1 // SPDX-License-Identifier: GPL-2.0 2 /* Copyright (c) Meta Platforms, Inc. and affiliates. */ 3 4 #include <linux/init.h> 5 #include <linux/module.h> 6 #include <linux/pci.h> 7 #include <linux/rtnetlink.h> 8 #include <linux/types.h> 9 #include <net/devlink.h> 10 11 #include "fbnic.h" 12 #include "fbnic_drvinfo.h" 13 #include "fbnic_hw_stats.h" 14 #include "fbnic_netdev.h" 15 16 char fbnic_driver_name[] = DRV_NAME; 17 18 MODULE_DESCRIPTION(DRV_SUMMARY); 19 MODULE_LICENSE("GPL"); 20 21 static const struct fbnic_info fbnic_asic_info = { 22 .max_num_queues = FBNIC_MAX_QUEUES, 23 .bar_mask = BIT(0) | BIT(4) 24 }; 25 26 static const struct fbnic_info *fbnic_info_tbl[] = { 27 [fbnic_board_asic] = &fbnic_asic_info, 28 }; 29 30 static const struct pci_device_id fbnic_pci_tbl[] = { 31 { PCI_DEVICE_DATA(META, FBNIC_ASIC, fbnic_board_asic) }, 32 /* Required last entry */ 33 {0, } 34 }; 35 MODULE_DEVICE_TABLE(pci, fbnic_pci_tbl); 36 37 u32 fbnic_rd32(struct fbnic_dev *fbd, u32 reg) 38 { 39 u32 __iomem *csr = READ_ONCE(fbd->uc_addr0); 40 u32 value; 41 42 if (!csr) 43 return ~0U; 44 45 value = readl(csr + reg); 46 47 /* If any bits are 0 value should be valid */ 48 if (~value) 49 return value; 50 51 /* All 1's may be valid if ZEROs register still works */ 52 if (reg != FBNIC_MASTER_SPARE_0 && ~readl(csr + FBNIC_MASTER_SPARE_0)) 53 return value; 54 55 /* Hardware is giving us all 1's reads, assume it is gone */ 56 WRITE_ONCE(fbd->uc_addr0, NULL); 57 WRITE_ONCE(fbd->uc_addr4, NULL); 58 59 dev_err(fbd->dev, 60 "Failed read (idx 0x%x AKA addr 0x%x), disabled CSR access, awaiting reset\n", 61 reg, reg << 2); 62 63 /* Notify stack that device has lost (PCIe) link */ 64 if (!fbnic_init_failure(fbd)) 65 netif_device_detach(fbd->netdev); 66 67 return ~0U; 68 } 69 70 bool fbnic_fw_present(struct fbnic_dev *fbd) 71 { 72 return !!READ_ONCE(fbd->uc_addr4); 73 } 74 75 void fbnic_fw_wr32(struct fbnic_dev *fbd, u32 reg, u32 val) 76 { 77 u32 __iomem *csr = READ_ONCE(fbd->uc_addr4); 78 79 if (csr) 80 writel(val, csr + reg); 81 } 82 83 u32 fbnic_fw_rd32(struct fbnic_dev *fbd, u32 reg) 84 { 85 u32 __iomem *csr = READ_ONCE(fbd->uc_addr4); 86 u32 value; 87 88 if (!csr) 89 return ~0U; 90 91 value = readl(csr + reg); 92 93 /* If any bits are 0 value should be valid */ 94 if (~value) 95 return value; 96 97 /* All 1's may be valid if ZEROs register still works */ 98 if (reg != FBNIC_FW_ZERO_REG && ~readl(csr + FBNIC_FW_ZERO_REG)) 99 return value; 100 101 /* Hardware is giving us all 1's reads, assume it is gone */ 102 WRITE_ONCE(fbd->uc_addr0, NULL); 103 WRITE_ONCE(fbd->uc_addr4, NULL); 104 105 dev_err(fbd->dev, 106 "Failed read (idx 0x%x AKA addr 0x%x), disabled CSR access, awaiting reset\n", 107 reg, reg << 2); 108 109 /* Notify stack that device has lost (PCIe) link */ 110 if (!fbnic_init_failure(fbd)) 111 netif_device_detach(fbd->netdev); 112 113 return ~0U; 114 } 115 116 static void fbnic_service_task_start(struct fbnic_net *fbn) 117 { 118 struct fbnic_dev *fbd = fbn->fbd; 119 120 schedule_delayed_work(&fbd->service_task, HZ); 121 } 122 123 static void fbnic_service_task_stop(struct fbnic_net *fbn) 124 { 125 struct fbnic_dev *fbd = fbn->fbd; 126 127 cancel_delayed_work(&fbd->service_task); 128 } 129 130 void fbnic_up(struct fbnic_net *fbn) 131 { 132 fbnic_enable(fbn); 133 134 fbnic_fill(fbn); 135 136 fbnic_rss_reinit_hw(fbn->fbd, fbn); 137 138 __fbnic_set_rx_mode(fbn->netdev); 139 140 /* Enable Tx/Rx processing */ 141 fbnic_napi_enable(fbn); 142 netif_tx_start_all_queues(fbn->netdev); 143 144 fbnic_service_task_start(fbn); 145 } 146 147 void fbnic_down_noidle(struct fbnic_net *fbn) 148 { 149 fbnic_service_task_stop(fbn); 150 151 /* Disable Tx/Rx Processing */ 152 fbnic_napi_disable(fbn); 153 netif_tx_disable(fbn->netdev); 154 155 fbnic_clear_rx_mode(fbn->netdev); 156 fbnic_clear_rules(fbn->fbd); 157 fbnic_rss_disable_hw(fbn->fbd); 158 fbnic_disable(fbn); 159 } 160 161 void fbnic_down(struct fbnic_net *fbn) 162 { 163 fbnic_down_noidle(fbn); 164 165 fbnic_wait_all_queues_idle(fbn->fbd, false); 166 167 fbnic_flush(fbn); 168 } 169 170 static void fbnic_health_check(struct fbnic_dev *fbd) 171 { 172 struct fbnic_fw_mbx *tx_mbx = &fbd->mbx[FBNIC_IPC_MBX_TX_IDX]; 173 174 /* As long as the heart is beating the FW is healty */ 175 if (fbd->fw_heartbeat_enabled) 176 return; 177 178 /* If the Tx mailbox still has messages sitting in it then there likely 179 * isn't anything we can do. We will wait until the mailbox is empty to 180 * report the fault so we can collect the crashlog. 181 */ 182 if (tx_mbx->head != tx_mbx->tail) 183 return; 184 185 /* TBD: Need to add a more thorough recovery here. 186 * Specifically I need to verify what all the firmware will have 187 * changed since we had setup and it rebooted. May just need to 188 * perform a down/up. For now we will just reclaim ownership so 189 * the heartbeat can catch the next fault. 190 */ 191 fbnic_fw_xmit_ownership_msg(fbd, true); 192 } 193 194 static void fbnic_service_task(struct work_struct *work) 195 { 196 struct fbnic_dev *fbd = container_of(to_delayed_work(work), 197 struct fbnic_dev, service_task); 198 199 rtnl_lock(); 200 201 fbnic_get_hw_stats32(fbd); 202 203 fbnic_fw_check_heartbeat(fbd); 204 205 fbnic_health_check(fbd); 206 207 if (netif_carrier_ok(fbd->netdev)) 208 fbnic_napi_depletion_check(fbd->netdev); 209 210 if (netif_running(fbd->netdev)) 211 schedule_delayed_work(&fbd->service_task, HZ); 212 213 rtnl_unlock(); 214 } 215 216 /** 217 * fbnic_probe - Device Initialization Routine 218 * @pdev: PCI device information struct 219 * @ent: entry in fbnic_pci_tbl 220 * 221 * Initializes a PCI device identified by a pci_dev structure. 222 * The OS initialization, configuring of the adapter private structure, 223 * and a hardware reset occur. 224 * 225 * Return: 0 on success, negative on failure 226 **/ 227 static int fbnic_probe(struct pci_dev *pdev, const struct pci_device_id *ent) 228 { 229 const struct fbnic_info *info = fbnic_info_tbl[ent->driver_data]; 230 struct net_device *netdev; 231 struct fbnic_dev *fbd; 232 int err; 233 234 if (pdev->error_state != pci_channel_io_normal) { 235 dev_err(&pdev->dev, 236 "PCI device still in an error state. Unable to load...\n"); 237 return -EIO; 238 } 239 240 err = pcim_enable_device(pdev); 241 if (err) { 242 dev_err(&pdev->dev, "PCI enable device failed: %d\n", err); 243 return err; 244 } 245 246 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(46)); 247 if (err) 248 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32)); 249 if (err) { 250 dev_err(&pdev->dev, "DMA configuration failed: %d\n", err); 251 return err; 252 } 253 254 err = pcim_iomap_regions(pdev, info->bar_mask, fbnic_driver_name); 255 if (err) { 256 dev_err(&pdev->dev, 257 "pci_request_selected_regions failed: %d\n", err); 258 return err; 259 } 260 261 fbd = fbnic_devlink_alloc(pdev); 262 if (!fbd) { 263 dev_err(&pdev->dev, "Devlink allocation failed\n"); 264 return -ENOMEM; 265 } 266 267 /* Populate driver with hardware-specific info and handlers */ 268 fbd->max_num_queues = info->max_num_queues; 269 270 pci_set_master(pdev); 271 pci_save_state(pdev); 272 273 INIT_DELAYED_WORK(&fbd->service_task, fbnic_service_task); 274 275 err = fbnic_alloc_irqs(fbd); 276 if (err) 277 goto free_fbd; 278 279 err = fbnic_mac_init(fbd); 280 if (err) { 281 dev_err(&pdev->dev, "Failed to initialize MAC: %d\n", err); 282 goto free_irqs; 283 } 284 285 err = fbnic_fw_request_mbx(fbd); 286 if (err) { 287 dev_err(&pdev->dev, 288 "Firmware mailbox initialization failure\n"); 289 goto free_irqs; 290 } 291 292 /* Send the request to enable the FW logging to host. Note if this 293 * fails we ignore the error and just display a message as it is 294 * possible the FW is just too old to support the logging and needs 295 * to be updated. 296 */ 297 err = fbnic_fw_log_init(fbd); 298 if (err) 299 dev_warn(fbd->dev, 300 "Unable to initialize firmware log buffer: %d\n", 301 err); 302 303 fbnic_devlink_register(fbd); 304 fbnic_dbg_fbd_init(fbd); 305 spin_lock_init(&fbd->hw_stats_lock); 306 307 /* Capture snapshot of hardware stats so netdev can calculate delta */ 308 fbnic_reset_hw_stats(fbd); 309 310 fbnic_hwmon_register(fbd); 311 312 if (!fbd->dsn) { 313 dev_warn(&pdev->dev, "Reading serial number failed\n"); 314 goto init_failure_mode; 315 } 316 317 netdev = fbnic_netdev_alloc(fbd); 318 if (!netdev) { 319 dev_err(&pdev->dev, "Netdev allocation failed\n"); 320 goto init_failure_mode; 321 } 322 323 err = fbnic_ptp_setup(fbd); 324 if (err) 325 goto ifm_free_netdev; 326 327 err = fbnic_netdev_register(netdev); 328 if (err) { 329 dev_err(&pdev->dev, "Netdev registration failed: %d\n", err); 330 goto ifm_destroy_ptp; 331 } 332 333 return 0; 334 335 ifm_destroy_ptp: 336 fbnic_ptp_destroy(fbd); 337 ifm_free_netdev: 338 fbnic_netdev_free(fbd); 339 init_failure_mode: 340 dev_warn(&pdev->dev, "Probe error encountered, entering init failure mode. Normal networking functionality will not be available.\n"); 341 /* Always return 0 even on error so devlink is registered to allow 342 * firmware updates for fixes. 343 */ 344 return 0; 345 free_irqs: 346 fbnic_free_irqs(fbd); 347 free_fbd: 348 fbnic_devlink_free(fbd); 349 350 return err; 351 } 352 353 /** 354 * fbnic_remove - Device Removal Routine 355 * @pdev: PCI device information struct 356 * 357 * Called by the PCI subsystem to alert the driver that it should release 358 * a PCI device. The could be caused by a Hot-Plug event, or because the 359 * driver is going to be removed from memory. 360 **/ 361 static void fbnic_remove(struct pci_dev *pdev) 362 { 363 struct fbnic_dev *fbd = pci_get_drvdata(pdev); 364 365 if (!fbnic_init_failure(fbd)) { 366 struct net_device *netdev = fbd->netdev; 367 368 fbnic_netdev_unregister(netdev); 369 cancel_delayed_work_sync(&fbd->service_task); 370 fbnic_ptp_destroy(fbd); 371 fbnic_netdev_free(fbd); 372 } 373 374 fbnic_hwmon_unregister(fbd); 375 fbnic_dbg_fbd_exit(fbd); 376 fbnic_devlink_unregister(fbd); 377 fbnic_fw_log_free(fbd); 378 fbnic_fw_free_mbx(fbd); 379 fbnic_free_irqs(fbd); 380 381 fbnic_devlink_free(fbd); 382 } 383 384 static int fbnic_pm_suspend(struct device *dev) 385 { 386 struct fbnic_dev *fbd = dev_get_drvdata(dev); 387 struct net_device *netdev = fbd->netdev; 388 389 if (fbnic_init_failure(fbd)) 390 goto null_uc_addr; 391 392 rtnl_lock(); 393 394 netif_device_detach(netdev); 395 396 if (netif_running(netdev)) 397 netdev->netdev_ops->ndo_stop(netdev); 398 399 rtnl_unlock(); 400 401 null_uc_addr: 402 fbnic_fw_log_disable(fbd); 403 404 devl_lock(priv_to_devlink(fbd)); 405 406 fbnic_fw_free_mbx(fbd); 407 408 devl_unlock(priv_to_devlink(fbd)); 409 410 /* Free the IRQs so they aren't trying to occupy sleeping CPUs */ 411 fbnic_free_irqs(fbd); 412 413 /* Hardware is about to go away, so switch off MMIO access internally */ 414 WRITE_ONCE(fbd->uc_addr0, NULL); 415 WRITE_ONCE(fbd->uc_addr4, NULL); 416 417 return 0; 418 } 419 420 static int __fbnic_pm_resume(struct device *dev) 421 { 422 struct fbnic_dev *fbd = dev_get_drvdata(dev); 423 struct net_device *netdev = fbd->netdev; 424 void __iomem * const *iomap_table; 425 struct fbnic_net *fbn; 426 int err; 427 428 /* Restore MMIO access */ 429 iomap_table = pcim_iomap_table(to_pci_dev(dev)); 430 fbd->uc_addr0 = iomap_table[0]; 431 fbd->uc_addr4 = iomap_table[4]; 432 433 /* Rerequest the IRQs */ 434 err = fbnic_alloc_irqs(fbd); 435 if (err) 436 goto err_invalidate_uc_addr; 437 438 fbd->mac->init_regs(fbd); 439 440 devl_lock(priv_to_devlink(fbd)); 441 442 /* Re-enable mailbox */ 443 err = fbnic_fw_request_mbx(fbd); 444 devl_unlock(priv_to_devlink(fbd)); 445 if (err) 446 goto err_free_irqs; 447 448 /* Only send log history if log buffer is empty to prevent duplicate 449 * log entries. 450 */ 451 fbnic_fw_log_enable(fbd, list_empty(&fbd->fw_log.entries)); 452 453 /* No netdev means there isn't a network interface to bring up */ 454 if (fbnic_init_failure(fbd)) 455 return 0; 456 457 fbn = netdev_priv(netdev); 458 459 /* Reset the queues if needed */ 460 fbnic_reset_queues(fbn, fbn->num_tx_queues, fbn->num_rx_queues); 461 462 rtnl_lock(); 463 464 if (netif_running(netdev)) 465 err = __fbnic_open(fbn); 466 467 rtnl_unlock(); 468 if (err) 469 goto err_free_mbx; 470 471 return 0; 472 err_free_mbx: 473 fbnic_fw_log_disable(fbd); 474 475 devl_lock(priv_to_devlink(fbd)); 476 fbnic_fw_free_mbx(fbd); 477 devl_unlock(priv_to_devlink(fbd)); 478 err_free_irqs: 479 fbnic_free_irqs(fbd); 480 err_invalidate_uc_addr: 481 WRITE_ONCE(fbd->uc_addr0, NULL); 482 WRITE_ONCE(fbd->uc_addr4, NULL); 483 return err; 484 } 485 486 static void __fbnic_pm_attach(struct device *dev) 487 { 488 struct fbnic_dev *fbd = dev_get_drvdata(dev); 489 struct net_device *netdev = fbd->netdev; 490 struct fbnic_net *fbn; 491 492 if (fbnic_init_failure(fbd)) 493 return; 494 495 fbn = netdev_priv(netdev); 496 497 if (netif_running(netdev)) 498 fbnic_up(fbn); 499 500 netif_device_attach(netdev); 501 } 502 503 static int __maybe_unused fbnic_pm_resume(struct device *dev) 504 { 505 int err; 506 507 err = __fbnic_pm_resume(dev); 508 if (!err) 509 __fbnic_pm_attach(dev); 510 511 return err; 512 } 513 514 static const struct dev_pm_ops fbnic_pm_ops = { 515 SET_SYSTEM_SLEEP_PM_OPS(fbnic_pm_suspend, fbnic_pm_resume) 516 }; 517 518 static void fbnic_shutdown(struct pci_dev *pdev) 519 { 520 fbnic_pm_suspend(&pdev->dev); 521 } 522 523 static pci_ers_result_t fbnic_err_error_detected(struct pci_dev *pdev, 524 pci_channel_state_t state) 525 { 526 /* Disconnect device if failure is not recoverable via reset */ 527 if (state == pci_channel_io_perm_failure) 528 return PCI_ERS_RESULT_DISCONNECT; 529 530 fbnic_pm_suspend(&pdev->dev); 531 532 /* Request a slot reset */ 533 return PCI_ERS_RESULT_NEED_RESET; 534 } 535 536 static pci_ers_result_t fbnic_err_slot_reset(struct pci_dev *pdev) 537 { 538 int err; 539 540 pci_set_power_state(pdev, PCI_D0); 541 pci_restore_state(pdev); 542 pci_save_state(pdev); 543 544 if (pci_enable_device_mem(pdev)) { 545 dev_err(&pdev->dev, 546 "Cannot re-enable PCI device after reset.\n"); 547 return PCI_ERS_RESULT_DISCONNECT; 548 } 549 550 /* Restore device to previous state */ 551 err = __fbnic_pm_resume(&pdev->dev); 552 553 return err ? PCI_ERS_RESULT_DISCONNECT : PCI_ERS_RESULT_RECOVERED; 554 } 555 556 static void fbnic_err_resume(struct pci_dev *pdev) 557 { 558 __fbnic_pm_attach(&pdev->dev); 559 } 560 561 static const struct pci_error_handlers fbnic_err_handler = { 562 .error_detected = fbnic_err_error_detected, 563 .slot_reset = fbnic_err_slot_reset, 564 .resume = fbnic_err_resume, 565 }; 566 567 static struct pci_driver fbnic_driver = { 568 .name = fbnic_driver_name, 569 .id_table = fbnic_pci_tbl, 570 .probe = fbnic_probe, 571 .remove = fbnic_remove, 572 .driver.pm = &fbnic_pm_ops, 573 .shutdown = fbnic_shutdown, 574 .err_handler = &fbnic_err_handler, 575 }; 576 577 /** 578 * fbnic_init_module - Driver Registration Routine 579 * 580 * The first routine called when the driver is loaded. All it does is 581 * register with the PCI subsystem. 582 * 583 * Return: 0 on success, negative on failure 584 **/ 585 static int __init fbnic_init_module(void) 586 { 587 int err; 588 589 fbnic_dbg_init(); 590 591 err = pci_register_driver(&fbnic_driver); 592 if (err) { 593 fbnic_dbg_exit(); 594 goto out; 595 } 596 597 pr_info(DRV_SUMMARY " (%s)", fbnic_driver.name); 598 out: 599 return err; 600 } 601 module_init(fbnic_init_module); 602 603 /** 604 * fbnic_exit_module - Driver Exit Cleanup Routine 605 * 606 * Called just before the driver is removed from memory. 607 **/ 608 static void __exit fbnic_exit_module(void) 609 { 610 pci_unregister_driver(&fbnic_driver); 611 612 fbnic_dbg_exit(); 613 } 614 module_exit(fbnic_exit_module); 615