Merge git://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi-misc-2.6
[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-k2"DRIVERNAPI
40 const char ixgb_driver_version[] = DRV_VERSION;
41 static const char ixgb_copyright[] = "Copyright (c) 1999-2006 Intel Corporation.";
42
43 /* ixgb_pci_tbl - PCI Device ID Table
44  *
45  * Wildcard entries (PCI_ANY_ID) should come last
46  * Last entry must be all 0s
47  *
48  * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
49  *   Class, Class Mask, private data (not used) }
50  */
51 static struct pci_device_id ixgb_pci_tbl[] = {
52         {INTEL_VENDOR_ID, IXGB_DEVICE_ID_82597EX,
53          PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0},
54         {INTEL_VENDOR_ID, IXGB_DEVICE_ID_82597EX_CX4,
55          PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0},
56         {INTEL_VENDOR_ID, IXGB_DEVICE_ID_82597EX_SR,
57          PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0},
58         {INTEL_VENDOR_ID, IXGB_DEVICE_ID_82597EX_LR,  
59          PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0},
60
61         /* required last entry */
62         {0,}
63 };
64
65 MODULE_DEVICE_TABLE(pci, ixgb_pci_tbl);
66
67 /* Local Function Prototypes */
68
69 int ixgb_up(struct ixgb_adapter *adapter);
70 void ixgb_down(struct ixgb_adapter *adapter, boolean_t kill_watchdog);
71 void ixgb_reset(struct ixgb_adapter *adapter);
72 int ixgb_setup_tx_resources(struct ixgb_adapter *adapter);
73 int ixgb_setup_rx_resources(struct ixgb_adapter *adapter);
74 void ixgb_free_tx_resources(struct ixgb_adapter *adapter);
75 void ixgb_free_rx_resources(struct ixgb_adapter *adapter);
76 void ixgb_update_stats(struct ixgb_adapter *adapter);
77
78 static int ixgb_init_module(void);
79 static void ixgb_exit_module(void);
80 static int ixgb_probe(struct pci_dev *pdev, const struct pci_device_id *ent);
81 static void __devexit ixgb_remove(struct pci_dev *pdev);
82 static int ixgb_sw_init(struct ixgb_adapter *adapter);
83 static int ixgb_open(struct net_device *netdev);
84 static int ixgb_close(struct net_device *netdev);
85 static void ixgb_configure_tx(struct ixgb_adapter *adapter);
86 static void ixgb_configure_rx(struct ixgb_adapter *adapter);
87 static void ixgb_setup_rctl(struct ixgb_adapter *adapter);
88 static void ixgb_clean_tx_ring(struct ixgb_adapter *adapter);
89 static void ixgb_clean_rx_ring(struct ixgb_adapter *adapter);
90 static void ixgb_set_multi(struct net_device *netdev);
91 static void ixgb_watchdog(unsigned long data);
92 static int ixgb_xmit_frame(struct sk_buff *skb, struct net_device *netdev);
93 static struct net_device_stats *ixgb_get_stats(struct net_device *netdev);
94 static int ixgb_change_mtu(struct net_device *netdev, int new_mtu);
95 static int ixgb_set_mac(struct net_device *netdev, void *p);
96 static irqreturn_t ixgb_intr(int irq, void *data);
97 static boolean_t ixgb_clean_tx_irq(struct ixgb_adapter *adapter);
98
99 #ifdef CONFIG_IXGB_NAPI
100 static int ixgb_clean(struct napi_struct *napi, int budget);
101 static boolean_t ixgb_clean_rx_irq(struct ixgb_adapter *adapter,
102                                    int *work_done, int work_to_do);
103 #else
104 static boolean_t ixgb_clean_rx_irq(struct ixgb_adapter *adapter);
105 #endif
106 static void ixgb_alloc_rx_buffers(struct ixgb_adapter *adapter);
107 static void ixgb_tx_timeout(struct net_device *dev);
108 static void ixgb_tx_timeout_task(struct work_struct *work);
109 static void ixgb_vlan_rx_register(struct net_device *netdev,
110                                   struct vlan_group *grp);
111 static void ixgb_vlan_rx_add_vid(struct net_device *netdev, uint16_t vid);
112 static void ixgb_vlan_rx_kill_vid(struct net_device *netdev, uint16_t vid);
113 static void ixgb_restore_vlan(struct ixgb_adapter *adapter);
114
115 #ifdef CONFIG_NET_POLL_CONTROLLER
116 /* for netdump / net console */
117 static void ixgb_netpoll(struct net_device *dev);
118 #endif
119
120 static pci_ers_result_t ixgb_io_error_detected (struct pci_dev *pdev,
121                              enum pci_channel_state state);
122 static pci_ers_result_t ixgb_io_slot_reset (struct pci_dev *pdev);
123 static void ixgb_io_resume (struct pci_dev *pdev);
124
125 static struct pci_error_handlers ixgb_err_handler = {
126         .error_detected = ixgb_io_error_detected,
127         .slot_reset = ixgb_io_slot_reset,
128         .resume = ixgb_io_resume,
129 };
130
131 static struct pci_driver ixgb_driver = {
132         .name     = ixgb_driver_name,
133         .id_table = ixgb_pci_tbl,
134         .probe    = ixgb_probe,
135         .remove   = __devexit_p(ixgb_remove),
136         .err_handler = &ixgb_err_handler
137 };
138
139 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
140 MODULE_DESCRIPTION("Intel(R) PRO/10GbE Network Driver");
141 MODULE_LICENSE("GPL");
142 MODULE_VERSION(DRV_VERSION);
143
144 #define DEFAULT_DEBUG_LEVEL_SHIFT 3
145 static int debug = DEFAULT_DEBUG_LEVEL_SHIFT;
146 module_param(debug, int, 0);
147 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
148
149 /* some defines for controlling descriptor fetches in h/w */
150 #define RXDCTL_WTHRESH_DEFAULT 15  /* chip writes back at this many or RXT0 */
151 #define RXDCTL_PTHRESH_DEFAULT 0   /* chip considers prefech below
152                                     * this */
153 #define RXDCTL_HTHRESH_DEFAULT 0   /* chip will only prefetch if tail
154                                     * is pushed this many descriptors
155                                     * from head */
156
157 /**
158  * ixgb_init_module - Driver Registration Routine
159  *
160  * ixgb_init_module is the first routine called when the driver is
161  * loaded. All it does is register with the PCI subsystem.
162  **/
163
164 static int __init
165 ixgb_init_module(void)
166 {
167         printk(KERN_INFO "%s - version %s\n",
168                ixgb_driver_string, ixgb_driver_version);
169
170         printk(KERN_INFO "%s\n", ixgb_copyright);
171
172         return pci_register_driver(&ixgb_driver);
173 }
174
175 module_init(ixgb_init_module);
176
177 /**
178  * ixgb_exit_module - Driver Exit Cleanup Routine
179  *
180  * ixgb_exit_module is called just before the driver is removed
181  * from memory.
182  **/
183
184 static void __exit
185 ixgb_exit_module(void)
186 {
187         pci_unregister_driver(&ixgb_driver);
188 }
189
190 module_exit(ixgb_exit_module);
191
192 /**
193  * ixgb_irq_disable - Mask off interrupt generation on the NIC
194  * @adapter: board private structure
195  **/
196
197 static void
198 ixgb_irq_disable(struct ixgb_adapter *adapter)
199 {
200         atomic_inc(&adapter->irq_sem);
201         IXGB_WRITE_REG(&adapter->hw, IMC, ~0);
202         IXGB_WRITE_FLUSH(&adapter->hw);
203         synchronize_irq(adapter->pdev->irq);
204 }
205
206 /**
207  * ixgb_irq_enable - Enable default interrupt generation settings
208  * @adapter: board private structure
209  **/
210
211 static void
212 ixgb_irq_enable(struct ixgb_adapter *adapter)
213 {
214         if(atomic_dec_and_test(&adapter->irq_sem)) {
215                 IXGB_WRITE_REG(&adapter->hw, IMS,
216                                IXGB_INT_RXT0 | IXGB_INT_RXDMT0 | IXGB_INT_TXDW |
217                                IXGB_INT_LSC);
218                 IXGB_WRITE_FLUSH(&adapter->hw);
219         }
220 }
221
222 int
223 ixgb_up(struct ixgb_adapter *adapter)
224 {
225         struct net_device *netdev = adapter->netdev;
226         int err, irq_flags = IRQF_SHARED;
227         int max_frame = netdev->mtu + ENET_HEADER_SIZE + ENET_FCS_LENGTH;
228         struct ixgb_hw *hw = &adapter->hw;
229
230         /* hardware has been reset, we need to reload some things */
231
232         ixgb_rar_set(hw, netdev->dev_addr, 0);
233         ixgb_set_multi(netdev);
234
235         ixgb_restore_vlan(adapter);
236
237         ixgb_configure_tx(adapter);
238         ixgb_setup_rctl(adapter);
239         ixgb_configure_rx(adapter);
240         ixgb_alloc_rx_buffers(adapter);
241
242         /* disable interrupts and get the hardware into a known state */
243         IXGB_WRITE_REG(&adapter->hw, IMC, 0xffffffff);
244
245         /* only enable MSI if bus is in PCI-X mode */
246         if (IXGB_READ_REG(&adapter->hw, STATUS) & IXGB_STATUS_PCIX_MODE) {
247                 err = pci_enable_msi(adapter->pdev);
248                 if (!err) {
249                         adapter->have_msi = 1;
250                         irq_flags = 0;
251                 }
252                 /* proceed to try to request regular interrupt */
253         }
254
255         err = request_irq(adapter->pdev->irq, &ixgb_intr, irq_flags,
256                           netdev->name, netdev);
257         if (err) {
258                 if (adapter->have_msi)
259                         pci_disable_msi(adapter->pdev);
260                 DPRINTK(PROBE, ERR,
261                  "Unable to allocate interrupt Error: %d\n", err);
262                 return err;
263         }
264
265         if((hw->max_frame_size != max_frame) ||
266                 (hw->max_frame_size !=
267                 (IXGB_READ_REG(hw, MFS) >> IXGB_MFS_SHIFT))) {
268
269                 hw->max_frame_size = max_frame;
270
271                 IXGB_WRITE_REG(hw, MFS, hw->max_frame_size << IXGB_MFS_SHIFT);
272
273                 if(hw->max_frame_size >
274                    IXGB_MAX_ENET_FRAME_SIZE_WITHOUT_FCS + ENET_FCS_LENGTH) {
275                         uint32_t ctrl0 = IXGB_READ_REG(hw, CTRL0);
276
277                         if(!(ctrl0 & IXGB_CTRL0_JFE)) {
278                                 ctrl0 |= IXGB_CTRL0_JFE;
279                                 IXGB_WRITE_REG(hw, CTRL0, ctrl0);
280                         }
281                 }
282         }
283
284         mod_timer(&adapter->watchdog_timer, jiffies);
285
286 #ifdef CONFIG_IXGB_NAPI
287         napi_enable(&adapter->napi);
288 #endif
289         ixgb_irq_enable(adapter);
290
291         return 0;
292 }
293
294 void
295 ixgb_down(struct ixgb_adapter *adapter, boolean_t kill_watchdog)
296 {
297         struct net_device *netdev = adapter->netdev;
298
299 #ifdef CONFIG_IXGB_NAPI
300         napi_disable(&adapter->napi);
301         atomic_set(&adapter->irq_sem, 0);
302 #endif
303
304         ixgb_irq_disable(adapter);
305         free_irq(adapter->pdev->irq, netdev);
306
307         if (adapter->have_msi)
308                 pci_disable_msi(adapter->pdev);
309
310         if(kill_watchdog)
311                 del_timer_sync(&adapter->watchdog_timer);
312
313         adapter->link_speed = 0;
314         adapter->link_duplex = 0;
315         netif_carrier_off(netdev);
316         netif_stop_queue(netdev);
317
318         ixgb_reset(adapter);
319         ixgb_clean_tx_ring(adapter);
320         ixgb_clean_rx_ring(adapter);
321 }
322
323 void
324 ixgb_reset(struct ixgb_adapter *adapter)
325 {
326         struct ixgb_hw *hw = &adapter->hw;
327
328         ixgb_adapter_stop(hw);
329         if (!ixgb_init_hw(hw))
330                 DPRINTK(PROBE, ERR, "ixgb_init_hw failed.\n");
331
332         /* restore frame size information */
333         IXGB_WRITE_REG(hw, MFS, hw->max_frame_size << IXGB_MFS_SHIFT);
334         if (hw->max_frame_size >
335             IXGB_MAX_ENET_FRAME_SIZE_WITHOUT_FCS + ENET_FCS_LENGTH) {
336                 u32 ctrl0 = IXGB_READ_REG(hw, CTRL0);
337                 if (!(ctrl0 & IXGB_CTRL0_JFE)) {
338                         ctrl0 |= IXGB_CTRL0_JFE;
339                         IXGB_WRITE_REG(hw, CTRL0, ctrl0);
340                 }
341         }
342 }
343
344 /**
345  * ixgb_probe - Device Initialization Routine
346  * @pdev: PCI device information struct
347  * @ent: entry in ixgb_pci_tbl
348  *
349  * Returns 0 on success, negative on failure
350  *
351  * ixgb_probe initializes an adapter identified by a pci_dev structure.
352  * The OS initialization, configuring of the adapter private structure,
353  * and a hardware reset occur.
354  **/
355
356 static int __devinit
357 ixgb_probe(struct pci_dev *pdev,
358                 const struct pci_device_id *ent)
359 {
360         struct net_device *netdev = NULL;
361         struct ixgb_adapter *adapter;
362         static int cards_found = 0;
363         unsigned long mmio_start;
364         int mmio_len;
365         int pci_using_dac;
366         int i;
367         int err;
368
369         if((err = pci_enable_device(pdev)))
370                 return err;
371
372         if(!(err = pci_set_dma_mask(pdev, DMA_64BIT_MASK)) &&
373            !(err = pci_set_consistent_dma_mask(pdev, DMA_64BIT_MASK))) {
374                 pci_using_dac = 1;
375         } else {
376                 if((err = pci_set_dma_mask(pdev, DMA_32BIT_MASK)) ||
377                    (err = pci_set_consistent_dma_mask(pdev, DMA_32BIT_MASK))) {
378                         printk(KERN_ERR
379                          "ixgb: No usable DMA configuration, aborting\n");
380                         goto err_dma_mask;
381                 }
382                 pci_using_dac = 0;
383         }
384
385         if((err = pci_request_regions(pdev, ixgb_driver_name)))
386                 goto err_request_regions;
387
388         pci_set_master(pdev);
389
390         netdev = alloc_etherdev(sizeof(struct ixgb_adapter));
391         if(!netdev) {
392                 err = -ENOMEM;
393                 goto err_alloc_etherdev;
394         }
395
396         SET_NETDEV_DEV(netdev, &pdev->dev);
397
398         pci_set_drvdata(pdev, netdev);
399         adapter = netdev_priv(netdev);
400         adapter->netdev = netdev;
401         adapter->pdev = pdev;
402         adapter->hw.back = adapter;
403         adapter->msg_enable = netif_msg_init(debug, DEFAULT_DEBUG_LEVEL_SHIFT);
404
405         mmio_start = pci_resource_start(pdev, BAR_0);
406         mmio_len = pci_resource_len(pdev, BAR_0);
407
408         adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
409         if(!adapter->hw.hw_addr) {
410                 err = -EIO;
411                 goto err_ioremap;
412         }
413
414         for(i = BAR_1; i <= BAR_5; i++) {
415                 if(pci_resource_len(pdev, i) == 0)
416                         continue;
417                 if(pci_resource_flags(pdev, i) & IORESOURCE_IO) {
418                         adapter->hw.io_base = pci_resource_start(pdev, i);
419                         break;
420                 }
421         }
422
423         netdev->open = &ixgb_open;
424         netdev->stop = &ixgb_close;
425         netdev->hard_start_xmit = &ixgb_xmit_frame;
426         netdev->get_stats = &ixgb_get_stats;
427         netdev->set_multicast_list = &ixgb_set_multi;
428         netdev->set_mac_address = &ixgb_set_mac;
429         netdev->change_mtu = &ixgb_change_mtu;
430         ixgb_set_ethtool_ops(netdev);
431         netdev->tx_timeout = &ixgb_tx_timeout;
432         netdev->watchdog_timeo = 5 * HZ;
433 #ifdef CONFIG_IXGB_NAPI
434         netif_napi_add(netdev, &adapter->napi, ixgb_clean, 64);
435 #endif
436         netdev->vlan_rx_register = ixgb_vlan_rx_register;
437         netdev->vlan_rx_add_vid = ixgb_vlan_rx_add_vid;
438         netdev->vlan_rx_kill_vid = ixgb_vlan_rx_kill_vid;
439 #ifdef CONFIG_NET_POLL_CONTROLLER
440         netdev->poll_controller = ixgb_netpoll;
441 #endif
442
443         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
444         netdev->mem_start = mmio_start;
445         netdev->mem_end = mmio_start + mmio_len;
446         netdev->base_addr = adapter->hw.io_base;
447
448         adapter->bd_number = cards_found;
449         adapter->link_speed = 0;
450         adapter->link_duplex = 0;
451
452         /* setup the private structure */
453
454         if((err = ixgb_sw_init(adapter)))
455                 goto err_sw_init;
456
457         netdev->features = NETIF_F_SG |
458                            NETIF_F_HW_CSUM |
459                            NETIF_F_HW_VLAN_TX |
460                            NETIF_F_HW_VLAN_RX |
461                            NETIF_F_HW_VLAN_FILTER;
462         netdev->features |= NETIF_F_TSO;
463 #ifdef NETIF_F_LLTX
464         netdev->features |= NETIF_F_LLTX;
465 #endif
466
467         if(pci_using_dac)
468                 netdev->features |= NETIF_F_HIGHDMA;
469
470         /* make sure the EEPROM is good */
471
472         if(!ixgb_validate_eeprom_checksum(&adapter->hw)) {
473                 DPRINTK(PROBE, ERR, "The EEPROM Checksum Is Not Valid\n");
474                 err = -EIO;
475                 goto err_eeprom;
476         }
477
478         ixgb_get_ee_mac_addr(&adapter->hw, netdev->dev_addr);
479         memcpy(netdev->perm_addr, netdev->dev_addr, netdev->addr_len);
480
481         if(!is_valid_ether_addr(netdev->perm_addr)) {
482                 DPRINTK(PROBE, ERR, "Invalid MAC Address\n");
483                 err = -EIO;
484                 goto err_eeprom;
485         }
486
487         adapter->part_num = ixgb_get_ee_pba_number(&adapter->hw);
488
489         init_timer(&adapter->watchdog_timer);
490         adapter->watchdog_timer.function = &ixgb_watchdog;
491         adapter->watchdog_timer.data = (unsigned long)adapter;
492
493         INIT_WORK(&adapter->tx_timeout_task, ixgb_tx_timeout_task);
494
495         strcpy(netdev->name, "eth%d");
496         if((err = register_netdev(netdev)))
497                 goto err_register;
498
499         /* we're going to reset, so assume we have no link for now */
500
501         netif_carrier_off(netdev);
502         netif_stop_queue(netdev);
503
504         DPRINTK(PROBE, INFO, "Intel(R) PRO/10GbE Network Connection\n");
505         ixgb_check_options(adapter);
506         /* reset the hardware with the new settings */
507
508         ixgb_reset(adapter);
509
510         cards_found++;
511         return 0;
512
513 err_register:
514 err_sw_init:
515 err_eeprom:
516         iounmap(adapter->hw.hw_addr);
517 err_ioremap:
518         free_netdev(netdev);
519 err_alloc_etherdev:
520         pci_release_regions(pdev);
521 err_request_regions:
522 err_dma_mask:
523         pci_disable_device(pdev);
524         return err;
525 }
526
527 /**
528  * ixgb_remove - Device Removal Routine
529  * @pdev: PCI device information struct
530  *
531  * ixgb_remove is called by the PCI subsystem to alert the driver
532  * that it should release a PCI device.  The could be caused by a
533  * Hot-Plug event, or because the driver is going to be removed from
534  * memory.
535  **/
536
537 static void __devexit
538 ixgb_remove(struct pci_dev *pdev)
539 {
540         struct net_device *netdev = pci_get_drvdata(pdev);
541         struct ixgb_adapter *adapter = netdev_priv(netdev);
542
543         unregister_netdev(netdev);
544
545         iounmap(adapter->hw.hw_addr);
546         pci_release_regions(pdev);
547
548         free_netdev(netdev);
549 }
550
551 /**
552  * ixgb_sw_init - Initialize general software structures (struct ixgb_adapter)
553  * @adapter: board private structure to initialize
554  *
555  * ixgb_sw_init initializes the Adapter private data structure.
556  * Fields are initialized based on PCI device information and
557  * OS network device settings (MTU size).
558  **/
559
560 static int __devinit
561 ixgb_sw_init(struct ixgb_adapter *adapter)
562 {
563         struct ixgb_hw *hw = &adapter->hw;
564         struct net_device *netdev = adapter->netdev;
565         struct pci_dev *pdev = adapter->pdev;
566
567         /* PCI config space info */
568
569         hw->vendor_id = pdev->vendor;
570         hw->device_id = pdev->device;
571         hw->subsystem_vendor_id = pdev->subsystem_vendor;
572         hw->subsystem_id = pdev->subsystem_device;
573
574         hw->max_frame_size = netdev->mtu + ENET_HEADER_SIZE + ENET_FCS_LENGTH;
575         adapter->rx_buffer_len = hw->max_frame_size;
576
577         if((hw->device_id == IXGB_DEVICE_ID_82597EX)
578            || (hw->device_id == IXGB_DEVICE_ID_82597EX_CX4)
579            || (hw->device_id == IXGB_DEVICE_ID_82597EX_LR)
580            || (hw->device_id == IXGB_DEVICE_ID_82597EX_SR))
581                         hw->mac_type = ixgb_82597;
582         else {
583                 /* should never have loaded on this device */
584                 DPRINTK(PROBE, ERR, "unsupported device id\n");
585         }
586
587         /* enable flow control to be programmed */
588         hw->fc.send_xon = 1;
589
590         atomic_set(&adapter->irq_sem, 1);
591         spin_lock_init(&adapter->tx_lock);
592
593         return 0;
594 }
595
596 /**
597  * ixgb_open - Called when a network interface is made active
598  * @netdev: network interface device structure
599  *
600  * Returns 0 on success, negative value on failure
601  *
602  * The open entry point is called when a network interface is made
603  * active by the system (IFF_UP).  At this point all resources needed
604  * for transmit and receive operations are allocated, the interrupt
605  * handler is registered with the OS, the watchdog timer is started,
606  * and the stack is notified that the interface is ready.
607  **/
608
609 static int
610 ixgb_open(struct net_device *netdev)
611 {
612         struct ixgb_adapter *adapter = netdev_priv(netdev);
613         int err;
614
615         /* allocate transmit descriptors */
616
617         if((err = ixgb_setup_tx_resources(adapter)))
618                 goto err_setup_tx;
619
620         /* allocate receive descriptors */
621
622         if((err = ixgb_setup_rx_resources(adapter)))
623                 goto err_setup_rx;
624
625         if((err = ixgb_up(adapter)))
626                 goto err_up;
627
628         return 0;
629
630 err_up:
631         ixgb_free_rx_resources(adapter);
632 err_setup_rx:
633         ixgb_free_tx_resources(adapter);
634 err_setup_tx:
635         ixgb_reset(adapter);
636
637         return err;
638 }
639
640 /**
641  * ixgb_close - Disables a network interface
642  * @netdev: network interface device structure
643  *
644  * Returns 0, this is not allowed to fail
645  *
646  * The close entry point is called when an interface is de-activated
647  * by the OS.  The hardware is still under the drivers control, but
648  * needs to be disabled.  A global MAC reset is issued to stop the
649  * hardware, and all transmit and receive resources are freed.
650  **/
651
652 static int
653 ixgb_close(struct net_device *netdev)
654 {
655         struct ixgb_adapter *adapter = netdev_priv(netdev);
656
657         ixgb_down(adapter, TRUE);
658
659         ixgb_free_tx_resources(adapter);
660         ixgb_free_rx_resources(adapter);
661
662         return 0;
663 }
664
665 /**
666  * ixgb_setup_tx_resources - allocate Tx resources (Descriptors)
667  * @adapter: board private structure
668  *
669  * Return 0 on success, negative on failure
670  **/
671
672 int
673 ixgb_setup_tx_resources(struct ixgb_adapter *adapter)
674 {
675         struct ixgb_desc_ring *txdr = &adapter->tx_ring;
676         struct pci_dev *pdev = adapter->pdev;
677         int size;
678
679         size = sizeof(struct ixgb_buffer) * txdr->count;
680         txdr->buffer_info = vmalloc(size);
681         if(!txdr->buffer_info) {
682                 DPRINTK(PROBE, ERR,
683                  "Unable to allocate transmit descriptor ring memory\n");
684                 return -ENOMEM;
685         }
686         memset(txdr->buffer_info, 0, size);
687
688         /* round up to nearest 4K */
689
690         txdr->size = txdr->count * sizeof(struct ixgb_tx_desc);
691         txdr->size = ALIGN(txdr->size, 4096);
692
693         txdr->desc = pci_alloc_consistent(pdev, txdr->size, &txdr->dma);
694         if(!txdr->desc) {
695                 vfree(txdr->buffer_info);
696                 DPRINTK(PROBE, ERR,
697                  "Unable to allocate transmit descriptor memory\n");
698                 return -ENOMEM;
699         }
700         memset(txdr->desc, 0, txdr->size);
701
702         txdr->next_to_use = 0;
703         txdr->next_to_clean = 0;
704
705         return 0;
706 }
707
708 /**
709  * ixgb_configure_tx - Configure 82597 Transmit Unit after Reset.
710  * @adapter: board private structure
711  *
712  * Configure the Tx unit of the MAC after a reset.
713  **/
714
715 static void
716 ixgb_configure_tx(struct ixgb_adapter *adapter)
717 {
718         uint64_t tdba = adapter->tx_ring.dma;
719         uint32_t tdlen = adapter->tx_ring.count * sizeof(struct ixgb_tx_desc);
720         uint32_t tctl;
721         struct ixgb_hw *hw = &adapter->hw;
722
723         /* Setup the Base and Length of the Tx Descriptor Ring 
724          * tx_ring.dma can be either a 32 or 64 bit value 
725          */
726
727         IXGB_WRITE_REG(hw, TDBAL, (tdba & 0x00000000ffffffffULL));
728         IXGB_WRITE_REG(hw, TDBAH, (tdba >> 32));
729
730         IXGB_WRITE_REG(hw, TDLEN, tdlen);
731
732         /* Setup the HW Tx Head and Tail descriptor pointers */
733
734         IXGB_WRITE_REG(hw, TDH, 0);
735         IXGB_WRITE_REG(hw, TDT, 0);
736
737         /* don't set up txdctl, it induces performance problems if configured
738          * incorrectly */
739         /* Set the Tx Interrupt Delay register */
740
741         IXGB_WRITE_REG(hw, TIDV, adapter->tx_int_delay);
742
743         /* Program the Transmit Control Register */
744
745         tctl = IXGB_TCTL_TCE | IXGB_TCTL_TXEN | IXGB_TCTL_TPDE;
746         IXGB_WRITE_REG(hw, TCTL, tctl);
747
748         /* Setup Transmit Descriptor Settings for this adapter */
749         adapter->tx_cmd_type =
750                 IXGB_TX_DESC_TYPE 
751                 | (adapter->tx_int_delay_enable ? IXGB_TX_DESC_CMD_IDE : 0);
752 }
753
754 /**
755  * ixgb_setup_rx_resources - allocate Rx resources (Descriptors)
756  * @adapter: board private structure
757  *
758  * Returns 0 on success, negative on failure
759  **/
760
761 int
762 ixgb_setup_rx_resources(struct ixgb_adapter *adapter)
763 {
764         struct ixgb_desc_ring *rxdr = &adapter->rx_ring;
765         struct pci_dev *pdev = adapter->pdev;
766         int size;
767
768         size = sizeof(struct ixgb_buffer) * rxdr->count;
769         rxdr->buffer_info = vmalloc(size);
770         if(!rxdr->buffer_info) {
771                 DPRINTK(PROBE, ERR,
772                  "Unable to allocate receive descriptor ring\n");
773                 return -ENOMEM;
774         }
775         memset(rxdr->buffer_info, 0, size);
776
777         /* Round up to nearest 4K */
778
779         rxdr->size = rxdr->count * sizeof(struct ixgb_rx_desc);
780         rxdr->size = ALIGN(rxdr->size, 4096);
781
782         rxdr->desc = pci_alloc_consistent(pdev, rxdr->size, &rxdr->dma);
783
784         if(!rxdr->desc) {
785                 vfree(rxdr->buffer_info);
786                 DPRINTK(PROBE, ERR,
787                  "Unable to allocate receive descriptors\n");
788                 return -ENOMEM;
789         }
790         memset(rxdr->desc, 0, rxdr->size);
791
792         rxdr->next_to_clean = 0;
793         rxdr->next_to_use = 0;
794
795         return 0;
796 }
797
798 /**
799  * ixgb_setup_rctl - configure the receive control register
800  * @adapter: Board private structure
801  **/
802
803 static void
804 ixgb_setup_rctl(struct ixgb_adapter *adapter)
805 {
806         uint32_t rctl;
807
808         rctl = IXGB_READ_REG(&adapter->hw, RCTL);
809
810         rctl &= ~(3 << IXGB_RCTL_MO_SHIFT);
811
812         rctl |=
813                 IXGB_RCTL_BAM | IXGB_RCTL_RDMTS_1_2 | 
814                 IXGB_RCTL_RXEN | IXGB_RCTL_CFF | 
815                 (adapter->hw.mc_filter_type << IXGB_RCTL_MO_SHIFT);
816
817         rctl |= IXGB_RCTL_SECRC;
818
819         if (adapter->rx_buffer_len <= IXGB_RXBUFFER_2048)
820                 rctl |= IXGB_RCTL_BSIZE_2048;
821         else if (adapter->rx_buffer_len <= IXGB_RXBUFFER_4096)
822                 rctl |= IXGB_RCTL_BSIZE_4096;
823         else if (adapter->rx_buffer_len <= IXGB_RXBUFFER_8192)
824                 rctl |= IXGB_RCTL_BSIZE_8192;
825         else if (adapter->rx_buffer_len <= IXGB_RXBUFFER_16384)
826                 rctl |= IXGB_RCTL_BSIZE_16384;
827
828         IXGB_WRITE_REG(&adapter->hw, RCTL, rctl);
829 }
830
831 /**
832  * ixgb_configure_rx - Configure 82597 Receive Unit after Reset.
833  * @adapter: board private structure
834  *
835  * Configure the Rx unit of the MAC after a reset.
836  **/
837
838 static void
839 ixgb_configure_rx(struct ixgb_adapter *adapter)
840 {
841         uint64_t rdba = adapter->rx_ring.dma;
842         uint32_t rdlen = adapter->rx_ring.count * sizeof(struct ixgb_rx_desc);
843         struct ixgb_hw *hw = &adapter->hw;
844         uint32_t rctl;
845         uint32_t rxcsum;
846         uint32_t rxdctl;
847
848         /* make sure receives are disabled while setting up the descriptors */
849
850         rctl = IXGB_READ_REG(hw, RCTL);
851         IXGB_WRITE_REG(hw, RCTL, rctl & ~IXGB_RCTL_RXEN);
852
853         /* set the Receive Delay Timer Register */
854
855         IXGB_WRITE_REG(hw, RDTR, adapter->rx_int_delay);
856
857         /* Setup the Base and Length of the Rx Descriptor Ring */
858
859         IXGB_WRITE_REG(hw, RDBAL, (rdba & 0x00000000ffffffffULL));
860         IXGB_WRITE_REG(hw, RDBAH, (rdba >> 32));
861
862         IXGB_WRITE_REG(hw, RDLEN, rdlen);
863
864         /* Setup the HW Rx Head and Tail Descriptor Pointers */
865         IXGB_WRITE_REG(hw, RDH, 0);
866         IXGB_WRITE_REG(hw, RDT, 0);
867
868         /* set up pre-fetching of receive buffers so we get some before we
869          * run out (default hardware behavior is to run out before fetching
870          * more).  This sets up to fetch if HTHRESH rx descriptors are avail
871          * and the descriptors in hw cache are below PTHRESH.  This avoids
872          * the hardware behavior of fetching <=512 descriptors in a single
873          * burst that pre-empts all other activity, usually causing fifo
874          * overflows. */
875         /* use WTHRESH to burst write 16 descriptors or burst when RXT0 */
876         rxdctl = RXDCTL_WTHRESH_DEFAULT << IXGB_RXDCTL_WTHRESH_SHIFT |
877                  RXDCTL_HTHRESH_DEFAULT << IXGB_RXDCTL_HTHRESH_SHIFT |
878                  RXDCTL_PTHRESH_DEFAULT << IXGB_RXDCTL_PTHRESH_SHIFT;
879         IXGB_WRITE_REG(hw, RXDCTL, rxdctl);
880
881         /* Enable Receive Checksum Offload for TCP and UDP */
882         if(adapter->rx_csum == TRUE) {
883                 rxcsum = IXGB_READ_REG(hw, RXCSUM);
884                 rxcsum |= IXGB_RXCSUM_TUOFL;
885                 IXGB_WRITE_REG(hw, RXCSUM, rxcsum);
886         }
887
888         /* Enable Receives */
889
890         IXGB_WRITE_REG(hw, RCTL, rctl);
891 }
892
893 /**
894  * ixgb_free_tx_resources - Free Tx Resources
895  * @adapter: board private structure
896  *
897  * Free all transmit software resources
898  **/
899
900 void
901 ixgb_free_tx_resources(struct ixgb_adapter *adapter)
902 {
903         struct pci_dev *pdev = adapter->pdev;
904
905         ixgb_clean_tx_ring(adapter);
906
907         vfree(adapter->tx_ring.buffer_info);
908         adapter->tx_ring.buffer_info = NULL;
909
910         pci_free_consistent(pdev, adapter->tx_ring.size,
911                             adapter->tx_ring.desc, adapter->tx_ring.dma);
912
913         adapter->tx_ring.desc = NULL;
914 }
915
916 static void
917 ixgb_unmap_and_free_tx_resource(struct ixgb_adapter *adapter,
918                                         struct ixgb_buffer *buffer_info)
919 {
920         struct pci_dev *pdev = adapter->pdev;
921
922         if (buffer_info->dma)
923                 pci_unmap_page(pdev, buffer_info->dma, buffer_info->length,
924                                PCI_DMA_TODEVICE);
925
926         if (buffer_info->skb)
927                 dev_kfree_skb_any(buffer_info->skb);
928
929         buffer_info->skb = NULL;
930         buffer_info->dma = 0;
931         buffer_info->time_stamp = 0;
932         /* these fields must always be initialized in tx
933          * buffer_info->length = 0;
934          * buffer_info->next_to_watch = 0; */
935 }
936
937 /**
938  * ixgb_clean_tx_ring - Free Tx Buffers
939  * @adapter: board private structure
940  **/
941
942 static void
943 ixgb_clean_tx_ring(struct ixgb_adapter *adapter)
944 {
945         struct ixgb_desc_ring *tx_ring = &adapter->tx_ring;
946         struct ixgb_buffer *buffer_info;
947         unsigned long size;
948         unsigned int i;
949
950         /* Free all the Tx ring sk_buffs */
951
952         for(i = 0; i < tx_ring->count; i++) {
953                 buffer_info = &tx_ring->buffer_info[i];
954                 ixgb_unmap_and_free_tx_resource(adapter, buffer_info);
955         }
956
957         size = sizeof(struct ixgb_buffer) * tx_ring->count;
958         memset(tx_ring->buffer_info, 0, size);
959
960         /* Zero out the descriptor ring */
961
962         memset(tx_ring->desc, 0, tx_ring->size);
963
964         tx_ring->next_to_use = 0;
965         tx_ring->next_to_clean = 0;
966
967         IXGB_WRITE_REG(&adapter->hw, TDH, 0);
968         IXGB_WRITE_REG(&adapter->hw, TDT, 0);
969 }
970
971 /**
972  * ixgb_free_rx_resources - Free Rx Resources
973  * @adapter: board private structure
974  *
975  * Free all receive software resources
976  **/
977
978 void
979 ixgb_free_rx_resources(struct ixgb_adapter *adapter)
980 {
981         struct ixgb_desc_ring *rx_ring = &adapter->rx_ring;
982         struct pci_dev *pdev = adapter->pdev;
983
984         ixgb_clean_rx_ring(adapter);
985
986         vfree(rx_ring->buffer_info);
987         rx_ring->buffer_info = NULL;
988
989         pci_free_consistent(pdev, rx_ring->size, rx_ring->desc, rx_ring->dma);
990
991         rx_ring->desc = NULL;
992 }
993
994 /**
995  * ixgb_clean_rx_ring - Free Rx Buffers
996  * @adapter: board private structure
997  **/
998
999 static void
1000 ixgb_clean_rx_ring(struct ixgb_adapter *adapter)
1001 {
1002         struct ixgb_desc_ring *rx_ring = &adapter->rx_ring;
1003         struct ixgb_buffer *buffer_info;
1004         struct pci_dev *pdev = adapter->pdev;
1005         unsigned long size;
1006         unsigned int i;
1007
1008         /* Free all the Rx ring sk_buffs */
1009
1010         for(i = 0; i < rx_ring->count; i++) {
1011                 buffer_info = &rx_ring->buffer_info[i];
1012                 if(buffer_info->skb) {
1013
1014                         pci_unmap_single(pdev,
1015                                          buffer_info->dma,
1016                                          buffer_info->length,
1017                                          PCI_DMA_FROMDEVICE);
1018
1019                         dev_kfree_skb(buffer_info->skb);
1020
1021                         buffer_info->skb = NULL;
1022                 }
1023         }
1024
1025         size = sizeof(struct ixgb_buffer) * rx_ring->count;
1026         memset(rx_ring->buffer_info, 0, size);
1027
1028         /* Zero out the descriptor ring */
1029
1030         memset(rx_ring->desc, 0, rx_ring->size);
1031
1032         rx_ring->next_to_clean = 0;
1033         rx_ring->next_to_use = 0;
1034
1035         IXGB_WRITE_REG(&adapter->hw, RDH, 0);
1036         IXGB_WRITE_REG(&adapter->hw, RDT, 0);
1037 }
1038
1039 /**
1040  * ixgb_set_mac - Change the Ethernet Address of the NIC
1041  * @netdev: network interface device structure
1042  * @p: pointer to an address structure
1043  *
1044  * Returns 0 on success, negative on failure
1045  **/
1046
1047 static int
1048 ixgb_set_mac(struct net_device *netdev, void *p)
1049 {
1050         struct ixgb_adapter *adapter = netdev_priv(netdev);
1051         struct sockaddr *addr = p;
1052
1053         if(!is_valid_ether_addr(addr->sa_data))
1054                 return -EADDRNOTAVAIL;
1055
1056         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
1057
1058         ixgb_rar_set(&adapter->hw, addr->sa_data, 0);
1059
1060         return 0;
1061 }
1062
1063 /**
1064  * ixgb_set_multi - Multicast and Promiscuous mode set
1065  * @netdev: network interface device structure
1066  *
1067  * The set_multi entry point is called whenever the multicast address
1068  * list or the network interface flags are updated.  This routine is
1069  * responsible for configuring the hardware for proper multicast,
1070  * promiscuous mode, and all-multi behavior.
1071  **/
1072
1073 static void
1074 ixgb_set_multi(struct net_device *netdev)
1075 {
1076         struct ixgb_adapter *adapter = netdev_priv(netdev);
1077         struct ixgb_hw *hw = &adapter->hw;
1078         struct dev_mc_list *mc_ptr;
1079         uint32_t rctl;
1080         int i;
1081
1082         /* Check for Promiscuous and All Multicast modes */
1083
1084         rctl = IXGB_READ_REG(hw, RCTL);
1085
1086         if(netdev->flags & IFF_PROMISC) {
1087                 rctl |= (IXGB_RCTL_UPE | IXGB_RCTL_MPE);
1088         } else if(netdev->flags & IFF_ALLMULTI) {
1089                 rctl |= IXGB_RCTL_MPE;
1090                 rctl &= ~IXGB_RCTL_UPE;
1091         } else {
1092                 rctl &= ~(IXGB_RCTL_UPE | IXGB_RCTL_MPE);
1093         }
1094
1095         if(netdev->mc_count > IXGB_MAX_NUM_MULTICAST_ADDRESSES) {
1096                 rctl |= IXGB_RCTL_MPE;
1097                 IXGB_WRITE_REG(hw, RCTL, rctl);
1098         } else {
1099                 uint8_t mta[IXGB_MAX_NUM_MULTICAST_ADDRESSES *
1100                             IXGB_ETH_LENGTH_OF_ADDRESS];
1101
1102                 IXGB_WRITE_REG(hw, RCTL, rctl);
1103
1104                 for(i = 0, mc_ptr = netdev->mc_list; mc_ptr;
1105                         i++, mc_ptr = mc_ptr->next)
1106                         memcpy(&mta[i * IXGB_ETH_LENGTH_OF_ADDRESS],
1107                                    mc_ptr->dmi_addr, IXGB_ETH_LENGTH_OF_ADDRESS);
1108
1109                 ixgb_mc_addr_list_update(hw, mta, netdev->mc_count, 0);
1110         }
1111 }
1112
1113 /**
1114  * ixgb_watchdog - Timer Call-back
1115  * @data: pointer to netdev cast into an unsigned long
1116  **/
1117
1118 static void
1119 ixgb_watchdog(unsigned long data)
1120 {
1121         struct ixgb_adapter *adapter = (struct ixgb_adapter *)data;
1122         struct net_device *netdev = adapter->netdev;
1123         struct ixgb_desc_ring *txdr = &adapter->tx_ring;
1124
1125         ixgb_check_for_link(&adapter->hw);
1126
1127         if (ixgb_check_for_bad_link(&adapter->hw)) {
1128                 /* force the reset path */
1129                 netif_stop_queue(netdev);
1130         }
1131
1132         if(adapter->hw.link_up) {
1133                 if(!netif_carrier_ok(netdev)) {
1134                         DPRINTK(LINK, INFO,
1135                                 "NIC Link is Up 10000 Mbps Full Duplex\n");
1136                         adapter->link_speed = 10000;
1137                         adapter->link_duplex = FULL_DUPLEX;
1138                         netif_carrier_on(netdev);
1139                         netif_wake_queue(netdev);
1140                 }
1141         } else {
1142                 if(netif_carrier_ok(netdev)) {
1143                         adapter->link_speed = 0;
1144                         adapter->link_duplex = 0;
1145                         DPRINTK(LINK, INFO, "NIC Link is Down\n");
1146                         netif_carrier_off(netdev);
1147                         netif_stop_queue(netdev);
1148
1149                 }
1150         }
1151
1152         ixgb_update_stats(adapter);
1153
1154         if(!netif_carrier_ok(netdev)) {
1155                 if(IXGB_DESC_UNUSED(txdr) + 1 < txdr->count) {
1156                         /* We've lost link, so the controller stops DMA,
1157                          * but we've got queued Tx work that's never going
1158                          * to get done, so reset controller to flush Tx.
1159                          * (Do the reset outside of interrupt context). */
1160                         schedule_work(&adapter->tx_timeout_task);
1161                 }
1162         }
1163
1164         /* Force detection of hung controller every watchdog period */
1165         adapter->detect_tx_hung = TRUE;
1166
1167         /* generate an interrupt to force clean up of any stragglers */
1168         IXGB_WRITE_REG(&adapter->hw, ICS, IXGB_INT_TXDW);
1169
1170         /* Reset the timer */
1171         mod_timer(&adapter->watchdog_timer, jiffies + 2 * HZ);
1172 }
1173
1174 #define IXGB_TX_FLAGS_CSUM              0x00000001
1175 #define IXGB_TX_FLAGS_VLAN              0x00000002
1176 #define IXGB_TX_FLAGS_TSO               0x00000004
1177
1178 static int
1179 ixgb_tso(struct ixgb_adapter *adapter, struct sk_buff *skb)
1180 {
1181         struct ixgb_context_desc *context_desc;
1182         unsigned int i;
1183         uint8_t ipcss, ipcso, tucss, tucso, hdr_len;
1184         uint16_t ipcse, tucse, mss;
1185         int err;
1186
1187         if (likely(skb_is_gso(skb))) {
1188                 struct ixgb_buffer *buffer_info;
1189                 struct iphdr *iph;
1190
1191                 if (skb_header_cloned(skb)) {
1192                         err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1193                         if (err)
1194                                 return err;
1195                 }
1196
1197                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
1198                 mss = skb_shinfo(skb)->gso_size;
1199                 iph = ip_hdr(skb);
1200                 iph->tot_len = 0;
1201                 iph->check = 0;
1202                 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
1203                                                          iph->daddr, 0,
1204                                                          IPPROTO_TCP, 0);
1205                 ipcss = skb_network_offset(skb);
1206                 ipcso = (void *)&(iph->check) - (void *)skb->data;
1207                 ipcse = skb_transport_offset(skb) - 1;
1208                 tucss = skb_transport_offset(skb);
1209                 tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
1210                 tucse = 0;
1211
1212                 i = adapter->tx_ring.next_to_use;
1213                 context_desc = IXGB_CONTEXT_DESC(adapter->tx_ring, i);
1214                 buffer_info = &adapter->tx_ring.buffer_info[i];
1215                 WARN_ON(buffer_info->dma != 0);
1216
1217                 context_desc->ipcss = ipcss;
1218                 context_desc->ipcso = ipcso;
1219                 context_desc->ipcse = cpu_to_le16(ipcse);
1220                 context_desc->tucss = tucss;
1221                 context_desc->tucso = tucso;
1222                 context_desc->tucse = cpu_to_le16(tucse);
1223                 context_desc->mss = cpu_to_le16(mss);
1224                 context_desc->hdr_len = hdr_len;
1225                 context_desc->status = 0;
1226                 context_desc->cmd_type_len = cpu_to_le32(
1227                                                   IXGB_CONTEXT_DESC_TYPE 
1228                                                 | IXGB_CONTEXT_DESC_CMD_TSE
1229                                                 | IXGB_CONTEXT_DESC_CMD_IP
1230                                                 | IXGB_CONTEXT_DESC_CMD_TCP
1231                                                 | IXGB_CONTEXT_DESC_CMD_IDE
1232                                                 | (skb->len - (hdr_len)));
1233
1234
1235                 if(++i == adapter->tx_ring.count) i = 0;
1236                 adapter->tx_ring.next_to_use = i;
1237
1238                 return 1;
1239         }
1240
1241         return 0;
1242 }
1243
1244 static boolean_t
1245 ixgb_tx_csum(struct ixgb_adapter *adapter, struct sk_buff *skb)
1246 {
1247         struct ixgb_context_desc *context_desc;
1248         unsigned int i;
1249         uint8_t css, cso;
1250
1251         if(likely(skb->ip_summed == CHECKSUM_PARTIAL)) {
1252                 struct ixgb_buffer *buffer_info;
1253                 css = skb_transport_offset(skb);
1254                 cso = css + skb->csum_offset;
1255
1256                 i = adapter->tx_ring.next_to_use;
1257                 context_desc = IXGB_CONTEXT_DESC(adapter->tx_ring, i);
1258                 buffer_info = &adapter->tx_ring.buffer_info[i];
1259                 WARN_ON(buffer_info->dma != 0);
1260
1261                 context_desc->tucss = css;
1262                 context_desc->tucso = cso;
1263                 context_desc->tucse = 0;
1264                 /* zero out any previously existing data in one instruction */
1265                 *(uint32_t *)&(context_desc->ipcss) = 0;
1266                 context_desc->status = 0;
1267                 context_desc->hdr_len = 0;
1268                 context_desc->mss = 0;
1269                 context_desc->cmd_type_len =
1270                         cpu_to_le32(IXGB_CONTEXT_DESC_TYPE
1271                                     | IXGB_TX_DESC_CMD_IDE);
1272
1273                 if(++i == adapter->tx_ring.count) i = 0;
1274                 adapter->tx_ring.next_to_use = i;
1275
1276                 return TRUE;
1277         }
1278
1279         return FALSE;
1280 }
1281
1282 #define IXGB_MAX_TXD_PWR        14
1283 #define IXGB_MAX_DATA_PER_TXD   (1<<IXGB_MAX_TXD_PWR)
1284
1285 static int
1286 ixgb_tx_map(struct ixgb_adapter *adapter, struct sk_buff *skb,
1287             unsigned int first)
1288 {
1289         struct ixgb_desc_ring *tx_ring = &adapter->tx_ring;
1290         struct ixgb_buffer *buffer_info;
1291         int len = skb->len;
1292         unsigned int offset = 0, size, count = 0, i;
1293         unsigned int mss = skb_shinfo(skb)->gso_size;
1294
1295         unsigned int nr_frags = skb_shinfo(skb)->nr_frags;
1296         unsigned int f;
1297
1298         len -= skb->data_len;
1299
1300         i = tx_ring->next_to_use;
1301
1302         while(len) {
1303                 buffer_info = &tx_ring->buffer_info[i];
1304                 size = min(len, IXGB_MAX_DATA_PER_TXD);
1305                 /* Workaround for premature desc write-backs
1306                  * in TSO mode.  Append 4-byte sentinel desc */
1307                 if (unlikely(mss && !nr_frags && size == len && size > 8))
1308                         size -= 4;
1309
1310                 buffer_info->length = size;
1311                 WARN_ON(buffer_info->dma != 0);
1312                 buffer_info->dma =
1313                         pci_map_single(adapter->pdev,
1314                                 skb->data + offset,
1315                                 size,
1316                                 PCI_DMA_TODEVICE);
1317                 buffer_info->time_stamp = jiffies;
1318                 buffer_info->next_to_watch = 0;
1319
1320                 len -= size;
1321                 offset += size;
1322                 count++;
1323                 if(++i == tx_ring->count) i = 0;
1324         }
1325
1326         for(f = 0; f < nr_frags; f++) {
1327                 struct skb_frag_struct *frag;
1328
1329                 frag = &skb_shinfo(skb)->frags[f];
1330                 len = frag->size;
1331                 offset = 0;
1332
1333                 while(len) {
1334                         buffer_info = &tx_ring->buffer_info[i];
1335                         size = min(len, IXGB_MAX_DATA_PER_TXD);
1336
1337                         /* Workaround for premature desc write-backs
1338                          * in TSO mode.  Append 4-byte sentinel desc */
1339                         if (unlikely(mss && (f == (nr_frags - 1))
1340                                      && size == len && size > 8))
1341                                 size -= 4;
1342
1343                         buffer_info->length = size;
1344                         buffer_info->dma =
1345                                 pci_map_page(adapter->pdev,
1346                                         frag->page,
1347                                         frag->page_offset + offset,
1348                                         size,
1349                                         PCI_DMA_TODEVICE);
1350                         buffer_info->time_stamp = jiffies;
1351                         buffer_info->next_to_watch = 0;
1352
1353                         len -= size;
1354                         offset += size;
1355                         count++;
1356                         if(++i == tx_ring->count) i = 0;
1357                 }
1358         }
1359         i = (i == 0) ? tx_ring->count - 1 : i - 1;
1360         tx_ring->buffer_info[i].skb = skb;
1361         tx_ring->buffer_info[first].next_to_watch = i;
1362
1363         return count;
1364 }
1365
1366 static void
1367 ixgb_tx_queue(struct ixgb_adapter *adapter, int count, int vlan_id,int tx_flags)
1368 {
1369         struct ixgb_desc_ring *tx_ring = &adapter->tx_ring;
1370         struct ixgb_tx_desc *tx_desc = NULL;
1371         struct ixgb_buffer *buffer_info;
1372         uint32_t cmd_type_len = adapter->tx_cmd_type;
1373         uint8_t status = 0;
1374         uint8_t popts = 0;
1375         unsigned int i;
1376
1377         if(tx_flags & IXGB_TX_FLAGS_TSO) {
1378                 cmd_type_len |= IXGB_TX_DESC_CMD_TSE;
1379                 popts |= (IXGB_TX_DESC_POPTS_IXSM | IXGB_TX_DESC_POPTS_TXSM);
1380         }
1381
1382         if(tx_flags & IXGB_TX_FLAGS_CSUM)
1383                 popts |= IXGB_TX_DESC_POPTS_TXSM;
1384
1385         if(tx_flags & IXGB_TX_FLAGS_VLAN) {
1386                 cmd_type_len |= IXGB_TX_DESC_CMD_VLE;
1387         }
1388
1389         i = tx_ring->next_to_use;
1390
1391         while(count--) {
1392                 buffer_info = &tx_ring->buffer_info[i];
1393                 tx_desc = IXGB_TX_DESC(*tx_ring, i);
1394                 tx_desc->buff_addr = cpu_to_le64(buffer_info->dma);
1395                 tx_desc->cmd_type_len =
1396                         cpu_to_le32(cmd_type_len | buffer_info->length);
1397                 tx_desc->status = status;
1398                 tx_desc->popts = popts;
1399                 tx_desc->vlan = cpu_to_le16(vlan_id);
1400
1401                 if(++i == tx_ring->count) i = 0;
1402         }
1403
1404         tx_desc->cmd_type_len |= cpu_to_le32(IXGB_TX_DESC_CMD_EOP 
1405                                 | IXGB_TX_DESC_CMD_RS );
1406
1407         /* Force memory writes to complete before letting h/w
1408          * know there are new descriptors to fetch.  (Only
1409          * applicable for weak-ordered memory model archs,
1410          * such as IA-64). */
1411         wmb();
1412
1413         tx_ring->next_to_use = i;
1414         IXGB_WRITE_REG(&adapter->hw, TDT, i);
1415 }
1416
1417 static int __ixgb_maybe_stop_tx(struct net_device *netdev, int size)
1418 {
1419         struct ixgb_adapter *adapter = netdev_priv(netdev);
1420         struct ixgb_desc_ring *tx_ring = &adapter->tx_ring;
1421
1422         netif_stop_queue(netdev);
1423         /* Herbert's original patch had:
1424          *  smp_mb__after_netif_stop_queue();
1425          * but since that doesn't exist yet, just open code it. */
1426         smp_mb();
1427
1428         /* We need to check again in a case another CPU has just
1429          * made room available. */
1430         if (likely(IXGB_DESC_UNUSED(tx_ring) < size))
1431                 return -EBUSY;
1432
1433         /* A reprieve! */
1434         netif_start_queue(netdev);
1435         ++adapter->restart_queue;
1436         return 0;
1437 }
1438
1439 static int ixgb_maybe_stop_tx(struct net_device *netdev,
1440                               struct ixgb_desc_ring *tx_ring, int size)
1441 {
1442         if (likely(IXGB_DESC_UNUSED(tx_ring) >= size))
1443                 return 0;
1444         return __ixgb_maybe_stop_tx(netdev, size);
1445 }
1446
1447
1448 /* Tx Descriptors needed, worst case */
1449 #define TXD_USE_COUNT(S) (((S) >> IXGB_MAX_TXD_PWR) + \
1450                          (((S) & (IXGB_MAX_DATA_PER_TXD - 1)) ? 1 : 0))
1451 #define DESC_NEEDED TXD_USE_COUNT(IXGB_MAX_DATA_PER_TXD) /* skb->date */ + \
1452         MAX_SKB_FRAGS * TXD_USE_COUNT(PAGE_SIZE) + 1 /* for context */ \
1453         + 1 /* one more needed for sentinel TSO workaround */
1454
1455 static int
1456 ixgb_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
1457 {
1458         struct ixgb_adapter *adapter = netdev_priv(netdev);
1459         unsigned int first;
1460         unsigned int tx_flags = 0;
1461         unsigned long flags;
1462         int vlan_id = 0;
1463         int tso;
1464
1465         if(skb->len <= 0) {
1466                 dev_kfree_skb_any(skb);
1467                 return 0;
1468         }
1469
1470 #ifdef NETIF_F_LLTX
1471         local_irq_save(flags);
1472         if (!spin_trylock(&adapter->tx_lock)) {
1473                 /* Collision - tell upper layer to requeue */
1474                 local_irq_restore(flags);
1475                 return NETDEV_TX_LOCKED;
1476         }
1477 #else
1478         spin_lock_irqsave(&adapter->tx_lock, flags);
1479 #endif
1480
1481         if (unlikely(ixgb_maybe_stop_tx(netdev, &adapter->tx_ring,
1482                      DESC_NEEDED))) {
1483                 netif_stop_queue(netdev);
1484                 spin_unlock_irqrestore(&adapter->tx_lock, flags);
1485                 return NETDEV_TX_BUSY;
1486         }
1487
1488 #ifndef NETIF_F_LLTX
1489         spin_unlock_irqrestore(&adapter->tx_lock, flags);
1490 #endif
1491
1492         if(adapter->vlgrp && vlan_tx_tag_present(skb)) {
1493                 tx_flags |= IXGB_TX_FLAGS_VLAN;
1494                 vlan_id = vlan_tx_tag_get(skb);
1495         }
1496
1497         first = adapter->tx_ring.next_to_use;
1498         
1499         tso = ixgb_tso(adapter, skb);
1500         if (tso < 0) {
1501                 dev_kfree_skb_any(skb);
1502 #ifdef NETIF_F_LLTX
1503                 spin_unlock_irqrestore(&adapter->tx_lock, flags);
1504 #endif
1505                 return NETDEV_TX_OK;
1506         }
1507
1508         if (likely(tso))
1509                 tx_flags |= IXGB_TX_FLAGS_TSO;
1510         else if(ixgb_tx_csum(adapter, skb))
1511                 tx_flags |= IXGB_TX_FLAGS_CSUM;
1512
1513         ixgb_tx_queue(adapter, ixgb_tx_map(adapter, skb, first), vlan_id,
1514                         tx_flags);
1515
1516         netdev->trans_start = jiffies;
1517
1518 #ifdef NETIF_F_LLTX
1519         /* Make sure there is space in the ring for the next send. */
1520         ixgb_maybe_stop_tx(netdev, &adapter->tx_ring, DESC_NEEDED);
1521
1522         spin_unlock_irqrestore(&adapter->tx_lock, flags);
1523
1524 #endif
1525         return NETDEV_TX_OK;
1526 }
1527
1528 /**
1529  * ixgb_tx_timeout - Respond to a Tx Hang
1530  * @netdev: network interface device structure
1531  **/
1532
1533 static void
1534 ixgb_tx_timeout(struct net_device *netdev)
1535 {
1536         struct ixgb_adapter *adapter = netdev_priv(netdev);
1537
1538         /* Do the reset outside of interrupt context */
1539         schedule_work(&adapter->tx_timeout_task);
1540 }
1541
1542 static void
1543 ixgb_tx_timeout_task(struct work_struct *work)
1544 {
1545         struct ixgb_adapter *adapter =
1546                 container_of(work, struct ixgb_adapter, tx_timeout_task);
1547
1548         adapter->tx_timeout_count++;
1549         ixgb_down(adapter, TRUE);
1550         ixgb_up(adapter);
1551 }
1552
1553 /**
1554  * ixgb_get_stats - Get System Network Statistics
1555  * @netdev: network interface device structure
1556  *
1557  * Returns the address of the device statistics structure.
1558  * The statistics are actually updated from the timer callback.
1559  **/
1560
1561 static struct net_device_stats *
1562 ixgb_get_stats(struct net_device *netdev)
1563 {
1564         struct ixgb_adapter *adapter = netdev_priv(netdev);
1565
1566         return &adapter->net_stats;
1567 }
1568
1569 /**
1570  * ixgb_change_mtu - Change the Maximum Transfer Unit
1571  * @netdev: network interface device structure
1572  * @new_mtu: new value for maximum frame size
1573  *
1574  * Returns 0 on success, negative on failure
1575  **/
1576
1577 static int
1578 ixgb_change_mtu(struct net_device *netdev, int new_mtu)
1579 {
1580         struct ixgb_adapter *adapter = netdev_priv(netdev);
1581         int max_frame = new_mtu + ENET_HEADER_SIZE + ENET_FCS_LENGTH;
1582         int old_max_frame = netdev->mtu + ENET_HEADER_SIZE + ENET_FCS_LENGTH;
1583
1584
1585         if((max_frame < IXGB_MIN_ENET_FRAME_SIZE_WITHOUT_FCS + ENET_FCS_LENGTH)
1586            || (max_frame > IXGB_MAX_JUMBO_FRAME_SIZE + ENET_FCS_LENGTH)) {
1587                 DPRINTK(PROBE, ERR, "Invalid MTU setting %d\n", new_mtu);
1588                 return -EINVAL;
1589         }
1590
1591         adapter->rx_buffer_len = max_frame;
1592
1593         netdev->mtu = new_mtu;
1594
1595         if ((old_max_frame != max_frame) && netif_running(netdev)) {
1596                 ixgb_down(adapter, TRUE);
1597                 ixgb_up(adapter);
1598         }
1599
1600         return 0;
1601 }
1602
1603 /**
1604  * ixgb_update_stats - Update the board statistics counters.
1605  * @adapter: board private structure
1606  **/
1607
1608 void
1609 ixgb_update_stats(struct ixgb_adapter *adapter)
1610 {
1611         struct net_device *netdev = adapter->netdev;
1612         struct pci_dev *pdev = adapter->pdev;
1613
1614         /* Prevent stats update while adapter is being reset */
1615         if (pci_channel_offline(pdev))
1616                 return;
1617
1618         if((netdev->flags & IFF_PROMISC) || (netdev->flags & IFF_ALLMULTI) ||
1619            (netdev->mc_count > IXGB_MAX_NUM_MULTICAST_ADDRESSES)) {
1620                 u64 multi = IXGB_READ_REG(&adapter->hw, MPRCL);
1621                 u32 bcast_l = IXGB_READ_REG(&adapter->hw, BPRCL);
1622                 u32 bcast_h = IXGB_READ_REG(&adapter->hw, BPRCH);
1623                 u64 bcast = ((u64)bcast_h << 32) | bcast_l; 
1624
1625                 multi |= ((u64)IXGB_READ_REG(&adapter->hw, MPRCH) << 32);
1626                 /* fix up multicast stats by removing broadcasts */
1627                 if(multi >= bcast)
1628                         multi -= bcast;
1629                 
1630                 adapter->stats.mprcl += (multi & 0xFFFFFFFF);
1631                 adapter->stats.mprch += (multi >> 32);
1632                 adapter->stats.bprcl += bcast_l; 
1633                 adapter->stats.bprch += bcast_h;
1634         } else {
1635                 adapter->stats.mprcl += IXGB_READ_REG(&adapter->hw, MPRCL);
1636                 adapter->stats.mprch += IXGB_READ_REG(&adapter->hw, MPRCH);
1637                 adapter->stats.bprcl += IXGB_READ_REG(&adapter->hw, BPRCL);
1638                 adapter->stats.bprch += IXGB_READ_REG(&adapter->hw, BPRCH);
1639         }
1640         adapter->stats.tprl += IXGB_READ_REG(&adapter->hw, TPRL);
1641         adapter->stats.tprh += IXGB_READ_REG(&adapter->hw, TPRH);
1642         adapter->stats.gprcl += IXGB_READ_REG(&adapter->hw, GPRCL);
1643         adapter->stats.gprch += IXGB_READ_REG(&adapter->hw, GPRCH);
1644         adapter->stats.uprcl += IXGB_READ_REG(&adapter->hw, UPRCL);
1645         adapter->stats.uprch += IXGB_READ_REG(&adapter->hw, UPRCH);
1646         adapter->stats.vprcl += IXGB_READ_REG(&adapter->hw, VPRCL);
1647         adapter->stats.vprch += IXGB_READ_REG(&adapter->hw, VPRCH);
1648         adapter->stats.jprcl += IXGB_READ_REG(&adapter->hw, JPRCL);
1649         adapter->stats.jprch += IXGB_READ_REG(&adapter->hw, JPRCH);
1650         adapter->stats.gorcl += IXGB_READ_REG(&adapter->hw, GORCL);
1651         adapter->stats.gorch += IXGB_READ_REG(&adapter->hw, GORCH);
1652         adapter->stats.torl += IXGB_READ_REG(&adapter->hw, TORL);
1653         adapter->stats.torh += IXGB_READ_REG(&adapter->hw, TORH);
1654         adapter->stats.rnbc += IXGB_READ_REG(&adapter->hw, RNBC);
1655         adapter->stats.ruc += IXGB_READ_REG(&adapter->hw, RUC);
1656         adapter->stats.roc += IXGB_READ_REG(&adapter->hw, ROC);
1657         adapter->stats.rlec += IXGB_READ_REG(&adapter->hw, RLEC);
1658         adapter->stats.crcerrs += IXGB_READ_REG(&adapter->hw, CRCERRS);
1659         adapter->stats.icbc += IXGB_READ_REG(&adapter->hw, ICBC);
1660         adapter->stats.ecbc += IXGB_READ_REG(&adapter->hw, ECBC);
1661         adapter->stats.mpc += IXGB_READ_REG(&adapter->hw, MPC);
1662         adapter->stats.tptl += IXGB_READ_REG(&adapter->hw, TPTL);
1663         adapter->stats.tpth += IXGB_READ_REG(&adapter->hw, TPTH);
1664         adapter->stats.gptcl += IXGB_READ_REG(&adapter->hw, GPTCL);
1665         adapter->stats.gptch += IXGB_READ_REG(&adapter->hw, GPTCH);
1666         adapter->stats.bptcl += IXGB_READ_REG(&adapter->hw, BPTCL);
1667         adapter->stats.bptch += IXGB_READ_REG(&adapter->hw, BPTCH);
1668         adapter->stats.mptcl += IXGB_READ_REG(&adapter->hw, MPTCL);
1669         adapter->stats.mptch += IXGB_READ_REG(&adapter->hw, MPTCH);
1670         adapter->stats.uptcl += IXGB_READ_REG(&adapter->hw, UPTCL);
1671         adapter->stats.uptch += IXGB_READ_REG(&adapter->hw, UPTCH);
1672         adapter->stats.vptcl += IXGB_READ_REG(&adapter->hw, VPTCL);
1673         adapter->stats.vptch += IXGB_READ_REG(&adapter->hw, VPTCH);
1674         adapter->stats.jptcl += IXGB_READ_REG(&adapter->hw, JPTCL);
1675         adapter->stats.jptch += IXGB_READ_REG(&adapter->hw, JPTCH);
1676         adapter->stats.gotcl += IXGB_READ_REG(&adapter->hw, GOTCL);
1677         adapter->stats.gotch += IXGB_READ_REG(&adapter->hw, GOTCH);
1678         adapter->stats.totl += IXGB_READ_REG(&adapter->hw, TOTL);
1679         adapter->stats.toth += IXGB_READ_REG(&adapter->hw, TOTH);
1680         adapter->stats.dc += IXGB_READ_REG(&adapter->hw, DC);
1681         adapter->stats.plt64c += IXGB_READ_REG(&adapter->hw, PLT64C);
1682         adapter->stats.tsctc += IXGB_READ_REG(&adapter->hw, TSCTC);
1683         adapter->stats.tsctfc += IXGB_READ_REG(&adapter->hw, TSCTFC);
1684         adapter->stats.ibic += IXGB_READ_REG(&adapter->hw, IBIC);
1685         adapter->stats.rfc += IXGB_READ_REG(&adapter->hw, RFC);
1686         adapter->stats.lfc += IXGB_READ_REG(&adapter->hw, LFC);
1687         adapter->stats.pfrc += IXGB_READ_REG(&adapter->hw, PFRC);
1688         adapter->stats.pftc += IXGB_READ_REG(&adapter->hw, PFTC);
1689         adapter->stats.mcfrc += IXGB_READ_REG(&adapter->hw, MCFRC);
1690         adapter->stats.mcftc += IXGB_READ_REG(&adapter->hw, MCFTC);
1691         adapter->stats.xonrxc += IXGB_READ_REG(&adapter->hw, XONRXC);
1692         adapter->stats.xontxc += IXGB_READ_REG(&adapter->hw, XONTXC);
1693         adapter->stats.xoffrxc += IXGB_READ_REG(&adapter->hw, XOFFRXC);
1694         adapter->stats.xofftxc += IXGB_READ_REG(&adapter->hw, XOFFTXC);
1695         adapter->stats.rjc += IXGB_READ_REG(&adapter->hw, RJC);
1696
1697         /* Fill out the OS statistics structure */
1698
1699         adapter->net_stats.rx_packets = adapter->stats.gprcl;
1700         adapter->net_stats.tx_packets = adapter->stats.gptcl;
1701         adapter->net_stats.rx_bytes = adapter->stats.gorcl;
1702         adapter->net_stats.tx_bytes = adapter->stats.gotcl;
1703         adapter->net_stats.multicast = adapter->stats.mprcl;
1704         adapter->net_stats.collisions = 0;
1705
1706         /* ignore RLEC as it reports errors for padded (<64bytes) frames
1707          * with a length in the type/len field */
1708         adapter->net_stats.rx_errors =
1709             /* adapter->stats.rnbc + */ adapter->stats.crcerrs +
1710             adapter->stats.ruc +
1711             adapter->stats.roc /*+ adapter->stats.rlec */  +
1712             adapter->stats.icbc +
1713             adapter->stats.ecbc + adapter->stats.mpc;
1714
1715         /* see above
1716          * adapter->net_stats.rx_length_errors = adapter->stats.rlec;
1717          */
1718
1719         adapter->net_stats.rx_crc_errors = adapter->stats.crcerrs;
1720         adapter->net_stats.rx_fifo_errors = adapter->stats.mpc;
1721         adapter->net_stats.rx_missed_errors = adapter->stats.mpc;
1722         adapter->net_stats.rx_over_errors = adapter->stats.mpc;
1723
1724         adapter->net_stats.tx_errors = 0;
1725         adapter->net_stats.rx_frame_errors = 0;
1726         adapter->net_stats.tx_aborted_errors = 0;
1727         adapter->net_stats.tx_carrier_errors = 0;
1728         adapter->net_stats.tx_fifo_errors = 0;
1729         adapter->net_stats.tx_heartbeat_errors = 0;
1730         adapter->net_stats.tx_window_errors = 0;
1731 }
1732
1733 #define IXGB_MAX_INTR 10
1734 /**
1735  * ixgb_intr - Interrupt Handler
1736  * @irq: interrupt number
1737  * @data: pointer to a network interface device structure
1738  **/
1739
1740 static irqreturn_t
1741 ixgb_intr(int irq, void *data)
1742 {
1743         struct net_device *netdev = data;
1744         struct ixgb_adapter *adapter = netdev_priv(netdev);
1745         struct ixgb_hw *hw = &adapter->hw;
1746         uint32_t icr = IXGB_READ_REG(hw, ICR);
1747 #ifndef CONFIG_IXGB_NAPI
1748         unsigned int i;
1749 #endif
1750
1751         if(unlikely(!icr))
1752                 return IRQ_NONE;  /* Not our interrupt */
1753
1754         if(unlikely(icr & (IXGB_INT_RXSEQ | IXGB_INT_LSC))) {
1755                 mod_timer(&adapter->watchdog_timer, jiffies);
1756         }
1757
1758 #ifdef CONFIG_IXGB_NAPI
1759         if (netif_rx_schedule_prep(netdev, &adapter->napi)) {
1760
1761                 /* Disable interrupts and register for poll. The flush 
1762                   of the posted write is intentionally left out.
1763                 */
1764
1765                 atomic_inc(&adapter->irq_sem);
1766                 IXGB_WRITE_REG(&adapter->hw, IMC, ~0);
1767                 __netif_rx_schedule(netdev, &adapter->napi);
1768         }
1769 #else
1770         /* yes, that is actually a & and it is meant to make sure that
1771          * every pass through this for loop checks both receive and
1772          * transmit queues for completed descriptors, intended to
1773          * avoid starvation issues and assist tx/rx fairness. */
1774         for(i = 0; i < IXGB_MAX_INTR; i++)
1775                 if(!ixgb_clean_rx_irq(adapter) &
1776                    !ixgb_clean_tx_irq(adapter))
1777                         break;
1778 #endif 
1779         return IRQ_HANDLED;
1780 }
1781
1782 #ifdef CONFIG_IXGB_NAPI
1783 /**
1784  * ixgb_clean - NAPI Rx polling callback
1785  * @adapter: board private structure
1786  **/
1787
1788 static int
1789 ixgb_clean(struct napi_struct *napi, int budget)
1790 {
1791         struct ixgb_adapter *adapter = container_of(napi, struct ixgb_adapter, napi);
1792         struct net_device *netdev = adapter->netdev;
1793         int work_done = 0;
1794
1795         ixgb_clean_tx_irq(adapter);
1796         ixgb_clean_rx_irq(adapter, &work_done, budget);
1797
1798         /* If budget not fully consumed, exit the polling mode */
1799         if (work_done < budget) {
1800                 netif_rx_complete(netdev, napi);
1801                 ixgb_irq_enable(adapter);
1802         }
1803
1804         return work_done;
1805 }
1806 #endif
1807
1808 /**
1809  * ixgb_clean_tx_irq - Reclaim resources after transmit completes
1810  * @adapter: board private structure
1811  **/
1812
1813 static boolean_t
1814 ixgb_clean_tx_irq(struct ixgb_adapter *adapter)
1815 {
1816         struct ixgb_desc_ring *tx_ring = &adapter->tx_ring;
1817         struct net_device *netdev = adapter->netdev;
1818         struct ixgb_tx_desc *tx_desc, *eop_desc;
1819         struct ixgb_buffer *buffer_info;
1820         unsigned int i, eop;
1821         boolean_t cleaned = FALSE;
1822
1823         i = tx_ring->next_to_clean;
1824         eop = tx_ring->buffer_info[i].next_to_watch;
1825         eop_desc = IXGB_TX_DESC(*tx_ring, eop);
1826
1827         while(eop_desc->status & IXGB_TX_DESC_STATUS_DD) {
1828
1829                 for(cleaned = FALSE; !cleaned; ) {
1830                         tx_desc = IXGB_TX_DESC(*tx_ring, i);
1831                         buffer_info = &tx_ring->buffer_info[i];
1832
1833                         if (tx_desc->popts
1834                             & (IXGB_TX_DESC_POPTS_TXSM |
1835                                IXGB_TX_DESC_POPTS_IXSM))
1836                                 adapter->hw_csum_tx_good++;
1837
1838                         ixgb_unmap_and_free_tx_resource(adapter, buffer_info);
1839
1840                         *(uint32_t *)&(tx_desc->status) = 0;
1841
1842                         cleaned = (i == eop);
1843                         if(++i == tx_ring->count) i = 0;
1844                 }
1845
1846                 eop = tx_ring->buffer_info[i].next_to_watch;
1847                 eop_desc = IXGB_TX_DESC(*tx_ring, eop);
1848         }
1849
1850         tx_ring->next_to_clean = i;
1851
1852         if (unlikely(netif_queue_stopped(netdev))) {
1853                 spin_lock(&adapter->tx_lock);
1854                 if (netif_queue_stopped(netdev) && netif_carrier_ok(netdev) &&
1855                     (IXGB_DESC_UNUSED(tx_ring) >= DESC_NEEDED))
1856                         netif_wake_queue(netdev);
1857                 spin_unlock(&adapter->tx_lock);
1858         }
1859
1860         if(adapter->detect_tx_hung) {
1861                 /* detect a transmit hang in hardware, this serializes the
1862                  * check with the clearing of time_stamp and movement of i */
1863                 adapter->detect_tx_hung = FALSE;
1864                 if (tx_ring->buffer_info[eop].dma &&
1865                    time_after(jiffies, tx_ring->buffer_info[eop].time_stamp + HZ)
1866                    && !(IXGB_READ_REG(&adapter->hw, STATUS) &
1867                         IXGB_STATUS_TXOFF)) {
1868                         /* detected Tx unit hang */
1869                         DPRINTK(DRV, ERR, "Detected Tx Unit Hang\n"
1870                                         "  TDH                  <%x>\n"
1871                                         "  TDT                  <%x>\n"
1872                                         "  next_to_use          <%x>\n"
1873                                         "  next_to_clean        <%x>\n"
1874                                         "buffer_info[next_to_clean]\n"
1875                                         "  time_stamp           <%lx>\n"
1876                                         "  next_to_watch        <%x>\n"
1877                                         "  jiffies              <%lx>\n"
1878                                         "  next_to_watch.status <%x>\n",
1879                                 IXGB_READ_REG(&adapter->hw, TDH),
1880                                 IXGB_READ_REG(&adapter->hw, TDT),
1881                                 tx_ring->next_to_use,
1882                                 tx_ring->next_to_clean,
1883                                 tx_ring->buffer_info[eop].time_stamp,
1884                                 eop,
1885                                 jiffies,
1886                                 eop_desc->status);
1887                         netif_stop_queue(netdev);
1888                 }
1889         }
1890
1891         return cleaned;
1892 }
1893
1894 /**
1895  * ixgb_rx_checksum - Receive Checksum Offload for 82597.
1896  * @adapter: board private structure
1897  * @rx_desc: receive descriptor
1898  * @sk_buff: socket buffer with received data
1899  **/
1900
1901 static void
1902 ixgb_rx_checksum(struct ixgb_adapter *adapter,
1903                  struct ixgb_rx_desc *rx_desc,
1904                  struct sk_buff *skb)
1905 {
1906         /* Ignore Checksum bit is set OR
1907          * TCP Checksum has not been calculated
1908          */
1909         if((rx_desc->status & IXGB_RX_DESC_STATUS_IXSM) ||
1910            (!(rx_desc->status & IXGB_RX_DESC_STATUS_TCPCS))) {
1911                 skb->ip_summed = CHECKSUM_NONE;
1912                 return;
1913         }
1914
1915         /* At this point we know the hardware did the TCP checksum */
1916         /* now look at the TCP checksum error bit */
1917         if(rx_desc->errors & IXGB_RX_DESC_ERRORS_TCPE) {
1918                 /* let the stack verify checksum errors */
1919                 skb->ip_summed = CHECKSUM_NONE;
1920                 adapter->hw_csum_rx_error++;
1921         } else {
1922                 /* TCP checksum is good */
1923                 skb->ip_summed = CHECKSUM_UNNECESSARY;
1924                 adapter->hw_csum_rx_good++;
1925         }
1926 }
1927
1928 /**
1929  * ixgb_clean_rx_irq - Send received data up the network stack,
1930  * @adapter: board private structure
1931  **/
1932
1933 static boolean_t
1934 #ifdef CONFIG_IXGB_NAPI
1935 ixgb_clean_rx_irq(struct ixgb_adapter *adapter, int *work_done, int work_to_do)
1936 #else
1937 ixgb_clean_rx_irq(struct ixgb_adapter *adapter)
1938 #endif
1939 {
1940         struct ixgb_desc_ring *rx_ring = &adapter->rx_ring;
1941         struct net_device *netdev = adapter->netdev;
1942         struct pci_dev *pdev = adapter->pdev;
1943         struct ixgb_rx_desc *rx_desc, *next_rxd;
1944         struct ixgb_buffer *buffer_info, *next_buffer, *next2_buffer;
1945         uint32_t length;
1946         unsigned int i, j;
1947         boolean_t cleaned = FALSE;
1948
1949         i = rx_ring->next_to_clean;
1950         rx_desc = IXGB_RX_DESC(*rx_ring, i);
1951         buffer_info = &rx_ring->buffer_info[i];
1952
1953         while(rx_desc->status & IXGB_RX_DESC_STATUS_DD) {
1954                 struct sk_buff *skb, *next_skb;
1955                 u8 status;
1956
1957 #ifdef CONFIG_IXGB_NAPI
1958                 if(*work_done >= work_to_do)
1959                         break;
1960
1961                 (*work_done)++;
1962 #endif
1963                 status = rx_desc->status;
1964                 skb = buffer_info->skb;
1965                 buffer_info->skb = NULL;
1966
1967                 prefetch(skb->data);
1968
1969                 if(++i == rx_ring->count) i = 0;
1970                 next_rxd = IXGB_RX_DESC(*rx_ring, i);
1971                 prefetch(next_rxd);
1972
1973                 if((j = i + 1) == rx_ring->count) j = 0;
1974                 next2_buffer = &rx_ring->buffer_info[j];
1975                 prefetch(next2_buffer);
1976
1977                 next_buffer = &rx_ring->buffer_info[i];
1978                 next_skb = next_buffer->skb;
1979                 prefetch(next_skb);
1980
1981                 cleaned = TRUE;
1982
1983                 pci_unmap_single(pdev,
1984                                  buffer_info->dma,
1985                                  buffer_info->length,
1986                                  PCI_DMA_FROMDEVICE);
1987
1988                 length = le16_to_cpu(rx_desc->length);
1989
1990                 if(unlikely(!(status & IXGB_RX_DESC_STATUS_EOP))) {
1991
1992                         /* All receives must fit into a single buffer */
1993
1994                         IXGB_DBG("Receive packet consumed multiple buffers "
1995                                          "length<%x>\n", length);
1996
1997                         dev_kfree_skb_irq(skb);
1998                         goto rxdesc_done;
1999                 }
2000
2001                 if (unlikely(rx_desc->errors
2002                              & (IXGB_RX_DESC_ERRORS_CE | IXGB_RX_DESC_ERRORS_SE
2003                                 | IXGB_RX_DESC_ERRORS_P |
2004                                 IXGB_RX_DESC_ERRORS_RXE))) {
2005
2006                         dev_kfree_skb_irq(skb);
2007                         goto rxdesc_done;
2008                 }
2009
2010                 /* code added for copybreak, this should improve
2011                  * performance for small packets with large amounts
2012                  * of reassembly being done in the stack */
2013 #define IXGB_CB_LENGTH 256
2014                 if (length < IXGB_CB_LENGTH) {
2015                         struct sk_buff *new_skb =
2016                             netdev_alloc_skb(netdev, length + NET_IP_ALIGN);
2017                         if (new_skb) {
2018                                 skb_reserve(new_skb, NET_IP_ALIGN);
2019                                 skb_copy_to_linear_data_offset(new_skb,
2020                                                                -NET_IP_ALIGN,
2021                                                                (skb->data -
2022                                                                 NET_IP_ALIGN),
2023                                                                (length +
2024                                                                 NET_IP_ALIGN));
2025                                 /* save the skb in buffer_info as good */
2026                                 buffer_info->skb = skb;
2027                                 skb = new_skb;
2028                         }
2029                 }
2030                 /* end copybreak code */
2031
2032                 /* Good Receive */
2033                 skb_put(skb, length);
2034
2035                 /* Receive Checksum Offload */
2036                 ixgb_rx_checksum(adapter, rx_desc, skb);
2037
2038                 skb->protocol = eth_type_trans(skb, netdev);
2039 #ifdef CONFIG_IXGB_NAPI
2040                 if(adapter->vlgrp && (status & IXGB_RX_DESC_STATUS_VP)) {
2041                         vlan_hwaccel_receive_skb(skb, adapter->vlgrp,
2042                                 le16_to_cpu(rx_desc->special) &
2043                                         IXGB_RX_DESC_SPECIAL_VLAN_MASK);
2044                 } else {
2045                         netif_receive_skb(skb);
2046                 }
2047 #else /* CONFIG_IXGB_NAPI */
2048                 if(adapter->vlgrp && (status & IXGB_RX_DESC_STATUS_VP)) {
2049                         vlan_hwaccel_rx(skb, adapter->vlgrp,
2050                                 le16_to_cpu(rx_desc->special) &
2051                                         IXGB_RX_DESC_SPECIAL_VLAN_MASK);
2052                 } else {
2053                         netif_rx(skb);
2054                 }
2055 #endif /* CONFIG_IXGB_NAPI */
2056                 netdev->last_rx = jiffies;
2057
2058 rxdesc_done:
2059                 /* clean up descriptor, might be written over by hw */
2060                 rx_desc->status = 0;
2061
2062                 /* use prefetched values */
2063                 rx_desc = next_rxd;
2064                 buffer_info = next_buffer;
2065         }
2066
2067         rx_ring->next_to_clean = i;
2068
2069         ixgb_alloc_rx_buffers(adapter);
2070
2071         return cleaned;
2072 }
2073
2074 /**
2075  * ixgb_alloc_rx_buffers - Replace used receive buffers
2076  * @adapter: address of board private structure
2077  **/
2078
2079 static void
2080 ixgb_alloc_rx_buffers(struct ixgb_adapter *adapter)
2081 {
2082         struct ixgb_desc_ring *rx_ring = &adapter->rx_ring;
2083         struct net_device *netdev = adapter->netdev;
2084         struct pci_dev *pdev = adapter->pdev;
2085         struct ixgb_rx_desc *rx_desc;
2086         struct ixgb_buffer *buffer_info;
2087         struct sk_buff *skb;
2088         unsigned int i;
2089         int num_group_tail_writes;
2090         long cleancount;
2091
2092         i = rx_ring->next_to_use;
2093         buffer_info = &rx_ring->buffer_info[i];
2094         cleancount = IXGB_DESC_UNUSED(rx_ring);
2095
2096         num_group_tail_writes = IXGB_RX_BUFFER_WRITE;
2097
2098         /* leave three descriptors unused */
2099         while(--cleancount > 2) {
2100                 /* recycle! its good for you */
2101                 skb = buffer_info->skb;
2102                 if (skb) {
2103                         skb_trim(skb, 0);
2104                         goto map_skb;
2105                 }
2106
2107                 skb = netdev_alloc_skb(netdev, adapter->rx_buffer_len
2108                                        + NET_IP_ALIGN);
2109                 if (unlikely(!skb)) {
2110                         /* Better luck next round */
2111                         adapter->alloc_rx_buff_failed++;
2112                         break;
2113                 }
2114
2115                 /* Make buffer alignment 2 beyond a 16 byte boundary
2116                  * this will result in a 16 byte aligned IP header after
2117                  * the 14 byte MAC header is removed
2118                  */
2119                 skb_reserve(skb, NET_IP_ALIGN);
2120
2121                 buffer_info->skb = skb;
2122                 buffer_info->length = adapter->rx_buffer_len;
2123 map_skb:
2124                 buffer_info->dma = pci_map_single(pdev,
2125                                                   skb->data,
2126                                                   adapter->rx_buffer_len,
2127                                                   PCI_DMA_FROMDEVICE);
2128
2129                 rx_desc = IXGB_RX_DESC(*rx_ring, i);
2130                 rx_desc->buff_addr = cpu_to_le64(buffer_info->dma);
2131                 /* guarantee DD bit not set now before h/w gets descriptor
2132                  * this is the rest of the workaround for h/w double 
2133                  * writeback. */
2134                 rx_desc->status = 0;
2135
2136
2137                 if(++i == rx_ring->count) i = 0;
2138                 buffer_info = &rx_ring->buffer_info[i];
2139         }
2140
2141         if (likely(rx_ring->next_to_use != i)) {
2142                 rx_ring->next_to_use = i;
2143                 if (unlikely(i-- == 0))
2144                         i = (rx_ring->count - 1);
2145
2146                 /* Force memory writes to complete before letting h/w
2147                  * know there are new descriptors to fetch.  (Only
2148                  * applicable for weak-ordered memory model archs, such
2149                  * as IA-64). */
2150                 wmb();
2151                 IXGB_WRITE_REG(&adapter->hw, RDT, i);
2152         }
2153 }
2154
2155 /**
2156  * ixgb_vlan_rx_register - enables or disables vlan tagging/stripping.
2157  * 
2158  * @param netdev network interface device structure
2159  * @param grp indicates to enable or disable tagging/stripping
2160  **/
2161 static void
2162 ixgb_vlan_rx_register(struct net_device *netdev, struct vlan_group *grp)
2163 {
2164         struct ixgb_adapter *adapter = netdev_priv(netdev);
2165         uint32_t ctrl, rctl;
2166
2167         ixgb_irq_disable(adapter);
2168         adapter->vlgrp = grp;
2169
2170         if(grp) {
2171                 /* enable VLAN tag insert/strip */
2172                 ctrl = IXGB_READ_REG(&adapter->hw, CTRL0);
2173                 ctrl |= IXGB_CTRL0_VME;
2174                 IXGB_WRITE_REG(&adapter->hw, CTRL0, ctrl);
2175
2176                 /* enable VLAN receive filtering */
2177
2178                 rctl = IXGB_READ_REG(&adapter->hw, RCTL);
2179                 rctl |= IXGB_RCTL_VFE;
2180                 rctl &= ~IXGB_RCTL_CFIEN;
2181                 IXGB_WRITE_REG(&adapter->hw, RCTL, rctl);
2182         } else {
2183                 /* disable VLAN tag insert/strip */
2184
2185                 ctrl = IXGB_READ_REG(&adapter->hw, CTRL0);
2186                 ctrl &= ~IXGB_CTRL0_VME;
2187                 IXGB_WRITE_REG(&adapter->hw, CTRL0, ctrl);
2188
2189                 /* disable VLAN filtering */
2190
2191                 rctl = IXGB_READ_REG(&adapter->hw, RCTL);
2192                 rctl &= ~IXGB_RCTL_VFE;
2193                 IXGB_WRITE_REG(&adapter->hw, RCTL, rctl);
2194         }
2195
2196         ixgb_irq_enable(adapter);
2197 }
2198
2199 static void
2200 ixgb_vlan_rx_add_vid(struct net_device *netdev, uint16_t vid)
2201 {
2202         struct ixgb_adapter *adapter = netdev_priv(netdev);
2203         uint32_t vfta, index;
2204
2205         /* add VID to filter table */
2206
2207         index = (vid >> 5) & 0x7F;
2208         vfta = IXGB_READ_REG_ARRAY(&adapter->hw, VFTA, index);
2209         vfta |= (1 << (vid & 0x1F));
2210         ixgb_write_vfta(&adapter->hw, index, vfta);
2211 }
2212
2213 static void
2214 ixgb_vlan_rx_kill_vid(struct net_device *netdev, uint16_t vid)
2215 {
2216         struct ixgb_adapter *adapter = netdev_priv(netdev);
2217         uint32_t vfta, index;
2218
2219         ixgb_irq_disable(adapter);
2220
2221         vlan_group_set_device(adapter->vlgrp, vid, NULL);
2222
2223         ixgb_irq_enable(adapter);
2224
2225         /* remove VID from filter table*/
2226
2227         index = (vid >> 5) & 0x7F;
2228         vfta = IXGB_READ_REG_ARRAY(&adapter->hw, VFTA, index);
2229         vfta &= ~(1 << (vid & 0x1F));
2230         ixgb_write_vfta(&adapter->hw, index, vfta);
2231 }
2232
2233 static void
2234 ixgb_restore_vlan(struct ixgb_adapter *adapter)
2235 {
2236         ixgb_vlan_rx_register(adapter->netdev, adapter->vlgrp);
2237
2238         if(adapter->vlgrp) {
2239                 uint16_t vid;
2240                 for(vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
2241                         if(!vlan_group_get_device(adapter->vlgrp, vid))
2242                                 continue;
2243                         ixgb_vlan_rx_add_vid(adapter->netdev, vid);
2244                 }
2245         }
2246 }
2247
2248 #ifdef CONFIG_NET_POLL_CONTROLLER
2249 /*
2250  * Polling 'interrupt' - used by things like netconsole to send skbs
2251  * without having to re-enable interrupts. It's not called while
2252  * the interrupt routine is executing.
2253  */
2254
2255 static void ixgb_netpoll(struct net_device *dev)
2256 {
2257         struct ixgb_adapter *adapter = netdev_priv(dev);
2258
2259         disable_irq(adapter->pdev->irq);
2260         ixgb_intr(adapter->pdev->irq, dev);
2261         enable_irq(adapter->pdev->irq);
2262 }
2263 #endif
2264
2265 /**
2266  * ixgb_io_error_detected() - called when PCI error is detected
2267  * @pdev    pointer to pci device with error
2268  * @state   pci channel state after error
2269  *
2270  * This callback is called by the PCI subsystem whenever
2271  * a PCI bus error is detected.
2272  */
2273 static pci_ers_result_t ixgb_io_error_detected (struct pci_dev *pdev,
2274                                      enum pci_channel_state state)
2275 {
2276         struct net_device *netdev = pci_get_drvdata(pdev);
2277         struct ixgb_adapter *adapter = netdev_priv(netdev);
2278
2279         if(netif_running(netdev))
2280                 ixgb_down(adapter, TRUE);
2281
2282         pci_disable_device(pdev);
2283
2284         /* Request a slot reset. */
2285         return PCI_ERS_RESULT_NEED_RESET;
2286 }
2287
2288 /**
2289  * ixgb_io_slot_reset - called after the pci bus has been reset.
2290  * @pdev    pointer to pci device with error
2291  *
2292  * This callback is called after the PCI buss has been reset.
2293  * Basically, this tries to restart the card from scratch.
2294  * This is a shortened version of the device probe/discovery code,
2295  * it resembles the first-half of the ixgb_probe() routine.
2296  */
2297 static pci_ers_result_t ixgb_io_slot_reset (struct pci_dev *pdev)
2298 {
2299         struct net_device *netdev = pci_get_drvdata(pdev);
2300         struct ixgb_adapter *adapter = netdev_priv(netdev);
2301
2302         if(pci_enable_device(pdev)) {
2303                 DPRINTK(PROBE, ERR, "Cannot re-enable PCI device after reset.\n");
2304                 return PCI_ERS_RESULT_DISCONNECT;
2305         }
2306
2307         /* Perform card reset only on one instance of the card */
2308         if (0 != PCI_FUNC (pdev->devfn))
2309                 return PCI_ERS_RESULT_RECOVERED;
2310
2311         pci_set_master(pdev);
2312
2313         netif_carrier_off(netdev);
2314         netif_stop_queue(netdev);
2315         ixgb_reset(adapter);
2316
2317         /* Make sure the EEPROM is good */
2318         if(!ixgb_validate_eeprom_checksum(&adapter->hw)) {
2319                 DPRINTK(PROBE, ERR, "After reset, the EEPROM checksum is not valid.\n");
2320                 return PCI_ERS_RESULT_DISCONNECT;
2321         }
2322         ixgb_get_ee_mac_addr(&adapter->hw, netdev->dev_addr);
2323         memcpy(netdev->perm_addr, netdev->dev_addr, netdev->addr_len);
2324
2325         if(!is_valid_ether_addr(netdev->perm_addr)) {
2326                 DPRINTK(PROBE, ERR, "After reset, invalid MAC address.\n");
2327                 return PCI_ERS_RESULT_DISCONNECT;
2328         }
2329
2330         return PCI_ERS_RESULT_RECOVERED;
2331 }
2332
2333 /**
2334  * ixgb_io_resume - called when its OK to resume normal operations
2335  * @pdev    pointer to pci device with error
2336  *
2337  * The error recovery driver tells us that its OK to resume
2338  * normal operation. Implementation resembles the second-half
2339  * of the ixgb_probe() routine.
2340  */
2341 static void ixgb_io_resume (struct pci_dev *pdev)
2342 {
2343         struct net_device *netdev = pci_get_drvdata(pdev);
2344         struct ixgb_adapter *adapter = netdev_priv(netdev);
2345
2346         pci_set_master(pdev);
2347
2348         if(netif_running(netdev)) {
2349                 if(ixgb_up(adapter)) {
2350                         printk ("ixgb: can't bring device back up after reset\n");
2351                         return;
2352                 }
2353         }
2354
2355         netif_device_attach(netdev);
2356         mod_timer(&adapter->watchdog_timer, jiffies);
2357 }
2358
2359 /* ixgb_main.c */