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/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 void fm10k_service_event_schedule(struct fm10k_intfc *interface) 95 { 96 if (!test_bit(__FM10K_SERVICE_DISABLE, &interface->state) && 97 !test_and_set_bit(__FM10K_SERVICE_SCHED, &interface->state)) 98 queue_work(fm10k_workqueue, &interface->service_task); 99 } 100 101 static void fm10k_service_event_complete(struct fm10k_intfc *interface) 102 { 103 WARN_ON(!test_bit(__FM10K_SERVICE_SCHED, &interface->state)); 104 105 /* flush memory to make sure state is correct before next watchog */ 106 smp_mb__before_atomic(); 107 clear_bit(__FM10K_SERVICE_SCHED, &interface->state); 108 } 109 110 /** 111 * fm10k_service_timer - Timer Call-back 112 * @data: pointer to interface cast into an unsigned long 113 **/ 114 static void fm10k_service_timer(unsigned long data) 115 { 116 struct fm10k_intfc *interface = (struct fm10k_intfc *)data; 117 118 /* Reset the timer */ 119 mod_timer(&interface->service_timer, (HZ * 2) + jiffies); 120 121 fm10k_service_event_schedule(interface); 122 } 123 124 static void fm10k_detach_subtask(struct fm10k_intfc *interface) 125 { 126 struct net_device *netdev = interface->netdev; 127 u32 __iomem *hw_addr; 128 u32 value; 129 130 /* do nothing if device is still present or hw_addr is set */ 131 if (netif_device_present(netdev) || interface->hw.hw_addr) 132 return; 133 134 /* check the real address space to see if we've recovered */ 135 hw_addr = READ_ONCE(interface->uc_addr); 136 value = readl(hw_addr); 137 if (~value) { 138 interface->hw.hw_addr = interface->uc_addr; 139 netif_device_attach(netdev); 140 interface->flags |= FM10K_FLAG_RESET_REQUESTED; 141 netdev_warn(netdev, "PCIe link restored, device now attached\n"); 142 return; 143 } 144 145 rtnl_lock(); 146 147 if (netif_running(netdev)) 148 dev_close(netdev); 149 150 rtnl_unlock(); 151 } 152 153 static void fm10k_prepare_for_reset(struct fm10k_intfc *interface) 154 { 155 struct net_device *netdev = interface->netdev; 156 157 WARN_ON(in_interrupt()); 158 159 /* put off any impending NetWatchDogTimeout */ 160 netif_trans_update(netdev); 161 162 while (test_and_set_bit(__FM10K_RESETTING, &interface->state)) 163 usleep_range(1000, 2000); 164 165 rtnl_lock(); 166 167 fm10k_iov_suspend(interface->pdev); 168 169 if (netif_running(netdev)) 170 fm10k_close(netdev); 171 172 fm10k_mbx_free_irq(interface); 173 174 /* free interrupts */ 175 fm10k_clear_queueing_scheme(interface); 176 177 /* delay any future reset requests */ 178 interface->last_reset = jiffies + (10 * HZ); 179 180 rtnl_unlock(); 181 } 182 183 static int fm10k_handle_reset(struct fm10k_intfc *interface) 184 { 185 struct net_device *netdev = interface->netdev; 186 struct fm10k_hw *hw = &interface->hw; 187 int err; 188 189 rtnl_lock(); 190 191 pci_set_master(interface->pdev); 192 193 /* reset and initialize the hardware so it is in a known state */ 194 err = hw->mac.ops.reset_hw(hw); 195 if (err) { 196 dev_err(&interface->pdev->dev, "reset_hw failed: %d\n", err); 197 goto reinit_err; 198 } 199 200 err = hw->mac.ops.init_hw(hw); 201 if (err) { 202 dev_err(&interface->pdev->dev, "init_hw failed: %d\n", err); 203 goto reinit_err; 204 } 205 206 err = fm10k_init_queueing_scheme(interface); 207 if (err) { 208 dev_err(&interface->pdev->dev, 209 "init_queueing_scheme failed: %d\n", err); 210 goto reinit_err; 211 } 212 213 /* re-associate interrupts */ 214 err = fm10k_mbx_request_irq(interface); 215 if (err) 216 goto err_mbx_irq; 217 218 err = fm10k_hw_ready(interface); 219 if (err) 220 goto err_open; 221 222 /* update hardware address for VFs if perm_addr has changed */ 223 if (hw->mac.type == fm10k_mac_vf) { 224 if (is_valid_ether_addr(hw->mac.perm_addr)) { 225 ether_addr_copy(hw->mac.addr, hw->mac.perm_addr); 226 ether_addr_copy(netdev->perm_addr, hw->mac.perm_addr); 227 ether_addr_copy(netdev->dev_addr, hw->mac.perm_addr); 228 netdev->addr_assign_type &= ~NET_ADDR_RANDOM; 229 } 230 231 if (hw->mac.vlan_override) 232 netdev->features &= ~NETIF_F_HW_VLAN_CTAG_RX; 233 else 234 netdev->features |= NETIF_F_HW_VLAN_CTAG_RX; 235 } 236 237 err = netif_running(netdev) ? fm10k_open(netdev) : 0; 238 if (err) 239 goto err_open; 240 241 fm10k_iov_resume(interface->pdev); 242 243 rtnl_unlock(); 244 245 clear_bit(__FM10K_RESETTING, &interface->state); 246 247 return err; 248 err_open: 249 fm10k_mbx_free_irq(interface); 250 err_mbx_irq: 251 fm10k_clear_queueing_scheme(interface); 252 reinit_err: 253 netif_device_detach(netdev); 254 255 rtnl_unlock(); 256 257 clear_bit(__FM10K_RESETTING, &interface->state); 258 259 return err; 260 } 261 262 static void fm10k_reinit(struct fm10k_intfc *interface) 263 { 264 int err; 265 266 fm10k_prepare_for_reset(interface); 267 268 err = fm10k_handle_reset(interface); 269 if (err) 270 dev_err(&interface->pdev->dev, 271 "fm10k_handle_reset failed: %d\n", err); 272 } 273 274 static void fm10k_reset_subtask(struct fm10k_intfc *interface) 275 { 276 if (!(interface->flags & FM10K_FLAG_RESET_REQUESTED)) 277 return; 278 279 interface->flags &= ~FM10K_FLAG_RESET_REQUESTED; 280 281 netdev_err(interface->netdev, "Reset interface\n"); 282 283 fm10k_reinit(interface); 284 } 285 286 /** 287 * fm10k_configure_swpri_map - Configure Receive SWPRI to PC mapping 288 * @interface: board private structure 289 * 290 * Configure the SWPRI to PC mapping for the port. 291 **/ 292 static void fm10k_configure_swpri_map(struct fm10k_intfc *interface) 293 { 294 struct net_device *netdev = interface->netdev; 295 struct fm10k_hw *hw = &interface->hw; 296 int i; 297 298 /* clear flag indicating update is needed */ 299 interface->flags &= ~FM10K_FLAG_SWPRI_CONFIG; 300 301 /* these registers are only available on the PF */ 302 if (hw->mac.type != fm10k_mac_pf) 303 return; 304 305 /* configure SWPRI to PC map */ 306 for (i = 0; i < FM10K_SWPRI_MAX; i++) 307 fm10k_write_reg(hw, FM10K_SWPRI_MAP(i), 308 netdev_get_prio_tc_map(netdev, i)); 309 } 310 311 /** 312 * fm10k_watchdog_update_host_state - Update the link status based on host. 313 * @interface: board private structure 314 **/ 315 static void fm10k_watchdog_update_host_state(struct fm10k_intfc *interface) 316 { 317 struct fm10k_hw *hw = &interface->hw; 318 s32 err; 319 320 if (test_bit(__FM10K_LINK_DOWN, &interface->state)) { 321 interface->host_ready = false; 322 if (time_is_after_jiffies(interface->link_down_event)) 323 return; 324 clear_bit(__FM10K_LINK_DOWN, &interface->state); 325 } 326 327 if (interface->flags & FM10K_FLAG_SWPRI_CONFIG) { 328 if (rtnl_trylock()) { 329 fm10k_configure_swpri_map(interface); 330 rtnl_unlock(); 331 } 332 } 333 334 /* lock the mailbox for transmit and receive */ 335 fm10k_mbx_lock(interface); 336 337 err = hw->mac.ops.get_host_state(hw, &interface->host_ready); 338 if (err && time_is_before_jiffies(interface->last_reset)) 339 interface->flags |= FM10K_FLAG_RESET_REQUESTED; 340 341 /* free the lock */ 342 fm10k_mbx_unlock(interface); 343 } 344 345 /** 346 * fm10k_mbx_subtask - Process upstream and downstream mailboxes 347 * @interface: board private structure 348 * 349 * This function will process both the upstream and downstream mailboxes. 350 **/ 351 static void fm10k_mbx_subtask(struct fm10k_intfc *interface) 352 { 353 /* process upstream mailbox and update device state */ 354 fm10k_watchdog_update_host_state(interface); 355 356 /* process downstream mailboxes */ 357 fm10k_iov_mbx(interface); 358 } 359 360 /** 361 * fm10k_watchdog_host_is_ready - Update netdev status based on host ready 362 * @interface: board private structure 363 **/ 364 static void fm10k_watchdog_host_is_ready(struct fm10k_intfc *interface) 365 { 366 struct net_device *netdev = interface->netdev; 367 368 /* only continue if link state is currently down */ 369 if (netif_carrier_ok(netdev)) 370 return; 371 372 netif_info(interface, drv, netdev, "NIC Link is up\n"); 373 374 netif_carrier_on(netdev); 375 netif_tx_wake_all_queues(netdev); 376 } 377 378 /** 379 * fm10k_watchdog_host_not_ready - Update netdev status based on host not ready 380 * @interface: board private structure 381 **/ 382 static void fm10k_watchdog_host_not_ready(struct fm10k_intfc *interface) 383 { 384 struct net_device *netdev = interface->netdev; 385 386 /* only continue if link state is currently up */ 387 if (!netif_carrier_ok(netdev)) 388 return; 389 390 netif_info(interface, drv, netdev, "NIC Link is down\n"); 391 392 netif_carrier_off(netdev); 393 netif_tx_stop_all_queues(netdev); 394 } 395 396 /** 397 * fm10k_update_stats - Update the board statistics counters. 398 * @interface: board private structure 399 **/ 400 void fm10k_update_stats(struct fm10k_intfc *interface) 401 { 402 struct net_device_stats *net_stats = &interface->netdev->stats; 403 struct fm10k_hw *hw = &interface->hw; 404 u64 hw_csum_tx_good = 0, hw_csum_rx_good = 0, rx_length_errors = 0; 405 u64 rx_switch_errors = 0, rx_drops = 0, rx_pp_errors = 0; 406 u64 rx_link_errors = 0; 407 u64 rx_errors = 0, rx_csum_errors = 0, tx_csum_errors = 0; 408 u64 restart_queue = 0, tx_busy = 0, alloc_failed = 0; 409 u64 rx_bytes_nic = 0, rx_pkts_nic = 0, rx_drops_nic = 0; 410 u64 tx_bytes_nic = 0, tx_pkts_nic = 0; 411 u64 bytes, pkts; 412 int i; 413 414 /* ensure only one thread updates stats at a time */ 415 if (test_and_set_bit(__FM10K_UPDATING_STATS, &interface->state)) 416 return; 417 418 /* do not allow stats update via service task for next second */ 419 interface->next_stats_update = jiffies + HZ; 420 421 /* gather some stats to the interface struct that are per queue */ 422 for (bytes = 0, pkts = 0, i = 0; i < interface->num_tx_queues; i++) { 423 struct fm10k_ring *tx_ring = READ_ONCE(interface->tx_ring[i]); 424 425 if (!tx_ring) 426 continue; 427 428 restart_queue += tx_ring->tx_stats.restart_queue; 429 tx_busy += tx_ring->tx_stats.tx_busy; 430 tx_csum_errors += tx_ring->tx_stats.csum_err; 431 bytes += tx_ring->stats.bytes; 432 pkts += tx_ring->stats.packets; 433 hw_csum_tx_good += tx_ring->tx_stats.csum_good; 434 } 435 436 interface->restart_queue = restart_queue; 437 interface->tx_busy = tx_busy; 438 net_stats->tx_bytes = bytes; 439 net_stats->tx_packets = pkts; 440 interface->tx_csum_errors = tx_csum_errors; 441 interface->hw_csum_tx_good = hw_csum_tx_good; 442 443 /* gather some stats to the interface struct that are per queue */ 444 for (bytes = 0, pkts = 0, i = 0; i < interface->num_rx_queues; i++) { 445 struct fm10k_ring *rx_ring = READ_ONCE(interface->rx_ring[i]); 446 447 if (!rx_ring) 448 continue; 449 450 bytes += rx_ring->stats.bytes; 451 pkts += rx_ring->stats.packets; 452 alloc_failed += rx_ring->rx_stats.alloc_failed; 453 rx_csum_errors += rx_ring->rx_stats.csum_err; 454 rx_errors += rx_ring->rx_stats.errors; 455 hw_csum_rx_good += rx_ring->rx_stats.csum_good; 456 rx_switch_errors += rx_ring->rx_stats.switch_errors; 457 rx_drops += rx_ring->rx_stats.drops; 458 rx_pp_errors += rx_ring->rx_stats.pp_errors; 459 rx_link_errors += rx_ring->rx_stats.link_errors; 460 rx_length_errors += rx_ring->rx_stats.length_errors; 461 } 462 463 net_stats->rx_bytes = bytes; 464 net_stats->rx_packets = pkts; 465 interface->alloc_failed = alloc_failed; 466 interface->rx_csum_errors = rx_csum_errors; 467 interface->hw_csum_rx_good = hw_csum_rx_good; 468 interface->rx_switch_errors = rx_switch_errors; 469 interface->rx_drops = rx_drops; 470 interface->rx_pp_errors = rx_pp_errors; 471 interface->rx_link_errors = rx_link_errors; 472 interface->rx_length_errors = rx_length_errors; 473 474 hw->mac.ops.update_hw_stats(hw, &interface->stats); 475 476 for (i = 0; i < hw->mac.max_queues; i++) { 477 struct fm10k_hw_stats_q *q = &interface->stats.q[i]; 478 479 tx_bytes_nic += q->tx_bytes.count; 480 tx_pkts_nic += q->tx_packets.count; 481 rx_bytes_nic += q->rx_bytes.count; 482 rx_pkts_nic += q->rx_packets.count; 483 rx_drops_nic += q->rx_drops.count; 484 } 485 486 interface->tx_bytes_nic = tx_bytes_nic; 487 interface->tx_packets_nic = tx_pkts_nic; 488 interface->rx_bytes_nic = rx_bytes_nic; 489 interface->rx_packets_nic = rx_pkts_nic; 490 interface->rx_drops_nic = rx_drops_nic; 491 492 /* Fill out the OS statistics structure */ 493 net_stats->rx_errors = rx_errors; 494 net_stats->rx_dropped = interface->stats.nodesc_drop.count; 495 496 clear_bit(__FM10K_UPDATING_STATS, &interface->state); 497 } 498 499 /** 500 * fm10k_watchdog_flush_tx - flush queues on host not ready 501 * @interface - pointer to the device interface structure 502 **/ 503 static void fm10k_watchdog_flush_tx(struct fm10k_intfc *interface) 504 { 505 int some_tx_pending = 0; 506 int i; 507 508 /* nothing to do if carrier is up */ 509 if (netif_carrier_ok(interface->netdev)) 510 return; 511 512 for (i = 0; i < interface->num_tx_queues; i++) { 513 struct fm10k_ring *tx_ring = interface->tx_ring[i]; 514 515 if (tx_ring->next_to_use != tx_ring->next_to_clean) { 516 some_tx_pending = 1; 517 break; 518 } 519 } 520 521 /* We've lost link, so the controller stops DMA, but we've got 522 * queued Tx work that's never going to get done, so reset 523 * controller to flush Tx. 524 */ 525 if (some_tx_pending) 526 interface->flags |= FM10K_FLAG_RESET_REQUESTED; 527 } 528 529 /** 530 * fm10k_watchdog_subtask - check and bring link up 531 * @interface - pointer to the device interface structure 532 **/ 533 static void fm10k_watchdog_subtask(struct fm10k_intfc *interface) 534 { 535 /* if interface is down do nothing */ 536 if (test_bit(__FM10K_DOWN, &interface->state) || 537 test_bit(__FM10K_RESETTING, &interface->state)) 538 return; 539 540 if (interface->host_ready) 541 fm10k_watchdog_host_is_ready(interface); 542 else 543 fm10k_watchdog_host_not_ready(interface); 544 545 /* update stats only once every second */ 546 if (time_is_before_jiffies(interface->next_stats_update)) 547 fm10k_update_stats(interface); 548 549 /* flush any uncompleted work */ 550 fm10k_watchdog_flush_tx(interface); 551 } 552 553 /** 554 * fm10k_check_hang_subtask - check for hung queues and dropped interrupts 555 * @interface - pointer to the device interface structure 556 * 557 * This function serves two purposes. First it strobes the interrupt lines 558 * in order to make certain interrupts are occurring. Secondly it sets the 559 * bits needed to check for TX hangs. As a result we should immediately 560 * determine if a hang has occurred. 561 */ 562 static void fm10k_check_hang_subtask(struct fm10k_intfc *interface) 563 { 564 int i; 565 566 /* If we're down or resetting, just bail */ 567 if (test_bit(__FM10K_DOWN, &interface->state) || 568 test_bit(__FM10K_RESETTING, &interface->state)) 569 return; 570 571 /* rate limit tx hang checks to only once every 2 seconds */ 572 if (time_is_after_eq_jiffies(interface->next_tx_hang_check)) 573 return; 574 interface->next_tx_hang_check = jiffies + (2 * HZ); 575 576 if (netif_carrier_ok(interface->netdev)) { 577 /* Force detection of hung controller */ 578 for (i = 0; i < interface->num_tx_queues; i++) 579 set_check_for_tx_hang(interface->tx_ring[i]); 580 581 /* Rearm all in-use q_vectors for immediate firing */ 582 for (i = 0; i < interface->num_q_vectors; i++) { 583 struct fm10k_q_vector *qv = interface->q_vector[i]; 584 585 if (!qv->tx.count && !qv->rx.count) 586 continue; 587 writel(FM10K_ITR_ENABLE | FM10K_ITR_PENDING2, qv->itr); 588 } 589 } 590 } 591 592 /** 593 * fm10k_service_task - manages and runs subtasks 594 * @work: pointer to work_struct containing our data 595 **/ 596 static void fm10k_service_task(struct work_struct *work) 597 { 598 struct fm10k_intfc *interface; 599 600 interface = container_of(work, struct fm10k_intfc, service_task); 601 602 /* tasks run even when interface is down */ 603 fm10k_mbx_subtask(interface); 604 fm10k_detach_subtask(interface); 605 fm10k_reset_subtask(interface); 606 607 /* tasks only run when interface is up */ 608 fm10k_watchdog_subtask(interface); 609 fm10k_check_hang_subtask(interface); 610 611 /* release lock on service events to allow scheduling next event */ 612 fm10k_service_event_complete(interface); 613 } 614 615 /** 616 * fm10k_configure_tx_ring - Configure Tx ring after Reset 617 * @interface: board private structure 618 * @ring: structure containing ring specific data 619 * 620 * Configure the Tx descriptor ring after a reset. 621 **/ 622 static void fm10k_configure_tx_ring(struct fm10k_intfc *interface, 623 struct fm10k_ring *ring) 624 { 625 struct fm10k_hw *hw = &interface->hw; 626 u64 tdba = ring->dma; 627 u32 size = ring->count * sizeof(struct fm10k_tx_desc); 628 u32 txint = FM10K_INT_MAP_DISABLE; 629 u32 txdctl = BIT(FM10K_TXDCTL_MAX_TIME_SHIFT) | FM10K_TXDCTL_ENABLE; 630 u8 reg_idx = ring->reg_idx; 631 632 /* disable queue to avoid issues while updating state */ 633 fm10k_write_reg(hw, FM10K_TXDCTL(reg_idx), 0); 634 fm10k_write_flush(hw); 635 636 /* possible poll here to verify ring resources have been cleaned */ 637 638 /* set location and size for descriptor ring */ 639 fm10k_write_reg(hw, FM10K_TDBAL(reg_idx), tdba & DMA_BIT_MASK(32)); 640 fm10k_write_reg(hw, FM10K_TDBAH(reg_idx), tdba >> 32); 641 fm10k_write_reg(hw, FM10K_TDLEN(reg_idx), size); 642 643 /* reset head and tail pointers */ 644 fm10k_write_reg(hw, FM10K_TDH(reg_idx), 0); 645 fm10k_write_reg(hw, FM10K_TDT(reg_idx), 0); 646 647 /* store tail pointer */ 648 ring->tail = &interface->uc_addr[FM10K_TDT(reg_idx)]; 649 650 /* reset ntu and ntc to place SW in sync with hardware */ 651 ring->next_to_clean = 0; 652 ring->next_to_use = 0; 653 654 /* Map interrupt */ 655 if (ring->q_vector) { 656 txint = ring->q_vector->v_idx + NON_Q_VECTORS(hw); 657 txint |= FM10K_INT_MAP_TIMER0; 658 } 659 660 fm10k_write_reg(hw, FM10K_TXINT(reg_idx), txint); 661 662 /* enable use of FTAG bit in Tx descriptor, register is RO for VF */ 663 fm10k_write_reg(hw, FM10K_PFVTCTL(reg_idx), 664 FM10K_PFVTCTL_FTAG_DESC_ENABLE); 665 666 /* Initialize XPS */ 667 if (!test_and_set_bit(__FM10K_TX_XPS_INIT_DONE, &ring->state) && 668 ring->q_vector) 669 netif_set_xps_queue(ring->netdev, 670 &ring->q_vector->affinity_mask, 671 ring->queue_index); 672 673 /* enable queue */ 674 fm10k_write_reg(hw, FM10K_TXDCTL(reg_idx), txdctl); 675 } 676 677 /** 678 * fm10k_enable_tx_ring - Verify Tx ring is enabled after configuration 679 * @interface: board private structure 680 * @ring: structure containing ring specific data 681 * 682 * Verify the Tx descriptor ring is ready for transmit. 683 **/ 684 static void fm10k_enable_tx_ring(struct fm10k_intfc *interface, 685 struct fm10k_ring *ring) 686 { 687 struct fm10k_hw *hw = &interface->hw; 688 int wait_loop = 10; 689 u32 txdctl; 690 u8 reg_idx = ring->reg_idx; 691 692 /* if we are already enabled just exit */ 693 if (fm10k_read_reg(hw, FM10K_TXDCTL(reg_idx)) & FM10K_TXDCTL_ENABLE) 694 return; 695 696 /* poll to verify queue is enabled */ 697 do { 698 usleep_range(1000, 2000); 699 txdctl = fm10k_read_reg(hw, FM10K_TXDCTL(reg_idx)); 700 } while (!(txdctl & FM10K_TXDCTL_ENABLE) && --wait_loop); 701 if (!wait_loop) 702 netif_err(interface, drv, interface->netdev, 703 "Could not enable Tx Queue %d\n", reg_idx); 704 } 705 706 /** 707 * fm10k_configure_tx - Configure Transmit Unit after Reset 708 * @interface: board private structure 709 * 710 * Configure the Tx unit of the MAC after a reset. 711 **/ 712 static void fm10k_configure_tx(struct fm10k_intfc *interface) 713 { 714 int i; 715 716 /* Setup the HW Tx Head and Tail descriptor pointers */ 717 for (i = 0; i < interface->num_tx_queues; i++) 718 fm10k_configure_tx_ring(interface, interface->tx_ring[i]); 719 720 /* poll here to verify that Tx rings are now enabled */ 721 for (i = 0; i < interface->num_tx_queues; i++) 722 fm10k_enable_tx_ring(interface, interface->tx_ring[i]); 723 } 724 725 /** 726 * fm10k_configure_rx_ring - Configure Rx ring after Reset 727 * @interface: board private structure 728 * @ring: structure containing ring specific data 729 * 730 * Configure the Rx descriptor ring after a reset. 731 **/ 732 static void fm10k_configure_rx_ring(struct fm10k_intfc *interface, 733 struct fm10k_ring *ring) 734 { 735 u64 rdba = ring->dma; 736 struct fm10k_hw *hw = &interface->hw; 737 u32 size = ring->count * sizeof(union fm10k_rx_desc); 738 u32 rxqctl, rxdctl = FM10K_RXDCTL_WRITE_BACK_MIN_DELAY; 739 u32 srrctl = FM10K_SRRCTL_BUFFER_CHAINING_EN; 740 u32 rxint = FM10K_INT_MAP_DISABLE; 741 u8 rx_pause = interface->rx_pause; 742 u8 reg_idx = ring->reg_idx; 743 744 /* disable queue to avoid issues while updating state */ 745 rxqctl = fm10k_read_reg(hw, FM10K_RXQCTL(reg_idx)); 746 rxqctl &= ~FM10K_RXQCTL_ENABLE; 747 fm10k_write_flush(hw); 748 749 /* possible poll here to verify ring resources have been cleaned */ 750 751 /* set location and size for descriptor ring */ 752 fm10k_write_reg(hw, FM10K_RDBAL(reg_idx), rdba & DMA_BIT_MASK(32)); 753 fm10k_write_reg(hw, FM10K_RDBAH(reg_idx), rdba >> 32); 754 fm10k_write_reg(hw, FM10K_RDLEN(reg_idx), size); 755 756 /* reset head and tail pointers */ 757 fm10k_write_reg(hw, FM10K_RDH(reg_idx), 0); 758 fm10k_write_reg(hw, FM10K_RDT(reg_idx), 0); 759 760 /* store tail pointer */ 761 ring->tail = &interface->uc_addr[FM10K_RDT(reg_idx)]; 762 763 /* reset ntu and ntc to place SW in sync with hardware */ 764 ring->next_to_clean = 0; 765 ring->next_to_use = 0; 766 ring->next_to_alloc = 0; 767 768 /* Configure the Rx buffer size for one buff without split */ 769 srrctl |= FM10K_RX_BUFSZ >> FM10K_SRRCTL_BSIZEPKT_SHIFT; 770 771 /* Configure the Rx ring to suppress loopback packets */ 772 srrctl |= FM10K_SRRCTL_LOOPBACK_SUPPRESS; 773 fm10k_write_reg(hw, FM10K_SRRCTL(reg_idx), srrctl); 774 775 /* Enable drop on empty */ 776 #ifdef CONFIG_DCB 777 if (interface->pfc_en) 778 rx_pause = interface->pfc_en; 779 #endif 780 if (!(rx_pause & BIT(ring->qos_pc))) 781 rxdctl |= FM10K_RXDCTL_DROP_ON_EMPTY; 782 783 fm10k_write_reg(hw, FM10K_RXDCTL(reg_idx), rxdctl); 784 785 /* assign default VLAN to queue */ 786 ring->vid = hw->mac.default_vid; 787 788 /* if we have an active VLAN, disable default VLAN ID */ 789 if (test_bit(hw->mac.default_vid, interface->active_vlans)) 790 ring->vid |= FM10K_VLAN_CLEAR; 791 792 /* Map interrupt */ 793 if (ring->q_vector) { 794 rxint = ring->q_vector->v_idx + NON_Q_VECTORS(hw); 795 rxint |= FM10K_INT_MAP_TIMER1; 796 } 797 798 fm10k_write_reg(hw, FM10K_RXINT(reg_idx), rxint); 799 800 /* enable queue */ 801 rxqctl = fm10k_read_reg(hw, FM10K_RXQCTL(reg_idx)); 802 rxqctl |= FM10K_RXQCTL_ENABLE; 803 fm10k_write_reg(hw, FM10K_RXQCTL(reg_idx), rxqctl); 804 805 /* place buffers on ring for receive data */ 806 fm10k_alloc_rx_buffers(ring, fm10k_desc_unused(ring)); 807 } 808 809 /** 810 * fm10k_update_rx_drop_en - Configures the drop enable bits for Rx rings 811 * @interface: board private structure 812 * 813 * Configure the drop enable bits for the Rx rings. 814 **/ 815 void fm10k_update_rx_drop_en(struct fm10k_intfc *interface) 816 { 817 struct fm10k_hw *hw = &interface->hw; 818 u8 rx_pause = interface->rx_pause; 819 int i; 820 821 #ifdef CONFIG_DCB 822 if (interface->pfc_en) 823 rx_pause = interface->pfc_en; 824 825 #endif 826 for (i = 0; i < interface->num_rx_queues; i++) { 827 struct fm10k_ring *ring = interface->rx_ring[i]; 828 u32 rxdctl = FM10K_RXDCTL_WRITE_BACK_MIN_DELAY; 829 u8 reg_idx = ring->reg_idx; 830 831 if (!(rx_pause & BIT(ring->qos_pc))) 832 rxdctl |= FM10K_RXDCTL_DROP_ON_EMPTY; 833 834 fm10k_write_reg(hw, FM10K_RXDCTL(reg_idx), rxdctl); 835 } 836 } 837 838 /** 839 * fm10k_configure_dglort - Configure Receive DGLORT after reset 840 * @interface: board private structure 841 * 842 * Configure the DGLORT description and RSS tables. 843 **/ 844 static void fm10k_configure_dglort(struct fm10k_intfc *interface) 845 { 846 struct fm10k_dglort_cfg dglort = { 0 }; 847 struct fm10k_hw *hw = &interface->hw; 848 int i; 849 u32 mrqc; 850 851 /* Fill out hash function seeds */ 852 for (i = 0; i < FM10K_RSSRK_SIZE; i++) 853 fm10k_write_reg(hw, FM10K_RSSRK(0, i), interface->rssrk[i]); 854 855 /* Write RETA table to hardware */ 856 for (i = 0; i < FM10K_RETA_SIZE; i++) 857 fm10k_write_reg(hw, FM10K_RETA(0, i), interface->reta[i]); 858 859 /* Generate RSS hash based on packet types, TCP/UDP 860 * port numbers and/or IPv4/v6 src and dst addresses 861 */ 862 mrqc = FM10K_MRQC_IPV4 | 863 FM10K_MRQC_TCP_IPV4 | 864 FM10K_MRQC_IPV6 | 865 FM10K_MRQC_TCP_IPV6; 866 867 if (interface->flags & FM10K_FLAG_RSS_FIELD_IPV4_UDP) 868 mrqc |= FM10K_MRQC_UDP_IPV4; 869 if (interface->flags & FM10K_FLAG_RSS_FIELD_IPV6_UDP) 870 mrqc |= FM10K_MRQC_UDP_IPV6; 871 872 fm10k_write_reg(hw, FM10K_MRQC(0), mrqc); 873 874 /* configure default DGLORT mapping for RSS/DCB */ 875 dglort.inner_rss = 1; 876 dglort.rss_l = fls(interface->ring_feature[RING_F_RSS].mask); 877 dglort.pc_l = fls(interface->ring_feature[RING_F_QOS].mask); 878 hw->mac.ops.configure_dglort_map(hw, &dglort); 879 880 /* assign GLORT per queue for queue mapped testing */ 881 if (interface->glort_count > 64) { 882 memset(&dglort, 0, sizeof(dglort)); 883 dglort.inner_rss = 1; 884 dglort.glort = interface->glort + 64; 885 dglort.idx = fm10k_dglort_pf_queue; 886 dglort.queue_l = fls(interface->num_rx_queues - 1); 887 hw->mac.ops.configure_dglort_map(hw, &dglort); 888 } 889 890 /* assign glort value for RSS/DCB specific to this interface */ 891 memset(&dglort, 0, sizeof(dglort)); 892 dglort.inner_rss = 1; 893 dglort.glort = interface->glort; 894 dglort.rss_l = fls(interface->ring_feature[RING_F_RSS].mask); 895 dglort.pc_l = fls(interface->ring_feature[RING_F_QOS].mask); 896 /* configure DGLORT mapping for RSS/DCB */ 897 dglort.idx = fm10k_dglort_pf_rss; 898 if (interface->l2_accel) 899 dglort.shared_l = fls(interface->l2_accel->size); 900 hw->mac.ops.configure_dglort_map(hw, &dglort); 901 } 902 903 /** 904 * fm10k_configure_rx - Configure Receive Unit after Reset 905 * @interface: board private structure 906 * 907 * Configure the Rx unit of the MAC after a reset. 908 **/ 909 static void fm10k_configure_rx(struct fm10k_intfc *interface) 910 { 911 int i; 912 913 /* Configure SWPRI to PC map */ 914 fm10k_configure_swpri_map(interface); 915 916 /* Configure RSS and DGLORT map */ 917 fm10k_configure_dglort(interface); 918 919 /* Setup the HW Rx Head and Tail descriptor pointers */ 920 for (i = 0; i < interface->num_rx_queues; i++) 921 fm10k_configure_rx_ring(interface, interface->rx_ring[i]); 922 923 /* possible poll here to verify that Rx rings are now enabled */ 924 } 925 926 static void fm10k_napi_enable_all(struct fm10k_intfc *interface) 927 { 928 struct fm10k_q_vector *q_vector; 929 int q_idx; 930 931 for (q_idx = 0; q_idx < interface->num_q_vectors; q_idx++) { 932 q_vector = interface->q_vector[q_idx]; 933 napi_enable(&q_vector->napi); 934 } 935 } 936 937 static irqreturn_t fm10k_msix_clean_rings(int __always_unused irq, void *data) 938 { 939 struct fm10k_q_vector *q_vector = data; 940 941 if (q_vector->rx.count || q_vector->tx.count) 942 napi_schedule_irqoff(&q_vector->napi); 943 944 return IRQ_HANDLED; 945 } 946 947 static irqreturn_t fm10k_msix_mbx_vf(int __always_unused irq, void *data) 948 { 949 struct fm10k_intfc *interface = data; 950 struct fm10k_hw *hw = &interface->hw; 951 struct fm10k_mbx_info *mbx = &hw->mbx; 952 953 /* re-enable mailbox interrupt and indicate 20us delay */ 954 fm10k_write_reg(hw, FM10K_VFITR(FM10K_MBX_VECTOR), 955 (FM10K_MBX_INT_DELAY >> hw->mac.itr_scale) | 956 FM10K_ITR_ENABLE); 957 958 /* service upstream mailbox */ 959 if (fm10k_mbx_trylock(interface)) { 960 mbx->ops.process(hw, mbx); 961 fm10k_mbx_unlock(interface); 962 } 963 964 hw->mac.get_host_state = true; 965 fm10k_service_event_schedule(interface); 966 967 return IRQ_HANDLED; 968 } 969 970 #ifdef CONFIG_NET_POLL_CONTROLLER 971 /** 972 * fm10k_netpoll - A Polling 'interrupt' handler 973 * @netdev: network interface device structure 974 * 975 * This is used by netconsole to send skbs without having to re-enable 976 * interrupts. It's not called while the normal interrupt routine is executing. 977 **/ 978 void fm10k_netpoll(struct net_device *netdev) 979 { 980 struct fm10k_intfc *interface = netdev_priv(netdev); 981 int i; 982 983 /* if interface is down do nothing */ 984 if (test_bit(__FM10K_DOWN, &interface->state)) 985 return; 986 987 for (i = 0; i < interface->num_q_vectors; i++) 988 fm10k_msix_clean_rings(0, interface->q_vector[i]); 989 } 990 991 #endif 992 #define FM10K_ERR_MSG(type) case (type): error = #type; break 993 static void fm10k_handle_fault(struct fm10k_intfc *interface, int type, 994 struct fm10k_fault *fault) 995 { 996 struct pci_dev *pdev = interface->pdev; 997 struct fm10k_hw *hw = &interface->hw; 998 struct fm10k_iov_data *iov_data = interface->iov_data; 999 char *error; 1000 1001 switch (type) { 1002 case FM10K_PCA_FAULT: 1003 switch (fault->type) { 1004 default: 1005 error = "Unknown PCA error"; 1006 break; 1007 FM10K_ERR_MSG(PCA_NO_FAULT); 1008 FM10K_ERR_MSG(PCA_UNMAPPED_ADDR); 1009 FM10K_ERR_MSG(PCA_BAD_QACCESS_PF); 1010 FM10K_ERR_MSG(PCA_BAD_QACCESS_VF); 1011 FM10K_ERR_MSG(PCA_MALICIOUS_REQ); 1012 FM10K_ERR_MSG(PCA_POISONED_TLP); 1013 FM10K_ERR_MSG(PCA_TLP_ABORT); 1014 } 1015 break; 1016 case FM10K_THI_FAULT: 1017 switch (fault->type) { 1018 default: 1019 error = "Unknown THI error"; 1020 break; 1021 FM10K_ERR_MSG(THI_NO_FAULT); 1022 FM10K_ERR_MSG(THI_MAL_DIS_Q_FAULT); 1023 } 1024 break; 1025 case FM10K_FUM_FAULT: 1026 switch (fault->type) { 1027 default: 1028 error = "Unknown FUM error"; 1029 break; 1030 FM10K_ERR_MSG(FUM_NO_FAULT); 1031 FM10K_ERR_MSG(FUM_UNMAPPED_ADDR); 1032 FM10K_ERR_MSG(FUM_BAD_VF_QACCESS); 1033 FM10K_ERR_MSG(FUM_ADD_DECODE_ERR); 1034 FM10K_ERR_MSG(FUM_RO_ERROR); 1035 FM10K_ERR_MSG(FUM_QPRC_CRC_ERROR); 1036 FM10K_ERR_MSG(FUM_CSR_TIMEOUT); 1037 FM10K_ERR_MSG(FUM_INVALID_TYPE); 1038 FM10K_ERR_MSG(FUM_INVALID_LENGTH); 1039 FM10K_ERR_MSG(FUM_INVALID_BE); 1040 FM10K_ERR_MSG(FUM_INVALID_ALIGN); 1041 } 1042 break; 1043 default: 1044 error = "Undocumented fault"; 1045 break; 1046 } 1047 1048 dev_warn(&pdev->dev, 1049 "%s Address: 0x%llx SpecInfo: 0x%x Func: %02x.%0x\n", 1050 error, fault->address, fault->specinfo, 1051 PCI_SLOT(fault->func), PCI_FUNC(fault->func)); 1052 1053 /* For VF faults, clear out the respective LPORT, reset the queue 1054 * resources, and then reconnect to the mailbox. This allows the 1055 * VF in question to resume behavior. For transient faults that are 1056 * the result of non-malicious behavior this will log the fault and 1057 * allow the VF to resume functionality. Obviously for malicious VFs 1058 * they will be able to attempt malicious behavior again. In this 1059 * case, the system administrator will need to step in and manually 1060 * remove or disable the VF in question. 1061 */ 1062 if (fault->func && iov_data) { 1063 int vf = fault->func - 1; 1064 struct fm10k_vf_info *vf_info = &iov_data->vf_info[vf]; 1065 1066 hw->iov.ops.reset_lport(hw, vf_info); 1067 hw->iov.ops.reset_resources(hw, vf_info); 1068 1069 /* reset_lport disables the VF, so re-enable it */ 1070 hw->iov.ops.set_lport(hw, vf_info, vf, 1071 FM10K_VF_FLAG_MULTI_CAPABLE); 1072 1073 /* reset_resources will disconnect from the mbx */ 1074 vf_info->mbx.ops.connect(hw, &vf_info->mbx); 1075 } 1076 } 1077 1078 static void fm10k_report_fault(struct fm10k_intfc *interface, u32 eicr) 1079 { 1080 struct fm10k_hw *hw = &interface->hw; 1081 struct fm10k_fault fault = { 0 }; 1082 int type, err; 1083 1084 for (eicr &= FM10K_EICR_FAULT_MASK, type = FM10K_PCA_FAULT; 1085 eicr; 1086 eicr >>= 1, type += FM10K_FAULT_SIZE) { 1087 /* only check if there is an error reported */ 1088 if (!(eicr & 0x1)) 1089 continue; 1090 1091 /* retrieve fault info */ 1092 err = hw->mac.ops.get_fault(hw, type, &fault); 1093 if (err) { 1094 dev_err(&interface->pdev->dev, 1095 "error reading fault\n"); 1096 continue; 1097 } 1098 1099 fm10k_handle_fault(interface, type, &fault); 1100 } 1101 } 1102 1103 static void fm10k_reset_drop_on_empty(struct fm10k_intfc *interface, u32 eicr) 1104 { 1105 struct fm10k_hw *hw = &interface->hw; 1106 const u32 rxdctl = FM10K_RXDCTL_WRITE_BACK_MIN_DELAY; 1107 u32 maxholdq; 1108 int q; 1109 1110 if (!(eicr & FM10K_EICR_MAXHOLDTIME)) 1111 return; 1112 1113 maxholdq = fm10k_read_reg(hw, FM10K_MAXHOLDQ(7)); 1114 if (maxholdq) 1115 fm10k_write_reg(hw, FM10K_MAXHOLDQ(7), maxholdq); 1116 for (q = 255;;) { 1117 if (maxholdq & BIT(31)) { 1118 if (q < FM10K_MAX_QUEUES_PF) { 1119 interface->rx_overrun_pf++; 1120 fm10k_write_reg(hw, FM10K_RXDCTL(q), rxdctl); 1121 } else { 1122 interface->rx_overrun_vf++; 1123 } 1124 } 1125 1126 maxholdq *= 2; 1127 if (!maxholdq) 1128 q &= ~(32 - 1); 1129 1130 if (!q) 1131 break; 1132 1133 if (q-- % 32) 1134 continue; 1135 1136 maxholdq = fm10k_read_reg(hw, FM10K_MAXHOLDQ(q / 32)); 1137 if (maxholdq) 1138 fm10k_write_reg(hw, FM10K_MAXHOLDQ(q / 32), maxholdq); 1139 } 1140 } 1141 1142 static irqreturn_t fm10k_msix_mbx_pf(int __always_unused irq, void *data) 1143 { 1144 struct fm10k_intfc *interface = data; 1145 struct fm10k_hw *hw = &interface->hw; 1146 struct fm10k_mbx_info *mbx = &hw->mbx; 1147 u32 eicr; 1148 s32 err = 0; 1149 1150 /* unmask any set bits related to this interrupt */ 1151 eicr = fm10k_read_reg(hw, FM10K_EICR); 1152 fm10k_write_reg(hw, FM10K_EICR, eicr & (FM10K_EICR_MAILBOX | 1153 FM10K_EICR_SWITCHREADY | 1154 FM10K_EICR_SWITCHNOTREADY)); 1155 1156 /* report any faults found to the message log */ 1157 fm10k_report_fault(interface, eicr); 1158 1159 /* reset any queues disabled due to receiver overrun */ 1160 fm10k_reset_drop_on_empty(interface, eicr); 1161 1162 /* service mailboxes */ 1163 if (fm10k_mbx_trylock(interface)) { 1164 err = mbx->ops.process(hw, mbx); 1165 /* handle VFLRE events */ 1166 fm10k_iov_event(interface); 1167 fm10k_mbx_unlock(interface); 1168 } 1169 1170 if (err == FM10K_ERR_RESET_REQUESTED) 1171 interface->flags |= FM10K_FLAG_RESET_REQUESTED; 1172 1173 /* if switch toggled state we should reset GLORTs */ 1174 if (eicr & FM10K_EICR_SWITCHNOTREADY) { 1175 /* force link down for at least 4 seconds */ 1176 interface->link_down_event = jiffies + (4 * HZ); 1177 set_bit(__FM10K_LINK_DOWN, &interface->state); 1178 1179 /* reset dglort_map back to no config */ 1180 hw->mac.dglort_map = FM10K_DGLORTMAP_NONE; 1181 } 1182 1183 /* we should validate host state after interrupt event */ 1184 hw->mac.get_host_state = true; 1185 1186 /* validate host state, and handle VF mailboxes in the service task */ 1187 fm10k_service_event_schedule(interface); 1188 1189 /* re-enable mailbox interrupt and indicate 20us delay */ 1190 fm10k_write_reg(hw, FM10K_ITR(FM10K_MBX_VECTOR), 1191 (FM10K_MBX_INT_DELAY >> hw->mac.itr_scale) | 1192 FM10K_ITR_ENABLE); 1193 1194 return IRQ_HANDLED; 1195 } 1196 1197 void fm10k_mbx_free_irq(struct fm10k_intfc *interface) 1198 { 1199 struct fm10k_hw *hw = &interface->hw; 1200 struct msix_entry *entry; 1201 int itr_reg; 1202 1203 /* no mailbox IRQ to free if MSI-X is not enabled */ 1204 if (!interface->msix_entries) 1205 return; 1206 1207 entry = &interface->msix_entries[FM10K_MBX_VECTOR]; 1208 1209 /* disconnect the mailbox */ 1210 hw->mbx.ops.disconnect(hw, &hw->mbx); 1211 1212 /* disable Mailbox cause */ 1213 if (hw->mac.type == fm10k_mac_pf) { 1214 fm10k_write_reg(hw, FM10K_EIMR, 1215 FM10K_EIMR_DISABLE(PCA_FAULT) | 1216 FM10K_EIMR_DISABLE(FUM_FAULT) | 1217 FM10K_EIMR_DISABLE(MAILBOX) | 1218 FM10K_EIMR_DISABLE(SWITCHREADY) | 1219 FM10K_EIMR_DISABLE(SWITCHNOTREADY) | 1220 FM10K_EIMR_DISABLE(SRAMERROR) | 1221 FM10K_EIMR_DISABLE(VFLR) | 1222 FM10K_EIMR_DISABLE(MAXHOLDTIME)); 1223 itr_reg = FM10K_ITR(FM10K_MBX_VECTOR); 1224 } else { 1225 itr_reg = FM10K_VFITR(FM10K_MBX_VECTOR); 1226 } 1227 1228 fm10k_write_reg(hw, itr_reg, FM10K_ITR_MASK_SET); 1229 1230 free_irq(entry->vector, interface); 1231 } 1232 1233 static s32 fm10k_mbx_mac_addr(struct fm10k_hw *hw, u32 **results, 1234 struct fm10k_mbx_info *mbx) 1235 { 1236 bool vlan_override = hw->mac.vlan_override; 1237 u16 default_vid = hw->mac.default_vid; 1238 struct fm10k_intfc *interface; 1239 s32 err; 1240 1241 err = fm10k_msg_mac_vlan_vf(hw, results, mbx); 1242 if (err) 1243 return err; 1244 1245 interface = container_of(hw, struct fm10k_intfc, hw); 1246 1247 /* MAC was changed so we need reset */ 1248 if (is_valid_ether_addr(hw->mac.perm_addr) && 1249 !ether_addr_equal(hw->mac.perm_addr, hw->mac.addr)) 1250 interface->flags |= FM10K_FLAG_RESET_REQUESTED; 1251 1252 /* VLAN override was changed, or default VLAN changed */ 1253 if ((vlan_override != hw->mac.vlan_override) || 1254 (default_vid != hw->mac.default_vid)) 1255 interface->flags |= FM10K_FLAG_RESET_REQUESTED; 1256 1257 return 0; 1258 } 1259 1260 /* generic error handler for mailbox issues */ 1261 static s32 fm10k_mbx_error(struct fm10k_hw *hw, u32 **results, 1262 struct fm10k_mbx_info __always_unused *mbx) 1263 { 1264 struct fm10k_intfc *interface; 1265 struct pci_dev *pdev; 1266 1267 interface = container_of(hw, struct fm10k_intfc, hw); 1268 pdev = interface->pdev; 1269 1270 dev_err(&pdev->dev, "Unknown message ID %u\n", 1271 **results & FM10K_TLV_ID_MASK); 1272 1273 return 0; 1274 } 1275 1276 static const struct fm10k_msg_data vf_mbx_data[] = { 1277 FM10K_TLV_MSG_TEST_HANDLER(fm10k_tlv_msg_test), 1278 FM10K_VF_MSG_MAC_VLAN_HANDLER(fm10k_mbx_mac_addr), 1279 FM10K_VF_MSG_LPORT_STATE_HANDLER(fm10k_msg_lport_state_vf), 1280 FM10K_TLV_MSG_ERROR_HANDLER(fm10k_mbx_error), 1281 }; 1282 1283 static int fm10k_mbx_request_irq_vf(struct fm10k_intfc *interface) 1284 { 1285 struct msix_entry *entry = &interface->msix_entries[FM10K_MBX_VECTOR]; 1286 struct net_device *dev = interface->netdev; 1287 struct fm10k_hw *hw = &interface->hw; 1288 int err; 1289 1290 /* Use timer0 for interrupt moderation on the mailbox */ 1291 u32 itr = entry->entry | FM10K_INT_MAP_TIMER0; 1292 1293 /* register mailbox handlers */ 1294 err = hw->mbx.ops.register_handlers(&hw->mbx, vf_mbx_data); 1295 if (err) 1296 return err; 1297 1298 /* request the IRQ */ 1299 err = request_irq(entry->vector, fm10k_msix_mbx_vf, 0, 1300 dev->name, interface); 1301 if (err) { 1302 netif_err(interface, probe, dev, 1303 "request_irq for msix_mbx failed: %d\n", err); 1304 return err; 1305 } 1306 1307 /* map all of the interrupt sources */ 1308 fm10k_write_reg(hw, FM10K_VFINT_MAP, itr); 1309 1310 /* enable interrupt */ 1311 fm10k_write_reg(hw, FM10K_VFITR(entry->entry), FM10K_ITR_ENABLE); 1312 1313 return 0; 1314 } 1315 1316 static s32 fm10k_lport_map(struct fm10k_hw *hw, u32 **results, 1317 struct fm10k_mbx_info *mbx) 1318 { 1319 struct fm10k_intfc *interface; 1320 u32 dglort_map = hw->mac.dglort_map; 1321 s32 err; 1322 1323 interface = container_of(hw, struct fm10k_intfc, hw); 1324 1325 err = fm10k_msg_err_pf(hw, results, mbx); 1326 if (!err && hw->swapi.status) { 1327 /* force link down for a reasonable delay */ 1328 interface->link_down_event = jiffies + (2 * HZ); 1329 set_bit(__FM10K_LINK_DOWN, &interface->state); 1330 1331 /* reset dglort_map back to no config */ 1332 hw->mac.dglort_map = FM10K_DGLORTMAP_NONE; 1333 1334 fm10k_service_event_schedule(interface); 1335 1336 /* prevent overloading kernel message buffer */ 1337 if (interface->lport_map_failed) 1338 return 0; 1339 1340 interface->lport_map_failed = true; 1341 1342 if (hw->swapi.status == FM10K_MSG_ERR_PEP_NOT_SCHEDULED) 1343 dev_warn(&interface->pdev->dev, 1344 "cannot obtain link because the host interface is configured for a PCIe host interface bandwidth of zero\n"); 1345 dev_warn(&interface->pdev->dev, 1346 "request logical port map failed: %d\n", 1347 hw->swapi.status); 1348 1349 return 0; 1350 } 1351 1352 err = fm10k_msg_lport_map_pf(hw, results, mbx); 1353 if (err) 1354 return err; 1355 1356 interface->lport_map_failed = false; 1357 1358 /* we need to reset if port count was just updated */ 1359 if (dglort_map != hw->mac.dglort_map) 1360 interface->flags |= FM10K_FLAG_RESET_REQUESTED; 1361 1362 return 0; 1363 } 1364 1365 static s32 fm10k_update_pvid(struct fm10k_hw *hw, u32 **results, 1366 struct fm10k_mbx_info __always_unused *mbx) 1367 { 1368 struct fm10k_intfc *interface; 1369 u16 glort, pvid; 1370 u32 pvid_update; 1371 s32 err; 1372 1373 err = fm10k_tlv_attr_get_u32(results[FM10K_PF_ATTR_ID_UPDATE_PVID], 1374 &pvid_update); 1375 if (err) 1376 return err; 1377 1378 /* extract values from the pvid update */ 1379 glort = FM10K_MSG_HDR_FIELD_GET(pvid_update, UPDATE_PVID_GLORT); 1380 pvid = FM10K_MSG_HDR_FIELD_GET(pvid_update, UPDATE_PVID_PVID); 1381 1382 /* if glort is not valid return error */ 1383 if (!fm10k_glort_valid_pf(hw, glort)) 1384 return FM10K_ERR_PARAM; 1385 1386 /* verify VLAN ID is valid */ 1387 if (pvid >= FM10K_VLAN_TABLE_VID_MAX) 1388 return FM10K_ERR_PARAM; 1389 1390 interface = container_of(hw, struct fm10k_intfc, hw); 1391 1392 /* check to see if this belongs to one of the VFs */ 1393 err = fm10k_iov_update_pvid(interface, glort, pvid); 1394 if (!err) 1395 return 0; 1396 1397 /* we need to reset if default VLAN was just updated */ 1398 if (pvid != hw->mac.default_vid) 1399 interface->flags |= FM10K_FLAG_RESET_REQUESTED; 1400 1401 hw->mac.default_vid = pvid; 1402 1403 return 0; 1404 } 1405 1406 static const struct fm10k_msg_data pf_mbx_data[] = { 1407 FM10K_PF_MSG_ERR_HANDLER(XCAST_MODES, fm10k_msg_err_pf), 1408 FM10K_PF_MSG_ERR_HANDLER(UPDATE_MAC_FWD_RULE, fm10k_msg_err_pf), 1409 FM10K_PF_MSG_LPORT_MAP_HANDLER(fm10k_lport_map), 1410 FM10K_PF_MSG_ERR_HANDLER(LPORT_CREATE, fm10k_msg_err_pf), 1411 FM10K_PF_MSG_ERR_HANDLER(LPORT_DELETE, fm10k_msg_err_pf), 1412 FM10K_PF_MSG_UPDATE_PVID_HANDLER(fm10k_update_pvid), 1413 FM10K_TLV_MSG_ERROR_HANDLER(fm10k_mbx_error), 1414 }; 1415 1416 static int fm10k_mbx_request_irq_pf(struct fm10k_intfc *interface) 1417 { 1418 struct msix_entry *entry = &interface->msix_entries[FM10K_MBX_VECTOR]; 1419 struct net_device *dev = interface->netdev; 1420 struct fm10k_hw *hw = &interface->hw; 1421 int err; 1422 1423 /* Use timer0 for interrupt moderation on the mailbox */ 1424 u32 mbx_itr = entry->entry | FM10K_INT_MAP_TIMER0; 1425 u32 other_itr = entry->entry | FM10K_INT_MAP_IMMEDIATE; 1426 1427 /* register mailbox handlers */ 1428 err = hw->mbx.ops.register_handlers(&hw->mbx, pf_mbx_data); 1429 if (err) 1430 return err; 1431 1432 /* request the IRQ */ 1433 err = request_irq(entry->vector, fm10k_msix_mbx_pf, 0, 1434 dev->name, interface); 1435 if (err) { 1436 netif_err(interface, probe, dev, 1437 "request_irq for msix_mbx failed: %d\n", err); 1438 return err; 1439 } 1440 1441 /* Enable interrupts w/ no moderation for "other" interrupts */ 1442 fm10k_write_reg(hw, FM10K_INT_MAP(fm10k_int_pcie_fault), other_itr); 1443 fm10k_write_reg(hw, FM10K_INT_MAP(fm10k_int_switch_up_down), other_itr); 1444 fm10k_write_reg(hw, FM10K_INT_MAP(fm10k_int_sram), other_itr); 1445 fm10k_write_reg(hw, FM10K_INT_MAP(fm10k_int_max_hold_time), other_itr); 1446 fm10k_write_reg(hw, FM10K_INT_MAP(fm10k_int_vflr), other_itr); 1447 1448 /* Enable interrupts w/ moderation for mailbox */ 1449 fm10k_write_reg(hw, FM10K_INT_MAP(fm10k_int_mailbox), mbx_itr); 1450 1451 /* Enable individual interrupt causes */ 1452 fm10k_write_reg(hw, FM10K_EIMR, FM10K_EIMR_ENABLE(PCA_FAULT) | 1453 FM10K_EIMR_ENABLE(FUM_FAULT) | 1454 FM10K_EIMR_ENABLE(MAILBOX) | 1455 FM10K_EIMR_ENABLE(SWITCHREADY) | 1456 FM10K_EIMR_ENABLE(SWITCHNOTREADY) | 1457 FM10K_EIMR_ENABLE(SRAMERROR) | 1458 FM10K_EIMR_ENABLE(VFLR) | 1459 FM10K_EIMR_ENABLE(MAXHOLDTIME)); 1460 1461 /* enable interrupt */ 1462 fm10k_write_reg(hw, FM10K_ITR(entry->entry), FM10K_ITR_ENABLE); 1463 1464 return 0; 1465 } 1466 1467 int fm10k_mbx_request_irq(struct fm10k_intfc *interface) 1468 { 1469 struct fm10k_hw *hw = &interface->hw; 1470 int err; 1471 1472 /* enable Mailbox cause */ 1473 if (hw->mac.type == fm10k_mac_pf) 1474 err = fm10k_mbx_request_irq_pf(interface); 1475 else 1476 err = fm10k_mbx_request_irq_vf(interface); 1477 if (err) 1478 return err; 1479 1480 /* connect mailbox */ 1481 err = hw->mbx.ops.connect(hw, &hw->mbx); 1482 1483 /* if the mailbox failed to connect, then free IRQ */ 1484 if (err) 1485 fm10k_mbx_free_irq(interface); 1486 1487 return err; 1488 } 1489 1490 /** 1491 * fm10k_qv_free_irq - release interrupts associated with queue vectors 1492 * @interface: board private structure 1493 * 1494 * Release all interrupts associated with this interface 1495 **/ 1496 void fm10k_qv_free_irq(struct fm10k_intfc *interface) 1497 { 1498 int vector = interface->num_q_vectors; 1499 struct fm10k_hw *hw = &interface->hw; 1500 struct msix_entry *entry; 1501 1502 entry = &interface->msix_entries[NON_Q_VECTORS(hw) + vector]; 1503 1504 while (vector) { 1505 struct fm10k_q_vector *q_vector; 1506 1507 vector--; 1508 entry--; 1509 q_vector = interface->q_vector[vector]; 1510 1511 if (!q_vector->tx.count && !q_vector->rx.count) 1512 continue; 1513 1514 /* clear the affinity_mask in the IRQ descriptor */ 1515 irq_set_affinity_hint(entry->vector, NULL); 1516 1517 /* disable interrupts */ 1518 writel(FM10K_ITR_MASK_SET, q_vector->itr); 1519 1520 free_irq(entry->vector, q_vector); 1521 } 1522 } 1523 1524 /** 1525 * fm10k_qv_request_irq - initialize interrupts for queue vectors 1526 * @interface: board private structure 1527 * 1528 * Attempts to configure interrupts using the best available 1529 * capabilities of the hardware and kernel. 1530 **/ 1531 int fm10k_qv_request_irq(struct fm10k_intfc *interface) 1532 { 1533 struct net_device *dev = interface->netdev; 1534 struct fm10k_hw *hw = &interface->hw; 1535 struct msix_entry *entry; 1536 int ri = 0, ti = 0; 1537 int vector, err; 1538 1539 entry = &interface->msix_entries[NON_Q_VECTORS(hw)]; 1540 1541 for (vector = 0; vector < interface->num_q_vectors; vector++) { 1542 struct fm10k_q_vector *q_vector = interface->q_vector[vector]; 1543 1544 /* name the vector */ 1545 if (q_vector->tx.count && q_vector->rx.count) { 1546 snprintf(q_vector->name, sizeof(q_vector->name) - 1, 1547 "%s-TxRx-%d", dev->name, ri++); 1548 ti++; 1549 } else if (q_vector->rx.count) { 1550 snprintf(q_vector->name, sizeof(q_vector->name) - 1, 1551 "%s-rx-%d", dev->name, ri++); 1552 } else if (q_vector->tx.count) { 1553 snprintf(q_vector->name, sizeof(q_vector->name) - 1, 1554 "%s-tx-%d", dev->name, ti++); 1555 } else { 1556 /* skip this unused q_vector */ 1557 continue; 1558 } 1559 1560 /* Assign ITR register to q_vector */ 1561 q_vector->itr = (hw->mac.type == fm10k_mac_pf) ? 1562 &interface->uc_addr[FM10K_ITR(entry->entry)] : 1563 &interface->uc_addr[FM10K_VFITR(entry->entry)]; 1564 1565 /* request the IRQ */ 1566 err = request_irq(entry->vector, &fm10k_msix_clean_rings, 0, 1567 q_vector->name, q_vector); 1568 if (err) { 1569 netif_err(interface, probe, dev, 1570 "request_irq failed for MSIX interrupt Error: %d\n", 1571 err); 1572 goto err_out; 1573 } 1574 1575 /* assign the mask for this irq */ 1576 irq_set_affinity_hint(entry->vector, &q_vector->affinity_mask); 1577 1578 /* Enable q_vector */ 1579 writel(FM10K_ITR_ENABLE, q_vector->itr); 1580 1581 entry++; 1582 } 1583 1584 return 0; 1585 1586 err_out: 1587 /* wind through the ring freeing all entries and vectors */ 1588 while (vector) { 1589 struct fm10k_q_vector *q_vector; 1590 1591 entry--; 1592 vector--; 1593 q_vector = interface->q_vector[vector]; 1594 1595 if (!q_vector->tx.count && !q_vector->rx.count) 1596 continue; 1597 1598 /* clear the affinity_mask in the IRQ descriptor */ 1599 irq_set_affinity_hint(entry->vector, NULL); 1600 1601 /* disable interrupts */ 1602 writel(FM10K_ITR_MASK_SET, q_vector->itr); 1603 1604 free_irq(entry->vector, q_vector); 1605 } 1606 1607 return err; 1608 } 1609 1610 void fm10k_up(struct fm10k_intfc *interface) 1611 { 1612 struct fm10k_hw *hw = &interface->hw; 1613 1614 /* Enable Tx/Rx DMA */ 1615 hw->mac.ops.start_hw(hw); 1616 1617 /* configure Tx descriptor rings */ 1618 fm10k_configure_tx(interface); 1619 1620 /* configure Rx descriptor rings */ 1621 fm10k_configure_rx(interface); 1622 1623 /* configure interrupts */ 1624 hw->mac.ops.update_int_moderator(hw); 1625 1626 /* enable statistics capture again */ 1627 clear_bit(__FM10K_UPDATING_STATS, &interface->state); 1628 1629 /* clear down bit to indicate we are ready to go */ 1630 clear_bit(__FM10K_DOWN, &interface->state); 1631 1632 /* enable polling cleanups */ 1633 fm10k_napi_enable_all(interface); 1634 1635 /* re-establish Rx filters */ 1636 fm10k_restore_rx_state(interface); 1637 1638 /* enable transmits */ 1639 netif_tx_start_all_queues(interface->netdev); 1640 1641 /* kick off the service timer now */ 1642 hw->mac.get_host_state = true; 1643 mod_timer(&interface->service_timer, jiffies); 1644 } 1645 1646 static void fm10k_napi_disable_all(struct fm10k_intfc *interface) 1647 { 1648 struct fm10k_q_vector *q_vector; 1649 int q_idx; 1650 1651 for (q_idx = 0; q_idx < interface->num_q_vectors; q_idx++) { 1652 q_vector = interface->q_vector[q_idx]; 1653 napi_disable(&q_vector->napi); 1654 } 1655 } 1656 1657 void fm10k_down(struct fm10k_intfc *interface) 1658 { 1659 struct net_device *netdev = interface->netdev; 1660 struct fm10k_hw *hw = &interface->hw; 1661 int err, i = 0, count = 0; 1662 1663 /* signal that we are down to the interrupt handler and service task */ 1664 if (test_and_set_bit(__FM10K_DOWN, &interface->state)) 1665 return; 1666 1667 /* call carrier off first to avoid false dev_watchdog timeouts */ 1668 netif_carrier_off(netdev); 1669 1670 /* disable transmits */ 1671 netif_tx_stop_all_queues(netdev); 1672 netif_tx_disable(netdev); 1673 1674 /* reset Rx filters */ 1675 fm10k_reset_rx_state(interface); 1676 1677 /* disable polling routines */ 1678 fm10k_napi_disable_all(interface); 1679 1680 /* capture stats one last time before stopping interface */ 1681 fm10k_update_stats(interface); 1682 1683 /* prevent updating statistics while we're down */ 1684 while (test_and_set_bit(__FM10K_UPDATING_STATS, &interface->state)) 1685 usleep_range(1000, 2000); 1686 1687 /* skip waiting for TX DMA if we lost PCIe link */ 1688 if (FM10K_REMOVED(hw->hw_addr)) 1689 goto skip_tx_dma_drain; 1690 1691 /* In some rare circumstances it can take a while for Tx queues to 1692 * quiesce and be fully disabled. Attempt to .stop_hw() first, and 1693 * then if we get ERR_REQUESTS_PENDING, go ahead and wait in a loop 1694 * until the Tx queues have emptied, or until a number of retries. If 1695 * we fail to clear within the retry loop, we will issue a warning 1696 * indicating that Tx DMA is probably hung. Note this means we call 1697 * .stop_hw() twice but this shouldn't cause any problems. 1698 */ 1699 err = hw->mac.ops.stop_hw(hw); 1700 if (err != FM10K_ERR_REQUESTS_PENDING) 1701 goto skip_tx_dma_drain; 1702 1703 #define TX_DMA_DRAIN_RETRIES 25 1704 for (count = 0; count < TX_DMA_DRAIN_RETRIES; count++) { 1705 usleep_range(10000, 20000); 1706 1707 /* start checking at the last ring to have pending Tx */ 1708 for (; i < interface->num_tx_queues; i++) 1709 if (fm10k_get_tx_pending(interface->tx_ring[i], false)) 1710 break; 1711 1712 /* if all the queues are drained, we can break now */ 1713 if (i == interface->num_tx_queues) 1714 break; 1715 } 1716 1717 if (count >= TX_DMA_DRAIN_RETRIES) 1718 dev_err(&interface->pdev->dev, 1719 "Tx queues failed to drain after %d tries. Tx DMA is probably hung.\n", 1720 count); 1721 skip_tx_dma_drain: 1722 /* Disable DMA engine for Tx/Rx */ 1723 err = hw->mac.ops.stop_hw(hw); 1724 if (err == FM10K_ERR_REQUESTS_PENDING) 1725 dev_err(&interface->pdev->dev, 1726 "due to pending requests hw was not shut down gracefully\n"); 1727 else if (err) 1728 dev_err(&interface->pdev->dev, "stop_hw failed: %d\n", err); 1729 1730 /* free any buffers still on the rings */ 1731 fm10k_clean_all_tx_rings(interface); 1732 fm10k_clean_all_rx_rings(interface); 1733 } 1734 1735 /** 1736 * fm10k_sw_init - Initialize general software structures 1737 * @interface: host interface private structure to initialize 1738 * 1739 * fm10k_sw_init initializes the interface private data structure. 1740 * Fields are initialized based on PCI device information and 1741 * OS network device settings (MTU size). 1742 **/ 1743 static int fm10k_sw_init(struct fm10k_intfc *interface, 1744 const struct pci_device_id *ent) 1745 { 1746 const struct fm10k_info *fi = fm10k_info_tbl[ent->driver_data]; 1747 struct fm10k_hw *hw = &interface->hw; 1748 struct pci_dev *pdev = interface->pdev; 1749 struct net_device *netdev = interface->netdev; 1750 u32 rss_key[FM10K_RSSRK_SIZE]; 1751 unsigned int rss; 1752 int err; 1753 1754 /* initialize back pointer */ 1755 hw->back = interface; 1756 hw->hw_addr = interface->uc_addr; 1757 1758 /* PCI config space info */ 1759 hw->vendor_id = pdev->vendor; 1760 hw->device_id = pdev->device; 1761 hw->revision_id = pdev->revision; 1762 hw->subsystem_vendor_id = pdev->subsystem_vendor; 1763 hw->subsystem_device_id = pdev->subsystem_device; 1764 1765 /* Setup hw api */ 1766 memcpy(&hw->mac.ops, fi->mac_ops, sizeof(hw->mac.ops)); 1767 hw->mac.type = fi->mac; 1768 1769 /* Setup IOV handlers */ 1770 if (fi->iov_ops) 1771 memcpy(&hw->iov.ops, fi->iov_ops, sizeof(hw->iov.ops)); 1772 1773 /* Set common capability flags and settings */ 1774 rss = min_t(int, FM10K_MAX_RSS_INDICES, num_online_cpus()); 1775 interface->ring_feature[RING_F_RSS].limit = rss; 1776 fi->get_invariants(hw); 1777 1778 /* pick up the PCIe bus settings for reporting later */ 1779 if (hw->mac.ops.get_bus_info) 1780 hw->mac.ops.get_bus_info(hw); 1781 1782 /* limit the usable DMA range */ 1783 if (hw->mac.ops.set_dma_mask) 1784 hw->mac.ops.set_dma_mask(hw, dma_get_mask(&pdev->dev)); 1785 1786 /* update netdev with DMA restrictions */ 1787 if (dma_get_mask(&pdev->dev) > DMA_BIT_MASK(32)) { 1788 netdev->features |= NETIF_F_HIGHDMA; 1789 netdev->vlan_features |= NETIF_F_HIGHDMA; 1790 } 1791 1792 /* delay any future reset requests */ 1793 interface->last_reset = jiffies + (10 * HZ); 1794 1795 /* reset and initialize the hardware so it is in a known state */ 1796 err = hw->mac.ops.reset_hw(hw); 1797 if (err) { 1798 dev_err(&pdev->dev, "reset_hw failed: %d\n", err); 1799 return err; 1800 } 1801 1802 err = hw->mac.ops.init_hw(hw); 1803 if (err) { 1804 dev_err(&pdev->dev, "init_hw failed: %d\n", err); 1805 return err; 1806 } 1807 1808 /* initialize hardware statistics */ 1809 hw->mac.ops.update_hw_stats(hw, &interface->stats); 1810 1811 /* Set upper limit on IOV VFs that can be allocated */ 1812 pci_sriov_set_totalvfs(pdev, hw->iov.total_vfs); 1813 1814 /* Start with random Ethernet address */ 1815 eth_random_addr(hw->mac.addr); 1816 1817 /* Initialize MAC address from hardware */ 1818 err = hw->mac.ops.read_mac_addr(hw); 1819 if (err) { 1820 dev_warn(&pdev->dev, 1821 "Failed to obtain MAC address defaulting to random\n"); 1822 /* tag address assignment as random */ 1823 netdev->addr_assign_type |= NET_ADDR_RANDOM; 1824 } 1825 1826 ether_addr_copy(netdev->dev_addr, hw->mac.addr); 1827 ether_addr_copy(netdev->perm_addr, hw->mac.addr); 1828 1829 if (!is_valid_ether_addr(netdev->perm_addr)) { 1830 dev_err(&pdev->dev, "Invalid MAC Address\n"); 1831 return -EIO; 1832 } 1833 1834 /* initialize DCBNL interface */ 1835 fm10k_dcbnl_set_ops(netdev); 1836 1837 /* set default ring sizes */ 1838 interface->tx_ring_count = FM10K_DEFAULT_TXD; 1839 interface->rx_ring_count = FM10K_DEFAULT_RXD; 1840 1841 /* set default interrupt moderation */ 1842 interface->tx_itr = FM10K_TX_ITR_DEFAULT; 1843 interface->rx_itr = FM10K_ITR_ADAPTIVE | FM10K_RX_ITR_DEFAULT; 1844 1845 /* initialize udp port lists */ 1846 INIT_LIST_HEAD(&interface->vxlan_port); 1847 INIT_LIST_HEAD(&interface->geneve_port); 1848 1849 netdev_rss_key_fill(rss_key, sizeof(rss_key)); 1850 memcpy(interface->rssrk, rss_key, sizeof(rss_key)); 1851 1852 /* Start off interface as being down */ 1853 set_bit(__FM10K_DOWN, &interface->state); 1854 set_bit(__FM10K_UPDATING_STATS, &interface->state); 1855 1856 return 0; 1857 } 1858 1859 static void fm10k_slot_warn(struct fm10k_intfc *interface) 1860 { 1861 enum pcie_link_width width = PCIE_LNK_WIDTH_UNKNOWN; 1862 enum pci_bus_speed speed = PCI_SPEED_UNKNOWN; 1863 struct fm10k_hw *hw = &interface->hw; 1864 int max_gts = 0, expected_gts = 0; 1865 1866 if (pcie_get_minimum_link(interface->pdev, &speed, &width) || 1867 speed == PCI_SPEED_UNKNOWN || width == PCIE_LNK_WIDTH_UNKNOWN) { 1868 dev_warn(&interface->pdev->dev, 1869 "Unable to determine PCI Express bandwidth.\n"); 1870 return; 1871 } 1872 1873 switch (speed) { 1874 case PCIE_SPEED_2_5GT: 1875 /* 8b/10b encoding reduces max throughput by 20% */ 1876 max_gts = 2 * width; 1877 break; 1878 case PCIE_SPEED_5_0GT: 1879 /* 8b/10b encoding reduces max throughput by 20% */ 1880 max_gts = 4 * width; 1881 break; 1882 case PCIE_SPEED_8_0GT: 1883 /* 128b/130b encoding has less than 2% impact on throughput */ 1884 max_gts = 8 * width; 1885 break; 1886 default: 1887 dev_warn(&interface->pdev->dev, 1888 "Unable to determine PCI Express bandwidth.\n"); 1889 return; 1890 } 1891 1892 dev_info(&interface->pdev->dev, 1893 "PCI Express bandwidth of %dGT/s available\n", 1894 max_gts); 1895 dev_info(&interface->pdev->dev, 1896 "(Speed:%s, Width: x%d, Encoding Loss:%s, Payload:%s)\n", 1897 (speed == PCIE_SPEED_8_0GT ? "8.0GT/s" : 1898 speed == PCIE_SPEED_5_0GT ? "5.0GT/s" : 1899 speed == PCIE_SPEED_2_5GT ? "2.5GT/s" : 1900 "Unknown"), 1901 hw->bus.width, 1902 (speed == PCIE_SPEED_2_5GT ? "20%" : 1903 speed == PCIE_SPEED_5_0GT ? "20%" : 1904 speed == PCIE_SPEED_8_0GT ? "<2%" : 1905 "Unknown"), 1906 (hw->bus.payload == fm10k_bus_payload_128 ? "128B" : 1907 hw->bus.payload == fm10k_bus_payload_256 ? "256B" : 1908 hw->bus.payload == fm10k_bus_payload_512 ? "512B" : 1909 "Unknown")); 1910 1911 switch (hw->bus_caps.speed) { 1912 case fm10k_bus_speed_2500: 1913 /* 8b/10b encoding reduces max throughput by 20% */ 1914 expected_gts = 2 * hw->bus_caps.width; 1915 break; 1916 case fm10k_bus_speed_5000: 1917 /* 8b/10b encoding reduces max throughput by 20% */ 1918 expected_gts = 4 * hw->bus_caps.width; 1919 break; 1920 case fm10k_bus_speed_8000: 1921 /* 128b/130b encoding has less than 2% impact on throughput */ 1922 expected_gts = 8 * hw->bus_caps.width; 1923 break; 1924 default: 1925 dev_warn(&interface->pdev->dev, 1926 "Unable to determine expected PCI Express bandwidth.\n"); 1927 return; 1928 } 1929 1930 if (max_gts >= expected_gts) 1931 return; 1932 1933 dev_warn(&interface->pdev->dev, 1934 "This device requires %dGT/s of bandwidth for optimal performance.\n", 1935 expected_gts); 1936 dev_warn(&interface->pdev->dev, 1937 "A %sslot with x%d lanes is suggested.\n", 1938 (hw->bus_caps.speed == fm10k_bus_speed_2500 ? "2.5GT/s " : 1939 hw->bus_caps.speed == fm10k_bus_speed_5000 ? "5.0GT/s " : 1940 hw->bus_caps.speed == fm10k_bus_speed_8000 ? "8.0GT/s " : ""), 1941 hw->bus_caps.width); 1942 } 1943 1944 /** 1945 * fm10k_probe - Device Initialization Routine 1946 * @pdev: PCI device information struct 1947 * @ent: entry in fm10k_pci_tbl 1948 * 1949 * Returns 0 on success, negative on failure 1950 * 1951 * fm10k_probe initializes an interface identified by a pci_dev structure. 1952 * The OS initialization, configuring of the interface private structure, 1953 * and a hardware reset occur. 1954 **/ 1955 static int fm10k_probe(struct pci_dev *pdev, const struct pci_device_id *ent) 1956 { 1957 struct net_device *netdev; 1958 struct fm10k_intfc *interface; 1959 int err; 1960 1961 if (pdev->error_state != pci_channel_io_normal) { 1962 dev_err(&pdev->dev, 1963 "PCI device still in an error state. Unable to load...\n"); 1964 return -EIO; 1965 } 1966 1967 err = pci_enable_device_mem(pdev); 1968 if (err) { 1969 dev_err(&pdev->dev, 1970 "PCI enable device failed: %d\n", err); 1971 return err; 1972 } 1973 1974 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(48)); 1975 if (err) 1976 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32)); 1977 if (err) { 1978 dev_err(&pdev->dev, 1979 "DMA configuration failed: %d\n", err); 1980 goto err_dma; 1981 } 1982 1983 err = pci_request_mem_regions(pdev, fm10k_driver_name); 1984 if (err) { 1985 dev_err(&pdev->dev, 1986 "pci_request_selected_regions failed: %d\n", err); 1987 goto err_pci_reg; 1988 } 1989 1990 pci_enable_pcie_error_reporting(pdev); 1991 1992 pci_set_master(pdev); 1993 pci_save_state(pdev); 1994 1995 netdev = fm10k_alloc_netdev(fm10k_info_tbl[ent->driver_data]); 1996 if (!netdev) { 1997 err = -ENOMEM; 1998 goto err_alloc_netdev; 1999 } 2000 2001 SET_NETDEV_DEV(netdev, &pdev->dev); 2002 2003 interface = netdev_priv(netdev); 2004 pci_set_drvdata(pdev, interface); 2005 2006 interface->netdev = netdev; 2007 interface->pdev = pdev; 2008 2009 interface->uc_addr = ioremap(pci_resource_start(pdev, 0), 2010 FM10K_UC_ADDR_SIZE); 2011 if (!interface->uc_addr) { 2012 err = -EIO; 2013 goto err_ioremap; 2014 } 2015 2016 err = fm10k_sw_init(interface, ent); 2017 if (err) 2018 goto err_sw_init; 2019 2020 /* enable debugfs support */ 2021 fm10k_dbg_intfc_init(interface); 2022 2023 err = fm10k_init_queueing_scheme(interface); 2024 if (err) 2025 goto err_sw_init; 2026 2027 /* the mbx interrupt might attempt to schedule the service task, so we 2028 * must ensure it is disabled since we haven't yet requested the timer 2029 * or work item. 2030 */ 2031 set_bit(__FM10K_SERVICE_DISABLE, &interface->state); 2032 2033 err = fm10k_mbx_request_irq(interface); 2034 if (err) 2035 goto err_mbx_interrupt; 2036 2037 /* final check of hardware state before registering the interface */ 2038 err = fm10k_hw_ready(interface); 2039 if (err) 2040 goto err_register; 2041 2042 err = register_netdev(netdev); 2043 if (err) 2044 goto err_register; 2045 2046 /* carrier off reporting is important to ethtool even BEFORE open */ 2047 netif_carrier_off(netdev); 2048 2049 /* stop all the transmit queues from transmitting until link is up */ 2050 netif_tx_stop_all_queues(netdev); 2051 2052 /* Initialize service timer and service task late in order to avoid 2053 * cleanup issues. 2054 */ 2055 setup_timer(&interface->service_timer, &fm10k_service_timer, 2056 (unsigned long)interface); 2057 INIT_WORK(&interface->service_task, fm10k_service_task); 2058 2059 /* kick off service timer now, even when interface is down */ 2060 mod_timer(&interface->service_timer, (HZ * 2) + jiffies); 2061 2062 /* print warning for non-optimal configurations */ 2063 fm10k_slot_warn(interface); 2064 2065 /* report MAC address for logging */ 2066 dev_info(&pdev->dev, "%pM\n", netdev->dev_addr); 2067 2068 /* enable SR-IOV after registering netdev to enforce PF/VF ordering */ 2069 fm10k_iov_configure(pdev, 0); 2070 2071 /* clear the service task disable bit to allow service task to start */ 2072 clear_bit(__FM10K_SERVICE_DISABLE, &interface->state); 2073 2074 return 0; 2075 2076 err_register: 2077 fm10k_mbx_free_irq(interface); 2078 err_mbx_interrupt: 2079 fm10k_clear_queueing_scheme(interface); 2080 err_sw_init: 2081 if (interface->sw_addr) 2082 iounmap(interface->sw_addr); 2083 iounmap(interface->uc_addr); 2084 err_ioremap: 2085 free_netdev(netdev); 2086 err_alloc_netdev: 2087 pci_release_mem_regions(pdev); 2088 err_pci_reg: 2089 err_dma: 2090 pci_disable_device(pdev); 2091 return err; 2092 } 2093 2094 /** 2095 * fm10k_remove - Device Removal Routine 2096 * @pdev: PCI device information struct 2097 * 2098 * fm10k_remove is called by the PCI subsystem to alert the driver 2099 * that it should release a PCI device. The could be caused by a 2100 * Hot-Plug event, or because the driver is going to be removed from 2101 * memory. 2102 **/ 2103 static void fm10k_remove(struct pci_dev *pdev) 2104 { 2105 struct fm10k_intfc *interface = pci_get_drvdata(pdev); 2106 struct net_device *netdev = interface->netdev; 2107 2108 del_timer_sync(&interface->service_timer); 2109 2110 set_bit(__FM10K_SERVICE_DISABLE, &interface->state); 2111 cancel_work_sync(&interface->service_task); 2112 2113 /* free netdev, this may bounce the interrupts due to setup_tc */ 2114 if (netdev->reg_state == NETREG_REGISTERED) 2115 unregister_netdev(netdev); 2116 2117 /* release VFs */ 2118 fm10k_iov_disable(pdev); 2119 2120 /* disable mailbox interrupt */ 2121 fm10k_mbx_free_irq(interface); 2122 2123 /* free interrupts */ 2124 fm10k_clear_queueing_scheme(interface); 2125 2126 /* remove any debugfs interfaces */ 2127 fm10k_dbg_intfc_exit(interface); 2128 2129 if (interface->sw_addr) 2130 iounmap(interface->sw_addr); 2131 iounmap(interface->uc_addr); 2132 2133 free_netdev(netdev); 2134 2135 pci_release_mem_regions(pdev); 2136 2137 pci_disable_pcie_error_reporting(pdev); 2138 2139 pci_disable_device(pdev); 2140 } 2141 2142 static void fm10k_prepare_suspend(struct fm10k_intfc *interface) 2143 { 2144 /* the watchdog task reads from registers, which might appear like 2145 * a surprise remove if the PCIe device is disabled while we're 2146 * stopped. We stop the watchdog task until after we resume software 2147 * activity. 2148 */ 2149 set_bit(__FM10K_SERVICE_DISABLE, &interface->state); 2150 cancel_work_sync(&interface->service_task); 2151 2152 fm10k_prepare_for_reset(interface); 2153 } 2154 2155 static int fm10k_handle_resume(struct fm10k_intfc *interface) 2156 { 2157 struct fm10k_hw *hw = &interface->hw; 2158 int err; 2159 2160 /* reset statistics starting values */ 2161 hw->mac.ops.rebind_hw_stats(hw, &interface->stats); 2162 2163 err = fm10k_handle_reset(interface); 2164 if (err) 2165 return err; 2166 2167 /* assume host is not ready, to prevent race with watchdog in case we 2168 * actually don't have connection to the switch 2169 */ 2170 interface->host_ready = false; 2171 fm10k_watchdog_host_not_ready(interface); 2172 2173 /* force link to stay down for a second to prevent link flutter */ 2174 interface->link_down_event = jiffies + (HZ); 2175 set_bit(__FM10K_LINK_DOWN, &interface->state); 2176 2177 /* clear the service task disable bit to allow service task to start */ 2178 clear_bit(__FM10K_SERVICE_DISABLE, &interface->state); 2179 fm10k_service_event_schedule(interface); 2180 2181 return err; 2182 } 2183 2184 #ifdef CONFIG_PM 2185 /** 2186 * fm10k_resume - Restore device to pre-sleep state 2187 * @pdev: PCI device information struct 2188 * 2189 * fm10k_resume is called after the system has powered back up from a sleep 2190 * state and is ready to resume operation. This function is meant to restore 2191 * the device back to its pre-sleep state. 2192 **/ 2193 static int fm10k_resume(struct pci_dev *pdev) 2194 { 2195 struct fm10k_intfc *interface = pci_get_drvdata(pdev); 2196 struct net_device *netdev = interface->netdev; 2197 struct fm10k_hw *hw = &interface->hw; 2198 u32 err; 2199 2200 pci_set_power_state(pdev, PCI_D0); 2201 pci_restore_state(pdev); 2202 2203 /* pci_restore_state clears dev->state_saved so call 2204 * pci_save_state to restore it. 2205 */ 2206 pci_save_state(pdev); 2207 2208 err = pci_enable_device_mem(pdev); 2209 if (err) { 2210 dev_err(&pdev->dev, "Cannot enable PCI device from suspend\n"); 2211 return err; 2212 } 2213 pci_set_master(pdev); 2214 2215 pci_wake_from_d3(pdev, false); 2216 2217 /* refresh hw_addr in case it was dropped */ 2218 hw->hw_addr = interface->uc_addr; 2219 2220 err = fm10k_handle_resume(interface); 2221 if (err) 2222 return err; 2223 2224 netif_device_attach(netdev); 2225 2226 return 0; 2227 } 2228 2229 /** 2230 * fm10k_suspend - Prepare the device for a system sleep state 2231 * @pdev: PCI device information struct 2232 * 2233 * fm10k_suspend is meant to shutdown the device prior to the system entering 2234 * a sleep state. The fm10k hardware does not support wake on lan so the 2235 * driver simply needs to shut down the device so it is in a low power state. 2236 **/ 2237 static int fm10k_suspend(struct pci_dev *pdev, 2238 pm_message_t __always_unused state) 2239 { 2240 struct fm10k_intfc *interface = pci_get_drvdata(pdev); 2241 struct net_device *netdev = interface->netdev; 2242 int err = 0; 2243 2244 netif_device_detach(netdev); 2245 2246 fm10k_prepare_suspend(interface); 2247 2248 err = pci_save_state(pdev); 2249 if (err) 2250 return err; 2251 2252 pci_disable_device(pdev); 2253 pci_wake_from_d3(pdev, false); 2254 pci_set_power_state(pdev, PCI_D3hot); 2255 2256 return 0; 2257 } 2258 2259 #endif /* CONFIG_PM */ 2260 /** 2261 * fm10k_io_error_detected - called when PCI error is detected 2262 * @pdev: Pointer to PCI device 2263 * @state: The current pci connection state 2264 * 2265 * This function is called after a PCI bus error affecting 2266 * this device has been detected. 2267 */ 2268 static pci_ers_result_t fm10k_io_error_detected(struct pci_dev *pdev, 2269 pci_channel_state_t state) 2270 { 2271 struct fm10k_intfc *interface = pci_get_drvdata(pdev); 2272 struct net_device *netdev = interface->netdev; 2273 2274 netif_device_detach(netdev); 2275 2276 if (state == pci_channel_io_perm_failure) 2277 return PCI_ERS_RESULT_DISCONNECT; 2278 2279 fm10k_prepare_suspend(interface); 2280 2281 /* Request a slot reset. */ 2282 return PCI_ERS_RESULT_NEED_RESET; 2283 } 2284 2285 /** 2286 * fm10k_io_slot_reset - called after the pci bus has been reset. 2287 * @pdev: Pointer to PCI device 2288 * 2289 * Restart the card from scratch, as if from a cold-boot. 2290 */ 2291 static pci_ers_result_t fm10k_io_slot_reset(struct pci_dev *pdev) 2292 { 2293 pci_ers_result_t result; 2294 2295 if (pci_reenable_device(pdev)) { 2296 dev_err(&pdev->dev, 2297 "Cannot re-enable PCI device after reset.\n"); 2298 result = PCI_ERS_RESULT_DISCONNECT; 2299 } else { 2300 pci_set_master(pdev); 2301 pci_restore_state(pdev); 2302 2303 /* After second error pci->state_saved is false, this 2304 * resets it so EEH doesn't break. 2305 */ 2306 pci_save_state(pdev); 2307 2308 pci_wake_from_d3(pdev, false); 2309 2310 result = PCI_ERS_RESULT_RECOVERED; 2311 } 2312 2313 pci_cleanup_aer_uncorrect_error_status(pdev); 2314 2315 return result; 2316 } 2317 2318 /** 2319 * fm10k_io_resume - called when traffic can start flowing again. 2320 * @pdev: Pointer to PCI device 2321 * 2322 * This callback is called when the error recovery driver tells us that 2323 * its OK to resume normal operation. 2324 */ 2325 static void fm10k_io_resume(struct pci_dev *pdev) 2326 { 2327 struct fm10k_intfc *interface = pci_get_drvdata(pdev); 2328 struct net_device *netdev = interface->netdev; 2329 int err; 2330 2331 err = fm10k_handle_resume(interface); 2332 2333 if (err) 2334 dev_warn(&pdev->dev, 2335 "fm10k_io_resume failed: %d\n", err); 2336 else 2337 netif_device_attach(netdev); 2338 } 2339 2340 /** 2341 * fm10k_io_reset_notify - called when PCI function is reset 2342 * @pdev: Pointer to PCI device 2343 * 2344 * This callback is called when the PCI function is reset such as from 2345 * /sys/class/net/<enpX>/device/reset or similar. When prepare is true, it 2346 * means we should prepare for a function reset. If prepare is false, it means 2347 * the function reset just occurred. 2348 */ 2349 static void fm10k_io_reset_notify(struct pci_dev *pdev, bool prepare) 2350 { 2351 struct fm10k_intfc *interface = pci_get_drvdata(pdev); 2352 int err = 0; 2353 2354 if (prepare) { 2355 /* warn incase we have any active VF devices */ 2356 if (pci_num_vf(pdev)) 2357 dev_warn(&pdev->dev, 2358 "PCIe FLR may cause issues for any active VF devices\n"); 2359 2360 fm10k_prepare_suspend(interface); 2361 } else { 2362 err = fm10k_handle_resume(interface); 2363 } 2364 2365 if (err) { 2366 dev_warn(&pdev->dev, 2367 "fm10k_io_reset_notify failed: %d\n", err); 2368 netif_device_detach(interface->netdev); 2369 } 2370 } 2371 2372 static const struct pci_error_handlers fm10k_err_handler = { 2373 .error_detected = fm10k_io_error_detected, 2374 .slot_reset = fm10k_io_slot_reset, 2375 .resume = fm10k_io_resume, 2376 .reset_notify = fm10k_io_reset_notify, 2377 }; 2378 2379 static struct pci_driver fm10k_driver = { 2380 .name = fm10k_driver_name, 2381 .id_table = fm10k_pci_tbl, 2382 .probe = fm10k_probe, 2383 .remove = fm10k_remove, 2384 #ifdef CONFIG_PM 2385 .suspend = fm10k_suspend, 2386 .resume = fm10k_resume, 2387 #endif 2388 .sriov_configure = fm10k_iov_configure, 2389 .err_handler = &fm10k_err_handler 2390 }; 2391 2392 /** 2393 * fm10k_register_pci_driver - register driver interface 2394 * 2395 * This function is called on module load in order to register the driver. 2396 **/ 2397 int fm10k_register_pci_driver(void) 2398 { 2399 return pci_register_driver(&fm10k_driver); 2400 } 2401 2402 /** 2403 * fm10k_unregister_pci_driver - unregister driver interface 2404 * 2405 * This function is called on module unload in order to remove the driver. 2406 **/ 2407 void fm10k_unregister_pci_driver(void) 2408 { 2409 pci_unregister_driver(&fm10k_driver); 2410 } 2411