1 /* Intel(R) Ethernet Switch Host Interface Driver 2 * Copyright(c) 2013 - 2016 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/aer.h> 23 24 #include "fm10k.h" 25 26 static const struct fm10k_info *fm10k_info_tbl[] = { 27 [fm10k_device_pf] = &fm10k_pf_info, 28 [fm10k_device_vf] = &fm10k_vf_info, 29 }; 30 31 /** 32 * fm10k_pci_tbl - PCI Device ID Table 33 * 34 * Wildcard entries (PCI_ANY_ID) should come last 35 * Last entry must be all 0s 36 * 37 * { Vendor ID, Device ID, SubVendor ID, SubDevice ID, 38 * Class, Class Mask, private data (not used) } 39 */ 40 static const struct pci_device_id fm10k_pci_tbl[] = { 41 { PCI_VDEVICE(INTEL, FM10K_DEV_ID_PF), fm10k_device_pf }, 42 { PCI_VDEVICE(INTEL, FM10K_DEV_ID_VF), fm10k_device_vf }, 43 /* required last entry */ 44 { 0, } 45 }; 46 MODULE_DEVICE_TABLE(pci, fm10k_pci_tbl); 47 48 u16 fm10k_read_pci_cfg_word(struct fm10k_hw *hw, u32 reg) 49 { 50 struct fm10k_intfc *interface = hw->back; 51 u16 value = 0; 52 53 if (FM10K_REMOVED(hw->hw_addr)) 54 return ~value; 55 56 pci_read_config_word(interface->pdev, reg, &value); 57 if (value == 0xFFFF) 58 fm10k_write_flush(hw); 59 60 return value; 61 } 62 63 u32 fm10k_read_reg(struct fm10k_hw *hw, int reg) 64 { 65 u32 __iomem *hw_addr = READ_ONCE(hw->hw_addr); 66 u32 value = 0; 67 68 if (FM10K_REMOVED(hw_addr)) 69 return ~value; 70 71 value = readl(&hw_addr[reg]); 72 if (!(~value) && (!reg || !(~readl(hw_addr)))) { 73 struct fm10k_intfc *interface = hw->back; 74 struct net_device *netdev = interface->netdev; 75 76 hw->hw_addr = NULL; 77 netif_device_detach(netdev); 78 netdev_err(netdev, "PCIe link lost, device now detached\n"); 79 } 80 81 return value; 82 } 83 84 static int fm10k_hw_ready(struct fm10k_intfc *interface) 85 { 86 struct fm10k_hw *hw = &interface->hw; 87 88 fm10k_write_flush(hw); 89 90 return FM10K_REMOVED(hw->hw_addr) ? -ENODEV : 0; 91 } 92 93 void fm10k_service_event_schedule(struct fm10k_intfc *interface) 94 { 95 if (!test_bit(__FM10K_SERVICE_DISABLE, &interface->state) && 96 !test_and_set_bit(__FM10K_SERVICE_SCHED, &interface->state)) 97 queue_work(fm10k_workqueue, &interface->service_task); 98 } 99 100 static void fm10k_service_event_complete(struct fm10k_intfc *interface) 101 { 102 WARN_ON(!test_bit(__FM10K_SERVICE_SCHED, &interface->state)); 103 104 /* flush memory to make sure state is correct before next watchog */ 105 smp_mb__before_atomic(); 106 clear_bit(__FM10K_SERVICE_SCHED, &interface->state); 107 } 108 109 /** 110 * fm10k_service_timer - Timer Call-back 111 * @data: pointer to interface cast into an unsigned long 112 **/ 113 static void fm10k_service_timer(unsigned long data) 114 { 115 struct fm10k_intfc *interface = (struct fm10k_intfc *)data; 116 117 /* Reset the timer */ 118 mod_timer(&interface->service_timer, (HZ * 2) + jiffies); 119 120 fm10k_service_event_schedule(interface); 121 } 122 123 static void fm10k_detach_subtask(struct fm10k_intfc *interface) 124 { 125 struct net_device *netdev = interface->netdev; 126 u32 __iomem *hw_addr; 127 u32 value; 128 129 /* do nothing if device is still present or hw_addr is set */ 130 if (netif_device_present(netdev) || interface->hw.hw_addr) 131 return; 132 133 /* check the real address space to see if we've recovered */ 134 hw_addr = READ_ONCE(interface->uc_addr); 135 value = readl(hw_addr); 136 if (~value) { 137 interface->hw.hw_addr = interface->uc_addr; 138 netif_device_attach(netdev); 139 interface->flags |= FM10K_FLAG_RESET_REQUESTED; 140 netdev_warn(netdev, "PCIe link restored, device now attached\n"); 141 return; 142 } 143 144 rtnl_lock(); 145 146 if (netif_running(netdev)) 147 dev_close(netdev); 148 149 rtnl_unlock(); 150 } 151 152 static void fm10k_prepare_for_reset(struct fm10k_intfc *interface) 153 { 154 struct net_device *netdev = interface->netdev; 155 156 WARN_ON(in_interrupt()); 157 158 /* put off any impending NetWatchDogTimeout */ 159 netif_trans_update(netdev); 160 161 while (test_and_set_bit(__FM10K_RESETTING, &interface->state)) 162 usleep_range(1000, 2000); 163 164 rtnl_lock(); 165 166 fm10k_iov_suspend(interface->pdev); 167 168 if (netif_running(netdev)) 169 fm10k_close(netdev); 170 171 fm10k_mbx_free_irq(interface); 172 173 /* free interrupts */ 174 fm10k_clear_queueing_scheme(interface); 175 176 /* delay any future reset requests */ 177 interface->last_reset = jiffies + (10 * HZ); 178 179 rtnl_unlock(); 180 } 181 182 static int fm10k_handle_reset(struct fm10k_intfc *interface) 183 { 184 struct net_device *netdev = interface->netdev; 185 struct fm10k_hw *hw = &interface->hw; 186 int err; 187 188 rtnl_lock(); 189 190 pci_set_master(interface->pdev); 191 192 /* reset and initialize the hardware so it is in a known state */ 193 err = hw->mac.ops.reset_hw(hw); 194 if (err) { 195 dev_err(&interface->pdev->dev, "reset_hw failed: %d\n", err); 196 goto reinit_err; 197 } 198 199 err = hw->mac.ops.init_hw(hw); 200 if (err) { 201 dev_err(&interface->pdev->dev, "init_hw failed: %d\n", err); 202 goto reinit_err; 203 } 204 205 err = fm10k_init_queueing_scheme(interface); 206 if (err) { 207 dev_err(&interface->pdev->dev, 208 "init_queueing_scheme failed: %d\n", err); 209 goto reinit_err; 210 } 211 212 /* re-associate interrupts */ 213 err = fm10k_mbx_request_irq(interface); 214 if (err) 215 goto err_mbx_irq; 216 217 err = fm10k_hw_ready(interface); 218 if (err) 219 goto err_open; 220 221 /* update hardware address for VFs if perm_addr has changed */ 222 if (hw->mac.type == fm10k_mac_vf) { 223 if (is_valid_ether_addr(hw->mac.perm_addr)) { 224 ether_addr_copy(hw->mac.addr, hw->mac.perm_addr); 225 ether_addr_copy(netdev->perm_addr, hw->mac.perm_addr); 226 ether_addr_copy(netdev->dev_addr, hw->mac.perm_addr); 227 netdev->addr_assign_type &= ~NET_ADDR_RANDOM; 228 } 229 230 if (hw->mac.vlan_override) 231 netdev->features &= ~NETIF_F_HW_VLAN_CTAG_RX; 232 else 233 netdev->features |= NETIF_F_HW_VLAN_CTAG_RX; 234 } 235 236 err = netif_running(netdev) ? fm10k_open(netdev) : 0; 237 if (err) 238 goto err_open; 239 240 fm10k_iov_resume(interface->pdev); 241 242 rtnl_unlock(); 243 244 clear_bit(__FM10K_RESETTING, &interface->state); 245 246 return err; 247 err_open: 248 fm10k_mbx_free_irq(interface); 249 err_mbx_irq: 250 fm10k_clear_queueing_scheme(interface); 251 reinit_err: 252 netif_device_detach(netdev); 253 254 rtnl_unlock(); 255 256 clear_bit(__FM10K_RESETTING, &interface->state); 257 258 return err; 259 } 260 261 static void fm10k_reinit(struct fm10k_intfc *interface) 262 { 263 int err; 264 265 fm10k_prepare_for_reset(interface); 266 267 err = fm10k_handle_reset(interface); 268 if (err) 269 dev_err(&interface->pdev->dev, 270 "fm10k_handle_reset failed: %d\n", err); 271 } 272 273 static void fm10k_reset_subtask(struct fm10k_intfc *interface) 274 { 275 if (!(interface->flags & FM10K_FLAG_RESET_REQUESTED)) 276 return; 277 278 interface->flags &= ~FM10K_FLAG_RESET_REQUESTED; 279 280 netdev_err(interface->netdev, "Reset interface\n"); 281 282 fm10k_reinit(interface); 283 } 284 285 /** 286 * fm10k_configure_swpri_map - Configure Receive SWPRI to PC mapping 287 * @interface: board private structure 288 * 289 * Configure the SWPRI to PC mapping for the port. 290 **/ 291 static void fm10k_configure_swpri_map(struct fm10k_intfc *interface) 292 { 293 struct net_device *netdev = interface->netdev; 294 struct fm10k_hw *hw = &interface->hw; 295 int i; 296 297 /* clear flag indicating update is needed */ 298 interface->flags &= ~FM10K_FLAG_SWPRI_CONFIG; 299 300 /* these registers are only available on the PF */ 301 if (hw->mac.type != fm10k_mac_pf) 302 return; 303 304 /* configure SWPRI to PC map */ 305 for (i = 0; i < FM10K_SWPRI_MAX; i++) 306 fm10k_write_reg(hw, FM10K_SWPRI_MAP(i), 307 netdev_get_prio_tc_map(netdev, i)); 308 } 309 310 /** 311 * fm10k_watchdog_update_host_state - Update the link status based on host. 312 * @interface: board private structure 313 **/ 314 static void fm10k_watchdog_update_host_state(struct fm10k_intfc *interface) 315 { 316 struct fm10k_hw *hw = &interface->hw; 317 s32 err; 318 319 if (test_bit(__FM10K_LINK_DOWN, &interface->state)) { 320 interface->host_ready = false; 321 if (time_is_after_jiffies(interface->link_down_event)) 322 return; 323 clear_bit(__FM10K_LINK_DOWN, &interface->state); 324 } 325 326 if (interface->flags & FM10K_FLAG_SWPRI_CONFIG) { 327 if (rtnl_trylock()) { 328 fm10k_configure_swpri_map(interface); 329 rtnl_unlock(); 330 } 331 } 332 333 /* lock the mailbox for transmit and receive */ 334 fm10k_mbx_lock(interface); 335 336 err = hw->mac.ops.get_host_state(hw, &interface->host_ready); 337 if (err && time_is_before_jiffies(interface->last_reset)) 338 interface->flags |= FM10K_FLAG_RESET_REQUESTED; 339 340 /* free the lock */ 341 fm10k_mbx_unlock(interface); 342 } 343 344 /** 345 * fm10k_mbx_subtask - Process upstream and downstream mailboxes 346 * @interface: board private structure 347 * 348 * This function will process both the upstream and downstream mailboxes. 349 **/ 350 static void fm10k_mbx_subtask(struct fm10k_intfc *interface) 351 { 352 /* process upstream mailbox and update device state */ 353 fm10k_watchdog_update_host_state(interface); 354 355 /* process downstream mailboxes */ 356 fm10k_iov_mbx(interface); 357 } 358 359 /** 360 * fm10k_watchdog_host_is_ready - Update netdev status based on host ready 361 * @interface: board private structure 362 **/ 363 static void fm10k_watchdog_host_is_ready(struct fm10k_intfc *interface) 364 { 365 struct net_device *netdev = interface->netdev; 366 367 /* only continue if link state is currently down */ 368 if (netif_carrier_ok(netdev)) 369 return; 370 371 netif_info(interface, drv, netdev, "NIC Link is up\n"); 372 373 netif_carrier_on(netdev); 374 netif_tx_wake_all_queues(netdev); 375 } 376 377 /** 378 * fm10k_watchdog_host_not_ready - Update netdev status based on host not ready 379 * @interface: board private structure 380 **/ 381 static void fm10k_watchdog_host_not_ready(struct fm10k_intfc *interface) 382 { 383 struct net_device *netdev = interface->netdev; 384 385 /* only continue if link state is currently up */ 386 if (!netif_carrier_ok(netdev)) 387 return; 388 389 netif_info(interface, drv, netdev, "NIC Link is down\n"); 390 391 netif_carrier_off(netdev); 392 netif_tx_stop_all_queues(netdev); 393 } 394 395 /** 396 * fm10k_update_stats - Update the board statistics counters. 397 * @interface: board private structure 398 **/ 399 void fm10k_update_stats(struct fm10k_intfc *interface) 400 { 401 struct net_device_stats *net_stats = &interface->netdev->stats; 402 struct fm10k_hw *hw = &interface->hw; 403 u64 hw_csum_tx_good = 0, hw_csum_rx_good = 0, rx_length_errors = 0; 404 u64 rx_switch_errors = 0, rx_drops = 0, rx_pp_errors = 0; 405 u64 rx_link_errors = 0; 406 u64 rx_errors = 0, rx_csum_errors = 0, tx_csum_errors = 0; 407 u64 restart_queue = 0, tx_busy = 0, alloc_failed = 0; 408 u64 rx_bytes_nic = 0, rx_pkts_nic = 0, rx_drops_nic = 0; 409 u64 tx_bytes_nic = 0, tx_pkts_nic = 0; 410 u64 bytes, pkts; 411 int i; 412 413 /* ensure only one thread updates stats at a time */ 414 if (test_and_set_bit(__FM10K_UPDATING_STATS, &interface->state)) 415 return; 416 417 /* do not allow stats update via service task for next second */ 418 interface->next_stats_update = jiffies + HZ; 419 420 /* gather some stats to the interface struct that are per queue */ 421 for (bytes = 0, pkts = 0, i = 0; i < interface->num_tx_queues; i++) { 422 struct fm10k_ring *tx_ring = READ_ONCE(interface->tx_ring[i]); 423 424 if (!tx_ring) 425 continue; 426 427 restart_queue += tx_ring->tx_stats.restart_queue; 428 tx_busy += tx_ring->tx_stats.tx_busy; 429 tx_csum_errors += tx_ring->tx_stats.csum_err; 430 bytes += tx_ring->stats.bytes; 431 pkts += tx_ring->stats.packets; 432 hw_csum_tx_good += tx_ring->tx_stats.csum_good; 433 } 434 435 interface->restart_queue = restart_queue; 436 interface->tx_busy = tx_busy; 437 net_stats->tx_bytes = bytes; 438 net_stats->tx_packets = pkts; 439 interface->tx_csum_errors = tx_csum_errors; 440 interface->hw_csum_tx_good = hw_csum_tx_good; 441 442 /* gather some stats to the interface struct that are per queue */ 443 for (bytes = 0, pkts = 0, i = 0; i < interface->num_rx_queues; i++) { 444 struct fm10k_ring *rx_ring = READ_ONCE(interface->rx_ring[i]); 445 446 if (!rx_ring) 447 continue; 448 449 bytes += rx_ring->stats.bytes; 450 pkts += rx_ring->stats.packets; 451 alloc_failed += rx_ring->rx_stats.alloc_failed; 452 rx_csum_errors += rx_ring->rx_stats.csum_err; 453 rx_errors += rx_ring->rx_stats.errors; 454 hw_csum_rx_good += rx_ring->rx_stats.csum_good; 455 rx_switch_errors += rx_ring->rx_stats.switch_errors; 456 rx_drops += rx_ring->rx_stats.drops; 457 rx_pp_errors += rx_ring->rx_stats.pp_errors; 458 rx_link_errors += rx_ring->rx_stats.link_errors; 459 rx_length_errors += rx_ring->rx_stats.length_errors; 460 } 461 462 net_stats->rx_bytes = bytes; 463 net_stats->rx_packets = pkts; 464 interface->alloc_failed = alloc_failed; 465 interface->rx_csum_errors = rx_csum_errors; 466 interface->hw_csum_rx_good = hw_csum_rx_good; 467 interface->rx_switch_errors = rx_switch_errors; 468 interface->rx_drops = rx_drops; 469 interface->rx_pp_errors = rx_pp_errors; 470 interface->rx_link_errors = rx_link_errors; 471 interface->rx_length_errors = rx_length_errors; 472 473 hw->mac.ops.update_hw_stats(hw, &interface->stats); 474 475 for (i = 0; i < hw->mac.max_queues; i++) { 476 struct fm10k_hw_stats_q *q = &interface->stats.q[i]; 477 478 tx_bytes_nic += q->tx_bytes.count; 479 tx_pkts_nic += q->tx_packets.count; 480 rx_bytes_nic += q->rx_bytes.count; 481 rx_pkts_nic += q->rx_packets.count; 482 rx_drops_nic += q->rx_drops.count; 483 } 484 485 interface->tx_bytes_nic = tx_bytes_nic; 486 interface->tx_packets_nic = tx_pkts_nic; 487 interface->rx_bytes_nic = rx_bytes_nic; 488 interface->rx_packets_nic = rx_pkts_nic; 489 interface->rx_drops_nic = rx_drops_nic; 490 491 /* Fill out the OS statistics structure */ 492 net_stats->rx_errors = rx_errors; 493 net_stats->rx_dropped = interface->stats.nodesc_drop.count; 494 495 clear_bit(__FM10K_UPDATING_STATS, &interface->state); 496 } 497 498 /** 499 * fm10k_watchdog_flush_tx - flush queues on host not ready 500 * @interface - pointer to the device interface structure 501 **/ 502 static void fm10k_watchdog_flush_tx(struct fm10k_intfc *interface) 503 { 504 int some_tx_pending = 0; 505 int i; 506 507 /* nothing to do if carrier is up */ 508 if (netif_carrier_ok(interface->netdev)) 509 return; 510 511 for (i = 0; i < interface->num_tx_queues; i++) { 512 struct fm10k_ring *tx_ring = interface->tx_ring[i]; 513 514 if (tx_ring->next_to_use != tx_ring->next_to_clean) { 515 some_tx_pending = 1; 516 break; 517 } 518 } 519 520 /* We've lost link, so the controller stops DMA, but we've got 521 * queued Tx work that's never going to get done, so reset 522 * controller to flush Tx. 523 */ 524 if (some_tx_pending) 525 interface->flags |= FM10K_FLAG_RESET_REQUESTED; 526 } 527 528 /** 529 * fm10k_watchdog_subtask - check and bring link up 530 * @interface - pointer to the device interface structure 531 **/ 532 static void fm10k_watchdog_subtask(struct fm10k_intfc *interface) 533 { 534 /* if interface is down do nothing */ 535 if (test_bit(__FM10K_DOWN, &interface->state) || 536 test_bit(__FM10K_RESETTING, &interface->state)) 537 return; 538 539 if (interface->host_ready) 540 fm10k_watchdog_host_is_ready(interface); 541 else 542 fm10k_watchdog_host_not_ready(interface); 543 544 /* update stats only once every second */ 545 if (time_is_before_jiffies(interface->next_stats_update)) 546 fm10k_update_stats(interface); 547 548 /* flush any uncompleted work */ 549 fm10k_watchdog_flush_tx(interface); 550 } 551 552 /** 553 * fm10k_check_hang_subtask - check for hung queues and dropped interrupts 554 * @interface - pointer to the device interface structure 555 * 556 * This function serves two purposes. First it strobes the interrupt lines 557 * in order to make certain interrupts are occurring. Secondly it sets the 558 * bits needed to check for TX hangs. As a result we should immediately 559 * determine if a hang has occurred. 560 */ 561 static void fm10k_check_hang_subtask(struct fm10k_intfc *interface) 562 { 563 int i; 564 565 /* If we're down or resetting, just bail */ 566 if (test_bit(__FM10K_DOWN, &interface->state) || 567 test_bit(__FM10K_RESETTING, &interface->state)) 568 return; 569 570 /* rate limit tx hang checks to only once every 2 seconds */ 571 if (time_is_after_eq_jiffies(interface->next_tx_hang_check)) 572 return; 573 interface->next_tx_hang_check = jiffies + (2 * HZ); 574 575 if (netif_carrier_ok(interface->netdev)) { 576 /* Force detection of hung controller */ 577 for (i = 0; i < interface->num_tx_queues; i++) 578 set_check_for_tx_hang(interface->tx_ring[i]); 579 580 /* Rearm all in-use q_vectors for immediate firing */ 581 for (i = 0; i < interface->num_q_vectors; i++) { 582 struct fm10k_q_vector *qv = interface->q_vector[i]; 583 584 if (!qv->tx.count && !qv->rx.count) 585 continue; 586 writel(FM10K_ITR_ENABLE | FM10K_ITR_PENDING2, qv->itr); 587 } 588 } 589 } 590 591 /** 592 * fm10k_service_task - manages and runs subtasks 593 * @work: pointer to work_struct containing our data 594 **/ 595 static void fm10k_service_task(struct work_struct *work) 596 { 597 struct fm10k_intfc *interface; 598 599 interface = container_of(work, struct fm10k_intfc, service_task); 600 601 /* tasks run even when interface is down */ 602 fm10k_mbx_subtask(interface); 603 fm10k_detach_subtask(interface); 604 fm10k_reset_subtask(interface); 605 606 /* tasks only run when interface is up */ 607 fm10k_watchdog_subtask(interface); 608 fm10k_check_hang_subtask(interface); 609 610 /* release lock on service events to allow scheduling next event */ 611 fm10k_service_event_complete(interface); 612 } 613 614 /** 615 * fm10k_configure_tx_ring - Configure Tx ring after Reset 616 * @interface: board private structure 617 * @ring: structure containing ring specific data 618 * 619 * Configure the Tx descriptor ring after a reset. 620 **/ 621 static void fm10k_configure_tx_ring(struct fm10k_intfc *interface, 622 struct fm10k_ring *ring) 623 { 624 struct fm10k_hw *hw = &interface->hw; 625 u64 tdba = ring->dma; 626 u32 size = ring->count * sizeof(struct fm10k_tx_desc); 627 u32 txint = FM10K_INT_MAP_DISABLE; 628 u32 txdctl = BIT(FM10K_TXDCTL_MAX_TIME_SHIFT) | FM10K_TXDCTL_ENABLE; 629 u8 reg_idx = ring->reg_idx; 630 631 /* disable queue to avoid issues while updating state */ 632 fm10k_write_reg(hw, FM10K_TXDCTL(reg_idx), 0); 633 fm10k_write_flush(hw); 634 635 /* possible poll here to verify ring resources have been cleaned */ 636 637 /* set location and size for descriptor ring */ 638 fm10k_write_reg(hw, FM10K_TDBAL(reg_idx), tdba & DMA_BIT_MASK(32)); 639 fm10k_write_reg(hw, FM10K_TDBAH(reg_idx), tdba >> 32); 640 fm10k_write_reg(hw, FM10K_TDLEN(reg_idx), size); 641 642 /* reset head and tail pointers */ 643 fm10k_write_reg(hw, FM10K_TDH(reg_idx), 0); 644 fm10k_write_reg(hw, FM10K_TDT(reg_idx), 0); 645 646 /* store tail pointer */ 647 ring->tail = &interface->uc_addr[FM10K_TDT(reg_idx)]; 648 649 /* reset ntu and ntc to place SW in sync with hardware */ 650 ring->next_to_clean = 0; 651 ring->next_to_use = 0; 652 653 /* Map interrupt */ 654 if (ring->q_vector) { 655 txint = ring->q_vector->v_idx + NON_Q_VECTORS(hw); 656 txint |= FM10K_INT_MAP_TIMER0; 657 } 658 659 fm10k_write_reg(hw, FM10K_TXINT(reg_idx), txint); 660 661 /* enable use of FTAG bit in Tx descriptor, register is RO for VF */ 662 fm10k_write_reg(hw, FM10K_PFVTCTL(reg_idx), 663 FM10K_PFVTCTL_FTAG_DESC_ENABLE); 664 665 /* Initialize XPS */ 666 if (!test_and_set_bit(__FM10K_TX_XPS_INIT_DONE, &ring->state) && 667 ring->q_vector) 668 netif_set_xps_queue(ring->netdev, 669 &ring->q_vector->affinity_mask, 670 ring->queue_index); 671 672 /* enable queue */ 673 fm10k_write_reg(hw, FM10K_TXDCTL(reg_idx), txdctl); 674 } 675 676 /** 677 * fm10k_enable_tx_ring - Verify Tx ring is enabled after configuration 678 * @interface: board private structure 679 * @ring: structure containing ring specific data 680 * 681 * Verify the Tx descriptor ring is ready for transmit. 682 **/ 683 static void fm10k_enable_tx_ring(struct fm10k_intfc *interface, 684 struct fm10k_ring *ring) 685 { 686 struct fm10k_hw *hw = &interface->hw; 687 int wait_loop = 10; 688 u32 txdctl; 689 u8 reg_idx = ring->reg_idx; 690 691 /* if we are already enabled just exit */ 692 if (fm10k_read_reg(hw, FM10K_TXDCTL(reg_idx)) & FM10K_TXDCTL_ENABLE) 693 return; 694 695 /* poll to verify queue is enabled */ 696 do { 697 usleep_range(1000, 2000); 698 txdctl = fm10k_read_reg(hw, FM10K_TXDCTL(reg_idx)); 699 } while (!(txdctl & FM10K_TXDCTL_ENABLE) && --wait_loop); 700 if (!wait_loop) 701 netif_err(interface, drv, interface->netdev, 702 "Could not enable Tx Queue %d\n", reg_idx); 703 } 704 705 /** 706 * fm10k_configure_tx - Configure Transmit Unit after Reset 707 * @interface: board private structure 708 * 709 * Configure the Tx unit of the MAC after a reset. 710 **/ 711 static void fm10k_configure_tx(struct fm10k_intfc *interface) 712 { 713 int i; 714 715 /* Setup the HW Tx Head and Tail descriptor pointers */ 716 for (i = 0; i < interface->num_tx_queues; i++) 717 fm10k_configure_tx_ring(interface, interface->tx_ring[i]); 718 719 /* poll here to verify that Tx rings are now enabled */ 720 for (i = 0; i < interface->num_tx_queues; i++) 721 fm10k_enable_tx_ring(interface, interface->tx_ring[i]); 722 } 723 724 /** 725 * fm10k_configure_rx_ring - Configure Rx ring after Reset 726 * @interface: board private structure 727 * @ring: structure containing ring specific data 728 * 729 * Configure the Rx descriptor ring after a reset. 730 **/ 731 static void fm10k_configure_rx_ring(struct fm10k_intfc *interface, 732 struct fm10k_ring *ring) 733 { 734 u64 rdba = ring->dma; 735 struct fm10k_hw *hw = &interface->hw; 736 u32 size = ring->count * sizeof(union fm10k_rx_desc); 737 u32 rxqctl, rxdctl = FM10K_RXDCTL_WRITE_BACK_MIN_DELAY; 738 u32 srrctl = FM10K_SRRCTL_BUFFER_CHAINING_EN; 739 u32 rxint = FM10K_INT_MAP_DISABLE; 740 u8 rx_pause = interface->rx_pause; 741 u8 reg_idx = ring->reg_idx; 742 743 /* disable queue to avoid issues while updating state */ 744 rxqctl = fm10k_read_reg(hw, FM10K_RXQCTL(reg_idx)); 745 rxqctl &= ~FM10K_RXQCTL_ENABLE; 746 fm10k_write_flush(hw); 747 748 /* possible poll here to verify ring resources have been cleaned */ 749 750 /* set location and size for descriptor ring */ 751 fm10k_write_reg(hw, FM10K_RDBAL(reg_idx), rdba & DMA_BIT_MASK(32)); 752 fm10k_write_reg(hw, FM10K_RDBAH(reg_idx), rdba >> 32); 753 fm10k_write_reg(hw, FM10K_RDLEN(reg_idx), size); 754 755 /* reset head and tail pointers */ 756 fm10k_write_reg(hw, FM10K_RDH(reg_idx), 0); 757 fm10k_write_reg(hw, FM10K_RDT(reg_idx), 0); 758 759 /* store tail pointer */ 760 ring->tail = &interface->uc_addr[FM10K_RDT(reg_idx)]; 761 762 /* reset ntu and ntc to place SW in sync with hardware */ 763 ring->next_to_clean = 0; 764 ring->next_to_use = 0; 765 ring->next_to_alloc = 0; 766 767 /* Configure the Rx buffer size for one buff without split */ 768 srrctl |= FM10K_RX_BUFSZ >> FM10K_SRRCTL_BSIZEPKT_SHIFT; 769 770 /* Configure the Rx ring to suppress loopback packets */ 771 srrctl |= FM10K_SRRCTL_LOOPBACK_SUPPRESS; 772 fm10k_write_reg(hw, FM10K_SRRCTL(reg_idx), srrctl); 773 774 /* Enable drop on empty */ 775 #ifdef CONFIG_DCB 776 if (interface->pfc_en) 777 rx_pause = interface->pfc_en; 778 #endif 779 if (!(rx_pause & BIT(ring->qos_pc))) 780 rxdctl |= FM10K_RXDCTL_DROP_ON_EMPTY; 781 782 fm10k_write_reg(hw, FM10K_RXDCTL(reg_idx), rxdctl); 783 784 /* assign default VLAN to queue */ 785 ring->vid = hw->mac.default_vid; 786 787 /* if we have an active VLAN, disable default VLAN ID */ 788 if (test_bit(hw->mac.default_vid, interface->active_vlans)) 789 ring->vid |= FM10K_VLAN_CLEAR; 790 791 /* Map interrupt */ 792 if (ring->q_vector) { 793 rxint = ring->q_vector->v_idx + NON_Q_VECTORS(hw); 794 rxint |= FM10K_INT_MAP_TIMER1; 795 } 796 797 fm10k_write_reg(hw, FM10K_RXINT(reg_idx), rxint); 798 799 /* enable queue */ 800 rxqctl = fm10k_read_reg(hw, FM10K_RXQCTL(reg_idx)); 801 rxqctl |= FM10K_RXQCTL_ENABLE; 802 fm10k_write_reg(hw, FM10K_RXQCTL(reg_idx), rxqctl); 803 804 /* place buffers on ring for receive data */ 805 fm10k_alloc_rx_buffers(ring, fm10k_desc_unused(ring)); 806 } 807 808 /** 809 * fm10k_update_rx_drop_en - Configures the drop enable bits for Rx rings 810 * @interface: board private structure 811 * 812 * Configure the drop enable bits for the Rx rings. 813 **/ 814 void fm10k_update_rx_drop_en(struct fm10k_intfc *interface) 815 { 816 struct fm10k_hw *hw = &interface->hw; 817 u8 rx_pause = interface->rx_pause; 818 int i; 819 820 #ifdef CONFIG_DCB 821 if (interface->pfc_en) 822 rx_pause = interface->pfc_en; 823 824 #endif 825 for (i = 0; i < interface->num_rx_queues; i++) { 826 struct fm10k_ring *ring = interface->rx_ring[i]; 827 u32 rxdctl = FM10K_RXDCTL_WRITE_BACK_MIN_DELAY; 828 u8 reg_idx = ring->reg_idx; 829 830 if (!(rx_pause & BIT(ring->qos_pc))) 831 rxdctl |= FM10K_RXDCTL_DROP_ON_EMPTY; 832 833 fm10k_write_reg(hw, FM10K_RXDCTL(reg_idx), rxdctl); 834 } 835 } 836 837 /** 838 * fm10k_configure_dglort - Configure Receive DGLORT after reset 839 * @interface: board private structure 840 * 841 * Configure the DGLORT description and RSS tables. 842 **/ 843 static void fm10k_configure_dglort(struct fm10k_intfc *interface) 844 { 845 struct fm10k_dglort_cfg dglort = { 0 }; 846 struct fm10k_hw *hw = &interface->hw; 847 int i; 848 u32 mrqc; 849 850 /* Fill out hash function seeds */ 851 for (i = 0; i < FM10K_RSSRK_SIZE; i++) 852 fm10k_write_reg(hw, FM10K_RSSRK(0, i), interface->rssrk[i]); 853 854 /* Write RETA table to hardware */ 855 for (i = 0; i < FM10K_RETA_SIZE; i++) 856 fm10k_write_reg(hw, FM10K_RETA(0, i), interface->reta[i]); 857 858 /* Generate RSS hash based on packet types, TCP/UDP 859 * port numbers and/or IPv4/v6 src and dst addresses 860 */ 861 mrqc = FM10K_MRQC_IPV4 | 862 FM10K_MRQC_TCP_IPV4 | 863 FM10K_MRQC_IPV6 | 864 FM10K_MRQC_TCP_IPV6; 865 866 if (interface->flags & FM10K_FLAG_RSS_FIELD_IPV4_UDP) 867 mrqc |= FM10K_MRQC_UDP_IPV4; 868 if (interface->flags & FM10K_FLAG_RSS_FIELD_IPV6_UDP) 869 mrqc |= FM10K_MRQC_UDP_IPV6; 870 871 fm10k_write_reg(hw, FM10K_MRQC(0), mrqc); 872 873 /* configure default DGLORT mapping for RSS/DCB */ 874 dglort.inner_rss = 1; 875 dglort.rss_l = fls(interface->ring_feature[RING_F_RSS].mask); 876 dglort.pc_l = fls(interface->ring_feature[RING_F_QOS].mask); 877 hw->mac.ops.configure_dglort_map(hw, &dglort); 878 879 /* assign GLORT per queue for queue mapped testing */ 880 if (interface->glort_count > 64) { 881 memset(&dglort, 0, sizeof(dglort)); 882 dglort.inner_rss = 1; 883 dglort.glort = interface->glort + 64; 884 dglort.idx = fm10k_dglort_pf_queue; 885 dglort.queue_l = fls(interface->num_rx_queues - 1); 886 hw->mac.ops.configure_dglort_map(hw, &dglort); 887 } 888 889 /* assign glort value for RSS/DCB specific to this interface */ 890 memset(&dglort, 0, sizeof(dglort)); 891 dglort.inner_rss = 1; 892 dglort.glort = interface->glort; 893 dglort.rss_l = fls(interface->ring_feature[RING_F_RSS].mask); 894 dglort.pc_l = fls(interface->ring_feature[RING_F_QOS].mask); 895 /* configure DGLORT mapping for RSS/DCB */ 896 dglort.idx = fm10k_dglort_pf_rss; 897 if (interface->l2_accel) 898 dglort.shared_l = fls(interface->l2_accel->size); 899 hw->mac.ops.configure_dglort_map(hw, &dglort); 900 } 901 902 /** 903 * fm10k_configure_rx - Configure Receive Unit after Reset 904 * @interface: board private structure 905 * 906 * Configure the Rx unit of the MAC after a reset. 907 **/ 908 static void fm10k_configure_rx(struct fm10k_intfc *interface) 909 { 910 int i; 911 912 /* Configure SWPRI to PC map */ 913 fm10k_configure_swpri_map(interface); 914 915 /* Configure RSS and DGLORT map */ 916 fm10k_configure_dglort(interface); 917 918 /* Setup the HW Rx Head and Tail descriptor pointers */ 919 for (i = 0; i < interface->num_rx_queues; i++) 920 fm10k_configure_rx_ring(interface, interface->rx_ring[i]); 921 922 /* possible poll here to verify that Rx rings are now enabled */ 923 } 924 925 static void fm10k_napi_enable_all(struct fm10k_intfc *interface) 926 { 927 struct fm10k_q_vector *q_vector; 928 int q_idx; 929 930 for (q_idx = 0; q_idx < interface->num_q_vectors; q_idx++) { 931 q_vector = interface->q_vector[q_idx]; 932 napi_enable(&q_vector->napi); 933 } 934 } 935 936 static irqreturn_t fm10k_msix_clean_rings(int __always_unused irq, void *data) 937 { 938 struct fm10k_q_vector *q_vector = data; 939 940 if (q_vector->rx.count || q_vector->tx.count) 941 napi_schedule_irqoff(&q_vector->napi); 942 943 return IRQ_HANDLED; 944 } 945 946 static irqreturn_t fm10k_msix_mbx_vf(int __always_unused irq, void *data) 947 { 948 struct fm10k_intfc *interface = data; 949 struct fm10k_hw *hw = &interface->hw; 950 struct fm10k_mbx_info *mbx = &hw->mbx; 951 952 /* re-enable mailbox interrupt and indicate 20us delay */ 953 fm10k_write_reg(hw, FM10K_VFITR(FM10K_MBX_VECTOR), 954 (FM10K_MBX_INT_DELAY >> hw->mac.itr_scale) | 955 FM10K_ITR_ENABLE); 956 957 /* service upstream mailbox */ 958 if (fm10k_mbx_trylock(interface)) { 959 mbx->ops.process(hw, mbx); 960 fm10k_mbx_unlock(interface); 961 } 962 963 hw->mac.get_host_state = true; 964 fm10k_service_event_schedule(interface); 965 966 return IRQ_HANDLED; 967 } 968 969 #ifdef CONFIG_NET_POLL_CONTROLLER 970 /** 971 * fm10k_netpoll - A Polling 'interrupt' handler 972 * @netdev: network interface device structure 973 * 974 * This is used by netconsole to send skbs without having to re-enable 975 * interrupts. It's not called while the normal interrupt routine is executing. 976 **/ 977 void fm10k_netpoll(struct net_device *netdev) 978 { 979 struct fm10k_intfc *interface = netdev_priv(netdev); 980 int i; 981 982 /* if interface is down do nothing */ 983 if (test_bit(__FM10K_DOWN, &interface->state)) 984 return; 985 986 for (i = 0; i < interface->num_q_vectors; i++) 987 fm10k_msix_clean_rings(0, interface->q_vector[i]); 988 } 989 990 #endif 991 #define FM10K_ERR_MSG(type) case (type): error = #type; break 992 static void fm10k_handle_fault(struct fm10k_intfc *interface, int type, 993 struct fm10k_fault *fault) 994 { 995 struct pci_dev *pdev = interface->pdev; 996 struct fm10k_hw *hw = &interface->hw; 997 struct fm10k_iov_data *iov_data = interface->iov_data; 998 char *error; 999 1000 switch (type) { 1001 case FM10K_PCA_FAULT: 1002 switch (fault->type) { 1003 default: 1004 error = "Unknown PCA error"; 1005 break; 1006 FM10K_ERR_MSG(PCA_NO_FAULT); 1007 FM10K_ERR_MSG(PCA_UNMAPPED_ADDR); 1008 FM10K_ERR_MSG(PCA_BAD_QACCESS_PF); 1009 FM10K_ERR_MSG(PCA_BAD_QACCESS_VF); 1010 FM10K_ERR_MSG(PCA_MALICIOUS_REQ); 1011 FM10K_ERR_MSG(PCA_POISONED_TLP); 1012 FM10K_ERR_MSG(PCA_TLP_ABORT); 1013 } 1014 break; 1015 case FM10K_THI_FAULT: 1016 switch (fault->type) { 1017 default: 1018 error = "Unknown THI error"; 1019 break; 1020 FM10K_ERR_MSG(THI_NO_FAULT); 1021 FM10K_ERR_MSG(THI_MAL_DIS_Q_FAULT); 1022 } 1023 break; 1024 case FM10K_FUM_FAULT: 1025 switch (fault->type) { 1026 default: 1027 error = "Unknown FUM error"; 1028 break; 1029 FM10K_ERR_MSG(FUM_NO_FAULT); 1030 FM10K_ERR_MSG(FUM_UNMAPPED_ADDR); 1031 FM10K_ERR_MSG(FUM_BAD_VF_QACCESS); 1032 FM10K_ERR_MSG(FUM_ADD_DECODE_ERR); 1033 FM10K_ERR_MSG(FUM_RO_ERROR); 1034 FM10K_ERR_MSG(FUM_QPRC_CRC_ERROR); 1035 FM10K_ERR_MSG(FUM_CSR_TIMEOUT); 1036 FM10K_ERR_MSG(FUM_INVALID_TYPE); 1037 FM10K_ERR_MSG(FUM_INVALID_LENGTH); 1038 FM10K_ERR_MSG(FUM_INVALID_BE); 1039 FM10K_ERR_MSG(FUM_INVALID_ALIGN); 1040 } 1041 break; 1042 default: 1043 error = "Undocumented fault"; 1044 break; 1045 } 1046 1047 dev_warn(&pdev->dev, 1048 "%s Address: 0x%llx SpecInfo: 0x%x Func: %02x.%0x\n", 1049 error, fault->address, fault->specinfo, 1050 PCI_SLOT(fault->func), PCI_FUNC(fault->func)); 1051 1052 /* For VF faults, clear out the respective LPORT, reset the queue 1053 * resources, and then reconnect to the mailbox. This allows the 1054 * VF in question to resume behavior. For transient faults that are 1055 * the result of non-malicious behavior this will log the fault and 1056 * allow the VF to resume functionality. Obviously for malicious VFs 1057 * they will be able to attempt malicious behavior again. In this 1058 * case, the system administrator will need to step in and manually 1059 * remove or disable the VF in question. 1060 */ 1061 if (fault->func && iov_data) { 1062 int vf = fault->func - 1; 1063 struct fm10k_vf_info *vf_info = &iov_data->vf_info[vf]; 1064 1065 hw->iov.ops.reset_lport(hw, vf_info); 1066 hw->iov.ops.reset_resources(hw, vf_info); 1067 1068 /* reset_lport disables the VF, so re-enable it */ 1069 hw->iov.ops.set_lport(hw, vf_info, vf, 1070 FM10K_VF_FLAG_MULTI_CAPABLE); 1071 1072 /* reset_resources will disconnect from the mbx */ 1073 vf_info->mbx.ops.connect(hw, &vf_info->mbx); 1074 } 1075 } 1076 1077 static void fm10k_report_fault(struct fm10k_intfc *interface, u32 eicr) 1078 { 1079 struct fm10k_hw *hw = &interface->hw; 1080 struct fm10k_fault fault = { 0 }; 1081 int type, err; 1082 1083 for (eicr &= FM10K_EICR_FAULT_MASK, type = FM10K_PCA_FAULT; 1084 eicr; 1085 eicr >>= 1, type += FM10K_FAULT_SIZE) { 1086 /* only check if there is an error reported */ 1087 if (!(eicr & 0x1)) 1088 continue; 1089 1090 /* retrieve fault info */ 1091 err = hw->mac.ops.get_fault(hw, type, &fault); 1092 if (err) { 1093 dev_err(&interface->pdev->dev, 1094 "error reading fault\n"); 1095 continue; 1096 } 1097 1098 fm10k_handle_fault(interface, type, &fault); 1099 } 1100 } 1101 1102 static void fm10k_reset_drop_on_empty(struct fm10k_intfc *interface, u32 eicr) 1103 { 1104 struct fm10k_hw *hw = &interface->hw; 1105 const u32 rxdctl = FM10K_RXDCTL_WRITE_BACK_MIN_DELAY; 1106 u32 maxholdq; 1107 int q; 1108 1109 if (!(eicr & FM10K_EICR_MAXHOLDTIME)) 1110 return; 1111 1112 maxholdq = fm10k_read_reg(hw, FM10K_MAXHOLDQ(7)); 1113 if (maxholdq) 1114 fm10k_write_reg(hw, FM10K_MAXHOLDQ(7), maxholdq); 1115 for (q = 255;;) { 1116 if (maxholdq & BIT(31)) { 1117 if (q < FM10K_MAX_QUEUES_PF) { 1118 interface->rx_overrun_pf++; 1119 fm10k_write_reg(hw, FM10K_RXDCTL(q), rxdctl); 1120 } else { 1121 interface->rx_overrun_vf++; 1122 } 1123 } 1124 1125 maxholdq *= 2; 1126 if (!maxholdq) 1127 q &= ~(32 - 1); 1128 1129 if (!q) 1130 break; 1131 1132 if (q-- % 32) 1133 continue; 1134 1135 maxholdq = fm10k_read_reg(hw, FM10K_MAXHOLDQ(q / 32)); 1136 if (maxholdq) 1137 fm10k_write_reg(hw, FM10K_MAXHOLDQ(q / 32), maxholdq); 1138 } 1139 } 1140 1141 static irqreturn_t fm10k_msix_mbx_pf(int __always_unused irq, void *data) 1142 { 1143 struct fm10k_intfc *interface = data; 1144 struct fm10k_hw *hw = &interface->hw; 1145 struct fm10k_mbx_info *mbx = &hw->mbx; 1146 u32 eicr; 1147 s32 err = 0; 1148 1149 /* unmask any set bits related to this interrupt */ 1150 eicr = fm10k_read_reg(hw, FM10K_EICR); 1151 fm10k_write_reg(hw, FM10K_EICR, eicr & (FM10K_EICR_MAILBOX | 1152 FM10K_EICR_SWITCHREADY | 1153 FM10K_EICR_SWITCHNOTREADY)); 1154 1155 /* report any faults found to the message log */ 1156 fm10k_report_fault(interface, eicr); 1157 1158 /* reset any queues disabled due to receiver overrun */ 1159 fm10k_reset_drop_on_empty(interface, eicr); 1160 1161 /* service mailboxes */ 1162 if (fm10k_mbx_trylock(interface)) { 1163 err = mbx->ops.process(hw, mbx); 1164 /* handle VFLRE events */ 1165 fm10k_iov_event(interface); 1166 fm10k_mbx_unlock(interface); 1167 } 1168 1169 if (err == FM10K_ERR_RESET_REQUESTED) 1170 interface->flags |= FM10K_FLAG_RESET_REQUESTED; 1171 1172 /* if switch toggled state we should reset GLORTs */ 1173 if (eicr & FM10K_EICR_SWITCHNOTREADY) { 1174 /* force link down for at least 4 seconds */ 1175 interface->link_down_event = jiffies + (4 * HZ); 1176 set_bit(__FM10K_LINK_DOWN, &interface->state); 1177 1178 /* reset dglort_map back to no config */ 1179 hw->mac.dglort_map = FM10K_DGLORTMAP_NONE; 1180 } 1181 1182 /* we should validate host state after interrupt event */ 1183 hw->mac.get_host_state = true; 1184 1185 /* validate host state, and handle VF mailboxes in the service task */ 1186 fm10k_service_event_schedule(interface); 1187 1188 /* re-enable mailbox interrupt and indicate 20us delay */ 1189 fm10k_write_reg(hw, FM10K_ITR(FM10K_MBX_VECTOR), 1190 (FM10K_MBX_INT_DELAY >> hw->mac.itr_scale) | 1191 FM10K_ITR_ENABLE); 1192 1193 return IRQ_HANDLED; 1194 } 1195 1196 void fm10k_mbx_free_irq(struct fm10k_intfc *interface) 1197 { 1198 struct fm10k_hw *hw = &interface->hw; 1199 struct msix_entry *entry; 1200 int itr_reg; 1201 1202 /* no mailbox IRQ to free if MSI-X is not enabled */ 1203 if (!interface->msix_entries) 1204 return; 1205 1206 entry = &interface->msix_entries[FM10K_MBX_VECTOR]; 1207 1208 /* disconnect the mailbox */ 1209 hw->mbx.ops.disconnect(hw, &hw->mbx); 1210 1211 /* disable Mailbox cause */ 1212 if (hw->mac.type == fm10k_mac_pf) { 1213 fm10k_write_reg(hw, FM10K_EIMR, 1214 FM10K_EIMR_DISABLE(PCA_FAULT) | 1215 FM10K_EIMR_DISABLE(FUM_FAULT) | 1216 FM10K_EIMR_DISABLE(MAILBOX) | 1217 FM10K_EIMR_DISABLE(SWITCHREADY) | 1218 FM10K_EIMR_DISABLE(SWITCHNOTREADY) | 1219 FM10K_EIMR_DISABLE(SRAMERROR) | 1220 FM10K_EIMR_DISABLE(VFLR) | 1221 FM10K_EIMR_DISABLE(MAXHOLDTIME)); 1222 itr_reg = FM10K_ITR(FM10K_MBX_VECTOR); 1223 } else { 1224 itr_reg = FM10K_VFITR(FM10K_MBX_VECTOR); 1225 } 1226 1227 fm10k_write_reg(hw, itr_reg, FM10K_ITR_MASK_SET); 1228 1229 free_irq(entry->vector, interface); 1230 } 1231 1232 static s32 fm10k_mbx_mac_addr(struct fm10k_hw *hw, u32 **results, 1233 struct fm10k_mbx_info *mbx) 1234 { 1235 bool vlan_override = hw->mac.vlan_override; 1236 u16 default_vid = hw->mac.default_vid; 1237 struct fm10k_intfc *interface; 1238 s32 err; 1239 1240 err = fm10k_msg_mac_vlan_vf(hw, results, mbx); 1241 if (err) 1242 return err; 1243 1244 interface = container_of(hw, struct fm10k_intfc, hw); 1245 1246 /* MAC was changed so we need reset */ 1247 if (is_valid_ether_addr(hw->mac.perm_addr) && 1248 !ether_addr_equal(hw->mac.perm_addr, hw->mac.addr)) 1249 interface->flags |= FM10K_FLAG_RESET_REQUESTED; 1250 1251 /* VLAN override was changed, or default VLAN changed */ 1252 if ((vlan_override != hw->mac.vlan_override) || 1253 (default_vid != hw->mac.default_vid)) 1254 interface->flags |= FM10K_FLAG_RESET_REQUESTED; 1255 1256 return 0; 1257 } 1258 1259 /* generic error handler for mailbox issues */ 1260 static s32 fm10k_mbx_error(struct fm10k_hw *hw, u32 **results, 1261 struct fm10k_mbx_info __always_unused *mbx) 1262 { 1263 struct fm10k_intfc *interface; 1264 struct pci_dev *pdev; 1265 1266 interface = container_of(hw, struct fm10k_intfc, hw); 1267 pdev = interface->pdev; 1268 1269 dev_err(&pdev->dev, "Unknown message ID %u\n", 1270 **results & FM10K_TLV_ID_MASK); 1271 1272 return 0; 1273 } 1274 1275 static const struct fm10k_msg_data vf_mbx_data[] = { 1276 FM10K_TLV_MSG_TEST_HANDLER(fm10k_tlv_msg_test), 1277 FM10K_VF_MSG_MAC_VLAN_HANDLER(fm10k_mbx_mac_addr), 1278 FM10K_VF_MSG_LPORT_STATE_HANDLER(fm10k_msg_lport_state_vf), 1279 FM10K_TLV_MSG_ERROR_HANDLER(fm10k_mbx_error), 1280 }; 1281 1282 static int fm10k_mbx_request_irq_vf(struct fm10k_intfc *interface) 1283 { 1284 struct msix_entry *entry = &interface->msix_entries[FM10K_MBX_VECTOR]; 1285 struct net_device *dev = interface->netdev; 1286 struct fm10k_hw *hw = &interface->hw; 1287 int err; 1288 1289 /* Use timer0 for interrupt moderation on the mailbox */ 1290 u32 itr = entry->entry | FM10K_INT_MAP_TIMER0; 1291 1292 /* register mailbox handlers */ 1293 err = hw->mbx.ops.register_handlers(&hw->mbx, vf_mbx_data); 1294 if (err) 1295 return err; 1296 1297 /* request the IRQ */ 1298 err = request_irq(entry->vector, fm10k_msix_mbx_vf, 0, 1299 dev->name, interface); 1300 if (err) { 1301 netif_err(interface, probe, dev, 1302 "request_irq for msix_mbx failed: %d\n", err); 1303 return err; 1304 } 1305 1306 /* map all of the interrupt sources */ 1307 fm10k_write_reg(hw, FM10K_VFINT_MAP, itr); 1308 1309 /* enable interrupt */ 1310 fm10k_write_reg(hw, FM10K_VFITR(entry->entry), FM10K_ITR_ENABLE); 1311 1312 return 0; 1313 } 1314 1315 static s32 fm10k_lport_map(struct fm10k_hw *hw, u32 **results, 1316 struct fm10k_mbx_info *mbx) 1317 { 1318 struct fm10k_intfc *interface; 1319 u32 dglort_map = hw->mac.dglort_map; 1320 s32 err; 1321 1322 interface = container_of(hw, struct fm10k_intfc, hw); 1323 1324 err = fm10k_msg_err_pf(hw, results, mbx); 1325 if (!err && hw->swapi.status) { 1326 /* force link down for a reasonable delay */ 1327 interface->link_down_event = jiffies + (2 * HZ); 1328 set_bit(__FM10K_LINK_DOWN, &interface->state); 1329 1330 /* reset dglort_map back to no config */ 1331 hw->mac.dglort_map = FM10K_DGLORTMAP_NONE; 1332 1333 fm10k_service_event_schedule(interface); 1334 1335 /* prevent overloading kernel message buffer */ 1336 if (interface->lport_map_failed) 1337 return 0; 1338 1339 interface->lport_map_failed = true; 1340 1341 if (hw->swapi.status == FM10K_MSG_ERR_PEP_NOT_SCHEDULED) 1342 dev_warn(&interface->pdev->dev, 1343 "cannot obtain link because the host interface is configured for a PCIe host interface bandwidth of zero\n"); 1344 dev_warn(&interface->pdev->dev, 1345 "request logical port map failed: %d\n", 1346 hw->swapi.status); 1347 1348 return 0; 1349 } 1350 1351 err = fm10k_msg_lport_map_pf(hw, results, mbx); 1352 if (err) 1353 return err; 1354 1355 interface->lport_map_failed = false; 1356 1357 /* we need to reset if port count was just updated */ 1358 if (dglort_map != hw->mac.dglort_map) 1359 interface->flags |= FM10K_FLAG_RESET_REQUESTED; 1360 1361 return 0; 1362 } 1363 1364 static s32 fm10k_update_pvid(struct fm10k_hw *hw, u32 **results, 1365 struct fm10k_mbx_info __always_unused *mbx) 1366 { 1367 struct fm10k_intfc *interface; 1368 u16 glort, pvid; 1369 u32 pvid_update; 1370 s32 err; 1371 1372 err = fm10k_tlv_attr_get_u32(results[FM10K_PF_ATTR_ID_UPDATE_PVID], 1373 &pvid_update); 1374 if (err) 1375 return err; 1376 1377 /* extract values from the pvid update */ 1378 glort = FM10K_MSG_HDR_FIELD_GET(pvid_update, UPDATE_PVID_GLORT); 1379 pvid = FM10K_MSG_HDR_FIELD_GET(pvid_update, UPDATE_PVID_PVID); 1380 1381 /* if glort is not valid return error */ 1382 if (!fm10k_glort_valid_pf(hw, glort)) 1383 return FM10K_ERR_PARAM; 1384 1385 /* verify VLAN ID is valid */ 1386 if (pvid >= FM10K_VLAN_TABLE_VID_MAX) 1387 return FM10K_ERR_PARAM; 1388 1389 interface = container_of(hw, struct fm10k_intfc, hw); 1390 1391 /* check to see if this belongs to one of the VFs */ 1392 err = fm10k_iov_update_pvid(interface, glort, pvid); 1393 if (!err) 1394 return 0; 1395 1396 /* we need to reset if default VLAN was just updated */ 1397 if (pvid != hw->mac.default_vid) 1398 interface->flags |= FM10K_FLAG_RESET_REQUESTED; 1399 1400 hw->mac.default_vid = pvid; 1401 1402 return 0; 1403 } 1404 1405 static const struct fm10k_msg_data pf_mbx_data[] = { 1406 FM10K_PF_MSG_ERR_HANDLER(XCAST_MODES, fm10k_msg_err_pf), 1407 FM10K_PF_MSG_ERR_HANDLER(UPDATE_MAC_FWD_RULE, fm10k_msg_err_pf), 1408 FM10K_PF_MSG_LPORT_MAP_HANDLER(fm10k_lport_map), 1409 FM10K_PF_MSG_ERR_HANDLER(LPORT_CREATE, fm10k_msg_err_pf), 1410 FM10K_PF_MSG_ERR_HANDLER(LPORT_DELETE, fm10k_msg_err_pf), 1411 FM10K_PF_MSG_UPDATE_PVID_HANDLER(fm10k_update_pvid), 1412 FM10K_TLV_MSG_ERROR_HANDLER(fm10k_mbx_error), 1413 }; 1414 1415 static int fm10k_mbx_request_irq_pf(struct fm10k_intfc *interface) 1416 { 1417 struct msix_entry *entry = &interface->msix_entries[FM10K_MBX_VECTOR]; 1418 struct net_device *dev = interface->netdev; 1419 struct fm10k_hw *hw = &interface->hw; 1420 int err; 1421 1422 /* Use timer0 for interrupt moderation on the mailbox */ 1423 u32 mbx_itr = entry->entry | FM10K_INT_MAP_TIMER0; 1424 u32 other_itr = entry->entry | FM10K_INT_MAP_IMMEDIATE; 1425 1426 /* register mailbox handlers */ 1427 err = hw->mbx.ops.register_handlers(&hw->mbx, pf_mbx_data); 1428 if (err) 1429 return err; 1430 1431 /* request the IRQ */ 1432 err = request_irq(entry->vector, fm10k_msix_mbx_pf, 0, 1433 dev->name, interface); 1434 if (err) { 1435 netif_err(interface, probe, dev, 1436 "request_irq for msix_mbx failed: %d\n", err); 1437 return err; 1438 } 1439 1440 /* Enable interrupts w/ no moderation for "other" interrupts */ 1441 fm10k_write_reg(hw, FM10K_INT_MAP(fm10k_int_pcie_fault), other_itr); 1442 fm10k_write_reg(hw, FM10K_INT_MAP(fm10k_int_switch_up_down), other_itr); 1443 fm10k_write_reg(hw, FM10K_INT_MAP(fm10k_int_sram), other_itr); 1444 fm10k_write_reg(hw, FM10K_INT_MAP(fm10k_int_max_hold_time), other_itr); 1445 fm10k_write_reg(hw, FM10K_INT_MAP(fm10k_int_vflr), other_itr); 1446 1447 /* Enable interrupts w/ moderation for mailbox */ 1448 fm10k_write_reg(hw, FM10K_INT_MAP(fm10k_int_mailbox), mbx_itr); 1449 1450 /* Enable individual interrupt causes */ 1451 fm10k_write_reg(hw, FM10K_EIMR, FM10K_EIMR_ENABLE(PCA_FAULT) | 1452 FM10K_EIMR_ENABLE(FUM_FAULT) | 1453 FM10K_EIMR_ENABLE(MAILBOX) | 1454 FM10K_EIMR_ENABLE(SWITCHREADY) | 1455 FM10K_EIMR_ENABLE(SWITCHNOTREADY) | 1456 FM10K_EIMR_ENABLE(SRAMERROR) | 1457 FM10K_EIMR_ENABLE(VFLR) | 1458 FM10K_EIMR_ENABLE(MAXHOLDTIME)); 1459 1460 /* enable interrupt */ 1461 fm10k_write_reg(hw, FM10K_ITR(entry->entry), FM10K_ITR_ENABLE); 1462 1463 return 0; 1464 } 1465 1466 int fm10k_mbx_request_irq(struct fm10k_intfc *interface) 1467 { 1468 struct fm10k_hw *hw = &interface->hw; 1469 int err; 1470 1471 /* enable Mailbox cause */ 1472 if (hw->mac.type == fm10k_mac_pf) 1473 err = fm10k_mbx_request_irq_pf(interface); 1474 else 1475 err = fm10k_mbx_request_irq_vf(interface); 1476 if (err) 1477 return err; 1478 1479 /* connect mailbox */ 1480 err = hw->mbx.ops.connect(hw, &hw->mbx); 1481 1482 /* if the mailbox failed to connect, then free IRQ */ 1483 if (err) 1484 fm10k_mbx_free_irq(interface); 1485 1486 return err; 1487 } 1488 1489 /** 1490 * fm10k_qv_free_irq - release interrupts associated with queue vectors 1491 * @interface: board private structure 1492 * 1493 * Release all interrupts associated with this interface 1494 **/ 1495 void fm10k_qv_free_irq(struct fm10k_intfc *interface) 1496 { 1497 int vector = interface->num_q_vectors; 1498 struct fm10k_hw *hw = &interface->hw; 1499 struct msix_entry *entry; 1500 1501 entry = &interface->msix_entries[NON_Q_VECTORS(hw) + vector]; 1502 1503 while (vector) { 1504 struct fm10k_q_vector *q_vector; 1505 1506 vector--; 1507 entry--; 1508 q_vector = interface->q_vector[vector]; 1509 1510 if (!q_vector->tx.count && !q_vector->rx.count) 1511 continue; 1512 1513 /* clear the affinity_mask in the IRQ descriptor */ 1514 irq_set_affinity_hint(entry->vector, NULL); 1515 1516 /* disable interrupts */ 1517 writel(FM10K_ITR_MASK_SET, q_vector->itr); 1518 1519 free_irq(entry->vector, q_vector); 1520 } 1521 } 1522 1523 /** 1524 * fm10k_qv_request_irq - initialize interrupts for queue vectors 1525 * @interface: board private structure 1526 * 1527 * Attempts to configure interrupts using the best available 1528 * capabilities of the hardware and kernel. 1529 **/ 1530 int fm10k_qv_request_irq(struct fm10k_intfc *interface) 1531 { 1532 struct net_device *dev = interface->netdev; 1533 struct fm10k_hw *hw = &interface->hw; 1534 struct msix_entry *entry; 1535 int ri = 0, ti = 0; 1536 int vector, err; 1537 1538 entry = &interface->msix_entries[NON_Q_VECTORS(hw)]; 1539 1540 for (vector = 0; vector < interface->num_q_vectors; vector++) { 1541 struct fm10k_q_vector *q_vector = interface->q_vector[vector]; 1542 1543 /* name the vector */ 1544 if (q_vector->tx.count && q_vector->rx.count) { 1545 snprintf(q_vector->name, sizeof(q_vector->name) - 1, 1546 "%s-TxRx-%d", dev->name, ri++); 1547 ti++; 1548 } else if (q_vector->rx.count) { 1549 snprintf(q_vector->name, sizeof(q_vector->name) - 1, 1550 "%s-rx-%d", dev->name, ri++); 1551 } else if (q_vector->tx.count) { 1552 snprintf(q_vector->name, sizeof(q_vector->name) - 1, 1553 "%s-tx-%d", dev->name, ti++); 1554 } else { 1555 /* skip this unused q_vector */ 1556 continue; 1557 } 1558 1559 /* Assign ITR register to q_vector */ 1560 q_vector->itr = (hw->mac.type == fm10k_mac_pf) ? 1561 &interface->uc_addr[FM10K_ITR(entry->entry)] : 1562 &interface->uc_addr[FM10K_VFITR(entry->entry)]; 1563 1564 /* request the IRQ */ 1565 err = request_irq(entry->vector, &fm10k_msix_clean_rings, 0, 1566 q_vector->name, q_vector); 1567 if (err) { 1568 netif_err(interface, probe, dev, 1569 "request_irq failed for MSIX interrupt Error: %d\n", 1570 err); 1571 goto err_out; 1572 } 1573 1574 /* assign the mask for this irq */ 1575 irq_set_affinity_hint(entry->vector, &q_vector->affinity_mask); 1576 1577 /* Enable q_vector */ 1578 writel(FM10K_ITR_ENABLE, q_vector->itr); 1579 1580 entry++; 1581 } 1582 1583 return 0; 1584 1585 err_out: 1586 /* wind through the ring freeing all entries and vectors */ 1587 while (vector) { 1588 struct fm10k_q_vector *q_vector; 1589 1590 entry--; 1591 vector--; 1592 q_vector = interface->q_vector[vector]; 1593 1594 if (!q_vector->tx.count && !q_vector->rx.count) 1595 continue; 1596 1597 /* clear the affinity_mask in the IRQ descriptor */ 1598 irq_set_affinity_hint(entry->vector, NULL); 1599 1600 /* disable interrupts */ 1601 writel(FM10K_ITR_MASK_SET, q_vector->itr); 1602 1603 free_irq(entry->vector, q_vector); 1604 } 1605 1606 return err; 1607 } 1608 1609 void fm10k_up(struct fm10k_intfc *interface) 1610 { 1611 struct fm10k_hw *hw = &interface->hw; 1612 1613 /* Enable Tx/Rx DMA */ 1614 hw->mac.ops.start_hw(hw); 1615 1616 /* configure Tx descriptor rings */ 1617 fm10k_configure_tx(interface); 1618 1619 /* configure Rx descriptor rings */ 1620 fm10k_configure_rx(interface); 1621 1622 /* configure interrupts */ 1623 hw->mac.ops.update_int_moderator(hw); 1624 1625 /* enable statistics capture again */ 1626 clear_bit(__FM10K_UPDATING_STATS, &interface->state); 1627 1628 /* clear down bit to indicate we are ready to go */ 1629 clear_bit(__FM10K_DOWN, &interface->state); 1630 1631 /* enable polling cleanups */ 1632 fm10k_napi_enable_all(interface); 1633 1634 /* re-establish Rx filters */ 1635 fm10k_restore_rx_state(interface); 1636 1637 /* enable transmits */ 1638 netif_tx_start_all_queues(interface->netdev); 1639 1640 /* kick off the service timer now */ 1641 hw->mac.get_host_state = true; 1642 mod_timer(&interface->service_timer, jiffies); 1643 } 1644 1645 static void fm10k_napi_disable_all(struct fm10k_intfc *interface) 1646 { 1647 struct fm10k_q_vector *q_vector; 1648 int q_idx; 1649 1650 for (q_idx = 0; q_idx < interface->num_q_vectors; q_idx++) { 1651 q_vector = interface->q_vector[q_idx]; 1652 napi_disable(&q_vector->napi); 1653 } 1654 } 1655 1656 void fm10k_down(struct fm10k_intfc *interface) 1657 { 1658 struct net_device *netdev = interface->netdev; 1659 struct fm10k_hw *hw = &interface->hw; 1660 int err, i = 0, count = 0; 1661 1662 /* signal that we are down to the interrupt handler and service task */ 1663 if (test_and_set_bit(__FM10K_DOWN, &interface->state)) 1664 return; 1665 1666 /* call carrier off first to avoid false dev_watchdog timeouts */ 1667 netif_carrier_off(netdev); 1668 1669 /* disable transmits */ 1670 netif_tx_stop_all_queues(netdev); 1671 netif_tx_disable(netdev); 1672 1673 /* reset Rx filters */ 1674 fm10k_reset_rx_state(interface); 1675 1676 /* disable polling routines */ 1677 fm10k_napi_disable_all(interface); 1678 1679 /* capture stats one last time before stopping interface */ 1680 fm10k_update_stats(interface); 1681 1682 /* prevent updating statistics while we're down */ 1683 while (test_and_set_bit(__FM10K_UPDATING_STATS, &interface->state)) 1684 usleep_range(1000, 2000); 1685 1686 /* skip waiting for TX DMA if we lost PCIe link */ 1687 if (FM10K_REMOVED(hw->hw_addr)) 1688 goto skip_tx_dma_drain; 1689 1690 /* In some rare circumstances it can take a while for Tx queues to 1691 * quiesce and be fully disabled. Attempt to .stop_hw() first, and 1692 * then if we get ERR_REQUESTS_PENDING, go ahead and wait in a loop 1693 * until the Tx queues have emptied, or until a number of retries. If 1694 * we fail to clear within the retry loop, we will issue a warning 1695 * indicating that Tx DMA is probably hung. Note this means we call 1696 * .stop_hw() twice but this shouldn't cause any problems. 1697 */ 1698 err = hw->mac.ops.stop_hw(hw); 1699 if (err != FM10K_ERR_REQUESTS_PENDING) 1700 goto skip_tx_dma_drain; 1701 1702 #define TX_DMA_DRAIN_RETRIES 25 1703 for (count = 0; count < TX_DMA_DRAIN_RETRIES; count++) { 1704 usleep_range(10000, 20000); 1705 1706 /* start checking at the last ring to have pending Tx */ 1707 for (; i < interface->num_tx_queues; i++) 1708 if (fm10k_get_tx_pending(interface->tx_ring[i], false)) 1709 break; 1710 1711 /* if all the queues are drained, we can break now */ 1712 if (i == interface->num_tx_queues) 1713 break; 1714 } 1715 1716 if (count >= TX_DMA_DRAIN_RETRIES) 1717 dev_err(&interface->pdev->dev, 1718 "Tx queues failed to drain after %d tries. Tx DMA is probably hung.\n", 1719 count); 1720 skip_tx_dma_drain: 1721 /* Disable DMA engine for Tx/Rx */ 1722 err = hw->mac.ops.stop_hw(hw); 1723 if (err == FM10K_ERR_REQUESTS_PENDING) 1724 dev_err(&interface->pdev->dev, 1725 "due to pending requests hw was not shut down gracefully\n"); 1726 else if (err) 1727 dev_err(&interface->pdev->dev, "stop_hw failed: %d\n", err); 1728 1729 /* free any buffers still on the rings */ 1730 fm10k_clean_all_tx_rings(interface); 1731 fm10k_clean_all_rx_rings(interface); 1732 } 1733 1734 /** 1735 * fm10k_sw_init - Initialize general software structures 1736 * @interface: host interface private structure to initialize 1737 * 1738 * fm10k_sw_init initializes the interface private data structure. 1739 * Fields are initialized based on PCI device information and 1740 * OS network device settings (MTU size). 1741 **/ 1742 static int fm10k_sw_init(struct fm10k_intfc *interface, 1743 const struct pci_device_id *ent) 1744 { 1745 const struct fm10k_info *fi = fm10k_info_tbl[ent->driver_data]; 1746 struct fm10k_hw *hw = &interface->hw; 1747 struct pci_dev *pdev = interface->pdev; 1748 struct net_device *netdev = interface->netdev; 1749 u32 rss_key[FM10K_RSSRK_SIZE]; 1750 unsigned int rss; 1751 int err; 1752 1753 /* initialize back pointer */ 1754 hw->back = interface; 1755 hw->hw_addr = interface->uc_addr; 1756 1757 /* PCI config space info */ 1758 hw->vendor_id = pdev->vendor; 1759 hw->device_id = pdev->device; 1760 hw->revision_id = pdev->revision; 1761 hw->subsystem_vendor_id = pdev->subsystem_vendor; 1762 hw->subsystem_device_id = pdev->subsystem_device; 1763 1764 /* Setup hw api */ 1765 memcpy(&hw->mac.ops, fi->mac_ops, sizeof(hw->mac.ops)); 1766 hw->mac.type = fi->mac; 1767 1768 /* Setup IOV handlers */ 1769 if (fi->iov_ops) 1770 memcpy(&hw->iov.ops, fi->iov_ops, sizeof(hw->iov.ops)); 1771 1772 /* Set common capability flags and settings */ 1773 rss = min_t(int, FM10K_MAX_RSS_INDICES, num_online_cpus()); 1774 interface->ring_feature[RING_F_RSS].limit = rss; 1775 fi->get_invariants(hw); 1776 1777 /* pick up the PCIe bus settings for reporting later */ 1778 if (hw->mac.ops.get_bus_info) 1779 hw->mac.ops.get_bus_info(hw); 1780 1781 /* limit the usable DMA range */ 1782 if (hw->mac.ops.set_dma_mask) 1783 hw->mac.ops.set_dma_mask(hw, dma_get_mask(&pdev->dev)); 1784 1785 /* update netdev with DMA restrictions */ 1786 if (dma_get_mask(&pdev->dev) > DMA_BIT_MASK(32)) { 1787 netdev->features |= NETIF_F_HIGHDMA; 1788 netdev->vlan_features |= NETIF_F_HIGHDMA; 1789 } 1790 1791 /* delay any future reset requests */ 1792 interface->last_reset = jiffies + (10 * HZ); 1793 1794 /* reset and initialize the hardware so it is in a known state */ 1795 err = hw->mac.ops.reset_hw(hw); 1796 if (err) { 1797 dev_err(&pdev->dev, "reset_hw failed: %d\n", err); 1798 return err; 1799 } 1800 1801 err = hw->mac.ops.init_hw(hw); 1802 if (err) { 1803 dev_err(&pdev->dev, "init_hw failed: %d\n", err); 1804 return err; 1805 } 1806 1807 /* initialize hardware statistics */ 1808 hw->mac.ops.update_hw_stats(hw, &interface->stats); 1809 1810 /* Set upper limit on IOV VFs that can be allocated */ 1811 pci_sriov_set_totalvfs(pdev, hw->iov.total_vfs); 1812 1813 /* Start with random Ethernet address */ 1814 eth_random_addr(hw->mac.addr); 1815 1816 /* Initialize MAC address from hardware */ 1817 err = hw->mac.ops.read_mac_addr(hw); 1818 if (err) { 1819 dev_warn(&pdev->dev, 1820 "Failed to obtain MAC address defaulting to random\n"); 1821 /* tag address assignment as random */ 1822 netdev->addr_assign_type |= NET_ADDR_RANDOM; 1823 } 1824 1825 ether_addr_copy(netdev->dev_addr, hw->mac.addr); 1826 ether_addr_copy(netdev->perm_addr, hw->mac.addr); 1827 1828 if (!is_valid_ether_addr(netdev->perm_addr)) { 1829 dev_err(&pdev->dev, "Invalid MAC Address\n"); 1830 return -EIO; 1831 } 1832 1833 /* initialize DCBNL interface */ 1834 fm10k_dcbnl_set_ops(netdev); 1835 1836 /* set default ring sizes */ 1837 interface->tx_ring_count = FM10K_DEFAULT_TXD; 1838 interface->rx_ring_count = FM10K_DEFAULT_RXD; 1839 1840 /* set default interrupt moderation */ 1841 interface->tx_itr = FM10K_TX_ITR_DEFAULT; 1842 interface->rx_itr = FM10K_ITR_ADAPTIVE | FM10K_RX_ITR_DEFAULT; 1843 1844 /* initialize udp port lists */ 1845 INIT_LIST_HEAD(&interface->vxlan_port); 1846 INIT_LIST_HEAD(&interface->geneve_port); 1847 1848 netdev_rss_key_fill(rss_key, sizeof(rss_key)); 1849 memcpy(interface->rssrk, rss_key, sizeof(rss_key)); 1850 1851 /* Start off interface as being down */ 1852 set_bit(__FM10K_DOWN, &interface->state); 1853 set_bit(__FM10K_UPDATING_STATS, &interface->state); 1854 1855 return 0; 1856 } 1857 1858 static void fm10k_slot_warn(struct fm10k_intfc *interface) 1859 { 1860 enum pcie_link_width width = PCIE_LNK_WIDTH_UNKNOWN; 1861 enum pci_bus_speed speed = PCI_SPEED_UNKNOWN; 1862 struct fm10k_hw *hw = &interface->hw; 1863 int max_gts = 0, expected_gts = 0; 1864 1865 if (pcie_get_minimum_link(interface->pdev, &speed, &width) || 1866 speed == PCI_SPEED_UNKNOWN || width == PCIE_LNK_WIDTH_UNKNOWN) { 1867 dev_warn(&interface->pdev->dev, 1868 "Unable to determine PCI Express bandwidth.\n"); 1869 return; 1870 } 1871 1872 switch (speed) { 1873 case PCIE_SPEED_2_5GT: 1874 /* 8b/10b encoding reduces max throughput by 20% */ 1875 max_gts = 2 * width; 1876 break; 1877 case PCIE_SPEED_5_0GT: 1878 /* 8b/10b encoding reduces max throughput by 20% */ 1879 max_gts = 4 * width; 1880 break; 1881 case PCIE_SPEED_8_0GT: 1882 /* 128b/130b encoding has less than 2% impact on throughput */ 1883 max_gts = 8 * width; 1884 break; 1885 default: 1886 dev_warn(&interface->pdev->dev, 1887 "Unable to determine PCI Express bandwidth.\n"); 1888 return; 1889 } 1890 1891 dev_info(&interface->pdev->dev, 1892 "PCI Express bandwidth of %dGT/s available\n", 1893 max_gts); 1894 dev_info(&interface->pdev->dev, 1895 "(Speed:%s, Width: x%d, Encoding Loss:%s, Payload:%s)\n", 1896 (speed == PCIE_SPEED_8_0GT ? "8.0GT/s" : 1897 speed == PCIE_SPEED_5_0GT ? "5.0GT/s" : 1898 speed == PCIE_SPEED_2_5GT ? "2.5GT/s" : 1899 "Unknown"), 1900 hw->bus.width, 1901 (speed == PCIE_SPEED_2_5GT ? "20%" : 1902 speed == PCIE_SPEED_5_0GT ? "20%" : 1903 speed == PCIE_SPEED_8_0GT ? "<2%" : 1904 "Unknown"), 1905 (hw->bus.payload == fm10k_bus_payload_128 ? "128B" : 1906 hw->bus.payload == fm10k_bus_payload_256 ? "256B" : 1907 hw->bus.payload == fm10k_bus_payload_512 ? "512B" : 1908 "Unknown")); 1909 1910 switch (hw->bus_caps.speed) { 1911 case fm10k_bus_speed_2500: 1912 /* 8b/10b encoding reduces max throughput by 20% */ 1913 expected_gts = 2 * hw->bus_caps.width; 1914 break; 1915 case fm10k_bus_speed_5000: 1916 /* 8b/10b encoding reduces max throughput by 20% */ 1917 expected_gts = 4 * hw->bus_caps.width; 1918 break; 1919 case fm10k_bus_speed_8000: 1920 /* 128b/130b encoding has less than 2% impact on throughput */ 1921 expected_gts = 8 * hw->bus_caps.width; 1922 break; 1923 default: 1924 dev_warn(&interface->pdev->dev, 1925 "Unable to determine expected PCI Express bandwidth.\n"); 1926 return; 1927 } 1928 1929 if (max_gts >= expected_gts) 1930 return; 1931 1932 dev_warn(&interface->pdev->dev, 1933 "This device requires %dGT/s of bandwidth for optimal performance.\n", 1934 expected_gts); 1935 dev_warn(&interface->pdev->dev, 1936 "A %sslot with x%d lanes is suggested.\n", 1937 (hw->bus_caps.speed == fm10k_bus_speed_2500 ? "2.5GT/s " : 1938 hw->bus_caps.speed == fm10k_bus_speed_5000 ? "5.0GT/s " : 1939 hw->bus_caps.speed == fm10k_bus_speed_8000 ? "8.0GT/s " : ""), 1940 hw->bus_caps.width); 1941 } 1942 1943 /** 1944 * fm10k_probe - Device Initialization Routine 1945 * @pdev: PCI device information struct 1946 * @ent: entry in fm10k_pci_tbl 1947 * 1948 * Returns 0 on success, negative on failure 1949 * 1950 * fm10k_probe initializes an interface identified by a pci_dev structure. 1951 * The OS initialization, configuring of the interface private structure, 1952 * and a hardware reset occur. 1953 **/ 1954 static int fm10k_probe(struct pci_dev *pdev, const struct pci_device_id *ent) 1955 { 1956 struct net_device *netdev; 1957 struct fm10k_intfc *interface; 1958 int err; 1959 1960 if (pdev->error_state != pci_channel_io_normal) { 1961 dev_err(&pdev->dev, 1962 "PCI device still in an error state. Unable to load...\n"); 1963 return -EIO; 1964 } 1965 1966 err = pci_enable_device_mem(pdev); 1967 if (err) { 1968 dev_err(&pdev->dev, 1969 "PCI enable device failed: %d\n", err); 1970 return err; 1971 } 1972 1973 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(48)); 1974 if (err) 1975 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32)); 1976 if (err) { 1977 dev_err(&pdev->dev, 1978 "DMA configuration failed: %d\n", err); 1979 goto err_dma; 1980 } 1981 1982 err = pci_request_mem_regions(pdev, fm10k_driver_name); 1983 if (err) { 1984 dev_err(&pdev->dev, 1985 "pci_request_selected_regions failed: %d\n", err); 1986 goto err_pci_reg; 1987 } 1988 1989 pci_enable_pcie_error_reporting(pdev); 1990 1991 pci_set_master(pdev); 1992 pci_save_state(pdev); 1993 1994 netdev = fm10k_alloc_netdev(fm10k_info_tbl[ent->driver_data]); 1995 if (!netdev) { 1996 err = -ENOMEM; 1997 goto err_alloc_netdev; 1998 } 1999 2000 SET_NETDEV_DEV(netdev, &pdev->dev); 2001 2002 interface = netdev_priv(netdev); 2003 pci_set_drvdata(pdev, interface); 2004 2005 interface->netdev = netdev; 2006 interface->pdev = pdev; 2007 2008 interface->uc_addr = ioremap(pci_resource_start(pdev, 0), 2009 FM10K_UC_ADDR_SIZE); 2010 if (!interface->uc_addr) { 2011 err = -EIO; 2012 goto err_ioremap; 2013 } 2014 2015 err = fm10k_sw_init(interface, ent); 2016 if (err) 2017 goto err_sw_init; 2018 2019 /* enable debugfs support */ 2020 fm10k_dbg_intfc_init(interface); 2021 2022 err = fm10k_init_queueing_scheme(interface); 2023 if (err) 2024 goto err_sw_init; 2025 2026 /* the mbx interrupt might attempt to schedule the service task, so we 2027 * must ensure it is disabled since we haven't yet requested the timer 2028 * or work item. 2029 */ 2030 set_bit(__FM10K_SERVICE_DISABLE, &interface->state); 2031 2032 err = fm10k_mbx_request_irq(interface); 2033 if (err) 2034 goto err_mbx_interrupt; 2035 2036 /* final check of hardware state before registering the interface */ 2037 err = fm10k_hw_ready(interface); 2038 if (err) 2039 goto err_register; 2040 2041 err = register_netdev(netdev); 2042 if (err) 2043 goto err_register; 2044 2045 /* carrier off reporting is important to ethtool even BEFORE open */ 2046 netif_carrier_off(netdev); 2047 2048 /* stop all the transmit queues from transmitting until link is up */ 2049 netif_tx_stop_all_queues(netdev); 2050 2051 /* Initialize service timer and service task late in order to avoid 2052 * cleanup issues. 2053 */ 2054 setup_timer(&interface->service_timer, &fm10k_service_timer, 2055 (unsigned long)interface); 2056 INIT_WORK(&interface->service_task, fm10k_service_task); 2057 2058 /* kick off service timer now, even when interface is down */ 2059 mod_timer(&interface->service_timer, (HZ * 2) + jiffies); 2060 2061 /* print warning for non-optimal configurations */ 2062 fm10k_slot_warn(interface); 2063 2064 /* report MAC address for logging */ 2065 dev_info(&pdev->dev, "%pM\n", netdev->dev_addr); 2066 2067 /* enable SR-IOV after registering netdev to enforce PF/VF ordering */ 2068 fm10k_iov_configure(pdev, 0); 2069 2070 /* clear the service task disable bit to allow service task to start */ 2071 clear_bit(__FM10K_SERVICE_DISABLE, &interface->state); 2072 2073 return 0; 2074 2075 err_register: 2076 fm10k_mbx_free_irq(interface); 2077 err_mbx_interrupt: 2078 fm10k_clear_queueing_scheme(interface); 2079 err_sw_init: 2080 if (interface->sw_addr) 2081 iounmap(interface->sw_addr); 2082 iounmap(interface->uc_addr); 2083 err_ioremap: 2084 free_netdev(netdev); 2085 err_alloc_netdev: 2086 pci_release_mem_regions(pdev); 2087 err_pci_reg: 2088 err_dma: 2089 pci_disable_device(pdev); 2090 return err; 2091 } 2092 2093 /** 2094 * fm10k_remove - Device Removal Routine 2095 * @pdev: PCI device information struct 2096 * 2097 * fm10k_remove is called by the PCI subsystem to alert the driver 2098 * that it should release a PCI device. The could be caused by a 2099 * Hot-Plug event, or because the driver is going to be removed from 2100 * memory. 2101 **/ 2102 static void fm10k_remove(struct pci_dev *pdev) 2103 { 2104 struct fm10k_intfc *interface = pci_get_drvdata(pdev); 2105 struct net_device *netdev = interface->netdev; 2106 2107 del_timer_sync(&interface->service_timer); 2108 2109 set_bit(__FM10K_SERVICE_DISABLE, &interface->state); 2110 cancel_work_sync(&interface->service_task); 2111 2112 /* free netdev, this may bounce the interrupts due to setup_tc */ 2113 if (netdev->reg_state == NETREG_REGISTERED) 2114 unregister_netdev(netdev); 2115 2116 /* release VFs */ 2117 fm10k_iov_disable(pdev); 2118 2119 /* disable mailbox interrupt */ 2120 fm10k_mbx_free_irq(interface); 2121 2122 /* free interrupts */ 2123 fm10k_clear_queueing_scheme(interface); 2124 2125 /* remove any debugfs interfaces */ 2126 fm10k_dbg_intfc_exit(interface); 2127 2128 if (interface->sw_addr) 2129 iounmap(interface->sw_addr); 2130 iounmap(interface->uc_addr); 2131 2132 free_netdev(netdev); 2133 2134 pci_release_mem_regions(pdev); 2135 2136 pci_disable_pcie_error_reporting(pdev); 2137 2138 pci_disable_device(pdev); 2139 } 2140 2141 static void fm10k_prepare_suspend(struct fm10k_intfc *interface) 2142 { 2143 /* the watchdog task reads from registers, which might appear like 2144 * a surprise remove if the PCIe device is disabled while we're 2145 * stopped. We stop the watchdog task until after we resume software 2146 * activity. 2147 */ 2148 set_bit(__FM10K_SERVICE_DISABLE, &interface->state); 2149 cancel_work_sync(&interface->service_task); 2150 2151 fm10k_prepare_for_reset(interface); 2152 } 2153 2154 static int fm10k_handle_resume(struct fm10k_intfc *interface) 2155 { 2156 struct fm10k_hw *hw = &interface->hw; 2157 int err; 2158 2159 /* reset statistics starting values */ 2160 hw->mac.ops.rebind_hw_stats(hw, &interface->stats); 2161 2162 err = fm10k_handle_reset(interface); 2163 if (err) 2164 return err; 2165 2166 /* assume host is not ready, to prevent race with watchdog in case we 2167 * actually don't have connection to the switch 2168 */ 2169 interface->host_ready = false; 2170 fm10k_watchdog_host_not_ready(interface); 2171 2172 /* force link to stay down for a second to prevent link flutter */ 2173 interface->link_down_event = jiffies + (HZ); 2174 set_bit(__FM10K_LINK_DOWN, &interface->state); 2175 2176 /* clear the service task disable bit to allow service task to start */ 2177 clear_bit(__FM10K_SERVICE_DISABLE, &interface->state); 2178 fm10k_service_event_schedule(interface); 2179 2180 return err; 2181 } 2182 2183 #ifdef CONFIG_PM 2184 /** 2185 * fm10k_resume - Restore device to pre-sleep state 2186 * @pdev: PCI device information struct 2187 * 2188 * fm10k_resume is called after the system has powered back up from a sleep 2189 * state and is ready to resume operation. This function is meant to restore 2190 * the device back to its pre-sleep state. 2191 **/ 2192 static int fm10k_resume(struct pci_dev *pdev) 2193 { 2194 struct fm10k_intfc *interface = pci_get_drvdata(pdev); 2195 struct net_device *netdev = interface->netdev; 2196 struct fm10k_hw *hw = &interface->hw; 2197 u32 err; 2198 2199 pci_set_power_state(pdev, PCI_D0); 2200 pci_restore_state(pdev); 2201 2202 /* pci_restore_state clears dev->state_saved so call 2203 * pci_save_state to restore it. 2204 */ 2205 pci_save_state(pdev); 2206 2207 err = pci_enable_device_mem(pdev); 2208 if (err) { 2209 dev_err(&pdev->dev, "Cannot enable PCI device from suspend\n"); 2210 return err; 2211 } 2212 pci_set_master(pdev); 2213 2214 pci_wake_from_d3(pdev, false); 2215 2216 /* refresh hw_addr in case it was dropped */ 2217 hw->hw_addr = interface->uc_addr; 2218 2219 err = fm10k_handle_resume(interface); 2220 if (err) 2221 return err; 2222 2223 netif_device_attach(netdev); 2224 2225 return 0; 2226 } 2227 2228 /** 2229 * fm10k_suspend - Prepare the device for a system sleep state 2230 * @pdev: PCI device information struct 2231 * 2232 * fm10k_suspend is meant to shutdown the device prior to the system entering 2233 * a sleep state. The fm10k hardware does not support wake on lan so the 2234 * driver simply needs to shut down the device so it is in a low power state. 2235 **/ 2236 static int fm10k_suspend(struct pci_dev *pdev, 2237 pm_message_t __always_unused state) 2238 { 2239 struct fm10k_intfc *interface = pci_get_drvdata(pdev); 2240 struct net_device *netdev = interface->netdev; 2241 int err = 0; 2242 2243 netif_device_detach(netdev); 2244 2245 fm10k_prepare_suspend(interface); 2246 2247 err = pci_save_state(pdev); 2248 if (err) 2249 return err; 2250 2251 pci_disable_device(pdev); 2252 pci_wake_from_d3(pdev, false); 2253 pci_set_power_state(pdev, PCI_D3hot); 2254 2255 return 0; 2256 } 2257 2258 #endif /* CONFIG_PM */ 2259 /** 2260 * fm10k_io_error_detected - called when PCI error is detected 2261 * @pdev: Pointer to PCI device 2262 * @state: The current pci connection state 2263 * 2264 * This function is called after a PCI bus error affecting 2265 * this device has been detected. 2266 */ 2267 static pci_ers_result_t fm10k_io_error_detected(struct pci_dev *pdev, 2268 pci_channel_state_t state) 2269 { 2270 struct fm10k_intfc *interface = pci_get_drvdata(pdev); 2271 struct net_device *netdev = interface->netdev; 2272 2273 netif_device_detach(netdev); 2274 2275 if (state == pci_channel_io_perm_failure) 2276 return PCI_ERS_RESULT_DISCONNECT; 2277 2278 fm10k_prepare_suspend(interface); 2279 2280 /* Request a slot reset. */ 2281 return PCI_ERS_RESULT_NEED_RESET; 2282 } 2283 2284 /** 2285 * fm10k_io_slot_reset - called after the pci bus has been reset. 2286 * @pdev: Pointer to PCI device 2287 * 2288 * Restart the card from scratch, as if from a cold-boot. 2289 */ 2290 static pci_ers_result_t fm10k_io_slot_reset(struct pci_dev *pdev) 2291 { 2292 pci_ers_result_t result; 2293 2294 if (pci_reenable_device(pdev)) { 2295 dev_err(&pdev->dev, 2296 "Cannot re-enable PCI device after reset.\n"); 2297 result = PCI_ERS_RESULT_DISCONNECT; 2298 } else { 2299 pci_set_master(pdev); 2300 pci_restore_state(pdev); 2301 2302 /* After second error pci->state_saved is false, this 2303 * resets it so EEH doesn't break. 2304 */ 2305 pci_save_state(pdev); 2306 2307 pci_wake_from_d3(pdev, false); 2308 2309 result = PCI_ERS_RESULT_RECOVERED; 2310 } 2311 2312 pci_cleanup_aer_uncorrect_error_status(pdev); 2313 2314 return result; 2315 } 2316 2317 /** 2318 * fm10k_io_resume - called when traffic can start flowing again. 2319 * @pdev: Pointer to PCI device 2320 * 2321 * This callback is called when the error recovery driver tells us that 2322 * its OK to resume normal operation. 2323 */ 2324 static void fm10k_io_resume(struct pci_dev *pdev) 2325 { 2326 struct fm10k_intfc *interface = pci_get_drvdata(pdev); 2327 struct net_device *netdev = interface->netdev; 2328 int err; 2329 2330 err = fm10k_handle_resume(interface); 2331 2332 if (err) 2333 dev_warn(&pdev->dev, 2334 "fm10k_io_resume failed: %d\n", err); 2335 else 2336 netif_device_attach(netdev); 2337 } 2338 2339 /** 2340 * fm10k_io_reset_notify - called when PCI function is reset 2341 * @pdev: Pointer to PCI device 2342 * 2343 * This callback is called when the PCI function is reset such as from 2344 * /sys/class/net/<enpX>/device/reset or similar. When prepare is true, it 2345 * means we should prepare for a function reset. If prepare is false, it means 2346 * the function reset just occurred. 2347 */ 2348 static void fm10k_io_reset_notify(struct pci_dev *pdev, bool prepare) 2349 { 2350 struct fm10k_intfc *interface = pci_get_drvdata(pdev); 2351 int err = 0; 2352 2353 if (prepare) { 2354 /* warn incase we have any active VF devices */ 2355 if (pci_num_vf(pdev)) 2356 dev_warn(&pdev->dev, 2357 "PCIe FLR may cause issues for any active VF devices\n"); 2358 2359 fm10k_prepare_suspend(interface); 2360 } else { 2361 err = fm10k_handle_resume(interface); 2362 } 2363 2364 if (err) { 2365 dev_warn(&pdev->dev, 2366 "fm10k_io_reset_notify failed: %d\n", err); 2367 netif_device_detach(interface->netdev); 2368 } 2369 } 2370 2371 static const struct pci_error_handlers fm10k_err_handler = { 2372 .error_detected = fm10k_io_error_detected, 2373 .slot_reset = fm10k_io_slot_reset, 2374 .resume = fm10k_io_resume, 2375 .reset_notify = fm10k_io_reset_notify, 2376 }; 2377 2378 static struct pci_driver fm10k_driver = { 2379 .name = fm10k_driver_name, 2380 .id_table = fm10k_pci_tbl, 2381 .probe = fm10k_probe, 2382 .remove = fm10k_remove, 2383 #ifdef CONFIG_PM 2384 .suspend = fm10k_suspend, 2385 .resume = fm10k_resume, 2386 #endif 2387 .sriov_configure = fm10k_iov_configure, 2388 .err_handler = &fm10k_err_handler 2389 }; 2390 2391 /** 2392 * fm10k_register_pci_driver - register driver interface 2393 * 2394 * This function is called on module load in order to register the driver. 2395 **/ 2396 int fm10k_register_pci_driver(void) 2397 { 2398 return pci_register_driver(&fm10k_driver); 2399 } 2400 2401 /** 2402 * fm10k_unregister_pci_driver - unregister driver interface 2403 * 2404 * This function is called on module unload in order to remove the driver. 2405 **/ 2406 void fm10k_unregister_pci_driver(void) 2407 { 2408 pci_unregister_driver(&fm10k_driver); 2409 } 2410