drivers/infiniband/hw/efa/efa_main.c

   1 // SPDX-License-Identifier: GPL-2.0 OR BSD-2-Clause
   2 /*
   3  * Copyright 2018-2019 Amazon.com, Inc. or its affiliates. All rights reserved.
   4  */
   5
   6 #include <linux/module.h>
   7 #include <linux/pci.h>
   8
   9 #include <rdma/ib_user_verbs.h>
  10
  11 #include "efa.h"
  12
  13 #define PCI_DEV_ID_EFA_VF 0xefa0
  14
  15 static const struct pci_device_id efa_pci_tbl[] = {
  16         { PCI_VDEVICE(AMAZON, PCI_DEV_ID_EFA_VF) },
  17         { }
  18 };
  19
  20 MODULE_AUTHOR("Amazon.com, Inc. or its affiliates");
  21 MODULE_LICENSE("Dual BSD/GPL");
  22 MODULE_DESCRIPTION(DEVICE_NAME);
  23 MODULE_DEVICE_TABLE(pci, efa_pci_tbl);
  24
  25 #define EFA_REG_BAR 0
  26 #define EFA_MEM_BAR 2
  27 #define EFA_BASE_BAR_MASK (BIT(EFA_REG_BAR) | BIT(EFA_MEM_BAR))
  28
  29 #define EFA_AENQ_ENABLED_GROUPS \
  30         (BIT(EFA_ADMIN_FATAL_ERROR) | BIT(EFA_ADMIN_WARNING) | \
  31          BIT(EFA_ADMIN_NOTIFICATION) | BIT(EFA_ADMIN_KEEP_ALIVE))
  32
  33 /* This handler will called for unknown event group or unimplemented handlers */
  34 static void unimplemented_aenq_handler(void *data,
  35                                        struct efa_admin_aenq_entry *aenq_e)
  36 {
  37         struct efa_dev *dev = (struct efa_dev *)data;
  38
  39         ibdev_err(&dev->ibdev,
  40                   "Unknown event was received or event with unimplemented handler\n");
  41 }
  42
  43 static void efa_keep_alive(void *data, struct efa_admin_aenq_entry *aenq_e)
  44 {
  45         struct efa_dev *dev = (struct efa_dev *)data;
  46
  47         atomic64_inc(&dev->stats.keep_alive_rcvd);
  48 }
  49
  50 static struct efa_aenq_handlers aenq_handlers = {
  51         .handlers = {
  52                 [EFA_ADMIN_KEEP_ALIVE] = efa_keep_alive,
  53         },
  54         .unimplemented_handler = unimplemented_aenq_handler
  55 };
  56
  57 static void efa_release_bars(struct efa_dev *dev, int bars_mask)
  58 {
  59         struct pci_dev *pdev = dev->pdev;
  60         int release_bars;
  61
  62         release_bars = pci_select_bars(pdev, IORESOURCE_MEM) & bars_mask;
  63         pci_release_selected_regions(pdev, release_bars);
  64 }
  65
  66 static irqreturn_t efa_intr_msix_mgmnt(int irq, void *data)
  67 {
  68         struct efa_dev *dev = data;
  69
  70         efa_com_admin_q_comp_intr_handler(&dev->edev);
  71         efa_com_aenq_intr_handler(&dev->edev, data);
  72
  73         return IRQ_HANDLED;
  74 }
  75
  76 static int efa_request_mgmnt_irq(struct efa_dev *dev)
  77 {
  78         struct efa_irq *irq;
  79         int err;
  80
  81         irq = &dev->admin_irq;
  82         err = request_irq(irq->vector, irq->handler, 0, irq->name,
  83                           irq->data);
  84         if (err) {
  85                 dev_err(&dev->pdev->dev, "Failed to request admin irq (%d)\n",
  86                         err);
  87                 return err;
  88         }
  89
  90         dev_dbg(&dev->pdev->dev, "Set affinity hint of mgmnt irq to %*pbl (irq vector: %d)\n",
  91                 nr_cpumask_bits, &irq->affinity_hint_mask, irq->vector);
  92         irq_set_affinity_hint(irq->vector, &irq->affinity_hint_mask);
  93
  94         return 0;
  95 }
  96
  97 static void efa_setup_mgmnt_irq(struct efa_dev *dev)
  98 {
  99         u32 cpu;
 100
 101         snprintf(dev->admin_irq.name, EFA_IRQNAME_SIZE,
 102                  "efa-mgmnt@pci:%s", pci_name(dev->pdev));
 103         dev->admin_irq.handler = efa_intr_msix_mgmnt;
 104         dev->admin_irq.data = dev;
 105         dev->admin_irq.vector =
 106                 pci_irq_vector(dev->pdev, dev->admin_msix_vector_idx);
 107         cpu = cpumask_first(cpu_online_mask);
 108         dev->admin_irq.cpu = cpu;
 109         cpumask_set_cpu(cpu,
 110                         &dev->admin_irq.affinity_hint_mask);
 111         dev_info(&dev->pdev->dev, "Setup irq:0x%p vector:%d name:%s\n",
 112                  &dev->admin_irq,
 113                  dev->admin_irq.vector,
 114                  dev->admin_irq.name);
 115 }
 116
 117 static void efa_free_mgmnt_irq(struct efa_dev *dev)
 118 {
 119         struct efa_irq *irq;
 120
 121         irq = &dev->admin_irq;
 122         irq_set_affinity_hint(irq->vector, NULL);
 123         free_irq(irq->vector, irq->data);
 124 }
 125
 126 static int efa_set_mgmnt_irq(struct efa_dev *dev)
 127 {
 128         efa_setup_mgmnt_irq(dev);
 129
 130         return efa_request_mgmnt_irq(dev);
 131 }
 132
 133 static int efa_request_doorbell_bar(struct efa_dev *dev)
 134 {
 135         u8 db_bar_idx = dev->dev_attr.db_bar;
 136         struct pci_dev *pdev = dev->pdev;
 137         int bars;
 138         int err;
 139
 140         if (!(BIT(db_bar_idx) & EFA_BASE_BAR_MASK)) {
 141                 bars = pci_select_bars(pdev, IORESOURCE_MEM) & BIT(db_bar_idx);
 142
 143                 err = pci_request_selected_regions(pdev, bars, DRV_MODULE_NAME);
 144                 if (err) {
 145                         dev_err(&dev->pdev->dev,
 146                                 "pci_request_selected_regions for bar %d failed %d\n",
 147                                 db_bar_idx, err);
 148                         return err;
 149                 }
 150         }
 151
 152         dev->db_bar_addr = pci_resource_start(dev->pdev, db_bar_idx);
 153         dev->db_bar_len = pci_resource_len(dev->pdev, db_bar_idx);
 154
 155         return 0;
 156 }
 157
 158 static void efa_release_doorbell_bar(struct efa_dev *dev)
 159 {
 160         if (!(BIT(dev->dev_attr.db_bar) & EFA_BASE_BAR_MASK))
 161                 efa_release_bars(dev, BIT(dev->dev_attr.db_bar));
 162 }
 163
 164 static void efa_update_hw_hints(struct efa_dev *dev,
 165                                 struct efa_com_get_hw_hints_result *hw_hints)
 166 {
 167         struct efa_com_dev *edev = &dev->edev;
 168
 169         if (hw_hints->mmio_read_timeout)
 170                 edev->mmio_read.mmio_read_timeout =
 171                         hw_hints->mmio_read_timeout * 1000;
 172
 173         if (hw_hints->poll_interval)
 174                 edev->aq.poll_interval = hw_hints->poll_interval;
 175
 176         if (hw_hints->admin_completion_timeout)
 177                 edev->aq.completion_timeout =
 178                         hw_hints->admin_completion_timeout;
 179 }
 180
 181 static void efa_stats_init(struct efa_dev *dev)
 182 {
 183         atomic64_t *s = (atomic64_t *)&dev->stats;
 184         int i;
 185
 186         for (i = 0; i < sizeof(dev->stats) / sizeof(*s); i++, s++)
 187                 atomic64_set(s, 0);
 188 }
 189
 190 static const struct ib_device_ops efa_dev_ops = {
 191         .owner = THIS_MODULE,
 192         .driver_id = RDMA_DRIVER_EFA,
 193         .uverbs_abi_ver = EFA_UVERBS_ABI_VERSION,
 194
 195         .alloc_hw_stats = efa_alloc_hw_stats,
 196         .alloc_pd = efa_alloc_pd,
 197         .alloc_ucontext = efa_alloc_ucontext,
 198         .create_ah = efa_create_ah,
 199         .create_cq = efa_create_cq,
 200         .create_qp = efa_create_qp,
 201         .dealloc_pd = efa_dealloc_pd,
 202         .dealloc_ucontext = efa_dealloc_ucontext,
 203         .dereg_mr = efa_dereg_mr,
 204         .destroy_ah = efa_destroy_ah,
 205         .destroy_cq = efa_destroy_cq,
 206         .destroy_qp = efa_destroy_qp,
 207         .get_hw_stats = efa_get_hw_stats,
 208         .get_link_layer = efa_port_link_layer,
 209         .get_port_immutable = efa_get_port_immutable,
 210         .mmap = efa_mmap,
 211         .mmap_free = efa_mmap_free,
 212         .modify_qp = efa_modify_qp,
 213         .query_device = efa_query_device,
 214         .query_gid = efa_query_gid,
 215         .query_pkey = efa_query_pkey,
 216         .query_port = efa_query_port,
 217         .query_qp = efa_query_qp,
 218         .reg_user_mr = efa_reg_mr,
 219
 220         INIT_RDMA_OBJ_SIZE(ib_ah, efa_ah, ibah),
 221         INIT_RDMA_OBJ_SIZE(ib_cq, efa_cq, ibcq),
 222         INIT_RDMA_OBJ_SIZE(ib_pd, efa_pd, ibpd),
 223         INIT_RDMA_OBJ_SIZE(ib_ucontext, efa_ucontext, ibucontext),
 224 };
 225
 226 static int efa_ib_device_add(struct efa_dev *dev)
 227 {
 228         struct efa_com_get_hw_hints_result hw_hints;
 229         struct pci_dev *pdev = dev->pdev;
 230         int err;
 231
 232         efa_stats_init(dev);
 233
 234         err = efa_com_get_device_attr(&dev->edev, &dev->dev_attr);
 235         if (err)
 236                 return err;
 237
 238         dev_dbg(&dev->pdev->dev, "Doorbells bar (%d)\n", dev->dev_attr.db_bar);
 239         err = efa_request_doorbell_bar(dev);
 240         if (err)
 241                 return err;
 242
 243         err = efa_com_get_hw_hints(&dev->edev, &hw_hints);
 244         if (err)
 245                 goto err_release_doorbell_bar;
 246
 247         efa_update_hw_hints(dev, &hw_hints);
 248
 249         /* Try to enable all the available aenq groups */
 250         err = efa_com_set_aenq_config(&dev->edev, EFA_AENQ_ENABLED_GROUPS);
 251         if (err)
 252                 goto err_release_doorbell_bar;
 253
 254         dev->ibdev.node_type = RDMA_NODE_UNSPECIFIED;
 255         dev->ibdev.phys_port_cnt = 1;
 256         dev->ibdev.num_comp_vectors = 1;
 257         dev->ibdev.dev.parent = &pdev->dev;
 258
 259         dev->ibdev.uverbs_cmd_mask =
 260                 (1ull << IB_USER_VERBS_CMD_GET_CONTEXT) |
 261                 (1ull << IB_USER_VERBS_CMD_QUERY_DEVICE) |
 262                 (1ull << IB_USER_VERBS_CMD_QUERY_PORT) |
 263                 (1ull << IB_USER_VERBS_CMD_ALLOC_PD) |
 264                 (1ull << IB_USER_VERBS_CMD_DEALLOC_PD) |
 265                 (1ull << IB_USER_VERBS_CMD_REG_MR) |
 266                 (1ull << IB_USER_VERBS_CMD_DEREG_MR) |
 267                 (1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL) |
 268                 (1ull << IB_USER_VERBS_CMD_CREATE_CQ) |
 269                 (1ull << IB_USER_VERBS_CMD_DESTROY_CQ) |
 270                 (1ull << IB_USER_VERBS_CMD_CREATE_QP) |
 271                 (1ull << IB_USER_VERBS_CMD_MODIFY_QP) |
 272                 (1ull << IB_USER_VERBS_CMD_QUERY_QP) |
 273                 (1ull << IB_USER_VERBS_CMD_DESTROY_QP) |
 274                 (1ull << IB_USER_VERBS_CMD_CREATE_AH) |
 275                 (1ull << IB_USER_VERBS_CMD_DESTROY_AH);
 276
 277         dev->ibdev.uverbs_ex_cmd_mask =
 278                 (1ull << IB_USER_VERBS_EX_CMD_QUERY_DEVICE);
 279
 280         ib_set_device_ops(&dev->ibdev, &efa_dev_ops);
 281
 282         err = ib_register_device(&dev->ibdev, "efa_%d");
 283         if (err)
 284                 goto err_release_doorbell_bar;
 285
 286         ibdev_info(&dev->ibdev, "IB device registered\n");
 287
 288         return 0;
 289
 290 err_release_doorbell_bar:
 291         efa_release_doorbell_bar(dev);
 292         return err;
 293 }
 294
 295 static void efa_ib_device_remove(struct efa_dev *dev)
 296 {
 297         efa_com_dev_reset(&dev->edev, EFA_REGS_RESET_NORMAL);
 298         ibdev_info(&dev->ibdev, "Unregister ib device\n");
 299         ib_unregister_device(&dev->ibdev);
 300         efa_release_doorbell_bar(dev);
 301 }
 302
 303 static void efa_disable_msix(struct efa_dev *dev)
 304 {
 305         pci_free_irq_vectors(dev->pdev);
 306 }
 307
 308 static int efa_enable_msix(struct efa_dev *dev)
 309 {
 310         int msix_vecs, irq_num;
 311
 312         /* Reserve the max msix vectors we might need */
 313         msix_vecs = EFA_NUM_MSIX_VEC;
 314         dev_dbg(&dev->pdev->dev, "Trying to enable MSI-X, vectors %d\n",
 315                 msix_vecs);
 316
 317         dev->admin_msix_vector_idx = EFA_MGMNT_MSIX_VEC_IDX;
 318         irq_num = pci_alloc_irq_vectors(dev->pdev, msix_vecs,
 319                                         msix_vecs, PCI_IRQ_MSIX);
 320
 321         if (irq_num < 0) {
 322                 dev_err(&dev->pdev->dev, "Failed to enable MSI-X. irq_num %d\n",
 323                         irq_num);
 324                 return -ENOSPC;
 325         }
 326
 327         if (irq_num != msix_vecs) {
 328                 dev_err(&dev->pdev->dev,
 329                         "Allocated %d MSI-X (out of %d requested)\n",
 330                         irq_num, msix_vecs);
 331                 return -ENOSPC;
 332         }
 333
 334         return 0;
 335 }
 336
 337 static int efa_device_init(struct efa_com_dev *edev, struct pci_dev *pdev)
 338 {
 339         int dma_width;
 340         int err;
 341
 342         err = efa_com_dev_reset(edev, EFA_REGS_RESET_NORMAL);
 343         if (err)
 344                 return err;
 345
 346         err = efa_com_validate_version(edev);
 347         if (err)
 348                 return err;
 349
 350         dma_width = efa_com_get_dma_width(edev);
 351         if (dma_width < 0) {
 352                 err = dma_width;
 353                 return err;
 354         }
 355
 356         err = pci_set_dma_mask(pdev, DMA_BIT_MASK(dma_width));
 357         if (err) {
 358                 dev_err(&pdev->dev, "pci_set_dma_mask failed %d\n", err);
 359                 return err;
 360         }
 361
 362         err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(dma_width));
 363         if (err) {
 364                 dev_err(&pdev->dev,
 365                         "err_pci_set_consistent_dma_mask failed %d\n",
 366                         err);
 367                 return err;
 368         }
 369
 370         return 0;
 371 }
 372
 373 static struct efa_dev *efa_probe_device(struct pci_dev *pdev)
 374 {
 375         struct efa_com_dev *edev;
 376         struct efa_dev *dev;
 377         int bars;
 378         int err;
 379
 380         err = pci_enable_device_mem(pdev);
 381         if (err) {
 382                 dev_err(&pdev->dev, "pci_enable_device_mem() failed!\n");
 383                 return ERR_PTR(err);
 384         }
 385
 386         pci_set_master(pdev);
 387
 388         dev = ib_alloc_device(efa_dev, ibdev);
 389         if (!dev) {
 390                 dev_err(&pdev->dev, "Device alloc failed\n");
 391                 err = -ENOMEM;
 392                 goto err_disable_device;
 393         }
 394
 395         pci_set_drvdata(pdev, dev);
 396         edev = &dev->edev;
 397         edev->efa_dev = dev;
 398         edev->dmadev = &pdev->dev;
 399         dev->pdev = pdev;
 400
 401         bars = pci_select_bars(pdev, IORESOURCE_MEM) & EFA_BASE_BAR_MASK;
 402         err = pci_request_selected_regions(pdev, bars, DRV_MODULE_NAME);
 403         if (err) {
 404                 dev_err(&pdev->dev, "pci_request_selected_regions failed %d\n",
 405                         err);
 406                 goto err_ibdev_destroy;
 407         }
 408
 409         dev->reg_bar_addr = pci_resource_start(pdev, EFA_REG_BAR);
 410         dev->reg_bar_len = pci_resource_len(pdev, EFA_REG_BAR);
 411         dev->mem_bar_addr = pci_resource_start(pdev, EFA_MEM_BAR);
 412         dev->mem_bar_len = pci_resource_len(pdev, EFA_MEM_BAR);
 413
 414         edev->reg_bar = devm_ioremap(&pdev->dev,
 415                                      dev->reg_bar_addr,
 416                                      dev->reg_bar_len);
 417         if (!edev->reg_bar) {
 418                 dev_err(&pdev->dev, "Failed to remap register bar\n");
 419                 err = -EFAULT;
 420                 goto err_release_bars;
 421         }
 422
 423         err = efa_com_mmio_reg_read_init(edev);
 424         if (err) {
 425                 dev_err(&pdev->dev, "Failed to init readless MMIO\n");
 426                 goto err_iounmap;
 427         }
 428
 429         err = efa_device_init(edev, pdev);
 430         if (err) {
 431                 dev_err(&pdev->dev, "EFA device init failed\n");
 432                 if (err == -ETIME)
 433                         err = -EPROBE_DEFER;
 434                 goto err_reg_read_destroy;
 435         }
 436
 437         err = efa_enable_msix(dev);
 438         if (err)
 439                 goto err_reg_read_destroy;
 440
 441         edev->aq.msix_vector_idx = dev->admin_msix_vector_idx;
 442         edev->aenq.msix_vector_idx = dev->admin_msix_vector_idx;
 443
 444         err = efa_set_mgmnt_irq(dev);
 445         if (err)
 446                 goto err_disable_msix;
 447
 448         err = efa_com_admin_init(edev, &aenq_handlers);
 449         if (err)
 450                 goto err_free_mgmnt_irq;
 451
 452         return dev;
 453
 454 err_free_mgmnt_irq:
 455         efa_free_mgmnt_irq(dev);
 456 err_disable_msix:
 457         efa_disable_msix(dev);
 458 err_reg_read_destroy:
 459         efa_com_mmio_reg_read_destroy(edev);
 460 err_iounmap:
 461         devm_iounmap(&pdev->dev, edev->reg_bar);
 462 err_release_bars:
 463         efa_release_bars(dev, EFA_BASE_BAR_MASK);
 464 err_ibdev_destroy:
 465         ib_dealloc_device(&dev->ibdev);
 466 err_disable_device:
 467         pci_disable_device(pdev);
 468         return ERR_PTR(err);
 469 }
 470
 471 static void efa_remove_device(struct pci_dev *pdev)
 472 {
 473         struct efa_dev *dev = pci_get_drvdata(pdev);
 474         struct efa_com_dev *edev;
 475
 476         edev = &dev->edev;
 477         efa_com_admin_destroy(edev);
 478         efa_free_mgmnt_irq(dev);
 479         efa_disable_msix(dev);
 480         efa_com_mmio_reg_read_destroy(edev);
 481         devm_iounmap(&pdev->dev, edev->reg_bar);
 482         efa_release_bars(dev, EFA_BASE_BAR_MASK);
 483         ib_dealloc_device(&dev->ibdev);
 484         pci_disable_device(pdev);
 485 }
 486
 487 static int efa_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
 488 {
 489         struct efa_dev *dev;
 490         int err;
 491
 492         dev = efa_probe_device(pdev);
 493         if (IS_ERR(dev))
 494                 return PTR_ERR(dev);
 495
 496         err = efa_ib_device_add(dev);
 497         if (err)
 498                 goto err_remove_device;
 499
 500         return 0;
 501
 502 err_remove_device:
 503         efa_remove_device(pdev);
 504         return err;
 505 }
 506
 507 static void efa_remove(struct pci_dev *pdev)
 508 {
 509         struct efa_dev *dev = pci_get_drvdata(pdev);
 510
 511         efa_ib_device_remove(dev);
 512         efa_remove_device(pdev);
 513 }
 514
 515 static struct pci_driver efa_pci_driver = {
 516         .name           = DRV_MODULE_NAME,
 517         .id_table       = efa_pci_tbl,
 518         .probe          = efa_probe,
 519         .remove         = efa_remove,
 520 };
 521
 522 module_pci_driver(efa_pci_driver);