1 /*******************************************************************************
3 Intel(R) Gigabit Ethernet Linux driver
4 Copyright(c) 2007 Intel Corporation.
6 This program is free software; you can redistribute it and/or modify it
7 under the terms and conditions of the GNU General Public License,
8 version 2, as published by the Free Software Foundation.
10 This program is distributed in the hope it will be useful, but WITHOUT
11 ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
15 You should have received a copy of the GNU General Public License along with
16 this program; if not, write to the Free Software Foundation, Inc.,
17 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
19 The full GNU General Public License is included in this distribution in
20 the file called "COPYING".
23 e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
24 Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
26 *******************************************************************************/
28 #include <linux/module.h>
29 #include <linux/types.h>
30 #include <linux/init.h>
31 #include <linux/vmalloc.h>
32 #include <linux/pagemap.h>
33 #include <linux/netdevice.h>
34 #include <linux/ipv6.h>
35 #include <net/checksum.h>
36 #include <net/ip6_checksum.h>
37 #include <linux/mii.h>
38 #include <linux/ethtool.h>
39 #include <linux/if_vlan.h>
40 #include <linux/pci.h>
41 #include <linux/delay.h>
42 #include <linux/interrupt.h>
43 #include <linux/if_ether.h>
47 #define DRV_VERSION "1.0.8-k2"
48 char igb_driver_name[] = "igb";
49 char igb_driver_version[] = DRV_VERSION;
50 static const char igb_driver_string[] =
51 "Intel(R) Gigabit Ethernet Network Driver";
52 static const char igb_copyright[] = "Copyright (c) 2007 Intel Corporation.";
55 static const struct e1000_info *igb_info_tbl[] = {
56 [board_82575] = &e1000_82575_info,
59 static struct pci_device_id igb_pci_tbl[] = {
60 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575EB_COPPER), board_82575 },
61 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575EB_FIBER_SERDES), board_82575 },
62 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575GB_QUAD_COPPER), board_82575 },
63 /* required last entry */
67 MODULE_DEVICE_TABLE(pci, igb_pci_tbl);
69 void igb_reset(struct igb_adapter *);
70 static int igb_setup_all_tx_resources(struct igb_adapter *);
71 static int igb_setup_all_rx_resources(struct igb_adapter *);
72 static void igb_free_all_tx_resources(struct igb_adapter *);
73 static void igb_free_all_rx_resources(struct igb_adapter *);
74 static void igb_free_tx_resources(struct igb_adapter *, struct igb_ring *);
75 static void igb_free_rx_resources(struct igb_adapter *, struct igb_ring *);
76 void igb_update_stats(struct igb_adapter *);
77 static int igb_probe(struct pci_dev *, const struct pci_device_id *);
78 static void __devexit igb_remove(struct pci_dev *pdev);
79 static int igb_sw_init(struct igb_adapter *);
80 static int igb_open(struct net_device *);
81 static int igb_close(struct net_device *);
82 static void igb_configure_tx(struct igb_adapter *);
83 static void igb_configure_rx(struct igb_adapter *);
84 static void igb_setup_rctl(struct igb_adapter *);
85 static void igb_clean_all_tx_rings(struct igb_adapter *);
86 static void igb_clean_all_rx_rings(struct igb_adapter *);
87 static void igb_clean_tx_ring(struct igb_adapter *, struct igb_ring *);
88 static void igb_clean_rx_ring(struct igb_adapter *, struct igb_ring *);
89 static void igb_set_multi(struct net_device *);
90 static void igb_update_phy_info(unsigned long);
91 static void igb_watchdog(unsigned long);
92 static void igb_watchdog_task(struct work_struct *);
93 static int igb_xmit_frame_ring_adv(struct sk_buff *, struct net_device *,
95 static int igb_xmit_frame_adv(struct sk_buff *skb, struct net_device *);
96 static struct net_device_stats *igb_get_stats(struct net_device *);
97 static int igb_change_mtu(struct net_device *, int);
98 static int igb_set_mac(struct net_device *, void *);
99 static irqreturn_t igb_intr(int irq, void *);
100 static irqreturn_t igb_intr_msi(int irq, void *);
101 static irqreturn_t igb_msix_other(int irq, void *);
102 static irqreturn_t igb_msix_rx(int irq, void *);
103 static irqreturn_t igb_msix_tx(int irq, void *);
104 static int igb_clean_rx_ring_msix(struct napi_struct *, int);
105 static bool igb_clean_tx_irq(struct igb_adapter *, struct igb_ring *);
106 static int igb_clean(struct napi_struct *, int);
107 static bool igb_clean_rx_irq_adv(struct igb_adapter *,
108 struct igb_ring *, int *, int);
109 static void igb_alloc_rx_buffers_adv(struct igb_adapter *,
110 struct igb_ring *, int);
111 static int igb_ioctl(struct net_device *, struct ifreq *, int cmd);
112 static void igb_tx_timeout(struct net_device *);
113 static void igb_reset_task(struct work_struct *);
114 static void igb_vlan_rx_register(struct net_device *, struct vlan_group *);
115 static void igb_vlan_rx_add_vid(struct net_device *, u16);
116 static void igb_vlan_rx_kill_vid(struct net_device *, u16);
117 static void igb_restore_vlan(struct igb_adapter *);
119 static int igb_suspend(struct pci_dev *, pm_message_t);
121 static int igb_resume(struct pci_dev *);
123 static void igb_shutdown(struct pci_dev *);
125 #ifdef CONFIG_NET_POLL_CONTROLLER
126 /* for netdump / net console */
127 static void igb_netpoll(struct net_device *);
130 static pci_ers_result_t igb_io_error_detected(struct pci_dev *,
131 pci_channel_state_t);
132 static pci_ers_result_t igb_io_slot_reset(struct pci_dev *);
133 static void igb_io_resume(struct pci_dev *);
135 static struct pci_error_handlers igb_err_handler = {
136 .error_detected = igb_io_error_detected,
137 .slot_reset = igb_io_slot_reset,
138 .resume = igb_io_resume,
142 static struct pci_driver igb_driver = {
143 .name = igb_driver_name,
144 .id_table = igb_pci_tbl,
146 .remove = __devexit_p(igb_remove),
148 /* Power Managment Hooks */
149 .suspend = igb_suspend,
150 .resume = igb_resume,
152 .shutdown = igb_shutdown,
153 .err_handler = &igb_err_handler
156 MODULE_AUTHOR("Intel Corporation, <e1000-devel@lists.sourceforge.net>");
157 MODULE_DESCRIPTION("Intel(R) Gigabit Ethernet Network Driver");
158 MODULE_LICENSE("GPL");
159 MODULE_VERSION(DRV_VERSION);
163 * igb_get_hw_dev_name - return device name string
164 * used by hardware layer to print debugging information
166 char *igb_get_hw_dev_name(struct e1000_hw *hw)
168 struct igb_adapter *adapter = hw->back;
169 return adapter->netdev->name;
174 * igb_init_module - Driver Registration Routine
176 * igb_init_module is the first routine called when the driver is
177 * loaded. All it does is register with the PCI subsystem.
179 static int __init igb_init_module(void)
182 printk(KERN_INFO "%s - version %s\n",
183 igb_driver_string, igb_driver_version);
185 printk(KERN_INFO "%s\n", igb_copyright);
187 ret = pci_register_driver(&igb_driver);
191 module_init(igb_init_module);
194 * igb_exit_module - Driver Exit Cleanup Routine
196 * igb_exit_module is called just before the driver is removed
199 static void __exit igb_exit_module(void)
201 pci_unregister_driver(&igb_driver);
204 module_exit(igb_exit_module);
207 * igb_alloc_queues - Allocate memory for all rings
208 * @adapter: board private structure to initialize
210 * We allocate one ring per queue at run-time since we don't know the
211 * number of queues at compile-time.
213 static int igb_alloc_queues(struct igb_adapter *adapter)
217 adapter->tx_ring = kcalloc(adapter->num_tx_queues,
218 sizeof(struct igb_ring), GFP_KERNEL);
219 if (!adapter->tx_ring)
222 adapter->rx_ring = kcalloc(adapter->num_rx_queues,
223 sizeof(struct igb_ring), GFP_KERNEL);
224 if (!adapter->rx_ring) {
225 kfree(adapter->tx_ring);
229 for (i = 0; i < adapter->num_rx_queues; i++) {
230 struct igb_ring *ring = &(adapter->rx_ring[i]);
231 ring->adapter = adapter;
232 ring->itr_register = E1000_ITR;
234 if (!ring->napi.poll)
235 netif_napi_add(adapter->netdev, &ring->napi, igb_clean,
236 adapter->napi.weight /
237 adapter->num_rx_queues);
242 #define IGB_N0_QUEUE -1
243 static void igb_assign_vector(struct igb_adapter *adapter, int rx_queue,
244 int tx_queue, int msix_vector)
247 struct e1000_hw *hw = &adapter->hw;
248 /* The 82575 assigns vectors using a bitmask, which matches the
249 bitmask for the EICR/EIMS/EIMC registers. To assign one
250 or more queues to a vector, we write the appropriate bits
251 into the MSIXBM register for that vector. */
252 if (rx_queue > IGB_N0_QUEUE) {
253 msixbm = E1000_EICR_RX_QUEUE0 << rx_queue;
254 adapter->rx_ring[rx_queue].eims_value = msixbm;
256 if (tx_queue > IGB_N0_QUEUE) {
257 msixbm |= E1000_EICR_TX_QUEUE0 << tx_queue;
258 adapter->tx_ring[tx_queue].eims_value =
259 E1000_EICR_TX_QUEUE0 << tx_queue;
261 array_wr32(E1000_MSIXBM(0), msix_vector, msixbm);
265 * igb_configure_msix - Configure MSI-X hardware
267 * igb_configure_msix sets up the hardware to properly
268 * generate MSI-X interrupts.
270 static void igb_configure_msix(struct igb_adapter *adapter)
274 struct e1000_hw *hw = &adapter->hw;
276 adapter->eims_enable_mask = 0;
278 for (i = 0; i < adapter->num_tx_queues; i++) {
279 struct igb_ring *tx_ring = &adapter->tx_ring[i];
280 igb_assign_vector(adapter, IGB_N0_QUEUE, i, vector++);
281 adapter->eims_enable_mask |= tx_ring->eims_value;
282 if (tx_ring->itr_val)
283 writel(1000000000 / (tx_ring->itr_val * 256),
284 hw->hw_addr + tx_ring->itr_register);
286 writel(1, hw->hw_addr + tx_ring->itr_register);
289 for (i = 0; i < adapter->num_rx_queues; i++) {
290 struct igb_ring *rx_ring = &adapter->rx_ring[i];
291 igb_assign_vector(adapter, i, IGB_N0_QUEUE, vector++);
292 adapter->eims_enable_mask |= rx_ring->eims_value;
293 if (rx_ring->itr_val)
294 writel(1000000000 / (rx_ring->itr_val * 256),
295 hw->hw_addr + rx_ring->itr_register);
297 writel(1, hw->hw_addr + rx_ring->itr_register);
301 /* set vector for other causes, i.e. link changes */
302 array_wr32(E1000_MSIXBM(0), vector++,
305 /* disable IAM for ICR interrupt bits */
308 tmp = rd32(E1000_CTRL_EXT);
309 /* enable MSI-X PBA support*/
310 tmp |= E1000_CTRL_EXT_PBA_CLR;
312 /* Auto-Mask interrupts upon ICR read. */
313 tmp |= E1000_CTRL_EXT_EIAME;
314 tmp |= E1000_CTRL_EXT_IRCA;
316 wr32(E1000_CTRL_EXT, tmp);
317 adapter->eims_enable_mask |= E1000_EIMS_OTHER;
323 * igb_request_msix - Initialize MSI-X interrupts
325 * igb_request_msix allocates MSI-X vectors and requests interrupts from the
328 static int igb_request_msix(struct igb_adapter *adapter)
330 struct net_device *netdev = adapter->netdev;
331 int i, err = 0, vector = 0;
335 for (i = 0; i < adapter->num_tx_queues; i++) {
336 struct igb_ring *ring = &(adapter->tx_ring[i]);
337 sprintf(ring->name, "%s-tx%d", netdev->name, i);
338 err = request_irq(adapter->msix_entries[vector].vector,
339 &igb_msix_tx, 0, ring->name,
340 &(adapter->tx_ring[i]));
343 ring->itr_register = E1000_EITR(0) + (vector << 2);
344 ring->itr_val = adapter->itr;
347 for (i = 0; i < adapter->num_rx_queues; i++) {
348 struct igb_ring *ring = &(adapter->rx_ring[i]);
349 if (strlen(netdev->name) < (IFNAMSIZ - 5))
350 sprintf(ring->name, "%s-rx%d", netdev->name, i);
352 memcpy(ring->name, netdev->name, IFNAMSIZ);
353 err = request_irq(adapter->msix_entries[vector].vector,
354 &igb_msix_rx, 0, ring->name,
355 &(adapter->rx_ring[i]));
358 ring->itr_register = E1000_EITR(0) + (vector << 2);
359 ring->itr_val = adapter->itr;
363 err = request_irq(adapter->msix_entries[vector].vector,
364 &igb_msix_other, 0, netdev->name, netdev);
368 adapter->napi.poll = igb_clean_rx_ring_msix;
369 for (i = 0; i < adapter->num_rx_queues; i++)
370 adapter->rx_ring[i].napi.poll = adapter->napi.poll;
371 igb_configure_msix(adapter);
377 static void igb_reset_interrupt_capability(struct igb_adapter *adapter)
379 if (adapter->msix_entries) {
380 pci_disable_msix(adapter->pdev);
381 kfree(adapter->msix_entries);
382 adapter->msix_entries = NULL;
383 } else if (adapter->msi_enabled)
384 pci_disable_msi(adapter->pdev);
390 * igb_set_interrupt_capability - set MSI or MSI-X if supported
392 * Attempt to configure interrupts using the best available
393 * capabilities of the hardware and kernel.
395 static void igb_set_interrupt_capability(struct igb_adapter *adapter)
400 numvecs = adapter->num_tx_queues + adapter->num_rx_queues + 1;
401 adapter->msix_entries = kcalloc(numvecs, sizeof(struct msix_entry),
403 if (!adapter->msix_entries)
406 for (i = 0; i < numvecs; i++)
407 adapter->msix_entries[i].entry = i;
409 err = pci_enable_msix(adapter->pdev,
410 adapter->msix_entries,
415 igb_reset_interrupt_capability(adapter);
417 /* If we can't do MSI-X, try MSI */
419 adapter->num_rx_queues = 1;
420 if (!pci_enable_msi(adapter->pdev))
421 adapter->msi_enabled = 1;
426 * igb_request_irq - initialize interrupts
428 * Attempts to configure interrupts using the best available
429 * capabilities of the hardware and kernel.
431 static int igb_request_irq(struct igb_adapter *adapter)
433 struct net_device *netdev = adapter->netdev;
434 struct e1000_hw *hw = &adapter->hw;
437 if (adapter->msix_entries) {
438 err = igb_request_msix(adapter);
440 /* enable IAM, auto-mask,
441 * DO NOT USE EIAM or IAM in legacy mode */
442 wr32(E1000_IAM, IMS_ENABLE_MASK);
445 /* fall back to MSI */
446 igb_reset_interrupt_capability(adapter);
447 if (!pci_enable_msi(adapter->pdev))
448 adapter->msi_enabled = 1;
449 igb_free_all_tx_resources(adapter);
450 igb_free_all_rx_resources(adapter);
451 adapter->num_rx_queues = 1;
452 igb_alloc_queues(adapter);
454 if (adapter->msi_enabled) {
455 err = request_irq(adapter->pdev->irq, &igb_intr_msi, 0,
456 netdev->name, netdev);
459 /* fall back to legacy interrupts */
460 igb_reset_interrupt_capability(adapter);
461 adapter->msi_enabled = 0;
464 err = request_irq(adapter->pdev->irq, &igb_intr, IRQF_SHARED,
465 netdev->name, netdev);
468 dev_err(&adapter->pdev->dev, "Error %d getting interrupt\n",
475 static void igb_free_irq(struct igb_adapter *adapter)
477 struct net_device *netdev = adapter->netdev;
479 if (adapter->msix_entries) {
482 for (i = 0; i < adapter->num_tx_queues; i++)
483 free_irq(adapter->msix_entries[vector++].vector,
484 &(adapter->tx_ring[i]));
485 for (i = 0; i < adapter->num_rx_queues; i++)
486 free_irq(adapter->msix_entries[vector++].vector,
487 &(adapter->rx_ring[i]));
489 free_irq(adapter->msix_entries[vector++].vector, netdev);
493 free_irq(adapter->pdev->irq, netdev);
497 * igb_irq_disable - Mask off interrupt generation on the NIC
498 * @adapter: board private structure
500 static void igb_irq_disable(struct igb_adapter *adapter)
502 struct e1000_hw *hw = &adapter->hw;
504 if (adapter->msix_entries) {
505 wr32(E1000_EIMC, ~0);
510 synchronize_irq(adapter->pdev->irq);
514 * igb_irq_enable - Enable default interrupt generation settings
515 * @adapter: board private structure
517 static void igb_irq_enable(struct igb_adapter *adapter)
519 struct e1000_hw *hw = &adapter->hw;
521 if (adapter->msix_entries) {
523 adapter->eims_enable_mask);
525 adapter->eims_enable_mask);
526 wr32(E1000_IMS, E1000_IMS_LSC);
528 wr32(E1000_IMS, IMS_ENABLE_MASK);
531 static void igb_update_mng_vlan(struct igb_adapter *adapter)
533 struct net_device *netdev = adapter->netdev;
534 u16 vid = adapter->hw.mng_cookie.vlan_id;
535 u16 old_vid = adapter->mng_vlan_id;
536 if (adapter->vlgrp) {
537 if (!vlan_group_get_device(adapter->vlgrp, vid)) {
538 if (adapter->hw.mng_cookie.status &
539 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
540 igb_vlan_rx_add_vid(netdev, vid);
541 adapter->mng_vlan_id = vid;
543 adapter->mng_vlan_id = IGB_MNG_VLAN_NONE;
545 if ((old_vid != (u16)IGB_MNG_VLAN_NONE) &&
547 !vlan_group_get_device(adapter->vlgrp, old_vid))
548 igb_vlan_rx_kill_vid(netdev, old_vid);
550 adapter->mng_vlan_id = vid;
555 * igb_release_hw_control - release control of the h/w to f/w
556 * @adapter: address of board private structure
558 * igb_release_hw_control resets CTRL_EXT:DRV_LOAD bit.
559 * For ASF and Pass Through versions of f/w this means that the
560 * driver is no longer loaded.
563 static void igb_release_hw_control(struct igb_adapter *adapter)
565 struct e1000_hw *hw = &adapter->hw;
568 /* Let firmware take over control of h/w */
569 ctrl_ext = rd32(E1000_CTRL_EXT);
571 ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
576 * igb_get_hw_control - get control of the h/w from f/w
577 * @adapter: address of board private structure
579 * igb_get_hw_control sets CTRL_EXT:DRV_LOAD bit.
580 * For ASF and Pass Through versions of f/w this means that
581 * the driver is loaded.
584 static void igb_get_hw_control(struct igb_adapter *adapter)
586 struct e1000_hw *hw = &adapter->hw;
589 /* Let firmware know the driver has taken over */
590 ctrl_ext = rd32(E1000_CTRL_EXT);
592 ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
595 static void igb_init_manageability(struct igb_adapter *adapter)
597 struct e1000_hw *hw = &adapter->hw;
599 if (adapter->en_mng_pt) {
600 u32 manc2h = rd32(E1000_MANC2H);
601 u32 manc = rd32(E1000_MANC);
603 /* enable receiving management packets to the host */
604 /* this will probably generate destination unreachable messages
605 * from the host OS, but the packets will be handled on SMBUS */
606 manc |= E1000_MANC_EN_MNG2HOST;
607 #define E1000_MNG2HOST_PORT_623 (1 << 5)
608 #define E1000_MNG2HOST_PORT_664 (1 << 6)
609 manc2h |= E1000_MNG2HOST_PORT_623;
610 manc2h |= E1000_MNG2HOST_PORT_664;
611 wr32(E1000_MANC2H, manc2h);
613 wr32(E1000_MANC, manc);
618 * igb_configure - configure the hardware for RX and TX
619 * @adapter: private board structure
621 static void igb_configure(struct igb_adapter *adapter)
623 struct net_device *netdev = adapter->netdev;
626 igb_get_hw_control(adapter);
627 igb_set_multi(netdev);
629 igb_restore_vlan(adapter);
630 igb_init_manageability(adapter);
632 igb_configure_tx(adapter);
633 igb_setup_rctl(adapter);
634 igb_configure_rx(adapter);
635 /* call IGB_DESC_UNUSED which always leaves
636 * at least 1 descriptor unused to make sure
637 * next_to_use != next_to_clean */
638 for (i = 0; i < adapter->num_rx_queues; i++) {
639 struct igb_ring *ring = &adapter->rx_ring[i];
640 igb_alloc_rx_buffers_adv(adapter, ring, IGB_DESC_UNUSED(ring));
644 adapter->tx_queue_len = netdev->tx_queue_len;
649 * igb_up - Open the interface and prepare it to handle traffic
650 * @adapter: board private structure
653 int igb_up(struct igb_adapter *adapter)
655 struct e1000_hw *hw = &adapter->hw;
658 /* hardware has been reset, we need to reload some things */
659 igb_configure(adapter);
661 clear_bit(__IGB_DOWN, &adapter->state);
663 napi_enable(&adapter->napi);
665 if (adapter->msix_entries) {
666 for (i = 0; i < adapter->num_rx_queues; i++)
667 napi_enable(&adapter->rx_ring[i].napi);
668 igb_configure_msix(adapter);
671 /* Clear any pending interrupts. */
673 igb_irq_enable(adapter);
675 /* Fire a link change interrupt to start the watchdog. */
676 wr32(E1000_ICS, E1000_ICS_LSC);
680 void igb_down(struct igb_adapter *adapter)
682 struct e1000_hw *hw = &adapter->hw;
683 struct net_device *netdev = adapter->netdev;
687 /* signal that we're down so the interrupt handler does not
688 * reschedule our watchdog timer */
689 set_bit(__IGB_DOWN, &adapter->state);
691 /* disable receives in the hardware */
692 rctl = rd32(E1000_RCTL);
693 wr32(E1000_RCTL, rctl & ~E1000_RCTL_EN);
694 /* flush and sleep below */
696 netif_stop_queue(netdev);
698 /* disable transmits in the hardware */
699 tctl = rd32(E1000_TCTL);
700 tctl &= ~E1000_TCTL_EN;
701 wr32(E1000_TCTL, tctl);
702 /* flush both disables and wait for them to finish */
706 napi_disable(&adapter->napi);
708 if (adapter->msix_entries)
709 for (i = 0; i < adapter->num_rx_queues; i++)
710 napi_disable(&adapter->rx_ring[i].napi);
711 igb_irq_disable(adapter);
713 del_timer_sync(&adapter->watchdog_timer);
714 del_timer_sync(&adapter->phy_info_timer);
716 netdev->tx_queue_len = adapter->tx_queue_len;
717 netif_carrier_off(netdev);
718 adapter->link_speed = 0;
719 adapter->link_duplex = 0;
721 if (!pci_channel_offline(adapter->pdev))
723 igb_clean_all_tx_rings(adapter);
724 igb_clean_all_rx_rings(adapter);
727 void igb_reinit_locked(struct igb_adapter *adapter)
729 WARN_ON(in_interrupt());
730 while (test_and_set_bit(__IGB_RESETTING, &adapter->state))
734 clear_bit(__IGB_RESETTING, &adapter->state);
737 void igb_reset(struct igb_adapter *adapter)
739 struct e1000_hw *hw = &adapter->hw;
740 struct e1000_fc_info *fc = &adapter->hw.fc;
741 u32 pba = 0, tx_space, min_tx_space, min_rx_space;
744 /* Repartition Pba for greater than 9k mtu
745 * To take effect CTRL.RST is required.
749 if (adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) {
750 /* adjust PBA for jumbo frames */
751 wr32(E1000_PBA, pba);
753 /* To maintain wire speed transmits, the Tx FIFO should be
754 * large enough to accommodate two full transmit packets,
755 * rounded up to the next 1KB and expressed in KB. Likewise,
756 * the Rx FIFO should be large enough to accommodate at least
757 * one full receive packet and is similarly rounded up and
758 * expressed in KB. */
759 pba = rd32(E1000_PBA);
760 /* upper 16 bits has Tx packet buffer allocation size in KB */
761 tx_space = pba >> 16;
762 /* lower 16 bits has Rx packet buffer allocation size in KB */
764 /* the tx fifo also stores 16 bytes of information about the tx
765 * but don't include ethernet FCS because hardware appends it */
766 min_tx_space = (adapter->max_frame_size +
767 sizeof(struct e1000_tx_desc) -
769 min_tx_space = ALIGN(min_tx_space, 1024);
771 /* software strips receive CRC, so leave room for it */
772 min_rx_space = adapter->max_frame_size;
773 min_rx_space = ALIGN(min_rx_space, 1024);
776 /* If current Tx allocation is less than the min Tx FIFO size,
777 * and the min Tx FIFO size is less than the current Rx FIFO
778 * allocation, take space away from current Rx allocation */
779 if (tx_space < min_tx_space &&
780 ((min_tx_space - tx_space) < pba)) {
781 pba = pba - (min_tx_space - tx_space);
783 /* if short on rx space, rx wins and must trump tx
785 if (pba < min_rx_space)
789 wr32(E1000_PBA, pba);
791 /* flow control settings */
792 /* The high water mark must be low enough to fit one full frame
793 * (or the size used for early receive) above it in the Rx FIFO.
794 * Set it to the lower of:
795 * - 90% of the Rx FIFO size, or
796 * - the full Rx FIFO size minus one full frame */
797 hwm = min(((pba << 10) * 9 / 10),
798 ((pba << 10) - adapter->max_frame_size));
800 fc->high_water = hwm & 0xFFF8; /* 8-byte granularity */
801 fc->low_water = fc->high_water - 8;
802 fc->pause_time = 0xFFFF;
804 fc->type = fc->original_type;
806 /* Allow time for pending master requests to run */
807 adapter->hw.mac.ops.reset_hw(&adapter->hw);
810 if (adapter->hw.mac.ops.init_hw(&adapter->hw))
811 dev_err(&adapter->pdev->dev, "Hardware Error\n");
813 igb_update_mng_vlan(adapter);
815 /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
816 wr32(E1000_VET, ETHERNET_IEEE_VLAN_TYPE);
818 igb_reset_adaptive(&adapter->hw);
819 if (adapter->hw.phy.ops.get_phy_info)
820 adapter->hw.phy.ops.get_phy_info(&adapter->hw);
824 * igb_is_need_ioport - determine if an adapter needs ioport resources or not
825 * @pdev: PCI device information struct
827 * Returns true if an adapter needs ioport resources
829 static int igb_is_need_ioport(struct pci_dev *pdev)
831 switch (pdev->device) {
832 /* Currently there are no adapters that need ioport resources */
839 * igb_probe - Device Initialization Routine
840 * @pdev: PCI device information struct
841 * @ent: entry in igb_pci_tbl
843 * Returns 0 on success, negative on failure
845 * igb_probe initializes an adapter identified by a pci_dev structure.
846 * The OS initialization, configuring of the adapter private structure,
847 * and a hardware reset occur.
849 static int __devinit igb_probe(struct pci_dev *pdev,
850 const struct pci_device_id *ent)
852 struct net_device *netdev;
853 struct igb_adapter *adapter;
855 const struct e1000_info *ei = igb_info_tbl[ent->driver_data];
856 unsigned long mmio_start, mmio_len;
857 static int cards_found;
858 int i, err, pci_using_dac;
860 u16 eeprom_apme_mask = IGB_EEPROM_APME;
862 int bars, need_ioport;
864 /* do not allocate ioport bars when not needed */
865 need_ioport = igb_is_need_ioport(pdev);
867 bars = pci_select_bars(pdev, IORESOURCE_MEM | IORESOURCE_IO);
868 err = pci_enable_device(pdev);
870 bars = pci_select_bars(pdev, IORESOURCE_MEM);
871 err = pci_enable_device_mem(pdev);
877 err = pci_set_dma_mask(pdev, DMA_64BIT_MASK);
879 err = pci_set_consistent_dma_mask(pdev, DMA_64BIT_MASK);
883 err = pci_set_dma_mask(pdev, DMA_32BIT_MASK);
885 err = pci_set_consistent_dma_mask(pdev, DMA_32BIT_MASK);
887 dev_err(&pdev->dev, "No usable DMA "
888 "configuration, aborting\n");
894 err = pci_request_selected_regions(pdev, bars, igb_driver_name);
898 pci_set_master(pdev);
899 pci_save_state(pdev);
902 netdev = alloc_etherdev(sizeof(struct igb_adapter));
904 goto err_alloc_etherdev;
906 SET_NETDEV_DEV(netdev, &pdev->dev);
908 pci_set_drvdata(pdev, netdev);
909 adapter = netdev_priv(netdev);
910 adapter->netdev = netdev;
911 adapter->pdev = pdev;
914 adapter->msg_enable = NETIF_MSG_DRV | NETIF_MSG_PROBE;
915 adapter->bars = bars;
916 adapter->need_ioport = need_ioport;
918 mmio_start = pci_resource_start(pdev, 0);
919 mmio_len = pci_resource_len(pdev, 0);
922 adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
923 if (!adapter->hw.hw_addr)
926 netdev->open = &igb_open;
927 netdev->stop = &igb_close;
928 netdev->get_stats = &igb_get_stats;
929 netdev->set_multicast_list = &igb_set_multi;
930 netdev->set_mac_address = &igb_set_mac;
931 netdev->change_mtu = &igb_change_mtu;
932 netdev->do_ioctl = &igb_ioctl;
933 igb_set_ethtool_ops(netdev);
934 netdev->tx_timeout = &igb_tx_timeout;
935 netdev->watchdog_timeo = 5 * HZ;
936 netif_napi_add(netdev, &adapter->napi, igb_clean, 64);
937 netdev->vlan_rx_register = igb_vlan_rx_register;
938 netdev->vlan_rx_add_vid = igb_vlan_rx_add_vid;
939 netdev->vlan_rx_kill_vid = igb_vlan_rx_kill_vid;
940 #ifdef CONFIG_NET_POLL_CONTROLLER
941 netdev->poll_controller = igb_netpoll;
943 netdev->hard_start_xmit = &igb_xmit_frame_adv;
945 strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
947 netdev->mem_start = mmio_start;
948 netdev->mem_end = mmio_start + mmio_len;
950 adapter->bd_number = cards_found;
952 /* PCI config space info */
953 hw->vendor_id = pdev->vendor;
954 hw->device_id = pdev->device;
955 hw->revision_id = pdev->revision;
956 hw->subsystem_vendor_id = pdev->subsystem_vendor;
957 hw->subsystem_device_id = pdev->subsystem_device;
959 /* setup the private structure */
961 /* Copy the default MAC, PHY and NVM function pointers */
962 memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
963 memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
964 memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
965 /* Initialize skew-specific constants */
966 err = ei->get_invariants(hw);
970 err = igb_sw_init(adapter);
974 igb_get_bus_info_pcie(hw);
976 hw->phy.autoneg_wait_to_complete = false;
977 hw->mac.adaptive_ifs = true;
980 if (hw->phy.media_type == e1000_media_type_copper) {
981 hw->phy.mdix = AUTO_ALL_MODES;
982 hw->phy.disable_polarity_correction = false;
983 hw->phy.ms_type = e1000_ms_hw_default;
986 if (igb_check_reset_block(hw))
988 "PHY reset is blocked due to SOL/IDER session.\n");
990 netdev->features = NETIF_F_SG |
994 NETIF_F_HW_VLAN_FILTER;
996 netdev->features |= NETIF_F_TSO;
997 netdev->features |= NETIF_F_TSO6;
999 netdev->vlan_features |= NETIF_F_TSO;
1000 netdev->vlan_features |= NETIF_F_TSO6;
1001 netdev->vlan_features |= NETIF_F_HW_CSUM;
1002 netdev->vlan_features |= NETIF_F_SG;
1005 netdev->features |= NETIF_F_HIGHDMA;
1007 netdev->features |= NETIF_F_LLTX;
1008 adapter->en_mng_pt = igb_enable_mng_pass_thru(&adapter->hw);
1010 /* before reading the NVM, reset the controller to put the device in a
1011 * known good starting state */
1012 hw->mac.ops.reset_hw(hw);
1014 /* make sure the NVM is good */
1015 if (igb_validate_nvm_checksum(hw) < 0) {
1016 dev_err(&pdev->dev, "The NVM Checksum Is Not Valid\n");
1021 /* copy the MAC address out of the NVM */
1022 if (hw->mac.ops.read_mac_addr(hw))
1023 dev_err(&pdev->dev, "NVM Read Error\n");
1025 memcpy(netdev->dev_addr, hw->mac.addr, netdev->addr_len);
1026 memcpy(netdev->perm_addr, hw->mac.addr, netdev->addr_len);
1028 if (!is_valid_ether_addr(netdev->perm_addr)) {
1029 dev_err(&pdev->dev, "Invalid MAC Address\n");
1034 init_timer(&adapter->watchdog_timer);
1035 adapter->watchdog_timer.function = &igb_watchdog;
1036 adapter->watchdog_timer.data = (unsigned long) adapter;
1038 init_timer(&adapter->phy_info_timer);
1039 adapter->phy_info_timer.function = &igb_update_phy_info;
1040 adapter->phy_info_timer.data = (unsigned long) adapter;
1042 INIT_WORK(&adapter->reset_task, igb_reset_task);
1043 INIT_WORK(&adapter->watchdog_task, igb_watchdog_task);
1045 /* Initialize link & ring properties that are user-changeable */
1046 adapter->tx_ring->count = 256;
1047 for (i = 0; i < adapter->num_tx_queues; i++)
1048 adapter->tx_ring[i].count = adapter->tx_ring->count;
1049 adapter->rx_ring->count = 256;
1050 for (i = 0; i < adapter->num_rx_queues; i++)
1051 adapter->rx_ring[i].count = adapter->rx_ring->count;
1053 adapter->fc_autoneg = true;
1054 hw->mac.autoneg = true;
1055 hw->phy.autoneg_advertised = 0x2f;
1057 hw->fc.original_type = e1000_fc_default;
1058 hw->fc.type = e1000_fc_default;
1060 adapter->itr_setting = 3;
1061 adapter->itr = IGB_START_ITR;
1063 igb_validate_mdi_setting(hw);
1065 adapter->rx_csum = 1;
1067 /* Initial Wake on LAN setting If APM wake is enabled in the EEPROM,
1068 * enable the ACPI Magic Packet filter
1071 if (hw->bus.func == 0 ||
1072 hw->device_id == E1000_DEV_ID_82575EB_COPPER)
1073 hw->nvm.ops.read_nvm(hw, NVM_INIT_CONTROL3_PORT_A, 1,
1076 if (eeprom_data & eeprom_apme_mask)
1077 adapter->eeprom_wol |= E1000_WUFC_MAG;
1079 /* now that we have the eeprom settings, apply the special cases where
1080 * the eeprom may be wrong or the board simply won't support wake on
1081 * lan on a particular port */
1082 switch (pdev->device) {
1083 case E1000_DEV_ID_82575GB_QUAD_COPPER:
1084 adapter->eeprom_wol = 0;
1086 case E1000_DEV_ID_82575EB_FIBER_SERDES:
1087 /* Wake events only supported on port A for dual fiber
1088 * regardless of eeprom setting */
1089 if (rd32(E1000_STATUS) & E1000_STATUS_FUNC_1)
1090 adapter->eeprom_wol = 0;
1094 /* initialize the wol settings based on the eeprom settings */
1095 adapter->wol = adapter->eeprom_wol;
1097 /* reset the hardware with the new settings */
1100 /* let the f/w know that the h/w is now under the control of the
1102 igb_get_hw_control(adapter);
1104 /* tell the stack to leave us alone until igb_open() is called */
1105 netif_carrier_off(netdev);
1106 netif_stop_queue(netdev);
1108 strcpy(netdev->name, "eth%d");
1109 err = register_netdev(netdev);
1113 dev_info(&pdev->dev, "Intel(R) Gigabit Ethernet Network Connection\n");
1114 /* print bus type/speed/width info */
1115 dev_info(&pdev->dev,
1116 "%s: (PCIe:%s:%s) %02x:%02x:%02x:%02x:%02x:%02x\n",
1118 ((hw->bus.speed == e1000_bus_speed_2500)
1119 ? "2.5Gb/s" : "unknown"),
1120 ((hw->bus.width == e1000_bus_width_pcie_x4)
1121 ? "Width x4" : (hw->bus.width == e1000_bus_width_pcie_x1)
1122 ? "Width x1" : "unknown"),
1123 netdev->dev_addr[0], netdev->dev_addr[1], netdev->dev_addr[2],
1124 netdev->dev_addr[3], netdev->dev_addr[4], netdev->dev_addr[5]);
1126 igb_read_part_num(hw, &part_num);
1127 dev_info(&pdev->dev, "%s: PBA No: %06x-%03x\n", netdev->name,
1128 (part_num >> 8), (part_num & 0xff));
1130 dev_info(&pdev->dev,
1131 "Using %s interrupts. %d rx queue(s), %d tx queue(s)\n",
1132 adapter->msix_entries ? "MSI-X" :
1133 adapter->msi_enabled ? "MSI" : "legacy",
1134 adapter->num_rx_queues, adapter->num_tx_queues);
1140 igb_release_hw_control(adapter);
1142 if (!igb_check_reset_block(hw))
1143 hw->phy.ops.reset_phy(hw);
1145 if (hw->flash_address)
1146 iounmap(hw->flash_address);
1148 igb_remove_device(hw);
1149 kfree(adapter->tx_ring);
1150 kfree(adapter->rx_ring);
1153 iounmap(hw->hw_addr);
1155 free_netdev(netdev);
1157 pci_release_selected_regions(pdev, bars);
1160 pci_disable_device(pdev);
1165 * igb_remove - Device Removal Routine
1166 * @pdev: PCI device information struct
1168 * igb_remove is called by the PCI subsystem to alert the driver
1169 * that it should release a PCI device. The could be caused by a
1170 * Hot-Plug event, or because the driver is going to be removed from
1173 static void __devexit igb_remove(struct pci_dev *pdev)
1175 struct net_device *netdev = pci_get_drvdata(pdev);
1176 struct igb_adapter *adapter = netdev_priv(netdev);
1178 /* flush_scheduled work may reschedule our watchdog task, so
1179 * explicitly disable watchdog tasks from being rescheduled */
1180 set_bit(__IGB_DOWN, &adapter->state);
1181 del_timer_sync(&adapter->watchdog_timer);
1182 del_timer_sync(&adapter->phy_info_timer);
1184 flush_scheduled_work();
1186 /* Release control of h/w to f/w. If f/w is AMT enabled, this
1187 * would have already happened in close and is redundant. */
1188 igb_release_hw_control(adapter);
1190 unregister_netdev(netdev);
1192 if (!igb_check_reset_block(&adapter->hw))
1193 adapter->hw.phy.ops.reset_phy(&adapter->hw);
1195 igb_remove_device(&adapter->hw);
1196 igb_reset_interrupt_capability(adapter);
1198 kfree(adapter->tx_ring);
1199 kfree(adapter->rx_ring);
1201 iounmap(adapter->hw.hw_addr);
1202 if (adapter->hw.flash_address)
1203 iounmap(adapter->hw.flash_address);
1204 pci_release_selected_regions(pdev, adapter->bars);
1206 free_netdev(netdev);
1208 pci_disable_device(pdev);
1212 * igb_sw_init - Initialize general software structures (struct igb_adapter)
1213 * @adapter: board private structure to initialize
1215 * igb_sw_init initializes the Adapter private data structure.
1216 * Fields are initialized based on PCI device information and
1217 * OS network device settings (MTU size).
1219 static int __devinit igb_sw_init(struct igb_adapter *adapter)
1221 struct e1000_hw *hw = &adapter->hw;
1222 struct net_device *netdev = adapter->netdev;
1223 struct pci_dev *pdev = adapter->pdev;
1225 pci_read_config_word(pdev, PCI_COMMAND, &hw->bus.pci_cmd_word);
1227 adapter->rx_buffer_len = MAXIMUM_ETHERNET_VLAN_SIZE;
1228 adapter->rx_ps_hdr_size = 0; /* disable packet split */
1229 adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
1230 adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
1232 /* Number of supported queues. */
1233 /* Having more queues than CPUs doesn't make sense. */
1234 adapter->num_tx_queues = 1;
1235 adapter->num_rx_queues = min(IGB_MAX_RX_QUEUES, num_online_cpus());
1237 igb_set_interrupt_capability(adapter);
1239 if (igb_alloc_queues(adapter)) {
1240 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
1244 /* Explicitly disable IRQ since the NIC can be in any state. */
1245 igb_irq_disable(adapter);
1247 set_bit(__IGB_DOWN, &adapter->state);
1252 * igb_open - Called when a network interface is made active
1253 * @netdev: network interface device structure
1255 * Returns 0 on success, negative value on failure
1257 * The open entry point is called when a network interface is made
1258 * active by the system (IFF_UP). At this point all resources needed
1259 * for transmit and receive operations are allocated, the interrupt
1260 * handler is registered with the OS, the watchdog timer is started,
1261 * and the stack is notified that the interface is ready.
1263 static int igb_open(struct net_device *netdev)
1265 struct igb_adapter *adapter = netdev_priv(netdev);
1266 struct e1000_hw *hw = &adapter->hw;
1270 /* disallow open during test */
1271 if (test_bit(__IGB_TESTING, &adapter->state))
1274 /* allocate transmit descriptors */
1275 err = igb_setup_all_tx_resources(adapter);
1279 /* allocate receive descriptors */
1280 err = igb_setup_all_rx_resources(adapter);
1284 /* e1000_power_up_phy(adapter); */
1286 adapter->mng_vlan_id = IGB_MNG_VLAN_NONE;
1287 if ((adapter->hw.mng_cookie.status &
1288 E1000_MNG_DHCP_COOKIE_STATUS_VLAN))
1289 igb_update_mng_vlan(adapter);
1291 /* before we allocate an interrupt, we must be ready to handle it.
1292 * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
1293 * as soon as we call pci_request_irq, so we have to setup our
1294 * clean_rx handler before we do so. */
1295 igb_configure(adapter);
1297 err = igb_request_irq(adapter);
1301 /* From here on the code is the same as igb_up() */
1302 clear_bit(__IGB_DOWN, &adapter->state);
1304 napi_enable(&adapter->napi);
1305 if (adapter->msix_entries)
1306 for (i = 0; i < adapter->num_rx_queues; i++)
1307 napi_enable(&adapter->rx_ring[i].napi);
1309 igb_irq_enable(adapter);
1311 /* Clear any pending interrupts. */
1313 /* Fire a link status change interrupt to start the watchdog. */
1314 wr32(E1000_ICS, E1000_ICS_LSC);
1319 igb_release_hw_control(adapter);
1320 /* e1000_power_down_phy(adapter); */
1321 igb_free_all_rx_resources(adapter);
1323 igb_free_all_tx_resources(adapter);
1331 * igb_close - Disables a network interface
1332 * @netdev: network interface device structure
1334 * Returns 0, this is not allowed to fail
1336 * The close entry point is called when an interface is de-activated
1337 * by the OS. The hardware is still under the driver's control, but
1338 * needs to be disabled. A global MAC reset is issued to stop the
1339 * hardware, and all transmit and receive resources are freed.
1341 static int igb_close(struct net_device *netdev)
1343 struct igb_adapter *adapter = netdev_priv(netdev);
1345 WARN_ON(test_bit(__IGB_RESETTING, &adapter->state));
1348 igb_free_irq(adapter);
1350 igb_free_all_tx_resources(adapter);
1351 igb_free_all_rx_resources(adapter);
1353 /* kill manageability vlan ID if supported, but not if a vlan with
1354 * the same ID is registered on the host OS (let 8021q kill it) */
1355 if ((adapter->hw.mng_cookie.status &
1356 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
1358 vlan_group_get_device(adapter->vlgrp, adapter->mng_vlan_id)))
1359 igb_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
1365 * igb_setup_tx_resources - allocate Tx resources (Descriptors)
1366 * @adapter: board private structure
1367 * @tx_ring: tx descriptor ring (for a specific queue) to setup
1369 * Return 0 on success, negative on failure
1372 int igb_setup_tx_resources(struct igb_adapter *adapter,
1373 struct igb_ring *tx_ring)
1375 struct pci_dev *pdev = adapter->pdev;
1378 size = sizeof(struct igb_buffer) * tx_ring->count;
1379 tx_ring->buffer_info = vmalloc(size);
1380 if (!tx_ring->buffer_info)
1382 memset(tx_ring->buffer_info, 0, size);
1384 /* round up to nearest 4K */
1385 tx_ring->size = tx_ring->count * sizeof(struct e1000_tx_desc)
1387 tx_ring->size = ALIGN(tx_ring->size, 4096);
1389 tx_ring->desc = pci_alloc_consistent(pdev, tx_ring->size,
1395 tx_ring->adapter = adapter;
1396 tx_ring->next_to_use = 0;
1397 tx_ring->next_to_clean = 0;
1398 spin_lock_init(&tx_ring->tx_clean_lock);
1399 spin_lock_init(&tx_ring->tx_lock);
1403 vfree(tx_ring->buffer_info);
1404 dev_err(&adapter->pdev->dev,
1405 "Unable to allocate memory for the transmit descriptor ring\n");
1410 * igb_setup_all_tx_resources - wrapper to allocate Tx resources
1411 * (Descriptors) for all queues
1412 * @adapter: board private structure
1414 * Return 0 on success, negative on failure
1416 static int igb_setup_all_tx_resources(struct igb_adapter *adapter)
1420 for (i = 0; i < adapter->num_tx_queues; i++) {
1421 err = igb_setup_tx_resources(adapter, &adapter->tx_ring[i]);
1423 dev_err(&adapter->pdev->dev,
1424 "Allocation for Tx Queue %u failed\n", i);
1425 for (i--; i >= 0; i--)
1426 igb_free_tx_resources(adapter,
1427 &adapter->tx_ring[i]);
1436 * igb_configure_tx - Configure transmit Unit after Reset
1437 * @adapter: board private structure
1439 * Configure the Tx unit of the MAC after a reset.
1441 static void igb_configure_tx(struct igb_adapter *adapter)
1444 struct e1000_hw *hw = &adapter->hw;
1449 for (i = 0; i < adapter->num_tx_queues; i++) {
1450 struct igb_ring *ring = &(adapter->tx_ring[i]);
1452 wr32(E1000_TDLEN(i),
1453 ring->count * sizeof(struct e1000_tx_desc));
1455 wr32(E1000_TDBAL(i),
1456 tdba & 0x00000000ffffffffULL);
1457 wr32(E1000_TDBAH(i), tdba >> 32);
1459 tdwba = ring->dma + ring->count * sizeof(struct e1000_tx_desc);
1460 tdwba |= 1; /* enable head wb */
1461 wr32(E1000_TDWBAL(i),
1462 tdwba & 0x00000000ffffffffULL);
1463 wr32(E1000_TDWBAH(i), tdwba >> 32);
1465 ring->head = E1000_TDH(i);
1466 ring->tail = E1000_TDT(i);
1467 writel(0, hw->hw_addr + ring->tail);
1468 writel(0, hw->hw_addr + ring->head);
1469 txdctl = rd32(E1000_TXDCTL(i));
1470 txdctl |= E1000_TXDCTL_QUEUE_ENABLE;
1471 wr32(E1000_TXDCTL(i), txdctl);
1473 /* Turn off Relaxed Ordering on head write-backs. The
1474 * writebacks MUST be delivered in order or it will
1475 * completely screw up our bookeeping.
1477 txctrl = rd32(E1000_DCA_TXCTRL(i));
1478 txctrl &= ~E1000_DCA_TXCTRL_TX_WB_RO_EN;
1479 wr32(E1000_DCA_TXCTRL(i), txctrl);
1484 /* Use the default values for the Tx Inter Packet Gap (IPG) timer */
1486 /* Program the Transmit Control Register */
1488 tctl = rd32(E1000_TCTL);
1489 tctl &= ~E1000_TCTL_CT;
1490 tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
1491 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
1493 igb_config_collision_dist(hw);
1495 /* Setup Transmit Descriptor Settings for eop descriptor */
1496 adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_RS;
1498 /* Enable transmits */
1499 tctl |= E1000_TCTL_EN;
1501 wr32(E1000_TCTL, tctl);
1505 * igb_setup_rx_resources - allocate Rx resources (Descriptors)
1506 * @adapter: board private structure
1507 * @rx_ring: rx descriptor ring (for a specific queue) to setup
1509 * Returns 0 on success, negative on failure
1512 int igb_setup_rx_resources(struct igb_adapter *adapter,
1513 struct igb_ring *rx_ring)
1515 struct pci_dev *pdev = adapter->pdev;
1518 size = sizeof(struct igb_buffer) * rx_ring->count;
1519 rx_ring->buffer_info = vmalloc(size);
1520 if (!rx_ring->buffer_info)
1522 memset(rx_ring->buffer_info, 0, size);
1524 desc_len = sizeof(union e1000_adv_rx_desc);
1526 /* Round up to nearest 4K */
1527 rx_ring->size = rx_ring->count * desc_len;
1528 rx_ring->size = ALIGN(rx_ring->size, 4096);
1530 rx_ring->desc = pci_alloc_consistent(pdev, rx_ring->size,
1536 rx_ring->next_to_clean = 0;
1537 rx_ring->next_to_use = 0;
1538 rx_ring->pending_skb = NULL;
1540 rx_ring->adapter = adapter;
1541 /* FIXME: do we want to setup ring->napi->poll here? */
1542 rx_ring->napi.poll = adapter->napi.poll;
1547 vfree(rx_ring->buffer_info);
1548 dev_err(&adapter->pdev->dev, "Unable to allocate memory for "
1549 "the receive descriptor ring\n");
1554 * igb_setup_all_rx_resources - wrapper to allocate Rx resources
1555 * (Descriptors) for all queues
1556 * @adapter: board private structure
1558 * Return 0 on success, negative on failure
1560 static int igb_setup_all_rx_resources(struct igb_adapter *adapter)
1564 for (i = 0; i < adapter->num_rx_queues; i++) {
1565 err = igb_setup_rx_resources(adapter, &adapter->rx_ring[i]);
1567 dev_err(&adapter->pdev->dev,
1568 "Allocation for Rx Queue %u failed\n", i);
1569 for (i--; i >= 0; i--)
1570 igb_free_rx_resources(adapter,
1571 &adapter->rx_ring[i]);
1580 * igb_setup_rctl - configure the receive control registers
1581 * @adapter: Board private structure
1583 static void igb_setup_rctl(struct igb_adapter *adapter)
1585 struct e1000_hw *hw = &adapter->hw;
1590 rctl = rd32(E1000_RCTL);
1592 rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
1594 rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
1595 E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
1596 (adapter->hw.mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
1598 /* disable the stripping of CRC because it breaks
1599 * BMC firmware connected over SMBUS
1600 rctl |= E1000_RCTL_SECRC;
1603 rctl &= ~E1000_RCTL_SBP;
1605 if (adapter->netdev->mtu <= ETH_DATA_LEN)
1606 rctl &= ~E1000_RCTL_LPE;
1608 rctl |= E1000_RCTL_LPE;
1609 if (adapter->rx_buffer_len <= IGB_RXBUFFER_2048) {
1610 /* Setup buffer sizes */
1611 rctl &= ~E1000_RCTL_SZ_4096;
1612 rctl |= E1000_RCTL_BSEX;
1613 switch (adapter->rx_buffer_len) {
1614 case IGB_RXBUFFER_256:
1615 rctl |= E1000_RCTL_SZ_256;
1616 rctl &= ~E1000_RCTL_BSEX;
1618 case IGB_RXBUFFER_512:
1619 rctl |= E1000_RCTL_SZ_512;
1620 rctl &= ~E1000_RCTL_BSEX;
1622 case IGB_RXBUFFER_1024:
1623 rctl |= E1000_RCTL_SZ_1024;
1624 rctl &= ~E1000_RCTL_BSEX;
1626 case IGB_RXBUFFER_2048:
1628 rctl |= E1000_RCTL_SZ_2048;
1629 rctl &= ~E1000_RCTL_BSEX;
1631 case IGB_RXBUFFER_4096:
1632 rctl |= E1000_RCTL_SZ_4096;
1634 case IGB_RXBUFFER_8192:
1635 rctl |= E1000_RCTL_SZ_8192;
1637 case IGB_RXBUFFER_16384:
1638 rctl |= E1000_RCTL_SZ_16384;
1642 rctl &= ~E1000_RCTL_BSEX;
1643 srrctl = adapter->rx_buffer_len >> E1000_SRRCTL_BSIZEPKT_SHIFT;
1646 /* 82575 and greater support packet-split where the protocol
1647 * header is placed in skb->data and the packet data is
1648 * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
1649 * In the case of a non-split, skb->data is linearly filled,
1650 * followed by the page buffers. Therefore, skb->data is
1651 * sized to hold the largest protocol header.
1653 /* allocations using alloc_page take too long for regular MTU
1654 * so only enable packet split for jumbo frames */
1655 if (rctl & E1000_RCTL_LPE) {
1656 adapter->rx_ps_hdr_size = IGB_RXBUFFER_128;
1657 srrctl = adapter->rx_ps_hdr_size <<
1658 E1000_SRRCTL_BSIZEHDRSIZE_SHIFT;
1659 /* buffer size is ALWAYS one page */
1660 srrctl |= PAGE_SIZE >> E1000_SRRCTL_BSIZEPKT_SHIFT;
1661 srrctl |= E1000_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
1663 adapter->rx_ps_hdr_size = 0;
1664 srrctl |= E1000_SRRCTL_DESCTYPE_ADV_ONEBUF;
1667 for (i = 0; i < adapter->num_rx_queues; i++)
1668 wr32(E1000_SRRCTL(i), srrctl);
1670 wr32(E1000_RCTL, rctl);
1674 * igb_configure_rx - Configure receive Unit after Reset
1675 * @adapter: board private structure
1677 * Configure the Rx unit of the MAC after a reset.
1679 static void igb_configure_rx(struct igb_adapter *adapter)
1682 struct e1000_hw *hw = &adapter->hw;
1687 /* disable receives while setting up the descriptors */
1688 rctl = rd32(E1000_RCTL);
1689 wr32(E1000_RCTL, rctl & ~E1000_RCTL_EN);
1693 if (adapter->itr_setting > 3)
1695 1000000000 / (adapter->itr * 256));
1697 /* Setup the HW Rx Head and Tail Descriptor Pointers and
1698 * the Base and Length of the Rx Descriptor Ring */
1699 for (i = 0; i < adapter->num_rx_queues; i++) {
1700 struct igb_ring *ring = &(adapter->rx_ring[i]);
1702 wr32(E1000_RDBAL(i),
1703 rdba & 0x00000000ffffffffULL);
1704 wr32(E1000_RDBAH(i), rdba >> 32);
1705 wr32(E1000_RDLEN(i),
1706 ring->count * sizeof(union e1000_adv_rx_desc));
1708 ring->head = E1000_RDH(i);
1709 ring->tail = E1000_RDT(i);
1710 writel(0, hw->hw_addr + ring->tail);
1711 writel(0, hw->hw_addr + ring->head);
1713 rxdctl = rd32(E1000_RXDCTL(i));
1714 rxdctl |= E1000_RXDCTL_QUEUE_ENABLE;
1715 rxdctl &= 0xFFF00000;
1716 rxdctl |= IGB_RX_PTHRESH;
1717 rxdctl |= IGB_RX_HTHRESH << 8;
1718 rxdctl |= IGB_RX_WTHRESH << 16;
1719 wr32(E1000_RXDCTL(i), rxdctl);
1722 if (adapter->num_rx_queues > 1) {
1731 get_random_bytes(&random[0], 40);
1734 for (j = 0; j < (32 * 4); j++) {
1736 (j % adapter->num_rx_queues) << shift;
1739 hw->hw_addr + E1000_RETA(0) + (j & ~3));
1741 mrqc = E1000_MRQC_ENABLE_RSS_4Q;
1743 /* Fill out hash function seeds */
1744 for (j = 0; j < 10; j++)
1745 array_wr32(E1000_RSSRK(0), j, random[j]);
1747 mrqc |= (E1000_MRQC_RSS_FIELD_IPV4 |
1748 E1000_MRQC_RSS_FIELD_IPV4_TCP);
1749 mrqc |= (E1000_MRQC_RSS_FIELD_IPV6 |
1750 E1000_MRQC_RSS_FIELD_IPV6_TCP);
1751 mrqc |= (E1000_MRQC_RSS_FIELD_IPV4_UDP |
1752 E1000_MRQC_RSS_FIELD_IPV6_UDP);
1753 mrqc |= (E1000_MRQC_RSS_FIELD_IPV6_UDP_EX |
1754 E1000_MRQC_RSS_FIELD_IPV6_TCP_EX);
1757 wr32(E1000_MRQC, mrqc);
1759 /* Multiqueue and raw packet checksumming are mutually
1760 * exclusive. Note that this not the same as TCP/IP
1761 * checksumming, which works fine. */
1762 rxcsum = rd32(E1000_RXCSUM);
1763 rxcsum |= E1000_RXCSUM_PCSD;
1764 wr32(E1000_RXCSUM, rxcsum);
1766 /* Enable Receive Checksum Offload for TCP and UDP */
1767 rxcsum = rd32(E1000_RXCSUM);
1768 if (adapter->rx_csum) {
1769 rxcsum |= E1000_RXCSUM_TUOFL;
1771 /* Enable IPv4 payload checksum for UDP fragments
1772 * Must be used in conjunction with packet-split. */
1773 if (adapter->rx_ps_hdr_size)
1774 rxcsum |= E1000_RXCSUM_IPPCSE;
1776 rxcsum &= ~E1000_RXCSUM_TUOFL;
1777 /* don't need to clear IPPCSE as it defaults to 0 */
1779 wr32(E1000_RXCSUM, rxcsum);
1784 adapter->max_frame_size + VLAN_TAG_SIZE);
1786 wr32(E1000_RLPML, adapter->max_frame_size);
1788 /* Enable Receives */
1789 wr32(E1000_RCTL, rctl);
1793 * igb_free_tx_resources - Free Tx Resources per Queue
1794 * @adapter: board private structure
1795 * @tx_ring: Tx descriptor ring for a specific queue
1797 * Free all transmit software resources
1799 static void igb_free_tx_resources(struct igb_adapter *adapter,
1800 struct igb_ring *tx_ring)
1802 struct pci_dev *pdev = adapter->pdev;
1804 igb_clean_tx_ring(adapter, tx_ring);
1806 vfree(tx_ring->buffer_info);
1807 tx_ring->buffer_info = NULL;
1809 pci_free_consistent(pdev, tx_ring->size, tx_ring->desc, tx_ring->dma);
1811 tx_ring->desc = NULL;
1815 * igb_free_all_tx_resources - Free Tx Resources for All Queues
1816 * @adapter: board private structure
1818 * Free all transmit software resources
1820 static void igb_free_all_tx_resources(struct igb_adapter *adapter)
1824 for (i = 0; i < adapter->num_tx_queues; i++)
1825 igb_free_tx_resources(adapter, &adapter->tx_ring[i]);
1828 static void igb_unmap_and_free_tx_resource(struct igb_adapter *adapter,
1829 struct igb_buffer *buffer_info)
1831 if (buffer_info->dma) {
1832 pci_unmap_page(adapter->pdev,
1834 buffer_info->length,
1836 buffer_info->dma = 0;
1838 if (buffer_info->skb) {
1839 dev_kfree_skb_any(buffer_info->skb);
1840 buffer_info->skb = NULL;
1842 buffer_info->time_stamp = 0;
1843 /* buffer_info must be completely set up in the transmit path */
1847 * igb_clean_tx_ring - Free Tx Buffers
1848 * @adapter: board private structure
1849 * @tx_ring: ring to be cleaned
1851 static void igb_clean_tx_ring(struct igb_adapter *adapter,
1852 struct igb_ring *tx_ring)
1854 struct igb_buffer *buffer_info;
1858 if (!tx_ring->buffer_info)
1860 /* Free all the Tx ring sk_buffs */
1862 for (i = 0; i < tx_ring->count; i++) {
1863 buffer_info = &tx_ring->buffer_info[i];
1864 igb_unmap_and_free_tx_resource(adapter, buffer_info);
1867 size = sizeof(struct igb_buffer) * tx_ring->count;
1868 memset(tx_ring->buffer_info, 0, size);
1870 /* Zero out the descriptor ring */
1872 memset(tx_ring->desc, 0, tx_ring->size);
1874 tx_ring->next_to_use = 0;
1875 tx_ring->next_to_clean = 0;
1877 writel(0, adapter->hw.hw_addr + tx_ring->head);
1878 writel(0, adapter->hw.hw_addr + tx_ring->tail);
1882 * igb_clean_all_tx_rings - Free Tx Buffers for all queues
1883 * @adapter: board private structure
1885 static void igb_clean_all_tx_rings(struct igb_adapter *adapter)
1889 for (i = 0; i < adapter->num_tx_queues; i++)
1890 igb_clean_tx_ring(adapter, &adapter->tx_ring[i]);
1894 * igb_free_rx_resources - Free Rx Resources
1895 * @adapter: board private structure
1896 * @rx_ring: ring to clean the resources from
1898 * Free all receive software resources
1900 static void igb_free_rx_resources(struct igb_adapter *adapter,
1901 struct igb_ring *rx_ring)
1903 struct pci_dev *pdev = adapter->pdev;
1905 igb_clean_rx_ring(adapter, rx_ring);
1907 vfree(rx_ring->buffer_info);
1908 rx_ring->buffer_info = NULL;
1910 pci_free_consistent(pdev, rx_ring->size, rx_ring->desc, rx_ring->dma);
1912 rx_ring->desc = NULL;
1916 * igb_free_all_rx_resources - Free Rx Resources for All Queues
1917 * @adapter: board private structure
1919 * Free all receive software resources
1921 static void igb_free_all_rx_resources(struct igb_adapter *adapter)
1925 for (i = 0; i < adapter->num_rx_queues; i++)
1926 igb_free_rx_resources(adapter, &adapter->rx_ring[i]);
1930 * igb_clean_rx_ring - Free Rx Buffers per Queue
1931 * @adapter: board private structure
1932 * @rx_ring: ring to free buffers from
1934 static void igb_clean_rx_ring(struct igb_adapter *adapter,
1935 struct igb_ring *rx_ring)
1937 struct igb_buffer *buffer_info;
1938 struct pci_dev *pdev = adapter->pdev;
1942 if (!rx_ring->buffer_info)
1944 /* Free all the Rx ring sk_buffs */
1945 for (i = 0; i < rx_ring->count; i++) {
1946 buffer_info = &rx_ring->buffer_info[i];
1947 if (buffer_info->dma) {
1948 if (adapter->rx_ps_hdr_size)
1949 pci_unmap_single(pdev, buffer_info->dma,
1950 adapter->rx_ps_hdr_size,
1951 PCI_DMA_FROMDEVICE);
1953 pci_unmap_single(pdev, buffer_info->dma,
1954 adapter->rx_buffer_len,
1955 PCI_DMA_FROMDEVICE);
1956 buffer_info->dma = 0;
1959 if (buffer_info->skb) {
1960 dev_kfree_skb(buffer_info->skb);
1961 buffer_info->skb = NULL;
1963 if (buffer_info->page) {
1964 pci_unmap_page(pdev, buffer_info->page_dma,
1965 PAGE_SIZE, PCI_DMA_FROMDEVICE);
1966 put_page(buffer_info->page);
1967 buffer_info->page = NULL;
1968 buffer_info->page_dma = 0;
1972 /* there also may be some cached data from a chained receive */
1973 if (rx_ring->pending_skb) {
1974 dev_kfree_skb(rx_ring->pending_skb);
1975 rx_ring->pending_skb = NULL;
1978 size = sizeof(struct igb_buffer) * rx_ring->count;
1979 memset(rx_ring->buffer_info, 0, size);
1981 /* Zero out the descriptor ring */
1982 memset(rx_ring->desc, 0, rx_ring->size);
1984 rx_ring->next_to_clean = 0;
1985 rx_ring->next_to_use = 0;
1987 writel(0, adapter->hw.hw_addr + rx_ring->head);
1988 writel(0, adapter->hw.hw_addr + rx_ring->tail);
1992 * igb_clean_all_rx_rings - Free Rx Buffers for all queues
1993 * @adapter: board private structure
1995 static void igb_clean_all_rx_rings(struct igb_adapter *adapter)
1999 for (i = 0; i < adapter->num_rx_queues; i++)
2000 igb_clean_rx_ring(adapter, &adapter->rx_ring[i]);
2004 * igb_set_mac - Change the Ethernet Address of the NIC
2005 * @netdev: network interface device structure
2006 * @p: pointer to an address structure
2008 * Returns 0 on success, negative on failure
2010 static int igb_set_mac(struct net_device *netdev, void *p)
2012 struct igb_adapter *adapter = netdev_priv(netdev);
2013 struct sockaddr *addr = p;
2015 if (!is_valid_ether_addr(addr->sa_data))
2016 return -EADDRNOTAVAIL;
2018 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
2019 memcpy(adapter->hw.mac.addr, addr->sa_data, netdev->addr_len);
2021 adapter->hw.mac.ops.rar_set(&adapter->hw, adapter->hw.mac.addr, 0);
2027 * igb_set_multi - Multicast and Promiscuous mode set
2028 * @netdev: network interface device structure
2030 * The set_multi entry point is called whenever the multicast address
2031 * list or the network interface flags are updated. This routine is
2032 * responsible for configuring the hardware for proper multicast,
2033 * promiscuous mode, and all-multi behavior.
2035 static void igb_set_multi(struct net_device *netdev)
2037 struct igb_adapter *adapter = netdev_priv(netdev);
2038 struct e1000_hw *hw = &adapter->hw;
2039 struct e1000_mac_info *mac = &hw->mac;
2040 struct dev_mc_list *mc_ptr;
2045 /* Check for Promiscuous and All Multicast modes */
2047 rctl = rd32(E1000_RCTL);
2049 if (netdev->flags & IFF_PROMISC)
2050 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
2051 else if (netdev->flags & IFF_ALLMULTI) {
2052 rctl |= E1000_RCTL_MPE;
2053 rctl &= ~E1000_RCTL_UPE;
2055 rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
2057 wr32(E1000_RCTL, rctl);
2059 if (!netdev->mc_count) {
2060 /* nothing to program, so clear mc list */
2061 igb_update_mc_addr_list(hw, NULL, 0, 1,
2062 mac->rar_entry_count);
2066 mta_list = kzalloc(netdev->mc_count * 6, GFP_ATOMIC);
2070 /* The shared function expects a packed array of only addresses. */
2071 mc_ptr = netdev->mc_list;
2073 for (i = 0; i < netdev->mc_count; i++) {
2076 memcpy(mta_list + (i*ETH_ALEN), mc_ptr->dmi_addr, ETH_ALEN);
2077 mc_ptr = mc_ptr->next;
2079 igb_update_mc_addr_list(hw, mta_list, i, 1, mac->rar_entry_count);
2083 /* Need to wait a few seconds after link up to get diagnostic information from
2085 static void igb_update_phy_info(unsigned long data)
2087 struct igb_adapter *adapter = (struct igb_adapter *) data;
2088 if (adapter->hw.phy.ops.get_phy_info)
2089 adapter->hw.phy.ops.get_phy_info(&adapter->hw);
2093 * igb_watchdog - Timer Call-back
2094 * @data: pointer to adapter cast into an unsigned long
2096 static void igb_watchdog(unsigned long data)
2098 struct igb_adapter *adapter = (struct igb_adapter *)data;
2099 /* Do the rest outside of interrupt context */
2100 schedule_work(&adapter->watchdog_task);
2103 static void igb_watchdog_task(struct work_struct *work)
2105 struct igb_adapter *adapter = container_of(work,
2106 struct igb_adapter, watchdog_task);
2107 struct e1000_hw *hw = &adapter->hw;
2109 struct net_device *netdev = adapter->netdev;
2110 struct igb_ring *tx_ring = adapter->tx_ring;
2111 struct e1000_mac_info *mac = &adapter->hw.mac;
2115 if ((netif_carrier_ok(netdev)) &&
2116 (rd32(E1000_STATUS) & E1000_STATUS_LU))
2119 ret_val = hw->mac.ops.check_for_link(&adapter->hw);
2120 if ((ret_val == E1000_ERR_PHY) &&
2121 (hw->phy.type == e1000_phy_igp_3) &&
2123 E1000_PHY_CTRL_GBE_DISABLE))
2124 dev_info(&adapter->pdev->dev,
2125 "Gigabit has been disabled, downgrading speed\n");
2127 if ((hw->phy.media_type == e1000_media_type_internal_serdes) &&
2128 !(rd32(E1000_TXCW) & E1000_TXCW_ANE))
2129 link = mac->serdes_has_link;
2131 link = rd32(E1000_STATUS) &
2135 if (!netif_carrier_ok(netdev)) {
2137 hw->mac.ops.get_speed_and_duplex(&adapter->hw,
2138 &adapter->link_speed,
2139 &adapter->link_duplex);
2141 ctrl = rd32(E1000_CTRL);
2142 dev_info(&adapter->pdev->dev,
2143 "NIC Link is Up %d Mbps %s, "
2144 "Flow Control: %s\n",
2145 adapter->link_speed,
2146 adapter->link_duplex == FULL_DUPLEX ?
2147 "Full Duplex" : "Half Duplex",
2148 ((ctrl & E1000_CTRL_TFCE) && (ctrl &
2149 E1000_CTRL_RFCE)) ? "RX/TX" : ((ctrl &
2150 E1000_CTRL_RFCE) ? "RX" : ((ctrl &
2151 E1000_CTRL_TFCE) ? "TX" : "None")));
2153 /* tweak tx_queue_len according to speed/duplex and
2154 * adjust the timeout factor */
2155 netdev->tx_queue_len = adapter->tx_queue_len;
2156 adapter->tx_timeout_factor = 1;
2157 switch (adapter->link_speed) {
2159 netdev->tx_queue_len = 10;
2160 adapter->tx_timeout_factor = 14;
2163 netdev->tx_queue_len = 100;
2164 /* maybe add some timeout factor ? */
2168 netif_carrier_on(netdev);
2169 netif_wake_queue(netdev);
2171 if (!test_bit(__IGB_DOWN, &adapter->state))
2172 mod_timer(&adapter->phy_info_timer,
2173 round_jiffies(jiffies + 2 * HZ));
2176 if (netif_carrier_ok(netdev)) {
2177 adapter->link_speed = 0;
2178 adapter->link_duplex = 0;
2179 dev_info(&adapter->pdev->dev, "NIC Link is Down\n");
2180 netif_carrier_off(netdev);
2181 netif_stop_queue(netdev);
2182 if (!test_bit(__IGB_DOWN, &adapter->state))
2183 mod_timer(&adapter->phy_info_timer,
2184 round_jiffies(jiffies + 2 * HZ));
2189 igb_update_stats(adapter);
2191 mac->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
2192 adapter->tpt_old = adapter->stats.tpt;
2193 mac->collision_delta = adapter->stats.colc - adapter->colc_old;
2194 adapter->colc_old = adapter->stats.colc;
2196 adapter->gorc = adapter->stats.gorc - adapter->gorc_old;
2197 adapter->gorc_old = adapter->stats.gorc;
2198 adapter->gotc = adapter->stats.gotc - adapter->gotc_old;
2199 adapter->gotc_old = adapter->stats.gotc;
2201 igb_update_adaptive(&adapter->hw);
2203 if (!netif_carrier_ok(netdev)) {
2204 if (IGB_DESC_UNUSED(tx_ring) + 1 < tx_ring->count) {
2205 /* We've lost link, so the controller stops DMA,
2206 * but we've got queued Tx work that's never going
2207 * to get done, so reset controller to flush Tx.
2208 * (Do the reset outside of interrupt context). */
2209 adapter->tx_timeout_count++;
2210 schedule_work(&adapter->reset_task);
2214 /* Cause software interrupt to ensure rx ring is cleaned */
2215 wr32(E1000_ICS, E1000_ICS_RXDMT0);
2217 /* Force detection of hung controller every watchdog period */
2218 tx_ring->detect_tx_hung = true;
2220 /* Reset the timer */
2221 if (!test_bit(__IGB_DOWN, &adapter->state))
2222 mod_timer(&adapter->watchdog_timer,
2223 round_jiffies(jiffies + 2 * HZ));
2226 enum latency_range {
2230 latency_invalid = 255
2234 static void igb_lower_rx_eitr(struct igb_adapter *adapter,
2235 struct igb_ring *rx_ring)
2237 struct e1000_hw *hw = &adapter->hw;
2240 new_val = rx_ring->itr_val / 2;
2241 if (new_val < IGB_MIN_DYN_ITR)
2242 new_val = IGB_MIN_DYN_ITR;
2244 if (new_val != rx_ring->itr_val) {
2245 rx_ring->itr_val = new_val;
2246 wr32(rx_ring->itr_register,
2247 1000000000 / (new_val * 256));
2251 static void igb_raise_rx_eitr(struct igb_adapter *adapter,
2252 struct igb_ring *rx_ring)
2254 struct e1000_hw *hw = &adapter->hw;
2257 new_val = rx_ring->itr_val * 2;
2258 if (new_val > IGB_MAX_DYN_ITR)
2259 new_val = IGB_MAX_DYN_ITR;
2261 if (new_val != rx_ring->itr_val) {
2262 rx_ring->itr_val = new_val;
2263 wr32(rx_ring->itr_register,
2264 1000000000 / (new_val * 256));
2269 * igb_update_itr - update the dynamic ITR value based on statistics
2270 * Stores a new ITR value based on packets and byte
2271 * counts during the last interrupt. The advantage of per interrupt
2272 * computation is faster updates and more accurate ITR for the current
2273 * traffic pattern. Constants in this function were computed
2274 * based on theoretical maximum wire speed and thresholds were set based
2275 * on testing data as well as attempting to minimize response time
2276 * while increasing bulk throughput.
2277 * this functionality is controlled by the InterruptThrottleRate module
2278 * parameter (see igb_param.c)
2279 * NOTE: These calculations are only valid when operating in a single-
2280 * queue environment.
2281 * @adapter: pointer to adapter
2282 * @itr_setting: current adapter->itr
2283 * @packets: the number of packets during this measurement interval
2284 * @bytes: the number of bytes during this measurement interval
2286 static unsigned int igb_update_itr(struct igb_adapter *adapter, u16 itr_setting,
2287 int packets, int bytes)
2289 unsigned int retval = itr_setting;
2292 goto update_itr_done;
2294 switch (itr_setting) {
2295 case lowest_latency:
2296 /* handle TSO and jumbo frames */
2297 if (bytes/packets > 8000)
2298 retval = bulk_latency;
2299 else if ((packets < 5) && (bytes > 512))
2300 retval = low_latency;
2302 case low_latency: /* 50 usec aka 20000 ints/s */
2303 if (bytes > 10000) {
2304 /* this if handles the TSO accounting */
2305 if (bytes/packets > 8000) {
2306 retval = bulk_latency;
2307 } else if ((packets < 10) || ((bytes/packets) > 1200)) {
2308 retval = bulk_latency;
2309 } else if ((packets > 35)) {
2310 retval = lowest_latency;
2312 } else if (bytes/packets > 2000) {
2313 retval = bulk_latency;
2314 } else if (packets <= 2 && bytes < 512) {
2315 retval = lowest_latency;
2318 case bulk_latency: /* 250 usec aka 4000 ints/s */
2319 if (bytes > 25000) {
2321 retval = low_latency;
2322 } else if (bytes < 6000) {
2323 retval = low_latency;
2332 static void igb_set_itr(struct igb_adapter *adapter, u16 itr_register,
2336 u32 new_itr = adapter->itr;
2338 /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
2339 if (adapter->link_speed != SPEED_1000) {
2345 adapter->rx_itr = igb_update_itr(adapter,
2347 adapter->rx_ring->total_packets,
2348 adapter->rx_ring->total_bytes);
2349 /* conservative mode (itr 3) eliminates the lowest_latency setting */
2350 if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
2351 adapter->rx_itr = low_latency;
2354 adapter->tx_itr = igb_update_itr(adapter,
2356 adapter->tx_ring->total_packets,
2357 adapter->tx_ring->total_bytes);
2358 /* conservative mode (itr 3) eliminates the
2359 * lowest_latency setting */
2360 if (adapter->itr_setting == 3 &&
2361 adapter->tx_itr == lowest_latency)
2362 adapter->tx_itr = low_latency;
2364 current_itr = max(adapter->rx_itr, adapter->tx_itr);
2366 current_itr = adapter->rx_itr;
2369 switch (current_itr) {
2370 /* counts and packets in update_itr are dependent on these numbers */
2371 case lowest_latency:
2375 new_itr = 20000; /* aka hwitr = ~200 */
2385 if (new_itr != adapter->itr) {
2386 /* this attempts to bias the interrupt rate towards Bulk
2387 * by adding intermediate steps when interrupt rate is
2389 new_itr = new_itr > adapter->itr ?
2390 min(adapter->itr + (new_itr >> 2), new_itr) :
2392 /* Don't write the value here; it resets the adapter's
2393 * internal timer, and causes us to delay far longer than
2394 * we should between interrupts. Instead, we write the ITR
2395 * value at the beginning of the next interrupt so the timing
2396 * ends up being correct.
2398 adapter->itr = new_itr;
2399 adapter->set_itr = 1;
2406 #define IGB_TX_FLAGS_CSUM 0x00000001
2407 #define IGB_TX_FLAGS_VLAN 0x00000002
2408 #define IGB_TX_FLAGS_TSO 0x00000004
2409 #define IGB_TX_FLAGS_IPV4 0x00000008
2410 #define IGB_TX_FLAGS_VLAN_MASK 0xffff0000
2411 #define IGB_TX_FLAGS_VLAN_SHIFT 16
2413 static inline int igb_tso_adv(struct igb_adapter *adapter,
2414 struct igb_ring *tx_ring,
2415 struct sk_buff *skb, u32 tx_flags, u8 *hdr_len)
2417 struct e1000_adv_tx_context_desc *context_desc;
2420 struct igb_buffer *buffer_info;
2421 u32 info = 0, tu_cmd = 0;
2422 u32 mss_l4len_idx, l4len;
2425 if (skb_header_cloned(skb)) {
2426 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2431 l4len = tcp_hdrlen(skb);
2434 if (skb->protocol == htons(ETH_P_IP)) {
2435 struct iphdr *iph = ip_hdr(skb);
2438 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
2442 } else if (skb_shinfo(skb)->gso_type == SKB_GSO_TCPV6) {
2443 ipv6_hdr(skb)->payload_len = 0;
2444 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
2445 &ipv6_hdr(skb)->daddr,
2449 i = tx_ring->next_to_use;
2451 buffer_info = &tx_ring->buffer_info[i];
2452 context_desc = E1000_TX_CTXTDESC_ADV(*tx_ring, i);
2453 /* VLAN MACLEN IPLEN */
2454 if (tx_flags & IGB_TX_FLAGS_VLAN)
2455 info |= (tx_flags & IGB_TX_FLAGS_VLAN_MASK);
2456 info |= (skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT);
2457 *hdr_len += skb_network_offset(skb);
2458 info |= skb_network_header_len(skb);
2459 *hdr_len += skb_network_header_len(skb);
2460 context_desc->vlan_macip_lens = cpu_to_le32(info);
2462 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
2463 tu_cmd |= (E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT);
2465 if (skb->protocol == htons(ETH_P_IP))
2466 tu_cmd |= E1000_ADVTXD_TUCMD_IPV4;
2467 tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
2469 context_desc->type_tucmd_mlhl = cpu_to_le32(tu_cmd);
2472 mss_l4len_idx = (skb_shinfo(skb)->gso_size << E1000_ADVTXD_MSS_SHIFT);
2473 mss_l4len_idx |= (l4len << E1000_ADVTXD_L4LEN_SHIFT);
2475 /* Context index must be unique per ring. Luckily, so is the interrupt
2477 mss_l4len_idx |= tx_ring->eims_value >> 4;
2479 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
2480 context_desc->seqnum_seed = 0;
2482 buffer_info->time_stamp = jiffies;
2483 buffer_info->dma = 0;
2485 if (i == tx_ring->count)
2488 tx_ring->next_to_use = i;
2493 static inline bool igb_tx_csum_adv(struct igb_adapter *adapter,
2494 struct igb_ring *tx_ring,
2495 struct sk_buff *skb, u32 tx_flags)
2497 struct e1000_adv_tx_context_desc *context_desc;
2499 struct igb_buffer *buffer_info;
2500 u32 info = 0, tu_cmd = 0;
2502 if ((skb->ip_summed == CHECKSUM_PARTIAL) ||
2503 (tx_flags & IGB_TX_FLAGS_VLAN)) {
2504 i = tx_ring->next_to_use;
2505 buffer_info = &tx_ring->buffer_info[i];
2506 context_desc = E1000_TX_CTXTDESC_ADV(*tx_ring, i);
2508 if (tx_flags & IGB_TX_FLAGS_VLAN)
2509 info |= (tx_flags & IGB_TX_FLAGS_VLAN_MASK);
2510 info |= (skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT);
2511 if (skb->ip_summed == CHECKSUM_PARTIAL)
2512 info |= skb_network_header_len(skb);
2514 context_desc->vlan_macip_lens = cpu_to_le32(info);
2516 tu_cmd |= (E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT);
2518 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2519 switch (skb->protocol) {
2520 case __constant_htons(ETH_P_IP):
2521 tu_cmd |= E1000_ADVTXD_TUCMD_IPV4;
2522 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
2523 tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
2525 case __constant_htons(ETH_P_IPV6):
2526 /* XXX what about other V6 headers?? */
2527 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
2528 tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
2531 if (unlikely(net_ratelimit()))
2532 dev_warn(&adapter->pdev->dev,
2533 "partial checksum but proto=%x!\n",
2539 context_desc->type_tucmd_mlhl = cpu_to_le32(tu_cmd);
2540 context_desc->seqnum_seed = 0;
2541 context_desc->mss_l4len_idx =
2542 cpu_to_le32(tx_ring->eims_value >> 4);
2544 buffer_info->time_stamp = jiffies;
2545 buffer_info->dma = 0;
2548 if (i == tx_ring->count)
2550 tx_ring->next_to_use = i;
2559 #define IGB_MAX_TXD_PWR 16
2560 #define IGB_MAX_DATA_PER_TXD (1<<IGB_MAX_TXD_PWR)
2562 static inline int igb_tx_map_adv(struct igb_adapter *adapter,
2563 struct igb_ring *tx_ring,
2564 struct sk_buff *skb)
2566 struct igb_buffer *buffer_info;
2567 unsigned int len = skb_headlen(skb);
2568 unsigned int count = 0, i;
2571 i = tx_ring->next_to_use;
2573 buffer_info = &tx_ring->buffer_info[i];
2574 BUG_ON(len >= IGB_MAX_DATA_PER_TXD);
2575 buffer_info->length = len;
2576 /* set time_stamp *before* dma to help avoid a possible race */
2577 buffer_info->time_stamp = jiffies;
2578 buffer_info->dma = pci_map_single(adapter->pdev, skb->data, len,
2582 if (i == tx_ring->count)
2585 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++) {
2586 struct skb_frag_struct *frag;
2588 frag = &skb_shinfo(skb)->frags[f];
2591 buffer_info = &tx_ring->buffer_info[i];
2592 BUG_ON(len >= IGB_MAX_DATA_PER_TXD);
2593 buffer_info->length = len;
2594 buffer_info->time_stamp = jiffies;
2595 buffer_info->dma = pci_map_page(adapter->pdev,
2603 if (i == tx_ring->count)
2607 i = (i == 0) ? tx_ring->count - 1 : i - 1;
2608 tx_ring->buffer_info[i].skb = skb;
2613 static inline void igb_tx_queue_adv(struct igb_adapter *adapter,
2614 struct igb_ring *tx_ring,
2615 int tx_flags, int count, u32 paylen,
2618 union e1000_adv_tx_desc *tx_desc = NULL;
2619 struct igb_buffer *buffer_info;
2620 u32 olinfo_status = 0, cmd_type_len;
2623 cmd_type_len = (E1000_ADVTXD_DTYP_DATA | E1000_ADVTXD_DCMD_IFCS |
2624 E1000_ADVTXD_DCMD_DEXT);
2626 if (tx_flags & IGB_TX_FLAGS_VLAN)
2627 cmd_type_len |= E1000_ADVTXD_DCMD_VLE;
2629 if (tx_flags & IGB_TX_FLAGS_TSO) {
2630 cmd_type_len |= E1000_ADVTXD_DCMD_TSE;
2632 /* insert tcp checksum */
2633 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
2635 /* insert ip checksum */
2636 if (tx_flags & IGB_TX_FLAGS_IPV4)
2637 olinfo_status |= E1000_TXD_POPTS_IXSM << 8;
2639 } else if (tx_flags & IGB_TX_FLAGS_CSUM) {
2640 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
2643 if (tx_flags & (IGB_TX_FLAGS_CSUM | IGB_TX_FLAGS_TSO |
2645 olinfo_status |= tx_ring->eims_value >> 4;
2647 olinfo_status |= ((paylen - hdr_len) << E1000_ADVTXD_PAYLEN_SHIFT);
2649 i = tx_ring->next_to_use;
2651 buffer_info = &tx_ring->buffer_info[i];
2652 tx_desc = E1000_TX_DESC_ADV(*tx_ring, i);
2653 tx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma);
2654 tx_desc->read.cmd_type_len =
2655 cpu_to_le32(cmd_type_len | buffer_info->length);
2656 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
2658 if (i == tx_ring->count)
2662 tx_desc->read.cmd_type_len |= cpu_to_le32(adapter->txd_cmd);
2663 /* Force memory writes to complete before letting h/w
2664 * know there are new descriptors to fetch. (Only
2665 * applicable for weak-ordered memory model archs,
2666 * such as IA-64). */
2669 tx_ring->next_to_use = i;
2670 writel(i, adapter->hw.hw_addr + tx_ring->tail);
2671 /* we need this if more than one processor can write to our tail
2672 * at a time, it syncronizes IO on IA64/Altix systems */
2676 static int __igb_maybe_stop_tx(struct net_device *netdev,
2677 struct igb_ring *tx_ring, int size)
2679 struct igb_adapter *adapter = netdev_priv(netdev);
2681 netif_stop_queue(netdev);
2682 /* Herbert's original patch had:
2683 * smp_mb__after_netif_stop_queue();
2684 * but since that doesn't exist yet, just open code it. */
2687 /* We need to check again in a case another CPU has just
2688 * made room available. */
2689 if (IGB_DESC_UNUSED(tx_ring) < size)
2693 netif_start_queue(netdev);
2694 ++adapter->restart_queue;
2698 static int igb_maybe_stop_tx(struct net_device *netdev,
2699 struct igb_ring *tx_ring, int size)
2701 if (IGB_DESC_UNUSED(tx_ring) >= size)
2703 return __igb_maybe_stop_tx(netdev, tx_ring, size);
2706 #define TXD_USE_COUNT(S) (((S) >> (IGB_MAX_TXD_PWR)) + 1)
2708 static int igb_xmit_frame_ring_adv(struct sk_buff *skb,
2709 struct net_device *netdev,
2710 struct igb_ring *tx_ring)
2712 struct igb_adapter *adapter = netdev_priv(netdev);
2713 unsigned int tx_flags = 0;
2715 unsigned long irq_flags;
2719 len = skb_headlen(skb);
2721 if (test_bit(__IGB_DOWN, &adapter->state)) {
2722 dev_kfree_skb_any(skb);
2723 return NETDEV_TX_OK;
2726 if (skb->len <= 0) {
2727 dev_kfree_skb_any(skb);
2728 return NETDEV_TX_OK;
2731 if (!spin_trylock_irqsave(&tx_ring->tx_lock, irq_flags))
2732 /* Collision - tell upper layer to requeue */
2733 return NETDEV_TX_LOCKED;
2735 /* need: 1 descriptor per page,
2736 * + 2 desc gap to keep tail from touching head,
2737 * + 1 desc for skb->data,
2738 * + 1 desc for context descriptor,
2739 * otherwise try next time */
2740 if (igb_maybe_stop_tx(netdev, tx_ring, skb_shinfo(skb)->nr_frags + 4)) {
2741 /* this is a hard error */
2742 spin_unlock_irqrestore(&tx_ring->tx_lock, irq_flags);
2743 return NETDEV_TX_BUSY;
2746 if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
2747 tx_flags |= IGB_TX_FLAGS_VLAN;
2748 tx_flags |= (vlan_tx_tag_get(skb) << IGB_TX_FLAGS_VLAN_SHIFT);
2751 tso = skb_is_gso(skb) ? igb_tso_adv(adapter, tx_ring, skb, tx_flags,
2755 dev_kfree_skb_any(skb);
2756 spin_unlock_irqrestore(&tx_ring->tx_lock, irq_flags);
2757 return NETDEV_TX_OK;
2761 tx_flags |= IGB_TX_FLAGS_TSO;
2762 else if (igb_tx_csum_adv(adapter, tx_ring, skb, tx_flags))
2763 if (skb->ip_summed == CHECKSUM_PARTIAL)
2764 tx_flags |= IGB_TX_FLAGS_CSUM;
2766 if (skb->protocol == htons(ETH_P_IP))
2767 tx_flags |= IGB_TX_FLAGS_IPV4;
2769 igb_tx_queue_adv(adapter, tx_ring, tx_flags,
2770 igb_tx_map_adv(adapter, tx_ring, skb),
2773 netdev->trans_start = jiffies;
2775 /* Make sure there is space in the ring for the next send. */
2776 igb_maybe_stop_tx(netdev, tx_ring, MAX_SKB_FRAGS + 4);
2778 spin_unlock_irqrestore(&tx_ring->tx_lock, irq_flags);
2779 return NETDEV_TX_OK;
2782 static int igb_xmit_frame_adv(struct sk_buff *skb, struct net_device *netdev)
2784 struct igb_adapter *adapter = netdev_priv(netdev);
2785 struct igb_ring *tx_ring = &adapter->tx_ring[0];
2787 /* This goes back to the question of how to logically map a tx queue
2788 * to a flow. Right now, performance is impacted slightly negatively
2789 * if using multiple tx queues. If the stack breaks away from a
2790 * single qdisc implementation, we can look at this again. */
2791 return (igb_xmit_frame_ring_adv(skb, netdev, tx_ring));
2795 * igb_tx_timeout - Respond to a Tx Hang
2796 * @netdev: network interface device structure
2798 static void igb_tx_timeout(struct net_device *netdev)
2800 struct igb_adapter *adapter = netdev_priv(netdev);
2801 struct e1000_hw *hw = &adapter->hw;
2803 /* Do the reset outside of interrupt context */
2804 adapter->tx_timeout_count++;
2805 schedule_work(&adapter->reset_task);
2806 wr32(E1000_EICS, adapter->eims_enable_mask &
2807 ~(E1000_EIMS_TCP_TIMER | E1000_EIMS_OTHER));
2810 static void igb_reset_task(struct work_struct *work)
2812 struct igb_adapter *adapter;
2813 adapter = container_of(work, struct igb_adapter, reset_task);
2815 igb_reinit_locked(adapter);
2819 * igb_get_stats - Get System Network Statistics
2820 * @netdev: network interface device structure
2822 * Returns the address of the device statistics structure.
2823 * The statistics are actually updated from the timer callback.
2825 static struct net_device_stats *
2826 igb_get_stats(struct net_device *netdev)
2828 struct igb_adapter *adapter = netdev_priv(netdev);
2830 /* only return the current stats */
2831 return &adapter->net_stats;
2835 * igb_change_mtu - Change the Maximum Transfer Unit
2836 * @netdev: network interface device structure
2837 * @new_mtu: new value for maximum frame size
2839 * Returns 0 on success, negative on failure
2841 static int igb_change_mtu(struct net_device *netdev, int new_mtu)
2843 struct igb_adapter *adapter = netdev_priv(netdev);
2844 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
2846 if ((max_frame < ETH_ZLEN + ETH_FCS_LEN) ||
2847 (max_frame > MAX_JUMBO_FRAME_SIZE)) {
2848 dev_err(&adapter->pdev->dev, "Invalid MTU setting\n");
2852 #define MAX_STD_JUMBO_FRAME_SIZE 9234
2853 if (max_frame > MAX_STD_JUMBO_FRAME_SIZE) {
2854 dev_err(&adapter->pdev->dev, "MTU > 9216 not supported.\n");
2858 while (test_and_set_bit(__IGB_RESETTING, &adapter->state))
2860 /* igb_down has a dependency on max_frame_size */
2861 adapter->max_frame_size = max_frame;
2862 if (netif_running(netdev))
2865 /* NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
2866 * means we reserve 2 more, this pushes us to allocate from the next
2868 * i.e. RXBUFFER_2048 --> size-4096 slab
2871 if (max_frame <= IGB_RXBUFFER_256)
2872 adapter->rx_buffer_len = IGB_RXBUFFER_256;
2873 else if (max_frame <= IGB_RXBUFFER_512)
2874 adapter->rx_buffer_len = IGB_RXBUFFER_512;
2875 else if (max_frame <= IGB_RXBUFFER_1024)
2876 adapter->rx_buffer_len = IGB_RXBUFFER_1024;
2877 else if (max_frame <= IGB_RXBUFFER_2048)
2878 adapter->rx_buffer_len = IGB_RXBUFFER_2048;
2880 adapter->rx_buffer_len = IGB_RXBUFFER_4096;
2881 /* adjust allocation if LPE protects us, and we aren't using SBP */
2882 if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
2883 (max_frame == MAXIMUM_ETHERNET_VLAN_SIZE))
2884 adapter->rx_buffer_len = MAXIMUM_ETHERNET_VLAN_SIZE;
2886 dev_info(&adapter->pdev->dev, "changing MTU from %d to %d\n",
2887 netdev->mtu, new_mtu);
2888 netdev->mtu = new_mtu;
2890 if (netif_running(netdev))
2895 clear_bit(__IGB_RESETTING, &adapter->state);
2901 * igb_update_stats - Update the board statistics counters
2902 * @adapter: board private structure
2905 void igb_update_stats(struct igb_adapter *adapter)
2907 struct e1000_hw *hw = &adapter->hw;
2908 struct pci_dev *pdev = adapter->pdev;
2911 #define PHY_IDLE_ERROR_COUNT_MASK 0x00FF
2914 * Prevent stats update while adapter is being reset, or if the pci
2915 * connection is down.
2917 if (adapter->link_speed == 0)
2919 if (pci_channel_offline(pdev))
2922 adapter->stats.crcerrs += rd32(E1000_CRCERRS);
2923 adapter->stats.gprc += rd32(E1000_GPRC);
2924 adapter->stats.gorc += rd32(E1000_GORCL);
2925 rd32(E1000_GORCH); /* clear GORCL */
2926 adapter->stats.bprc += rd32(E1000_BPRC);
2927 adapter->stats.mprc += rd32(E1000_MPRC);
2928 adapter->stats.roc += rd32(E1000_ROC);
2930 adapter->stats.prc64 += rd32(E1000_PRC64);
2931 adapter->stats.prc127 += rd32(E1000_PRC127);
2932 adapter->stats.prc255 += rd32(E1000_PRC255);
2933 adapter->stats.prc511 += rd32(E1000_PRC511);
2934 adapter->stats.prc1023 += rd32(E1000_PRC1023);
2935 adapter->stats.prc1522 += rd32(E1000_PRC1522);
2936 adapter->stats.symerrs += rd32(E1000_SYMERRS);
2937 adapter->stats.sec += rd32(E1000_SEC);
2939 adapter->stats.mpc += rd32(E1000_MPC);
2940 adapter->stats.scc += rd32(E1000_SCC);
2941 adapter->stats.ecol += rd32(E1000_ECOL);
2942 adapter->stats.mcc += rd32(E1000_MCC);
2943 adapter->stats.latecol += rd32(E1000_LATECOL);
2944 adapter->stats.dc += rd32(E1000_DC);
2945 adapter->stats.rlec += rd32(E1000_RLEC);
2946 adapter->stats.xonrxc += rd32(E1000_XONRXC);
2947 adapter->stats.xontxc += rd32(E1000_XONTXC);
2948 adapter->stats.xoffrxc += rd32(E1000_XOFFRXC);
2949 adapter->stats.xofftxc += rd32(E1000_XOFFTXC);
2950 adapter->stats.fcruc += rd32(E1000_FCRUC);
2951 adapter->stats.gptc += rd32(E1000_GPTC);
2952 adapter->stats.gotc += rd32(E1000_GOTCL);
2953 rd32(E1000_GOTCH); /* clear GOTCL */
2954 adapter->stats.rnbc += rd32(E1000_RNBC);
2955 adapter->stats.ruc += rd32(E1000_RUC);
2956 adapter->stats.rfc += rd32(E1000_RFC);
2957 adapter->stats.rjc += rd32(E1000_RJC);
2958 adapter->stats.tor += rd32(E1000_TORH);
2959 adapter->stats.tot += rd32(E1000_TOTH);
2960 adapter->stats.tpr += rd32(E1000_TPR);
2962 adapter->stats.ptc64 += rd32(E1000_PTC64);
2963 adapter->stats.ptc127 += rd32(E1000_PTC127);
2964 adapter->stats.ptc255 += rd32(E1000_PTC255);
2965 adapter->stats.ptc511 += rd32(E1000_PTC511);
2966 adapter->stats.ptc1023 += rd32(E1000_PTC1023);
2967 adapter->stats.ptc1522 += rd32(E1000_PTC1522);
2969 adapter->stats.mptc += rd32(E1000_MPTC);
2970 adapter->stats.bptc += rd32(E1000_BPTC);
2972 /* used for adaptive IFS */
2974 hw->mac.tx_packet_delta = rd32(E1000_TPT);
2975 adapter->stats.tpt += hw->mac.tx_packet_delta;
2976 hw->mac.collision_delta = rd32(E1000_COLC);
2977 adapter->stats.colc += hw->mac.collision_delta;
2979 adapter->stats.algnerrc += rd32(E1000_ALGNERRC);
2980 adapter->stats.rxerrc += rd32(E1000_RXERRC);
2981 adapter->stats.tncrs += rd32(E1000_TNCRS);
2982 adapter->stats.tsctc += rd32(E1000_TSCTC);
2983 adapter->stats.tsctfc += rd32(E1000_TSCTFC);
2985 adapter->stats.iac += rd32(E1000_IAC);
2986 adapter->stats.icrxoc += rd32(E1000_ICRXOC);
2987 adapter->stats.icrxptc += rd32(E1000_ICRXPTC);
2988 adapter->stats.icrxatc += rd32(E1000_ICRXATC);
2989 adapter->stats.ictxptc += rd32(E1000_ICTXPTC);
2990 adapter->stats.ictxatc += rd32(E1000_ICTXATC);
2991 adapter->stats.ictxqec += rd32(E1000_ICTXQEC);
2992 adapter->stats.ictxqmtc += rd32(E1000_ICTXQMTC);
2993 adapter->stats.icrxdmtc += rd32(E1000_ICRXDMTC);
2995 /* Fill out the OS statistics structure */
2996 adapter->net_stats.multicast = adapter->stats.mprc;
2997 adapter->net_stats.collisions = adapter->stats.colc;
3001 /* RLEC on some newer hardware can be incorrect so build
3002 * our own version based on RUC and ROC */
3003 adapter->net_stats.rx_errors = adapter->stats.rxerrc +
3004 adapter->stats.crcerrs + adapter->stats.algnerrc +
3005 adapter->stats.ruc + adapter->stats.roc +
3006 adapter->stats.cexterr;
3007 adapter->net_stats.rx_length_errors = adapter->stats.ruc +
3009 adapter->net_stats.rx_crc_errors = adapter->stats.crcerrs;
3010 adapter->net_stats.rx_frame_errors = adapter->stats.algnerrc;
3011 adapter->net_stats.rx_missed_errors = adapter->stats.mpc;
3014 adapter->net_stats.tx_errors = adapter->stats.ecol +
3015 adapter->stats.latecol;
3016 adapter->net_stats.tx_aborted_errors = adapter->stats.ecol;
3017 adapter->net_stats.tx_window_errors = adapter->stats.latecol;
3018 adapter->net_stats.tx_carrier_errors = adapter->stats.tncrs;
3020 /* Tx Dropped needs to be maintained elsewhere */
3023 if (hw->phy.media_type == e1000_media_type_copper) {
3024 if ((adapter->link_speed == SPEED_1000) &&
3025 (!hw->phy.ops.read_phy_reg(hw, PHY_1000T_STATUS,
3027 phy_tmp &= PHY_IDLE_ERROR_COUNT_MASK;
3028 adapter->phy_stats.idle_errors += phy_tmp;
3032 /* Management Stats */
3033 adapter->stats.mgptc += rd32(E1000_MGTPTC);
3034 adapter->stats.mgprc += rd32(E1000_MGTPRC);
3035 adapter->stats.mgpdc += rd32(E1000_MGTPDC);
3039 static irqreturn_t igb_msix_other(int irq, void *data)
3041 struct net_device *netdev = data;
3042 struct igb_adapter *adapter = netdev_priv(netdev);
3043 struct e1000_hw *hw = &adapter->hw;
3045 /* disable interrupts from the "other" bit, avoid re-entry */
3046 wr32(E1000_EIMC, E1000_EIMS_OTHER);
3048 eicr = rd32(E1000_EICR);
3050 if (eicr & E1000_EIMS_OTHER) {
3051 u32 icr = rd32(E1000_ICR);
3052 /* reading ICR causes bit 31 of EICR to be cleared */
3053 if (!(icr & E1000_ICR_LSC))
3054 goto no_link_interrupt;
3055 hw->mac.get_link_status = 1;
3056 /* guard against interrupt when we're going down */
3057 if (!test_bit(__IGB_DOWN, &adapter->state))
3058 mod_timer(&adapter->watchdog_timer, jiffies + 1);
3062 wr32(E1000_IMS, E1000_IMS_LSC);
3063 wr32(E1000_EIMS, E1000_EIMS_OTHER);
3068 static irqreturn_t igb_msix_tx(int irq, void *data)
3070 struct igb_ring *tx_ring = data;
3071 struct igb_adapter *adapter = tx_ring->adapter;
3072 struct e1000_hw *hw = &adapter->hw;
3074 if (!tx_ring->itr_val)
3075 wr32(E1000_EIMC, tx_ring->eims_value);
3077 tx_ring->total_bytes = 0;
3078 tx_ring->total_packets = 0;
3079 if (!igb_clean_tx_irq(adapter, tx_ring))
3080 /* Ring was not completely cleaned, so fire another interrupt */
3081 wr32(E1000_EICS, tx_ring->eims_value);
3083 if (!tx_ring->itr_val)
3084 wr32(E1000_EIMS, tx_ring->eims_value);
3088 static irqreturn_t igb_msix_rx(int irq, void *data)
3090 struct igb_ring *rx_ring = data;
3091 struct igb_adapter *adapter = rx_ring->adapter;
3092 struct e1000_hw *hw = &adapter->hw;
3094 if (!rx_ring->itr_val)
3095 wr32(E1000_EIMC, rx_ring->eims_value);
3097 if (netif_rx_schedule_prep(adapter->netdev, &rx_ring->napi)) {
3098 rx_ring->total_bytes = 0;
3099 rx_ring->total_packets = 0;
3100 rx_ring->no_itr_adjust = 0;
3101 __netif_rx_schedule(adapter->netdev, &rx_ring->napi);
3103 if (!rx_ring->no_itr_adjust) {
3104 igb_lower_rx_eitr(adapter, rx_ring);
3105 rx_ring->no_itr_adjust = 1;
3114 * igb_intr_msi - Interrupt Handler
3115 * @irq: interrupt number
3116 * @data: pointer to a network interface device structure
3118 static irqreturn_t igb_intr_msi(int irq, void *data)
3120 struct net_device *netdev = data;
3121 struct igb_adapter *adapter = netdev_priv(netdev);
3122 struct napi_struct *napi = &adapter->napi;
3123 struct e1000_hw *hw = &adapter->hw;
3124 /* read ICR disables interrupts using IAM */
3125 u32 icr = rd32(E1000_ICR);
3127 /* Write the ITR value calculated at the end of the
3128 * previous interrupt.
3130 if (adapter->set_itr) {
3132 1000000000 / (adapter->itr * 256));
3133 adapter->set_itr = 0;
3136 /* read ICR disables interrupts using IAM */
3137 if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
3138 hw->mac.get_link_status = 1;
3139 if (!test_bit(__IGB_DOWN, &adapter->state))
3140 mod_timer(&adapter->watchdog_timer, jiffies + 1);
3143 if (netif_rx_schedule_prep(netdev, napi)) {
3144 adapter->tx_ring->total_bytes = 0;
3145 adapter->tx_ring->total_packets = 0;
3146 adapter->rx_ring->total_bytes = 0;
3147 adapter->rx_ring->total_packets = 0;
3148 __netif_rx_schedule(netdev, napi);
3155 * igb_intr - Interrupt Handler
3156 * @irq: interrupt number
3157 * @data: pointer to a network interface device structure
3159 static irqreturn_t igb_intr(int irq, void *data)
3161 struct net_device *netdev = data;
3162 struct igb_adapter *adapter = netdev_priv(netdev);
3163 struct napi_struct *napi = &adapter->napi;
3164 struct e1000_hw *hw = &adapter->hw;
3165 /* Interrupt Auto-Mask...upon reading ICR, interrupts are masked. No
3166 * need for the IMC write */
3167 u32 icr = rd32(E1000_ICR);
3170 return IRQ_NONE; /* Not our interrupt */
3172 /* Write the ITR value calculated at the end of the
3173 * previous interrupt.
3175 if (adapter->set_itr) {
3177 1000000000 / (adapter->itr * 256));
3178 adapter->set_itr = 0;
3181 /* IMS will not auto-mask if INT_ASSERTED is not set, and if it is
3182 * not set, then the adapter didn't send an interrupt */
3183 if (!(icr & E1000_ICR_INT_ASSERTED))
3186 eicr = rd32(E1000_EICR);
3188 if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
3189 hw->mac.get_link_status = 1;
3190 /* guard against interrupt when we're going down */
3191 if (!test_bit(__IGB_DOWN, &adapter->state))
3192 mod_timer(&adapter->watchdog_timer, jiffies + 1);
3195 if (netif_rx_schedule_prep(netdev, napi)) {
3196 adapter->tx_ring->total_bytes = 0;
3197 adapter->rx_ring->total_bytes = 0;
3198 adapter->tx_ring->total_packets = 0;
3199 adapter->rx_ring->total_packets = 0;
3200 __netif_rx_schedule(netdev, napi);
3207 * igb_clean - NAPI Rx polling callback
3208 * @adapter: board private structure
3210 static int igb_clean(struct napi_struct *napi, int budget)
3212 struct igb_adapter *adapter = container_of(napi, struct igb_adapter,
3214 struct net_device *netdev = adapter->netdev;
3215 int tx_clean_complete = 1, work_done = 0;
3218 /* Must NOT use netdev_priv macro here. */
3219 adapter = netdev->priv;
3221 /* Keep link state information with original netdev */
3222 if (!netif_carrier_ok(netdev))
3225 /* igb_clean is called per-cpu. This lock protects tx_ring[i] from
3226 * being cleaned by multiple cpus simultaneously. A failure obtaining
3227 * the lock means tx_ring[i] is currently being cleaned anyway. */
3228 for (i = 0; i < adapter->num_tx_queues; i++) {
3229 if (spin_trylock(&adapter->tx_ring[i].tx_clean_lock)) {
3230 tx_clean_complete &= igb_clean_tx_irq(adapter,
3231 &adapter->tx_ring[i]);
3232 spin_unlock(&adapter->tx_ring[i].tx_clean_lock);
3236 for (i = 0; i < adapter->num_rx_queues; i++)
3237 igb_clean_rx_irq_adv(adapter, &adapter->rx_ring[i], &work_done,
3238 adapter->rx_ring[i].napi.weight);
3240 /* If no Tx and not enough Rx work done, exit the polling mode */
3241 if ((tx_clean_complete && (work_done < budget)) ||
3242 !netif_running(netdev)) {
3244 if (adapter->itr_setting & 3)
3245 igb_set_itr(adapter, E1000_ITR, false);
3246 netif_rx_complete(netdev, napi);
3247 if (!test_bit(__IGB_DOWN, &adapter->state))
3248 igb_irq_enable(adapter);
3255 static int igb_clean_rx_ring_msix(struct napi_struct *napi, int budget)
3257 struct igb_ring *rx_ring = container_of(napi, struct igb_ring, napi);
3258 struct igb_adapter *adapter = rx_ring->adapter;
3259 struct e1000_hw *hw = &adapter->hw;
3260 struct net_device *netdev = adapter->netdev;
3263 /* Keep link state information with original netdev */
3264 if (!netif_carrier_ok(netdev))
3267 igb_clean_rx_irq_adv(adapter, rx_ring, &work_done, budget);
3270 /* If not enough Rx work done, exit the polling mode */
3271 if ((work_done == 0) || !netif_running(netdev)) {
3273 netif_rx_complete(netdev, napi);
3275 wr32(E1000_EIMS, rx_ring->eims_value);
3276 if ((adapter->itr_setting & 3) && !rx_ring->no_itr_adjust &&
3277 (rx_ring->total_packets > IGB_DYN_ITR_PACKET_THRESHOLD)) {
3278 int mean_size = rx_ring->total_bytes /
3279 rx_ring->total_packets;
3280 if (mean_size < IGB_DYN_ITR_LENGTH_LOW)
3281 igb_raise_rx_eitr(adapter, rx_ring);
3282 else if (mean_size > IGB_DYN_ITR_LENGTH_HIGH)
3283 igb_lower_rx_eitr(adapter, rx_ring);
3291 static inline u32 get_head(struct igb_ring *tx_ring)
3293 void *end = (struct e1000_tx_desc *)tx_ring->desc + tx_ring->count;
3294 return le32_to_cpu(*(volatile __le32 *)end);
3298 * igb_clean_tx_irq - Reclaim resources after transmit completes
3299 * @adapter: board private structure
3300 * returns true if ring is completely cleaned
3302 static bool igb_clean_tx_irq(struct igb_adapter *adapter,
3303 struct igb_ring *tx_ring)
3305 struct net_device *netdev = adapter->netdev;
3306 struct e1000_hw *hw = &adapter->hw;
3307 struct e1000_tx_desc *tx_desc;
3308 struct igb_buffer *buffer_info;
3309 struct sk_buff *skb;
3312 unsigned int count = 0;
3313 bool cleaned = false;
3315 unsigned int total_bytes = 0, total_packets = 0;
3318 head = get_head(tx_ring);
3319 i = tx_ring->next_to_clean;
3323 tx_desc = E1000_TX_DESC(*tx_ring, i);
3324 buffer_info = &tx_ring->buffer_info[i];
3325 skb = buffer_info->skb;
3328 unsigned int segs, bytecount;
3329 /* gso_segs is currently only valid for tcp */
3330 segs = skb_shinfo(skb)->gso_segs ?: 1;
3331 /* multiply data chunks by size of headers */
3332 bytecount = ((segs - 1) * skb_headlen(skb)) +
3334 total_packets += segs;
3335 total_bytes += bytecount;
3338 igb_unmap_and_free_tx_resource(adapter, buffer_info);
3339 tx_desc->upper.data = 0;
3342 if (i == tx_ring->count)
3346 if (count == IGB_MAX_TX_CLEAN) {
3353 head = get_head(tx_ring);
3354 if (head == oldhead)
3359 tx_ring->next_to_clean = i;
3361 if (unlikely(cleaned &&
3362 netif_carrier_ok(netdev) &&
3363 IGB_DESC_UNUSED(tx_ring) >= IGB_TX_QUEUE_WAKE)) {
3364 /* Make sure that anybody stopping the queue after this
3365 * sees the new next_to_clean.
3368 if (netif_queue_stopped(netdev) &&
3369 !(test_bit(__IGB_DOWN, &adapter->state))) {
3370 netif_wake_queue(netdev);
3371 ++adapter->restart_queue;
3375 if (tx_ring->detect_tx_hung) {
3376 /* Detect a transmit hang in hardware, this serializes the
3377 * check with the clearing of time_stamp and movement of i */
3378 tx_ring->detect_tx_hung = false;
3379 if (tx_ring->buffer_info[i].time_stamp &&
3380 time_after(jiffies, tx_ring->buffer_info[i].time_stamp +
3381 (adapter->tx_timeout_factor * HZ))
3382 && !(rd32(E1000_STATUS) &
3383 E1000_STATUS_TXOFF)) {
3385 tx_desc = E1000_TX_DESC(*tx_ring, i);
3386 /* detected Tx unit hang */
3387 dev_err(&adapter->pdev->dev,
3388 "Detected Tx Unit Hang\n"
3392 " next_to_use <%x>\n"
3393 " next_to_clean <%x>\n"
3395 "buffer_info[next_to_clean]\n"
3396 " time_stamp <%lx>\n"
3398 " desc.status <%x>\n",
3399 (unsigned long)((tx_ring - adapter->tx_ring) /
3400 sizeof(struct igb_ring)),
3401 readl(adapter->hw.hw_addr + tx_ring->head),
3402 readl(adapter->hw.hw_addr + tx_ring->tail),
3403 tx_ring->next_to_use,
3404 tx_ring->next_to_clean,
3406 tx_ring->buffer_info[i].time_stamp,
3408 tx_desc->upper.fields.status);
3409 netif_stop_queue(netdev);
3412 tx_ring->total_bytes += total_bytes;
3413 tx_ring->total_packets += total_packets;
3414 adapter->net_stats.tx_bytes += total_bytes;
3415 adapter->net_stats.tx_packets += total_packets;
3421 * igb_receive_skb - helper function to handle rx indications
3422 * @adapter: board private structure
3423 * @status: descriptor status field as written by hardware
3424 * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
3425 * @skb: pointer to sk_buff to be indicated to stack
3427 static void igb_receive_skb(struct igb_adapter *adapter, u8 status, __le16 vlan,
3428 struct sk_buff *skb)
3430 if (adapter->vlgrp && (status & E1000_RXD_STAT_VP))
3431 vlan_hwaccel_receive_skb(skb, adapter->vlgrp,
3433 E1000_RXD_SPC_VLAN_MASK);
3435 netif_receive_skb(skb);
3439 static inline void igb_rx_checksum_adv(struct igb_adapter *adapter,
3440 u32 status_err, struct sk_buff *skb)
3442 skb->ip_summed = CHECKSUM_NONE;
3444 /* Ignore Checksum bit is set or checksum is disabled through ethtool */
3445 if ((status_err & E1000_RXD_STAT_IXSM) || !adapter->rx_csum)
3447 /* TCP/UDP checksum error bit is set */
3449 (E1000_RXDEXT_STATERR_TCPE | E1000_RXDEXT_STATERR_IPE)) {
3450 /* let the stack verify checksum errors */
3451 adapter->hw_csum_err++;
3454 /* It must be a TCP or UDP packet with a valid checksum */
3455 if (status_err & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS))
3456 skb->ip_summed = CHECKSUM_UNNECESSARY;
3458 adapter->hw_csum_good++;
3461 static bool igb_clean_rx_irq_adv(struct igb_adapter *adapter,
3462 struct igb_ring *rx_ring,
3463 int *work_done, int budget)
3465 struct net_device *netdev = adapter->netdev;
3466 struct pci_dev *pdev = adapter->pdev;
3467 union e1000_adv_rx_desc *rx_desc , *next_rxd;
3468 struct igb_buffer *buffer_info , *next_buffer;
3469 struct sk_buff *skb;
3471 u32 length, hlen, staterr;
3472 bool cleaned = false;
3473 int cleaned_count = 0;
3474 unsigned int total_bytes = 0, total_packets = 0;
3476 i = rx_ring->next_to_clean;
3477 rx_desc = E1000_RX_DESC_ADV(*rx_ring, i);
3478 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
3480 while (staterr & E1000_RXD_STAT_DD) {
3481 if (*work_done >= budget)
3484 buffer_info = &rx_ring->buffer_info[i];
3486 /* HW will not DMA in data larger than the given buffer, even
3487 * if it parses the (NFS, of course) header to be larger. In
3488 * that case, it fills the header buffer and spills the rest
3491 hlen = (le16_to_cpu(rx_desc->wb.lower.lo_dword.hdr_info) &
3492 E1000_RXDADV_HDRBUFLEN_MASK) >> E1000_RXDADV_HDRBUFLEN_SHIFT;
3493 if (hlen > adapter->rx_ps_hdr_size)
3494 hlen = adapter->rx_ps_hdr_size;
3496 length = le16_to_cpu(rx_desc->wb.upper.length);
3500 if (rx_ring->pending_skb != NULL) {
3501 skb = rx_ring->pending_skb;
3502 rx_ring->pending_skb = NULL;
3503 j = rx_ring->pending_skb_page;
3505 skb = buffer_info->skb;
3506 prefetch(skb->data - NET_IP_ALIGN);
3507 buffer_info->skb = NULL;
3509 pci_unmap_single(pdev, buffer_info->dma,
3510 adapter->rx_ps_hdr_size +
3512 PCI_DMA_FROMDEVICE);
3515 pci_unmap_single(pdev, buffer_info->dma,
3516 adapter->rx_buffer_len +
3518 PCI_DMA_FROMDEVICE);
3519 skb_put(skb, length);
3526 pci_unmap_page(pdev, buffer_info->page_dma,
3527 PAGE_SIZE, PCI_DMA_FROMDEVICE);
3528 buffer_info->page_dma = 0;
3529 skb_fill_page_desc(skb, j, buffer_info->page,
3531 buffer_info->page = NULL;
3534 skb->data_len += length;
3535 skb->truesize += length;
3536 rx_desc->wb.upper.status_error = 0;
3537 if (staterr & E1000_RXD_STAT_EOP)
3543 if (i == rx_ring->count)
3546 buffer_info = &rx_ring->buffer_info[i];
3547 rx_desc = E1000_RX_DESC_ADV(*rx_ring, i);
3548 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
3549 length = le16_to_cpu(rx_desc->wb.upper.length);
3550 if (!(staterr & E1000_RXD_STAT_DD)) {
3551 rx_ring->pending_skb = skb;
3552 rx_ring->pending_skb_page = j;
3557 pskb_trim(skb, skb->len - 4);
3559 if (i == rx_ring->count)
3561 next_rxd = E1000_RX_DESC_ADV(*rx_ring, i);
3563 next_buffer = &rx_ring->buffer_info[i];
3565 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
3566 dev_kfree_skb_irq(skb);
3569 rx_ring->no_itr_adjust |= (staterr & E1000_RXD_STAT_DYNINT);
3571 total_bytes += skb->len;
3574 igb_rx_checksum_adv(adapter, staterr, skb);
3576 skb->protocol = eth_type_trans(skb, netdev);
3578 igb_receive_skb(adapter, staterr, rx_desc->wb.upper.vlan, skb);
3580 netdev->last_rx = jiffies;
3583 rx_desc->wb.upper.status_error = 0;
3585 /* return some buffers to hardware, one at a time is too slow */
3586 if (cleaned_count >= IGB_RX_BUFFER_WRITE) {
3587 igb_alloc_rx_buffers_adv(adapter, rx_ring,
3592 /* use prefetched values */
3594 buffer_info = next_buffer;
3596 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
3599 rx_ring->next_to_clean = i;
3600 cleaned_count = IGB_DESC_UNUSED(rx_ring);
3603 igb_alloc_rx_buffers_adv(adapter, rx_ring, cleaned_count);
3605 rx_ring->total_packets += total_packets;
3606 rx_ring->total_bytes += total_bytes;
3607 rx_ring->rx_stats.packets += total_packets;
3608 rx_ring->rx_stats.bytes += total_bytes;
3609 adapter->net_stats.rx_bytes += total_bytes;
3610 adapter->net_stats.rx_packets += total_packets;
3616 * igb_alloc_rx_buffers_adv - Replace used receive buffers; packet split
3617 * @adapter: address of board private structure
3619 static void igb_alloc_rx_buffers_adv(struct igb_adapter *adapter,
3620 struct igb_ring *rx_ring,
3623 struct net_device *netdev = adapter->netdev;
3624 struct pci_dev *pdev = adapter->pdev;
3625 union e1000_adv_rx_desc *rx_desc;
3626 struct igb_buffer *buffer_info;
3627 struct sk_buff *skb;
3630 i = rx_ring->next_to_use;
3631 buffer_info = &rx_ring->buffer_info[i];
3633 while (cleaned_count--) {
3634 rx_desc = E1000_RX_DESC_ADV(*rx_ring, i);
3636 if (adapter->rx_ps_hdr_size && !buffer_info->page) {
3637 buffer_info->page = alloc_page(GFP_ATOMIC);
3638 if (!buffer_info->page) {
3639 adapter->alloc_rx_buff_failed++;
3642 buffer_info->page_dma =
3646 PCI_DMA_FROMDEVICE);
3649 if (!buffer_info->skb) {
3652 if (adapter->rx_ps_hdr_size)
3653 bufsz = adapter->rx_ps_hdr_size;
3655 bufsz = adapter->rx_buffer_len;
3656 bufsz += NET_IP_ALIGN;
3657 skb = netdev_alloc_skb(netdev, bufsz);
3660 adapter->alloc_rx_buff_failed++;
3664 /* Make buffer alignment 2 beyond a 16 byte boundary
3665 * this will result in a 16 byte aligned IP header after
3666 * the 14 byte MAC header is removed
3668 skb_reserve(skb, NET_IP_ALIGN);
3670 buffer_info->skb = skb;
3671 buffer_info->dma = pci_map_single(pdev, skb->data,
3673 PCI_DMA_FROMDEVICE);
3676 /* Refresh the desc even if buffer_addrs didn't change because
3677 * each write-back erases this info. */
3678 if (adapter->rx_ps_hdr_size) {
3679 rx_desc->read.pkt_addr =
3680 cpu_to_le64(buffer_info->page_dma);
3681 rx_desc->read.hdr_addr = cpu_to_le64(buffer_info->dma);
3683 rx_desc->read.pkt_addr =
3684 cpu_to_le64(buffer_info->dma);
3685 rx_desc->read.hdr_addr = 0;
3689 if (i == rx_ring->count)
3691 buffer_info = &rx_ring->buffer_info[i];
3695 if (rx_ring->next_to_use != i) {
3696 rx_ring->next_to_use = i;
3698 i = (rx_ring->count - 1);
3702 /* Force memory writes to complete before letting h/w
3703 * know there are new descriptors to fetch. (Only
3704 * applicable for weak-ordered memory model archs,
3705 * such as IA-64). */
3707 writel(i, adapter->hw.hw_addr + rx_ring->tail);
3717 static int igb_mii_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
3719 struct igb_adapter *adapter = netdev_priv(netdev);
3720 struct mii_ioctl_data *data = if_mii(ifr);
3722 if (adapter->hw.phy.media_type != e1000_media_type_copper)
3727 data->phy_id = adapter->hw.phy.addr;
3730 if (!capable(CAP_NET_ADMIN))
3732 if (adapter->hw.phy.ops.read_phy_reg(&adapter->hw,
3734 & 0x1F, &data->val_out))
3750 static int igb_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
3756 return igb_mii_ioctl(netdev, ifr, cmd);
3762 static void igb_vlan_rx_register(struct net_device *netdev,
3763 struct vlan_group *grp)
3765 struct igb_adapter *adapter = netdev_priv(netdev);
3766 struct e1000_hw *hw = &adapter->hw;
3769 igb_irq_disable(adapter);
3770 adapter->vlgrp = grp;
3773 /* enable VLAN tag insert/strip */
3774 ctrl = rd32(E1000_CTRL);
3775 ctrl |= E1000_CTRL_VME;
3776 wr32(E1000_CTRL, ctrl);
3778 /* enable VLAN receive filtering */
3779 rctl = rd32(E1000_RCTL);
3780 rctl |= E1000_RCTL_VFE;
3781 rctl &= ~E1000_RCTL_CFIEN;
3782 wr32(E1000_RCTL, rctl);
3783 igb_update_mng_vlan(adapter);
3785 adapter->max_frame_size + VLAN_TAG_SIZE);
3787 /* disable VLAN tag insert/strip */
3788 ctrl = rd32(E1000_CTRL);
3789 ctrl &= ~E1000_CTRL_VME;
3790 wr32(E1000_CTRL, ctrl);
3792 /* disable VLAN filtering */
3793 rctl = rd32(E1000_RCTL);
3794 rctl &= ~E1000_RCTL_VFE;
3795 wr32(E1000_RCTL, rctl);
3796 if (adapter->mng_vlan_id != (u16)IGB_MNG_VLAN_NONE) {
3797 igb_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
3798 adapter->mng_vlan_id = IGB_MNG_VLAN_NONE;
3801 adapter->max_frame_size);
3804 if (!test_bit(__IGB_DOWN, &adapter->state))
3805 igb_irq_enable(adapter);
3808 static void igb_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
3810 struct igb_adapter *adapter = netdev_priv(netdev);
3811 struct e1000_hw *hw = &adapter->hw;
3814 if ((adapter->hw.mng_cookie.status &
3815 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
3816 (vid == adapter->mng_vlan_id))
3818 /* add VID to filter table */
3819 index = (vid >> 5) & 0x7F;
3820 vfta = array_rd32(E1000_VFTA, index);
3821 vfta |= (1 << (vid & 0x1F));
3822 igb_write_vfta(&adapter->hw, index, vfta);
3825 static void igb_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
3827 struct igb_adapter *adapter = netdev_priv(netdev);
3828 struct e1000_hw *hw = &adapter->hw;
3831 igb_irq_disable(adapter);
3832 vlan_group_set_device(adapter->vlgrp, vid, NULL);
3834 if (!test_bit(__IGB_DOWN, &adapter->state))
3835 igb_irq_enable(adapter);
3837 if ((adapter->hw.mng_cookie.status &
3838 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
3839 (vid == adapter->mng_vlan_id)) {
3840 /* release control to f/w */
3841 igb_release_hw_control(adapter);
3845 /* remove VID from filter table */
3846 index = (vid >> 5) & 0x7F;
3847 vfta = array_rd32(E1000_VFTA, index);
3848 vfta &= ~(1 << (vid & 0x1F));
3849 igb_write_vfta(&adapter->hw, index, vfta);
3852 static void igb_restore_vlan(struct igb_adapter *adapter)
3854 igb_vlan_rx_register(adapter->netdev, adapter->vlgrp);
3856 if (adapter->vlgrp) {
3858 for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
3859 if (!vlan_group_get_device(adapter->vlgrp, vid))
3861 igb_vlan_rx_add_vid(adapter->netdev, vid);
3866 int igb_set_spd_dplx(struct igb_adapter *adapter, u16 spddplx)
3868 struct e1000_mac_info *mac = &adapter->hw.mac;
3872 /* Fiber NICs only allow 1000 gbps Full duplex */
3873 if ((adapter->hw.phy.media_type == e1000_media_type_fiber) &&
3874 spddplx != (SPEED_1000 + DUPLEX_FULL)) {
3875 dev_err(&adapter->pdev->dev,
3876 "Unsupported Speed/Duplex configuration\n");
3881 case SPEED_10 + DUPLEX_HALF:
3882 mac->forced_speed_duplex = ADVERTISE_10_HALF;
3884 case SPEED_10 + DUPLEX_FULL:
3885 mac->forced_speed_duplex = ADVERTISE_10_FULL;
3887 case SPEED_100 + DUPLEX_HALF:
3888 mac->forced_speed_duplex = ADVERTISE_100_HALF;
3890 case SPEED_100 + DUPLEX_FULL:
3891 mac->forced_speed_duplex = ADVERTISE_100_FULL;
3893 case SPEED_1000 + DUPLEX_FULL:
3895 adapter->hw.phy.autoneg_advertised = ADVERTISE_1000_FULL;
3897 case SPEED_1000 + DUPLEX_HALF: /* not supported */
3899 dev_err(&adapter->pdev->dev,
3900 "Unsupported Speed/Duplex configuration\n");
3907 static int igb_suspend(struct pci_dev *pdev, pm_message_t state)
3909 struct net_device *netdev = pci_get_drvdata(pdev);
3910 struct igb_adapter *adapter = netdev_priv(netdev);
3911 struct e1000_hw *hw = &adapter->hw;
3912 u32 ctrl, ctrl_ext, rctl, status;
3913 u32 wufc = adapter->wol;
3918 netif_device_detach(netdev);
3920 if (netif_running(netdev)) {
3921 WARN_ON(test_bit(__IGB_RESETTING, &adapter->state));
3923 igb_free_irq(adapter);
3927 retval = pci_save_state(pdev);
3932 status = rd32(E1000_STATUS);
3933 if (status & E1000_STATUS_LU)
3934 wufc &= ~E1000_WUFC_LNKC;
3937 igb_setup_rctl(adapter);
3938 igb_set_multi(netdev);
3940 /* turn on all-multi mode if wake on multicast is enabled */
3941 if (wufc & E1000_WUFC_MC) {
3942 rctl = rd32(E1000_RCTL);
3943 rctl |= E1000_RCTL_MPE;
3944 wr32(E1000_RCTL, rctl);
3947 ctrl = rd32(E1000_CTRL);
3948 /* advertise wake from D3Cold */
3949 #define E1000_CTRL_ADVD3WUC 0x00100000
3950 /* phy power management enable */
3951 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
3952 ctrl |= E1000_CTRL_ADVD3WUC;
3953 wr32(E1000_CTRL, ctrl);
3955 if (adapter->hw.phy.media_type == e1000_media_type_fiber ||
3956 adapter->hw.phy.media_type ==
3957 e1000_media_type_internal_serdes) {
3958 /* keep the laser running in D3 */
3959 ctrl_ext = rd32(E1000_CTRL_EXT);
3960 ctrl_ext |= E1000_CTRL_EXT_SDP7_DATA;
3961 wr32(E1000_CTRL_EXT, ctrl_ext);
3964 /* Allow time for pending master requests to run */
3965 igb_disable_pcie_master(&adapter->hw);
3967 wr32(E1000_WUC, E1000_WUC_PME_EN);
3968 wr32(E1000_WUFC, wufc);
3969 pci_enable_wake(pdev, PCI_D3hot, 1);
3970 pci_enable_wake(pdev, PCI_D3cold, 1);
3973 wr32(E1000_WUFC, 0);
3974 pci_enable_wake(pdev, PCI_D3hot, 0);
3975 pci_enable_wake(pdev, PCI_D3cold, 0);
3978 /* make sure adapter isn't asleep if manageability is enabled */
3979 if (adapter->en_mng_pt) {
3980 pci_enable_wake(pdev, PCI_D3hot, 1);
3981 pci_enable_wake(pdev, PCI_D3cold, 1);
3984 /* Release control of h/w to f/w. If f/w is AMT enabled, this
3985 * would have already happened in close and is redundant. */
3986 igb_release_hw_control(adapter);
3988 pci_disable_device(pdev);
3990 pci_set_power_state(pdev, pci_choose_state(pdev, state));
3996 static int igb_resume(struct pci_dev *pdev)
3998 struct net_device *netdev = pci_get_drvdata(pdev);
3999 struct igb_adapter *adapter = netdev_priv(netdev);
4000 struct e1000_hw *hw = &adapter->hw;
4003 pci_set_power_state(pdev, PCI_D0);
4004 pci_restore_state(pdev);
4006 if (adapter->need_ioport)
4007 err = pci_enable_device(pdev);
4009 err = pci_enable_device_mem(pdev);
4012 "igb: Cannot enable PCI device from suspend\n");
4015 pci_set_master(pdev);
4017 pci_enable_wake(pdev, PCI_D3hot, 0);
4018 pci_enable_wake(pdev, PCI_D3cold, 0);
4020 if (netif_running(netdev)) {
4021 err = igb_request_irq(adapter);
4026 /* e1000_power_up_phy(adapter); */
4029 wr32(E1000_WUS, ~0);
4031 igb_init_manageability(adapter);
4033 if (netif_running(netdev))
4036 netif_device_attach(netdev);
4038 /* let the f/w know that the h/w is now under the control of the
4040 igb_get_hw_control(adapter);
4046 static void igb_shutdown(struct pci_dev *pdev)
4048 igb_suspend(pdev, PMSG_SUSPEND);
4051 #ifdef CONFIG_NET_POLL_CONTROLLER
4053 * Polling 'interrupt' - used by things like netconsole to send skbs
4054 * without having to re-enable interrupts. It's not called while
4055 * the interrupt routine is executing.
4057 static void igb_netpoll(struct net_device *netdev)
4059 struct igb_adapter *adapter = netdev_priv(netdev);
4063 igb_irq_disable(adapter);
4064 for (i = 0; i < adapter->num_tx_queues; i++)
4065 igb_clean_tx_irq(adapter, &adapter->tx_ring[i]);
4067 for (i = 0; i < adapter->num_rx_queues; i++)
4068 igb_clean_rx_irq_adv(adapter, &adapter->rx_ring[i],
4070 adapter->rx_ring[i].napi.weight);
4072 igb_irq_enable(adapter);
4074 #endif /* CONFIG_NET_POLL_CONTROLLER */
4077 * igb_io_error_detected - called when PCI error is detected
4078 * @pdev: Pointer to PCI device
4079 * @state: The current pci connection state
4081 * This function is called after a PCI bus error affecting
4082 * this device has been detected.
4084 static pci_ers_result_t igb_io_error_detected(struct pci_dev *pdev,
4085 pci_channel_state_t state)
4087 struct net_device *netdev = pci_get_drvdata(pdev);
4088 struct igb_adapter *adapter = netdev_priv(netdev);
4090 netif_device_detach(netdev);
4092 if (netif_running(netdev))
4094 pci_disable_device(pdev);
4096 /* Request a slot slot reset. */
4097 return PCI_ERS_RESULT_NEED_RESET;
4101 * igb_io_slot_reset - called after the pci bus has been reset.
4102 * @pdev: Pointer to PCI device
4104 * Restart the card from scratch, as if from a cold-boot. Implementation
4105 * resembles the first-half of the igb_resume routine.
4107 static pci_ers_result_t igb_io_slot_reset(struct pci_dev *pdev)
4109 struct net_device *netdev = pci_get_drvdata(pdev);
4110 struct igb_adapter *adapter = netdev_priv(netdev);
4111 struct e1000_hw *hw = &adapter->hw;
4114 if (adapter->need_ioport)
4115 err = pci_enable_device(pdev);
4117 err = pci_enable_device_mem(pdev);
4120 "Cannot re-enable PCI device after reset.\n");
4121 return PCI_ERS_RESULT_DISCONNECT;
4123 pci_set_master(pdev);
4124 pci_restore_state(pdev);
4126 pci_enable_wake(pdev, PCI_D3hot, 0);
4127 pci_enable_wake(pdev, PCI_D3cold, 0);
4130 wr32(E1000_WUS, ~0);
4132 return PCI_ERS_RESULT_RECOVERED;
4136 * igb_io_resume - called when traffic can start flowing again.
4137 * @pdev: Pointer to PCI device
4139 * This callback is called when the error recovery driver tells us that
4140 * its OK to resume normal operation. Implementation resembles the
4141 * second-half of the igb_resume routine.
4143 static void igb_io_resume(struct pci_dev *pdev)
4145 struct net_device *netdev = pci_get_drvdata(pdev);
4146 struct igb_adapter *adapter = netdev_priv(netdev);
4148 igb_init_manageability(adapter);
4150 if (netif_running(netdev)) {
4151 if (igb_up(adapter)) {
4152 dev_err(&pdev->dev, "igb_up failed after reset\n");
4157 netif_device_attach(netdev);
4159 /* let the f/w know that the h/w is now under the control of the
4161 igb_get_hw_control(adapter);