1 /* Intel(R) Ethernet Switch Host Interface Driver 2 * Copyright(c) 2013 - 2017 Intel Corporation. 3 * 4 * This program is free software; you can redistribute it and/or modify it 5 * under the terms and conditions of the GNU General Public License, 6 * version 2, as published by the Free Software Foundation. 7 * 8 * This program is distributed in the hope it will be useful, but WITHOUT 9 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 10 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for 11 * more details. 12 * 13 * The full GNU General Public License is included in this distribution in 14 * the file called "COPYING". 15 * 16 * Contact Information: 17 * e1000-devel Mailing List <e1000-devel@lists.sourceforge.net> 18 * Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497 19 */ 20 21 #include <linux/module.h> 22 #include <linux/interrupt.h> 23 #include <linux/aer.h> 24 25 #include "fm10k.h" 26 27 static const struct fm10k_info *fm10k_info_tbl[] = { 28 [fm10k_device_pf] = &fm10k_pf_info, 29 [fm10k_device_vf] = &fm10k_vf_info, 30 }; 31 32 /** 33 * fm10k_pci_tbl - PCI Device ID Table 34 * 35 * Wildcard entries (PCI_ANY_ID) should come last 36 * Last entry must be all 0s 37 * 38 * { Vendor ID, Device ID, SubVendor ID, SubDevice ID, 39 * Class, Class Mask, private data (not used) } 40 */ 41 static const struct pci_device_id fm10k_pci_tbl[] = { 42 { PCI_VDEVICE(INTEL, FM10K_DEV_ID_PF), fm10k_device_pf }, 43 { PCI_VDEVICE(INTEL, FM10K_DEV_ID_VF), fm10k_device_vf }, 44 /* required last entry */ 45 { 0, } 46 }; 47 MODULE_DEVICE_TABLE(pci, fm10k_pci_tbl); 48 49 u16 fm10k_read_pci_cfg_word(struct fm10k_hw *hw, u32 reg) 50 { 51 struct fm10k_intfc *interface = hw->back; 52 u16 value = 0; 53 54 if (FM10K_REMOVED(hw->hw_addr)) 55 return ~value; 56 57 pci_read_config_word(interface->pdev, reg, &value); 58 if (value == 0xFFFF) 59 fm10k_write_flush(hw); 60 61 return value; 62 } 63 64 u32 fm10k_read_reg(struct fm10k_hw *hw, int reg) 65 { 66 u32 __iomem *hw_addr = READ_ONCE(hw->hw_addr); 67 u32 value = 0; 68 69 if (FM10K_REMOVED(hw_addr)) 70 return ~value; 71 72 value = readl(&hw_addr[reg]); 73 if (!(~value) && (!reg || !(~readl(hw_addr)))) { 74 struct fm10k_intfc *interface = hw->back; 75 struct net_device *netdev = interface->netdev; 76 77 hw->hw_addr = NULL; 78 netif_device_detach(netdev); 79 netdev_err(netdev, "PCIe link lost, device now detached\n"); 80 } 81 82 return value; 83 } 84 85 static int fm10k_hw_ready(struct fm10k_intfc *interface) 86 { 87 struct fm10k_hw *hw = &interface->hw; 88 89 fm10k_write_flush(hw); 90 91 return FM10K_REMOVED(hw->hw_addr) ? -ENODEV : 0; 92 } 93 94 /** 95 * fm10k_macvlan_schedule - Schedule MAC/VLAN queue task 96 * @interface: fm10k private interface structure 97 * 98 * Schedule the MAC/VLAN queue monitor task. If the MAC/VLAN task cannot be 99 * started immediately, request that it be restarted when possible. 100 */ 101 void fm10k_macvlan_schedule(struct fm10k_intfc *interface) 102 { 103 /* Avoid processing the MAC/VLAN queue when the service task is 104 * disabled, or when we're resetting the device. 105 */ 106 if (!test_bit(__FM10K_MACVLAN_DISABLE, interface->state) && 107 !test_and_set_bit(__FM10K_MACVLAN_SCHED, interface->state)) { 108 clear_bit(__FM10K_MACVLAN_REQUEST, interface->state); 109 /* We delay the actual start of execution in order to allow 110 * multiple MAC/VLAN updates to accumulate before handling 111 * them, and to allow some time to let the mailbox drain 112 * between runs. 113 */ 114 queue_delayed_work(fm10k_workqueue, 115 &interface->macvlan_task, 10); 116 } else { 117 set_bit(__FM10K_MACVLAN_REQUEST, interface->state); 118 } 119 } 120 121 /** 122 * fm10k_stop_macvlan_task - Stop the MAC/VLAN queue monitor 123 * @interface: fm10k private interface structure 124 * 125 * Wait until the MAC/VLAN queue task has stopped, and cancel any future 126 * requests. 127 */ 128 static void fm10k_stop_macvlan_task(struct fm10k_intfc *interface) 129 { 130 /* Disable the MAC/VLAN work item */ 131 set_bit(__FM10K_MACVLAN_DISABLE, interface->state); 132 133 /* Make sure we waited until any current invocations have stopped */ 134 cancel_delayed_work_sync(&interface->macvlan_task); 135 136 /* We set the __FM10K_MACVLAN_SCHED bit when we schedule the task. 137 * However, it may not be unset of the MAC/VLAN task never actually 138 * got a chance to run. Since we've canceled the task here, and it 139 * cannot be rescheuled right now, we need to ensure the scheduled bit 140 * gets unset. 141 */ 142 clear_bit(__FM10K_MACVLAN_SCHED, interface->state); 143 } 144 145 /** 146 * fm10k_resume_macvlan_task - Restart the MAC/VLAN queue monitor 147 * @interface: fm10k private interface structure 148 * 149 * Clear the __FM10K_MACVLAN_DISABLE bit and, if a request occurred, schedule 150 * the MAC/VLAN work monitor. 151 */ 152 static void fm10k_resume_macvlan_task(struct fm10k_intfc *interface) 153 { 154 /* Re-enable the MAC/VLAN work item */ 155 clear_bit(__FM10K_MACVLAN_DISABLE, interface->state); 156 157 /* We might have received a MAC/VLAN request while disabled. If so, 158 * kick off the queue now. 159 */ 160 if (test_bit(__FM10K_MACVLAN_REQUEST, interface->state)) 161 fm10k_macvlan_schedule(interface); 162 } 163 164 void fm10k_service_event_schedule(struct fm10k_intfc *interface) 165 { 166 if (!test_bit(__FM10K_SERVICE_DISABLE, interface->state) && 167 !test_and_set_bit(__FM10K_SERVICE_SCHED, interface->state)) { 168 clear_bit(__FM10K_SERVICE_REQUEST, interface->state); 169 queue_work(fm10k_workqueue, &interface->service_task); 170 } else { 171 set_bit(__FM10K_SERVICE_REQUEST, interface->state); 172 } 173 } 174 175 static void fm10k_service_event_complete(struct fm10k_intfc *interface) 176 { 177 WARN_ON(!test_bit(__FM10K_SERVICE_SCHED, interface->state)); 178 179 /* flush memory to make sure state is correct before next watchog */ 180 smp_mb__before_atomic(); 181 clear_bit(__FM10K_SERVICE_SCHED, interface->state); 182 183 /* If a service event was requested since we started, immediately 184 * re-schedule now. This ensures we don't drop a request until the 185 * next timer event. 186 */ 187 if (test_bit(__FM10K_SERVICE_REQUEST, interface->state)) 188 fm10k_service_event_schedule(interface); 189 } 190 191 static void fm10k_stop_service_event(struct fm10k_intfc *interface) 192 { 193 set_bit(__FM10K_SERVICE_DISABLE, interface->state); 194 cancel_work_sync(&interface->service_task); 195 196 /* It's possible that cancel_work_sync stopped the service task from 197 * running before it could actually start. In this case the 198 * __FM10K_SERVICE_SCHED bit will never be cleared. Since we know that 199 * the service task cannot be running at this point, we need to clear 200 * the scheduled bit, as otherwise the service task may never be 201 * restarted. 202 */ 203 clear_bit(__FM10K_SERVICE_SCHED, interface->state); 204 } 205 206 static void fm10k_start_service_event(struct fm10k_intfc *interface) 207 { 208 clear_bit(__FM10K_SERVICE_DISABLE, interface->state); 209 fm10k_service_event_schedule(interface); 210 } 211 212 /** 213 * fm10k_service_timer - Timer Call-back 214 * @data: pointer to interface cast into an unsigned long 215 **/ 216 static void fm10k_service_timer(struct timer_list *t) 217 { 218 struct fm10k_intfc *interface = from_timer(interface, t, 219 service_timer); 220 221 /* Reset the timer */ 222 mod_timer(&interface->service_timer, (HZ * 2) + jiffies); 223 224 fm10k_service_event_schedule(interface); 225 } 226 227 /** 228 * fm10k_prepare_for_reset - Prepare the driver and device for a pending reset 229 * @interface: fm10k private data structure 230 * 231 * This function prepares for a device reset by shutting as much down as we 232 * can. It does nothing and returns false if __FM10K_RESETTING was already set 233 * prior to calling this function. It returns true if it actually did work. 234 */ 235 static bool fm10k_prepare_for_reset(struct fm10k_intfc *interface) 236 { 237 struct net_device *netdev = interface->netdev; 238 239 WARN_ON(in_interrupt()); 240 241 /* put off any impending NetWatchDogTimeout */ 242 netif_trans_update(netdev); 243 244 /* Nothing to do if a reset is already in progress */ 245 if (test_and_set_bit(__FM10K_RESETTING, interface->state)) 246 return false; 247 248 /* As the MAC/VLAN task will be accessing registers it must not be 249 * running while we reset. Although the task will not be scheduled 250 * once we start resetting it may already be running 251 */ 252 fm10k_stop_macvlan_task(interface); 253 254 rtnl_lock(); 255 256 fm10k_iov_suspend(interface->pdev); 257 258 if (netif_running(netdev)) 259 fm10k_close(netdev); 260 261 fm10k_mbx_free_irq(interface); 262 263 /* free interrupts */ 264 fm10k_clear_queueing_scheme(interface); 265 266 /* delay any future reset requests */ 267 interface->last_reset = jiffies + (10 * HZ); 268 269 rtnl_unlock(); 270 271 return true; 272 } 273 274 static int fm10k_handle_reset(struct fm10k_intfc *interface) 275 { 276 struct net_device *netdev = interface->netdev; 277 struct fm10k_hw *hw = &interface->hw; 278 int err; 279 280 WARN_ON(!test_bit(__FM10K_RESETTING, interface->state)); 281 282 rtnl_lock(); 283 284 pci_set_master(interface->pdev); 285 286 /* reset and initialize the hardware so it is in a known state */ 287 err = hw->mac.ops.reset_hw(hw); 288 if (err) { 289 dev_err(&interface->pdev->dev, "reset_hw failed: %d\n", err); 290 goto reinit_err; 291 } 292 293 err = hw->mac.ops.init_hw(hw); 294 if (err) { 295 dev_err(&interface->pdev->dev, "init_hw failed: %d\n", err); 296 goto reinit_err; 297 } 298 299 err = fm10k_init_queueing_scheme(interface); 300 if (err) { 301 dev_err(&interface->pdev->dev, 302 "init_queueing_scheme failed: %d\n", err); 303 goto reinit_err; 304 } 305 306 /* re-associate interrupts */ 307 err = fm10k_mbx_request_irq(interface); 308 if (err) 309 goto err_mbx_irq; 310 311 err = fm10k_hw_ready(interface); 312 if (err) 313 goto err_open; 314 315 /* update hardware address for VFs if perm_addr has changed */ 316 if (hw->mac.type == fm10k_mac_vf) { 317 if (is_valid_ether_addr(hw->mac.perm_addr)) { 318 ether_addr_copy(hw->mac.addr, hw->mac.perm_addr); 319 ether_addr_copy(netdev->perm_addr, hw->mac.perm_addr); 320 ether_addr_copy(netdev->dev_addr, hw->mac.perm_addr); 321 netdev->addr_assign_type &= ~NET_ADDR_RANDOM; 322 } 323 324 if (hw->mac.vlan_override) 325 netdev->features &= ~NETIF_F_HW_VLAN_CTAG_RX; 326 else 327 netdev->features |= NETIF_F_HW_VLAN_CTAG_RX; 328 } 329 330 err = netif_running(netdev) ? fm10k_open(netdev) : 0; 331 if (err) 332 goto err_open; 333 334 fm10k_iov_resume(interface->pdev); 335 336 rtnl_unlock(); 337 338 fm10k_resume_macvlan_task(interface); 339 340 clear_bit(__FM10K_RESETTING, interface->state); 341 342 return err; 343 err_open: 344 fm10k_mbx_free_irq(interface); 345 err_mbx_irq: 346 fm10k_clear_queueing_scheme(interface); 347 reinit_err: 348 netif_device_detach(netdev); 349 350 rtnl_unlock(); 351 352 clear_bit(__FM10K_RESETTING, interface->state); 353 354 return err; 355 } 356 357 static void fm10k_detach_subtask(struct fm10k_intfc *interface) 358 { 359 struct net_device *netdev = interface->netdev; 360 u32 __iomem *hw_addr; 361 u32 value; 362 int err; 363 364 /* do nothing if netdev is still present or hw_addr is set */ 365 if (netif_device_present(netdev) || interface->hw.hw_addr) 366 return; 367 368 /* We've lost the PCIe register space, and can no longer access the 369 * device. Shut everything except the detach subtask down and prepare 370 * to reset the device in case we recover. If we actually prepare for 371 * reset, indicate that we're detached. 372 */ 373 if (fm10k_prepare_for_reset(interface)) 374 set_bit(__FM10K_RESET_DETACHED, interface->state); 375 376 /* check the real address space to see if we've recovered */ 377 hw_addr = READ_ONCE(interface->uc_addr); 378 value = readl(hw_addr); 379 if (~value) { 380 /* Make sure the reset was initiated because we detached, 381 * otherwise we might race with a different reset flow. 382 */ 383 if (!test_and_clear_bit(__FM10K_RESET_DETACHED, 384 interface->state)) 385 return; 386 387 /* Restore the hardware address */ 388 interface->hw.hw_addr = interface->uc_addr; 389 390 /* PCIe link has been restored, and the device is active 391 * again. Restore everything and reset the device. 392 */ 393 err = fm10k_handle_reset(interface); 394 if (err) { 395 netdev_err(netdev, "Unable to reset device: %d\n", err); 396 interface->hw.hw_addr = NULL; 397 return; 398 } 399 400 /* Re-attach the netdev */ 401 netif_device_attach(netdev); 402 netdev_warn(netdev, "PCIe link restored, device now attached\n"); 403 return; 404 } 405 } 406 407 static void fm10k_reset_subtask(struct fm10k_intfc *interface) 408 { 409 int err; 410 411 if (!test_and_clear_bit(FM10K_FLAG_RESET_REQUESTED, 412 interface->flags)) 413 return; 414 415 /* If another thread has already prepared to reset the device, we 416 * should not attempt to handle a reset here, since we'd race with 417 * that thread. This may happen if we suspend the device or if the 418 * PCIe link is lost. In this case, we'll just ignore the RESET 419 * request, as it will (eventually) be taken care of when the thread 420 * which actually started the reset is finished. 421 */ 422 if (!fm10k_prepare_for_reset(interface)) 423 return; 424 425 netdev_err(interface->netdev, "Reset interface\n"); 426 427 err = fm10k_handle_reset(interface); 428 if (err) 429 dev_err(&interface->pdev->dev, 430 "fm10k_handle_reset failed: %d\n", err); 431 } 432 433 /** 434 * fm10k_configure_swpri_map - Configure Receive SWPRI to PC mapping 435 * @interface: board private structure 436 * 437 * Configure the SWPRI to PC mapping for the port. 438 **/ 439 static void fm10k_configure_swpri_map(struct fm10k_intfc *interface) 440 { 441 struct net_device *netdev = interface->netdev; 442 struct fm10k_hw *hw = &interface->hw; 443 int i; 444 445 /* clear flag indicating update is needed */ 446 clear_bit(FM10K_FLAG_SWPRI_CONFIG, interface->flags); 447 448 /* these registers are only available on the PF */ 449 if (hw->mac.type != fm10k_mac_pf) 450 return; 451 452 /* configure SWPRI to PC map */ 453 for (i = 0; i < FM10K_SWPRI_MAX; i++) 454 fm10k_write_reg(hw, FM10K_SWPRI_MAP(i), 455 netdev_get_prio_tc_map(netdev, i)); 456 } 457 458 /** 459 * fm10k_watchdog_update_host_state - Update the link status based on host. 460 * @interface: board private structure 461 **/ 462 static void fm10k_watchdog_update_host_state(struct fm10k_intfc *interface) 463 { 464 struct fm10k_hw *hw = &interface->hw; 465 s32 err; 466 467 if (test_bit(__FM10K_LINK_DOWN, interface->state)) { 468 interface->host_ready = false; 469 if (time_is_after_jiffies(interface->link_down_event)) 470 return; 471 clear_bit(__FM10K_LINK_DOWN, interface->state); 472 } 473 474 if (test_bit(FM10K_FLAG_SWPRI_CONFIG, interface->flags)) { 475 if (rtnl_trylock()) { 476 fm10k_configure_swpri_map(interface); 477 rtnl_unlock(); 478 } 479 } 480 481 /* lock the mailbox for transmit and receive */ 482 fm10k_mbx_lock(interface); 483 484 err = hw->mac.ops.get_host_state(hw, &interface->host_ready); 485 if (err && time_is_before_jiffies(interface->last_reset)) 486 set_bit(FM10K_FLAG_RESET_REQUESTED, interface->flags); 487 488 /* free the lock */ 489 fm10k_mbx_unlock(interface); 490 } 491 492 /** 493 * fm10k_mbx_subtask - Process upstream and downstream mailboxes 494 * @interface: board private structure 495 * 496 * This function will process both the upstream and downstream mailboxes. 497 **/ 498 static void fm10k_mbx_subtask(struct fm10k_intfc *interface) 499 { 500 /* If we're resetting, bail out */ 501 if (test_bit(__FM10K_RESETTING, interface->state)) 502 return; 503 504 /* process upstream mailbox and update device state */ 505 fm10k_watchdog_update_host_state(interface); 506 507 /* process downstream mailboxes */ 508 fm10k_iov_mbx(interface); 509 } 510 511 /** 512 * fm10k_watchdog_host_is_ready - Update netdev status based on host ready 513 * @interface: board private structure 514 **/ 515 static void fm10k_watchdog_host_is_ready(struct fm10k_intfc *interface) 516 { 517 struct net_device *netdev = interface->netdev; 518 519 /* only continue if link state is currently down */ 520 if (netif_carrier_ok(netdev)) 521 return; 522 523 netif_info(interface, drv, netdev, "NIC Link is up\n"); 524 525 netif_carrier_on(netdev); 526 netif_tx_wake_all_queues(netdev); 527 } 528 529 /** 530 * fm10k_watchdog_host_not_ready - Update netdev status based on host not ready 531 * @interface: board private structure 532 **/ 533 static void fm10k_watchdog_host_not_ready(struct fm10k_intfc *interface) 534 { 535 struct net_device *netdev = interface->netdev; 536 537 /* only continue if link state is currently up */ 538 if (!netif_carrier_ok(netdev)) 539 return; 540 541 netif_info(interface, drv, netdev, "NIC Link is down\n"); 542 543 netif_carrier_off(netdev); 544 netif_tx_stop_all_queues(netdev); 545 } 546 547 /** 548 * fm10k_update_stats - Update the board statistics counters. 549 * @interface: board private structure 550 **/ 551 void fm10k_update_stats(struct fm10k_intfc *interface) 552 { 553 struct net_device_stats *net_stats = &interface->netdev->stats; 554 struct fm10k_hw *hw = &interface->hw; 555 u64 hw_csum_tx_good = 0, hw_csum_rx_good = 0, rx_length_errors = 0; 556 u64 rx_switch_errors = 0, rx_drops = 0, rx_pp_errors = 0; 557 u64 rx_link_errors = 0; 558 u64 rx_errors = 0, rx_csum_errors = 0, tx_csum_errors = 0; 559 u64 restart_queue = 0, tx_busy = 0, alloc_failed = 0; 560 u64 rx_bytes_nic = 0, rx_pkts_nic = 0, rx_drops_nic = 0; 561 u64 tx_bytes_nic = 0, tx_pkts_nic = 0; 562 u64 bytes, pkts; 563 int i; 564 565 /* ensure only one thread updates stats at a time */ 566 if (test_and_set_bit(__FM10K_UPDATING_STATS, interface->state)) 567 return; 568 569 /* do not allow stats update via service task for next second */ 570 interface->next_stats_update = jiffies + HZ; 571 572 /* gather some stats to the interface struct that are per queue */ 573 for (bytes = 0, pkts = 0, i = 0; i < interface->num_tx_queues; i++) { 574 struct fm10k_ring *tx_ring = READ_ONCE(interface->tx_ring[i]); 575 576 if (!tx_ring) 577 continue; 578 579 restart_queue += tx_ring->tx_stats.restart_queue; 580 tx_busy += tx_ring->tx_stats.tx_busy; 581 tx_csum_errors += tx_ring->tx_stats.csum_err; 582 bytes += tx_ring->stats.bytes; 583 pkts += tx_ring->stats.packets; 584 hw_csum_tx_good += tx_ring->tx_stats.csum_good; 585 } 586 587 interface->restart_queue = restart_queue; 588 interface->tx_busy = tx_busy; 589 net_stats->tx_bytes = bytes; 590 net_stats->tx_packets = pkts; 591 interface->tx_csum_errors = tx_csum_errors; 592 interface->hw_csum_tx_good = hw_csum_tx_good; 593 594 /* gather some stats to the interface struct that are per queue */ 595 for (bytes = 0, pkts = 0, i = 0; i < interface->num_rx_queues; i++) { 596 struct fm10k_ring *rx_ring = READ_ONCE(interface->rx_ring[i]); 597 598 if (!rx_ring) 599 continue; 600 601 bytes += rx_ring->stats.bytes; 602 pkts += rx_ring->stats.packets; 603 alloc_failed += rx_ring->rx_stats.alloc_failed; 604 rx_csum_errors += rx_ring->rx_stats.csum_err; 605 rx_errors += rx_ring->rx_stats.errors; 606 hw_csum_rx_good += rx_ring->rx_stats.csum_good; 607 rx_switch_errors += rx_ring->rx_stats.switch_errors; 608 rx_drops += rx_ring->rx_stats.drops; 609 rx_pp_errors += rx_ring->rx_stats.pp_errors; 610 rx_link_errors += rx_ring->rx_stats.link_errors; 611 rx_length_errors += rx_ring->rx_stats.length_errors; 612 } 613 614 net_stats->rx_bytes = bytes; 615 net_stats->rx_packets = pkts; 616 interface->alloc_failed = alloc_failed; 617 interface->rx_csum_errors = rx_csum_errors; 618 interface->hw_csum_rx_good = hw_csum_rx_good; 619 interface->rx_switch_errors = rx_switch_errors; 620 interface->rx_drops = rx_drops; 621 interface->rx_pp_errors = rx_pp_errors; 622 interface->rx_link_errors = rx_link_errors; 623 interface->rx_length_errors = rx_length_errors; 624 625 hw->mac.ops.update_hw_stats(hw, &interface->stats); 626 627 for (i = 0; i < hw->mac.max_queues; i++) { 628 struct fm10k_hw_stats_q *q = &interface->stats.q[i]; 629 630 tx_bytes_nic += q->tx_bytes.count; 631 tx_pkts_nic += q->tx_packets.count; 632 rx_bytes_nic += q->rx_bytes.count; 633 rx_pkts_nic += q->rx_packets.count; 634 rx_drops_nic += q->rx_drops.count; 635 } 636 637 interface->tx_bytes_nic = tx_bytes_nic; 638 interface->tx_packets_nic = tx_pkts_nic; 639 interface->rx_bytes_nic = rx_bytes_nic; 640 interface->rx_packets_nic = rx_pkts_nic; 641 interface->rx_drops_nic = rx_drops_nic; 642 643 /* Fill out the OS statistics structure */ 644 net_stats->rx_errors = rx_errors; 645 net_stats->rx_dropped = interface->stats.nodesc_drop.count; 646 647 clear_bit(__FM10K_UPDATING_STATS, interface->state); 648 } 649 650 /** 651 * fm10k_watchdog_flush_tx - flush queues on host not ready 652 * @interface - pointer to the device interface structure 653 **/ 654 static void fm10k_watchdog_flush_tx(struct fm10k_intfc *interface) 655 { 656 int some_tx_pending = 0; 657 int i; 658 659 /* nothing to do if carrier is up */ 660 if (netif_carrier_ok(interface->netdev)) 661 return; 662 663 for (i = 0; i < interface->num_tx_queues; i++) { 664 struct fm10k_ring *tx_ring = interface->tx_ring[i]; 665 666 if (tx_ring->next_to_use != tx_ring->next_to_clean) { 667 some_tx_pending = 1; 668 break; 669 } 670 } 671 672 /* We've lost link, so the controller stops DMA, but we've got 673 * queued Tx work that's never going to get done, so reset 674 * controller to flush Tx. 675 */ 676 if (some_tx_pending) 677 set_bit(FM10K_FLAG_RESET_REQUESTED, interface->flags); 678 } 679 680 /** 681 * fm10k_watchdog_subtask - check and bring link up 682 * @interface - pointer to the device interface structure 683 **/ 684 static void fm10k_watchdog_subtask(struct fm10k_intfc *interface) 685 { 686 /* if interface is down do nothing */ 687 if (test_bit(__FM10K_DOWN, interface->state) || 688 test_bit(__FM10K_RESETTING, interface->state)) 689 return; 690 691 if (interface->host_ready) 692 fm10k_watchdog_host_is_ready(interface); 693 else 694 fm10k_watchdog_host_not_ready(interface); 695 696 /* update stats only once every second */ 697 if (time_is_before_jiffies(interface->next_stats_update)) 698 fm10k_update_stats(interface); 699 700 /* flush any uncompleted work */ 701 fm10k_watchdog_flush_tx(interface); 702 } 703 704 /** 705 * fm10k_check_hang_subtask - check for hung queues and dropped interrupts 706 * @interface - pointer to the device interface structure 707 * 708 * This function serves two purposes. First it strobes the interrupt lines 709 * in order to make certain interrupts are occurring. Secondly it sets the 710 * bits needed to check for TX hangs. As a result we should immediately 711 * determine if a hang has occurred. 712 */ 713 static void fm10k_check_hang_subtask(struct fm10k_intfc *interface) 714 { 715 int i; 716 717 /* If we're down or resetting, just bail */ 718 if (test_bit(__FM10K_DOWN, interface->state) || 719 test_bit(__FM10K_RESETTING, interface->state)) 720 return; 721 722 /* rate limit tx hang checks to only once every 2 seconds */ 723 if (time_is_after_eq_jiffies(interface->next_tx_hang_check)) 724 return; 725 interface->next_tx_hang_check = jiffies + (2 * HZ); 726 727 if (netif_carrier_ok(interface->netdev)) { 728 /* Force detection of hung controller */ 729 for (i = 0; i < interface->num_tx_queues; i++) 730 set_check_for_tx_hang(interface->tx_ring[i]); 731 732 /* Rearm all in-use q_vectors for immediate firing */ 733 for (i = 0; i < interface->num_q_vectors; i++) { 734 struct fm10k_q_vector *qv = interface->q_vector[i]; 735 736 if (!qv->tx.count && !qv->rx.count) 737 continue; 738 writel(FM10K_ITR_ENABLE | FM10K_ITR_PENDING2, qv->itr); 739 } 740 } 741 } 742 743 /** 744 * fm10k_service_task - manages and runs subtasks 745 * @work: pointer to work_struct containing our data 746 **/ 747 static void fm10k_service_task(struct work_struct *work) 748 { 749 struct fm10k_intfc *interface; 750 751 interface = container_of(work, struct fm10k_intfc, service_task); 752 753 /* Check whether we're detached first */ 754 fm10k_detach_subtask(interface); 755 756 /* tasks run even when interface is down */ 757 fm10k_mbx_subtask(interface); 758 fm10k_reset_subtask(interface); 759 760 /* tasks only run when interface is up */ 761 fm10k_watchdog_subtask(interface); 762 fm10k_check_hang_subtask(interface); 763 764 /* release lock on service events to allow scheduling next event */ 765 fm10k_service_event_complete(interface); 766 } 767 768 /** 769 * fm10k_macvlan_task - send queued MAC/VLAN requests to switch manager 770 * @work: pointer to work_struct containing our data 771 * 772 * This work item handles sending MAC/VLAN updates to the switch manager. When 773 * the interface is up, it will attempt to queue mailbox messages to the 774 * switch manager requesting updates for MAC/VLAN pairs. If the Tx fifo of the 775 * mailbox is full, it will reschedule itself to try again in a short while. 776 * This ensures that the driver does not overload the switch mailbox with too 777 * many simultaneous requests, causing an unnecessary reset. 778 **/ 779 static void fm10k_macvlan_task(struct work_struct *work) 780 { 781 struct fm10k_macvlan_request *item; 782 struct fm10k_intfc *interface; 783 struct delayed_work *dwork; 784 struct list_head *requests; 785 struct fm10k_hw *hw; 786 unsigned long flags; 787 788 dwork = to_delayed_work(work); 789 interface = container_of(dwork, struct fm10k_intfc, macvlan_task); 790 hw = &interface->hw; 791 requests = &interface->macvlan_requests; 792 793 do { 794 /* Pop the first item off the list */ 795 spin_lock_irqsave(&interface->macvlan_lock, flags); 796 item = list_first_entry_or_null(requests, 797 struct fm10k_macvlan_request, 798 list); 799 if (item) 800 list_del_init(&item->list); 801 802 spin_unlock_irqrestore(&interface->macvlan_lock, flags); 803 804 /* We have no more items to process */ 805 if (!item) 806 goto done; 807 808 fm10k_mbx_lock(interface); 809 810 /* Check that we have plenty of space to send the message. We 811 * want to ensure that the mailbox stays low enough to avoid a 812 * change in the host state, otherwise we may see spurious 813 * link up / link down notifications. 814 */ 815 if (!hw->mbx.ops.tx_ready(&hw->mbx, FM10K_VFMBX_MSG_MTU + 5)) { 816 hw->mbx.ops.process(hw, &hw->mbx); 817 set_bit(__FM10K_MACVLAN_REQUEST, interface->state); 818 fm10k_mbx_unlock(interface); 819 820 /* Put the request back on the list */ 821 spin_lock_irqsave(&interface->macvlan_lock, flags); 822 list_add(&item->list, requests); 823 spin_unlock_irqrestore(&interface->macvlan_lock, flags); 824 break; 825 } 826 827 switch (item->type) { 828 case FM10K_MC_MAC_REQUEST: 829 hw->mac.ops.update_mc_addr(hw, 830 item->mac.glort, 831 item->mac.addr, 832 item->mac.vid, 833 item->set); 834 break; 835 case FM10K_UC_MAC_REQUEST: 836 hw->mac.ops.update_uc_addr(hw, 837 item->mac.glort, 838 item->mac.addr, 839 item->mac.vid, 840 item->set, 841 0); 842 break; 843 case FM10K_VLAN_REQUEST: 844 hw->mac.ops.update_vlan(hw, 845 item->vlan.vid, 846 item->vlan.vsi, 847 item->set); 848 break; 849 default: 850 break; 851 } 852 853 fm10k_mbx_unlock(interface); 854 855 /* Free the item now that we've sent the update */ 856 kfree(item); 857 } while (true); 858 859 done: 860 WARN_ON(!test_bit(__FM10K_MACVLAN_SCHED, interface->state)); 861 862 /* flush memory to make sure state is correct */ 863 smp_mb__before_atomic(); 864 clear_bit(__FM10K_MACVLAN_SCHED, interface->state); 865 866 /* If a MAC/VLAN request was scheduled since we started, we should 867 * re-schedule. However, there is no reason to re-schedule if there is 868 * no work to do. 869 */ 870 if (test_bit(__FM10K_MACVLAN_REQUEST, interface->state)) 871 fm10k_macvlan_schedule(interface); 872 } 873 874 /** 875 * fm10k_configure_tx_ring - Configure Tx ring after Reset 876 * @interface: board private structure 877 * @ring: structure containing ring specific data 878 * 879 * Configure the Tx descriptor ring after a reset. 880 **/ 881 static void fm10k_configure_tx_ring(struct fm10k_intfc *interface, 882 struct fm10k_ring *ring) 883 { 884 struct fm10k_hw *hw = &interface->hw; 885 u64 tdba = ring->dma; 886 u32 size = ring->count * sizeof(struct fm10k_tx_desc); 887 u32 txint = FM10K_INT_MAP_DISABLE; 888 u32 txdctl = BIT(FM10K_TXDCTL_MAX_TIME_SHIFT) | FM10K_TXDCTL_ENABLE; 889 u8 reg_idx = ring->reg_idx; 890 891 /* disable queue to avoid issues while updating state */ 892 fm10k_write_reg(hw, FM10K_TXDCTL(reg_idx), 0); 893 fm10k_write_flush(hw); 894 895 /* possible poll here to verify ring resources have been cleaned */ 896 897 /* set location and size for descriptor ring */ 898 fm10k_write_reg(hw, FM10K_TDBAL(reg_idx), tdba & DMA_BIT_MASK(32)); 899 fm10k_write_reg(hw, FM10K_TDBAH(reg_idx), tdba >> 32); 900 fm10k_write_reg(hw, FM10K_TDLEN(reg_idx), size); 901 902 /* reset head and tail pointers */ 903 fm10k_write_reg(hw, FM10K_TDH(reg_idx), 0); 904 fm10k_write_reg(hw, FM10K_TDT(reg_idx), 0); 905 906 /* store tail pointer */ 907 ring->tail = &interface->uc_addr[FM10K_TDT(reg_idx)]; 908 909 /* reset ntu and ntc to place SW in sync with hardware */ 910 ring->next_to_clean = 0; 911 ring->next_to_use = 0; 912 913 /* Map interrupt */ 914 if (ring->q_vector) { 915 txint = ring->q_vector->v_idx + NON_Q_VECTORS(hw); 916 txint |= FM10K_INT_MAP_TIMER0; 917 } 918 919 fm10k_write_reg(hw, FM10K_TXINT(reg_idx), txint); 920 921 /* enable use of FTAG bit in Tx descriptor, register is RO for VF */ 922 fm10k_write_reg(hw, FM10K_PFVTCTL(reg_idx), 923 FM10K_PFVTCTL_FTAG_DESC_ENABLE); 924 925 /* Initialize XPS */ 926 if (!test_and_set_bit(__FM10K_TX_XPS_INIT_DONE, ring->state) && 927 ring->q_vector) 928 netif_set_xps_queue(ring->netdev, 929 &ring->q_vector->affinity_mask, 930 ring->queue_index); 931 932 /* enable queue */ 933 fm10k_write_reg(hw, FM10K_TXDCTL(reg_idx), txdctl); 934 } 935 936 /** 937 * fm10k_enable_tx_ring - Verify Tx ring is enabled after configuration 938 * @interface: board private structure 939 * @ring: structure containing ring specific data 940 * 941 * Verify the Tx descriptor ring is ready for transmit. 942 **/ 943 static void fm10k_enable_tx_ring(struct fm10k_intfc *interface, 944 struct fm10k_ring *ring) 945 { 946 struct fm10k_hw *hw = &interface->hw; 947 int wait_loop = 10; 948 u32 txdctl; 949 u8 reg_idx = ring->reg_idx; 950 951 /* if we are already enabled just exit */ 952 if (fm10k_read_reg(hw, FM10K_TXDCTL(reg_idx)) & FM10K_TXDCTL_ENABLE) 953 return; 954 955 /* poll to verify queue is enabled */ 956 do { 957 usleep_range(1000, 2000); 958 txdctl = fm10k_read_reg(hw, FM10K_TXDCTL(reg_idx)); 959 } while (!(txdctl & FM10K_TXDCTL_ENABLE) && --wait_loop); 960 if (!wait_loop) 961 netif_err(interface, drv, interface->netdev, 962 "Could not enable Tx Queue %d\n", reg_idx); 963 } 964 965 /** 966 * fm10k_configure_tx - Configure Transmit Unit after Reset 967 * @interface: board private structure 968 * 969 * Configure the Tx unit of the MAC after a reset. 970 **/ 971 static void fm10k_configure_tx(struct fm10k_intfc *interface) 972 { 973 int i; 974 975 /* Setup the HW Tx Head and Tail descriptor pointers */ 976 for (i = 0; i < interface->num_tx_queues; i++) 977 fm10k_configure_tx_ring(interface, interface->tx_ring[i]); 978 979 /* poll here to verify that Tx rings are now enabled */ 980 for (i = 0; i < interface->num_tx_queues; i++) 981 fm10k_enable_tx_ring(interface, interface->tx_ring[i]); 982 } 983 984 /** 985 * fm10k_configure_rx_ring - Configure Rx ring after Reset 986 * @interface: board private structure 987 * @ring: structure containing ring specific data 988 * 989 * Configure the Rx descriptor ring after a reset. 990 **/ 991 static void fm10k_configure_rx_ring(struct fm10k_intfc *interface, 992 struct fm10k_ring *ring) 993 { 994 u64 rdba = ring->dma; 995 struct fm10k_hw *hw = &interface->hw; 996 u32 size = ring->count * sizeof(union fm10k_rx_desc); 997 u32 rxqctl, rxdctl = FM10K_RXDCTL_WRITE_BACK_MIN_DELAY; 998 u32 srrctl = FM10K_SRRCTL_BUFFER_CHAINING_EN; 999 u32 rxint = FM10K_INT_MAP_DISABLE; 1000 u8 rx_pause = interface->rx_pause; 1001 u8 reg_idx = ring->reg_idx; 1002 1003 /* disable queue to avoid issues while updating state */ 1004 rxqctl = fm10k_read_reg(hw, FM10K_RXQCTL(reg_idx)); 1005 rxqctl &= ~FM10K_RXQCTL_ENABLE; 1006 fm10k_write_reg(hw, FM10K_RXQCTL(reg_idx), rxqctl); 1007 fm10k_write_flush(hw); 1008 1009 /* possible poll here to verify ring resources have been cleaned */ 1010 1011 /* set location and size for descriptor ring */ 1012 fm10k_write_reg(hw, FM10K_RDBAL(reg_idx), rdba & DMA_BIT_MASK(32)); 1013 fm10k_write_reg(hw, FM10K_RDBAH(reg_idx), rdba >> 32); 1014 fm10k_write_reg(hw, FM10K_RDLEN(reg_idx), size); 1015 1016 /* reset head and tail pointers */ 1017 fm10k_write_reg(hw, FM10K_RDH(reg_idx), 0); 1018 fm10k_write_reg(hw, FM10K_RDT(reg_idx), 0); 1019 1020 /* store tail pointer */ 1021 ring->tail = &interface->uc_addr[FM10K_RDT(reg_idx)]; 1022 1023 /* reset ntu and ntc to place SW in sync with hardware */ 1024 ring->next_to_clean = 0; 1025 ring->next_to_use = 0; 1026 ring->next_to_alloc = 0; 1027 1028 /* Configure the Rx buffer size for one buff without split */ 1029 srrctl |= FM10K_RX_BUFSZ >> FM10K_SRRCTL_BSIZEPKT_SHIFT; 1030 1031 /* Configure the Rx ring to suppress loopback packets */ 1032 srrctl |= FM10K_SRRCTL_LOOPBACK_SUPPRESS; 1033 fm10k_write_reg(hw, FM10K_SRRCTL(reg_idx), srrctl); 1034 1035 /* Enable drop on empty */ 1036 #ifdef CONFIG_DCB 1037 if (interface->pfc_en) 1038 rx_pause = interface->pfc_en; 1039 #endif 1040 if (!(rx_pause & BIT(ring->qos_pc))) 1041 rxdctl |= FM10K_RXDCTL_DROP_ON_EMPTY; 1042 1043 fm10k_write_reg(hw, FM10K_RXDCTL(reg_idx), rxdctl); 1044 1045 /* assign default VLAN to queue */ 1046 ring->vid = hw->mac.default_vid; 1047 1048 /* if we have an active VLAN, disable default VLAN ID */ 1049 if (test_bit(hw->mac.default_vid, interface->active_vlans)) 1050 ring->vid |= FM10K_VLAN_CLEAR; 1051 1052 /* Map interrupt */ 1053 if (ring->q_vector) { 1054 rxint = ring->q_vector->v_idx + NON_Q_VECTORS(hw); 1055 rxint |= FM10K_INT_MAP_TIMER1; 1056 } 1057 1058 fm10k_write_reg(hw, FM10K_RXINT(reg_idx), rxint); 1059 1060 /* enable queue */ 1061 rxqctl = fm10k_read_reg(hw, FM10K_RXQCTL(reg_idx)); 1062 rxqctl |= FM10K_RXQCTL_ENABLE; 1063 fm10k_write_reg(hw, FM10K_RXQCTL(reg_idx), rxqctl); 1064 1065 /* place buffers on ring for receive data */ 1066 fm10k_alloc_rx_buffers(ring, fm10k_desc_unused(ring)); 1067 } 1068 1069 /** 1070 * fm10k_update_rx_drop_en - Configures the drop enable bits for Rx rings 1071 * @interface: board private structure 1072 * 1073 * Configure the drop enable bits for the Rx rings. 1074 **/ 1075 void fm10k_update_rx_drop_en(struct fm10k_intfc *interface) 1076 { 1077 struct fm10k_hw *hw = &interface->hw; 1078 u8 rx_pause = interface->rx_pause; 1079 int i; 1080 1081 #ifdef CONFIG_DCB 1082 if (interface->pfc_en) 1083 rx_pause = interface->pfc_en; 1084 1085 #endif 1086 for (i = 0; i < interface->num_rx_queues; i++) { 1087 struct fm10k_ring *ring = interface->rx_ring[i]; 1088 u32 rxdctl = FM10K_RXDCTL_WRITE_BACK_MIN_DELAY; 1089 u8 reg_idx = ring->reg_idx; 1090 1091 if (!(rx_pause & BIT(ring->qos_pc))) 1092 rxdctl |= FM10K_RXDCTL_DROP_ON_EMPTY; 1093 1094 fm10k_write_reg(hw, FM10K_RXDCTL(reg_idx), rxdctl); 1095 } 1096 } 1097 1098 /** 1099 * fm10k_configure_dglort - Configure Receive DGLORT after reset 1100 * @interface: board private structure 1101 * 1102 * Configure the DGLORT description and RSS tables. 1103 **/ 1104 static void fm10k_configure_dglort(struct fm10k_intfc *interface) 1105 { 1106 struct fm10k_dglort_cfg dglort = { 0 }; 1107 struct fm10k_hw *hw = &interface->hw; 1108 int i; 1109 u32 mrqc; 1110 1111 /* Fill out hash function seeds */ 1112 for (i = 0; i < FM10K_RSSRK_SIZE; i++) 1113 fm10k_write_reg(hw, FM10K_RSSRK(0, i), interface->rssrk[i]); 1114 1115 /* Write RETA table to hardware */ 1116 for (i = 0; i < FM10K_RETA_SIZE; i++) 1117 fm10k_write_reg(hw, FM10K_RETA(0, i), interface->reta[i]); 1118 1119 /* Generate RSS hash based on packet types, TCP/UDP 1120 * port numbers and/or IPv4/v6 src and dst addresses 1121 */ 1122 mrqc = FM10K_MRQC_IPV4 | 1123 FM10K_MRQC_TCP_IPV4 | 1124 FM10K_MRQC_IPV6 | 1125 FM10K_MRQC_TCP_IPV6; 1126 1127 if (test_bit(FM10K_FLAG_RSS_FIELD_IPV4_UDP, interface->flags)) 1128 mrqc |= FM10K_MRQC_UDP_IPV4; 1129 if (test_bit(FM10K_FLAG_RSS_FIELD_IPV6_UDP, interface->flags)) 1130 mrqc |= FM10K_MRQC_UDP_IPV6; 1131 1132 fm10k_write_reg(hw, FM10K_MRQC(0), mrqc); 1133 1134 /* configure default DGLORT mapping for RSS/DCB */ 1135 dglort.inner_rss = 1; 1136 dglort.rss_l = fls(interface->ring_feature[RING_F_RSS].mask); 1137 dglort.pc_l = fls(interface->ring_feature[RING_F_QOS].mask); 1138 hw->mac.ops.configure_dglort_map(hw, &dglort); 1139 1140 /* assign GLORT per queue for queue mapped testing */ 1141 if (interface->glort_count > 64) { 1142 memset(&dglort, 0, sizeof(dglort)); 1143 dglort.inner_rss = 1; 1144 dglort.glort = interface->glort + 64; 1145 dglort.idx = fm10k_dglort_pf_queue; 1146 dglort.queue_l = fls(interface->num_rx_queues - 1); 1147 hw->mac.ops.configure_dglort_map(hw, &dglort); 1148 } 1149 1150 /* assign glort value for RSS/DCB specific to this interface */ 1151 memset(&dglort, 0, sizeof(dglort)); 1152 dglort.inner_rss = 1; 1153 dglort.glort = interface->glort; 1154 dglort.rss_l = fls(interface->ring_feature[RING_F_RSS].mask); 1155 dglort.pc_l = fls(interface->ring_feature[RING_F_QOS].mask); 1156 /* configure DGLORT mapping for RSS/DCB */ 1157 dglort.idx = fm10k_dglort_pf_rss; 1158 if (interface->l2_accel) 1159 dglort.shared_l = fls(interface->l2_accel->size); 1160 hw->mac.ops.configure_dglort_map(hw, &dglort); 1161 } 1162 1163 /** 1164 * fm10k_configure_rx - Configure Receive Unit after Reset 1165 * @interface: board private structure 1166 * 1167 * Configure the Rx unit of the MAC after a reset. 1168 **/ 1169 static void fm10k_configure_rx(struct fm10k_intfc *interface) 1170 { 1171 int i; 1172 1173 /* Configure SWPRI to PC map */ 1174 fm10k_configure_swpri_map(interface); 1175 1176 /* Configure RSS and DGLORT map */ 1177 fm10k_configure_dglort(interface); 1178 1179 /* Setup the HW Rx Head and Tail descriptor pointers */ 1180 for (i = 0; i < interface->num_rx_queues; i++) 1181 fm10k_configure_rx_ring(interface, interface->rx_ring[i]); 1182 1183 /* possible poll here to verify that Rx rings are now enabled */ 1184 } 1185 1186 static void fm10k_napi_enable_all(struct fm10k_intfc *interface) 1187 { 1188 struct fm10k_q_vector *q_vector; 1189 int q_idx; 1190 1191 for (q_idx = 0; q_idx < interface->num_q_vectors; q_idx++) { 1192 q_vector = interface->q_vector[q_idx]; 1193 napi_enable(&q_vector->napi); 1194 } 1195 } 1196 1197 static irqreturn_t fm10k_msix_clean_rings(int __always_unused irq, void *data) 1198 { 1199 struct fm10k_q_vector *q_vector = data; 1200 1201 if (q_vector->rx.count || q_vector->tx.count) 1202 napi_schedule_irqoff(&q_vector->napi); 1203 1204 return IRQ_HANDLED; 1205 } 1206 1207 static irqreturn_t fm10k_msix_mbx_vf(int __always_unused irq, void *data) 1208 { 1209 struct fm10k_intfc *interface = data; 1210 struct fm10k_hw *hw = &interface->hw; 1211 struct fm10k_mbx_info *mbx = &hw->mbx; 1212 1213 /* re-enable mailbox interrupt and indicate 20us delay */ 1214 fm10k_write_reg(hw, FM10K_VFITR(FM10K_MBX_VECTOR), 1215 (FM10K_MBX_INT_DELAY >> hw->mac.itr_scale) | 1216 FM10K_ITR_ENABLE); 1217 1218 /* service upstream mailbox */ 1219 if (fm10k_mbx_trylock(interface)) { 1220 mbx->ops.process(hw, mbx); 1221 fm10k_mbx_unlock(interface); 1222 } 1223 1224 hw->mac.get_host_state = true; 1225 fm10k_service_event_schedule(interface); 1226 1227 return IRQ_HANDLED; 1228 } 1229 1230 #ifdef CONFIG_NET_POLL_CONTROLLER 1231 /** 1232 * fm10k_netpoll - A Polling 'interrupt' handler 1233 * @netdev: network interface device structure 1234 * 1235 * This is used by netconsole to send skbs without having to re-enable 1236 * interrupts. It's not called while the normal interrupt routine is executing. 1237 **/ 1238 void fm10k_netpoll(struct net_device *netdev) 1239 { 1240 struct fm10k_intfc *interface = netdev_priv(netdev); 1241 int i; 1242 1243 /* if interface is down do nothing */ 1244 if (test_bit(__FM10K_DOWN, interface->state)) 1245 return; 1246 1247 for (i = 0; i < interface->num_q_vectors; i++) 1248 fm10k_msix_clean_rings(0, interface->q_vector[i]); 1249 } 1250 1251 #endif 1252 #define FM10K_ERR_MSG(type) case (type): error = #type; break 1253 static void fm10k_handle_fault(struct fm10k_intfc *interface, int type, 1254 struct fm10k_fault *fault) 1255 { 1256 struct pci_dev *pdev = interface->pdev; 1257 struct fm10k_hw *hw = &interface->hw; 1258 struct fm10k_iov_data *iov_data = interface->iov_data; 1259 char *error; 1260 1261 switch (type) { 1262 case FM10K_PCA_FAULT: 1263 switch (fault->type) { 1264 default: 1265 error = "Unknown PCA error"; 1266 break; 1267 FM10K_ERR_MSG(PCA_NO_FAULT); 1268 FM10K_ERR_MSG(PCA_UNMAPPED_ADDR); 1269 FM10K_ERR_MSG(PCA_BAD_QACCESS_PF); 1270 FM10K_ERR_MSG(PCA_BAD_QACCESS_VF); 1271 FM10K_ERR_MSG(PCA_MALICIOUS_REQ); 1272 FM10K_ERR_MSG(PCA_POISONED_TLP); 1273 FM10K_ERR_MSG(PCA_TLP_ABORT); 1274 } 1275 break; 1276 case FM10K_THI_FAULT: 1277 switch (fault->type) { 1278 default: 1279 error = "Unknown THI error"; 1280 break; 1281 FM10K_ERR_MSG(THI_NO_FAULT); 1282 FM10K_ERR_MSG(THI_MAL_DIS_Q_FAULT); 1283 } 1284 break; 1285 case FM10K_FUM_FAULT: 1286 switch (fault->type) { 1287 default: 1288 error = "Unknown FUM error"; 1289 break; 1290 FM10K_ERR_MSG(FUM_NO_FAULT); 1291 FM10K_ERR_MSG(FUM_UNMAPPED_ADDR); 1292 FM10K_ERR_MSG(FUM_BAD_VF_QACCESS); 1293 FM10K_ERR_MSG(FUM_ADD_DECODE_ERR); 1294 FM10K_ERR_MSG(FUM_RO_ERROR); 1295 FM10K_ERR_MSG(FUM_QPRC_CRC_ERROR); 1296 FM10K_ERR_MSG(FUM_CSR_TIMEOUT); 1297 FM10K_ERR_MSG(FUM_INVALID_TYPE); 1298 FM10K_ERR_MSG(FUM_INVALID_LENGTH); 1299 FM10K_ERR_MSG(FUM_INVALID_BE); 1300 FM10K_ERR_MSG(FUM_INVALID_ALIGN); 1301 } 1302 break; 1303 default: 1304 error = "Undocumented fault"; 1305 break; 1306 } 1307 1308 dev_warn(&pdev->dev, 1309 "%s Address: 0x%llx SpecInfo: 0x%x Func: %02x.%0x\n", 1310 error, fault->address, fault->specinfo, 1311 PCI_SLOT(fault->func), PCI_FUNC(fault->func)); 1312 1313 /* For VF faults, clear out the respective LPORT, reset the queue 1314 * resources, and then reconnect to the mailbox. This allows the 1315 * VF in question to resume behavior. For transient faults that are 1316 * the result of non-malicious behavior this will log the fault and 1317 * allow the VF to resume functionality. Obviously for malicious VFs 1318 * they will be able to attempt malicious behavior again. In this 1319 * case, the system administrator will need to step in and manually 1320 * remove or disable the VF in question. 1321 */ 1322 if (fault->func && iov_data) { 1323 int vf = fault->func - 1; 1324 struct fm10k_vf_info *vf_info = &iov_data->vf_info[vf]; 1325 1326 hw->iov.ops.reset_lport(hw, vf_info); 1327 hw->iov.ops.reset_resources(hw, vf_info); 1328 1329 /* reset_lport disables the VF, so re-enable it */ 1330 hw->iov.ops.set_lport(hw, vf_info, vf, 1331 FM10K_VF_FLAG_MULTI_CAPABLE); 1332 1333 /* reset_resources will disconnect from the mbx */ 1334 vf_info->mbx.ops.connect(hw, &vf_info->mbx); 1335 } 1336 } 1337 1338 static void fm10k_report_fault(struct fm10k_intfc *interface, u32 eicr) 1339 { 1340 struct fm10k_hw *hw = &interface->hw; 1341 struct fm10k_fault fault = { 0 }; 1342 int type, err; 1343 1344 for (eicr &= FM10K_EICR_FAULT_MASK, type = FM10K_PCA_FAULT; 1345 eicr; 1346 eicr >>= 1, type += FM10K_FAULT_SIZE) { 1347 /* only check if there is an error reported */ 1348 if (!(eicr & 0x1)) 1349 continue; 1350 1351 /* retrieve fault info */ 1352 err = hw->mac.ops.get_fault(hw, type, &fault); 1353 if (err) { 1354 dev_err(&interface->pdev->dev, 1355 "error reading fault\n"); 1356 continue; 1357 } 1358 1359 fm10k_handle_fault(interface, type, &fault); 1360 } 1361 } 1362 1363 static void fm10k_reset_drop_on_empty(struct fm10k_intfc *interface, u32 eicr) 1364 { 1365 struct fm10k_hw *hw = &interface->hw; 1366 const u32 rxdctl = FM10K_RXDCTL_WRITE_BACK_MIN_DELAY; 1367 u32 maxholdq; 1368 int q; 1369 1370 if (!(eicr & FM10K_EICR_MAXHOLDTIME)) 1371 return; 1372 1373 maxholdq = fm10k_read_reg(hw, FM10K_MAXHOLDQ(7)); 1374 if (maxholdq) 1375 fm10k_write_reg(hw, FM10K_MAXHOLDQ(7), maxholdq); 1376 for (q = 255;;) { 1377 if (maxholdq & BIT(31)) { 1378 if (q < FM10K_MAX_QUEUES_PF) { 1379 interface->rx_overrun_pf++; 1380 fm10k_write_reg(hw, FM10K_RXDCTL(q), rxdctl); 1381 } else { 1382 interface->rx_overrun_vf++; 1383 } 1384 } 1385 1386 maxholdq *= 2; 1387 if (!maxholdq) 1388 q &= ~(32 - 1); 1389 1390 if (!q) 1391 break; 1392 1393 if (q-- % 32) 1394 continue; 1395 1396 maxholdq = fm10k_read_reg(hw, FM10K_MAXHOLDQ(q / 32)); 1397 if (maxholdq) 1398 fm10k_write_reg(hw, FM10K_MAXHOLDQ(q / 32), maxholdq); 1399 } 1400 } 1401 1402 static irqreturn_t fm10k_msix_mbx_pf(int __always_unused irq, void *data) 1403 { 1404 struct fm10k_intfc *interface = data; 1405 struct fm10k_hw *hw = &interface->hw; 1406 struct fm10k_mbx_info *mbx = &hw->mbx; 1407 u32 eicr; 1408 s32 err = 0; 1409 1410 /* unmask any set bits related to this interrupt */ 1411 eicr = fm10k_read_reg(hw, FM10K_EICR); 1412 fm10k_write_reg(hw, FM10K_EICR, eicr & (FM10K_EICR_MAILBOX | 1413 FM10K_EICR_SWITCHREADY | 1414 FM10K_EICR_SWITCHNOTREADY)); 1415 1416 /* report any faults found to the message log */ 1417 fm10k_report_fault(interface, eicr); 1418 1419 /* reset any queues disabled due to receiver overrun */ 1420 fm10k_reset_drop_on_empty(interface, eicr); 1421 1422 /* service mailboxes */ 1423 if (fm10k_mbx_trylock(interface)) { 1424 err = mbx->ops.process(hw, mbx); 1425 /* handle VFLRE events */ 1426 fm10k_iov_event(interface); 1427 fm10k_mbx_unlock(interface); 1428 } 1429 1430 if (err == FM10K_ERR_RESET_REQUESTED) 1431 set_bit(FM10K_FLAG_RESET_REQUESTED, interface->flags); 1432 1433 /* if switch toggled state we should reset GLORTs */ 1434 if (eicr & FM10K_EICR_SWITCHNOTREADY) { 1435 /* force link down for at least 4 seconds */ 1436 interface->link_down_event = jiffies + (4 * HZ); 1437 set_bit(__FM10K_LINK_DOWN, interface->state); 1438 1439 /* reset dglort_map back to no config */ 1440 hw->mac.dglort_map = FM10K_DGLORTMAP_NONE; 1441 } 1442 1443 /* we should validate host state after interrupt event */ 1444 hw->mac.get_host_state = true; 1445 1446 /* validate host state, and handle VF mailboxes in the service task */ 1447 fm10k_service_event_schedule(interface); 1448 1449 /* re-enable mailbox interrupt and indicate 20us delay */ 1450 fm10k_write_reg(hw, FM10K_ITR(FM10K_MBX_VECTOR), 1451 (FM10K_MBX_INT_DELAY >> hw->mac.itr_scale) | 1452 FM10K_ITR_ENABLE); 1453 1454 return IRQ_HANDLED; 1455 } 1456 1457 void fm10k_mbx_free_irq(struct fm10k_intfc *interface) 1458 { 1459 struct fm10k_hw *hw = &interface->hw; 1460 struct msix_entry *entry; 1461 int itr_reg; 1462 1463 /* no mailbox IRQ to free if MSI-X is not enabled */ 1464 if (!interface->msix_entries) 1465 return; 1466 1467 entry = &interface->msix_entries[FM10K_MBX_VECTOR]; 1468 1469 /* disconnect the mailbox */ 1470 hw->mbx.ops.disconnect(hw, &hw->mbx); 1471 1472 /* disable Mailbox cause */ 1473 if (hw->mac.type == fm10k_mac_pf) { 1474 fm10k_write_reg(hw, FM10K_EIMR, 1475 FM10K_EIMR_DISABLE(PCA_FAULT) | 1476 FM10K_EIMR_DISABLE(FUM_FAULT) | 1477 FM10K_EIMR_DISABLE(MAILBOX) | 1478 FM10K_EIMR_DISABLE(SWITCHREADY) | 1479 FM10K_EIMR_DISABLE(SWITCHNOTREADY) | 1480 FM10K_EIMR_DISABLE(SRAMERROR) | 1481 FM10K_EIMR_DISABLE(VFLR) | 1482 FM10K_EIMR_DISABLE(MAXHOLDTIME)); 1483 itr_reg = FM10K_ITR(FM10K_MBX_VECTOR); 1484 } else { 1485 itr_reg = FM10K_VFITR(FM10K_MBX_VECTOR); 1486 } 1487 1488 fm10k_write_reg(hw, itr_reg, FM10K_ITR_MASK_SET); 1489 1490 free_irq(entry->vector, interface); 1491 } 1492 1493 static s32 fm10k_mbx_mac_addr(struct fm10k_hw *hw, u32 **results, 1494 struct fm10k_mbx_info *mbx) 1495 { 1496 bool vlan_override = hw->mac.vlan_override; 1497 u16 default_vid = hw->mac.default_vid; 1498 struct fm10k_intfc *interface; 1499 s32 err; 1500 1501 err = fm10k_msg_mac_vlan_vf(hw, results, mbx); 1502 if (err) 1503 return err; 1504 1505 interface = container_of(hw, struct fm10k_intfc, hw); 1506 1507 /* MAC was changed so we need reset */ 1508 if (is_valid_ether_addr(hw->mac.perm_addr) && 1509 !ether_addr_equal(hw->mac.perm_addr, hw->mac.addr)) 1510 set_bit(FM10K_FLAG_RESET_REQUESTED, interface->flags); 1511 1512 /* VLAN override was changed, or default VLAN changed */ 1513 if ((vlan_override != hw->mac.vlan_override) || 1514 (default_vid != hw->mac.default_vid)) 1515 set_bit(FM10K_FLAG_RESET_REQUESTED, interface->flags); 1516 1517 return 0; 1518 } 1519 1520 /* generic error handler for mailbox issues */ 1521 static s32 fm10k_mbx_error(struct fm10k_hw *hw, u32 **results, 1522 struct fm10k_mbx_info __always_unused *mbx) 1523 { 1524 struct fm10k_intfc *interface; 1525 struct pci_dev *pdev; 1526 1527 interface = container_of(hw, struct fm10k_intfc, hw); 1528 pdev = interface->pdev; 1529 1530 dev_err(&pdev->dev, "Unknown message ID %u\n", 1531 **results & FM10K_TLV_ID_MASK); 1532 1533 return 0; 1534 } 1535 1536 static const struct fm10k_msg_data vf_mbx_data[] = { 1537 FM10K_TLV_MSG_TEST_HANDLER(fm10k_tlv_msg_test), 1538 FM10K_VF_MSG_MAC_VLAN_HANDLER(fm10k_mbx_mac_addr), 1539 FM10K_VF_MSG_LPORT_STATE_HANDLER(fm10k_msg_lport_state_vf), 1540 FM10K_TLV_MSG_ERROR_HANDLER(fm10k_mbx_error), 1541 }; 1542 1543 static int fm10k_mbx_request_irq_vf(struct fm10k_intfc *interface) 1544 { 1545 struct msix_entry *entry = &interface->msix_entries[FM10K_MBX_VECTOR]; 1546 struct net_device *dev = interface->netdev; 1547 struct fm10k_hw *hw = &interface->hw; 1548 int err; 1549 1550 /* Use timer0 for interrupt moderation on the mailbox */ 1551 u32 itr = entry->entry | FM10K_INT_MAP_TIMER0; 1552 1553 /* register mailbox handlers */ 1554 err = hw->mbx.ops.register_handlers(&hw->mbx, vf_mbx_data); 1555 if (err) 1556 return err; 1557 1558 /* request the IRQ */ 1559 err = request_irq(entry->vector, fm10k_msix_mbx_vf, 0, 1560 dev->name, interface); 1561 if (err) { 1562 netif_err(interface, probe, dev, 1563 "request_irq for msix_mbx failed: %d\n", err); 1564 return err; 1565 } 1566 1567 /* map all of the interrupt sources */ 1568 fm10k_write_reg(hw, FM10K_VFINT_MAP, itr); 1569 1570 /* enable interrupt */ 1571 fm10k_write_reg(hw, FM10K_VFITR(entry->entry), FM10K_ITR_ENABLE); 1572 1573 return 0; 1574 } 1575 1576 static s32 fm10k_lport_map(struct fm10k_hw *hw, u32 **results, 1577 struct fm10k_mbx_info *mbx) 1578 { 1579 struct fm10k_intfc *interface; 1580 u32 dglort_map = hw->mac.dglort_map; 1581 s32 err; 1582 1583 interface = container_of(hw, struct fm10k_intfc, hw); 1584 1585 err = fm10k_msg_err_pf(hw, results, mbx); 1586 if (!err && hw->swapi.status) { 1587 /* force link down for a reasonable delay */ 1588 interface->link_down_event = jiffies + (2 * HZ); 1589 set_bit(__FM10K_LINK_DOWN, interface->state); 1590 1591 /* reset dglort_map back to no config */ 1592 hw->mac.dglort_map = FM10K_DGLORTMAP_NONE; 1593 1594 fm10k_service_event_schedule(interface); 1595 1596 /* prevent overloading kernel message buffer */ 1597 if (interface->lport_map_failed) 1598 return 0; 1599 1600 interface->lport_map_failed = true; 1601 1602 if (hw->swapi.status == FM10K_MSG_ERR_PEP_NOT_SCHEDULED) 1603 dev_warn(&interface->pdev->dev, 1604 "cannot obtain link because the host interface is configured for a PCIe host interface bandwidth of zero\n"); 1605 dev_warn(&interface->pdev->dev, 1606 "request logical port map failed: %d\n", 1607 hw->swapi.status); 1608 1609 return 0; 1610 } 1611 1612 err = fm10k_msg_lport_map_pf(hw, results, mbx); 1613 if (err) 1614 return err; 1615 1616 interface->lport_map_failed = false; 1617 1618 /* we need to reset if port count was just updated */ 1619 if (dglort_map != hw->mac.dglort_map) 1620 set_bit(FM10K_FLAG_RESET_REQUESTED, interface->flags); 1621 1622 return 0; 1623 } 1624 1625 static s32 fm10k_update_pvid(struct fm10k_hw *hw, u32 **results, 1626 struct fm10k_mbx_info __always_unused *mbx) 1627 { 1628 struct fm10k_intfc *interface; 1629 u16 glort, pvid; 1630 u32 pvid_update; 1631 s32 err; 1632 1633 err = fm10k_tlv_attr_get_u32(results[FM10K_PF_ATTR_ID_UPDATE_PVID], 1634 &pvid_update); 1635 if (err) 1636 return err; 1637 1638 /* extract values from the pvid update */ 1639 glort = FM10K_MSG_HDR_FIELD_GET(pvid_update, UPDATE_PVID_GLORT); 1640 pvid = FM10K_MSG_HDR_FIELD_GET(pvid_update, UPDATE_PVID_PVID); 1641 1642 /* if glort is not valid return error */ 1643 if (!fm10k_glort_valid_pf(hw, glort)) 1644 return FM10K_ERR_PARAM; 1645 1646 /* verify VLAN ID is valid */ 1647 if (pvid >= FM10K_VLAN_TABLE_VID_MAX) 1648 return FM10K_ERR_PARAM; 1649 1650 interface = container_of(hw, struct fm10k_intfc, hw); 1651 1652 /* check to see if this belongs to one of the VFs */ 1653 err = fm10k_iov_update_pvid(interface, glort, pvid); 1654 if (!err) 1655 return 0; 1656 1657 /* we need to reset if default VLAN was just updated */ 1658 if (pvid != hw->mac.default_vid) 1659 set_bit(FM10K_FLAG_RESET_REQUESTED, interface->flags); 1660 1661 hw->mac.default_vid = pvid; 1662 1663 return 0; 1664 } 1665 1666 static const struct fm10k_msg_data pf_mbx_data[] = { 1667 FM10K_PF_MSG_ERR_HANDLER(XCAST_MODES, fm10k_msg_err_pf), 1668 FM10K_PF_MSG_ERR_HANDLER(UPDATE_MAC_FWD_RULE, fm10k_msg_err_pf), 1669 FM10K_PF_MSG_LPORT_MAP_HANDLER(fm10k_lport_map), 1670 FM10K_PF_MSG_ERR_HANDLER(LPORT_CREATE, fm10k_msg_err_pf), 1671 FM10K_PF_MSG_ERR_HANDLER(LPORT_DELETE, fm10k_msg_err_pf), 1672 FM10K_PF_MSG_UPDATE_PVID_HANDLER(fm10k_update_pvid), 1673 FM10K_TLV_MSG_ERROR_HANDLER(fm10k_mbx_error), 1674 }; 1675 1676 static int fm10k_mbx_request_irq_pf(struct fm10k_intfc *interface) 1677 { 1678 struct msix_entry *entry = &interface->msix_entries[FM10K_MBX_VECTOR]; 1679 struct net_device *dev = interface->netdev; 1680 struct fm10k_hw *hw = &interface->hw; 1681 int err; 1682 1683 /* Use timer0 for interrupt moderation on the mailbox */ 1684 u32 mbx_itr = entry->entry | FM10K_INT_MAP_TIMER0; 1685 u32 other_itr = entry->entry | FM10K_INT_MAP_IMMEDIATE; 1686 1687 /* register mailbox handlers */ 1688 err = hw->mbx.ops.register_handlers(&hw->mbx, pf_mbx_data); 1689 if (err) 1690 return err; 1691 1692 /* request the IRQ */ 1693 err = request_irq(entry->vector, fm10k_msix_mbx_pf, 0, 1694 dev->name, interface); 1695 if (err) { 1696 netif_err(interface, probe, dev, 1697 "request_irq for msix_mbx failed: %d\n", err); 1698 return err; 1699 } 1700 1701 /* Enable interrupts w/ no moderation for "other" interrupts */ 1702 fm10k_write_reg(hw, FM10K_INT_MAP(fm10k_int_pcie_fault), other_itr); 1703 fm10k_write_reg(hw, FM10K_INT_MAP(fm10k_int_switch_up_down), other_itr); 1704 fm10k_write_reg(hw, FM10K_INT_MAP(fm10k_int_sram), other_itr); 1705 fm10k_write_reg(hw, FM10K_INT_MAP(fm10k_int_max_hold_time), other_itr); 1706 fm10k_write_reg(hw, FM10K_INT_MAP(fm10k_int_vflr), other_itr); 1707 1708 /* Enable interrupts w/ moderation for mailbox */ 1709 fm10k_write_reg(hw, FM10K_INT_MAP(fm10k_int_mailbox), mbx_itr); 1710 1711 /* Enable individual interrupt causes */ 1712 fm10k_write_reg(hw, FM10K_EIMR, FM10K_EIMR_ENABLE(PCA_FAULT) | 1713 FM10K_EIMR_ENABLE(FUM_FAULT) | 1714 FM10K_EIMR_ENABLE(MAILBOX) | 1715 FM10K_EIMR_ENABLE(SWITCHREADY) | 1716 FM10K_EIMR_ENABLE(SWITCHNOTREADY) | 1717 FM10K_EIMR_ENABLE(SRAMERROR) | 1718 FM10K_EIMR_ENABLE(VFLR) | 1719 FM10K_EIMR_ENABLE(MAXHOLDTIME)); 1720 1721 /* enable interrupt */ 1722 fm10k_write_reg(hw, FM10K_ITR(entry->entry), FM10K_ITR_ENABLE); 1723 1724 return 0; 1725 } 1726 1727 int fm10k_mbx_request_irq(struct fm10k_intfc *interface) 1728 { 1729 struct fm10k_hw *hw = &interface->hw; 1730 int err; 1731 1732 /* enable Mailbox cause */ 1733 if (hw->mac.type == fm10k_mac_pf) 1734 err = fm10k_mbx_request_irq_pf(interface); 1735 else 1736 err = fm10k_mbx_request_irq_vf(interface); 1737 if (err) 1738 return err; 1739 1740 /* connect mailbox */ 1741 err = hw->mbx.ops.connect(hw, &hw->mbx); 1742 1743 /* if the mailbox failed to connect, then free IRQ */ 1744 if (err) 1745 fm10k_mbx_free_irq(interface); 1746 1747 return err; 1748 } 1749 1750 /** 1751 * fm10k_qv_free_irq - release interrupts associated with queue vectors 1752 * @interface: board private structure 1753 * 1754 * Release all interrupts associated with this interface 1755 **/ 1756 void fm10k_qv_free_irq(struct fm10k_intfc *interface) 1757 { 1758 int vector = interface->num_q_vectors; 1759 struct fm10k_hw *hw = &interface->hw; 1760 struct msix_entry *entry; 1761 1762 entry = &interface->msix_entries[NON_Q_VECTORS(hw) + vector]; 1763 1764 while (vector) { 1765 struct fm10k_q_vector *q_vector; 1766 1767 vector--; 1768 entry--; 1769 q_vector = interface->q_vector[vector]; 1770 1771 if (!q_vector->tx.count && !q_vector->rx.count) 1772 continue; 1773 1774 /* clear the affinity_mask in the IRQ descriptor */ 1775 irq_set_affinity_hint(entry->vector, NULL); 1776 1777 /* disable interrupts */ 1778 writel(FM10K_ITR_MASK_SET, q_vector->itr); 1779 1780 free_irq(entry->vector, q_vector); 1781 } 1782 } 1783 1784 /** 1785 * fm10k_qv_request_irq - initialize interrupts for queue vectors 1786 * @interface: board private structure 1787 * 1788 * Attempts to configure interrupts using the best available 1789 * capabilities of the hardware and kernel. 1790 **/ 1791 int fm10k_qv_request_irq(struct fm10k_intfc *interface) 1792 { 1793 struct net_device *dev = interface->netdev; 1794 struct fm10k_hw *hw = &interface->hw; 1795 struct msix_entry *entry; 1796 unsigned int ri = 0, ti = 0; 1797 int vector, err; 1798 1799 entry = &interface->msix_entries[NON_Q_VECTORS(hw)]; 1800 1801 for (vector = 0; vector < interface->num_q_vectors; vector++) { 1802 struct fm10k_q_vector *q_vector = interface->q_vector[vector]; 1803 1804 /* name the vector */ 1805 if (q_vector->tx.count && q_vector->rx.count) { 1806 snprintf(q_vector->name, sizeof(q_vector->name), 1807 "%s-TxRx-%u", dev->name, ri++); 1808 ti++; 1809 } else if (q_vector->rx.count) { 1810 snprintf(q_vector->name, sizeof(q_vector->name), 1811 "%s-rx-%u", dev->name, ri++); 1812 } else if (q_vector->tx.count) { 1813 snprintf(q_vector->name, sizeof(q_vector->name), 1814 "%s-tx-%u", dev->name, ti++); 1815 } else { 1816 /* skip this unused q_vector */ 1817 continue; 1818 } 1819 1820 /* Assign ITR register to q_vector */ 1821 q_vector->itr = (hw->mac.type == fm10k_mac_pf) ? 1822 &interface->uc_addr[FM10K_ITR(entry->entry)] : 1823 &interface->uc_addr[FM10K_VFITR(entry->entry)]; 1824 1825 /* request the IRQ */ 1826 err = request_irq(entry->vector, &fm10k_msix_clean_rings, 0, 1827 q_vector->name, q_vector); 1828 if (err) { 1829 netif_err(interface, probe, dev, 1830 "request_irq failed for MSIX interrupt Error: %d\n", 1831 err); 1832 goto err_out; 1833 } 1834 1835 /* assign the mask for this irq */ 1836 irq_set_affinity_hint(entry->vector, &q_vector->affinity_mask); 1837 1838 /* Enable q_vector */ 1839 writel(FM10K_ITR_ENABLE, q_vector->itr); 1840 1841 entry++; 1842 } 1843 1844 return 0; 1845 1846 err_out: 1847 /* wind through the ring freeing all entries and vectors */ 1848 while (vector) { 1849 struct fm10k_q_vector *q_vector; 1850 1851 entry--; 1852 vector--; 1853 q_vector = interface->q_vector[vector]; 1854 1855 if (!q_vector->tx.count && !q_vector->rx.count) 1856 continue; 1857 1858 /* clear the affinity_mask in the IRQ descriptor */ 1859 irq_set_affinity_hint(entry->vector, NULL); 1860 1861 /* disable interrupts */ 1862 writel(FM10K_ITR_MASK_SET, q_vector->itr); 1863 1864 free_irq(entry->vector, q_vector); 1865 } 1866 1867 return err; 1868 } 1869 1870 void fm10k_up(struct fm10k_intfc *interface) 1871 { 1872 struct fm10k_hw *hw = &interface->hw; 1873 1874 /* Enable Tx/Rx DMA */ 1875 hw->mac.ops.start_hw(hw); 1876 1877 /* configure Tx descriptor rings */ 1878 fm10k_configure_tx(interface); 1879 1880 /* configure Rx descriptor rings */ 1881 fm10k_configure_rx(interface); 1882 1883 /* configure interrupts */ 1884 hw->mac.ops.update_int_moderator(hw); 1885 1886 /* enable statistics capture again */ 1887 clear_bit(__FM10K_UPDATING_STATS, interface->state); 1888 1889 /* clear down bit to indicate we are ready to go */ 1890 clear_bit(__FM10K_DOWN, interface->state); 1891 1892 /* enable polling cleanups */ 1893 fm10k_napi_enable_all(interface); 1894 1895 /* re-establish Rx filters */ 1896 fm10k_restore_rx_state(interface); 1897 1898 /* enable transmits */ 1899 netif_tx_start_all_queues(interface->netdev); 1900 1901 /* kick off the service timer now */ 1902 hw->mac.get_host_state = true; 1903 mod_timer(&interface->service_timer, jiffies); 1904 } 1905 1906 static void fm10k_napi_disable_all(struct fm10k_intfc *interface) 1907 { 1908 struct fm10k_q_vector *q_vector; 1909 int q_idx; 1910 1911 for (q_idx = 0; q_idx < interface->num_q_vectors; q_idx++) { 1912 q_vector = interface->q_vector[q_idx]; 1913 napi_disable(&q_vector->napi); 1914 } 1915 } 1916 1917 void fm10k_down(struct fm10k_intfc *interface) 1918 { 1919 struct net_device *netdev = interface->netdev; 1920 struct fm10k_hw *hw = &interface->hw; 1921 int err, i = 0, count = 0; 1922 1923 /* signal that we are down to the interrupt handler and service task */ 1924 if (test_and_set_bit(__FM10K_DOWN, interface->state)) 1925 return; 1926 1927 /* call carrier off first to avoid false dev_watchdog timeouts */ 1928 netif_carrier_off(netdev); 1929 1930 /* disable transmits */ 1931 netif_tx_stop_all_queues(netdev); 1932 netif_tx_disable(netdev); 1933 1934 /* reset Rx filters */ 1935 fm10k_reset_rx_state(interface); 1936 1937 /* disable polling routines */ 1938 fm10k_napi_disable_all(interface); 1939 1940 /* capture stats one last time before stopping interface */ 1941 fm10k_update_stats(interface); 1942 1943 /* prevent updating statistics while we're down */ 1944 while (test_and_set_bit(__FM10K_UPDATING_STATS, interface->state)) 1945 usleep_range(1000, 2000); 1946 1947 /* skip waiting for TX DMA if we lost PCIe link */ 1948 if (FM10K_REMOVED(hw->hw_addr)) 1949 goto skip_tx_dma_drain; 1950 1951 /* In some rare circumstances it can take a while for Tx queues to 1952 * quiesce and be fully disabled. Attempt to .stop_hw() first, and 1953 * then if we get ERR_REQUESTS_PENDING, go ahead and wait in a loop 1954 * until the Tx queues have emptied, or until a number of retries. If 1955 * we fail to clear within the retry loop, we will issue a warning 1956 * indicating that Tx DMA is probably hung. Note this means we call 1957 * .stop_hw() twice but this shouldn't cause any problems. 1958 */ 1959 err = hw->mac.ops.stop_hw(hw); 1960 if (err != FM10K_ERR_REQUESTS_PENDING) 1961 goto skip_tx_dma_drain; 1962 1963 #define TX_DMA_DRAIN_RETRIES 25 1964 for (count = 0; count < TX_DMA_DRAIN_RETRIES; count++) { 1965 usleep_range(10000, 20000); 1966 1967 /* start checking at the last ring to have pending Tx */ 1968 for (; i < interface->num_tx_queues; i++) 1969 if (fm10k_get_tx_pending(interface->tx_ring[i], false)) 1970 break; 1971 1972 /* if all the queues are drained, we can break now */ 1973 if (i == interface->num_tx_queues) 1974 break; 1975 } 1976 1977 if (count >= TX_DMA_DRAIN_RETRIES) 1978 dev_err(&interface->pdev->dev, 1979 "Tx queues failed to drain after %d tries. Tx DMA is probably hung.\n", 1980 count); 1981 skip_tx_dma_drain: 1982 /* Disable DMA engine for Tx/Rx */ 1983 err = hw->mac.ops.stop_hw(hw); 1984 if (err == FM10K_ERR_REQUESTS_PENDING) 1985 dev_err(&interface->pdev->dev, 1986 "due to pending requests hw was not shut down gracefully\n"); 1987 else if (err) 1988 dev_err(&interface->pdev->dev, "stop_hw failed: %d\n", err); 1989 1990 /* free any buffers still on the rings */ 1991 fm10k_clean_all_tx_rings(interface); 1992 fm10k_clean_all_rx_rings(interface); 1993 } 1994 1995 /** 1996 * fm10k_sw_init - Initialize general software structures 1997 * @interface: host interface private structure to initialize 1998 * 1999 * fm10k_sw_init initializes the interface private data structure. 2000 * Fields are initialized based on PCI device information and 2001 * OS network device settings (MTU size). 2002 **/ 2003 static int fm10k_sw_init(struct fm10k_intfc *interface, 2004 const struct pci_device_id *ent) 2005 { 2006 const struct fm10k_info *fi = fm10k_info_tbl[ent->driver_data]; 2007 struct fm10k_hw *hw = &interface->hw; 2008 struct pci_dev *pdev = interface->pdev; 2009 struct net_device *netdev = interface->netdev; 2010 u32 rss_key[FM10K_RSSRK_SIZE]; 2011 unsigned int rss; 2012 int err; 2013 2014 /* initialize back pointer */ 2015 hw->back = interface; 2016 hw->hw_addr = interface->uc_addr; 2017 2018 /* PCI config space info */ 2019 hw->vendor_id = pdev->vendor; 2020 hw->device_id = pdev->device; 2021 hw->revision_id = pdev->revision; 2022 hw->subsystem_vendor_id = pdev->subsystem_vendor; 2023 hw->subsystem_device_id = pdev->subsystem_device; 2024 2025 /* Setup hw api */ 2026 memcpy(&hw->mac.ops, fi->mac_ops, sizeof(hw->mac.ops)); 2027 hw->mac.type = fi->mac; 2028 2029 /* Setup IOV handlers */ 2030 if (fi->iov_ops) 2031 memcpy(&hw->iov.ops, fi->iov_ops, sizeof(hw->iov.ops)); 2032 2033 /* Set common capability flags and settings */ 2034 rss = min_t(int, FM10K_MAX_RSS_INDICES, num_online_cpus()); 2035 interface->ring_feature[RING_F_RSS].limit = rss; 2036 fi->get_invariants(hw); 2037 2038 /* pick up the PCIe bus settings for reporting later */ 2039 if (hw->mac.ops.get_bus_info) 2040 hw->mac.ops.get_bus_info(hw); 2041 2042 /* limit the usable DMA range */ 2043 if (hw->mac.ops.set_dma_mask) 2044 hw->mac.ops.set_dma_mask(hw, dma_get_mask(&pdev->dev)); 2045 2046 /* update netdev with DMA restrictions */ 2047 if (dma_get_mask(&pdev->dev) > DMA_BIT_MASK(32)) { 2048 netdev->features |= NETIF_F_HIGHDMA; 2049 netdev->vlan_features |= NETIF_F_HIGHDMA; 2050 } 2051 2052 /* reset and initialize the hardware so it is in a known state */ 2053 err = hw->mac.ops.reset_hw(hw); 2054 if (err) { 2055 dev_err(&pdev->dev, "reset_hw failed: %d\n", err); 2056 return err; 2057 } 2058 2059 err = hw->mac.ops.init_hw(hw); 2060 if (err) { 2061 dev_err(&pdev->dev, "init_hw failed: %d\n", err); 2062 return err; 2063 } 2064 2065 /* initialize hardware statistics */ 2066 hw->mac.ops.update_hw_stats(hw, &interface->stats); 2067 2068 /* Set upper limit on IOV VFs that can be allocated */ 2069 pci_sriov_set_totalvfs(pdev, hw->iov.total_vfs); 2070 2071 /* Start with random Ethernet address */ 2072 eth_random_addr(hw->mac.addr); 2073 2074 /* Initialize MAC address from hardware */ 2075 err = hw->mac.ops.read_mac_addr(hw); 2076 if (err) { 2077 dev_warn(&pdev->dev, 2078 "Failed to obtain MAC address defaulting to random\n"); 2079 /* tag address assignment as random */ 2080 netdev->addr_assign_type |= NET_ADDR_RANDOM; 2081 } 2082 2083 ether_addr_copy(netdev->dev_addr, hw->mac.addr); 2084 ether_addr_copy(netdev->perm_addr, hw->mac.addr); 2085 2086 if (!is_valid_ether_addr(netdev->perm_addr)) { 2087 dev_err(&pdev->dev, "Invalid MAC Address\n"); 2088 return -EIO; 2089 } 2090 2091 /* initialize DCBNL interface */ 2092 fm10k_dcbnl_set_ops(netdev); 2093 2094 /* set default ring sizes */ 2095 interface->tx_ring_count = FM10K_DEFAULT_TXD; 2096 interface->rx_ring_count = FM10K_DEFAULT_RXD; 2097 2098 /* set default interrupt moderation */ 2099 interface->tx_itr = FM10K_TX_ITR_DEFAULT; 2100 interface->rx_itr = FM10K_ITR_ADAPTIVE | FM10K_RX_ITR_DEFAULT; 2101 2102 /* initialize udp port lists */ 2103 INIT_LIST_HEAD(&interface->vxlan_port); 2104 INIT_LIST_HEAD(&interface->geneve_port); 2105 2106 /* Initialize the MAC/VLAN queue */ 2107 INIT_LIST_HEAD(&interface->macvlan_requests); 2108 2109 netdev_rss_key_fill(rss_key, sizeof(rss_key)); 2110 memcpy(interface->rssrk, rss_key, sizeof(rss_key)); 2111 2112 /* Initialize the mailbox lock */ 2113 spin_lock_init(&interface->mbx_lock); 2114 spin_lock_init(&interface->macvlan_lock); 2115 2116 /* Start off interface as being down */ 2117 set_bit(__FM10K_DOWN, interface->state); 2118 set_bit(__FM10K_UPDATING_STATS, interface->state); 2119 2120 return 0; 2121 } 2122 2123 static void fm10k_slot_warn(struct fm10k_intfc *interface) 2124 { 2125 enum pcie_link_width width = PCIE_LNK_WIDTH_UNKNOWN; 2126 enum pci_bus_speed speed = PCI_SPEED_UNKNOWN; 2127 struct fm10k_hw *hw = &interface->hw; 2128 int max_gts = 0, expected_gts = 0; 2129 2130 if (pcie_get_minimum_link(interface->pdev, &speed, &width) || 2131 speed == PCI_SPEED_UNKNOWN || width == PCIE_LNK_WIDTH_UNKNOWN) { 2132 dev_warn(&interface->pdev->dev, 2133 "Unable to determine PCI Express bandwidth.\n"); 2134 return; 2135 } 2136 2137 switch (speed) { 2138 case PCIE_SPEED_2_5GT: 2139 /* 8b/10b encoding reduces max throughput by 20% */ 2140 max_gts = 2 * width; 2141 break; 2142 case PCIE_SPEED_5_0GT: 2143 /* 8b/10b encoding reduces max throughput by 20% */ 2144 max_gts = 4 * width; 2145 break; 2146 case PCIE_SPEED_8_0GT: 2147 /* 128b/130b encoding has less than 2% impact on throughput */ 2148 max_gts = 8 * width; 2149 break; 2150 default: 2151 dev_warn(&interface->pdev->dev, 2152 "Unable to determine PCI Express bandwidth.\n"); 2153 return; 2154 } 2155 2156 dev_info(&interface->pdev->dev, 2157 "PCI Express bandwidth of %dGT/s available\n", 2158 max_gts); 2159 dev_info(&interface->pdev->dev, 2160 "(Speed:%s, Width: x%d, Encoding Loss:%s, Payload:%s)\n", 2161 (speed == PCIE_SPEED_8_0GT ? "8.0GT/s" : 2162 speed == PCIE_SPEED_5_0GT ? "5.0GT/s" : 2163 speed == PCIE_SPEED_2_5GT ? "2.5GT/s" : 2164 "Unknown"), 2165 hw->bus.width, 2166 (speed == PCIE_SPEED_2_5GT ? "20%" : 2167 speed == PCIE_SPEED_5_0GT ? "20%" : 2168 speed == PCIE_SPEED_8_0GT ? "<2%" : 2169 "Unknown"), 2170 (hw->bus.payload == fm10k_bus_payload_128 ? "128B" : 2171 hw->bus.payload == fm10k_bus_payload_256 ? "256B" : 2172 hw->bus.payload == fm10k_bus_payload_512 ? "512B" : 2173 "Unknown")); 2174 2175 switch (hw->bus_caps.speed) { 2176 case fm10k_bus_speed_2500: 2177 /* 8b/10b encoding reduces max throughput by 20% */ 2178 expected_gts = 2 * hw->bus_caps.width; 2179 break; 2180 case fm10k_bus_speed_5000: 2181 /* 8b/10b encoding reduces max throughput by 20% */ 2182 expected_gts = 4 * hw->bus_caps.width; 2183 break; 2184 case fm10k_bus_speed_8000: 2185 /* 128b/130b encoding has less than 2% impact on throughput */ 2186 expected_gts = 8 * hw->bus_caps.width; 2187 break; 2188 default: 2189 dev_warn(&interface->pdev->dev, 2190 "Unable to determine expected PCI Express bandwidth.\n"); 2191 return; 2192 } 2193 2194 if (max_gts >= expected_gts) 2195 return; 2196 2197 dev_warn(&interface->pdev->dev, 2198 "This device requires %dGT/s of bandwidth for optimal performance.\n", 2199 expected_gts); 2200 dev_warn(&interface->pdev->dev, 2201 "A %sslot with x%d lanes is suggested.\n", 2202 (hw->bus_caps.speed == fm10k_bus_speed_2500 ? "2.5GT/s " : 2203 hw->bus_caps.speed == fm10k_bus_speed_5000 ? "5.0GT/s " : 2204 hw->bus_caps.speed == fm10k_bus_speed_8000 ? "8.0GT/s " : ""), 2205 hw->bus_caps.width); 2206 } 2207 2208 /** 2209 * fm10k_probe - Device Initialization Routine 2210 * @pdev: PCI device information struct 2211 * @ent: entry in fm10k_pci_tbl 2212 * 2213 * Returns 0 on success, negative on failure 2214 * 2215 * fm10k_probe initializes an interface identified by a pci_dev structure. 2216 * The OS initialization, configuring of the interface private structure, 2217 * and a hardware reset occur. 2218 **/ 2219 static int fm10k_probe(struct pci_dev *pdev, const struct pci_device_id *ent) 2220 { 2221 struct net_device *netdev; 2222 struct fm10k_intfc *interface; 2223 int err; 2224 2225 if (pdev->error_state != pci_channel_io_normal) { 2226 dev_err(&pdev->dev, 2227 "PCI device still in an error state. Unable to load...\n"); 2228 return -EIO; 2229 } 2230 2231 err = pci_enable_device_mem(pdev); 2232 if (err) { 2233 dev_err(&pdev->dev, 2234 "PCI enable device failed: %d\n", err); 2235 return err; 2236 } 2237 2238 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(48)); 2239 if (err) 2240 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32)); 2241 if (err) { 2242 dev_err(&pdev->dev, 2243 "DMA configuration failed: %d\n", err); 2244 goto err_dma; 2245 } 2246 2247 err = pci_request_mem_regions(pdev, fm10k_driver_name); 2248 if (err) { 2249 dev_err(&pdev->dev, 2250 "pci_request_selected_regions failed: %d\n", err); 2251 goto err_pci_reg; 2252 } 2253 2254 pci_enable_pcie_error_reporting(pdev); 2255 2256 pci_set_master(pdev); 2257 pci_save_state(pdev); 2258 2259 netdev = fm10k_alloc_netdev(fm10k_info_tbl[ent->driver_data]); 2260 if (!netdev) { 2261 err = -ENOMEM; 2262 goto err_alloc_netdev; 2263 } 2264 2265 SET_NETDEV_DEV(netdev, &pdev->dev); 2266 2267 interface = netdev_priv(netdev); 2268 pci_set_drvdata(pdev, interface); 2269 2270 interface->netdev = netdev; 2271 interface->pdev = pdev; 2272 2273 interface->uc_addr = ioremap(pci_resource_start(pdev, 0), 2274 FM10K_UC_ADDR_SIZE); 2275 if (!interface->uc_addr) { 2276 err = -EIO; 2277 goto err_ioremap; 2278 } 2279 2280 err = fm10k_sw_init(interface, ent); 2281 if (err) 2282 goto err_sw_init; 2283 2284 /* enable debugfs support */ 2285 fm10k_dbg_intfc_init(interface); 2286 2287 err = fm10k_init_queueing_scheme(interface); 2288 if (err) 2289 goto err_sw_init; 2290 2291 /* the mbx interrupt might attempt to schedule the service task, so we 2292 * must ensure it is disabled since we haven't yet requested the timer 2293 * or work item. 2294 */ 2295 set_bit(__FM10K_SERVICE_DISABLE, interface->state); 2296 2297 err = fm10k_mbx_request_irq(interface); 2298 if (err) 2299 goto err_mbx_interrupt; 2300 2301 /* final check of hardware state before registering the interface */ 2302 err = fm10k_hw_ready(interface); 2303 if (err) 2304 goto err_register; 2305 2306 err = register_netdev(netdev); 2307 if (err) 2308 goto err_register; 2309 2310 /* carrier off reporting is important to ethtool even BEFORE open */ 2311 netif_carrier_off(netdev); 2312 2313 /* stop all the transmit queues from transmitting until link is up */ 2314 netif_tx_stop_all_queues(netdev); 2315 2316 /* Initialize service timer and service task late in order to avoid 2317 * cleanup issues. 2318 */ 2319 timer_setup(&interface->service_timer, fm10k_service_timer, 0); 2320 INIT_WORK(&interface->service_task, fm10k_service_task); 2321 2322 /* Setup the MAC/VLAN queue */ 2323 INIT_DELAYED_WORK(&interface->macvlan_task, fm10k_macvlan_task); 2324 2325 /* kick off service timer now, even when interface is down */ 2326 mod_timer(&interface->service_timer, (HZ * 2) + jiffies); 2327 2328 /* print warning for non-optimal configurations */ 2329 fm10k_slot_warn(interface); 2330 2331 /* report MAC address for logging */ 2332 dev_info(&pdev->dev, "%pM\n", netdev->dev_addr); 2333 2334 /* enable SR-IOV after registering netdev to enforce PF/VF ordering */ 2335 fm10k_iov_configure(pdev, 0); 2336 2337 /* clear the service task disable bit and kick off service task */ 2338 clear_bit(__FM10K_SERVICE_DISABLE, interface->state); 2339 fm10k_service_event_schedule(interface); 2340 2341 return 0; 2342 2343 err_register: 2344 fm10k_mbx_free_irq(interface); 2345 err_mbx_interrupt: 2346 fm10k_clear_queueing_scheme(interface); 2347 err_sw_init: 2348 if (interface->sw_addr) 2349 iounmap(interface->sw_addr); 2350 iounmap(interface->uc_addr); 2351 err_ioremap: 2352 free_netdev(netdev); 2353 err_alloc_netdev: 2354 pci_release_mem_regions(pdev); 2355 err_pci_reg: 2356 err_dma: 2357 pci_disable_device(pdev); 2358 return err; 2359 } 2360 2361 /** 2362 * fm10k_remove - Device Removal Routine 2363 * @pdev: PCI device information struct 2364 * 2365 * fm10k_remove is called by the PCI subsystem to alert the driver 2366 * that it should release a PCI device. The could be caused by a 2367 * Hot-Plug event, or because the driver is going to be removed from 2368 * memory. 2369 **/ 2370 static void fm10k_remove(struct pci_dev *pdev) 2371 { 2372 struct fm10k_intfc *interface = pci_get_drvdata(pdev); 2373 struct net_device *netdev = interface->netdev; 2374 2375 del_timer_sync(&interface->service_timer); 2376 2377 fm10k_stop_service_event(interface); 2378 fm10k_stop_macvlan_task(interface); 2379 2380 /* Remove all pending MAC/VLAN requests */ 2381 fm10k_clear_macvlan_queue(interface, interface->glort, true); 2382 2383 /* free netdev, this may bounce the interrupts due to setup_tc */ 2384 if (netdev->reg_state == NETREG_REGISTERED) 2385 unregister_netdev(netdev); 2386 2387 /* release VFs */ 2388 fm10k_iov_disable(pdev); 2389 2390 /* disable mailbox interrupt */ 2391 fm10k_mbx_free_irq(interface); 2392 2393 /* free interrupts */ 2394 fm10k_clear_queueing_scheme(interface); 2395 2396 /* remove any debugfs interfaces */ 2397 fm10k_dbg_intfc_exit(interface); 2398 2399 if (interface->sw_addr) 2400 iounmap(interface->sw_addr); 2401 iounmap(interface->uc_addr); 2402 2403 free_netdev(netdev); 2404 2405 pci_release_mem_regions(pdev); 2406 2407 pci_disable_pcie_error_reporting(pdev); 2408 2409 pci_disable_device(pdev); 2410 } 2411 2412 static void fm10k_prepare_suspend(struct fm10k_intfc *interface) 2413 { 2414 /* the watchdog task reads from registers, which might appear like 2415 * a surprise remove if the PCIe device is disabled while we're 2416 * stopped. We stop the watchdog task until after we resume software 2417 * activity. 2418 * 2419 * Note that the MAC/VLAN task will be stopped as part of preparing 2420 * for reset so we don't need to handle it here. 2421 */ 2422 fm10k_stop_service_event(interface); 2423 2424 if (fm10k_prepare_for_reset(interface)) 2425 set_bit(__FM10K_RESET_SUSPENDED, interface->state); 2426 } 2427 2428 static int fm10k_handle_resume(struct fm10k_intfc *interface) 2429 { 2430 struct fm10k_hw *hw = &interface->hw; 2431 int err; 2432 2433 /* Even if we didn't properly prepare for reset in 2434 * fm10k_prepare_suspend, we'll attempt to resume anyways. 2435 */ 2436 if (!test_and_clear_bit(__FM10K_RESET_SUSPENDED, interface->state)) 2437 dev_warn(&interface->pdev->dev, 2438 "Device was shut down as part of suspend... Attempting to recover\n"); 2439 2440 /* reset statistics starting values */ 2441 hw->mac.ops.rebind_hw_stats(hw, &interface->stats); 2442 2443 err = fm10k_handle_reset(interface); 2444 if (err) 2445 return err; 2446 2447 /* assume host is not ready, to prevent race with watchdog in case we 2448 * actually don't have connection to the switch 2449 */ 2450 interface->host_ready = false; 2451 fm10k_watchdog_host_not_ready(interface); 2452 2453 /* force link to stay down for a second to prevent link flutter */ 2454 interface->link_down_event = jiffies + (HZ); 2455 set_bit(__FM10K_LINK_DOWN, interface->state); 2456 2457 /* restart the service task */ 2458 fm10k_start_service_event(interface); 2459 2460 /* Restart the MAC/VLAN request queue in-case of outstanding events */ 2461 fm10k_macvlan_schedule(interface); 2462 2463 return err; 2464 } 2465 2466 /** 2467 * fm10k_resume - Generic PM resume hook 2468 * @dev: generic device structure 2469 * 2470 * Generic PM hook used when waking the device from a low power state after 2471 * suspend or hibernation. This function does not need to handle lower PCIe 2472 * device state as the stack takes care of that for us. 2473 **/ 2474 static int __maybe_unused fm10k_resume(struct device *dev) 2475 { 2476 struct fm10k_intfc *interface = pci_get_drvdata(to_pci_dev(dev)); 2477 struct net_device *netdev = interface->netdev; 2478 struct fm10k_hw *hw = &interface->hw; 2479 int err; 2480 2481 /* refresh hw_addr in case it was dropped */ 2482 hw->hw_addr = interface->uc_addr; 2483 2484 err = fm10k_handle_resume(interface); 2485 if (err) 2486 return err; 2487 2488 netif_device_attach(netdev); 2489 2490 return 0; 2491 } 2492 2493 /** 2494 * fm10k_suspend - Generic PM suspend hook 2495 * @dev: generic device structure 2496 * 2497 * Generic PM hook used when setting the device into a low power state for 2498 * system suspend or hibernation. This function does not need to handle lower 2499 * PCIe device state as the stack takes care of that for us. 2500 **/ 2501 static int __maybe_unused fm10k_suspend(struct device *dev) 2502 { 2503 struct fm10k_intfc *interface = pci_get_drvdata(to_pci_dev(dev)); 2504 struct net_device *netdev = interface->netdev; 2505 2506 netif_device_detach(netdev); 2507 2508 fm10k_prepare_suspend(interface); 2509 2510 return 0; 2511 } 2512 2513 /** 2514 * fm10k_io_error_detected - called when PCI error is detected 2515 * @pdev: Pointer to PCI device 2516 * @state: The current pci connection state 2517 * 2518 * This function is called after a PCI bus error affecting 2519 * this device has been detected. 2520 */ 2521 static pci_ers_result_t fm10k_io_error_detected(struct pci_dev *pdev, 2522 pci_channel_state_t state) 2523 { 2524 struct fm10k_intfc *interface = pci_get_drvdata(pdev); 2525 struct net_device *netdev = interface->netdev; 2526 2527 netif_device_detach(netdev); 2528 2529 if (state == pci_channel_io_perm_failure) 2530 return PCI_ERS_RESULT_DISCONNECT; 2531 2532 fm10k_prepare_suspend(interface); 2533 2534 /* Request a slot reset. */ 2535 return PCI_ERS_RESULT_NEED_RESET; 2536 } 2537 2538 /** 2539 * fm10k_io_slot_reset - called after the pci bus has been reset. 2540 * @pdev: Pointer to PCI device 2541 * 2542 * Restart the card from scratch, as if from a cold-boot. 2543 */ 2544 static pci_ers_result_t fm10k_io_slot_reset(struct pci_dev *pdev) 2545 { 2546 pci_ers_result_t result; 2547 2548 if (pci_reenable_device(pdev)) { 2549 dev_err(&pdev->dev, 2550 "Cannot re-enable PCI device after reset.\n"); 2551 result = PCI_ERS_RESULT_DISCONNECT; 2552 } else { 2553 pci_set_master(pdev); 2554 pci_restore_state(pdev); 2555 2556 /* After second error pci->state_saved is false, this 2557 * resets it so EEH doesn't break. 2558 */ 2559 pci_save_state(pdev); 2560 2561 pci_wake_from_d3(pdev, false); 2562 2563 result = PCI_ERS_RESULT_RECOVERED; 2564 } 2565 2566 pci_cleanup_aer_uncorrect_error_status(pdev); 2567 2568 return result; 2569 } 2570 2571 /** 2572 * fm10k_io_resume - called when traffic can start flowing again. 2573 * @pdev: Pointer to PCI device 2574 * 2575 * This callback is called when the error recovery driver tells us that 2576 * its OK to resume normal operation. 2577 */ 2578 static void fm10k_io_resume(struct pci_dev *pdev) 2579 { 2580 struct fm10k_intfc *interface = pci_get_drvdata(pdev); 2581 struct net_device *netdev = interface->netdev; 2582 int err; 2583 2584 err = fm10k_handle_resume(interface); 2585 2586 if (err) 2587 dev_warn(&pdev->dev, 2588 "%s failed: %d\n", __func__, err); 2589 else 2590 netif_device_attach(netdev); 2591 } 2592 2593 /** 2594 * fm10k_io_reset_prepare - called when PCI function is about to be reset 2595 * @pdev: Pointer to PCI device 2596 * 2597 * This callback is called when the PCI function is about to be reset, 2598 * allowing the device driver to prepare for it. 2599 */ 2600 static void fm10k_io_reset_prepare(struct pci_dev *pdev) 2601 { 2602 /* warn incase we have any active VF devices */ 2603 if (pci_num_vf(pdev)) 2604 dev_warn(&pdev->dev, 2605 "PCIe FLR may cause issues for any active VF devices\n"); 2606 fm10k_prepare_suspend(pci_get_drvdata(pdev)); 2607 } 2608 2609 /** 2610 * fm10k_io_reset_done - called when PCI function has finished resetting 2611 * @pdev: Pointer to PCI device 2612 * 2613 * This callback is called just after the PCI function is reset, such as via 2614 * /sys/class/net/<enpX>/device/reset or similar. 2615 */ 2616 static void fm10k_io_reset_done(struct pci_dev *pdev) 2617 { 2618 struct fm10k_intfc *interface = pci_get_drvdata(pdev); 2619 int err = fm10k_handle_resume(interface); 2620 2621 if (err) { 2622 dev_warn(&pdev->dev, 2623 "%s failed: %d\n", __func__, err); 2624 netif_device_detach(interface->netdev); 2625 } 2626 } 2627 2628 static const struct pci_error_handlers fm10k_err_handler = { 2629 .error_detected = fm10k_io_error_detected, 2630 .slot_reset = fm10k_io_slot_reset, 2631 .resume = fm10k_io_resume, 2632 .reset_prepare = fm10k_io_reset_prepare, 2633 .reset_done = fm10k_io_reset_done, 2634 }; 2635 2636 static SIMPLE_DEV_PM_OPS(fm10k_pm_ops, fm10k_suspend, fm10k_resume); 2637 2638 static struct pci_driver fm10k_driver = { 2639 .name = fm10k_driver_name, 2640 .id_table = fm10k_pci_tbl, 2641 .probe = fm10k_probe, 2642 .remove = fm10k_remove, 2643 .driver = { 2644 .pm = &fm10k_pm_ops, 2645 }, 2646 .sriov_configure = fm10k_iov_configure, 2647 .err_handler = &fm10k_err_handler 2648 }; 2649 2650 /** 2651 * fm10k_register_pci_driver - register driver interface 2652 * 2653 * This function is called on module load in order to register the driver. 2654 **/ 2655 int fm10k_register_pci_driver(void) 2656 { 2657 return pci_register_driver(&fm10k_driver); 2658 } 2659 2660 /** 2661 * fm10k_unregister_pci_driver - unregister driver interface 2662 * 2663 * This function is called on module unload in order to remove the driver. 2664 **/ 2665 void fm10k_unregister_pci_driver(void) 2666 { 2667 pci_unregister_driver(&fm10k_driver); 2668 } 2669