Merge branch 'master' of git://git.kernel.org/pub/scm/linux/kernel/git/jk/spufs into...
[linux-2.6] / drivers / net / igb / igb_main.c
1 /*******************************************************************************
2
3   Intel(R) Gigabit Ethernet Linux driver
4   Copyright(c) 2007 Intel Corporation.
5
6   This program is free software; you can redistribute it and/or modify it
7   under the terms and conditions of the GNU General Public License,
8   version 2, as published by the Free Software Foundation.
9
10   This program is distributed in the hope it will be useful, but WITHOUT
11   ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12   FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
13   more details.
14
15   You should have received a copy of the GNU General Public License along with
16   this program; if not, write to the Free Software Foundation, Inc.,
17   51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
18
19   The full GNU General Public License is included in this distribution in
20   the file called "COPYING".
21
22   Contact Information:
23   e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
24   Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
25
26 *******************************************************************************/
27
28 #include <linux/module.h>
29 #include <linux/types.h>
30 #include <linux/init.h>
31 #include <linux/vmalloc.h>
32 #include <linux/pagemap.h>
33 #include <linux/netdevice.h>
34 #include <linux/tcp.h>
35 #include <linux/ipv6.h>
36 #include <net/checksum.h>
37 #include <net/ip6_checksum.h>
38 #include <linux/mii.h>
39 #include <linux/ethtool.h>
40 #include <linux/if_vlan.h>
41 #include <linux/pci.h>
42 #include <linux/delay.h>
43 #include <linux/interrupt.h>
44 #include <linux/if_ether.h>
45
46 #include "igb.h"
47
48 #define DRV_VERSION "1.0.8-k2"
49 char igb_driver_name[] = "igb";
50 char igb_driver_version[] = DRV_VERSION;
51 static const char igb_driver_string[] =
52                                 "Intel(R) Gigabit Ethernet Network Driver";
53 static const char igb_copyright[] = "Copyright (c) 2007 Intel Corporation.";
54
55
56 static const struct e1000_info *igb_info_tbl[] = {
57         [board_82575] = &e1000_82575_info,
58 };
59
60 static struct pci_device_id igb_pci_tbl[] = {
61         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575EB_COPPER), board_82575 },
62         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575EB_FIBER_SERDES), board_82575 },
63         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575GB_QUAD_COPPER), board_82575 },
64         /* required last entry */
65         {0, }
66 };
67
68 MODULE_DEVICE_TABLE(pci, igb_pci_tbl);
69
70 void igb_reset(struct igb_adapter *);
71 static int igb_setup_all_tx_resources(struct igb_adapter *);
72 static int igb_setup_all_rx_resources(struct igb_adapter *);
73 static void igb_free_all_tx_resources(struct igb_adapter *);
74 static void igb_free_all_rx_resources(struct igb_adapter *);
75 static void igb_free_tx_resources(struct igb_adapter *, struct igb_ring *);
76 static void igb_free_rx_resources(struct igb_adapter *, struct igb_ring *);
77 void igb_update_stats(struct igb_adapter *);
78 static int igb_probe(struct pci_dev *, const struct pci_device_id *);
79 static void __devexit igb_remove(struct pci_dev *pdev);
80 static int igb_sw_init(struct igb_adapter *);
81 static int igb_open(struct net_device *);
82 static int igb_close(struct net_device *);
83 static void igb_configure_tx(struct igb_adapter *);
84 static void igb_configure_rx(struct igb_adapter *);
85 static void igb_setup_rctl(struct igb_adapter *);
86 static void igb_clean_all_tx_rings(struct igb_adapter *);
87 static void igb_clean_all_rx_rings(struct igb_adapter *);
88 static void igb_clean_tx_ring(struct igb_adapter *, struct igb_ring *);
89 static void igb_clean_rx_ring(struct igb_adapter *, struct igb_ring *);
90 static void igb_set_multi(struct net_device *);
91 static void igb_update_phy_info(unsigned long);
92 static void igb_watchdog(unsigned long);
93 static void igb_watchdog_task(struct work_struct *);
94 static int igb_xmit_frame_ring_adv(struct sk_buff *, struct net_device *,
95                                   struct igb_ring *);
96 static int igb_xmit_frame_adv(struct sk_buff *skb, struct net_device *);
97 static struct net_device_stats *igb_get_stats(struct net_device *);
98 static int igb_change_mtu(struct net_device *, int);
99 static int igb_set_mac(struct net_device *, void *);
100 static irqreturn_t igb_intr(int irq, void *);
101 static irqreturn_t igb_intr_msi(int irq, void *);
102 static irqreturn_t igb_msix_other(int irq, void *);
103 static irqreturn_t igb_msix_rx(int irq, void *);
104 static irqreturn_t igb_msix_tx(int irq, void *);
105 static int igb_clean_rx_ring_msix(struct napi_struct *, int);
106 static bool igb_clean_tx_irq(struct igb_adapter *, struct igb_ring *);
107 static int igb_clean(struct napi_struct *, int);
108 static bool igb_clean_rx_irq_adv(struct igb_adapter *,
109                                  struct igb_ring *, int *, int);
110 static void igb_alloc_rx_buffers_adv(struct igb_adapter *,
111                                      struct igb_ring *, int);
112 static int igb_ioctl(struct net_device *, struct ifreq *, int cmd);
113 static void igb_tx_timeout(struct net_device *);
114 static void igb_reset_task(struct work_struct *);
115 static void igb_vlan_rx_register(struct net_device *, struct vlan_group *);
116 static void igb_vlan_rx_add_vid(struct net_device *, u16);
117 static void igb_vlan_rx_kill_vid(struct net_device *, u16);
118 static void igb_restore_vlan(struct igb_adapter *);
119
120 static int igb_suspend(struct pci_dev *, pm_message_t);
121 #ifdef CONFIG_PM
122 static int igb_resume(struct pci_dev *);
123 #endif
124 static void igb_shutdown(struct pci_dev *);
125
126 #ifdef CONFIG_NET_POLL_CONTROLLER
127 /* for netdump / net console */
128 static void igb_netpoll(struct net_device *);
129 #endif
130
131 static pci_ers_result_t igb_io_error_detected(struct pci_dev *,
132                      pci_channel_state_t);
133 static pci_ers_result_t igb_io_slot_reset(struct pci_dev *);
134 static void igb_io_resume(struct pci_dev *);
135
136 static struct pci_error_handlers igb_err_handler = {
137         .error_detected = igb_io_error_detected,
138         .slot_reset = igb_io_slot_reset,
139         .resume = igb_io_resume,
140 };
141
142
143 static struct pci_driver igb_driver = {
144         .name     = igb_driver_name,
145         .id_table = igb_pci_tbl,
146         .probe    = igb_probe,
147         .remove   = __devexit_p(igb_remove),
148 #ifdef CONFIG_PM
149         /* Power Managment Hooks */
150         .suspend  = igb_suspend,
151         .resume   = igb_resume,
152 #endif
153         .shutdown = igb_shutdown,
154         .err_handler = &igb_err_handler
155 };
156
157 MODULE_AUTHOR("Intel Corporation, <e1000-devel@lists.sourceforge.net>");
158 MODULE_DESCRIPTION("Intel(R) Gigabit Ethernet Network Driver");
159 MODULE_LICENSE("GPL");
160 MODULE_VERSION(DRV_VERSION);
161
162 #ifdef DEBUG
163 /**
164  * igb_get_hw_dev_name - return device name string
165  * used by hardware layer to print debugging information
166  **/
167 char *igb_get_hw_dev_name(struct e1000_hw *hw)
168 {
169         struct igb_adapter *adapter = hw->back;
170         return adapter->netdev->name;
171 }
172 #endif
173
174 /**
175  * igb_init_module - Driver Registration Routine
176  *
177  * igb_init_module is the first routine called when the driver is
178  * loaded. All it does is register with the PCI subsystem.
179  **/
180 static int __init igb_init_module(void)
181 {
182         int ret;
183         printk(KERN_INFO "%s - version %s\n",
184                igb_driver_string, igb_driver_version);
185
186         printk(KERN_INFO "%s\n", igb_copyright);
187
188         ret = pci_register_driver(&igb_driver);
189         return ret;
190 }
191
192 module_init(igb_init_module);
193
194 /**
195  * igb_exit_module - Driver Exit Cleanup Routine
196  *
197  * igb_exit_module is called just before the driver is removed
198  * from memory.
199  **/
200 static void __exit igb_exit_module(void)
201 {
202         pci_unregister_driver(&igb_driver);
203 }
204
205 module_exit(igb_exit_module);
206
207 /**
208  * igb_alloc_queues - Allocate memory for all rings
209  * @adapter: board private structure to initialize
210  *
211  * We allocate one ring per queue at run-time since we don't know the
212  * number of queues at compile-time.
213  **/
214 static int igb_alloc_queues(struct igb_adapter *adapter)
215 {
216         int i;
217
218         adapter->tx_ring = kcalloc(adapter->num_tx_queues,
219                                    sizeof(struct igb_ring), GFP_KERNEL);
220         if (!adapter->tx_ring)
221                 return -ENOMEM;
222
223         adapter->rx_ring = kcalloc(adapter->num_rx_queues,
224                                    sizeof(struct igb_ring), GFP_KERNEL);
225         if (!adapter->rx_ring) {
226                 kfree(adapter->tx_ring);
227                 return -ENOMEM;
228         }
229
230         for (i = 0; i < adapter->num_rx_queues; i++) {
231                 struct igb_ring *ring = &(adapter->rx_ring[i]);
232                 ring->adapter = adapter;
233                 ring->itr_register = E1000_ITR;
234
235                 if (!ring->napi.poll)
236                         netif_napi_add(adapter->netdev, &ring->napi, igb_clean,
237                                        adapter->napi.weight /
238                                        adapter->num_rx_queues);
239         }
240         return 0;
241 }
242
243 #define IGB_N0_QUEUE -1
244 static void igb_assign_vector(struct igb_adapter *adapter, int rx_queue,
245                               int tx_queue, int msix_vector)
246 {
247         u32 msixbm = 0;
248         struct e1000_hw *hw = &adapter->hw;
249                 /* The 82575 assigns vectors using a bitmask, which matches the
250                    bitmask for the EICR/EIMS/EIMC registers.  To assign one
251                    or more queues to a vector, we write the appropriate bits
252                    into the MSIXBM register for that vector. */
253                 if (rx_queue > IGB_N0_QUEUE) {
254                         msixbm = E1000_EICR_RX_QUEUE0 << rx_queue;
255                         adapter->rx_ring[rx_queue].eims_value = msixbm;
256                 }
257                 if (tx_queue > IGB_N0_QUEUE) {
258                         msixbm |= E1000_EICR_TX_QUEUE0 << tx_queue;
259                         adapter->tx_ring[tx_queue].eims_value =
260                                   E1000_EICR_TX_QUEUE0 << tx_queue;
261                 }
262                 array_wr32(E1000_MSIXBM(0), msix_vector, msixbm);
263 }
264
265 /**
266  * igb_configure_msix - Configure MSI-X hardware
267  *
268  * igb_configure_msix sets up the hardware to properly
269  * generate MSI-X interrupts.
270  **/
271 static void igb_configure_msix(struct igb_adapter *adapter)
272 {
273         u32 tmp;
274         int i, vector = 0;
275         struct e1000_hw *hw = &adapter->hw;
276
277         adapter->eims_enable_mask = 0;
278
279         for (i = 0; i < adapter->num_tx_queues; i++) {
280                 struct igb_ring *tx_ring = &adapter->tx_ring[i];
281                 igb_assign_vector(adapter, IGB_N0_QUEUE, i, vector++);
282                 adapter->eims_enable_mask |= tx_ring->eims_value;
283                 if (tx_ring->itr_val)
284                         writel(1000000000 / (tx_ring->itr_val * 256),
285                                hw->hw_addr + tx_ring->itr_register);
286                 else
287                         writel(1, hw->hw_addr + tx_ring->itr_register);
288         }
289
290         for (i = 0; i < adapter->num_rx_queues; i++) {
291                 struct igb_ring *rx_ring = &adapter->rx_ring[i];
292                 igb_assign_vector(adapter, i, IGB_N0_QUEUE, vector++);
293                 adapter->eims_enable_mask |= rx_ring->eims_value;
294                 if (rx_ring->itr_val)
295                         writel(1000000000 / (rx_ring->itr_val * 256),
296                                hw->hw_addr + rx_ring->itr_register);
297                 else
298                         writel(1, hw->hw_addr + rx_ring->itr_register);
299         }
300
301
302         /* set vector for other causes, i.e. link changes */
303                 array_wr32(E1000_MSIXBM(0), vector++,
304                                       E1000_EIMS_OTHER);
305
306                 /* disable IAM for ICR interrupt bits */
307                 wr32(E1000_IAM, 0);
308
309                 tmp = rd32(E1000_CTRL_EXT);
310                 /* enable MSI-X PBA support*/
311                 tmp |= E1000_CTRL_EXT_PBA_CLR;
312
313                 /* Auto-Mask interrupts upon ICR read. */
314                 tmp |= E1000_CTRL_EXT_EIAME;
315                 tmp |= E1000_CTRL_EXT_IRCA;
316
317                 wr32(E1000_CTRL_EXT, tmp);
318                 adapter->eims_enable_mask |= E1000_EIMS_OTHER;
319
320         wrfl();
321 }
322
323 /**
324  * igb_request_msix - Initialize MSI-X interrupts
325  *
326  * igb_request_msix allocates MSI-X vectors and requests interrupts from the
327  * kernel.
328  **/
329 static int igb_request_msix(struct igb_adapter *adapter)
330 {
331         struct net_device *netdev = adapter->netdev;
332         int i, err = 0, vector = 0;
333
334         vector = 0;
335
336         for (i = 0; i < adapter->num_tx_queues; i++) {
337                 struct igb_ring *ring = &(adapter->tx_ring[i]);
338                 sprintf(ring->name, "%s-tx%d", netdev->name, i);
339                 err = request_irq(adapter->msix_entries[vector].vector,
340                                   &igb_msix_tx, 0, ring->name,
341                                   &(adapter->tx_ring[i]));
342                 if (err)
343                         goto out;
344                 ring->itr_register = E1000_EITR(0) + (vector << 2);
345                 ring->itr_val = adapter->itr;
346                 vector++;
347         }
348         for (i = 0; i < adapter->num_rx_queues; i++) {
349                 struct igb_ring *ring = &(adapter->rx_ring[i]);
350                 if (strlen(netdev->name) < (IFNAMSIZ - 5))
351                         sprintf(ring->name, "%s-rx%d", netdev->name, i);
352                 else
353                         memcpy(ring->name, netdev->name, IFNAMSIZ);
354                 err = request_irq(adapter->msix_entries[vector].vector,
355                                   &igb_msix_rx, 0, ring->name,
356                                   &(adapter->rx_ring[i]));
357                 if (err)
358                         goto out;
359                 ring->itr_register = E1000_EITR(0) + (vector << 2);
360                 ring->itr_val = adapter->itr;
361                 vector++;
362         }
363
364         err = request_irq(adapter->msix_entries[vector].vector,
365                           &igb_msix_other, 0, netdev->name, netdev);
366         if (err)
367                 goto out;
368
369         adapter->napi.poll = igb_clean_rx_ring_msix;
370         for (i = 0; i < adapter->num_rx_queues; i++)
371                 adapter->rx_ring[i].napi.poll = adapter->napi.poll;
372         igb_configure_msix(adapter);
373         return 0;
374 out:
375         return err;
376 }
377
378 static void igb_reset_interrupt_capability(struct igb_adapter *adapter)
379 {
380         if (adapter->msix_entries) {
381                 pci_disable_msix(adapter->pdev);
382                 kfree(adapter->msix_entries);
383                 adapter->msix_entries = NULL;
384         } else if (adapter->msi_enabled)
385                 pci_disable_msi(adapter->pdev);
386         return;
387 }
388
389
390 /**
391  * igb_set_interrupt_capability - set MSI or MSI-X if supported
392  *
393  * Attempt to configure interrupts using the best available
394  * capabilities of the hardware and kernel.
395  **/
396 static void igb_set_interrupt_capability(struct igb_adapter *adapter)
397 {
398         int err;
399         int numvecs, i;
400
401         numvecs = adapter->num_tx_queues + adapter->num_rx_queues + 1;
402         adapter->msix_entries = kcalloc(numvecs, sizeof(struct msix_entry),
403                                         GFP_KERNEL);
404         if (!adapter->msix_entries)
405                 goto msi_only;
406
407         for (i = 0; i < numvecs; i++)
408                 adapter->msix_entries[i].entry = i;
409
410         err = pci_enable_msix(adapter->pdev,
411                               adapter->msix_entries,
412                               numvecs);
413         if (err == 0)
414                 return;
415
416         igb_reset_interrupt_capability(adapter);
417
418         /* If we can't do MSI-X, try MSI */
419 msi_only:
420         adapter->num_rx_queues = 1;
421         if (!pci_enable_msi(adapter->pdev))
422                 adapter->msi_enabled = 1;
423         return;
424 }
425
426 /**
427  * igb_request_irq - initialize interrupts
428  *
429  * Attempts to configure interrupts using the best available
430  * capabilities of the hardware and kernel.
431  **/
432 static int igb_request_irq(struct igb_adapter *adapter)
433 {
434         struct net_device *netdev = adapter->netdev;
435         struct e1000_hw *hw = &adapter->hw;
436         int err = 0;
437
438         if (adapter->msix_entries) {
439                 err = igb_request_msix(adapter);
440                 if (!err) {
441                         /* enable IAM, auto-mask,
442                          * DO NOT USE EIAME or IAME in legacy mode */
443                         wr32(E1000_IAM, IMS_ENABLE_MASK);
444                         goto request_done;
445                 }
446                 /* fall back to MSI */
447                 igb_reset_interrupt_capability(adapter);
448                 if (!pci_enable_msi(adapter->pdev))
449                         adapter->msi_enabled = 1;
450                 igb_free_all_tx_resources(adapter);
451                 igb_free_all_rx_resources(adapter);
452                 adapter->num_rx_queues = 1;
453                 igb_alloc_queues(adapter);
454         }
455         if (adapter->msi_enabled) {
456                 err = request_irq(adapter->pdev->irq, &igb_intr_msi, 0,
457                                   netdev->name, netdev);
458                 if (!err)
459                         goto request_done;
460                 /* fall back to legacy interrupts */
461                 igb_reset_interrupt_capability(adapter);
462                 adapter->msi_enabled = 0;
463         }
464
465         err = request_irq(adapter->pdev->irq, &igb_intr, IRQF_SHARED,
466                           netdev->name, netdev);
467
468         if (err) {
469                 dev_err(&adapter->pdev->dev, "Error %d getting interrupt\n",
470                         err);
471                 goto request_done;
472         }
473
474         /* enable IAM, auto-mask */
475         wr32(E1000_IAM, IMS_ENABLE_MASK);
476
477 request_done:
478         return err;
479 }
480
481 static void igb_free_irq(struct igb_adapter *adapter)
482 {
483         struct net_device *netdev = adapter->netdev;
484
485         if (adapter->msix_entries) {
486                 int vector = 0, i;
487
488                 for (i = 0; i < adapter->num_tx_queues; i++)
489                         free_irq(adapter->msix_entries[vector++].vector,
490                                 &(adapter->tx_ring[i]));
491                 for (i = 0; i < adapter->num_rx_queues; i++)
492                         free_irq(adapter->msix_entries[vector++].vector,
493                                 &(adapter->rx_ring[i]));
494
495                 free_irq(adapter->msix_entries[vector++].vector, netdev);
496                 return;
497         }
498
499         free_irq(adapter->pdev->irq, netdev);
500 }
501
502 /**
503  * igb_irq_disable - Mask off interrupt generation on the NIC
504  * @adapter: board private structure
505  **/
506 static void igb_irq_disable(struct igb_adapter *adapter)
507 {
508         struct e1000_hw *hw = &adapter->hw;
509
510         if (adapter->msix_entries) {
511                 wr32(E1000_EIMC, ~0);
512                 wr32(E1000_EIAC, 0);
513         }
514         wr32(E1000_IMC, ~0);
515         wrfl();
516         synchronize_irq(adapter->pdev->irq);
517 }
518
519 /**
520  * igb_irq_enable - Enable default interrupt generation settings
521  * @adapter: board private structure
522  **/
523 static void igb_irq_enable(struct igb_adapter *adapter)
524 {
525         struct e1000_hw *hw = &adapter->hw;
526
527         if (adapter->msix_entries) {
528                 wr32(E1000_EIMS,
529                                 adapter->eims_enable_mask);
530                 wr32(E1000_EIAC,
531                                 adapter->eims_enable_mask);
532                 wr32(E1000_IMS, E1000_IMS_LSC);
533         } else
534         wr32(E1000_IMS, IMS_ENABLE_MASK);
535 }
536
537 static void igb_update_mng_vlan(struct igb_adapter *adapter)
538 {
539         struct net_device *netdev = adapter->netdev;
540         u16 vid = adapter->hw.mng_cookie.vlan_id;
541         u16 old_vid = adapter->mng_vlan_id;
542         if (adapter->vlgrp) {
543                 if (!vlan_group_get_device(adapter->vlgrp, vid)) {
544                         if (adapter->hw.mng_cookie.status &
545                                 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
546                                 igb_vlan_rx_add_vid(netdev, vid);
547                                 adapter->mng_vlan_id = vid;
548                         } else
549                                 adapter->mng_vlan_id = IGB_MNG_VLAN_NONE;
550
551                         if ((old_vid != (u16)IGB_MNG_VLAN_NONE) &&
552                                         (vid != old_vid) &&
553                             !vlan_group_get_device(adapter->vlgrp, old_vid))
554                                 igb_vlan_rx_kill_vid(netdev, old_vid);
555                 } else
556                         adapter->mng_vlan_id = vid;
557         }
558 }
559
560 /**
561  * igb_release_hw_control - release control of the h/w to f/w
562  * @adapter: address of board private structure
563  *
564  * igb_release_hw_control resets CTRL_EXT:DRV_LOAD bit.
565  * For ASF and Pass Through versions of f/w this means that the
566  * driver is no longer loaded.
567  *
568  **/
569 static void igb_release_hw_control(struct igb_adapter *adapter)
570 {
571         struct e1000_hw *hw = &adapter->hw;
572         u32 ctrl_ext;
573
574         /* Let firmware take over control of h/w */
575         ctrl_ext = rd32(E1000_CTRL_EXT);
576         wr32(E1000_CTRL_EXT,
577                         ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
578 }
579
580
581 /**
582  * igb_get_hw_control - get control of the h/w from f/w
583  * @adapter: address of board private structure
584  *
585  * igb_get_hw_control sets CTRL_EXT:DRV_LOAD bit.
586  * For ASF and Pass Through versions of f/w this means that
587  * the driver is loaded.
588  *
589  **/
590 static void igb_get_hw_control(struct igb_adapter *adapter)
591 {
592         struct e1000_hw *hw = &adapter->hw;
593         u32 ctrl_ext;
594
595         /* Let firmware know the driver has taken over */
596         ctrl_ext = rd32(E1000_CTRL_EXT);
597         wr32(E1000_CTRL_EXT,
598                         ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
599 }
600
601 static void igb_init_manageability(struct igb_adapter *adapter)
602 {
603         struct e1000_hw *hw = &adapter->hw;
604
605         if (adapter->en_mng_pt) {
606                 u32 manc2h = rd32(E1000_MANC2H);
607                 u32 manc = rd32(E1000_MANC);
608
609                 /* enable receiving management packets to the host */
610                 /* this will probably generate destination unreachable messages
611                  * from the host OS, but the packets will be handled on SMBUS */
612                 manc |= E1000_MANC_EN_MNG2HOST;
613 #define E1000_MNG2HOST_PORT_623 (1 << 5)
614 #define E1000_MNG2HOST_PORT_664 (1 << 6)
615                 manc2h |= E1000_MNG2HOST_PORT_623;
616                 manc2h |= E1000_MNG2HOST_PORT_664;
617                 wr32(E1000_MANC2H, manc2h);
618
619                 wr32(E1000_MANC, manc);
620         }
621 }
622
623 /**
624  * igb_configure - configure the hardware for RX and TX
625  * @adapter: private board structure
626  **/
627 static void igb_configure(struct igb_adapter *adapter)
628 {
629         struct net_device *netdev = adapter->netdev;
630         int i;
631
632         igb_get_hw_control(adapter);
633         igb_set_multi(netdev);
634
635         igb_restore_vlan(adapter);
636         igb_init_manageability(adapter);
637
638         igb_configure_tx(adapter);
639         igb_setup_rctl(adapter);
640         igb_configure_rx(adapter);
641         /* call IGB_DESC_UNUSED which always leaves
642          * at least 1 descriptor unused to make sure
643          * next_to_use != next_to_clean */
644         for (i = 0; i < adapter->num_rx_queues; i++) {
645                 struct igb_ring *ring = &adapter->rx_ring[i];
646                 igb_alloc_rx_buffers_adv(adapter, ring, IGB_DESC_UNUSED(ring));
647         }
648
649
650         adapter->tx_queue_len = netdev->tx_queue_len;
651 }
652
653
654 /**
655  * igb_up - Open the interface and prepare it to handle traffic
656  * @adapter: board private structure
657  **/
658
659 int igb_up(struct igb_adapter *adapter)
660 {
661         struct e1000_hw *hw = &adapter->hw;
662         int i;
663
664         /* hardware has been reset, we need to reload some things */
665         igb_configure(adapter);
666
667         clear_bit(__IGB_DOWN, &adapter->state);
668
669         napi_enable(&adapter->napi);
670
671         if (adapter->msix_entries) {
672                 for (i = 0; i < adapter->num_rx_queues; i++)
673                         napi_enable(&adapter->rx_ring[i].napi);
674                 igb_configure_msix(adapter);
675         }
676
677         /* Clear any pending interrupts. */
678         rd32(E1000_ICR);
679         igb_irq_enable(adapter);
680
681         /* Fire a link change interrupt to start the watchdog. */
682         wr32(E1000_ICS, E1000_ICS_LSC);
683         return 0;
684 }
685
686 void igb_down(struct igb_adapter *adapter)
687 {
688         struct e1000_hw *hw = &adapter->hw;
689         struct net_device *netdev = adapter->netdev;
690         u32 tctl, rctl;
691         int i;
692
693         /* signal that we're down so the interrupt handler does not
694          * reschedule our watchdog timer */
695         set_bit(__IGB_DOWN, &adapter->state);
696
697         /* disable receives in the hardware */
698         rctl = rd32(E1000_RCTL);
699         wr32(E1000_RCTL, rctl & ~E1000_RCTL_EN);
700         /* flush and sleep below */
701
702         netif_stop_queue(netdev);
703
704         /* disable transmits in the hardware */
705         tctl = rd32(E1000_TCTL);
706         tctl &= ~E1000_TCTL_EN;
707         wr32(E1000_TCTL, tctl);
708         /* flush both disables and wait for them to finish */
709         wrfl();
710         msleep(10);
711
712         napi_disable(&adapter->napi);
713
714         if (adapter->msix_entries)
715                 for (i = 0; i < adapter->num_rx_queues; i++)
716                         napi_disable(&adapter->rx_ring[i].napi);
717         igb_irq_disable(adapter);
718
719         del_timer_sync(&adapter->watchdog_timer);
720         del_timer_sync(&adapter->phy_info_timer);
721
722         netdev->tx_queue_len = adapter->tx_queue_len;
723         netif_carrier_off(netdev);
724         adapter->link_speed = 0;
725         adapter->link_duplex = 0;
726
727         igb_reset(adapter);
728         igb_clean_all_tx_rings(adapter);
729         igb_clean_all_rx_rings(adapter);
730 }
731
732 void igb_reinit_locked(struct igb_adapter *adapter)
733 {
734         WARN_ON(in_interrupt());
735         while (test_and_set_bit(__IGB_RESETTING, &adapter->state))
736                 msleep(1);
737         igb_down(adapter);
738         igb_up(adapter);
739         clear_bit(__IGB_RESETTING, &adapter->state);
740 }
741
742 void igb_reset(struct igb_adapter *adapter)
743 {
744         struct e1000_hw *hw = &adapter->hw;
745         struct e1000_fc_info *fc = &adapter->hw.fc;
746         u32 pba = 0, tx_space, min_tx_space, min_rx_space;
747         u16 hwm;
748
749         /* Repartition Pba for greater than 9k mtu
750          * To take effect CTRL.RST is required.
751          */
752         pba = E1000_PBA_34K;
753
754         if (adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) {
755                 /* adjust PBA for jumbo frames */
756                 wr32(E1000_PBA, pba);
757
758                 /* To maintain wire speed transmits, the Tx FIFO should be
759                  * large enough to accommodate two full transmit packets,
760                  * rounded up to the next 1KB and expressed in KB.  Likewise,
761                  * the Rx FIFO should be large enough to accommodate at least
762                  * one full receive packet and is similarly rounded up and
763                  * expressed in KB. */
764                 pba = rd32(E1000_PBA);
765                 /* upper 16 bits has Tx packet buffer allocation size in KB */
766                 tx_space = pba >> 16;
767                 /* lower 16 bits has Rx packet buffer allocation size in KB */
768                 pba &= 0xffff;
769                 /* the tx fifo also stores 16 bytes of information about the tx
770                  * but don't include ethernet FCS because hardware appends it */
771                 min_tx_space = (adapter->max_frame_size +
772                                 sizeof(struct e1000_tx_desc) -
773                                 ETH_FCS_LEN) * 2;
774                 min_tx_space = ALIGN(min_tx_space, 1024);
775                 min_tx_space >>= 10;
776                 /* software strips receive CRC, so leave room for it */
777                 min_rx_space = adapter->max_frame_size;
778                 min_rx_space = ALIGN(min_rx_space, 1024);
779                 min_rx_space >>= 10;
780
781                 /* If current Tx allocation is less than the min Tx FIFO size,
782                  * and the min Tx FIFO size is less than the current Rx FIFO
783                  * allocation, take space away from current Rx allocation */
784                 if (tx_space < min_tx_space &&
785                     ((min_tx_space - tx_space) < pba)) {
786                         pba = pba - (min_tx_space - tx_space);
787
788                         /* if short on rx space, rx wins and must trump tx
789                          * adjustment */
790                         if (pba < min_rx_space)
791                                 pba = min_rx_space;
792                 }
793         }
794         wr32(E1000_PBA, pba);
795
796         /* flow control settings */
797         /* The high water mark must be low enough to fit one full frame
798          * (or the size used for early receive) above it in the Rx FIFO.
799          * Set it to the lower of:
800          * - 90% of the Rx FIFO size, or
801          * - the full Rx FIFO size minus one full frame */
802         hwm = min(((pba << 10) * 9 / 10),
803                   ((pba << 10) - adapter->max_frame_size));
804
805         fc->high_water = hwm & 0xFFF8;  /* 8-byte granularity */
806         fc->low_water = fc->high_water - 8;
807         fc->pause_time = 0xFFFF;
808         fc->send_xon = 1;
809         fc->type = fc->original_type;
810
811         /* Allow time for pending master requests to run */
812         adapter->hw.mac.ops.reset_hw(&adapter->hw);
813         wr32(E1000_WUC, 0);
814
815         if (adapter->hw.mac.ops.init_hw(&adapter->hw))
816                 dev_err(&adapter->pdev->dev, "Hardware Error\n");
817
818         igb_update_mng_vlan(adapter);
819
820         /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
821         wr32(E1000_VET, ETHERNET_IEEE_VLAN_TYPE);
822
823         igb_reset_adaptive(&adapter->hw);
824         adapter->hw.phy.ops.get_phy_info(&adapter->hw);
825 }
826
827 /**
828  * igb_probe - Device Initialization Routine
829  * @pdev: PCI device information struct
830  * @ent: entry in igb_pci_tbl
831  *
832  * Returns 0 on success, negative on failure
833  *
834  * igb_probe initializes an adapter identified by a pci_dev structure.
835  * The OS initialization, configuring of the adapter private structure,
836  * and a hardware reset occur.
837  **/
838 static int __devinit igb_probe(struct pci_dev *pdev,
839                                const struct pci_device_id *ent)
840 {
841         struct net_device *netdev;
842         struct igb_adapter *adapter;
843         struct e1000_hw *hw;
844         const struct e1000_info *ei = igb_info_tbl[ent->driver_data];
845         unsigned long mmio_start, mmio_len;
846         static int cards_found;
847         int i, err, pci_using_dac;
848         u16 eeprom_data = 0;
849         u16 eeprom_apme_mask = IGB_EEPROM_APME;
850         u32 part_num;
851
852         err = pci_enable_device(pdev);
853         if (err)
854                 return err;
855
856         pci_using_dac = 0;
857         err = pci_set_dma_mask(pdev, DMA_64BIT_MASK);
858         if (!err) {
859                 err = pci_set_consistent_dma_mask(pdev, DMA_64BIT_MASK);
860                 if (!err)
861                         pci_using_dac = 1;
862         } else {
863                 err = pci_set_dma_mask(pdev, DMA_32BIT_MASK);
864                 if (err) {
865                         err = pci_set_consistent_dma_mask(pdev, DMA_32BIT_MASK);
866                         if (err) {
867                                 dev_err(&pdev->dev, "No usable DMA "
868                                         "configuration, aborting\n");
869                                 goto err_dma;
870                         }
871                 }
872         }
873
874         err = pci_request_regions(pdev, igb_driver_name);
875         if (err)
876                 goto err_pci_reg;
877
878         pci_set_master(pdev);
879
880         err = -ENOMEM;
881         netdev = alloc_etherdev(sizeof(struct igb_adapter));
882         if (!netdev)
883                 goto err_alloc_etherdev;
884
885         SET_NETDEV_DEV(netdev, &pdev->dev);
886
887         pci_set_drvdata(pdev, netdev);
888         adapter = netdev_priv(netdev);
889         adapter->netdev = netdev;
890         adapter->pdev = pdev;
891         hw = &adapter->hw;
892         hw->back = adapter;
893         adapter->msg_enable = NETIF_MSG_DRV | NETIF_MSG_PROBE;
894
895         mmio_start = pci_resource_start(pdev, 0);
896         mmio_len = pci_resource_len(pdev, 0);
897
898         err = -EIO;
899         adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
900         if (!adapter->hw.hw_addr)
901                 goto err_ioremap;
902
903         netdev->open = &igb_open;
904         netdev->stop = &igb_close;
905         netdev->get_stats = &igb_get_stats;
906         netdev->set_multicast_list = &igb_set_multi;
907         netdev->set_mac_address = &igb_set_mac;
908         netdev->change_mtu = &igb_change_mtu;
909         netdev->do_ioctl = &igb_ioctl;
910         igb_set_ethtool_ops(netdev);
911         netdev->tx_timeout = &igb_tx_timeout;
912         netdev->watchdog_timeo = 5 * HZ;
913         netif_napi_add(netdev, &adapter->napi, igb_clean, 64);
914         netdev->vlan_rx_register = igb_vlan_rx_register;
915         netdev->vlan_rx_add_vid = igb_vlan_rx_add_vid;
916         netdev->vlan_rx_kill_vid = igb_vlan_rx_kill_vid;
917 #ifdef CONFIG_NET_POLL_CONTROLLER
918         netdev->poll_controller = igb_netpoll;
919 #endif
920         netdev->hard_start_xmit = &igb_xmit_frame_adv;
921
922         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
923
924         netdev->mem_start = mmio_start;
925         netdev->mem_end = mmio_start + mmio_len;
926
927         adapter->bd_number = cards_found;
928
929         /* PCI config space info */
930         hw->vendor_id = pdev->vendor;
931         hw->device_id = pdev->device;
932         hw->revision_id = pdev->revision;
933         hw->subsystem_vendor_id = pdev->subsystem_vendor;
934         hw->subsystem_device_id = pdev->subsystem_device;
935
936         /* setup the private structure */
937         hw->back = adapter;
938         /* Copy the default MAC, PHY and NVM function pointers */
939         memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
940         memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
941         memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
942         /* Initialize skew-specific constants */
943         err = ei->get_invariants(hw);
944         if (err)
945                 goto err_hw_init;
946
947         err = igb_sw_init(adapter);
948         if (err)
949                 goto err_sw_init;
950
951         igb_get_bus_info_pcie(hw);
952
953         hw->phy.autoneg_wait_to_complete = false;
954         hw->mac.adaptive_ifs = true;
955
956         /* Copper options */
957         if (hw->phy.media_type == e1000_media_type_copper) {
958                 hw->phy.mdix = AUTO_ALL_MODES;
959                 hw->phy.disable_polarity_correction = false;
960                 hw->phy.ms_type = e1000_ms_hw_default;
961         }
962
963         if (igb_check_reset_block(hw))
964                 dev_info(&pdev->dev,
965                         "PHY reset is blocked due to SOL/IDER session.\n");
966
967         netdev->features = NETIF_F_SG |
968                            NETIF_F_HW_CSUM |
969                            NETIF_F_HW_VLAN_TX |
970                            NETIF_F_HW_VLAN_RX |
971                            NETIF_F_HW_VLAN_FILTER;
972
973         netdev->features |= NETIF_F_TSO;
974
975         netdev->features |= NETIF_F_TSO6;
976         if (pci_using_dac)
977                 netdev->features |= NETIF_F_HIGHDMA;
978
979         netdev->features |= NETIF_F_LLTX;
980         adapter->en_mng_pt = igb_enable_mng_pass_thru(&adapter->hw);
981
982         /* before reading the NVM, reset the controller to put the device in a
983          * known good starting state */
984         hw->mac.ops.reset_hw(hw);
985
986         /* make sure the NVM is good */
987         if (igb_validate_nvm_checksum(hw) < 0) {
988                 dev_err(&pdev->dev, "The NVM Checksum Is Not Valid\n");
989                 err = -EIO;
990                 goto err_eeprom;
991         }
992
993         /* copy the MAC address out of the NVM */
994         if (hw->mac.ops.read_mac_addr(hw))
995                 dev_err(&pdev->dev, "NVM Read Error\n");
996
997         memcpy(netdev->dev_addr, hw->mac.addr, netdev->addr_len);
998         memcpy(netdev->perm_addr, hw->mac.addr, netdev->addr_len);
999
1000         if (!is_valid_ether_addr(netdev->perm_addr)) {
1001                 dev_err(&pdev->dev, "Invalid MAC Address\n");
1002                 err = -EIO;
1003                 goto err_eeprom;
1004         }
1005
1006         init_timer(&adapter->watchdog_timer);
1007         adapter->watchdog_timer.function = &igb_watchdog;
1008         adapter->watchdog_timer.data = (unsigned long) adapter;
1009
1010         init_timer(&adapter->phy_info_timer);
1011         adapter->phy_info_timer.function = &igb_update_phy_info;
1012         adapter->phy_info_timer.data = (unsigned long) adapter;
1013
1014         INIT_WORK(&adapter->reset_task, igb_reset_task);
1015         INIT_WORK(&adapter->watchdog_task, igb_watchdog_task);
1016
1017         /* Initialize link & ring properties that are user-changeable */
1018         adapter->tx_ring->count = 256;
1019         for (i = 0; i < adapter->num_tx_queues; i++)
1020                 adapter->tx_ring[i].count = adapter->tx_ring->count;
1021         adapter->rx_ring->count = 256;
1022         for (i = 0; i < adapter->num_rx_queues; i++)
1023                 adapter->rx_ring[i].count = adapter->rx_ring->count;
1024
1025         adapter->fc_autoneg = true;
1026         hw->mac.autoneg = true;
1027         hw->phy.autoneg_advertised = 0x2f;
1028
1029         hw->fc.original_type = e1000_fc_default;
1030         hw->fc.type = e1000_fc_default;
1031
1032         adapter->itr_setting = 3;
1033         adapter->itr = IGB_START_ITR;
1034
1035         igb_validate_mdi_setting(hw);
1036
1037         adapter->rx_csum = 1;
1038
1039         /* Initial Wake on LAN setting If APM wake is enabled in the EEPROM,
1040          * enable the ACPI Magic Packet filter
1041          */
1042
1043         if (hw->bus.func == 0 ||
1044             hw->device_id == E1000_DEV_ID_82575EB_COPPER)
1045                 hw->nvm.ops.read_nvm(hw, NVM_INIT_CONTROL3_PORT_A, 1,
1046                                      &eeprom_data);
1047
1048         if (eeprom_data & eeprom_apme_mask)
1049                 adapter->eeprom_wol |= E1000_WUFC_MAG;
1050
1051         /* now that we have the eeprom settings, apply the special cases where
1052          * the eeprom may be wrong or the board simply won't support wake on
1053          * lan on a particular port */
1054         switch (pdev->device) {
1055         case E1000_DEV_ID_82575GB_QUAD_COPPER:
1056                 adapter->eeprom_wol = 0;
1057                 break;
1058         case E1000_DEV_ID_82575EB_FIBER_SERDES:
1059                 /* Wake events only supported on port A for dual fiber
1060                  * regardless of eeprom setting */
1061                 if (rd32(E1000_STATUS) & E1000_STATUS_FUNC_1)
1062                         adapter->eeprom_wol = 0;
1063                 break;
1064         }
1065
1066         /* initialize the wol settings based on the eeprom settings */
1067         adapter->wol = adapter->eeprom_wol;
1068
1069         /* reset the hardware with the new settings */
1070         igb_reset(adapter);
1071
1072         /* let the f/w know that the h/w is now under the control of the
1073          * driver. */
1074         igb_get_hw_control(adapter);
1075
1076         /* tell the stack to leave us alone until igb_open() is called */
1077         netif_carrier_off(netdev);
1078         netif_stop_queue(netdev);
1079
1080         strcpy(netdev->name, "eth%d");
1081         err = register_netdev(netdev);
1082         if (err)
1083                 goto err_register;
1084
1085         dev_info(&pdev->dev, "Intel(R) Gigabit Ethernet Network Connection\n");
1086         /* print bus type/speed/width info */
1087         dev_info(&pdev->dev,
1088                  "%s: (PCIe:%s:%s) %02x:%02x:%02x:%02x:%02x:%02x\n",
1089                  netdev->name,
1090                  ((hw->bus.speed == e1000_bus_speed_2500)
1091                   ? "2.5Gb/s" : "unknown"),
1092                  ((hw->bus.width == e1000_bus_width_pcie_x4)
1093                   ? "Width x4" : (hw->bus.width == e1000_bus_width_pcie_x1)
1094                   ? "Width x1" : "unknown"),
1095                  netdev->dev_addr[0], netdev->dev_addr[1], netdev->dev_addr[2],
1096                  netdev->dev_addr[3], netdev->dev_addr[4], netdev->dev_addr[5]);
1097
1098         igb_read_part_num(hw, &part_num);
1099         dev_info(&pdev->dev, "%s: PBA No: %06x-%03x\n", netdev->name,
1100                 (part_num >> 8), (part_num & 0xff));
1101
1102         dev_info(&pdev->dev,
1103                 "Using %s interrupts. %d rx queue(s), %d tx queue(s)\n",
1104                 adapter->msix_entries ? "MSI-X" :
1105                 adapter->msi_enabled ? "MSI" : "legacy",
1106                 adapter->num_rx_queues, adapter->num_tx_queues);
1107
1108         cards_found++;
1109         return 0;
1110
1111 err_register:
1112         igb_release_hw_control(adapter);
1113 err_eeprom:
1114         if (!igb_check_reset_block(hw))
1115                 hw->phy.ops.reset_phy(hw);
1116
1117         if (hw->flash_address)
1118                 iounmap(hw->flash_address);
1119
1120         igb_remove_device(hw);
1121         kfree(adapter->tx_ring);
1122         kfree(adapter->rx_ring);
1123 err_sw_init:
1124 err_hw_init:
1125         iounmap(hw->hw_addr);
1126 err_ioremap:
1127         free_netdev(netdev);
1128 err_alloc_etherdev:
1129         pci_release_regions(pdev);
1130 err_pci_reg:
1131 err_dma:
1132         pci_disable_device(pdev);
1133         return err;
1134 }
1135
1136 /**
1137  * igb_remove - Device Removal Routine
1138  * @pdev: PCI device information struct
1139  *
1140  * igb_remove is called by the PCI subsystem to alert the driver
1141  * that it should release a PCI device.  The could be caused by a
1142  * Hot-Plug event, or because the driver is going to be removed from
1143  * memory.
1144  **/
1145 static void __devexit igb_remove(struct pci_dev *pdev)
1146 {
1147         struct net_device *netdev = pci_get_drvdata(pdev);
1148         struct igb_adapter *adapter = netdev_priv(netdev);
1149
1150         /* flush_scheduled work may reschedule our watchdog task, so
1151          * explicitly disable watchdog tasks from being rescheduled  */
1152         set_bit(__IGB_DOWN, &adapter->state);
1153         del_timer_sync(&adapter->watchdog_timer);
1154         del_timer_sync(&adapter->phy_info_timer);
1155
1156         flush_scheduled_work();
1157
1158         /* Release control of h/w to f/w.  If f/w is AMT enabled, this
1159          * would have already happened in close and is redundant. */
1160         igb_release_hw_control(adapter);
1161
1162         unregister_netdev(netdev);
1163
1164         if (!igb_check_reset_block(&adapter->hw))
1165                 adapter->hw.phy.ops.reset_phy(&adapter->hw);
1166
1167         igb_remove_device(&adapter->hw);
1168         igb_reset_interrupt_capability(adapter);
1169
1170         kfree(adapter->tx_ring);
1171         kfree(adapter->rx_ring);
1172
1173         iounmap(adapter->hw.hw_addr);
1174         if (adapter->hw.flash_address)
1175                 iounmap(adapter->hw.flash_address);
1176         pci_release_regions(pdev);
1177
1178         free_netdev(netdev);
1179
1180         pci_disable_device(pdev);
1181 }
1182
1183 /**
1184  * igb_sw_init - Initialize general software structures (struct igb_adapter)
1185  * @adapter: board private structure to initialize
1186  *
1187  * igb_sw_init initializes the Adapter private data structure.
1188  * Fields are initialized based on PCI device information and
1189  * OS network device settings (MTU size).
1190  **/
1191 static int __devinit igb_sw_init(struct igb_adapter *adapter)
1192 {
1193         struct e1000_hw *hw = &adapter->hw;
1194         struct net_device *netdev = adapter->netdev;
1195         struct pci_dev *pdev = adapter->pdev;
1196
1197         pci_read_config_word(pdev, PCI_COMMAND, &hw->bus.pci_cmd_word);
1198
1199         adapter->rx_buffer_len = MAXIMUM_ETHERNET_VLAN_SIZE;
1200         adapter->rx_ps_hdr_size = 0; /* disable packet split */
1201         adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
1202         adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
1203
1204         /* Number of supported queues. */
1205         /* Having more queues than CPUs doesn't make sense. */
1206         adapter->num_tx_queues = 1;
1207         adapter->num_rx_queues = min(IGB_MAX_RX_QUEUES, num_online_cpus());
1208
1209         igb_set_interrupt_capability(adapter);
1210
1211         if (igb_alloc_queues(adapter)) {
1212                 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
1213                 return -ENOMEM;
1214         }
1215
1216         /* Explicitly disable IRQ since the NIC can be in any state. */
1217         igb_irq_disable(adapter);
1218
1219         set_bit(__IGB_DOWN, &adapter->state);
1220         return 0;
1221 }
1222
1223 /**
1224  * igb_open - Called when a network interface is made active
1225  * @netdev: network interface device structure
1226  *
1227  * Returns 0 on success, negative value on failure
1228  *
1229  * The open entry point is called when a network interface is made
1230  * active by the system (IFF_UP).  At this point all resources needed
1231  * for transmit and receive operations are allocated, the interrupt
1232  * handler is registered with the OS, the watchdog timer is started,
1233  * and the stack is notified that the interface is ready.
1234  **/
1235 static int igb_open(struct net_device *netdev)
1236 {
1237         struct igb_adapter *adapter = netdev_priv(netdev);
1238         struct e1000_hw *hw = &adapter->hw;
1239         int err;
1240         int i;
1241
1242         /* disallow open during test */
1243         if (test_bit(__IGB_TESTING, &adapter->state))
1244                 return -EBUSY;
1245
1246         /* allocate transmit descriptors */
1247         err = igb_setup_all_tx_resources(adapter);
1248         if (err)
1249                 goto err_setup_tx;
1250
1251         /* allocate receive descriptors */
1252         err = igb_setup_all_rx_resources(adapter);
1253         if (err)
1254                 goto err_setup_rx;
1255
1256         /* e1000_power_up_phy(adapter); */
1257
1258         adapter->mng_vlan_id = IGB_MNG_VLAN_NONE;
1259         if ((adapter->hw.mng_cookie.status &
1260              E1000_MNG_DHCP_COOKIE_STATUS_VLAN))
1261                 igb_update_mng_vlan(adapter);
1262
1263         /* before we allocate an interrupt, we must be ready to handle it.
1264          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
1265          * as soon as we call pci_request_irq, so we have to setup our
1266          * clean_rx handler before we do so.  */
1267         igb_configure(adapter);
1268
1269         err = igb_request_irq(adapter);
1270         if (err)
1271                 goto err_req_irq;
1272
1273         /* From here on the code is the same as igb_up() */
1274         clear_bit(__IGB_DOWN, &adapter->state);
1275
1276         napi_enable(&adapter->napi);
1277         if (adapter->msix_entries)
1278                 for (i = 0; i < adapter->num_rx_queues; i++)
1279                         napi_enable(&adapter->rx_ring[i].napi);
1280
1281         igb_irq_enable(adapter);
1282
1283         /* Clear any pending interrupts. */
1284         rd32(E1000_ICR);
1285         /* Fire a link status change interrupt to start the watchdog. */
1286         wr32(E1000_ICS, E1000_ICS_LSC);
1287
1288         return 0;
1289
1290 err_req_irq:
1291         igb_release_hw_control(adapter);
1292         /* e1000_power_down_phy(adapter); */
1293         igb_free_all_rx_resources(adapter);
1294 err_setup_rx:
1295         igb_free_all_tx_resources(adapter);
1296 err_setup_tx:
1297         igb_reset(adapter);
1298
1299         return err;
1300 }
1301
1302 /**
1303  * igb_close - Disables a network interface
1304  * @netdev: network interface device structure
1305  *
1306  * Returns 0, this is not allowed to fail
1307  *
1308  * The close entry point is called when an interface is de-activated
1309  * by the OS.  The hardware is still under the driver's control, but
1310  * needs to be disabled.  A global MAC reset is issued to stop the
1311  * hardware, and all transmit and receive resources are freed.
1312  **/
1313 static int igb_close(struct net_device *netdev)
1314 {
1315         struct igb_adapter *adapter = netdev_priv(netdev);
1316
1317         WARN_ON(test_bit(__IGB_RESETTING, &adapter->state));
1318         igb_down(adapter);
1319
1320         igb_free_irq(adapter);
1321
1322         igb_free_all_tx_resources(adapter);
1323         igb_free_all_rx_resources(adapter);
1324
1325         /* kill manageability vlan ID if supported, but not if a vlan with
1326          * the same ID is registered on the host OS (let 8021q kill it) */
1327         if ((adapter->hw.mng_cookie.status &
1328                           E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
1329              !(adapter->vlgrp &&
1330                vlan_group_get_device(adapter->vlgrp, adapter->mng_vlan_id)))
1331                 igb_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
1332
1333         return 0;
1334 }
1335
1336 /**
1337  * igb_setup_tx_resources - allocate Tx resources (Descriptors)
1338  * @adapter: board private structure
1339  * @tx_ring: tx descriptor ring (for a specific queue) to setup
1340  *
1341  * Return 0 on success, negative on failure
1342  **/
1343
1344 int igb_setup_tx_resources(struct igb_adapter *adapter,
1345                            struct igb_ring *tx_ring)
1346 {
1347         struct pci_dev *pdev = adapter->pdev;
1348         int size;
1349
1350         size = sizeof(struct igb_buffer) * tx_ring->count;
1351         tx_ring->buffer_info = vmalloc(size);
1352         if (!tx_ring->buffer_info)
1353                 goto err;
1354         memset(tx_ring->buffer_info, 0, size);
1355
1356         /* round up to nearest 4K */
1357         tx_ring->size = tx_ring->count * sizeof(struct e1000_tx_desc)
1358                         + sizeof(u32);
1359         tx_ring->size = ALIGN(tx_ring->size, 4096);
1360
1361         tx_ring->desc = pci_alloc_consistent(pdev, tx_ring->size,
1362                                              &tx_ring->dma);
1363
1364         if (!tx_ring->desc)
1365                 goto err;
1366
1367         tx_ring->adapter = adapter;
1368         tx_ring->next_to_use = 0;
1369         tx_ring->next_to_clean = 0;
1370         spin_lock_init(&tx_ring->tx_clean_lock);
1371         spin_lock_init(&tx_ring->tx_lock);
1372         return 0;
1373
1374 err:
1375         vfree(tx_ring->buffer_info);
1376         dev_err(&adapter->pdev->dev,
1377                 "Unable to allocate memory for the transmit descriptor ring\n");
1378         return -ENOMEM;
1379 }
1380
1381 /**
1382  * igb_setup_all_tx_resources - wrapper to allocate Tx resources
1383  *                                (Descriptors) for all queues
1384  * @adapter: board private structure
1385  *
1386  * Return 0 on success, negative on failure
1387  **/
1388 static int igb_setup_all_tx_resources(struct igb_adapter *adapter)
1389 {
1390         int i, err = 0;
1391
1392         for (i = 0; i < adapter->num_tx_queues; i++) {
1393                 err = igb_setup_tx_resources(adapter, &adapter->tx_ring[i]);
1394                 if (err) {
1395                         dev_err(&adapter->pdev->dev,
1396                                 "Allocation for Tx Queue %u failed\n", i);
1397                         for (i--; i >= 0; i--)
1398                                 igb_free_tx_resources(adapter,
1399                                                         &adapter->tx_ring[i]);
1400                         break;
1401                 }
1402         }
1403
1404         return err;
1405 }
1406
1407 /**
1408  * igb_configure_tx - Configure transmit Unit after Reset
1409  * @adapter: board private structure
1410  *
1411  * Configure the Tx unit of the MAC after a reset.
1412  **/
1413 static void igb_configure_tx(struct igb_adapter *adapter)
1414 {
1415         u64 tdba, tdwba;
1416         struct e1000_hw *hw = &adapter->hw;
1417         u32 tctl;
1418         u32 txdctl, txctrl;
1419         int i;
1420
1421         for (i = 0; i < adapter->num_tx_queues; i++) {
1422                 struct igb_ring *ring = &(adapter->tx_ring[i]);
1423
1424                 wr32(E1000_TDLEN(i),
1425                                 ring->count * sizeof(struct e1000_tx_desc));
1426                 tdba = ring->dma;
1427                 wr32(E1000_TDBAL(i),
1428                                 tdba & 0x00000000ffffffffULL);
1429                 wr32(E1000_TDBAH(i), tdba >> 32);
1430
1431                 tdwba = ring->dma + ring->count * sizeof(struct e1000_tx_desc);
1432                 tdwba |= 1; /* enable head wb */
1433                 wr32(E1000_TDWBAL(i),
1434                                 tdwba & 0x00000000ffffffffULL);
1435                 wr32(E1000_TDWBAH(i), tdwba >> 32);
1436
1437                 ring->head = E1000_TDH(i);
1438                 ring->tail = E1000_TDT(i);
1439                 writel(0, hw->hw_addr + ring->tail);
1440                 writel(0, hw->hw_addr + ring->head);
1441                 txdctl = rd32(E1000_TXDCTL(i));
1442                 txdctl |= E1000_TXDCTL_QUEUE_ENABLE;
1443                 wr32(E1000_TXDCTL(i), txdctl);
1444
1445                 /* Turn off Relaxed Ordering on head write-backs.  The
1446                  * writebacks MUST be delivered in order or it will
1447                  * completely screw up our bookeeping.
1448                  */
1449                 txctrl = rd32(E1000_DCA_TXCTRL(i));
1450                 txctrl &= ~E1000_DCA_TXCTRL_TX_WB_RO_EN;
1451                 wr32(E1000_DCA_TXCTRL(i), txctrl);
1452         }
1453
1454
1455
1456         /* Use the default values for the Tx Inter Packet Gap (IPG) timer */
1457
1458         /* Program the Transmit Control Register */
1459
1460         tctl = rd32(E1000_TCTL);
1461         tctl &= ~E1000_TCTL_CT;
1462         tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
1463                 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
1464
1465         igb_config_collision_dist(hw);
1466
1467         /* Setup Transmit Descriptor Settings for eop descriptor */
1468         adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_RS;
1469
1470         /* Enable transmits */
1471         tctl |= E1000_TCTL_EN;
1472
1473         wr32(E1000_TCTL, tctl);
1474 }
1475
1476 /**
1477  * igb_setup_rx_resources - allocate Rx resources (Descriptors)
1478  * @adapter: board private structure
1479  * @rx_ring:    rx descriptor ring (for a specific queue) to setup
1480  *
1481  * Returns 0 on success, negative on failure
1482  **/
1483
1484 int igb_setup_rx_resources(struct igb_adapter *adapter,
1485                            struct igb_ring *rx_ring)
1486 {
1487         struct pci_dev *pdev = adapter->pdev;
1488         int size, desc_len;
1489
1490         size = sizeof(struct igb_buffer) * rx_ring->count;
1491         rx_ring->buffer_info = vmalloc(size);
1492         if (!rx_ring->buffer_info)
1493                 goto err;
1494         memset(rx_ring->buffer_info, 0, size);
1495
1496         desc_len = sizeof(union e1000_adv_rx_desc);
1497
1498         /* Round up to nearest 4K */
1499         rx_ring->size = rx_ring->count * desc_len;
1500         rx_ring->size = ALIGN(rx_ring->size, 4096);
1501
1502         rx_ring->desc = pci_alloc_consistent(pdev, rx_ring->size,
1503                                              &rx_ring->dma);
1504
1505         if (!rx_ring->desc)
1506                 goto err;
1507
1508         rx_ring->next_to_clean = 0;
1509         rx_ring->next_to_use = 0;
1510         rx_ring->pending_skb = NULL;
1511
1512         rx_ring->adapter = adapter;
1513         /* FIXME: do we want to setup ring->napi->poll here? */
1514         rx_ring->napi.poll = adapter->napi.poll;
1515
1516         return 0;
1517
1518 err:
1519         vfree(rx_ring->buffer_info);
1520         dev_err(&adapter->pdev->dev, "Unable to allocate memory for "
1521                 "the receive descriptor ring\n");
1522         return -ENOMEM;
1523 }
1524
1525 /**
1526  * igb_setup_all_rx_resources - wrapper to allocate Rx resources
1527  *                                (Descriptors) for all queues
1528  * @adapter: board private structure
1529  *
1530  * Return 0 on success, negative on failure
1531  **/
1532 static int igb_setup_all_rx_resources(struct igb_adapter *adapter)
1533 {
1534         int i, err = 0;
1535
1536         for (i = 0; i < adapter->num_rx_queues; i++) {
1537                 err = igb_setup_rx_resources(adapter, &adapter->rx_ring[i]);
1538                 if (err) {
1539                         dev_err(&adapter->pdev->dev,
1540                                 "Allocation for Rx Queue %u failed\n", i);
1541                         for (i--; i >= 0; i--)
1542                                 igb_free_rx_resources(adapter,
1543                                                         &adapter->rx_ring[i]);
1544                         break;
1545                 }
1546         }
1547
1548         return err;
1549 }
1550
1551 /**
1552  * igb_setup_rctl - configure the receive control registers
1553  * @adapter: Board private structure
1554  **/
1555 static void igb_setup_rctl(struct igb_adapter *adapter)
1556 {
1557         struct e1000_hw *hw = &adapter->hw;
1558         u32 rctl;
1559         u32 srrctl = 0;
1560         int i;
1561
1562         rctl = rd32(E1000_RCTL);
1563
1564         rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
1565
1566         rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
1567                 E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
1568                 (adapter->hw.mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
1569
1570         /* disable the stripping of CRC because it breaks
1571          * BMC firmware connected over SMBUS
1572         rctl |= E1000_RCTL_SECRC;
1573         */
1574
1575         rctl &= ~E1000_RCTL_SBP;
1576
1577         if (adapter->netdev->mtu <= ETH_DATA_LEN)
1578                 rctl &= ~E1000_RCTL_LPE;
1579         else
1580                 rctl |= E1000_RCTL_LPE;
1581         if (adapter->rx_buffer_len <= IGB_RXBUFFER_2048) {
1582                 /* Setup buffer sizes */
1583                 rctl &= ~E1000_RCTL_SZ_4096;
1584                 rctl |= E1000_RCTL_BSEX;
1585                 switch (adapter->rx_buffer_len) {
1586                 case IGB_RXBUFFER_256:
1587                         rctl |= E1000_RCTL_SZ_256;
1588                         rctl &= ~E1000_RCTL_BSEX;
1589                         break;
1590                 case IGB_RXBUFFER_512:
1591                         rctl |= E1000_RCTL_SZ_512;
1592                         rctl &= ~E1000_RCTL_BSEX;
1593                         break;
1594                 case IGB_RXBUFFER_1024:
1595                         rctl |= E1000_RCTL_SZ_1024;
1596                         rctl &= ~E1000_RCTL_BSEX;
1597                         break;
1598                 case IGB_RXBUFFER_2048:
1599                 default:
1600                         rctl |= E1000_RCTL_SZ_2048;
1601                         rctl &= ~E1000_RCTL_BSEX;
1602                         break;
1603                 case IGB_RXBUFFER_4096:
1604                         rctl |= E1000_RCTL_SZ_4096;
1605                         break;
1606                 case IGB_RXBUFFER_8192:
1607                         rctl |= E1000_RCTL_SZ_8192;
1608                         break;
1609                 case IGB_RXBUFFER_16384:
1610                         rctl |= E1000_RCTL_SZ_16384;
1611                         break;
1612                 }
1613         } else {
1614                 rctl &= ~E1000_RCTL_BSEX;
1615                 srrctl = adapter->rx_buffer_len >> E1000_SRRCTL_BSIZEPKT_SHIFT;
1616         }
1617
1618         /* 82575 and greater support packet-split where the protocol
1619          * header is placed in skb->data and the packet data is
1620          * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
1621          * In the case of a non-split, skb->data is linearly filled,
1622          * followed by the page buffers.  Therefore, skb->data is
1623          * sized to hold the largest protocol header.
1624          */
1625         /* allocations using alloc_page take too long for regular MTU
1626          * so only enable packet split for jumbo frames */
1627         if (rctl & E1000_RCTL_LPE) {
1628                 adapter->rx_ps_hdr_size = IGB_RXBUFFER_128;
1629                 srrctl = adapter->rx_ps_hdr_size <<
1630                          E1000_SRRCTL_BSIZEHDRSIZE_SHIFT;
1631                 /* buffer size is ALWAYS one page */
1632                 srrctl |= PAGE_SIZE >> E1000_SRRCTL_BSIZEPKT_SHIFT;
1633                 srrctl |= E1000_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
1634         } else {
1635                 adapter->rx_ps_hdr_size = 0;
1636                 srrctl |= E1000_SRRCTL_DESCTYPE_ADV_ONEBUF;
1637         }
1638
1639         for (i = 0; i < adapter->num_rx_queues; i++)
1640                 wr32(E1000_SRRCTL(i), srrctl);
1641
1642         wr32(E1000_RCTL, rctl);
1643 }
1644
1645 /**
1646  * igb_configure_rx - Configure receive Unit after Reset
1647  * @adapter: board private structure
1648  *
1649  * Configure the Rx unit of the MAC after a reset.
1650  **/
1651 static void igb_configure_rx(struct igb_adapter *adapter)
1652 {
1653         u64 rdba;
1654         struct e1000_hw *hw = &adapter->hw;
1655         u32 rctl, rxcsum;
1656         u32 rxdctl;
1657         int i;
1658
1659         /* disable receives while setting up the descriptors */
1660         rctl = rd32(E1000_RCTL);
1661         wr32(E1000_RCTL, rctl & ~E1000_RCTL_EN);
1662         wrfl();
1663         mdelay(10);
1664
1665         if (adapter->itr_setting > 3)
1666                 wr32(E1000_ITR,
1667                                 1000000000 / (adapter->itr * 256));
1668
1669         /* Setup the HW Rx Head and Tail Descriptor Pointers and
1670          * the Base and Length of the Rx Descriptor Ring */
1671         for (i = 0; i < adapter->num_rx_queues; i++) {
1672                 struct igb_ring *ring = &(adapter->rx_ring[i]);
1673                 rdba = ring->dma;
1674                 wr32(E1000_RDBAL(i),
1675                                 rdba & 0x00000000ffffffffULL);
1676                 wr32(E1000_RDBAH(i), rdba >> 32);
1677                 wr32(E1000_RDLEN(i),
1678                                ring->count * sizeof(union e1000_adv_rx_desc));
1679
1680                 ring->head = E1000_RDH(i);
1681                 ring->tail = E1000_RDT(i);
1682                 writel(0, hw->hw_addr + ring->tail);
1683                 writel(0, hw->hw_addr + ring->head);
1684
1685                 rxdctl = rd32(E1000_RXDCTL(i));
1686                 rxdctl |= E1000_RXDCTL_QUEUE_ENABLE;
1687                 rxdctl &= 0xFFF00000;
1688                 rxdctl |= IGB_RX_PTHRESH;
1689                 rxdctl |= IGB_RX_HTHRESH << 8;
1690                 rxdctl |= IGB_RX_WTHRESH << 16;
1691                 wr32(E1000_RXDCTL(i), rxdctl);
1692         }
1693
1694         if (adapter->num_rx_queues > 1) {
1695                 u32 random[10];
1696                 u32 mrqc;
1697                 u32 j, shift;
1698                 union e1000_reta {
1699                         u32 dword;
1700                         u8  bytes[4];
1701                 } reta;
1702
1703                 get_random_bytes(&random[0], 40);
1704
1705                 shift = 6;
1706                 for (j = 0; j < (32 * 4); j++) {
1707                         reta.bytes[j & 3] =
1708                                 (j % adapter->num_rx_queues) << shift;
1709                         if ((j & 3) == 3)
1710                                 writel(reta.dword,
1711                                        hw->hw_addr + E1000_RETA(0) + (j & ~3));
1712                 }
1713                 mrqc = E1000_MRQC_ENABLE_RSS_4Q;
1714
1715                 /* Fill out hash function seeds */
1716                 for (j = 0; j < 10; j++)
1717                         array_wr32(E1000_RSSRK(0), j, random[j]);
1718
1719                 mrqc |= (E1000_MRQC_RSS_FIELD_IPV4 |
1720                          E1000_MRQC_RSS_FIELD_IPV4_TCP);
1721                 mrqc |= (E1000_MRQC_RSS_FIELD_IPV6 |
1722                          E1000_MRQC_RSS_FIELD_IPV6_TCP);
1723                 mrqc |= (E1000_MRQC_RSS_FIELD_IPV4_UDP |
1724                          E1000_MRQC_RSS_FIELD_IPV6_UDP);
1725                 mrqc |= (E1000_MRQC_RSS_FIELD_IPV6_UDP_EX |
1726                          E1000_MRQC_RSS_FIELD_IPV6_TCP_EX);
1727
1728
1729                 wr32(E1000_MRQC, mrqc);
1730
1731                 /* Multiqueue and raw packet checksumming are mutually
1732                  * exclusive.  Note that this not the same as TCP/IP
1733                  * checksumming, which works fine. */
1734                 rxcsum = rd32(E1000_RXCSUM);
1735                 rxcsum |= E1000_RXCSUM_PCSD;
1736                 wr32(E1000_RXCSUM, rxcsum);
1737         } else {
1738                 /* Enable Receive Checksum Offload for TCP and UDP */
1739                 rxcsum = rd32(E1000_RXCSUM);
1740                 if (adapter->rx_csum) {
1741                         rxcsum |= E1000_RXCSUM_TUOFL;
1742
1743                         /* Enable IPv4 payload checksum for UDP fragments
1744                          * Must be used in conjunction with packet-split. */
1745                         if (adapter->rx_ps_hdr_size)
1746                                 rxcsum |= E1000_RXCSUM_IPPCSE;
1747                 } else {
1748                         rxcsum &= ~E1000_RXCSUM_TUOFL;
1749                         /* don't need to clear IPPCSE as it defaults to 0 */
1750                 }
1751                 wr32(E1000_RXCSUM, rxcsum);
1752         }
1753
1754         if (adapter->vlgrp)
1755                 wr32(E1000_RLPML,
1756                                 adapter->max_frame_size + VLAN_TAG_SIZE);
1757         else
1758                 wr32(E1000_RLPML, adapter->max_frame_size);
1759
1760         /* Enable Receives */
1761         wr32(E1000_RCTL, rctl);
1762 }
1763
1764 /**
1765  * igb_free_tx_resources - Free Tx Resources per Queue
1766  * @adapter: board private structure
1767  * @tx_ring: Tx descriptor ring for a specific queue
1768  *
1769  * Free all transmit software resources
1770  **/
1771 static void igb_free_tx_resources(struct igb_adapter *adapter,
1772                                   struct igb_ring *tx_ring)
1773 {
1774         struct pci_dev *pdev = adapter->pdev;
1775
1776         igb_clean_tx_ring(adapter, tx_ring);
1777
1778         vfree(tx_ring->buffer_info);
1779         tx_ring->buffer_info = NULL;
1780
1781         pci_free_consistent(pdev, tx_ring->size, tx_ring->desc, tx_ring->dma);
1782
1783         tx_ring->desc = NULL;
1784 }
1785
1786 /**
1787  * igb_free_all_tx_resources - Free Tx Resources for All Queues
1788  * @adapter: board private structure
1789  *
1790  * Free all transmit software resources
1791  **/
1792 static void igb_free_all_tx_resources(struct igb_adapter *adapter)
1793 {
1794         int i;
1795
1796         for (i = 0; i < adapter->num_tx_queues; i++)
1797                 igb_free_tx_resources(adapter, &adapter->tx_ring[i]);
1798 }
1799
1800 static void igb_unmap_and_free_tx_resource(struct igb_adapter *adapter,
1801                                            struct igb_buffer *buffer_info)
1802 {
1803         if (buffer_info->dma) {
1804                 pci_unmap_page(adapter->pdev,
1805                                 buffer_info->dma,
1806                                 buffer_info->length,
1807                                 PCI_DMA_TODEVICE);
1808                 buffer_info->dma = 0;
1809         }
1810         if (buffer_info->skb) {
1811                 dev_kfree_skb_any(buffer_info->skb);
1812                 buffer_info->skb = NULL;
1813         }
1814         buffer_info->time_stamp = 0;
1815         /* buffer_info must be completely set up in the transmit path */
1816 }
1817
1818 /**
1819  * igb_clean_tx_ring - Free Tx Buffers
1820  * @adapter: board private structure
1821  * @tx_ring: ring to be cleaned
1822  **/
1823 static void igb_clean_tx_ring(struct igb_adapter *adapter,
1824                               struct igb_ring *tx_ring)
1825 {
1826         struct igb_buffer *buffer_info;
1827         unsigned long size;
1828         unsigned int i;
1829
1830         if (!tx_ring->buffer_info)
1831                 return;
1832         /* Free all the Tx ring sk_buffs */
1833
1834         for (i = 0; i < tx_ring->count; i++) {
1835                 buffer_info = &tx_ring->buffer_info[i];
1836                 igb_unmap_and_free_tx_resource(adapter, buffer_info);
1837         }
1838
1839         size = sizeof(struct igb_buffer) * tx_ring->count;
1840         memset(tx_ring->buffer_info, 0, size);
1841
1842         /* Zero out the descriptor ring */
1843
1844         memset(tx_ring->desc, 0, tx_ring->size);
1845
1846         tx_ring->next_to_use = 0;
1847         tx_ring->next_to_clean = 0;
1848
1849         writel(0, adapter->hw.hw_addr + tx_ring->head);
1850         writel(0, adapter->hw.hw_addr + tx_ring->tail);
1851 }
1852
1853 /**
1854  * igb_clean_all_tx_rings - Free Tx Buffers for all queues
1855  * @adapter: board private structure
1856  **/
1857 static void igb_clean_all_tx_rings(struct igb_adapter *adapter)
1858 {
1859         int i;
1860
1861         for (i = 0; i < adapter->num_tx_queues; i++)
1862                 igb_clean_tx_ring(adapter, &adapter->tx_ring[i]);
1863 }
1864
1865 /**
1866  * igb_free_rx_resources - Free Rx Resources
1867  * @adapter: board private structure
1868  * @rx_ring: ring to clean the resources from
1869  *
1870  * Free all receive software resources
1871  **/
1872 static void igb_free_rx_resources(struct igb_adapter *adapter,
1873                                   struct igb_ring *rx_ring)
1874 {
1875         struct pci_dev *pdev = adapter->pdev;
1876
1877         igb_clean_rx_ring(adapter, rx_ring);
1878
1879         vfree(rx_ring->buffer_info);
1880         rx_ring->buffer_info = NULL;
1881
1882         pci_free_consistent(pdev, rx_ring->size, rx_ring->desc, rx_ring->dma);
1883
1884         rx_ring->desc = NULL;
1885 }
1886
1887 /**
1888  * igb_free_all_rx_resources - Free Rx Resources for All Queues
1889  * @adapter: board private structure
1890  *
1891  * Free all receive software resources
1892  **/
1893 static void igb_free_all_rx_resources(struct igb_adapter *adapter)
1894 {
1895         int i;
1896
1897         for (i = 0; i < adapter->num_rx_queues; i++)
1898                 igb_free_rx_resources(adapter, &adapter->rx_ring[i]);
1899 }
1900
1901 /**
1902  * igb_clean_rx_ring - Free Rx Buffers per Queue
1903  * @adapter: board private structure
1904  * @rx_ring: ring to free buffers from
1905  **/
1906 static void igb_clean_rx_ring(struct igb_adapter *adapter,
1907                               struct igb_ring *rx_ring)
1908 {
1909         struct igb_buffer *buffer_info;
1910         struct pci_dev *pdev = adapter->pdev;
1911         unsigned long size;
1912         unsigned int i;
1913
1914         if (!rx_ring->buffer_info)
1915                 return;
1916         /* Free all the Rx ring sk_buffs */
1917         for (i = 0; i < rx_ring->count; i++) {
1918                 buffer_info = &rx_ring->buffer_info[i];
1919                 if (buffer_info->dma) {
1920                         if (adapter->rx_ps_hdr_size)
1921                                 pci_unmap_single(pdev, buffer_info->dma,
1922                                                  adapter->rx_ps_hdr_size,
1923                                                  PCI_DMA_FROMDEVICE);
1924                         else
1925                                 pci_unmap_single(pdev, buffer_info->dma,
1926                                                  adapter->rx_buffer_len,
1927                                                  PCI_DMA_FROMDEVICE);
1928                         buffer_info->dma = 0;
1929                 }
1930
1931                 if (buffer_info->skb) {
1932                         dev_kfree_skb(buffer_info->skb);
1933                         buffer_info->skb = NULL;
1934                 }
1935                 if (buffer_info->page) {
1936                         pci_unmap_page(pdev, buffer_info->page_dma,
1937                                        PAGE_SIZE, PCI_DMA_FROMDEVICE);
1938                         put_page(buffer_info->page);
1939                         buffer_info->page = NULL;
1940                         buffer_info->page_dma = 0;
1941                 }
1942         }
1943
1944         /* there also may be some cached data from a chained receive */
1945         if (rx_ring->pending_skb) {
1946                 dev_kfree_skb(rx_ring->pending_skb);
1947                 rx_ring->pending_skb = NULL;
1948         }
1949
1950         size = sizeof(struct igb_buffer) * rx_ring->count;
1951         memset(rx_ring->buffer_info, 0, size);
1952
1953         /* Zero out the descriptor ring */
1954         memset(rx_ring->desc, 0, rx_ring->size);
1955
1956         rx_ring->next_to_clean = 0;
1957         rx_ring->next_to_use = 0;
1958
1959         writel(0, adapter->hw.hw_addr + rx_ring->head);
1960         writel(0, adapter->hw.hw_addr + rx_ring->tail);
1961 }
1962
1963 /**
1964  * igb_clean_all_rx_rings - Free Rx Buffers for all queues
1965  * @adapter: board private structure
1966  **/
1967 static void igb_clean_all_rx_rings(struct igb_adapter *adapter)
1968 {
1969         int i;
1970
1971         for (i = 0; i < adapter->num_rx_queues; i++)
1972                 igb_clean_rx_ring(adapter, &adapter->rx_ring[i]);
1973 }
1974
1975 /**
1976  * igb_set_mac - Change the Ethernet Address of the NIC
1977  * @netdev: network interface device structure
1978  * @p: pointer to an address structure
1979  *
1980  * Returns 0 on success, negative on failure
1981  **/
1982 static int igb_set_mac(struct net_device *netdev, void *p)
1983 {
1984         struct igb_adapter *adapter = netdev_priv(netdev);
1985         struct sockaddr *addr = p;
1986
1987         if (!is_valid_ether_addr(addr->sa_data))
1988                 return -EADDRNOTAVAIL;
1989
1990         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
1991         memcpy(adapter->hw.mac.addr, addr->sa_data, netdev->addr_len);
1992
1993         adapter->hw.mac.ops.rar_set(&adapter->hw, adapter->hw.mac.addr, 0);
1994
1995         return 0;
1996 }
1997
1998 /**
1999  * igb_set_multi - Multicast and Promiscuous mode set
2000  * @netdev: network interface device structure
2001  *
2002  * The set_multi entry point is called whenever the multicast address
2003  * list or the network interface flags are updated.  This routine is
2004  * responsible for configuring the hardware for proper multicast,
2005  * promiscuous mode, and all-multi behavior.
2006  **/
2007 static void igb_set_multi(struct net_device *netdev)
2008 {
2009         struct igb_adapter *adapter = netdev_priv(netdev);
2010         struct e1000_hw *hw = &adapter->hw;
2011         struct e1000_mac_info *mac = &hw->mac;
2012         struct dev_mc_list *mc_ptr;
2013         u8  *mta_list;
2014         u32 rctl;
2015         int i;
2016
2017         /* Check for Promiscuous and All Multicast modes */
2018
2019         rctl = rd32(E1000_RCTL);
2020
2021         if (netdev->flags & IFF_PROMISC)
2022                 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
2023         else if (netdev->flags & IFF_ALLMULTI) {
2024                 rctl |= E1000_RCTL_MPE;
2025                 rctl &= ~E1000_RCTL_UPE;
2026         } else
2027                 rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
2028
2029         wr32(E1000_RCTL, rctl);
2030
2031         if (!netdev->mc_count) {
2032                 /* nothing to program, so clear mc list */
2033                 igb_update_mc_addr_list(hw, NULL, 0, 1,
2034                                           mac->rar_entry_count);
2035                 return;
2036         }
2037
2038         mta_list = kzalloc(netdev->mc_count * 6, GFP_ATOMIC);
2039         if (!mta_list)
2040                 return;
2041
2042         /* The shared function expects a packed array of only addresses. */
2043         mc_ptr = netdev->mc_list;
2044
2045         for (i = 0; i < netdev->mc_count; i++) {
2046                 if (!mc_ptr)
2047                         break;
2048                 memcpy(mta_list + (i*ETH_ALEN), mc_ptr->dmi_addr, ETH_ALEN);
2049                 mc_ptr = mc_ptr->next;
2050         }
2051         igb_update_mc_addr_list(hw, mta_list, i, 1, mac->rar_entry_count);
2052         kfree(mta_list);
2053 }
2054
2055 /* Need to wait a few seconds after link up to get diagnostic information from
2056  * the phy */
2057 static void igb_update_phy_info(unsigned long data)
2058 {
2059         struct igb_adapter *adapter = (struct igb_adapter *) data;
2060         adapter->hw.phy.ops.get_phy_info(&adapter->hw);
2061 }
2062
2063 /**
2064  * igb_watchdog - Timer Call-back
2065  * @data: pointer to adapter cast into an unsigned long
2066  **/
2067 static void igb_watchdog(unsigned long data)
2068 {
2069         struct igb_adapter *adapter = (struct igb_adapter *)data;
2070         /* Do the rest outside of interrupt context */
2071         schedule_work(&adapter->watchdog_task);
2072 }
2073
2074 static void igb_watchdog_task(struct work_struct *work)
2075 {
2076         struct igb_adapter *adapter = container_of(work,
2077                                         struct igb_adapter, watchdog_task);
2078         struct e1000_hw *hw = &adapter->hw;
2079
2080         struct net_device *netdev = adapter->netdev;
2081         struct igb_ring *tx_ring = adapter->tx_ring;
2082         struct e1000_mac_info *mac = &adapter->hw.mac;
2083         u32 link;
2084         s32 ret_val;
2085
2086         if ((netif_carrier_ok(netdev)) &&
2087             (rd32(E1000_STATUS) & E1000_STATUS_LU))
2088                 goto link_up;
2089
2090         ret_val = hw->mac.ops.check_for_link(&adapter->hw);
2091         if ((ret_val == E1000_ERR_PHY) &&
2092             (hw->phy.type == e1000_phy_igp_3) &&
2093             (rd32(E1000_CTRL) &
2094              E1000_PHY_CTRL_GBE_DISABLE))
2095                 dev_info(&adapter->pdev->dev,
2096                          "Gigabit has been disabled, downgrading speed\n");
2097
2098         if ((hw->phy.media_type == e1000_media_type_internal_serdes) &&
2099             !(rd32(E1000_TXCW) & E1000_TXCW_ANE))
2100                 link = mac->serdes_has_link;
2101         else
2102                 link = rd32(E1000_STATUS) &
2103                                       E1000_STATUS_LU;
2104
2105         if (link) {
2106                 if (!netif_carrier_ok(netdev)) {
2107                         u32 ctrl;
2108                         hw->mac.ops.get_speed_and_duplex(&adapter->hw,
2109                                                    &adapter->link_speed,
2110                                                    &adapter->link_duplex);
2111
2112                         ctrl = rd32(E1000_CTRL);
2113                         dev_info(&adapter->pdev->dev,
2114                                  "NIC Link is Up %d Mbps %s, "
2115                                  "Flow Control: %s\n",
2116                                  adapter->link_speed,
2117                                  adapter->link_duplex == FULL_DUPLEX ?
2118                                  "Full Duplex" : "Half Duplex",
2119                                  ((ctrl & E1000_CTRL_TFCE) && (ctrl &
2120                                  E1000_CTRL_RFCE)) ? "RX/TX" : ((ctrl &
2121                                  E1000_CTRL_RFCE) ? "RX" : ((ctrl &
2122                                  E1000_CTRL_TFCE) ? "TX" : "None")));
2123
2124                         /* tweak tx_queue_len according to speed/duplex and
2125                          * adjust the timeout factor */
2126                         netdev->tx_queue_len = adapter->tx_queue_len;
2127                         adapter->tx_timeout_factor = 1;
2128                         switch (adapter->link_speed) {
2129                         case SPEED_10:
2130                                 netdev->tx_queue_len = 10;
2131                                 adapter->tx_timeout_factor = 14;
2132                                 break;
2133                         case SPEED_100:
2134                                 netdev->tx_queue_len = 100;
2135                                 /* maybe add some timeout factor ? */
2136                                 break;
2137                         }
2138
2139                         netif_carrier_on(netdev);
2140                         netif_wake_queue(netdev);
2141
2142                         if (!test_bit(__IGB_DOWN, &adapter->state))
2143                                 mod_timer(&adapter->phy_info_timer,
2144                                           round_jiffies(jiffies + 2 * HZ));
2145                 }
2146         } else {
2147                 if (netif_carrier_ok(netdev)) {
2148                         adapter->link_speed = 0;
2149                         adapter->link_duplex = 0;
2150                         dev_info(&adapter->pdev->dev, "NIC Link is Down\n");
2151                         netif_carrier_off(netdev);
2152                         netif_stop_queue(netdev);
2153                         if (!test_bit(__IGB_DOWN, &adapter->state))
2154                                 mod_timer(&adapter->phy_info_timer,
2155                                           round_jiffies(jiffies + 2 * HZ));
2156                 }
2157         }
2158
2159 link_up:
2160         igb_update_stats(adapter);
2161
2162         mac->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
2163         adapter->tpt_old = adapter->stats.tpt;
2164         mac->collision_delta = adapter->stats.colc - adapter->colc_old;
2165         adapter->colc_old = adapter->stats.colc;
2166
2167         adapter->gorc = adapter->stats.gorc - adapter->gorc_old;
2168         adapter->gorc_old = adapter->stats.gorc;
2169         adapter->gotc = adapter->stats.gotc - adapter->gotc_old;
2170         adapter->gotc_old = adapter->stats.gotc;
2171
2172         igb_update_adaptive(&adapter->hw);
2173
2174         if (!netif_carrier_ok(netdev)) {
2175                 if (IGB_DESC_UNUSED(tx_ring) + 1 < tx_ring->count) {
2176                         /* We've lost link, so the controller stops DMA,
2177                          * but we've got queued Tx work that's never going
2178                          * to get done, so reset controller to flush Tx.
2179                          * (Do the reset outside of interrupt context). */
2180                         adapter->tx_timeout_count++;
2181                         schedule_work(&adapter->reset_task);
2182                 }
2183         }
2184
2185         /* Cause software interrupt to ensure rx ring is cleaned */
2186         wr32(E1000_ICS, E1000_ICS_RXDMT0);
2187
2188         /* Force detection of hung controller every watchdog period */
2189         tx_ring->detect_tx_hung = true;
2190
2191         /* Reset the timer */
2192         if (!test_bit(__IGB_DOWN, &adapter->state))
2193                 mod_timer(&adapter->watchdog_timer,
2194                           round_jiffies(jiffies + 2 * HZ));
2195 }
2196
2197 enum latency_range {
2198         lowest_latency = 0,
2199         low_latency = 1,
2200         bulk_latency = 2,
2201         latency_invalid = 255
2202 };
2203
2204
2205 static void igb_lower_rx_eitr(struct igb_adapter *adapter,
2206                               struct igb_ring *rx_ring)
2207 {
2208         struct e1000_hw *hw = &adapter->hw;
2209         int new_val;
2210
2211         new_val = rx_ring->itr_val / 2;
2212         if (new_val < IGB_MIN_DYN_ITR)
2213                 new_val = IGB_MIN_DYN_ITR;
2214
2215         if (new_val != rx_ring->itr_val) {
2216                 rx_ring->itr_val = new_val;
2217                 wr32(rx_ring->itr_register,
2218                                 1000000000 / (new_val * 256));
2219         }
2220 }
2221
2222 static void igb_raise_rx_eitr(struct igb_adapter *adapter,
2223                               struct igb_ring *rx_ring)
2224 {
2225         struct e1000_hw *hw = &adapter->hw;
2226         int new_val;
2227
2228         new_val = rx_ring->itr_val * 2;
2229         if (new_val > IGB_MAX_DYN_ITR)
2230                 new_val = IGB_MAX_DYN_ITR;
2231
2232         if (new_val != rx_ring->itr_val) {
2233                 rx_ring->itr_val = new_val;
2234                 wr32(rx_ring->itr_register,
2235                                 1000000000 / (new_val * 256));
2236         }
2237 }
2238
2239 /**
2240  * igb_update_itr - update the dynamic ITR value based on statistics
2241  *      Stores a new ITR value based on packets and byte
2242  *      counts during the last interrupt.  The advantage of per interrupt
2243  *      computation is faster updates and more accurate ITR for the current
2244  *      traffic pattern.  Constants in this function were computed
2245  *      based on theoretical maximum wire speed and thresholds were set based
2246  *      on testing data as well as attempting to minimize response time
2247  *      while increasing bulk throughput.
2248  *      this functionality is controlled by the InterruptThrottleRate module
2249  *      parameter (see igb_param.c)
2250  *      NOTE:  These calculations are only valid when operating in a single-
2251  *             queue environment.
2252  * @adapter: pointer to adapter
2253  * @itr_setting: current adapter->itr
2254  * @packets: the number of packets during this measurement interval
2255  * @bytes: the number of bytes during this measurement interval
2256  **/
2257 static unsigned int igb_update_itr(struct igb_adapter *adapter, u16 itr_setting,
2258                                    int packets, int bytes)
2259 {
2260         unsigned int retval = itr_setting;
2261
2262         if (packets == 0)
2263                 goto update_itr_done;
2264
2265         switch (itr_setting) {
2266         case lowest_latency:
2267                 /* handle TSO and jumbo frames */
2268                 if (bytes/packets > 8000)
2269                         retval = bulk_latency;
2270                 else if ((packets < 5) && (bytes > 512))
2271                         retval = low_latency;
2272                 break;
2273         case low_latency:  /* 50 usec aka 20000 ints/s */
2274                 if (bytes > 10000) {
2275                         /* this if handles the TSO accounting */
2276                         if (bytes/packets > 8000) {
2277                                 retval = bulk_latency;
2278                         } else if ((packets < 10) || ((bytes/packets) > 1200)) {
2279                                 retval = bulk_latency;
2280                         } else if ((packets > 35)) {
2281                                 retval = lowest_latency;
2282                         }
2283                 } else if (bytes/packets > 2000) {
2284                         retval = bulk_latency;
2285                 } else if (packets <= 2 && bytes < 512) {
2286                         retval = lowest_latency;
2287                 }
2288                 break;
2289         case bulk_latency: /* 250 usec aka 4000 ints/s */
2290                 if (bytes > 25000) {
2291                         if (packets > 35)
2292                                 retval = low_latency;
2293                 } else if (bytes < 6000) {
2294                         retval = low_latency;
2295                 }
2296                 break;
2297         }
2298
2299 update_itr_done:
2300         return retval;
2301 }
2302
2303 static void igb_set_itr(struct igb_adapter *adapter, u16 itr_register,
2304                         int rx_only)
2305 {
2306         u16 current_itr;
2307         u32 new_itr = adapter->itr;
2308
2309         /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
2310         if (adapter->link_speed != SPEED_1000) {
2311                 current_itr = 0;
2312                 new_itr = 4000;
2313                 goto set_itr_now;
2314         }
2315
2316         adapter->rx_itr = igb_update_itr(adapter,
2317                                     adapter->rx_itr,
2318                                     adapter->rx_ring->total_packets,
2319                                     adapter->rx_ring->total_bytes);
2320         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2321         if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
2322                 adapter->rx_itr = low_latency;
2323
2324         if (!rx_only) {
2325                 adapter->tx_itr = igb_update_itr(adapter,
2326                                             adapter->tx_itr,
2327                                             adapter->tx_ring->total_packets,
2328                                             adapter->tx_ring->total_bytes);
2329                 /* conservative mode (itr 3) eliminates the
2330                  * lowest_latency setting */
2331                 if (adapter->itr_setting == 3 &&
2332                     adapter->tx_itr == lowest_latency)
2333                         adapter->tx_itr = low_latency;
2334
2335                 current_itr = max(adapter->rx_itr, adapter->tx_itr);
2336         } else {
2337                 current_itr = adapter->rx_itr;
2338         }
2339
2340         switch (current_itr) {
2341         /* counts and packets in update_itr are dependent on these numbers */
2342         case lowest_latency:
2343                 new_itr = 70000;
2344                 break;
2345         case low_latency:
2346                 new_itr = 20000; /* aka hwitr = ~200 */
2347                 break;
2348         case bulk_latency:
2349                 new_itr = 4000;
2350                 break;
2351         default:
2352                 break;
2353         }
2354
2355 set_itr_now:
2356         if (new_itr != adapter->itr) {
2357                 /* this attempts to bias the interrupt rate towards Bulk
2358                  * by adding intermediate steps when interrupt rate is
2359                  * increasing */
2360                 new_itr = new_itr > adapter->itr ?
2361                              min(adapter->itr + (new_itr >> 2), new_itr) :
2362                              new_itr;
2363                 /* Don't write the value here; it resets the adapter's
2364                  * internal timer, and causes us to delay far longer than
2365                  * we should between interrupts.  Instead, we write the ITR
2366                  * value at the beginning of the next interrupt so the timing
2367                  * ends up being correct.
2368                  */
2369                 adapter->itr = new_itr;
2370                 adapter->set_itr = 1;
2371         }
2372
2373         return;
2374 }
2375
2376
2377 #define IGB_TX_FLAGS_CSUM               0x00000001
2378 #define IGB_TX_FLAGS_VLAN               0x00000002
2379 #define IGB_TX_FLAGS_TSO                0x00000004
2380 #define IGB_TX_FLAGS_IPV4               0x00000008
2381 #define IGB_TX_FLAGS_VLAN_MASK  0xffff0000
2382 #define IGB_TX_FLAGS_VLAN_SHIFT 16
2383
2384 static inline int igb_tso_adv(struct igb_adapter *adapter,
2385                               struct igb_ring *tx_ring,
2386                               struct sk_buff *skb, u32 tx_flags, u8 *hdr_len)
2387 {
2388         struct e1000_adv_tx_context_desc *context_desc;
2389         unsigned int i;
2390         int err;
2391         struct igb_buffer *buffer_info;
2392         u32 info = 0, tu_cmd = 0;
2393         u32 mss_l4len_idx, l4len;
2394         *hdr_len = 0;
2395
2396         if (skb_header_cloned(skb)) {
2397                 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2398                 if (err)
2399                         return err;
2400         }
2401
2402         l4len = tcp_hdrlen(skb);
2403         *hdr_len += l4len;
2404
2405         if (skb->protocol == htons(ETH_P_IP)) {
2406                 struct iphdr *iph = ip_hdr(skb);
2407                 iph->tot_len = 0;
2408                 iph->check = 0;
2409                 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
2410                                                          iph->daddr, 0,
2411                                                          IPPROTO_TCP,
2412                                                          0);
2413         } else if (skb_shinfo(skb)->gso_type == SKB_GSO_TCPV6) {
2414                 ipv6_hdr(skb)->payload_len = 0;
2415                 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
2416                                                        &ipv6_hdr(skb)->daddr,
2417                                                        0, IPPROTO_TCP, 0);
2418         }
2419
2420         i = tx_ring->next_to_use;
2421
2422         buffer_info = &tx_ring->buffer_info[i];
2423         context_desc = E1000_TX_CTXTDESC_ADV(*tx_ring, i);
2424         /* VLAN MACLEN IPLEN */
2425         if (tx_flags & IGB_TX_FLAGS_VLAN)
2426                 info |= (tx_flags & IGB_TX_FLAGS_VLAN_MASK);
2427         info |= (skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT);
2428         *hdr_len += skb_network_offset(skb);
2429         info |= skb_network_header_len(skb);
2430         *hdr_len += skb_network_header_len(skb);
2431         context_desc->vlan_macip_lens = cpu_to_le32(info);
2432
2433         /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
2434         tu_cmd |= (E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT);
2435
2436         if (skb->protocol == htons(ETH_P_IP))
2437                 tu_cmd |= E1000_ADVTXD_TUCMD_IPV4;
2438         tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
2439
2440         context_desc->type_tucmd_mlhl = cpu_to_le32(tu_cmd);
2441
2442         /* MSS L4LEN IDX */
2443         mss_l4len_idx = (skb_shinfo(skb)->gso_size << E1000_ADVTXD_MSS_SHIFT);
2444         mss_l4len_idx |= (l4len << E1000_ADVTXD_L4LEN_SHIFT);
2445
2446         /* Context index must be unique per ring.  Luckily, so is the interrupt
2447          * mask value. */
2448         mss_l4len_idx |= tx_ring->eims_value >> 4;
2449
2450         context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
2451         context_desc->seqnum_seed = 0;
2452
2453         buffer_info->time_stamp = jiffies;
2454         buffer_info->dma = 0;
2455         i++;
2456         if (i == tx_ring->count)
2457                 i = 0;
2458
2459         tx_ring->next_to_use = i;
2460
2461         return true;
2462 }
2463
2464 static inline bool igb_tx_csum_adv(struct igb_adapter *adapter,
2465                                         struct igb_ring *tx_ring,
2466                                         struct sk_buff *skb, u32 tx_flags)
2467 {
2468         struct e1000_adv_tx_context_desc *context_desc;
2469         unsigned int i;
2470         struct igb_buffer *buffer_info;
2471         u32 info = 0, tu_cmd = 0;
2472
2473         if ((skb->ip_summed == CHECKSUM_PARTIAL) ||
2474             (tx_flags & IGB_TX_FLAGS_VLAN)) {
2475                 i = tx_ring->next_to_use;
2476                 buffer_info = &tx_ring->buffer_info[i];
2477                 context_desc = E1000_TX_CTXTDESC_ADV(*tx_ring, i);
2478
2479                 if (tx_flags & IGB_TX_FLAGS_VLAN)
2480                         info |= (tx_flags & IGB_TX_FLAGS_VLAN_MASK);
2481                 info |= (skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT);
2482                 if (skb->ip_summed == CHECKSUM_PARTIAL)
2483                         info |= skb_network_header_len(skb);
2484
2485                 context_desc->vlan_macip_lens = cpu_to_le32(info);
2486
2487                 tu_cmd |= (E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT);
2488
2489                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2490                         if (skb->protocol == htons(ETH_P_IP))
2491                                 tu_cmd |= E1000_ADVTXD_TUCMD_IPV4;
2492                         if (skb->sk && (skb->sk->sk_protocol == IPPROTO_TCP))
2493                                 tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
2494                 }
2495
2496                 context_desc->type_tucmd_mlhl = cpu_to_le32(tu_cmd);
2497                 context_desc->seqnum_seed = 0;
2498                 context_desc->mss_l4len_idx =
2499                                           cpu_to_le32(tx_ring->eims_value >> 4);
2500
2501                 buffer_info->time_stamp = jiffies;
2502                 buffer_info->dma = 0;
2503
2504                 i++;
2505                 if (i == tx_ring->count)
2506                         i = 0;
2507                 tx_ring->next_to_use = i;
2508
2509                 return true;
2510         }
2511
2512
2513         return false;
2514 }
2515
2516 #define IGB_MAX_TXD_PWR 16
2517 #define IGB_MAX_DATA_PER_TXD    (1<<IGB_MAX_TXD_PWR)
2518
2519 static inline int igb_tx_map_adv(struct igb_adapter *adapter,
2520                                  struct igb_ring *tx_ring,
2521                                  struct sk_buff *skb)
2522 {
2523         struct igb_buffer *buffer_info;
2524         unsigned int len = skb_headlen(skb);
2525         unsigned int count = 0, i;
2526         unsigned int f;
2527
2528         i = tx_ring->next_to_use;
2529
2530         buffer_info = &tx_ring->buffer_info[i];
2531         BUG_ON(len >= IGB_MAX_DATA_PER_TXD);
2532         buffer_info->length = len;
2533         /* set time_stamp *before* dma to help avoid a possible race */
2534         buffer_info->time_stamp = jiffies;
2535         buffer_info->dma = pci_map_single(adapter->pdev, skb->data, len,
2536                                           PCI_DMA_TODEVICE);
2537         count++;
2538         i++;
2539         if (i == tx_ring->count)
2540                 i = 0;
2541
2542         for (f = 0; f < skb_shinfo(skb)->nr_frags; f++) {
2543                 struct skb_frag_struct *frag;
2544
2545                 frag = &skb_shinfo(skb)->frags[f];
2546                 len = frag->size;
2547
2548                 buffer_info = &tx_ring->buffer_info[i];
2549                 BUG_ON(len >= IGB_MAX_DATA_PER_TXD);
2550                 buffer_info->length = len;
2551                 buffer_info->time_stamp = jiffies;
2552                 buffer_info->dma = pci_map_page(adapter->pdev,
2553                                                 frag->page,
2554                                                 frag->page_offset,
2555                                                 len,
2556                                                 PCI_DMA_TODEVICE);
2557
2558                 count++;
2559                 i++;
2560                 if (i == tx_ring->count)
2561                         i = 0;
2562         }
2563
2564         i = (i == 0) ? tx_ring->count - 1 : i - 1;
2565         tx_ring->buffer_info[i].skb = skb;
2566
2567         return count;
2568 }
2569
2570 static inline void igb_tx_queue_adv(struct igb_adapter *adapter,
2571                                     struct igb_ring *tx_ring,
2572                                     int tx_flags, int count, u32 paylen,
2573                                     u8 hdr_len)
2574 {
2575         union e1000_adv_tx_desc *tx_desc = NULL;
2576         struct igb_buffer *buffer_info;
2577         u32 olinfo_status = 0, cmd_type_len;
2578         unsigned int i;
2579
2580         cmd_type_len = (E1000_ADVTXD_DTYP_DATA | E1000_ADVTXD_DCMD_IFCS |
2581                         E1000_ADVTXD_DCMD_DEXT);
2582
2583         if (tx_flags & IGB_TX_FLAGS_VLAN)
2584                 cmd_type_len |= E1000_ADVTXD_DCMD_VLE;
2585
2586         if (tx_flags & IGB_TX_FLAGS_TSO) {
2587                 cmd_type_len |= E1000_ADVTXD_DCMD_TSE;
2588
2589                 /* insert tcp checksum */
2590                 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
2591
2592                 /* insert ip checksum */
2593                 if (tx_flags & IGB_TX_FLAGS_IPV4)
2594                         olinfo_status |= E1000_TXD_POPTS_IXSM << 8;
2595
2596         } else if (tx_flags & IGB_TX_FLAGS_CSUM) {
2597                 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
2598         }
2599
2600         if (tx_flags & (IGB_TX_FLAGS_CSUM | IGB_TX_FLAGS_TSO |
2601                         IGB_TX_FLAGS_VLAN))
2602                 olinfo_status |= tx_ring->eims_value >> 4;
2603
2604         olinfo_status |= ((paylen - hdr_len) << E1000_ADVTXD_PAYLEN_SHIFT);
2605
2606         i = tx_ring->next_to_use;
2607         while (count--) {
2608                 buffer_info = &tx_ring->buffer_info[i];
2609                 tx_desc = E1000_TX_DESC_ADV(*tx_ring, i);
2610                 tx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma);
2611                 tx_desc->read.cmd_type_len =
2612                         cpu_to_le32(cmd_type_len | buffer_info->length);
2613                 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
2614                 i++;
2615                 if (i == tx_ring->count)
2616                         i = 0;
2617         }
2618
2619         tx_desc->read.cmd_type_len |= cpu_to_le32(adapter->txd_cmd);
2620         /* Force memory writes to complete before letting h/w
2621          * know there are new descriptors to fetch.  (Only
2622          * applicable for weak-ordered memory model archs,
2623          * such as IA-64). */
2624         wmb();
2625
2626         tx_ring->next_to_use = i;
2627         writel(i, adapter->hw.hw_addr + tx_ring->tail);
2628         /* we need this if more than one processor can write to our tail
2629          * at a time, it syncronizes IO on IA64/Altix systems */
2630         mmiowb();
2631 }
2632
2633 static int __igb_maybe_stop_tx(struct net_device *netdev,
2634                                struct igb_ring *tx_ring, int size)
2635 {
2636         struct igb_adapter *adapter = netdev_priv(netdev);
2637
2638         netif_stop_queue(netdev);
2639         /* Herbert's original patch had:
2640          *  smp_mb__after_netif_stop_queue();
2641          * but since that doesn't exist yet, just open code it. */
2642         smp_mb();
2643
2644         /* We need to check again in a case another CPU has just
2645          * made room available. */
2646         if (IGB_DESC_UNUSED(tx_ring) < size)
2647                 return -EBUSY;
2648
2649         /* A reprieve! */
2650         netif_start_queue(netdev);
2651         ++adapter->restart_queue;
2652         return 0;
2653 }
2654
2655 static int igb_maybe_stop_tx(struct net_device *netdev,
2656                              struct igb_ring *tx_ring, int size)
2657 {
2658         if (IGB_DESC_UNUSED(tx_ring) >= size)
2659                 return 0;
2660         return __igb_maybe_stop_tx(netdev, tx_ring, size);
2661 }
2662
2663 #define TXD_USE_COUNT(S) (((S) >> (IGB_MAX_TXD_PWR)) + 1)
2664
2665 static int igb_xmit_frame_ring_adv(struct sk_buff *skb,
2666                                    struct net_device *netdev,
2667                                    struct igb_ring *tx_ring)
2668 {
2669         struct igb_adapter *adapter = netdev_priv(netdev);
2670         unsigned int tx_flags = 0;
2671         unsigned int len;
2672         unsigned long irq_flags;
2673         u8 hdr_len = 0;
2674         int tso = 0;
2675
2676         len = skb_headlen(skb);
2677
2678         if (test_bit(__IGB_DOWN, &adapter->state)) {
2679                 dev_kfree_skb_any(skb);
2680                 return NETDEV_TX_OK;
2681         }
2682
2683         if (skb->len <= 0) {
2684                 dev_kfree_skb_any(skb);
2685                 return NETDEV_TX_OK;
2686         }
2687
2688         if (!spin_trylock_irqsave(&tx_ring->tx_lock, irq_flags))
2689                 /* Collision - tell upper layer to requeue */
2690                 return NETDEV_TX_LOCKED;
2691
2692         /* need: 1 descriptor per page,
2693          *       + 2 desc gap to keep tail from touching head,
2694          *       + 1 desc for skb->data,
2695          *       + 1 desc for context descriptor,
2696          * otherwise try next time */
2697         if (igb_maybe_stop_tx(netdev, tx_ring, skb_shinfo(skb)->nr_frags + 4)) {
2698                 /* this is a hard error */
2699                 spin_unlock_irqrestore(&tx_ring->tx_lock, irq_flags);
2700                 return NETDEV_TX_BUSY;
2701         }
2702
2703         if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
2704                 tx_flags |= IGB_TX_FLAGS_VLAN;
2705                 tx_flags |= (vlan_tx_tag_get(skb) << IGB_TX_FLAGS_VLAN_SHIFT);
2706         }
2707
2708         tso = skb_is_gso(skb) ? igb_tso_adv(adapter, tx_ring, skb, tx_flags,
2709                                               &hdr_len) : 0;
2710
2711         if (tso < 0) {
2712                 dev_kfree_skb_any(skb);
2713                 spin_unlock_irqrestore(&tx_ring->tx_lock, irq_flags);
2714                 return NETDEV_TX_OK;
2715         }
2716
2717         if (tso)
2718                 tx_flags |= IGB_TX_FLAGS_TSO;
2719         else if (igb_tx_csum_adv(adapter, tx_ring, skb, tx_flags))
2720                         if (skb->ip_summed == CHECKSUM_PARTIAL)
2721                                 tx_flags |= IGB_TX_FLAGS_CSUM;
2722
2723         if (skb->protocol == htons(ETH_P_IP))
2724                 tx_flags |= IGB_TX_FLAGS_IPV4;
2725
2726         igb_tx_queue_adv(adapter, tx_ring, tx_flags,
2727                          igb_tx_map_adv(adapter, tx_ring, skb),
2728                          skb->len, hdr_len);
2729
2730         netdev->trans_start = jiffies;
2731
2732         /* Make sure there is space in the ring for the next send. */
2733         igb_maybe_stop_tx(netdev, tx_ring, MAX_SKB_FRAGS + 4);
2734
2735         spin_unlock_irqrestore(&tx_ring->tx_lock, irq_flags);
2736         return NETDEV_TX_OK;
2737 }
2738
2739 static int igb_xmit_frame_adv(struct sk_buff *skb, struct net_device *netdev)
2740 {
2741         struct igb_adapter *adapter = netdev_priv(netdev);
2742         struct igb_ring *tx_ring = &adapter->tx_ring[0];
2743
2744         /* This goes back to the question of how to logically map a tx queue
2745          * to a flow.  Right now, performance is impacted slightly negatively
2746          * if using multiple tx queues.  If the stack breaks away from a
2747          * single qdisc implementation, we can look at this again. */
2748         return (igb_xmit_frame_ring_adv(skb, netdev, tx_ring));
2749 }
2750
2751 /**
2752  * igb_tx_timeout - Respond to a Tx Hang
2753  * @netdev: network interface device structure
2754  **/
2755 static void igb_tx_timeout(struct net_device *netdev)
2756 {
2757         struct igb_adapter *adapter = netdev_priv(netdev);
2758         struct e1000_hw *hw = &adapter->hw;
2759
2760         /* Do the reset outside of interrupt context */
2761         adapter->tx_timeout_count++;
2762         schedule_work(&adapter->reset_task);
2763         wr32(E1000_EICS, adapter->eims_enable_mask &
2764                 ~(E1000_EIMS_TCP_TIMER | E1000_EIMS_OTHER));
2765 }
2766
2767 static void igb_reset_task(struct work_struct *work)
2768 {
2769         struct igb_adapter *adapter;
2770         adapter = container_of(work, struct igb_adapter, reset_task);
2771
2772         igb_reinit_locked(adapter);
2773 }
2774
2775 /**
2776  * igb_get_stats - Get System Network Statistics
2777  * @netdev: network interface device structure
2778  *
2779  * Returns the address of the device statistics structure.
2780  * The statistics are actually updated from the timer callback.
2781  **/
2782 static struct net_device_stats *
2783 igb_get_stats(struct net_device *netdev)
2784 {
2785         struct igb_adapter *adapter = netdev_priv(netdev);
2786
2787         /* only return the current stats */
2788         return &adapter->net_stats;
2789 }
2790
2791 /**
2792  * igb_change_mtu - Change the Maximum Transfer Unit
2793  * @netdev: network interface device structure
2794  * @new_mtu: new value for maximum frame size
2795  *
2796  * Returns 0 on success, negative on failure
2797  **/
2798 static int igb_change_mtu(struct net_device *netdev, int new_mtu)
2799 {
2800         struct igb_adapter *adapter = netdev_priv(netdev);
2801         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
2802
2803         if ((max_frame < ETH_ZLEN + ETH_FCS_LEN) ||
2804             (max_frame > MAX_JUMBO_FRAME_SIZE)) {
2805                 dev_err(&adapter->pdev->dev, "Invalid MTU setting\n");
2806                 return -EINVAL;
2807         }
2808
2809 #define MAX_STD_JUMBO_FRAME_SIZE 9234
2810         if (max_frame > MAX_STD_JUMBO_FRAME_SIZE) {
2811                 dev_err(&adapter->pdev->dev, "MTU > 9216 not supported.\n");
2812                 return -EINVAL;
2813         }
2814
2815         while (test_and_set_bit(__IGB_RESETTING, &adapter->state))
2816                 msleep(1);
2817         /* igb_down has a dependency on max_frame_size */
2818         adapter->max_frame_size = max_frame;
2819         if (netif_running(netdev))
2820                 igb_down(adapter);
2821
2822         /* NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
2823          * means we reserve 2 more, this pushes us to allocate from the next
2824          * larger slab size.
2825          * i.e. RXBUFFER_2048 --> size-4096 slab
2826          */
2827
2828         if (max_frame <= IGB_RXBUFFER_256)
2829                 adapter->rx_buffer_len = IGB_RXBUFFER_256;
2830         else if (max_frame <= IGB_RXBUFFER_512)
2831                 adapter->rx_buffer_len = IGB_RXBUFFER_512;
2832         else if (max_frame <= IGB_RXBUFFER_1024)
2833                 adapter->rx_buffer_len = IGB_RXBUFFER_1024;
2834         else if (max_frame <= IGB_RXBUFFER_2048)
2835                 adapter->rx_buffer_len = IGB_RXBUFFER_2048;
2836         else
2837                 adapter->rx_buffer_len = IGB_RXBUFFER_4096;
2838         /* adjust allocation if LPE protects us, and we aren't using SBP */
2839         if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
2840              (max_frame == MAXIMUM_ETHERNET_VLAN_SIZE))
2841                 adapter->rx_buffer_len = MAXIMUM_ETHERNET_VLAN_SIZE;
2842
2843         dev_info(&adapter->pdev->dev, "changing MTU from %d to %d\n",
2844                  netdev->mtu, new_mtu);
2845         netdev->mtu = new_mtu;
2846
2847         if (netif_running(netdev))
2848                 igb_up(adapter);
2849         else
2850                 igb_reset(adapter);
2851
2852         clear_bit(__IGB_RESETTING, &adapter->state);
2853
2854         return 0;
2855 }
2856
2857 /**
2858  * igb_update_stats - Update the board statistics counters
2859  * @adapter: board private structure
2860  **/
2861
2862 void igb_update_stats(struct igb_adapter *adapter)
2863 {
2864         struct e1000_hw *hw = &adapter->hw;
2865         struct pci_dev *pdev = adapter->pdev;
2866         u16 phy_tmp;
2867
2868 #define PHY_IDLE_ERROR_COUNT_MASK 0x00FF
2869
2870         /*
2871          * Prevent stats update while adapter is being reset, or if the pci
2872          * connection is down.
2873          */
2874         if (adapter->link_speed == 0)
2875                 return;
2876         if (pci_channel_offline(pdev))
2877                 return;
2878
2879         adapter->stats.crcerrs += rd32(E1000_CRCERRS);
2880         adapter->stats.gprc += rd32(E1000_GPRC);
2881         adapter->stats.gorc += rd32(E1000_GORCL);
2882         rd32(E1000_GORCH); /* clear GORCL */
2883         adapter->stats.bprc += rd32(E1000_BPRC);
2884         adapter->stats.mprc += rd32(E1000_MPRC);
2885         adapter->stats.roc += rd32(E1000_ROC);
2886
2887         adapter->stats.prc64 += rd32(E1000_PRC64);
2888         adapter->stats.prc127 += rd32(E1000_PRC127);
2889         adapter->stats.prc255 += rd32(E1000_PRC255);
2890         adapter->stats.prc511 += rd32(E1000_PRC511);
2891         adapter->stats.prc1023 += rd32(E1000_PRC1023);
2892         adapter->stats.prc1522 += rd32(E1000_PRC1522);
2893         adapter->stats.symerrs += rd32(E1000_SYMERRS);
2894         adapter->stats.sec += rd32(E1000_SEC);
2895
2896         adapter->stats.mpc += rd32(E1000_MPC);
2897         adapter->stats.scc += rd32(E1000_SCC);
2898         adapter->stats.ecol += rd32(E1000_ECOL);
2899         adapter->stats.mcc += rd32(E1000_MCC);
2900         adapter->stats.latecol += rd32(E1000_LATECOL);
2901         adapter->stats.dc += rd32(E1000_DC);
2902         adapter->stats.rlec += rd32(E1000_RLEC);
2903         adapter->stats.xonrxc += rd32(E1000_XONRXC);
2904         adapter->stats.xontxc += rd32(E1000_XONTXC);
2905         adapter->stats.xoffrxc += rd32(E1000_XOFFRXC);
2906         adapter->stats.xofftxc += rd32(E1000_XOFFTXC);
2907         adapter->stats.fcruc += rd32(E1000_FCRUC);
2908         adapter->stats.gptc += rd32(E1000_GPTC);
2909         adapter->stats.gotc += rd32(E1000_GOTCL);
2910         rd32(E1000_GOTCH); /* clear GOTCL */
2911         adapter->stats.rnbc += rd32(E1000_RNBC);
2912         adapter->stats.ruc += rd32(E1000_RUC);
2913         adapter->stats.rfc += rd32(E1000_RFC);
2914         adapter->stats.rjc += rd32(E1000_RJC);
2915         adapter->stats.tor += rd32(E1000_TORH);
2916         adapter->stats.tot += rd32(E1000_TOTH);
2917         adapter->stats.tpr += rd32(E1000_TPR);
2918
2919         adapter->stats.ptc64 += rd32(E1000_PTC64);
2920         adapter->stats.ptc127 += rd32(E1000_PTC127);
2921         adapter->stats.ptc255 += rd32(E1000_PTC255);
2922         adapter->stats.ptc511 += rd32(E1000_PTC511);
2923         adapter->stats.ptc1023 += rd32(E1000_PTC1023);
2924         adapter->stats.ptc1522 += rd32(E1000_PTC1522);
2925
2926         adapter->stats.mptc += rd32(E1000_MPTC);
2927         adapter->stats.bptc += rd32(E1000_BPTC);
2928
2929         /* used for adaptive IFS */
2930
2931         hw->mac.tx_packet_delta = rd32(E1000_TPT);
2932         adapter->stats.tpt += hw->mac.tx_packet_delta;
2933         hw->mac.collision_delta = rd32(E1000_COLC);
2934         adapter->stats.colc += hw->mac.collision_delta;
2935
2936         adapter->stats.algnerrc += rd32(E1000_ALGNERRC);
2937         adapter->stats.rxerrc += rd32(E1000_RXERRC);
2938         adapter->stats.tncrs += rd32(E1000_TNCRS);
2939         adapter->stats.tsctc += rd32(E1000_TSCTC);
2940         adapter->stats.tsctfc += rd32(E1000_TSCTFC);
2941
2942         adapter->stats.iac += rd32(E1000_IAC);
2943         adapter->stats.icrxoc += rd32(E1000_ICRXOC);
2944         adapter->stats.icrxptc += rd32(E1000_ICRXPTC);
2945         adapter->stats.icrxatc += rd32(E1000_ICRXATC);
2946         adapter->stats.ictxptc += rd32(E1000_ICTXPTC);
2947         adapter->stats.ictxatc += rd32(E1000_ICTXATC);
2948         adapter->stats.ictxqec += rd32(E1000_ICTXQEC);
2949         adapter->stats.ictxqmtc += rd32(E1000_ICTXQMTC);
2950         adapter->stats.icrxdmtc += rd32(E1000_ICRXDMTC);
2951
2952         /* Fill out the OS statistics structure */
2953         adapter->net_stats.multicast = adapter->stats.mprc;
2954         adapter->net_stats.collisions = adapter->stats.colc;
2955
2956         /* Rx Errors */
2957
2958         /* RLEC on some newer hardware can be incorrect so build
2959         * our own version based on RUC and ROC */
2960         adapter->net_stats.rx_errors = adapter->stats.rxerrc +
2961                 adapter->stats.crcerrs + adapter->stats.algnerrc +
2962                 adapter->stats.ruc + adapter->stats.roc +
2963                 adapter->stats.cexterr;
2964         adapter->net_stats.rx_length_errors = adapter->stats.ruc +
2965                                               adapter->stats.roc;
2966         adapter->net_stats.rx_crc_errors = adapter->stats.crcerrs;
2967         adapter->net_stats.rx_frame_errors = adapter->stats.algnerrc;
2968         adapter->net_stats.rx_missed_errors = adapter->stats.mpc;
2969
2970         /* Tx Errors */
2971         adapter->net_stats.tx_errors = adapter->stats.ecol +
2972                                        adapter->stats.latecol;
2973         adapter->net_stats.tx_aborted_errors = adapter->stats.ecol;
2974         adapter->net_stats.tx_window_errors = adapter->stats.latecol;
2975         adapter->net_stats.tx_carrier_errors = adapter->stats.tncrs;
2976
2977         /* Tx Dropped needs to be maintained elsewhere */
2978
2979         /* Phy Stats */
2980         if (hw->phy.media_type == e1000_media_type_copper) {
2981                 if ((adapter->link_speed == SPEED_1000) &&
2982                    (!hw->phy.ops.read_phy_reg(hw, PHY_1000T_STATUS,
2983                                               &phy_tmp))) {
2984                         phy_tmp &= PHY_IDLE_ERROR_COUNT_MASK;
2985                         adapter->phy_stats.idle_errors += phy_tmp;
2986                 }
2987         }
2988
2989         /* Management Stats */
2990         adapter->stats.mgptc += rd32(E1000_MGTPTC);
2991         adapter->stats.mgprc += rd32(E1000_MGTPRC);
2992         adapter->stats.mgpdc += rd32(E1000_MGTPDC);
2993 }
2994
2995
2996 static irqreturn_t igb_msix_other(int irq, void *data)
2997 {
2998         struct net_device *netdev = data;
2999         struct igb_adapter *adapter = netdev_priv(netdev);
3000         struct e1000_hw *hw = &adapter->hw;
3001         u32 eicr;
3002         /* disable interrupts from the "other" bit, avoid re-entry */
3003         wr32(E1000_EIMC, E1000_EIMS_OTHER);
3004
3005         eicr = rd32(E1000_EICR);
3006
3007         if (eicr & E1000_EIMS_OTHER) {
3008                 u32 icr = rd32(E1000_ICR);
3009                 /* reading ICR causes bit 31 of EICR to be cleared */
3010                 if (!(icr & E1000_ICR_LSC))
3011                         goto no_link_interrupt;
3012                 hw->mac.get_link_status = 1;
3013                 /* guard against interrupt when we're going down */
3014                 if (!test_bit(__IGB_DOWN, &adapter->state))
3015                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
3016         }
3017
3018 no_link_interrupt:
3019         wr32(E1000_IMS, E1000_IMS_LSC);
3020         wr32(E1000_EIMS, E1000_EIMS_OTHER);
3021
3022         return IRQ_HANDLED;
3023 }
3024
3025 static irqreturn_t igb_msix_tx(int irq, void *data)
3026 {
3027         struct igb_ring *tx_ring = data;
3028         struct igb_adapter *adapter = tx_ring->adapter;
3029         struct e1000_hw *hw = &adapter->hw;
3030
3031         if (!tx_ring->itr_val)
3032                 wr32(E1000_EIMC, tx_ring->eims_value);
3033
3034         tx_ring->total_bytes = 0;
3035         tx_ring->total_packets = 0;
3036         if (!igb_clean_tx_irq(adapter, tx_ring))
3037                 /* Ring was not completely cleaned, so fire another interrupt */
3038                 wr32(E1000_EICS, tx_ring->eims_value);
3039
3040         if (!tx_ring->itr_val)
3041                 wr32(E1000_EIMS, tx_ring->eims_value);
3042         return IRQ_HANDLED;
3043 }
3044
3045 static irqreturn_t igb_msix_rx(int irq, void *data)
3046 {
3047         struct igb_ring *rx_ring = data;
3048         struct igb_adapter *adapter = rx_ring->adapter;
3049         struct e1000_hw *hw = &adapter->hw;
3050
3051         if (!rx_ring->itr_val)
3052                 wr32(E1000_EIMC, rx_ring->eims_value);
3053
3054         if (netif_rx_schedule_prep(adapter->netdev, &rx_ring->napi)) {
3055                 rx_ring->total_bytes = 0;
3056                 rx_ring->total_packets = 0;
3057                 rx_ring->no_itr_adjust = 0;
3058                 __netif_rx_schedule(adapter->netdev, &rx_ring->napi);
3059         } else {
3060                 if (!rx_ring->no_itr_adjust) {
3061                         igb_lower_rx_eitr(adapter, rx_ring);
3062                         rx_ring->no_itr_adjust = 1;
3063                 }
3064         }
3065
3066         return IRQ_HANDLED;
3067 }
3068
3069
3070 /**
3071  * igb_intr_msi - Interrupt Handler
3072  * @irq: interrupt number
3073  * @data: pointer to a network interface device structure
3074  **/
3075 static irqreturn_t igb_intr_msi(int irq, void *data)
3076 {
3077         struct net_device *netdev = data;
3078         struct igb_adapter *adapter = netdev_priv(netdev);
3079         struct napi_struct *napi = &adapter->napi;
3080         struct e1000_hw *hw = &adapter->hw;
3081         /* read ICR disables interrupts using IAM */
3082         u32 icr = rd32(E1000_ICR);
3083
3084         /* Write the ITR value calculated at the end of the
3085          * previous interrupt.
3086          */
3087         if (adapter->set_itr) {
3088                 wr32(E1000_ITR,
3089                         1000000000 / (adapter->itr * 256));
3090                 adapter->set_itr = 0;
3091         }
3092
3093         /* read ICR disables interrupts using IAM */
3094         if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
3095                 hw->mac.get_link_status = 1;
3096                 if (!test_bit(__IGB_DOWN, &adapter->state))
3097                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
3098         }
3099
3100         if (netif_rx_schedule_prep(netdev, napi)) {
3101                 adapter->tx_ring->total_bytes = 0;
3102                 adapter->tx_ring->total_packets = 0;
3103                 adapter->rx_ring->total_bytes = 0;
3104                 adapter->rx_ring->total_packets = 0;
3105                 __netif_rx_schedule(netdev, napi);
3106         }
3107
3108         return IRQ_HANDLED;
3109 }
3110
3111 /**
3112  * igb_intr - Interrupt Handler
3113  * @irq: interrupt number
3114  * @data: pointer to a network interface device structure
3115  **/
3116 static irqreturn_t igb_intr(int irq, void *data)
3117 {
3118         struct net_device *netdev = data;
3119         struct igb_adapter *adapter = netdev_priv(netdev);
3120         struct napi_struct *napi = &adapter->napi;
3121         struct e1000_hw *hw = &adapter->hw;
3122         /* Interrupt Auto-Mask...upon reading ICR, interrupts are masked.  No
3123          * need for the IMC write */
3124         u32 icr = rd32(E1000_ICR);
3125         u32 eicr = 0;
3126         if (!icr)
3127                 return IRQ_NONE;  /* Not our interrupt */
3128
3129         /* Write the ITR value calculated at the end of the
3130          * previous interrupt.
3131          */
3132         if (adapter->set_itr) {
3133                 wr32(E1000_ITR,
3134                         1000000000 / (adapter->itr * 256));
3135                 adapter->set_itr = 0;
3136         }
3137
3138         /* IMS will not auto-mask if INT_ASSERTED is not set, and if it is
3139          * not set, then the adapter didn't send an interrupt */
3140         if (!(icr & E1000_ICR_INT_ASSERTED))
3141                 return IRQ_NONE;
3142
3143         eicr = rd32(E1000_EICR);
3144
3145         if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
3146                 hw->mac.get_link_status = 1;
3147                 /* guard against interrupt when we're going down */
3148                 if (!test_bit(__IGB_DOWN, &adapter->state))
3149                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
3150         }
3151
3152         if (netif_rx_schedule_prep(netdev, napi)) {
3153                 adapter->tx_ring->total_bytes = 0;
3154                 adapter->rx_ring->total_bytes = 0;
3155                 adapter->tx_ring->total_packets = 0;
3156                 adapter->rx_ring->total_packets = 0;
3157                 __netif_rx_schedule(netdev, napi);
3158         }
3159
3160         return IRQ_HANDLED;
3161 }
3162
3163 /**
3164  * igb_clean - NAPI Rx polling callback
3165  * @adapter: board private structure
3166  **/
3167 static int igb_clean(struct napi_struct *napi, int budget)
3168 {
3169         struct igb_adapter *adapter = container_of(napi, struct igb_adapter,
3170                                                    napi);
3171         struct net_device *netdev = adapter->netdev;
3172         int tx_clean_complete = 1, work_done = 0;
3173         int i;
3174
3175         /* Must NOT use netdev_priv macro here. */
3176         adapter = netdev->priv;
3177
3178         /* Keep link state information with original netdev */
3179         if (!netif_carrier_ok(netdev))
3180                 goto quit_polling;
3181
3182         /* igb_clean is called per-cpu.  This lock protects tx_ring[i] from
3183          * being cleaned by multiple cpus simultaneously.  A failure obtaining
3184          * the lock means tx_ring[i] is currently being cleaned anyway. */
3185         for (i = 0; i < adapter->num_tx_queues; i++) {
3186                 if (spin_trylock(&adapter->tx_ring[i].tx_clean_lock)) {
3187                         tx_clean_complete &= igb_clean_tx_irq(adapter,
3188                                                         &adapter->tx_ring[i]);
3189                         spin_unlock(&adapter->tx_ring[i].tx_clean_lock);
3190                 }
3191         }
3192
3193         for (i = 0; i < adapter->num_rx_queues; i++)
3194                 igb_clean_rx_irq_adv(adapter, &adapter->rx_ring[i], &work_done,
3195                                      adapter->rx_ring[i].napi.weight);
3196
3197         /* If no Tx and not enough Rx work done, exit the polling mode */
3198         if ((tx_clean_complete && (work_done < budget)) ||
3199             !netif_running(netdev)) {
3200 quit_polling:
3201                 if (adapter->itr_setting & 3)
3202                         igb_set_itr(adapter, E1000_ITR, false);
3203                 netif_rx_complete(netdev, napi);
3204                 if (!test_bit(__IGB_DOWN, &adapter->state))
3205                         igb_irq_enable(adapter);
3206                 return 0;
3207         }
3208
3209         return 1;
3210 }
3211
3212 static int igb_clean_rx_ring_msix(struct napi_struct *napi, int budget)
3213 {
3214         struct igb_ring *rx_ring = container_of(napi, struct igb_ring, napi);
3215         struct igb_adapter *adapter = rx_ring->adapter;
3216         struct e1000_hw *hw = &adapter->hw;
3217         struct net_device *netdev = adapter->netdev;
3218         int work_done = 0;
3219
3220         /* Keep link state information with original netdev */
3221         if (!netif_carrier_ok(netdev))
3222                 goto quit_polling;
3223
3224         igb_clean_rx_irq_adv(adapter, rx_ring, &work_done, budget);
3225
3226
3227         /* If not enough Rx work done, exit the polling mode */
3228         if ((work_done == 0) || !netif_running(netdev)) {
3229 quit_polling:
3230                 netif_rx_complete(netdev, napi);
3231
3232                 wr32(E1000_EIMS, rx_ring->eims_value);
3233                 if ((adapter->itr_setting & 3) && !rx_ring->no_itr_adjust &&
3234                     (rx_ring->total_packets > IGB_DYN_ITR_PACKET_THRESHOLD)) {
3235                         int mean_size = rx_ring->total_bytes /
3236                                         rx_ring->total_packets;
3237                         if (mean_size < IGB_DYN_ITR_LENGTH_LOW)
3238                                 igb_raise_rx_eitr(adapter, rx_ring);
3239                         else if (mean_size > IGB_DYN_ITR_LENGTH_HIGH)
3240                                 igb_lower_rx_eitr(adapter, rx_ring);
3241                 }
3242                 return 0;
3243         }
3244
3245         return 1;
3246 }
3247 /**
3248  * igb_clean_tx_irq - Reclaim resources after transmit completes
3249  * @adapter: board private structure
3250  * returns true if ring is completely cleaned
3251  **/
3252 static bool igb_clean_tx_irq(struct igb_adapter *adapter,
3253                                   struct igb_ring *tx_ring)
3254 {
3255         struct net_device *netdev = adapter->netdev;
3256         struct e1000_hw *hw = &adapter->hw;
3257         struct e1000_tx_desc *tx_desc;
3258         struct igb_buffer *buffer_info;
3259         struct sk_buff *skb;
3260         unsigned int i;
3261         u32 head, oldhead;
3262         unsigned int count = 0;
3263         bool cleaned = false;
3264         bool retval = true;
3265         unsigned int total_bytes = 0, total_packets = 0;
3266
3267         rmb();
3268         head = *(volatile u32 *)((struct e1000_tx_desc *)tx_ring->desc
3269                                  + tx_ring->count);
3270         head = le32_to_cpu(head);
3271         i = tx_ring->next_to_clean;
3272         while (1) {
3273                 while (i != head) {
3274                         cleaned = true;
3275                         tx_desc = E1000_TX_DESC(*tx_ring, i);
3276                         buffer_info = &tx_ring->buffer_info[i];
3277                         skb = buffer_info->skb;
3278
3279                         if (skb) {
3280                                 unsigned int segs, bytecount;
3281                                 /* gso_segs is currently only valid for tcp */
3282                                 segs = skb_shinfo(skb)->gso_segs ?: 1;
3283                                 /* multiply data chunks by size of headers */
3284                                 bytecount = ((segs - 1) * skb_headlen(skb)) +
3285                                             skb->len;
3286                                 total_packets += segs;
3287                                 total_bytes += bytecount;
3288                         }
3289
3290                         igb_unmap_and_free_tx_resource(adapter, buffer_info);
3291                         tx_desc->upper.data = 0;
3292
3293                         i++;
3294                         if (i == tx_ring->count)
3295                                 i = 0;
3296
3297                         count++;
3298                         if (count == IGB_MAX_TX_CLEAN) {
3299                                 retval = false;
3300                                 goto done_cleaning;
3301                         }
3302                 }
3303                 oldhead = head;
3304                 rmb();
3305                 head = *(volatile u32 *)((struct e1000_tx_desc *)tx_ring->desc
3306                                          + tx_ring->count);
3307                 head = le32_to_cpu(head);
3308                 if (head == oldhead)
3309                         goto done_cleaning;
3310         }  /* while (1) */
3311
3312 done_cleaning:
3313         tx_ring->next_to_clean = i;
3314
3315         if (unlikely(cleaned &&
3316                      netif_carrier_ok(netdev) &&
3317                      IGB_DESC_UNUSED(tx_ring) >= IGB_TX_QUEUE_WAKE)) {
3318                 /* Make sure that anybody stopping the queue after this
3319                  * sees the new next_to_clean.
3320                  */
3321                 smp_mb();
3322                 if (netif_queue_stopped(netdev) &&
3323                     !(test_bit(__IGB_DOWN, &adapter->state))) {
3324                         netif_wake_queue(netdev);
3325                         ++adapter->restart_queue;
3326                 }
3327         }
3328
3329         if (tx_ring->detect_tx_hung) {
3330                 /* Detect a transmit hang in hardware, this serializes the
3331                  * check with the clearing of time_stamp and movement of i */
3332                 tx_ring->detect_tx_hung = false;
3333                 if (tx_ring->buffer_info[i].time_stamp &&
3334                     time_after(jiffies, tx_ring->buffer_info[i].time_stamp +
3335                                (adapter->tx_timeout_factor * HZ))
3336                     && !(rd32(E1000_STATUS) &
3337                          E1000_STATUS_TXOFF)) {
3338
3339                         tx_desc = E1000_TX_DESC(*tx_ring, i);
3340                         /* detected Tx unit hang */
3341                         dev_err(&adapter->pdev->dev,
3342                                 "Detected Tx Unit Hang\n"
3343                                 "  Tx Queue             <%lu>\n"
3344                                 "  TDH                  <%x>\n"
3345                                 "  TDT                  <%x>\n"
3346                                 "  next_to_use          <%x>\n"
3347                                 "  next_to_clean        <%x>\n"
3348                                 "  head (WB)            <%x>\n"
3349                                 "buffer_info[next_to_clean]\n"
3350                                 "  time_stamp           <%lx>\n"
3351                                 "  jiffies              <%lx>\n"
3352                                 "  desc.status          <%x>\n",
3353                                 (unsigned long)((tx_ring - adapter->tx_ring) /
3354                                         sizeof(struct igb_ring)),
3355                                 readl(adapter->hw.hw_addr + tx_ring->head),
3356                                 readl(adapter->hw.hw_addr + tx_ring->tail),
3357                                 tx_ring->next_to_use,
3358                                 tx_ring->next_to_clean,
3359                                 head,
3360                                 tx_ring->buffer_info[i].time_stamp,
3361                                 jiffies,
3362                                 tx_desc->upper.fields.status);
3363                         netif_stop_queue(netdev);
3364                 }
3365         }
3366         tx_ring->total_bytes += total_bytes;
3367         tx_ring->total_packets += total_packets;
3368         adapter->net_stats.tx_bytes += total_bytes;
3369         adapter->net_stats.tx_packets += total_packets;
3370         return retval;
3371 }
3372
3373
3374 /**
3375  * igb_receive_skb - helper function to handle rx indications
3376  * @adapter: board private structure
3377  * @status: descriptor status field as written by hardware
3378  * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
3379  * @skb: pointer to sk_buff to be indicated to stack
3380  **/
3381 static void igb_receive_skb(struct igb_adapter *adapter, u8 status, u16 vlan,
3382                             struct sk_buff *skb)
3383 {
3384         if (adapter->vlgrp && (status & E1000_RXD_STAT_VP))
3385                 vlan_hwaccel_receive_skb(skb, adapter->vlgrp,
3386                                          le16_to_cpu(vlan) &
3387                                          E1000_RXD_SPC_VLAN_MASK);
3388         else
3389                 netif_receive_skb(skb);
3390 }
3391
3392
3393 static inline void igb_rx_checksum_adv(struct igb_adapter *adapter,
3394                                        u32 status_err, struct sk_buff *skb)
3395 {
3396         skb->ip_summed = CHECKSUM_NONE;
3397
3398         /* Ignore Checksum bit is set or checksum is disabled through ethtool */
3399         if ((status_err & E1000_RXD_STAT_IXSM) || !adapter->rx_csum)
3400                 return;
3401         /* TCP/UDP checksum error bit is set */
3402         if (status_err &
3403             (E1000_RXDEXT_STATERR_TCPE | E1000_RXDEXT_STATERR_IPE)) {
3404                 /* let the stack verify checksum errors */
3405                 adapter->hw_csum_err++;
3406                 return;
3407         }
3408         /* It must be a TCP or UDP packet with a valid checksum */
3409         if (status_err & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS))
3410                 skb->ip_summed = CHECKSUM_UNNECESSARY;
3411
3412         adapter->hw_csum_good++;
3413 }
3414
3415 static bool igb_clean_rx_irq_adv(struct igb_adapter *adapter,
3416                                       struct igb_ring *rx_ring,
3417                                       int *work_done, int budget)
3418 {
3419         struct net_device *netdev = adapter->netdev;
3420         struct pci_dev *pdev = adapter->pdev;
3421         union e1000_adv_rx_desc *rx_desc , *next_rxd;
3422         struct igb_buffer *buffer_info , *next_buffer;
3423         struct sk_buff *skb;
3424         unsigned int i, j;
3425         u32 length, hlen, staterr;
3426         bool cleaned = false;
3427         int cleaned_count = 0;
3428         unsigned int total_bytes = 0, total_packets = 0;
3429
3430         i = rx_ring->next_to_clean;
3431         rx_desc = E1000_RX_DESC_ADV(*rx_ring, i);
3432         staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
3433
3434         while (staterr & E1000_RXD_STAT_DD) {
3435                 if (*work_done >= budget)
3436                         break;
3437                 (*work_done)++;
3438                 buffer_info = &rx_ring->buffer_info[i];
3439
3440                 /* HW will not DMA in data larger than the given buffer, even
3441                  * if it parses the (NFS, of course) header to be larger.  In
3442                  * that case, it fills the header buffer and spills the rest
3443                  * into the page.
3444                  */
3445                 hlen = le16_to_cpu((rx_desc->wb.lower.lo_dword.hdr_info &
3446                   E1000_RXDADV_HDRBUFLEN_MASK) >> E1000_RXDADV_HDRBUFLEN_SHIFT);
3447                 if (hlen > adapter->rx_ps_hdr_size)
3448                         hlen = adapter->rx_ps_hdr_size;
3449
3450                 length = le16_to_cpu(rx_desc->wb.upper.length);
3451                 cleaned = true;
3452                 cleaned_count++;
3453
3454                 if (rx_ring->pending_skb != NULL) {
3455                         skb = rx_ring->pending_skb;
3456                         rx_ring->pending_skb = NULL;
3457                         j = rx_ring->pending_skb_page;
3458                 } else {
3459                         skb = buffer_info->skb;
3460                         prefetch(skb->data - NET_IP_ALIGN);
3461                         buffer_info->skb = NULL;
3462                         if (hlen) {
3463                                 pci_unmap_single(pdev, buffer_info->dma,
3464                                                  adapter->rx_ps_hdr_size +
3465                                                    NET_IP_ALIGN,
3466                                                  PCI_DMA_FROMDEVICE);
3467                                 skb_put(skb, hlen);
3468                         } else {
3469                                 pci_unmap_single(pdev, buffer_info->dma,
3470                                                  adapter->rx_buffer_len +
3471                                                    NET_IP_ALIGN,
3472                                                  PCI_DMA_FROMDEVICE);
3473                                 skb_put(skb, length);
3474                                 goto send_up;
3475                         }
3476                         j = 0;
3477                 }
3478
3479                 while (length) {
3480                         pci_unmap_page(pdev, buffer_info->page_dma,
3481                                 PAGE_SIZE, PCI_DMA_FROMDEVICE);
3482                         buffer_info->page_dma = 0;
3483                         skb_fill_page_desc(skb, j, buffer_info->page,
3484                                                 0, length);
3485                         buffer_info->page = NULL;
3486
3487                         skb->len += length;
3488                         skb->data_len += length;
3489                         skb->truesize += length;
3490                         rx_desc->wb.upper.status_error = 0;
3491                         if (staterr & E1000_RXD_STAT_EOP)
3492                                 break;
3493
3494                         j++;
3495                         cleaned_count++;
3496                         i++;
3497                         if (i == rx_ring->count)
3498                                 i = 0;
3499
3500                         buffer_info = &rx_ring->buffer_info[i];
3501                         rx_desc = E1000_RX_DESC_ADV(*rx_ring, i);
3502                         staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
3503                         length = le16_to_cpu(rx_desc->wb.upper.length);
3504                         if (!(staterr & E1000_RXD_STAT_DD)) {
3505                                 rx_ring->pending_skb = skb;
3506                                 rx_ring->pending_skb_page = j;
3507                                 goto out;
3508                         }
3509                 }
3510 send_up:
3511                 pskb_trim(skb, skb->len - 4);
3512                 i++;
3513                 if (i == rx_ring->count)
3514                         i = 0;
3515                 next_rxd = E1000_RX_DESC_ADV(*rx_ring, i);
3516                 prefetch(next_rxd);
3517                 next_buffer = &rx_ring->buffer_info[i];
3518
3519                 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
3520                         dev_kfree_skb_irq(skb);
3521                         goto next_desc;
3522                 }
3523                 rx_ring->no_itr_adjust |= (staterr & E1000_RXD_STAT_DYNINT);
3524
3525                 total_bytes += skb->len;
3526                 total_packets++;
3527
3528                 igb_rx_checksum_adv(adapter, staterr, skb);
3529
3530                 skb->protocol = eth_type_trans(skb, netdev);
3531
3532                 igb_receive_skb(adapter, staterr, rx_desc->wb.upper.vlan, skb);
3533
3534                 netdev->last_rx = jiffies;
3535
3536 next_desc:
3537                 rx_desc->wb.upper.status_error = 0;
3538
3539                 /* return some buffers to hardware, one at a time is too slow */
3540                 if (cleaned_count >= IGB_RX_BUFFER_WRITE) {
3541                         igb_alloc_rx_buffers_adv(adapter, rx_ring,
3542                                                  cleaned_count);
3543                         cleaned_count = 0;
3544                 }
3545
3546                 /* use prefetched values */
3547                 rx_desc = next_rxd;
3548                 buffer_info = next_buffer;
3549
3550                 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
3551         }
3552 out:
3553         rx_ring->next_to_clean = i;
3554         cleaned_count = IGB_DESC_UNUSED(rx_ring);
3555
3556         if (cleaned_count)
3557                 igb_alloc_rx_buffers_adv(adapter, rx_ring, cleaned_count);
3558
3559         rx_ring->total_packets += total_packets;
3560         rx_ring->total_bytes += total_bytes;
3561         rx_ring->rx_stats.packets += total_packets;
3562         rx_ring->rx_stats.bytes += total_bytes;
3563         adapter->net_stats.rx_bytes += total_bytes;
3564         adapter->net_stats.rx_packets += total_packets;
3565         return cleaned;
3566 }
3567
3568
3569 /**
3570  * igb_alloc_rx_buffers_adv - Replace used receive buffers; packet split
3571  * @adapter: address of board private structure
3572  **/
3573 static void igb_alloc_rx_buffers_adv(struct igb_adapter *adapter,
3574                                      struct igb_ring *rx_ring,
3575                                      int cleaned_count)
3576 {
3577         struct net_device *netdev = adapter->netdev;
3578         struct pci_dev *pdev = adapter->pdev;
3579         union e1000_adv_rx_desc *rx_desc;
3580         struct igb_buffer *buffer_info;
3581         struct sk_buff *skb;
3582         unsigned int i;
3583
3584         i = rx_ring->next_to_use;
3585         buffer_info = &rx_ring->buffer_info[i];
3586
3587         while (cleaned_count--) {
3588                 rx_desc = E1000_RX_DESC_ADV(*rx_ring, i);
3589
3590                 if (adapter->rx_ps_hdr_size && !buffer_info->page) {
3591                         buffer_info->page = alloc_page(GFP_ATOMIC);
3592                         if (!buffer_info->page) {
3593                                 adapter->alloc_rx_buff_failed++;
3594                                 goto no_buffers;
3595                         }
3596                         buffer_info->page_dma =
3597                                 pci_map_page(pdev,
3598                                              buffer_info->page,
3599                                              0, PAGE_SIZE,
3600                                              PCI_DMA_FROMDEVICE);
3601                 }
3602
3603                 if (!buffer_info->skb) {
3604                         int bufsz;
3605
3606                         if (adapter->rx_ps_hdr_size)
3607                                 bufsz = adapter->rx_ps_hdr_size;
3608                         else
3609                                 bufsz = adapter->rx_buffer_len;
3610                         bufsz += NET_IP_ALIGN;
3611                         skb = netdev_alloc_skb(netdev, bufsz);
3612
3613                         if (!skb) {
3614                                 adapter->alloc_rx_buff_failed++;
3615                                 goto no_buffers;
3616                         }
3617
3618                         /* Make buffer alignment 2 beyond a 16 byte boundary
3619                          * this will result in a 16 byte aligned IP header after
3620                          * the 14 byte MAC header is removed
3621                          */
3622                         skb_reserve(skb, NET_IP_ALIGN);
3623
3624                         buffer_info->skb = skb;
3625                         buffer_info->dma = pci_map_single(pdev, skb->data,
3626                                                           bufsz,
3627                                                           PCI_DMA_FROMDEVICE);
3628
3629                 }
3630                 /* Refresh the desc even if buffer_addrs didn't change because
3631                  * each write-back erases this info. */
3632                 if (adapter->rx_ps_hdr_size) {
3633                         rx_desc->read.pkt_addr =
3634                              cpu_to_le64(buffer_info->page_dma);
3635                         rx_desc->read.hdr_addr = cpu_to_le64(buffer_info->dma);
3636                 } else {
3637                         rx_desc->read.pkt_addr =
3638                              cpu_to_le64(buffer_info->dma);
3639                         rx_desc->read.hdr_addr = 0;
3640                 }
3641
3642                 i++;
3643                 if (i == rx_ring->count)
3644                         i = 0;
3645                 buffer_info = &rx_ring->buffer_info[i];
3646         }
3647
3648 no_buffers:
3649         if (rx_ring->next_to_use != i) {
3650                 rx_ring->next_to_use = i;
3651                 if (i == 0)
3652                         i = (rx_ring->count - 1);
3653                 else
3654                         i--;
3655
3656                 /* Force memory writes to complete before letting h/w
3657                  * know there are new descriptors to fetch.  (Only
3658                  * applicable for weak-ordered memory model archs,
3659                  * such as IA-64). */
3660                 wmb();
3661                 writel(i, adapter->hw.hw_addr + rx_ring->tail);
3662         }
3663 }
3664
3665 /**
3666  * igb_mii_ioctl -
3667  * @netdev:
3668  * @ifreq:
3669  * @cmd:
3670  **/
3671 static int igb_mii_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
3672 {
3673         struct igb_adapter *adapter = netdev_priv(netdev);
3674         struct mii_ioctl_data *data = if_mii(ifr);
3675
3676         if (adapter->hw.phy.media_type != e1000_media_type_copper)
3677                 return -EOPNOTSUPP;
3678
3679         switch (cmd) {
3680         case SIOCGMIIPHY:
3681                 data->phy_id = adapter->hw.phy.addr;
3682                 break;
3683         case SIOCGMIIREG:
3684                 if (!capable(CAP_NET_ADMIN))
3685                         return -EPERM;
3686                 if (adapter->hw.phy.ops.read_phy_reg(&adapter->hw,
3687                                                      data->reg_num
3688                                                      & 0x1F, &data->val_out))
3689                         return -EIO;
3690                 break;
3691         case SIOCSMIIREG:
3692         default:
3693                 return -EOPNOTSUPP;
3694         }
3695         return 0;
3696 }
3697
3698 /**
3699  * igb_ioctl -
3700  * @netdev:
3701  * @ifreq:
3702  * @cmd:
3703  **/
3704 static int igb_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
3705 {
3706         switch (cmd) {
3707         case SIOCGMIIPHY:
3708         case SIOCGMIIREG:
3709         case SIOCSMIIREG:
3710                 return igb_mii_ioctl(netdev, ifr, cmd);
3711         default:
3712                 return -EOPNOTSUPP;
3713         }
3714 }
3715
3716 static void igb_vlan_rx_register(struct net_device *netdev,
3717                                  struct vlan_group *grp)
3718 {
3719         struct igb_adapter *adapter = netdev_priv(netdev);
3720         struct e1000_hw *hw = &adapter->hw;
3721         u32 ctrl, rctl;
3722
3723         igb_irq_disable(adapter);
3724         adapter->vlgrp = grp;
3725
3726         if (grp) {
3727                 /* enable VLAN tag insert/strip */
3728                 ctrl = rd32(E1000_CTRL);
3729                 ctrl |= E1000_CTRL_VME;
3730                 wr32(E1000_CTRL, ctrl);
3731
3732                 /* enable VLAN receive filtering */
3733                 rctl = rd32(E1000_RCTL);
3734                 rctl |= E1000_RCTL_VFE;
3735                 rctl &= ~E1000_RCTL_CFIEN;
3736                 wr32(E1000_RCTL, rctl);
3737                 igb_update_mng_vlan(adapter);
3738                 wr32(E1000_RLPML,
3739                                 adapter->max_frame_size + VLAN_TAG_SIZE);
3740         } else {
3741                 /* disable VLAN tag insert/strip */
3742                 ctrl = rd32(E1000_CTRL);
3743                 ctrl &= ~E1000_CTRL_VME;
3744                 wr32(E1000_CTRL, ctrl);
3745
3746                 /* disable VLAN filtering */
3747                 rctl = rd32(E1000_RCTL);
3748                 rctl &= ~E1000_RCTL_VFE;
3749                 wr32(E1000_RCTL, rctl);
3750                 if (adapter->mng_vlan_id != (u16)IGB_MNG_VLAN_NONE) {
3751                         igb_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
3752                         adapter->mng_vlan_id = IGB_MNG_VLAN_NONE;
3753                 }
3754                 wr32(E1000_RLPML,
3755                                 adapter->max_frame_size);
3756         }
3757
3758         if (!test_bit(__IGB_DOWN, &adapter->state))
3759                 igb_irq_enable(adapter);
3760 }
3761
3762 static void igb_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
3763 {
3764         struct igb_adapter *adapter = netdev_priv(netdev);
3765         struct e1000_hw *hw = &adapter->hw;
3766         u32 vfta, index;
3767
3768         if ((adapter->hw.mng_cookie.status &
3769              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
3770             (vid == adapter->mng_vlan_id))
3771                 return;
3772         /* add VID to filter table */
3773         index = (vid >> 5) & 0x7F;
3774         vfta = array_rd32(E1000_VFTA, index);
3775         vfta |= (1 << (vid & 0x1F));
3776         igb_write_vfta(&adapter->hw, index, vfta);
3777 }
3778
3779 static void igb_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
3780 {
3781         struct igb_adapter *adapter = netdev_priv(netdev);
3782         struct e1000_hw *hw = &adapter->hw;
3783         u32 vfta, index;
3784
3785         igb_irq_disable(adapter);
3786         vlan_group_set_device(adapter->vlgrp, vid, NULL);
3787
3788         if (!test_bit(__IGB_DOWN, &adapter->state))
3789                 igb_irq_enable(adapter);
3790
3791         if ((adapter->hw.mng_cookie.status &
3792              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
3793             (vid == adapter->mng_vlan_id)) {
3794                 /* release control to f/w */
3795                 igb_release_hw_control(adapter);
3796                 return;
3797         }
3798
3799         /* remove VID from filter table */
3800         index = (vid >> 5) & 0x7F;
3801         vfta = array_rd32(E1000_VFTA, index);
3802         vfta &= ~(1 << (vid & 0x1F));
3803         igb_write_vfta(&adapter->hw, index, vfta);
3804 }
3805
3806 static void igb_restore_vlan(struct igb_adapter *adapter)
3807 {
3808         igb_vlan_rx_register(adapter->netdev, adapter->vlgrp);
3809
3810         if (adapter->vlgrp) {
3811                 u16 vid;
3812                 for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
3813                         if (!vlan_group_get_device(adapter->vlgrp, vid))
3814                                 continue;
3815                         igb_vlan_rx_add_vid(adapter->netdev, vid);
3816                 }
3817         }
3818 }
3819
3820 int igb_set_spd_dplx(struct igb_adapter *adapter, u16 spddplx)
3821 {
3822         struct e1000_mac_info *mac = &adapter->hw.mac;
3823
3824         mac->autoneg = 0;
3825
3826         /* Fiber NICs only allow 1000 gbps Full duplex */
3827         if ((adapter->hw.phy.media_type == e1000_media_type_fiber) &&
3828                 spddplx != (SPEED_1000 + DUPLEX_FULL)) {
3829                 dev_err(&adapter->pdev->dev,
3830                         "Unsupported Speed/Duplex configuration\n");
3831                 return -EINVAL;
3832         }
3833
3834         switch (spddplx) {
3835         case SPEED_10 + DUPLEX_HALF:
3836                 mac->forced_speed_duplex = ADVERTISE_10_HALF;
3837                 break;
3838         case SPEED_10 + DUPLEX_FULL:
3839                 mac->forced_speed_duplex = ADVERTISE_10_FULL;
3840                 break;
3841         case SPEED_100 + DUPLEX_HALF:
3842                 mac->forced_speed_duplex = ADVERTISE_100_HALF;
3843                 break;
3844         case SPEED_100 + DUPLEX_FULL:
3845                 mac->forced_speed_duplex = ADVERTISE_100_FULL;
3846                 break;
3847         case SPEED_1000 + DUPLEX_FULL:
3848                 mac->autoneg = 1;
3849                 adapter->hw.phy.autoneg_advertised = ADVERTISE_1000_FULL;
3850                 break;
3851         case SPEED_1000 + DUPLEX_HALF: /* not supported */
3852         default:
3853                 dev_err(&adapter->pdev->dev,
3854                         "Unsupported Speed/Duplex configuration\n");
3855                 return -EINVAL;
3856         }
3857         return 0;
3858 }
3859
3860
3861 static int igb_suspend(struct pci_dev *pdev, pm_message_t state)
3862 {
3863         struct net_device *netdev = pci_get_drvdata(pdev);
3864         struct igb_adapter *adapter = netdev_priv(netdev);
3865         struct e1000_hw *hw = &adapter->hw;
3866         u32 ctrl, ctrl_ext, rctl, status;
3867         u32 wufc = adapter->wol;
3868 #ifdef CONFIG_PM
3869         int retval = 0;
3870 #endif
3871
3872         netif_device_detach(netdev);
3873
3874         if (netif_running(netdev)) {
3875                 WARN_ON(test_bit(__IGB_RESETTING, &adapter->state));
3876                 igb_down(adapter);
3877                 igb_free_irq(adapter);
3878         }
3879
3880 #ifdef CONFIG_PM
3881         retval = pci_save_state(pdev);
3882         if (retval)
3883                 return retval;
3884 #endif
3885
3886         status = rd32(E1000_STATUS);
3887         if (status & E1000_STATUS_LU)
3888                 wufc &= ~E1000_WUFC_LNKC;
3889
3890         if (wufc) {
3891                 igb_setup_rctl(adapter);
3892                 igb_set_multi(netdev);
3893
3894                 /* turn on all-multi mode if wake on multicast is enabled */
3895                 if (wufc & E1000_WUFC_MC) {
3896                         rctl = rd32(E1000_RCTL);
3897                         rctl |= E1000_RCTL_MPE;
3898                         wr32(E1000_RCTL, rctl);
3899                 }
3900
3901                 ctrl = rd32(E1000_CTRL);
3902                 /* advertise wake from D3Cold */
3903                 #define E1000_CTRL_ADVD3WUC 0x00100000
3904                 /* phy power management enable */
3905                 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
3906                 ctrl |= E1000_CTRL_ADVD3WUC;
3907                 wr32(E1000_CTRL, ctrl);
3908
3909                 if (adapter->hw.phy.media_type == e1000_media_type_fiber ||
3910                    adapter->hw.phy.media_type ==
3911                                         e1000_media_type_internal_serdes) {
3912                         /* keep the laser running in D3 */
3913                         ctrl_ext = rd32(E1000_CTRL_EXT);
3914                         ctrl_ext |= E1000_CTRL_EXT_SDP7_DATA;
3915                         wr32(E1000_CTRL_EXT, ctrl_ext);
3916                 }
3917
3918                 /* Allow time for pending master requests to run */
3919                 igb_disable_pcie_master(&adapter->hw);
3920
3921                 wr32(E1000_WUC, E1000_WUC_PME_EN);
3922                 wr32(E1000_WUFC, wufc);
3923                 pci_enable_wake(pdev, PCI_D3hot, 1);
3924                 pci_enable_wake(pdev, PCI_D3cold, 1);
3925         } else {
3926                 wr32(E1000_WUC, 0);
3927                 wr32(E1000_WUFC, 0);
3928                 pci_enable_wake(pdev, PCI_D3hot, 0);
3929                 pci_enable_wake(pdev, PCI_D3cold, 0);
3930         }
3931
3932         /* make sure adapter isn't asleep if manageability is enabled */
3933         if (adapter->en_mng_pt) {
3934                 pci_enable_wake(pdev, PCI_D3hot, 1);
3935                 pci_enable_wake(pdev, PCI_D3cold, 1);
3936         }
3937
3938         /* Release control of h/w to f/w.  If f/w is AMT enabled, this
3939          * would have already happened in close and is redundant. */
3940         igb_release_hw_control(adapter);
3941
3942         pci_disable_device(pdev);
3943
3944         pci_set_power_state(pdev, pci_choose_state(pdev, state));
3945
3946         return 0;
3947 }
3948
3949 #ifdef CONFIG_PM
3950 static int igb_resume(struct pci_dev *pdev)
3951 {
3952         struct net_device *netdev = pci_get_drvdata(pdev);
3953         struct igb_adapter *adapter = netdev_priv(netdev);
3954         struct e1000_hw *hw = &adapter->hw;
3955         u32 err;
3956
3957         pci_set_power_state(pdev, PCI_D0);
3958         pci_restore_state(pdev);
3959         err = pci_enable_device(pdev);
3960         if (err) {
3961                 dev_err(&pdev->dev,
3962                         "igb: Cannot enable PCI device from suspend\n");
3963                 return err;
3964         }
3965         pci_set_master(pdev);
3966
3967         pci_enable_wake(pdev, PCI_D3hot, 0);
3968         pci_enable_wake(pdev, PCI_D3cold, 0);
3969
3970         if (netif_running(netdev)) {
3971                 err = igb_request_irq(adapter);
3972                 if (err)
3973                         return err;
3974         }
3975
3976         /* e1000_power_up_phy(adapter); */
3977
3978         igb_reset(adapter);
3979         wr32(E1000_WUS, ~0);
3980
3981         igb_init_manageability(adapter);
3982
3983         if (netif_running(netdev))
3984                 igb_up(adapter);
3985
3986         netif_device_attach(netdev);
3987
3988         /* let the f/w know that the h/w is now under the control of the
3989          * driver. */
3990         igb_get_hw_control(adapter);
3991
3992         return 0;
3993 }
3994 #endif
3995
3996 static void igb_shutdown(struct pci_dev *pdev)
3997 {
3998         igb_suspend(pdev, PMSG_SUSPEND);
3999 }
4000
4001 #ifdef CONFIG_NET_POLL_CONTROLLER
4002 /*
4003  * Polling 'interrupt' - used by things like netconsole to send skbs
4004  * without having to re-enable interrupts. It's not called while
4005  * the interrupt routine is executing.
4006  */
4007 static void igb_netpoll(struct net_device *netdev)
4008 {
4009         struct igb_adapter *adapter = netdev_priv(netdev);
4010         int i;
4011         int work_done = 0;
4012
4013         igb_irq_disable(adapter);
4014         for (i = 0; i < adapter->num_tx_queues; i++)
4015                 igb_clean_tx_irq(adapter, &adapter->tx_ring[i]);
4016
4017         for (i = 0; i < adapter->num_rx_queues; i++)
4018                 igb_clean_rx_irq_adv(adapter, &adapter->rx_ring[i],
4019                                      &work_done,
4020                                      adapter->rx_ring[i].napi.weight);
4021
4022         igb_irq_enable(adapter);
4023 }
4024 #endif /* CONFIG_NET_POLL_CONTROLLER */
4025
4026 /**
4027  * igb_io_error_detected - called when PCI error is detected
4028  * @pdev: Pointer to PCI device
4029  * @state: The current pci connection state
4030  *
4031  * This function is called after a PCI bus error affecting
4032  * this device has been detected.
4033  */
4034 static pci_ers_result_t igb_io_error_detected(struct pci_dev *pdev,
4035                                               pci_channel_state_t state)
4036 {
4037         struct net_device *netdev = pci_get_drvdata(pdev);
4038         struct igb_adapter *adapter = netdev_priv(netdev);
4039
4040         netif_device_detach(netdev);
4041
4042         if (netif_running(netdev))
4043                 igb_down(adapter);
4044         pci_disable_device(pdev);
4045
4046         /* Request a slot slot reset. */
4047         return PCI_ERS_RESULT_NEED_RESET;
4048 }
4049
4050 /**
4051  * igb_io_slot_reset - called after the pci bus has been reset.
4052  * @pdev: Pointer to PCI device
4053  *
4054  * Restart the card from scratch, as if from a cold-boot. Implementation
4055  * resembles the first-half of the igb_resume routine.
4056  */
4057 static pci_ers_result_t igb_io_slot_reset(struct pci_dev *pdev)
4058 {
4059         struct net_device *netdev = pci_get_drvdata(pdev);
4060         struct igb_adapter *adapter = netdev_priv(netdev);
4061         struct e1000_hw *hw = &adapter->hw;
4062
4063         if (pci_enable_device(pdev)) {
4064                 dev_err(&pdev->dev,
4065                         "Cannot re-enable PCI device after reset.\n");
4066                 return PCI_ERS_RESULT_DISCONNECT;
4067         }
4068         pci_set_master(pdev);
4069
4070         pci_enable_wake(pdev, PCI_D3hot, 0);
4071         pci_enable_wake(pdev, PCI_D3cold, 0);
4072
4073         igb_reset(adapter);
4074         wr32(E1000_WUS, ~0);
4075
4076         return PCI_ERS_RESULT_RECOVERED;
4077 }
4078
4079 /**
4080  * igb_io_resume - called when traffic can start flowing again.
4081  * @pdev: Pointer to PCI device
4082  *
4083  * This callback is called when the error recovery driver tells us that
4084  * its OK to resume normal operation. Implementation resembles the
4085  * second-half of the igb_resume routine.
4086  */
4087 static void igb_io_resume(struct pci_dev *pdev)
4088 {
4089         struct net_device *netdev = pci_get_drvdata(pdev);
4090         struct igb_adapter *adapter = netdev_priv(netdev);
4091
4092         igb_init_manageability(adapter);
4093
4094         if (netif_running(netdev)) {
4095                 if (igb_up(adapter)) {
4096                         dev_err(&pdev->dev, "igb_up failed after reset\n");
4097                         return;
4098                 }
4099         }
4100
4101         netif_device_attach(netdev);
4102
4103         /* let the f/w know that the h/w is now under the control of the
4104          * driver. */
4105         igb_get_hw_control(adapter);
4106
4107 }
4108
4109 /* igb_main.c */