netdrv: don't truncate VLAN TCI with VLAN stripping
[linux-2.6.git] / 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/delay.h>
42 #include <linux/interrupt.h>
43 #include <linux/if_ether.h>
44
45 #include "igb.h"
46
47 #define DRV_VERSION "1.0.8-k2"
48 char igb_driver_name[] = "igb";
49 char igb_driver_version[] = DRV_VERSION;
50 static const char igb_driver_string[] =
51                                 "Intel(R) Gigabit Ethernet Network Driver";
52 static const char igb_copyright[] = "Copyright (c) 2007 Intel Corporation.";
53
54
55 static const struct e1000_info *igb_info_tbl[] = {
56         [board_82575] = &e1000_82575_info,
57 };
58
59 static struct pci_device_id igb_pci_tbl[] = {
60         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575EB_COPPER), board_82575 },
61         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575EB_FIBER_SERDES), board_82575 },
62         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575GB_QUAD_COPPER), board_82575 },
63         /* required last entry */
64         {0, }
65 };
66
67 MODULE_DEVICE_TABLE(pci, igb_pci_tbl);
68
69 void igb_reset(struct igb_adapter *);
70 static int igb_setup_all_tx_resources(struct igb_adapter *);
71 static int igb_setup_all_rx_resources(struct igb_adapter *);
72 static void igb_free_all_tx_resources(struct igb_adapter *);
73 static void igb_free_all_rx_resources(struct igb_adapter *);
74 static void igb_free_tx_resources(struct igb_ring *);
75 static void igb_free_rx_resources(struct igb_ring *);
76 void igb_update_stats(struct igb_adapter *);
77 static int igb_probe(struct pci_dev *, const struct pci_device_id *);
78 static void __devexit igb_remove(struct pci_dev *pdev);
79 static int igb_sw_init(struct igb_adapter *);
80 static int igb_open(struct net_device *);
81 static int igb_close(struct net_device *);
82 static void igb_configure_tx(struct igb_adapter *);
83 static void igb_configure_rx(struct igb_adapter *);
84 static void igb_setup_rctl(struct igb_adapter *);
85 static void igb_clean_all_tx_rings(struct igb_adapter *);
86 static void igb_clean_all_rx_rings(struct igb_adapter *);
87 static void igb_clean_tx_ring(struct igb_ring *);
88 static void igb_clean_rx_ring(struct igb_ring *);
89 static void igb_set_multi(struct net_device *);
90 static void igb_update_phy_info(unsigned long);
91 static void igb_watchdog(unsigned long);
92 static void igb_watchdog_task(struct work_struct *);
93 static int igb_xmit_frame_ring_adv(struct sk_buff *, struct net_device *,
94                                   struct igb_ring *);
95 static int igb_xmit_frame_adv(struct sk_buff *skb, struct net_device *);
96 static struct net_device_stats *igb_get_stats(struct net_device *);
97 static int igb_change_mtu(struct net_device *, int);
98 static int igb_set_mac(struct net_device *, void *);
99 static irqreturn_t igb_intr(int irq, void *);
100 static irqreturn_t igb_intr_msi(int irq, void *);
101 static irqreturn_t igb_msix_other(int irq, void *);
102 static irqreturn_t igb_msix_rx(int irq, void *);
103 static irqreturn_t igb_msix_tx(int irq, void *);
104 static int igb_clean_rx_ring_msix(struct napi_struct *, int);
105 static bool igb_clean_tx_irq(struct igb_ring *);
106 static int igb_clean(struct napi_struct *, int);
107 static bool igb_clean_rx_irq_adv(struct igb_ring *, int *, int);
108 static void igb_alloc_rx_buffers_adv(struct igb_ring *, int);
109 static int igb_ioctl(struct net_device *, struct ifreq *, int cmd);
110 static void igb_tx_timeout(struct net_device *);
111 static void igb_reset_task(struct work_struct *);
112 static void igb_vlan_rx_register(struct net_device *, struct vlan_group *);
113 static void igb_vlan_rx_add_vid(struct net_device *, u16);
114 static void igb_vlan_rx_kill_vid(struct net_device *, u16);
115 static void igb_restore_vlan(struct igb_adapter *);
116
117 static int igb_suspend(struct pci_dev *, pm_message_t);
118 #ifdef CONFIG_PM
119 static int igb_resume(struct pci_dev *);
120 #endif
121 static void igb_shutdown(struct pci_dev *);
122
123 #ifdef CONFIG_NET_POLL_CONTROLLER
124 /* for netdump / net console */
125 static void igb_netpoll(struct net_device *);
126 #endif
127
128 static pci_ers_result_t igb_io_error_detected(struct pci_dev *,
129                      pci_channel_state_t);
130 static pci_ers_result_t igb_io_slot_reset(struct pci_dev *);
131 static void igb_io_resume(struct pci_dev *);
132
133 static struct pci_error_handlers igb_err_handler = {
134         .error_detected = igb_io_error_detected,
135         .slot_reset = igb_io_slot_reset,
136         .resume = igb_io_resume,
137 };
138
139
140 static struct pci_driver igb_driver = {
141         .name     = igb_driver_name,
142         .id_table = igb_pci_tbl,
143         .probe    = igb_probe,
144         .remove   = __devexit_p(igb_remove),
145 #ifdef CONFIG_PM
146         /* Power Managment Hooks */
147         .suspend  = igb_suspend,
148         .resume   = igb_resume,
149 #endif
150         .shutdown = igb_shutdown,
151         .err_handler = &igb_err_handler
152 };
153
154 MODULE_AUTHOR("Intel Corporation, <e1000-devel@lists.sourceforge.net>");
155 MODULE_DESCRIPTION("Intel(R) Gigabit Ethernet Network Driver");
156 MODULE_LICENSE("GPL");
157 MODULE_VERSION(DRV_VERSION);
158
159 #ifdef DEBUG
160 /**
161  * igb_get_hw_dev_name - return device name string
162  * used by hardware layer to print debugging information
163  **/
164 char *igb_get_hw_dev_name(struct e1000_hw *hw)
165 {
166         struct igb_adapter *adapter = hw->back;
167         return adapter->netdev->name;
168 }
169 #endif
170
171 /**
172  * igb_init_module - Driver Registration Routine
173  *
174  * igb_init_module is the first routine called when the driver is
175  * loaded. All it does is register with the PCI subsystem.
176  **/
177 static int __init igb_init_module(void)
178 {
179         int ret;
180         printk(KERN_INFO "%s - version %s\n",
181                igb_driver_string, igb_driver_version);
182
183         printk(KERN_INFO "%s\n", igb_copyright);
184
185         ret = pci_register_driver(&igb_driver);
186         return ret;
187 }
188
189 module_init(igb_init_module);
190
191 /**
192  * igb_exit_module - Driver Exit Cleanup Routine
193  *
194  * igb_exit_module is called just before the driver is removed
195  * from memory.
196  **/
197 static void __exit igb_exit_module(void)
198 {
199         pci_unregister_driver(&igb_driver);
200 }
201
202 module_exit(igb_exit_module);
203
204 /**
205  * igb_alloc_queues - Allocate memory for all rings
206  * @adapter: board private structure to initialize
207  *
208  * We allocate one ring per queue at run-time since we don't know the
209  * number of queues at compile-time.
210  **/
211 static int igb_alloc_queues(struct igb_adapter *adapter)
212 {
213         int i;
214
215         adapter->tx_ring = kcalloc(adapter->num_tx_queues,
216                                    sizeof(struct igb_ring), GFP_KERNEL);
217         if (!adapter->tx_ring)
218                 return -ENOMEM;
219
220         adapter->rx_ring = kcalloc(adapter->num_rx_queues,
221                                    sizeof(struct igb_ring), GFP_KERNEL);
222         if (!adapter->rx_ring) {
223                 kfree(adapter->tx_ring);
224                 return -ENOMEM;
225         }
226
227         for (i = 0; i < adapter->num_rx_queues; i++) {
228                 struct igb_ring *ring = &(adapter->rx_ring[i]);
229                 ring->adapter = adapter;
230                 ring->queue_index = i;
231                 ring->itr_register = E1000_ITR;
232
233                 /* set a default napi handler for each rx_ring */
234                 netif_napi_add(adapter->netdev, &ring->napi, igb_clean, 64);
235         }
236         return 0;
237 }
238
239 #define IGB_N0_QUEUE -1
240 static void igb_assign_vector(struct igb_adapter *adapter, int rx_queue,
241                               int tx_queue, int msix_vector)
242 {
243         u32 msixbm = 0;
244         struct e1000_hw *hw = &adapter->hw;
245                 /* The 82575 assigns vectors using a bitmask, which matches the
246                    bitmask for the EICR/EIMS/EIMC registers.  To assign one
247                    or more queues to a vector, we write the appropriate bits
248                    into the MSIXBM register for that vector. */
249                 if (rx_queue > IGB_N0_QUEUE) {
250                         msixbm = E1000_EICR_RX_QUEUE0 << rx_queue;
251                         adapter->rx_ring[rx_queue].eims_value = msixbm;
252                 }
253                 if (tx_queue > IGB_N0_QUEUE) {
254                         msixbm |= E1000_EICR_TX_QUEUE0 << tx_queue;
255                         adapter->tx_ring[tx_queue].eims_value =
256                                   E1000_EICR_TX_QUEUE0 << tx_queue;
257                 }
258                 array_wr32(E1000_MSIXBM(0), msix_vector, msixbm);
259 }
260
261 /**
262  * igb_configure_msix - Configure MSI-X hardware
263  *
264  * igb_configure_msix sets up the hardware to properly
265  * generate MSI-X interrupts.
266  **/
267 static void igb_configure_msix(struct igb_adapter *adapter)
268 {
269         u32 tmp;
270         int i, vector = 0;
271         struct e1000_hw *hw = &adapter->hw;
272
273         adapter->eims_enable_mask = 0;
274
275         for (i = 0; i < adapter->num_tx_queues; i++) {
276                 struct igb_ring *tx_ring = &adapter->tx_ring[i];
277                 igb_assign_vector(adapter, IGB_N0_QUEUE, i, vector++);
278                 adapter->eims_enable_mask |= tx_ring->eims_value;
279                 if (tx_ring->itr_val)
280                         writel(1000000000 / (tx_ring->itr_val * 256),
281                                hw->hw_addr + tx_ring->itr_register);
282                 else
283                         writel(1, hw->hw_addr + tx_ring->itr_register);
284         }
285
286         for (i = 0; i < adapter->num_rx_queues; i++) {
287                 struct igb_ring *rx_ring = &adapter->rx_ring[i];
288                 igb_assign_vector(adapter, i, IGB_N0_QUEUE, vector++);
289                 adapter->eims_enable_mask |= rx_ring->eims_value;
290                 if (rx_ring->itr_val)
291                         writel(1000000000 / (rx_ring->itr_val * 256),
292                                hw->hw_addr + rx_ring->itr_register);
293                 else
294                         writel(1, hw->hw_addr + rx_ring->itr_register);
295         }
296
297
298         /* set vector for other causes, i.e. link changes */
299                 array_wr32(E1000_MSIXBM(0), vector++,
300                                       E1000_EIMS_OTHER);
301
302                 tmp = rd32(E1000_CTRL_EXT);
303                 /* enable MSI-X PBA support*/
304                 tmp |= E1000_CTRL_EXT_PBA_CLR;
305
306                 /* Auto-Mask interrupts upon ICR read. */
307                 tmp |= E1000_CTRL_EXT_EIAME;
308                 tmp |= E1000_CTRL_EXT_IRCA;
309
310                 wr32(E1000_CTRL_EXT, tmp);
311                 adapter->eims_enable_mask |= E1000_EIMS_OTHER;
312                 adapter->eims_other = E1000_EIMS_OTHER;
313
314         wrfl();
315 }
316
317 /**
318  * igb_request_msix - Initialize MSI-X interrupts
319  *
320  * igb_request_msix allocates MSI-X vectors and requests interrupts from the
321  * kernel.
322  **/
323 static int igb_request_msix(struct igb_adapter *adapter)
324 {
325         struct net_device *netdev = adapter->netdev;
326         int i, err = 0, vector = 0;
327
328         vector = 0;
329
330         for (i = 0; i < adapter->num_tx_queues; i++) {
331                 struct igb_ring *ring = &(adapter->tx_ring[i]);
332                 sprintf(ring->name, "%s-tx%d", netdev->name, i);
333                 err = request_irq(adapter->msix_entries[vector].vector,
334                                   &igb_msix_tx, 0, ring->name,
335                                   &(adapter->tx_ring[i]));
336                 if (err)
337                         goto out;
338                 ring->itr_register = E1000_EITR(0) + (vector << 2);
339                 ring->itr_val = adapter->itr;
340                 vector++;
341         }
342         for (i = 0; i < adapter->num_rx_queues; i++) {
343                 struct igb_ring *ring = &(adapter->rx_ring[i]);
344                 if (strlen(netdev->name) < (IFNAMSIZ - 5))
345                         sprintf(ring->name, "%s-rx%d", netdev->name, i);
346                 else
347                         memcpy(ring->name, netdev->name, IFNAMSIZ);
348                 err = request_irq(adapter->msix_entries[vector].vector,
349                                   &igb_msix_rx, 0, ring->name,
350                                   &(adapter->rx_ring[i]));
351                 if (err)
352                         goto out;
353                 ring->itr_register = E1000_EITR(0) + (vector << 2);
354                 ring->itr_val = adapter->itr;
355                 /* overwrite the poll routine for MSIX, we've already done
356                  * netif_napi_add */
357                 ring->napi.poll = &igb_clean_rx_ring_msix;
358                 vector++;
359         }
360
361         err = request_irq(adapter->msix_entries[vector].vector,
362                           &igb_msix_other, 0, netdev->name, netdev);
363         if (err)
364                 goto out;
365
366         igb_configure_msix(adapter);
367         return 0;
368 out:
369         return err;
370 }
371
372 static void igb_reset_interrupt_capability(struct igb_adapter *adapter)
373 {
374         if (adapter->msix_entries) {
375                 pci_disable_msix(adapter->pdev);
376                 kfree(adapter->msix_entries);
377                 adapter->msix_entries = NULL;
378         } else if (adapter->msi_enabled)
379                 pci_disable_msi(adapter->pdev);
380         return;
381 }
382
383
384 /**
385  * igb_set_interrupt_capability - set MSI or MSI-X if supported
386  *
387  * Attempt to configure interrupts using the best available
388  * capabilities of the hardware and kernel.
389  **/
390 static void igb_set_interrupt_capability(struct igb_adapter *adapter)
391 {
392         int err;
393         int numvecs, i;
394
395         numvecs = adapter->num_tx_queues + adapter->num_rx_queues + 1;
396         adapter->msix_entries = kcalloc(numvecs, sizeof(struct msix_entry),
397                                         GFP_KERNEL);
398         if (!adapter->msix_entries)
399                 goto msi_only;
400
401         for (i = 0; i < numvecs; i++)
402                 adapter->msix_entries[i].entry = i;
403
404         err = pci_enable_msix(adapter->pdev,
405                               adapter->msix_entries,
406                               numvecs);
407         if (err == 0)
408                 return;
409
410         igb_reset_interrupt_capability(adapter);
411
412         /* If we can't do MSI-X, try MSI */
413 msi_only:
414         adapter->num_rx_queues = 1;
415         if (!pci_enable_msi(adapter->pdev))
416                 adapter->msi_enabled = 1;
417         return;
418 }
419
420 /**
421  * igb_request_irq - initialize interrupts
422  *
423  * Attempts to configure interrupts using the best available
424  * capabilities of the hardware and kernel.
425  **/
426 static int igb_request_irq(struct igb_adapter *adapter)
427 {
428         struct net_device *netdev = adapter->netdev;
429         struct e1000_hw *hw = &adapter->hw;
430         int err = 0;
431
432         if (adapter->msix_entries) {
433                 err = igb_request_msix(adapter);
434                 if (!err)
435                         goto request_done;
436                 /* fall back to MSI */
437                 igb_reset_interrupt_capability(adapter);
438                 if (!pci_enable_msi(adapter->pdev))
439                         adapter->msi_enabled = 1;
440                 igb_free_all_tx_resources(adapter);
441                 igb_free_all_rx_resources(adapter);
442                 adapter->num_rx_queues = 1;
443                 igb_alloc_queues(adapter);
444         } else {
445                 wr32(E1000_MSIXBM(0), (E1000_EICR_RX_QUEUE0 |
446                                        E1000_EIMS_OTHER));
447         }
448
449         if (adapter->msi_enabled) {
450                 err = request_irq(adapter->pdev->irq, &igb_intr_msi, 0,
451                                   netdev->name, netdev);
452                 if (!err)
453                         goto request_done;
454                 /* fall back to legacy interrupts */
455                 igb_reset_interrupt_capability(adapter);
456                 adapter->msi_enabled = 0;
457         }
458
459         err = request_irq(adapter->pdev->irq, &igb_intr, IRQF_SHARED,
460                           netdev->name, netdev);
461
462         if (err)
463                 dev_err(&adapter->pdev->dev, "Error %d getting interrupt\n",
464                         err);
465
466 request_done:
467         return err;
468 }
469
470 static void igb_free_irq(struct igb_adapter *adapter)
471 {
472         struct net_device *netdev = adapter->netdev;
473
474         if (adapter->msix_entries) {
475                 int vector = 0, i;
476
477                 for (i = 0; i < adapter->num_tx_queues; i++)
478                         free_irq(adapter->msix_entries[vector++].vector,
479                                 &(adapter->tx_ring[i]));
480                 for (i = 0; i < adapter->num_rx_queues; i++)
481                         free_irq(adapter->msix_entries[vector++].vector,
482                                 &(adapter->rx_ring[i]));
483
484                 free_irq(adapter->msix_entries[vector++].vector, netdev);
485                 return;
486         }
487
488         free_irq(adapter->pdev->irq, netdev);
489 }
490
491 /**
492  * igb_irq_disable - Mask off interrupt generation on the NIC
493  * @adapter: board private structure
494  **/
495 static void igb_irq_disable(struct igb_adapter *adapter)
496 {
497         struct e1000_hw *hw = &adapter->hw;
498
499         if (adapter->msix_entries) {
500                 wr32(E1000_EIAM, 0);
501                 wr32(E1000_EIMC, ~0);
502                 wr32(E1000_EIAC, 0);
503         }
504
505         wr32(E1000_IAM, 0);
506         wr32(E1000_IMC, ~0);
507         wrfl();
508         synchronize_irq(adapter->pdev->irq);
509 }
510
511 /**
512  * igb_irq_enable - Enable default interrupt generation settings
513  * @adapter: board private structure
514  **/
515 static void igb_irq_enable(struct igb_adapter *adapter)
516 {
517         struct e1000_hw *hw = &adapter->hw;
518
519         if (adapter->msix_entries) {
520                 wr32(E1000_EIAC, adapter->eims_enable_mask);
521                 wr32(E1000_EIAM, adapter->eims_enable_mask);
522                 wr32(E1000_EIMS, adapter->eims_enable_mask);
523                 wr32(E1000_IMS, E1000_IMS_LSC);
524         } else {
525                 wr32(E1000_IMS, IMS_ENABLE_MASK);
526                 wr32(E1000_IAM, IMS_ENABLE_MASK);
527         }
528 }
529
530 static void igb_update_mng_vlan(struct igb_adapter *adapter)
531 {
532         struct net_device *netdev = adapter->netdev;
533         u16 vid = adapter->hw.mng_cookie.vlan_id;
534         u16 old_vid = adapter->mng_vlan_id;
535         if (adapter->vlgrp) {
536                 if (!vlan_group_get_device(adapter->vlgrp, vid)) {
537                         if (adapter->hw.mng_cookie.status &
538                                 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
539                                 igb_vlan_rx_add_vid(netdev, vid);
540                                 adapter->mng_vlan_id = vid;
541                         } else
542                                 adapter->mng_vlan_id = IGB_MNG_VLAN_NONE;
543
544                         if ((old_vid != (u16)IGB_MNG_VLAN_NONE) &&
545                                         (vid != old_vid) &&
546                             !vlan_group_get_device(adapter->vlgrp, old_vid))
547                                 igb_vlan_rx_kill_vid(netdev, old_vid);
548                 } else
549                         adapter->mng_vlan_id = vid;
550         }
551 }
552
553 /**
554  * igb_release_hw_control - release control of the h/w to f/w
555  * @adapter: address of board private structure
556  *
557  * igb_release_hw_control resets CTRL_EXT:DRV_LOAD bit.
558  * For ASF and Pass Through versions of f/w this means that the
559  * driver is no longer loaded.
560  *
561  **/
562 static void igb_release_hw_control(struct igb_adapter *adapter)
563 {
564         struct e1000_hw *hw = &adapter->hw;
565         u32 ctrl_ext;
566
567         /* Let firmware take over control of h/w */
568         ctrl_ext = rd32(E1000_CTRL_EXT);
569         wr32(E1000_CTRL_EXT,
570                         ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
571 }
572
573
574 /**
575  * igb_get_hw_control - get control of the h/w from f/w
576  * @adapter: address of board private structure
577  *
578  * igb_get_hw_control sets CTRL_EXT:DRV_LOAD bit.
579  * For ASF and Pass Through versions of f/w this means that
580  * the driver is loaded.
581  *
582  **/
583 static void igb_get_hw_control(struct igb_adapter *adapter)
584 {
585         struct e1000_hw *hw = &adapter->hw;
586         u32 ctrl_ext;
587
588         /* Let firmware know the driver has taken over */
589         ctrl_ext = rd32(E1000_CTRL_EXT);
590         wr32(E1000_CTRL_EXT,
591                         ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
592 }
593
594 static void igb_init_manageability(struct igb_adapter *adapter)
595 {
596         struct e1000_hw *hw = &adapter->hw;
597
598         if (adapter->en_mng_pt) {
599                 u32 manc2h = rd32(E1000_MANC2H);
600                 u32 manc = rd32(E1000_MANC);
601
602                 /* enable receiving management packets to the host */
603                 /* this will probably generate destination unreachable messages
604                  * from the host OS, but the packets will be handled on SMBUS */
605                 manc |= E1000_MANC_EN_MNG2HOST;
606 #define E1000_MNG2HOST_PORT_623 (1 << 5)
607 #define E1000_MNG2HOST_PORT_664 (1 << 6)
608                 manc2h |= E1000_MNG2HOST_PORT_623;
609                 manc2h |= E1000_MNG2HOST_PORT_664;
610                 wr32(E1000_MANC2H, manc2h);
611
612                 wr32(E1000_MANC, manc);
613         }
614 }
615
616 /**
617  * igb_configure - configure the hardware for RX and TX
618  * @adapter: private board structure
619  **/
620 static void igb_configure(struct igb_adapter *adapter)
621 {
622         struct net_device *netdev = adapter->netdev;
623         int i;
624
625         igb_get_hw_control(adapter);
626         igb_set_multi(netdev);
627
628         igb_restore_vlan(adapter);
629         igb_init_manageability(adapter);
630
631         igb_configure_tx(adapter);
632         igb_setup_rctl(adapter);
633         igb_configure_rx(adapter);
634
635         igb_rx_fifo_flush_82575(&adapter->hw);
636
637         /* call IGB_DESC_UNUSED which always leaves
638          * at least 1 descriptor unused to make sure
639          * next_to_use != next_to_clean */
640         for (i = 0; i < adapter->num_rx_queues; i++) {
641                 struct igb_ring *ring = &adapter->rx_ring[i];
642                 igb_alloc_rx_buffers_adv(ring, IGB_DESC_UNUSED(ring));
643         }
644
645
646         adapter->tx_queue_len = netdev->tx_queue_len;
647 }
648
649
650 /**
651  * igb_up - Open the interface and prepare it to handle traffic
652  * @adapter: board private structure
653  **/
654
655 int igb_up(struct igb_adapter *adapter)
656 {
657         struct e1000_hw *hw = &adapter->hw;
658         int i;
659
660         /* hardware has been reset, we need to reload some things */
661         igb_configure(adapter);
662
663         clear_bit(__IGB_DOWN, &adapter->state);
664
665         for (i = 0; i < adapter->num_rx_queues; i++)
666                 napi_enable(&adapter->rx_ring[i].napi);
667         if (adapter->msix_entries)
668                 igb_configure_msix(adapter);
669
670         /* Clear any pending interrupts. */
671         rd32(E1000_ICR);
672         igb_irq_enable(adapter);
673
674         /* Fire a link change interrupt to start the watchdog. */
675         wr32(E1000_ICS, E1000_ICS_LSC);
676         return 0;
677 }
678
679 void igb_down(struct igb_adapter *adapter)
680 {
681         struct e1000_hw *hw = &adapter->hw;
682         struct net_device *netdev = adapter->netdev;
683         u32 tctl, rctl;
684         int i;
685
686         /* signal that we're down so the interrupt handler does not
687          * reschedule our watchdog timer */
688         set_bit(__IGB_DOWN, &adapter->state);
689
690         /* disable receives in the hardware */
691         rctl = rd32(E1000_RCTL);
692         wr32(E1000_RCTL, rctl & ~E1000_RCTL_EN);
693         /* flush and sleep below */
694
695         netif_stop_queue(netdev);
696
697         /* disable transmits in the hardware */
698         tctl = rd32(E1000_TCTL);
699         tctl &= ~E1000_TCTL_EN;
700         wr32(E1000_TCTL, tctl);
701         /* flush both disables and wait for them to finish */
702         wrfl();
703         msleep(10);
704
705         for (i = 0; i < adapter->num_rx_queues; i++)
706                 napi_disable(&adapter->rx_ring[i].napi);
707
708         igb_irq_disable(adapter);
709
710         del_timer_sync(&adapter->watchdog_timer);
711         del_timer_sync(&adapter->phy_info_timer);
712
713         netdev->tx_queue_len = adapter->tx_queue_len;
714         netif_carrier_off(netdev);
715         adapter->link_speed = 0;
716         adapter->link_duplex = 0;
717
718         if (!pci_channel_offline(adapter->pdev))
719                 igb_reset(adapter);
720         igb_clean_all_tx_rings(adapter);
721         igb_clean_all_rx_rings(adapter);
722 }
723
724 void igb_reinit_locked(struct igb_adapter *adapter)
725 {
726         WARN_ON(in_interrupt());
727         while (test_and_set_bit(__IGB_RESETTING, &adapter->state))
728                 msleep(1);
729         igb_down(adapter);
730         igb_up(adapter);
731         clear_bit(__IGB_RESETTING, &adapter->state);
732 }
733
734 void igb_reset(struct igb_adapter *adapter)
735 {
736         struct e1000_hw *hw = &adapter->hw;
737         struct e1000_fc_info *fc = &adapter->hw.fc;
738         u32 pba = 0, tx_space, min_tx_space, min_rx_space;
739         u16 hwm;
740
741         /* Repartition Pba for greater than 9k mtu
742          * To take effect CTRL.RST is required.
743          */
744         pba = E1000_PBA_34K;
745
746         if (adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) {
747                 /* adjust PBA for jumbo frames */
748                 wr32(E1000_PBA, pba);
749
750                 /* To maintain wire speed transmits, the Tx FIFO should be
751                  * large enough to accommodate two full transmit packets,
752                  * rounded up to the next 1KB and expressed in KB.  Likewise,
753                  * the Rx FIFO should be large enough to accommodate at least
754                  * one full receive packet and is similarly rounded up and
755                  * expressed in KB. */
756                 pba = rd32(E1000_PBA);
757                 /* upper 16 bits has Tx packet buffer allocation size in KB */
758                 tx_space = pba >> 16;
759                 /* lower 16 bits has Rx packet buffer allocation size in KB */
760                 pba &= 0xffff;
761                 /* the tx fifo also stores 16 bytes of information about the tx
762                  * but don't include ethernet FCS because hardware appends it */
763                 min_tx_space = (adapter->max_frame_size +
764                                 sizeof(struct e1000_tx_desc) -
765                                 ETH_FCS_LEN) * 2;
766                 min_tx_space = ALIGN(min_tx_space, 1024);
767                 min_tx_space >>= 10;
768                 /* software strips receive CRC, so leave room for it */
769                 min_rx_space = adapter->max_frame_size;
770                 min_rx_space = ALIGN(min_rx_space, 1024);
771                 min_rx_space >>= 10;
772
773                 /* If current Tx allocation is less than the min Tx FIFO size,
774                  * and the min Tx FIFO size is less than the current Rx FIFO
775                  * allocation, take space away from current Rx allocation */
776                 if (tx_space < min_tx_space &&
777                     ((min_tx_space - tx_space) < pba)) {
778                         pba = pba - (min_tx_space - tx_space);
779
780                         /* if short on rx space, rx wins and must trump tx
781                          * adjustment */
782                         if (pba < min_rx_space)
783                                 pba = min_rx_space;
784                 }
785         }
786         wr32(E1000_PBA, pba);
787
788         /* flow control settings */
789         /* The high water mark must be low enough to fit one full frame
790          * (or the size used for early receive) above it in the Rx FIFO.
791          * Set it to the lower of:
792          * - 90% of the Rx FIFO size, or
793          * - the full Rx FIFO size minus one full frame */
794         hwm = min(((pba << 10) * 9 / 10),
795                   ((pba << 10) - adapter->max_frame_size));
796
797         fc->high_water = hwm & 0xFFF8;  /* 8-byte granularity */
798         fc->low_water = fc->high_water - 8;
799         fc->pause_time = 0xFFFF;
800         fc->send_xon = 1;
801         fc->type = fc->original_type;
802
803         /* Allow time for pending master requests to run */
804         adapter->hw.mac.ops.reset_hw(&adapter->hw);
805         wr32(E1000_WUC, 0);
806
807         if (adapter->hw.mac.ops.init_hw(&adapter->hw))
808                 dev_err(&adapter->pdev->dev, "Hardware Error\n");
809
810         igb_update_mng_vlan(adapter);
811
812         /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
813         wr32(E1000_VET, ETHERNET_IEEE_VLAN_TYPE);
814
815         igb_reset_adaptive(&adapter->hw);
816         if (adapter->hw.phy.ops.get_phy_info)
817                 adapter->hw.phy.ops.get_phy_info(&adapter->hw);
818 }
819
820 /**
821  * igb_is_need_ioport - determine if an adapter needs ioport resources or not
822  * @pdev: PCI device information struct
823  *
824  * Returns true if an adapter needs ioport resources
825  **/
826 static int igb_is_need_ioport(struct pci_dev *pdev)
827 {
828         switch (pdev->device) {
829         /* Currently there are no adapters that need ioport resources */
830         default:
831                 return false;
832         }
833 }
834
835 /**
836  * igb_probe - Device Initialization Routine
837  * @pdev: PCI device information struct
838  * @ent: entry in igb_pci_tbl
839  *
840  * Returns 0 on success, negative on failure
841  *
842  * igb_probe initializes an adapter identified by a pci_dev structure.
843  * The OS initialization, configuring of the adapter private structure,
844  * and a hardware reset occur.
845  **/
846 static int __devinit igb_probe(struct pci_dev *pdev,
847                                const struct pci_device_id *ent)
848 {
849         struct net_device *netdev;
850         struct igb_adapter *adapter;
851         struct e1000_hw *hw;
852         const struct e1000_info *ei = igb_info_tbl[ent->driver_data];
853         unsigned long mmio_start, mmio_len;
854         int i, err, pci_using_dac;
855         u16 eeprom_data = 0;
856         u16 eeprom_apme_mask = IGB_EEPROM_APME;
857         u32 part_num;
858         int bars, need_ioport;
859
860         /* do not allocate ioport bars when not needed */
861         need_ioport = igb_is_need_ioport(pdev);
862         if (need_ioport) {
863                 bars = pci_select_bars(pdev, IORESOURCE_MEM | IORESOURCE_IO);
864                 err = pci_enable_device(pdev);
865         } else {
866                 bars = pci_select_bars(pdev, IORESOURCE_MEM);
867                 err = pci_enable_device_mem(pdev);
868         }
869         if (err)
870                 return err;
871
872         pci_using_dac = 0;
873         err = pci_set_dma_mask(pdev, DMA_64BIT_MASK);
874         if (!err) {
875                 err = pci_set_consistent_dma_mask(pdev, DMA_64BIT_MASK);
876                 if (!err)
877                         pci_using_dac = 1;
878         } else {
879                 err = pci_set_dma_mask(pdev, DMA_32BIT_MASK);
880                 if (err) {
881                         err = pci_set_consistent_dma_mask(pdev, DMA_32BIT_MASK);
882                         if (err) {
883                                 dev_err(&pdev->dev, "No usable DMA "
884                                         "configuration, aborting\n");
885                                 goto err_dma;
886                         }
887                 }
888         }
889
890         err = pci_request_selected_regions(pdev, bars, igb_driver_name);
891         if (err)
892                 goto err_pci_reg;
893
894         pci_set_master(pdev);
895         pci_save_state(pdev);
896
897         err = -ENOMEM;
898         netdev = alloc_etherdev(sizeof(struct igb_adapter));
899         if (!netdev)
900                 goto err_alloc_etherdev;
901
902         SET_NETDEV_DEV(netdev, &pdev->dev);
903
904         pci_set_drvdata(pdev, netdev);
905         adapter = netdev_priv(netdev);
906         adapter->netdev = netdev;
907         adapter->pdev = pdev;
908         hw = &adapter->hw;
909         hw->back = adapter;
910         adapter->msg_enable = NETIF_MSG_DRV | NETIF_MSG_PROBE;
911         adapter->bars = bars;
912         adapter->need_ioport = need_ioport;
913
914         mmio_start = pci_resource_start(pdev, 0);
915         mmio_len = pci_resource_len(pdev, 0);
916
917         err = -EIO;
918         adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
919         if (!adapter->hw.hw_addr)
920                 goto err_ioremap;
921
922         netdev->open = &igb_open;
923         netdev->stop = &igb_close;
924         netdev->get_stats = &igb_get_stats;
925         netdev->set_multicast_list = &igb_set_multi;
926         netdev->set_mac_address = &igb_set_mac;
927         netdev->change_mtu = &igb_change_mtu;
928         netdev->do_ioctl = &igb_ioctl;
929         igb_set_ethtool_ops(netdev);
930         netdev->tx_timeout = &igb_tx_timeout;
931         netdev->watchdog_timeo = 5 * HZ;
932         netdev->vlan_rx_register = igb_vlan_rx_register;
933         netdev->vlan_rx_add_vid = igb_vlan_rx_add_vid;
934         netdev->vlan_rx_kill_vid = igb_vlan_rx_kill_vid;
935 #ifdef CONFIG_NET_POLL_CONTROLLER
936         netdev->poll_controller = igb_netpoll;
937 #endif
938         netdev->hard_start_xmit = &igb_xmit_frame_adv;
939
940         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
941
942         netdev->mem_start = mmio_start;
943         netdev->mem_end = mmio_start + mmio_len;
944
945         /* PCI config space info */
946         hw->vendor_id = pdev->vendor;
947         hw->device_id = pdev->device;
948         hw->revision_id = pdev->revision;
949         hw->subsystem_vendor_id = pdev->subsystem_vendor;
950         hw->subsystem_device_id = pdev->subsystem_device;
951
952         /* setup the private structure */
953         hw->back = adapter;
954         /* Copy the default MAC, PHY and NVM function pointers */
955         memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
956         memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
957         memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
958         /* Initialize skew-specific constants */
959         err = ei->get_invariants(hw);
960         if (err)
961                 goto err_hw_init;
962
963         err = igb_sw_init(adapter);
964         if (err)
965                 goto err_sw_init;
966
967         igb_get_bus_info_pcie(hw);
968
969         hw->phy.autoneg_wait_to_complete = false;
970         hw->mac.adaptive_ifs = true;
971
972         /* Copper options */
973         if (hw->phy.media_type == e1000_media_type_copper) {
974                 hw->phy.mdix = AUTO_ALL_MODES;
975                 hw->phy.disable_polarity_correction = false;
976                 hw->phy.ms_type = e1000_ms_hw_default;
977         }
978
979         if (igb_check_reset_block(hw))
980                 dev_info(&pdev->dev,
981                         "PHY reset is blocked due to SOL/IDER session.\n");
982
983         netdev->features = NETIF_F_SG |
984                            NETIF_F_HW_CSUM |
985                            NETIF_F_HW_VLAN_TX |
986                            NETIF_F_HW_VLAN_RX |
987                            NETIF_F_HW_VLAN_FILTER;
988
989         netdev->features |= NETIF_F_TSO;
990         netdev->features |= NETIF_F_TSO6;
991
992         netdev->vlan_features |= NETIF_F_TSO;
993         netdev->vlan_features |= NETIF_F_TSO6;
994         netdev->vlan_features |= NETIF_F_HW_CSUM;
995         netdev->vlan_features |= NETIF_F_SG;
996
997         if (pci_using_dac)
998                 netdev->features |= NETIF_F_HIGHDMA;
999
1000         netdev->features |= NETIF_F_LLTX;
1001         adapter->en_mng_pt = igb_enable_mng_pass_thru(&adapter->hw);
1002
1003         /* before reading the NVM, reset the controller to put the device in a
1004          * known good starting state */
1005         hw->mac.ops.reset_hw(hw);
1006
1007         /* make sure the NVM is good */
1008         if (igb_validate_nvm_checksum(hw) < 0) {
1009                 dev_err(&pdev->dev, "The NVM Checksum Is Not Valid\n");
1010                 err = -EIO;
1011                 goto err_eeprom;
1012         }
1013
1014         /* copy the MAC address out of the NVM */
1015         if (hw->mac.ops.read_mac_addr(hw))
1016                 dev_err(&pdev->dev, "NVM Read Error\n");
1017
1018         memcpy(netdev->dev_addr, hw->mac.addr, netdev->addr_len);
1019         memcpy(netdev->perm_addr, hw->mac.addr, netdev->addr_len);
1020
1021         if (!is_valid_ether_addr(netdev->perm_addr)) {
1022                 dev_err(&pdev->dev, "Invalid MAC Address\n");
1023                 err = -EIO;
1024                 goto err_eeprom;
1025         }
1026
1027         init_timer(&adapter->watchdog_timer);
1028         adapter->watchdog_timer.function = &igb_watchdog;
1029         adapter->watchdog_timer.data = (unsigned long) adapter;
1030
1031         init_timer(&adapter->phy_info_timer);
1032         adapter->phy_info_timer.function = &igb_update_phy_info;
1033         adapter->phy_info_timer.data = (unsigned long) adapter;
1034
1035         INIT_WORK(&adapter->reset_task, igb_reset_task);
1036         INIT_WORK(&adapter->watchdog_task, igb_watchdog_task);
1037
1038         /* Initialize link & ring properties that are user-changeable */
1039         adapter->tx_ring->count = 256;
1040         for (i = 0; i < adapter->num_tx_queues; i++)
1041                 adapter->tx_ring[i].count = adapter->tx_ring->count;
1042         adapter->rx_ring->count = 256;
1043         for (i = 0; i < adapter->num_rx_queues; i++)
1044                 adapter->rx_ring[i].count = adapter->rx_ring->count;
1045
1046         adapter->fc_autoneg = true;
1047         hw->mac.autoneg = true;
1048         hw->phy.autoneg_advertised = 0x2f;
1049
1050         hw->fc.original_type = e1000_fc_default;
1051         hw->fc.type = e1000_fc_default;
1052
1053         adapter->itr_setting = 3;
1054         adapter->itr = IGB_START_ITR;
1055
1056         igb_validate_mdi_setting(hw);
1057
1058         adapter->rx_csum = 1;
1059
1060         /* Initial Wake on LAN setting If APM wake is enabled in the EEPROM,
1061          * enable the ACPI Magic Packet filter
1062          */
1063
1064         if (hw->bus.func == 0 ||
1065             hw->device_id == E1000_DEV_ID_82575EB_COPPER)
1066                 hw->nvm.ops.read_nvm(hw, NVM_INIT_CONTROL3_PORT_A, 1,
1067                                      &eeprom_data);
1068
1069         if (eeprom_data & eeprom_apme_mask)
1070                 adapter->eeprom_wol |= E1000_WUFC_MAG;
1071
1072         /* now that we have the eeprom settings, apply the special cases where
1073          * the eeprom may be wrong or the board simply won't support wake on
1074          * lan on a particular port */
1075         switch (pdev->device) {
1076         case E1000_DEV_ID_82575GB_QUAD_COPPER:
1077                 adapter->eeprom_wol = 0;
1078                 break;
1079         case E1000_DEV_ID_82575EB_FIBER_SERDES:
1080                 /* Wake events only supported on port A for dual fiber
1081                  * regardless of eeprom setting */
1082                 if (rd32(E1000_STATUS) & E1000_STATUS_FUNC_1)
1083                         adapter->eeprom_wol = 0;
1084                 break;
1085         }
1086
1087         /* initialize the wol settings based on the eeprom settings */
1088         adapter->wol = adapter->eeprom_wol;
1089
1090         /* reset the hardware with the new settings */
1091         igb_reset(adapter);
1092
1093         /* let the f/w know that the h/w is now under the control of the
1094          * driver. */
1095         igb_get_hw_control(adapter);
1096
1097         /* tell the stack to leave us alone until igb_open() is called */
1098         netif_carrier_off(netdev);
1099         netif_stop_queue(netdev);
1100
1101         strcpy(netdev->name, "eth%d");
1102         err = register_netdev(netdev);
1103         if (err)
1104                 goto err_register;
1105
1106         dev_info(&pdev->dev, "Intel(R) Gigabit Ethernet Network Connection\n");
1107         /* print bus type/speed/width info */
1108         dev_info(&pdev->dev,
1109                  "%s: (PCIe:%s:%s) %02x:%02x:%02x:%02x:%02x:%02x\n",
1110                  netdev->name,
1111                  ((hw->bus.speed == e1000_bus_speed_2500)
1112                   ? "2.5Gb/s" : "unknown"),
1113                  ((hw->bus.width == e1000_bus_width_pcie_x4)
1114                   ? "Width x4" : (hw->bus.width == e1000_bus_width_pcie_x1)
1115                   ? "Width x1" : "unknown"),
1116                  netdev->dev_addr[0], netdev->dev_addr[1], netdev->dev_addr[2],
1117                  netdev->dev_addr[3], netdev->dev_addr[4], netdev->dev_addr[5]);
1118
1119         igb_read_part_num(hw, &part_num);
1120         dev_info(&pdev->dev, "%s: PBA No: %06x-%03x\n", netdev->name,
1121                 (part_num >> 8), (part_num & 0xff));
1122
1123         dev_info(&pdev->dev,
1124                 "Using %s interrupts. %d rx queue(s), %d tx queue(s)\n",
1125                 adapter->msix_entries ? "MSI-X" :
1126                 adapter->msi_enabled ? "MSI" : "legacy",
1127                 adapter->num_rx_queues, adapter->num_tx_queues);
1128
1129         return 0;
1130
1131 err_register:
1132         igb_release_hw_control(adapter);
1133 err_eeprom:
1134         if (!igb_check_reset_block(hw))
1135                 hw->phy.ops.reset_phy(hw);
1136
1137         if (hw->flash_address)
1138                 iounmap(hw->flash_address);
1139
1140         igb_remove_device(hw);
1141         kfree(adapter->tx_ring);
1142         kfree(adapter->rx_ring);
1143 err_sw_init:
1144 err_hw_init:
1145         iounmap(hw->hw_addr);
1146 err_ioremap:
1147         free_netdev(netdev);
1148 err_alloc_etherdev:
1149         pci_release_selected_regions(pdev, bars);
1150 err_pci_reg:
1151 err_dma:
1152         pci_disable_device(pdev);
1153         return err;
1154 }
1155
1156 /**
1157  * igb_remove - Device Removal Routine
1158  * @pdev: PCI device information struct
1159  *
1160  * igb_remove is called by the PCI subsystem to alert the driver
1161  * that it should release a PCI device.  The could be caused by a
1162  * Hot-Plug event, or because the driver is going to be removed from
1163  * memory.
1164  **/
1165 static void __devexit igb_remove(struct pci_dev *pdev)
1166 {
1167         struct net_device *netdev = pci_get_drvdata(pdev);
1168         struct igb_adapter *adapter = netdev_priv(netdev);
1169
1170         /* flush_scheduled work may reschedule our watchdog task, so
1171          * explicitly disable watchdog tasks from being rescheduled  */
1172         set_bit(__IGB_DOWN, &adapter->state);
1173         del_timer_sync(&adapter->watchdog_timer);
1174         del_timer_sync(&adapter->phy_info_timer);
1175
1176         flush_scheduled_work();
1177
1178         /* Release control of h/w to f/w.  If f/w is AMT enabled, this
1179          * would have already happened in close and is redundant. */
1180         igb_release_hw_control(adapter);
1181
1182         unregister_netdev(netdev);
1183
1184         if (!igb_check_reset_block(&adapter->hw))
1185                 adapter->hw.phy.ops.reset_phy(&adapter->hw);
1186
1187         igb_remove_device(&adapter->hw);
1188         igb_reset_interrupt_capability(adapter);
1189
1190         kfree(adapter->tx_ring);
1191         kfree(adapter->rx_ring);
1192
1193         iounmap(adapter->hw.hw_addr);
1194         if (adapter->hw.flash_address)
1195                 iounmap(adapter->hw.flash_address);
1196         pci_release_selected_regions(pdev, adapter->bars);
1197
1198         free_netdev(netdev);
1199
1200         pci_disable_device(pdev);
1201 }
1202
1203 /**
1204  * igb_sw_init - Initialize general software structures (struct igb_adapter)
1205  * @adapter: board private structure to initialize
1206  *
1207  * igb_sw_init initializes the Adapter private data structure.
1208  * Fields are initialized based on PCI device information and
1209  * OS network device settings (MTU size).
1210  **/
1211 static int __devinit igb_sw_init(struct igb_adapter *adapter)
1212 {
1213         struct e1000_hw *hw = &adapter->hw;
1214         struct net_device *netdev = adapter->netdev;
1215         struct pci_dev *pdev = adapter->pdev;
1216
1217         pci_read_config_word(pdev, PCI_COMMAND, &hw->bus.pci_cmd_word);
1218
1219         adapter->rx_buffer_len = MAXIMUM_ETHERNET_VLAN_SIZE;
1220         adapter->rx_ps_hdr_size = 0; /* disable packet split */
1221         adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
1222         adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
1223
1224         /* Number of supported queues. */
1225         /* Having more queues than CPUs doesn't make sense. */
1226         adapter->num_tx_queues = 1;
1227         adapter->num_rx_queues = min(IGB_MAX_RX_QUEUES, num_online_cpus());
1228
1229         igb_set_interrupt_capability(adapter);
1230
1231         if (igb_alloc_queues(adapter)) {
1232                 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
1233                 return -ENOMEM;
1234         }
1235
1236         /* Explicitly disable IRQ since the NIC can be in any state. */
1237         igb_irq_disable(adapter);
1238
1239         set_bit(__IGB_DOWN, &adapter->state);
1240         return 0;
1241 }
1242
1243 /**
1244  * igb_open - Called when a network interface is made active
1245  * @netdev: network interface device structure
1246  *
1247  * Returns 0 on success, negative value on failure
1248  *
1249  * The open entry point is called when a network interface is made
1250  * active by the system (IFF_UP).  At this point all resources needed
1251  * for transmit and receive operations are allocated, the interrupt
1252  * handler is registered with the OS, the watchdog timer is started,
1253  * and the stack is notified that the interface is ready.
1254  **/
1255 static int igb_open(struct net_device *netdev)
1256 {
1257         struct igb_adapter *adapter = netdev_priv(netdev);
1258         struct e1000_hw *hw = &adapter->hw;
1259         int err;
1260         int i;
1261
1262         /* disallow open during test */
1263         if (test_bit(__IGB_TESTING, &adapter->state))
1264                 return -EBUSY;
1265
1266         /* allocate transmit descriptors */
1267         err = igb_setup_all_tx_resources(adapter);
1268         if (err)
1269                 goto err_setup_tx;
1270
1271         /* allocate receive descriptors */
1272         err = igb_setup_all_rx_resources(adapter);
1273         if (err)
1274                 goto err_setup_rx;
1275
1276         /* e1000_power_up_phy(adapter); */
1277
1278         adapter->mng_vlan_id = IGB_MNG_VLAN_NONE;
1279         if ((adapter->hw.mng_cookie.status &
1280              E1000_MNG_DHCP_COOKIE_STATUS_VLAN))
1281                 igb_update_mng_vlan(adapter);
1282
1283         /* before we allocate an interrupt, we must be ready to handle it.
1284          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
1285          * as soon as we call pci_request_irq, so we have to setup our
1286          * clean_rx handler before we do so.  */
1287         igb_configure(adapter);
1288
1289         err = igb_request_irq(adapter);
1290         if (err)
1291                 goto err_req_irq;
1292
1293         /* From here on the code is the same as igb_up() */
1294         clear_bit(__IGB_DOWN, &adapter->state);
1295
1296         for (i = 0; i < adapter->num_rx_queues; i++)
1297                 napi_enable(&adapter->rx_ring[i].napi);
1298
1299         /* Clear any pending interrupts. */
1300         rd32(E1000_ICR);
1301
1302         igb_irq_enable(adapter);
1303
1304         /* Fire a link status change interrupt to start the watchdog. */
1305         wr32(E1000_ICS, E1000_ICS_LSC);
1306
1307         return 0;
1308
1309 err_req_irq:
1310         igb_release_hw_control(adapter);
1311         /* e1000_power_down_phy(adapter); */
1312         igb_free_all_rx_resources(adapter);
1313 err_setup_rx:
1314         igb_free_all_tx_resources(adapter);
1315 err_setup_tx:
1316         igb_reset(adapter);
1317
1318         return err;
1319 }
1320
1321 /**
1322  * igb_close - Disables a network interface
1323  * @netdev: network interface device structure
1324  *
1325  * Returns 0, this is not allowed to fail
1326  *
1327  * The close entry point is called when an interface is de-activated
1328  * by the OS.  The hardware is still under the driver's control, but
1329  * needs to be disabled.  A global MAC reset is issued to stop the
1330  * hardware, and all transmit and receive resources are freed.
1331  **/
1332 static int igb_close(struct net_device *netdev)
1333 {
1334         struct igb_adapter *adapter = netdev_priv(netdev);
1335
1336         WARN_ON(test_bit(__IGB_RESETTING, &adapter->state));
1337         igb_down(adapter);
1338
1339         igb_free_irq(adapter);
1340
1341         igb_free_all_tx_resources(adapter);
1342         igb_free_all_rx_resources(adapter);
1343
1344         /* kill manageability vlan ID if supported, but not if a vlan with
1345          * the same ID is registered on the host OS (let 8021q kill it) */
1346         if ((adapter->hw.mng_cookie.status &
1347                           E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
1348              !(adapter->vlgrp &&
1349                vlan_group_get_device(adapter->vlgrp, adapter->mng_vlan_id)))
1350                 igb_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
1351
1352         return 0;
1353 }
1354
1355 /**
1356  * igb_setup_tx_resources - allocate Tx resources (Descriptors)
1357  * @adapter: board private structure
1358  * @tx_ring: tx descriptor ring (for a specific queue) to setup
1359  *
1360  * Return 0 on success, negative on failure
1361  **/
1362
1363 int igb_setup_tx_resources(struct igb_adapter *adapter,
1364                            struct igb_ring *tx_ring)
1365 {
1366         struct pci_dev *pdev = adapter->pdev;
1367         int size;
1368
1369         size = sizeof(struct igb_buffer) * tx_ring->count;
1370         tx_ring->buffer_info = vmalloc(size);
1371         if (!tx_ring->buffer_info)
1372                 goto err;
1373         memset(tx_ring->buffer_info, 0, size);
1374
1375         /* round up to nearest 4K */
1376         tx_ring->size = tx_ring->count * sizeof(struct e1000_tx_desc)
1377                         + sizeof(u32);
1378         tx_ring->size = ALIGN(tx_ring->size, 4096);
1379
1380         tx_ring->desc = pci_alloc_consistent(pdev, tx_ring->size,
1381                                              &tx_ring->dma);
1382
1383         if (!tx_ring->desc)
1384                 goto err;
1385
1386         tx_ring->adapter = adapter;
1387         tx_ring->next_to_use = 0;
1388         tx_ring->next_to_clean = 0;
1389         spin_lock_init(&tx_ring->tx_clean_lock);
1390         spin_lock_init(&tx_ring->tx_lock);
1391         return 0;
1392
1393 err:
1394         vfree(tx_ring->buffer_info);
1395         dev_err(&adapter->pdev->dev,
1396                 "Unable to allocate memory for the transmit descriptor ring\n");
1397         return -ENOMEM;
1398 }
1399
1400 /**
1401  * igb_setup_all_tx_resources - wrapper to allocate Tx resources
1402  *                                (Descriptors) for all queues
1403  * @adapter: board private structure
1404  *
1405  * Return 0 on success, negative on failure
1406  **/
1407 static int igb_setup_all_tx_resources(struct igb_adapter *adapter)
1408 {
1409         int i, err = 0;
1410
1411         for (i = 0; i < adapter->num_tx_queues; i++) {
1412                 err = igb_setup_tx_resources(adapter, &adapter->tx_ring[i]);
1413                 if (err) {
1414                         dev_err(&adapter->pdev->dev,
1415                                 "Allocation for Tx Queue %u failed\n", i);
1416                         for (i--; i >= 0; i--)
1417                                 igb_free_tx_resources(&adapter->tx_ring[i]);
1418                         break;
1419                 }
1420         }
1421
1422         return err;
1423 }
1424
1425 /**
1426  * igb_configure_tx - Configure transmit Unit after Reset
1427  * @adapter: board private structure
1428  *
1429  * Configure the Tx unit of the MAC after a reset.
1430  **/
1431 static void igb_configure_tx(struct igb_adapter *adapter)
1432 {
1433         u64 tdba, tdwba;
1434         struct e1000_hw *hw = &adapter->hw;
1435         u32 tctl;
1436         u32 txdctl, txctrl;
1437         int i;
1438
1439         for (i = 0; i < adapter->num_tx_queues; i++) {
1440                 struct igb_ring *ring = &(adapter->tx_ring[i]);
1441
1442                 wr32(E1000_TDLEN(i),
1443                                 ring->count * sizeof(struct e1000_tx_desc));
1444                 tdba = ring->dma;
1445                 wr32(E1000_TDBAL(i),
1446                                 tdba & 0x00000000ffffffffULL);
1447                 wr32(E1000_TDBAH(i), tdba >> 32);
1448
1449                 tdwba = ring->dma + ring->count * sizeof(struct e1000_tx_desc);
1450                 tdwba |= 1; /* enable head wb */
1451                 wr32(E1000_TDWBAL(i),
1452                                 tdwba & 0x00000000ffffffffULL);
1453                 wr32(E1000_TDWBAH(i), tdwba >> 32);
1454
1455                 ring->head = E1000_TDH(i);
1456                 ring->tail = E1000_TDT(i);
1457                 writel(0, hw->hw_addr + ring->tail);
1458                 writel(0, hw->hw_addr + ring->head);
1459                 txdctl = rd32(E1000_TXDCTL(i));
1460                 txdctl |= E1000_TXDCTL_QUEUE_ENABLE;
1461                 wr32(E1000_TXDCTL(i), txdctl);
1462
1463                 /* Turn off Relaxed Ordering on head write-backs.  The
1464                  * writebacks MUST be delivered in order or it will
1465                  * completely screw up our bookeeping.
1466                  */
1467                 txctrl = rd32(E1000_DCA_TXCTRL(i));
1468                 txctrl &= ~E1000_DCA_TXCTRL_TX_WB_RO_EN;
1469                 wr32(E1000_DCA_TXCTRL(i), txctrl);
1470         }
1471
1472
1473
1474         /* Use the default values for the Tx Inter Packet Gap (IPG) timer */
1475
1476         /* Program the Transmit Control Register */
1477
1478         tctl = rd32(E1000_TCTL);
1479         tctl &= ~E1000_TCTL_CT;
1480         tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
1481                 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
1482
1483         igb_config_collision_dist(hw);
1484
1485         /* Setup Transmit Descriptor Settings for eop descriptor */
1486         adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_RS;
1487
1488         /* Enable transmits */
1489         tctl |= E1000_TCTL_EN;
1490
1491         wr32(E1000_TCTL, tctl);
1492 }
1493
1494 /**
1495  * igb_setup_rx_resources - allocate Rx resources (Descriptors)
1496  * @adapter: board private structure
1497  * @rx_ring:    rx descriptor ring (for a specific queue) to setup
1498  *
1499  * Returns 0 on success, negative on failure
1500  **/
1501
1502 int igb_setup_rx_resources(struct igb_adapter *adapter,
1503                            struct igb_ring *rx_ring)
1504 {
1505         struct pci_dev *pdev = adapter->pdev;
1506         int size, desc_len;
1507
1508         size = sizeof(struct igb_buffer) * rx_ring->count;
1509         rx_ring->buffer_info = vmalloc(size);
1510         if (!rx_ring->buffer_info)
1511                 goto err;
1512         memset(rx_ring->buffer_info, 0, size);
1513
1514         desc_len = sizeof(union e1000_adv_rx_desc);
1515
1516         /* Round up to nearest 4K */
1517         rx_ring->size = rx_ring->count * desc_len;
1518         rx_ring->size = ALIGN(rx_ring->size, 4096);
1519
1520         rx_ring->desc = pci_alloc_consistent(pdev, rx_ring->size,
1521                                              &rx_ring->dma);
1522
1523         if (!rx_ring->desc)
1524                 goto err;
1525
1526         rx_ring->next_to_clean = 0;
1527         rx_ring->next_to_use = 0;
1528         rx_ring->pending_skb = NULL;
1529
1530         rx_ring->adapter = adapter;
1531
1532         return 0;
1533
1534 err:
1535         vfree(rx_ring->buffer_info);
1536         dev_err(&adapter->pdev->dev, "Unable to allocate memory for "
1537                 "the receive descriptor ring\n");
1538         return -ENOMEM;
1539 }
1540
1541 /**
1542  * igb_setup_all_rx_resources - wrapper to allocate Rx resources
1543  *                                (Descriptors) for all queues
1544  * @adapter: board private structure
1545  *
1546  * Return 0 on success, negative on failure
1547  **/
1548 static int igb_setup_all_rx_resources(struct igb_adapter *adapter)
1549 {
1550         int i, err = 0;
1551
1552         for (i = 0; i < adapter->num_rx_queues; i++) {
1553                 err = igb_setup_rx_resources(adapter, &adapter->rx_ring[i]);
1554                 if (err) {
1555                         dev_err(&adapter->pdev->dev,
1556                                 "Allocation for Rx Queue %u failed\n", i);
1557                         for (i--; i >= 0; i--)
1558                                 igb_free_rx_resources(&adapter->rx_ring[i]);
1559                         break;
1560                 }
1561         }
1562
1563         return err;
1564 }
1565
1566 /**
1567  * igb_setup_rctl - configure the receive control registers
1568  * @adapter: Board private structure
1569  **/
1570 static void igb_setup_rctl(struct igb_adapter *adapter)
1571 {
1572         struct e1000_hw *hw = &adapter->hw;
1573         u32 rctl;
1574         u32 srrctl = 0;
1575         int i;
1576
1577         rctl = rd32(E1000_RCTL);
1578
1579         rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
1580
1581         rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
1582                 E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
1583                 (adapter->hw.mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
1584
1585         /* disable the stripping of CRC because it breaks
1586          * BMC firmware connected over SMBUS
1587         rctl |= E1000_RCTL_SECRC;
1588         */
1589
1590         rctl &= ~E1000_RCTL_SBP;
1591
1592         if (adapter->netdev->mtu <= ETH_DATA_LEN)
1593                 rctl &= ~E1000_RCTL_LPE;
1594         else
1595                 rctl |= E1000_RCTL_LPE;
1596         if (adapter->rx_buffer_len <= IGB_RXBUFFER_2048) {
1597                 /* Setup buffer sizes */
1598                 rctl &= ~E1000_RCTL_SZ_4096;
1599                 rctl |= E1000_RCTL_BSEX;
1600                 switch (adapter->rx_buffer_len) {
1601                 case IGB_RXBUFFER_256:
1602                         rctl |= E1000_RCTL_SZ_256;
1603                         rctl &= ~E1000_RCTL_BSEX;
1604                         break;
1605                 case IGB_RXBUFFER_512:
1606                         rctl |= E1000_RCTL_SZ_512;
1607                         rctl &= ~E1000_RCTL_BSEX;
1608                         break;
1609                 case IGB_RXBUFFER_1024:
1610                         rctl |= E1000_RCTL_SZ_1024;
1611                         rctl &= ~E1000_RCTL_BSEX;
1612                         break;
1613                 case IGB_RXBUFFER_2048:
1614                 default:
1615                         rctl |= E1000_RCTL_SZ_2048;
1616                         rctl &= ~E1000_RCTL_BSEX;
1617                         break;
1618                 case IGB_RXBUFFER_4096:
1619                         rctl |= E1000_RCTL_SZ_4096;
1620                         break;
1621                 case IGB_RXBUFFER_8192:
1622                         rctl |= E1000_RCTL_SZ_8192;
1623                         break;
1624                 case IGB_RXBUFFER_16384:
1625                         rctl |= E1000_RCTL_SZ_16384;
1626                         break;
1627                 }
1628         } else {
1629                 rctl &= ~E1000_RCTL_BSEX;
1630                 srrctl = adapter->rx_buffer_len >> E1000_SRRCTL_BSIZEPKT_SHIFT;
1631         }
1632
1633         /* 82575 and greater support packet-split where the protocol
1634          * header is placed in skb->data and the packet data is
1635          * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
1636          * In the case of a non-split, skb->data is linearly filled,
1637          * followed by the page buffers.  Therefore, skb->data is
1638          * sized to hold the largest protocol header.
1639          */
1640         /* allocations using alloc_page take too long for regular MTU
1641          * so only enable packet split for jumbo frames */
1642         if (rctl & E1000_RCTL_LPE) {
1643                 adapter->rx_ps_hdr_size = IGB_RXBUFFER_128;
1644                 srrctl = adapter->rx_ps_hdr_size <<
1645                          E1000_SRRCTL_BSIZEHDRSIZE_SHIFT;
1646                 /* buffer size is ALWAYS one page */
1647                 srrctl |= PAGE_SIZE >> E1000_SRRCTL_BSIZEPKT_SHIFT;
1648                 srrctl |= E1000_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
1649         } else {
1650                 adapter->rx_ps_hdr_size = 0;
1651                 srrctl |= E1000_SRRCTL_DESCTYPE_ADV_ONEBUF;
1652         }
1653
1654         for (i = 0; i < adapter->num_rx_queues; i++)
1655                 wr32(E1000_SRRCTL(i), srrctl);
1656
1657         wr32(E1000_RCTL, rctl);
1658 }
1659
1660 /**
1661  * igb_configure_rx - Configure receive Unit after Reset
1662  * @adapter: board private structure
1663  *
1664  * Configure the Rx unit of the MAC after a reset.
1665  **/
1666 static void igb_configure_rx(struct igb_adapter *adapter)
1667 {
1668         u64 rdba;
1669         struct e1000_hw *hw = &adapter->hw;
1670         u32 rctl, rxcsum;
1671         u32 rxdctl;
1672         int i;
1673
1674         /* disable receives while setting up the descriptors */
1675         rctl = rd32(E1000_RCTL);
1676         wr32(E1000_RCTL, rctl & ~E1000_RCTL_EN);
1677         wrfl();
1678         mdelay(10);
1679
1680         if (adapter->itr_setting > 3)
1681                 wr32(E1000_ITR,
1682                                 1000000000 / (adapter->itr * 256));
1683
1684         /* Setup the HW Rx Head and Tail Descriptor Pointers and
1685          * the Base and Length of the Rx Descriptor Ring */
1686         for (i = 0; i < adapter->num_rx_queues; i++) {
1687                 struct igb_ring *ring = &(adapter->rx_ring[i]);
1688                 rdba = ring->dma;
1689                 wr32(E1000_RDBAL(i),
1690                                 rdba & 0x00000000ffffffffULL);
1691                 wr32(E1000_RDBAH(i), rdba >> 32);
1692                 wr32(E1000_RDLEN(i),
1693                                ring->count * sizeof(union e1000_adv_rx_desc));
1694
1695                 ring->head = E1000_RDH(i);
1696                 ring->tail = E1000_RDT(i);
1697                 writel(0, hw->hw_addr + ring->tail);
1698                 writel(0, hw->hw_addr + ring->head);
1699
1700                 rxdctl = rd32(E1000_RXDCTL(i));
1701                 rxdctl |= E1000_RXDCTL_QUEUE_ENABLE;
1702                 rxdctl &= 0xFFF00000;
1703                 rxdctl |= IGB_RX_PTHRESH;
1704                 rxdctl |= IGB_RX_HTHRESH << 8;
1705                 rxdctl |= IGB_RX_WTHRESH << 16;
1706                 wr32(E1000_RXDCTL(i), rxdctl);
1707         }
1708
1709         if (adapter->num_rx_queues > 1) {
1710                 u32 random[10];
1711                 u32 mrqc;
1712                 u32 j, shift;
1713                 union e1000_reta {
1714                         u32 dword;
1715                         u8  bytes[4];
1716                 } reta;
1717
1718                 get_random_bytes(&random[0], 40);
1719
1720                 shift = 6;
1721                 for (j = 0; j < (32 * 4); j++) {
1722                         reta.bytes[j & 3] =
1723                                 (j % adapter->num_rx_queues) << shift;
1724                         if ((j & 3) == 3)
1725                                 writel(reta.dword,
1726                                        hw->hw_addr + E1000_RETA(0) + (j & ~3));
1727                 }
1728                 mrqc = E1000_MRQC_ENABLE_RSS_4Q;
1729
1730                 /* Fill out hash function seeds */
1731                 for (j = 0; j < 10; j++)
1732                         array_wr32(E1000_RSSRK(0), j, random[j]);
1733
1734                 mrqc |= (E1000_MRQC_RSS_FIELD_IPV4 |
1735                          E1000_MRQC_RSS_FIELD_IPV4_TCP);
1736                 mrqc |= (E1000_MRQC_RSS_FIELD_IPV6 |
1737                          E1000_MRQC_RSS_FIELD_IPV6_TCP);
1738                 mrqc |= (E1000_MRQC_RSS_FIELD_IPV4_UDP |
1739                          E1000_MRQC_RSS_FIELD_IPV6_UDP);
1740                 mrqc |= (E1000_MRQC_RSS_FIELD_IPV6_UDP_EX |
1741                          E1000_MRQC_RSS_FIELD_IPV6_TCP_EX);
1742
1743
1744                 wr32(E1000_MRQC, mrqc);
1745
1746                 /* Multiqueue and raw packet checksumming are mutually
1747                  * exclusive.  Note that this not the same as TCP/IP
1748                  * checksumming, which works fine. */
1749                 rxcsum = rd32(E1000_RXCSUM);
1750                 rxcsum |= E1000_RXCSUM_PCSD;
1751                 wr32(E1000_RXCSUM, rxcsum);
1752         } else {
1753                 /* Enable Receive Checksum Offload for TCP and UDP */
1754                 rxcsum = rd32(E1000_RXCSUM);
1755                 if (adapter->rx_csum) {
1756                         rxcsum |= E1000_RXCSUM_TUOFL;
1757
1758                         /* Enable IPv4 payload checksum for UDP fragments
1759                          * Must be used in conjunction with packet-split. */
1760                         if (adapter->rx_ps_hdr_size)
1761                                 rxcsum |= E1000_RXCSUM_IPPCSE;
1762                 } else {
1763                         rxcsum &= ~E1000_RXCSUM_TUOFL;
1764                         /* don't need to clear IPPCSE as it defaults to 0 */
1765                 }
1766                 wr32(E1000_RXCSUM, rxcsum);
1767         }
1768
1769         if (adapter->vlgrp)
1770                 wr32(E1000_RLPML,
1771                                 adapter->max_frame_size + VLAN_TAG_SIZE);
1772         else
1773                 wr32(E1000_RLPML, adapter->max_frame_size);
1774
1775         /* Enable Receives */
1776         wr32(E1000_RCTL, rctl);
1777 }
1778
1779 /**
1780  * igb_free_tx_resources - Free Tx Resources per Queue
1781  * @adapter: board private structure
1782  * @tx_ring: Tx descriptor ring for a specific queue
1783  *
1784  * Free all transmit software resources
1785  **/
1786 static void igb_free_tx_resources(struct igb_ring *tx_ring)
1787 {
1788         struct pci_dev *pdev = tx_ring->adapter->pdev;
1789
1790         igb_clean_tx_ring(tx_ring);
1791
1792         vfree(tx_ring->buffer_info);
1793         tx_ring->buffer_info = NULL;
1794
1795         pci_free_consistent(pdev, tx_ring->size, tx_ring->desc, tx_ring->dma);
1796
1797         tx_ring->desc = NULL;
1798 }
1799
1800 /**
1801  * igb_free_all_tx_resources - Free Tx Resources for All Queues
1802  * @adapter: board private structure
1803  *
1804  * Free all transmit software resources
1805  **/
1806 static void igb_free_all_tx_resources(struct igb_adapter *adapter)
1807 {
1808         int i;
1809
1810         for (i = 0; i < adapter->num_tx_queues; i++)
1811                 igb_free_tx_resources(&adapter->tx_ring[i]);
1812 }
1813
1814 static void igb_unmap_and_free_tx_resource(struct igb_adapter *adapter,
1815                                            struct igb_buffer *buffer_info)
1816 {
1817         if (buffer_info->dma) {
1818                 pci_unmap_page(adapter->pdev,
1819                                 buffer_info->dma,
1820                                 buffer_info->length,
1821                                 PCI_DMA_TODEVICE);
1822                 buffer_info->dma = 0;
1823         }
1824         if (buffer_info->skb) {
1825                 dev_kfree_skb_any(buffer_info->skb);
1826                 buffer_info->skb = NULL;
1827         }
1828         buffer_info->time_stamp = 0;
1829         /* buffer_info must be completely set up in the transmit path */
1830 }
1831
1832 /**
1833  * igb_clean_tx_ring - Free Tx Buffers
1834  * @adapter: board private structure
1835  * @tx_ring: ring to be cleaned
1836  **/
1837 static void igb_clean_tx_ring(struct igb_ring *tx_ring)
1838 {
1839         struct igb_adapter *adapter = tx_ring->adapter;
1840         struct igb_buffer *buffer_info;
1841         unsigned long size;
1842         unsigned int i;
1843
1844         if (!tx_ring->buffer_info)
1845                 return;
1846         /* Free all the Tx ring sk_buffs */
1847
1848         for (i = 0; i < tx_ring->count; i++) {
1849                 buffer_info = &tx_ring->buffer_info[i];
1850                 igb_unmap_and_free_tx_resource(adapter, buffer_info);
1851         }
1852
1853         size = sizeof(struct igb_buffer) * tx_ring->count;
1854         memset(tx_ring->buffer_info, 0, size);
1855
1856         /* Zero out the descriptor ring */
1857
1858         memset(tx_ring->desc, 0, tx_ring->size);
1859
1860         tx_ring->next_to_use = 0;
1861         tx_ring->next_to_clean = 0;
1862
1863         writel(0, adapter->hw.hw_addr + tx_ring->head);
1864         writel(0, adapter->hw.hw_addr + tx_ring->tail);
1865 }
1866
1867 /**
1868  * igb_clean_all_tx_rings - Free Tx Buffers for all queues
1869  * @adapter: board private structure
1870  **/
1871 static void igb_clean_all_tx_rings(struct igb_adapter *adapter)
1872 {
1873         int i;
1874
1875         for (i = 0; i < adapter->num_tx_queues; i++)
1876                 igb_clean_tx_ring(&adapter->tx_ring[i]);
1877 }
1878
1879 /**
1880  * igb_free_rx_resources - Free Rx Resources
1881  * @adapter: board private structure
1882  * @rx_ring: ring to clean the resources from
1883  *
1884  * Free all receive software resources
1885  **/
1886 static void igb_free_rx_resources(struct igb_ring *rx_ring)
1887 {
1888         struct pci_dev *pdev = rx_ring->adapter->pdev;
1889
1890         igb_clean_rx_ring(rx_ring);
1891
1892         vfree(rx_ring->buffer_info);
1893         rx_ring->buffer_info = NULL;
1894
1895         pci_free_consistent(pdev, rx_ring->size, rx_ring->desc, rx_ring->dma);
1896
1897         rx_ring->desc = NULL;
1898 }
1899
1900 /**
1901  * igb_free_all_rx_resources - Free Rx Resources for All Queues
1902  * @adapter: board private structure
1903  *
1904  * Free all receive software resources
1905  **/
1906 static void igb_free_all_rx_resources(struct igb_adapter *adapter)
1907 {
1908         int i;
1909
1910         for (i = 0; i < adapter->num_rx_queues; i++)
1911                 igb_free_rx_resources(&adapter->rx_ring[i]);
1912 }
1913
1914 /**
1915  * igb_clean_rx_ring - Free Rx Buffers per Queue
1916  * @adapter: board private structure
1917  * @rx_ring: ring to free buffers from
1918  **/
1919 static void igb_clean_rx_ring(struct igb_ring *rx_ring)
1920 {
1921         struct igb_adapter *adapter = rx_ring->adapter;
1922         struct igb_buffer *buffer_info;
1923         struct pci_dev *pdev = adapter->pdev;
1924         unsigned long size;
1925         unsigned int i;
1926
1927         if (!rx_ring->buffer_info)
1928                 return;
1929         /* Free all the Rx ring sk_buffs */
1930         for (i = 0; i < rx_ring->count; i++) {
1931                 buffer_info = &rx_ring->buffer_info[i];
1932                 if (buffer_info->dma) {
1933                         if (adapter->rx_ps_hdr_size)
1934                                 pci_unmap_single(pdev, buffer_info->dma,
1935                                                  adapter->rx_ps_hdr_size,
1936                                                  PCI_DMA_FROMDEVICE);
1937                         else
1938                                 pci_unmap_single(pdev, buffer_info->dma,
1939                                                  adapter->rx_buffer_len,
1940                                                  PCI_DMA_FROMDEVICE);
1941                         buffer_info->dma = 0;
1942                 }
1943
1944                 if (buffer_info->skb) {
1945                         dev_kfree_skb(buffer_info->skb);
1946                         buffer_info->skb = NULL;
1947                 }
1948                 if (buffer_info->page) {
1949                         pci_unmap_page(pdev, buffer_info->page_dma,
1950                                        PAGE_SIZE, PCI_DMA_FROMDEVICE);
1951                         put_page(buffer_info->page);
1952                         buffer_info->page = NULL;
1953                         buffer_info->page_dma = 0;
1954                 }
1955         }
1956
1957         /* there also may be some cached data from a chained receive */
1958         if (rx_ring->pending_skb) {
1959                 dev_kfree_skb(rx_ring->pending_skb);
1960                 rx_ring->pending_skb = NULL;
1961         }
1962
1963         size = sizeof(struct igb_buffer) * rx_ring->count;
1964         memset(rx_ring->buffer_info, 0, size);
1965
1966         /* Zero out the descriptor ring */
1967         memset(rx_ring->desc, 0, rx_ring->size);
1968
1969         rx_ring->next_to_clean = 0;
1970         rx_ring->next_to_use = 0;
1971
1972         writel(0, adapter->hw.hw_addr + rx_ring->head);
1973         writel(0, adapter->hw.hw_addr + rx_ring->tail);
1974 }
1975
1976 /**
1977  * igb_clean_all_rx_rings - Free Rx Buffers for all queues
1978  * @adapter: board private structure
1979  **/
1980 static void igb_clean_all_rx_rings(struct igb_adapter *adapter)
1981 {
1982         int i;
1983
1984         for (i = 0; i < adapter->num_rx_queues; i++)
1985                 igb_clean_rx_ring(&adapter->rx_ring[i]);
1986 }
1987
1988 /**
1989  * igb_set_mac - Change the Ethernet Address of the NIC
1990  * @netdev: network interface device structure
1991  * @p: pointer to an address structure
1992  *
1993  * Returns 0 on success, negative on failure
1994  **/
1995 static int igb_set_mac(struct net_device *netdev, void *p)
1996 {
1997         struct igb_adapter *adapter = netdev_priv(netdev);
1998         struct sockaddr *addr = p;
1999
2000         if (!is_valid_ether_addr(addr->sa_data))
2001                 return -EADDRNOTAVAIL;
2002
2003         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
2004         memcpy(adapter->hw.mac.addr, addr->sa_data, netdev->addr_len);
2005
2006         adapter->hw.mac.ops.rar_set(&adapter->hw, adapter->hw.mac.addr, 0);
2007
2008         return 0;
2009 }
2010
2011 /**
2012  * igb_set_multi - Multicast and Promiscuous mode set
2013  * @netdev: network interface device structure
2014  *
2015  * The set_multi entry point is called whenever the multicast address
2016  * list or the network interface flags are updated.  This routine is
2017  * responsible for configuring the hardware for proper multicast,
2018  * promiscuous mode, and all-multi behavior.
2019  **/
2020 static void igb_set_multi(struct net_device *netdev)
2021 {
2022         struct igb_adapter *adapter = netdev_priv(netdev);
2023         struct e1000_hw *hw = &adapter->hw;
2024         struct e1000_mac_info *mac = &hw->mac;
2025         struct dev_mc_list *mc_ptr;
2026         u8  *mta_list;
2027         u32 rctl;
2028         int i;
2029
2030         /* Check for Promiscuous and All Multicast modes */
2031
2032         rctl = rd32(E1000_RCTL);
2033
2034         if (netdev->flags & IFF_PROMISC)
2035                 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
2036         else if (netdev->flags & IFF_ALLMULTI) {
2037                 rctl |= E1000_RCTL_MPE;
2038                 rctl &= ~E1000_RCTL_UPE;
2039         } else
2040                 rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
2041
2042         wr32(E1000_RCTL, rctl);
2043
2044         if (!netdev->mc_count) {
2045                 /* nothing to program, so clear mc list */
2046                 igb_update_mc_addr_list(hw, NULL, 0, 1,
2047                                           mac->rar_entry_count);
2048                 return;
2049         }
2050
2051         mta_list = kzalloc(netdev->mc_count * 6, GFP_ATOMIC);
2052         if (!mta_list)
2053                 return;
2054
2055         /* The shared function expects a packed array of only addresses. */
2056         mc_ptr = netdev->mc_list;
2057
2058         for (i = 0; i < netdev->mc_count; i++) {
2059                 if (!mc_ptr)
2060                         break;
2061                 memcpy(mta_list + (i*ETH_ALEN), mc_ptr->dmi_addr, ETH_ALEN);
2062                 mc_ptr = mc_ptr->next;
2063         }
2064         igb_update_mc_addr_list(hw, mta_list, i, 1, mac->rar_entry_count);
2065         kfree(mta_list);
2066 }
2067
2068 /* Need to wait a few seconds after link up to get diagnostic information from
2069  * the phy */
2070 static void igb_update_phy_info(unsigned long data)
2071 {
2072         struct igb_adapter *adapter = (struct igb_adapter *) data;
2073         if (adapter->hw.phy.ops.get_phy_info)
2074                 adapter->hw.phy.ops.get_phy_info(&adapter->hw);
2075 }
2076
2077 /**
2078  * igb_watchdog - Timer Call-back
2079  * @data: pointer to adapter cast into an unsigned long
2080  **/
2081 static void igb_watchdog(unsigned long data)
2082 {
2083         struct igb_adapter *adapter = (struct igb_adapter *)data;
2084         /* Do the rest outside of interrupt context */
2085         schedule_work(&adapter->watchdog_task);
2086 }
2087
2088 static void igb_watchdog_task(struct work_struct *work)
2089 {
2090         struct igb_adapter *adapter = container_of(work,
2091                                         struct igb_adapter, watchdog_task);
2092         struct e1000_hw *hw = &adapter->hw;
2093
2094         struct net_device *netdev = adapter->netdev;
2095         struct igb_ring *tx_ring = adapter->tx_ring;
2096         struct e1000_mac_info *mac = &adapter->hw.mac;
2097         u32 link;
2098         s32 ret_val;
2099
2100         if ((netif_carrier_ok(netdev)) &&
2101             (rd32(E1000_STATUS) & E1000_STATUS_LU))
2102                 goto link_up;
2103
2104         ret_val = hw->mac.ops.check_for_link(&adapter->hw);
2105         if ((ret_val == E1000_ERR_PHY) &&
2106             (hw->phy.type == e1000_phy_igp_3) &&
2107             (rd32(E1000_CTRL) &
2108              E1000_PHY_CTRL_GBE_DISABLE))
2109                 dev_info(&adapter->pdev->dev,
2110                          "Gigabit has been disabled, downgrading speed\n");
2111
2112         if ((hw->phy.media_type == e1000_media_type_internal_serdes) &&
2113             !(rd32(E1000_TXCW) & E1000_TXCW_ANE))
2114                 link = mac->serdes_has_link;
2115         else
2116                 link = rd32(E1000_STATUS) &
2117                                       E1000_STATUS_LU;
2118
2119         if (link) {
2120                 if (!netif_carrier_ok(netdev)) {
2121                         u32 ctrl;
2122                         hw->mac.ops.get_speed_and_duplex(&adapter->hw,
2123                                                    &adapter->link_speed,
2124                                                    &adapter->link_duplex);
2125
2126                         ctrl = rd32(E1000_CTRL);
2127                         dev_info(&adapter->pdev->dev,
2128                                  "NIC Link is Up %d Mbps %s, "
2129                                  "Flow Control: %s\n",
2130                                  adapter->link_speed,
2131                                  adapter->link_duplex == FULL_DUPLEX ?
2132                                  "Full Duplex" : "Half Duplex",
2133                                  ((ctrl & E1000_CTRL_TFCE) && (ctrl &
2134                                  E1000_CTRL_RFCE)) ? "RX/TX" : ((ctrl &
2135                                  E1000_CTRL_RFCE) ? "RX" : ((ctrl &
2136                                  E1000_CTRL_TFCE) ? "TX" : "None")));
2137
2138                         /* tweak tx_queue_len according to speed/duplex and
2139                          * adjust the timeout factor */
2140                         netdev->tx_queue_len = adapter->tx_queue_len;
2141                         adapter->tx_timeout_factor = 1;
2142                         switch (adapter->link_speed) {
2143                         case SPEED_10:
2144                                 netdev->tx_queue_len = 10;
2145                                 adapter->tx_timeout_factor = 14;
2146                                 break;
2147                         case SPEED_100:
2148                                 netdev->tx_queue_len = 100;
2149                                 /* maybe add some timeout factor ? */
2150                                 break;
2151                         }
2152
2153                         netif_carrier_on(netdev);
2154                         netif_wake_queue(netdev);
2155
2156                         if (!test_bit(__IGB_DOWN, &adapter->state))
2157                                 mod_timer(&adapter->phy_info_timer,
2158                                           round_jiffies(jiffies + 2 * HZ));
2159                 }
2160         } else {
2161                 if (netif_carrier_ok(netdev)) {
2162                         adapter->link_speed = 0;
2163                         adapter->link_duplex = 0;
2164                         dev_info(&adapter->pdev->dev, "NIC Link is Down\n");
2165                         netif_carrier_off(netdev);
2166                         netif_stop_queue(netdev);
2167                         if (!test_bit(__IGB_DOWN, &adapter->state))
2168                                 mod_timer(&adapter->phy_info_timer,
2169                                           round_jiffies(jiffies + 2 * HZ));
2170                 }
2171         }
2172
2173 link_up:
2174         igb_update_stats(adapter);
2175
2176         mac->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
2177         adapter->tpt_old = adapter->stats.tpt;
2178         mac->collision_delta = adapter->stats.colc - adapter->colc_old;
2179         adapter->colc_old = adapter->stats.colc;
2180
2181         adapter->gorc = adapter->stats.gorc - adapter->gorc_old;
2182         adapter->gorc_old = adapter->stats.gorc;
2183         adapter->gotc = adapter->stats.gotc - adapter->gotc_old;
2184         adapter->gotc_old = adapter->stats.gotc;
2185
2186         igb_update_adaptive(&adapter->hw);
2187
2188         if (!netif_carrier_ok(netdev)) {
2189                 if (IGB_DESC_UNUSED(tx_ring) + 1 < tx_ring->count) {
2190                         /* We've lost link, so the controller stops DMA,
2191                          * but we've got queued Tx work that's never going
2192                          * to get done, so reset controller to flush Tx.
2193                          * (Do the reset outside of interrupt context). */
2194                         adapter->tx_timeout_count++;
2195                         schedule_work(&adapter->reset_task);
2196                 }
2197         }
2198
2199         /* Cause software interrupt to ensure rx ring is cleaned */
2200         wr32(E1000_ICS, E1000_ICS_RXDMT0);
2201
2202         /* Force detection of hung controller every watchdog period */
2203         tx_ring->detect_tx_hung = true;
2204
2205         /* Reset the timer */
2206         if (!test_bit(__IGB_DOWN, &adapter->state))
2207                 mod_timer(&adapter->watchdog_timer,
2208                           round_jiffies(jiffies + 2 * HZ));
2209 }
2210
2211 enum latency_range {
2212         lowest_latency = 0,
2213         low_latency = 1,
2214         bulk_latency = 2,
2215         latency_invalid = 255
2216 };
2217
2218
2219 static void igb_lower_rx_eitr(struct igb_adapter *adapter,
2220                               struct igb_ring *rx_ring)
2221 {
2222         struct e1000_hw *hw = &adapter->hw;
2223         int new_val;
2224
2225         new_val = rx_ring->itr_val / 2;
2226         if (new_val < IGB_MIN_DYN_ITR)
2227                 new_val = IGB_MIN_DYN_ITR;
2228
2229         if (new_val != rx_ring->itr_val) {
2230                 rx_ring->itr_val = new_val;
2231                 wr32(rx_ring->itr_register,
2232                                 1000000000 / (new_val * 256));
2233         }
2234 }
2235
2236 static void igb_raise_rx_eitr(struct igb_adapter *adapter,
2237                               struct igb_ring *rx_ring)
2238 {
2239         struct e1000_hw *hw = &adapter->hw;
2240         int new_val;
2241
2242         new_val = rx_ring->itr_val * 2;
2243         if (new_val > IGB_MAX_DYN_ITR)
2244                 new_val = IGB_MAX_DYN_ITR;
2245
2246         if (new_val != rx_ring->itr_val) {
2247                 rx_ring->itr_val = new_val;
2248                 wr32(rx_ring->itr_register,
2249                                 1000000000 / (new_val * 256));
2250         }
2251 }
2252
2253 /**
2254  * igb_update_itr - update the dynamic ITR value based on statistics
2255  *      Stores a new ITR value based on packets and byte
2256  *      counts during the last interrupt.  The advantage of per interrupt
2257  *      computation is faster updates and more accurate ITR for the current
2258  *      traffic pattern.  Constants in this function were computed
2259  *      based on theoretical maximum wire speed and thresholds were set based
2260  *      on testing data as well as attempting to minimize response time
2261  *      while increasing bulk throughput.
2262  *      this functionality is controlled by the InterruptThrottleRate module
2263  *      parameter (see igb_param.c)
2264  *      NOTE:  These calculations are only valid when operating in a single-
2265  *             queue environment.
2266  * @adapter: pointer to adapter
2267  * @itr_setting: current adapter->itr
2268  * @packets: the number of packets during this measurement interval
2269  * @bytes: the number of bytes during this measurement interval
2270  **/
2271 static unsigned int igb_update_itr(struct igb_adapter *adapter, u16 itr_setting,
2272                                    int packets, int bytes)
2273 {
2274         unsigned int retval = itr_setting;
2275
2276         if (packets == 0)
2277                 goto update_itr_done;
2278
2279         switch (itr_setting) {
2280         case lowest_latency:
2281                 /* handle TSO and jumbo frames */
2282                 if (bytes/packets > 8000)
2283                         retval = bulk_latency;
2284                 else if ((packets < 5) && (bytes > 512))
2285                         retval = low_latency;
2286                 break;
2287         case low_latency:  /* 50 usec aka 20000 ints/s */
2288                 if (bytes > 10000) {
2289                         /* this if handles the TSO accounting */
2290                         if (bytes/packets > 8000) {
2291                                 retval = bulk_latency;
2292                         } else if ((packets < 10) || ((bytes/packets) > 1200)) {
2293                                 retval = bulk_latency;
2294                         } else if ((packets > 35)) {
2295                                 retval = lowest_latency;
2296                         }
2297                 } else if (bytes/packets > 2000) {
2298                         retval = bulk_latency;
2299                 } else if (packets <= 2 && bytes < 512) {
2300                         retval = lowest_latency;
2301                 }
2302                 break;
2303         case bulk_latency: /* 250 usec aka 4000 ints/s */
2304                 if (bytes > 25000) {
2305                         if (packets > 35)
2306                                 retval = low_latency;
2307                 } else if (bytes < 6000) {
2308                         retval = low_latency;
2309                 }
2310                 break;
2311         }
2312
2313 update_itr_done:
2314         return retval;
2315 }
2316
2317 static void igb_set_itr(struct igb_adapter *adapter, u16 itr_register,
2318                         int rx_only)
2319 {
2320         u16 current_itr;
2321         u32 new_itr = adapter->itr;
2322
2323         /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
2324         if (adapter->link_speed != SPEED_1000) {
2325                 current_itr = 0;
2326                 new_itr = 4000;
2327                 goto set_itr_now;
2328         }
2329
2330         adapter->rx_itr = igb_update_itr(adapter,
2331                                     adapter->rx_itr,
2332                                     adapter->rx_ring->total_packets,
2333                                     adapter->rx_ring->total_bytes);
2334         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2335         if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
2336                 adapter->rx_itr = low_latency;
2337
2338         if (!rx_only) {
2339                 adapter->tx_itr = igb_update_itr(adapter,
2340                                             adapter->tx_itr,
2341                                             adapter->tx_ring->total_packets,
2342                                             adapter->tx_ring->total_bytes);
2343                 /* conservative mode (itr 3) eliminates the
2344                  * lowest_latency setting */
2345                 if (adapter->itr_setting == 3 &&
2346                     adapter->tx_itr == lowest_latency)
2347                         adapter->tx_itr = low_latency;
2348
2349                 current_itr = max(adapter->rx_itr, adapter->tx_itr);
2350         } else {
2351                 current_itr = adapter->rx_itr;
2352         }
2353
2354         switch (current_itr) {
2355         /* counts and packets in update_itr are dependent on these numbers */
2356         case lowest_latency:
2357                 new_itr = 70000;
2358                 break;
2359         case low_latency:
2360                 new_itr = 20000; /* aka hwitr = ~200 */
2361                 break;
2362         case bulk_latency:
2363                 new_itr = 4000;
2364                 break;
2365         default:
2366                 break;
2367         }
2368
2369 set_itr_now:
2370         if (new_itr != adapter->itr) {
2371                 /* this attempts to bias the interrupt rate towards Bulk
2372                  * by adding intermediate steps when interrupt rate is
2373                  * increasing */
2374                 new_itr = new_itr > adapter->itr ?
2375                              min(adapter->itr + (new_itr >> 2), new_itr) :
2376                              new_itr;
2377                 /* Don't write the value here; it resets the adapter's
2378                  * internal timer, and causes us to delay far longer than
2379                  * we should between interrupts.  Instead, we write the ITR
2380                  * value at the beginning of the next interrupt so the timing
2381                  * ends up being correct.
2382                  */
2383                 adapter->itr = new_itr;
2384                 adapter->set_itr = 1;
2385         }
2386
2387         return;
2388 }
2389
2390
2391 #define IGB_TX_FLAGS_CSUM               0x00000001
2392 #define IGB_TX_FLAGS_VLAN               0x00000002
2393 #define IGB_TX_FLAGS_TSO                0x00000004
2394 #define IGB_TX_FLAGS_IPV4               0x00000008
2395 #define IGB_TX_FLAGS_VLAN_MASK  0xffff0000
2396 #define IGB_TX_FLAGS_VLAN_SHIFT 16
2397
2398 static inline int igb_tso_adv(struct igb_adapter *adapter,
2399                               struct igb_ring *tx_ring,
2400                               struct sk_buff *skb, u32 tx_flags, u8 *hdr_len)
2401 {
2402         struct e1000_adv_tx_context_desc *context_desc;
2403         unsigned int i;
2404         int err;
2405         struct igb_buffer *buffer_info;
2406         u32 info = 0, tu_cmd = 0;
2407         u32 mss_l4len_idx, l4len;
2408         *hdr_len = 0;
2409
2410         if (skb_header_cloned(skb)) {
2411                 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2412                 if (err)
2413                         return err;
2414         }
2415
2416         l4len = tcp_hdrlen(skb);
2417         *hdr_len += l4len;
2418
2419         if (skb->protocol == htons(ETH_P_IP)) {
2420                 struct iphdr *iph = ip_hdr(skb);
2421                 iph->tot_len = 0;
2422                 iph->check = 0;
2423                 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
2424                                                          iph->daddr, 0,
2425                                                          IPPROTO_TCP,
2426                                                          0);
2427         } else if (skb_shinfo(skb)->gso_type == SKB_GSO_TCPV6) {
2428                 ipv6_hdr(skb)->payload_len = 0;
2429                 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
2430                                                        &ipv6_hdr(skb)->daddr,
2431                                                        0, IPPROTO_TCP, 0);
2432         }
2433
2434         i = tx_ring->next_to_use;
2435
2436         buffer_info = &tx_ring->buffer_info[i];
2437         context_desc = E1000_TX_CTXTDESC_ADV(*tx_ring, i);
2438         /* VLAN MACLEN IPLEN */
2439         if (tx_flags & IGB_TX_FLAGS_VLAN)
2440                 info |= (tx_flags & IGB_TX_FLAGS_VLAN_MASK);
2441         info |= (skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT);
2442         *hdr_len += skb_network_offset(skb);
2443         info |= skb_network_header_len(skb);
2444         *hdr_len += skb_network_header_len(skb);
2445         context_desc->vlan_macip_lens = cpu_to_le32(info);
2446
2447         /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
2448         tu_cmd |= (E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT);
2449
2450         if (skb->protocol == htons(ETH_P_IP))
2451                 tu_cmd |= E1000_ADVTXD_TUCMD_IPV4;
2452         tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
2453
2454         context_desc->type_tucmd_mlhl = cpu_to_le32(tu_cmd);
2455
2456         /* MSS L4LEN IDX */
2457         mss_l4len_idx = (skb_shinfo(skb)->gso_size << E1000_ADVTXD_MSS_SHIFT);
2458         mss_l4len_idx |= (l4len << E1000_ADVTXD_L4LEN_SHIFT);
2459
2460         /* Context index must be unique per ring.  Luckily, so is the interrupt
2461          * mask value. */
2462         mss_l4len_idx |= tx_ring->eims_value >> 4;
2463
2464         context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
2465         context_desc->seqnum_seed = 0;
2466
2467         buffer_info->time_stamp = jiffies;
2468         buffer_info->dma = 0;
2469         i++;
2470         if (i == tx_ring->count)
2471                 i = 0;
2472
2473         tx_ring->next_to_use = i;
2474
2475         return true;
2476 }
2477
2478 static inline bool igb_tx_csum_adv(struct igb_adapter *adapter,
2479                                         struct igb_ring *tx_ring,
2480                                         struct sk_buff *skb, u32 tx_flags)
2481 {
2482         struct e1000_adv_tx_context_desc *context_desc;
2483         unsigned int i;
2484         struct igb_buffer *buffer_info;
2485         u32 info = 0, tu_cmd = 0;
2486
2487         if ((skb->ip_summed == CHECKSUM_PARTIAL) ||
2488             (tx_flags & IGB_TX_FLAGS_VLAN)) {
2489                 i = tx_ring->next_to_use;
2490                 buffer_info = &tx_ring->buffer_info[i];
2491                 context_desc = E1000_TX_CTXTDESC_ADV(*tx_ring, i);
2492
2493                 if (tx_flags & IGB_TX_FLAGS_VLAN)
2494                         info |= (tx_flags & IGB_TX_FLAGS_VLAN_MASK);
2495                 info |= (skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT);
2496                 if (skb->ip_summed == CHECKSUM_PARTIAL)
2497                         info |= skb_network_header_len(skb);
2498
2499                 context_desc->vlan_macip_lens = cpu_to_le32(info);
2500
2501                 tu_cmd |= (E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT);
2502
2503                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2504                         switch (skb->protocol) {
2505                         case __constant_htons(ETH_P_IP):
2506                                 tu_cmd |= E1000_ADVTXD_TUCMD_IPV4;
2507                                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
2508                                         tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
2509                                 break;
2510                         case __constant_htons(ETH_P_IPV6):
2511                                 /* XXX what about other V6 headers?? */
2512                                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
2513                                         tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
2514                                 break;
2515                         default:
2516                                 if (unlikely(net_ratelimit()))
2517                                         dev_warn(&adapter->pdev->dev,
2518                                             "partial checksum but proto=%x!\n",
2519                                             skb->protocol);
2520                                 break;
2521                         }
2522                 }
2523
2524                 context_desc->type_tucmd_mlhl = cpu_to_le32(tu_cmd);
2525                 context_desc->seqnum_seed = 0;
2526                 context_desc->mss_l4len_idx =
2527                                           cpu_to_le32(tx_ring->eims_value >> 4);
2528
2529                 buffer_info->time_stamp = jiffies;
2530                 buffer_info->dma = 0;
2531
2532                 i++;
2533                 if (i == tx_ring->count)
2534                         i = 0;
2535                 tx_ring->next_to_use = i;
2536
2537                 return true;
2538         }
2539
2540
2541         return false;
2542 }
2543
2544 #define IGB_MAX_TXD_PWR 16
2545 #define IGB_MAX_DATA_PER_TXD    (1<<IGB_MAX_TXD_PWR)
2546
2547 static inline int igb_tx_map_adv(struct igb_adapter *adapter,
2548                                  struct igb_ring *tx_ring,
2549                                  struct sk_buff *skb)
2550 {
2551         struct igb_buffer *buffer_info;
2552         unsigned int len = skb_headlen(skb);
2553         unsigned int count = 0, i;
2554         unsigned int f;
2555
2556         i = tx_ring->next_to_use;
2557
2558         buffer_info = &tx_ring->buffer_info[i];
2559         BUG_ON(len >= IGB_MAX_DATA_PER_TXD);
2560         buffer_info->length = len;
2561         /* set time_stamp *before* dma to help avoid a possible race */
2562         buffer_info->time_stamp = jiffies;
2563         buffer_info->dma = pci_map_single(adapter->pdev, skb->data, len,
2564                                           PCI_DMA_TODEVICE);
2565         count++;
2566         i++;
2567         if (i == tx_ring->count)
2568                 i = 0;
2569
2570         for (f = 0; f < skb_shinfo(skb)->nr_frags; f++) {
2571                 struct skb_frag_struct *frag;
2572
2573                 frag = &skb_shinfo(skb)->frags[f];
2574                 len = frag->size;
2575
2576                 buffer_info = &tx_ring->buffer_info[i];
2577                 BUG_ON(len >= IGB_MAX_DATA_PER_TXD);
2578                 buffer_info->length = len;
2579                 buffer_info->time_stamp = jiffies;
2580                 buffer_info->dma = pci_map_page(adapter->pdev,
2581                                                 frag->page,
2582                                                 frag->page_offset,
2583                                                 len,
2584                                                 PCI_DMA_TODEVICE);
2585
2586                 count++;
2587                 i++;
2588                 if (i == tx_ring->count)
2589                         i = 0;
2590         }
2591
2592         i = (i == 0) ? tx_ring->count - 1 : i - 1;
2593         tx_ring->buffer_info[i].skb = skb;
2594
2595         return count;
2596 }
2597
2598 static inline void igb_tx_queue_adv(struct igb_adapter *adapter,
2599                                     struct igb_ring *tx_ring,
2600                                     int tx_flags, int count, u32 paylen,
2601                                     u8 hdr_len)
2602 {
2603         union e1000_adv_tx_desc *tx_desc = NULL;
2604         struct igb_buffer *buffer_info;
2605         u32 olinfo_status = 0, cmd_type_len;
2606         unsigned int i;
2607
2608         cmd_type_len = (E1000_ADVTXD_DTYP_DATA | E1000_ADVTXD_DCMD_IFCS |
2609                         E1000_ADVTXD_DCMD_DEXT);
2610
2611         if (tx_flags & IGB_TX_FLAGS_VLAN)
2612                 cmd_type_len |= E1000_ADVTXD_DCMD_VLE;
2613
2614         if (tx_flags & IGB_TX_FLAGS_TSO) {
2615                 cmd_type_len |= E1000_ADVTXD_DCMD_TSE;
2616
2617                 /* insert tcp checksum */
2618                 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
2619
2620                 /* insert ip checksum */
2621                 if (tx_flags & IGB_TX_FLAGS_IPV4)
2622                         olinfo_status |= E1000_TXD_POPTS_IXSM << 8;
2623
2624         } else if (tx_flags & IGB_TX_FLAGS_CSUM) {
2625                 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
2626         }
2627
2628         if (tx_flags & (IGB_TX_FLAGS_CSUM | IGB_TX_FLAGS_TSO |
2629                         IGB_TX_FLAGS_VLAN))
2630                 olinfo_status |= tx_ring->eims_value >> 4;
2631
2632         olinfo_status |= ((paylen - hdr_len) << E1000_ADVTXD_PAYLEN_SHIFT);
2633
2634         i = tx_ring->next_to_use;
2635         while (count--) {
2636                 buffer_info = &tx_ring->buffer_info[i];
2637                 tx_desc = E1000_TX_DESC_ADV(*tx_ring, i);
2638                 tx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma);
2639                 tx_desc->read.cmd_type_len =
2640                         cpu_to_le32(cmd_type_len | buffer_info->length);
2641                 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
2642                 i++;
2643                 if (i == tx_ring->count)
2644                         i = 0;
2645         }
2646
2647         tx_desc->read.cmd_type_len |= cpu_to_le32(adapter->txd_cmd);
2648         /* Force memory writes to complete before letting h/w
2649          * know there are new descriptors to fetch.  (Only
2650          * applicable for weak-ordered memory model archs,
2651          * such as IA-64). */
2652         wmb();
2653
2654         tx_ring->next_to_use = i;
2655         writel(i, adapter->hw.hw_addr + tx_ring->tail);
2656         /* we need this if more than one processor can write to our tail
2657          * at a time, it syncronizes IO on IA64/Altix systems */
2658         mmiowb();
2659 }
2660
2661 static int __igb_maybe_stop_tx(struct net_device *netdev,
2662                                struct igb_ring *tx_ring, int size)
2663 {
2664         struct igb_adapter *adapter = netdev_priv(netdev);
2665
2666         netif_stop_queue(netdev);
2667         /* Herbert's original patch had:
2668          *  smp_mb__after_netif_stop_queue();
2669          * but since that doesn't exist yet, just open code it. */
2670         smp_mb();
2671
2672         /* We need to check again in a case another CPU has just
2673          * made room available. */
2674         if (IGB_DESC_UNUSED(tx_ring) < size)
2675                 return -EBUSY;
2676
2677         /* A reprieve! */
2678         netif_start_queue(netdev);
2679         ++adapter->restart_queue;
2680         return 0;
2681 }
2682
2683 static int igb_maybe_stop_tx(struct net_device *netdev,
2684                              struct igb_ring *tx_ring, int size)
2685 {
2686         if (IGB_DESC_UNUSED(tx_ring) >= size)
2687                 return 0;
2688         return __igb_maybe_stop_tx(netdev, tx_ring, size);
2689 }
2690
2691 #define TXD_USE_COUNT(S) (((S) >> (IGB_MAX_TXD_PWR)) + 1)
2692
2693 static int igb_xmit_frame_ring_adv(struct sk_buff *skb,
2694                                    struct net_device *netdev,
2695                                    struct igb_ring *tx_ring)
2696 {
2697         struct igb_adapter *adapter = netdev_priv(netdev);
2698         unsigned int tx_flags = 0;
2699         unsigned int len;
2700         unsigned long irq_flags;
2701         u8 hdr_len = 0;
2702         int tso = 0;
2703
2704         len = skb_headlen(skb);
2705
2706         if (test_bit(__IGB_DOWN, &adapter->state)) {
2707                 dev_kfree_skb_any(skb);
2708                 return NETDEV_TX_OK;
2709         }
2710
2711         if (skb->len <= 0) {
2712                 dev_kfree_skb_any(skb);
2713                 return NETDEV_TX_OK;
2714         }
2715
2716         if (!spin_trylock_irqsave(&tx_ring->tx_lock, irq_flags))
2717                 /* Collision - tell upper layer to requeue */
2718                 return NETDEV_TX_LOCKED;
2719
2720         /* need: 1 descriptor per page,
2721          *       + 2 desc gap to keep tail from touching head,
2722          *       + 1 desc for skb->data,
2723          *       + 1 desc for context descriptor,
2724          * otherwise try next time */
2725         if (igb_maybe_stop_tx(netdev, tx_ring, skb_shinfo(skb)->nr_frags + 4)) {
2726                 /* this is a hard error */
2727                 spin_unlock_irqrestore(&tx_ring->tx_lock, irq_flags);
2728                 return NETDEV_TX_BUSY;
2729         }
2730
2731         if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
2732                 tx_flags |= IGB_TX_FLAGS_VLAN;
2733                 tx_flags |= (vlan_tx_tag_get(skb) << IGB_TX_FLAGS_VLAN_SHIFT);
2734         }
2735
2736         tso = skb_is_gso(skb) ? igb_tso_adv(adapter, tx_ring, skb, tx_flags,
2737                                               &hdr_len) : 0;
2738
2739         if (tso < 0) {
2740                 dev_kfree_skb_any(skb);
2741                 spin_unlock_irqrestore(&tx_ring->tx_lock, irq_flags);
2742                 return NETDEV_TX_OK;
2743         }
2744
2745         if (tso)
2746                 tx_flags |= IGB_TX_FLAGS_TSO;
2747         else if (igb_tx_csum_adv(adapter, tx_ring, skb, tx_flags))
2748                         if (skb->ip_summed == CHECKSUM_PARTIAL)
2749                                 tx_flags |= IGB_TX_FLAGS_CSUM;
2750
2751         if (skb->protocol == htons(ETH_P_IP))
2752                 tx_flags |= IGB_TX_FLAGS_IPV4;
2753
2754         igb_tx_queue_adv(adapter, tx_ring, tx_flags,
2755                          igb_tx_map_adv(adapter, tx_ring, skb),
2756                          skb->len, hdr_len);
2757
2758         netdev->trans_start = jiffies;
2759
2760         /* Make sure there is space in the ring for the next send. */
2761         igb_maybe_stop_tx(netdev, tx_ring, MAX_SKB_FRAGS + 4);
2762
2763         spin_unlock_irqrestore(&tx_ring->tx_lock, irq_flags);
2764         return NETDEV_TX_OK;
2765 }
2766
2767 static int igb_xmit_frame_adv(struct sk_buff *skb, struct net_device *netdev)
2768 {
2769         struct igb_adapter *adapter = netdev_priv(netdev);
2770         struct igb_ring *tx_ring = &adapter->tx_ring[0];
2771
2772         /* This goes back to the question of how to logically map a tx queue
2773          * to a flow.  Right now, performance is impacted slightly negatively
2774          * if using multiple tx queues.  If the stack breaks away from a
2775          * single qdisc implementation, we can look at this again. */
2776         return (igb_xmit_frame_ring_adv(skb, netdev, tx_ring));
2777 }
2778
2779 /**
2780  * igb_tx_timeout - Respond to a Tx Hang
2781  * @netdev: network interface device structure
2782  **/
2783 static void igb_tx_timeout(struct net_device *netdev)
2784 {
2785         struct igb_adapter *adapter = netdev_priv(netdev);
2786         struct e1000_hw *hw = &adapter->hw;
2787
2788         /* Do the reset outside of interrupt context */
2789         adapter->tx_timeout_count++;
2790         schedule_work(&adapter->reset_task);
2791         wr32(E1000_EICS, adapter->eims_enable_mask &
2792                 ~(E1000_EIMS_TCP_TIMER | E1000_EIMS_OTHER));
2793 }
2794
2795 static void igb_reset_task(struct work_struct *work)
2796 {
2797         struct igb_adapter *adapter;
2798         adapter = container_of(work, struct igb_adapter, reset_task);
2799
2800         igb_reinit_locked(adapter);
2801 }
2802
2803 /**
2804  * igb_get_stats - Get System Network Statistics
2805  * @netdev: network interface device structure
2806  *
2807  * Returns the address of the device statistics structure.
2808  * The statistics are actually updated from the timer callback.
2809  **/
2810 static struct net_device_stats *
2811 igb_get_stats(struct net_device *netdev)
2812 {
2813         struct igb_adapter *adapter = netdev_priv(netdev);
2814
2815         /* only return the current stats */
2816         return &adapter->net_stats;
2817 }
2818
2819 /**
2820  * igb_change_mtu - Change the Maximum Transfer Unit
2821  * @netdev: network interface device structure
2822  * @new_mtu: new value for maximum frame size
2823  *
2824  * Returns 0 on success, negative on failure
2825  **/
2826 static int igb_change_mtu(struct net_device *netdev, int new_mtu)
2827 {
2828         struct igb_adapter *adapter = netdev_priv(netdev);
2829         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
2830
2831         if ((max_frame < ETH_ZLEN + ETH_FCS_LEN) ||
2832             (max_frame > MAX_JUMBO_FRAME_SIZE)) {
2833                 dev_err(&adapter->pdev->dev, "Invalid MTU setting\n");
2834                 return -EINVAL;
2835         }
2836
2837 #define MAX_STD_JUMBO_FRAME_SIZE 9234
2838         if (max_frame > MAX_STD_JUMBO_FRAME_SIZE) {
2839                 dev_err(&adapter->pdev->dev, "MTU > 9216 not supported.\n");
2840                 return -EINVAL;
2841         }
2842
2843         while (test_and_set_bit(__IGB_RESETTING, &adapter->state))
2844                 msleep(1);
2845         /* igb_down has a dependency on max_frame_size */
2846         adapter->max_frame_size = max_frame;
2847         if (netif_running(netdev))
2848                 igb_down(adapter);
2849
2850         /* NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
2851          * means we reserve 2 more, this pushes us to allocate from the next
2852          * larger slab size.
2853          * i.e. RXBUFFER_2048 --> size-4096 slab
2854          */
2855
2856         if (max_frame <= IGB_RXBUFFER_256)
2857                 adapter->rx_buffer_len = IGB_RXBUFFER_256;
2858         else if (max_frame <= IGB_RXBUFFER_512)
2859                 adapter->rx_buffer_len = IGB_RXBUFFER_512;
2860         else if (max_frame <= IGB_RXBUFFER_1024)
2861                 adapter->rx_buffer_len = IGB_RXBUFFER_1024;
2862         else if (max_frame <= IGB_RXBUFFER_2048)
2863                 adapter->rx_buffer_len = IGB_RXBUFFER_2048;
2864         else
2865                 adapter->rx_buffer_len = IGB_RXBUFFER_4096;
2866         /* adjust allocation if LPE protects us, and we aren't using SBP */
2867         if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
2868              (max_frame == MAXIMUM_ETHERNET_VLAN_SIZE))
2869                 adapter->rx_buffer_len = MAXIMUM_ETHERNET_VLAN_SIZE;
2870
2871         dev_info(&adapter->pdev->dev, "changing MTU from %d to %d\n",
2872                  netdev->mtu, new_mtu);
2873         netdev->mtu = new_mtu;
2874
2875         if (netif_running(netdev))
2876                 igb_up(adapter);
2877         else
2878                 igb_reset(adapter);
2879
2880         clear_bit(__IGB_RESETTING, &adapter->state);
2881
2882         return 0;
2883 }
2884
2885 /**
2886  * igb_update_stats - Update the board statistics counters
2887  * @adapter: board private structure
2888  **/
2889
2890 void igb_update_stats(struct igb_adapter *adapter)
2891 {
2892         struct e1000_hw *hw = &adapter->hw;
2893         struct pci_dev *pdev = adapter->pdev;
2894         u16 phy_tmp;
2895
2896 #define PHY_IDLE_ERROR_COUNT_MASK 0x00FF
2897
2898         /*
2899          * Prevent stats update while adapter is being reset, or if the pci
2900          * connection is down.
2901          */
2902         if (adapter->link_speed == 0)
2903                 return;
2904         if (pci_channel_offline(pdev))
2905                 return;
2906
2907         adapter->stats.crcerrs += rd32(E1000_CRCERRS);
2908         adapter->stats.gprc += rd32(E1000_GPRC);
2909         adapter->stats.gorc += rd32(E1000_GORCL);
2910         rd32(E1000_GORCH); /* clear GORCL */
2911         adapter->stats.bprc += rd32(E1000_BPRC);
2912         adapter->stats.mprc += rd32(E1000_MPRC);
2913         adapter->stats.roc += rd32(E1000_ROC);
2914
2915         adapter->stats.prc64 += rd32(E1000_PRC64);
2916         adapter->stats.prc127 += rd32(E1000_PRC127);
2917         adapter->stats.prc255 += rd32(E1000_PRC255);
2918         adapter->stats.prc511 += rd32(E1000_PRC511);
2919         adapter->stats.prc1023 += rd32(E1000_PRC1023);
2920         adapter->stats.prc1522 += rd32(E1000_PRC1522);
2921         adapter->stats.symerrs += rd32(E1000_SYMERRS);
2922         adapter->stats.sec += rd32(E1000_SEC);
2923
2924         adapter->stats.mpc += rd32(E1000_MPC);
2925         adapter->stats.scc += rd32(E1000_SCC);
2926         adapter->stats.ecol += rd32(E1000_ECOL);
2927         adapter->stats.mcc += rd32(E1000_MCC);
2928         adapter->stats.latecol += rd32(E1000_LATECOL);
2929         adapter->stats.dc += rd32(E1000_DC);
2930         adapter->stats.rlec += rd32(E1000_RLEC);
2931         adapter->stats.xonrxc += rd32(E1000_XONRXC);
2932         adapter->stats.xontxc += rd32(E1000_XONTXC);
2933         adapter->stats.xoffrxc += rd32(E1000_XOFFRXC);
2934         adapter->stats.xofftxc += rd32(E1000_XOFFTXC);
2935         adapter->stats.fcruc += rd32(E1000_FCRUC);
2936         adapter->stats.gptc += rd32(E1000_GPTC);
2937         adapter->stats.gotc += rd32(E1000_GOTCL);
2938         rd32(E1000_GOTCH); /* clear GOTCL */
2939         adapter->stats.rnbc += rd32(E1000_RNBC);
2940         adapter->stats.ruc += rd32(E1000_RUC);
2941         adapter->stats.rfc += rd32(E1000_RFC);
2942         adapter->stats.rjc += rd32(E1000_RJC);
2943         adapter->stats.tor += rd32(E1000_TORH);
2944         adapter->stats.tot += rd32(E1000_TOTH);
2945         adapter->stats.tpr += rd32(E1000_TPR);
2946
2947         adapter->stats.ptc64 += rd32(E1000_PTC64);
2948         adapter->stats.ptc127 += rd32(E1000_PTC127);
2949         adapter->stats.ptc255 += rd32(E1000_PTC255);
2950         adapter->stats.ptc511 += rd32(E1000_PTC511);
2951         adapter->stats.ptc1023 += rd32(E1000_PTC1023);
2952         adapter->stats.ptc1522 += rd32(E1000_PTC1522);
2953
2954         adapter->stats.mptc += rd32(E1000_MPTC);
2955         adapter->stats.bptc += rd32(E1000_BPTC);
2956
2957         /* used for adaptive IFS */
2958
2959         hw->mac.tx_packet_delta = rd32(E1000_TPT);
2960         adapter->stats.tpt += hw->mac.tx_packet_delta;
2961         hw->mac.collision_delta = rd32(E1000_COLC);
2962         adapter->stats.colc += hw->mac.collision_delta;
2963
2964         adapter->stats.algnerrc += rd32(E1000_ALGNERRC);
2965         adapter->stats.rxerrc += rd32(E1000_RXERRC);
2966         adapter->stats.tncrs += rd32(E1000_TNCRS);
2967         adapter->stats.tsctc += rd32(E1000_TSCTC);
2968         adapter->stats.tsctfc += rd32(E1000_TSCTFC);
2969
2970         adapter->stats.iac += rd32(E1000_IAC);
2971         adapter->stats.icrxoc += rd32(E1000_ICRXOC);
2972         adapter->stats.icrxptc += rd32(E1000_ICRXPTC);
2973         adapter->stats.icrxatc += rd32(E1000_ICRXATC);
2974         adapter->stats.ictxptc += rd32(E1000_ICTXPTC);
2975         adapter->stats.ictxatc += rd32(E1000_ICTXATC);
2976         adapter->stats.ictxqec += rd32(E1000_ICTXQEC);
2977         adapter->stats.ictxqmtc += rd32(E1000_ICTXQMTC);
2978         adapter->stats.icrxdmtc += rd32(E1000_ICRXDMTC);
2979
2980         /* Fill out the OS statistics structure */
2981         adapter->net_stats.multicast = adapter->stats.mprc;
2982         adapter->net_stats.collisions = adapter->stats.colc;
2983
2984         /* Rx Errors */
2985
2986         /* RLEC on some newer hardware can be incorrect so build
2987         * our own version based on RUC and ROC */
2988         adapter->net_stats.rx_errors = adapter->stats.rxerrc +
2989                 adapter->stats.crcerrs + adapter->stats.algnerrc +
2990                 adapter->stats.ruc + adapter->stats.roc +
2991                 adapter->stats.cexterr;
2992         adapter->net_stats.rx_length_errors = adapter->stats.ruc +
2993                                               adapter->stats.roc;
2994         adapter->net_stats.rx_crc_errors = adapter->stats.crcerrs;
2995         adapter->net_stats.rx_frame_errors = adapter->stats.algnerrc;
2996         adapter->net_stats.rx_missed_errors = adapter->stats.mpc;
2997
2998         /* Tx Errors */
2999         adapter->net_stats.tx_errors = adapter->stats.ecol +
3000                                        adapter->stats.latecol;
3001         adapter->net_stats.tx_aborted_errors = adapter->stats.ecol;
3002         adapter->net_stats.tx_window_errors = adapter->stats.latecol;
3003         adapter->net_stats.tx_carrier_errors = adapter->stats.tncrs;
3004
3005         /* Tx Dropped needs to be maintained elsewhere */
3006
3007         /* Phy Stats */
3008         if (hw->phy.media_type == e1000_media_type_copper) {
3009                 if ((adapter->link_speed == SPEED_1000) &&
3010                    (!hw->phy.ops.read_phy_reg(hw, PHY_1000T_STATUS,
3011                                               &phy_tmp))) {
3012                         phy_tmp &= PHY_IDLE_ERROR_COUNT_MASK;
3013                         adapter->phy_stats.idle_errors += phy_tmp;
3014                 }
3015         }
3016
3017         /* Management Stats */
3018         adapter->stats.mgptc += rd32(E1000_MGTPTC);
3019         adapter->stats.mgprc += rd32(E1000_MGTPRC);
3020         adapter->stats.mgpdc += rd32(E1000_MGTPDC);
3021 }
3022
3023
3024 static irqreturn_t igb_msix_other(int irq, void *data)
3025 {
3026         struct net_device *netdev = data;
3027         struct igb_adapter *adapter = netdev_priv(netdev);
3028         struct e1000_hw *hw = &adapter->hw;
3029         u32 icr = rd32(E1000_ICR);
3030
3031         /* reading ICR causes bit 31 of EICR to be cleared */
3032         if (!(icr & E1000_ICR_LSC))
3033                 goto no_link_interrupt;
3034         hw->mac.get_link_status = 1;
3035         /* guard against interrupt when we're going down */
3036         if (!test_bit(__IGB_DOWN, &adapter->state))
3037                 mod_timer(&adapter->watchdog_timer, jiffies + 1);
3038
3039 no_link_interrupt:
3040         wr32(E1000_IMS, E1000_IMS_LSC);
3041         wr32(E1000_EIMS, adapter->eims_other);
3042
3043         return IRQ_HANDLED;
3044 }
3045
3046 static irqreturn_t igb_msix_tx(int irq, void *data)
3047 {
3048         struct igb_ring *tx_ring = data;
3049         struct igb_adapter *adapter = tx_ring->adapter;
3050         struct e1000_hw *hw = &adapter->hw;
3051
3052         if (!tx_ring->itr_val)
3053                 wr32(E1000_EIMC, tx_ring->eims_value);
3054
3055         tx_ring->total_bytes = 0;
3056         tx_ring->total_packets = 0;
3057         if (!igb_clean_tx_irq(tx_ring))
3058                 /* Ring was not completely cleaned, so fire another interrupt */
3059                 wr32(E1000_EICS, tx_ring->eims_value);
3060
3061         if (!tx_ring->itr_val)
3062                 wr32(E1000_EIMS, tx_ring->eims_value);
3063         return IRQ_HANDLED;
3064 }
3065
3066 static irqreturn_t igb_msix_rx(int irq, void *data)
3067 {
3068         struct igb_ring *rx_ring = data;
3069         struct igb_adapter *adapter = rx_ring->adapter;
3070         struct e1000_hw *hw = &adapter->hw;
3071
3072         /* Write the ITR value calculated at the end of the
3073          * previous interrupt.
3074          */
3075
3076         if (adapter->set_itr) {
3077                 wr32(rx_ring->itr_register,
3078                      1000000000 / (rx_ring->itr_val * 256));
3079                 adapter->set_itr = 0;
3080         }
3081
3082         if (netif_rx_schedule_prep(adapter->netdev, &rx_ring->napi))
3083                 __netif_rx_schedule(adapter->netdev, &rx_ring->napi);
3084
3085         return IRQ_HANDLED;
3086 }
3087
3088
3089 /**
3090  * igb_intr_msi - Interrupt Handler
3091  * @irq: interrupt number
3092  * @data: pointer to a network interface device structure
3093  **/
3094 static irqreturn_t igb_intr_msi(int irq, void *data)
3095 {
3096         struct net_device *netdev = data;
3097         struct igb_adapter *adapter = netdev_priv(netdev);
3098         struct e1000_hw *hw = &adapter->hw;
3099         /* read ICR disables interrupts using IAM */
3100         u32 icr = rd32(E1000_ICR);
3101
3102         /* Write the ITR value calculated at the end of the
3103          * previous interrupt.
3104          */
3105         if (adapter->set_itr) {
3106                 wr32(E1000_ITR, 1000000000 / (adapter->itr * 256));
3107                 adapter->set_itr = 0;
3108         }
3109
3110         if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
3111                 hw->mac.get_link_status = 1;
3112                 if (!test_bit(__IGB_DOWN, &adapter->state))
3113                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
3114         }
3115
3116         netif_rx_schedule(netdev, &adapter->rx_ring[0].napi);
3117
3118         return IRQ_HANDLED;
3119 }
3120
3121 /**
3122  * igb_intr - Interrupt Handler
3123  * @irq: interrupt number
3124  * @data: pointer to a network interface device structure
3125  **/
3126 static irqreturn_t igb_intr(int irq, void *data)
3127 {
3128         struct net_device *netdev = data;
3129         struct igb_adapter *adapter = netdev_priv(netdev);
3130         struct e1000_hw *hw = &adapter->hw;
3131         /* Interrupt Auto-Mask...upon reading ICR, interrupts are masked.  No
3132          * need for the IMC write */
3133         u32 icr = rd32(E1000_ICR);
3134         u32 eicr = 0;
3135         if (!icr)
3136                 return IRQ_NONE;  /* Not our interrupt */
3137
3138         /* Write the ITR value calculated at the end of the
3139          * previous interrupt.
3140          */
3141         if (adapter->set_itr) {
3142                 wr32(E1000_ITR, 1000000000 / (adapter->itr * 256));
3143                 adapter->set_itr = 0;
3144         }
3145
3146         /* IMS will not auto-mask if INT_ASSERTED is not set, and if it is
3147          * not set, then the adapter didn't send an interrupt */
3148         if (!(icr & E1000_ICR_INT_ASSERTED))
3149                 return IRQ_NONE;
3150
3151         eicr = rd32(E1000_EICR);
3152
3153         if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
3154                 hw->mac.get_link_status = 1;
3155                 /* guard against interrupt when we're going down */
3156                 if (!test_bit(__IGB_DOWN, &adapter->state))
3157                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
3158         }
3159
3160         netif_rx_schedule(netdev, &adapter->rx_ring[0].napi);
3161
3162         return IRQ_HANDLED;
3163 }
3164
3165 /**
3166  * igb_clean - NAPI Rx polling callback
3167  * @adapter: board private structure
3168  **/
3169 static int igb_clean(struct napi_struct *napi, int budget)
3170 {
3171         struct igb_adapter *adapter = container_of(napi, struct igb_adapter,
3172                                                    napi);
3173         struct net_device *netdev = adapter->netdev;
3174         int tx_clean_complete = 1, work_done = 0;
3175         int i;
3176
3177         /* Must NOT use netdev_priv macro here. */
3178         adapter = netdev->priv;
3179
3180         /* Keep link state information with original netdev */
3181         if (!netif_carrier_ok(netdev))
3182                 goto quit_polling;
3183
3184         /* igb_clean is called per-cpu.  This lock protects tx_ring[i] from
3185          * being cleaned by multiple cpus simultaneously.  A failure obtaining
3186          * the lock means tx_ring[i] is currently being cleaned anyway. */
3187         for (i = 0; i < adapter->num_tx_queues; i++) {
3188                 if (spin_trylock(&adapter->tx_ring[i].tx_clean_lock)) {
3189                         tx_clean_complete &= igb_clean_tx_irq(&adapter->tx_ring[i]);
3190                         spin_unlock(&adapter->tx_ring[i].tx_clean_lock);
3191                 }
3192         }
3193
3194         for (i = 0; i < adapter->num_rx_queues; i++)
3195                 igb_clean_rx_irq_adv(&adapter->rx_ring[i], &work_done,
3196                                      adapter->rx_ring[i].napi.weight);
3197
3198         /* If no Tx and not enough Rx work done, exit the polling mode */
3199         if ((tx_clean_complete && (work_done < budget)) ||
3200             !netif_running(netdev)) {
3201 quit_polling:
3202                 if (adapter->itr_setting & 3)
3203                         igb_set_itr(adapter, E1000_ITR, false);
3204                 netif_rx_complete(netdev, napi);
3205                 if (!test_bit(__IGB_DOWN, &adapter->state))
3206                         igb_irq_enable(adapter);
3207                 return 0;
3208         }
3209
3210         return 1;
3211 }
3212
3213 static int igb_clean_rx_ring_msix(struct napi_struct *napi, int budget)
3214 {
3215         struct igb_ring *rx_ring = container_of(napi, struct igb_ring, napi);
3216         struct igb_adapter *adapter = rx_ring->adapter;
3217         struct e1000_hw *hw = &adapter->hw;
3218         struct net_device *netdev = adapter->netdev;
3219         int work_done = 0;
3220
3221         /* Keep link state information with original netdev */
3222         if (!netif_carrier_ok(netdev))
3223                 goto quit_polling;
3224
3225         igb_clean_rx_irq_adv(rx_ring, &work_done, budget);
3226
3227
3228         /* If not enough Rx work done, exit the polling mode */
3229         if ((work_done == 0) || !netif_running(netdev)) {
3230 quit_polling:
3231                 netif_rx_complete(netdev, napi);
3232
3233                 wr32(E1000_EIMS, rx_ring->eims_value);
3234                 if ((adapter->itr_setting & 3) && !rx_ring->no_itr_adjust &&
3235                     (rx_ring->total_packets > IGB_DYN_ITR_PACKET_THRESHOLD)) {
3236                         int mean_size = rx_ring->total_bytes /
3237                                         rx_ring->total_packets;
3238                         if (mean_size < IGB_DYN_ITR_LENGTH_LOW)
3239                                 igb_raise_rx_eitr(adapter, rx_ring);
3240                         else if (mean_size > IGB_DYN_ITR_LENGTH_HIGH)
3241                                 igb_lower_rx_eitr(adapter, rx_ring);
3242                 }
3243
3244                 if (!test_bit(__IGB_DOWN, &adapter->state))
3245                         wr32(E1000_EIMS, rx_ring->eims_value);
3246
3247                 return 0;
3248         }
3249
3250         return 1;
3251 }
3252
3253 static inline u32 get_head(struct igb_ring *tx_ring)
3254 {
3255         void *end = (struct e1000_tx_desc *)tx_ring->desc + tx_ring->count;
3256         return le32_to_cpu(*(volatile __le32 *)end);
3257 }
3258
3259 /**
3260  * igb_clean_tx_irq - Reclaim resources after transmit completes
3261  * @adapter: board private structure
3262  * returns true if ring is completely cleaned
3263  **/
3264 static bool igb_clean_tx_irq(struct igb_ring *tx_ring)
3265 {
3266         struct igb_adapter *adapter = tx_ring->adapter;
3267         struct e1000_hw *hw = &adapter->hw;
3268         struct net_device *netdev = adapter->netdev;
3269         struct e1000_tx_desc *tx_desc;
3270         struct igb_buffer *buffer_info;
3271         struct sk_buff *skb;
3272         unsigned int i;
3273         u32 head, oldhead;
3274         unsigned int count = 0;
3275         bool cleaned = false;
3276         bool retval = true;
3277         unsigned int total_bytes = 0, total_packets = 0;
3278
3279         rmb();
3280         head = get_head(tx_ring);
3281         i = tx_ring->next_to_clean;
3282         while (1) {
3283                 while (i != head) {
3284                         cleaned = true;
3285                         tx_desc = E1000_TX_DESC(*tx_ring, i);
3286                         buffer_info = &tx_ring->buffer_info[i];
3287                         skb = buffer_info->skb;
3288
3289                         if (skb) {
3290                                 unsigned int segs, bytecount;
3291                                 /* gso_segs is currently only valid for tcp */
3292                                 segs = skb_shinfo(skb)->gso_segs ?: 1;
3293                                 /* multiply data chunks by size of headers */
3294                                 bytecount = ((segs - 1) * skb_headlen(skb)) +
3295                                             skb->len;
3296                                 total_packets += segs;
3297                                 total_bytes += bytecount;
3298                         }
3299
3300                         igb_unmap_and_free_tx_resource(adapter, buffer_info);
3301                         tx_desc->upper.data = 0;
3302
3303                         i++;
3304                         if (i == tx_ring->count)
3305                                 i = 0;
3306
3307                         count++;
3308                         if (count == IGB_MAX_TX_CLEAN) {
3309                                 retval = false;
3310                                 goto done_cleaning;
3311                         }
3312                 }
3313                 oldhead = head;
3314                 rmb();
3315                 head = get_head(tx_ring);
3316                 if (head == oldhead)
3317                         goto done_cleaning;
3318         }  /* while (1) */
3319
3320 done_cleaning:
3321         tx_ring->next_to_clean = i;
3322
3323         if (unlikely(cleaned &&
3324                      netif_carrier_ok(netdev) &&
3325                      IGB_DESC_UNUSED(tx_ring) >= IGB_TX_QUEUE_WAKE)) {
3326                 /* Make sure that anybody stopping the queue after this
3327                  * sees the new next_to_clean.
3328                  */
3329                 smp_mb();
3330                 if (netif_queue_stopped(netdev) &&
3331                     !(test_bit(__IGB_DOWN, &adapter->state))) {
3332                         netif_wake_queue(netdev);
3333                         ++adapter->restart_queue;
3334                 }
3335         }
3336
3337         if (tx_ring->detect_tx_hung) {
3338                 /* Detect a transmit hang in hardware, this serializes the
3339                  * check with the clearing of time_stamp and movement of i */
3340                 tx_ring->detect_tx_hung = false;
3341                 if (tx_ring->buffer_info[i].time_stamp &&
3342                     time_after(jiffies, tx_ring->buffer_info[i].time_stamp +
3343                                (adapter->tx_timeout_factor * HZ))
3344                     && !(rd32(E1000_STATUS) &
3345                          E1000_STATUS_TXOFF)) {
3346
3347                         tx_desc = E1000_TX_DESC(*tx_ring, i);
3348                         /* detected Tx unit hang */
3349                         dev_err(&adapter->pdev->dev,
3350                                 "Detected Tx Unit Hang\n"
3351                                 "  Tx Queue             <%lu>\n"
3352                                 "  TDH                  <%x>\n"
3353                                 "  TDT                  <%x>\n"
3354                                 "  next_to_use          <%x>\n"
3355                                 "  next_to_clean        <%x>\n"
3356                                 "  head (WB)            <%x>\n"
3357                                 "buffer_info[next_to_clean]\n"
3358                                 "  time_stamp           <%lx>\n"
3359                                 "  jiffies              <%lx>\n"
3360                                 "  desc.status          <%x>\n",
3361                                 (unsigned long)((tx_ring - adapter->tx_ring) /
3362                                         sizeof(struct igb_ring)),
3363                                 readl(adapter->hw.hw_addr + tx_ring->head),
3364                                 readl(adapter->hw.hw_addr + tx_ring->tail),
3365                                 tx_ring->next_to_use,
3366                                 tx_ring->next_to_clean,
3367                                 head,
3368                                 tx_ring->buffer_info[i].time_stamp,
3369                                 jiffies,
3370                                 tx_desc->upper.fields.status);
3371                         netif_stop_queue(netdev);
3372                 }
3373         }
3374         tx_ring->total_bytes += total_bytes;
3375         tx_ring->total_packets += total_packets;
3376         adapter->net_stats.tx_bytes += total_bytes;
3377         adapter->net_stats.tx_packets += total_packets;
3378         return retval;
3379 }
3380
3381
3382 /**
3383  * igb_receive_skb - helper function to handle rx indications
3384  * @adapter: board private structure
3385  * @status: descriptor status field as written by hardware
3386  * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
3387  * @skb: pointer to sk_buff to be indicated to stack
3388  **/
3389 static void igb_receive_skb(struct igb_adapter *adapter, u8 status, __le16 vlan,
3390                             struct sk_buff *skb)
3391 {
3392         if (adapter->vlgrp && (status & E1000_RXD_STAT_VP))
3393                 vlan_hwaccel_receive_skb(skb, adapter->vlgrp,
3394                                          le16_to_cpu(vlan));
3395         else
3396                 netif_receive_skb(skb);
3397 }
3398
3399
3400 static inline void igb_rx_checksum_adv(struct igb_adapter *adapter,
3401                                        u32 status_err, struct sk_buff *skb)
3402 {
3403         skb->ip_summed = CHECKSUM_NONE;
3404
3405         /* Ignore Checksum bit is set or checksum is disabled through ethtool */
3406         if ((status_err & E1000_RXD_STAT_IXSM) || !adapter->rx_csum)
3407                 return;
3408         /* TCP/UDP checksum error bit is set */
3409         if (status_err &
3410             (E1000_RXDEXT_STATERR_TCPE | E1000_RXDEXT_STATERR_IPE)) {
3411                 /* let the stack verify checksum errors */
3412                 adapter->hw_csum_err++;
3413                 return;
3414         }
3415         /* It must be a TCP or UDP packet with a valid checksum */
3416         if (status_err & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS))
3417                 skb->ip_summed = CHECKSUM_UNNECESSARY;
3418
3419         adapter->hw_csum_good++;
3420 }
3421
3422 static bool igb_clean_rx_irq_adv(struct igb_ring *rx_ring,
3423                                  int *work_done, int budget)
3424 {
3425         struct igb_adapter *adapter = rx_ring->adapter;
3426         struct net_device *netdev = adapter->netdev;
3427         struct pci_dev *pdev = adapter->pdev;
3428         union e1000_adv_rx_desc *rx_desc , *next_rxd;
3429         struct igb_buffer *buffer_info , *next_buffer;
3430         struct sk_buff *skb;
3431         unsigned int i, j;
3432         u32 length, hlen, staterr;
3433         bool cleaned = false;
3434         int cleaned_count = 0;
3435         unsigned int total_bytes = 0, total_packets = 0;
3436
3437         i = rx_ring->next_to_clean;
3438         rx_desc = E1000_RX_DESC_ADV(*rx_ring, i);
3439         staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
3440
3441         while (staterr & E1000_RXD_STAT_DD) {
3442                 if (*work_done >= budget)
3443                         break;
3444                 (*work_done)++;
3445                 buffer_info = &rx_ring->buffer_info[i];
3446
3447                 /* HW will not DMA in data larger than the given buffer, even
3448                  * if it parses the (NFS, of course) header to be larger.  In
3449                  * that case, it fills the header buffer and spills the rest
3450                  * into the page.
3451                  */
3452                 hlen = (le16_to_cpu(rx_desc->wb.lower.lo_dword.hdr_info) &
3453                   E1000_RXDADV_HDRBUFLEN_MASK) >> E1000_RXDADV_HDRBUFLEN_SHIFT;
3454                 if (hlen > adapter->rx_ps_hdr_size)
3455                         hlen = adapter->rx_ps_hdr_size;
3456
3457                 length = le16_to_cpu(rx_desc->wb.upper.length);
3458                 cleaned = true;
3459                 cleaned_count++;
3460
3461                 if (rx_ring->pending_skb != NULL) {
3462                         skb = rx_ring->pending_skb;
3463                         rx_ring->pending_skb = NULL;
3464                         j = rx_ring->pending_skb_page;
3465                 } else {
3466                         skb = buffer_info->skb;
3467                         prefetch(skb->data - NET_IP_ALIGN);
3468                         buffer_info->skb = NULL;
3469                         if (hlen) {
3470                                 pci_unmap_single(pdev, buffer_info->dma,
3471                                                  adapter->rx_ps_hdr_size +
3472                                                    NET_IP_ALIGN,
3473                                                  PCI_DMA_FROMDEVICE);
3474                                 skb_put(skb, hlen);
3475                         } else {
3476                                 pci_unmap_single(pdev, buffer_info->dma,
3477                                                  adapter->rx_buffer_len +
3478                                                    NET_IP_ALIGN,
3479                                                  PCI_DMA_FROMDEVICE);
3480                                 skb_put(skb, length);
3481                                 goto send_up;
3482                         }
3483                         j = 0;
3484                 }
3485
3486                 while (length) {
3487                         pci_unmap_page(pdev, buffer_info->page_dma,
3488                                 PAGE_SIZE, PCI_DMA_FROMDEVICE);
3489                         buffer_info->page_dma = 0;
3490                         skb_fill_page_desc(skb, j, buffer_info->page,
3491                                                 0, length);
3492                         buffer_info->page = NULL;
3493
3494                         skb->len += length;
3495                         skb->data_len += length;
3496                         skb->truesize += length;
3497                         rx_desc->wb.upper.status_error = 0;
3498                         if (staterr & E1000_RXD_STAT_EOP)
3499                                 break;
3500
3501                         j++;
3502                         cleaned_count++;
3503                         i++;
3504                         if (i == rx_ring->count)
3505                                 i = 0;
3506
3507                         buffer_info = &rx_ring->buffer_info[i];
3508                         rx_desc = E1000_RX_DESC_ADV(*rx_ring, i);
3509                         staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
3510                         length = le16_to_cpu(rx_desc->wb.upper.length);
3511                         if (!(staterr & E1000_RXD_STAT_DD)) {
3512                                 rx_ring->pending_skb = skb;
3513                                 rx_ring->pending_skb_page = j;
3514                                 goto out;
3515                         }
3516                 }
3517 send_up:
3518                 pskb_trim(skb, skb->len - 4);
3519                 i++;
3520                 if (i == rx_ring->count)
3521                         i = 0;
3522                 next_rxd = E1000_RX_DESC_ADV(*rx_ring, i);
3523                 prefetch(next_rxd);
3524                 next_buffer = &rx_ring->buffer_info[i];
3525
3526                 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
3527                         dev_kfree_skb_irq(skb);
3528                         goto next_desc;
3529                 }
3530                 rx_ring->no_itr_adjust |= (staterr & E1000_RXD_STAT_DYNINT);
3531
3532                 total_bytes += skb->len;
3533                 total_packets++;
3534
3535                 igb_rx_checksum_adv(adapter, staterr, skb);
3536
3537                 skb->protocol = eth_type_trans(skb, netdev);
3538
3539                 igb_receive_skb(adapter, staterr, rx_desc->wb.upper.vlan, skb);
3540
3541                 netdev->last_rx = jiffies;
3542
3543 next_desc:
3544                 rx_desc->wb.upper.status_error = 0;
3545
3546                 /* return some buffers to hardware, one at a time is too slow */
3547                 if (cleaned_count >= IGB_RX_BUFFER_WRITE) {
3548                         igb_alloc_rx_buffers_adv(rx_ring, cleaned_count);
3549                         cleaned_count = 0;
3550                 }
3551
3552                 /* use prefetched values */
3553                 rx_desc = next_rxd;
3554                 buffer_info = next_buffer;
3555
3556                 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
3557         }
3558 out:
3559         rx_ring->next_to_clean = i;
3560         cleaned_count = IGB_DESC_UNUSED(rx_ring);
3561
3562         if (cleaned_count)
3563                 igb_alloc_rx_buffers_adv(rx_ring, cleaned_count);
3564
3565         rx_ring->total_packets += total_packets;
3566         rx_ring->total_bytes += total_bytes;
3567         rx_ring->rx_stats.packets += total_packets;
3568         rx_ring->rx_stats.bytes += total_bytes;
3569         adapter->net_stats.rx_bytes += total_bytes;
3570         adapter->net_stats.rx_packets += total_packets;
3571         return cleaned;
3572 }
3573
3574
3575 /**
3576  * igb_alloc_rx_buffers_adv - Replace used receive buffers; packet split
3577  * @adapter: address of board private structure
3578  **/
3579 static void igb_alloc_rx_buffers_adv(struct igb_ring *rx_ring,
3580                                      int cleaned_count)
3581 {
3582         struct igb_adapter *adapter = rx_ring->adapter;
3583         struct net_device *netdev = adapter->netdev;
3584         struct pci_dev *pdev = adapter->pdev;
3585         union e1000_adv_rx_desc *rx_desc;
3586         struct igb_buffer *buffer_info;
3587         struct sk_buff *skb;
3588         unsigned int i;
3589
3590         i = rx_ring->next_to_use;
3591         buffer_info = &rx_ring->buffer_info[i];
3592
3593         while (cleaned_count--) {
3594                 rx_desc = E1000_RX_DESC_ADV(*rx_ring, i);
3595
3596                 if (adapter->rx_ps_hdr_size && !buffer_info->page) {
3597                         buffer_info->page = alloc_page(GFP_ATOMIC);
3598                         if (!buffer_info->page) {
3599                                 adapter->alloc_rx_buff_failed++;
3600                                 goto no_buffers;
3601                         }
3602                         buffer_info->page_dma =
3603                                 pci_map_page(pdev,
3604                                              buffer_info->page,
3605                                              0, PAGE_SIZE,
3606                                              PCI_DMA_FROMDEVICE);
3607                 }
3608
3609                 if (!buffer_info->skb) {
3610                         int bufsz;
3611
3612                         if (adapter->rx_ps_hdr_size)
3613                                 bufsz = adapter->rx_ps_hdr_size;
3614                         else
3615                                 bufsz = adapter->rx_buffer_len;
3616                         bufsz += NET_IP_ALIGN;
3617                         skb = netdev_alloc_skb(netdev, bufsz);
3618
3619                         if (!skb) {
3620                                 adapter->alloc_rx_buff_failed++;
3621                                 goto no_buffers;
3622                         }
3623
3624                         /* Make buffer alignment 2 beyond a 16 byte boundary
3625                          * this will result in a 16 byte aligned IP header after
3626                          * the 14 byte MAC header is removed
3627                          */
3628                         skb_reserve(skb, NET_IP_ALIGN);
3629
3630                         buffer_info->skb = skb;
3631                         buffer_info->dma = pci_map_single(pdev, skb->data,
3632                                                           bufsz,
3633                                                           PCI_DMA_FROMDEVICE);
3634
3635                 }
3636                 /* Refresh the desc even if buffer_addrs didn't change because
3637                  * each write-back erases this info. */
3638                 if (adapter->rx_ps_hdr_size) {
3639                         rx_desc->read.pkt_addr =
3640                              cpu_to_le64(buffer_info->page_dma);
3641                         rx_desc->read.hdr_addr = cpu_to_le64(buffer_info->dma);
3642                 } else {
3643                         rx_desc->read.pkt_addr =
3644                              cpu_to_le64(buffer_info->dma);
3645                         rx_desc->read.hdr_addr = 0;
3646                 }
3647
3648                 i++;
3649                 if (i == rx_ring->count)
3650                         i = 0;
3651                 buffer_info = &rx_ring->buffer_info[i];
3652         }
3653
3654 no_buffers:
3655         if (rx_ring->next_to_use != i) {
3656                 rx_ring->next_to_use = i;
3657                 if (i == 0)
3658                         i = (rx_ring->count - 1);
3659                 else
3660                         i--;
3661
3662                 /* Force memory writes to complete before letting h/w
3663                  * know there are new descriptors to fetch.  (Only
3664                  * applicable for weak-ordered memory model archs,
3665                  * such as IA-64). */
3666                 wmb();
3667                 writel(i, adapter->hw.hw_addr + rx_ring->tail);
3668         }
3669 }
3670
3671 /**
3672  * igb_mii_ioctl -
3673  * @netdev:
3674  * @ifreq:
3675  * @cmd:
3676  **/
3677 static int igb_mii_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
3678 {
3679         struct igb_adapter *adapter = netdev_priv(netdev);
3680         struct mii_ioctl_data *data = if_mii(ifr);
3681
3682         if (adapter->hw.phy.media_type != e1000_media_type_copper)
3683                 return -EOPNOTSUPP;
3684
3685         switch (cmd) {
3686         case SIOCGMIIPHY:
3687                 data->phy_id = adapter->hw.phy.addr;
3688                 break;
3689         case SIOCGMIIREG:
3690                 if (!capable(CAP_NET_ADMIN))
3691                         return -EPERM;
3692                 if (adapter->hw.phy.ops.read_phy_reg(&adapter->hw,
3693                                                      data->reg_num
3694                                                      & 0x1F, &data->val_out))
3695                         return -EIO;
3696                 break;
3697         case SIOCSMIIREG:
3698         default:
3699                 return -EOPNOTSUPP;
3700         }
3701         return 0;
3702 }
3703
3704 /**
3705  * igb_ioctl -
3706  * @netdev:
3707  * @ifreq:
3708  * @cmd:
3709  **/
3710 static int igb_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
3711 {
3712         switch (cmd) {
3713         case SIOCGMIIPHY:
3714         case SIOCGMIIREG:
3715         case SIOCSMIIREG:
3716                 return igb_mii_ioctl(netdev, ifr, cmd);
3717         default:
3718                 return -EOPNOTSUPP;
3719         }
3720 }
3721
3722 static void igb_vlan_rx_register(struct net_device *netdev,
3723                                  struct vlan_group *grp)
3724 {
3725         struct igb_adapter *adapter = netdev_priv(netdev);
3726         struct e1000_hw *hw = &adapter->hw;
3727         u32 ctrl, rctl;
3728
3729         igb_irq_disable(adapter);
3730         adapter->vlgrp = grp;
3731
3732         if (grp) {
3733                 /* enable VLAN tag insert/strip */
3734                 ctrl = rd32(E1000_CTRL);
3735                 ctrl |= E1000_CTRL_VME;
3736                 wr32(E1000_CTRL, ctrl);
3737
3738                 /* enable VLAN receive filtering */
3739                 rctl = rd32(E1000_RCTL);
3740                 rctl |= E1000_RCTL_VFE;
3741                 rctl &= ~E1000_RCTL_CFIEN;
3742                 wr32(E1000_RCTL, rctl);
3743                 igb_update_mng_vlan(adapter);
3744                 wr32(E1000_RLPML,
3745                                 adapter->max_frame_size + VLAN_TAG_SIZE);
3746         } else {
3747                 /* disable VLAN tag insert/strip */
3748                 ctrl = rd32(E1000_CTRL);
3749                 ctrl &= ~E1000_CTRL_VME;
3750                 wr32(E1000_CTRL, ctrl);
3751
3752                 /* disable VLAN filtering */
3753                 rctl = rd32(E1000_RCTL);
3754                 rctl &= ~E1000_RCTL_VFE;
3755                 wr32(E1000_RCTL, rctl);
3756                 if (adapter->mng_vlan_id != (u16)IGB_MNG_VLAN_NONE) {
3757                         igb_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
3758                         adapter->mng_vlan_id = IGB_MNG_VLAN_NONE;
3759                 }
3760                 wr32(E1000_RLPML,
3761                                 adapter->max_frame_size);
3762         }
3763
3764         if (!test_bit(__IGB_DOWN, &adapter->state))
3765                 igb_irq_enable(adapter);
3766 }
3767
3768 static void igb_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
3769 {
3770         struct igb_adapter *adapter = netdev_priv(netdev);
3771         struct e1000_hw *hw = &adapter->hw;
3772         u32 vfta, index;
3773
3774         if ((adapter->hw.mng_cookie.status &
3775              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
3776             (vid == adapter->mng_vlan_id))
3777                 return;
3778         /* add VID to filter table */
3779         index = (vid >> 5) & 0x7F;
3780         vfta = array_rd32(E1000_VFTA, index);
3781         vfta |= (1 << (vid & 0x1F));
3782         igb_write_vfta(&adapter->hw, index, vfta);
3783 }
3784
3785 static void igb_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
3786 {
3787         struct igb_adapter *adapter = netdev_priv(netdev);
3788         struct e1000_hw *hw = &adapter->hw;
3789         u32 vfta, index;
3790
3791         igb_irq_disable(adapter);
3792         vlan_group_set_device(adapter->vlgrp, vid, NULL);
3793
3794         if (!test_bit(__IGB_DOWN, &adapter->state))
3795                 igb_irq_enable(adapter);
3796
3797         if ((adapter->hw.mng_cookie.status &
3798              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
3799             (vid == adapter->mng_vlan_id)) {
3800                 /* release control to f/w */
3801                 igb_release_hw_control(adapter);
3802                 return;
3803         }
3804
3805         /* remove VID from filter table */
3806         index = (vid >> 5) & 0x7F;
3807         vfta = array_rd32(E1000_VFTA, index);
3808         vfta &= ~(1 << (vid & 0x1F));
3809         igb_write_vfta(&adapter->hw, index, vfta);
3810 }
3811
3812 static void igb_restore_vlan(struct igb_adapter *adapter)
3813 {
3814         igb_vlan_rx_register(adapter->netdev, adapter->vlgrp);
3815
3816         if (adapter->vlgrp) {
3817                 u16 vid;
3818                 for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
3819                         if (!vlan_group_get_device(adapter->vlgrp, vid))
3820                                 continue;
3821                         igb_vlan_rx_add_vid(adapter->netdev, vid);
3822                 }
3823         }
3824 }
3825
3826 int igb_set_spd_dplx(struct igb_adapter *adapter, u16 spddplx)
3827 {
3828         struct e1000_mac_info *mac = &adapter->hw.mac;
3829
3830         mac->autoneg = 0;
3831
3832         /* Fiber NICs only allow 1000 gbps Full duplex */
3833         if ((adapter->hw.phy.media_type == e1000_media_type_fiber) &&
3834                 spddplx != (SPEED_1000 + DUPLEX_FULL)) {
3835                 dev_err(&adapter->pdev->dev,
3836                         "Unsupported Speed/Duplex configuration\n");
3837                 return -EINVAL;
3838         }
3839
3840         switch (spddplx) {
3841         case SPEED_10 + DUPLEX_HALF:
3842                 mac->forced_speed_duplex = ADVERTISE_10_HALF;
3843                 break;
3844         case SPEED_10 + DUPLEX_FULL:
3845                 mac->forced_speed_duplex = ADVERTISE_10_FULL;
3846                 break;
3847         case SPEED_100 + DUPLEX_HALF:
3848                 mac->forced_speed_duplex = ADVERTISE_100_HALF;
3849                 break;
3850         case SPEED_100 + DUPLEX_FULL:
3851                 mac->forced_speed_duplex = ADVERTISE_100_FULL;
3852                 break;
3853         case SPEED_1000 + DUPLEX_FULL:
3854                 mac->autoneg = 1;
3855                 adapter->hw.phy.autoneg_advertised = ADVERTISE_1000_FULL;
3856                 break;
3857         case SPEED_1000 + DUPLEX_HALF: /* not supported */
3858         default:
3859                 dev_err(&adapter->pdev->dev,
3860                         "Unsupported Speed/Duplex configuration\n");
3861                 return -EINVAL;
3862         }
3863         return 0;
3864 }
3865
3866
3867 static int igb_suspend(struct pci_dev *pdev, pm_message_t state)
3868 {
3869         struct net_device *netdev = pci_get_drvdata(pdev);
3870         struct igb_adapter *adapter = netdev_priv(netdev);
3871         struct e1000_hw *hw = &adapter->hw;
3872         u32 ctrl, ctrl_ext, rctl, status;
3873         u32 wufc = adapter->wol;
3874 #ifdef CONFIG_PM
3875         int retval = 0;
3876 #endif
3877
3878         netif_device_detach(netdev);
3879
3880         if (netif_running(netdev)) {
3881                 WARN_ON(test_bit(__IGB_RESETTING, &adapter->state));
3882                 igb_down(adapter);
3883                 igb_free_irq(adapter);
3884         }
3885
3886 #ifdef CONFIG_PM
3887         retval = pci_save_state(pdev);
3888         if (retval)
3889                 return retval;
3890 #endif
3891
3892         status = rd32(E1000_STATUS);
3893         if (status & E1000_STATUS_LU)
3894                 wufc &= ~E1000_WUFC_LNKC;
3895
3896         if (wufc) {
3897                 igb_setup_rctl(adapter);
3898                 igb_set_multi(netdev);
3899
3900                 /* turn on all-multi mode if wake on multicast is enabled */
3901                 if (wufc & E1000_WUFC_MC) {
3902                         rctl = rd32(E1000_RCTL);
3903                         rctl |= E1000_RCTL_MPE;
3904                         wr32(E1000_RCTL, rctl);
3905                 }
3906
3907                 ctrl = rd32(E1000_CTRL);
3908                 /* advertise wake from D3Cold */
3909                 #define E1000_CTRL_ADVD3WUC 0x00100000
3910                 /* phy power management enable */
3911                 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
3912                 ctrl |= E1000_CTRL_ADVD3WUC;
3913                 wr32(E1000_CTRL, ctrl);
3914
3915                 if (adapter->hw.phy.media_type == e1000_media_type_fiber ||
3916                    adapter->hw.phy.media_type ==
3917                                         e1000_media_type_internal_serdes) {
3918                         /* keep the laser running in D3 */
3919                         ctrl_ext = rd32(E1000_CTRL_EXT);
3920                         ctrl_ext |= E1000_CTRL_EXT_SDP7_DATA;
3921                         wr32(E1000_CTRL_EXT, ctrl_ext);
3922                 }
3923
3924                 /* Allow time for pending master requests to run */
3925                 igb_disable_pcie_master(&adapter->hw);
3926
3927                 wr32(E1000_WUC, E1000_WUC_PME_EN);
3928                 wr32(E1000_WUFC, wufc);
3929                 pci_enable_wake(pdev, PCI_D3hot, 1);
3930                 pci_enable_wake(pdev, PCI_D3cold, 1);
3931         } else {
3932                 wr32(E1000_WUC, 0);
3933                 wr32(E1000_WUFC, 0);
3934                 pci_enable_wake(pdev, PCI_D3hot, 0);
3935                 pci_enable_wake(pdev, PCI_D3cold, 0);
3936         }
3937
3938         /* make sure adapter isn't asleep if manageability is enabled */
3939         if (adapter->en_mng_pt) {
3940                 pci_enable_wake(pdev, PCI_D3hot, 1);
3941                 pci_enable_wake(pdev, PCI_D3cold, 1);
3942         }
3943
3944         /* Release control of h/w to f/w.  If f/w is AMT enabled, this
3945          * would have already happened in close and is redundant. */
3946         igb_release_hw_control(adapter);
3947
3948         pci_disable_device(pdev);
3949
3950         pci_set_power_state(pdev, pci_choose_state(pdev, state));
3951
3952         return 0;
3953 }
3954
3955 #ifdef CONFIG_PM
3956 static int igb_resume(struct pci_dev *pdev)
3957 {
3958         struct net_device *netdev = pci_get_drvdata(pdev);
3959         struct igb_adapter *adapter = netdev_priv(netdev);
3960         struct e1000_hw *hw = &adapter->hw;
3961         u32 err;
3962
3963         pci_set_power_state(pdev, PCI_D0);
3964         pci_restore_state(pdev);
3965
3966         if (adapter->need_ioport)
3967                 err = pci_enable_device(pdev);
3968         else
3969                 err = pci_enable_device_mem(pdev);
3970         if (err) {
3971                 dev_err(&pdev->dev,
3972                         "igb: Cannot enable PCI device from suspend\n");
3973                 return err;
3974         }
3975         pci_set_master(pdev);
3976
3977         pci_enable_wake(pdev, PCI_D3hot, 0);
3978         pci_enable_wake(pdev, PCI_D3cold, 0);
3979
3980         if (netif_running(netdev)) {
3981                 err = igb_request_irq(adapter);
3982                 if (err)
3983                         return err;
3984         }
3985
3986         /* e1000_power_up_phy(adapter); */
3987
3988         igb_reset(adapter);
3989         wr32(E1000_WUS, ~0);
3990
3991         igb_init_manageability(adapter);
3992
3993         if (netif_running(netdev))
3994                 igb_up(adapter);
3995
3996         netif_device_attach(netdev);
3997
3998         /* let the f/w know that the h/w is now under the control of the
3999          * driver. */
4000         igb_get_hw_control(adapter);
4001
4002         return 0;
4003 }
4004 #endif
4005
4006 static void igb_shutdown(struct pci_dev *pdev)
4007 {
4008         igb_suspend(pdev, PMSG_SUSPEND);
4009 }
4010
4011 #ifdef CONFIG_NET_POLL_CONTROLLER
4012 /*
4013  * Polling 'interrupt' - used by things like netconsole to send skbs
4014  * without having to re-enable interrupts. It's not called while
4015  * the interrupt routine is executing.
4016  */
4017 static void igb_netpoll(struct net_device *netdev)
4018 {
4019         struct igb_adapter *adapter = netdev_priv(netdev);
4020         int i;
4021         int work_done = 0;
4022
4023         igb_irq_disable(adapter);
4024         for (i = 0; i < adapter->num_tx_queues; i++)
4025                 igb_clean_tx_irq(&adapter->tx_ring[i]);
4026
4027         for (i = 0; i < adapter->num_rx_queues; i++)
4028                 igb_clean_rx_irq_adv(&adapter->rx_ring[i],
4029                                      &work_done,
4030                                      adapter->rx_ring[i].napi.weight);
4031
4032         igb_irq_enable(adapter);
4033 }
4034 #endif /* CONFIG_NET_POLL_CONTROLLER */
4035
4036 /**
4037  * igb_io_error_detected - called when PCI error is detected
4038  * @pdev: Pointer to PCI device
4039  * @state: The current pci connection state
4040  *
4041  * This function is called after a PCI bus error affecting
4042  * this device has been detected.
4043  */
4044 static pci_ers_result_t igb_io_error_detected(struct pci_dev *pdev,
4045                                               pci_channel_state_t state)
4046 {
4047         struct net_device *netdev = pci_get_drvdata(pdev);
4048         struct igb_adapter *adapter = netdev_priv(netdev);
4049
4050         netif_device_detach(netdev);
4051
4052         if (netif_running(netdev))
4053                 igb_down(adapter);
4054         pci_disable_device(pdev);
4055
4056         /* Request a slot slot reset. */
4057         return PCI_ERS_RESULT_NEED_RESET;
4058 }
4059
4060 /**
4061  * igb_io_slot_reset - called after the pci bus has been reset.
4062  * @pdev: Pointer to PCI device
4063  *
4064  * Restart the card from scratch, as if from a cold-boot. Implementation
4065  * resembles the first-half of the igb_resume routine.
4066  */
4067 static pci_ers_result_t igb_io_slot_reset(struct pci_dev *pdev)
4068 {
4069         struct net_device *netdev = pci_get_drvdata(pdev);
4070         struct igb_adapter *adapter = netdev_priv(netdev);
4071         struct e1000_hw *hw = &adapter->hw;
4072         int err;
4073
4074         if (adapter->need_ioport)
4075                 err = pci_enable_device(pdev);
4076         else
4077                 err = pci_enable_device_mem(pdev);
4078         if (err) {
4079                 dev_err(&pdev->dev,
4080                         "Cannot re-enable PCI device after reset.\n");
4081                 return PCI_ERS_RESULT_DISCONNECT;
4082         }
4083         pci_set_master(pdev);
4084         pci_restore_state(pdev);
4085
4086         pci_enable_wake(pdev, PCI_D3hot, 0);
4087         pci_enable_wake(pdev, PCI_D3cold, 0);
4088
4089         igb_reset(adapter);
4090         wr32(E1000_WUS, ~0);
4091
4092         return PCI_ERS_RESULT_RECOVERED;
4093 }
4094
4095 /**
4096  * igb_io_resume - called when traffic can start flowing again.
4097  * @pdev: Pointer to PCI device
4098  *
4099  * This callback is called when the error recovery driver tells us that
4100  * its OK to resume normal operation. Implementation resembles the
4101  * second-half of the igb_resume routine.
4102  */
4103 static void igb_io_resume(struct pci_dev *pdev)
4104 {
4105         struct net_device *netdev = pci_get_drvdata(pdev);
4106         struct igb_adapter *adapter = netdev_priv(netdev);
4107
4108         igb_init_manageability(adapter);
4109
4110         if (netif_running(netdev)) {
4111                 if (igb_up(adapter)) {
4112                         dev_err(&pdev->dev, "igb_up failed after reset\n");
4113                         return;
4114                 }
4115         }
4116
4117         netif_device_attach(netdev);
4118
4119         /* let the f/w know that the h/w is now under the control of the
4120          * driver. */
4121         igb_get_hw_control(adapter);
4122
4123 }
4124
4125 /* igb_main.c */