]> err.no Git - linux-2.6/blob - drivers/net/ixgb/ixgb_main.c
3e857c022b6626b6d5c07d584fbcfa830ea5f515
[linux-2.6] / drivers / net / ixgb / ixgb_main.c
1 /*******************************************************************************
2
3   Intel PRO/10GbE Linux driver
4   Copyright(c) 1999 - 2006 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 #ifndef CONFIG_IXGB_NAPI
35 #define DRIVERNAPI
36 #else
37 #define DRIVERNAPI "-NAPI"
38 #endif
39 #define DRV_VERSION             "1.0.126-k4"DRIVERNAPI
40 const char ixgb_driver_version[] = DRV_VERSION;
41 static const char ixgb_copyright[] = "Copyright (c) 1999-2006 Intel Corporation.";
42
43 #define IXGB_CB_LENGTH 256
44 static unsigned int copybreak __read_mostly = IXGB_CB_LENGTH;
45 module_param(copybreak, uint, 0644);
46 MODULE_PARM_DESC(copybreak,
47         "Maximum size of packet that is copied to a new buffer on receive");
48
49 /* ixgb_pci_tbl - PCI Device ID Table
50  *
51  * Wildcard entries (PCI_ANY_ID) should come last
52  * Last entry must be all 0s
53  *
54  * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
55  *   Class, Class Mask, private data (not used) }
56  */
57 static struct pci_device_id ixgb_pci_tbl[] = {
58         {INTEL_VENDOR_ID, IXGB_DEVICE_ID_82597EX,
59          PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0},
60         {INTEL_VENDOR_ID, IXGB_DEVICE_ID_82597EX_CX4,
61          PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0},
62         {INTEL_VENDOR_ID, IXGB_DEVICE_ID_82597EX_SR,
63          PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0},
64         {INTEL_VENDOR_ID, IXGB_DEVICE_ID_82597EX_LR,
65          PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0},
66
67         /* required last entry */
68         {0,}
69 };
70
71 MODULE_DEVICE_TABLE(pci, ixgb_pci_tbl);
72
73 /* Local Function Prototypes */
74 static int ixgb_init_module(void);
75 static void ixgb_exit_module(void);
76 static int ixgb_probe(struct pci_dev *pdev, const struct pci_device_id *ent);
77 static void __devexit ixgb_remove(struct pci_dev *pdev);
78 static int ixgb_sw_init(struct ixgb_adapter *adapter);
79 static int ixgb_open(struct net_device *netdev);
80 static int ixgb_close(struct net_device *netdev);
81 static void ixgb_configure_tx(struct ixgb_adapter *adapter);
82 static void ixgb_configure_rx(struct ixgb_adapter *adapter);
83 static void ixgb_setup_rctl(struct ixgb_adapter *adapter);
84 static void ixgb_clean_tx_ring(struct ixgb_adapter *adapter);
85 static void ixgb_clean_rx_ring(struct ixgb_adapter *adapter);
86 static void ixgb_set_multi(struct net_device *netdev);
87 static void ixgb_watchdog(unsigned long data);
88 static int ixgb_xmit_frame(struct sk_buff *skb, struct net_device *netdev);
89 static struct net_device_stats *ixgb_get_stats(struct net_device *netdev);
90 static int ixgb_change_mtu(struct net_device *netdev, int new_mtu);
91 static int ixgb_set_mac(struct net_device *netdev, void *p);
92 static irqreturn_t ixgb_intr(int irq, void *data);
93 static bool ixgb_clean_tx_irq(struct ixgb_adapter *adapter);
94
95 #ifdef CONFIG_IXGB_NAPI
96 static int ixgb_clean(struct napi_struct *, int);
97 static bool ixgb_clean_rx_irq(struct ixgb_adapter *, int *, int);
98 #else
99 static bool ixgb_clean_rx_irq(struct ixgb_adapter *);
100 #endif
101 static void ixgb_alloc_rx_buffers(struct ixgb_adapter *, int);
102
103 static void ixgb_tx_timeout(struct net_device *dev);
104 static void ixgb_tx_timeout_task(struct work_struct *work);
105
106 static void ixgb_vlan_rx_register(struct net_device *netdev,
107                                   struct vlan_group *grp);
108 static void ixgb_vlan_rx_add_vid(struct net_device *netdev, u16 vid);
109 static void ixgb_vlan_rx_kill_vid(struct net_device *netdev, u16 vid);
110 static void ixgb_restore_vlan(struct ixgb_adapter *adapter);
111
112 #ifdef CONFIG_NET_POLL_CONTROLLER
113 /* for netdump / net console */
114 static void ixgb_netpoll(struct net_device *dev);
115 #endif
116
117 static pci_ers_result_t ixgb_io_error_detected (struct pci_dev *pdev,
118                              enum pci_channel_state state);
119 static pci_ers_result_t ixgb_io_slot_reset (struct pci_dev *pdev);
120 static void ixgb_io_resume (struct pci_dev *pdev);
121
122 static struct pci_error_handlers ixgb_err_handler = {
123         .error_detected = ixgb_io_error_detected,
124         .slot_reset = ixgb_io_slot_reset,
125         .resume = ixgb_io_resume,
126 };
127
128 static struct pci_driver ixgb_driver = {
129         .name     = ixgb_driver_name,
130         .id_table = ixgb_pci_tbl,
131         .probe    = ixgb_probe,
132         .remove   = __devexit_p(ixgb_remove),
133         .err_handler = &ixgb_err_handler
134 };
135
136 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
137 MODULE_DESCRIPTION("Intel(R) PRO/10GbE Network Driver");
138 MODULE_LICENSE("GPL");
139 MODULE_VERSION(DRV_VERSION);
140
141 #define DEFAULT_DEBUG_LEVEL_SHIFT 3
142 static int debug = DEFAULT_DEBUG_LEVEL_SHIFT;
143 module_param(debug, int, 0);
144 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
145
146 /**
147  * ixgb_init_module - Driver Registration Routine
148  *
149  * ixgb_init_module is the first routine called when the driver is
150  * loaded. All it does is register with the PCI subsystem.
151  **/
152
153 static int __init
154 ixgb_init_module(void)
155 {
156         printk(KERN_INFO "%s - version %s\n",
157                ixgb_driver_string, ixgb_driver_version);
158
159         printk(KERN_INFO "%s\n", ixgb_copyright);
160
161         return pci_register_driver(&ixgb_driver);
162 }
163
164 module_init(ixgb_init_module);
165
166 /**
167  * ixgb_exit_module - Driver Exit Cleanup Routine
168  *
169  * ixgb_exit_module is called just before the driver is removed
170  * from memory.
171  **/
172
173 static void __exit
174 ixgb_exit_module(void)
175 {
176         pci_unregister_driver(&ixgb_driver);
177 }
178
179 module_exit(ixgb_exit_module);
180
181 /**
182  * ixgb_irq_disable - Mask off interrupt generation on the NIC
183  * @adapter: board private structure
184  **/
185
186 static void
187 ixgb_irq_disable(struct ixgb_adapter *adapter)
188 {
189         IXGB_WRITE_REG(&adapter->hw, IMC, ~0);
190         IXGB_WRITE_FLUSH(&adapter->hw);
191         synchronize_irq(adapter->pdev->irq);
192 }
193
194 /**
195  * ixgb_irq_enable - Enable default interrupt generation settings
196  * @adapter: board private structure
197  **/
198
199 static void
200 ixgb_irq_enable(struct ixgb_adapter *adapter)
201 {
202         u32 val = IXGB_INT_RXT0 | IXGB_INT_RXDMT0 |
203                   IXGB_INT_TXDW | IXGB_INT_LSC;
204         if (adapter->hw.subsystem_vendor_id == SUN_SUBVENDOR_ID)
205                 val |= IXGB_INT_GPI0;
206         IXGB_WRITE_REG(&adapter->hw, IMS, val);
207         IXGB_WRITE_FLUSH(&adapter->hw);
208 }
209
210 int
211 ixgb_up(struct ixgb_adapter *adapter)
212 {
213         struct net_device *netdev = adapter->netdev;
214         int err, irq_flags = IRQF_SHARED;
215         int max_frame = netdev->mtu + ENET_HEADER_SIZE + ENET_FCS_LENGTH;
216         struct ixgb_hw *hw = &adapter->hw;
217
218         /* hardware has been reset, we need to reload some things */
219
220         ixgb_rar_set(hw, netdev->dev_addr, 0);
221         ixgb_set_multi(netdev);
222
223         ixgb_restore_vlan(adapter);
224
225         ixgb_configure_tx(adapter);
226         ixgb_setup_rctl(adapter);
227         ixgb_configure_rx(adapter);
228         ixgb_alloc_rx_buffers(adapter, IXGB_DESC_UNUSED(&adapter->rx_ring));
229
230         /* disable interrupts and get the hardware into a known state */
231         IXGB_WRITE_REG(&adapter->hw, IMC, 0xffffffff);
232
233         /* only enable MSI if bus is in PCI-X mode */
234         if (IXGB_READ_REG(&adapter->hw, STATUS) & IXGB_STATUS_PCIX_MODE) {
235                 err = pci_enable_msi(adapter->pdev);
236                 if (!err) {
237                         adapter->have_msi = 1;
238                         irq_flags = 0;
239                 }
240                 /* proceed to try to request regular interrupt */
241         }
242
243         err = request_irq(adapter->pdev->irq, &ixgb_intr, irq_flags,
244                           netdev->name, netdev);
245         if (err) {
246                 if (adapter->have_msi)
247                         pci_disable_msi(adapter->pdev);
248                 DPRINTK(PROBE, ERR,
249                  "Unable to allocate interrupt Error: %d\n", err);
250                 return err;
251         }
252
253         if ((hw->max_frame_size != max_frame) ||
254                 (hw->max_frame_size !=
255                 (IXGB_READ_REG(hw, MFS) >> IXGB_MFS_SHIFT))) {
256
257                 hw->max_frame_size = max_frame;
258
259                 IXGB_WRITE_REG(hw, MFS, hw->max_frame_size << IXGB_MFS_SHIFT);
260
261                 if (hw->max_frame_size >
262                    IXGB_MAX_ENET_FRAME_SIZE_WITHOUT_FCS + ENET_FCS_LENGTH) {
263                         u32 ctrl0 = IXGB_READ_REG(hw, CTRL0);
264
265                         if (!(ctrl0 & IXGB_CTRL0_JFE)) {
266                                 ctrl0 |= IXGB_CTRL0_JFE;
267                                 IXGB_WRITE_REG(hw, CTRL0, ctrl0);
268                         }
269                 }
270         }
271
272         clear_bit(__IXGB_DOWN, &adapter->flags);
273
274 #ifdef CONFIG_IXGB_NAPI
275         napi_enable(&adapter->napi);
276 #endif
277         ixgb_irq_enable(adapter);
278
279         mod_timer(&adapter->watchdog_timer, jiffies);
280
281         return 0;
282 }
283
284 void
285 ixgb_down(struct ixgb_adapter *adapter, bool kill_watchdog)
286 {
287         struct net_device *netdev = adapter->netdev;
288
289         /* prevent the interrupt handler from restarting watchdog */
290         set_bit(__IXGB_DOWN, &adapter->flags);
291
292 #ifdef CONFIG_IXGB_NAPI
293         napi_disable(&adapter->napi);
294 #endif
295         /* waiting for NAPI to complete can re-enable interrupts */
296         ixgb_irq_disable(adapter);
297         free_irq(adapter->pdev->irq, netdev);
298
299         if (adapter->have_msi)
300                 pci_disable_msi(adapter->pdev);
301
302         if (kill_watchdog)
303                 del_timer_sync(&adapter->watchdog_timer);
304
305         adapter->link_speed = 0;
306         adapter->link_duplex = 0;
307         netif_carrier_off(netdev);
308         netif_stop_queue(netdev);
309
310         ixgb_reset(adapter);
311         ixgb_clean_tx_ring(adapter);
312         ixgb_clean_rx_ring(adapter);
313 }
314
315 void
316 ixgb_reset(struct ixgb_adapter *adapter)
317 {
318         struct ixgb_hw *hw = &adapter->hw;
319
320         ixgb_adapter_stop(hw);
321         if (!ixgb_init_hw(hw))
322                 DPRINTK(PROBE, ERR, "ixgb_init_hw failed.\n");
323
324         /* restore frame size information */
325         IXGB_WRITE_REG(hw, MFS, hw->max_frame_size << IXGB_MFS_SHIFT);
326         if (hw->max_frame_size >
327             IXGB_MAX_ENET_FRAME_SIZE_WITHOUT_FCS + ENET_FCS_LENGTH) {
328                 u32 ctrl0 = IXGB_READ_REG(hw, CTRL0);
329                 if (!(ctrl0 & IXGB_CTRL0_JFE)) {
330                         ctrl0 |= IXGB_CTRL0_JFE;
331                         IXGB_WRITE_REG(hw, CTRL0, ctrl0);
332                 }
333         }
334 }
335
336 /**
337  * ixgb_probe - Device Initialization Routine
338  * @pdev: PCI device information struct
339  * @ent: entry in ixgb_pci_tbl
340  *
341  * Returns 0 on success, negative on failure
342  *
343  * ixgb_probe initializes an adapter identified by a pci_dev structure.
344  * The OS initialization, configuring of the adapter private structure,
345  * and a hardware reset occur.
346  **/
347
348 static int __devinit
349 ixgb_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
350 {
351         struct net_device *netdev = NULL;
352         struct ixgb_adapter *adapter;
353         static int cards_found = 0;
354         int pci_using_dac;
355         int i;
356         int err;
357
358         err = pci_enable_device(pdev);
359         if (err)
360                 return err;
361
362         if (!(err = pci_set_dma_mask(pdev, DMA_64BIT_MASK)) &&
363             !(err = pci_set_consistent_dma_mask(pdev, DMA_64BIT_MASK))) {
364                 pci_using_dac = 1;
365         } else {
366                 if ((err = pci_set_dma_mask(pdev, DMA_32BIT_MASK)) ||
367                     (err = pci_set_consistent_dma_mask(pdev, DMA_32BIT_MASK))) {
368                         printk(KERN_ERR
369                          "ixgb: No usable DMA configuration, aborting\n");
370                         goto err_dma_mask;
371                 }
372                 pci_using_dac = 0;
373         }
374
375         err = pci_request_regions(pdev, ixgb_driver_name);
376         if (err)
377                 goto err_request_regions;
378
379         pci_set_master(pdev);
380
381         netdev = alloc_etherdev(sizeof(struct ixgb_adapter));
382         if (!netdev) {
383                 err = -ENOMEM;
384                 goto err_alloc_etherdev;
385         }
386
387         SET_NETDEV_DEV(netdev, &pdev->dev);
388
389         pci_set_drvdata(pdev, netdev);
390         adapter = netdev_priv(netdev);
391         adapter->netdev = netdev;
392         adapter->pdev = pdev;
393         adapter->hw.back = adapter;
394         adapter->msg_enable = netif_msg_init(debug, DEFAULT_DEBUG_LEVEL_SHIFT);
395
396         adapter->hw.hw_addr = ioremap(pci_resource_start(pdev, BAR_0),
397                                       pci_resource_len(pdev, BAR_0));
398         if (!adapter->hw.hw_addr) {
399                 err = -EIO;
400                 goto err_ioremap;
401         }
402
403         for (i = BAR_1; i <= BAR_5; i++) {
404                 if (pci_resource_len(pdev, i) == 0)
405                         continue;
406                 if (pci_resource_flags(pdev, i) & IORESOURCE_IO) {
407                         adapter->hw.io_base = pci_resource_start(pdev, i);
408                         break;
409                 }
410         }
411
412         netdev->open = &ixgb_open;
413         netdev->stop = &ixgb_close;
414         netdev->hard_start_xmit = &ixgb_xmit_frame;
415         netdev->get_stats = &ixgb_get_stats;
416         netdev->set_multicast_list = &ixgb_set_multi;
417         netdev->set_mac_address = &ixgb_set_mac;
418         netdev->change_mtu = &ixgb_change_mtu;
419         ixgb_set_ethtool_ops(netdev);
420         netdev->tx_timeout = &ixgb_tx_timeout;
421         netdev->watchdog_timeo = 5 * HZ;
422 #ifdef CONFIG_IXGB_NAPI
423         netif_napi_add(netdev, &adapter->napi, ixgb_clean, 64);
424 #endif
425         netdev->vlan_rx_register = ixgb_vlan_rx_register;
426         netdev->vlan_rx_add_vid = ixgb_vlan_rx_add_vid;
427         netdev->vlan_rx_kill_vid = ixgb_vlan_rx_kill_vid;
428 #ifdef CONFIG_NET_POLL_CONTROLLER
429         netdev->poll_controller = ixgb_netpoll;
430 #endif
431
432         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
433
434         adapter->bd_number = cards_found;
435         adapter->link_speed = 0;
436         adapter->link_duplex = 0;
437
438         /* setup the private structure */
439
440         err = ixgb_sw_init(adapter);
441         if (err)
442                 goto err_sw_init;
443
444         netdev->features = NETIF_F_SG |
445                            NETIF_F_HW_CSUM |
446                            NETIF_F_HW_VLAN_TX |
447                            NETIF_F_HW_VLAN_RX |
448                            NETIF_F_HW_VLAN_FILTER;
449         netdev->features |= NETIF_F_TSO;
450
451         if (pci_using_dac)
452                 netdev->features |= NETIF_F_HIGHDMA;
453
454         /* make sure the EEPROM is good */
455
456         if (!ixgb_validate_eeprom_checksum(&adapter->hw)) {
457                 DPRINTK(PROBE, ERR, "The EEPROM Checksum Is Not Valid\n");
458                 err = -EIO;
459                 goto err_eeprom;
460         }
461
462         ixgb_get_ee_mac_addr(&adapter->hw, netdev->dev_addr);
463         memcpy(netdev->perm_addr, netdev->dev_addr, netdev->addr_len);
464
465         if (!is_valid_ether_addr(netdev->perm_addr)) {
466                 DPRINTK(PROBE, ERR, "Invalid MAC Address\n");
467                 err = -EIO;
468                 goto err_eeprom;
469         }
470
471         adapter->part_num = ixgb_get_ee_pba_number(&adapter->hw);
472
473         init_timer(&adapter->watchdog_timer);
474         adapter->watchdog_timer.function = &ixgb_watchdog;
475         adapter->watchdog_timer.data = (unsigned long)adapter;
476
477         INIT_WORK(&adapter->tx_timeout_task, ixgb_tx_timeout_task);
478
479         strcpy(netdev->name, "eth%d");
480         err = register_netdev(netdev);
481         if (err)
482                 goto err_register;
483
484         /* we're going to reset, so assume we have no link for now */
485
486         netif_carrier_off(netdev);
487         netif_stop_queue(netdev);
488
489         DPRINTK(PROBE, INFO, "Intel(R) PRO/10GbE Network Connection\n");
490         ixgb_check_options(adapter);
491         /* reset the hardware with the new settings */
492
493         ixgb_reset(adapter);
494
495         cards_found++;
496         return 0;
497
498 err_register:
499 err_sw_init:
500 err_eeprom:
501         iounmap(adapter->hw.hw_addr);
502 err_ioremap:
503         free_netdev(netdev);
504 err_alloc_etherdev:
505         pci_release_regions(pdev);
506 err_request_regions:
507 err_dma_mask:
508         pci_disable_device(pdev);
509         return err;
510 }
511
512 /**
513  * ixgb_remove - Device Removal Routine
514  * @pdev: PCI device information struct
515  *
516  * ixgb_remove is called by the PCI subsystem to alert the driver
517  * that it should release a PCI device.  The could be caused by a
518  * Hot-Plug event, or because the driver is going to be removed from
519  * memory.
520  **/
521
522 static void __devexit
523 ixgb_remove(struct pci_dev *pdev)
524 {
525         struct net_device *netdev = pci_get_drvdata(pdev);
526         struct ixgb_adapter *adapter = netdev_priv(netdev);
527
528         flush_scheduled_work();
529
530         unregister_netdev(netdev);
531
532         iounmap(adapter->hw.hw_addr);
533         pci_release_regions(pdev);
534
535         free_netdev(netdev);
536 }
537
538 /**
539  * ixgb_sw_init - Initialize general software structures (struct ixgb_adapter)
540  * @adapter: board private structure to initialize
541  *
542  * ixgb_sw_init initializes the Adapter private data structure.
543  * Fields are initialized based on PCI device information and
544  * OS network device settings (MTU size).
545  **/
546
547 static int __devinit
548 ixgb_sw_init(struct ixgb_adapter *adapter)
549 {
550         struct ixgb_hw *hw = &adapter->hw;
551         struct net_device *netdev = adapter->netdev;
552         struct pci_dev *pdev = adapter->pdev;
553
554         /* PCI config space info */
555
556         hw->vendor_id = pdev->vendor;
557         hw->device_id = pdev->device;
558         hw->subsystem_vendor_id = pdev->subsystem_vendor;
559         hw->subsystem_id = pdev->subsystem_device;
560
561         hw->max_frame_size = netdev->mtu + ENET_HEADER_SIZE + ENET_FCS_LENGTH;
562         adapter->rx_buffer_len = hw->max_frame_size + 8; /* + 8 for errata */
563
564         if ((hw->device_id == IXGB_DEVICE_ID_82597EX)
565            || (hw->device_id == IXGB_DEVICE_ID_82597EX_CX4)
566            || (hw->device_id == IXGB_DEVICE_ID_82597EX_LR)
567            || (hw->device_id == IXGB_DEVICE_ID_82597EX_SR))
568                 hw->mac_type = ixgb_82597;
569         else {
570                 /* should never have loaded on this device */
571                 DPRINTK(PROBE, ERR, "unsupported device id\n");
572         }
573
574         /* enable flow control to be programmed */
575         hw->fc.send_xon = 1;
576
577         set_bit(__IXGB_DOWN, &adapter->flags);
578         return 0;
579 }
580
581 /**
582  * ixgb_open - Called when a network interface is made active
583  * @netdev: network interface device structure
584  *
585  * Returns 0 on success, negative value on failure
586  *
587  * The open entry point is called when a network interface is made
588  * active by the system (IFF_UP).  At this point all resources needed
589  * for transmit and receive operations are allocated, the interrupt
590  * handler is registered with the OS, the watchdog timer is started,
591  * and the stack is notified that the interface is ready.
592  **/
593
594 static int
595 ixgb_open(struct net_device *netdev)
596 {
597         struct ixgb_adapter *adapter = netdev_priv(netdev);
598         int err;
599
600         /* allocate transmit descriptors */
601         err = ixgb_setup_tx_resources(adapter);
602         if (err)
603                 goto err_setup_tx;
604
605         /* allocate receive descriptors */
606
607         err = ixgb_setup_rx_resources(adapter);
608         if (err)
609                 goto err_setup_rx;
610
611         err = ixgb_up(adapter);
612         if (err)
613                 goto err_up;
614
615         return 0;
616
617 err_up:
618         ixgb_free_rx_resources(adapter);
619 err_setup_rx:
620         ixgb_free_tx_resources(adapter);
621 err_setup_tx:
622         ixgb_reset(adapter);
623
624         return err;
625 }
626
627 /**
628  * ixgb_close - Disables a network interface
629  * @netdev: network interface device structure
630  *
631  * Returns 0, this is not allowed to fail
632  *
633  * The close entry point is called when an interface is de-activated
634  * by the OS.  The hardware is still under the drivers control, but
635  * needs to be disabled.  A global MAC reset is issued to stop the
636  * hardware, and all transmit and receive resources are freed.
637  **/
638
639 static int
640 ixgb_close(struct net_device *netdev)
641 {
642         struct ixgb_adapter *adapter = netdev_priv(netdev);
643
644         ixgb_down(adapter, true);
645
646         ixgb_free_tx_resources(adapter);
647         ixgb_free_rx_resources(adapter);
648
649         return 0;
650 }
651
652 /**
653  * ixgb_setup_tx_resources - allocate Tx resources (Descriptors)
654  * @adapter: board private structure
655  *
656  * Return 0 on success, negative on failure
657  **/
658
659 int
660 ixgb_setup_tx_resources(struct ixgb_adapter *adapter)
661 {
662         struct ixgb_desc_ring *txdr = &adapter->tx_ring;
663         struct pci_dev *pdev = adapter->pdev;
664         int size;
665
666         size = sizeof(struct ixgb_buffer) * txdr->count;
667         txdr->buffer_info = vmalloc(size);
668         if (!txdr->buffer_info) {
669                 DPRINTK(PROBE, ERR,
670                  "Unable to allocate transmit descriptor ring memory\n");
671                 return -ENOMEM;
672         }
673         memset(txdr->buffer_info, 0, size);
674
675         /* round up to nearest 4K */
676
677         txdr->size = txdr->count * sizeof(struct ixgb_tx_desc);
678         txdr->size = ALIGN(txdr->size, 4096);
679
680         txdr->desc = pci_alloc_consistent(pdev, txdr->size, &txdr->dma);
681         if (!txdr->desc) {
682                 vfree(txdr->buffer_info);
683                 DPRINTK(PROBE, ERR,
684                  "Unable to allocate transmit descriptor memory\n");
685                 return -ENOMEM;
686         }
687         memset(txdr->desc, 0, txdr->size);
688
689         txdr->next_to_use = 0;
690         txdr->next_to_clean = 0;
691
692         return 0;
693 }
694
695 /**
696  * ixgb_configure_tx - Configure 82597 Transmit Unit after Reset.
697  * @adapter: board private structure
698  *
699  * Configure the Tx unit of the MAC after a reset.
700  **/
701
702 static void
703 ixgb_configure_tx(struct ixgb_adapter *adapter)
704 {
705         u64 tdba = adapter->tx_ring.dma;
706         u32 tdlen = adapter->tx_ring.count * sizeof(struct ixgb_tx_desc);
707         u32 tctl;
708         struct ixgb_hw *hw = &adapter->hw;
709
710         /* Setup the Base and Length of the Tx Descriptor Ring
711          * tx_ring.dma can be either a 32 or 64 bit value
712          */
713
714         IXGB_WRITE_REG(hw, TDBAL, (tdba & 0x00000000ffffffffULL));
715         IXGB_WRITE_REG(hw, TDBAH, (tdba >> 32));
716
717         IXGB_WRITE_REG(hw, TDLEN, tdlen);
718
719         /* Setup the HW Tx Head and Tail descriptor pointers */
720
721         IXGB_WRITE_REG(hw, TDH, 0);
722         IXGB_WRITE_REG(hw, TDT, 0);
723
724         /* don't set up txdctl, it induces performance problems if configured
725          * incorrectly */
726         /* Set the Tx Interrupt Delay register */
727
728         IXGB_WRITE_REG(hw, TIDV, adapter->tx_int_delay);
729
730         /* Program the Transmit Control Register */
731
732         tctl = IXGB_TCTL_TCE | IXGB_TCTL_TXEN | IXGB_TCTL_TPDE;
733         IXGB_WRITE_REG(hw, TCTL, tctl);
734
735         /* Setup Transmit Descriptor Settings for this adapter */
736         adapter->tx_cmd_type =
737                 IXGB_TX_DESC_TYPE |
738                 (adapter->tx_int_delay_enable ? IXGB_TX_DESC_CMD_IDE : 0);
739 }
740
741 /**
742  * ixgb_setup_rx_resources - allocate Rx resources (Descriptors)
743  * @adapter: board private structure
744  *
745  * Returns 0 on success, negative on failure
746  **/
747
748 int
749 ixgb_setup_rx_resources(struct ixgb_adapter *adapter)
750 {
751         struct ixgb_desc_ring *rxdr = &adapter->rx_ring;
752         struct pci_dev *pdev = adapter->pdev;
753         int size;
754
755         size = sizeof(struct ixgb_buffer) * rxdr->count;
756         rxdr->buffer_info = vmalloc(size);
757         if (!rxdr->buffer_info) {
758                 DPRINTK(PROBE, ERR,
759                  "Unable to allocate receive descriptor ring\n");
760                 return -ENOMEM;
761         }
762         memset(rxdr->buffer_info, 0, size);
763
764         /* Round up to nearest 4K */
765
766         rxdr->size = rxdr->count * sizeof(struct ixgb_rx_desc);
767         rxdr->size = ALIGN(rxdr->size, 4096);
768
769         rxdr->desc = pci_alloc_consistent(pdev, rxdr->size, &rxdr->dma);
770
771         if (!rxdr->desc) {
772                 vfree(rxdr->buffer_info);
773                 DPRINTK(PROBE, ERR,
774                  "Unable to allocate receive descriptors\n");
775                 return -ENOMEM;
776         }
777         memset(rxdr->desc, 0, rxdr->size);
778
779         rxdr->next_to_clean = 0;
780         rxdr->next_to_use = 0;
781
782         return 0;
783 }
784
785 /**
786  * ixgb_setup_rctl - configure the receive control register
787  * @adapter: Board private structure
788  **/
789
790 static void
791 ixgb_setup_rctl(struct ixgb_adapter *adapter)
792 {
793         u32 rctl;
794
795         rctl = IXGB_READ_REG(&adapter->hw, RCTL);
796
797         rctl &= ~(3 << IXGB_RCTL_MO_SHIFT);
798
799         rctl |=
800                 IXGB_RCTL_BAM | IXGB_RCTL_RDMTS_1_2 |
801                 IXGB_RCTL_RXEN | IXGB_RCTL_CFF |
802                 (adapter->hw.mc_filter_type << IXGB_RCTL_MO_SHIFT);
803
804         rctl |= IXGB_RCTL_SECRC;
805
806         if (adapter->rx_buffer_len <= IXGB_RXBUFFER_2048)
807                 rctl |= IXGB_RCTL_BSIZE_2048;
808         else if (adapter->rx_buffer_len <= IXGB_RXBUFFER_4096)
809                 rctl |= IXGB_RCTL_BSIZE_4096;
810         else if (adapter->rx_buffer_len <= IXGB_RXBUFFER_8192)
811                 rctl |= IXGB_RCTL_BSIZE_8192;
812         else if (adapter->rx_buffer_len <= IXGB_RXBUFFER_16384)
813                 rctl |= IXGB_RCTL_BSIZE_16384;
814
815         IXGB_WRITE_REG(&adapter->hw, RCTL, rctl);
816 }
817
818 /**
819  * ixgb_configure_rx - Configure 82597 Receive Unit after Reset.
820  * @adapter: board private structure
821  *
822  * Configure the Rx unit of the MAC after a reset.
823  **/
824
825 static void
826 ixgb_configure_rx(struct ixgb_adapter *adapter)
827 {
828         u64 rdba = adapter->rx_ring.dma;
829         u32 rdlen = adapter->rx_ring.count * sizeof(struct ixgb_rx_desc);
830         struct ixgb_hw *hw = &adapter->hw;
831         u32 rctl;
832         u32 rxcsum;
833
834         /* make sure receives are disabled while setting up the descriptors */
835
836         rctl = IXGB_READ_REG(hw, RCTL);
837         IXGB_WRITE_REG(hw, RCTL, rctl & ~IXGB_RCTL_RXEN);
838
839         /* set the Receive Delay Timer Register */
840
841         IXGB_WRITE_REG(hw, RDTR, adapter->rx_int_delay);
842
843         /* Setup the Base and Length of the Rx Descriptor Ring */
844
845         IXGB_WRITE_REG(hw, RDBAL, (rdba & 0x00000000ffffffffULL));
846         IXGB_WRITE_REG(hw, RDBAH, (rdba >> 32));
847
848         IXGB_WRITE_REG(hw, RDLEN, rdlen);
849
850         /* Setup the HW Rx Head and Tail Descriptor Pointers */
851         IXGB_WRITE_REG(hw, RDH, 0);
852         IXGB_WRITE_REG(hw, RDT, 0);
853
854         /* due to the hardware errata with RXDCTL, we are unable to use any of
855          * the performance enhancing features of it without causing other
856          * subtle bugs, some of the bugs could include receive length
857          * corruption at high data rates (WTHRESH > 0) and/or receive
858          * descriptor ring irregularites (particularly in hardware cache) */
859         IXGB_WRITE_REG(hw, RXDCTL, 0);
860
861         /* Enable Receive Checksum Offload for TCP and UDP */
862         if (adapter->rx_csum) {
863                 rxcsum = IXGB_READ_REG(hw, RXCSUM);
864                 rxcsum |= IXGB_RXCSUM_TUOFL;
865                 IXGB_WRITE_REG(hw, RXCSUM, rxcsum);
866         }
867
868         /* Enable Receives */
869
870         IXGB_WRITE_REG(hw, RCTL, rctl);
871 }
872
873 /**
874  * ixgb_free_tx_resources - Free Tx Resources
875  * @adapter: board private structure
876  *
877  * Free all transmit software resources
878  **/
879
880 void
881 ixgb_free_tx_resources(struct ixgb_adapter *adapter)
882 {
883         struct pci_dev *pdev = adapter->pdev;
884
885         ixgb_clean_tx_ring(adapter);
886
887         vfree(adapter->tx_ring.buffer_info);
888         adapter->tx_ring.buffer_info = NULL;
889
890         pci_free_consistent(pdev, adapter->tx_ring.size,
891                             adapter->tx_ring.desc, adapter->tx_ring.dma);
892
893         adapter->tx_ring.desc = NULL;
894 }
895
896 static void
897 ixgb_unmap_and_free_tx_resource(struct ixgb_adapter *adapter,
898                                 struct ixgb_buffer *buffer_info)
899 {
900         struct pci_dev *pdev = adapter->pdev;
901
902         if (buffer_info->dma)
903                 pci_unmap_page(pdev, buffer_info->dma, buffer_info->length,
904                                PCI_DMA_TODEVICE);
905
906         /* okay to call kfree_skb here instead of kfree_skb_any because
907          * this is never called in interrupt context */
908         if (buffer_info->skb)
909                 dev_kfree_skb(buffer_info->skb);
910
911         buffer_info->skb = NULL;
912         buffer_info->dma = 0;
913         buffer_info->time_stamp = 0;
914         /* these fields must always be initialized in tx
915          * buffer_info->length = 0;
916          * buffer_info->next_to_watch = 0; */
917 }
918
919 /**
920  * ixgb_clean_tx_ring - Free Tx Buffers
921  * @adapter: board private structure
922  **/
923
924 static void
925 ixgb_clean_tx_ring(struct ixgb_adapter *adapter)
926 {
927         struct ixgb_desc_ring *tx_ring = &adapter->tx_ring;
928         struct ixgb_buffer *buffer_info;
929         unsigned long size;
930         unsigned int i;
931
932         /* Free all the Tx ring sk_buffs */
933
934         for (i = 0; i < tx_ring->count; i++) {
935                 buffer_info = &tx_ring->buffer_info[i];
936                 ixgb_unmap_and_free_tx_resource(adapter, buffer_info);
937         }
938
939         size = sizeof(struct ixgb_buffer) * tx_ring->count;
940         memset(tx_ring->buffer_info, 0, size);
941
942         /* Zero out the descriptor ring */
943
944         memset(tx_ring->desc, 0, tx_ring->size);
945
946         tx_ring->next_to_use = 0;
947         tx_ring->next_to_clean = 0;
948
949         IXGB_WRITE_REG(&adapter->hw, TDH, 0);
950         IXGB_WRITE_REG(&adapter->hw, TDT, 0);
951 }
952
953 /**
954  * ixgb_free_rx_resources - Free Rx Resources
955  * @adapter: board private structure
956  *
957  * Free all receive software resources
958  **/
959
960 void
961 ixgb_free_rx_resources(struct ixgb_adapter *adapter)
962 {
963         struct ixgb_desc_ring *rx_ring = &adapter->rx_ring;
964         struct pci_dev *pdev = adapter->pdev;
965
966         ixgb_clean_rx_ring(adapter);
967
968         vfree(rx_ring->buffer_info);
969         rx_ring->buffer_info = NULL;
970
971         pci_free_consistent(pdev, rx_ring->size, rx_ring->desc, rx_ring->dma);
972
973         rx_ring->desc = NULL;
974 }
975
976 /**
977  * ixgb_clean_rx_ring - Free Rx Buffers
978  * @adapter: board private structure
979  **/
980
981 static void
982 ixgb_clean_rx_ring(struct ixgb_adapter *adapter)
983 {
984         struct ixgb_desc_ring *rx_ring = &adapter->rx_ring;
985         struct ixgb_buffer *buffer_info;
986         struct pci_dev *pdev = adapter->pdev;
987         unsigned long size;
988         unsigned int i;
989
990         /* Free all the Rx ring sk_buffs */
991
992         for (i = 0; i < rx_ring->count; i++) {
993                 buffer_info = &rx_ring->buffer_info[i];
994                 if (buffer_info->skb) {
995
996                         pci_unmap_single(pdev,
997                                          buffer_info->dma,
998                                          buffer_info->length,
999                                          PCI_DMA_FROMDEVICE);
1000
1001                         dev_kfree_skb(buffer_info->skb);
1002
1003                         buffer_info->skb = NULL;
1004                 }
1005         }
1006
1007         size = sizeof(struct ixgb_buffer) * rx_ring->count;
1008         memset(rx_ring->buffer_info, 0, size);
1009
1010         /* Zero out the descriptor ring */
1011
1012         memset(rx_ring->desc, 0, rx_ring->size);
1013
1014         rx_ring->next_to_clean = 0;
1015         rx_ring->next_to_use = 0;
1016
1017         IXGB_WRITE_REG(&adapter->hw, RDH, 0);
1018         IXGB_WRITE_REG(&adapter->hw, RDT, 0);
1019 }
1020
1021 /**
1022  * ixgb_set_mac - Change the Ethernet Address of the NIC
1023  * @netdev: network interface device structure
1024  * @p: pointer to an address structure
1025  *
1026  * Returns 0 on success, negative on failure
1027  **/
1028
1029 static int
1030 ixgb_set_mac(struct net_device *netdev, void *p)
1031 {
1032         struct ixgb_adapter *adapter = netdev_priv(netdev);
1033         struct sockaddr *addr = p;
1034
1035         if (!is_valid_ether_addr(addr->sa_data))
1036                 return -EADDRNOTAVAIL;
1037
1038         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
1039
1040         ixgb_rar_set(&adapter->hw, addr->sa_data, 0);
1041
1042         return 0;
1043 }
1044
1045 /**
1046  * ixgb_set_multi - Multicast and Promiscuous mode set
1047  * @netdev: network interface device structure
1048  *
1049  * The set_multi entry point is called whenever the multicast address
1050  * list or the network interface flags are updated.  This routine is
1051  * responsible for configuring the hardware for proper multicast,
1052  * promiscuous mode, and all-multi behavior.
1053  **/
1054
1055 static void
1056 ixgb_set_multi(struct net_device *netdev)
1057 {
1058         struct ixgb_adapter *adapter = netdev_priv(netdev);
1059         struct ixgb_hw *hw = &adapter->hw;
1060         struct dev_mc_list *mc_ptr;
1061         u32 rctl;
1062         int i;
1063
1064         /* Check for Promiscuous and All Multicast modes */
1065
1066         rctl = IXGB_READ_REG(hw, RCTL);
1067
1068         if (netdev->flags & IFF_PROMISC) {
1069                 rctl |= (IXGB_RCTL_UPE | IXGB_RCTL_MPE);
1070         } else if (netdev->flags & IFF_ALLMULTI) {
1071                 rctl |= IXGB_RCTL_MPE;
1072                 rctl &= ~IXGB_RCTL_UPE;
1073         } else {
1074                 rctl &= ~(IXGB_RCTL_UPE | IXGB_RCTL_MPE);
1075         }
1076
1077         if (netdev->mc_count > IXGB_MAX_NUM_MULTICAST_ADDRESSES) {
1078                 rctl |= IXGB_RCTL_MPE;
1079                 IXGB_WRITE_REG(hw, RCTL, rctl);
1080         } else {
1081                 u8 mta[IXGB_MAX_NUM_MULTICAST_ADDRESSES *
1082                             IXGB_ETH_LENGTH_OF_ADDRESS];
1083
1084                 IXGB_WRITE_REG(hw, RCTL, rctl);
1085
1086                 for (i = 0, mc_ptr = netdev->mc_list;
1087                      mc_ptr;
1088                      i++, mc_ptr = mc_ptr->next)
1089                         memcpy(&mta[i * IXGB_ETH_LENGTH_OF_ADDRESS],
1090                                mc_ptr->dmi_addr, IXGB_ETH_LENGTH_OF_ADDRESS);
1091
1092                 ixgb_mc_addr_list_update(hw, mta, netdev->mc_count, 0);
1093         }
1094 }
1095
1096 /**
1097  * ixgb_watchdog - Timer Call-back
1098  * @data: pointer to netdev cast into an unsigned long
1099  **/
1100
1101 static void
1102 ixgb_watchdog(unsigned long data)
1103 {
1104         struct ixgb_adapter *adapter = (struct ixgb_adapter *)data;
1105         struct net_device *netdev = adapter->netdev;
1106         struct ixgb_desc_ring *txdr = &adapter->tx_ring;
1107
1108         ixgb_check_for_link(&adapter->hw);
1109
1110         if (ixgb_check_for_bad_link(&adapter->hw)) {
1111                 /* force the reset path */
1112                 netif_stop_queue(netdev);
1113         }
1114
1115         if (adapter->hw.link_up) {
1116                 if (!netif_carrier_ok(netdev)) {
1117                         DPRINTK(LINK, INFO,
1118                                 "NIC Link is Up 10000 Mbps Full Duplex\n");
1119                         adapter->link_speed = 10000;
1120                         adapter->link_duplex = FULL_DUPLEX;
1121                         netif_carrier_on(netdev);
1122                         netif_wake_queue(netdev);
1123                 }
1124         } else {
1125                 if (netif_carrier_ok(netdev)) {
1126                         adapter->link_speed = 0;
1127                         adapter->link_duplex = 0;
1128                         DPRINTK(LINK, INFO, "NIC Link is Down\n");
1129                         netif_carrier_off(netdev);
1130                         netif_stop_queue(netdev);
1131
1132                 }
1133         }
1134
1135         ixgb_update_stats(adapter);
1136
1137         if (!netif_carrier_ok(netdev)) {
1138                 if (IXGB_DESC_UNUSED(txdr) + 1 < txdr->count) {
1139                         /* We've lost link, so the controller stops DMA,
1140                          * but we've got queued Tx work that's never going
1141                          * to get done, so reset controller to flush Tx.
1142                          * (Do the reset outside of interrupt context). */
1143                         schedule_work(&adapter->tx_timeout_task);
1144                 }
1145         }
1146
1147         /* Force detection of hung controller every watchdog period */
1148         adapter->detect_tx_hung = true;
1149
1150         /* generate an interrupt to force clean up of any stragglers */
1151         IXGB_WRITE_REG(&adapter->hw, ICS, IXGB_INT_TXDW);
1152
1153         /* Reset the timer */
1154         mod_timer(&adapter->watchdog_timer, jiffies + 2 * HZ);
1155 }
1156
1157 #define IXGB_TX_FLAGS_CSUM              0x00000001
1158 #define IXGB_TX_FLAGS_VLAN              0x00000002
1159 #define IXGB_TX_FLAGS_TSO               0x00000004
1160
1161 static int
1162 ixgb_tso(struct ixgb_adapter *adapter, struct sk_buff *skb)
1163 {
1164         struct ixgb_context_desc *context_desc;
1165         unsigned int i;
1166         u8 ipcss, ipcso, tucss, tucso, hdr_len;
1167         u16 ipcse, tucse, mss;
1168         int err;
1169
1170         if (likely(skb_is_gso(skb))) {
1171                 struct ixgb_buffer *buffer_info;
1172                 struct iphdr *iph;
1173
1174                 if (skb_header_cloned(skb)) {
1175                         err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1176                         if (err)
1177                                 return err;
1178                 }
1179
1180                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
1181                 mss = skb_shinfo(skb)->gso_size;
1182                 iph = ip_hdr(skb);
1183                 iph->tot_len = 0;
1184                 iph->check = 0;
1185                 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
1186                                                          iph->daddr, 0,
1187                                                          IPPROTO_TCP, 0);
1188                 ipcss = skb_network_offset(skb);
1189                 ipcso = (void *)&(iph->check) - (void *)skb->data;
1190                 ipcse = skb_transport_offset(skb) - 1;
1191                 tucss = skb_transport_offset(skb);
1192                 tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
1193                 tucse = 0;
1194
1195                 i = adapter->tx_ring.next_to_use;
1196                 context_desc = IXGB_CONTEXT_DESC(adapter->tx_ring, i);
1197                 buffer_info = &adapter->tx_ring.buffer_info[i];
1198                 WARN_ON(buffer_info->dma != 0);
1199
1200                 context_desc->ipcss = ipcss;
1201                 context_desc->ipcso = ipcso;
1202                 context_desc->ipcse = cpu_to_le16(ipcse);
1203                 context_desc->tucss = tucss;
1204                 context_desc->tucso = tucso;
1205                 context_desc->tucse = cpu_to_le16(tucse);
1206                 context_desc->mss = cpu_to_le16(mss);
1207                 context_desc->hdr_len = hdr_len;
1208                 context_desc->status = 0;
1209                 context_desc->cmd_type_len = cpu_to_le32(
1210                                                   IXGB_CONTEXT_DESC_TYPE
1211                                                 | IXGB_CONTEXT_DESC_CMD_TSE
1212                                                 | IXGB_CONTEXT_DESC_CMD_IP
1213                                                 | IXGB_CONTEXT_DESC_CMD_TCP
1214                                                 | IXGB_CONTEXT_DESC_CMD_IDE
1215                                                 | (skb->len - (hdr_len)));
1216
1217
1218                 if (++i == adapter->tx_ring.count) i = 0;
1219                 adapter->tx_ring.next_to_use = i;
1220
1221                 return 1;
1222         }
1223
1224         return 0;
1225 }
1226
1227 static bool
1228 ixgb_tx_csum(struct ixgb_adapter *adapter, struct sk_buff *skb)
1229 {
1230         struct ixgb_context_desc *context_desc;
1231         unsigned int i;
1232         u8 css, cso;
1233
1234         if (likely(skb->ip_summed == CHECKSUM_PARTIAL)) {
1235                 struct ixgb_buffer *buffer_info;
1236                 css = skb_transport_offset(skb);
1237                 cso = css + skb->csum_offset;
1238
1239                 i = adapter->tx_ring.next_to_use;
1240                 context_desc = IXGB_CONTEXT_DESC(adapter->tx_ring, i);
1241                 buffer_info = &adapter->tx_ring.buffer_info[i];
1242                 WARN_ON(buffer_info->dma != 0);
1243
1244                 context_desc->tucss = css;
1245                 context_desc->tucso = cso;
1246                 context_desc->tucse = 0;
1247                 /* zero out any previously existing data in one instruction */
1248                 *(u32 *)&(context_desc->ipcss) = 0;
1249                 context_desc->status = 0;
1250                 context_desc->hdr_len = 0;
1251                 context_desc->mss = 0;
1252                 context_desc->cmd_type_len =
1253                         cpu_to_le32(IXGB_CONTEXT_DESC_TYPE
1254                                     | IXGB_TX_DESC_CMD_IDE);
1255
1256                 if (++i == adapter->tx_ring.count) i = 0;
1257                 adapter->tx_ring.next_to_use = i;
1258
1259                 return true;
1260         }
1261
1262         return false;
1263 }
1264
1265 #define IXGB_MAX_TXD_PWR        14
1266 #define IXGB_MAX_DATA_PER_TXD   (1<<IXGB_MAX_TXD_PWR)
1267
1268 static int
1269 ixgb_tx_map(struct ixgb_adapter *adapter, struct sk_buff *skb,
1270             unsigned int first)
1271 {
1272         struct ixgb_desc_ring *tx_ring = &adapter->tx_ring;
1273         struct ixgb_buffer *buffer_info;
1274         int len = skb->len;
1275         unsigned int offset = 0, size, count = 0, i;
1276         unsigned int mss = skb_shinfo(skb)->gso_size;
1277
1278         unsigned int nr_frags = skb_shinfo(skb)->nr_frags;
1279         unsigned int f;
1280
1281         len -= skb->data_len;
1282
1283         i = tx_ring->next_to_use;
1284
1285         while (len) {
1286                 buffer_info = &tx_ring->buffer_info[i];
1287                 size = min(len, IXGB_MAX_DATA_PER_TXD);
1288                 /* Workaround for premature desc write-backs
1289                  * in TSO mode.  Append 4-byte sentinel desc */
1290                 if (unlikely(mss && !nr_frags && size == len && size > 8))
1291                         size -= 4;
1292
1293                 buffer_info->length = size;
1294                 WARN_ON(buffer_info->dma != 0);
1295                 buffer_info->time_stamp = jiffies;
1296                 buffer_info->dma =
1297                         pci_map_single(adapter->pdev,
1298                                 skb->data + offset,
1299                                 size,
1300                                 PCI_DMA_TODEVICE);
1301                 buffer_info->next_to_watch = 0;
1302
1303                 len -= size;
1304                 offset += size;
1305                 count++;
1306                 if (++i == tx_ring->count) i = 0;
1307         }
1308
1309         for (f = 0; f < nr_frags; f++) {
1310                 struct skb_frag_struct *frag;
1311
1312                 frag = &skb_shinfo(skb)->frags[f];
1313                 len = frag->size;
1314                 offset = 0;
1315
1316                 while (len) {
1317                         buffer_info = &tx_ring->buffer_info[i];
1318                         size = min(len, IXGB_MAX_DATA_PER_TXD);
1319
1320                         /* Workaround for premature desc write-backs
1321                          * in TSO mode.  Append 4-byte sentinel desc */
1322                         if (unlikely(mss && (f == (nr_frags - 1))
1323                                      && size == len && size > 8))
1324                                 size -= 4;
1325
1326                         buffer_info->length = size;
1327                         buffer_info->time_stamp = jiffies;
1328                         buffer_info->dma =
1329                                 pci_map_page(adapter->pdev,
1330                                         frag->page,
1331                                         frag->page_offset + offset,
1332                                         size,
1333                                         PCI_DMA_TODEVICE);
1334                         buffer_info->next_to_watch = 0;
1335
1336                         len -= size;
1337                         offset += size;
1338                         count++;
1339                         if (++i == tx_ring->count) i = 0;
1340                 }
1341         }
1342         i = (i == 0) ? tx_ring->count - 1 : i - 1;
1343         tx_ring->buffer_info[i].skb = skb;
1344         tx_ring->buffer_info[first].next_to_watch = i;
1345
1346         return count;
1347 }
1348
1349 static void
1350 ixgb_tx_queue(struct ixgb_adapter *adapter, int count, int vlan_id,int tx_flags)
1351 {
1352         struct ixgb_desc_ring *tx_ring = &adapter->tx_ring;
1353         struct ixgb_tx_desc *tx_desc = NULL;
1354         struct ixgb_buffer *buffer_info;
1355         u32 cmd_type_len = adapter->tx_cmd_type;
1356         u8 status = 0;
1357         u8 popts = 0;
1358         unsigned int i;
1359
1360         if (tx_flags & IXGB_TX_FLAGS_TSO) {
1361                 cmd_type_len |= IXGB_TX_DESC_CMD_TSE;
1362                 popts |= (IXGB_TX_DESC_POPTS_IXSM | IXGB_TX_DESC_POPTS_TXSM);
1363         }
1364
1365         if (tx_flags & IXGB_TX_FLAGS_CSUM)
1366                 popts |= IXGB_TX_DESC_POPTS_TXSM;
1367
1368         if (tx_flags & IXGB_TX_FLAGS_VLAN)
1369                 cmd_type_len |= IXGB_TX_DESC_CMD_VLE;
1370
1371         i = tx_ring->next_to_use;
1372
1373         while (count--) {
1374                 buffer_info = &tx_ring->buffer_info[i];
1375                 tx_desc = IXGB_TX_DESC(*tx_ring, i);
1376                 tx_desc->buff_addr = cpu_to_le64(buffer_info->dma);
1377                 tx_desc->cmd_type_len =
1378                         cpu_to_le32(cmd_type_len | buffer_info->length);
1379                 tx_desc->status = status;
1380                 tx_desc->popts = popts;
1381                 tx_desc->vlan = cpu_to_le16(vlan_id);
1382
1383                 if (++i == tx_ring->count) i = 0;
1384         }
1385
1386         tx_desc->cmd_type_len |=
1387                 cpu_to_le32(IXGB_TX_DESC_CMD_EOP | IXGB_TX_DESC_CMD_RS);
1388
1389         /* Force memory writes to complete before letting h/w
1390          * know there are new descriptors to fetch.  (Only
1391          * applicable for weak-ordered memory model archs,
1392          * such as IA-64). */
1393         wmb();
1394
1395         tx_ring->next_to_use = i;
1396         IXGB_WRITE_REG(&adapter->hw, TDT, i);
1397 }
1398
1399 static int __ixgb_maybe_stop_tx(struct net_device *netdev, int size)
1400 {
1401         struct ixgb_adapter *adapter = netdev_priv(netdev);
1402         struct ixgb_desc_ring *tx_ring = &adapter->tx_ring;
1403
1404         netif_stop_queue(netdev);
1405         /* Herbert's original patch had:
1406          *  smp_mb__after_netif_stop_queue();
1407          * but since that doesn't exist yet, just open code it. */
1408         smp_mb();
1409
1410         /* We need to check again in a case another CPU has just
1411          * made room available. */
1412         if (likely(IXGB_DESC_UNUSED(tx_ring) < size))
1413                 return -EBUSY;
1414
1415         /* A reprieve! */
1416         netif_start_queue(netdev);
1417         ++adapter->restart_queue;
1418         return 0;
1419 }
1420
1421 static int ixgb_maybe_stop_tx(struct net_device *netdev,
1422                               struct ixgb_desc_ring *tx_ring, int size)
1423 {
1424         if (likely(IXGB_DESC_UNUSED(tx_ring) >= size))
1425                 return 0;
1426         return __ixgb_maybe_stop_tx(netdev, size);
1427 }
1428
1429
1430 /* Tx Descriptors needed, worst case */
1431 #define TXD_USE_COUNT(S) (((S) >> IXGB_MAX_TXD_PWR) + \
1432                          (((S) & (IXGB_MAX_DATA_PER_TXD - 1)) ? 1 : 0))
1433 #define DESC_NEEDED TXD_USE_COUNT(IXGB_MAX_DATA_PER_TXD) /* skb->date */ + \
1434         MAX_SKB_FRAGS * TXD_USE_COUNT(PAGE_SIZE) + 1 /* for context */ \
1435         + 1 /* one more needed for sentinel TSO workaround */
1436
1437 static int
1438 ixgb_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
1439 {
1440         struct ixgb_adapter *adapter = netdev_priv(netdev);
1441         unsigned int first;
1442         unsigned int tx_flags = 0;
1443         int vlan_id = 0;
1444         int tso;
1445
1446         if (test_bit(__IXGB_DOWN, &adapter->flags)) {
1447                 dev_kfree_skb(skb);
1448                 return NETDEV_TX_OK;
1449         }
1450
1451         if (skb->len <= 0) {
1452                 dev_kfree_skb(skb);
1453                 return 0;
1454         }
1455
1456         if (unlikely(ixgb_maybe_stop_tx(netdev, &adapter->tx_ring,
1457                      DESC_NEEDED)))
1458                 return NETDEV_TX_BUSY;
1459
1460         if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
1461                 tx_flags |= IXGB_TX_FLAGS_VLAN;
1462                 vlan_id = vlan_tx_tag_get(skb);
1463         }
1464
1465         first = adapter->tx_ring.next_to_use;
1466
1467         tso = ixgb_tso(adapter, skb);
1468         if (tso < 0) {
1469                 dev_kfree_skb(skb);
1470                 return NETDEV_TX_OK;
1471         }
1472
1473         if (likely(tso))
1474                 tx_flags |= IXGB_TX_FLAGS_TSO;
1475         else if (ixgb_tx_csum(adapter, skb))
1476                 tx_flags |= IXGB_TX_FLAGS_CSUM;
1477
1478         ixgb_tx_queue(adapter, ixgb_tx_map(adapter, skb, first), vlan_id,
1479                         tx_flags);
1480
1481         netdev->trans_start = jiffies;
1482
1483         /* Make sure there is space in the ring for the next send. */
1484         ixgb_maybe_stop_tx(netdev, &adapter->tx_ring, DESC_NEEDED);
1485
1486         return NETDEV_TX_OK;
1487 }
1488
1489 /**
1490  * ixgb_tx_timeout - Respond to a Tx Hang
1491  * @netdev: network interface device structure
1492  **/
1493
1494 static void
1495 ixgb_tx_timeout(struct net_device *netdev)
1496 {
1497         struct ixgb_adapter *adapter = netdev_priv(netdev);
1498
1499         /* Do the reset outside of interrupt context */
1500         schedule_work(&adapter->tx_timeout_task);
1501 }
1502
1503 static void
1504 ixgb_tx_timeout_task(struct work_struct *work)
1505 {
1506         struct ixgb_adapter *adapter =
1507                 container_of(work, struct ixgb_adapter, tx_timeout_task);
1508
1509         adapter->tx_timeout_count++;
1510         ixgb_down(adapter, true);
1511         ixgb_up(adapter);
1512 }
1513
1514 /**
1515  * ixgb_get_stats - Get System Network Statistics
1516  * @netdev: network interface device structure
1517  *
1518  * Returns the address of the device statistics structure.
1519  * The statistics are actually updated from the timer callback.
1520  **/
1521
1522 static struct net_device_stats *
1523 ixgb_get_stats(struct net_device *netdev)
1524 {
1525         struct ixgb_adapter *adapter = netdev_priv(netdev);
1526
1527         return &adapter->net_stats;
1528 }
1529
1530 /**
1531  * ixgb_change_mtu - Change the Maximum Transfer Unit
1532  * @netdev: network interface device structure
1533  * @new_mtu: new value for maximum frame size
1534  *
1535  * Returns 0 on success, negative on failure
1536  **/
1537
1538 static int
1539 ixgb_change_mtu(struct net_device *netdev, int new_mtu)
1540 {
1541         struct ixgb_adapter *adapter = netdev_priv(netdev);
1542         int max_frame = new_mtu + ENET_HEADER_SIZE + ENET_FCS_LENGTH;
1543         int old_max_frame = netdev->mtu + ENET_HEADER_SIZE + ENET_FCS_LENGTH;
1544
1545         /* MTU < 68 is an error for IPv4 traffic, just don't allow it */
1546         if ((new_mtu < 68) ||
1547             (max_frame > IXGB_MAX_JUMBO_FRAME_SIZE + ENET_FCS_LENGTH)) {
1548                 DPRINTK(PROBE, ERR, "Invalid MTU setting %d\n", new_mtu);
1549                 return -EINVAL;
1550         }
1551
1552         if (old_max_frame == max_frame)
1553                 return 0;
1554
1555         if (netif_running(netdev))
1556                 ixgb_down(adapter, true);
1557
1558         adapter->rx_buffer_len = max_frame + 8; /* + 8 for errata */
1559
1560         netdev->mtu = new_mtu;
1561
1562         if (netif_running(netdev))
1563                 ixgb_up(adapter);
1564
1565         return 0;
1566 }
1567
1568 /**
1569  * ixgb_update_stats - Update the board statistics counters.
1570  * @adapter: board private structure
1571  **/
1572
1573 void
1574 ixgb_update_stats(struct ixgb_adapter *adapter)
1575 {
1576         struct net_device *netdev = adapter->netdev;
1577         struct pci_dev *pdev = adapter->pdev;
1578
1579         /* Prevent stats update while adapter is being reset */
1580         if (pci_channel_offline(pdev))
1581                 return;
1582
1583         if ((netdev->flags & IFF_PROMISC) || (netdev->flags & IFF_ALLMULTI) ||
1584            (netdev->mc_count > IXGB_MAX_NUM_MULTICAST_ADDRESSES)) {
1585                 u64 multi = IXGB_READ_REG(&adapter->hw, MPRCL);
1586                 u32 bcast_l = IXGB_READ_REG(&adapter->hw, BPRCL);
1587                 u32 bcast_h = IXGB_READ_REG(&adapter->hw, BPRCH);
1588                 u64 bcast = ((u64)bcast_h << 32) | bcast_l;
1589
1590                 multi |= ((u64)IXGB_READ_REG(&adapter->hw, MPRCH) << 32);
1591                 /* fix up multicast stats by removing broadcasts */
1592                 if (multi >= bcast)
1593                         multi -= bcast;
1594
1595                 adapter->stats.mprcl += (multi & 0xFFFFFFFF);
1596                 adapter->stats.mprch += (multi >> 32);
1597                 adapter->stats.bprcl += bcast_l;
1598                 adapter->stats.bprch += bcast_h;
1599         } else {
1600                 adapter->stats.mprcl += IXGB_READ_REG(&adapter->hw, MPRCL);
1601                 adapter->stats.mprch += IXGB_READ_REG(&adapter->hw, MPRCH);
1602                 adapter->stats.bprcl += IXGB_READ_REG(&adapter->hw, BPRCL);
1603                 adapter->stats.bprch += IXGB_READ_REG(&adapter->hw, BPRCH);
1604         }
1605         adapter->stats.tprl += IXGB_READ_REG(&adapter->hw, TPRL);
1606         adapter->stats.tprh += IXGB_READ_REG(&adapter->hw, TPRH);
1607         adapter->stats.gprcl += IXGB_READ_REG(&adapter->hw, GPRCL);
1608         adapter->stats.gprch += IXGB_READ_REG(&adapter->hw, GPRCH);
1609         adapter->stats.uprcl += IXGB_READ_REG(&adapter->hw, UPRCL);
1610         adapter->stats.uprch += IXGB_READ_REG(&adapter->hw, UPRCH);
1611         adapter->stats.vprcl += IXGB_READ_REG(&adapter->hw, VPRCL);
1612         adapter->stats.vprch += IXGB_READ_REG(&adapter->hw, VPRCH);
1613         adapter->stats.jprcl += IXGB_READ_REG(&adapter->hw, JPRCL);
1614         adapter->stats.jprch += IXGB_READ_REG(&adapter->hw, JPRCH);
1615         adapter->stats.gorcl += IXGB_READ_REG(&adapter->hw, GORCL);
1616         adapter->stats.gorch += IXGB_READ_REG(&adapter->hw, GORCH);
1617         adapter->stats.torl += IXGB_READ_REG(&adapter->hw, TORL);
1618         adapter->stats.torh += IXGB_READ_REG(&adapter->hw, TORH);
1619         adapter->stats.rnbc += IXGB_READ_REG(&adapter->hw, RNBC);
1620         adapter->stats.ruc += IXGB_READ_REG(&adapter->hw, RUC);
1621         adapter->stats.roc += IXGB_READ_REG(&adapter->hw, ROC);
1622         adapter->stats.rlec += IXGB_READ_REG(&adapter->hw, RLEC);
1623         adapter->stats.crcerrs += IXGB_READ_REG(&adapter->hw, CRCERRS);
1624         adapter->stats.icbc += IXGB_READ_REG(&adapter->hw, ICBC);
1625         adapter->stats.ecbc += IXGB_READ_REG(&adapter->hw, ECBC);
1626         adapter->stats.mpc += IXGB_READ_REG(&adapter->hw, MPC);
1627         adapter->stats.tptl += IXGB_READ_REG(&adapter->hw, TPTL);
1628         adapter->stats.tpth += IXGB_READ_REG(&adapter->hw, TPTH);
1629         adapter->stats.gptcl += IXGB_READ_REG(&adapter->hw, GPTCL);
1630         adapter->stats.gptch += IXGB_READ_REG(&adapter->hw, GPTCH);
1631         adapter->stats.bptcl += IXGB_READ_REG(&adapter->hw, BPTCL);
1632         adapter->stats.bptch += IXGB_READ_REG(&adapter->hw, BPTCH);
1633         adapter->stats.mptcl += IXGB_READ_REG(&adapter->hw, MPTCL);
1634         adapter->stats.mptch += IXGB_READ_REG(&adapter->hw, MPTCH);
1635         adapter->stats.uptcl += IXGB_READ_REG(&adapter->hw, UPTCL);
1636         adapter->stats.uptch += IXGB_READ_REG(&adapter->hw, UPTCH);
1637         adapter->stats.vptcl += IXGB_READ_REG(&adapter->hw, VPTCL);
1638         adapter->stats.vptch += IXGB_READ_REG(&adapter->hw, VPTCH);
1639         adapter->stats.jptcl += IXGB_READ_REG(&adapter->hw, JPTCL);
1640         adapter->stats.jptch += IXGB_READ_REG(&adapter->hw, JPTCH);
1641         adapter->stats.gotcl += IXGB_READ_REG(&adapter->hw, GOTCL);
1642         adapter->stats.gotch += IXGB_READ_REG(&adapter->hw, GOTCH);
1643         adapter->stats.totl += IXGB_READ_REG(&adapter->hw, TOTL);
1644         adapter->stats.toth += IXGB_READ_REG(&adapter->hw, TOTH);
1645         adapter->stats.dc += IXGB_READ_REG(&adapter->hw, DC);
1646         adapter->stats.plt64c += IXGB_READ_REG(&adapter->hw, PLT64C);
1647         adapter->stats.tsctc += IXGB_READ_REG(&adapter->hw, TSCTC);
1648         adapter->stats.tsctfc += IXGB_READ_REG(&adapter->hw, TSCTFC);
1649         adapter->stats.ibic += IXGB_READ_REG(&adapter->hw, IBIC);
1650         adapter->stats.rfc += IXGB_READ_REG(&adapter->hw, RFC);
1651         adapter->stats.lfc += IXGB_READ_REG(&adapter->hw, LFC);
1652         adapter->stats.pfrc += IXGB_READ_REG(&adapter->hw, PFRC);
1653         adapter->stats.pftc += IXGB_READ_REG(&adapter->hw, PFTC);
1654         adapter->stats.mcfrc += IXGB_READ_REG(&adapter->hw, MCFRC);
1655         adapter->stats.mcftc += IXGB_READ_REG(&adapter->hw, MCFTC);
1656         adapter->stats.xonrxc += IXGB_READ_REG(&adapter->hw, XONRXC);
1657         adapter->stats.xontxc += IXGB_READ_REG(&adapter->hw, XONTXC);
1658         adapter->stats.xoffrxc += IXGB_READ_REG(&adapter->hw, XOFFRXC);
1659         adapter->stats.xofftxc += IXGB_READ_REG(&adapter->hw, XOFFTXC);
1660         adapter->stats.rjc += IXGB_READ_REG(&adapter->hw, RJC);
1661
1662         /* Fill out the OS statistics structure */
1663
1664         adapter->net_stats.rx_packets = adapter->stats.gprcl;
1665         adapter->net_stats.tx_packets = adapter->stats.gptcl;
1666         adapter->net_stats.rx_bytes = adapter->stats.gorcl;
1667         adapter->net_stats.tx_bytes = adapter->stats.gotcl;
1668         adapter->net_stats.multicast = adapter->stats.mprcl;
1669         adapter->net_stats.collisions = 0;
1670
1671         /* ignore RLEC as it reports errors for padded (<64bytes) frames
1672          * with a length in the type/len field */
1673         adapter->net_stats.rx_errors =
1674             /* adapter->stats.rnbc + */ adapter->stats.crcerrs +
1675             adapter->stats.ruc +
1676             adapter->stats.roc /*+ adapter->stats.rlec */  +
1677             adapter->stats.icbc +
1678             adapter->stats.ecbc + adapter->stats.mpc;
1679
1680         /* see above
1681          * adapter->net_stats.rx_length_errors = adapter->stats.rlec;
1682          */
1683
1684         adapter->net_stats.rx_crc_errors = adapter->stats.crcerrs;
1685         adapter->net_stats.rx_fifo_errors = adapter->stats.mpc;
1686         adapter->net_stats.rx_missed_errors = adapter->stats.mpc;
1687         adapter->net_stats.rx_over_errors = adapter->stats.mpc;
1688
1689         adapter->net_stats.tx_errors = 0;
1690         adapter->net_stats.rx_frame_errors = 0;
1691         adapter->net_stats.tx_aborted_errors = 0;
1692         adapter->net_stats.tx_carrier_errors = 0;
1693         adapter->net_stats.tx_fifo_errors = 0;
1694         adapter->net_stats.tx_heartbeat_errors = 0;
1695         adapter->net_stats.tx_window_errors = 0;
1696 }
1697
1698 #define IXGB_MAX_INTR 10
1699 /**
1700  * ixgb_intr - Interrupt Handler
1701  * @irq: interrupt number
1702  * @data: pointer to a network interface device structure
1703  **/
1704
1705 static irqreturn_t
1706 ixgb_intr(int irq, void *data)
1707 {
1708         struct net_device *netdev = data;
1709         struct ixgb_adapter *adapter = netdev_priv(netdev);
1710         struct ixgb_hw *hw = &adapter->hw;
1711         u32 icr = IXGB_READ_REG(hw, ICR);
1712 #ifndef CONFIG_IXGB_NAPI
1713         unsigned int i;
1714 #endif
1715
1716         if (unlikely(!icr))
1717                 return IRQ_NONE;  /* Not our interrupt */
1718
1719         if (unlikely(icr & (IXGB_INT_RXSEQ | IXGB_INT_LSC)))
1720                 if (!test_bit(__IXGB_DOWN, &adapter->flags))
1721                         mod_timer(&adapter->watchdog_timer, jiffies);
1722
1723 #ifdef CONFIG_IXGB_NAPI
1724         if (netif_rx_schedule_prep(netdev, &adapter->napi)) {
1725
1726                 /* Disable interrupts and register for poll. The flush
1727                   of the posted write is intentionally left out.
1728                 */
1729
1730                 IXGB_WRITE_REG(&adapter->hw, IMC, ~0);
1731                 __netif_rx_schedule(netdev, &adapter->napi);
1732         }
1733 #else
1734         /* yes, that is actually a & and it is meant to make sure that
1735          * every pass through this for loop checks both receive and
1736          * transmit queues for completed descriptors, intended to
1737          * avoid starvation issues and assist tx/rx fairness. */
1738         for (i = 0; i < IXGB_MAX_INTR; i++)
1739                 if (!ixgb_clean_rx_irq(adapter) &
1740                    !ixgb_clean_tx_irq(adapter))
1741                         break;
1742 #endif
1743         return IRQ_HANDLED;
1744 }
1745
1746 #ifdef CONFIG_IXGB_NAPI
1747 /**
1748  * ixgb_clean - NAPI Rx polling callback
1749  * @adapter: board private structure
1750  **/
1751
1752 static int
1753 ixgb_clean(struct napi_struct *napi, int budget)
1754 {
1755         struct ixgb_adapter *adapter = container_of(napi, struct ixgb_adapter, napi);
1756         struct net_device *netdev = adapter->netdev;
1757         int work_done = 0;
1758
1759         ixgb_clean_tx_irq(adapter);
1760         ixgb_clean_rx_irq(adapter, &work_done, budget);
1761
1762         /* If budget not fully consumed, exit the polling mode */
1763         if (work_done < budget) {
1764                 netif_rx_complete(netdev, napi);
1765                 if (!test_bit(__IXGB_DOWN, &adapter->flags))
1766                         ixgb_irq_enable(adapter);
1767         }
1768
1769         return work_done;
1770 }
1771 #endif
1772
1773 /**
1774  * ixgb_clean_tx_irq - Reclaim resources after transmit completes
1775  * @adapter: board private structure
1776  **/
1777
1778 static bool
1779 ixgb_clean_tx_irq(struct ixgb_adapter *adapter)
1780 {
1781         struct ixgb_desc_ring *tx_ring = &adapter->tx_ring;
1782         struct net_device *netdev = adapter->netdev;
1783         struct ixgb_tx_desc *tx_desc, *eop_desc;
1784         struct ixgb_buffer *buffer_info;
1785         unsigned int i, eop;
1786         bool cleaned = false;
1787
1788         i = tx_ring->next_to_clean;
1789         eop = tx_ring->buffer_info[i].next_to_watch;
1790         eop_desc = IXGB_TX_DESC(*tx_ring, eop);
1791
1792         while (eop_desc->status & IXGB_TX_DESC_STATUS_DD) {
1793
1794                 for (cleaned = false; !cleaned; ) {
1795                         tx_desc = IXGB_TX_DESC(*tx_ring, i);
1796                         buffer_info = &tx_ring->buffer_info[i];
1797
1798                         if (tx_desc->popts &
1799                            (IXGB_TX_DESC_POPTS_TXSM |
1800                             IXGB_TX_DESC_POPTS_IXSM))
1801                                 adapter->hw_csum_tx_good++;
1802
1803                         ixgb_unmap_and_free_tx_resource(adapter, buffer_info);
1804
1805                         *(u32 *)&(tx_desc->status) = 0;
1806
1807                         cleaned = (i == eop);
1808                         if (++i == tx_ring->count) i = 0;
1809                 }
1810
1811                 eop = tx_ring->buffer_info[i].next_to_watch;
1812                 eop_desc = IXGB_TX_DESC(*tx_ring, eop);
1813         }
1814
1815         tx_ring->next_to_clean = i;
1816
1817         if (unlikely(cleaned && netif_carrier_ok(netdev) &&
1818                      IXGB_DESC_UNUSED(tx_ring) >= DESC_NEEDED)) {
1819                 /* Make sure that anybody stopping the queue after this
1820                  * sees the new next_to_clean. */
1821                 smp_mb();
1822
1823                 if (netif_queue_stopped(netdev) &&
1824                     !(test_bit(__IXGB_DOWN, &adapter->flags))) {
1825                         netif_wake_queue(netdev);
1826                         ++adapter->restart_queue;
1827                 }
1828         }
1829
1830         if (adapter->detect_tx_hung) {
1831                 /* detect a transmit hang in hardware, this serializes the
1832                  * check with the clearing of time_stamp and movement of i */
1833                 adapter->detect_tx_hung = false;
1834                 if (tx_ring->buffer_info[eop].dma &&
1835                    time_after(jiffies, tx_ring->buffer_info[eop].time_stamp + HZ)
1836                    && !(IXGB_READ_REG(&adapter->hw, STATUS) &
1837                         IXGB_STATUS_TXOFF)) {
1838                         /* detected Tx unit hang */
1839                         DPRINTK(DRV, ERR, "Detected Tx Unit Hang\n"
1840                                         "  TDH                  <%x>\n"
1841                                         "  TDT                  <%x>\n"
1842                                         "  next_to_use          <%x>\n"
1843                                         "  next_to_clean        <%x>\n"
1844                                         "buffer_info[next_to_clean]\n"
1845                                         "  time_stamp           <%lx>\n"
1846                                         "  next_to_watch        <%x>\n"
1847                                         "  jiffies              <%lx>\n"
1848                                         "  next_to_watch.status <%x>\n",
1849                                 IXGB_READ_REG(&adapter->hw, TDH),
1850                                 IXGB_READ_REG(&adapter->hw, TDT),
1851                                 tx_ring->next_to_use,
1852                                 tx_ring->next_to_clean,
1853                                 tx_ring->buffer_info[eop].time_stamp,
1854                                 eop,
1855                                 jiffies,
1856                                 eop_desc->status);
1857                         netif_stop_queue(netdev);
1858                 }
1859         }
1860
1861         return cleaned;
1862 }
1863
1864 /**
1865  * ixgb_rx_checksum - Receive Checksum Offload for 82597.
1866  * @adapter: board private structure
1867  * @rx_desc: receive descriptor
1868  * @sk_buff: socket buffer with received data
1869  **/
1870
1871 static void
1872 ixgb_rx_checksum(struct ixgb_adapter *adapter,
1873                  struct ixgb_rx_desc *rx_desc,
1874                  struct sk_buff *skb)
1875 {
1876         /* Ignore Checksum bit is set OR
1877          * TCP Checksum has not been calculated
1878          */
1879         if ((rx_desc->status & IXGB_RX_DESC_STATUS_IXSM) ||
1880            (!(rx_desc->status & IXGB_RX_DESC_STATUS_TCPCS))) {
1881                 skb->ip_summed = CHECKSUM_NONE;
1882                 return;
1883         }
1884
1885         /* At this point we know the hardware did the TCP checksum */
1886         /* now look at the TCP checksum error bit */
1887         if (rx_desc->errors & IXGB_RX_DESC_ERRORS_TCPE) {
1888                 /* let the stack verify checksum errors */
1889                 skb->ip_summed = CHECKSUM_NONE;
1890                 adapter->hw_csum_rx_error++;
1891         } else {
1892                 /* TCP checksum is good */
1893                 skb->ip_summed = CHECKSUM_UNNECESSARY;
1894                 adapter->hw_csum_rx_good++;
1895         }
1896 }
1897
1898 /**
1899  * ixgb_clean_rx_irq - Send received data up the network stack,
1900  * @adapter: board private structure
1901  **/
1902
1903 static bool
1904 #ifdef CONFIG_IXGB_NAPI
1905 ixgb_clean_rx_irq(struct ixgb_adapter *adapter, int *work_done, int work_to_do)
1906 #else
1907 ixgb_clean_rx_irq(struct ixgb_adapter *adapter)
1908 #endif
1909 {
1910         struct ixgb_desc_ring *rx_ring = &adapter->rx_ring;
1911         struct net_device *netdev = adapter->netdev;
1912         struct pci_dev *pdev = adapter->pdev;
1913         struct ixgb_rx_desc *rx_desc, *next_rxd;
1914         struct ixgb_buffer *buffer_info, *next_buffer, *next2_buffer;
1915         u32 length;
1916         unsigned int i, j;
1917         int cleaned_count = 0;
1918         bool cleaned = false;
1919
1920         i = rx_ring->next_to_clean;
1921         rx_desc = IXGB_RX_DESC(*rx_ring, i);
1922         buffer_info = &rx_ring->buffer_info[i];
1923
1924         while (rx_desc->status & IXGB_RX_DESC_STATUS_DD) {
1925                 struct sk_buff *skb;
1926                 u8 status;
1927
1928 #ifdef CONFIG_IXGB_NAPI
1929                 if (*work_done >= work_to_do)
1930                         break;
1931
1932                 (*work_done)++;
1933 #endif
1934                 status = rx_desc->status;
1935                 skb = buffer_info->skb;
1936                 buffer_info->skb = NULL;
1937
1938                 prefetch(skb->data - NET_IP_ALIGN);
1939
1940                 if (++i == rx_ring->count) i = 0;
1941                 next_rxd = IXGB_RX_DESC(*rx_ring, i);
1942                 prefetch(next_rxd);
1943
1944                 if ((j = i + 1) == rx_ring->count) j = 0;
1945                 next2_buffer = &rx_ring->buffer_info[j];
1946                 prefetch(next2_buffer);
1947
1948                 next_buffer = &rx_ring->buffer_info[i];
1949
1950                 cleaned = true;
1951                 cleaned_count++;
1952
1953                 pci_unmap_single(pdev,
1954                                  buffer_info->dma,
1955                                  buffer_info->length,
1956                                  PCI_DMA_FROMDEVICE);
1957                 buffer_info->dma = 0;
1958
1959                 length = le16_to_cpu(rx_desc->length);
1960                 rx_desc->length = 0;
1961
1962                 if (unlikely(!(status & IXGB_RX_DESC_STATUS_EOP))) {
1963
1964                         /* All receives must fit into a single buffer */
1965
1966                         IXGB_DBG("Receive packet consumed multiple buffers "
1967                                          "length<%x>\n", length);
1968
1969                         dev_kfree_skb_irq(skb);
1970                         goto rxdesc_done;
1971                 }
1972
1973                 if (unlikely(rx_desc->errors &
1974                     (IXGB_RX_DESC_ERRORS_CE | IXGB_RX_DESC_ERRORS_SE |
1975                      IXGB_RX_DESC_ERRORS_P | IXGB_RX_DESC_ERRORS_RXE))) {
1976                         dev_kfree_skb_irq(skb);
1977                         goto rxdesc_done;
1978                 }
1979
1980                 /* code added for copybreak, this should improve
1981                  * performance for small packets with large amounts
1982                  * of reassembly being done in the stack */
1983                 if (length < copybreak) {
1984                         struct sk_buff *new_skb =
1985                             netdev_alloc_skb(netdev, length + NET_IP_ALIGN);
1986                         if (new_skb) {
1987                                 skb_reserve(new_skb, NET_IP_ALIGN);
1988                                 skb_copy_to_linear_data_offset(new_skb,
1989                                                                -NET_IP_ALIGN,
1990                                                                (skb->data -
1991                                                                 NET_IP_ALIGN),
1992                                                                (length +
1993                                                                 NET_IP_ALIGN));
1994                                 /* save the skb in buffer_info as good */
1995                                 buffer_info->skb = skb;
1996                                 skb = new_skb;
1997                         }
1998                 }
1999                 /* end copybreak code */
2000
2001                 /* Good Receive */
2002                 skb_put(skb, length);
2003
2004                 /* Receive Checksum Offload */
2005                 ixgb_rx_checksum(adapter, rx_desc, skb);
2006
2007                 skb->protocol = eth_type_trans(skb, netdev);
2008 #ifdef CONFIG_IXGB_NAPI
2009                 if (adapter->vlgrp && (status & IXGB_RX_DESC_STATUS_VP)) {
2010                         vlan_hwaccel_receive_skb(skb, adapter->vlgrp,
2011                                                 le16_to_cpu(rx_desc->special));
2012                 } else {
2013                         netif_receive_skb(skb);
2014                 }
2015 #else /* CONFIG_IXGB_NAPI */
2016                 if (adapter->vlgrp && (status & IXGB_RX_DESC_STATUS_VP)) {
2017                         vlan_hwaccel_rx(skb, adapter->vlgrp,
2018                                         le16_to_cpu(rx_desc->special));
2019                 } else {
2020                         netif_rx(skb);
2021                 }
2022 #endif /* CONFIG_IXGB_NAPI */
2023                 netdev->last_rx = jiffies;
2024
2025 rxdesc_done:
2026                 /* clean up descriptor, might be written over by hw */
2027                 rx_desc->status = 0;
2028
2029                 /* return some buffers to hardware, one at a time is too slow */
2030                 if (unlikely(cleaned_count >= IXGB_RX_BUFFER_WRITE)) {
2031                         ixgb_alloc_rx_buffers(adapter, cleaned_count);
2032                         cleaned_count = 0;
2033                 }
2034
2035                 /* use prefetched values */
2036                 rx_desc = next_rxd;
2037                 buffer_info = next_buffer;
2038         }
2039
2040         rx_ring->next_to_clean = i;
2041
2042         cleaned_count = IXGB_DESC_UNUSED(rx_ring);
2043         if (cleaned_count)
2044                 ixgb_alloc_rx_buffers(adapter, cleaned_count);
2045
2046         return cleaned;
2047 }
2048
2049 /**
2050  * ixgb_alloc_rx_buffers - Replace used receive buffers
2051  * @adapter: address of board private structure
2052  **/
2053
2054 static void
2055 ixgb_alloc_rx_buffers(struct ixgb_adapter *adapter, int cleaned_count)
2056 {
2057         struct ixgb_desc_ring *rx_ring = &adapter->rx_ring;
2058         struct net_device *netdev = adapter->netdev;
2059         struct pci_dev *pdev = adapter->pdev;
2060         struct ixgb_rx_desc *rx_desc;
2061         struct ixgb_buffer *buffer_info;
2062         struct sk_buff *skb;
2063         unsigned int i;
2064         long cleancount;
2065
2066         i = rx_ring->next_to_use;
2067         buffer_info = &rx_ring->buffer_info[i];
2068         cleancount = IXGB_DESC_UNUSED(rx_ring);
2069
2070
2071         /* leave three descriptors unused */
2072         while (--cleancount > 2 && cleaned_count--) {
2073                 /* recycle! its good for you */
2074                 skb = buffer_info->skb;
2075                 if (skb) {
2076                         skb_trim(skb, 0);
2077                         goto map_skb;
2078                 }
2079
2080                 skb = netdev_alloc_skb(netdev, adapter->rx_buffer_len
2081                                        + NET_IP_ALIGN);
2082                 if (unlikely(!skb)) {
2083                         /* Better luck next round */
2084                         adapter->alloc_rx_buff_failed++;
2085                         break;
2086                 }
2087
2088                 /* Make buffer alignment 2 beyond a 16 byte boundary
2089                  * this will result in a 16 byte aligned IP header after
2090                  * the 14 byte MAC header is removed
2091                  */
2092                 skb_reserve(skb, NET_IP_ALIGN);
2093
2094                 buffer_info->skb = skb;
2095                 buffer_info->length = adapter->rx_buffer_len;
2096 map_skb:
2097                 buffer_info->dma = pci_map_single(pdev,
2098                                                   skb->data,
2099                                                   adapter->rx_buffer_len,
2100                                                   PCI_DMA_FROMDEVICE);
2101
2102                 rx_desc = IXGB_RX_DESC(*rx_ring, i);
2103                 rx_desc->buff_addr = cpu_to_le64(buffer_info->dma);
2104                 /* guarantee DD bit not set now before h/w gets descriptor
2105                  * this is the rest of the workaround for h/w double
2106                  * writeback. */
2107                 rx_desc->status = 0;
2108
2109
2110                 if (++i == rx_ring->count) i = 0;
2111                 buffer_info = &rx_ring->buffer_info[i];
2112         }
2113
2114         if (likely(rx_ring->next_to_use != i)) {
2115                 rx_ring->next_to_use = i;
2116                 if (unlikely(i-- == 0))
2117                         i = (rx_ring->count - 1);
2118
2119                 /* Force memory writes to complete before letting h/w
2120                  * know there are new descriptors to fetch.  (Only
2121                  * applicable for weak-ordered memory model archs, such
2122                  * as IA-64). */
2123                 wmb();
2124                 IXGB_WRITE_REG(&adapter->hw, RDT, i);
2125         }
2126 }
2127
2128 /**
2129  * ixgb_vlan_rx_register - enables or disables vlan tagging/stripping.
2130  *
2131  * @param netdev network interface device structure
2132  * @param grp indicates to enable or disable tagging/stripping
2133  **/
2134 static void
2135 ixgb_vlan_rx_register(struct net_device *netdev, struct vlan_group *grp)
2136 {
2137         struct ixgb_adapter *adapter = netdev_priv(netdev);
2138         u32 ctrl, rctl;
2139
2140         ixgb_irq_disable(adapter);
2141         adapter->vlgrp = grp;
2142
2143         if (grp) {
2144                 /* enable VLAN tag insert/strip */
2145                 ctrl = IXGB_READ_REG(&adapter->hw, CTRL0);
2146                 ctrl |= IXGB_CTRL0_VME;
2147                 IXGB_WRITE_REG(&adapter->hw, CTRL0, ctrl);
2148
2149                 /* enable VLAN receive filtering */
2150
2151                 rctl = IXGB_READ_REG(&adapter->hw, RCTL);
2152                 rctl |= IXGB_RCTL_VFE;
2153                 rctl &= ~IXGB_RCTL_CFIEN;
2154                 IXGB_WRITE_REG(&adapter->hw, RCTL, rctl);
2155         } else {
2156                 /* disable VLAN tag insert/strip */
2157
2158                 ctrl = IXGB_READ_REG(&adapter->hw, CTRL0);
2159                 ctrl &= ~IXGB_CTRL0_VME;
2160                 IXGB_WRITE_REG(&adapter->hw, CTRL0, ctrl);
2161
2162                 /* disable VLAN filtering */
2163
2164                 rctl = IXGB_READ_REG(&adapter->hw, RCTL);
2165                 rctl &= ~IXGB_RCTL_VFE;
2166                 IXGB_WRITE_REG(&adapter->hw, RCTL, rctl);
2167         }
2168
2169         /* don't enable interrupts unless we are UP */
2170         if (adapter->netdev->flags & IFF_UP)
2171                 ixgb_irq_enable(adapter);
2172 }
2173
2174 static void
2175 ixgb_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
2176 {
2177         struct ixgb_adapter *adapter = netdev_priv(netdev);
2178         u32 vfta, index;
2179
2180         /* add VID to filter table */
2181
2182         index = (vid >> 5) & 0x7F;
2183         vfta = IXGB_READ_REG_ARRAY(&adapter->hw, VFTA, index);
2184         vfta |= (1 << (vid & 0x1F));
2185         ixgb_write_vfta(&adapter->hw, index, vfta);
2186 }
2187
2188 static void
2189 ixgb_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
2190 {
2191         struct ixgb_adapter *adapter = netdev_priv(netdev);
2192         u32 vfta, index;
2193
2194         ixgb_irq_disable(adapter);
2195
2196         vlan_group_set_device(adapter->vlgrp, vid, NULL);
2197
2198         /* don't enable interrupts unless we are UP */
2199         if (adapter->netdev->flags & IFF_UP)
2200                 ixgb_irq_enable(adapter);
2201
2202         /* remove VID from filter table */
2203
2204         index = (vid >> 5) & 0x7F;
2205         vfta = IXGB_READ_REG_ARRAY(&adapter->hw, VFTA, index);
2206         vfta &= ~(1 << (vid & 0x1F));
2207         ixgb_write_vfta(&adapter->hw, index, vfta);
2208 }
2209
2210 static void
2211 ixgb_restore_vlan(struct ixgb_adapter *adapter)
2212 {
2213         ixgb_vlan_rx_register(adapter->netdev, adapter->vlgrp);
2214
2215         if (adapter->vlgrp) {
2216                 u16 vid;
2217                 for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
2218                         if (!vlan_group_get_device(adapter->vlgrp, vid))
2219                                 continue;
2220                         ixgb_vlan_rx_add_vid(adapter->netdev, vid);
2221                 }
2222         }
2223 }
2224
2225 #ifdef CONFIG_NET_POLL_CONTROLLER
2226 /*
2227  * Polling 'interrupt' - used by things like netconsole to send skbs
2228  * without having to re-enable interrupts. It's not called while
2229  * the interrupt routine is executing.
2230  */
2231
2232 static void ixgb_netpoll(struct net_device *dev)
2233 {
2234         struct ixgb_adapter *adapter = netdev_priv(dev);
2235
2236         disable_irq(adapter->pdev->irq);
2237         ixgb_intr(adapter->pdev->irq, dev);
2238         enable_irq(adapter->pdev->irq);
2239 }
2240 #endif
2241
2242 /**
2243  * ixgb_io_error_detected() - called when PCI error is detected
2244  * @pdev    pointer to pci device with error
2245  * @state   pci channel state after error
2246  *
2247  * This callback is called by the PCI subsystem whenever
2248  * a PCI bus error is detected.
2249  */
2250 static pci_ers_result_t ixgb_io_error_detected(struct pci_dev *pdev,
2251                                                enum pci_channel_state state)
2252 {
2253         struct net_device *netdev = pci_get_drvdata(pdev);
2254         struct ixgb_adapter *adapter = netdev_priv(netdev);
2255
2256         if (netif_running(netdev))
2257                 ixgb_down(adapter, true);
2258
2259         pci_disable_device(pdev);
2260
2261         /* Request a slot reset. */
2262         return PCI_ERS_RESULT_NEED_RESET;
2263 }
2264
2265 /**
2266  * ixgb_io_slot_reset - called after the pci bus has been reset.
2267  * @pdev    pointer to pci device with error
2268  *
2269  * This callback is called after the PCI bus has been reset.
2270  * Basically, this tries to restart the card from scratch.
2271  * This is a shortened version of the device probe/discovery code,
2272  * it resembles the first-half of the ixgb_probe() routine.
2273  */
2274 static pci_ers_result_t ixgb_io_slot_reset(struct pci_dev *pdev)
2275 {
2276         struct net_device *netdev = pci_get_drvdata(pdev);
2277         struct ixgb_adapter *adapter = netdev_priv(netdev);
2278
2279         if (pci_enable_device(pdev)) {
2280                 DPRINTK(PROBE, ERR, "Cannot re-enable PCI device after reset.\n");
2281                 return PCI_ERS_RESULT_DISCONNECT;
2282         }
2283
2284         /* Perform card reset only on one instance of the card */
2285         if (0 != PCI_FUNC (pdev->devfn))
2286                 return PCI_ERS_RESULT_RECOVERED;
2287
2288         pci_set_master(pdev);
2289
2290         netif_carrier_off(netdev);
2291         netif_stop_queue(netdev);
2292         ixgb_reset(adapter);
2293
2294         /* Make sure the EEPROM is good */
2295         if (!ixgb_validate_eeprom_checksum(&adapter->hw)) {
2296                 DPRINTK(PROBE, ERR, "After reset, the EEPROM checksum is not valid.\n");
2297                 return PCI_ERS_RESULT_DISCONNECT;
2298         }
2299         ixgb_get_ee_mac_addr(&adapter->hw, netdev->dev_addr);
2300         memcpy(netdev->perm_addr, netdev->dev_addr, netdev->addr_len);
2301
2302         if (!is_valid_ether_addr(netdev->perm_addr)) {
2303                 DPRINTK(PROBE, ERR, "After reset, invalid MAC address.\n");
2304                 return PCI_ERS_RESULT_DISCONNECT;
2305         }
2306
2307         return PCI_ERS_RESULT_RECOVERED;
2308 }
2309
2310 /**
2311  * ixgb_io_resume - called when its OK to resume normal operations
2312  * @pdev    pointer to pci device with error
2313  *
2314  * The error recovery driver tells us that its OK to resume
2315  * normal operation. Implementation resembles the second-half
2316  * of the ixgb_probe() routine.
2317  */
2318 static void ixgb_io_resume(struct pci_dev *pdev)
2319 {
2320         struct net_device *netdev = pci_get_drvdata(pdev);
2321         struct ixgb_adapter *adapter = netdev_priv(netdev);
2322
2323         pci_set_master(pdev);
2324
2325         if (netif_running(netdev)) {
2326                 if (ixgb_up(adapter)) {
2327                         printk ("ixgb: can't bring device back up after reset\n");
2328                         return;
2329                 }
2330         }
2331
2332         netif_device_attach(netdev);
2333         mod_timer(&adapter->watchdog_timer, jiffies);
2334 }
2335
2336 /* ixgb_main.c */