Merge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-2.6
[linux-2.6] / drivers / net / igb / igb_main.c
1 /*******************************************************************************
2
3   Intel(R) Gigabit Ethernet Linux driver
4   Copyright(c) 2007 Intel Corporation.
5
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.
9
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
13   more details.
14
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.
18
19   The full GNU General Public License is included in this distribution in
20   the file called "COPYING".
21
22   Contact Information:
23   e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
24   Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
25
26 *******************************************************************************/
27
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/pci-aspm.h>
42 #include <linux/delay.h>
43 #include <linux/interrupt.h>
44 #include <linux/if_ether.h>
45 #include <linux/aer.h>
46 #ifdef CONFIG_IGB_DCA
47 #include <linux/dca.h>
48 #endif
49 #include "igb.h"
50
51 #define DRV_VERSION "1.2.45-k2"
52 char igb_driver_name[] = "igb";
53 char igb_driver_version[] = DRV_VERSION;
54 static const char igb_driver_string[] =
55                                 "Intel(R) Gigabit Ethernet Network Driver";
56 static const char igb_copyright[] = "Copyright (c) 2008 Intel Corporation.";
57
58 static const struct e1000_info *igb_info_tbl[] = {
59         [board_82575] = &e1000_82575_info,
60 };
61
62 static struct pci_device_id igb_pci_tbl[] = {
63         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576), board_82575 },
64         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_FIBER), board_82575 },
65         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_SERDES), board_82575 },
66         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575EB_COPPER), board_82575 },
67         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575EB_FIBER_SERDES), board_82575 },
68         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575GB_QUAD_COPPER), board_82575 },
69         /* required last entry */
70         {0, }
71 };
72
73 MODULE_DEVICE_TABLE(pci, igb_pci_tbl);
74
75 void igb_reset(struct igb_adapter *);
76 static int igb_setup_all_tx_resources(struct igb_adapter *);
77 static int igb_setup_all_rx_resources(struct igb_adapter *);
78 static void igb_free_all_tx_resources(struct igb_adapter *);
79 static void igb_free_all_rx_resources(struct igb_adapter *);
80 void igb_update_stats(struct igb_adapter *);
81 static int igb_probe(struct pci_dev *, const struct pci_device_id *);
82 static void __devexit igb_remove(struct pci_dev *pdev);
83 static int igb_sw_init(struct igb_adapter *);
84 static int igb_open(struct net_device *);
85 static int igb_close(struct net_device *);
86 static void igb_configure_tx(struct igb_adapter *);
87 static void igb_configure_rx(struct igb_adapter *);
88 static void igb_setup_rctl(struct igb_adapter *);
89 static void igb_clean_all_tx_rings(struct igb_adapter *);
90 static void igb_clean_all_rx_rings(struct igb_adapter *);
91 static void igb_clean_tx_ring(struct igb_ring *);
92 static void igb_clean_rx_ring(struct igb_ring *);
93 static void igb_set_multi(struct net_device *);
94 static void igb_update_phy_info(unsigned long);
95 static void igb_watchdog(unsigned long);
96 static void igb_watchdog_task(struct work_struct *);
97 static int igb_xmit_frame_ring_adv(struct sk_buff *, struct net_device *,
98                                   struct igb_ring *);
99 static int igb_xmit_frame_adv(struct sk_buff *skb, struct net_device *);
100 static struct net_device_stats *igb_get_stats(struct net_device *);
101 static int igb_change_mtu(struct net_device *, int);
102 static int igb_set_mac(struct net_device *, void *);
103 static irqreturn_t igb_intr(int irq, void *);
104 static irqreturn_t igb_intr_msi(int irq, void *);
105 static irqreturn_t igb_msix_other(int irq, void *);
106 static irqreturn_t igb_msix_rx(int irq, void *);
107 static irqreturn_t igb_msix_tx(int irq, void *);
108 static int igb_clean_rx_ring_msix(struct napi_struct *, int);
109 #ifdef CONFIG_IGB_DCA
110 static void igb_update_rx_dca(struct igb_ring *);
111 static void igb_update_tx_dca(struct igb_ring *);
112 static void igb_setup_dca(struct igb_adapter *);
113 #endif /* CONFIG_IGB_DCA */
114 static bool igb_clean_tx_irq(struct igb_ring *);
115 static int igb_poll(struct napi_struct *, int);
116 static bool igb_clean_rx_irq_adv(struct igb_ring *, int *, int);
117 static void igb_alloc_rx_buffers_adv(struct igb_ring *, int);
118 #ifdef CONFIG_IGB_LRO
119 static int igb_get_skb_hdr(struct sk_buff *skb, void **, void **, u64 *, void *);
120 #endif
121 static int igb_ioctl(struct net_device *, struct ifreq *, int cmd);
122 static void igb_tx_timeout(struct net_device *);
123 static void igb_reset_task(struct work_struct *);
124 static void igb_vlan_rx_register(struct net_device *, struct vlan_group *);
125 static void igb_vlan_rx_add_vid(struct net_device *, u16);
126 static void igb_vlan_rx_kill_vid(struct net_device *, u16);
127 static void igb_restore_vlan(struct igb_adapter *);
128
129 static int igb_suspend(struct pci_dev *, pm_message_t);
130 #ifdef CONFIG_PM
131 static int igb_resume(struct pci_dev *);
132 #endif
133 static void igb_shutdown(struct pci_dev *);
134 #ifdef CONFIG_IGB_DCA
135 static int igb_notify_dca(struct notifier_block *, unsigned long, void *);
136 static struct notifier_block dca_notifier = {
137         .notifier_call  = igb_notify_dca,
138         .next           = NULL,
139         .priority       = 0
140 };
141 #endif
142
143 #ifdef CONFIG_NET_POLL_CONTROLLER
144 /* for netdump / net console */
145 static void igb_netpoll(struct net_device *);
146 #endif
147
148 static pci_ers_result_t igb_io_error_detected(struct pci_dev *,
149                      pci_channel_state_t);
150 static pci_ers_result_t igb_io_slot_reset(struct pci_dev *);
151 static void igb_io_resume(struct pci_dev *);
152
153 static struct pci_error_handlers igb_err_handler = {
154         .error_detected = igb_io_error_detected,
155         .slot_reset = igb_io_slot_reset,
156         .resume = igb_io_resume,
157 };
158
159
160 static struct pci_driver igb_driver = {
161         .name     = igb_driver_name,
162         .id_table = igb_pci_tbl,
163         .probe    = igb_probe,
164         .remove   = __devexit_p(igb_remove),
165 #ifdef CONFIG_PM
166         /* Power Managment Hooks */
167         .suspend  = igb_suspend,
168         .resume   = igb_resume,
169 #endif
170         .shutdown = igb_shutdown,
171         .err_handler = &igb_err_handler
172 };
173
174 static int global_quad_port_a; /* global quad port a indication */
175
176 MODULE_AUTHOR("Intel Corporation, <e1000-devel@lists.sourceforge.net>");
177 MODULE_DESCRIPTION("Intel(R) Gigabit Ethernet Network Driver");
178 MODULE_LICENSE("GPL");
179 MODULE_VERSION(DRV_VERSION);
180
181 #ifdef DEBUG
182 /**
183  * igb_get_hw_dev_name - return device name string
184  * used by hardware layer to print debugging information
185  **/
186 char *igb_get_hw_dev_name(struct e1000_hw *hw)
187 {
188         struct igb_adapter *adapter = hw->back;
189         return adapter->netdev->name;
190 }
191 #endif
192
193 /**
194  * igb_init_module - Driver Registration Routine
195  *
196  * igb_init_module is the first routine called when the driver is
197  * loaded. All it does is register with the PCI subsystem.
198  **/
199 static int __init igb_init_module(void)
200 {
201         int ret;
202         printk(KERN_INFO "%s - version %s\n",
203                igb_driver_string, igb_driver_version);
204
205         printk(KERN_INFO "%s\n", igb_copyright);
206
207         global_quad_port_a = 0;
208
209 #ifdef CONFIG_IGB_DCA
210         dca_register_notify(&dca_notifier);
211 #endif
212
213         ret = pci_register_driver(&igb_driver);
214         return ret;
215 }
216
217 module_init(igb_init_module);
218
219 /**
220  * igb_exit_module - Driver Exit Cleanup Routine
221  *
222  * igb_exit_module is called just before the driver is removed
223  * from memory.
224  **/
225 static void __exit igb_exit_module(void)
226 {
227 #ifdef CONFIG_IGB_DCA
228         dca_unregister_notify(&dca_notifier);
229 #endif
230         pci_unregister_driver(&igb_driver);
231 }
232
233 module_exit(igb_exit_module);
234
235 #define Q_IDX_82576(i) (((i & 0x1) << 3) + (i >> 1))
236 /**
237  * igb_cache_ring_register - Descriptor ring to register mapping
238  * @adapter: board private structure to initialize
239  *
240  * Once we know the feature-set enabled for the device, we'll cache
241  * the register offset the descriptor ring is assigned to.
242  **/
243 static void igb_cache_ring_register(struct igb_adapter *adapter)
244 {
245         int i;
246
247         switch (adapter->hw.mac.type) {
248         case e1000_82576:
249                 /* The queues are allocated for virtualization such that VF 0
250                  * is allocated queues 0 and 8, VF 1 queues 1 and 9, etc.
251                  * In order to avoid collision we start at the first free queue
252                  * and continue consuming queues in the same sequence
253                  */
254                 for (i = 0; i < adapter->num_rx_queues; i++)
255                         adapter->rx_ring[i].reg_idx = Q_IDX_82576(i);
256                 for (i = 0; i < adapter->num_tx_queues; i++)
257                         adapter->tx_ring[i].reg_idx = Q_IDX_82576(i);
258                 break;
259         case e1000_82575:
260         default:
261                 for (i = 0; i < adapter->num_rx_queues; i++)
262                         adapter->rx_ring[i].reg_idx = i;
263                 for (i = 0; i < adapter->num_tx_queues; i++)
264                         adapter->tx_ring[i].reg_idx = i;
265                 break;
266         }
267 }
268
269 /**
270  * igb_alloc_queues - Allocate memory for all rings
271  * @adapter: board private structure to initialize
272  *
273  * We allocate one ring per queue at run-time since we don't know the
274  * number of queues at compile-time.
275  **/
276 static int igb_alloc_queues(struct igb_adapter *adapter)
277 {
278         int i;
279
280         adapter->tx_ring = kcalloc(adapter->num_tx_queues,
281                                    sizeof(struct igb_ring), GFP_KERNEL);
282         if (!adapter->tx_ring)
283                 return -ENOMEM;
284
285         adapter->rx_ring = kcalloc(adapter->num_rx_queues,
286                                    sizeof(struct igb_ring), GFP_KERNEL);
287         if (!adapter->rx_ring) {
288                 kfree(adapter->tx_ring);
289                 return -ENOMEM;
290         }
291
292         adapter->rx_ring->buddy = adapter->tx_ring;
293
294         for (i = 0; i < adapter->num_tx_queues; i++) {
295                 struct igb_ring *ring = &(adapter->tx_ring[i]);
296                 ring->count = adapter->tx_ring_count;
297                 ring->adapter = adapter;
298                 ring->queue_index = i;
299         }
300         for (i = 0; i < adapter->num_rx_queues; i++) {
301                 struct igb_ring *ring = &(adapter->rx_ring[i]);
302                 ring->count = adapter->rx_ring_count;
303                 ring->adapter = adapter;
304                 ring->queue_index = i;
305                 ring->itr_register = E1000_ITR;
306
307                 /* set a default napi handler for each rx_ring */
308                 netif_napi_add(adapter->netdev, &ring->napi, igb_poll, 64);
309         }
310
311         igb_cache_ring_register(adapter);
312         return 0;
313 }
314
315 static void igb_free_queues(struct igb_adapter *adapter)
316 {
317         int i;
318
319         for (i = 0; i < adapter->num_rx_queues; i++)
320                 netif_napi_del(&adapter->rx_ring[i].napi);
321
322         kfree(adapter->tx_ring);
323         kfree(adapter->rx_ring);
324 }
325
326 #define IGB_N0_QUEUE -1
327 static void igb_assign_vector(struct igb_adapter *adapter, int rx_queue,
328                               int tx_queue, int msix_vector)
329 {
330         u32 msixbm = 0;
331         struct e1000_hw *hw = &adapter->hw;
332         u32 ivar, index;
333
334         switch (hw->mac.type) {
335         case e1000_82575:
336                 /* The 82575 assigns vectors using a bitmask, which matches the
337                    bitmask for the EICR/EIMS/EIMC registers.  To assign one
338                    or more queues to a vector, we write the appropriate bits
339                    into the MSIXBM register for that vector. */
340                 if (rx_queue > IGB_N0_QUEUE) {
341                         msixbm = E1000_EICR_RX_QUEUE0 << rx_queue;
342                         adapter->rx_ring[rx_queue].eims_value = msixbm;
343                 }
344                 if (tx_queue > IGB_N0_QUEUE) {
345                         msixbm |= E1000_EICR_TX_QUEUE0 << tx_queue;
346                         adapter->tx_ring[tx_queue].eims_value =
347                                   E1000_EICR_TX_QUEUE0 << tx_queue;
348                 }
349                 array_wr32(E1000_MSIXBM(0), msix_vector, msixbm);
350                 break;
351         case e1000_82576:
352                 /* 82576 uses a table-based method for assigning vectors.
353                    Each queue has a single entry in the table to which we write
354                    a vector number along with a "valid" bit.  Sadly, the layout
355                    of the table is somewhat counterintuitive. */
356                 if (rx_queue > IGB_N0_QUEUE) {
357                         index = (rx_queue >> 1);
358                         ivar = array_rd32(E1000_IVAR0, index);
359                         if (rx_queue & 0x1) {
360                                 /* vector goes into third byte of register */
361                                 ivar = ivar & 0xFF00FFFF;
362                                 ivar |= (msix_vector | E1000_IVAR_VALID) << 16;
363                         } else {
364                                 /* vector goes into low byte of register */
365                                 ivar = ivar & 0xFFFFFF00;
366                                 ivar |= msix_vector | E1000_IVAR_VALID;
367                         }
368                         adapter->rx_ring[rx_queue].eims_value= 1 << msix_vector;
369                         array_wr32(E1000_IVAR0, index, ivar);
370                 }
371                 if (tx_queue > IGB_N0_QUEUE) {
372                         index = (tx_queue >> 1);
373                         ivar = array_rd32(E1000_IVAR0, index);
374                         if (tx_queue & 0x1) {
375                                 /* vector goes into high byte of register */
376                                 ivar = ivar & 0x00FFFFFF;
377                                 ivar |= (msix_vector | E1000_IVAR_VALID) << 24;
378                         } else {
379                                 /* vector goes into second byte of register */
380                                 ivar = ivar & 0xFFFF00FF;
381                                 ivar |= (msix_vector | E1000_IVAR_VALID) << 8;
382                         }
383                         adapter->tx_ring[tx_queue].eims_value= 1 << msix_vector;
384                         array_wr32(E1000_IVAR0, index, ivar);
385                 }
386                 break;
387         default:
388                 BUG();
389                 break;
390         }
391 }
392
393 /**
394  * igb_configure_msix - Configure MSI-X hardware
395  *
396  * igb_configure_msix sets up the hardware to properly
397  * generate MSI-X interrupts.
398  **/
399 static void igb_configure_msix(struct igb_adapter *adapter)
400 {
401         u32 tmp;
402         int i, vector = 0;
403         struct e1000_hw *hw = &adapter->hw;
404
405         adapter->eims_enable_mask = 0;
406         if (hw->mac.type == e1000_82576)
407                 /* Turn on MSI-X capability first, or our settings
408                  * won't stick.  And it will take days to debug. */
409                 wr32(E1000_GPIE, E1000_GPIE_MSIX_MODE |
410                                    E1000_GPIE_PBA | E1000_GPIE_EIAME | 
411                                    E1000_GPIE_NSICR);
412
413         for (i = 0; i < adapter->num_tx_queues; i++) {
414                 struct igb_ring *tx_ring = &adapter->tx_ring[i];
415                 igb_assign_vector(adapter, IGB_N0_QUEUE, i, vector++);
416                 adapter->eims_enable_mask |= tx_ring->eims_value;
417                 if (tx_ring->itr_val)
418                         writel(tx_ring->itr_val,
419                                hw->hw_addr + tx_ring->itr_register);
420                 else
421                         writel(1, hw->hw_addr + tx_ring->itr_register);
422         }
423
424         for (i = 0; i < adapter->num_rx_queues; i++) {
425                 struct igb_ring *rx_ring = &adapter->rx_ring[i];
426                 rx_ring->buddy = NULL;
427                 igb_assign_vector(adapter, i, IGB_N0_QUEUE, vector++);
428                 adapter->eims_enable_mask |= rx_ring->eims_value;
429                 if (rx_ring->itr_val)
430                         writel(rx_ring->itr_val,
431                                hw->hw_addr + rx_ring->itr_register);
432                 else
433                         writel(1, hw->hw_addr + rx_ring->itr_register);
434         }
435
436
437         /* set vector for other causes, i.e. link changes */
438         switch (hw->mac.type) {
439         case e1000_82575:
440                 array_wr32(E1000_MSIXBM(0), vector++,
441                                       E1000_EIMS_OTHER);
442
443                 tmp = rd32(E1000_CTRL_EXT);
444                 /* enable MSI-X PBA support*/
445                 tmp |= E1000_CTRL_EXT_PBA_CLR;
446
447                 /* Auto-Mask interrupts upon ICR read. */
448                 tmp |= E1000_CTRL_EXT_EIAME;
449                 tmp |= E1000_CTRL_EXT_IRCA;
450
451                 wr32(E1000_CTRL_EXT, tmp);
452                 adapter->eims_enable_mask |= E1000_EIMS_OTHER;
453                 adapter->eims_other = E1000_EIMS_OTHER;
454
455                 break;
456
457         case e1000_82576:
458                 tmp = (vector++ | E1000_IVAR_VALID) << 8;
459                 wr32(E1000_IVAR_MISC, tmp);
460
461                 adapter->eims_enable_mask = (1 << (vector)) - 1;
462                 adapter->eims_other = 1 << (vector - 1);
463                 break;
464         default:
465                 /* do nothing, since nothing else supports MSI-X */
466                 break;
467         } /* switch (hw->mac.type) */
468         wrfl();
469 }
470
471 /**
472  * igb_request_msix - Initialize MSI-X interrupts
473  *
474  * igb_request_msix allocates MSI-X vectors and requests interrupts from the
475  * kernel.
476  **/
477 static int igb_request_msix(struct igb_adapter *adapter)
478 {
479         struct net_device *netdev = adapter->netdev;
480         int i, err = 0, vector = 0;
481
482         vector = 0;
483
484         for (i = 0; i < adapter->num_tx_queues; i++) {
485                 struct igb_ring *ring = &(adapter->tx_ring[i]);
486                 sprintf(ring->name, "%s-tx-%d", netdev->name, i);
487                 err = request_irq(adapter->msix_entries[vector].vector,
488                                   &igb_msix_tx, 0, ring->name,
489                                   &(adapter->tx_ring[i]));
490                 if (err)
491                         goto out;
492                 ring->itr_register = E1000_EITR(0) + (vector << 2);
493                 ring->itr_val = 976; /* ~4000 ints/sec */
494                 vector++;
495         }
496         for (i = 0; i < adapter->num_rx_queues; i++) {
497                 struct igb_ring *ring = &(adapter->rx_ring[i]);
498                 if (strlen(netdev->name) < (IFNAMSIZ - 5))
499                         sprintf(ring->name, "%s-rx-%d", netdev->name, i);
500                 else
501                         memcpy(ring->name, netdev->name, IFNAMSIZ);
502                 err = request_irq(adapter->msix_entries[vector].vector,
503                                   &igb_msix_rx, 0, ring->name,
504                                   &(adapter->rx_ring[i]));
505                 if (err)
506                         goto out;
507                 ring->itr_register = E1000_EITR(0) + (vector << 2);
508                 ring->itr_val = adapter->itr;
509                 /* overwrite the poll routine for MSIX, we've already done
510                  * netif_napi_add */
511                 ring->napi.poll = &igb_clean_rx_ring_msix;
512                 vector++;
513         }
514
515         err = request_irq(adapter->msix_entries[vector].vector,
516                           &igb_msix_other, 0, netdev->name, netdev);
517         if (err)
518                 goto out;
519
520         igb_configure_msix(adapter);
521         return 0;
522 out:
523         return err;
524 }
525
526 static void igb_reset_interrupt_capability(struct igb_adapter *adapter)
527 {
528         if (adapter->msix_entries) {
529                 pci_disable_msix(adapter->pdev);
530                 kfree(adapter->msix_entries);
531                 adapter->msix_entries = NULL;
532         } else if (adapter->flags & IGB_FLAG_HAS_MSI)
533                 pci_disable_msi(adapter->pdev);
534         return;
535 }
536
537
538 /**
539  * igb_set_interrupt_capability - set MSI or MSI-X if supported
540  *
541  * Attempt to configure interrupts using the best available
542  * capabilities of the hardware and kernel.
543  **/
544 static void igb_set_interrupt_capability(struct igb_adapter *adapter)
545 {
546         int err;
547         int numvecs, i;
548
549         numvecs = adapter->num_tx_queues + adapter->num_rx_queues + 1;
550         adapter->msix_entries = kcalloc(numvecs, sizeof(struct msix_entry),
551                                         GFP_KERNEL);
552         if (!adapter->msix_entries)
553                 goto msi_only;
554
555         for (i = 0; i < numvecs; i++)
556                 adapter->msix_entries[i].entry = i;
557
558         err = pci_enable_msix(adapter->pdev,
559                               adapter->msix_entries,
560                               numvecs);
561         if (err == 0)
562                 goto out;
563
564         igb_reset_interrupt_capability(adapter);
565
566         /* If we can't do MSI-X, try MSI */
567 msi_only:
568         adapter->num_rx_queues = 1;
569         adapter->num_tx_queues = 1;
570         if (!pci_enable_msi(adapter->pdev))
571                 adapter->flags |= IGB_FLAG_HAS_MSI;
572 out:
573         /* Notify the stack of the (possibly) reduced Tx Queue count. */
574         adapter->netdev->real_num_tx_queues = adapter->num_tx_queues;
575         return;
576 }
577
578 /**
579  * igb_request_irq - initialize interrupts
580  *
581  * Attempts to configure interrupts using the best available
582  * capabilities of the hardware and kernel.
583  **/
584 static int igb_request_irq(struct igb_adapter *adapter)
585 {
586         struct net_device *netdev = adapter->netdev;
587         struct e1000_hw *hw = &adapter->hw;
588         int err = 0;
589
590         if (adapter->msix_entries) {
591                 err = igb_request_msix(adapter);
592                 if (!err)
593                         goto request_done;
594                 /* fall back to MSI */
595                 igb_reset_interrupt_capability(adapter);
596                 if (!pci_enable_msi(adapter->pdev))
597                         adapter->flags |= IGB_FLAG_HAS_MSI;
598                 igb_free_all_tx_resources(adapter);
599                 igb_free_all_rx_resources(adapter);
600                 adapter->num_rx_queues = 1;
601                 igb_alloc_queues(adapter);
602         } else {
603                 switch (hw->mac.type) {
604                 case e1000_82575:
605                         wr32(E1000_MSIXBM(0),
606                              (E1000_EICR_RX_QUEUE0 | E1000_EIMS_OTHER));
607                         break;
608                 case e1000_82576:
609                         wr32(E1000_IVAR0, E1000_IVAR_VALID);
610                         break;
611                 default:
612                         break;
613                 }
614         }
615
616         if (adapter->flags & IGB_FLAG_HAS_MSI) {
617                 err = request_irq(adapter->pdev->irq, &igb_intr_msi, 0,
618                                   netdev->name, netdev);
619                 if (!err)
620                         goto request_done;
621                 /* fall back to legacy interrupts */
622                 igb_reset_interrupt_capability(adapter);
623                 adapter->flags &= ~IGB_FLAG_HAS_MSI;
624         }
625
626         err = request_irq(adapter->pdev->irq, &igb_intr, IRQF_SHARED,
627                           netdev->name, netdev);
628
629         if (err)
630                 dev_err(&adapter->pdev->dev, "Error %d getting interrupt\n",
631                         err);
632
633 request_done:
634         return err;
635 }
636
637 static void igb_free_irq(struct igb_adapter *adapter)
638 {
639         struct net_device *netdev = adapter->netdev;
640
641         if (adapter->msix_entries) {
642                 int vector = 0, i;
643
644                 for (i = 0; i < adapter->num_tx_queues; i++)
645                         free_irq(adapter->msix_entries[vector++].vector,
646                                 &(adapter->tx_ring[i]));
647                 for (i = 0; i < adapter->num_rx_queues; i++)
648                         free_irq(adapter->msix_entries[vector++].vector,
649                                 &(adapter->rx_ring[i]));
650
651                 free_irq(adapter->msix_entries[vector++].vector, netdev);
652                 return;
653         }
654
655         free_irq(adapter->pdev->irq, netdev);
656 }
657
658 /**
659  * igb_irq_disable - Mask off interrupt generation on the NIC
660  * @adapter: board private structure
661  **/
662 static void igb_irq_disable(struct igb_adapter *adapter)
663 {
664         struct e1000_hw *hw = &adapter->hw;
665
666         if (adapter->msix_entries) {
667                 wr32(E1000_EIAM, 0);
668                 wr32(E1000_EIMC, ~0);
669                 wr32(E1000_EIAC, 0);
670         }
671
672         wr32(E1000_IAM, 0);
673         wr32(E1000_IMC, ~0);
674         wrfl();
675         synchronize_irq(adapter->pdev->irq);
676 }
677
678 /**
679  * igb_irq_enable - Enable default interrupt generation settings
680  * @adapter: board private structure
681  **/
682 static void igb_irq_enable(struct igb_adapter *adapter)
683 {
684         struct e1000_hw *hw = &adapter->hw;
685
686         if (adapter->msix_entries) {
687                 wr32(E1000_EIAC, adapter->eims_enable_mask);
688                 wr32(E1000_EIAM, adapter->eims_enable_mask);
689                 wr32(E1000_EIMS, adapter->eims_enable_mask);
690                 wr32(E1000_IMS, E1000_IMS_LSC);
691         } else {
692                 wr32(E1000_IMS, IMS_ENABLE_MASK);
693                 wr32(E1000_IAM, IMS_ENABLE_MASK);
694         }
695 }
696
697 static void igb_update_mng_vlan(struct igb_adapter *adapter)
698 {
699         struct net_device *netdev = adapter->netdev;
700         u16 vid = adapter->hw.mng_cookie.vlan_id;
701         u16 old_vid = adapter->mng_vlan_id;
702         if (adapter->vlgrp) {
703                 if (!vlan_group_get_device(adapter->vlgrp, vid)) {
704                         if (adapter->hw.mng_cookie.status &
705                                 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
706                                 igb_vlan_rx_add_vid(netdev, vid);
707                                 adapter->mng_vlan_id = vid;
708                         } else
709                                 adapter->mng_vlan_id = IGB_MNG_VLAN_NONE;
710
711                         if ((old_vid != (u16)IGB_MNG_VLAN_NONE) &&
712                                         (vid != old_vid) &&
713                             !vlan_group_get_device(adapter->vlgrp, old_vid))
714                                 igb_vlan_rx_kill_vid(netdev, old_vid);
715                 } else
716                         adapter->mng_vlan_id = vid;
717         }
718 }
719
720 /**
721  * igb_release_hw_control - release control of the h/w to f/w
722  * @adapter: address of board private structure
723  *
724  * igb_release_hw_control resets CTRL_EXT:DRV_LOAD bit.
725  * For ASF and Pass Through versions of f/w this means that the
726  * driver is no longer loaded.
727  *
728  **/
729 static void igb_release_hw_control(struct igb_adapter *adapter)
730 {
731         struct e1000_hw *hw = &adapter->hw;
732         u32 ctrl_ext;
733
734         /* Let firmware take over control of h/w */
735         ctrl_ext = rd32(E1000_CTRL_EXT);
736         wr32(E1000_CTRL_EXT,
737                         ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
738 }
739
740
741 /**
742  * igb_get_hw_control - get control of the h/w from f/w
743  * @adapter: address of board private structure
744  *
745  * igb_get_hw_control sets CTRL_EXT:DRV_LOAD bit.
746  * For ASF and Pass Through versions of f/w this means that
747  * the driver is loaded.
748  *
749  **/
750 static void igb_get_hw_control(struct igb_adapter *adapter)
751 {
752         struct e1000_hw *hw = &adapter->hw;
753         u32 ctrl_ext;
754
755         /* Let firmware know the driver has taken over */
756         ctrl_ext = rd32(E1000_CTRL_EXT);
757         wr32(E1000_CTRL_EXT,
758                         ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
759 }
760
761 /**
762  * igb_configure - configure the hardware for RX and TX
763  * @adapter: private board structure
764  **/
765 static void igb_configure(struct igb_adapter *adapter)
766 {
767         struct net_device *netdev = adapter->netdev;
768         int i;
769
770         igb_get_hw_control(adapter);
771         igb_set_multi(netdev);
772
773         igb_restore_vlan(adapter);
774
775         igb_configure_tx(adapter);
776         igb_setup_rctl(adapter);
777         igb_configure_rx(adapter);
778
779         igb_rx_fifo_flush_82575(&adapter->hw);
780
781         /* call IGB_DESC_UNUSED which always leaves
782          * at least 1 descriptor unused to make sure
783          * next_to_use != next_to_clean */
784         for (i = 0; i < adapter->num_rx_queues; i++) {
785                 struct igb_ring *ring = &adapter->rx_ring[i];
786                 igb_alloc_rx_buffers_adv(ring, IGB_DESC_UNUSED(ring));
787         }
788
789
790         adapter->tx_queue_len = netdev->tx_queue_len;
791 }
792
793
794 /**
795  * igb_up - Open the interface and prepare it to handle traffic
796  * @adapter: board private structure
797  **/
798
799 int igb_up(struct igb_adapter *adapter)
800 {
801         struct e1000_hw *hw = &adapter->hw;
802         int i;
803
804         /* hardware has been reset, we need to reload some things */
805         igb_configure(adapter);
806
807         clear_bit(__IGB_DOWN, &adapter->state);
808
809         for (i = 0; i < adapter->num_rx_queues; i++)
810                 napi_enable(&adapter->rx_ring[i].napi);
811         if (adapter->msix_entries)
812                 igb_configure_msix(adapter);
813
814         /* Clear any pending interrupts. */
815         rd32(E1000_ICR);
816         igb_irq_enable(adapter);
817
818         /* Fire a link change interrupt to start the watchdog. */
819         wr32(E1000_ICS, E1000_ICS_LSC);
820         return 0;
821 }
822
823 void igb_down(struct igb_adapter *adapter)
824 {
825         struct e1000_hw *hw = &adapter->hw;
826         struct net_device *netdev = adapter->netdev;
827         u32 tctl, rctl;
828         int i;
829
830         /* signal that we're down so the interrupt handler does not
831          * reschedule our watchdog timer */
832         set_bit(__IGB_DOWN, &adapter->state);
833
834         /* disable receives in the hardware */
835         rctl = rd32(E1000_RCTL);
836         wr32(E1000_RCTL, rctl & ~E1000_RCTL_EN);
837         /* flush and sleep below */
838
839         netif_tx_stop_all_queues(netdev);
840
841         /* disable transmits in the hardware */
842         tctl = rd32(E1000_TCTL);
843         tctl &= ~E1000_TCTL_EN;
844         wr32(E1000_TCTL, tctl);
845         /* flush both disables and wait for them to finish */
846         wrfl();
847         msleep(10);
848
849         for (i = 0; i < adapter->num_rx_queues; i++)
850                 napi_disable(&adapter->rx_ring[i].napi);
851
852         igb_irq_disable(adapter);
853
854         del_timer_sync(&adapter->watchdog_timer);
855         del_timer_sync(&adapter->phy_info_timer);
856
857         netdev->tx_queue_len = adapter->tx_queue_len;
858         netif_carrier_off(netdev);
859         adapter->link_speed = 0;
860         adapter->link_duplex = 0;
861
862         if (!pci_channel_offline(adapter->pdev))
863                 igb_reset(adapter);
864         igb_clean_all_tx_rings(adapter);
865         igb_clean_all_rx_rings(adapter);
866 }
867
868 void igb_reinit_locked(struct igb_adapter *adapter)
869 {
870         WARN_ON(in_interrupt());
871         while (test_and_set_bit(__IGB_RESETTING, &adapter->state))
872                 msleep(1);
873         igb_down(adapter);
874         igb_up(adapter);
875         clear_bit(__IGB_RESETTING, &adapter->state);
876 }
877
878 void igb_reset(struct igb_adapter *adapter)
879 {
880         struct e1000_hw *hw = &adapter->hw;
881         struct e1000_mac_info *mac = &hw->mac;
882         struct e1000_fc_info *fc = &hw->fc;
883         u32 pba = 0, tx_space, min_tx_space, min_rx_space;
884         u16 hwm;
885
886         /* Repartition Pba for greater than 9k mtu
887          * To take effect CTRL.RST is required.
888          */
889         if (mac->type != e1000_82576) {
890         pba = E1000_PBA_34K;
891         }
892         else {
893                 pba = E1000_PBA_64K;
894         }
895
896         if ((adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) &&
897             (mac->type < e1000_82576)) {
898                 /* adjust PBA for jumbo frames */
899                 wr32(E1000_PBA, pba);
900
901                 /* To maintain wire speed transmits, the Tx FIFO should be
902                  * large enough to accommodate two full transmit packets,
903                  * rounded up to the next 1KB and expressed in KB.  Likewise,
904                  * the Rx FIFO should be large enough to accommodate at least
905                  * one full receive packet and is similarly rounded up and
906                  * expressed in KB. */
907                 pba = rd32(E1000_PBA);
908                 /* upper 16 bits has Tx packet buffer allocation size in KB */
909                 tx_space = pba >> 16;
910                 /* lower 16 bits has Rx packet buffer allocation size in KB */
911                 pba &= 0xffff;
912                 /* the tx fifo also stores 16 bytes of information about the tx
913                  * but don't include ethernet FCS because hardware appends it */
914                 min_tx_space = (adapter->max_frame_size +
915                                 sizeof(struct e1000_tx_desc) -
916                                 ETH_FCS_LEN) * 2;
917                 min_tx_space = ALIGN(min_tx_space, 1024);
918                 min_tx_space >>= 10;
919                 /* software strips receive CRC, so leave room for it */
920                 min_rx_space = adapter->max_frame_size;
921                 min_rx_space = ALIGN(min_rx_space, 1024);
922                 min_rx_space >>= 10;
923
924                 /* If current Tx allocation is less than the min Tx FIFO size,
925                  * and the min Tx FIFO size is less than the current Rx FIFO
926                  * allocation, take space away from current Rx allocation */
927                 if (tx_space < min_tx_space &&
928                     ((min_tx_space - tx_space) < pba)) {
929                         pba = pba - (min_tx_space - tx_space);
930
931                         /* if short on rx space, rx wins and must trump tx
932                          * adjustment */
933                         if (pba < min_rx_space)
934                                 pba = min_rx_space;
935                 }
936                 wr32(E1000_PBA, pba);
937         }
938
939         /* flow control settings */
940         /* The high water mark must be low enough to fit one full frame
941          * (or the size used for early receive) above it in the Rx FIFO.
942          * Set it to the lower of:
943          * - 90% of the Rx FIFO size, or
944          * - the full Rx FIFO size minus one full frame */
945         hwm = min(((pba << 10) * 9 / 10),
946                         ((pba << 10) - 2 * adapter->max_frame_size));
947
948         if (mac->type < e1000_82576) {
949                 fc->high_water = hwm & 0xFFF8;  /* 8-byte granularity */
950                 fc->low_water = fc->high_water - 8;
951         } else {
952                 fc->high_water = hwm & 0xFFF0;  /* 16-byte granularity */
953                 fc->low_water = fc->high_water - 16;
954         }
955         fc->pause_time = 0xFFFF;
956         fc->send_xon = 1;
957         fc->type = fc->original_type;
958
959         /* Allow time for pending master requests to run */
960         adapter->hw.mac.ops.reset_hw(&adapter->hw);
961         wr32(E1000_WUC, 0);
962
963         if (adapter->hw.mac.ops.init_hw(&adapter->hw))
964                 dev_err(&adapter->pdev->dev, "Hardware Error\n");
965
966         igb_update_mng_vlan(adapter);
967
968         /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
969         wr32(E1000_VET, ETHERNET_IEEE_VLAN_TYPE);
970
971         igb_reset_adaptive(&adapter->hw);
972         igb_get_phy_info(&adapter->hw);
973 }
974
975 /**
976  * igb_is_need_ioport - determine if an adapter needs ioport resources or not
977  * @pdev: PCI device information struct
978  *
979  * Returns true if an adapter needs ioport resources
980  **/
981 static int igb_is_need_ioport(struct pci_dev *pdev)
982 {
983         switch (pdev->device) {
984         /* Currently there are no adapters that need ioport resources */
985         default:
986                 return false;
987         }
988 }
989
990 static const struct net_device_ops igb_netdev_ops = {
991         .ndo_open               = igb_open,
992         .ndo_stop               = igb_close,
993         .ndo_start_xmit         = igb_xmit_frame_adv,
994         .ndo_get_stats          = igb_get_stats,
995         .ndo_set_multicast_list = igb_set_multi,
996         .ndo_set_mac_address    = igb_set_mac,
997         .ndo_change_mtu         = igb_change_mtu,
998         .ndo_do_ioctl           = igb_ioctl,
999         .ndo_tx_timeout         = igb_tx_timeout,
1000         .ndo_validate_addr      = eth_validate_addr,
1001         .ndo_vlan_rx_register   = igb_vlan_rx_register,
1002         .ndo_vlan_rx_add_vid    = igb_vlan_rx_add_vid,
1003         .ndo_vlan_rx_kill_vid   = igb_vlan_rx_kill_vid,
1004 #ifdef CONFIG_NET_POLL_CONTROLLER
1005         .ndo_poll_controller    = igb_netpoll,
1006 #endif
1007 };
1008
1009 /**
1010  * igb_probe - Device Initialization Routine
1011  * @pdev: PCI device information struct
1012  * @ent: entry in igb_pci_tbl
1013  *
1014  * Returns 0 on success, negative on failure
1015  *
1016  * igb_probe initializes an adapter identified by a pci_dev structure.
1017  * The OS initialization, configuring of the adapter private structure,
1018  * and a hardware reset occur.
1019  **/
1020 static int __devinit igb_probe(struct pci_dev *pdev,
1021                                const struct pci_device_id *ent)
1022 {
1023         struct net_device *netdev;
1024         struct igb_adapter *adapter;
1025         struct e1000_hw *hw;
1026         const struct e1000_info *ei = igb_info_tbl[ent->driver_data];
1027         unsigned long mmio_start, mmio_len;
1028         int i, err, pci_using_dac;
1029         u16 eeprom_data = 0;
1030         u16 eeprom_apme_mask = IGB_EEPROM_APME;
1031         u32 part_num;
1032         int bars, need_ioport;
1033
1034         /* do not allocate ioport bars when not needed */
1035         need_ioport = igb_is_need_ioport(pdev);
1036         if (need_ioport) {
1037                 bars = pci_select_bars(pdev, IORESOURCE_MEM | IORESOURCE_IO);
1038                 err = pci_enable_device(pdev);
1039         } else {
1040                 bars = pci_select_bars(pdev, IORESOURCE_MEM);
1041                 err = pci_enable_device_mem(pdev);
1042         }
1043         if (err)
1044                 return err;
1045
1046         pci_using_dac = 0;
1047         err = pci_set_dma_mask(pdev, DMA_64BIT_MASK);
1048         if (!err) {
1049                 err = pci_set_consistent_dma_mask(pdev, DMA_64BIT_MASK);
1050                 if (!err)
1051                         pci_using_dac = 1;
1052         } else {
1053                 err = pci_set_dma_mask(pdev, DMA_32BIT_MASK);
1054                 if (err) {
1055                         err = pci_set_consistent_dma_mask(pdev, DMA_32BIT_MASK);
1056                         if (err) {
1057                                 dev_err(&pdev->dev, "No usable DMA "
1058                                         "configuration, aborting\n");
1059                                 goto err_dma;
1060                         }
1061                 }
1062         }
1063
1064         err = pci_request_selected_regions(pdev, bars, igb_driver_name);
1065         if (err)
1066                 goto err_pci_reg;
1067
1068         err = pci_enable_pcie_error_reporting(pdev);
1069         if (err) {
1070                 dev_err(&pdev->dev, "pci_enable_pcie_error_reporting failed "
1071                         "0x%x\n", err);
1072                 /* non-fatal, continue */
1073         }
1074
1075         pci_set_master(pdev);
1076         pci_save_state(pdev);
1077
1078         err = -ENOMEM;
1079         netdev = alloc_etherdev_mq(sizeof(struct igb_adapter), IGB_MAX_TX_QUEUES);
1080         if (!netdev)
1081                 goto err_alloc_etherdev;
1082
1083         SET_NETDEV_DEV(netdev, &pdev->dev);
1084
1085         pci_set_drvdata(pdev, netdev);
1086         adapter = netdev_priv(netdev);
1087         adapter->netdev = netdev;
1088         adapter->pdev = pdev;
1089         hw = &adapter->hw;
1090         hw->back = adapter;
1091         adapter->msg_enable = NETIF_MSG_DRV | NETIF_MSG_PROBE;
1092         adapter->bars = bars;
1093         adapter->need_ioport = need_ioport;
1094
1095         mmio_start = pci_resource_start(pdev, 0);
1096         mmio_len = pci_resource_len(pdev, 0);
1097
1098         err = -EIO;
1099         adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
1100         if (!adapter->hw.hw_addr)
1101                 goto err_ioremap;
1102
1103         netdev->netdev_ops = &igb_netdev_ops;
1104         igb_set_ethtool_ops(netdev);
1105         netdev->watchdog_timeo = 5 * HZ;
1106
1107         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
1108
1109         netdev->mem_start = mmio_start;
1110         netdev->mem_end = mmio_start + mmio_len;
1111
1112         /* PCI config space info */
1113         hw->vendor_id = pdev->vendor;
1114         hw->device_id = pdev->device;
1115         hw->revision_id = pdev->revision;
1116         hw->subsystem_vendor_id = pdev->subsystem_vendor;
1117         hw->subsystem_device_id = pdev->subsystem_device;
1118
1119         /* setup the private structure */
1120         hw->back = adapter;
1121         /* Copy the default MAC, PHY and NVM function pointers */
1122         memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
1123         memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
1124         memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
1125         /* Initialize skew-specific constants */
1126         err = ei->get_invariants(hw);
1127         if (err)
1128                 goto err_hw_init;
1129
1130         err = igb_sw_init(adapter);
1131         if (err)
1132                 goto err_sw_init;
1133
1134         igb_get_bus_info_pcie(hw);
1135
1136         /* set flags */
1137         switch (hw->mac.type) {
1138         case e1000_82575:
1139                 adapter->flags |= IGB_FLAG_NEED_CTX_IDX;
1140                 break;
1141         case e1000_82576:
1142         default:
1143                 break;
1144         }
1145
1146         hw->phy.autoneg_wait_to_complete = false;
1147         hw->mac.adaptive_ifs = true;
1148
1149         /* Copper options */
1150         if (hw->phy.media_type == e1000_media_type_copper) {
1151                 hw->phy.mdix = AUTO_ALL_MODES;
1152                 hw->phy.disable_polarity_correction = false;
1153                 hw->phy.ms_type = e1000_ms_hw_default;
1154         }
1155
1156         if (igb_check_reset_block(hw))
1157                 dev_info(&pdev->dev,
1158                         "PHY reset is blocked due to SOL/IDER session.\n");
1159
1160         netdev->features = NETIF_F_SG |
1161                            NETIF_F_HW_CSUM |
1162                            NETIF_F_HW_VLAN_TX |
1163                            NETIF_F_HW_VLAN_RX |
1164                            NETIF_F_HW_VLAN_FILTER;
1165
1166         netdev->features |= NETIF_F_TSO;
1167         netdev->features |= NETIF_F_TSO6;
1168
1169 #ifdef CONFIG_IGB_LRO
1170         netdev->features |= NETIF_F_LRO;
1171 #endif
1172
1173         netdev->vlan_features |= NETIF_F_TSO;
1174         netdev->vlan_features |= NETIF_F_TSO6;
1175         netdev->vlan_features |= NETIF_F_HW_CSUM;
1176         netdev->vlan_features |= NETIF_F_SG;
1177
1178         if (pci_using_dac)
1179                 netdev->features |= NETIF_F_HIGHDMA;
1180
1181         netdev->features |= NETIF_F_LLTX;
1182         adapter->en_mng_pt = igb_enable_mng_pass_thru(&adapter->hw);
1183
1184         /* before reading the NVM, reset the controller to put the device in a
1185          * known good starting state */
1186         hw->mac.ops.reset_hw(hw);
1187
1188         /* make sure the NVM is good */
1189         if (igb_validate_nvm_checksum(hw) < 0) {
1190                 dev_err(&pdev->dev, "The NVM Checksum Is Not Valid\n");
1191                 err = -EIO;
1192                 goto err_eeprom;
1193         }
1194
1195         /* copy the MAC address out of the NVM */
1196         if (hw->mac.ops.read_mac_addr(hw))
1197                 dev_err(&pdev->dev, "NVM Read Error\n");
1198
1199         memcpy(netdev->dev_addr, hw->mac.addr, netdev->addr_len);
1200         memcpy(netdev->perm_addr, hw->mac.addr, netdev->addr_len);
1201
1202         if (!is_valid_ether_addr(netdev->perm_addr)) {
1203                 dev_err(&pdev->dev, "Invalid MAC Address\n");
1204                 err = -EIO;
1205                 goto err_eeprom;
1206         }
1207
1208         init_timer(&adapter->watchdog_timer);
1209         adapter->watchdog_timer.function = &igb_watchdog;
1210         adapter->watchdog_timer.data = (unsigned long) adapter;
1211
1212         init_timer(&adapter->phy_info_timer);
1213         adapter->phy_info_timer.function = &igb_update_phy_info;
1214         adapter->phy_info_timer.data = (unsigned long) adapter;
1215
1216         INIT_WORK(&adapter->reset_task, igb_reset_task);
1217         INIT_WORK(&adapter->watchdog_task, igb_watchdog_task);
1218
1219         /* Initialize link & ring properties that are user-changeable */
1220         adapter->tx_ring->count = 256;
1221         for (i = 0; i < adapter->num_tx_queues; i++)
1222                 adapter->tx_ring[i].count = adapter->tx_ring->count;
1223         adapter->rx_ring->count = 256;
1224         for (i = 0; i < adapter->num_rx_queues; i++)
1225                 adapter->rx_ring[i].count = adapter->rx_ring->count;
1226
1227         adapter->fc_autoneg = true;
1228         hw->mac.autoneg = true;
1229         hw->phy.autoneg_advertised = 0x2f;
1230
1231         hw->fc.original_type = e1000_fc_default;
1232         hw->fc.type = e1000_fc_default;
1233
1234         adapter->itr_setting = 3;
1235         adapter->itr = IGB_START_ITR;
1236
1237         igb_validate_mdi_setting(hw);
1238
1239         adapter->rx_csum = 1;
1240
1241         /* Initial Wake on LAN setting If APM wake is enabled in the EEPROM,
1242          * enable the ACPI Magic Packet filter
1243          */
1244
1245         if (hw->bus.func == 0 ||
1246             hw->device_id == E1000_DEV_ID_82575EB_COPPER)
1247                 hw->nvm.ops.read_nvm(hw, NVM_INIT_CONTROL3_PORT_A, 1,
1248                                      &eeprom_data);
1249
1250         if (eeprom_data & eeprom_apme_mask)
1251                 adapter->eeprom_wol |= E1000_WUFC_MAG;
1252
1253         /* now that we have the eeprom settings, apply the special cases where
1254          * the eeprom may be wrong or the board simply won't support wake on
1255          * lan on a particular port */
1256         switch (pdev->device) {
1257         case E1000_DEV_ID_82575GB_QUAD_COPPER:
1258                 adapter->eeprom_wol = 0;
1259                 break;
1260         case E1000_DEV_ID_82575EB_FIBER_SERDES:
1261         case E1000_DEV_ID_82576_FIBER:
1262         case E1000_DEV_ID_82576_SERDES:
1263                 /* Wake events only supported on port A for dual fiber
1264                  * regardless of eeprom setting */
1265                 if (rd32(E1000_STATUS) & E1000_STATUS_FUNC_1)
1266                         adapter->eeprom_wol = 0;
1267                 break;
1268         }
1269
1270         /* initialize the wol settings based on the eeprom settings */
1271         adapter->wol = adapter->eeprom_wol;
1272         device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
1273
1274         /* reset the hardware with the new settings */
1275         igb_reset(adapter);
1276
1277         /* let the f/w know that the h/w is now under the control of the
1278          * driver. */
1279         igb_get_hw_control(adapter);
1280
1281         /* tell the stack to leave us alone until igb_open() is called */
1282         netif_carrier_off(netdev);
1283         netif_tx_stop_all_queues(netdev);
1284
1285         strcpy(netdev->name, "eth%d");
1286         err = register_netdev(netdev);
1287         if (err)
1288                 goto err_register;
1289
1290 #ifdef CONFIG_IGB_DCA
1291         if (dca_add_requester(&pdev->dev) == 0) {
1292                 adapter->flags |= IGB_FLAG_DCA_ENABLED;
1293                 dev_info(&pdev->dev, "DCA enabled\n");
1294                 /* Always use CB2 mode, difference is masked
1295                  * in the CB driver. */
1296                 wr32(E1000_DCA_CTRL, 2);
1297                 igb_setup_dca(adapter);
1298         }
1299 #endif
1300
1301         dev_info(&pdev->dev, "Intel(R) Gigabit Ethernet Network Connection\n");
1302         /* print bus type/speed/width info */
1303         dev_info(&pdev->dev, "%s: (PCIe:%s:%s) %pM\n",
1304                  netdev->name,
1305                  ((hw->bus.speed == e1000_bus_speed_2500)
1306                   ? "2.5Gb/s" : "unknown"),
1307                  ((hw->bus.width == e1000_bus_width_pcie_x4)
1308                   ? "Width x4" : (hw->bus.width == e1000_bus_width_pcie_x1)
1309                   ? "Width x1" : "unknown"),
1310                  netdev->dev_addr);
1311
1312         igb_read_part_num(hw, &part_num);
1313         dev_info(&pdev->dev, "%s: PBA No: %06x-%03x\n", netdev->name,
1314                 (part_num >> 8), (part_num & 0xff));
1315
1316         dev_info(&pdev->dev,
1317                 "Using %s interrupts. %d rx queue(s), %d tx queue(s)\n",
1318                 adapter->msix_entries ? "MSI-X" :
1319                 (adapter->flags & IGB_FLAG_HAS_MSI) ? "MSI" : "legacy",
1320                 adapter->num_rx_queues, adapter->num_tx_queues);
1321
1322         return 0;
1323
1324 err_register:
1325         igb_release_hw_control(adapter);
1326 err_eeprom:
1327         if (!igb_check_reset_block(hw))
1328                 igb_reset_phy(hw);
1329
1330         if (hw->flash_address)
1331                 iounmap(hw->flash_address);
1332
1333         igb_remove_device(hw);
1334         igb_free_queues(adapter);
1335 err_sw_init:
1336 err_hw_init:
1337         iounmap(hw->hw_addr);
1338 err_ioremap:
1339         free_netdev(netdev);
1340 err_alloc_etherdev:
1341         pci_release_selected_regions(pdev, bars);
1342 err_pci_reg:
1343 err_dma:
1344         pci_disable_device(pdev);
1345         return err;
1346 }
1347
1348 /**
1349  * igb_remove - Device Removal Routine
1350  * @pdev: PCI device information struct
1351  *
1352  * igb_remove is called by the PCI subsystem to alert the driver
1353  * that it should release a PCI device.  The could be caused by a
1354  * Hot-Plug event, or because the driver is going to be removed from
1355  * memory.
1356  **/
1357 static void __devexit igb_remove(struct pci_dev *pdev)
1358 {
1359         struct net_device *netdev = pci_get_drvdata(pdev);
1360         struct igb_adapter *adapter = netdev_priv(netdev);
1361 #ifdef CONFIG_IGB_DCA
1362         struct e1000_hw *hw = &adapter->hw;
1363 #endif
1364         int err;
1365
1366         /* flush_scheduled work may reschedule our watchdog task, so
1367          * explicitly disable watchdog tasks from being rescheduled  */
1368         set_bit(__IGB_DOWN, &adapter->state);
1369         del_timer_sync(&adapter->watchdog_timer);
1370         del_timer_sync(&adapter->phy_info_timer);
1371
1372         flush_scheduled_work();
1373
1374 #ifdef CONFIG_IGB_DCA
1375         if (adapter->flags & IGB_FLAG_DCA_ENABLED) {
1376                 dev_info(&pdev->dev, "DCA disabled\n");
1377                 dca_remove_requester(&pdev->dev);
1378                 adapter->flags &= ~IGB_FLAG_DCA_ENABLED;
1379                 wr32(E1000_DCA_CTRL, 1);
1380         }
1381 #endif
1382
1383         /* Release control of h/w to f/w.  If f/w is AMT enabled, this
1384          * would have already happened in close and is redundant. */
1385         igb_release_hw_control(adapter);
1386
1387         unregister_netdev(netdev);
1388
1389         if (!igb_check_reset_block(&adapter->hw))
1390                 igb_reset_phy(&adapter->hw);
1391
1392         igb_remove_device(&adapter->hw);
1393         igb_reset_interrupt_capability(adapter);
1394
1395         igb_free_queues(adapter);
1396
1397         iounmap(adapter->hw.hw_addr);
1398         if (adapter->hw.flash_address)
1399                 iounmap(adapter->hw.flash_address);
1400         pci_release_selected_regions(pdev, adapter->bars);
1401
1402         free_netdev(netdev);
1403
1404         err = pci_disable_pcie_error_reporting(pdev);
1405         if (err)
1406                 dev_err(&pdev->dev,
1407                         "pci_disable_pcie_error_reporting failed 0x%x\n", err);
1408
1409         pci_disable_device(pdev);
1410 }
1411
1412 /**
1413  * igb_sw_init - Initialize general software structures (struct igb_adapter)
1414  * @adapter: board private structure to initialize
1415  *
1416  * igb_sw_init initializes the Adapter private data structure.
1417  * Fields are initialized based on PCI device information and
1418  * OS network device settings (MTU size).
1419  **/
1420 static int __devinit igb_sw_init(struct igb_adapter *adapter)
1421 {
1422         struct e1000_hw *hw = &adapter->hw;
1423         struct net_device *netdev = adapter->netdev;
1424         struct pci_dev *pdev = adapter->pdev;
1425
1426         pci_read_config_word(pdev, PCI_COMMAND, &hw->bus.pci_cmd_word);
1427
1428         adapter->tx_ring_count = IGB_DEFAULT_TXD;
1429         adapter->rx_ring_count = IGB_DEFAULT_RXD;
1430         adapter->rx_buffer_len = MAXIMUM_ETHERNET_VLAN_SIZE;
1431         adapter->rx_ps_hdr_size = 0; /* disable packet split */
1432         adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
1433         adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
1434
1435         /* Number of supported queues. */
1436         /* Having more queues than CPUs doesn't make sense. */
1437         adapter->num_rx_queues = min_t(u32, IGB_MAX_RX_QUEUES, num_online_cpus());
1438         adapter->num_tx_queues = min_t(u32, IGB_MAX_TX_QUEUES, num_online_cpus());
1439
1440         /* This call may decrease the number of queues depending on
1441          * interrupt mode. */
1442         igb_set_interrupt_capability(adapter);
1443
1444         if (igb_alloc_queues(adapter)) {
1445                 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
1446                 return -ENOMEM;
1447         }
1448
1449         /* Explicitly disable IRQ since the NIC can be in any state. */
1450         igb_irq_disable(adapter);
1451
1452         set_bit(__IGB_DOWN, &adapter->state);
1453         return 0;
1454 }
1455
1456 /**
1457  * igb_open - Called when a network interface is made active
1458  * @netdev: network interface device structure
1459  *
1460  * Returns 0 on success, negative value on failure
1461  *
1462  * The open entry point is called when a network interface is made
1463  * active by the system (IFF_UP).  At this point all resources needed
1464  * for transmit and receive operations are allocated, the interrupt
1465  * handler is registered with the OS, the watchdog timer is started,
1466  * and the stack is notified that the interface is ready.
1467  **/
1468 static int igb_open(struct net_device *netdev)
1469 {
1470         struct igb_adapter *adapter = netdev_priv(netdev);
1471         struct e1000_hw *hw = &adapter->hw;
1472         int err;
1473         int i;
1474
1475         /* disallow open during test */
1476         if (test_bit(__IGB_TESTING, &adapter->state))
1477                 return -EBUSY;
1478
1479         /* allocate transmit descriptors */
1480         err = igb_setup_all_tx_resources(adapter);
1481         if (err)
1482                 goto err_setup_tx;
1483
1484         /* allocate receive descriptors */
1485         err = igb_setup_all_rx_resources(adapter);
1486         if (err)
1487                 goto err_setup_rx;
1488
1489         /* e1000_power_up_phy(adapter); */
1490
1491         adapter->mng_vlan_id = IGB_MNG_VLAN_NONE;
1492         if ((adapter->hw.mng_cookie.status &
1493              E1000_MNG_DHCP_COOKIE_STATUS_VLAN))
1494                 igb_update_mng_vlan(adapter);
1495
1496         /* before we allocate an interrupt, we must be ready to handle it.
1497          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
1498          * as soon as we call pci_request_irq, so we have to setup our
1499          * clean_rx handler before we do so.  */
1500         igb_configure(adapter);
1501
1502         err = igb_request_irq(adapter);
1503         if (err)
1504                 goto err_req_irq;
1505
1506         /* From here on the code is the same as igb_up() */
1507         clear_bit(__IGB_DOWN, &adapter->state);
1508
1509         for (i = 0; i < adapter->num_rx_queues; i++)
1510                 napi_enable(&adapter->rx_ring[i].napi);
1511
1512         /* Clear any pending interrupts. */
1513         rd32(E1000_ICR);
1514
1515         igb_irq_enable(adapter);
1516
1517         netif_tx_start_all_queues(netdev);
1518
1519         /* Fire a link status change interrupt to start the watchdog. */
1520         wr32(E1000_ICS, E1000_ICS_LSC);
1521
1522         return 0;
1523
1524 err_req_irq:
1525         igb_release_hw_control(adapter);
1526         /* e1000_power_down_phy(adapter); */
1527         igb_free_all_rx_resources(adapter);
1528 err_setup_rx:
1529         igb_free_all_tx_resources(adapter);
1530 err_setup_tx:
1531         igb_reset(adapter);
1532
1533         return err;
1534 }
1535
1536 /**
1537  * igb_close - Disables a network interface
1538  * @netdev: network interface device structure
1539  *
1540  * Returns 0, this is not allowed to fail
1541  *
1542  * The close entry point is called when an interface is de-activated
1543  * by the OS.  The hardware is still under the driver's control, but
1544  * needs to be disabled.  A global MAC reset is issued to stop the
1545  * hardware, and all transmit and receive resources are freed.
1546  **/
1547 static int igb_close(struct net_device *netdev)
1548 {
1549         struct igb_adapter *adapter = netdev_priv(netdev);
1550
1551         WARN_ON(test_bit(__IGB_RESETTING, &adapter->state));
1552         igb_down(adapter);
1553
1554         igb_free_irq(adapter);
1555
1556         igb_free_all_tx_resources(adapter);
1557         igb_free_all_rx_resources(adapter);
1558
1559         /* kill manageability vlan ID if supported, but not if a vlan with
1560          * the same ID is registered on the host OS (let 8021q kill it) */
1561         if ((adapter->hw.mng_cookie.status &
1562                           E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
1563              !(adapter->vlgrp &&
1564                vlan_group_get_device(adapter->vlgrp, adapter->mng_vlan_id)))
1565                 igb_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
1566
1567         return 0;
1568 }
1569
1570 /**
1571  * igb_setup_tx_resources - allocate Tx resources (Descriptors)
1572  * @adapter: board private structure
1573  * @tx_ring: tx descriptor ring (for a specific queue) to setup
1574  *
1575  * Return 0 on success, negative on failure
1576  **/
1577
1578 int igb_setup_tx_resources(struct igb_adapter *adapter,
1579                            struct igb_ring *tx_ring)
1580 {
1581         struct pci_dev *pdev = adapter->pdev;
1582         int size;
1583
1584         size = sizeof(struct igb_buffer) * tx_ring->count;
1585         tx_ring->buffer_info = vmalloc(size);
1586         if (!tx_ring->buffer_info)
1587                 goto err;
1588         memset(tx_ring->buffer_info, 0, size);
1589
1590         /* round up to nearest 4K */
1591         tx_ring->size = tx_ring->count * sizeof(struct e1000_tx_desc);
1592         tx_ring->size = ALIGN(tx_ring->size, 4096);
1593
1594         tx_ring->desc = pci_alloc_consistent(pdev, tx_ring->size,
1595                                              &tx_ring->dma);
1596
1597         if (!tx_ring->desc)
1598                 goto err;
1599
1600         tx_ring->adapter = adapter;
1601         tx_ring->next_to_use = 0;
1602         tx_ring->next_to_clean = 0;
1603         return 0;
1604
1605 err:
1606         vfree(tx_ring->buffer_info);
1607         dev_err(&adapter->pdev->dev,
1608                 "Unable to allocate memory for the transmit descriptor ring\n");
1609         return -ENOMEM;
1610 }
1611
1612 /**
1613  * igb_setup_all_tx_resources - wrapper to allocate Tx resources
1614  *                                (Descriptors) for all queues
1615  * @adapter: board private structure
1616  *
1617  * Return 0 on success, negative on failure
1618  **/
1619 static int igb_setup_all_tx_resources(struct igb_adapter *adapter)
1620 {
1621         int i, err = 0;
1622         int r_idx;
1623
1624         for (i = 0; i < adapter->num_tx_queues; i++) {
1625                 err = igb_setup_tx_resources(adapter, &adapter->tx_ring[i]);
1626                 if (err) {
1627                         dev_err(&adapter->pdev->dev,
1628                                 "Allocation for Tx Queue %u failed\n", i);
1629                         for (i--; i >= 0; i--)
1630                                 igb_free_tx_resources(&adapter->tx_ring[i]);
1631                         break;
1632                 }
1633         }
1634
1635         for (i = 0; i < IGB_MAX_TX_QUEUES; i++) {
1636                 r_idx = i % adapter->num_tx_queues;
1637                 adapter->multi_tx_table[i] = &adapter->tx_ring[r_idx];
1638         }       
1639         return err;
1640 }
1641
1642 /**
1643  * igb_configure_tx - Configure transmit Unit after Reset
1644  * @adapter: board private structure
1645  *
1646  * Configure the Tx unit of the MAC after a reset.
1647  **/
1648 static void igb_configure_tx(struct igb_adapter *adapter)
1649 {
1650         u64 tdba;
1651         struct e1000_hw *hw = &adapter->hw;
1652         u32 tctl;
1653         u32 txdctl, txctrl;
1654         int i, j;
1655
1656         for (i = 0; i < adapter->num_tx_queues; i++) {
1657                 struct igb_ring *ring = &(adapter->tx_ring[i]);
1658                 j = ring->reg_idx;
1659                 wr32(E1000_TDLEN(j),
1660                                 ring->count * sizeof(struct e1000_tx_desc));
1661                 tdba = ring->dma;
1662                 wr32(E1000_TDBAL(j),
1663                                 tdba & 0x00000000ffffffffULL);
1664                 wr32(E1000_TDBAH(j), tdba >> 32);
1665
1666                 ring->head = E1000_TDH(j);
1667                 ring->tail = E1000_TDT(j);
1668                 writel(0, hw->hw_addr + ring->tail);
1669                 writel(0, hw->hw_addr + ring->head);
1670                 txdctl = rd32(E1000_TXDCTL(j));
1671                 txdctl |= E1000_TXDCTL_QUEUE_ENABLE;
1672                 wr32(E1000_TXDCTL(j), txdctl);
1673
1674                 /* Turn off Relaxed Ordering on head write-backs.  The
1675                  * writebacks MUST be delivered in order or it will
1676                  * completely screw up our bookeeping.
1677                  */
1678                 txctrl = rd32(E1000_DCA_TXCTRL(j));
1679                 txctrl &= ~E1000_DCA_TXCTRL_TX_WB_RO_EN;
1680                 wr32(E1000_DCA_TXCTRL(j), txctrl);
1681         }
1682
1683
1684
1685         /* Use the default values for the Tx Inter Packet Gap (IPG) timer */
1686
1687         /* Program the Transmit Control Register */
1688
1689         tctl = rd32(E1000_TCTL);
1690         tctl &= ~E1000_TCTL_CT;
1691         tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
1692                 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
1693
1694         igb_config_collision_dist(hw);
1695
1696         /* Setup Transmit Descriptor Settings for eop descriptor */
1697         adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_RS;
1698
1699         /* Enable transmits */
1700         tctl |= E1000_TCTL_EN;
1701
1702         wr32(E1000_TCTL, tctl);
1703 }
1704
1705 /**
1706  * igb_setup_rx_resources - allocate Rx resources (Descriptors)
1707  * @adapter: board private structure
1708  * @rx_ring:    rx descriptor ring (for a specific queue) to setup
1709  *
1710  * Returns 0 on success, negative on failure
1711  **/
1712
1713 int igb_setup_rx_resources(struct igb_adapter *adapter,
1714                            struct igb_ring *rx_ring)
1715 {
1716         struct pci_dev *pdev = adapter->pdev;
1717         int size, desc_len;
1718
1719 #ifdef CONFIG_IGB_LRO
1720         size = sizeof(struct net_lro_desc) * MAX_LRO_DESCRIPTORS;
1721         rx_ring->lro_mgr.lro_arr = vmalloc(size);
1722         if (!rx_ring->lro_mgr.lro_arr)
1723                 goto err;
1724         memset(rx_ring->lro_mgr.lro_arr, 0, size);
1725 #endif
1726
1727         size = sizeof(struct igb_buffer) * rx_ring->count;
1728         rx_ring->buffer_info = vmalloc(size);
1729         if (!rx_ring->buffer_info)
1730                 goto err;
1731         memset(rx_ring->buffer_info, 0, size);
1732
1733         desc_len = sizeof(union e1000_adv_rx_desc);
1734
1735         /* Round up to nearest 4K */
1736         rx_ring->size = rx_ring->count * desc_len;
1737         rx_ring->size = ALIGN(rx_ring->size, 4096);
1738
1739         rx_ring->desc = pci_alloc_consistent(pdev, rx_ring->size,
1740                                              &rx_ring->dma);
1741
1742         if (!rx_ring->desc)
1743                 goto err;
1744
1745         rx_ring->next_to_clean = 0;
1746         rx_ring->next_to_use = 0;
1747
1748         rx_ring->adapter = adapter;
1749
1750         return 0;
1751
1752 err:
1753 #ifdef CONFIG_IGB_LRO
1754         vfree(rx_ring->lro_mgr.lro_arr);
1755         rx_ring->lro_mgr.lro_arr = NULL;
1756 #endif
1757         vfree(rx_ring->buffer_info);
1758         dev_err(&adapter->pdev->dev, "Unable to allocate memory for "
1759                 "the receive descriptor ring\n");
1760         return -ENOMEM;
1761 }
1762
1763 /**
1764  * igb_setup_all_rx_resources - wrapper to allocate Rx resources
1765  *                                (Descriptors) for all queues
1766  * @adapter: board private structure
1767  *
1768  * Return 0 on success, negative on failure
1769  **/
1770 static int igb_setup_all_rx_resources(struct igb_adapter *adapter)
1771 {
1772         int i, err = 0;
1773
1774         for (i = 0; i < adapter->num_rx_queues; i++) {
1775                 err = igb_setup_rx_resources(adapter, &adapter->rx_ring[i]);
1776                 if (err) {
1777                         dev_err(&adapter->pdev->dev,
1778                                 "Allocation for Rx Queue %u failed\n", i);
1779                         for (i--; i >= 0; i--)
1780                                 igb_free_rx_resources(&adapter->rx_ring[i]);
1781                         break;
1782                 }
1783         }
1784
1785         return err;
1786 }
1787
1788 /**
1789  * igb_setup_rctl - configure the receive control registers
1790  * @adapter: Board private structure
1791  **/
1792 static void igb_setup_rctl(struct igb_adapter *adapter)
1793 {
1794         struct e1000_hw *hw = &adapter->hw;
1795         u32 rctl;
1796         u32 srrctl = 0;
1797         int i, j;
1798
1799         rctl = rd32(E1000_RCTL);
1800
1801         rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
1802         rctl &= ~(E1000_RCTL_LBM_TCVR | E1000_RCTL_LBM_MAC);
1803
1804         rctl |= E1000_RCTL_EN | E1000_RCTL_BAM | E1000_RCTL_RDMTS_HALF |
1805                 (adapter->hw.mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
1806
1807         /*
1808          * enable stripping of CRC. It's unlikely this will break BMC
1809          * redirection as it did with e1000. Newer features require
1810          * that the HW strips the CRC.
1811         */
1812         rctl |= E1000_RCTL_SECRC;
1813
1814         /*
1815          * disable store bad packets and clear size bits.
1816          */
1817         rctl &= ~(E1000_RCTL_SBP | E1000_RCTL_SZ_256);
1818
1819         /* enable LPE when to prevent packets larger than max_frame_size */
1820                 rctl |= E1000_RCTL_LPE;
1821
1822         /* Setup buffer sizes */
1823         switch (adapter->rx_buffer_len) {
1824         case IGB_RXBUFFER_256:
1825                 rctl |= E1000_RCTL_SZ_256;
1826                 break;
1827         case IGB_RXBUFFER_512:
1828                 rctl |= E1000_RCTL_SZ_512;
1829                 break;
1830         default:
1831                 srrctl = ALIGN(adapter->rx_buffer_len, 1024)
1832                          >> E1000_SRRCTL_BSIZEPKT_SHIFT;
1833                 break;
1834         }
1835
1836         /* 82575 and greater support packet-split where the protocol
1837          * header is placed in skb->data and the packet data is
1838          * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
1839          * In the case of a non-split, skb->data is linearly filled,
1840          * followed by the page buffers.  Therefore, skb->data is
1841          * sized to hold the largest protocol header.
1842          */
1843         /* allocations using alloc_page take too long for regular MTU
1844          * so only enable packet split for jumbo frames */
1845         if (adapter->netdev->mtu > ETH_DATA_LEN) {
1846                 adapter->rx_ps_hdr_size = IGB_RXBUFFER_128;
1847                 srrctl |= adapter->rx_ps_hdr_size <<
1848                          E1000_SRRCTL_BSIZEHDRSIZE_SHIFT;
1849                 srrctl |= E1000_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
1850         } else {
1851                 adapter->rx_ps_hdr_size = 0;
1852                 srrctl |= E1000_SRRCTL_DESCTYPE_ADV_ONEBUF;
1853         }
1854
1855         for (i = 0; i < adapter->num_rx_queues; i++) {
1856                 j = adapter->rx_ring[i].reg_idx;
1857                 wr32(E1000_SRRCTL(j), srrctl);
1858         }
1859
1860         wr32(E1000_RCTL, rctl);
1861 }
1862
1863 /**
1864  * igb_configure_rx - Configure receive Unit after Reset
1865  * @adapter: board private structure
1866  *
1867  * Configure the Rx unit of the MAC after a reset.
1868  **/
1869 static void igb_configure_rx(struct igb_adapter *adapter)
1870 {
1871         u64 rdba;
1872         struct e1000_hw *hw = &adapter->hw;
1873         u32 rctl, rxcsum;
1874         u32 rxdctl;
1875         int i, j;
1876
1877         /* disable receives while setting up the descriptors */
1878         rctl = rd32(E1000_RCTL);
1879         wr32(E1000_RCTL, rctl & ~E1000_RCTL_EN);
1880         wrfl();
1881         mdelay(10);
1882
1883         if (adapter->itr_setting > 3)
1884                 wr32(E1000_ITR, adapter->itr);
1885
1886         /* Setup the HW Rx Head and Tail Descriptor Pointers and
1887          * the Base and Length of the Rx Descriptor Ring */
1888         for (i = 0; i < adapter->num_rx_queues; i++) {
1889                 struct igb_ring *ring = &(adapter->rx_ring[i]);
1890                 j = ring->reg_idx;
1891                 rdba = ring->dma;
1892                 wr32(E1000_RDBAL(j),
1893                                 rdba & 0x00000000ffffffffULL);
1894                 wr32(E1000_RDBAH(j), rdba >> 32);
1895                 wr32(E1000_RDLEN(j),
1896                                ring->count * sizeof(union e1000_adv_rx_desc));
1897
1898                 ring->head = E1000_RDH(j);
1899                 ring->tail = E1000_RDT(j);
1900                 writel(0, hw->hw_addr + ring->tail);
1901                 writel(0, hw->hw_addr + ring->head);
1902
1903                 rxdctl = rd32(E1000_RXDCTL(j));
1904                 rxdctl |= E1000_RXDCTL_QUEUE_ENABLE;
1905                 rxdctl &= 0xFFF00000;
1906                 rxdctl |= IGB_RX_PTHRESH;
1907                 rxdctl |= IGB_RX_HTHRESH << 8;
1908                 rxdctl |= IGB_RX_WTHRESH << 16;
1909                 wr32(E1000_RXDCTL(j), rxdctl);
1910 #ifdef CONFIG_IGB_LRO
1911                 /* Intitial LRO Settings */
1912                 ring->lro_mgr.max_aggr = MAX_LRO_AGGR;
1913                 ring->lro_mgr.max_desc = MAX_LRO_DESCRIPTORS;
1914                 ring->lro_mgr.get_skb_header = igb_get_skb_hdr;
1915                 ring->lro_mgr.features = LRO_F_NAPI | LRO_F_EXTRACT_VLAN_ID;
1916                 ring->lro_mgr.dev = adapter->netdev;
1917                 ring->lro_mgr.ip_summed = CHECKSUM_UNNECESSARY;
1918                 ring->lro_mgr.ip_summed_aggr = CHECKSUM_UNNECESSARY;
1919 #endif
1920         }
1921
1922         if (adapter->num_rx_queues > 1) {
1923                 u32 random[10];
1924                 u32 mrqc;
1925                 u32 j, shift;
1926                 union e1000_reta {
1927                         u32 dword;
1928                         u8  bytes[4];
1929                 } reta;
1930
1931                 get_random_bytes(&random[0], 40);
1932
1933                 if (hw->mac.type >= e1000_82576)
1934                         shift = 0;
1935                 else
1936                         shift = 6;
1937                 for (j = 0; j < (32 * 4); j++) {
1938                         reta.bytes[j & 3] =
1939                                 adapter->rx_ring[(j % adapter->num_rx_queues)].reg_idx << shift;
1940                         if ((j & 3) == 3)
1941                                 writel(reta.dword,
1942                                        hw->hw_addr + E1000_RETA(0) + (j & ~3));
1943                 }
1944                 mrqc = E1000_MRQC_ENABLE_RSS_4Q;
1945
1946                 /* Fill out hash function seeds */
1947                 for (j = 0; j < 10; j++)
1948                         array_wr32(E1000_RSSRK(0), j, random[j]);
1949
1950                 mrqc |= (E1000_MRQC_RSS_FIELD_IPV4 |
1951                          E1000_MRQC_RSS_FIELD_IPV4_TCP);
1952                 mrqc |= (E1000_MRQC_RSS_FIELD_IPV6 |
1953                          E1000_MRQC_RSS_FIELD_IPV6_TCP);
1954                 mrqc |= (E1000_MRQC_RSS_FIELD_IPV4_UDP |
1955                          E1000_MRQC_RSS_FIELD_IPV6_UDP);
1956                 mrqc |= (E1000_MRQC_RSS_FIELD_IPV6_UDP_EX |
1957                          E1000_MRQC_RSS_FIELD_IPV6_TCP_EX);
1958
1959
1960                 wr32(E1000_MRQC, mrqc);
1961
1962                 /* Multiqueue and raw packet checksumming are mutually
1963                  * exclusive.  Note that this not the same as TCP/IP
1964                  * checksumming, which works fine. */
1965                 rxcsum = rd32(E1000_RXCSUM);
1966                 rxcsum |= E1000_RXCSUM_PCSD;
1967                 wr32(E1000_RXCSUM, rxcsum);
1968         } else {
1969                 /* Enable Receive Checksum Offload for TCP and UDP */
1970                 rxcsum = rd32(E1000_RXCSUM);
1971                 if (adapter->rx_csum) {
1972                         rxcsum |= E1000_RXCSUM_TUOFL;
1973
1974                         /* Enable IPv4 payload checksum for UDP fragments
1975                          * Must be used in conjunction with packet-split. */
1976                         if (adapter->rx_ps_hdr_size)
1977                                 rxcsum |= E1000_RXCSUM_IPPCSE;
1978                 } else {
1979                         rxcsum &= ~E1000_RXCSUM_TUOFL;
1980                         /* don't need to clear IPPCSE as it defaults to 0 */
1981                 }
1982                 wr32(E1000_RXCSUM, rxcsum);
1983         }
1984
1985         if (adapter->vlgrp)
1986                 wr32(E1000_RLPML,
1987                                 adapter->max_frame_size + VLAN_TAG_SIZE);
1988         else
1989                 wr32(E1000_RLPML, adapter->max_frame_size);
1990
1991         /* Enable Receives */
1992         wr32(E1000_RCTL, rctl);
1993 }
1994
1995 /**
1996  * igb_free_tx_resources - Free Tx Resources per Queue
1997  * @tx_ring: Tx descriptor ring for a specific queue
1998  *
1999  * Free all transmit software resources
2000  **/
2001 void igb_free_tx_resources(struct igb_ring *tx_ring)
2002 {
2003         struct pci_dev *pdev = tx_ring->adapter->pdev;
2004
2005         igb_clean_tx_ring(tx_ring);
2006
2007         vfree(tx_ring->buffer_info);
2008         tx_ring->buffer_info = NULL;
2009
2010         pci_free_consistent(pdev, tx_ring->size, tx_ring->desc, tx_ring->dma);
2011
2012         tx_ring->desc = NULL;
2013 }
2014
2015 /**
2016  * igb_free_all_tx_resources - Free Tx Resources for All Queues
2017  * @adapter: board private structure
2018  *
2019  * Free all transmit software resources
2020  **/
2021 static void igb_free_all_tx_resources(struct igb_adapter *adapter)
2022 {
2023         int i;
2024
2025         for (i = 0; i < adapter->num_tx_queues; i++)
2026                 igb_free_tx_resources(&adapter->tx_ring[i]);
2027 }
2028
2029 static void igb_unmap_and_free_tx_resource(struct igb_adapter *adapter,
2030                                            struct igb_buffer *buffer_info)
2031 {
2032         if (buffer_info->dma) {
2033                 pci_unmap_page(adapter->pdev,
2034                                 buffer_info->dma,
2035                                 buffer_info->length,
2036                                 PCI_DMA_TODEVICE);
2037                 buffer_info->dma = 0;
2038         }
2039         if (buffer_info->skb) {
2040                 dev_kfree_skb_any(buffer_info->skb);
2041                 buffer_info->skb = NULL;
2042         }
2043         buffer_info->time_stamp = 0;
2044         /* buffer_info must be completely set up in the transmit path */
2045 }
2046
2047 /**
2048  * igb_clean_tx_ring - Free Tx Buffers
2049  * @tx_ring: ring to be cleaned
2050  **/
2051 static void igb_clean_tx_ring(struct igb_ring *tx_ring)
2052 {
2053         struct igb_adapter *adapter = tx_ring->adapter;
2054         struct igb_buffer *buffer_info;
2055         unsigned long size;
2056         unsigned int i;
2057
2058         if (!tx_ring->buffer_info)
2059                 return;
2060         /* Free all the Tx ring sk_buffs */
2061
2062         for (i = 0; i < tx_ring->count; i++) {
2063                 buffer_info = &tx_ring->buffer_info[i];
2064                 igb_unmap_and_free_tx_resource(adapter, buffer_info);
2065         }
2066
2067         size = sizeof(struct igb_buffer) * tx_ring->count;
2068         memset(tx_ring->buffer_info, 0, size);
2069
2070         /* Zero out the descriptor ring */
2071
2072         memset(tx_ring->desc, 0, tx_ring->size);
2073
2074         tx_ring->next_to_use = 0;
2075         tx_ring->next_to_clean = 0;
2076
2077         writel(0, adapter->hw.hw_addr + tx_ring->head);
2078         writel(0, adapter->hw.hw_addr + tx_ring->tail);
2079 }
2080
2081 /**
2082  * igb_clean_all_tx_rings - Free Tx Buffers for all queues
2083  * @adapter: board private structure
2084  **/
2085 static void igb_clean_all_tx_rings(struct igb_adapter *adapter)
2086 {
2087         int i;
2088
2089         for (i = 0; i < adapter->num_tx_queues; i++)
2090                 igb_clean_tx_ring(&adapter->tx_ring[i]);
2091 }
2092
2093 /**
2094  * igb_free_rx_resources - Free Rx Resources
2095  * @rx_ring: ring to clean the resources from
2096  *
2097  * Free all receive software resources
2098  **/
2099 void igb_free_rx_resources(struct igb_ring *rx_ring)
2100 {
2101         struct pci_dev *pdev = rx_ring->adapter->pdev;
2102
2103         igb_clean_rx_ring(rx_ring);
2104
2105         vfree(rx_ring->buffer_info);
2106         rx_ring->buffer_info = NULL;
2107
2108 #ifdef CONFIG_IGB_LRO
2109         vfree(rx_ring->lro_mgr.lro_arr);
2110         rx_ring->lro_mgr.lro_arr = NULL;
2111 #endif 
2112
2113         pci_free_consistent(pdev, rx_ring->size, rx_ring->desc, rx_ring->dma);
2114
2115         rx_ring->desc = NULL;
2116 }
2117
2118 /**
2119  * igb_free_all_rx_resources - Free Rx Resources for All Queues
2120  * @adapter: board private structure
2121  *
2122  * Free all receive software resources
2123  **/
2124 static void igb_free_all_rx_resources(struct igb_adapter *adapter)
2125 {
2126         int i;
2127
2128         for (i = 0; i < adapter->num_rx_queues; i++)
2129                 igb_free_rx_resources(&adapter->rx_ring[i]);
2130 }
2131
2132 /**
2133  * igb_clean_rx_ring - Free Rx Buffers per Queue
2134  * @rx_ring: ring to free buffers from
2135  **/
2136 static void igb_clean_rx_ring(struct igb_ring *rx_ring)
2137 {
2138         struct igb_adapter *adapter = rx_ring->adapter;
2139         struct igb_buffer *buffer_info;
2140         struct pci_dev *pdev = adapter->pdev;
2141         unsigned long size;
2142         unsigned int i;
2143
2144         if (!rx_ring->buffer_info)
2145                 return;
2146         /* Free all the Rx ring sk_buffs */
2147         for (i = 0; i < rx_ring->count; i++) {
2148                 buffer_info = &rx_ring->buffer_info[i];
2149                 if (buffer_info->dma) {
2150                         if (adapter->rx_ps_hdr_size)
2151                                 pci_unmap_single(pdev, buffer_info->dma,
2152                                                  adapter->rx_ps_hdr_size,
2153                                                  PCI_DMA_FROMDEVICE);
2154                         else
2155                                 pci_unmap_single(pdev, buffer_info->dma,
2156                                                  adapter->rx_buffer_len,
2157                                                  PCI_DMA_FROMDEVICE);
2158                         buffer_info->dma = 0;
2159                 }
2160
2161                 if (buffer_info->skb) {
2162                         dev_kfree_skb(buffer_info->skb);
2163                         buffer_info->skb = NULL;
2164                 }
2165                 if (buffer_info->page) {
2166                         if (buffer_info->page_dma)
2167                                 pci_unmap_page(pdev, buffer_info->page_dma,
2168                                                PAGE_SIZE / 2,
2169                                                PCI_DMA_FROMDEVICE);
2170                         put_page(buffer_info->page);
2171                         buffer_info->page = NULL;
2172                         buffer_info->page_dma = 0;
2173                         buffer_info->page_offset = 0;
2174                 }
2175         }
2176
2177         size = sizeof(struct igb_buffer) * rx_ring->count;
2178         memset(rx_ring->buffer_info, 0, size);
2179
2180         /* Zero out the descriptor ring */
2181         memset(rx_ring->desc, 0, rx_ring->size);
2182
2183         rx_ring->next_to_clean = 0;
2184         rx_ring->next_to_use = 0;
2185
2186         writel(0, adapter->hw.hw_addr + rx_ring->head);
2187         writel(0, adapter->hw.hw_addr + rx_ring->tail);
2188 }
2189
2190 /**
2191  * igb_clean_all_rx_rings - Free Rx Buffers for all queues
2192  * @adapter: board private structure
2193  **/
2194 static void igb_clean_all_rx_rings(struct igb_adapter *adapter)
2195 {
2196         int i;
2197
2198         for (i = 0; i < adapter->num_rx_queues; i++)
2199                 igb_clean_rx_ring(&adapter->rx_ring[i]);
2200 }
2201
2202 /**
2203  * igb_set_mac - Change the Ethernet Address of the NIC
2204  * @netdev: network interface device structure
2205  * @p: pointer to an address structure
2206  *
2207  * Returns 0 on success, negative on failure
2208  **/
2209 static int igb_set_mac(struct net_device *netdev, void *p)
2210 {
2211         struct igb_adapter *adapter = netdev_priv(netdev);
2212         struct sockaddr *addr = p;
2213
2214         if (!is_valid_ether_addr(addr->sa_data))
2215                 return -EADDRNOTAVAIL;
2216
2217         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
2218         memcpy(adapter->hw.mac.addr, addr->sa_data, netdev->addr_len);
2219
2220         adapter->hw.mac.ops.rar_set(&adapter->hw, adapter->hw.mac.addr, 0);
2221
2222         return 0;
2223 }
2224
2225 /**
2226  * igb_set_multi - Multicast and Promiscuous mode set
2227  * @netdev: network interface device structure
2228  *
2229  * The set_multi entry point is called whenever the multicast address
2230  * list or the network interface flags are updated.  This routine is
2231  * responsible for configuring the hardware for proper multicast,
2232  * promiscuous mode, and all-multi behavior.
2233  **/
2234 static void igb_set_multi(struct net_device *netdev)
2235 {
2236         struct igb_adapter *adapter = netdev_priv(netdev);
2237         struct e1000_hw *hw = &adapter->hw;
2238         struct e1000_mac_info *mac = &hw->mac;
2239         struct dev_mc_list *mc_ptr;
2240         u8  *mta_list;
2241         u32 rctl;
2242         int i;
2243
2244         /* Check for Promiscuous and All Multicast modes */
2245
2246         rctl = rd32(E1000_RCTL);
2247
2248         if (netdev->flags & IFF_PROMISC) {
2249                 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
2250                 rctl &= ~E1000_RCTL_VFE;
2251         } else {
2252                 if (netdev->flags & IFF_ALLMULTI) {
2253                         rctl |= E1000_RCTL_MPE;
2254                         rctl &= ~E1000_RCTL_UPE;
2255                 } else
2256                         rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
2257                 rctl |= E1000_RCTL_VFE;
2258         }
2259         wr32(E1000_RCTL, rctl);
2260
2261         if (!netdev->mc_count) {
2262                 /* nothing to program, so clear mc list */
2263                 igb_update_mc_addr_list_82575(hw, NULL, 0, 1,
2264                                           mac->rar_entry_count);
2265                 return;
2266         }
2267
2268         mta_list = kzalloc(netdev->mc_count * 6, GFP_ATOMIC);
2269         if (!mta_list)
2270                 return;
2271
2272         /* The shared function expects a packed array of only addresses. */
2273         mc_ptr = netdev->mc_list;
2274
2275         for (i = 0; i < netdev->mc_count; i++) {
2276                 if (!mc_ptr)
2277                         break;
2278                 memcpy(mta_list + (i*ETH_ALEN), mc_ptr->dmi_addr, ETH_ALEN);
2279                 mc_ptr = mc_ptr->next;
2280         }
2281         igb_update_mc_addr_list_82575(hw, mta_list, i, 1,
2282                                       mac->rar_entry_count);
2283         kfree(mta_list);
2284 }
2285
2286 /* Need to wait a few seconds after link up to get diagnostic information from
2287  * the phy */
2288 static void igb_update_phy_info(unsigned long data)
2289 {
2290         struct igb_adapter *adapter = (struct igb_adapter *) data;
2291         igb_get_phy_info(&adapter->hw);
2292 }
2293
2294 /**
2295  * igb_watchdog - Timer Call-back
2296  * @data: pointer to adapter cast into an unsigned long
2297  **/
2298 static void igb_watchdog(unsigned long data)
2299 {
2300         struct igb_adapter *adapter = (struct igb_adapter *)data;
2301         /* Do the rest outside of interrupt context */
2302         schedule_work(&adapter->watchdog_task);
2303 }
2304
2305 static void igb_watchdog_task(struct work_struct *work)
2306 {
2307         struct igb_adapter *adapter = container_of(work,
2308                                         struct igb_adapter, watchdog_task);
2309         struct e1000_hw *hw = &adapter->hw;
2310
2311         struct net_device *netdev = adapter->netdev;
2312         struct igb_ring *tx_ring = adapter->tx_ring;
2313         struct e1000_mac_info *mac = &adapter->hw.mac;
2314         u32 link;
2315         u32 eics = 0;
2316         s32 ret_val;
2317         int i;
2318
2319         if ((netif_carrier_ok(netdev)) &&
2320             (rd32(E1000_STATUS) & E1000_STATUS_LU))
2321                 goto link_up;
2322
2323         ret_val = hw->mac.ops.check_for_link(&adapter->hw);
2324         if ((ret_val == E1000_ERR_PHY) &&
2325             (hw->phy.type == e1000_phy_igp_3) &&
2326             (rd32(E1000_CTRL) &
2327              E1000_PHY_CTRL_GBE_DISABLE))
2328                 dev_info(&adapter->pdev->dev,
2329                          "Gigabit has been disabled, downgrading speed\n");
2330
2331         if ((hw->phy.media_type == e1000_media_type_internal_serdes) &&
2332             !(rd32(E1000_TXCW) & E1000_TXCW_ANE))
2333                 link = mac->serdes_has_link;
2334         else
2335                 link = rd32(E1000_STATUS) &
2336                                       E1000_STATUS_LU;
2337
2338         if (link) {
2339                 if (!netif_carrier_ok(netdev)) {
2340                         u32 ctrl;
2341                         hw->mac.ops.get_speed_and_duplex(&adapter->hw,
2342                                                    &adapter->link_speed,
2343                                                    &adapter->link_duplex);
2344
2345                         ctrl = rd32(E1000_CTRL);
2346                         /* Links status message must follow this format */
2347                         printk(KERN_INFO "igb: %s NIC Link is Up %d Mbps %s, "
2348                                  "Flow Control: %s\n",
2349                                  netdev->name,
2350                                  adapter->link_speed,
2351                                  adapter->link_duplex == FULL_DUPLEX ?
2352                                  "Full Duplex" : "Half Duplex",
2353                                  ((ctrl & E1000_CTRL_TFCE) && (ctrl &
2354                                  E1000_CTRL_RFCE)) ? "RX/TX" : ((ctrl &
2355                                  E1000_CTRL_RFCE) ? "RX" : ((ctrl &
2356                                  E1000_CTRL_TFCE) ? "TX" : "None")));
2357
2358                         /* tweak tx_queue_len according to speed/duplex and
2359                          * adjust the timeout factor */
2360                         netdev->tx_queue_len = adapter->tx_queue_len;
2361                         adapter->tx_timeout_factor = 1;
2362                         switch (adapter->link_speed) {
2363                         case SPEED_10:
2364                                 netdev->tx_queue_len = 10;
2365                                 adapter->tx_timeout_factor = 14;
2366                                 break;
2367                         case SPEED_100:
2368                                 netdev->tx_queue_len = 100;
2369                                 /* maybe add some timeout factor ? */
2370                                 break;
2371                         }
2372
2373                         netif_carrier_on(netdev);
2374                         netif_tx_wake_all_queues(netdev);
2375
2376                         if (!test_bit(__IGB_DOWN, &adapter->state))
2377                                 mod_timer(&adapter->phy_info_timer,
2378                                           round_jiffies(jiffies + 2 * HZ));
2379                 }
2380         } else {
2381                 if (netif_carrier_ok(netdev)) {
2382                         adapter->link_speed = 0;
2383                         adapter->link_duplex = 0;
2384                         /* Links status message must follow this format */
2385                         printk(KERN_INFO "igb: %s NIC Link is Down\n",
2386                                netdev->name);
2387                         netif_carrier_off(netdev);
2388                         netif_tx_stop_all_queues(netdev);
2389                         if (!test_bit(__IGB_DOWN, &adapter->state))
2390                                 mod_timer(&adapter->phy_info_timer,
2391                                           round_jiffies(jiffies + 2 * HZ));
2392                 }
2393         }
2394
2395 link_up:
2396         igb_update_stats(adapter);
2397
2398         mac->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
2399         adapter->tpt_old = adapter->stats.tpt;
2400         mac->collision_delta = adapter->stats.colc - adapter->colc_old;
2401         adapter->colc_old = adapter->stats.colc;
2402
2403         adapter->gorc = adapter->stats.gorc - adapter->gorc_old;
2404         adapter->gorc_old = adapter->stats.gorc;
2405         adapter->gotc = adapter->stats.gotc - adapter->gotc_old;
2406         adapter->gotc_old = adapter->stats.gotc;
2407
2408         igb_update_adaptive(&adapter->hw);
2409
2410         if (!netif_carrier_ok(netdev)) {
2411                 if (IGB_DESC_UNUSED(tx_ring) + 1 < tx_ring->count) {
2412                         /* We've lost link, so the controller stops DMA,
2413                          * but we've got queued Tx work that's never going
2414                          * to get done, so reset controller to flush Tx.
2415                          * (Do the reset outside of interrupt context). */
2416                         adapter->tx_timeout_count++;
2417                         schedule_work(&adapter->reset_task);
2418                 }
2419         }
2420
2421         /* Cause software interrupt to ensure rx ring is cleaned */
2422         if (adapter->msix_entries) {
2423                 for (i = 0; i < adapter->num_rx_queues; i++)
2424                         eics |= adapter->rx_ring[i].eims_value;
2425                 wr32(E1000_EICS, eics);
2426         } else {
2427                 wr32(E1000_ICS, E1000_ICS_RXDMT0);
2428         }
2429
2430         /* Force detection of hung controller every watchdog period */
2431         tx_ring->detect_tx_hung = true;
2432
2433         /* Reset the timer */
2434         if (!test_bit(__IGB_DOWN, &adapter->state))
2435                 mod_timer(&adapter->watchdog_timer,
2436                           round_jiffies(jiffies + 2 * HZ));
2437 }
2438
2439 enum latency_range {
2440         lowest_latency = 0,
2441         low_latency = 1,
2442         bulk_latency = 2,
2443         latency_invalid = 255
2444 };
2445
2446
2447 /**
2448  * igb_update_ring_itr - update the dynamic ITR value based on packet size
2449  *
2450  *      Stores a new ITR value based on strictly on packet size.  This
2451  *      algorithm is less sophisticated than that used in igb_update_itr,
2452  *      due to the difficulty of synchronizing statistics across multiple
2453  *      receive rings.  The divisors and thresholds used by this fuction
2454  *      were determined based on theoretical maximum wire speed and testing
2455  *      data, in order to minimize response time while increasing bulk
2456  *      throughput.
2457  *      This functionality is controlled by the InterruptThrottleRate module
2458  *      parameter (see igb_param.c)
2459  *      NOTE:  This function is called only when operating in a multiqueue
2460  *             receive environment.
2461  * @rx_ring: pointer to ring
2462  **/
2463 static void igb_update_ring_itr(struct igb_ring *rx_ring)
2464 {
2465         int new_val = rx_ring->itr_val;
2466         int avg_wire_size = 0;
2467         struct igb_adapter *adapter = rx_ring->adapter;
2468
2469         if (!rx_ring->total_packets)
2470                 goto clear_counts; /* no packets, so don't do anything */
2471
2472         /* For non-gigabit speeds, just fix the interrupt rate at 4000
2473          * ints/sec - ITR timer value of 120 ticks.
2474          */
2475         if (adapter->link_speed != SPEED_1000) {
2476                 new_val = 120;
2477                 goto set_itr_val;
2478         }
2479         avg_wire_size = rx_ring->total_bytes / rx_ring->total_packets;
2480
2481         /* Add 24 bytes to size to account for CRC, preamble, and gap */
2482         avg_wire_size += 24;
2483
2484         /* Don't starve jumbo frames */
2485         avg_wire_size = min(avg_wire_size, 3000);
2486
2487         /* Give a little boost to mid-size frames */
2488         if ((avg_wire_size > 300) && (avg_wire_size < 1200))
2489                 new_val = avg_wire_size / 3;
2490         else
2491                 new_val = avg_wire_size / 2;
2492
2493 set_itr_val:
2494         if (new_val != rx_ring->itr_val) {
2495                 rx_ring->itr_val = new_val;
2496                 rx_ring->set_itr = 1;
2497         }
2498 clear_counts:
2499         rx_ring->total_bytes = 0;
2500         rx_ring->total_packets = 0;
2501 }
2502
2503 /**
2504  * igb_update_itr - update the dynamic ITR value based on statistics
2505  *      Stores a new ITR value based on packets and byte
2506  *      counts during the last interrupt.  The advantage of per interrupt
2507  *      computation is faster updates and more accurate ITR for the current
2508  *      traffic pattern.  Constants in this function were computed
2509  *      based on theoretical maximum wire speed and thresholds were set based
2510  *      on testing data as well as attempting to minimize response time
2511  *      while increasing bulk throughput.
2512  *      this functionality is controlled by the InterruptThrottleRate module
2513  *      parameter (see igb_param.c)
2514  *      NOTE:  These calculations are only valid when operating in a single-
2515  *             queue environment.
2516  * @adapter: pointer to adapter
2517  * @itr_setting: current adapter->itr
2518  * @packets: the number of packets during this measurement interval
2519  * @bytes: the number of bytes during this measurement interval
2520  **/
2521 static unsigned int igb_update_itr(struct igb_adapter *adapter, u16 itr_setting,
2522                                    int packets, int bytes)
2523 {
2524         unsigned int retval = itr_setting;
2525
2526         if (packets == 0)
2527                 goto update_itr_done;
2528
2529         switch (itr_setting) {
2530         case lowest_latency:
2531                 /* handle TSO and jumbo frames */
2532                 if (bytes/packets > 8000)
2533                         retval = bulk_latency;
2534                 else if ((packets < 5) && (bytes > 512))
2535                         retval = low_latency;
2536                 break;
2537         case low_latency:  /* 50 usec aka 20000 ints/s */
2538                 if (bytes > 10000) {
2539                         /* this if handles the TSO accounting */
2540                         if (bytes/packets > 8000) {
2541                                 retval = bulk_latency;
2542                         } else if ((packets < 10) || ((bytes/packets) > 1200)) {
2543                                 retval = bulk_latency;
2544                         } else if ((packets > 35)) {
2545                                 retval = lowest_latency;
2546                         }
2547                 } else if (bytes/packets > 2000) {
2548                         retval = bulk_latency;
2549                 } else if (packets <= 2 && bytes < 512) {
2550                         retval = lowest_latency;
2551                 }
2552                 break;
2553         case bulk_latency: /* 250 usec aka 4000 ints/s */
2554                 if (bytes > 25000) {
2555                         if (packets > 35)
2556                                 retval = low_latency;
2557                 } else if (bytes < 6000) {
2558                         retval = low_latency;
2559                 }
2560                 break;
2561         }
2562
2563 update_itr_done:
2564         return retval;
2565 }
2566
2567 static void igb_set_itr(struct igb_adapter *adapter)
2568 {
2569         u16 current_itr;
2570         u32 new_itr = adapter->itr;
2571
2572         /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
2573         if (adapter->link_speed != SPEED_1000) {
2574                 current_itr = 0;
2575                 new_itr = 4000;
2576                 goto set_itr_now;
2577         }
2578
2579         adapter->rx_itr = igb_update_itr(adapter,
2580                                     adapter->rx_itr,
2581                                     adapter->rx_ring->total_packets,
2582                                     adapter->rx_ring->total_bytes);
2583
2584         if (adapter->rx_ring->buddy) {
2585                 adapter->tx_itr = igb_update_itr(adapter,
2586                                             adapter->tx_itr,
2587                                             adapter->tx_ring->total_packets,
2588                                             adapter->tx_ring->total_bytes);
2589
2590                 current_itr = max(adapter->rx_itr, adapter->tx_itr);
2591         } else {
2592                 current_itr = adapter->rx_itr;
2593         }
2594
2595         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2596         if (adapter->itr_setting == 3 &&
2597             current_itr == lowest_latency)
2598                 current_itr = low_latency;
2599
2600         switch (current_itr) {
2601         /* counts and packets in update_itr are dependent on these numbers */
2602         case lowest_latency:
2603                 new_itr = 70000;
2604                 break;
2605         case low_latency:
2606                 new_itr = 20000; /* aka hwitr = ~200 */
2607                 break;
2608         case bulk_latency:
2609                 new_itr = 4000;
2610                 break;
2611         default:
2612                 break;
2613         }
2614
2615 set_itr_now:
2616         adapter->rx_ring->total_bytes = 0;
2617         adapter->rx_ring->total_packets = 0;
2618         if (adapter->rx_ring->buddy) {
2619                 adapter->rx_ring->buddy->total_bytes = 0;
2620                 adapter->rx_ring->buddy->total_packets = 0;
2621         }
2622
2623         if (new_itr != adapter->itr) {
2624                 /* this attempts to bias the interrupt rate towards Bulk
2625                  * by adding intermediate steps when interrupt rate is
2626                  * increasing */
2627                 new_itr = new_itr > adapter->itr ?
2628                              min(adapter->itr + (new_itr >> 2), new_itr) :
2629                              new_itr;
2630                 /* Don't write the value here; it resets the adapter's
2631                  * internal timer, and causes us to delay far longer than
2632                  * we should between interrupts.  Instead, we write the ITR
2633                  * value at the beginning of the next interrupt so the timing
2634                  * ends up being correct.
2635                  */
2636                 adapter->itr = new_itr;
2637                 adapter->rx_ring->itr_val = 1000000000 / (new_itr * 256);
2638                 adapter->rx_ring->set_itr = 1;
2639         }
2640
2641         return;
2642 }
2643
2644
2645 #define IGB_TX_FLAGS_CSUM               0x00000001
2646 #define IGB_TX_FLAGS_VLAN               0x00000002
2647 #define IGB_TX_FLAGS_TSO                0x00000004
2648 #define IGB_TX_FLAGS_IPV4               0x00000008
2649 #define IGB_TX_FLAGS_VLAN_MASK  0xffff0000
2650 #define IGB_TX_FLAGS_VLAN_SHIFT 16
2651
2652 static inline int igb_tso_adv(struct igb_adapter *adapter,
2653                               struct igb_ring *tx_ring,
2654                               struct sk_buff *skb, u32 tx_flags, u8 *hdr_len)
2655 {
2656         struct e1000_adv_tx_context_desc *context_desc;
2657         unsigned int i;
2658         int err;
2659         struct igb_buffer *buffer_info;
2660         u32 info = 0, tu_cmd = 0;
2661         u32 mss_l4len_idx, l4len;
2662         *hdr_len = 0;
2663
2664         if (skb_header_cloned(skb)) {
2665                 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2666                 if (err)
2667                         return err;
2668         }
2669
2670         l4len = tcp_hdrlen(skb);
2671         *hdr_len += l4len;
2672
2673         if (skb->protocol == htons(ETH_P_IP)) {
2674                 struct iphdr *iph = ip_hdr(skb);
2675                 iph->tot_len = 0;
2676                 iph->check = 0;
2677                 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
2678                                                          iph->daddr, 0,
2679                                                          IPPROTO_TCP,
2680                                                          0);
2681         } else if (skb_shinfo(skb)->gso_type == SKB_GSO_TCPV6) {
2682                 ipv6_hdr(skb)->payload_len = 0;
2683                 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
2684                                                        &ipv6_hdr(skb)->daddr,
2685                                                        0, IPPROTO_TCP, 0);
2686         }
2687
2688         i = tx_ring->next_to_use;
2689
2690         buffer_info = &tx_ring->buffer_info[i];
2691         context_desc = E1000_TX_CTXTDESC_ADV(*tx_ring, i);
2692         /* VLAN MACLEN IPLEN */
2693         if (tx_flags & IGB_TX_FLAGS_VLAN)
2694                 info |= (tx_flags & IGB_TX_FLAGS_VLAN_MASK);
2695         info |= (skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT);
2696         *hdr_len += skb_network_offset(skb);
2697         info |= skb_network_header_len(skb);
2698         *hdr_len += skb_network_header_len(skb);
2699         context_desc->vlan_macip_lens = cpu_to_le32(info);
2700
2701         /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
2702         tu_cmd |= (E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT);
2703
2704         if (skb->protocol == htons(ETH_P_IP))
2705                 tu_cmd |= E1000_ADVTXD_TUCMD_IPV4;
2706         tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
2707
2708         context_desc->type_tucmd_mlhl = cpu_to_le32(tu_cmd);
2709
2710         /* MSS L4LEN IDX */
2711         mss_l4len_idx = (skb_shinfo(skb)->gso_size << E1000_ADVTXD_MSS_SHIFT);
2712         mss_l4len_idx |= (l4len << E1000_ADVTXD_L4LEN_SHIFT);
2713
2714         /* Context index must be unique per ring. */
2715         if (adapter->flags & IGB_FLAG_NEED_CTX_IDX)
2716                 mss_l4len_idx |= tx_ring->queue_index << 4;
2717
2718         context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
2719         context_desc->seqnum_seed = 0;
2720
2721         buffer_info->time_stamp = jiffies;
2722         buffer_info->next_to_watch = i;
2723         buffer_info->dma = 0;
2724         i++;
2725         if (i == tx_ring->count)
2726                 i = 0;
2727
2728         tx_ring->next_to_use = i;
2729
2730         return true;
2731 }
2732
2733 static inline bool igb_tx_csum_adv(struct igb_adapter *adapter,
2734                                         struct igb_ring *tx_ring,
2735                                         struct sk_buff *skb, u32 tx_flags)
2736 {
2737         struct e1000_adv_tx_context_desc *context_desc;
2738         unsigned int i;
2739         struct igb_buffer *buffer_info;
2740         u32 info = 0, tu_cmd = 0;
2741
2742         if ((skb->ip_summed == CHECKSUM_PARTIAL) ||
2743             (tx_flags & IGB_TX_FLAGS_VLAN)) {
2744                 i = tx_ring->next_to_use;
2745                 buffer_info = &tx_ring->buffer_info[i];
2746                 context_desc = E1000_TX_CTXTDESC_ADV(*tx_ring, i);
2747
2748                 if (tx_flags & IGB_TX_FLAGS_VLAN)
2749                         info |= (tx_flags & IGB_TX_FLAGS_VLAN_MASK);
2750                 info |= (skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT);
2751                 if (skb->ip_summed == CHECKSUM_PARTIAL)
2752                         info |= skb_network_header_len(skb);
2753
2754                 context_desc->vlan_macip_lens = cpu_to_le32(info);
2755
2756                 tu_cmd |= (E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT);
2757
2758                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2759                         switch (skb->protocol) {
2760                         case __constant_htons(ETH_P_IP):
2761                                 tu_cmd |= E1000_ADVTXD_TUCMD_IPV4;
2762                                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
2763                                         tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
2764                                 break;
2765                         case __constant_htons(ETH_P_IPV6):
2766                                 /* XXX what about other V6 headers?? */
2767                                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
2768                                         tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
2769                                 break;
2770                         default:
2771                                 if (unlikely(net_ratelimit()))
2772                                         dev_warn(&adapter->pdev->dev,
2773                                             "partial checksum but proto=%x!\n",
2774                                             skb->protocol);
2775                                 break;
2776                         }
2777                 }
2778
2779                 context_desc->type_tucmd_mlhl = cpu_to_le32(tu_cmd);
2780                 context_desc->seqnum_seed = 0;
2781                 if (adapter->flags & IGB_FLAG_NEED_CTX_IDX)
2782                         context_desc->mss_l4len_idx =
2783                                 cpu_to_le32(tx_ring->queue_index << 4);
2784
2785                 buffer_info->time_stamp = jiffies;
2786                 buffer_info->next_to_watch = i;
2787                 buffer_info->dma = 0;
2788
2789                 i++;
2790                 if (i == tx_ring->count)
2791                         i = 0;
2792                 tx_ring->next_to_use = i;
2793
2794                 return true;
2795         }
2796
2797
2798         return false;
2799 }
2800
2801 #define IGB_MAX_TXD_PWR 16
2802 #define IGB_MAX_DATA_PER_TXD    (1<<IGB_MAX_TXD_PWR)
2803
2804 static inline int igb_tx_map_adv(struct igb_adapter *adapter,
2805                                  struct igb_ring *tx_ring, struct sk_buff *skb,
2806                                  unsigned int first)
2807 {
2808         struct igb_buffer *buffer_info;
2809         unsigned int len = skb_headlen(skb);
2810         unsigned int count = 0, i;
2811         unsigned int f;
2812
2813         i = tx_ring->next_to_use;
2814
2815         buffer_info = &tx_ring->buffer_info[i];
2816         BUG_ON(len >= IGB_MAX_DATA_PER_TXD);
2817         buffer_info->length = len;
2818         /* set time_stamp *before* dma to help avoid a possible race */
2819         buffer_info->time_stamp = jiffies;
2820         buffer_info->next_to_watch = i;
2821         buffer_info->dma = pci_map_single(adapter->pdev, skb->data, len,
2822                                           PCI_DMA_TODEVICE);
2823         count++;
2824         i++;
2825         if (i == tx_ring->count)
2826                 i = 0;
2827
2828         for (f = 0; f < skb_shinfo(skb)->nr_frags; f++) {
2829                 struct skb_frag_struct *frag;
2830
2831                 frag = &skb_shinfo(skb)->frags[f];
2832                 len = frag->size;
2833
2834                 buffer_info = &tx_ring->buffer_info[i];
2835                 BUG_ON(len >= IGB_MAX_DATA_PER_TXD);
2836                 buffer_info->length = len;
2837                 buffer_info->time_stamp = jiffies;
2838                 buffer_info->next_to_watch = i;
2839                 buffer_info->dma = pci_map_page(adapter->pdev,
2840                                                 frag->page,
2841                                                 frag->page_offset,
2842                                                 len,
2843                                                 PCI_DMA_TODEVICE);
2844
2845                 count++;
2846                 i++;
2847                 if (i == tx_ring->count)
2848                         i = 0;
2849         }
2850
2851         i = ((i == 0) ? tx_ring->count - 1 : i - 1);
2852         tx_ring->buffer_info[i].skb = skb;
2853         tx_ring->buffer_info[first].next_to_watch = i;
2854
2855         return count;
2856 }
2857
2858 static inline void igb_tx_queue_adv(struct igb_adapter *adapter,
2859                                     struct igb_ring *tx_ring,
2860                                     int tx_flags, int count, u32 paylen,
2861                                     u8 hdr_len)
2862 {
2863         union e1000_adv_tx_desc *tx_desc = NULL;
2864         struct igb_buffer *buffer_info;
2865         u32 olinfo_status = 0, cmd_type_len;
2866         unsigned int i;
2867
2868         cmd_type_len = (E1000_ADVTXD_DTYP_DATA | E1000_ADVTXD_DCMD_IFCS |
2869                         E1000_ADVTXD_DCMD_DEXT);
2870
2871         if (tx_flags & IGB_TX_FLAGS_VLAN)
2872                 cmd_type_len |= E1000_ADVTXD_DCMD_VLE;
2873
2874         if (tx_flags & IGB_TX_FLAGS_TSO) {
2875                 cmd_type_len |= E1000_ADVTXD_DCMD_TSE;
2876
2877                 /* insert tcp checksum */
2878                 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
2879
2880                 /* insert ip checksum */
2881                 if (tx_flags & IGB_TX_FLAGS_IPV4)
2882                         olinfo_status |= E1000_TXD_POPTS_IXSM << 8;
2883
2884         } else if (tx_flags & IGB_TX_FLAGS_CSUM) {
2885                 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
2886         }
2887
2888         if ((adapter->flags & IGB_FLAG_NEED_CTX_IDX) &&
2889             (tx_flags & (IGB_TX_FLAGS_CSUM | IGB_TX_FLAGS_TSO |
2890                          IGB_TX_FLAGS_VLAN)))
2891                 olinfo_status |= tx_ring->queue_index << 4;
2892
2893         olinfo_status |= ((paylen - hdr_len) << E1000_ADVTXD_PAYLEN_SHIFT);
2894
2895         i = tx_ring->next_to_use;
2896         while (count--) {
2897                 buffer_info = &tx_ring->buffer_info[i];
2898                 tx_desc = E1000_TX_DESC_ADV(*tx_ring, i);
2899                 tx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma);
2900                 tx_desc->read.cmd_type_len =
2901                         cpu_to_le32(cmd_type_len | buffer_info->length);
2902                 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
2903                 i++;
2904                 if (i == tx_ring->count)
2905                         i = 0;
2906         }
2907
2908         tx_desc->read.cmd_type_len |= cpu_to_le32(adapter->txd_cmd);
2909         /* Force memory writes to complete before letting h/w
2910          * know there are new descriptors to fetch.  (Only
2911          * applicable for weak-ordered memory model archs,
2912          * such as IA-64). */
2913         wmb();
2914
2915         tx_ring->next_to_use = i;
2916         writel(i, adapter->hw.hw_addr + tx_ring->tail);
2917         /* we need this if more than one processor can write to our tail
2918          * at a time, it syncronizes IO on IA64/Altix systems */
2919         mmiowb();
2920 }
2921
2922 static int __igb_maybe_stop_tx(struct net_device *netdev,
2923                                struct igb_ring *tx_ring, int size)
2924 {
2925         struct igb_adapter *adapter = netdev_priv(netdev);
2926
2927         netif_stop_subqueue(netdev, tx_ring->queue_index);
2928
2929         /* Herbert's original patch had:
2930          *  smp_mb__after_netif_stop_queue();
2931          * but since that doesn't exist yet, just open code it. */
2932         smp_mb();
2933
2934         /* We need to check again in a case another CPU has just
2935          * made room available. */
2936         if (IGB_DESC_UNUSED(tx_ring) < size)
2937                 return -EBUSY;
2938
2939         /* A reprieve! */
2940         netif_wake_subqueue(netdev, tx_ring->queue_index);
2941         ++adapter->restart_queue;
2942         return 0;
2943 }
2944
2945 static int igb_maybe_stop_tx(struct net_device *netdev,
2946                              struct igb_ring *tx_ring, int size)
2947 {
2948         if (IGB_DESC_UNUSED(tx_ring) >= size)
2949                 return 0;
2950         return __igb_maybe_stop_tx(netdev, tx_ring, size);
2951 }
2952
2953 #define TXD_USE_COUNT(S) (((S) >> (IGB_MAX_TXD_PWR)) + 1)
2954
2955 static int igb_xmit_frame_ring_adv(struct sk_buff *skb,
2956                                    struct net_device *netdev,
2957                                    struct igb_ring *tx_ring)
2958 {
2959         struct igb_adapter *adapter = netdev_priv(netdev);
2960         unsigned int first;
2961         unsigned int tx_flags = 0;
2962         unsigned int len;
2963         u8 hdr_len = 0;
2964         int tso = 0;
2965
2966         len = skb_headlen(skb);
2967
2968         if (test_bit(__IGB_DOWN, &adapter->state)) {
2969                 dev_kfree_skb_any(skb);
2970                 return NETDEV_TX_OK;
2971         }
2972
2973         if (skb->len <= 0) {
2974                 dev_kfree_skb_any(skb);
2975                 return NETDEV_TX_OK;
2976         }
2977
2978         /* need: 1 descriptor per page,
2979          *       + 2 desc gap to keep tail from touching head,
2980          *       + 1 desc for skb->data,
2981          *       + 1 desc for context descriptor,
2982          * otherwise try next time */
2983         if (igb_maybe_stop_tx(netdev, tx_ring, skb_shinfo(skb)->nr_frags + 4)) {
2984                 /* this is a hard error */
2985                 return NETDEV_TX_BUSY;
2986         }
2987         skb_orphan(skb);
2988
2989         if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
2990                 tx_flags |= IGB_TX_FLAGS_VLAN;
2991                 tx_flags |= (vlan_tx_tag_get(skb) << IGB_TX_FLAGS_VLAN_SHIFT);
2992         }
2993
2994         if (skb->protocol == htons(ETH_P_IP))
2995                 tx_flags |= IGB_TX_FLAGS_IPV4;
2996
2997         first = tx_ring->next_to_use;
2998
2999         tso = skb_is_gso(skb) ? igb_tso_adv(adapter, tx_ring, skb, tx_flags,
3000                                               &hdr_len) : 0;
3001
3002         if (tso < 0) {
3003                 dev_kfree_skb_any(skb);
3004                 return NETDEV_TX_OK;
3005         }
3006
3007         if (tso)
3008                 tx_flags |= IGB_TX_FLAGS_TSO;
3009         else if (igb_tx_csum_adv(adapter, tx_ring, skb, tx_flags))
3010                         if (skb->ip_summed == CHECKSUM_PARTIAL)
3011                                 tx_flags |= IGB_TX_FLAGS_CSUM;
3012
3013         igb_tx_queue_adv(adapter, tx_ring, tx_flags,
3014                          igb_tx_map_adv(adapter, tx_ring, skb, first),
3015                          skb->len, hdr_len);
3016
3017         netdev->trans_start = jiffies;
3018
3019         /* Make sure there is space in the ring for the next send. */
3020         igb_maybe_stop_tx(netdev, tx_ring, MAX_SKB_FRAGS + 4);
3021
3022         return NETDEV_TX_OK;
3023 }
3024
3025 static int igb_xmit_frame_adv(struct sk_buff *skb, struct net_device *netdev)
3026 {
3027         struct igb_adapter *adapter = netdev_priv(netdev);
3028         struct igb_ring *tx_ring;
3029
3030         int r_idx = 0;
3031         r_idx = skb->queue_mapping & (IGB_MAX_TX_QUEUES - 1);
3032         tx_ring = adapter->multi_tx_table[r_idx];
3033
3034         /* This goes back to the question of how to logically map a tx queue
3035          * to a flow.  Right now, performance is impacted slightly negatively
3036          * if using multiple tx queues.  If the stack breaks away from a
3037          * single qdisc implementation, we can look at this again. */
3038         return (igb_xmit_frame_ring_adv(skb, netdev, tx_ring));
3039 }
3040
3041 /**
3042  * igb_tx_timeout - Respond to a Tx Hang
3043  * @netdev: network interface device structure
3044  **/
3045 static void igb_tx_timeout(struct net_device *netdev)
3046 {
3047         struct igb_adapter *adapter = netdev_priv(netdev);
3048         struct e1000_hw *hw = &adapter->hw;
3049
3050         /* Do the reset outside of interrupt context */
3051         adapter->tx_timeout_count++;
3052         schedule_work(&adapter->reset_task);
3053         wr32(E1000_EICS, adapter->eims_enable_mask &
3054                 ~(E1000_EIMS_TCP_TIMER | E1000_EIMS_OTHER));
3055 }
3056
3057 static void igb_reset_task(struct work_struct *work)
3058 {
3059         struct igb_adapter *adapter;
3060         adapter = container_of(work, struct igb_adapter, reset_task);
3061
3062         igb_reinit_locked(adapter);
3063 }
3064
3065 /**
3066  * igb_get_stats - Get System Network Statistics
3067  * @netdev: network interface device structure
3068  *
3069  * Returns the address of the device statistics structure.
3070  * The statistics are actually updated from the timer callback.
3071  **/
3072 static struct net_device_stats *
3073 igb_get_stats(struct net_device *netdev)
3074 {
3075         struct igb_adapter *adapter = netdev_priv(netdev);
3076
3077         /* only return the current stats */
3078         return &adapter->net_stats;
3079 }
3080
3081 /**
3082  * igb_change_mtu - Change the Maximum Transfer Unit
3083  * @netdev: network interface device structure
3084  * @new_mtu: new value for maximum frame size
3085  *
3086  * Returns 0 on success, negative on failure
3087  **/
3088 static int igb_change_mtu(struct net_device *netdev, int new_mtu)
3089 {
3090         struct igb_adapter *adapter = netdev_priv(netdev);
3091         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
3092
3093         if ((max_frame < ETH_ZLEN + ETH_FCS_LEN) ||
3094             (max_frame > MAX_JUMBO_FRAME_SIZE)) {
3095                 dev_err(&adapter->pdev->dev, "Invalid MTU setting\n");
3096                 return -EINVAL;
3097         }
3098
3099 #define MAX_STD_JUMBO_FRAME_SIZE 9234
3100         if (max_frame > MAX_STD_JUMBO_FRAME_SIZE) {
3101                 dev_err(&adapter->pdev->dev, "MTU > 9216 not supported.\n");
3102                 return -EINVAL;
3103         }
3104
3105         while (test_and_set_bit(__IGB_RESETTING, &adapter->state))
3106                 msleep(1);
3107         /* igb_down has a dependency on max_frame_size */
3108         adapter->max_frame_size = max_frame;
3109         if (netif_running(netdev))
3110                 igb_down(adapter);
3111
3112         /* NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
3113          * means we reserve 2 more, this pushes us to allocate from the next
3114          * larger slab size.
3115          * i.e. RXBUFFER_2048 --> size-4096 slab
3116          */
3117
3118         if (max_frame <= IGB_RXBUFFER_256)
3119                 adapter->rx_buffer_len = IGB_RXBUFFER_256;
3120         else if (max_frame <= IGB_RXBUFFER_512)
3121                 adapter->rx_buffer_len = IGB_RXBUFFER_512;
3122         else if (max_frame <= IGB_RXBUFFER_1024)
3123                 adapter->rx_buffer_len = IGB_RXBUFFER_1024;
3124         else if (max_frame <= IGB_RXBUFFER_2048)
3125                 adapter->rx_buffer_len = IGB_RXBUFFER_2048;
3126         else
3127 #if (PAGE_SIZE / 2) > IGB_RXBUFFER_16384
3128                 adapter->rx_buffer_len = IGB_RXBUFFER_16384;
3129 #else
3130                 adapter->rx_buffer_len = PAGE_SIZE / 2;
3131 #endif
3132         /* adjust allocation if LPE protects us, and we aren't using SBP */
3133         if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
3134              (max_frame == MAXIMUM_ETHERNET_VLAN_SIZE))
3135                 adapter->rx_buffer_len = MAXIMUM_ETHERNET_VLAN_SIZE;
3136
3137         dev_info(&adapter->pdev->dev, "changing MTU from %d to %d\n",
3138                  netdev->mtu, new_mtu);
3139         netdev->mtu = new_mtu;
3140
3141         if (netif_running(netdev))
3142                 igb_up(adapter);
3143         else
3144                 igb_reset(adapter);
3145
3146         clear_bit(__IGB_RESETTING, &adapter->state);
3147
3148         return 0;
3149 }
3150
3151 /**
3152  * igb_update_stats - Update the board statistics counters
3153  * @adapter: board private structure
3154  **/
3155
3156 void igb_update_stats(struct igb_adapter *adapter)
3157 {
3158         struct e1000_hw *hw = &adapter->hw;
3159         struct pci_dev *pdev = adapter->pdev;
3160         u16 phy_tmp;
3161
3162 #define PHY_IDLE_ERROR_COUNT_MASK 0x00FF
3163
3164         /*
3165          * Prevent stats update while adapter is being reset, or if the pci
3166          * connection is down.
3167          */
3168         if (adapter->link_speed == 0)
3169                 return;
3170         if (pci_channel_offline(pdev))
3171                 return;
3172
3173         adapter->stats.crcerrs += rd32(E1000_CRCERRS);
3174         adapter->stats.gprc += rd32(E1000_GPRC);
3175         adapter->stats.gorc += rd32(E1000_GORCL);
3176         rd32(E1000_GORCH); /* clear GORCL */
3177         adapter->stats.bprc += rd32(E1000_BPRC);
3178         adapter->stats.mprc += rd32(E1000_MPRC);
3179         adapter->stats.roc += rd32(E1000_ROC);
3180
3181         adapter->stats.prc64 += rd32(E1000_PRC64);
3182         adapter->stats.prc127 += rd32(E1000_PRC127);
3183         adapter->stats.prc255 += rd32(E1000_PRC255);
3184         adapter->stats.prc511 += rd32(E1000_PRC511);
3185         adapter->stats.prc1023 += rd32(E1000_PRC1023);
3186         adapter->stats.prc1522 += rd32(E1000_PRC1522);
3187         adapter->stats.symerrs += rd32(E1000_SYMERRS);
3188         adapter->stats.sec += rd32(E1000_SEC);
3189
3190         adapter->stats.mpc += rd32(E1000_MPC);
3191         adapter->stats.scc += rd32(E1000_SCC);
3192         adapter->stats.ecol += rd32(E1000_ECOL);
3193         adapter->stats.mcc += rd32(E1000_MCC);
3194         adapter->stats.latecol += rd32(E1000_LATECOL);
3195         adapter->stats.dc += rd32(E1000_DC);
3196         adapter->stats.rlec += rd32(E1000_RLEC);
3197         adapter->stats.xonrxc += rd32(E1000_XONRXC);
3198         adapter->stats.xontxc += rd32(E1000_XONTXC);
3199         adapter->stats.xoffrxc += rd32(E1000_XOFFRXC);
3200         adapter->stats.xofftxc += rd32(E1000_XOFFTXC);
3201         adapter->stats.fcruc += rd32(E1000_FCRUC);
3202         adapter->stats.gptc += rd32(E1000_GPTC);
3203         adapter->stats.gotc += rd32(E1000_GOTCL);
3204         rd32(E1000_GOTCH); /* clear GOTCL */
3205         adapter->stats.rnbc += rd32(E1000_RNBC);
3206         adapter->stats.ruc += rd32(E1000_RUC);
3207         adapter->stats.rfc += rd32(E1000_RFC);
3208         adapter->stats.rjc += rd32(E1000_RJC);
3209         adapter->stats.tor += rd32(E1000_TORH);
3210         adapter->stats.tot += rd32(E1000_TOTH);
3211         adapter->stats.tpr += rd32(E1000_TPR);
3212
3213         adapter->stats.ptc64 += rd32(E1000_PTC64);
3214         adapter->stats.ptc127 += rd32(E1000_PTC127);
3215         adapter->stats.ptc255 += rd32(E1000_PTC255);
3216         adapter->stats.ptc511 += rd32(E1000_PTC511);
3217         adapter->stats.ptc1023 += rd32(E1000_PTC1023);
3218         adapter->stats.ptc1522 += rd32(E1000_PTC1522);
3219
3220         adapter->stats.mptc += rd32(E1000_MPTC);
3221         adapter->stats.bptc += rd32(E1000_BPTC);
3222
3223         /* used for adaptive IFS */
3224
3225         hw->mac.tx_packet_delta = rd32(E1000_TPT);
3226         adapter->stats.tpt += hw->mac.tx_packet_delta;
3227         hw->mac.collision_delta = rd32(E1000_COLC);
3228         adapter->stats.colc += hw->mac.collision_delta;
3229
3230         adapter->stats.algnerrc += rd32(E1000_ALGNERRC);
3231         adapter->stats.rxerrc += rd32(E1000_RXERRC);
3232         adapter->stats.tncrs += rd32(E1000_TNCRS);
3233         adapter->stats.tsctc += rd32(E1000_TSCTC);
3234         adapter->stats.tsctfc += rd32(E1000_TSCTFC);
3235
3236         adapter->stats.iac += rd32(E1000_IAC);
3237         adapter->stats.icrxoc += rd32(E1000_ICRXOC);
3238         adapter->stats.icrxptc += rd32(E1000_ICRXPTC);
3239         adapter->stats.icrxatc += rd32(E1000_ICRXATC);
3240         adapter->stats.ictxptc += rd32(E1000_ICTXPTC);
3241         adapter->stats.ictxatc += rd32(E1000_ICTXATC);
3242         adapter->stats.ictxqec += rd32(E1000_ICTXQEC);
3243         adapter->stats.ictxqmtc += rd32(E1000_ICTXQMTC);
3244         adapter->stats.icrxdmtc += rd32(E1000_ICRXDMTC);
3245
3246         /* Fill out the OS statistics structure */
3247         adapter->net_stats.multicast = adapter->stats.mprc;
3248         adapter->net_stats.collisions = adapter->stats.colc;
3249
3250         /* Rx Errors */
3251
3252         /* RLEC on some newer hardware can be incorrect so build
3253         * our own version based on RUC and ROC */
3254         adapter->net_stats.rx_errors = adapter->stats.rxerrc +
3255                 adapter->stats.crcerrs + adapter->stats.algnerrc +
3256                 adapter->stats.ruc + adapter->stats.roc +
3257                 adapter->stats.cexterr;
3258         adapter->net_stats.rx_length_errors = adapter->stats.ruc +
3259                                               adapter->stats.roc;
3260         adapter->net_stats.rx_crc_errors = adapter->stats.crcerrs;
3261         adapter->net_stats.rx_frame_errors = adapter->stats.algnerrc;
3262         adapter->net_stats.rx_missed_errors = adapter->stats.mpc;
3263
3264         /* Tx Errors */
3265         adapter->net_stats.tx_errors = adapter->stats.ecol +
3266                                        adapter->stats.latecol;
3267         adapter->net_stats.tx_aborted_errors = adapter->stats.ecol;
3268         adapter->net_stats.tx_window_errors = adapter->stats.latecol;
3269         adapter->net_stats.tx_carrier_errors = adapter->stats.tncrs;
3270
3271         /* Tx Dropped needs to be maintained elsewhere */
3272
3273         /* Phy Stats */
3274         if (hw->phy.media_type == e1000_media_type_copper) {
3275                 if ((adapter->link_speed == SPEED_1000) &&
3276                    (!igb_read_phy_reg(hw, PHY_1000T_STATUS,
3277                                               &phy_tmp))) {
3278                         phy_tmp &= PHY_IDLE_ERROR_COUNT_MASK;
3279                         adapter->phy_stats.idle_errors += phy_tmp;
3280                 }
3281         }
3282
3283         /* Management Stats */
3284         adapter->stats.mgptc += rd32(E1000_MGTPTC);
3285         adapter->stats.mgprc += rd32(E1000_MGTPRC);
3286         adapter->stats.mgpdc += rd32(E1000_MGTPDC);
3287 }
3288
3289
3290 static irqreturn_t igb_msix_other(int irq, void *data)
3291 {
3292         struct net_device *netdev = data;
3293         struct igb_adapter *adapter = netdev_priv(netdev);
3294         struct e1000_hw *hw = &adapter->hw;
3295         u32 icr = rd32(E1000_ICR);
3296
3297         /* reading ICR causes bit 31 of EICR to be cleared */
3298         if (!(icr & E1000_ICR_LSC))
3299                 goto no_link_interrupt;
3300         hw->mac.get_link_status = 1;
3301         /* guard against interrupt when we're going down */
3302         if (!test_bit(__IGB_DOWN, &adapter->state))
3303                 mod_timer(&adapter->watchdog_timer, jiffies + 1);
3304         
3305 no_link_interrupt:
3306         wr32(E1000_IMS, E1000_IMS_LSC);
3307         wr32(E1000_EIMS, adapter->eims_other);
3308
3309         return IRQ_HANDLED;
3310 }
3311
3312 static irqreturn_t igb_msix_tx(int irq, void *data)
3313 {
3314         struct igb_ring *tx_ring = data;
3315         struct igb_adapter *adapter = tx_ring->adapter;
3316         struct e1000_hw *hw = &adapter->hw;
3317
3318 #ifdef CONFIG_IGB_DCA
3319         if (adapter->flags & IGB_FLAG_DCA_ENABLED)
3320                 igb_update_tx_dca(tx_ring);
3321 #endif
3322         tx_ring->total_bytes = 0;
3323         tx_ring->total_packets = 0;
3324
3325         /* auto mask will automatically reenable the interrupt when we write
3326          * EICS */
3327         if (!igb_clean_tx_irq(tx_ring))
3328                 /* Ring was not completely cleaned, so fire another interrupt */
3329                 wr32(E1000_EICS, tx_ring->eims_value);
3330         else
3331                 wr32(E1000_EIMS, tx_ring->eims_value);
3332
3333         return IRQ_HANDLED;
3334 }
3335
3336 static void igb_write_itr(struct igb_ring *ring)
3337 {
3338         struct e1000_hw *hw = &ring->adapter->hw;
3339         if ((ring->adapter->itr_setting & 3) && ring->set_itr) {
3340                 switch (hw->mac.type) {
3341                 case e1000_82576:
3342                         wr32(ring->itr_register,
3343                              ring->itr_val |
3344                              0x80000000);
3345                         break;
3346                 default:
3347                         wr32(ring->itr_register,
3348                              ring->itr_val |
3349                              (ring->itr_val << 16));
3350                         break;
3351                 }
3352                 ring->set_itr = 0;
3353         }
3354 }
3355
3356 static irqreturn_t igb_msix_rx(int irq, void *data)
3357 {
3358         struct igb_ring *rx_ring = data;
3359
3360         /* Write the ITR value calculated at the end of the
3361          * previous interrupt.
3362          */
3363
3364         igb_write_itr(rx_ring);
3365
3366         if (netif_rx_schedule_prep(&rx_ring->napi))
3367                 __netif_rx_schedule(&rx_ring->napi);
3368
3369 #ifdef CONFIG_IGB_DCA
3370         if (rx_ring->adapter->flags & IGB_FLAG_DCA_ENABLED)
3371                 igb_update_rx_dca(rx_ring);
3372 #endif
3373                 return IRQ_HANDLED;
3374 }
3375
3376 #ifdef CONFIG_IGB_DCA
3377 static void igb_update_rx_dca(struct igb_ring *rx_ring)
3378 {
3379         u32 dca_rxctrl;
3380         struct igb_adapter *adapter = rx_ring->adapter;
3381         struct e1000_hw *hw = &adapter->hw;
3382         int cpu = get_cpu();
3383         int q = rx_ring->reg_idx;
3384
3385         if (rx_ring->cpu != cpu) {
3386                 dca_rxctrl = rd32(E1000_DCA_RXCTRL(q));
3387                 if (hw->mac.type == e1000_82576) {
3388                         dca_rxctrl &= ~E1000_DCA_RXCTRL_CPUID_MASK_82576;
3389                         dca_rxctrl |= dca_get_tag(cpu) <<
3390                                       E1000_DCA_RXCTRL_CPUID_SHIFT;
3391                 } else {
3392                         dca_rxctrl &= ~E1000_DCA_RXCTRL_CPUID_MASK;
3393                         dca_rxctrl |= dca_get_tag(cpu);
3394                 }
3395                 dca_rxctrl |= E1000_DCA_RXCTRL_DESC_DCA_EN;
3396                 dca_rxctrl |= E1000_DCA_RXCTRL_HEAD_DCA_EN;
3397                 dca_rxctrl |= E1000_DCA_RXCTRL_DATA_DCA_EN;
3398                 wr32(E1000_DCA_RXCTRL(q), dca_rxctrl);
3399                 rx_ring->cpu = cpu;
3400         }
3401         put_cpu();
3402 }
3403
3404 static void igb_update_tx_dca(struct igb_ring *tx_ring)
3405 {
3406         u32 dca_txctrl;
3407         struct igb_adapter *adapter = tx_ring->adapter;
3408         struct e1000_hw *hw = &adapter->hw;
3409         int cpu = get_cpu();
3410         int q = tx_ring->reg_idx;
3411
3412         if (tx_ring->cpu != cpu) {
3413                 dca_txctrl = rd32(E1000_DCA_TXCTRL(q));
3414                 if (hw->mac.type == e1000_82576) {
3415                         dca_txctrl &= ~E1000_DCA_TXCTRL_CPUID_MASK_82576;
3416                         dca_txctrl |= dca_get_tag(cpu) <<
3417                                       E1000_DCA_TXCTRL_CPUID_SHIFT;
3418                 } else {
3419                         dca_txctrl &= ~E1000_DCA_TXCTRL_CPUID_MASK;
3420                         dca_txctrl |= dca_get_tag(cpu);
3421                 }
3422                 dca_txctrl |= E1000_DCA_TXCTRL_DESC_DCA_EN;
3423                 wr32(E1000_DCA_TXCTRL(q), dca_txctrl);
3424                 tx_ring->cpu = cpu;
3425         }
3426         put_cpu();
3427 }
3428
3429 static void igb_setup_dca(struct igb_adapter *adapter)
3430 {
3431         int i;
3432
3433         if (!(adapter->flags & IGB_FLAG_DCA_ENABLED))
3434                 return;
3435
3436         for (i = 0; i < adapter->num_tx_queues; i++) {
3437                 adapter->tx_ring[i].cpu = -1;
3438                 igb_update_tx_dca(&adapter->tx_ring[i]);
3439         }
3440         for (i = 0; i < adapter->num_rx_queues; i++) {
3441                 adapter->rx_ring[i].cpu = -1;
3442                 igb_update_rx_dca(&adapter->rx_ring[i]);
3443         }
3444 }
3445
3446 static int __igb_notify_dca(struct device *dev, void *data)
3447 {
3448         struct net_device *netdev = dev_get_drvdata(dev);
3449         struct igb_adapter *adapter = netdev_priv(netdev);
3450         struct e1000_hw *hw = &adapter->hw;
3451         unsigned long event = *(unsigned long *)data;
3452
3453         switch (event) {
3454         case DCA_PROVIDER_ADD:
3455                 /* if already enabled, don't do it again */
3456                 if (adapter->flags & IGB_FLAG_DCA_ENABLED)
3457                         break;
3458                 /* Always use CB2 mode, difference is masked
3459                  * in the CB driver. */
3460                 wr32(E1000_DCA_CTRL, 2);
3461                 if (dca_add_requester(dev) == 0) {
3462                         adapter->flags |= IGB_FLAG_DCA_ENABLED;
3463                         dev_info(&adapter->pdev->dev, "DCA enabled\n");
3464                         igb_setup_dca(adapter);
3465                         break;
3466                 }
3467                 /* Fall Through since DCA is disabled. */
3468         case DCA_PROVIDER_REMOVE:
3469                 if (adapter->flags & IGB_FLAG_DCA_ENABLED) {
3470                         /* without this a class_device is left
3471                          * hanging around in the sysfs model */
3472                         dca_remove_requester(dev);
3473                         dev_info(&adapter->pdev->dev, "DCA disabled\n");
3474                         adapter->flags &= ~IGB_FLAG_DCA_ENABLED;
3475                         wr32(E1000_DCA_CTRL, 1);
3476                 }
3477                 break;
3478         }
3479
3480         return 0;
3481 }
3482
3483 static int igb_notify_dca(struct notifier_block *nb, unsigned long event,
3484                           void *p)
3485 {
3486         int ret_val;
3487
3488         ret_val = driver_for_each_device(&igb_driver.driver, NULL, &event,
3489                                          __igb_notify_dca);
3490
3491         return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
3492 }
3493 #endif /* CONFIG_IGB_DCA */
3494
3495 /**
3496  * igb_intr_msi - Interrupt Handler
3497  * @irq: interrupt number
3498  * @data: pointer to a network interface device structure
3499  **/
3500 static irqreturn_t igb_intr_msi(int irq, void *data)
3501 {
3502         struct net_device *netdev = data;
3503         struct igb_adapter *adapter = netdev_priv(netdev);
3504         struct e1000_hw *hw = &adapter->hw;
3505         /* read ICR disables interrupts using IAM */
3506         u32 icr = rd32(E1000_ICR);
3507
3508         igb_write_itr(adapter->rx_ring);
3509
3510         if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
3511                 hw->mac.get_link_status = 1;
3512                 if (!test_bit(__IGB_DOWN, &adapter->state))
3513                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
3514         }
3515
3516         netif_rx_schedule(&adapter->rx_ring[0].napi);
3517
3518         return IRQ_HANDLED;
3519 }
3520
3521 /**
3522  * igb_intr - Interrupt Handler
3523  * @irq: interrupt number
3524  * @data: pointer to a network interface device structure
3525  **/
3526 static irqreturn_t igb_intr(int irq, void *data)
3527 {
3528         struct net_device *netdev = data;
3529         struct igb_adapter *adapter = netdev_priv(netdev);
3530         struct e1000_hw *hw = &adapter->hw;
3531         /* Interrupt Auto-Mask...upon reading ICR, interrupts are masked.  No
3532          * need for the IMC write */
3533         u32 icr = rd32(E1000_ICR);
3534         u32 eicr = 0;
3535         if (!icr)
3536                 return IRQ_NONE;  /* Not our interrupt */
3537
3538         igb_write_itr(adapter->rx_ring);
3539
3540         /* IMS will not auto-mask if INT_ASSERTED is not set, and if it is
3541          * not set, then the adapter didn't send an interrupt */
3542         if (!(icr & E1000_ICR_INT_ASSERTED))
3543                 return IRQ_NONE;
3544
3545         eicr = rd32(E1000_EICR);
3546
3547         if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
3548                 hw->mac.get_link_status = 1;
3549                 /* guard against interrupt when we're going down */
3550                 if (!test_bit(__IGB_DOWN, &adapter->state))
3551                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
3552         }
3553
3554         netif_rx_schedule(&adapter->rx_ring[0].napi);
3555
3556         return IRQ_HANDLED;
3557 }
3558
3559 /**
3560  * igb_poll - NAPI Rx polling callback
3561  * @napi: napi polling structure
3562  * @budget: count of how many packets we should handle
3563  **/
3564 static int igb_poll(struct napi_struct *napi, int budget)
3565 {
3566         struct igb_ring *rx_ring = container_of(napi, struct igb_ring, napi);
3567         struct igb_adapter *adapter = rx_ring->adapter;
3568         struct net_device *netdev = adapter->netdev;
3569         int tx_clean_complete, work_done = 0;
3570
3571         /* this poll routine only supports one tx and one rx queue */
3572 #ifdef CONFIG_IGB_DCA
3573         if (adapter->flags & IGB_FLAG_DCA_ENABLED)
3574                 igb_update_tx_dca(&adapter->tx_ring[0]);
3575 #endif
3576         tx_clean_complete = igb_clean_tx_irq(&adapter->tx_ring[0]);
3577
3578 #ifdef CONFIG_IGB_DCA
3579         if (adapter->flags & IGB_FLAG_DCA_ENABLED)
3580                 igb_update_rx_dca(&adapter->rx_ring[0]);
3581 #endif
3582         igb_clean_rx_irq_adv(&adapter->rx_ring[0], &work_done, budget);
3583
3584         /* If no Tx and not enough Rx work done, exit the polling mode */
3585         if ((tx_clean_complete && (work_done < budget)) ||
3586             !netif_running(netdev)) {
3587                 if (adapter->itr_setting & 3)
3588                         igb_set_itr(adapter);
3589                 netif_rx_complete(napi);
3590                 if (!test_bit(__IGB_DOWN, &adapter->state))
3591                         igb_irq_enable(adapter);
3592                 return 0;
3593         }
3594
3595         return 1;
3596 }
3597
3598 static int igb_clean_rx_ring_msix(struct napi_struct *napi, int budget)
3599 {
3600         struct igb_ring *rx_ring = container_of(napi, struct igb_ring, napi);
3601         struct igb_adapter *adapter = rx_ring->adapter;
3602         struct e1000_hw *hw = &adapter->hw;
3603         struct net_device *netdev = adapter->netdev;
3604         int work_done = 0;
3605
3606 #ifdef CONFIG_IGB_DCA
3607         if (adapter->flags & IGB_FLAG_DCA_ENABLED)
3608                 igb_update_rx_dca(rx_ring);
3609 #endif
3610         igb_clean_rx_irq_adv(rx_ring, &work_done, budget);
3611
3612
3613         /* If not enough Rx work done, exit the polling mode */
3614         if ((work_done == 0) || !netif_running(netdev)) {
3615                 netif_rx_complete(napi);
3616
3617                 if (adapter->itr_setting & 3) {
3618                         if (adapter->num_rx_queues == 1)
3619                                 igb_set_itr(adapter);
3620                         else
3621                                 igb_update_ring_itr(rx_ring);
3622                 }
3623
3624                 if (!test_bit(__IGB_DOWN, &adapter->state))
3625                         wr32(E1000_EIMS, rx_ring->eims_value);
3626
3627                 return 0;
3628         }
3629
3630         return 1;
3631 }
3632
3633 /**
3634  * igb_clean_tx_irq - Reclaim resources after transmit completes
3635  * @adapter: board private structure
3636  * returns true if ring is completely cleaned
3637  **/
3638 static bool igb_clean_tx_irq(struct igb_ring *tx_ring)
3639 {
3640         struct igb_adapter *adapter = tx_ring->adapter;
3641         struct net_device *netdev = adapter->netdev;
3642         struct e1000_hw *hw = &adapter->hw;
3643         struct igb_buffer *buffer_info;
3644         struct sk_buff *skb;
3645         union e1000_adv_tx_desc *tx_desc, *eop_desc;
3646         unsigned int total_bytes = 0, total_packets = 0;
3647         unsigned int i, eop, count = 0;
3648         bool cleaned = false;
3649
3650         i = tx_ring->next_to_clean;
3651         eop = tx_ring->buffer_info[i].next_to_watch;
3652         eop_desc = E1000_TX_DESC_ADV(*tx_ring, eop);
3653
3654         while ((eop_desc->wb.status & cpu_to_le32(E1000_TXD_STAT_DD)) &&
3655                (count < tx_ring->count)) {
3656                 for (cleaned = false; !cleaned; count++) {
3657                         tx_desc = E1000_TX_DESC_ADV(*tx_ring, i);
3658                         buffer_info = &tx_ring->buffer_info[i];
3659                         cleaned = (i == eop);
3660                         skb = buffer_info->skb;
3661
3662                         if (skb) {
3663                                 unsigned int segs, bytecount;
3664                                 /* gso_segs is currently only valid for tcp */
3665                                 segs = skb_shinfo(skb)->gso_segs ?: 1;
3666                                 /* multiply data chunks by size of headers */
3667                                 bytecount = ((segs - 1) * skb_headlen(skb)) +
3668                                             skb->len;
3669                                 total_packets += segs;
3670                                 total_bytes += bytecount;
3671                         }
3672
3673                         igb_unmap_and_free_tx_resource(adapter, buffer_info);
3674                         tx_desc->wb.status = 0;
3675
3676                         i++;
3677                         if (i == tx_ring->count)
3678                                 i = 0;
3679                 }
3680
3681                 eop = tx_ring->buffer_info[i].next_to_watch;
3682                 eop_desc = E1000_TX_DESC_ADV(*tx_ring, eop);
3683         }
3684
3685         tx_ring->next_to_clean = i;
3686
3687         if (unlikely(count &&
3688                      netif_carrier_ok(netdev) &&
3689                      IGB_DESC_UNUSED(tx_ring) >= IGB_TX_QUEUE_WAKE)) {
3690                 /* Make sure that anybody stopping the queue after this
3691                  * sees the new next_to_clean.
3692                  */
3693                 smp_mb();
3694                 if (__netif_subqueue_stopped(netdev, tx_ring->queue_index) &&
3695                     !(test_bit(__IGB_DOWN, &adapter->state))) {
3696                         netif_wake_subqueue(netdev, tx_ring->queue_index);
3697                         ++adapter->restart_queue;
3698                 }
3699         }
3700
3701         if (tx_ring->detect_tx_hung) {
3702                 /* Detect a transmit hang in hardware, this serializes the
3703                  * check with the clearing of time_stamp and movement of i */
3704                 tx_ring->detect_tx_hung = false;
3705                 if (tx_ring->buffer_info[i].time_stamp &&
3706                     time_after(jiffies, tx_ring->buffer_info[i].time_stamp +
3707                                (adapter->tx_timeout_factor * HZ))
3708                     && !(rd32(E1000_STATUS) &
3709                          E1000_STATUS_TXOFF)) {
3710
3711                         /* detected Tx unit hang */
3712                         dev_err(&adapter->pdev->dev,
3713                                 "Detected Tx Unit Hang\n"
3714                                 "  Tx Queue             <%d>\n"
3715                                 "  TDH                  <%x>\n"
3716                                 "  TDT                  <%x>\n"
3717                                 "  next_to_use          <%x>\n"
3718                                 "  next_to_clean        <%x>\n"
3719                                 "buffer_info[next_to_clean]\n"
3720                                 "  time_stamp           <%lx>\n"
3721                                 "  next_to_watch        <%x>\n"
3722                                 "  jiffies              <%lx>\n"
3723                                 "  desc.status          <%x>\n",
3724                                 tx_ring->queue_index,
3725                                 readl(adapter->hw.hw_addr + tx_ring->head),
3726                                 readl(adapter->hw.hw_addr + tx_ring->tail),
3727                                 tx_ring->next_to_use,
3728                                 tx_ring->next_to_clean,
3729                                 tx_ring->buffer_info[i].time_stamp,
3730                                 eop,
3731                                 jiffies,
3732                                 eop_desc->wb.status);
3733                         netif_stop_subqueue(netdev, tx_ring->queue_index);
3734                 }
3735         }
3736         tx_ring->total_bytes += total_bytes;
3737         tx_ring->total_packets += total_packets;
3738         tx_ring->tx_stats.bytes += total_bytes;
3739         tx_ring->tx_stats.packets += total_packets;
3740         adapter->net_stats.tx_bytes += total_bytes;
3741         adapter->net_stats.tx_packets += total_packets;
3742         return (count < tx_ring->count);
3743 }
3744
3745 #ifdef CONFIG_IGB_LRO
3746  /**
3747  * igb_get_skb_hdr - helper function for LRO header processing
3748  * @skb: pointer to sk_buff to be added to LRO packet
3749  * @iphdr: pointer to ip header structure
3750  * @tcph: pointer to tcp header structure
3751  * @hdr_flags: pointer to header flags
3752  * @priv: pointer to the receive descriptor for the current sk_buff
3753  **/
3754 static int igb_get_skb_hdr(struct sk_buff *skb, void **iphdr, void **tcph,
3755                            u64 *hdr_flags, void *priv)
3756 {
3757         union e1000_adv_rx_desc *rx_desc = priv;
3758         u16 pkt_type = rx_desc->wb.lower.lo_dword.pkt_info &
3759                        (E1000_RXDADV_PKTTYPE_IPV4 | E1000_RXDADV_PKTTYPE_TCP);
3760
3761         /* Verify that this is a valid IPv4 TCP packet */
3762         if (pkt_type != (E1000_RXDADV_PKTTYPE_IPV4 |
3763                           E1000_RXDADV_PKTTYPE_TCP))
3764                 return -1;
3765
3766         /* Set network headers */
3767         skb_reset_network_header(skb);
3768         skb_set_transport_header(skb, ip_hdrlen(skb));
3769         *iphdr = ip_hdr(skb);
3770         *tcph = tcp_hdr(skb);
3771         *hdr_flags = LRO_IPV4 | LRO_TCP;
3772
3773         return 0;
3774
3775 }
3776 #endif /* CONFIG_IGB_LRO */
3777
3778 /**
3779  * igb_receive_skb - helper function to handle rx indications
3780  * @ring: pointer to receive ring receving this packet 
3781  * @status: descriptor status field as written by hardware
3782  * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
3783  * @skb: pointer to sk_buff to be indicated to stack
3784  **/
3785 static void igb_receive_skb(struct igb_ring *ring, u8 status,
3786                             union e1000_adv_rx_desc * rx_desc,
3787                             struct sk_buff *skb)
3788 {
3789         struct igb_adapter * adapter = ring->adapter;
3790         bool vlan_extracted = (adapter->vlgrp && (status & E1000_RXD_STAT_VP));
3791
3792 #ifdef CONFIG_IGB_LRO
3793         if (adapter->netdev->features & NETIF_F_LRO &&
3794             skb->ip_summed == CHECKSUM_UNNECESSARY) {
3795                 if (vlan_extracted)
3796                         lro_vlan_hwaccel_receive_skb(&ring->lro_mgr, skb,
3797                                            adapter->vlgrp,
3798                                            le16_to_cpu(rx_desc->wb.upper.vlan),
3799                                            rx_desc);
3800                 else
3801                         lro_receive_skb(&ring->lro_mgr,skb, rx_desc);
3802                 ring->lro_used = 1;
3803         } else {
3804 #endif
3805                 if (vlan_extracted)
3806                         vlan_hwaccel_receive_skb(skb, adapter->vlgrp,
3807                                           le16_to_cpu(rx_desc->wb.upper.vlan));
3808                 else
3809
3810                         netif_receive_skb(skb);
3811 #ifdef CONFIG_IGB_LRO
3812         }
3813 #endif
3814 }
3815
3816
3817 static inline void igb_rx_checksum_adv(struct igb_adapter *adapter,
3818                                        u32 status_err, struct sk_buff *skb)
3819 {
3820         skb->ip_summed = CHECKSUM_NONE;
3821
3822         /* Ignore Checksum bit is set or checksum is disabled through ethtool */
3823         if ((status_err & E1000_RXD_STAT_IXSM) || !adapter->rx_csum)
3824                 return;
3825         /* TCP/UDP checksum error bit is set */
3826         if (status_err &
3827             (E1000_RXDEXT_STATERR_TCPE | E1000_RXDEXT_STATERR_IPE)) {
3828                 /* let the stack verify checksum errors */
3829                 adapter->hw_csum_err++;
3830                 return;
3831         }
3832         /* It must be a TCP or UDP packet with a valid checksum */
3833         if (status_err & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS))
3834                 skb->ip_summed = CHECKSUM_UNNECESSARY;
3835
3836         adapter->hw_csum_good++;
3837 }
3838
3839 static bool igb_clean_rx_irq_adv(struct igb_ring *rx_ring,
3840                                  int *work_done, int budget)
3841 {
3842         struct igb_adapter *adapter = rx_ring->adapter;
3843         struct net_device *netdev = adapter->netdev;
3844         struct pci_dev *pdev = adapter->pdev;
3845         union e1000_adv_rx_desc *rx_desc , *next_rxd;
3846         struct igb_buffer *buffer_info , *next_buffer;
3847         struct sk_buff *skb;
3848         unsigned int i;
3849         u32 length, hlen, staterr;
3850         bool cleaned = false;
3851         int cleaned_count = 0;
3852         unsigned int total_bytes = 0, total_packets = 0;
3853
3854         i = rx_ring->next_to_clean;
3855         rx_desc = E1000_RX_DESC_ADV(*rx_ring, i);
3856         staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
3857
3858         while (staterr & E1000_RXD_STAT_DD) {
3859                 if (*work_done >= budget)
3860                         break;
3861                 (*work_done)++;
3862                 buffer_info = &rx_ring->buffer_info[i];
3863
3864                 /* HW will not DMA in data larger than the given buffer, even
3865                  * if it parses the (NFS, of course) header to be larger.  In
3866                  * that case, it fills the header buffer and spills the rest
3867                  * into the page.
3868                  */
3869                 hlen = (le16_to_cpu(rx_desc->wb.lower.lo_dword.hdr_info) &
3870                   E1000_RXDADV_HDRBUFLEN_MASK) >> E1000_RXDADV_HDRBUFLEN_SHIFT;
3871                 if (hlen > adapter->rx_ps_hdr_size)
3872                         hlen = adapter->rx_ps_hdr_size;
3873
3874                 length = le16_to_cpu(rx_desc->wb.upper.length);
3875                 cleaned = true;
3876                 cleaned_count++;
3877
3878                 skb = buffer_info->skb;
3879                 prefetch(skb->data - NET_IP_ALIGN);
3880                 buffer_info->skb = NULL;
3881                 if (!adapter->rx_ps_hdr_size) {
3882                         pci_unmap_single(pdev, buffer_info->dma,
3883                                          adapter->rx_buffer_len +
3884                                            NET_IP_ALIGN,
3885                                          PCI_DMA_FROMDEVICE);
3886                         skb_put(skb, length);
3887                         goto send_up;
3888                 }
3889
3890                 if (!skb_shinfo(skb)->nr_frags) {
3891                         pci_unmap_single(pdev, buffer_info->dma,
3892                                          adapter->rx_ps_hdr_size +
3893                                            NET_IP_ALIGN,
3894                                          PCI_DMA_FROMDEVICE);
3895                         skb_put(skb, hlen);
3896                 }
3897
3898                 if (length) {
3899                         pci_unmap_page(pdev, buffer_info->page_dma,
3900                                        PAGE_SIZE / 2, PCI_DMA_FROMDEVICE);
3901                         buffer_info->page_dma = 0;
3902
3903                         skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags++,
3904                                                 buffer_info->page,
3905                                                 buffer_info->page_offset,
3906                                                 length);
3907
3908                         if ((adapter->rx_buffer_len > (PAGE_SIZE / 2)) ||
3909                             (page_count(buffer_info->page) != 1))
3910                                 buffer_info->page = NULL;
3911                         else
3912                                 get_page(buffer_info->page);
3913
3914                         skb->len += length;
3915                         skb->data_len += length;
3916
3917                         skb->truesize += length;
3918                 }
3919 send_up:
3920                 i++;
3921                 if (i == rx_ring->count)
3922                         i = 0;
3923                 next_rxd = E1000_RX_DESC_ADV(*rx_ring, i);
3924                 prefetch(next_rxd);
3925                 next_buffer = &rx_ring->buffer_info[i];
3926
3927                 if (!(staterr & E1000_RXD_STAT_EOP)) {
3928                         buffer_info->skb = next_buffer->skb;
3929                         buffer_info->dma = next_buffer->dma;
3930                         next_buffer->skb = skb;
3931                         next_buffer->dma = 0;
3932                         goto next_desc;
3933                 }
3934
3935                 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
3936                         dev_kfree_skb_irq(skb);
3937                         goto next_desc;
3938                 }
3939
3940                 total_bytes += skb->len;
3941                 total_packets++;
3942
3943                 igb_rx_checksum_adv(adapter, staterr, skb);
3944
3945                 skb->protocol = eth_type_trans(skb, netdev);
3946
3947                 igb_receive_skb(rx_ring, staterr, rx_desc, skb);
3948
3949 next_desc:
3950                 rx_desc->wb.upper.status_error = 0;
3951
3952                 /* return some buffers to hardware, one at a time is too slow */
3953                 if (cleaned_count >= IGB_RX_BUFFER_WRITE) {
3954                         igb_alloc_rx_buffers_adv(rx_ring, cleaned_count);
3955                         cleaned_count = 0;
3956                 }
3957
3958                 /* use prefetched values */
3959                 rx_desc = next_rxd;
3960                 buffer_info = next_buffer;
3961
3962                 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
3963         }
3964
3965         rx_ring->next_to_clean = i;
3966         cleaned_count = IGB_DESC_UNUSED(rx_ring);
3967
3968 #ifdef CONFIG_IGB_LRO
3969         if (rx_ring->lro_used) {
3970                 lro_flush_all(&rx_ring->lro_mgr);
3971                 rx_ring->lro_used = 0;
3972         }
3973 #endif
3974
3975         if (cleaned_count)
3976                 igb_alloc_rx_buffers_adv(rx_ring, cleaned_count);
3977
3978         rx_ring->total_packets += total_packets;
3979         rx_ring->total_bytes += total_bytes;
3980         rx_ring->rx_stats.packets += total_packets;
3981         rx_ring->rx_stats.bytes += total_bytes;
3982         adapter->net_stats.rx_bytes += total_bytes;
3983         adapter->net_stats.rx_packets += total_packets;
3984         return cleaned;
3985 }
3986
3987
3988 /**
3989  * igb_alloc_rx_buffers_adv - Replace used receive buffers; packet split
3990  * @adapter: address of board private structure
3991  **/
3992 static void igb_alloc_rx_buffers_adv(struct igb_ring *rx_ring,
3993                                      int cleaned_count)
3994 {
3995         struct igb_adapter *adapter = rx_ring->adapter;
3996         struct net_device *netdev = adapter->netdev;
3997         struct pci_dev *pdev = adapter->pdev;
3998         union e1000_adv_rx_desc *rx_desc;
3999         struct igb_buffer *buffer_info;
4000         struct sk_buff *skb;
4001         unsigned int i;
4002
4003         i = rx_ring->next_to_use;
4004         buffer_info = &rx_ring->buffer_info[i];
4005
4006         while (cleaned_count--) {
4007                 rx_desc = E1000_RX_DESC_ADV(*rx_ring, i);
4008
4009                 if (adapter->rx_ps_hdr_size && !buffer_info->page_dma) {
4010                         if (!buffer_info->page) {
4011                                 buffer_info->page = alloc_page(GFP_ATOMIC);
4012                                 if (!buffer_info->page) {
4013                                         adapter->alloc_rx_buff_failed++;
4014                                         goto no_buffers;
4015                                 }
4016                                 buffer_info->page_offset = 0;
4017                         } else {
4018                                 buffer_info->page_offset ^= PAGE_SIZE / 2;
4019                         }
4020                         buffer_info->page_dma =
4021                                 pci_map_page(pdev,
4022                                              buffer_info->page,
4023                                              buffer_info->page_offset,
4024                                              PAGE_SIZE / 2,
4025                                              PCI_DMA_FROMDEVICE);
4026                 }
4027
4028                 if (!buffer_info->skb) {
4029                         int bufsz;
4030
4031                         if (adapter->rx_ps_hdr_size)
4032                                 bufsz = adapter->rx_ps_hdr_size;
4033                         else
4034                                 bufsz = adapter->rx_buffer_len;
4035                         bufsz += NET_IP_ALIGN;
4036                         skb = netdev_alloc_skb(netdev, bufsz);
4037
4038                         if (!skb) {
4039                                 adapter->alloc_rx_buff_failed++;
4040                                 goto no_buffers;
4041                         }
4042
4043                         /* Make buffer alignment 2 beyond a 16 byte boundary
4044                          * this will result in a 16 byte aligned IP header after
4045                          * the 14 byte MAC header is removed
4046                          */
4047                         skb_reserve(skb, NET_IP_ALIGN);
4048
4049                         buffer_info->skb = skb;
4050                         buffer_info->dma = pci_map_single(pdev, skb->data,
4051                                                           bufsz,
4052                                                           PCI_DMA_FROMDEVICE);
4053
4054                 }
4055                 /* Refresh the desc even if buffer_addrs didn't change because
4056                  * each write-back erases this info. */
4057                 if (adapter->rx_ps_hdr_size) {
4058                         rx_desc->read.pkt_addr =
4059                              cpu_to_le64(buffer_info->page_dma);
4060                         rx_desc->read.hdr_addr = cpu_to_le64(buffer_info->dma);
4061                 } else {
4062                         rx_desc->read.pkt_addr =
4063                              cpu_to_le64(buffer_info->dma);
4064                         rx_desc->read.hdr_addr = 0;
4065                 }
4066
4067                 i++;
4068                 if (i == rx_ring->count)
4069                         i = 0;
4070                 buffer_info = &rx_ring->buffer_info[i];
4071         }
4072
4073 no_buffers:
4074         if (rx_ring->next_to_use != i) {
4075                 rx_ring->next_to_use = i;
4076                 if (i == 0)
4077                         i = (rx_ring->count - 1);
4078                 else
4079                         i--;
4080
4081                 /* Force memory writes to complete before letting h/w
4082                  * know there are new descriptors to fetch.  (Only
4083                  * applicable for weak-ordered memory model archs,
4084                  * such as IA-64). */
4085                 wmb();
4086                 writel(i, adapter->hw.hw_addr + rx_ring->tail);
4087         }
4088 }
4089
4090 /**
4091  * igb_mii_ioctl -
4092  * @netdev:
4093  * @ifreq:
4094  * @cmd:
4095  **/
4096 static int igb_mii_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
4097 {
4098         struct igb_adapter *adapter = netdev_priv(netdev);
4099         struct mii_ioctl_data *data = if_mii(ifr);
4100
4101         if (adapter->hw.phy.media_type != e1000_media_type_copper)
4102                 return -EOPNOTSUPP;
4103
4104         switch (cmd) {
4105         case SIOCGMIIPHY:
4106                 data->phy_id = adapter->hw.phy.addr;
4107                 break;
4108         case SIOCGMIIREG:
4109                 if (!capable(CAP_NET_ADMIN))
4110                         return -EPERM;
4111                 if (igb_read_phy_reg(&adapter->hw, data->reg_num & 0x1F,
4112                                      &data->val_out))
4113                         return -EIO;
4114                 break;
4115         case SIOCSMIIREG:
4116         default:
4117                 return -EOPNOTSUPP;
4118         }
4119         return 0;
4120 }
4121
4122 /**
4123  * igb_ioctl -
4124  * @netdev:
4125  * @ifreq:
4126  * @cmd:
4127  **/
4128 static int igb_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
4129 {
4130         switch (cmd) {
4131         case SIOCGMIIPHY:
4132         case SIOCGMIIREG:
4133         case SIOCSMIIREG:
4134                 return igb_mii_ioctl(netdev, ifr, cmd);
4135         default:
4136                 return -EOPNOTSUPP;
4137         }
4138 }
4139
4140 static void igb_vlan_rx_register(struct net_device *netdev,
4141                                  struct vlan_group *grp)
4142 {
4143         struct igb_adapter *adapter = netdev_priv(netdev);
4144         struct e1000_hw *hw = &adapter->hw;
4145         u32 ctrl, rctl;
4146
4147         igb_irq_disable(adapter);
4148         adapter->vlgrp = grp;
4149
4150         if (grp) {
4151                 /* enable VLAN tag insert/strip */
4152                 ctrl = rd32(E1000_CTRL);
4153                 ctrl |= E1000_CTRL_VME;
4154                 wr32(E1000_CTRL, ctrl);
4155
4156                 /* enable VLAN receive filtering */
4157                 rctl = rd32(E1000_RCTL);
4158                 rctl &= ~E1000_RCTL_CFIEN;
4159                 wr32(E1000_RCTL, rctl);
4160                 igb_update_mng_vlan(adapter);
4161                 wr32(E1000_RLPML,
4162                                 adapter->max_frame_size + VLAN_TAG_SIZE);
4163         } else {
4164                 /* disable VLAN tag insert/strip */
4165                 ctrl = rd32(E1000_CTRL);
4166                 ctrl &= ~E1000_CTRL_VME;
4167                 wr32(E1000_CTRL, ctrl);
4168
4169                 if (adapter->mng_vlan_id != (u16)IGB_MNG_VLAN_NONE) {
4170                         igb_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
4171                         adapter->mng_vlan_id = IGB_MNG_VLAN_NONE;
4172                 }
4173                 wr32(E1000_RLPML,
4174                                 adapter->max_frame_size);
4175         }
4176
4177         if (!test_bit(__IGB_DOWN, &adapter->state))
4178                 igb_irq_enable(adapter);
4179 }
4180
4181 static void igb_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
4182 {
4183         struct igb_adapter *adapter = netdev_priv(netdev);
4184         struct e1000_hw *hw = &adapter->hw;
4185         u32 vfta, index;
4186
4187         if ((adapter->hw.mng_cookie.status &
4188              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
4189             (vid == adapter->mng_vlan_id))
4190                 return;
4191         /* add VID to filter table */
4192         index = (vid >> 5) & 0x7F;
4193         vfta = array_rd32(E1000_VFTA, index);
4194         vfta |= (1 << (vid & 0x1F));
4195         igb_write_vfta(&adapter->hw, index, vfta);
4196 }
4197
4198 static void igb_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
4199 {
4200         struct igb_adapter *adapter = netdev_priv(netdev);
4201         struct e1000_hw *hw = &adapter->hw;
4202         u32 vfta, index;
4203
4204         igb_irq_disable(adapter);
4205         vlan_group_set_device(adapter->vlgrp, vid, NULL);
4206
4207         if (!test_bit(__IGB_DOWN, &adapter->state))
4208                 igb_irq_enable(adapter);
4209
4210         if ((adapter->hw.mng_cookie.status &
4211              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
4212             (vid == adapter->mng_vlan_id)) {
4213                 /* release control to f/w */
4214                 igb_release_hw_control(adapter);
4215                 return;
4216         }
4217
4218         /* remove VID from filter table */
4219         index = (vid >> 5) & 0x7F;
4220         vfta = array_rd32(E1000_VFTA, index);
4221         vfta &= ~(1 << (vid & 0x1F));
4222         igb_write_vfta(&adapter->hw, index, vfta);
4223 }
4224
4225 static void igb_restore_vlan(struct igb_adapter *adapter)
4226 {
4227         igb_vlan_rx_register(adapter->netdev, adapter->vlgrp);
4228
4229         if (adapter->vlgrp) {
4230                 u16 vid;
4231                 for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
4232                         if (!vlan_group_get_device(adapter->vlgrp, vid))
4233                                 continue;
4234                         igb_vlan_rx_add_vid(adapter->netdev, vid);
4235                 }
4236         }
4237 }
4238
4239 int igb_set_spd_dplx(struct igb_adapter *adapter, u16 spddplx)
4240 {
4241         struct e1000_mac_info *mac = &adapter->hw.mac;
4242
4243         mac->autoneg = 0;
4244
4245         /* Fiber NICs only allow 1000 gbps Full duplex */
4246         if ((adapter->hw.phy.media_type == e1000_media_type_fiber) &&
4247                 spddplx != (SPEED_1000 + DUPLEX_FULL)) {
4248                 dev_err(&adapter->pdev->dev,
4249                         "Unsupported Speed/Duplex configuration\n");
4250                 return -EINVAL;
4251         }
4252
4253         switch (spddplx) {
4254         case SPEED_10 + DUPLEX_HALF:
4255                 mac->forced_speed_duplex = ADVERTISE_10_HALF;
4256                 break;
4257         case SPEED_10 + DUPLEX_FULL:
4258                 mac->forced_speed_duplex = ADVERTISE_10_FULL;
4259                 break;
4260         case SPEED_100 + DUPLEX_HALF:
4261                 mac->forced_speed_duplex = ADVERTISE_100_HALF;
4262                 break;
4263         case SPEED_100 + DUPLEX_FULL:
4264                 mac->forced_speed_duplex = ADVERTISE_100_FULL;
4265                 break;
4266         case SPEED_1000 + DUPLEX_FULL:
4267                 mac->autoneg = 1;
4268                 adapter->hw.phy.autoneg_advertised = ADVERTISE_1000_FULL;
4269                 break;
4270         case SPEED_1000 + DUPLEX_HALF: /* not supported */
4271         default:
4272                 dev_err(&adapter->pdev->dev,
4273                         "Unsupported Speed/Duplex configuration\n");
4274                 return -EINVAL;
4275         }
4276         return 0;
4277 }
4278
4279
4280 static int igb_suspend(struct pci_dev *pdev, pm_message_t state)
4281 {
4282         struct net_device *netdev = pci_get_drvdata(pdev);
4283         struct igb_adapter *adapter = netdev_priv(netdev);
4284         struct e1000_hw *hw = &adapter->hw;
4285         u32 ctrl, rctl, status;
4286         u32 wufc = adapter->wol;
4287 #ifdef CONFIG_PM
4288         int retval = 0;
4289 #endif
4290
4291         netif_device_detach(netdev);
4292
4293         if (netif_running(netdev))
4294                 igb_close(netdev);
4295
4296         igb_reset_interrupt_capability(adapter);
4297
4298         igb_free_queues(adapter);
4299
4300 #ifdef CONFIG_PM
4301         retval = pci_save_state(pdev);
4302         if (retval)
4303                 return retval;
4304 #endif
4305
4306         status = rd32(E1000_STATUS);
4307         if (status & E1000_STATUS_LU)
4308                 wufc &= ~E1000_WUFC_LNKC;
4309
4310         if (wufc) {
4311                 igb_setup_rctl(adapter);
4312                 igb_set_multi(netdev);
4313
4314                 /* turn on all-multi mode if wake on multicast is enabled */
4315                 if (wufc & E1000_WUFC_MC) {
4316                         rctl = rd32(E1000_RCTL);
4317                         rctl |= E1000_RCTL_MPE;
4318                         wr32(E1000_RCTL, rctl);
4319                 }
4320
4321                 ctrl = rd32(E1000_CTRL);
4322                 /* advertise wake from D3Cold */
4323                 #define E1000_CTRL_ADVD3WUC 0x00100000
4324                 /* phy power management enable */
4325                 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
4326                 ctrl |= E1000_CTRL_ADVD3WUC;
4327                 wr32(E1000_CTRL, ctrl);
4328
4329                 /* Allow time for pending master requests to run */
4330                 igb_disable_pcie_master(&adapter->hw);
4331
4332                 wr32(E1000_WUC, E1000_WUC_PME_EN);
4333                 wr32(E1000_WUFC, wufc);
4334         } else {
4335                 wr32(E1000_WUC, 0);
4336                 wr32(E1000_WUFC, 0);
4337         }
4338
4339         /* make sure adapter isn't asleep if manageability/wol is enabled */
4340         if (wufc || adapter->en_mng_pt) {
4341                 pci_enable_wake(pdev, PCI_D3hot, 1);
4342                 pci_enable_wake(pdev, PCI_D3cold, 1);
4343         } else {
4344                 igb_shutdown_fiber_serdes_link_82575(hw);
4345                 pci_enable_wake(pdev, PCI_D3hot, 0);
4346                 pci_enable_wake(pdev, PCI_D3cold, 0);
4347         }
4348
4349         /* Release control of h/w to f/w.  If f/w is AMT enabled, this
4350          * would have already happened in close and is redundant. */
4351         igb_release_hw_control(adapter);
4352
4353         pci_disable_device(pdev);
4354
4355         pci_set_power_state(pdev, pci_choose_state(pdev, state));
4356
4357         return 0;
4358 }
4359
4360 #ifdef CONFIG_PM
4361 static int igb_resume(struct pci_dev *pdev)
4362 {
4363         struct net_device *netdev = pci_get_drvdata(pdev);
4364         struct igb_adapter *adapter = netdev_priv(netdev);
4365         struct e1000_hw *hw = &adapter->hw;
4366         u32 err;
4367
4368         pci_set_power_state(pdev, PCI_D0);
4369         pci_restore_state(pdev);
4370
4371         if (adapter->need_ioport)
4372                 err = pci_enable_device(pdev);
4373         else
4374                 err = pci_enable_device_mem(pdev);
4375         if (err) {
4376                 dev_err(&pdev->dev,
4377                         "igb: Cannot enable PCI device from suspend\n");
4378                 return err;
4379         }
4380         pci_set_master(pdev);
4381
4382         pci_enable_wake(pdev, PCI_D3hot, 0);
4383         pci_enable_wake(pdev, PCI_D3cold, 0);
4384
4385         igb_set_interrupt_capability(adapter);
4386
4387         if (igb_alloc_queues(adapter)) {
4388                 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
4389                 return -ENOMEM;
4390         }
4391
4392         /* e1000_power_up_phy(adapter); */
4393
4394         igb_reset(adapter);
4395         wr32(E1000_WUS, ~0);
4396
4397         if (netif_running(netdev)) {
4398                 err = igb_open(netdev);
4399                 if (err)
4400                         return err;
4401         }
4402
4403         netif_device_attach(netdev);
4404
4405         /* let the f/w know that the h/w is now under the control of the
4406          * driver. */
4407         igb_get_hw_control(adapter);
4408
4409         return 0;
4410 }
4411 #endif
4412
4413 static void igb_shutdown(struct pci_dev *pdev)
4414 {
4415         igb_suspend(pdev, PMSG_SUSPEND);
4416 }
4417
4418 #ifdef CONFIG_NET_POLL_CONTROLLER
4419 /*
4420  * Polling 'interrupt' - used by things like netconsole to send skbs
4421  * without having to re-enable interrupts. It's not called while
4422  * the interrupt routine is executing.
4423  */
4424 static void igb_netpoll(struct net_device *netdev)
4425 {
4426         struct igb_adapter *adapter = netdev_priv(netdev);
4427         int i;
4428         int work_done = 0;
4429
4430         igb_irq_disable(adapter);
4431         adapter->flags |= IGB_FLAG_IN_NETPOLL;
4432
4433         for (i = 0; i < adapter->num_tx_queues; i++)
4434                 igb_clean_tx_irq(&adapter->tx_ring[i]);
4435
4436         for (i = 0; i < adapter->num_rx_queues; i++)
4437                 igb_clean_rx_irq_adv(&adapter->rx_ring[i],
4438                                      &work_done,
4439                                      adapter->rx_ring[i].napi.weight);
4440
4441         adapter->flags &= ~IGB_FLAG_IN_NETPOLL;
4442         igb_irq_enable(adapter);
4443 }
4444 #endif /* CONFIG_NET_POLL_CONTROLLER */
4445
4446 /**
4447  * igb_io_error_detected - called when PCI error is detected
4448  * @pdev: Pointer to PCI device
4449  * @state: The current pci connection state
4450  *
4451  * This function is called after a PCI bus error affecting
4452  * this device has been detected.
4453  */
4454 static pci_ers_result_t igb_io_error_detected(struct pci_dev *pdev,
4455                                               pci_channel_state_t state)
4456 {
4457         struct net_device *netdev = pci_get_drvdata(pdev);
4458         struct igb_adapter *adapter = netdev_priv(netdev);
4459
4460         netif_device_detach(netdev);
4461
4462         if (netif_running(netdev))
4463                 igb_down(adapter);
4464         pci_disable_device(pdev);
4465
4466         /* Request a slot slot reset. */
4467         return PCI_ERS_RESULT_NEED_RESET;
4468 }
4469
4470 /**
4471  * igb_io_slot_reset - called after the pci bus has been reset.
4472  * @pdev: Pointer to PCI device
4473  *
4474  * Restart the card from scratch, as if from a cold-boot. Implementation
4475  * resembles the first-half of the igb_resume routine.
4476  */
4477 static pci_ers_result_t igb_io_slot_reset(struct pci_dev *pdev)
4478 {
4479         struct net_device *netdev = pci_get_drvdata(pdev);
4480         struct igb_adapter *adapter = netdev_priv(netdev);
4481         struct e1000_hw *hw = &adapter->hw;
4482         pci_ers_result_t result;
4483         int err;
4484
4485         if (adapter->need_ioport)
4486                 err = pci_enable_device(pdev);
4487         else
4488                 err = pci_enable_device_mem(pdev);
4489
4490         if (err) {
4491                 dev_err(&pdev->dev,
4492                         "Cannot re-enable PCI device after reset.\n");
4493                 result = PCI_ERS_RESULT_DISCONNECT;
4494         } else {
4495                 pci_set_master(pdev);
4496                 pci_restore_state(pdev);
4497
4498                 pci_enable_wake(pdev, PCI_D3hot, 0);
4499                 pci_enable_wake(pdev, PCI_D3cold, 0);
4500
4501                 igb_reset(adapter);
4502                 wr32(E1000_WUS, ~0);
4503                 result = PCI_ERS_RESULT_RECOVERED;
4504         }
4505
4506         err = pci_cleanup_aer_uncorrect_error_status(pdev);
4507         if (err) {
4508                 dev_err(&pdev->dev, "pci_cleanup_aer_uncorrect_error_status "
4509                         "failed 0x%0x\n", err);
4510                 /* non-fatal, continue */
4511         }
4512
4513         return result;
4514 }
4515
4516 /**
4517  * igb_io_resume - called when traffic can start flowing again.
4518  * @pdev: Pointer to PCI device
4519  *
4520  * This callback is called when the error recovery driver tells us that
4521  * its OK to resume normal operation. Implementation resembles the
4522  * second-half of the igb_resume routine.
4523  */
4524 static void igb_io_resume(struct pci_dev *pdev)
4525 {
4526         struct net_device *netdev = pci_get_drvdata(pdev);
4527         struct igb_adapter *adapter = netdev_priv(netdev);
4528
4529         if (netif_running(netdev)) {
4530                 if (igb_up(adapter)) {
4531                         dev_err(&pdev->dev, "igb_up failed after reset\n");
4532                         return;
4533                 }
4534         }
4535
4536         netif_device_attach(netdev);
4537
4538         /* let the f/w know that the h/w is now under the control of the
4539          * driver. */
4540         igb_get_hw_control(adapter);
4541 }
4542
4543 /* igb_main.c */