMA_EVENT_SRQ_LIMIT_REACHED:
 462                         pvrdma_srq_event(dev, eqe->info, eqe->type);
 463                         break;
 464
 465                 case PVRDMA_EVENT_PORT_ACTIVE:
 466                 case PVRDMA_EVENT_PORT_ERR:
 467                 case PVRDMA_EVENT_LID_CHANGE:
 468                 case PVRDMA_EVENT_PKEY_CHANGE:
 469                 case PVRDMA_EVENT_SM_CHANGE:
 470                 case PVRDMA_EVENT_CLIENT_REREGISTER:
 471                 case PVRDMA_EVENT_GID_CHANGE:
 472                         pvrdma_dev_event(dev, eqe->info, eqe->type);
 473                         break;
 474
 475                 case PVRDMA_EVENT_DEVICE_FATAL:
 476                         pvrdma_dev_event(dev, 1, eqe->type);
 477                         break;
 478
 479                 default:
 480                         break;
 481                 }
 482
 483                 pvrdma_idx_ring_inc(&ring->cons_head, ring_slots);
 484         }
 485
 486         return IRQ_HANDLED;
 487 }
 488
 489 static inline struct pvrdma_cqne *get_cqne(struct pvrdma_dev *dev,
 490                                            unsigned int i)
 491 {
 492         return (struct pvrdma_cqne *)pvrdma_page_dir_get_ptr(
 493                                         &dev->cq_pdir,
 494                                         PAGE_SIZE +
 495                                         sizeof(struct pvrdma_cqne) * i);
 496 }
 497
 498 static irqreturn_t pvrdma_intrx_handler(int irq, void *dev_id)
 499 {
 500         struct pvrdma_dev *dev = dev_id;
 501         struct pvrdma_ring *ring = &dev->cq_ring_state->rx;
 502         int ring_slots = (dev->dsr->cq_ring_pages.num_pages - 1) * PAGE_SIZE /
 503                          sizeof(struct pvrdma_cqne);
 504         unsigned int head;
 505         unsigned long flags;
 506
 507         dev_dbg(&dev->pdev->dev, "interrupt x (completion) handler\n");
 508
 509         while (pvrdma_idx_ring_has_data(ring, ring_slots, &head) > 0) {
 510                 struct pvrdma_cqne *cqne;
 511                 struct pvrdma_cq *cq;
 512
 513                 cqne = get_cqne(dev, head);
 514                 spin_lock_irqsave(&dev->cq_tbl_lock, flags);
 515                 cq = dev->cq_tbl[cqne->info % dev->dsr->caps.max_cq];
 516                 if (cq)
 517                         refcount_inc(&cq->refcnt);
 518                 spin_unlock_irqrestore(&dev->cq_tbl_lock, flags);
 519
 520                 if (cq && cq->ibcq.comp_handler)
 521                         cq->ibcq.comp_handler(&cq->ibcq, cq->ibcq.cq_context);
 522                 if (cq) {
 523                         if (refcount_dec_and_test(&cq->refcnt))
 524                                 complete(&cq->free);
 525                 }
 526                 pvrdma_idx_ring_inc(&ring->cons_head, ring_slots);
 527         }
 528
 529         return IRQ_HANDLED;
 530 }
 531
 532 static void pvrdma_free_irq(struct pvrdma_dev *dev)
 533 {
 534         int i;
 535
 536         dev_dbg(&dev->pdev->dev, "freeing interrupts\n");
 537         for (i = 0; i < dev->nr_vectors; i++)
 538                 free_irq(pci_irq_vector(dev->pdev, i), dev);
 539 }
 540
 541 static void pvrdma_enable_intrs(struct pvrdma_dev *dev)
 542 {
 543         dev_dbg(&dev->pdev->dev, "enable interrupts\n");
 544         pvrdma_write_reg(dev, PVRDMA_REG_IMR, 0);
 545 }
 546
 547 static void pvrdma_disable_intrs(struct pvrdma_dev *dev)
 548 {
 549         dev_dbg(&dev->pdev->dev, "disable interrupts\n");
 550         pvrdma_write_reg(dev, PVRDMA_REG_IMR, ~0);
 551 }
 552
 553 static int pvrdma_alloc_intrs(struct pvrdma_dev *dev)
 554 {
 555         struct pci_dev *pdev = dev->pdev;
 556         int ret = 0, i;
 557
 558         ret = pci_alloc_irq_vectors(pdev, 1, PVRDMA_MAX_INTERRUPTS,
 559                         PCI_IRQ_MSIX);
 560         if (ret < 0) {
 561                 ret = pci_alloc_irq_vectors(pdev, 1, 1,
 562                                 PCI_IRQ_MSI | PCI_IRQ_LEGACY);
 563                 if (ret < 0)
 564                         return ret;
 565         }
 566         dev->nr_vectors = ret;
 567
 568         ret = request_irq(pci_irq_vector(dev->pdev, 0), pvrdma_intr0_handler,
 569                         pdev->msix_enabled ? 0 : IRQF_SHARED, DRV_NAME, dev);
 570         if (ret) {
 571                 dev_err(&dev->pdev->dev,
 572                         "failed to request interrupt 0\n");
 573                 goto out_free_vectors;
 574         }
 575
 576         for (i = 1; i < dev->nr_vectors; i++) {
 577                 ret = request_irq(pci_irq_vector(dev->pdev, i),
 578                                 i == 1 ? pvrdma_intr1_handler :
 579                                          pvrdma_intrx_handler,
 580                                 0, DRV_NAME, dev);
 581                 if (ret) {
 582                         dev_err(&dev->pdev->dev,
 583                                 "failed to request interrupt %d\n", i);
 584                         goto free_irqs;
 585                 }
 586         }
 587
 588         return 0;
 589
 590 free_irqs:
 591         while (--i >= 0)
 592                 free_irq(pci_irq_vector(dev->pdev, i), dev);
 593 out_free_vectors:
 594         pci_free_irq_vectors(pdev);
 595         return ret;
 596 }
 597
 598 static void pvrdma_free_slots(struct pvrdma_dev *dev)
 599 {
 600         struct pci_dev *pdev = dev->pdev;
 601
 602         if (dev->resp_slot)
 603                 dma_free_coherent(&pdev->dev, PAGE_SIZE, dev->resp_slot,
 604                                   dev->dsr->resp_slot_dma);
 605         if (dev->cmd_slot)
 606                 dma_free_coherent(&pdev->dev, PAGE_SIZE, dev->cmd_slot,
 607                                   dev->dsr->cmd_slot_dma);
 608 }
 609
 610 static int pvrdma_add_gid_at_index(struct pvrdma_dev *dev,
 611                                    const union ib_gid *gid,
 612                                    u8 gid_type,
 613                                    int index)
 614 {
 615         int ret;
 616         union pvrdma_cmd_req req;
 617         struct pvrdma_cmd_create_bind *cmd_bind = &req.create_bind;
 618
 619         if (!dev->sgid_tbl) {
 620                 dev_warn(&dev->pdev->dev, "sgid table not initialized\n");
 621                 return -EINVAL;
 622         }
 623
 624         memset(cmd_bind, 0, sizeof(*cmd_bind));
 625         cmd_bind->hdr.cmd = PVRDMA_CMD_CREATE_BIND;
 626         memcpy(cmd_bind->new_gid, gid->raw, 16);
 627         cmd_bind->mtu = ib_mtu_enum_to_int(IB_MTU_1024);
 628         cmd_bind->vlan = 0xfff;
 629         cmd_bind->index = index;
 630         cmd_bind->gid_type = gid_type;
 631
 632         ret = pvrdma_cmd_post(dev, &req, NULL, 0);
 633         if (ret < 0) {
 634                 dev_warn(&dev->pdev->dev,
 635                          "could not create binding, error: %d\n", ret);
 636                 return -EFAULT;
 637         }
 638         memcpy(&dev->sgid_tbl[index], gid, sizeof(*gid));
 639         return 0;
 640 }
 641
 642 static int pvrdma_add_gid(const struct ib_gid_attr *attr, void **context)
 643 {
 644         struct pvrdma_dev *dev = to_vdev(attr->device);
 645
 646         return pvrdma_add_gid_at_index(dev, &attr->gid,
 647                                        ib_gid_type_to_pvrdma(attr->gid_type),
 648                                        attr->index);
 649 }
 650
 651 static int pvrdma_del_gid_at_index(struct pvrdma_dev *dev, int index)
 652 {
 653         int ret;
 654         union pvrdma_cmd_req req;
 655         struct pvrdma_cmd_destroy_bind *cmd_dest = &req.destroy_bind;
 656
 657         /* Update sgid table. */
 658         if (!dev->sgid_tbl) {
 659                 dev_warn(&dev->pdev->dev, "sgid table not initialized\n");
 660                 return -EINVAL;
 661         }
 662
 663         memset(cmd_dest, 0, sizeof(*cmd_dest));
 664         cmd_dest->hdr.cmd = PVRDMA_CMD_DESTROY_BIND;
 665         memcpy(cmd_dest->dest_gid, &dev->sgid_tbl[index], 16);
 666         cmd_dest->index = index;
 667
 668         ret = pvrdma_cmd_post(dev, &req, NULL, 0);
 669         if (ret < 0) {
 670                 dev_warn(&dev->pdev->dev,
 671                          "could not destroy binding, error: %d\n", ret);
 672                 return ret;
 673         }
 674         memset(&dev->sgid_tbl[index], 0, 16);
 675         return 0;
 676 }
 677
 678 static int pvrdma_del_gid(const struct ib_gid_attr *attr, void **context)
 679 {
 680         struct pvrdma_dev *dev = to_vdev(attr->device);
 681
 682         dev_dbg(&dev->pdev->dev, "removing gid at index %u from %s",
 683                 attr->index, dev->netdev->name);
 684
 685         return pvrdma_del_gid_at_index(dev, attr->index);
 686 }
 687
 688 static void pvrdma_netdevice_event_handle(struct pvrdma_dev *dev,
 689                                           struct net_device *ndev,
 690                                           unsigned long event)
 691 {
 692         struct pci_dev *pdev_net;
 693         unsigned int slot;
 694
 695         switch (event) {
 696         case NETDEV_REBOOT:
 697         case NETDEV_DOWN:
 698                 pvrdma_dispatch_event(dev, 1, IB_EVENT_PORT_ERR);
 699                 break;
 700         case NETDEV_UP:
 701                 pvrdma_write_reg(dev, PVRDMA_REG_CTL,
 702                                  PVRDMA_DEVICE_CTL_UNQUIESCE);
 703
 704                 mb();
 705
 706                 if (pvrdma_read_reg(dev, PVRDMA_REG_ERR))
 707                         dev_err(&dev->pdev->dev,
 708                                 "failed to activate device during link up\n");
 709                 else
 710                         pvrdma_dispatch_event(dev, 1, IB_EVENT_PORT_ACTIVE);
 711                 break;
 712         case NETDEV_UNREGISTER:
 713                 ib_device_set_netdev(&dev->ib_dev, NULL, 1);
 714                 dev_put(dev->netdev);
 715                 dev->netdev = NULL;
 716                 break;
 717         case NETDEV_REGISTER:
 718                 /* vmxnet3 will have same bus, slot. But func will be 0 */
 719                 slot = PCI_SLOT(dev->pdev->devfn);
 720                 pdev_net = pci_get_slot(dev->pdev->bus,
 721                                         PCI_DEVFN(slot, 0));
 722                 if ((dev->netdev == NULL) &&
 723                     (pci_get_drvdata(pdev_net) == ndev)) {
 724                         /* this is our netdev */
 725                         ib_device_set_netdev(&dev->ib_dev, ndev, 1);
 726                         dev->netdev = ndev;
 727                         dev_hold(ndev);
 728                 }
 729                 pci_dev_put(pdev_net);
 730                 break;
 731
 732         default:
 733                 dev_dbg(&dev->pdev->dev, "ignore netdevice event %ld on %s\n",
 734                         event, dev_name(&dev->ib_dev.dev));
 735                 break;
 736         }
 737 }
 738
 739 static void pvrdma_netdevice_event_work(struct work_struct *work)
 740 {
 741         struct pvrdma_netdevice_work *netdev_work;
 742         struct pvrdma_dev *dev;
 743
 744         netdev_work = container_of(work, struct pvrdma_netdevice_work, work);
 745
 746         mutex_lock(&pvrdma_device_list_lock);
 747         list_for_each_entry(dev, &pvrdma_device_list, device_link) {
 748                 if ((netdev_work->event == NETDEV_REGISTER) ||
 749                     (dev->netdev == netdev_work->event_netdev)) {
 750                         pvrdma_netdevice_event_handle(dev,
 751                                                       netdev_work->event_netdev,
 752                                                       netdev_work->event);
 753                         break;
 754                 }
 755         }
 756         mutex_unlock(&pvrdma_device_list_lock);
 757
 758         kfree(netdev_work);
 759 }
 760
 761 static int pvrdma_netdevice_event(struct notifier_block *this,
 762                                   unsigned long event, void *ptr)
 763 {
 764         struct net_device *event_netdev = netdev_notifier_info_to_dev(ptr);
 765         struct pvrdma_netdevice_work *netdev_work;
 766
 767         netdev_work = kmalloc(sizeof(*netdev_work), GFP_ATOMIC);
 768         if (!netdev_work)
 769                 return NOTIFY_BAD;
 770
 771         INIT_WORK(&netdev_work->work, pvrdma_netdevice_event_work);
 772         netdev_work->event_netdev = event_netdev;
 773         netdev_work->event = event;
 774         queue_work(event_wq, &netdev_work->work);
 775
 776         return NOTIFY_DONE;
 777 }
 778
 779 static int pvrdma_pci_probe(struct pci_dev *pdev,
 780                             const struct pci_device_id *id)
 781 {
 782         struct pci_dev *pdev_net;
 783         struct pvrdma_dev *dev;
 784         int ret;
 785         unsigned long start;
 786         unsigned long len;
 787         dma_addr_t slot_dma = 0;
 788
 789         dev_dbg(&pdev->dev, "initializing driver %s\n", pci_name(pdev));
 790
 791         /* Allocate zero-out device */
 792         dev = ib_alloc_device(pvrdma_dev, ib_dev);
 793         if (!dev) {
 794                 dev_err(&pdev->dev, "failed to allocate IB device\n");
 795                 return -ENOMEM;
 796         }
 797
 798         mutex_lock(&pvrdma_device_list_lock);
 799         list_add(&dev->device_link, &pvrdma_device_list);
 800         mutex_unlock(&pvrdma_device_list_lock);
 801
 802         ret = pvrdma_init_device(dev);
 803         if (ret)
 804                 goto err_free_device;
 805
 806         dev->pdev = pdev;
 807         pci_set_drvdata(pdev, dev);
 808
 809         ret = pci_enable_device(pdev);
 810         if (ret) {
 811                 dev_err(&pdev->dev, "cannot enable PCI device\n");
 812                 goto err_free_device;
 813         }
 814
 815         dev_dbg(&pdev->dev, "PCI resource flags BAR0 %#lx\n",
 816                 pci_resource_flags(pdev, 0));
 817         dev_dbg(&pdev->dev, "PCI resource len %#llx\n",
 818                 (unsigned long long)pci_resource_len(pdev, 0));
 819         dev_dbg(&pdev->dev, "PCI resource start %#llx\n",
 820                 (unsigned long long)pci_resource_start(pdev, 0));
 821         dev_dbg(&pdev->dev, "PCI resource flags BAR1 %#lx\n",
 822                 pci_resource_flags(pdev, 1));
 823         dev_dbg(&pdev->dev, "PCI resource len %#llx\n",
 824                 (unsigned long long)pci_resource_len(pdev, 1));
 825         dev_dbg(&pdev->dev, "PCI resource start %#llx\n",
 826                 (unsigned long long)pci_resource_start(pdev, 1));
 827
 828         if (!(pci_resource_flags(pdev, 0) & IORESOURCE_MEM) ||
 829             !(pci_resource_flags(pdev, 1) & IORESOURCE_MEM)) {
 830                 dev_err(&pdev->dev, "PCI BAR region not MMIO\n");
 831                 ret = -ENOMEM;
 832                 goto err_free_device;
 833         }
 834
 835         ret = pci_request_regions(pdev, DRV_NAME);
 836         if (ret) {
 837                 dev_err(&pdev->dev, "cannot request PCI resources\n");
 838                 goto err_disable_pdev;
 839         }
 840
 841         /* Enable 64-Bit DMA */
 842         if (pci_set_dma_mask(pdev, DMA_BIT_MASK(64)) == 0) {
 843                 ret = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
 844                 if (ret != 0) {
 845                         dev_err(&pdev->dev,
 846                                 "pci_set_consistent_dma_mask failed\n");
 847                         goto err_free_resource;
 848                 }
 849         } else {
 850                 ret = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
 851                 if (ret != 0) {
 852                         dev_err(&pdev->dev,
 853                                 "pci_set_dma_mask failed\n");
 854                         goto err_free_resource;
 855                 }
 856         }
 857
 858         pci_set_master(pdev);
 859
 860         /* Map register space */
 861         start = pci_resource_start(dev->pdev, PVRDMA_PCI_RESOURCE_REG);
 862         len = pci_resource_len(dev->pdev, PVRDMA_PCI_RESOURCE_REG);
 863         dev->regs = ioremap(start, len);
 864         if (!dev->regs) {
 865                 dev_err(&pdev->dev, "register mapping failed\n");
 866                 ret = -ENOMEM;
 867                 goto err_free_resource;
 868         }
 869
 870         /* Setup per-device UAR. */
 871         dev->driver_uar.index = 0;
 872         dev->driver_uar.pfn =
 873                 pci_resource_start(dev->pdev, PVRDMA_PCI_RESOURCE_UAR) >>
 874                 PAGE_SHIFT;
 875         dev->driver_uar.map =
 876                 ioremap(dev->driver_uar.pfn << PAGE_SHIFT, PAGE_SIZE);
 877         if (!dev->driver_uar.map) {
 878                 dev_err(&pdev->dev, "failed to remap UAR pages\n");
 879                 ret = -ENOMEM;
 880                 goto err_unmap_regs;
 881         }
 882
 883         dev->dsr_version = pvrdma_read_reg(dev, PVRDMA_REG_VERSION);
 884         dev_info(&pdev->dev, "device version %d, driver version %d\n",
 885                  dev->dsr_version, PVRDMA_VERSION);
 886
 887         dev->dsr = dma_alloc_coherent(&pdev->dev, sizeof(*dev->dsr),
 888                                       &dev->dsrbase, GFP_KERNEL);
 889         if (!dev->dsr) {
 890                 dev_err(&pdev->dev, "failed to allocate shared region\n");
 891                 ret = -ENOMEM;
 892                 goto err_uar_unmap;
 893         }
 894
 895         /* Setup the shared region */
 896         dev->dsr->driver_version = PVRDMA_VERSION;
 897         dev->dsr->gos_info.gos_bits = sizeof(void *) == 4 ?
 898                 PVRDMA_GOS_BITS_32 :
 899                 PVRDMA_GOS_BITS_64;
 900         dev->dsr->gos_info.gos_type = PVRDMA_GOS_TYPE_LINUX;
 901         dev->dsr->gos_info.gos_ver = 1;
 902
 903         if (dev->dsr_version < PVRDMA_PPN64_VERSION)
 904                 dev->dsr->uar_pfn = dev->driver_uar.pfn;
 905         else
 906                 dev->dsr->uar_pfn64 = dev->driver_uar.pfn;
 907
 908         /* Command slot. */
 909         dev->cmd_slot = dma_alloc_coherent(&pdev->dev, PAGE_SIZE,
 910                                            &slot_dma, GFP_KERNEL);
 911         if (!dev->cmd_slot) {
 912                 ret = -ENOMEM;
 913                 goto err_free_dsr;
 914         }
 915
 916         dev->dsr->cmd_slot_dma = (u64)slot_dma;
 917
 918         /* Response slot. */
 919         dev->resp_slot = dma_alloc_coherent(&pdev->dev, PAGE_SIZE,
 920                                             &slot_dma, GFP_KERNEL);
 921         if (!dev->resp_slot) {
 922                 ret = -ENOMEM;
 923                 goto err_free_slots;
 924         }
 925
 926         dev->dsr->resp_slot_dma = (u64)slot_dma;
 927
 928         /* Async event ring */
 929         dev->dsr->async_ring_pages.num_pages = PVRDMA_NUM_RING_PAGES;
 930         ret = pvrdma_page_dir_init(dev, &dev->async_pdir,
 931                                    dev->dsr->async_ring_pages.num_pages, true);
 932         if (ret)
 933                 goto err_free_slots;
 934         dev->async_ring_state = dev->async_pdir.pages[0];
 935         dev->dsr->async_ring_pages.pdir_dma = dev->async_pdir.dir_dma;
 936
 937         /* CQ notification ring */
 938         dev->dsr->cq_ring_pages.num_pages = PVRDMA_NUM_RING_PAGES;
 939         ret = pvrdma_page_dir_init(dev, &dev->cq_pdir,
 940                                    dev->dsr->cq_ring_pages.num_pages, true);
 941         if (ret)
 942                 goto err_free_async_ring;
 943         dev->cq_ring_state = dev->cq_pdir.pages[0];
 944         dev->dsr->cq_ring_pages.pdir_dma = dev->cq_pdir.dir_dma;
 945
 946         /*
 947          * Write the PA of the shared region to the device. The writes must be
 948          * ordered such that the high bits are written last. When the writes
 949          * complete, the device will have filled out the capabilities.
 950          */
 951
 952         pvrdma_write_reg(dev, PVRDMA_REG_DSRLOW, (u32)dev->dsrbase);
 953         pvrdma_write_reg(dev, PVRDMA_REG_DSRHIGH,
 954                          (u32)((u64)(dev->dsrbase) >> 32));
 955
 956         /* Make sure the write is complete before reading status. */
 957         mb();
 958
 959         /* The driver supports RoCE V1 and V2. */
 960         if (!PVRDMA_SUPPORTED(dev)) {
 961                 dev_err(&pdev->dev, "driver needs RoCE v1 or v2 support\n");
 962                 ret = -EFAULT;
 963                 goto err_free_cq_ring;
 964         }
 965
 966         /* Paired vmxnet3 will have same bus, slot. But func will be 0 */
 967         pdev_net = pci_get_slot(pdev->bus, PCI_DEVFN(PCI_SLOT(pdev->devfn), 0));
 968         if (!pdev_net) {
 969                 dev_err(&pdev->dev, "failed to find paired net device\n");
 970                 ret = -ENODEV;
 971                 goto err_free_cq_ring;
 972         }
 973
 974         if (pdev_net->vendor != PCI_VENDOR_ID_VMWARE ||
 975             pdev_net->device != PCI_DEVICE_ID_VMWARE_VMXNET3) {
 976                 dev_err(&pdev->dev, "failed to find paired vmxnet3 device\n");
 977                 pci_dev_put(pdev_net);
 978                 ret = -ENODEV;
 979                 goto err_free_cq_ring;
 980         }
 981
 982         dev->netdev = pci_get_drvdata(pdev_net);
 983         pci_dev_put(pdev_net);
 984         if (!dev->netdev) {
 985                 dev_err(&pdev->dev, "failed to get vmxnet3 device\n");
 986                 ret = -ENODEV;
 987                 goto err_free_cq_ring;
 988         }
 989         dev_hold(dev->netdev);
 990
 991         dev_info(&pdev->dev, "paired device to %s\n", dev->netdev->name);
 992
 993         /* Interrupt setup */
 994         ret = pvrdma_alloc_intrs(dev);
 995         if (ret) {
 996                 dev_err(&pdev->dev, "failed to allocate interrupts\n");
 997                 ret = -ENOMEM;
 998                 goto err_free_cq_ring;
 999         }
1000
1001         /* Allocate UAR table. */
1002         ret = pvrdma_uar_table_init(dev);
1003         if (ret) {
1004                 dev_err(&pdev->dev, "failed to allocate UAR table\n");
1005                 ret = -ENOMEM;
1006                 goto err_free_intrs;
1007         }
1008
1009         /* Allocate GID table */
1010         dev->sgid_tbl = kcalloc(dev->dsr->caps.gid_tbl_len,
1011                                 sizeof(union ib_gid), GFP_KERNEL);
1012         if (!dev->sgid_tbl) {
1013                 ret = -ENOMEM;
1014                 goto err_free_uar_table;
1015         }
1016         dev_dbg(&pdev->dev, "gid table len %d\n", dev->dsr->caps.gid_tbl_len);
1017
1018         pvrdma_enable_intrs(dev);
1019
1020         /* Activate pvrdma device */
1021         pvrdma_write_reg(dev, PVRDMA_REG_CTL, PVRDMA_DEVICE_CTL_ACTIVATE);
1022
1023         /* Make sure the write is complete before reading status. */
1024         mb();
1025
1026         /* Check if device was successfully activated */
1027         ret = pvrdma_read_reg(dev, PVRDMA_REG_ERR);
1028         if (ret != 0) {
1029                 dev_err(&pdev->dev, "failed to activate device\n");
1030                 ret = -EFAULT;
1031                 goto err_disable_intr;
1032         }
1033
1034         /* Register IB device */
1035         ret = pvrdma_register_device(dev);
1036         if (ret) {
1037                 dev_err(&pdev->dev, "failed to register IB device\n");
1038                 goto err_disable_intr;
1039         }
1040
1041         dev->nb_netdev.notifier_call = pvrdma_netdevice_event;
1042         ret = register_netdevice_notifier(&dev->nb_netdev);
1043         if (ret) {
1044                 dev_err(&pdev->dev, "failed to register netdevice events\n");
1045                 goto err_unreg_ibdev;
1046         }
1047
1048         dev_info(&pdev->dev, "attached to device\n");
1049         return 0;
1050
1051 err_unreg_ibdev:
1052         ib_unregister_device(&dev->ib_dev);
1053 err_disable_intr:
1054         pvrdma_disable_intrs(dev);
1055         kfree(dev->sgid_tbl);
1056 err_free_uar_table:
1057         pvrdma_uar_table_cleanup(dev);
1058 err_free_intrs:
1059         pvrdma_free_irq(dev);
1060         pci_free_irq_vectors(pdev);
1061 err_free_cq_ring:
1062         if (dev->netdev) {
1063                 dev_put(dev->netdev);
1064                 dev->netdev = NULL;
1065         }
1066         pvrdma_page_dir_cleanup(dev, &dev->cq_pdir);
1067 err_free_async_ring:
1068         pvrdma_page_dir_cleanup(dev, &dev->async_pdir);
1069 err_free_slots:
1070         pvrdma_free_slots(dev);
1071 err_free_dsr:
1072         dma_free_coherent(&pdev->dev, sizeof(*dev->dsr), dev->dsr,
1073                           dev->dsrbase);
1074 err_uar_unmap:
1075         iounmap(dev->driver_uar.map);
1076 err_unmap_regs:
1077         iounmap(dev->regs);
1078 err_free_resource:
1079         pci_release_regions(pdev);
1080 err_disable_pdev:
1081         pci_disable_device(pdev);
1082         pci_set_drvdata(pdev, NULL);
1083 err_free_device:
1084         mutex_lock(&pvrdma_device_list_lock);
1085         list_del(&dev->device_link);
1086         mutex_unlock(&pvrdma_device_list_lock);
1087         ib_dealloc_device(&dev->ib_dev);
1088         return ret;
1089 }
1090
1091 static void pvrdma_pci_remove(struct pci_dev *pdev)
1092 {
1093         struct pvrdma_dev *dev = pci_get_drvdata(pdev);
1094
1095         if (!dev)
1096                 return;
1097
1098         dev_info(&pdev->dev, "detaching from device\n");
1099
1100         unregister_netdevice_notifier(&dev->nb_netdev);
1101         dev->nb_netdev.notifier_call = NULL;
1102
1103         flush_workqueue(event_wq);
1104
1105         if (dev->netdev) {
1106                 dev_put(dev->netdev);
1107                 dev->netdev = NULL;
1108         }
1109
1110         /* Unregister ib device */
1111         ib_unregister_device(&dev->ib_dev);
1112
1113         mutex_lock(&pvrdma_device_list_lock);
1114         list_del(&dev->device_link);
1115         mutex_unlock(&pvrdma_device_list_lock);
1116
1117         pvrdma_disable_intrs(dev);
1118         pvrdma_free_irq(dev);
1119         pci_free_irq_vectors(pdev);
1120
1121         /* Deactivate pvrdma device */
1122         pvrdma_write_reg(dev, PVRDMA_REG_CTL, PVRDMA_DEVICE_CTL_RESET);
1123         pvrdma_page_dir_cleanup(dev, &dev->cq_pdir);
1124         pvrdma_page_dir_cleanup(dev, &dev->async_pdir);
1125         pvrdma_free_slots(dev);
1126         dma_free_coherent(&pdev->dev, sizeof(*dev->dsr), dev->dsr,
1127                           dev->dsrbase);
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");