| 1 | // SPDX-License-Identifier: GPL-2.0 OR BSD-2-Clause |
| 2 | /* |
| 3 | * Copyright 2018-2025 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(ibdev: &dev->ibdev, |
| 51 | format: "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(v: &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(dev: 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, index: 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, eeq: &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(edev: &dev->edev); |
| 123 | efa_com_aenq_intr_handler(edev: &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: irq->irqn, handler: irq->handler, flags: 0, name: irq->name, dev: 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: irq->irqn, m: &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, u32 vector) |
| 145 | { |
| 146 | u32 cpu; |
| 147 | |
| 148 | cpu = vector - EFA_COMP_EQS_VEC_BASE; |
| 149 | snprintf(buf: eq->irq.name, EFA_IRQNAME_SIZE, fmt: "efa-comp%d@pci:%s" , cpu, |
| 150 | pci_name(pdev: dev->pdev)); |
| 151 | eq->irq.handler = efa_intr_msix_comp; |
| 152 | eq->irq.data = eq; |
| 153 | eq->irq.vector = vector; |
| 154 | eq->irq.irqn = pci_irq_vector(dev: dev->pdev, nr: vector); |
| 155 | cpumask_set_cpu(cpu, dstp: &eq->irq.affinity_hint_mask); |
| 156 | } |
| 157 | |
| 158 | static void efa_free_irq(struct efa_dev *dev, struct efa_irq *irq) |
| 159 | { |
| 160 | irq_set_affinity_hint(irq: irq->irqn, NULL); |
| 161 | free_irq(irq->irqn, irq->data); |
| 162 | } |
| 163 | |
| 164 | static void efa_setup_mgmnt_irq(struct efa_dev *dev) |
| 165 | { |
| 166 | u32 cpu; |
| 167 | |
| 168 | snprintf(buf: dev->admin_irq.name, EFA_IRQNAME_SIZE, |
| 169 | fmt: "efa-mgmnt@pci:%s" , pci_name(pdev: dev->pdev)); |
| 170 | dev->admin_irq.handler = efa_intr_msix_mgmnt; |
| 171 | dev->admin_irq.data = dev; |
| 172 | dev->admin_irq.vector = dev->admin_msix_vector_idx; |
| 173 | dev->admin_irq.irqn = pci_irq_vector(dev: dev->pdev, |
| 174 | nr: dev->admin_msix_vector_idx); |
| 175 | cpu = cpumask_first(cpu_online_mask); |
| 176 | cpumask_set_cpu(cpu, |
| 177 | dstp: &dev->admin_irq.affinity_hint_mask); |
| 178 | dev_info(&dev->pdev->dev, "Setup irq:%d name:%s\n" , |
| 179 | dev->admin_irq.irqn, |
| 180 | dev->admin_irq.name); |
| 181 | } |
| 182 | |
| 183 | static int efa_set_mgmnt_irq(struct efa_dev *dev) |
| 184 | { |
| 185 | efa_setup_mgmnt_irq(dev); |
| 186 | |
| 187 | return efa_request_irq(dev, irq: &dev->admin_irq); |
| 188 | } |
| 189 | |
| 190 | static int efa_request_doorbell_bar(struct efa_dev *dev) |
| 191 | { |
| 192 | u8 db_bar_idx = dev->dev_attr.db_bar; |
| 193 | struct pci_dev *pdev = dev->pdev; |
| 194 | int pci_mem_bars; |
| 195 | int db_bar; |
| 196 | int err; |
| 197 | |
| 198 | db_bar = BIT(db_bar_idx); |
| 199 | if (!(db_bar & EFA_BASE_BAR_MASK)) { |
| 200 | pci_mem_bars = pci_select_bars(dev: pdev, IORESOURCE_MEM); |
| 201 | if (db_bar & ~pci_mem_bars) { |
| 202 | dev_err(&pdev->dev, |
| 203 | "Doorbells BAR unavailable. Requested %#x, available %#x\n" , |
| 204 | db_bar, pci_mem_bars); |
| 205 | return -ENODEV; |
| 206 | } |
| 207 | |
| 208 | err = pci_request_selected_regions(pdev, db_bar, DRV_MODULE_NAME); |
| 209 | if (err) { |
| 210 | dev_err(&pdev->dev, |
| 211 | "pci_request_selected_regions for bar %d failed %d\n" , |
| 212 | db_bar_idx, err); |
| 213 | return err; |
| 214 | } |
| 215 | } |
| 216 | |
| 217 | dev->db_bar_addr = pci_resource_start(dev->pdev, db_bar_idx); |
| 218 | dev->db_bar_len = pci_resource_len(dev->pdev, db_bar_idx); |
| 219 | |
| 220 | return 0; |
| 221 | } |
| 222 | |
| 223 | static void efa_release_doorbell_bar(struct efa_dev *dev) |
| 224 | { |
| 225 | if (!(BIT(dev->dev_attr.db_bar) & EFA_BASE_BAR_MASK)) |
| 226 | efa_release_bars(dev, BIT(dev->dev_attr.db_bar)); |
| 227 | } |
| 228 | |
| 229 | static void efa_update_hw_hints(struct efa_dev *dev, |
| 230 | struct efa_com_get_hw_hints_result *hw_hints) |
| 231 | { |
| 232 | struct efa_com_dev *edev = &dev->edev; |
| 233 | |
| 234 | if (hw_hints->mmio_read_timeout) |
| 235 | edev->mmio_read.mmio_read_timeout = |
| 236 | hw_hints->mmio_read_timeout * 1000; |
| 237 | |
| 238 | if (hw_hints->poll_interval) |
| 239 | edev->aq.poll_interval = hw_hints->poll_interval; |
| 240 | |
| 241 | if (hw_hints->admin_completion_timeout) |
| 242 | edev->aq.completion_timeout = |
| 243 | hw_hints->admin_completion_timeout; |
| 244 | } |
| 245 | |
| 246 | static void efa_stats_init(struct efa_dev *dev) |
| 247 | { |
| 248 | atomic64_t *s = (atomic64_t *)&dev->stats; |
| 249 | int i; |
| 250 | |
| 251 | for (i = 0; i < sizeof(dev->stats) / sizeof(*s); i++, s++) |
| 252 | atomic64_set(v: s, i: 0); |
| 253 | } |
| 254 | |
| 255 | static void efa_set_host_info(struct efa_dev *dev) |
| 256 | { |
| 257 | struct efa_admin_set_feature_resp resp = {}; |
| 258 | struct efa_admin_set_feature_cmd cmd = {}; |
| 259 | struct efa_admin_host_info *hinf; |
| 260 | u32 bufsz = sizeof(*hinf); |
| 261 | dma_addr_t hinf_dma; |
| 262 | |
| 263 | if (!efa_com_check_supported_feature_id(edev: &dev->edev, |
| 264 | feature_id: EFA_ADMIN_HOST_INFO)) |
| 265 | return; |
| 266 | |
| 267 | /* Failures in host info set shall not disturb probe */ |
| 268 | hinf = dma_alloc_coherent(dev: &dev->pdev->dev, size: bufsz, dma_handle: &hinf_dma, |
| 269 | GFP_KERNEL); |
| 270 | if (!hinf) |
| 271 | return; |
| 272 | |
| 273 | strscpy(hinf->os_dist_str, utsname()->release, |
| 274 | sizeof(hinf->os_dist_str)); |
| 275 | hinf->os_type = EFA_ADMIN_OS_LINUX; |
| 276 | strscpy(hinf->kernel_ver_str, utsname()->version, |
| 277 | sizeof(hinf->kernel_ver_str)); |
| 278 | hinf->kernel_ver = LINUX_VERSION_CODE; |
| 279 | EFA_SET(&hinf->driver_ver, EFA_ADMIN_HOST_INFO_DRIVER_MAJOR, 0); |
| 280 | EFA_SET(&hinf->driver_ver, EFA_ADMIN_HOST_INFO_DRIVER_MINOR, 0); |
| 281 | EFA_SET(&hinf->driver_ver, EFA_ADMIN_HOST_INFO_DRIVER_SUB_MINOR, 0); |
| 282 | EFA_SET(&hinf->driver_ver, EFA_ADMIN_HOST_INFO_DRIVER_MODULE_TYPE, 0); |
| 283 | EFA_SET(&hinf->bdf, EFA_ADMIN_HOST_INFO_BUS, dev->pdev->bus->number); |
| 284 | EFA_SET(&hinf->bdf, EFA_ADMIN_HOST_INFO_DEVICE, |
| 285 | PCI_SLOT(dev->pdev->devfn)); |
| 286 | EFA_SET(&hinf->bdf, EFA_ADMIN_HOST_INFO_FUNCTION, |
| 287 | PCI_FUNC(dev->pdev->devfn)); |
| 288 | EFA_SET(&hinf->spec_ver, EFA_ADMIN_HOST_INFO_SPEC_MAJOR, |
| 289 | EFA_COMMON_SPEC_VERSION_MAJOR); |
| 290 | EFA_SET(&hinf->spec_ver, EFA_ADMIN_HOST_INFO_SPEC_MINOR, |
| 291 | EFA_COMMON_SPEC_VERSION_MINOR); |
| 292 | EFA_SET(&hinf->flags, EFA_ADMIN_HOST_INFO_INTREE, 1); |
| 293 | EFA_SET(&hinf->flags, EFA_ADMIN_HOST_INFO_GDR, 0); |
| 294 | |
| 295 | efa_com_set_feature_ex(edev: &dev->edev, set_resp: &resp, set_cmd: &cmd, feature_id: EFA_ADMIN_HOST_INFO, |
| 296 | control_buf_dma_addr: hinf_dma, control_buff_size: bufsz); |
| 297 | |
| 298 | dma_free_coherent(dev: &dev->pdev->dev, size: bufsz, cpu_addr: hinf, dma_handle: hinf_dma); |
| 299 | } |
| 300 | |
| 301 | static void efa_destroy_eq(struct efa_dev *dev, struct efa_eq *eq) |
| 302 | { |
| 303 | efa_com_eq_destroy(edev: &dev->edev, eeq: &eq->eeq); |
| 304 | efa_free_irq(dev, irq: &eq->irq); |
| 305 | } |
| 306 | |
| 307 | static int efa_create_eq(struct efa_dev *dev, struct efa_eq *eq, u32 msix_vec) |
| 308 | { |
| 309 | int err; |
| 310 | |
| 311 | efa_setup_comp_irq(dev, eq, vector: msix_vec); |
| 312 | err = efa_request_irq(dev, irq: &eq->irq); |
| 313 | if (err) |
| 314 | return err; |
| 315 | |
| 316 | err = efa_com_eq_init(edev: &dev->edev, eeq: &eq->eeq, cb: efa_process_eqe, |
| 317 | depth: dev->dev_attr.max_eq_depth, msix_vec); |
| 318 | if (err) |
| 319 | goto err_free_comp_irq; |
| 320 | |
| 321 | return 0; |
| 322 | |
| 323 | err_free_comp_irq: |
| 324 | efa_free_irq(dev, irq: &eq->irq); |
| 325 | return err; |
| 326 | } |
| 327 | |
| 328 | static int efa_create_eqs(struct efa_dev *dev) |
| 329 | { |
| 330 | u32 neqs = dev->dev_attr.max_eq; |
| 331 | int err, i; |
| 332 | |
| 333 | neqs = min_t(u32, neqs, dev->num_irq_vectors - EFA_COMP_EQS_VEC_BASE); |
| 334 | dev->neqs = neqs; |
| 335 | dev->eqs = kcalloc(neqs, sizeof(*dev->eqs), GFP_KERNEL); |
| 336 | if (!dev->eqs) |
| 337 | return -ENOMEM; |
| 338 | |
| 339 | for (i = 0; i < neqs; i++) { |
| 340 | err = efa_create_eq(dev, eq: &dev->eqs[i], msix_vec: i + EFA_COMP_EQS_VEC_BASE); |
| 341 | if (err) |
| 342 | goto err_destroy_eqs; |
| 343 | } |
| 344 | |
| 345 | return 0; |
| 346 | |
| 347 | err_destroy_eqs: |
| 348 | for (i--; i >= 0; i--) |
| 349 | efa_destroy_eq(dev, eq: &dev->eqs[i]); |
| 350 | kfree(objp: dev->eqs); |
| 351 | |
| 352 | return err; |
| 353 | } |
| 354 | |
| 355 | static void efa_destroy_eqs(struct efa_dev *dev) |
| 356 | { |
| 357 | int i; |
| 358 | |
| 359 | for (i = 0; i < dev->neqs; i++) |
| 360 | efa_destroy_eq(dev, eq: &dev->eqs[i]); |
| 361 | |
| 362 | kfree(objp: dev->eqs); |
| 363 | } |
| 364 | |
| 365 | static const struct ib_device_ops efa_dev_ops = { |
| 366 | .owner = THIS_MODULE, |
| 367 | .driver_id = RDMA_DRIVER_EFA, |
| 368 | .uverbs_abi_ver = EFA_UVERBS_ABI_VERSION, |
| 369 | |
| 370 | .alloc_hw_port_stats = efa_alloc_hw_port_stats, |
| 371 | .alloc_hw_device_stats = efa_alloc_hw_device_stats, |
| 372 | .alloc_pd = efa_alloc_pd, |
| 373 | .alloc_ucontext = efa_alloc_ucontext, |
| 374 | .create_cq = efa_create_cq, |
| 375 | .create_cq_umem = efa_create_cq_umem, |
| 376 | .create_qp = efa_create_qp, |
| 377 | .create_user_ah = efa_create_ah, |
| 378 | .dealloc_pd = efa_dealloc_pd, |
| 379 | .dealloc_ucontext = efa_dealloc_ucontext, |
| 380 | .dereg_mr = efa_dereg_mr, |
| 381 | .destroy_ah = efa_destroy_ah, |
| 382 | .destroy_cq = efa_destroy_cq, |
| 383 | .destroy_qp = efa_destroy_qp, |
| 384 | .get_hw_stats = efa_get_hw_stats, |
| 385 | .get_link_layer = efa_port_link_layer, |
| 386 | .get_port_immutable = efa_get_port_immutable, |
| 387 | .mmap = efa_mmap, |
| 388 | .mmap_free = efa_mmap_free, |
| 389 | .modify_qp = efa_modify_qp, |
| 390 | .query_device = efa_query_device, |
| 391 | .query_gid = efa_query_gid, |
| 392 | .query_pkey = efa_query_pkey, |
| 393 | .query_port = efa_query_port, |
| 394 | .query_qp = efa_query_qp, |
| 395 | .reg_user_mr = efa_reg_mr, |
| 396 | .reg_user_mr_dmabuf = efa_reg_user_mr_dmabuf, |
| 397 | |
| 398 | INIT_RDMA_OBJ_SIZE(ib_ah, efa_ah, ibah), |
| 399 | INIT_RDMA_OBJ_SIZE(ib_cq, efa_cq, ibcq), |
| 400 | INIT_RDMA_OBJ_SIZE(ib_pd, efa_pd, ibpd), |
| 401 | INIT_RDMA_OBJ_SIZE(ib_qp, efa_qp, ibqp), |
| 402 | INIT_RDMA_OBJ_SIZE(ib_ucontext, efa_ucontext, ibucontext), |
| 403 | }; |
| 404 | |
| 405 | static int efa_ib_device_add(struct efa_dev *dev) |
| 406 | { |
| 407 | struct efa_com_get_hw_hints_result hw_hints; |
| 408 | struct pci_dev *pdev = dev->pdev; |
| 409 | int err; |
| 410 | |
| 411 | efa_stats_init(dev); |
| 412 | |
| 413 | err = efa_com_get_device_attr(edev: &dev->edev, result: &dev->dev_attr); |
| 414 | if (err) |
| 415 | return err; |
| 416 | |
| 417 | dev_dbg(&dev->pdev->dev, "Doorbells bar (%d)\n" , dev->dev_attr.db_bar); |
| 418 | err = efa_request_doorbell_bar(dev); |
| 419 | if (err) |
| 420 | return err; |
| 421 | |
| 422 | err = efa_com_get_hw_hints(edev: &dev->edev, result: &hw_hints); |
| 423 | if (err) |
| 424 | goto err_release_doorbell_bar; |
| 425 | |
| 426 | efa_update_hw_hints(dev, hw_hints: &hw_hints); |
| 427 | |
| 428 | /* Try to enable all the available aenq groups */ |
| 429 | err = efa_com_set_aenq_config(edev: &dev->edev, EFA_AENQ_ENABLED_GROUPS); |
| 430 | if (err) |
| 431 | goto err_release_doorbell_bar; |
| 432 | |
| 433 | err = efa_create_eqs(dev); |
| 434 | if (err) |
| 435 | goto err_release_doorbell_bar; |
| 436 | |
| 437 | efa_set_host_info(dev); |
| 438 | |
| 439 | dev->ibdev.node_type = RDMA_NODE_UNSPECIFIED; |
| 440 | dev->ibdev.node_guid = dev->dev_attr.guid; |
| 441 | dev->ibdev.phys_port_cnt = 1; |
| 442 | dev->ibdev.num_comp_vectors = dev->neqs ?: 1; |
| 443 | dev->ibdev.dev.parent = &pdev->dev; |
| 444 | |
| 445 | ib_set_device_ops(device: &dev->ibdev, ops: &efa_dev_ops); |
| 446 | |
| 447 | dev->ibdev.driver_def = efa_uapi_defs; |
| 448 | |
| 449 | err = ib_register_device(device: &dev->ibdev, name: "efa_%d" , dma_device: &pdev->dev); |
| 450 | if (err) |
| 451 | goto err_destroy_eqs; |
| 452 | |
| 453 | ibdev_info(ibdev: &dev->ibdev, format: "IB device registered\n" ); |
| 454 | |
| 455 | return 0; |
| 456 | |
| 457 | err_destroy_eqs: |
| 458 | efa_destroy_eqs(dev); |
| 459 | err_release_doorbell_bar: |
| 460 | efa_release_doorbell_bar(dev); |
| 461 | return err; |
| 462 | } |
| 463 | |
| 464 | static void efa_ib_device_remove(struct efa_dev *dev) |
| 465 | { |
| 466 | ibdev_info(ibdev: &dev->ibdev, format: "Unregister ib device\n" ); |
| 467 | ib_unregister_device(device: &dev->ibdev); |
| 468 | efa_destroy_eqs(dev); |
| 469 | efa_release_doorbell_bar(dev); |
| 470 | } |
| 471 | |
| 472 | static void efa_disable_msix(struct efa_dev *dev) |
| 473 | { |
| 474 | pci_free_irq_vectors(dev: dev->pdev); |
| 475 | } |
| 476 | |
| 477 | static int efa_enable_msix(struct efa_dev *dev) |
| 478 | { |
| 479 | int max_vecs, num_vecs; |
| 480 | |
| 481 | /* |
| 482 | * Reserve the max msix vectors we might need, one vector is reserved |
| 483 | * for admin. |
| 484 | */ |
| 485 | max_vecs = min_t(int, pci_msix_vec_count(dev->pdev), |
| 486 | num_online_cpus() + 1); |
| 487 | dev_dbg(&dev->pdev->dev, "Trying to enable MSI-X, vectors %d\n" , |
| 488 | max_vecs); |
| 489 | |
| 490 | dev->admin_msix_vector_idx = EFA_MGMNT_MSIX_VEC_IDX; |
| 491 | num_vecs = pci_alloc_irq_vectors(dev: dev->pdev, min_vecs: 1, |
| 492 | max_vecs, PCI_IRQ_MSIX); |
| 493 | |
| 494 | if (num_vecs < 0) { |
| 495 | dev_err(&dev->pdev->dev, "Failed to enable MSI-X. error %d\n" , |
| 496 | num_vecs); |
| 497 | return -ENOSPC; |
| 498 | } |
| 499 | |
| 500 | dev_dbg(&dev->pdev->dev, "Allocated %d MSI-X vectors\n" , num_vecs); |
| 501 | |
| 502 | dev->num_irq_vectors = num_vecs; |
| 503 | |
| 504 | return 0; |
| 505 | } |
| 506 | |
| 507 | static int efa_device_init(struct efa_com_dev *edev, struct pci_dev *pdev) |
| 508 | { |
| 509 | int dma_width; |
| 510 | int err; |
| 511 | |
| 512 | err = efa_com_dev_reset(edev, reset_reason: EFA_REGS_RESET_NORMAL); |
| 513 | if (err) |
| 514 | return err; |
| 515 | |
| 516 | err = efa_com_validate_version(edev); |
| 517 | if (err) |
| 518 | return err; |
| 519 | |
| 520 | dma_width = efa_com_get_dma_width(edev); |
| 521 | if (dma_width < 0) { |
| 522 | err = dma_width; |
| 523 | return err; |
| 524 | } |
| 525 | |
| 526 | err = dma_set_mask_and_coherent(dev: &pdev->dev, DMA_BIT_MASK(dma_width)); |
| 527 | if (err) { |
| 528 | dev_err(&pdev->dev, "dma_set_mask_and_coherent failed %d\n" , err); |
| 529 | return err; |
| 530 | } |
| 531 | |
| 532 | dma_set_max_seg_size(dev: &pdev->dev, UINT_MAX); |
| 533 | return 0; |
| 534 | } |
| 535 | |
| 536 | static struct efa_dev *efa_probe_device(struct pci_dev *pdev) |
| 537 | { |
| 538 | struct efa_com_dev *edev; |
| 539 | struct efa_dev *dev; |
| 540 | int pci_mem_bars; |
| 541 | int err; |
| 542 | |
| 543 | err = pci_enable_device_mem(dev: pdev); |
| 544 | if (err) { |
| 545 | dev_err(&pdev->dev, "pci_enable_device_mem() failed!\n" ); |
| 546 | return ERR_PTR(error: err); |
| 547 | } |
| 548 | |
| 549 | pci_set_master(dev: pdev); |
| 550 | |
| 551 | dev = ib_alloc_device(efa_dev, ibdev); |
| 552 | if (!dev) { |
| 553 | dev_err(&pdev->dev, "Device alloc failed\n" ); |
| 554 | err = -ENOMEM; |
| 555 | goto err_disable_device; |
| 556 | } |
| 557 | |
| 558 | pci_set_drvdata(pdev, data: dev); |
| 559 | edev = &dev->edev; |
| 560 | edev->efa_dev = dev; |
| 561 | edev->dmadev = &pdev->dev; |
| 562 | dev->pdev = pdev; |
| 563 | xa_init(xa: &dev->cqs_xa); |
| 564 | |
| 565 | pci_mem_bars = pci_select_bars(dev: pdev, IORESOURCE_MEM); |
| 566 | if (EFA_BASE_BAR_MASK & ~pci_mem_bars) { |
| 567 | dev_err(&pdev->dev, "BARs unavailable. Requested %#x, available %#x\n" , |
| 568 | (int)EFA_BASE_BAR_MASK, pci_mem_bars); |
| 569 | err = -ENODEV; |
| 570 | goto err_ibdev_destroy; |
| 571 | } |
| 572 | err = pci_request_selected_regions(pdev, EFA_BASE_BAR_MASK, DRV_MODULE_NAME); |
| 573 | if (err) { |
| 574 | dev_err(&pdev->dev, "pci_request_selected_regions failed %d\n" , |
| 575 | err); |
| 576 | goto err_ibdev_destroy; |
| 577 | } |
| 578 | |
| 579 | dev->reg_bar_addr = pci_resource_start(pdev, EFA_REG_BAR); |
| 580 | dev->reg_bar_len = pci_resource_len(pdev, EFA_REG_BAR); |
| 581 | dev->mem_bar_addr = pci_resource_start(pdev, EFA_MEM_BAR); |
| 582 | dev->mem_bar_len = pci_resource_len(pdev, EFA_MEM_BAR); |
| 583 | |
| 584 | edev->reg_bar = devm_ioremap(dev: &pdev->dev, |
| 585 | offset: dev->reg_bar_addr, |
| 586 | size: dev->reg_bar_len); |
| 587 | if (!edev->reg_bar) { |
| 588 | dev_err(&pdev->dev, "Failed to remap register bar\n" ); |
| 589 | err = -EFAULT; |
| 590 | goto err_release_bars; |
| 591 | } |
| 592 | |
| 593 | err = efa_com_mmio_reg_read_init(edev); |
| 594 | if (err) { |
| 595 | dev_err(&pdev->dev, "Failed to init readless MMIO\n" ); |
| 596 | goto err_iounmap; |
| 597 | } |
| 598 | |
| 599 | err = efa_device_init(edev, pdev); |
| 600 | if (err) { |
| 601 | dev_err(&pdev->dev, "EFA device init failed\n" ); |
| 602 | if (err == -ETIME) |
| 603 | err = -EPROBE_DEFER; |
| 604 | goto err_reg_read_destroy; |
| 605 | } |
| 606 | |
| 607 | err = efa_enable_msix(dev); |
| 608 | if (err) |
| 609 | goto err_reg_read_destroy; |
| 610 | |
| 611 | edev->aq.msix_vector_idx = dev->admin_msix_vector_idx; |
| 612 | edev->aenq.msix_vector_idx = dev->admin_msix_vector_idx; |
| 613 | |
| 614 | err = efa_set_mgmnt_irq(dev); |
| 615 | if (err) |
| 616 | goto err_disable_msix; |
| 617 | |
| 618 | err = efa_com_admin_init(edev, aenq_handlers: &aenq_handlers); |
| 619 | if (err) |
| 620 | goto err_free_mgmnt_irq; |
| 621 | |
| 622 | return dev; |
| 623 | |
| 624 | err_free_mgmnt_irq: |
| 625 | efa_free_irq(dev, irq: &dev->admin_irq); |
| 626 | err_disable_msix: |
| 627 | efa_disable_msix(dev); |
| 628 | err_reg_read_destroy: |
| 629 | efa_com_mmio_reg_read_destroy(edev); |
| 630 | err_iounmap: |
| 631 | devm_iounmap(dev: &pdev->dev, addr: edev->reg_bar); |
| 632 | err_release_bars: |
| 633 | efa_release_bars(dev, EFA_BASE_BAR_MASK); |
| 634 | err_ibdev_destroy: |
| 635 | ib_dealloc_device(device: &dev->ibdev); |
| 636 | err_disable_device: |
| 637 | pci_disable_device(dev: pdev); |
| 638 | return ERR_PTR(error: err); |
| 639 | } |
| 640 | |
| 641 | static void efa_remove_device(struct pci_dev *pdev, |
| 642 | enum efa_regs_reset_reason_types reset_reason) |
| 643 | { |
| 644 | struct efa_dev *dev = pci_get_drvdata(pdev); |
| 645 | struct efa_com_dev *edev; |
| 646 | |
| 647 | edev = &dev->edev; |
| 648 | efa_com_dev_reset(edev, reset_reason); |
| 649 | efa_com_admin_destroy(edev); |
| 650 | efa_free_irq(dev, irq: &dev->admin_irq); |
| 651 | efa_disable_msix(dev); |
| 652 | efa_com_mmio_reg_read_destroy(edev); |
| 653 | devm_iounmap(dev: &pdev->dev, addr: edev->reg_bar); |
| 654 | efa_release_bars(dev, EFA_BASE_BAR_MASK); |
| 655 | xa_destroy(&dev->cqs_xa); |
| 656 | ib_dealloc_device(device: &dev->ibdev); |
| 657 | pci_disable_device(dev: pdev); |
| 658 | } |
| 659 | |
| 660 | static int efa_probe(struct pci_dev *pdev, const struct pci_device_id *ent) |
| 661 | { |
| 662 | struct efa_dev *dev; |
| 663 | int err; |
| 664 | |
| 665 | dev = efa_probe_device(pdev); |
| 666 | if (IS_ERR(ptr: dev)) |
| 667 | return PTR_ERR(ptr: dev); |
| 668 | |
| 669 | err = efa_ib_device_add(dev); |
| 670 | if (err) |
| 671 | goto err_remove_device; |
| 672 | |
| 673 | return 0; |
| 674 | |
| 675 | err_remove_device: |
| 676 | efa_remove_device(pdev, reset_reason: EFA_REGS_RESET_INIT_ERR); |
| 677 | return err; |
| 678 | } |
| 679 | |
| 680 | static void efa_remove(struct pci_dev *pdev) |
| 681 | { |
| 682 | struct efa_dev *dev = pci_get_drvdata(pdev); |
| 683 | |
| 684 | efa_ib_device_remove(dev); |
| 685 | efa_remove_device(pdev, reset_reason: EFA_REGS_RESET_NORMAL); |
| 686 | } |
| 687 | |
| 688 | static void efa_shutdown(struct pci_dev *pdev) |
| 689 | { |
| 690 | struct efa_dev *dev = pci_get_drvdata(pdev); |
| 691 | |
| 692 | efa_destroy_eqs(dev); |
| 693 | efa_com_dev_reset(edev: &dev->edev, reset_reason: EFA_REGS_RESET_SHUTDOWN); |
| 694 | efa_free_irq(dev, irq: &dev->admin_irq); |
| 695 | efa_disable_msix(dev); |
| 696 | } |
| 697 | |
| 698 | static struct pci_driver efa_pci_driver = { |
| 699 | .name = DRV_MODULE_NAME, |
| 700 | .id_table = efa_pci_tbl, |
| 701 | .probe = efa_probe, |
| 702 | .remove = efa_remove, |
| 703 | .shutdown = efa_shutdown, |
| 704 | }; |
| 705 | |
| 706 | module_pci_driver(efa_pci_driver); |
| 707 | |