1 // SPDX-License-Identifier: GPL-2.0
2 /* Copyright(c) 2013 - 2018 Intel Corporation. */
4 #include <linux/module.h>
5 #include <linux/interrupt.h>
10 static const struct fm10k_info *fm10k_info_tbl[] = {
11 [fm10k_device_pf] = &fm10k_pf_info,
12 [fm10k_device_vf] = &fm10k_vf_info,
16 * fm10k_pci_tbl - PCI Device ID Table
18 * Wildcard entries (PCI_ANY_ID) should come last
19 * Last entry must be all 0s
21 * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
22 * Class, Class Mask, private data (not used) }
24 static const struct pci_device_id fm10k_pci_tbl[] = {
25 { PCI_VDEVICE(INTEL, FM10K_DEV_ID_PF), fm10k_device_pf },
26 { PCI_VDEVICE(INTEL, FM10K_DEV_ID_VF), fm10k_device_vf },
27 /* required last entry */
30 MODULE_DEVICE_TABLE(pci, fm10k_pci_tbl);
32 u16 fm10k_read_pci_cfg_word(struct fm10k_hw *hw, u32 reg)
34 struct fm10k_intfc *interface = hw->back;
37 if (FM10K_REMOVED(hw->hw_addr))
40 pci_read_config_word(interface->pdev, reg, &value);
42 fm10k_write_flush(hw);
47 u32 fm10k_read_reg(struct fm10k_hw *hw, int reg)
49 u32 __iomem *hw_addr = READ_ONCE(hw->hw_addr);
52 if (FM10K_REMOVED(hw_addr))
55 value = readl(&hw_addr[reg]);
56 if (!(~value) && (!reg || !(~readl(hw_addr)))) {
57 struct fm10k_intfc *interface = hw->back;
58 struct net_device *netdev = interface->netdev;
61 netif_device_detach(netdev);
62 netdev_err(netdev, "PCIe link lost, device now detached\n");
68 static int fm10k_hw_ready(struct fm10k_intfc *interface)
70 struct fm10k_hw *hw = &interface->hw;
72 fm10k_write_flush(hw);
74 return FM10K_REMOVED(hw->hw_addr) ? -ENODEV : 0;
78 * fm10k_macvlan_schedule - Schedule MAC/VLAN queue task
79 * @interface: fm10k private interface structure
81 * Schedule the MAC/VLAN queue monitor task. If the MAC/VLAN task cannot be
82 * started immediately, request that it be restarted when possible.
84 void fm10k_macvlan_schedule(struct fm10k_intfc *interface)
86 /* Avoid processing the MAC/VLAN queue when the service task is
87 * disabled, or when we're resetting the device.
89 if (!test_bit(__FM10K_MACVLAN_DISABLE, interface->state) &&
90 !test_and_set_bit(__FM10K_MACVLAN_SCHED, interface->state)) {
91 clear_bit(__FM10K_MACVLAN_REQUEST, interface->state);
92 /* We delay the actual start of execution in order to allow
93 * multiple MAC/VLAN updates to accumulate before handling
94 * them, and to allow some time to let the mailbox drain
97 queue_delayed_work(fm10k_workqueue,
98 &interface->macvlan_task, 10);
100 set_bit(__FM10K_MACVLAN_REQUEST, interface->state);
105 * fm10k_stop_macvlan_task - Stop the MAC/VLAN queue monitor
106 * @interface: fm10k private interface structure
108 * Wait until the MAC/VLAN queue task has stopped, and cancel any future
111 static void fm10k_stop_macvlan_task(struct fm10k_intfc *interface)
113 /* Disable the MAC/VLAN work item */
114 set_bit(__FM10K_MACVLAN_DISABLE, interface->state);
116 /* Make sure we waited until any current invocations have stopped */
117 cancel_delayed_work_sync(&interface->macvlan_task);
119 /* We set the __FM10K_MACVLAN_SCHED bit when we schedule the task.
120 * However, it may not be unset of the MAC/VLAN task never actually
121 * got a chance to run. Since we've canceled the task here, and it
122 * cannot be rescheuled right now, we need to ensure the scheduled bit
125 clear_bit(__FM10K_MACVLAN_SCHED, interface->state);
129 * fm10k_resume_macvlan_task - Restart the MAC/VLAN queue monitor
130 * @interface: fm10k private interface structure
132 * Clear the __FM10K_MACVLAN_DISABLE bit and, if a request occurred, schedule
133 * the MAC/VLAN work monitor.
135 static void fm10k_resume_macvlan_task(struct fm10k_intfc *interface)
137 /* Re-enable the MAC/VLAN work item */
138 clear_bit(__FM10K_MACVLAN_DISABLE, interface->state);
140 /* We might have received a MAC/VLAN request while disabled. If so,
141 * kick off the queue now.
143 if (test_bit(__FM10K_MACVLAN_REQUEST, interface->state))
144 fm10k_macvlan_schedule(interface);
147 void fm10k_service_event_schedule(struct fm10k_intfc *interface)
149 if (!test_bit(__FM10K_SERVICE_DISABLE, interface->state) &&
150 !test_and_set_bit(__FM10K_SERVICE_SCHED, interface->state)) {
151 clear_bit(__FM10K_SERVICE_REQUEST, interface->state);
152 queue_work(fm10k_workqueue, &interface->service_task);
154 set_bit(__FM10K_SERVICE_REQUEST, interface->state);
158 static void fm10k_service_event_complete(struct fm10k_intfc *interface)
160 WARN_ON(!test_bit(__FM10K_SERVICE_SCHED, interface->state));
162 /* flush memory to make sure state is correct before next watchog */
163 smp_mb__before_atomic();
164 clear_bit(__FM10K_SERVICE_SCHED, interface->state);
166 /* If a service event was requested since we started, immediately
167 * re-schedule now. This ensures we don't drop a request until the
170 if (test_bit(__FM10K_SERVICE_REQUEST, interface->state))
171 fm10k_service_event_schedule(interface);
174 static void fm10k_stop_service_event(struct fm10k_intfc *interface)
176 set_bit(__FM10K_SERVICE_DISABLE, interface->state);
177 cancel_work_sync(&interface->service_task);
179 /* It's possible that cancel_work_sync stopped the service task from
180 * running before it could actually start. In this case the
181 * __FM10K_SERVICE_SCHED bit will never be cleared. Since we know that
182 * the service task cannot be running at this point, we need to clear
183 * the scheduled bit, as otherwise the service task may never be
186 clear_bit(__FM10K_SERVICE_SCHED, interface->state);
189 static void fm10k_start_service_event(struct fm10k_intfc *interface)
191 clear_bit(__FM10K_SERVICE_DISABLE, interface->state);
192 fm10k_service_event_schedule(interface);
196 * fm10k_service_timer - Timer Call-back
197 * @t: pointer to timer data
199 static void fm10k_service_timer(struct timer_list *t)
201 struct fm10k_intfc *interface = from_timer(interface, t,
204 /* Reset the timer */
205 mod_timer(&interface->service_timer, (HZ * 2) + jiffies);
207 fm10k_service_event_schedule(interface);
211 * fm10k_prepare_for_reset - Prepare the driver and device for a pending reset
212 * @interface: fm10k private data structure
214 * This function prepares for a device reset by shutting as much down as we
215 * can. It does nothing and returns false if __FM10K_RESETTING was already set
216 * prior to calling this function. It returns true if it actually did work.
218 static bool fm10k_prepare_for_reset(struct fm10k_intfc *interface)
220 struct net_device *netdev = interface->netdev;
222 WARN_ON(in_interrupt());
224 /* put off any impending NetWatchDogTimeout */
225 netif_trans_update(netdev);
227 /* Nothing to do if a reset is already in progress */
228 if (test_and_set_bit(__FM10K_RESETTING, interface->state))
231 /* As the MAC/VLAN task will be accessing registers it must not be
232 * running while we reset. Although the task will not be scheduled
233 * once we start resetting it may already be running
235 fm10k_stop_macvlan_task(interface);
239 fm10k_iov_suspend(interface->pdev);
241 if (netif_running(netdev))
244 fm10k_mbx_free_irq(interface);
246 /* free interrupts */
247 fm10k_clear_queueing_scheme(interface);
249 /* delay any future reset requests */
250 interface->last_reset = jiffies + (10 * HZ);
257 static int fm10k_handle_reset(struct fm10k_intfc *interface)
259 struct net_device *netdev = interface->netdev;
260 struct fm10k_hw *hw = &interface->hw;
263 WARN_ON(!test_bit(__FM10K_RESETTING, interface->state));
267 pci_set_master(interface->pdev);
269 /* reset and initialize the hardware so it is in a known state */
270 err = hw->mac.ops.reset_hw(hw);
272 dev_err(&interface->pdev->dev, "reset_hw failed: %d\n", err);
276 err = hw->mac.ops.init_hw(hw);
278 dev_err(&interface->pdev->dev, "init_hw failed: %d\n", err);
282 err = fm10k_init_queueing_scheme(interface);
284 dev_err(&interface->pdev->dev,
285 "init_queueing_scheme failed: %d\n", err);
289 /* re-associate interrupts */
290 err = fm10k_mbx_request_irq(interface);
294 err = fm10k_hw_ready(interface);
298 /* update hardware address for VFs if perm_addr has changed */
299 if (hw->mac.type == fm10k_mac_vf) {
300 if (is_valid_ether_addr(hw->mac.perm_addr)) {
301 ether_addr_copy(hw->mac.addr, hw->mac.perm_addr);
302 ether_addr_copy(netdev->perm_addr, hw->mac.perm_addr);
303 ether_addr_copy(netdev->dev_addr, hw->mac.perm_addr);
304 netdev->addr_assign_type &= ~NET_ADDR_RANDOM;
307 if (hw->mac.vlan_override)
308 netdev->features &= ~NETIF_F_HW_VLAN_CTAG_RX;
310 netdev->features |= NETIF_F_HW_VLAN_CTAG_RX;
313 err = netif_running(netdev) ? fm10k_open(netdev) : 0;
317 fm10k_iov_resume(interface->pdev);
321 fm10k_resume_macvlan_task(interface);
323 clear_bit(__FM10K_RESETTING, interface->state);
327 fm10k_mbx_free_irq(interface);
329 fm10k_clear_queueing_scheme(interface);
331 netif_device_detach(netdev);
335 clear_bit(__FM10K_RESETTING, interface->state);
340 static void fm10k_detach_subtask(struct fm10k_intfc *interface)
342 struct net_device *netdev = interface->netdev;
343 u32 __iomem *hw_addr;
347 /* do nothing if netdev is still present or hw_addr is set */
348 if (netif_device_present(netdev) || interface->hw.hw_addr)
351 /* We've lost the PCIe register space, and can no longer access the
352 * device. Shut everything except the detach subtask down and prepare
353 * to reset the device in case we recover. If we actually prepare for
354 * reset, indicate that we're detached.
356 if (fm10k_prepare_for_reset(interface))
357 set_bit(__FM10K_RESET_DETACHED, interface->state);
359 /* check the real address space to see if we've recovered */
360 hw_addr = READ_ONCE(interface->uc_addr);
361 value = readl(hw_addr);
363 /* Make sure the reset was initiated because we detached,
364 * otherwise we might race with a different reset flow.
366 if (!test_and_clear_bit(__FM10K_RESET_DETACHED,
370 /* Restore the hardware address */
371 interface->hw.hw_addr = interface->uc_addr;
373 /* PCIe link has been restored, and the device is active
374 * again. Restore everything and reset the device.
376 err = fm10k_handle_reset(interface);
378 netdev_err(netdev, "Unable to reset device: %d\n", err);
379 interface->hw.hw_addr = NULL;
383 /* Re-attach the netdev */
384 netif_device_attach(netdev);
385 netdev_warn(netdev, "PCIe link restored, device now attached\n");
390 static void fm10k_reset_subtask(struct fm10k_intfc *interface)
394 if (!test_and_clear_bit(FM10K_FLAG_RESET_REQUESTED,
398 /* If another thread has already prepared to reset the device, we
399 * should not attempt to handle a reset here, since we'd race with
400 * that thread. This may happen if we suspend the device or if the
401 * PCIe link is lost. In this case, we'll just ignore the RESET
402 * request, as it will (eventually) be taken care of when the thread
403 * which actually started the reset is finished.
405 if (!fm10k_prepare_for_reset(interface))
408 netdev_err(interface->netdev, "Reset interface\n");
410 err = fm10k_handle_reset(interface);
412 dev_err(&interface->pdev->dev,
413 "fm10k_handle_reset failed: %d\n", err);
417 * fm10k_configure_swpri_map - Configure Receive SWPRI to PC mapping
418 * @interface: board private structure
420 * Configure the SWPRI to PC mapping for the port.
422 static void fm10k_configure_swpri_map(struct fm10k_intfc *interface)
424 struct net_device *netdev = interface->netdev;
425 struct fm10k_hw *hw = &interface->hw;
428 /* clear flag indicating update is needed */
429 clear_bit(FM10K_FLAG_SWPRI_CONFIG, interface->flags);
431 /* these registers are only available on the PF */
432 if (hw->mac.type != fm10k_mac_pf)
435 /* configure SWPRI to PC map */
436 for (i = 0; i < FM10K_SWPRI_MAX; i++)
437 fm10k_write_reg(hw, FM10K_SWPRI_MAP(i),
438 netdev_get_prio_tc_map(netdev, i));
442 * fm10k_watchdog_update_host_state - Update the link status based on host.
443 * @interface: board private structure
445 static void fm10k_watchdog_update_host_state(struct fm10k_intfc *interface)
447 struct fm10k_hw *hw = &interface->hw;
450 if (test_bit(__FM10K_LINK_DOWN, interface->state)) {
451 interface->host_ready = false;
452 if (time_is_after_jiffies(interface->link_down_event))
454 clear_bit(__FM10K_LINK_DOWN, interface->state);
457 if (test_bit(FM10K_FLAG_SWPRI_CONFIG, interface->flags)) {
458 if (rtnl_trylock()) {
459 fm10k_configure_swpri_map(interface);
464 /* lock the mailbox for transmit and receive */
465 fm10k_mbx_lock(interface);
467 err = hw->mac.ops.get_host_state(hw, &interface->host_ready);
468 if (err && time_is_before_jiffies(interface->last_reset))
469 set_bit(FM10K_FLAG_RESET_REQUESTED, interface->flags);
472 fm10k_mbx_unlock(interface);
476 * fm10k_mbx_subtask - Process upstream and downstream mailboxes
477 * @interface: board private structure
479 * This function will process both the upstream and downstream mailboxes.
481 static void fm10k_mbx_subtask(struct fm10k_intfc *interface)
483 /* If we're resetting, bail out */
484 if (test_bit(__FM10K_RESETTING, interface->state))
487 /* process upstream mailbox and update device state */
488 fm10k_watchdog_update_host_state(interface);
490 /* process downstream mailboxes */
491 fm10k_iov_mbx(interface);
495 * fm10k_watchdog_host_is_ready - Update netdev status based on host ready
496 * @interface: board private structure
498 static void fm10k_watchdog_host_is_ready(struct fm10k_intfc *interface)
500 struct net_device *netdev = interface->netdev;
502 /* only continue if link state is currently down */
503 if (netif_carrier_ok(netdev))
506 netif_info(interface, drv, netdev, "NIC Link is up\n");
508 netif_carrier_on(netdev);
509 netif_tx_wake_all_queues(netdev);
513 * fm10k_watchdog_host_not_ready - Update netdev status based on host not ready
514 * @interface: board private structure
516 static void fm10k_watchdog_host_not_ready(struct fm10k_intfc *interface)
518 struct net_device *netdev = interface->netdev;
520 /* only continue if link state is currently up */
521 if (!netif_carrier_ok(netdev))
524 netif_info(interface, drv, netdev, "NIC Link is down\n");
526 netif_carrier_off(netdev);
527 netif_tx_stop_all_queues(netdev);
531 * fm10k_update_stats - Update the board statistics counters.
532 * @interface: board private structure
534 void fm10k_update_stats(struct fm10k_intfc *interface)
536 struct net_device_stats *net_stats = &interface->netdev->stats;
537 struct fm10k_hw *hw = &interface->hw;
538 u64 hw_csum_tx_good = 0, hw_csum_rx_good = 0, rx_length_errors = 0;
539 u64 rx_switch_errors = 0, rx_drops = 0, rx_pp_errors = 0;
540 u64 rx_link_errors = 0;
541 u64 rx_errors = 0, rx_csum_errors = 0, tx_csum_errors = 0;
542 u64 restart_queue = 0, tx_busy = 0, alloc_failed = 0;
543 u64 rx_bytes_nic = 0, rx_pkts_nic = 0, rx_drops_nic = 0;
544 u64 tx_bytes_nic = 0, tx_pkts_nic = 0;
548 /* ensure only one thread updates stats at a time */
549 if (test_and_set_bit(__FM10K_UPDATING_STATS, interface->state))
552 /* do not allow stats update via service task for next second */
553 interface->next_stats_update = jiffies + HZ;
555 /* gather some stats to the interface struct that are per queue */
556 for (bytes = 0, pkts = 0, i = 0; i < interface->num_tx_queues; i++) {
557 struct fm10k_ring *tx_ring = READ_ONCE(interface->tx_ring[i]);
562 restart_queue += tx_ring->tx_stats.restart_queue;
563 tx_busy += tx_ring->tx_stats.tx_busy;
564 tx_csum_errors += tx_ring->tx_stats.csum_err;
565 bytes += tx_ring->stats.bytes;
566 pkts += tx_ring->stats.packets;
567 hw_csum_tx_good += tx_ring->tx_stats.csum_good;
570 interface->restart_queue = restart_queue;
571 interface->tx_busy = tx_busy;
572 net_stats->tx_bytes = bytes;
573 net_stats->tx_packets = pkts;
574 interface->tx_csum_errors = tx_csum_errors;
575 interface->hw_csum_tx_good = hw_csum_tx_good;
577 /* gather some stats to the interface struct that are per queue */
578 for (bytes = 0, pkts = 0, i = 0; i < interface->num_rx_queues; i++) {
579 struct fm10k_ring *rx_ring = READ_ONCE(interface->rx_ring[i]);
584 bytes += rx_ring->stats.bytes;
585 pkts += rx_ring->stats.packets;
586 alloc_failed += rx_ring->rx_stats.alloc_failed;
587 rx_csum_errors += rx_ring->rx_stats.csum_err;
588 rx_errors += rx_ring->rx_stats.errors;
589 hw_csum_rx_good += rx_ring->rx_stats.csum_good;
590 rx_switch_errors += rx_ring->rx_stats.switch_errors;
591 rx_drops += rx_ring->rx_stats.drops;
592 rx_pp_errors += rx_ring->rx_stats.pp_errors;
593 rx_link_errors += rx_ring->rx_stats.link_errors;
594 rx_length_errors += rx_ring->rx_stats.length_errors;
597 net_stats->rx_bytes = bytes;
598 net_stats->rx_packets = pkts;
599 interface->alloc_failed = alloc_failed;
600 interface->rx_csum_errors = rx_csum_errors;
601 interface->hw_csum_rx_good = hw_csum_rx_good;
602 interface->rx_switch_errors = rx_switch_errors;
603 interface->rx_drops = rx_drops;
604 interface->rx_pp_errors = rx_pp_errors;
605 interface->rx_link_errors = rx_link_errors;
606 interface->rx_length_errors = rx_length_errors;
608 hw->mac.ops.update_hw_stats(hw, &interface->stats);
610 for (i = 0; i < hw->mac.max_queues; i++) {
611 struct fm10k_hw_stats_q *q = &interface->stats.q[i];
613 tx_bytes_nic += q->tx_bytes.count;
614 tx_pkts_nic += q->tx_packets.count;
615 rx_bytes_nic += q->rx_bytes.count;
616 rx_pkts_nic += q->rx_packets.count;
617 rx_drops_nic += q->rx_drops.count;
620 interface->tx_bytes_nic = tx_bytes_nic;
621 interface->tx_packets_nic = tx_pkts_nic;
622 interface->rx_bytes_nic = rx_bytes_nic;
623 interface->rx_packets_nic = rx_pkts_nic;
624 interface->rx_drops_nic = rx_drops_nic;
626 /* Fill out the OS statistics structure */
627 net_stats->rx_errors = rx_errors;
628 net_stats->rx_dropped = interface->stats.nodesc_drop.count;
630 clear_bit(__FM10K_UPDATING_STATS, interface->state);
634 * fm10k_watchdog_flush_tx - flush queues on host not ready
635 * @interface: pointer to the device interface structure
637 static void fm10k_watchdog_flush_tx(struct fm10k_intfc *interface)
639 int some_tx_pending = 0;
642 /* nothing to do if carrier is up */
643 if (netif_carrier_ok(interface->netdev))
646 for (i = 0; i < interface->num_tx_queues; i++) {
647 struct fm10k_ring *tx_ring = interface->tx_ring[i];
649 if (tx_ring->next_to_use != tx_ring->next_to_clean) {
655 /* We've lost link, so the controller stops DMA, but we've got
656 * queued Tx work that's never going to get done, so reset
657 * controller to flush Tx.
660 set_bit(FM10K_FLAG_RESET_REQUESTED, interface->flags);
664 * fm10k_watchdog_subtask - check and bring link up
665 * @interface: pointer to the device interface structure
667 static void fm10k_watchdog_subtask(struct fm10k_intfc *interface)
669 /* if interface is down do nothing */
670 if (test_bit(__FM10K_DOWN, interface->state) ||
671 test_bit(__FM10K_RESETTING, interface->state))
674 if (interface->host_ready)
675 fm10k_watchdog_host_is_ready(interface);
677 fm10k_watchdog_host_not_ready(interface);
679 /* update stats only once every second */
680 if (time_is_before_jiffies(interface->next_stats_update))
681 fm10k_update_stats(interface);
683 /* flush any uncompleted work */
684 fm10k_watchdog_flush_tx(interface);
688 * fm10k_check_hang_subtask - check for hung queues and dropped interrupts
689 * @interface: pointer to the device interface structure
691 * This function serves two purposes. First it strobes the interrupt lines
692 * in order to make certain interrupts are occurring. Secondly it sets the
693 * bits needed to check for TX hangs. As a result we should immediately
694 * determine if a hang has occurred.
696 static void fm10k_check_hang_subtask(struct fm10k_intfc *interface)
700 /* If we're down or resetting, just bail */
701 if (test_bit(__FM10K_DOWN, interface->state) ||
702 test_bit(__FM10K_RESETTING, interface->state))
705 /* rate limit tx hang checks to only once every 2 seconds */
706 if (time_is_after_eq_jiffies(interface->next_tx_hang_check))
708 interface->next_tx_hang_check = jiffies + (2 * HZ);
710 if (netif_carrier_ok(interface->netdev)) {
711 /* Force detection of hung controller */
712 for (i = 0; i < interface->num_tx_queues; i++)
713 set_check_for_tx_hang(interface->tx_ring[i]);
715 /* Rearm all in-use q_vectors for immediate firing */
716 for (i = 0; i < interface->num_q_vectors; i++) {
717 struct fm10k_q_vector *qv = interface->q_vector[i];
719 if (!qv->tx.count && !qv->rx.count)
721 writel(FM10K_ITR_ENABLE | FM10K_ITR_PENDING2, qv->itr);
727 * fm10k_service_task - manages and runs subtasks
728 * @work: pointer to work_struct containing our data
730 static void fm10k_service_task(struct work_struct *work)
732 struct fm10k_intfc *interface;
734 interface = container_of(work, struct fm10k_intfc, service_task);
736 /* Check whether we're detached first */
737 fm10k_detach_subtask(interface);
739 /* tasks run even when interface is down */
740 fm10k_mbx_subtask(interface);
741 fm10k_reset_subtask(interface);
743 /* tasks only run when interface is up */
744 fm10k_watchdog_subtask(interface);
745 fm10k_check_hang_subtask(interface);
747 /* release lock on service events to allow scheduling next event */
748 fm10k_service_event_complete(interface);
752 * fm10k_macvlan_task - send queued MAC/VLAN requests to switch manager
753 * @work: pointer to work_struct containing our data
755 * This work item handles sending MAC/VLAN updates to the switch manager. When
756 * the interface is up, it will attempt to queue mailbox messages to the
757 * switch manager requesting updates for MAC/VLAN pairs. If the Tx fifo of the
758 * mailbox is full, it will reschedule itself to try again in a short while.
759 * This ensures that the driver does not overload the switch mailbox with too
760 * many simultaneous requests, causing an unnecessary reset.
762 static void fm10k_macvlan_task(struct work_struct *work)
764 struct fm10k_macvlan_request *item;
765 struct fm10k_intfc *interface;
766 struct delayed_work *dwork;
767 struct list_head *requests;
771 dwork = to_delayed_work(work);
772 interface = container_of(dwork, struct fm10k_intfc, macvlan_task);
774 requests = &interface->macvlan_requests;
777 /* Pop the first item off the list */
778 spin_lock_irqsave(&interface->macvlan_lock, flags);
779 item = list_first_entry_or_null(requests,
780 struct fm10k_macvlan_request,
783 list_del_init(&item->list);
785 spin_unlock_irqrestore(&interface->macvlan_lock, flags);
787 /* We have no more items to process */
791 fm10k_mbx_lock(interface);
793 /* Check that we have plenty of space to send the message. We
794 * want to ensure that the mailbox stays low enough to avoid a
795 * change in the host state, otherwise we may see spurious
796 * link up / link down notifications.
798 if (!hw->mbx.ops.tx_ready(&hw->mbx, FM10K_VFMBX_MSG_MTU + 5)) {
799 hw->mbx.ops.process(hw, &hw->mbx);
800 set_bit(__FM10K_MACVLAN_REQUEST, interface->state);
801 fm10k_mbx_unlock(interface);
803 /* Put the request back on the list */
804 spin_lock_irqsave(&interface->macvlan_lock, flags);
805 list_add(&item->list, requests);
806 spin_unlock_irqrestore(&interface->macvlan_lock, flags);
810 switch (item->type) {
811 case FM10K_MC_MAC_REQUEST:
812 hw->mac.ops.update_mc_addr(hw,
818 case FM10K_UC_MAC_REQUEST:
819 hw->mac.ops.update_uc_addr(hw,
826 case FM10K_VLAN_REQUEST:
827 hw->mac.ops.update_vlan(hw,
836 fm10k_mbx_unlock(interface);
838 /* Free the item now that we've sent the update */
843 WARN_ON(!test_bit(__FM10K_MACVLAN_SCHED, interface->state));
845 /* flush memory to make sure state is correct */
846 smp_mb__before_atomic();
847 clear_bit(__FM10K_MACVLAN_SCHED, interface->state);
849 /* If a MAC/VLAN request was scheduled since we started, we should
850 * re-schedule. However, there is no reason to re-schedule if there is
853 if (test_bit(__FM10K_MACVLAN_REQUEST, interface->state))
854 fm10k_macvlan_schedule(interface);
858 * fm10k_configure_tx_ring - Configure Tx ring after Reset
859 * @interface: board private structure
860 * @ring: structure containing ring specific data
862 * Configure the Tx descriptor ring after a reset.
864 static void fm10k_configure_tx_ring(struct fm10k_intfc *interface,
865 struct fm10k_ring *ring)
867 struct fm10k_hw *hw = &interface->hw;
868 u64 tdba = ring->dma;
869 u32 size = ring->count * sizeof(struct fm10k_tx_desc);
870 u32 txint = FM10K_INT_MAP_DISABLE;
871 u32 txdctl = BIT(FM10K_TXDCTL_MAX_TIME_SHIFT) | FM10K_TXDCTL_ENABLE;
872 u8 reg_idx = ring->reg_idx;
874 /* disable queue to avoid issues while updating state */
875 fm10k_write_reg(hw, FM10K_TXDCTL(reg_idx), 0);
876 fm10k_write_flush(hw);
878 /* possible poll here to verify ring resources have been cleaned */
880 /* set location and size for descriptor ring */
881 fm10k_write_reg(hw, FM10K_TDBAL(reg_idx), tdba & DMA_BIT_MASK(32));
882 fm10k_write_reg(hw, FM10K_TDBAH(reg_idx), tdba >> 32);
883 fm10k_write_reg(hw, FM10K_TDLEN(reg_idx), size);
885 /* reset head and tail pointers */
886 fm10k_write_reg(hw, FM10K_TDH(reg_idx), 0);
887 fm10k_write_reg(hw, FM10K_TDT(reg_idx), 0);
889 /* store tail pointer */
890 ring->tail = &interface->uc_addr[FM10K_TDT(reg_idx)];
892 /* reset ntu and ntc to place SW in sync with hardware */
893 ring->next_to_clean = 0;
894 ring->next_to_use = 0;
897 if (ring->q_vector) {
898 txint = ring->q_vector->v_idx + NON_Q_VECTORS(hw);
899 txint |= FM10K_INT_MAP_TIMER0;
902 fm10k_write_reg(hw, FM10K_TXINT(reg_idx), txint);
904 /* enable use of FTAG bit in Tx descriptor, register is RO for VF */
905 fm10k_write_reg(hw, FM10K_PFVTCTL(reg_idx),
906 FM10K_PFVTCTL_FTAG_DESC_ENABLE);
909 if (!test_and_set_bit(__FM10K_TX_XPS_INIT_DONE, ring->state) &&
911 netif_set_xps_queue(ring->netdev,
912 &ring->q_vector->affinity_mask,
916 fm10k_write_reg(hw, FM10K_TXDCTL(reg_idx), txdctl);
920 * fm10k_enable_tx_ring - Verify Tx ring is enabled after configuration
921 * @interface: board private structure
922 * @ring: structure containing ring specific data
924 * Verify the Tx descriptor ring is ready for transmit.
926 static void fm10k_enable_tx_ring(struct fm10k_intfc *interface,
927 struct fm10k_ring *ring)
929 struct fm10k_hw *hw = &interface->hw;
932 u8 reg_idx = ring->reg_idx;
934 /* if we are already enabled just exit */
935 if (fm10k_read_reg(hw, FM10K_TXDCTL(reg_idx)) & FM10K_TXDCTL_ENABLE)
938 /* poll to verify queue is enabled */
940 usleep_range(1000, 2000);
941 txdctl = fm10k_read_reg(hw, FM10K_TXDCTL(reg_idx));
942 } while (!(txdctl & FM10K_TXDCTL_ENABLE) && --wait_loop);
944 netif_err(interface, drv, interface->netdev,
945 "Could not enable Tx Queue %d\n", reg_idx);
949 * fm10k_configure_tx - Configure Transmit Unit after Reset
950 * @interface: board private structure
952 * Configure the Tx unit of the MAC after a reset.
954 static void fm10k_configure_tx(struct fm10k_intfc *interface)
958 /* Setup the HW Tx Head and Tail descriptor pointers */
959 for (i = 0; i < interface->num_tx_queues; i++)
960 fm10k_configure_tx_ring(interface, interface->tx_ring[i]);
962 /* poll here to verify that Tx rings are now enabled */
963 for (i = 0; i < interface->num_tx_queues; i++)
964 fm10k_enable_tx_ring(interface, interface->tx_ring[i]);
968 * fm10k_configure_rx_ring - Configure Rx ring after Reset
969 * @interface: board private structure
970 * @ring: structure containing ring specific data
972 * Configure the Rx descriptor ring after a reset.
974 static void fm10k_configure_rx_ring(struct fm10k_intfc *interface,
975 struct fm10k_ring *ring)
977 u64 rdba = ring->dma;
978 struct fm10k_hw *hw = &interface->hw;
979 u32 size = ring->count * sizeof(union fm10k_rx_desc);
980 u32 rxqctl, rxdctl = FM10K_RXDCTL_WRITE_BACK_MIN_DELAY;
981 u32 srrctl = FM10K_SRRCTL_BUFFER_CHAINING_EN;
982 u32 rxint = FM10K_INT_MAP_DISABLE;
983 u8 rx_pause = interface->rx_pause;
984 u8 reg_idx = ring->reg_idx;
986 /* disable queue to avoid issues while updating state */
987 rxqctl = fm10k_read_reg(hw, FM10K_RXQCTL(reg_idx));
988 rxqctl &= ~FM10K_RXQCTL_ENABLE;
989 fm10k_write_reg(hw, FM10K_RXQCTL(reg_idx), rxqctl);
990 fm10k_write_flush(hw);
992 /* possible poll here to verify ring resources have been cleaned */
994 /* set location and size for descriptor ring */
995 fm10k_write_reg(hw, FM10K_RDBAL(reg_idx), rdba & DMA_BIT_MASK(32));
996 fm10k_write_reg(hw, FM10K_RDBAH(reg_idx), rdba >> 32);
997 fm10k_write_reg(hw, FM10K_RDLEN(reg_idx), size);
999 /* reset head and tail pointers */
1000 fm10k_write_reg(hw, FM10K_RDH(reg_idx), 0);
1001 fm10k_write_reg(hw, FM10K_RDT(reg_idx), 0);
1003 /* store tail pointer */
1004 ring->tail = &interface->uc_addr[FM10K_RDT(reg_idx)];
1006 /* reset ntu and ntc to place SW in sync with hardware */
1007 ring->next_to_clean = 0;
1008 ring->next_to_use = 0;
1009 ring->next_to_alloc = 0;
1011 /* Configure the Rx buffer size for one buff without split */
1012 srrctl |= FM10K_RX_BUFSZ >> FM10K_SRRCTL_BSIZEPKT_SHIFT;
1014 /* Configure the Rx ring to suppress loopback packets */
1015 srrctl |= FM10K_SRRCTL_LOOPBACK_SUPPRESS;
1016 fm10k_write_reg(hw, FM10K_SRRCTL(reg_idx), srrctl);
1018 /* Enable drop on empty */
1020 if (interface->pfc_en)
1021 rx_pause = interface->pfc_en;
1023 if (!(rx_pause & BIT(ring->qos_pc)))
1024 rxdctl |= FM10K_RXDCTL_DROP_ON_EMPTY;
1026 fm10k_write_reg(hw, FM10K_RXDCTL(reg_idx), rxdctl);
1028 /* assign default VLAN to queue */
1029 ring->vid = hw->mac.default_vid;
1031 /* if we have an active VLAN, disable default VLAN ID */
1032 if (test_bit(hw->mac.default_vid, interface->active_vlans))
1033 ring->vid |= FM10K_VLAN_CLEAR;
1036 if (ring->q_vector) {
1037 rxint = ring->q_vector->v_idx + NON_Q_VECTORS(hw);
1038 rxint |= FM10K_INT_MAP_TIMER1;
1041 fm10k_write_reg(hw, FM10K_RXINT(reg_idx), rxint);
1044 rxqctl = fm10k_read_reg(hw, FM10K_RXQCTL(reg_idx));
1045 rxqctl |= FM10K_RXQCTL_ENABLE;
1046 fm10k_write_reg(hw, FM10K_RXQCTL(reg_idx), rxqctl);
1048 /* place buffers on ring for receive data */
1049 fm10k_alloc_rx_buffers(ring, fm10k_desc_unused(ring));
1053 * fm10k_update_rx_drop_en - Configures the drop enable bits for Rx rings
1054 * @interface: board private structure
1056 * Configure the drop enable bits for the Rx rings.
1058 void fm10k_update_rx_drop_en(struct fm10k_intfc *interface)
1060 struct fm10k_hw *hw = &interface->hw;
1061 u8 rx_pause = interface->rx_pause;
1065 if (interface->pfc_en)
1066 rx_pause = interface->pfc_en;
1069 for (i = 0; i < interface->num_rx_queues; i++) {
1070 struct fm10k_ring *ring = interface->rx_ring[i];
1071 u32 rxdctl = FM10K_RXDCTL_WRITE_BACK_MIN_DELAY;
1072 u8 reg_idx = ring->reg_idx;
1074 if (!(rx_pause & BIT(ring->qos_pc)))
1075 rxdctl |= FM10K_RXDCTL_DROP_ON_EMPTY;
1077 fm10k_write_reg(hw, FM10K_RXDCTL(reg_idx), rxdctl);
1082 * fm10k_configure_dglort - Configure Receive DGLORT after reset
1083 * @interface: board private structure
1085 * Configure the DGLORT description and RSS tables.
1087 static void fm10k_configure_dglort(struct fm10k_intfc *interface)
1089 struct fm10k_dglort_cfg dglort = { 0 };
1090 struct fm10k_hw *hw = &interface->hw;
1094 /* Fill out hash function seeds */
1095 for (i = 0; i < FM10K_RSSRK_SIZE; i++)
1096 fm10k_write_reg(hw, FM10K_RSSRK(0, i), interface->rssrk[i]);
1098 /* Write RETA table to hardware */
1099 for (i = 0; i < FM10K_RETA_SIZE; i++)
1100 fm10k_write_reg(hw, FM10K_RETA(0, i), interface->reta[i]);
1102 /* Generate RSS hash based on packet types, TCP/UDP
1103 * port numbers and/or IPv4/v6 src and dst addresses
1105 mrqc = FM10K_MRQC_IPV4 |
1106 FM10K_MRQC_TCP_IPV4 |
1108 FM10K_MRQC_TCP_IPV6;
1110 if (test_bit(FM10K_FLAG_RSS_FIELD_IPV4_UDP, interface->flags))
1111 mrqc |= FM10K_MRQC_UDP_IPV4;
1112 if (test_bit(FM10K_FLAG_RSS_FIELD_IPV6_UDP, interface->flags))
1113 mrqc |= FM10K_MRQC_UDP_IPV6;
1115 fm10k_write_reg(hw, FM10K_MRQC(0), mrqc);
1117 /* configure default DGLORT mapping for RSS/DCB */
1118 dglort.inner_rss = 1;
1119 dglort.rss_l = fls(interface->ring_feature[RING_F_RSS].mask);
1120 dglort.pc_l = fls(interface->ring_feature[RING_F_QOS].mask);
1121 hw->mac.ops.configure_dglort_map(hw, &dglort);
1123 /* assign GLORT per queue for queue mapped testing */
1124 if (interface->glort_count > 64) {
1125 memset(&dglort, 0, sizeof(dglort));
1126 dglort.inner_rss = 1;
1127 dglort.glort = interface->glort + 64;
1128 dglort.idx = fm10k_dglort_pf_queue;
1129 dglort.queue_l = fls(interface->num_rx_queues - 1);
1130 hw->mac.ops.configure_dglort_map(hw, &dglort);
1133 /* assign glort value for RSS/DCB specific to this interface */
1134 memset(&dglort, 0, sizeof(dglort));
1135 dglort.inner_rss = 1;
1136 dglort.glort = interface->glort;
1137 dglort.rss_l = fls(interface->ring_feature[RING_F_RSS].mask);
1138 dglort.pc_l = fls(interface->ring_feature[RING_F_QOS].mask);
1139 /* configure DGLORT mapping for RSS/DCB */
1140 dglort.idx = fm10k_dglort_pf_rss;
1141 if (interface->l2_accel)
1142 dglort.shared_l = fls(interface->l2_accel->size);
1143 hw->mac.ops.configure_dglort_map(hw, &dglort);
1147 * fm10k_configure_rx - Configure Receive Unit after Reset
1148 * @interface: board private structure
1150 * Configure the Rx unit of the MAC after a reset.
1152 static void fm10k_configure_rx(struct fm10k_intfc *interface)
1156 /* Configure SWPRI to PC map */
1157 fm10k_configure_swpri_map(interface);
1159 /* Configure RSS and DGLORT map */
1160 fm10k_configure_dglort(interface);
1162 /* Setup the HW Rx Head and Tail descriptor pointers */
1163 for (i = 0; i < interface->num_rx_queues; i++)
1164 fm10k_configure_rx_ring(interface, interface->rx_ring[i]);
1166 /* possible poll here to verify that Rx rings are now enabled */
1169 static void fm10k_napi_enable_all(struct fm10k_intfc *interface)
1171 struct fm10k_q_vector *q_vector;
1174 for (q_idx = 0; q_idx < interface->num_q_vectors; q_idx++) {
1175 q_vector = interface->q_vector[q_idx];
1176 napi_enable(&q_vector->napi);
1180 static irqreturn_t fm10k_msix_clean_rings(int __always_unused irq, void *data)
1182 struct fm10k_q_vector *q_vector = data;
1184 if (q_vector->rx.count || q_vector->tx.count)
1185 napi_schedule_irqoff(&q_vector->napi);
1190 static irqreturn_t fm10k_msix_mbx_vf(int __always_unused irq, void *data)
1192 struct fm10k_intfc *interface = data;
1193 struct fm10k_hw *hw = &interface->hw;
1194 struct fm10k_mbx_info *mbx = &hw->mbx;
1196 /* re-enable mailbox interrupt and indicate 20us delay */
1197 fm10k_write_reg(hw, FM10K_VFITR(FM10K_MBX_VECTOR),
1198 (FM10K_MBX_INT_DELAY >> hw->mac.itr_scale) |
1201 /* service upstream mailbox */
1202 if (fm10k_mbx_trylock(interface)) {
1203 mbx->ops.process(hw, mbx);
1204 fm10k_mbx_unlock(interface);
1207 hw->mac.get_host_state = true;
1208 fm10k_service_event_schedule(interface);
1213 #ifdef CONFIG_NET_POLL_CONTROLLER
1215 * fm10k_netpoll - A Polling 'interrupt' handler
1216 * @netdev: network interface device structure
1218 * This is used by netconsole to send skbs without having to re-enable
1219 * interrupts. It's not called while the normal interrupt routine is executing.
1221 void fm10k_netpoll(struct net_device *netdev)
1223 struct fm10k_intfc *interface = netdev_priv(netdev);
1226 /* if interface is down do nothing */
1227 if (test_bit(__FM10K_DOWN, interface->state))
1230 for (i = 0; i < interface->num_q_vectors; i++)
1231 fm10k_msix_clean_rings(0, interface->q_vector[i]);
1235 #define FM10K_ERR_MSG(type) case (type): error = #type; break
1236 static void fm10k_handle_fault(struct fm10k_intfc *interface, int type,
1237 struct fm10k_fault *fault)
1239 struct pci_dev *pdev = interface->pdev;
1240 struct fm10k_hw *hw = &interface->hw;
1241 struct fm10k_iov_data *iov_data = interface->iov_data;
1245 case FM10K_PCA_FAULT:
1246 switch (fault->type) {
1248 error = "Unknown PCA error";
1250 FM10K_ERR_MSG(PCA_NO_FAULT);
1251 FM10K_ERR_MSG(PCA_UNMAPPED_ADDR);
1252 FM10K_ERR_MSG(PCA_BAD_QACCESS_PF);
1253 FM10K_ERR_MSG(PCA_BAD_QACCESS_VF);
1254 FM10K_ERR_MSG(PCA_MALICIOUS_REQ);
1255 FM10K_ERR_MSG(PCA_POISONED_TLP);
1256 FM10K_ERR_MSG(PCA_TLP_ABORT);
1259 case FM10K_THI_FAULT:
1260 switch (fault->type) {
1262 error = "Unknown THI error";
1264 FM10K_ERR_MSG(THI_NO_FAULT);
1265 FM10K_ERR_MSG(THI_MAL_DIS_Q_FAULT);
1268 case FM10K_FUM_FAULT:
1269 switch (fault->type) {
1271 error = "Unknown FUM error";
1273 FM10K_ERR_MSG(FUM_NO_FAULT);
1274 FM10K_ERR_MSG(FUM_UNMAPPED_ADDR);
1275 FM10K_ERR_MSG(FUM_BAD_VF_QACCESS);
1276 FM10K_ERR_MSG(FUM_ADD_DECODE_ERR);
1277 FM10K_ERR_MSG(FUM_RO_ERROR);
1278 FM10K_ERR_MSG(FUM_QPRC_CRC_ERROR);
1279 FM10K_ERR_MSG(FUM_CSR_TIMEOUT);
1280 FM10K_ERR_MSG(FUM_INVALID_TYPE);
1281 FM10K_ERR_MSG(FUM_INVALID_LENGTH);
1282 FM10K_ERR_MSG(FUM_INVALID_BE);
1283 FM10K_ERR_MSG(FUM_INVALID_ALIGN);
1287 error = "Undocumented fault";
1291 dev_warn(&pdev->dev,
1292 "%s Address: 0x%llx SpecInfo: 0x%x Func: %02x.%0x\n",
1293 error, fault->address, fault->specinfo,
1294 PCI_SLOT(fault->func), PCI_FUNC(fault->func));
1296 /* For VF faults, clear out the respective LPORT, reset the queue
1297 * resources, and then reconnect to the mailbox. This allows the
1298 * VF in question to resume behavior. For transient faults that are
1299 * the result of non-malicious behavior this will log the fault and
1300 * allow the VF to resume functionality. Obviously for malicious VFs
1301 * they will be able to attempt malicious behavior again. In this
1302 * case, the system administrator will need to step in and manually
1303 * remove or disable the VF in question.
1305 if (fault->func && iov_data) {
1306 int vf = fault->func - 1;
1307 struct fm10k_vf_info *vf_info = &iov_data->vf_info[vf];
1309 hw->iov.ops.reset_lport(hw, vf_info);
1310 hw->iov.ops.reset_resources(hw, vf_info);
1312 /* reset_lport disables the VF, so re-enable it */
1313 hw->iov.ops.set_lport(hw, vf_info, vf,
1314 FM10K_VF_FLAG_MULTI_CAPABLE);
1316 /* reset_resources will disconnect from the mbx */
1317 vf_info->mbx.ops.connect(hw, &vf_info->mbx);
1321 static void fm10k_report_fault(struct fm10k_intfc *interface, u32 eicr)
1323 struct fm10k_hw *hw = &interface->hw;
1324 struct fm10k_fault fault = { 0 };
1327 for (eicr &= FM10K_EICR_FAULT_MASK, type = FM10K_PCA_FAULT;
1329 eicr >>= 1, type += FM10K_FAULT_SIZE) {
1330 /* only check if there is an error reported */
1334 /* retrieve fault info */
1335 err = hw->mac.ops.get_fault(hw, type, &fault);
1337 dev_err(&interface->pdev->dev,
1338 "error reading fault\n");
1342 fm10k_handle_fault(interface, type, &fault);
1346 static void fm10k_reset_drop_on_empty(struct fm10k_intfc *interface, u32 eicr)
1348 struct fm10k_hw *hw = &interface->hw;
1349 const u32 rxdctl = FM10K_RXDCTL_WRITE_BACK_MIN_DELAY;
1353 if (!(eicr & FM10K_EICR_MAXHOLDTIME))
1356 maxholdq = fm10k_read_reg(hw, FM10K_MAXHOLDQ(7));
1358 fm10k_write_reg(hw, FM10K_MAXHOLDQ(7), maxholdq);
1360 if (maxholdq & BIT(31)) {
1361 if (q < FM10K_MAX_QUEUES_PF) {
1362 interface->rx_overrun_pf++;
1363 fm10k_write_reg(hw, FM10K_RXDCTL(q), rxdctl);
1365 interface->rx_overrun_vf++;
1379 maxholdq = fm10k_read_reg(hw, FM10K_MAXHOLDQ(q / 32));
1381 fm10k_write_reg(hw, FM10K_MAXHOLDQ(q / 32), maxholdq);
1385 static irqreturn_t fm10k_msix_mbx_pf(int __always_unused irq, void *data)
1387 struct fm10k_intfc *interface = data;
1388 struct fm10k_hw *hw = &interface->hw;
1389 struct fm10k_mbx_info *mbx = &hw->mbx;
1393 /* unmask any set bits related to this interrupt */
1394 eicr = fm10k_read_reg(hw, FM10K_EICR);
1395 fm10k_write_reg(hw, FM10K_EICR, eicr & (FM10K_EICR_MAILBOX |
1396 FM10K_EICR_SWITCHREADY |
1397 FM10K_EICR_SWITCHNOTREADY));
1399 /* report any faults found to the message log */
1400 fm10k_report_fault(interface, eicr);
1402 /* reset any queues disabled due to receiver overrun */
1403 fm10k_reset_drop_on_empty(interface, eicr);
1405 /* service mailboxes */
1406 if (fm10k_mbx_trylock(interface)) {
1407 err = mbx->ops.process(hw, mbx);
1408 /* handle VFLRE events */
1409 fm10k_iov_event(interface);
1410 fm10k_mbx_unlock(interface);
1413 if (err == FM10K_ERR_RESET_REQUESTED)
1414 set_bit(FM10K_FLAG_RESET_REQUESTED, interface->flags);
1416 /* if switch toggled state we should reset GLORTs */
1417 if (eicr & FM10K_EICR_SWITCHNOTREADY) {
1418 /* force link down for at least 4 seconds */
1419 interface->link_down_event = jiffies + (4 * HZ);
1420 set_bit(__FM10K_LINK_DOWN, interface->state);
1422 /* reset dglort_map back to no config */
1423 hw->mac.dglort_map = FM10K_DGLORTMAP_NONE;
1426 /* we should validate host state after interrupt event */
1427 hw->mac.get_host_state = true;
1429 /* validate host state, and handle VF mailboxes in the service task */
1430 fm10k_service_event_schedule(interface);
1432 /* re-enable mailbox interrupt and indicate 20us delay */
1433 fm10k_write_reg(hw, FM10K_ITR(FM10K_MBX_VECTOR),
1434 (FM10K_MBX_INT_DELAY >> hw->mac.itr_scale) |
1440 void fm10k_mbx_free_irq(struct fm10k_intfc *interface)
1442 struct fm10k_hw *hw = &interface->hw;
1443 struct msix_entry *entry;
1446 /* no mailbox IRQ to free if MSI-X is not enabled */
1447 if (!interface->msix_entries)
1450 entry = &interface->msix_entries[FM10K_MBX_VECTOR];
1452 /* disconnect the mailbox */
1453 hw->mbx.ops.disconnect(hw, &hw->mbx);
1455 /* disable Mailbox cause */
1456 if (hw->mac.type == fm10k_mac_pf) {
1457 fm10k_write_reg(hw, FM10K_EIMR,
1458 FM10K_EIMR_DISABLE(PCA_FAULT) |
1459 FM10K_EIMR_DISABLE(FUM_FAULT) |
1460 FM10K_EIMR_DISABLE(MAILBOX) |
1461 FM10K_EIMR_DISABLE(SWITCHREADY) |
1462 FM10K_EIMR_DISABLE(SWITCHNOTREADY) |
1463 FM10K_EIMR_DISABLE(SRAMERROR) |
1464 FM10K_EIMR_DISABLE(VFLR) |
1465 FM10K_EIMR_DISABLE(MAXHOLDTIME));
1466 itr_reg = FM10K_ITR(FM10K_MBX_VECTOR);
1468 itr_reg = FM10K_VFITR(FM10K_MBX_VECTOR);
1471 fm10k_write_reg(hw, itr_reg, FM10K_ITR_MASK_SET);
1473 free_irq(entry->vector, interface);
1476 static s32 fm10k_mbx_mac_addr(struct fm10k_hw *hw, u32 **results,
1477 struct fm10k_mbx_info *mbx)
1479 bool vlan_override = hw->mac.vlan_override;
1480 u16 default_vid = hw->mac.default_vid;
1481 struct fm10k_intfc *interface;
1484 err = fm10k_msg_mac_vlan_vf(hw, results, mbx);
1488 interface = container_of(hw, struct fm10k_intfc, hw);
1490 /* MAC was changed so we need reset */
1491 if (is_valid_ether_addr(hw->mac.perm_addr) &&
1492 !ether_addr_equal(hw->mac.perm_addr, hw->mac.addr))
1493 set_bit(FM10K_FLAG_RESET_REQUESTED, interface->flags);
1495 /* VLAN override was changed, or default VLAN changed */
1496 if ((vlan_override != hw->mac.vlan_override) ||
1497 (default_vid != hw->mac.default_vid))
1498 set_bit(FM10K_FLAG_RESET_REQUESTED, interface->flags);
1503 /* generic error handler for mailbox issues */
1504 static s32 fm10k_mbx_error(struct fm10k_hw *hw, u32 **results,
1505 struct fm10k_mbx_info __always_unused *mbx)
1507 struct fm10k_intfc *interface;
1508 struct pci_dev *pdev;
1510 interface = container_of(hw, struct fm10k_intfc, hw);
1511 pdev = interface->pdev;
1513 dev_err(&pdev->dev, "Unknown message ID %u\n",
1514 **results & FM10K_TLV_ID_MASK);
1519 static const struct fm10k_msg_data vf_mbx_data[] = {
1520 FM10K_TLV_MSG_TEST_HANDLER(fm10k_tlv_msg_test),
1521 FM10K_VF_MSG_MAC_VLAN_HANDLER(fm10k_mbx_mac_addr),
1522 FM10K_VF_MSG_LPORT_STATE_HANDLER(fm10k_msg_lport_state_vf),
1523 FM10K_TLV_MSG_ERROR_HANDLER(fm10k_mbx_error),
1526 static int fm10k_mbx_request_irq_vf(struct fm10k_intfc *interface)
1528 struct msix_entry *entry = &interface->msix_entries[FM10K_MBX_VECTOR];
1529 struct net_device *dev = interface->netdev;
1530 struct fm10k_hw *hw = &interface->hw;
1533 /* Use timer0 for interrupt moderation on the mailbox */
1534 u32 itr = entry->entry | FM10K_INT_MAP_TIMER0;
1536 /* register mailbox handlers */
1537 err = hw->mbx.ops.register_handlers(&hw->mbx, vf_mbx_data);
1541 /* request the IRQ */
1542 err = request_irq(entry->vector, fm10k_msix_mbx_vf, 0,
1543 dev->name, interface);
1545 netif_err(interface, probe, dev,
1546 "request_irq for msix_mbx failed: %d\n", err);
1550 /* map all of the interrupt sources */
1551 fm10k_write_reg(hw, FM10K_VFINT_MAP, itr);
1553 /* enable interrupt */
1554 fm10k_write_reg(hw, FM10K_VFITR(entry->entry), FM10K_ITR_ENABLE);
1559 static s32 fm10k_lport_map(struct fm10k_hw *hw, u32 **results,
1560 struct fm10k_mbx_info *mbx)
1562 struct fm10k_intfc *interface;
1563 u32 dglort_map = hw->mac.dglort_map;
1566 interface = container_of(hw, struct fm10k_intfc, hw);
1568 err = fm10k_msg_err_pf(hw, results, mbx);
1569 if (!err && hw->swapi.status) {
1570 /* force link down for a reasonable delay */
1571 interface->link_down_event = jiffies + (2 * HZ);
1572 set_bit(__FM10K_LINK_DOWN, interface->state);
1574 /* reset dglort_map back to no config */
1575 hw->mac.dglort_map = FM10K_DGLORTMAP_NONE;
1577 fm10k_service_event_schedule(interface);
1579 /* prevent overloading kernel message buffer */
1580 if (interface->lport_map_failed)
1583 interface->lport_map_failed = true;
1585 if (hw->swapi.status == FM10K_MSG_ERR_PEP_NOT_SCHEDULED)
1586 dev_warn(&interface->pdev->dev,
1587 "cannot obtain link because the host interface is configured for a PCIe host interface bandwidth of zero\n");
1588 dev_warn(&interface->pdev->dev,
1589 "request logical port map failed: %d\n",
1595 err = fm10k_msg_lport_map_pf(hw, results, mbx);
1599 interface->lport_map_failed = false;
1601 /* we need to reset if port count was just updated */
1602 if (dglort_map != hw->mac.dglort_map)
1603 set_bit(FM10K_FLAG_RESET_REQUESTED, interface->flags);
1608 static s32 fm10k_update_pvid(struct fm10k_hw *hw, u32 **results,
1609 struct fm10k_mbx_info __always_unused *mbx)
1611 struct fm10k_intfc *interface;
1616 err = fm10k_tlv_attr_get_u32(results[FM10K_PF_ATTR_ID_UPDATE_PVID],
1621 /* extract values from the pvid update */
1622 glort = FM10K_MSG_HDR_FIELD_GET(pvid_update, UPDATE_PVID_GLORT);
1623 pvid = FM10K_MSG_HDR_FIELD_GET(pvid_update, UPDATE_PVID_PVID);
1625 /* if glort is not valid return error */
1626 if (!fm10k_glort_valid_pf(hw, glort))
1627 return FM10K_ERR_PARAM;
1629 /* verify VLAN ID is valid */
1630 if (pvid >= FM10K_VLAN_TABLE_VID_MAX)
1631 return FM10K_ERR_PARAM;
1633 interface = container_of(hw, struct fm10k_intfc, hw);
1635 /* check to see if this belongs to one of the VFs */
1636 err = fm10k_iov_update_pvid(interface, glort, pvid);
1640 /* we need to reset if default VLAN was just updated */
1641 if (pvid != hw->mac.default_vid)
1642 set_bit(FM10K_FLAG_RESET_REQUESTED, interface->flags);
1644 hw->mac.default_vid = pvid;
1649 static const struct fm10k_msg_data pf_mbx_data[] = {
1650 FM10K_PF_MSG_ERR_HANDLER(XCAST_MODES, fm10k_msg_err_pf),
1651 FM10K_PF_MSG_ERR_HANDLER(UPDATE_MAC_FWD_RULE, fm10k_msg_err_pf),
1652 FM10K_PF_MSG_LPORT_MAP_HANDLER(fm10k_lport_map),
1653 FM10K_PF_MSG_ERR_HANDLER(LPORT_CREATE, fm10k_msg_err_pf),
1654 FM10K_PF_MSG_ERR_HANDLER(LPORT_DELETE, fm10k_msg_err_pf),
1655 FM10K_PF_MSG_UPDATE_PVID_HANDLER(fm10k_update_pvid),
1656 FM10K_TLV_MSG_ERROR_HANDLER(fm10k_mbx_error),
1659 static int fm10k_mbx_request_irq_pf(struct fm10k_intfc *interface)
1661 struct msix_entry *entry = &interface->msix_entries[FM10K_MBX_VECTOR];
1662 struct net_device *dev = interface->netdev;
1663 struct fm10k_hw *hw = &interface->hw;
1666 /* Use timer0 for interrupt moderation on the mailbox */
1667 u32 mbx_itr = entry->entry | FM10K_INT_MAP_TIMER0;
1668 u32 other_itr = entry->entry | FM10K_INT_MAP_IMMEDIATE;
1670 /* register mailbox handlers */
1671 err = hw->mbx.ops.register_handlers(&hw->mbx, pf_mbx_data);
1675 /* request the IRQ */
1676 err = request_irq(entry->vector, fm10k_msix_mbx_pf, 0,
1677 dev->name, interface);
1679 netif_err(interface, probe, dev,
1680 "request_irq for msix_mbx failed: %d\n", err);
1684 /* Enable interrupts w/ no moderation for "other" interrupts */
1685 fm10k_write_reg(hw, FM10K_INT_MAP(fm10k_int_pcie_fault), other_itr);
1686 fm10k_write_reg(hw, FM10K_INT_MAP(fm10k_int_switch_up_down), other_itr);
1687 fm10k_write_reg(hw, FM10K_INT_MAP(fm10k_int_sram), other_itr);
1688 fm10k_write_reg(hw, FM10K_INT_MAP(fm10k_int_max_hold_time), other_itr);
1689 fm10k_write_reg(hw, FM10K_INT_MAP(fm10k_int_vflr), other_itr);
1691 /* Enable interrupts w/ moderation for mailbox */
1692 fm10k_write_reg(hw, FM10K_INT_MAP(fm10k_int_mailbox), mbx_itr);
1694 /* Enable individual interrupt causes */
1695 fm10k_write_reg(hw, FM10K_EIMR, FM10K_EIMR_ENABLE(PCA_FAULT) |
1696 FM10K_EIMR_ENABLE(FUM_FAULT) |
1697 FM10K_EIMR_ENABLE(MAILBOX) |
1698 FM10K_EIMR_ENABLE(SWITCHREADY) |
1699 FM10K_EIMR_ENABLE(SWITCHNOTREADY) |
1700 FM10K_EIMR_ENABLE(SRAMERROR) |
1701 FM10K_EIMR_ENABLE(VFLR) |
1702 FM10K_EIMR_ENABLE(MAXHOLDTIME));
1704 /* enable interrupt */
1705 fm10k_write_reg(hw, FM10K_ITR(entry->entry), FM10K_ITR_ENABLE);
1710 int fm10k_mbx_request_irq(struct fm10k_intfc *interface)
1712 struct fm10k_hw *hw = &interface->hw;
1715 /* enable Mailbox cause */
1716 if (hw->mac.type == fm10k_mac_pf)
1717 err = fm10k_mbx_request_irq_pf(interface);
1719 err = fm10k_mbx_request_irq_vf(interface);
1723 /* connect mailbox */
1724 err = hw->mbx.ops.connect(hw, &hw->mbx);
1726 /* if the mailbox failed to connect, then free IRQ */
1728 fm10k_mbx_free_irq(interface);
1734 * fm10k_qv_free_irq - release interrupts associated with queue vectors
1735 * @interface: board private structure
1737 * Release all interrupts associated with this interface
1739 void fm10k_qv_free_irq(struct fm10k_intfc *interface)
1741 int vector = interface->num_q_vectors;
1742 struct fm10k_hw *hw = &interface->hw;
1743 struct msix_entry *entry;
1745 entry = &interface->msix_entries[NON_Q_VECTORS(hw) + vector];
1748 struct fm10k_q_vector *q_vector;
1752 q_vector = interface->q_vector[vector];
1754 if (!q_vector->tx.count && !q_vector->rx.count)
1757 /* clear the affinity_mask in the IRQ descriptor */
1758 irq_set_affinity_hint(entry->vector, NULL);
1760 /* disable interrupts */
1761 writel(FM10K_ITR_MASK_SET, q_vector->itr);
1763 free_irq(entry->vector, q_vector);
1768 * fm10k_qv_request_irq - initialize interrupts for queue vectors
1769 * @interface: board private structure
1771 * Attempts to configure interrupts using the best available
1772 * capabilities of the hardware and kernel.
1774 int fm10k_qv_request_irq(struct fm10k_intfc *interface)
1776 struct net_device *dev = interface->netdev;
1777 struct fm10k_hw *hw = &interface->hw;
1778 struct msix_entry *entry;
1779 unsigned int ri = 0, ti = 0;
1782 entry = &interface->msix_entries[NON_Q_VECTORS(hw)];
1784 for (vector = 0; vector < interface->num_q_vectors; vector++) {
1785 struct fm10k_q_vector *q_vector = interface->q_vector[vector];
1787 /* name the vector */
1788 if (q_vector->tx.count && q_vector->rx.count) {
1789 snprintf(q_vector->name, sizeof(q_vector->name),
1790 "%s-TxRx-%u", dev->name, ri++);
1792 } else if (q_vector->rx.count) {
1793 snprintf(q_vector->name, sizeof(q_vector->name),
1794 "%s-rx-%u", dev->name, ri++);
1795 } else if (q_vector->tx.count) {
1796 snprintf(q_vector->name, sizeof(q_vector->name),
1797 "%s-tx-%u", dev->name, ti++);
1799 /* skip this unused q_vector */
1803 /* Assign ITR register to q_vector */
1804 q_vector->itr = (hw->mac.type == fm10k_mac_pf) ?
1805 &interface->uc_addr[FM10K_ITR(entry->entry)] :
1806 &interface->uc_addr[FM10K_VFITR(entry->entry)];
1808 /* request the IRQ */
1809 err = request_irq(entry->vector, &fm10k_msix_clean_rings, 0,
1810 q_vector->name, q_vector);
1812 netif_err(interface, probe, dev,
1813 "request_irq failed for MSIX interrupt Error: %d\n",
1818 /* assign the mask for this irq */
1819 irq_set_affinity_hint(entry->vector, &q_vector->affinity_mask);
1821 /* Enable q_vector */
1822 writel(FM10K_ITR_ENABLE, q_vector->itr);
1830 /* wind through the ring freeing all entries and vectors */
1832 struct fm10k_q_vector *q_vector;
1836 q_vector = interface->q_vector[vector];
1838 if (!q_vector->tx.count && !q_vector->rx.count)
1841 /* clear the affinity_mask in the IRQ descriptor */
1842 irq_set_affinity_hint(entry->vector, NULL);
1844 /* disable interrupts */
1845 writel(FM10K_ITR_MASK_SET, q_vector->itr);
1847 free_irq(entry->vector, q_vector);
1853 void fm10k_up(struct fm10k_intfc *interface)
1855 struct fm10k_hw *hw = &interface->hw;
1857 /* Enable Tx/Rx DMA */
1858 hw->mac.ops.start_hw(hw);
1860 /* configure Tx descriptor rings */
1861 fm10k_configure_tx(interface);
1863 /* configure Rx descriptor rings */
1864 fm10k_configure_rx(interface);
1866 /* configure interrupts */
1867 hw->mac.ops.update_int_moderator(hw);
1869 /* enable statistics capture again */
1870 clear_bit(__FM10K_UPDATING_STATS, interface->state);
1872 /* clear down bit to indicate we are ready to go */
1873 clear_bit(__FM10K_DOWN, interface->state);
1875 /* enable polling cleanups */
1876 fm10k_napi_enable_all(interface);
1878 /* re-establish Rx filters */
1879 fm10k_restore_rx_state(interface);
1881 /* enable transmits */
1882 netif_tx_start_all_queues(interface->netdev);
1884 /* kick off the service timer now */
1885 hw->mac.get_host_state = true;
1886 mod_timer(&interface->service_timer, jiffies);
1889 static void fm10k_napi_disable_all(struct fm10k_intfc *interface)
1891 struct fm10k_q_vector *q_vector;
1894 for (q_idx = 0; q_idx < interface->num_q_vectors; q_idx++) {
1895 q_vector = interface->q_vector[q_idx];
1896 napi_disable(&q_vector->napi);
1900 void fm10k_down(struct fm10k_intfc *interface)
1902 struct net_device *netdev = interface->netdev;
1903 struct fm10k_hw *hw = &interface->hw;
1904 int err, i = 0, count = 0;
1906 /* signal that we are down to the interrupt handler and service task */
1907 if (test_and_set_bit(__FM10K_DOWN, interface->state))
1910 /* call carrier off first to avoid false dev_watchdog timeouts */
1911 netif_carrier_off(netdev);
1913 /* disable transmits */
1914 netif_tx_stop_all_queues(netdev);
1915 netif_tx_disable(netdev);
1917 /* reset Rx filters */
1918 fm10k_reset_rx_state(interface);
1920 /* disable polling routines */
1921 fm10k_napi_disable_all(interface);
1923 /* capture stats one last time before stopping interface */
1924 fm10k_update_stats(interface);
1926 /* prevent updating statistics while we're down */
1927 while (test_and_set_bit(__FM10K_UPDATING_STATS, interface->state))
1928 usleep_range(1000, 2000);
1930 /* skip waiting for TX DMA if we lost PCIe link */
1931 if (FM10K_REMOVED(hw->hw_addr))
1932 goto skip_tx_dma_drain;
1934 /* In some rare circumstances it can take a while for Tx queues to
1935 * quiesce and be fully disabled. Attempt to .stop_hw() first, and
1936 * then if we get ERR_REQUESTS_PENDING, go ahead and wait in a loop
1937 * until the Tx queues have emptied, or until a number of retries. If
1938 * we fail to clear within the retry loop, we will issue a warning
1939 * indicating that Tx DMA is probably hung. Note this means we call
1940 * .stop_hw() twice but this shouldn't cause any problems.
1942 err = hw->mac.ops.stop_hw(hw);
1943 if (err != FM10K_ERR_REQUESTS_PENDING)
1944 goto skip_tx_dma_drain;
1946 #define TX_DMA_DRAIN_RETRIES 25
1947 for (count = 0; count < TX_DMA_DRAIN_RETRIES; count++) {
1948 usleep_range(10000, 20000);
1950 /* start checking at the last ring to have pending Tx */
1951 for (; i < interface->num_tx_queues; i++)
1952 if (fm10k_get_tx_pending(interface->tx_ring[i], false))
1955 /* if all the queues are drained, we can break now */
1956 if (i == interface->num_tx_queues)
1960 if (count >= TX_DMA_DRAIN_RETRIES)
1961 dev_err(&interface->pdev->dev,
1962 "Tx queues failed to drain after %d tries. Tx DMA is probably hung.\n",
1965 /* Disable DMA engine for Tx/Rx */
1966 err = hw->mac.ops.stop_hw(hw);
1967 if (err == FM10K_ERR_REQUESTS_PENDING)
1968 dev_err(&interface->pdev->dev,
1969 "due to pending requests hw was not shut down gracefully\n");
1971 dev_err(&interface->pdev->dev, "stop_hw failed: %d\n", err);
1973 /* free any buffers still on the rings */
1974 fm10k_clean_all_tx_rings(interface);
1975 fm10k_clean_all_rx_rings(interface);
1979 * fm10k_sw_init - Initialize general software structures
1980 * @interface: host interface private structure to initialize
1981 * @ent: PCI device ID entry
1983 * fm10k_sw_init initializes the interface private data structure.
1984 * Fields are initialized based on PCI device information and
1985 * OS network device settings (MTU size).
1987 static int fm10k_sw_init(struct fm10k_intfc *interface,
1988 const struct pci_device_id *ent)
1990 const struct fm10k_info *fi = fm10k_info_tbl[ent->driver_data];
1991 struct fm10k_hw *hw = &interface->hw;
1992 struct pci_dev *pdev = interface->pdev;
1993 struct net_device *netdev = interface->netdev;
1994 u32 rss_key[FM10K_RSSRK_SIZE];
1998 /* initialize back pointer */
1999 hw->back = interface;
2000 hw->hw_addr = interface->uc_addr;
2002 /* PCI config space info */
2003 hw->vendor_id = pdev->vendor;
2004 hw->device_id = pdev->device;
2005 hw->revision_id = pdev->revision;
2006 hw->subsystem_vendor_id = pdev->subsystem_vendor;
2007 hw->subsystem_device_id = pdev->subsystem_device;
2010 memcpy(&hw->mac.ops, fi->mac_ops, sizeof(hw->mac.ops));
2011 hw->mac.type = fi->mac;
2013 /* Setup IOV handlers */
2015 memcpy(&hw->iov.ops, fi->iov_ops, sizeof(hw->iov.ops));
2017 /* Set common capability flags and settings */
2018 rss = min_t(int, FM10K_MAX_RSS_INDICES, num_online_cpus());
2019 interface->ring_feature[RING_F_RSS].limit = rss;
2020 fi->get_invariants(hw);
2022 /* pick up the PCIe bus settings for reporting later */
2023 if (hw->mac.ops.get_bus_info)
2024 hw->mac.ops.get_bus_info(hw);
2026 /* limit the usable DMA range */
2027 if (hw->mac.ops.set_dma_mask)
2028 hw->mac.ops.set_dma_mask(hw, dma_get_mask(&pdev->dev));
2030 /* update netdev with DMA restrictions */
2031 if (dma_get_mask(&pdev->dev) > DMA_BIT_MASK(32)) {
2032 netdev->features |= NETIF_F_HIGHDMA;
2033 netdev->vlan_features |= NETIF_F_HIGHDMA;
2036 /* reset and initialize the hardware so it is in a known state */
2037 err = hw->mac.ops.reset_hw(hw);
2039 dev_err(&pdev->dev, "reset_hw failed: %d\n", err);
2043 err = hw->mac.ops.init_hw(hw);
2045 dev_err(&pdev->dev, "init_hw failed: %d\n", err);
2049 /* initialize hardware statistics */
2050 hw->mac.ops.update_hw_stats(hw, &interface->stats);
2052 /* Set upper limit on IOV VFs that can be allocated */
2053 pci_sriov_set_totalvfs(pdev, hw->iov.total_vfs);
2055 /* Start with random Ethernet address */
2056 eth_random_addr(hw->mac.addr);
2058 /* Initialize MAC address from hardware */
2059 err = hw->mac.ops.read_mac_addr(hw);
2061 dev_warn(&pdev->dev,
2062 "Failed to obtain MAC address defaulting to random\n");
2063 /* tag address assignment as random */
2064 netdev->addr_assign_type |= NET_ADDR_RANDOM;
2067 ether_addr_copy(netdev->dev_addr, hw->mac.addr);
2068 ether_addr_copy(netdev->perm_addr, hw->mac.addr);
2070 if (!is_valid_ether_addr(netdev->perm_addr)) {
2071 dev_err(&pdev->dev, "Invalid MAC Address\n");
2075 /* initialize DCBNL interface */
2076 fm10k_dcbnl_set_ops(netdev);
2078 /* set default ring sizes */
2079 interface->tx_ring_count = FM10K_DEFAULT_TXD;
2080 interface->rx_ring_count = FM10K_DEFAULT_RXD;
2082 /* set default interrupt moderation */
2083 interface->tx_itr = FM10K_TX_ITR_DEFAULT;
2084 interface->rx_itr = FM10K_ITR_ADAPTIVE | FM10K_RX_ITR_DEFAULT;
2086 /* initialize udp port lists */
2087 INIT_LIST_HEAD(&interface->vxlan_port);
2088 INIT_LIST_HEAD(&interface->geneve_port);
2090 /* Initialize the MAC/VLAN queue */
2091 INIT_LIST_HEAD(&interface->macvlan_requests);
2093 netdev_rss_key_fill(rss_key, sizeof(rss_key));
2094 memcpy(interface->rssrk, rss_key, sizeof(rss_key));
2096 /* Initialize the mailbox lock */
2097 spin_lock_init(&interface->mbx_lock);
2098 spin_lock_init(&interface->macvlan_lock);
2100 /* Start off interface as being down */
2101 set_bit(__FM10K_DOWN, interface->state);
2102 set_bit(__FM10K_UPDATING_STATS, interface->state);
2108 * fm10k_probe - Device Initialization Routine
2109 * @pdev: PCI device information struct
2110 * @ent: entry in fm10k_pci_tbl
2112 * Returns 0 on success, negative on failure
2114 * fm10k_probe initializes an interface identified by a pci_dev structure.
2115 * The OS initialization, configuring of the interface private structure,
2116 * and a hardware reset occur.
2118 static int fm10k_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
2120 struct net_device *netdev;
2121 struct fm10k_intfc *interface;
2124 if (pdev->error_state != pci_channel_io_normal) {
2126 "PCI device still in an error state. Unable to load...\n");
2130 err = pci_enable_device_mem(pdev);
2133 "PCI enable device failed: %d\n", err);
2137 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(48));
2139 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
2142 "DMA configuration failed: %d\n", err);
2146 err = pci_request_mem_regions(pdev, fm10k_driver_name);
2149 "pci_request_selected_regions failed: %d\n", err);
2153 pci_enable_pcie_error_reporting(pdev);
2155 pci_set_master(pdev);
2156 pci_save_state(pdev);
2158 netdev = fm10k_alloc_netdev(fm10k_info_tbl[ent->driver_data]);
2161 goto err_alloc_netdev;
2164 SET_NETDEV_DEV(netdev, &pdev->dev);
2166 interface = netdev_priv(netdev);
2167 pci_set_drvdata(pdev, interface);
2169 interface->netdev = netdev;
2170 interface->pdev = pdev;
2172 interface->uc_addr = ioremap(pci_resource_start(pdev, 0),
2173 FM10K_UC_ADDR_SIZE);
2174 if (!interface->uc_addr) {
2179 err = fm10k_sw_init(interface, ent);
2183 /* enable debugfs support */
2184 fm10k_dbg_intfc_init(interface);
2186 err = fm10k_init_queueing_scheme(interface);
2190 /* the mbx interrupt might attempt to schedule the service task, so we
2191 * must ensure it is disabled since we haven't yet requested the timer
2194 set_bit(__FM10K_SERVICE_DISABLE, interface->state);
2196 err = fm10k_mbx_request_irq(interface);
2198 goto err_mbx_interrupt;
2200 /* final check of hardware state before registering the interface */
2201 err = fm10k_hw_ready(interface);
2205 err = register_netdev(netdev);
2209 /* carrier off reporting is important to ethtool even BEFORE open */
2210 netif_carrier_off(netdev);
2212 /* stop all the transmit queues from transmitting until link is up */
2213 netif_tx_stop_all_queues(netdev);
2215 /* Initialize service timer and service task late in order to avoid
2218 timer_setup(&interface->service_timer, fm10k_service_timer, 0);
2219 INIT_WORK(&interface->service_task, fm10k_service_task);
2221 /* Setup the MAC/VLAN queue */
2222 INIT_DELAYED_WORK(&interface->macvlan_task, fm10k_macvlan_task);
2224 /* kick off service timer now, even when interface is down */
2225 mod_timer(&interface->service_timer, (HZ * 2) + jiffies);
2227 /* print warning for non-optimal configurations */
2228 pcie_print_link_status(interface->pdev);
2230 /* report MAC address for logging */
2231 dev_info(&pdev->dev, "%pM\n", netdev->dev_addr);
2233 /* enable SR-IOV after registering netdev to enforce PF/VF ordering */
2234 fm10k_iov_configure(pdev, 0);
2236 /* clear the service task disable bit and kick off service task */
2237 clear_bit(__FM10K_SERVICE_DISABLE, interface->state);
2238 fm10k_service_event_schedule(interface);
2243 fm10k_mbx_free_irq(interface);
2245 fm10k_clear_queueing_scheme(interface);
2247 if (interface->sw_addr)
2248 iounmap(interface->sw_addr);
2249 iounmap(interface->uc_addr);
2251 free_netdev(netdev);
2253 pci_release_mem_regions(pdev);
2256 pci_disable_device(pdev);
2261 * fm10k_remove - Device Removal Routine
2262 * @pdev: PCI device information struct
2264 * fm10k_remove is called by the PCI subsystem to alert the driver
2265 * that it should release a PCI device. The could be caused by a
2266 * Hot-Plug event, or because the driver is going to be removed from
2269 static void fm10k_remove(struct pci_dev *pdev)
2271 struct fm10k_intfc *interface = pci_get_drvdata(pdev);
2272 struct net_device *netdev = interface->netdev;
2274 del_timer_sync(&interface->service_timer);
2276 fm10k_stop_service_event(interface);
2277 fm10k_stop_macvlan_task(interface);
2279 /* Remove all pending MAC/VLAN requests */
2280 fm10k_clear_macvlan_queue(interface, interface->glort, true);
2282 /* free netdev, this may bounce the interrupts due to setup_tc */
2283 if (netdev->reg_state == NETREG_REGISTERED)
2284 unregister_netdev(netdev);
2287 fm10k_iov_disable(pdev);
2289 /* disable mailbox interrupt */
2290 fm10k_mbx_free_irq(interface);
2292 /* free interrupts */
2293 fm10k_clear_queueing_scheme(interface);
2295 /* remove any debugfs interfaces */
2296 fm10k_dbg_intfc_exit(interface);
2298 if (interface->sw_addr)
2299 iounmap(interface->sw_addr);
2300 iounmap(interface->uc_addr);
2302 free_netdev(netdev);
2304 pci_release_mem_regions(pdev);
2306 pci_disable_pcie_error_reporting(pdev);
2308 pci_disable_device(pdev);
2311 static void fm10k_prepare_suspend(struct fm10k_intfc *interface)
2313 /* the watchdog task reads from registers, which might appear like
2314 * a surprise remove if the PCIe device is disabled while we're
2315 * stopped. We stop the watchdog task until after we resume software
2318 * Note that the MAC/VLAN task will be stopped as part of preparing
2319 * for reset so we don't need to handle it here.
2321 fm10k_stop_service_event(interface);
2323 if (fm10k_prepare_for_reset(interface))
2324 set_bit(__FM10K_RESET_SUSPENDED, interface->state);
2327 static int fm10k_handle_resume(struct fm10k_intfc *interface)
2329 struct fm10k_hw *hw = &interface->hw;
2332 /* Even if we didn't properly prepare for reset in
2333 * fm10k_prepare_suspend, we'll attempt to resume anyways.
2335 if (!test_and_clear_bit(__FM10K_RESET_SUSPENDED, interface->state))
2336 dev_warn(&interface->pdev->dev,
2337 "Device was shut down as part of suspend... Attempting to recover\n");
2339 /* reset statistics starting values */
2340 hw->mac.ops.rebind_hw_stats(hw, &interface->stats);
2342 err = fm10k_handle_reset(interface);
2346 /* assume host is not ready, to prevent race with watchdog in case we
2347 * actually don't have connection to the switch
2349 interface->host_ready = false;
2350 fm10k_watchdog_host_not_ready(interface);
2352 /* force link to stay down for a second to prevent link flutter */
2353 interface->link_down_event = jiffies + (HZ);
2354 set_bit(__FM10K_LINK_DOWN, interface->state);
2356 /* restart the service task */
2357 fm10k_start_service_event(interface);
2359 /* Restart the MAC/VLAN request queue in-case of outstanding events */
2360 fm10k_macvlan_schedule(interface);
2366 * fm10k_resume - Generic PM resume hook
2367 * @dev: generic device structure
2369 * Generic PM hook used when waking the device from a low power state after
2370 * suspend or hibernation. This function does not need to handle lower PCIe
2371 * device state as the stack takes care of that for us.
2373 static int __maybe_unused fm10k_resume(struct device *dev)
2375 struct fm10k_intfc *interface = pci_get_drvdata(to_pci_dev(dev));
2376 struct net_device *netdev = interface->netdev;
2377 struct fm10k_hw *hw = &interface->hw;
2380 /* refresh hw_addr in case it was dropped */
2381 hw->hw_addr = interface->uc_addr;
2383 err = fm10k_handle_resume(interface);
2387 netif_device_attach(netdev);
2393 * fm10k_suspend - Generic PM suspend hook
2394 * @dev: generic device structure
2396 * Generic PM hook used when setting the device into a low power state for
2397 * system suspend or hibernation. This function does not need to handle lower
2398 * PCIe device state as the stack takes care of that for us.
2400 static int __maybe_unused fm10k_suspend(struct device *dev)
2402 struct fm10k_intfc *interface = pci_get_drvdata(to_pci_dev(dev));
2403 struct net_device *netdev = interface->netdev;
2405 netif_device_detach(netdev);
2407 fm10k_prepare_suspend(interface);
2413 * fm10k_io_error_detected - called when PCI error is detected
2414 * @pdev: Pointer to PCI device
2415 * @state: The current pci connection state
2417 * This function is called after a PCI bus error affecting
2418 * this device has been detected.
2420 static pci_ers_result_t fm10k_io_error_detected(struct pci_dev *pdev,
2421 pci_channel_state_t state)
2423 struct fm10k_intfc *interface = pci_get_drvdata(pdev);
2424 struct net_device *netdev = interface->netdev;
2426 netif_device_detach(netdev);
2428 if (state == pci_channel_io_perm_failure)
2429 return PCI_ERS_RESULT_DISCONNECT;
2431 fm10k_prepare_suspend(interface);
2433 /* Request a slot reset. */
2434 return PCI_ERS_RESULT_NEED_RESET;
2438 * fm10k_io_slot_reset - called after the pci bus has been reset.
2439 * @pdev: Pointer to PCI device
2441 * Restart the card from scratch, as if from a cold-boot.
2443 static pci_ers_result_t fm10k_io_slot_reset(struct pci_dev *pdev)
2445 pci_ers_result_t result;
2447 if (pci_reenable_device(pdev)) {
2449 "Cannot re-enable PCI device after reset.\n");
2450 result = PCI_ERS_RESULT_DISCONNECT;
2452 pci_set_master(pdev);
2453 pci_restore_state(pdev);
2455 /* After second error pci->state_saved is false, this
2456 * resets it so EEH doesn't break.
2458 pci_save_state(pdev);
2460 pci_wake_from_d3(pdev, false);
2462 result = PCI_ERS_RESULT_RECOVERED;
2465 pci_cleanup_aer_uncorrect_error_status(pdev);
2471 * fm10k_io_resume - called when traffic can start flowing again.
2472 * @pdev: Pointer to PCI device
2474 * This callback is called when the error recovery driver tells us that
2475 * its OK to resume normal operation.
2477 static void fm10k_io_resume(struct pci_dev *pdev)
2479 struct fm10k_intfc *interface = pci_get_drvdata(pdev);
2480 struct net_device *netdev = interface->netdev;
2483 err = fm10k_handle_resume(interface);
2486 dev_warn(&pdev->dev,
2487 "%s failed: %d\n", __func__, err);
2489 netif_device_attach(netdev);
2493 * fm10k_io_reset_prepare - called when PCI function is about to be reset
2494 * @pdev: Pointer to PCI device
2496 * This callback is called when the PCI function is about to be reset,
2497 * allowing the device driver to prepare for it.
2499 static void fm10k_io_reset_prepare(struct pci_dev *pdev)
2501 /* warn incase we have any active VF devices */
2502 if (pci_num_vf(pdev))
2503 dev_warn(&pdev->dev,
2504 "PCIe FLR may cause issues for any active VF devices\n");
2505 fm10k_prepare_suspend(pci_get_drvdata(pdev));
2509 * fm10k_io_reset_done - called when PCI function has finished resetting
2510 * @pdev: Pointer to PCI device
2512 * This callback is called just after the PCI function is reset, such as via
2513 * /sys/class/net/<enpX>/device/reset or similar.
2515 static void fm10k_io_reset_done(struct pci_dev *pdev)
2517 struct fm10k_intfc *interface = pci_get_drvdata(pdev);
2518 int err = fm10k_handle_resume(interface);
2521 dev_warn(&pdev->dev,
2522 "%s failed: %d\n", __func__, err);
2523 netif_device_detach(interface->netdev);
2527 static const struct pci_error_handlers fm10k_err_handler = {
2528 .error_detected = fm10k_io_error_detected,
2529 .slot_reset = fm10k_io_slot_reset,
2530 .resume = fm10k_io_resume,
2531 .reset_prepare = fm10k_io_reset_prepare,
2532 .reset_done = fm10k_io_reset_done,
2535 static SIMPLE_DEV_PM_OPS(fm10k_pm_ops, fm10k_suspend, fm10k_resume);
2537 static struct pci_driver fm10k_driver = {
2538 .name = fm10k_driver_name,
2539 .id_table = fm10k_pci_tbl,
2540 .probe = fm10k_probe,
2541 .remove = fm10k_remove,
2543 .pm = &fm10k_pm_ops,
2545 .sriov_configure = fm10k_iov_configure,
2546 .err_handler = &fm10k_err_handler
2550 * fm10k_register_pci_driver - register driver interface
2552 * This function is called on module load in order to register the driver.
2554 int fm10k_register_pci_driver(void)
2556 return pci_register_driver(&fm10k_driver);
2560 * fm10k_unregister_pci_driver - unregister driver interface
2562 * This function is called on module unload in order to remove the driver.
2564 void fm10k_unregister_pci_driver(void)
2566 pci_unregister_driver(&fm10k_driver);