drivers/infiniband/hw/efa/efa_main.c

   1 // SPDX-License-Identifier: GPL-2.0 OR BSD-2-Clause
   2 /*
   3  * Copyright 2018-2024 Amazon.com, Inc. or its affiliates. All rights reserved.
   4  */
   5
   6 #include <linux/module.h>
   7 #include <linux/pci.h>
   8 #include <linux/utsname.h>
   9 #include <linux/version.h>
  10
  11 #include <rdma/ib_user_verbs.h>
  12 #include <rdma/uverbs_ioctl.h>
  13
  14 #include "efa.h"
  15
  16 #define PCI_DEV_ID_EFA0_VF 0xefa0
  17 #define PCI_DEV_ID_EFA1_VF 0xefa1
  18 #define PCI_DEV_ID_EFA2_VF 0xefa2
  19 #define PCI_DEV_ID_EFA3_VF 0xefa3
  20
  21 static const struct pci_device_id efa_pci_tbl[] = {
  22         { PCI_VDEVICE(AMAZON, PCI_DEV_ID_EFA0_VF) },
  23         { PCI_VDEVICE(AMAZON, PCI_DEV_ID_EFA1_VF) },
  24         { PCI_VDEVICE(AMAZON, PCI_DEV_ID_EFA2_VF) },
  25         { PCI_VDEVICE(AMAZON, PCI_DEV_ID_EFA3_VF) },
  26         { }
  27 };
  28
  29 MODULE_AUTHOR("Amazon.com, Inc. or its affiliates");
  30 MODULE_LICENSE("Dual BSD/GPL");
  31 MODULE_DESCRIPTION(DEVICE_NAME);
  32 MODULE_DEVICE_TABLE(pci, efa_pci_tbl);
  33
  34 #define EFA_REG_BAR 0
  35 #define EFA_MEM_BAR 2
  36 #define EFA_BASE_BAR_MASK (BIT(EFA_REG_BAR) | BIT(EFA_MEM_BAR))
  37
  38 #define EFA_AENQ_ENABLED_GROUPS \
  39         (BIT(EFA_ADMIN_FATAL_ERROR) | BIT(EFA_ADMIN_WARNING) | \
  40          BIT(EFA_ADMIN_NOTIFICATION) | BIT(EFA_ADMIN_KEEP_ALIVE))
  41
  42 extern const struct uapi_definition efa_uapi_defs[];
  43
  44 /* This handler will called for unknown event group or unimplemented handlers */
  45 static void unimplemented_aenq_handler(void *data,
  46                                        struct efa_admin_aenq_entry *aenq_e)
  47 {
  48         struct efa_dev *dev = (struct efa_dev *)data;
  49
  50         ibdev_err(&dev->ibdev,
  51                   "Unknown event was received or event with unimplemented handler\n");
  52 }
  53
  54 static void efa_keep_alive(void *data, struct efa_admin_aenq_entry *aenq_e)
  55 {
  56         struct efa_dev *dev = (struct efa_dev *)data;
  57
  58         atomic64_inc(&dev->stats.keep_alive_rcvd);
  59 }
  60
  61 static struct efa_aenq_handlers aenq_handlers = {
  62         .handlers = {
  63                 [EFA_ADMIN_KEEP_ALIVE] = efa_keep_alive,
  64         },
  65         .unimplemented_handler = unimplemented_aenq_handler
  66 };
  67
  68 static void efa_release_bars(struct efa_dev *dev, int bars_mask)
  69 {
  70         struct pci_dev *pdev = dev->pdev;
  71         int release_bars;
  72
  73         release_bars = pci_select_bars(pdev, IORESOURCE_MEM) & bars_mask;
  74         pci_release_selected_regions(pdev, release_bars);
  75 }
  76
  77 static void efa_process_comp_eqe(struct efa_dev *dev, struct efa_admin_eqe *eqe)
  78 {
  79         u16 cqn = eqe->u.comp_event.cqn;
  80         struct efa_cq *cq;
  81
  82         /* Safe to load as we're in irq and removal calls synchronize_irq() */
  83         cq = xa_load(&dev->cqs_xa, cqn);
  84         if (unlikely(!cq)) {
  85                 ibdev_err_ratelimited(&dev->ibdev,
  86                                       "Completion event on non-existent CQ[%u]",
  87                                       cqn);
  88                 return;
  89         }
  90
  91         cq->ibcq.comp_handler(&cq->ibcq, cq->ibcq.cq_context);
  92 }
  93
  94 static void efa_process_eqe(struct efa_com_eq *eeq, struct efa_admin_eqe *eqe)
  95 {
  96         struct efa_dev *dev = container_of(eeq->edev, struct efa_dev, edev);
  97
  98         if (likely(EFA_GET(&eqe->common, EFA_ADMIN_EQE_EVENT_TYPE) ==
  99                            EFA_ADMIN_EQE_EVENT_TYPE_COMPLETION))
 100                 efa_process_comp_eqe(dev, eqe);
 101         else
 102                 ibdev_err_ratelimited(&dev->ibdev,
 103                                       "Unknown event type received %lu",
 104                                       EFA_GET(&eqe->common,
 105                                               EFA_ADMIN_EQE_EVENT_TYPE));
 106 }
 107
 108 static irqreturn_t efa_intr_msix_comp(int irq, void *data)
 109 {
 110         struct efa_eq *eq = data;
 111         struct efa_com_dev *edev = eq->eeq.edev;
 112
 113         efa_com_eq_comp_intr_handler(edev, &eq->eeq);
 114
 115         return IRQ_HANDLED;
 116 }
 117
 118 static irqreturn_t efa_intr_msix_mgmnt(int irq, void *data)
 119 {
 120         struct efa_dev *dev = data;
 121
 122         efa_com_admin_q_comp_intr_handler(&dev->edev);
 123         efa_com_aenq_intr_handler(&dev->edev, data);
 124
 125         return IRQ_HANDLED;
 126 }
 127
 128 static int efa_request_irq(struct efa_dev *dev, struct efa_irq *irq)
 129 {
 130         int err;
 131
 132         err = request_irq(irq->irqn, irq->handler, 0, irq->name, irq->data);
 133         if (err) {
 134                 dev_err(&dev->pdev->dev, "Failed to request irq %s (%d)\n",
 135                         irq->name, err);
 136                 return err;
 137         }
 138
 139         irq_set_affinity_hint(irq->irqn, &irq->affinity_hint_mask);
 140
 141         return 0;
 142 }
 143
 144 static void efa_setup_comp_irq(struct efa_dev *dev, struct efa_eq *eq,
 145                                int vector)
 146 {
 147         u32 cpu;
 148
 149         cpu = vector - EFA_COMP_EQS_VEC_BASE;
 150         snprintf(eq->irq.name, EFA_IRQNAME_SIZE, "efa-comp%d@pci:%s", cpu,
 151                  pci_name(dev->pdev));
 152         eq->irq.handler = efa_intr_msix_comp;
 153         eq->irq.data = eq;
 154         eq->irq.vector = vector;
 155         eq->irq.irqn = pci_irq_vector(dev->pdev, vector);
 156         cpumask_set_cpu(cpu, &eq->irq.affinity_hint_mask);
 157 }
 158
 159 static void efa_free_irq(struct efa_dev *dev, struct efa_irq *irq)
 160 {
 161         irq_set_affinity_hint(irq->irqn, NULL);
 162         free_irq(irq->irqn, irq->data);
 163 }
 164
 165 static void efa_setup_mgmnt_irq(struct efa_dev *dev)
 166 {
 167         u32 cpu;
 168
 169         snprintf(dev->admin_irq.name, EFA_IRQNAME_SIZE,
 170                  "efa-mgmnt@pci:%s", pci_name(dev->pdev));
 171         dev->admin_irq.handler = efa_intr_msix_mgmnt;
 172         dev->admin_irq.data = dev;
 173         dev->admin_irq.vector = dev->admin_msix_vector_idx;
 174         dev->admin_irq.irqn = pci_irq_vector(dev->pdev,
 175                                              dev->admin_msix_vector_idx);
 176         cpu = cpumask_first(cpu_online_mask);
 177         cpumask_set_cpu(cpu,
 178                         &dev->admin_irq.affinity_hint_mask);
 179         dev_info(&dev->pdev->dev, "Setup irq:%d name:%s\n",
 180                  dev->admin_irq.irqn,
 181                  dev->admin_irq.name);
 182 }
 183
 184 static int efa_set_mgmnt_irq(struct efa_dev *dev)
 185 {
 186         efa_setup_mgmnt_irq(dev);
 187
 188         return efa_request_irq(dev, &dev->admin_irq);
 189 }
 190
 191 static int efa_request_doorbell_bar(struct efa_dev *dev)
 192 {
 193         u8 db_bar_idx = dev->dev_attr.db_bar;
 194         struct pci_dev *pdev = dev->pdev;
 195         int pci_mem_bars;
 196         int db_bar;
 197         int err;
 198
 199         db_bar = BIT(db_bar_idx);
 200         if (!(db_bar & EFA_BASE_BAR_MASK)) {
 201                 pci_mem_bars = pci_select_bars(pdev, IORESOURCE_MEM);
 202                 if (db_bar & ~pci_mem_bars) {
 203                         dev_err(&pdev->dev,
 204                                 "Doorbells BAR unavailable. Requested %#x, available %#x\n",
 205                                 db_bar, pci_mem_bars);
 206                         return -ENODEV;
 207                 }
 208
 209                 err = pci_request_selected_regions(pdev, db_bar, DRV_MODULE_NAME);
 210                 if (err) {
 211                         dev_err(&pdev->dev,
 212                                 "pci_request_selected_regions for bar %d failed %d\n",
 213                                 db_bar_idx, err);
 214                         return err;
 215                 }
 216         }
 217
 218         dev->db_bar_addr = pci_resource_start(dev->pdev, db_bar_idx);
 219         dev->db_bar_len = pci_resource_len(dev->pdev, db_bar_idx);
 220
 221         return 0;
 222 }
 223
 224 static void efa_release_doorbell_bar(struct efa_dev *dev)
 225 {
 226         if (!(BIT(dev->dev_attr.db_bar) & EFA_BASE_BAR_MASK))
 227                 efa_release_bars(dev, BIT(dev->dev_attr.db_bar));
 228 }
 229
 230 static void efa_update_hw_hints(struct efa_dev *dev,
 231                                 struct efa_com_get_hw_hints_result *hw_hints)
 232 {
 233         struct efa_com_dev *edev = &dev->edev;
 234
 235         if (hw_hints->mmio_read_timeout)
 236                 edev->mmio_read.mmio_read_timeout =
 237                         hw_hints->mmio_read_timeout * 1000;
 238
 239         if (hw_hints->poll_interval)
 240                 edev->aq.poll_interval = hw_hints->poll_interval;
 241
 242         if (hw_hints->admin_completion_timeout)
 243                 edev->aq.completion_timeout =
 244                         hw_hints->admin_completion_timeout;
 245 }
 246
 247 static void efa_stats_init(struct efa_dev *dev)
 248 {
 249         atomic64_t *s = (atomic64_t *)&dev->stats;
 250         int i;
 251
 252         for (i = 0; i < sizeof(dev->stats) / sizeof(*s); i++, s++)
 253                 atomic64_set(s, 0);
 254 }
 255
 256 static void efa_set_host_info(struct efa_dev *dev)
 257 {
 258         struct efa_admin_set_feature_resp resp = {};
 259         struct efa_admin_set_feature_cmd cmd = {};
 260         struct efa_admin_host_info *hinf;
 261         u32 bufsz = sizeof(*hinf);
 262         dma_addr_t hinf_dma;
 263
 264         if (!efa_com_check_supported_feature_id(&dev->edev,
 265                                                 EFA_ADMIN_HOST_INFO))
 266                 return;
 267
 268         /* Failures in host info set shall not disturb probe */
 269         hinf = dma_alloc_coherent(&dev->pdev->dev, bufsz, &hinf_dma,
 270                                   GFP_KERNEL);
 271         if (!hinf)
 272                 return;
 273
 274         strscpy(hinf->os_dist_str, utsname()->release,
 275                 sizeof(hinf->os_dist_str));
 276         hinf->os_type = EFA_ADMIN_OS_LINUX;
 277         strscpy(hinf->kernel_ver_str, utsname()->version,
 278                 sizeof(hinf->kernel_ver_str));
 279         hinf->kernel_ver = LINUX_VERSION_CODE;
 280         EFA_SET(&hinf->driver_ver, EFA_ADMIN_HOST_INFO_DRIVER_MAJOR, 0);
 281         EFA_SET(&hinf->driver_ver, EFA_ADMIN_HOST_INFO_DRIVER_MINOR, 0);
 282         EFA_SET(&hinf->driver_ver, EFA_ADMIN_HOST_INFO_DRIVER_SUB_MINOR, 0);
 283         EFA_SET(&hinf->driver_ver, EFA_ADMIN_HOST_INFO_DRIVER_MODULE_TYPE, 0);
 284         EFA_SET(&hinf->bdf, EFA_ADMIN_HOST_INFO_BUS, dev->pdev->bus->number);
 285         EFA_SET(&hinf->bdf, EFA_ADMIN_HOST_INFO_DEVICE,
 286                 PCI_SLOT(dev->pdev->devfn));
 287         EFA_SET(&hinf->bdf, EFA_ADMIN_HOST_INFO_FUNCTION,
 288                 PCI_FUNC(dev->pdev->devfn));
 289         EFA_SET(&hinf->spec_ver, EFA_ADMIN_HOST_INFO_SPEC_MAJOR,
 290                 EFA_COMMON_SPEC_VERSION_MAJOR);
 291         EFA_SET(&hinf->spec_ver, EFA_ADMIN_HOST_INFO_SPEC_MINOR,
 292                 EFA_COMMON_SPEC_VERSION_MINOR);
 293         EFA_SET(&hinf->flags, EFA_ADMIN_HOST_INFO_INTREE, 1);
 294         EFA_SET(&hinf->flags, EFA_ADMIN_HOST_INFO_GDR, 0);
 295
 296         efa_com_set_feature_ex(&dev->edev, &resp, &cmd, EFA_ADMIN_HOST_INFO,
 297                                hinf_dma, bufsz);
 298
 299         dma_free_coherent(&dev->pdev->dev, bufsz, hinf, hinf_dma);
 300 }
 301
 302 static void efa_destroy_eq(struct efa_dev *dev, struct efa_eq *eq)
 303 {
 304         efa_com_eq_destroy(&dev->edev, &eq->eeq);
 305         efa_free_irq(dev, &eq->irq);
 306 }
 307
 308 static int efa_create_eq(struct efa_dev *dev, struct efa_eq *eq, u8 msix_vec)
 309 {
 310         int err;
 311
 312         efa_setup_comp_irq(dev, eq, msix_vec);
 313         err = efa_request_irq(dev, &eq->irq);
 314         if (err)
 315                 return err;
 316
 317         err = efa_com_eq_init(&dev->edev, &eq->eeq, efa_process_eqe,
 318                               dev->dev_attr.max_eq_depth, msix_vec);
 319         if (err)
 320                 goto err_free_comp_irq;
 321
 322         return 0;
 323
 324 err_free_comp_irq:
 325         efa_free_irq(dev, &eq->irq);
 326         return err;
 327 }
 328
 329 static int efa_create_eqs(struct efa_dev *dev)
 330 {
 331         unsigned int neqs = dev->dev_attr.max_eq;
 332         int err;
 333         int i;
 334
 335         neqs = min_t(unsigned int, neqs,
 336                      dev->num_irq_vectors - EFA_COMP_EQS_VEC_BASE);
 337
 338         dev->neqs = neqs;
 339         dev->eqs = kcalloc(neqs, sizeof(*dev->eqs), GFP_KERNEL);
 340         if (!dev->eqs)
 341                 return -ENOMEM;
 342
 343         for (i = 0; i < neqs; i++) {
 344                 err = efa_create_eq(dev, &dev->eqs[i],
 345                                     i + EFA_COMP_EQS_VEC_BASE);
 346                 if (err)
 347                         goto err_destroy_eqs;
 348         }
 349
 350         return 0;
 351
 352 err_destroy_eqs:
 353         for (i--; i >= 0; i--)
 354                 efa_destroy_eq(dev, &dev->eqs[i]);
 355         kfree(dev->eqs);
 356
 357         return err;
 358 }
 359
 360 static void efa_destroy_eqs(struct efa_dev *dev)
 361 {
 362         int i;
 363
 364         for (i = 0; i < dev->neqs; i++)
 365                 efa_destroy_eq(dev, &dev->eqs[i]);
 366
 367         kfree(dev->eqs);
 368 }
 369
 370 static const struct ib_device_ops efa_dev_ops = {
 371         .owner = THIS_MODULE,
 372         .driver_id = RDMA_DRIVER_EFA,
 373         .uverbs_abi_ver = EFA_UVERBS_ABI_VERSION,
 374
 375         .alloc_hw_port_stats = efa_alloc_hw_port_stats,
 376         .alloc_hw_device_stats = efa_alloc_hw_device_stats,
 377         .alloc_pd = efa_alloc_pd,
 378         .alloc_ucontext = efa_alloc_ucontext,
 379         .create_cq = efa_create_cq,
 380         .create_qp = efa_create_qp,
 381         .create_user_ah = efa_create_ah,
 382         .dealloc_pd = efa_dealloc_pd,
 383         .dealloc_ucontext = efa_dealloc_ucontext,
 384         .dereg_mr = efa_dereg_mr,
 385         .destroy_ah = efa_destroy_ah,
 386         .destroy_cq = efa_destroy_cq,
 387         .destroy_qp = efa_destroy_qp,
 388         .get_hw_stats = efa_get_hw_stats,
 389         .get_link_layer = efa_port_link_layer,
 390         .get_port_immutable = efa_get_port_immutable,
 391         .mmap = efa_mmap,
 392         .mmap_free = efa_mmap_free,
 393         .modify_qp = efa_modify_qp,
 394         .query_device = efa_query_device,
 395         .query_gid = efa_query_gid,
 396         .query_pkey = efa_query_pkey,
 397         .query_port = efa_query_port,
 398         .query_qp = efa_query_qp,
 399         .reg_user_mr = efa_reg_mr,
 400         .reg_user_mr_dmabuf = efa_reg_user_mr_dmabuf,
 401
 402         INIT_RDMA_OBJ_SIZE(ib_ah, efa_ah, ibah),
 403         INIT_RDMA_OBJ_SIZE(ib_cq, efa_cq, ibcq),
 404         INIT_RDMA_OBJ_SIZE(ib_pd, efa_pd, ibpd),
 405         INIT_RDMA_OBJ_SIZE(ib_qp, efa_qp, ibqp),
 406         INIT_RDMA_OBJ_SIZE(ib_ucontext, efa_ucontext, ibucontext),
 407 };
 408
 409 static int efa_ib_device_add(struct efa_dev *dev)
 410 {
 411         struct efa_com_get_hw_hints_result hw_hints;
 412         struct pci_dev *pdev = dev->pdev;
 413         int err;
 414
 415         efa_stats_init(dev);
 416
 417         err = efa_com_get_device_attr(&dev->edev, &dev->dev_attr);
 418         if (err)
 419                 return err;
 420
 421         dev_dbg(&dev->pdev->dev, "Doorbells bar (%d)\n", dev->dev_attr.db_bar);
 422         err = efa_request_doorbell_bar(dev);
 423         if (err)
 424                 return err;
 425
 426         err = efa_com_get_hw_hints(&dev->edev, &hw_hints);
 427         if (err)
 428                 goto err_release_doorbell_bar;
 429
 430         efa_update_hw_hints(dev, &hw_hints);
 431
 432         /* Try to enable all the available aenq groups */
 433         err = efa_com_set_aenq_config(&dev->edev, EFA_AENQ_ENABLED_GROUPS);
 434         if (err)
 435                 goto err_release_doorbell_bar;
 436
 437         err = efa_create_eqs(dev);
 438         if (err)
 439                 goto err_release_doorbell_bar;
 440
 441         efa_set_host_info(dev);
 442
 443         dev->ibdev.node_type = RDMA_NODE_UNSPECIFIED;
 444         dev->ibdev.node_guid = dev->dev_attr.guid;
 445         dev->ibdev.phys_port_cnt = 1;
 446         dev->ibdev.num_comp_vectors = dev->neqs ?: 1;
 447         dev->ibdev.dev.parent = &pdev->dev;
 448
 449         ib_set_device_ops(&dev->ibdev, &efa_dev_ops);
 450
 451         dev->ibdev.driver_def = efa_uapi_defs;
 452
 453         err = ib_register_device(&dev->ibdev, "efa_%d", &pdev->dev);
 454         if (err)
 455                 goto err_destroy_eqs;
 456
 457         ibdev_info(&dev->ibdev, "IB device registered\n");
 458
 459         return 0;
 460
 461 err_destroy_eqs:
 462         efa_destroy_eqs(dev);
 463 err_release_doorbell_bar:
 464         efa_release_doorbell_bar(dev);
 465         return err;
 466 }
 467
 468 static void efa_ib_device_remove(struct efa_dev *dev)
 469 {
 470         ibdev_info(&dev->ibdev, "Unregister ib device\n");
 471         ib_unregister_device(&dev->ibdev);
 472         efa_destroy_eqs(dev);
 473         efa_com_dev_reset(&dev->edev, EFA_REGS_RESET_NORMAL);
 474         efa_release_doorbell_bar(dev);
 475 }
 476
 477 static void efa_disable_msix(struct efa_dev *dev)
 478 {
 479         pci_free_irq_vectors(dev->pdev);
 480 }
 481
 482 static int efa_enable_msix(struct efa_dev *dev)
 483 {
 484         int max_vecs, num_vecs;
 485
 486         /*
 487          * Reserve the max msix vectors we might need, one vector is reserved
 488          * for admin.
 489          */
 490         max_vecs = min_t(int, pci_msix_vec_count(dev->pdev),
 491                          num_online_cpus() + 1);
 492         dev_dbg(&dev->pdev->dev, "Trying to enable MSI-X, vectors %d\n",
 493                 max_vecs);
 494
 495         dev->admin_msix_vector_idx = EFA_MGMNT_MSIX_VEC_IDX;
 496         num_vecs = pci_alloc_irq_vectors(dev->pdev, 1,
 497                                          max_vecs, PCI_IRQ_MSIX);
 498
 499         if (num_vecs < 0) {
 500                 dev_err(&dev->pdev->dev, "Failed to enable MSI-X. error %d\n",
 501                         num_vecs);
 502                 return -ENOSPC;
 503         }
 504
 505         dev_dbg(&dev->pdev->dev, "Allocated %d MSI-X vectors\n", num_vecs);
 506
 507         dev->num_irq_vectors = num_vecs;
 508
 509         return 0;
 510 }
 511
 512 static int efa_device_init(struct efa_com_dev *edev, struct pci_dev *pdev)
 513 {
 514         int dma_width;
 515         int err;
 516
 517         err = efa_com_dev_reset(edev, EFA_REGS_RESET_NORMAL);
 518         if (err)
 519                 return err;
 520
 521         err = efa_com_validate_version(edev);
 522         if (err)
 523                 return err;
 524
 525         dma_width = efa_com_get_dma_width(edev);
 526         if (dma_width < 0) {
 527                 err = dma_width;
 528                 return err;
 529         }
 530
 531         err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(dma_width));
 532         if (err) {
 533                 dev_err(&pdev->dev, "dma_set_mask_and_coherent failed %d\n", err);
 534                 return err;
 535         }
 536
 537         dma_set_max_seg_size(&pdev->dev, UINT_MAX);
 538         return 0;
 539 }
 540
 541 static struct efa_dev *efa_probe_device(struct pci_dev *pdev)
 542 {
 543         struct efa_com_dev *edev;
 544         struct efa_dev *dev;
 545         int pci_mem_bars;
 546         int err;
 547
 548         err = pci_enable_device_mem(pdev);
 549         if (err) {
 550                 dev_err(&pdev->dev, "pci_enable_device_mem() failed!\n");
 551                 return ERR_PTR(err);
 552         }
 553
 554         pci_set_master(pdev);
 555
 556         dev = ib_alloc_device(efa_dev, ibdev);
 557         if (!dev) {
 558                 dev_err(&pdev->dev, "Device alloc failed\n");
 559                 err = -ENOMEM;
 560                 goto err_disable_device;
 561         }
 562
 563         pci_set_drvdata(pdev, dev);
 564         edev = &dev->edev;
 565         edev->efa_dev = dev;
 566         edev->dmadev = &pdev->dev;
 567         dev->pdev = pdev;
 568         xa_init(&dev->cqs_xa);
 569
 570         pci_mem_bars = pci_select_bars(pdev, IORESOURCE_MEM);
 571         if (EFA_BASE_BAR_MASK & ~pci_mem_bars) {
 572                 dev_err(&pdev->dev, "BARs unavailable. Requested %#x, available %#x\n",
 573                         (int)EFA_BASE_BAR_MASK, pci_mem_bars);
 574                 err = -ENODEV;
 575                 goto err_ibdev_destroy;
 576         }
 577         err = pci_request_selected_regions(pdev, EFA_BASE_BAR_MASK, DRV_MODULE_NAME);
 578         if (err) {
 579                 dev_err(&pdev->dev, "pci_request_selected_regions failed %d\n",
 580                         err);
 581                 goto err_ibdev_destroy;
 582         }
 583
 584         dev->reg_bar_addr = pci_resource_start(pdev, EFA_REG_BAR);
 585         dev->reg_bar_len = pci_resource_len(pdev, EFA_REG_BAR);
 586         dev->mem_bar_addr = pci_resource_start(pdev, EFA_MEM_BAR);
 587         dev->mem_bar_len = pci_resource_len(pdev, EFA_MEM_BAR);
 588
 589         edev->reg_bar = devm_ioremap(&pdev->dev,
 590                                      dev->reg_bar_addr,
 591                                      dev->reg_bar_len);
 592         if (!edev->reg_bar) {
 593                 dev_err(&pdev->dev, "Failed to remap register bar\n");
 594                 err = -EFAULT;
 595                 goto err_release_bars;
 596         }
 597
 598         err = efa_com_mmio_reg_read_init(edev);
 599         if (err) {
 600                 dev_err(&pdev->dev, "Failed to init readless MMIO\n");
 601                 goto err_iounmap;
 602         }
 603
 604         err = efa_device_init(edev, pdev);
 605         if (err) {
 606                 dev_err(&pdev->dev, "EFA device init failed\n");
 607                 if (err == -ETIME)
 608                         err = -EPROBE_DEFER;
 609                 goto err_reg_read_destroy;
 610         }
 611
 612         err = efa_enable_msix(dev);
 613         if (err)
 614                 goto err_reg_read_destroy;
 615
 616         edev->aq.msix_vector_idx = dev->admin_msix_vector_idx;
 617         edev->aenq.msix_vector_idx = dev->admin_msix_vector_idx;
 618
 619         err = efa_set_mgmnt_irq(dev);
 620         if (err)
 621                 goto err_disable_msix;
 622
 623         err = efa_com_admin_init(edev, &aenq_handlers);
 624         if (err)
 625                 goto err_free_mgmnt_irq;
 626
 627         return dev;
 628
 629 err_free_mgmnt_irq:
 630         efa_free_irq(dev, &dev->admin_irq);
 631 err_disable_msix:
 632         efa_disable_msix(dev);
 633 err_reg_read_destroy:
 634         efa_com_mmio_reg_read_destroy(edev);
 635 err_iounmap:
 636         devm_iounmap(&pdev->dev, edev->reg_bar);
 637 err_release_bars:
 638         efa_release_bars(dev, EFA_BASE_BAR_MASK);
 639 err_ibdev_destroy:
 640         ib_dealloc_device(&dev->ibdev);
 641 err_disable_device:
 642         pci_disable_device(pdev);
 643         return ERR_PTR(err);
 644 }
 645
 646 static void efa_remove_device(struct pci_dev *pdev)
 647 {
 648         struct efa_dev *dev = pci_get_drvdata(pdev);
 649         struct efa_com_dev *edev;
 650
 651         edev = &dev->edev;
 652         efa_com_admin_destroy(edev);
 653         efa_free_irq(dev, &dev->admin_irq);
 654         efa_disable_msix(dev);
 655         efa_com_mmio_reg_read_destroy(edev);
 656         devm_iounmap(&pdev->dev, edev->reg_bar);
 657         efa_release_bars(dev, EFA_BASE_BAR_MASK);
 658         xa_destroy(&dev->cqs_xa);
 659         ib_dealloc_device(&dev->ibdev);
 660         pci_disable_device(pdev);
 661 }
 662
 663 static int efa_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
 664 {
 665         struct efa_dev *dev;
 666         int err;
 667
 668         dev = efa_probe_device(pdev);
 669         if (IS_ERR(dev))
 670                 return PTR_ERR(dev);
 671
 672         err = efa_ib_device_add(dev);
 673         if (err)
 674                 goto err_remove_device;
 675
 676         return 0;
 677
 678 err_remove_device:
 679         efa_remove_device(pdev);
 680         return err;
 681 }
 682
 683 static void efa_remove(struct pci_dev *pdev)
 684 {
 685         struct efa_dev *dev = pci_get_drvdata(pdev);
 686
 687         efa_ib_device_remove(dev);
 688         efa_remove_device(pdev);
 689 }
 690
 691 static void efa_shutdown(struct pci_dev *pdev)
 692 {
 693         struct efa_dev *dev = pci_get_drvdata(pdev);
 694
 695         efa_destroy_eqs(dev);
 696         efa_com_dev_reset(&dev->edev, EFA_REGS_RESET_SHUTDOWN);
 697         efa_free_irq(dev, &dev->admin_irq);
 698         efa_disable_msix(dev);
 699 }
 700
 701 static struct pci_driver efa_pci_driver = {
 702         .name           = DRV_MODULE_NAME,
 703         .id_table       = efa_pci_tbl,
 704         .probe          = efa_probe,
 705         .remove         = efa_remove,
 706         .shutdown       = efa_shutdown,
 707 };
 708
 709 module_pci_driver(efa_pci_driver);