netdev: add more functions to netdevice ops
[linux-2.6.git] / drivers / net / ixgb / ixgb_main.c
1 /*******************************************************************************
2
3   Intel PRO/10GbE Linux driver
4   Copyright(c) 1999 - 2008 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   Linux NICS <linux.nics@intel.com>
24   e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
25   Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
26
27 *******************************************************************************/
28
29 #include "ixgb.h"
30
31 char ixgb_driver_name[] = "ixgb";
32 static char ixgb_driver_string[] = "Intel(R) PRO/10GbE Network Driver";
33
34 #define DRIVERNAPI "-NAPI"
35 #define DRV_VERSION "1.0.135-k2" DRIVERNAPI
36 const char ixgb_driver_version[] = DRV_VERSION;
37 static const char ixgb_copyright[] = "Copyright (c) 1999-2008 Intel Corporation.";
38
39 #define IXGB_CB_LENGTH 256
40 static unsigned int copybreak __read_mostly = IXGB_CB_LENGTH;
41 module_param(copybreak, uint, 0644);
42 MODULE_PARM_DESC(copybreak,
43         "Maximum size of packet that is copied to a new buffer on receive");
44
45 /* ixgb_pci_tbl - PCI Device ID Table
46  *
47  * Wildcard entries (PCI_ANY_ID) should come last
48  * Last entry must be all 0s
49  *
50  * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
51  *   Class, Class Mask, private data (not used) }
52  */
53 static struct pci_device_id ixgb_pci_tbl[] = {
54         {INTEL_VENDOR_ID, IXGB_DEVICE_ID_82597EX,
55          PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0},
56         {INTEL_VENDOR_ID, IXGB_DEVICE_ID_82597EX_CX4,
57          PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0},
58         {INTEL_VENDOR_ID, IXGB_DEVICE_ID_82597EX_SR,
59          PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0},
60         {INTEL_VENDOR_ID, IXGB_DEVICE_ID_82597EX_LR,
61          PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0},
62
63         /* required last entry */
64         {0,}
65 };
66
67 MODULE_DEVICE_TABLE(pci, ixgb_pci_tbl);
68
69 /* Local Function Prototypes */
70 static int ixgb_init_module(void);
71 static void ixgb_exit_module(void);
72 static int ixgb_probe(struct pci_dev *pdev, const struct pci_device_id *ent);
73 static void __devexit ixgb_remove(struct pci_dev *pdev);
74 static int ixgb_sw_init(struct ixgb_adapter *adapter);
75 static int ixgb_open(struct net_device *netdev);
76 static int ixgb_close(struct net_device *netdev);
77 static void ixgb_configure_tx(struct ixgb_adapter *adapter);
78 static void ixgb_configure_rx(struct ixgb_adapter *adapter);
79 static void ixgb_setup_rctl(struct ixgb_adapter *adapter);
80 static void ixgb_clean_tx_ring(struct ixgb_adapter *adapter);
81 static void ixgb_clean_rx_ring(struct ixgb_adapter *adapter);
82 static void ixgb_set_multi(struct net_device *netdev);
83 static void ixgb_watchdog(unsigned long data);
84 static int ixgb_xmit_frame(struct sk_buff *skb, struct net_device *netdev);
85 static struct net_device_stats *ixgb_get_stats(struct net_device *netdev);
86 static int ixgb_change_mtu(struct net_device *netdev, int new_mtu);
87 static int ixgb_set_mac(struct net_device *netdev, void *p);
88 static irqreturn_t ixgb_intr(int irq, void *data);
89 static bool ixgb_clean_tx_irq(struct ixgb_adapter *adapter);
90
91 static int ixgb_clean(struct napi_struct *, int);
92 static bool ixgb_clean_rx_irq(struct ixgb_adapter *, int *, int);
93 static void ixgb_alloc_rx_buffers(struct ixgb_adapter *, int);
94
95 static void ixgb_tx_timeout(struct net_device *dev);
96 static void ixgb_tx_timeout_task(struct work_struct *work);
97
98 static void ixgb_vlan_rx_register(struct net_device *netdev,
99                                   struct vlan_group *grp);
100 static void ixgb_vlan_rx_add_vid(struct net_device *netdev, u16 vid);
101 static void ixgb_vlan_rx_kill_vid(struct net_device *netdev, u16 vid);
102 static void ixgb_restore_vlan(struct ixgb_adapter *adapter);
103
104 #ifdef CONFIG_NET_POLL_CONTROLLER
105 /* for netdump / net console */
106 static void ixgb_netpoll(struct net_device *dev);
107 #endif
108
109 static pci_ers_result_t ixgb_io_error_detected (struct pci_dev *pdev,
110                              enum pci_channel_state state);
111 static pci_ers_result_t ixgb_io_slot_reset (struct pci_dev *pdev);
112 static void ixgb_io_resume (struct pci_dev *pdev);
113
114 static struct pci_error_handlers ixgb_err_handler = {
115         .error_detected = ixgb_io_error_detected,
116         .slot_reset = ixgb_io_slot_reset,
117         .resume = ixgb_io_resume,
118 };
119
120 static struct pci_driver ixgb_driver = {
121         .name     = ixgb_driver_name,
122         .id_table = ixgb_pci_tbl,
123         .probe    = ixgb_probe,
124         .remove   = __devexit_p(ixgb_remove),
125         .err_handler = &ixgb_err_handler
126 };
127
128 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
129 MODULE_DESCRIPTION("Intel(R) PRO/10GbE Network Driver");
130 MODULE_LICENSE("GPL");
131 MODULE_VERSION(DRV_VERSION);
132
133 #define DEFAULT_DEBUG_LEVEL_SHIFT 3
134 static int debug = DEFAULT_DEBUG_LEVEL_SHIFT;
135 module_param(debug, int, 0);
136 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
137
138 /**
139  * ixgb_init_module - Driver Registration Routine
140  *
141  * ixgb_init_module is the first routine called when the driver is
142  * loaded. All it does is register with the PCI subsystem.
143  **/
144
145 static int __init
146 ixgb_init_module(void)
147 {
148         printk(KERN_INFO "%s - version %s\n",
149                ixgb_driver_string, ixgb_driver_version);
150
151         printk(KERN_INFO "%s\n", ixgb_copyright);
152
153         return pci_register_driver(&ixgb_driver);
154 }
155
156 module_init(ixgb_init_module);
157
158 /**
159  * ixgb_exit_module - Driver Exit Cleanup Routine
160  *
161  * ixgb_exit_module is called just before the driver is removed
162  * from memory.
163  **/
164
165 static void __exit
166 ixgb_exit_module(void)
167 {
168         pci_unregister_driver(&ixgb_driver);
169 }
170
171 module_exit(ixgb_exit_module);
172
173 /**
174  * ixgb_irq_disable - Mask off interrupt generation on the NIC
175  * @adapter: board private structure
176  **/
177
178 static void
179 ixgb_irq_disable(struct ixgb_adapter *adapter)
180 {
181         IXGB_WRITE_REG(&adapter->hw, IMC, ~0);
182         IXGB_WRITE_FLUSH(&adapter->hw);
183         synchronize_irq(adapter->pdev->irq);
184 }
185
186 /**
187  * ixgb_irq_enable - Enable default interrupt generation settings
188  * @adapter: board private structure
189  **/
190
191 static void
192 ixgb_irq_enable(struct ixgb_adapter *adapter)
193 {
194         u32 val = IXGB_INT_RXT0 | IXGB_INT_RXDMT0 |
195                   IXGB_INT_TXDW | IXGB_INT_LSC;
196         if (adapter->hw.subsystem_vendor_id == SUN_SUBVENDOR_ID)
197                 val |= IXGB_INT_GPI0;
198         IXGB_WRITE_REG(&adapter->hw, IMS, val);
199         IXGB_WRITE_FLUSH(&adapter->hw);
200 }
201
202 int
203 ixgb_up(struct ixgb_adapter *adapter)
204 {
205         struct net_device *netdev = adapter->netdev;
206         int err, irq_flags = IRQF_SHARED;
207         int max_frame = netdev->mtu + ENET_HEADER_SIZE + ENET_FCS_LENGTH;
208         struct ixgb_hw *hw = &adapter->hw;
209
210         /* hardware has been reset, we need to reload some things */
211
212         ixgb_rar_set(hw, netdev->dev_addr, 0);
213         ixgb_set_multi(netdev);
214
215         ixgb_restore_vlan(adapter);
216
217         ixgb_configure_tx(adapter);
218         ixgb_setup_rctl(adapter);
219         ixgb_configure_rx(adapter);
220         ixgb_alloc_rx_buffers(adapter, IXGB_DESC_UNUSED(&adapter->rx_ring));
221
222         /* disable interrupts and get the hardware into a known state */
223         IXGB_WRITE_REG(&adapter->hw, IMC, 0xffffffff);
224
225         /* only enable MSI if bus is in PCI-X mode */
226         if (IXGB_READ_REG(&adapter->hw, STATUS) & IXGB_STATUS_PCIX_MODE) {
227                 err = pci_enable_msi(adapter->pdev);
228                 if (!err) {
229                         adapter->have_msi = 1;
230                         irq_flags = 0;
231                 }
232                 /* proceed to try to request regular interrupt */
233         }
234
235         err = request_irq(adapter->pdev->irq, &ixgb_intr, irq_flags,
236                           netdev->name, netdev);
237         if (err) {
238                 if (adapter->have_msi)
239                         pci_disable_msi(adapter->pdev);
240                 DPRINTK(PROBE, ERR,
241                  "Unable to allocate interrupt Error: %d\n", err);
242                 return err;
243         }
244
245         if ((hw->max_frame_size != max_frame) ||
246                 (hw->max_frame_size !=
247                 (IXGB_READ_REG(hw, MFS) >> IXGB_MFS_SHIFT))) {
248
249                 hw->max_frame_size = max_frame;
250
251                 IXGB_WRITE_REG(hw, MFS, hw->max_frame_size << IXGB_MFS_SHIFT);
252
253                 if (hw->max_frame_size >
254                    IXGB_MAX_ENET_FRAME_SIZE_WITHOUT_FCS + ENET_FCS_LENGTH) {
255                         u32 ctrl0 = IXGB_READ_REG(hw, CTRL0);
256
257                         if (!(ctrl0 & IXGB_CTRL0_JFE)) {
258                                 ctrl0 |= IXGB_CTRL0_JFE;
259                                 IXGB_WRITE_REG(hw, CTRL0, ctrl0);
260                         }
261                 }
262         }
263
264         clear_bit(__IXGB_DOWN, &adapter->flags);
265
266         napi_enable(&adapter->napi);
267         ixgb_irq_enable(adapter);
268
269         mod_timer(&adapter->watchdog_timer, jiffies);
270
271         return 0;
272 }
273
274 void
275 ixgb_down(struct ixgb_adapter *adapter, bool kill_watchdog)
276 {
277         struct net_device *netdev = adapter->netdev;
278
279         /* prevent the interrupt handler from restarting watchdog */
280         set_bit(__IXGB_DOWN, &adapter->flags);
281
282         napi_disable(&adapter->napi);
283         /* waiting for NAPI to complete can re-enable interrupts */
284         ixgb_irq_disable(adapter);
285         free_irq(adapter->pdev->irq, netdev);
286
287         if (adapter->have_msi)
288                 pci_disable_msi(adapter->pdev);
289
290         if (kill_watchdog)
291                 del_timer_sync(&adapter->watchdog_timer);
292
293         adapter->link_speed = 0;
294         adapter->link_duplex = 0;
295         netif_carrier_off(netdev);
296         netif_stop_queue(netdev);
297
298         ixgb_reset(adapter);
299         ixgb_clean_tx_ring(adapter);
300         ixgb_clean_rx_ring(adapter);
301 }
302
303 void
304 ixgb_reset(struct ixgb_adapter *adapter)
305 {
306         struct ixgb_hw *hw = &adapter->hw;
307
308         ixgb_adapter_stop(hw);
309         if (!ixgb_init_hw(hw))
310                 DPRINTK(PROBE, ERR, "ixgb_init_hw failed.\n");
311
312         /* restore frame size information */
313         IXGB_WRITE_REG(hw, MFS, hw->max_frame_size << IXGB_MFS_SHIFT);
314         if (hw->max_frame_size >
315             IXGB_MAX_ENET_FRAME_SIZE_WITHOUT_FCS + ENET_FCS_LENGTH) {
316                 u32 ctrl0 = IXGB_READ_REG(hw, CTRL0);
317                 if (!(ctrl0 & IXGB_CTRL0_JFE)) {
318                         ctrl0 |= IXGB_CTRL0_JFE;
319                         IXGB_WRITE_REG(hw, CTRL0, ctrl0);
320                 }
321         }
322 }
323
324 static const struct net_device_ops ixgb_netdev_ops = {
325         .ndo_open               = ixgb_open,
326         .ndo_stop               = ixgb_close,
327         .ndo_start_xmit         = ixgb_xmit_frame,
328         .ndo_get_stats          = ixgb_get_stats,
329         .ndo_set_multicast_list = ixgb_set_multi,
330         .ndo_validate_addr      = eth_validate_addr,
331         .ndo_set_mac_address    = ixgb_set_mac,
332         .ndo_change_mtu         = ixgb_change_mtu,
333         .ndo_tx_timeout         = ixgb_tx_timeout,
334         .ndo_vlan_rx_register   = ixgb_vlan_rx_register,
335         .ndo_vlan_rx_add_vid    = ixgb_vlan_rx_add_vid,
336         .ndo_vlan_rx_kill_vid   = ixgb_vlan_rx_kill_vid,
337 #ifdef CONFIG_NET_POLL_CONTROLLER
338         .ndo_poll_controller    = ixgb_netpoll,
339 #endif
340 };
341
342 /**
343  * ixgb_probe - Device Initialization Routine
344  * @pdev: PCI device information struct
345  * @ent: entry in ixgb_pci_tbl
346  *
347  * Returns 0 on success, negative on failure
348  *
349  * ixgb_probe initializes an adapter identified by a pci_dev structure.
350  * The OS initialization, configuring of the adapter private structure,
351  * and a hardware reset occur.
352  **/
353
354 static int __devinit
355 ixgb_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
356 {
357         struct net_device *netdev = NULL;
358         struct ixgb_adapter *adapter;
359         static int cards_found = 0;
360         int pci_using_dac;
361         int i;
362         int err;
363
364         err = pci_enable_device(pdev);
365         if (err)
366                 return err;
367
368         if (!(err = pci_set_dma_mask(pdev, DMA_64BIT_MASK)) &&
369             !(err = pci_set_consistent_dma_mask(pdev, DMA_64BIT_MASK))) {
370                 pci_using_dac = 1;
371         } else {
372                 if ((err = pci_set_dma_mask(pdev, DMA_32BIT_MASK)) ||
373                     (err = pci_set_consistent_dma_mask(pdev, DMA_32BIT_MASK))) {
374                         printk(KERN_ERR
375                          "ixgb: No usable DMA configuration, aborting\n");
376                         goto err_dma_mask;
377                 }
378                 pci_using_dac = 0;
379         }
380
381         err = pci_request_regions(pdev, ixgb_driver_name);
382         if (err)
383                 goto err_request_regions;
384
385         pci_set_master(pdev);
386
387         netdev = alloc_etherdev(sizeof(struct ixgb_adapter));
388         if (!netdev) {
389                 err = -ENOMEM;
390                 goto err_alloc_etherdev;
391         }
392
393         SET_NETDEV_DEV(netdev, &pdev->dev);
394
395         pci_set_drvdata(pdev, netdev);
396         adapter = netdev_priv(netdev);
397         adapter->netdev = netdev;
398         adapter->pdev = pdev;
399         adapter->hw.back = adapter;
400         adapter->msg_enable = netif_msg_init(debug, DEFAULT_DEBUG_LEVEL_SHIFT);
401
402         adapter->hw.hw_addr = pci_ioremap_bar(pdev, BAR_0);
403         if (!adapter->hw.hw_addr) {
404                 err = -EIO;
405                 goto err_ioremap;
406         }
407
408         for (i = BAR_1; i <= BAR_5; i++) {
409                 if (pci_resource_len(pdev, i) == 0)
410                         continue;
411                 if (pci_resource_flags(pdev, i) & IORESOURCE_IO) {
412                         adapter->hw.io_base = pci_resource_start(pdev, i);
413                         break;
414                 }
415         }
416
417         netdev->netdev_ops = &ixgb_netdev_ops;
418         ixgb_set_ethtool_ops(netdev);
419         netdev->watchdog_timeo = 5 * HZ;
420         netif_napi_add(netdev, &adapter->napi, ixgb_clean, 64);
421
422         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
423
424         adapter->bd_number = cards_found;
425         adapter->link_speed = 0;
426         adapter->link_duplex = 0;
427
428         /* setup the private structure */
429
430         err = ixgb_sw_init(adapter);
431         if (err)
432                 goto err_sw_init;
433
434         netdev->features = NETIF_F_SG |
435                            NETIF_F_HW_CSUM |
436                            NETIF_F_HW_VLAN_TX |
437                            NETIF_F_HW_VLAN_RX |
438                            NETIF_F_HW_VLAN_FILTER;
439         netdev->features |= NETIF_F_TSO;
440
441         if (pci_using_dac)
442                 netdev->features |= NETIF_F_HIGHDMA;
443
444         /* make sure the EEPROM is good */
445
446         if (!ixgb_validate_eeprom_checksum(&adapter->hw)) {
447                 DPRINTK(PROBE, ERR, "The EEPROM Checksum Is Not Valid\n");
448                 err = -EIO;
449                 goto err_eeprom;
450         }
451
452         ixgb_get_ee_mac_addr(&adapter->hw, netdev->dev_addr);
453         memcpy(netdev->perm_addr, netdev->dev_addr, netdev->addr_len);
454
455         if (!is_valid_ether_addr(netdev->perm_addr)) {
456                 DPRINTK(PROBE, ERR, "Invalid MAC Address\n");
457                 err = -EIO;
458                 goto err_eeprom;
459         }
460
461         adapter->part_num = ixgb_get_ee_pba_number(&adapter->hw);
462
463         init_timer(&adapter->watchdog_timer);
464         adapter->watchdog_timer.function = &ixgb_watchdog;
465         adapter->watchdog_timer.data = (unsigned long)adapter;
466
467         INIT_WORK(&adapter->tx_timeout_task, ixgb_tx_timeout_task);
468
469         strcpy(netdev->name, "eth%d");
470         err = register_netdev(netdev);
471         if (err)
472                 goto err_register;
473
474         /* we're going to reset, so assume we have no link for now */
475
476         netif_carrier_off(netdev);
477         netif_stop_queue(netdev);
478
479         DPRINTK(PROBE, INFO, "Intel(R) PRO/10GbE Network Connection\n");
480         ixgb_check_options(adapter);
481         /* reset the hardware with the new settings */
482
483         ixgb_reset(adapter);
484
485         cards_found++;
486         return 0;
487
488 err_register:
489 err_sw_init:
490 err_eeprom:
491         iounmap(adapter->hw.hw_addr);
492 err_ioremap:
493         free_netdev(netdev);
494 err_alloc_etherdev:
495         pci_release_regions(pdev);
496 err_request_regions:
497 err_dma_mask:
498         pci_disable_device(pdev);
499         return err;
500 }
501
502 /**
503  * ixgb_remove - Device Removal Routine
504  * @pdev: PCI device information struct
505  *
506  * ixgb_remove is called by the PCI subsystem to alert the driver
507  * that it should release a PCI device.  The could be caused by a
508  * Hot-Plug event, or because the driver is going to be removed from
509  * memory.
510  **/
511
512 static void __devexit
513 ixgb_remove(struct pci_dev *pdev)
514 {
515         struct net_device *netdev = pci_get_drvdata(pdev);
516         struct ixgb_adapter *adapter = netdev_priv(netdev);
517
518         flush_scheduled_work();
519
520         unregister_netdev(netdev);
521
522         iounmap(adapter->hw.hw_addr);
523         pci_release_regions(pdev);
524
525         free_netdev(netdev);
526 }
527
528 /**
529  * ixgb_sw_init - Initialize general software structures (struct ixgb_adapter)
530  * @adapter: board private structure to initialize
531  *
532  * ixgb_sw_init initializes the Adapter private data structure.
533  * Fields are initialized based on PCI device information and
534  * OS network device settings (MTU size).
535  **/
536
537 static int __devinit
538 ixgb_sw_init(struct ixgb_adapter *adapter)
539 {
540         struct ixgb_hw *hw = &adapter->hw;
541         struct net_device *netdev = adapter->netdev;
542         struct pci_dev *pdev = adapter->pdev;
543
544         /* PCI config space info */
545
546         hw->vendor_id = pdev->vendor;
547         hw->device_id = pdev->device;
548         hw->subsystem_vendor_id = pdev->subsystem_vendor;
549         hw->subsystem_id = pdev->subsystem_device;
550
551         hw->max_frame_size = netdev->mtu + ENET_HEADER_SIZE + ENET_FCS_LENGTH;
552         adapter->rx_buffer_len = hw->max_frame_size + 8; /* + 8 for errata */
553
554         if ((hw->device_id == IXGB_DEVICE_ID_82597EX)
555            || (hw->device_id == IXGB_DEVICE_ID_82597EX_CX4)
556            || (hw->device_id == IXGB_DEVICE_ID_82597EX_LR)
557            || (hw->device_id == IXGB_DEVICE_ID_82597EX_SR))
558                 hw->mac_type = ixgb_82597;
559         else {
560                 /* should never have loaded on this device */
561                 DPRINTK(PROBE, ERR, "unsupported device id\n");
562         }
563
564         /* enable flow control to be programmed */
565         hw->fc.send_xon = 1;
566
567         set_bit(__IXGB_DOWN, &adapter->flags);
568         return 0;
569 }
570
571 /**
572  * ixgb_open - Called when a network interface is made active
573  * @netdev: network interface device structure
574  *
575  * Returns 0 on success, negative value on failure
576  *
577  * The open entry point is called when a network interface is made
578  * active by the system (IFF_UP).  At this point all resources needed
579  * for transmit and receive operations are allocated, the interrupt
580  * handler is registered with the OS, the watchdog timer is started,
581  * and the stack is notified that the interface is ready.
582  **/
583
584 static int
585 ixgb_open(struct net_device *netdev)
586 {
587         struct ixgb_adapter *adapter = netdev_priv(netdev);
588         int err;
589
590         /* allocate transmit descriptors */
591         err = ixgb_setup_tx_resources(adapter);
592         if (err)
593                 goto err_setup_tx;
594
595         /* allocate receive descriptors */
596
597         err = ixgb_setup_rx_resources(adapter);
598         if (err)
599                 goto err_setup_rx;
600
601         err = ixgb_up(adapter);
602         if (err)
603                 goto err_up;
604
605         return 0;
606
607 err_up:
608         ixgb_free_rx_resources(adapter);
609 err_setup_rx:
610         ixgb_free_tx_resources(adapter);
611 err_setup_tx:
612         ixgb_reset(adapter);
613
614         return err;
615 }
616
617 /**
618  * ixgb_close - Disables a network interface
619  * @netdev: network interface device structure
620  *
621  * Returns 0, this is not allowed to fail
622  *
623  * The close entry point is called when an interface is de-activated
624  * by the OS.  The hardware is still under the drivers control, but
625  * needs to be disabled.  A global MAC reset is issued to stop the
626  * hardware, and all transmit and receive resources are freed.
627  **/
628
629 static int
630 ixgb_close(struct net_device *netdev)
631 {
632         struct ixgb_adapter *adapter = netdev_priv(netdev);
633
634         ixgb_down(adapter, true);
635
636         ixgb_free_tx_resources(adapter);
637         ixgb_free_rx_resources(adapter);
638
639         return 0;
640 }
641
642 /**
643  * ixgb_setup_tx_resources - allocate Tx resources (Descriptors)
644  * @adapter: board private structure
645  *
646  * Return 0 on success, negative on failure
647  **/
648
649 int
650 ixgb_setup_tx_resources(struct ixgb_adapter *adapter)
651 {
652         struct ixgb_desc_ring *txdr = &adapter->tx_ring;
653         struct pci_dev *pdev = adapter->pdev;
654         int size;
655
656         size = sizeof(struct ixgb_buffer) * txdr->count;
657         txdr->buffer_info = vmalloc(size);
658         if (!txdr->buffer_info) {
659                 DPRINTK(PROBE, ERR,
660                  "Unable to allocate transmit descriptor ring memory\n");
661                 return -ENOMEM;
662         }
663         memset(txdr->buffer_info, 0, size);
664
665         /* round up to nearest 4K */
666
667         txdr->size = txdr->count * sizeof(struct ixgb_tx_desc);
668         txdr->size = ALIGN(txdr->size, 4096);
669
670         txdr->desc = pci_alloc_consistent(pdev, txdr->size, &txdr->dma);
671         if (!txdr->desc) {
672                 vfree(txdr->buffer_info);
673                 DPRINTK(PROBE, ERR,
674                  "Unable to allocate transmit descriptor memory\n");
675                 return -ENOMEM;
676         }
677         memset(txdr->desc, 0, txdr->size);
678
679         txdr->next_to_use = 0;
680         txdr->next_to_clean = 0;
681
682         return 0;
683 }
684
685 /**
686  * ixgb_configure_tx - Configure 82597 Transmit Unit after Reset.
687  * @adapter: board private structure
688  *
689  * Configure the Tx unit of the MAC after a reset.
690  **/
691
692 static void
693 ixgb_configure_tx(struct ixgb_adapter *adapter)
694 {
695         u64 tdba = adapter->tx_ring.dma;
696         u32 tdlen = adapter->tx_ring.count * sizeof(struct ixgb_tx_desc);
697         u32 tctl;
698         struct ixgb_hw *hw = &adapter->hw;
699
700         /* Setup the Base and Length of the Tx Descriptor Ring
701          * tx_ring.dma can be either a 32 or 64 bit value
702          */
703
704         IXGB_WRITE_REG(hw, TDBAL, (tdba & 0x00000000ffffffffULL));
705         IXGB_WRITE_REG(hw, TDBAH, (tdba >> 32));
706
707         IXGB_WRITE_REG(hw, TDLEN, tdlen);
708
709         /* Setup the HW Tx Head and Tail descriptor pointers */
710
711         IXGB_WRITE_REG(hw, TDH, 0);
712         IXGB_WRITE_REG(hw, TDT, 0);
713
714         /* don't set up txdctl, it induces performance problems if configured
715          * incorrectly */
716         /* Set the Tx Interrupt Delay register */
717
718         IXGB_WRITE_REG(hw, TIDV, adapter->tx_int_delay);
719
720         /* Program the Transmit Control Register */
721
722         tctl = IXGB_TCTL_TCE | IXGB_TCTL_TXEN | IXGB_TCTL_TPDE;
723         IXGB_WRITE_REG(hw, TCTL, tctl);
724
725         /* Setup Transmit Descriptor Settings for this adapter */
726         adapter->tx_cmd_type =
727                 IXGB_TX_DESC_TYPE |
728                 (adapter->tx_int_delay_enable ? IXGB_TX_DESC_CMD_IDE : 0);
729 }
730
731 /**
732  * ixgb_setup_rx_resources - allocate Rx resources (Descriptors)
733  * @adapter: board private structure
734  *
735  * Returns 0 on success, negative on failure
736  **/
737
738 int
739 ixgb_setup_rx_resources(struct ixgb_adapter *adapter)
740 {
741         struct ixgb_desc_ring *rxdr = &adapter->rx_ring;
742         struct pci_dev *pdev = adapter->pdev;
743         int size;
744
745         size = sizeof(struct ixgb_buffer) * rxdr->count;
746         rxdr->buffer_info = vmalloc(size);
747         if (!rxdr->buffer_info) {
748                 DPRINTK(PROBE, ERR,
749                  "Unable to allocate receive descriptor ring\n");
750                 return -ENOMEM;
751         }
752         memset(rxdr->buffer_info, 0, size);
753
754         /* Round up to nearest 4K */
755
756         rxdr->size = rxdr->count * sizeof(struct ixgb_rx_desc);
757         rxdr->size = ALIGN(rxdr->size, 4096);
758
759         rxdr->desc = pci_alloc_consistent(pdev, rxdr->size, &rxdr->dma);
760
761         if (!rxdr->desc) {
762                 vfree(rxdr->buffer_info);
763                 DPRINTK(PROBE, ERR,
764                  "Unable to allocate receive descriptors\n");
765                 return -ENOMEM;
766         }
767         memset(rxdr->desc, 0, rxdr->size);
768
769         rxdr->next_to_clean = 0;
770         rxdr->next_to_use = 0;
771
772         return 0;
773 }
774
775 /**
776  * ixgb_setup_rctl - configure the receive control register
777  * @adapter: Board private structure
778  **/
779
780 static void
781 ixgb_setup_rctl(struct ixgb_adapter *adapter)
782 {
783         u32 rctl;
784
785         rctl = IXGB_READ_REG(&adapter->hw, RCTL);
786
787         rctl &= ~(3 << IXGB_RCTL_MO_SHIFT);
788
789         rctl |=
790                 IXGB_RCTL_BAM | IXGB_RCTL_RDMTS_1_2 |
791                 IXGB_RCTL_RXEN | IXGB_RCTL_CFF |
792                 (adapter->hw.mc_filter_type << IXGB_RCTL_MO_SHIFT);
793
794         rctl |= IXGB_RCTL_SECRC;
795
796         if (adapter->rx_buffer_len <= IXGB_RXBUFFER_2048)
797                 rctl |= IXGB_RCTL_BSIZE_2048;
798         else if (adapter->rx_buffer_len <= IXGB_RXBUFFER_4096)
799                 rctl |= IXGB_RCTL_BSIZE_4096;
800         else if (adapter->rx_buffer_len <= IXGB_RXBUFFER_8192)
801                 rctl |= IXGB_RCTL_BSIZE_8192;
802         else if (adapter->rx_buffer_len <= IXGB_RXBUFFER_16384)
803                 rctl |= IXGB_RCTL_BSIZE_16384;
804
805         IXGB_WRITE_REG(&adapter->hw, RCTL, rctl);
806 }
807
808 /**
809  * ixgb_configure_rx - Configure 82597 Receive Unit after Reset.
810  * @adapter: board private structure
811  *
812  * Configure the Rx unit of the MAC after a reset.
813  **/
814
815 static void
816 ixgb_configure_rx(struct ixgb_adapter *adapter)
817 {
818         u64 rdba = adapter->rx_ring.dma;
819         u32 rdlen = adapter->rx_ring.count * sizeof(struct ixgb_rx_desc);
820         struct ixgb_hw *hw = &adapter->hw;
821         u32 rctl;
822         u32 rxcsum;
823
824         /* make sure receives are disabled while setting up the descriptors */
825
826         rctl = IXGB_READ_REG(hw, RCTL);
827         IXGB_WRITE_REG(hw, RCTL, rctl & ~IXGB_RCTL_RXEN);
828
829         /* set the Receive Delay Timer Register */
830
831         IXGB_WRITE_REG(hw, RDTR, adapter->rx_int_delay);
832
833         /* Setup the Base and Length of the Rx Descriptor Ring */
834
835         IXGB_WRITE_REG(hw, RDBAL, (rdba & 0x00000000ffffffffULL));
836         IXGB_WRITE_REG(hw, RDBAH, (rdba >> 32));
837
838         IXGB_WRITE_REG(hw, RDLEN, rdlen);
839
840         /* Setup the HW Rx Head and Tail Descriptor Pointers */
841         IXGB_WRITE_REG(hw, RDH, 0);
842         IXGB_WRITE_REG(hw, RDT, 0);
843
844         /* due to the hardware errata with RXDCTL, we are unable to use any of
845          * the performance enhancing features of it without causing other
846          * subtle bugs, some of the bugs could include receive length
847          * corruption at high data rates (WTHRESH > 0) and/or receive
848          * descriptor ring irregularites (particularly in hardware cache) */
849         IXGB_WRITE_REG(hw, RXDCTL, 0);
850
851         /* Enable Receive Checksum Offload for TCP and UDP */
852         if (adapter->rx_csum) {
853                 rxcsum = IXGB_READ_REG(hw, RXCSUM);
854                 rxcsum |= IXGB_RXCSUM_TUOFL;
855                 IXGB_WRITE_REG(hw, RXCSUM, rxcsum);
856         }
857
858         /* Enable Receives */
859
860         IXGB_WRITE_REG(hw, RCTL, rctl);
861 }
862
863 /**
864  * ixgb_free_tx_resources - Free Tx Resources
865  * @adapter: board private structure
866  *
867  * Free all transmit software resources
868  **/
869
870 void
871 ixgb_free_tx_resources(struct ixgb_adapter *adapter)
872 {
873         struct pci_dev *pdev = adapter->pdev;
874
875         ixgb_clean_tx_ring(adapter);
876
877         vfree(adapter->tx_ring.buffer_info);
878         adapter->tx_ring.buffer_info = NULL;
879
880         pci_free_consistent(pdev, adapter->tx_ring.size,
881                             adapter->tx_ring.desc, adapter->tx_ring.dma);
882
883         adapter->tx_ring.desc = NULL;
884 }
885
886 static void
887 ixgb_unmap_and_free_tx_resource(struct ixgb_adapter *adapter,
888                                 struct ixgb_buffer *buffer_info)
889 {
890         struct pci_dev *pdev = adapter->pdev;
891
892         if (buffer_info->dma)
893                 pci_unmap_page(pdev, buffer_info->dma, buffer_info->length,
894                                PCI_DMA_TODEVICE);
895
896         /* okay to call kfree_skb here instead of kfree_skb_any because
897          * this is never called in interrupt context */
898         if (buffer_info->skb)
899                 dev_kfree_skb(buffer_info->skb);
900
901         buffer_info->skb = NULL;
902         buffer_info->dma = 0;
903         buffer_info->time_stamp = 0;
904         /* these fields must always be initialized in tx
905          * buffer_info->length = 0;
906          * buffer_info->next_to_watch = 0; */
907 }
908
909 /**
910  * ixgb_clean_tx_ring - Free Tx Buffers
911  * @adapter: board private structure
912  **/
913
914 static void
915 ixgb_clean_tx_ring(struct ixgb_adapter *adapter)
916 {
917         struct ixgb_desc_ring *tx_ring = &adapter->tx_ring;
918         struct ixgb_buffer *buffer_info;
919         unsigned long size;
920         unsigned int i;
921
922         /* Free all the Tx ring sk_buffs */
923
924         for (i = 0; i < tx_ring->count; i++) {
925                 buffer_info = &tx_ring->buffer_info[i];
926                 ixgb_unmap_and_free_tx_resource(adapter, buffer_info);
927         }
928
929         size = sizeof(struct ixgb_buffer) * tx_ring->count;
930         memset(tx_ring->buffer_info, 0, size);
931
932         /* Zero out the descriptor ring */
933
934         memset(tx_ring->desc, 0, tx_ring->size);
935
936         tx_ring->next_to_use = 0;
937         tx_ring->next_to_clean = 0;
938
939         IXGB_WRITE_REG(&adapter->hw, TDH, 0);
940         IXGB_WRITE_REG(&adapter->hw, TDT, 0);
941 }
942
943 /**
944  * ixgb_free_rx_resources - Free Rx Resources
945  * @adapter: board private structure
946  *
947  * Free all receive software resources
948  **/
949
950 void
951 ixgb_free_rx_resources(struct ixgb_adapter *adapter)
952 {
953         struct ixgb_desc_ring *rx_ring = &adapter->rx_ring;
954         struct pci_dev *pdev = adapter->pdev;
955
956         ixgb_clean_rx_ring(adapter);
957
958         vfree(rx_ring->buffer_info);
959         rx_ring->buffer_info = NULL;
960
961         pci_free_consistent(pdev, rx_ring->size, rx_ring->desc, rx_ring->dma);
962
963         rx_ring->desc = NULL;
964 }
965
966 /**
967  * ixgb_clean_rx_ring - Free Rx Buffers
968  * @adapter: board private structure
969  **/
970
971 static void
972 ixgb_clean_rx_ring(struct ixgb_adapter *adapter)
973 {
974         struct ixgb_desc_ring *rx_ring = &adapter->rx_ring;
975         struct ixgb_buffer *buffer_info;
976         struct pci_dev *pdev = adapter->pdev;
977         unsigned long size;
978         unsigned int i;
979
980         /* Free all the Rx ring sk_buffs */
981
982         for (i = 0; i < rx_ring->count; i++) {
983                 buffer_info = &rx_ring->buffer_info[i];
984                 if (buffer_info->dma) {
985                         pci_unmap_single(pdev,
986                                          buffer_info->dma,
987                                          buffer_info->length,
988                                          PCI_DMA_FROMDEVICE);
989                         buffer_info->dma = 0;
990                         buffer_info->length = 0;
991                 }
992
993                 if (buffer_info->skb) {
994                         dev_kfree_skb(buffer_info->skb);
995                         buffer_info->skb = NULL;
996                 }
997         }
998
999         size = sizeof(struct ixgb_buffer) * rx_ring->count;
1000         memset(rx_ring->buffer_info, 0, size);
1001
1002         /* Zero out the descriptor ring */
1003
1004         memset(rx_ring->desc, 0, rx_ring->size);
1005
1006         rx_ring->next_to_clean = 0;
1007         rx_ring->next_to_use = 0;
1008
1009         IXGB_WRITE_REG(&adapter->hw, RDH, 0);
1010         IXGB_WRITE_REG(&adapter->hw, RDT, 0);
1011 }
1012
1013 /**
1014  * ixgb_set_mac - Change the Ethernet Address of the NIC
1015  * @netdev: network interface device structure
1016  * @p: pointer to an address structure
1017  *
1018  * Returns 0 on success, negative on failure
1019  **/
1020
1021 static int
1022 ixgb_set_mac(struct net_device *netdev, void *p)
1023 {
1024         struct ixgb_adapter *adapter = netdev_priv(netdev);
1025         struct sockaddr *addr = p;
1026
1027         if (!is_valid_ether_addr(addr->sa_data))
1028                 return -EADDRNOTAVAIL;
1029
1030         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
1031
1032         ixgb_rar_set(&adapter->hw, addr->sa_data, 0);
1033
1034         return 0;
1035 }
1036
1037 /**
1038  * ixgb_set_multi - Multicast and Promiscuous mode set
1039  * @netdev: network interface device structure
1040  *
1041  * The set_multi entry point is called whenever the multicast address
1042  * list or the network interface flags are updated.  This routine is
1043  * responsible for configuring the hardware for proper multicast,
1044  * promiscuous mode, and all-multi behavior.
1045  **/
1046
1047 static void
1048 ixgb_set_multi(struct net_device *netdev)
1049 {
1050         struct ixgb_adapter *adapter = netdev_priv(netdev);
1051         struct ixgb_hw *hw = &adapter->hw;
1052         struct dev_mc_list *mc_ptr;
1053         u32 rctl;
1054         int i;
1055
1056         /* Check for Promiscuous and All Multicast modes */
1057
1058         rctl = IXGB_READ_REG(hw, RCTL);
1059
1060         if (netdev->flags & IFF_PROMISC) {
1061                 rctl |= (IXGB_RCTL_UPE | IXGB_RCTL_MPE);
1062                 rctl &= ~IXGB_RCTL_VFE;
1063         } else {
1064                 if (netdev->flags & IFF_ALLMULTI) {
1065                         rctl |= IXGB_RCTL_MPE;
1066                         rctl &= ~IXGB_RCTL_UPE;
1067                 } else {
1068                         rctl &= ~(IXGB_RCTL_UPE | IXGB_RCTL_MPE);
1069                 }
1070                 rctl |= IXGB_RCTL_VFE;
1071         }
1072
1073         if (netdev->mc_count > IXGB_MAX_NUM_MULTICAST_ADDRESSES) {
1074                 rctl |= IXGB_RCTL_MPE;
1075                 IXGB_WRITE_REG(hw, RCTL, rctl);
1076         } else {
1077                 u8 mta[IXGB_MAX_NUM_MULTICAST_ADDRESSES *
1078                             IXGB_ETH_LENGTH_OF_ADDRESS];
1079
1080                 IXGB_WRITE_REG(hw, RCTL, rctl);
1081
1082                 for (i = 0, mc_ptr = netdev->mc_list;
1083                      mc_ptr;
1084                      i++, mc_ptr = mc_ptr->next)
1085                         memcpy(&mta[i * IXGB_ETH_LENGTH_OF_ADDRESS],
1086                                mc_ptr->dmi_addr, IXGB_ETH_LENGTH_OF_ADDRESS);
1087
1088                 ixgb_mc_addr_list_update(hw, mta, netdev->mc_count, 0);
1089         }
1090 }
1091
1092 /**
1093  * ixgb_watchdog - Timer Call-back
1094  * @data: pointer to netdev cast into an unsigned long
1095  **/
1096
1097 static void
1098 ixgb_watchdog(unsigned long data)
1099 {
1100         struct ixgb_adapter *adapter = (struct ixgb_adapter *)data;
1101         struct net_device *netdev = adapter->netdev;
1102         struct ixgb_desc_ring *txdr = &adapter->tx_ring;
1103
1104         ixgb_check_for_link(&adapter->hw);
1105
1106         if (ixgb_check_for_bad_link(&adapter->hw)) {
1107                 /* force the reset path */
1108                 netif_stop_queue(netdev);
1109         }
1110
1111         if (adapter->hw.link_up) {
1112                 if (!netif_carrier_ok(netdev)) {
1113                         DPRINTK(LINK, INFO,
1114                                 "NIC Link is Up 10000 Mbps Full Duplex\n");
1115                         adapter->link_speed = 10000;
1116                         adapter->link_duplex = FULL_DUPLEX;
1117                         netif_carrier_on(netdev);
1118                         netif_wake_queue(netdev);
1119                 }
1120         } else {
1121                 if (netif_carrier_ok(netdev)) {
1122                         adapter->link_speed = 0;
1123                         adapter->link_duplex = 0;
1124                         DPRINTK(LINK, INFO, "NIC Link is Down\n");
1125                         netif_carrier_off(netdev);
1126                         netif_stop_queue(netdev);
1127
1128                 }
1129         }
1130
1131         ixgb_update_stats(adapter);
1132
1133         if (!netif_carrier_ok(netdev)) {
1134                 if (IXGB_DESC_UNUSED(txdr) + 1 < txdr->count) {
1135                         /* We've lost link, so the controller stops DMA,
1136                          * but we've got queued Tx work that's never going
1137                          * to get done, so reset controller to flush Tx.
1138                          * (Do the reset outside of interrupt context). */
1139                         schedule_work(&adapter->tx_timeout_task);
1140                 }
1141         }
1142
1143         /* Force detection of hung controller every watchdog period */
1144         adapter->detect_tx_hung = true;
1145
1146         /* generate an interrupt to force clean up of any stragglers */
1147         IXGB_WRITE_REG(&adapter->hw, ICS, IXGB_INT_TXDW);
1148
1149         /* Reset the timer */
1150         mod_timer(&adapter->watchdog_timer, jiffies + 2 * HZ);
1151 }
1152
1153 #define IXGB_TX_FLAGS_CSUM              0x00000001
1154 #define IXGB_TX_FLAGS_VLAN              0x00000002
1155 #define IXGB_TX_FLAGS_TSO               0x00000004
1156
1157 static int
1158 ixgb_tso(struct ixgb_adapter *adapter, struct sk_buff *skb)
1159 {
1160         struct ixgb_context_desc *context_desc;
1161         unsigned int i;
1162         u8 ipcss, ipcso, tucss, tucso, hdr_len;
1163         u16 ipcse, tucse, mss;
1164         int err;
1165
1166         if (likely(skb_is_gso(skb))) {
1167                 struct ixgb_buffer *buffer_info;
1168                 struct iphdr *iph;
1169
1170                 if (skb_header_cloned(skb)) {
1171                         err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1172                         if (err)
1173                                 return err;
1174                 }
1175
1176                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
1177                 mss = skb_shinfo(skb)->gso_size;
1178                 iph = ip_hdr(skb);
1179                 iph->tot_len = 0;
1180                 iph->check = 0;
1181                 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
1182                                                          iph->daddr, 0,
1183                                                          IPPROTO_TCP, 0);
1184                 ipcss = skb_network_offset(skb);
1185                 ipcso = (void *)&(iph->check) - (void *)skb->data;
1186                 ipcse = skb_transport_offset(skb) - 1;
1187                 tucss = skb_transport_offset(skb);
1188                 tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
1189                 tucse = 0;
1190
1191                 i = adapter->tx_ring.next_to_use;
1192                 context_desc = IXGB_CONTEXT_DESC(adapter->tx_ring, i);
1193                 buffer_info = &adapter->tx_ring.buffer_info[i];
1194                 WARN_ON(buffer_info->dma != 0);
1195
1196                 context_desc->ipcss = ipcss;
1197                 context_desc->ipcso = ipcso;
1198                 context_desc->ipcse = cpu_to_le16(ipcse);
1199                 context_desc->tucss = tucss;
1200                 context_desc->tucso = tucso;
1201                 context_desc->tucse = cpu_to_le16(tucse);
1202                 context_desc->mss = cpu_to_le16(mss);
1203                 context_desc->hdr_len = hdr_len;
1204                 context_desc->status = 0;
1205                 context_desc->cmd_type_len = cpu_to_le32(
1206                                                   IXGB_CONTEXT_DESC_TYPE
1207                                                 | IXGB_CONTEXT_DESC_CMD_TSE
1208                                                 | IXGB_CONTEXT_DESC_CMD_IP
1209                                                 | IXGB_CONTEXT_DESC_CMD_TCP
1210                                                 | IXGB_CONTEXT_DESC_CMD_IDE
1211                                                 | (skb->len - (hdr_len)));
1212
1213
1214                 if (++i == adapter->tx_ring.count) i = 0;
1215                 adapter->tx_ring.next_to_use = i;
1216
1217                 return 1;
1218         }
1219
1220         return 0;
1221 }
1222
1223 static bool
1224 ixgb_tx_csum(struct ixgb_adapter *adapter, struct sk_buff *skb)
1225 {
1226         struct ixgb_context_desc *context_desc;
1227         unsigned int i;
1228         u8 css, cso;
1229
1230         if (likely(skb->ip_summed == CHECKSUM_PARTIAL)) {
1231                 struct ixgb_buffer *buffer_info;
1232                 css = skb_transport_offset(skb);
1233                 cso = css + skb->csum_offset;
1234
1235                 i = adapter->tx_ring.next_to_use;
1236                 context_desc = IXGB_CONTEXT_DESC(adapter->tx_ring, i);
1237                 buffer_info = &adapter->tx_ring.buffer_info[i];
1238                 WARN_ON(buffer_info->dma != 0);
1239
1240                 context_desc->tucss = css;
1241                 context_desc->tucso = cso;
1242                 context_desc->tucse = 0;
1243                 /* zero out any previously existing data in one instruction */
1244                 *(u32 *)&(context_desc->ipcss) = 0;
1245                 context_desc->status = 0;
1246                 context_desc->hdr_len = 0;
1247                 context_desc->mss = 0;
1248                 context_desc->cmd_type_len =
1249                         cpu_to_le32(IXGB_CONTEXT_DESC_TYPE
1250                                     | IXGB_TX_DESC_CMD_IDE);
1251
1252                 if (++i == adapter->tx_ring.count) i = 0;
1253                 adapter->tx_ring.next_to_use = i;
1254
1255                 return true;
1256         }
1257
1258         return false;
1259 }
1260
1261 #define IXGB_MAX_TXD_PWR        14
1262 #define IXGB_MAX_DATA_PER_TXD   (1<<IXGB_MAX_TXD_PWR)
1263
1264 static int
1265 ixgb_tx_map(struct ixgb_adapter *adapter, struct sk_buff *skb,
1266             unsigned int first)
1267 {
1268         struct ixgb_desc_ring *tx_ring = &adapter->tx_ring;
1269         struct ixgb_buffer *buffer_info;
1270         int len = skb->len;
1271         unsigned int offset = 0, size, count = 0, i;
1272         unsigned int mss = skb_shinfo(skb)->gso_size;
1273
1274         unsigned int nr_frags = skb_shinfo(skb)->nr_frags;
1275         unsigned int f;
1276
1277         len -= skb->data_len;
1278
1279         i = tx_ring->next_to_use;
1280
1281         while (len) {
1282                 buffer_info = &tx_ring->buffer_info[i];
1283                 size = min(len, IXGB_MAX_DATA_PER_TXD);
1284                 /* Workaround for premature desc write-backs
1285                  * in TSO mode.  Append 4-byte sentinel desc */
1286                 if (unlikely(mss && !nr_frags && size == len && size > 8))
1287                         size -= 4;
1288
1289                 buffer_info->length = size;
1290                 WARN_ON(buffer_info->dma != 0);
1291                 buffer_info->time_stamp = jiffies;
1292                 buffer_info->dma =
1293                         pci_map_single(adapter->pdev,
1294                                 skb->data + offset,
1295                                 size,
1296                                 PCI_DMA_TODEVICE);
1297                 buffer_info->next_to_watch = 0;
1298
1299                 len -= size;
1300                 offset += size;
1301                 count++;
1302                 if (++i == tx_ring->count) i = 0;
1303         }
1304
1305         for (f = 0; f < nr_frags; f++) {
1306                 struct skb_frag_struct *frag;
1307
1308                 frag = &skb_shinfo(skb)->frags[f];
1309                 len = frag->size;
1310                 offset = 0;
1311
1312                 while (len) {
1313                         buffer_info = &tx_ring->buffer_info[i];
1314                         size = min(len, IXGB_MAX_DATA_PER_TXD);
1315
1316                         /* Workaround for premature desc write-backs
1317                          * in TSO mode.  Append 4-byte sentinel desc */
1318                         if (unlikely(mss && (f == (nr_frags - 1))
1319                                      && size == len && size > 8))
1320                                 size -= 4;
1321
1322                         buffer_info->length = size;
1323                         buffer_info->time_stamp = jiffies;
1324                         buffer_info->dma =
1325                                 pci_map_page(adapter->pdev,
1326                                         frag->page,
1327                                         frag->page_offset + offset,
1328                                         size,
1329                                         PCI_DMA_TODEVICE);
1330                         buffer_info->next_to_watch = 0;
1331
1332                         len -= size;
1333                         offset += size;
1334                         count++;
1335                         if (++i == tx_ring->count) i = 0;
1336                 }
1337         }
1338         i = (i == 0) ? tx_ring->count - 1 : i - 1;
1339         tx_ring->buffer_info[i].skb = skb;
1340         tx_ring->buffer_info[first].next_to_watch = i;
1341
1342         return count;
1343 }
1344
1345 static void
1346 ixgb_tx_queue(struct ixgb_adapter *adapter, int count, int vlan_id,int tx_flags)
1347 {
1348         struct ixgb_desc_ring *tx_ring = &adapter->tx_ring;
1349         struct ixgb_tx_desc *tx_desc = NULL;
1350         struct ixgb_buffer *buffer_info;
1351         u32 cmd_type_len = adapter->tx_cmd_type;
1352         u8 status = 0;
1353         u8 popts = 0;
1354         unsigned int i;
1355
1356         if (tx_flags & IXGB_TX_FLAGS_TSO) {
1357                 cmd_type_len |= IXGB_TX_DESC_CMD_TSE;
1358                 popts |= (IXGB_TX_DESC_POPTS_IXSM | IXGB_TX_DESC_POPTS_TXSM);
1359         }
1360
1361         if (tx_flags & IXGB_TX_FLAGS_CSUM)
1362                 popts |= IXGB_TX_DESC_POPTS_TXSM;
1363
1364         if (tx_flags & IXGB_TX_FLAGS_VLAN)
1365                 cmd_type_len |= IXGB_TX_DESC_CMD_VLE;
1366
1367         i = tx_ring->next_to_use;
1368
1369         while (count--) {
1370                 buffer_info = &tx_ring->buffer_info[i];
1371                 tx_desc = IXGB_TX_DESC(*tx_ring, i);
1372                 tx_desc->buff_addr = cpu_to_le64(buffer_info->dma);
1373                 tx_desc->cmd_type_len =
1374                         cpu_to_le32(cmd_type_len | buffer_info->length);
1375                 tx_desc->status = status;
1376                 tx_desc->popts = popts;
1377                 tx_desc->vlan = cpu_to_le16(vlan_id);
1378
1379                 if (++i == tx_ring->count) i = 0;
1380         }
1381
1382         tx_desc->cmd_type_len |=
1383                 cpu_to_le32(IXGB_TX_DESC_CMD_EOP | IXGB_TX_DESC_CMD_RS);
1384
1385         /* Force memory writes to complete before letting h/w
1386          * know there are new descriptors to fetch.  (Only
1387          * applicable for weak-ordered memory model archs,
1388          * such as IA-64). */
1389         wmb();
1390
1391         tx_ring->next_to_use = i;
1392         IXGB_WRITE_REG(&adapter->hw, TDT, i);
1393 }
1394
1395 static int __ixgb_maybe_stop_tx(struct net_device *netdev, int size)
1396 {
1397         struct ixgb_adapter *adapter = netdev_priv(netdev);
1398         struct ixgb_desc_ring *tx_ring = &adapter->tx_ring;
1399
1400         netif_stop_queue(netdev);
1401         /* Herbert's original patch had:
1402          *  smp_mb__after_netif_stop_queue();
1403          * but since that doesn't exist yet, just open code it. */
1404         smp_mb();
1405
1406         /* We need to check again in a case another CPU has just
1407          * made room available. */
1408         if (likely(IXGB_DESC_UNUSED(tx_ring) < size))
1409                 return -EBUSY;
1410
1411         /* A reprieve! */
1412         netif_start_queue(netdev);
1413         ++adapter->restart_queue;
1414         return 0;
1415 }
1416
1417 static int ixgb_maybe_stop_tx(struct net_device *netdev,
1418                               struct ixgb_desc_ring *tx_ring, int size)
1419 {
1420         if (likely(IXGB_DESC_UNUSED(tx_ring) >= size))
1421                 return 0;
1422         return __ixgb_maybe_stop_tx(netdev, size);
1423 }
1424
1425
1426 /* Tx Descriptors needed, worst case */
1427 #define TXD_USE_COUNT(S) (((S) >> IXGB_MAX_TXD_PWR) + \
1428                          (((S) & (IXGB_MAX_DATA_PER_TXD - 1)) ? 1 : 0))
1429 #define DESC_NEEDED TXD_USE_COUNT(IXGB_MAX_DATA_PER_TXD) /* skb->date */ + \
1430         MAX_SKB_FRAGS * TXD_USE_COUNT(PAGE_SIZE) + 1 /* for context */ \
1431         + 1 /* one more needed for sentinel TSO workaround */
1432
1433 static int
1434 ixgb_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
1435 {
1436         struct ixgb_adapter *adapter = netdev_priv(netdev);
1437         unsigned int first;
1438         unsigned int tx_flags = 0;
1439         int vlan_id = 0;
1440         int tso;
1441
1442         if (test_bit(__IXGB_DOWN, &adapter->flags)) {
1443                 dev_kfree_skb(skb);
1444                 return NETDEV_TX_OK;
1445         }
1446
1447         if (skb->len <= 0) {
1448                 dev_kfree_skb(skb);
1449                 return 0;
1450         }
1451
1452         if (unlikely(ixgb_maybe_stop_tx(netdev, &adapter->tx_ring,
1453                      DESC_NEEDED)))
1454                 return NETDEV_TX_BUSY;
1455
1456         if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
1457                 tx_flags |= IXGB_TX_FLAGS_VLAN;
1458                 vlan_id = vlan_tx_tag_get(skb);
1459         }
1460
1461         first = adapter->tx_ring.next_to_use;
1462
1463         tso = ixgb_tso(adapter, skb);
1464         if (tso < 0) {
1465                 dev_kfree_skb(skb);
1466                 return NETDEV_TX_OK;
1467         }
1468
1469         if (likely(tso))
1470                 tx_flags |= IXGB_TX_FLAGS_TSO;
1471         else if (ixgb_tx_csum(adapter, skb))
1472                 tx_flags |= IXGB_TX_FLAGS_CSUM;
1473
1474         ixgb_tx_queue(adapter, ixgb_tx_map(adapter, skb, first), vlan_id,
1475                         tx_flags);
1476
1477         netdev->trans_start = jiffies;
1478
1479         /* Make sure there is space in the ring for the next send. */
1480         ixgb_maybe_stop_tx(netdev, &adapter->tx_ring, DESC_NEEDED);
1481
1482         return NETDEV_TX_OK;
1483 }
1484
1485 /**
1486  * ixgb_tx_timeout - Respond to a Tx Hang
1487  * @netdev: network interface device structure
1488  **/
1489
1490 static void
1491 ixgb_tx_timeout(struct net_device *netdev)
1492 {
1493         struct ixgb_adapter *adapter = netdev_priv(netdev);
1494
1495         /* Do the reset outside of interrupt context */
1496         schedule_work(&adapter->tx_timeout_task);
1497 }
1498
1499 static void
1500 ixgb_tx_timeout_task(struct work_struct *work)
1501 {
1502         struct ixgb_adapter *adapter =
1503                 container_of(work, struct ixgb_adapter, tx_timeout_task);
1504
1505         adapter->tx_timeout_count++;
1506         ixgb_down(adapter, true);
1507         ixgb_up(adapter);
1508 }
1509
1510 /**
1511  * ixgb_get_stats - Get System Network Statistics
1512  * @netdev: network interface device structure
1513  *
1514  * Returns the address of the device statistics structure.
1515  * The statistics are actually updated from the timer callback.
1516  **/
1517
1518 static struct net_device_stats *
1519 ixgb_get_stats(struct net_device *netdev)
1520 {
1521         struct ixgb_adapter *adapter = netdev_priv(netdev);
1522
1523         return &adapter->net_stats;
1524 }
1525
1526 /**
1527  * ixgb_change_mtu - Change the Maximum Transfer Unit
1528  * @netdev: network interface device structure
1529  * @new_mtu: new value for maximum frame size
1530  *
1531  * Returns 0 on success, negative on failure
1532  **/
1533
1534 static int
1535 ixgb_change_mtu(struct net_device *netdev, int new_mtu)
1536 {
1537         struct ixgb_adapter *adapter = netdev_priv(netdev);
1538         int max_frame = new_mtu + ENET_HEADER_SIZE + ENET_FCS_LENGTH;
1539         int old_max_frame = netdev->mtu + ENET_HEADER_SIZE + ENET_FCS_LENGTH;
1540
1541         /* MTU < 68 is an error for IPv4 traffic, just don't allow it */
1542         if ((new_mtu < 68) ||
1543             (max_frame > IXGB_MAX_JUMBO_FRAME_SIZE + ENET_FCS_LENGTH)) {
1544                 DPRINTK(PROBE, ERR, "Invalid MTU setting %d\n", new_mtu);
1545                 return -EINVAL;
1546         }
1547
1548         if (old_max_frame == max_frame)
1549                 return 0;
1550
1551         if (netif_running(netdev))
1552                 ixgb_down(adapter, true);
1553
1554         adapter->rx_buffer_len = max_frame + 8; /* + 8 for errata */
1555
1556         netdev->mtu = new_mtu;
1557
1558         if (netif_running(netdev))
1559                 ixgb_up(adapter);
1560
1561         return 0;
1562 }
1563
1564 /**
1565  * ixgb_update_stats - Update the board statistics counters.
1566  * @adapter: board private structure
1567  **/
1568
1569 void
1570 ixgb_update_stats(struct ixgb_adapter *adapter)
1571 {
1572         struct net_device *netdev = adapter->netdev;
1573         struct pci_dev *pdev = adapter->pdev;
1574
1575         /* Prevent stats update while adapter is being reset */
1576         if (pci_channel_offline(pdev))
1577                 return;
1578
1579         if ((netdev->flags & IFF_PROMISC) || (netdev->flags & IFF_ALLMULTI) ||
1580            (netdev->mc_count > IXGB_MAX_NUM_MULTICAST_ADDRESSES)) {
1581                 u64 multi = IXGB_READ_REG(&adapter->hw, MPRCL);
1582                 u32 bcast_l = IXGB_READ_REG(&adapter->hw, BPRCL);
1583                 u32 bcast_h = IXGB_READ_REG(&adapter->hw, BPRCH);
1584                 u64 bcast = ((u64)bcast_h << 32) | bcast_l;
1585
1586                 multi |= ((u64)IXGB_READ_REG(&adapter->hw, MPRCH) << 32);
1587                 /* fix up multicast stats by removing broadcasts */
1588                 if (multi >= bcast)
1589                         multi -= bcast;
1590
1591                 adapter->stats.mprcl += (multi & 0xFFFFFFFF);
1592                 adapter->stats.mprch += (multi >> 32);
1593                 adapter->stats.bprcl += bcast_l;
1594                 adapter->stats.bprch += bcast_h;
1595         } else {
1596                 adapter->stats.mprcl += IXGB_READ_REG(&adapter->hw, MPRCL);
1597                 adapter->stats.mprch += IXGB_READ_REG(&adapter->hw, MPRCH);
1598                 adapter->stats.bprcl += IXGB_READ_REG(&adapter->hw, BPRCL);
1599                 adapter->stats.bprch += IXGB_READ_REG(&adapter->hw, BPRCH);
1600         }
1601         adapter->stats.tprl += IXGB_READ_REG(&adapter->hw, TPRL);
1602         adapter->stats.tprh += IXGB_READ_REG(&adapter->hw, TPRH);
1603         adapter->stats.gprcl += IXGB_READ_REG(&adapter->hw, GPRCL);
1604         adapter->stats.gprch += IXGB_READ_REG(&adapter->hw, GPRCH);
1605         adapter->stats.uprcl += IXGB_READ_REG(&adapter->hw, UPRCL);
1606         adapter->stats.uprch += IXGB_READ_REG(&adapter->hw, UPRCH);
1607         adapter->stats.vprcl += IXGB_READ_REG(&adapter->hw, VPRCL);
1608         adapter->stats.vprch += IXGB_READ_REG(&adapter->hw, VPRCH);
1609         adapter->stats.jprcl += IXGB_READ_REG(&adapter->hw, JPRCL);
1610         adapter->stats.jprch += IXGB_READ_REG(&adapter->hw, JPRCH);
1611         adapter->stats.gorcl += IXGB_READ_REG(&adapter->hw, GORCL);
1612         adapter->stats.gorch += IXGB_READ_REG(&adapter->hw, GORCH);
1613         adapter->stats.torl += IXGB_READ_REG(&adapter->hw, TORL);
1614         adapter->stats.torh += IXGB_READ_REG(&adapter->hw, TORH);
1615         adapter->stats.rnbc += IXGB_READ_REG(&adapter->hw, RNBC);
1616         adapter->stats.ruc += IXGB_READ_REG(&adapter->hw, RUC);
1617         adapter->stats.roc += IXGB_READ_REG(&adapter->hw, ROC);
1618         adapter->stats.rlec += IXGB_READ_REG(&adapter->hw, RLEC);
1619         adapter->stats.crcerrs += IXGB_READ_REG(&adapter->hw, CRCERRS);
1620         adapter->stats.icbc += IXGB_READ_REG(&adapter->hw, ICBC);
1621         adapter->stats.ecbc += IXGB_READ_REG(&adapter->hw, ECBC);
1622         adapter->stats.mpc += IXGB_READ_REG(&adapter->hw, MPC);
1623         adapter->stats.tptl += IXGB_READ_REG(&adapter->hw, TPTL);
1624         adapter->stats.tpth += IXGB_READ_REG(&adapter->hw, TPTH);
1625         adapter->stats.gptcl += IXGB_READ_REG(&adapter->hw, GPTCL);
1626         adapter->stats.gptch += IXGB_READ_REG(&adapter->hw, GPTCH);
1627         adapter->stats.bptcl += IXGB_READ_REG(&adapter->hw, BPTCL);
1628         adapter->stats.bptch += IXGB_READ_REG(&adapter->hw, BPTCH);
1629         adapter->stats.mptcl += IXGB_READ_REG(&adapter->hw, MPTCL);
1630         adapter->stats.mptch += IXGB_READ_REG(&adapter->hw, MPTCH);
1631         adapter->stats.uptcl += IXGB_READ_REG(&adapter->hw, UPTCL);
1632         adapter->stats.uptch += IXGB_READ_REG(&adapter->hw, UPTCH);
1633         adapter->stats.vptcl += IXGB_READ_REG(&adapter->hw, VPTCL);
1634         adapter->stats.vptch += IXGB_READ_REG(&adapter->hw, VPTCH);
1635         adapter->stats.jptcl += IXGB_READ_REG(&adapter->hw, JPTCL);
1636         adapter->stats.jptch += IXGB_READ_REG(&adapter->hw, JPTCH);
1637         adapter->stats.gotcl += IXGB_READ_REG(&adapter->hw, GOTCL);
1638         adapter->stats.gotch += IXGB_READ_REG(&adapter->hw, GOTCH);
1639         adapter->stats.totl += IXGB_READ_REG(&adapter->hw, TOTL);
1640         adapter->stats.toth += IXGB_READ_REG(&adapter->hw, TOTH);
1641         adapter->stats.dc += IXGB_READ_REG(&adapter->hw, DC);
1642         adapter->stats.plt64c += IXGB_READ_REG(&adapter->hw, PLT64C);
1643         adapter->stats.tsctc += IXGB_READ_REG(&adapter->hw, TSCTC);
1644         adapter->stats.tsctfc += IXGB_READ_REG(&adapter->hw, TSCTFC);
1645         adapter->stats.ibic += IXGB_READ_REG(&adapter->hw, IBIC);
1646         adapter->stats.rfc += IXGB_READ_REG(&adapter->hw, RFC);
1647         adapter->stats.lfc += IXGB_READ_REG(&adapter->hw, LFC);
1648         adapter->stats.pfrc += IXGB_READ_REG(&adapter->hw, PFRC);
1649         adapter->stats.pftc += IXGB_READ_REG(&adapter->hw, PFTC);
1650         adapter->stats.mcfrc += IXGB_READ_REG(&adapter->hw, MCFRC);
1651         adapter->stats.mcftc += IXGB_READ_REG(&adapter->hw, MCFTC);
1652         adapter->stats.xonrxc += IXGB_READ_REG(&adapter->hw, XONRXC);
1653         adapter->stats.xontxc += IXGB_READ_REG(&adapter->hw, XONTXC);
1654         adapter->stats.xoffrxc += IXGB_READ_REG(&adapter->hw, XOFFRXC);
1655         adapter->stats.xofftxc += IXGB_READ_REG(&adapter->hw, XOFFTXC);
1656         adapter->stats.rjc += IXGB_READ_REG(&adapter->hw, RJC);
1657
1658         /* Fill out the OS statistics structure */
1659
1660         adapter->net_stats.rx_packets = adapter->stats.gprcl;
1661         adapter->net_stats.tx_packets = adapter->stats.gptcl;
1662         adapter->net_stats.rx_bytes = adapter->stats.gorcl;
1663         adapter->net_stats.tx_bytes = adapter->stats.gotcl;
1664         adapter->net_stats.multicast = adapter->stats.mprcl;
1665         adapter->net_stats.collisions = 0;
1666
1667         /* ignore RLEC as it reports errors for padded (<64bytes) frames
1668          * with a length in the type/len field */
1669         adapter->net_stats.rx_errors =
1670             /* adapter->stats.rnbc + */ adapter->stats.crcerrs +
1671             adapter->stats.ruc +
1672             adapter->stats.roc /*+ adapter->stats.rlec */  +
1673             adapter->stats.icbc +
1674             adapter->stats.ecbc + adapter->stats.mpc;
1675
1676         /* see above
1677          * adapter->net_stats.rx_length_errors = adapter->stats.rlec;
1678          */
1679
1680         adapter->net_stats.rx_crc_errors = adapter->stats.crcerrs;
1681         adapter->net_stats.rx_fifo_errors = adapter->stats.mpc;
1682         adapter->net_stats.rx_missed_errors = adapter->stats.mpc;
1683         adapter->net_stats.rx_over_errors = adapter->stats.mpc;
1684
1685         adapter->net_stats.tx_errors = 0;
1686         adapter->net_stats.rx_frame_errors = 0;
1687         adapter->net_stats.tx_aborted_errors = 0;
1688         adapter->net_stats.tx_carrier_errors = 0;
1689         adapter->net_stats.tx_fifo_errors = 0;
1690         adapter->net_stats.tx_heartbeat_errors = 0;
1691         adapter->net_stats.tx_window_errors = 0;
1692 }
1693
1694 #define IXGB_MAX_INTR 10
1695 /**
1696  * ixgb_intr - Interrupt Handler
1697  * @irq: interrupt number
1698  * @data: pointer to a network interface device structure
1699  **/
1700
1701 static irqreturn_t
1702 ixgb_intr(int irq, void *data)
1703 {
1704         struct net_device *netdev = data;
1705         struct ixgb_adapter *adapter = netdev_priv(netdev);
1706         struct ixgb_hw *hw = &adapter->hw;
1707         u32 icr = IXGB_READ_REG(hw, ICR);
1708
1709         if (unlikely(!icr))
1710                 return IRQ_NONE;  /* Not our interrupt */
1711
1712         if (unlikely(icr & (IXGB_INT_RXSEQ | IXGB_INT_LSC)))
1713                 if (!test_bit(__IXGB_DOWN, &adapter->flags))
1714                         mod_timer(&adapter->watchdog_timer, jiffies);
1715
1716         if (netif_rx_schedule_prep(netdev, &adapter->napi)) {
1717
1718                 /* Disable interrupts and register for poll. The flush
1719                   of the posted write is intentionally left out.
1720                 */
1721
1722                 IXGB_WRITE_REG(&adapter->hw, IMC, ~0);
1723                 __netif_rx_schedule(netdev, &adapter->napi);
1724         }
1725         return IRQ_HANDLED;
1726 }
1727
1728 /**
1729  * ixgb_clean - NAPI Rx polling callback
1730  * @adapter: board private structure
1731  **/
1732
1733 static int
1734 ixgb_clean(struct napi_struct *napi, int budget)
1735 {
1736         struct ixgb_adapter *adapter = container_of(napi, struct ixgb_adapter, napi);
1737         struct net_device *netdev = adapter->netdev;
1738         int work_done = 0;
1739
1740         ixgb_clean_tx_irq(adapter);
1741         ixgb_clean_rx_irq(adapter, &work_done, budget);
1742
1743         /* If budget not fully consumed, exit the polling mode */
1744         if (work_done < budget) {
1745                 netif_rx_complete(netdev, napi);
1746                 if (!test_bit(__IXGB_DOWN, &adapter->flags))
1747                         ixgb_irq_enable(adapter);
1748         }
1749
1750         return work_done;
1751 }
1752
1753 /**
1754  * ixgb_clean_tx_irq - Reclaim resources after transmit completes
1755  * @adapter: board private structure
1756  **/
1757
1758 static bool
1759 ixgb_clean_tx_irq(struct ixgb_adapter *adapter)
1760 {
1761         struct ixgb_desc_ring *tx_ring = &adapter->tx_ring;
1762         struct net_device *netdev = adapter->netdev;
1763         struct ixgb_tx_desc *tx_desc, *eop_desc;
1764         struct ixgb_buffer *buffer_info;
1765         unsigned int i, eop;
1766         bool cleaned = false;
1767
1768         i = tx_ring->next_to_clean;
1769         eop = tx_ring->buffer_info[i].next_to_watch;
1770         eop_desc = IXGB_TX_DESC(*tx_ring, eop);
1771
1772         while (eop_desc->status & IXGB_TX_DESC_STATUS_DD) {
1773
1774                 for (cleaned = false; !cleaned; ) {
1775                         tx_desc = IXGB_TX_DESC(*tx_ring, i);
1776                         buffer_info = &tx_ring->buffer_info[i];
1777
1778                         if (tx_desc->popts &
1779                            (IXGB_TX_DESC_POPTS_TXSM |
1780                             IXGB_TX_DESC_POPTS_IXSM))
1781                                 adapter->hw_csum_tx_good++;
1782
1783                         ixgb_unmap_and_free_tx_resource(adapter, buffer_info);
1784
1785                         *(u32 *)&(tx_desc->status) = 0;
1786
1787                         cleaned = (i == eop);
1788                         if (++i == tx_ring->count) i = 0;
1789                 }
1790
1791                 eop = tx_ring->buffer_info[i].next_to_watch;
1792                 eop_desc = IXGB_TX_DESC(*tx_ring, eop);
1793         }
1794
1795         tx_ring->next_to_clean = i;
1796
1797         if (unlikely(cleaned && netif_carrier_ok(netdev) &&
1798                      IXGB_DESC_UNUSED(tx_ring) >= DESC_NEEDED)) {
1799                 /* Make sure that anybody stopping the queue after this
1800                  * sees the new next_to_clean. */
1801                 smp_mb();
1802
1803                 if (netif_queue_stopped(netdev) &&
1804                     !(test_bit(__IXGB_DOWN, &adapter->flags))) {
1805                         netif_wake_queue(netdev);
1806                         ++adapter->restart_queue;
1807                 }
1808         }
1809
1810         if (adapter->detect_tx_hung) {
1811                 /* detect a transmit hang in hardware, this serializes the
1812                  * check with the clearing of time_stamp and movement of i */
1813                 adapter->detect_tx_hung = false;
1814                 if (tx_ring->buffer_info[eop].dma &&
1815                    time_after(jiffies, tx_ring->buffer_info[eop].time_stamp + HZ)
1816                    && !(IXGB_READ_REG(&adapter->hw, STATUS) &
1817                         IXGB_STATUS_TXOFF)) {
1818                         /* detected Tx unit hang */
1819                         DPRINTK(DRV, ERR, "Detected Tx Unit Hang\n"
1820                                         "  TDH                  <%x>\n"
1821                                         "  TDT                  <%x>\n"
1822                                         "  next_to_use          <%x>\n"
1823                                         "  next_to_clean        <%x>\n"
1824                                         "buffer_info[next_to_clean]\n"
1825                                         "  time_stamp           <%lx>\n"
1826                                         "  next_to_watch        <%x>\n"
1827                                         "  jiffies              <%lx>\n"
1828                                         "  next_to_watch.status <%x>\n",
1829                                 IXGB_READ_REG(&adapter->hw, TDH),
1830                                 IXGB_READ_REG(&adapter->hw, TDT),
1831                                 tx_ring->next_to_use,
1832                                 tx_ring->next_to_clean,
1833                                 tx_ring->buffer_info[eop].time_stamp,
1834                                 eop,
1835                                 jiffies,
1836                                 eop_desc->status);
1837                         netif_stop_queue(netdev);
1838                 }
1839         }
1840
1841         return cleaned;
1842 }
1843
1844 /**
1845  * ixgb_rx_checksum - Receive Checksum Offload for 82597.
1846  * @adapter: board private structure
1847  * @rx_desc: receive descriptor
1848  * @sk_buff: socket buffer with received data
1849  **/
1850
1851 static void
1852 ixgb_rx_checksum(struct ixgb_adapter *adapter,
1853                  struct ixgb_rx_desc *rx_desc,
1854                  struct sk_buff *skb)
1855 {
1856         /* Ignore Checksum bit is set OR
1857          * TCP Checksum has not been calculated
1858          */
1859         if ((rx_desc->status & IXGB_RX_DESC_STATUS_IXSM) ||
1860            (!(rx_desc->status & IXGB_RX_DESC_STATUS_TCPCS))) {
1861                 skb->ip_summed = CHECKSUM_NONE;
1862                 return;
1863         }
1864
1865         /* At this point we know the hardware did the TCP checksum */
1866         /* now look at the TCP checksum error bit */
1867         if (rx_desc->errors & IXGB_RX_DESC_ERRORS_TCPE) {
1868                 /* let the stack verify checksum errors */
1869                 skb->ip_summed = CHECKSUM_NONE;
1870                 adapter->hw_csum_rx_error++;
1871         } else {
1872                 /* TCP checksum is good */
1873                 skb->ip_summed = CHECKSUM_UNNECESSARY;
1874                 adapter->hw_csum_rx_good++;
1875         }
1876 }
1877
1878 /**
1879  * ixgb_clean_rx_irq - Send received data up the network stack,
1880  * @adapter: board private structure
1881  **/
1882
1883 static bool
1884 ixgb_clean_rx_irq(struct ixgb_adapter *adapter, int *work_done, int work_to_do)
1885 {
1886         struct ixgb_desc_ring *rx_ring = &adapter->rx_ring;
1887         struct net_device *netdev = adapter->netdev;
1888         struct pci_dev *pdev = adapter->pdev;
1889         struct ixgb_rx_desc *rx_desc, *next_rxd;
1890         struct ixgb_buffer *buffer_info, *next_buffer, *next2_buffer;
1891         u32 length;
1892         unsigned int i, j;
1893         int cleaned_count = 0;
1894         bool cleaned = false;
1895
1896         i = rx_ring->next_to_clean;
1897         rx_desc = IXGB_RX_DESC(*rx_ring, i);
1898         buffer_info = &rx_ring->buffer_info[i];
1899
1900         while (rx_desc->status & IXGB_RX_DESC_STATUS_DD) {
1901                 struct sk_buff *skb;
1902                 u8 status;
1903
1904                 if (*work_done >= work_to_do)
1905                         break;
1906
1907                 (*work_done)++;
1908                 status = rx_desc->status;
1909                 skb = buffer_info->skb;
1910                 buffer_info->skb = NULL;
1911
1912                 prefetch(skb->data - NET_IP_ALIGN);
1913
1914                 if (++i == rx_ring->count) i = 0;
1915                 next_rxd = IXGB_RX_DESC(*rx_ring, i);
1916                 prefetch(next_rxd);
1917
1918                 if ((j = i + 1) == rx_ring->count) j = 0;
1919                 next2_buffer = &rx_ring->buffer_info[j];
1920                 prefetch(next2_buffer);
1921
1922                 next_buffer = &rx_ring->buffer_info[i];
1923
1924                 cleaned = true;
1925                 cleaned_count++;
1926
1927                 pci_unmap_single(pdev,
1928                                  buffer_info->dma,
1929                                  buffer_info->length,
1930                                  PCI_DMA_FROMDEVICE);
1931                 buffer_info->dma = 0;
1932
1933                 length = le16_to_cpu(rx_desc->length);
1934                 rx_desc->length = 0;
1935
1936                 if (unlikely(!(status & IXGB_RX_DESC_STATUS_EOP))) {
1937
1938                         /* All receives must fit into a single buffer */
1939
1940                         IXGB_DBG("Receive packet consumed multiple buffers "
1941                                          "length<%x>\n", length);
1942
1943                         dev_kfree_skb_irq(skb);
1944                         goto rxdesc_done;
1945                 }
1946
1947                 if (unlikely(rx_desc->errors &
1948                     (IXGB_RX_DESC_ERRORS_CE | IXGB_RX_DESC_ERRORS_SE |
1949                      IXGB_RX_DESC_ERRORS_P | IXGB_RX_DESC_ERRORS_RXE))) {
1950                         dev_kfree_skb_irq(skb);
1951                         goto rxdesc_done;
1952                 }
1953
1954                 /* code added for copybreak, this should improve
1955                  * performance for small packets with large amounts
1956                  * of reassembly being done in the stack */
1957                 if (length < copybreak) {
1958                         struct sk_buff *new_skb =
1959                             netdev_alloc_skb(netdev, length + NET_IP_ALIGN);
1960                         if (new_skb) {
1961                                 skb_reserve(new_skb, NET_IP_ALIGN);
1962                                 skb_copy_to_linear_data_offset(new_skb,
1963                                                                -NET_IP_ALIGN,
1964                                                                (skb->data -
1965                                                                 NET_IP_ALIGN),
1966                                                                (length +
1967                                                                 NET_IP_ALIGN));
1968                                 /* save the skb in buffer_info as good */
1969                                 buffer_info->skb = skb;
1970                                 skb = new_skb;
1971                         }
1972                 }
1973                 /* end copybreak code */
1974
1975                 /* Good Receive */
1976                 skb_put(skb, length);
1977
1978                 /* Receive Checksum Offload */
1979                 ixgb_rx_checksum(adapter, rx_desc, skb);
1980
1981                 skb->protocol = eth_type_trans(skb, netdev);
1982                 if (adapter->vlgrp && (status & IXGB_RX_DESC_STATUS_VP)) {
1983                         vlan_hwaccel_receive_skb(skb, adapter->vlgrp,
1984                                                 le16_to_cpu(rx_desc->special));
1985                 } else {
1986                         netif_receive_skb(skb);
1987                 }
1988
1989 rxdesc_done:
1990                 /* clean up descriptor, might be written over by hw */
1991                 rx_desc->status = 0;
1992
1993                 /* return some buffers to hardware, one at a time is too slow */
1994                 if (unlikely(cleaned_count >= IXGB_RX_BUFFER_WRITE)) {
1995                         ixgb_alloc_rx_buffers(adapter, cleaned_count);
1996                         cleaned_count = 0;
1997                 }
1998
1999                 /* use prefetched values */
2000                 rx_desc = next_rxd;
2001                 buffer_info = next_buffer;
2002         }
2003
2004         rx_ring->next_to_clean = i;
2005
2006         cleaned_count = IXGB_DESC_UNUSED(rx_ring);
2007         if (cleaned_count)
2008                 ixgb_alloc_rx_buffers(adapter, cleaned_count);
2009
2010         return cleaned;
2011 }
2012
2013 /**
2014  * ixgb_alloc_rx_buffers - Replace used receive buffers
2015  * @adapter: address of board private structure
2016  **/
2017
2018 static void
2019 ixgb_alloc_rx_buffers(struct ixgb_adapter *adapter, int cleaned_count)
2020 {
2021         struct ixgb_desc_ring *rx_ring = &adapter->rx_ring;
2022         struct net_device *netdev = adapter->netdev;
2023         struct pci_dev *pdev = adapter->pdev;
2024         struct ixgb_rx_desc *rx_desc;
2025         struct ixgb_buffer *buffer_info;
2026         struct sk_buff *skb;
2027         unsigned int i;
2028         long cleancount;
2029
2030         i = rx_ring->next_to_use;
2031         buffer_info = &rx_ring->buffer_info[i];
2032         cleancount = IXGB_DESC_UNUSED(rx_ring);
2033
2034
2035         /* leave three descriptors unused */
2036         while (--cleancount > 2 && cleaned_count--) {
2037                 /* recycle! its good for you */
2038                 skb = buffer_info->skb;
2039                 if (skb) {
2040                         skb_trim(skb, 0);
2041                         goto map_skb;
2042                 }
2043
2044                 skb = netdev_alloc_skb(netdev, adapter->rx_buffer_len
2045                                        + NET_IP_ALIGN);
2046                 if (unlikely(!skb)) {
2047                         /* Better luck next round */
2048                         adapter->alloc_rx_buff_failed++;
2049                         break;
2050                 }
2051
2052                 /* Make buffer alignment 2 beyond a 16 byte boundary
2053                  * this will result in a 16 byte aligned IP header after
2054                  * the 14 byte MAC header is removed
2055                  */
2056                 skb_reserve(skb, NET_IP_ALIGN);
2057
2058                 buffer_info->skb = skb;
2059                 buffer_info->length = adapter->rx_buffer_len;
2060 map_skb:
2061                 buffer_info->dma = pci_map_single(pdev,
2062                                                   skb->data,
2063                                                   adapter->rx_buffer_len,
2064                                                   PCI_DMA_FROMDEVICE);
2065
2066                 rx_desc = IXGB_RX_DESC(*rx_ring, i);
2067                 rx_desc->buff_addr = cpu_to_le64(buffer_info->dma);
2068                 /* guarantee DD bit not set now before h/w gets descriptor
2069                  * this is the rest of the workaround for h/w double
2070                  * writeback. */
2071                 rx_desc->status = 0;
2072
2073
2074                 if (++i == rx_ring->count) i = 0;
2075                 buffer_info = &rx_ring->buffer_info[i];
2076         }
2077
2078         if (likely(rx_ring->next_to_use != i)) {
2079                 rx_ring->next_to_use = i;
2080                 if (unlikely(i-- == 0))
2081                         i = (rx_ring->count - 1);
2082
2083                 /* Force memory writes to complete before letting h/w
2084                  * know there are new descriptors to fetch.  (Only
2085                  * applicable for weak-ordered memory model archs, such
2086                  * as IA-64). */
2087                 wmb();
2088                 IXGB_WRITE_REG(&adapter->hw, RDT, i);
2089         }
2090 }
2091
2092 /**
2093  * ixgb_vlan_rx_register - enables or disables vlan tagging/stripping.
2094  *
2095  * @param netdev network interface device structure
2096  * @param grp indicates to enable or disable tagging/stripping
2097  **/
2098 static void
2099 ixgb_vlan_rx_register(struct net_device *netdev, struct vlan_group *grp)
2100 {
2101         struct ixgb_adapter *adapter = netdev_priv(netdev);
2102         u32 ctrl, rctl;
2103
2104         ixgb_irq_disable(adapter);
2105         adapter->vlgrp = grp;
2106
2107         if (grp) {
2108                 /* enable VLAN tag insert/strip */
2109                 ctrl = IXGB_READ_REG(&adapter->hw, CTRL0);
2110                 ctrl |= IXGB_CTRL0_VME;
2111                 IXGB_WRITE_REG(&adapter->hw, CTRL0, ctrl);
2112
2113                 /* enable VLAN receive filtering */
2114
2115                 rctl = IXGB_READ_REG(&adapter->hw, RCTL);
2116                 rctl &= ~IXGB_RCTL_CFIEN;
2117                 IXGB_WRITE_REG(&adapter->hw, RCTL, rctl);
2118         } else {
2119                 /* disable VLAN tag insert/strip */
2120
2121                 ctrl = IXGB_READ_REG(&adapter->hw, CTRL0);
2122                 ctrl &= ~IXGB_CTRL0_VME;
2123                 IXGB_WRITE_REG(&adapter->hw, CTRL0, ctrl);
2124         }
2125
2126         /* don't enable interrupts unless we are UP */
2127         if (adapter->netdev->flags & IFF_UP)
2128                 ixgb_irq_enable(adapter);
2129 }
2130
2131 static void
2132 ixgb_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
2133 {
2134         struct ixgb_adapter *adapter = netdev_priv(netdev);
2135         u32 vfta, index;
2136
2137         /* add VID to filter table */
2138
2139         index = (vid >> 5) & 0x7F;
2140         vfta = IXGB_READ_REG_ARRAY(&adapter->hw, VFTA, index);
2141         vfta |= (1 << (vid & 0x1F));
2142         ixgb_write_vfta(&adapter->hw, index, vfta);
2143 }
2144
2145 static void
2146 ixgb_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
2147 {
2148         struct ixgb_adapter *adapter = netdev_priv(netdev);
2149         u32 vfta, index;
2150
2151         ixgb_irq_disable(adapter);
2152
2153         vlan_group_set_device(adapter->vlgrp, vid, NULL);
2154
2155         /* don't enable interrupts unless we are UP */
2156         if (adapter->netdev->flags & IFF_UP)
2157                 ixgb_irq_enable(adapter);
2158
2159         /* remove VID from filter table */
2160
2161         index = (vid >> 5) & 0x7F;
2162         vfta = IXGB_READ_REG_ARRAY(&adapter->hw, VFTA, index);
2163         vfta &= ~(1 << (vid & 0x1F));
2164         ixgb_write_vfta(&adapter->hw, index, vfta);
2165 }
2166
2167 static void
2168 ixgb_restore_vlan(struct ixgb_adapter *adapter)
2169 {
2170         ixgb_vlan_rx_register(adapter->netdev, adapter->vlgrp);
2171
2172         if (adapter->vlgrp) {
2173                 u16 vid;
2174                 for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
2175                         if (!vlan_group_get_device(adapter->vlgrp, vid))
2176                                 continue;
2177                         ixgb_vlan_rx_add_vid(adapter->netdev, vid);
2178                 }
2179         }
2180 }
2181
2182 #ifdef CONFIG_NET_POLL_CONTROLLER
2183 /*
2184  * Polling 'interrupt' - used by things like netconsole to send skbs
2185  * without having to re-enable interrupts. It's not called while
2186  * the interrupt routine is executing.
2187  */
2188
2189 static void ixgb_netpoll(struct net_device *dev)
2190 {
2191         struct ixgb_adapter *adapter = netdev_priv(dev);
2192
2193         disable_irq(adapter->pdev->irq);
2194         ixgb_intr(adapter->pdev->irq, dev);
2195         enable_irq(adapter->pdev->irq);
2196 }
2197 #endif
2198
2199 /**
2200  * ixgb_io_error_detected() - called when PCI error is detected
2201  * @pdev    pointer to pci device with error
2202  * @state   pci channel state after error
2203  *
2204  * This callback is called by the PCI subsystem whenever
2205  * a PCI bus error is detected.
2206  */
2207 static pci_ers_result_t ixgb_io_error_detected(struct pci_dev *pdev,
2208                                                enum pci_channel_state state)
2209 {
2210         struct net_device *netdev = pci_get_drvdata(pdev);
2211         struct ixgb_adapter *adapter = netdev_priv(netdev);
2212
2213         if (netif_running(netdev))
2214                 ixgb_down(adapter, true);
2215
2216         pci_disable_device(pdev);
2217
2218         /* Request a slot reset. */
2219         return PCI_ERS_RESULT_NEED_RESET;
2220 }
2221
2222 /**
2223  * ixgb_io_slot_reset - called after the pci bus has been reset.
2224  * @pdev    pointer to pci device with error
2225  *
2226  * This callback is called after the PCI bus has been reset.
2227  * Basically, this tries to restart the card from scratch.
2228  * This is a shortened version of the device probe/discovery code,
2229  * it resembles the first-half of the ixgb_probe() routine.
2230  */
2231 static pci_ers_result_t ixgb_io_slot_reset(struct pci_dev *pdev)
2232 {
2233         struct net_device *netdev = pci_get_drvdata(pdev);
2234         struct ixgb_adapter *adapter = netdev_priv(netdev);
2235
2236         if (pci_enable_device(pdev)) {
2237                 DPRINTK(PROBE, ERR, "Cannot re-enable PCI device after reset.\n");
2238                 return PCI_ERS_RESULT_DISCONNECT;
2239         }
2240
2241         /* Perform card reset only on one instance of the card */
2242         if (0 != PCI_FUNC (pdev->devfn))
2243                 return PCI_ERS_RESULT_RECOVERED;
2244
2245         pci_set_master(pdev);
2246
2247         netif_carrier_off(netdev);
2248         netif_stop_queue(netdev);
2249         ixgb_reset(adapter);
2250
2251         /* Make sure the EEPROM is good */
2252         if (!ixgb_validate_eeprom_checksum(&adapter->hw)) {
2253                 DPRINTK(PROBE, ERR, "After reset, the EEPROM checksum is not valid.\n");
2254                 return PCI_ERS_RESULT_DISCONNECT;
2255         }
2256         ixgb_get_ee_mac_addr(&adapter->hw, netdev->dev_addr);
2257         memcpy(netdev->perm_addr, netdev->dev_addr, netdev->addr_len);
2258
2259         if (!is_valid_ether_addr(netdev->perm_addr)) {
2260                 DPRINTK(PROBE, ERR, "After reset, invalid MAC address.\n");
2261                 return PCI_ERS_RESULT_DISCONNECT;
2262         }
2263
2264         return PCI_ERS_RESULT_RECOVERED;
2265 }
2266
2267 /**
2268  * ixgb_io_resume - called when its OK to resume normal operations
2269  * @pdev    pointer to pci device with error
2270  *
2271  * The error recovery driver tells us that its OK to resume
2272  * normal operation. Implementation resembles the second-half
2273  * of the ixgb_probe() routine.
2274  */
2275 static void ixgb_io_resume(struct pci_dev *pdev)
2276 {
2277         struct net_device *netdev = pci_get_drvdata(pdev);
2278         struct ixgb_adapter *adapter = netdev_priv(netdev);
2279
2280         pci_set_master(pdev);
2281
2282         if (netif_running(netdev)) {
2283                 if (ixgb_up(adapter)) {
2284                         printk ("ixgb: can't bring device back up after reset\n");
2285                         return;
2286                 }
2287         }
2288
2289         netif_device_attach(netdev);
2290         mod_timer(&adapter->watchdog_timer, jiffies);
2291 }
2292
2293 /* ixgb_main.c */