a083a9189230d8f0a3dda97f7b3eba485862f1d1
[linux-2.6.git] / 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 char ixgb_driver_version[] = DRV_VERSION;
41 static 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 net_device *netdev, 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 void ixgb_set_ethtool_ops(struct net_device *netdev);
108 static void ixgb_tx_timeout(struct net_device *dev);
109 static void ixgb_tx_timeout_task(struct work_struct *work);
110 static void ixgb_vlan_rx_register(struct net_device *netdev,
111                                   struct vlan_group *grp);
112 static void ixgb_vlan_rx_add_vid(struct net_device *netdev, uint16_t vid);
113 static void ixgb_vlan_rx_kill_vid(struct net_device *netdev, uint16_t vid);
114 static void ixgb_restore_vlan(struct ixgb_adapter *adapter);
115
116 #ifdef CONFIG_NET_POLL_CONTROLLER
117 /* for netdump / net console */
118 static void ixgb_netpoll(struct net_device *dev);
119 #endif
120
121 static pci_ers_result_t ixgb_io_error_detected (struct pci_dev *pdev,
122                              enum pci_channel_state state);
123 static pci_ers_result_t ixgb_io_slot_reset (struct pci_dev *pdev);
124 static void ixgb_io_resume (struct pci_dev *pdev);
125
126 /* Exported from other modules */
127 extern void ixgb_check_options(struct ixgb_adapter *adapter);
128
129 static struct pci_error_handlers ixgb_err_handler = {
130         .error_detected = ixgb_io_error_detected,
131         .slot_reset = ixgb_io_slot_reset,
132         .resume = ixgb_io_resume,
133 };
134
135 static struct pci_driver ixgb_driver = {
136         .name     = ixgb_driver_name,
137         .id_table = ixgb_pci_tbl,
138         .probe    = ixgb_probe,
139         .remove   = __devexit_p(ixgb_remove),
140         .err_handler = &ixgb_err_handler
141 };
142
143 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
144 MODULE_DESCRIPTION("Intel(R) PRO/10GbE Network Driver");
145 MODULE_LICENSE("GPL");
146 MODULE_VERSION(DRV_VERSION);
147
148 #define DEFAULT_DEBUG_LEVEL_SHIFT 3
149 static int debug = DEFAULT_DEBUG_LEVEL_SHIFT;
150 module_param(debug, int, 0);
151 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
152
153 /* some defines for controlling descriptor fetches in h/w */
154 #define RXDCTL_WTHRESH_DEFAULT 15  /* chip writes back at this many or RXT0 */
155 #define RXDCTL_PTHRESH_DEFAULT 0   /* chip considers prefech below
156                                     * this */
157 #define RXDCTL_HTHRESH_DEFAULT 0   /* chip will only prefetch if tail
158                                     * is pushed this many descriptors
159                                     * from head */
160
161 /**
162  * ixgb_init_module - Driver Registration Routine
163  *
164  * ixgb_init_module is the first routine called when the driver is
165  * loaded. All it does is register with the PCI subsystem.
166  **/
167
168 static int __init
169 ixgb_init_module(void)
170 {
171         printk(KERN_INFO "%s - version %s\n",
172                ixgb_driver_string, ixgb_driver_version);
173
174         printk(KERN_INFO "%s\n", ixgb_copyright);
175
176         return pci_register_driver(&ixgb_driver);
177 }
178
179 module_init(ixgb_init_module);
180
181 /**
182  * ixgb_exit_module - Driver Exit Cleanup Routine
183  *
184  * ixgb_exit_module is called just before the driver is removed
185  * from memory.
186  **/
187
188 static void __exit
189 ixgb_exit_module(void)
190 {
191         pci_unregister_driver(&ixgb_driver);
192 }
193
194 module_exit(ixgb_exit_module);
195
196 /**
197  * ixgb_irq_disable - Mask off interrupt generation on the NIC
198  * @adapter: board private structure
199  **/
200
201 static void
202 ixgb_irq_disable(struct ixgb_adapter *adapter)
203 {
204         atomic_inc(&adapter->irq_sem);
205         IXGB_WRITE_REG(&adapter->hw, IMC, ~0);
206         IXGB_WRITE_FLUSH(&adapter->hw);
207         synchronize_irq(adapter->pdev->irq);
208 }
209
210 /**
211  * ixgb_irq_enable - Enable default interrupt generation settings
212  * @adapter: board private structure
213  **/
214
215 static void
216 ixgb_irq_enable(struct ixgb_adapter *adapter)
217 {
218         if(atomic_dec_and_test(&adapter->irq_sem)) {
219                 IXGB_WRITE_REG(&adapter->hw, IMS,
220                                IXGB_INT_RXT0 | IXGB_INT_RXDMT0 | IXGB_INT_TXDW |
221                                IXGB_INT_LSC);
222                 IXGB_WRITE_FLUSH(&adapter->hw);
223         }
224 }
225
226 int
227 ixgb_up(struct ixgb_adapter *adapter)
228 {
229         struct net_device *netdev = adapter->netdev;
230         int err;
231         int max_frame = netdev->mtu + ENET_HEADER_SIZE + ENET_FCS_LENGTH;
232         struct ixgb_hw *hw = &adapter->hw;
233
234         /* hardware has been reset, we need to reload some things */
235
236         ixgb_rar_set(hw, netdev->dev_addr, 0);
237         ixgb_set_multi(netdev);
238
239         ixgb_restore_vlan(adapter);
240
241         ixgb_configure_tx(adapter);
242         ixgb_setup_rctl(adapter);
243         ixgb_configure_rx(adapter);
244         ixgb_alloc_rx_buffers(adapter);
245
246         /* disable interrupts and get the hardware into a known state */
247         IXGB_WRITE_REG(&adapter->hw, IMC, 0xffffffff);
248
249 #ifdef CONFIG_PCI_MSI
250         {
251         boolean_t pcix = (IXGB_READ_REG(&adapter->hw, STATUS) & 
252                                                   IXGB_STATUS_PCIX_MODE) ? TRUE : FALSE;
253         adapter->have_msi = TRUE;
254
255         if (!pcix)
256            adapter->have_msi = FALSE;
257         else if((err = pci_enable_msi(adapter->pdev))) {
258                 DPRINTK(PROBE, ERR,
259                  "Unable to allocate MSI interrupt Error: %d\n", err);
260                 adapter->have_msi = FALSE;
261                 /* proceed to try to request regular interrupt */
262         }
263         }
264
265 #endif
266         if((err = request_irq(adapter->pdev->irq, &ixgb_intr,
267                                   IRQF_SHARED | IRQF_SAMPLE_RANDOM,
268                                   netdev->name, netdev))) {
269                 DPRINTK(PROBE, ERR,
270                  "Unable to allocate interrupt Error: %d\n", err);
271                 return err;
272         }
273
274         if((hw->max_frame_size != max_frame) ||
275                 (hw->max_frame_size !=
276                 (IXGB_READ_REG(hw, MFS) >> IXGB_MFS_SHIFT))) {
277
278                 hw->max_frame_size = max_frame;
279
280                 IXGB_WRITE_REG(hw, MFS, hw->max_frame_size << IXGB_MFS_SHIFT);
281
282                 if(hw->max_frame_size >
283                    IXGB_MAX_ENET_FRAME_SIZE_WITHOUT_FCS + ENET_FCS_LENGTH) {
284                         uint32_t ctrl0 = IXGB_READ_REG(hw, CTRL0);
285
286                         if(!(ctrl0 & IXGB_CTRL0_JFE)) {
287                                 ctrl0 |= IXGB_CTRL0_JFE;
288                                 IXGB_WRITE_REG(hw, CTRL0, ctrl0);
289                         }
290                 }
291         }
292
293         mod_timer(&adapter->watchdog_timer, jiffies);
294
295 #ifdef CONFIG_IXGB_NAPI
296         netif_poll_enable(netdev);
297 #endif
298         ixgb_irq_enable(adapter);
299
300         return 0;
301 }
302
303 void
304 ixgb_down(struct ixgb_adapter *adapter, boolean_t kill_watchdog)
305 {
306         struct net_device *netdev = adapter->netdev;
307
308         ixgb_irq_disable(adapter);
309         free_irq(adapter->pdev->irq, netdev);
310 #ifdef CONFIG_PCI_MSI
311         if(adapter->have_msi == TRUE)
312                 pci_disable_msi(adapter->pdev);
313
314 #endif
315         if(kill_watchdog)
316                 del_timer_sync(&adapter->watchdog_timer);
317 #ifdef CONFIG_IXGB_NAPI
318         netif_poll_disable(netdev);
319 #endif
320         adapter->link_speed = 0;
321         adapter->link_duplex = 0;
322         netif_carrier_off(netdev);
323         netif_stop_queue(netdev);
324
325         ixgb_reset(adapter);
326         ixgb_clean_tx_ring(adapter);
327         ixgb_clean_rx_ring(adapter);
328 }
329
330 void
331 ixgb_reset(struct ixgb_adapter *adapter)
332 {
333
334         ixgb_adapter_stop(&adapter->hw);
335         if(!ixgb_init_hw(&adapter->hw))
336                 DPRINTK(PROBE, ERR, "ixgb_init_hw failed.\n");
337 }
338
339 /**
340  * ixgb_probe - Device Initialization Routine
341  * @pdev: PCI device information struct
342  * @ent: entry in ixgb_pci_tbl
343  *
344  * Returns 0 on success, negative on failure
345  *
346  * ixgb_probe initializes an adapter identified by a pci_dev structure.
347  * The OS initialization, configuring of the adapter private structure,
348  * and a hardware reset occur.
349  **/
350
351 static int __devinit
352 ixgb_probe(struct pci_dev *pdev,
353                 const struct pci_device_id *ent)
354 {
355         struct net_device *netdev = NULL;
356         struct ixgb_adapter *adapter;
357         static int cards_found = 0;
358         unsigned long mmio_start;
359         int mmio_len;
360         int pci_using_dac;
361         int i;
362         int err;
363
364         if((err = pci_enable_device(pdev)))
365                 return err;
366
367         if(!(err = pci_set_dma_mask(pdev, DMA_64BIT_MASK)) &&
368            !(err = pci_set_consistent_dma_mask(pdev, DMA_64BIT_MASK))) {
369                 pci_using_dac = 1;
370         } else {
371                 if((err = pci_set_dma_mask(pdev, DMA_32BIT_MASK)) ||
372                    (err = pci_set_consistent_dma_mask(pdev, DMA_32BIT_MASK))) {
373                         printk(KERN_ERR
374                          "ixgb: No usable DMA configuration, aborting\n");
375                         goto err_dma_mask;
376                 }
377                 pci_using_dac = 0;
378         }
379
380         if((err = pci_request_regions(pdev, ixgb_driver_name)))
381                 goto err_request_regions;
382
383         pci_set_master(pdev);
384
385         netdev = alloc_etherdev(sizeof(struct ixgb_adapter));
386         if(!netdev) {
387                 err = -ENOMEM;
388                 goto err_alloc_etherdev;
389         }
390
391         SET_MODULE_OWNER(netdev);
392         SET_NETDEV_DEV(netdev, &pdev->dev);
393
394         pci_set_drvdata(pdev, netdev);
395         adapter = netdev_priv(netdev);
396         adapter->netdev = netdev;
397         adapter->pdev = pdev;
398         adapter->hw.back = adapter;
399         adapter->msg_enable = netif_msg_init(debug, DEFAULT_DEBUG_LEVEL_SHIFT);
400
401         mmio_start = pci_resource_start(pdev, BAR_0);
402         mmio_len = pci_resource_len(pdev, BAR_0);
403
404         adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
405         if(!adapter->hw.hw_addr) {
406                 err = -EIO;
407                 goto err_ioremap;
408         }
409
410         for(i = BAR_1; i <= BAR_5; i++) {
411                 if(pci_resource_len(pdev, i) == 0)
412                         continue;
413                 if(pci_resource_flags(pdev, i) & IORESOURCE_IO) {
414                         adapter->hw.io_base = pci_resource_start(pdev, i);
415                         break;
416                 }
417         }
418
419         netdev->open = &ixgb_open;
420         netdev->stop = &ixgb_close;
421         netdev->hard_start_xmit = &ixgb_xmit_frame;
422         netdev->get_stats = &ixgb_get_stats;
423         netdev->set_multicast_list = &ixgb_set_multi;
424         netdev->set_mac_address = &ixgb_set_mac;
425         netdev->change_mtu = &ixgb_change_mtu;
426         ixgb_set_ethtool_ops(netdev);
427         netdev->tx_timeout = &ixgb_tx_timeout;
428         netdev->watchdog_timeo = 5 * HZ;
429 #ifdef CONFIG_IXGB_NAPI
430         netdev->poll = &ixgb_clean;
431         netdev->weight = 64;
432 #endif
433         netdev->vlan_rx_register = ixgb_vlan_rx_register;
434         netdev->vlan_rx_add_vid = ixgb_vlan_rx_add_vid;
435         netdev->vlan_rx_kill_vid = ixgb_vlan_rx_kill_vid;
436 #ifdef CONFIG_NET_POLL_CONTROLLER
437         netdev->poll_controller = ixgb_netpoll;
438 #endif
439
440         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
441         netdev->mem_start = mmio_start;
442         netdev->mem_end = mmio_start + mmio_len;
443         netdev->base_addr = adapter->hw.io_base;
444
445         adapter->bd_number = cards_found;
446         adapter->link_speed = 0;
447         adapter->link_duplex = 0;
448
449         /* setup the private structure */
450
451         if((err = ixgb_sw_init(adapter)))
452                 goto err_sw_init;
453
454         netdev->features = NETIF_F_SG |
455                            NETIF_F_HW_CSUM |
456                            NETIF_F_HW_VLAN_TX |
457                            NETIF_F_HW_VLAN_RX |
458                            NETIF_F_HW_VLAN_FILTER;
459 #ifdef NETIF_F_TSO
460         netdev->features |= NETIF_F_TSO;
461 #endif
462 #ifdef NETIF_F_LLTX
463         netdev->features |= NETIF_F_LLTX;
464 #endif
465
466         if(pci_using_dac)
467                 netdev->features |= NETIF_F_HIGHDMA;
468
469         /* make sure the EEPROM is good */
470
471         if(!ixgb_validate_eeprom_checksum(&adapter->hw)) {
472                 DPRINTK(PROBE, ERR, "The EEPROM Checksum Is Not Valid\n");
473                 err = -EIO;
474                 goto err_eeprom;
475         }
476
477         ixgb_get_ee_mac_addr(&adapter->hw, netdev->dev_addr);
478         memcpy(netdev->perm_addr, netdev->dev_addr, netdev->addr_len);
479
480         if(!is_valid_ether_addr(netdev->perm_addr)) {
481                 DPRINTK(PROBE, ERR, "Invalid MAC Address\n");
482                 err = -EIO;
483                 goto err_eeprom;
484         }
485
486         adapter->part_num = ixgb_get_ee_pba_number(&adapter->hw);
487
488         init_timer(&adapter->watchdog_timer);
489         adapter->watchdog_timer.function = &ixgb_watchdog;
490         adapter->watchdog_timer.data = (unsigned long)adapter;
491
492         INIT_WORK(&adapter->tx_timeout_task, ixgb_tx_timeout_task);
493
494         strcpy(netdev->name, "eth%d");
495         if((err = register_netdev(netdev)))
496                 goto err_register;
497
498         /* we're going to reset, so assume we have no link for now */
499
500         netif_carrier_off(netdev);
501         netif_stop_queue(netdev);
502
503         DPRINTK(PROBE, INFO, "Intel(R) PRO/10GbE Network Connection\n");
504         ixgb_check_options(adapter);
505         /* reset the hardware with the new settings */
506
507         ixgb_reset(adapter);
508
509         cards_found++;
510         return 0;
511
512 err_register:
513 err_sw_init:
514 err_eeprom:
515         iounmap(adapter->hw.hw_addr);
516 err_ioremap:
517         free_netdev(netdev);
518 err_alloc_etherdev:
519         pci_release_regions(pdev);
520 err_request_regions:
521 err_dma_mask:
522         pci_disable_device(pdev);
523         return err;
524 }
525
526 /**
527  * ixgb_remove - Device Removal Routine
528  * @pdev: PCI device information struct
529  *
530  * ixgb_remove is called by the PCI subsystem to alert the driver
531  * that it should release a PCI device.  The could be caused by a
532  * Hot-Plug event, or because the driver is going to be removed from
533  * memory.
534  **/
535
536 static void __devexit
537 ixgb_remove(struct pci_dev *pdev)
538 {
539         struct net_device *netdev = pci_get_drvdata(pdev);
540         struct ixgb_adapter *adapter = netdev_priv(netdev);
541
542         unregister_netdev(netdev);
543
544         iounmap(adapter->hw.hw_addr);
545         pci_release_regions(pdev);
546
547         free_netdev(netdev);
548 }
549
550 /**
551  * ixgb_sw_init - Initialize general software structures (struct ixgb_adapter)
552  * @adapter: board private structure to initialize
553  *
554  * ixgb_sw_init initializes the Adapter private data structure.
555  * Fields are initialized based on PCI device information and
556  * OS network device settings (MTU size).
557  **/
558
559 static int __devinit
560 ixgb_sw_init(struct ixgb_adapter *adapter)
561 {
562         struct ixgb_hw *hw = &adapter->hw;
563         struct net_device *netdev = adapter->netdev;
564         struct pci_dev *pdev = adapter->pdev;
565
566         /* PCI config space info */
567
568         hw->vendor_id = pdev->vendor;
569         hw->device_id = pdev->device;
570         hw->subsystem_vendor_id = pdev->subsystem_vendor;
571         hw->subsystem_id = pdev->subsystem_device;
572
573         hw->max_frame_size = netdev->mtu + ENET_HEADER_SIZE + ENET_FCS_LENGTH;
574         adapter->rx_buffer_len = hw->max_frame_size;
575
576         if((hw->device_id == IXGB_DEVICE_ID_82597EX)
577            || (hw->device_id == IXGB_DEVICE_ID_82597EX_CX4)
578            || (hw->device_id == IXGB_DEVICE_ID_82597EX_LR)
579            || (hw->device_id == IXGB_DEVICE_ID_82597EX_SR))
580                         hw->mac_type = ixgb_82597;
581         else {
582                 /* should never have loaded on this device */
583                 DPRINTK(PROBE, ERR, "unsupported device id\n");
584         }
585
586         /* enable flow control to be programmed */
587         hw->fc.send_xon = 1;
588
589         atomic_set(&adapter->irq_sem, 1);
590         spin_lock_init(&adapter->tx_lock);
591
592         return 0;
593 }
594
595 /**
596  * ixgb_open - Called when a network interface is made active
597  * @netdev: network interface device structure
598  *
599  * Returns 0 on success, negative value on failure
600  *
601  * The open entry point is called when a network interface is made
602  * active by the system (IFF_UP).  At this point all resources needed
603  * for transmit and receive operations are allocated, the interrupt
604  * handler is registered with the OS, the watchdog timer is started,
605  * and the stack is notified that the interface is ready.
606  **/
607
608 static int
609 ixgb_open(struct net_device *netdev)
610 {
611         struct ixgb_adapter *adapter = netdev_priv(netdev);
612         int err;
613
614         /* allocate transmit descriptors */
615
616         if((err = ixgb_setup_tx_resources(adapter)))
617                 goto err_setup_tx;
618
619         /* allocate receive descriptors */
620
621         if((err = ixgb_setup_rx_resources(adapter)))
622                 goto err_setup_rx;
623
624         if((err = ixgb_up(adapter)))
625                 goto err_up;
626
627         return 0;
628
629 err_up:
630         ixgb_free_rx_resources(adapter);
631 err_setup_rx:
632         ixgb_free_tx_resources(adapter);
633 err_setup_tx:
634         ixgb_reset(adapter);
635
636         return err;
637 }
638
639 /**
640  * ixgb_close - Disables a network interface
641  * @netdev: network interface device structure
642  *
643  * Returns 0, this is not allowed to fail
644  *
645  * The close entry point is called when an interface is de-activated
646  * by the OS.  The hardware is still under the drivers control, but
647  * needs to be disabled.  A global MAC reset is issued to stop the
648  * hardware, and all transmit and receive resources are freed.
649  **/
650
651 static int
652 ixgb_close(struct net_device *netdev)
653 {
654         struct ixgb_adapter *adapter = netdev_priv(netdev);
655
656         ixgb_down(adapter, TRUE);
657
658         ixgb_free_tx_resources(adapter);
659         ixgb_free_rx_resources(adapter);
660
661         return 0;
662 }
663
664 /**
665  * ixgb_setup_tx_resources - allocate Tx resources (Descriptors)
666  * @adapter: board private structure
667  *
668  * Return 0 on success, negative on failure
669  **/
670
671 int
672 ixgb_setup_tx_resources(struct ixgb_adapter *adapter)
673 {
674         struct ixgb_desc_ring *txdr = &adapter->tx_ring;
675         struct pci_dev *pdev = adapter->pdev;
676         int size;
677
678         size = sizeof(struct ixgb_buffer) * txdr->count;
679         txdr->buffer_info = vmalloc(size);
680         if(!txdr->buffer_info) {
681                 DPRINTK(PROBE, ERR,
682                  "Unable to allocate transmit descriptor ring memory\n");
683                 return -ENOMEM;
684         }
685         memset(txdr->buffer_info, 0, size);
686
687         /* round up to nearest 4K */
688
689         txdr->size = txdr->count * sizeof(struct ixgb_tx_desc);
690         IXGB_ROUNDUP(txdr->size, 4096);
691
692         txdr->desc = pci_alloc_consistent(pdev, txdr->size, &txdr->dma);
693         if(!txdr->desc) {
694                 vfree(txdr->buffer_info);
695                 DPRINTK(PROBE, ERR,
696                  "Unable to allocate transmit descriptor memory\n");
697                 return -ENOMEM;
698         }
699         memset(txdr->desc, 0, txdr->size);
700
701         txdr->next_to_use = 0;
702         txdr->next_to_clean = 0;
703
704         return 0;
705 }
706
707 /**
708  * ixgb_configure_tx - Configure 82597 Transmit Unit after Reset.
709  * @adapter: board private structure
710  *
711  * Configure the Tx unit of the MAC after a reset.
712  **/
713
714 static void
715 ixgb_configure_tx(struct ixgb_adapter *adapter)
716 {
717         uint64_t tdba = adapter->tx_ring.dma;
718         uint32_t tdlen = adapter->tx_ring.count * sizeof(struct ixgb_tx_desc);
719         uint32_t tctl;
720         struct ixgb_hw *hw = &adapter->hw;
721
722         /* Setup the Base and Length of the Tx Descriptor Ring 
723          * tx_ring.dma can be either a 32 or 64 bit value 
724          */
725
726         IXGB_WRITE_REG(hw, TDBAL, (tdba & 0x00000000ffffffffULL));
727         IXGB_WRITE_REG(hw, TDBAH, (tdba >> 32));
728
729         IXGB_WRITE_REG(hw, TDLEN, tdlen);
730
731         /* Setup the HW Tx Head and Tail descriptor pointers */
732
733         IXGB_WRITE_REG(hw, TDH, 0);
734         IXGB_WRITE_REG(hw, TDT, 0);
735
736         /* don't set up txdctl, it induces performance problems if configured
737          * incorrectly */
738         /* Set the Tx Interrupt Delay register */
739
740         IXGB_WRITE_REG(hw, TIDV, adapter->tx_int_delay);
741
742         /* Program the Transmit Control Register */
743
744         tctl = IXGB_TCTL_TCE | IXGB_TCTL_TXEN | IXGB_TCTL_TPDE;
745         IXGB_WRITE_REG(hw, TCTL, tctl);
746
747         /* Setup Transmit Descriptor Settings for this adapter */
748         adapter->tx_cmd_type =
749                 IXGB_TX_DESC_TYPE 
750                 | (adapter->tx_int_delay_enable ? IXGB_TX_DESC_CMD_IDE : 0);
751 }
752
753 /**
754  * ixgb_setup_rx_resources - allocate Rx resources (Descriptors)
755  * @adapter: board private structure
756  *
757  * Returns 0 on success, negative on failure
758  **/
759
760 int
761 ixgb_setup_rx_resources(struct ixgb_adapter *adapter)
762 {
763         struct ixgb_desc_ring *rxdr = &adapter->rx_ring;
764         struct pci_dev *pdev = adapter->pdev;
765         int size;
766
767         size = sizeof(struct ixgb_buffer) * rxdr->count;
768         rxdr->buffer_info = vmalloc(size);
769         if(!rxdr->buffer_info) {
770                 DPRINTK(PROBE, ERR,
771                  "Unable to allocate receive descriptor ring\n");
772                 return -ENOMEM;
773         }
774         memset(rxdr->buffer_info, 0, size);
775
776         /* Round up to nearest 4K */
777
778         rxdr->size = rxdr->count * sizeof(struct ixgb_rx_desc);
779         IXGB_ROUNDUP(rxdr->size, 4096);
780
781         rxdr->desc = pci_alloc_consistent(pdev, rxdr->size, &rxdr->dma);
782
783         if(!rxdr->desc) {
784                 vfree(rxdr->buffer_info);
785                 DPRINTK(PROBE, ERR,
786                  "Unable to allocate receive descriptors\n");
787                 return -ENOMEM;
788         }
789         memset(rxdr->desc, 0, rxdr->size);
790
791         rxdr->next_to_clean = 0;
792         rxdr->next_to_use = 0;
793
794         return 0;
795 }
796
797 /**
798  * ixgb_setup_rctl - configure the receive control register
799  * @adapter: Board private structure
800  **/
801
802 static void
803 ixgb_setup_rctl(struct ixgb_adapter *adapter)
804 {
805         uint32_t rctl;
806
807         rctl = IXGB_READ_REG(&adapter->hw, RCTL);
808
809         rctl &= ~(3 << IXGB_RCTL_MO_SHIFT);
810
811         rctl |=
812                 IXGB_RCTL_BAM | IXGB_RCTL_RDMTS_1_2 | 
813                 IXGB_RCTL_RXEN | IXGB_RCTL_CFF | 
814                 (adapter->hw.mc_filter_type << IXGB_RCTL_MO_SHIFT);
815
816         rctl |= IXGB_RCTL_SECRC;
817
818         if (adapter->rx_buffer_len <= IXGB_RXBUFFER_2048)
819                 rctl |= IXGB_RCTL_BSIZE_2048;
820         else if (adapter->rx_buffer_len <= IXGB_RXBUFFER_4096)
821                 rctl |= IXGB_RCTL_BSIZE_4096;
822         else if (adapter->rx_buffer_len <= IXGB_RXBUFFER_8192)
823                 rctl |= IXGB_RCTL_BSIZE_8192;
824         else if (adapter->rx_buffer_len <= IXGB_RXBUFFER_16384)
825                 rctl |= IXGB_RCTL_BSIZE_16384;
826
827         IXGB_WRITE_REG(&adapter->hw, RCTL, rctl);
828 }
829
830 /**
831  * ixgb_configure_rx - Configure 82597 Receive Unit after Reset.
832  * @adapter: board private structure
833  *
834  * Configure the Rx unit of the MAC after a reset.
835  **/
836
837 static void
838 ixgb_configure_rx(struct ixgb_adapter *adapter)
839 {
840         uint64_t rdba = adapter->rx_ring.dma;
841         uint32_t rdlen = adapter->rx_ring.count * sizeof(struct ixgb_rx_desc);
842         struct ixgb_hw *hw = &adapter->hw;
843         uint32_t rctl;
844         uint32_t rxcsum;
845         uint32_t rxdctl;
846
847         /* make sure receives are disabled while setting up the descriptors */
848
849         rctl = IXGB_READ_REG(hw, RCTL);
850         IXGB_WRITE_REG(hw, RCTL, rctl & ~IXGB_RCTL_RXEN);
851
852         /* set the Receive Delay Timer Register */
853
854         IXGB_WRITE_REG(hw, RDTR, adapter->rx_int_delay);
855
856         /* Setup the Base and Length of the Rx Descriptor Ring */
857
858         IXGB_WRITE_REG(hw, RDBAL, (rdba & 0x00000000ffffffffULL));
859         IXGB_WRITE_REG(hw, RDBAH, (rdba >> 32));
860
861         IXGB_WRITE_REG(hw, RDLEN, rdlen);
862
863         /* Setup the HW Rx Head and Tail Descriptor Pointers */
864         IXGB_WRITE_REG(hw, RDH, 0);
865         IXGB_WRITE_REG(hw, RDT, 0);
866
867         /* set up pre-fetching of receive buffers so we get some before we
868          * run out (default hardware behavior is to run out before fetching
869          * more).  This sets up to fetch if HTHRESH rx descriptors are avail
870          * and the descriptors in hw cache are below PTHRESH.  This avoids
871          * the hardware behavior of fetching <=512 descriptors in a single
872          * burst that pre-empts all other activity, usually causing fifo
873          * overflows. */
874         /* use WTHRESH to burst write 16 descriptors or burst when RXT0 */
875         rxdctl = RXDCTL_WTHRESH_DEFAULT << IXGB_RXDCTL_WTHRESH_SHIFT |
876                  RXDCTL_HTHRESH_DEFAULT << IXGB_RXDCTL_HTHRESH_SHIFT |
877                  RXDCTL_PTHRESH_DEFAULT << IXGB_RXDCTL_PTHRESH_SHIFT;
878         IXGB_WRITE_REG(hw, RXDCTL, rxdctl);
879
880         /* Enable Receive Checksum Offload for TCP and UDP */
881         if(adapter->rx_csum == TRUE) {
882                 rxcsum = IXGB_READ_REG(hw, RXCSUM);
883                 rxcsum |= IXGB_RXCSUM_TUOFL;
884                 IXGB_WRITE_REG(hw, RXCSUM, rxcsum);
885         }
886
887         /* Enable Receives */
888
889         IXGB_WRITE_REG(hw, RCTL, rctl);
890 }
891
892 /**
893  * ixgb_free_tx_resources - Free Tx Resources
894  * @adapter: board private structure
895  *
896  * Free all transmit software resources
897  **/
898
899 void
900 ixgb_free_tx_resources(struct ixgb_adapter *adapter)
901 {
902         struct pci_dev *pdev = adapter->pdev;
903
904         ixgb_clean_tx_ring(adapter);
905
906         vfree(adapter->tx_ring.buffer_info);
907         adapter->tx_ring.buffer_info = NULL;
908
909         pci_free_consistent(pdev, adapter->tx_ring.size,
910                             adapter->tx_ring.desc, adapter->tx_ring.dma);
911
912         adapter->tx_ring.desc = NULL;
913 }
914
915 static void
916 ixgb_unmap_and_free_tx_resource(struct ixgb_adapter *adapter,
917                                         struct ixgb_buffer *buffer_info)
918 {
919         struct pci_dev *pdev = adapter->pdev;
920
921         if (buffer_info->dma)
922                 pci_unmap_page(pdev, buffer_info->dma, buffer_info->length,
923                                PCI_DMA_TODEVICE);
924
925         if (buffer_info->skb)
926                 dev_kfree_skb_any(buffer_info->skb);
927
928         buffer_info->skb = NULL;
929         buffer_info->dma = 0;
930         buffer_info->time_stamp = 0;
931         /* these fields must always be initialized in tx
932          * buffer_info->length = 0;
933          * buffer_info->next_to_watch = 0; */
934 }
935
936 /**
937  * ixgb_clean_tx_ring - Free Tx Buffers
938  * @adapter: board private structure
939  **/
940
941 static void
942 ixgb_clean_tx_ring(struct ixgb_adapter *adapter)
943 {
944         struct ixgb_desc_ring *tx_ring = &adapter->tx_ring;
945         struct ixgb_buffer *buffer_info;
946         unsigned long size;
947         unsigned int i;
948
949         /* Free all the Tx ring sk_buffs */
950
951         for(i = 0; i < tx_ring->count; i++) {
952                 buffer_info = &tx_ring->buffer_info[i];
953                 ixgb_unmap_and_free_tx_resource(adapter, buffer_info);
954         }
955
956         size = sizeof(struct ixgb_buffer) * tx_ring->count;
957         memset(tx_ring->buffer_info, 0, size);
958
959         /* Zero out the descriptor ring */
960
961         memset(tx_ring->desc, 0, tx_ring->size);
962
963         tx_ring->next_to_use = 0;
964         tx_ring->next_to_clean = 0;
965
966         IXGB_WRITE_REG(&adapter->hw, TDH, 0);
967         IXGB_WRITE_REG(&adapter->hw, TDT, 0);
968 }
969
970 /**
971  * ixgb_free_rx_resources - Free Rx Resources
972  * @adapter: board private structure
973  *
974  * Free all receive software resources
975  **/
976
977 void
978 ixgb_free_rx_resources(struct ixgb_adapter *adapter)
979 {
980         struct ixgb_desc_ring *rx_ring = &adapter->rx_ring;
981         struct pci_dev *pdev = adapter->pdev;
982
983         ixgb_clean_rx_ring(adapter);
984
985         vfree(rx_ring->buffer_info);
986         rx_ring->buffer_info = NULL;
987
988         pci_free_consistent(pdev, rx_ring->size, rx_ring->desc, rx_ring->dma);
989
990         rx_ring->desc = NULL;
991 }
992
993 /**
994  * ixgb_clean_rx_ring - Free Rx Buffers
995  * @adapter: board private structure
996  **/
997
998 static void
999 ixgb_clean_rx_ring(struct ixgb_adapter *adapter)
1000 {
1001         struct ixgb_desc_ring *rx_ring = &adapter->rx_ring;
1002         struct ixgb_buffer *buffer_info;
1003         struct pci_dev *pdev = adapter->pdev;
1004         unsigned long size;
1005         unsigned int i;
1006
1007         /* Free all the Rx ring sk_buffs */
1008
1009         for(i = 0; i < rx_ring->count; i++) {
1010                 buffer_info = &rx_ring->buffer_info[i];
1011                 if(buffer_info->skb) {
1012
1013                         pci_unmap_single(pdev,
1014                                          buffer_info->dma,
1015                                          buffer_info->length,
1016                                          PCI_DMA_FROMDEVICE);
1017
1018                         dev_kfree_skb(buffer_info->skb);
1019
1020                         buffer_info->skb = NULL;
1021                 }
1022         }
1023
1024         size = sizeof(struct ixgb_buffer) * rx_ring->count;
1025         memset(rx_ring->buffer_info, 0, size);
1026
1027         /* Zero out the descriptor ring */
1028
1029         memset(rx_ring->desc, 0, rx_ring->size);
1030
1031         rx_ring->next_to_clean = 0;
1032         rx_ring->next_to_use = 0;
1033
1034         IXGB_WRITE_REG(&adapter->hw, RDH, 0);
1035         IXGB_WRITE_REG(&adapter->hw, RDT, 0);
1036 }
1037
1038 /**
1039  * ixgb_set_mac - Change the Ethernet Address of the NIC
1040  * @netdev: network interface device structure
1041  * @p: pointer to an address structure
1042  *
1043  * Returns 0 on success, negative on failure
1044  **/
1045
1046 static int
1047 ixgb_set_mac(struct net_device *netdev, void *p)
1048 {
1049         struct ixgb_adapter *adapter = netdev_priv(netdev);
1050         struct sockaddr *addr = p;
1051
1052         if(!is_valid_ether_addr(addr->sa_data))
1053                 return -EADDRNOTAVAIL;
1054
1055         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
1056
1057         ixgb_rar_set(&adapter->hw, addr->sa_data, 0);
1058
1059         return 0;
1060 }
1061
1062 /**
1063  * ixgb_set_multi - Multicast and Promiscuous mode set
1064  * @netdev: network interface device structure
1065  *
1066  * The set_multi entry point is called whenever the multicast address
1067  * list or the network interface flags are updated.  This routine is
1068  * responsible for configuring the hardware for proper multicast,
1069  * promiscuous mode, and all-multi behavior.
1070  **/
1071
1072 static void
1073 ixgb_set_multi(struct net_device *netdev)
1074 {
1075         struct ixgb_adapter *adapter = netdev_priv(netdev);
1076         struct ixgb_hw *hw = &adapter->hw;
1077         struct dev_mc_list *mc_ptr;
1078         uint32_t rctl;
1079         int i;
1080
1081         /* Check for Promiscuous and All Multicast modes */
1082
1083         rctl = IXGB_READ_REG(hw, RCTL);
1084
1085         if(netdev->flags & IFF_PROMISC) {
1086                 rctl |= (IXGB_RCTL_UPE | IXGB_RCTL_MPE);
1087         } else if(netdev->flags & IFF_ALLMULTI) {
1088                 rctl |= IXGB_RCTL_MPE;
1089                 rctl &= ~IXGB_RCTL_UPE;
1090         } else {
1091                 rctl &= ~(IXGB_RCTL_UPE | IXGB_RCTL_MPE);
1092         }
1093
1094         if(netdev->mc_count > IXGB_MAX_NUM_MULTICAST_ADDRESSES) {
1095                 rctl |= IXGB_RCTL_MPE;
1096                 IXGB_WRITE_REG(hw, RCTL, rctl);
1097         } else {
1098                 uint8_t mta[netdev->mc_count * IXGB_ETH_LENGTH_OF_ADDRESS];
1099
1100                 IXGB_WRITE_REG(hw, RCTL, rctl);
1101
1102                 for(i = 0, mc_ptr = netdev->mc_list; mc_ptr;
1103                         i++, mc_ptr = mc_ptr->next)
1104                         memcpy(&mta[i * IXGB_ETH_LENGTH_OF_ADDRESS],
1105                                    mc_ptr->dmi_addr, IXGB_ETH_LENGTH_OF_ADDRESS);
1106
1107                 ixgb_mc_addr_list_update(hw, mta, netdev->mc_count, 0);
1108         }
1109 }
1110
1111 /**
1112  * ixgb_watchdog - Timer Call-back
1113  * @data: pointer to netdev cast into an unsigned long
1114  **/
1115
1116 static void
1117 ixgb_watchdog(unsigned long data)
1118 {
1119         struct ixgb_adapter *adapter = (struct ixgb_adapter *)data;
1120         struct net_device *netdev = adapter->netdev;
1121         struct ixgb_desc_ring *txdr = &adapter->tx_ring;
1122
1123         ixgb_check_for_link(&adapter->hw);
1124
1125         if (ixgb_check_for_bad_link(&adapter->hw)) {
1126                 /* force the reset path */
1127                 netif_stop_queue(netdev);
1128         }
1129
1130         if(adapter->hw.link_up) {
1131                 if(!netif_carrier_ok(netdev)) {
1132                         DPRINTK(LINK, INFO,
1133                                 "NIC Link is Up 10000 Mbps Full Duplex\n");
1134                         adapter->link_speed = 10000;
1135                         adapter->link_duplex = FULL_DUPLEX;
1136                         netif_carrier_on(netdev);
1137                         netif_wake_queue(netdev);
1138                 }
1139         } else {
1140                 if(netif_carrier_ok(netdev)) {
1141                         adapter->link_speed = 0;
1142                         adapter->link_duplex = 0;
1143                         DPRINTK(LINK, INFO, "NIC Link is Down\n");
1144                         netif_carrier_off(netdev);
1145                         netif_stop_queue(netdev);
1146
1147                 }
1148         }
1149
1150         ixgb_update_stats(adapter);
1151
1152         if(!netif_carrier_ok(netdev)) {
1153                 if(IXGB_DESC_UNUSED(txdr) + 1 < txdr->count) {
1154                         /* We've lost link, so the controller stops DMA,
1155                          * but we've got queued Tx work that's never going
1156                          * to get done, so reset controller to flush Tx.
1157                          * (Do the reset outside of interrupt context). */
1158                         schedule_work(&adapter->tx_timeout_task);
1159                 }
1160         }
1161
1162         /* Force detection of hung controller every watchdog period */
1163         adapter->detect_tx_hung = TRUE;
1164
1165         /* generate an interrupt to force clean up of any stragglers */
1166         IXGB_WRITE_REG(&adapter->hw, ICS, IXGB_INT_TXDW);
1167
1168         /* Reset the timer */
1169         mod_timer(&adapter->watchdog_timer, jiffies + 2 * HZ);
1170 }
1171
1172 #define IXGB_TX_FLAGS_CSUM              0x00000001
1173 #define IXGB_TX_FLAGS_VLAN              0x00000002
1174 #define IXGB_TX_FLAGS_TSO               0x00000004
1175
1176 static int
1177 ixgb_tso(struct ixgb_adapter *adapter, struct sk_buff *skb)
1178 {
1179 #ifdef NETIF_F_TSO
1180         struct ixgb_context_desc *context_desc;
1181         unsigned int i;
1182         uint8_t ipcss, ipcso, tucss, tucso, hdr_len;
1183         uint16_t ipcse, tucse, mss;
1184         int err;
1185
1186         if (likely(skb_is_gso(skb))) {
1187                 struct ixgb_buffer *buffer_info;
1188                 if (skb_header_cloned(skb)) {
1189                         err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1190                         if (err)
1191                                 return err;
1192                 }
1193
1194                 hdr_len = ((skb->h.raw - skb->data) + (skb->h.th->doff << 2));
1195                 mss = skb_shinfo(skb)->gso_size;
1196                 skb->nh.iph->tot_len = 0;
1197                 skb->nh.iph->check = 0;
1198                 skb->h.th->check = ~csum_tcpudp_magic(skb->nh.iph->saddr,
1199                                                       skb->nh.iph->daddr,
1200                                                       0, IPPROTO_TCP, 0);
1201                 ipcss = skb->nh.raw - skb->data;
1202                 ipcso = (void *)&(skb->nh.iph->check) - (void *)skb->data;
1203                 ipcse = skb->h.raw - skb->data - 1;
1204                 tucss = skb->h.raw - skb->data;
1205                 tucso = (void *)&(skb->h.th->check) - (void *)skb->data;
1206                 tucse = 0;
1207
1208                 i = adapter->tx_ring.next_to_use;
1209                 context_desc = IXGB_CONTEXT_DESC(adapter->tx_ring, i);
1210                 buffer_info = &adapter->tx_ring.buffer_info[i];
1211                 WARN_ON(buffer_info->dma != 0);
1212
1213                 context_desc->ipcss = ipcss;
1214                 context_desc->ipcso = ipcso;
1215                 context_desc->ipcse = cpu_to_le16(ipcse);
1216                 context_desc->tucss = tucss;
1217                 context_desc->tucso = tucso;
1218                 context_desc->tucse = cpu_to_le16(tucse);
1219                 context_desc->mss = cpu_to_le16(mss);
1220                 context_desc->hdr_len = hdr_len;
1221                 context_desc->status = 0;
1222                 context_desc->cmd_type_len = cpu_to_le32(
1223                                                   IXGB_CONTEXT_DESC_TYPE 
1224                                                 | IXGB_CONTEXT_DESC_CMD_TSE
1225                                                 | IXGB_CONTEXT_DESC_CMD_IP
1226                                                 | IXGB_CONTEXT_DESC_CMD_TCP
1227                                                 | IXGB_CONTEXT_DESC_CMD_IDE
1228                                                 | (skb->len - (hdr_len)));
1229
1230
1231                 if(++i == adapter->tx_ring.count) i = 0;
1232                 adapter->tx_ring.next_to_use = i;
1233
1234                 return 1;
1235         }
1236 #endif
1237
1238         return 0;
1239 }
1240
1241 static boolean_t
1242 ixgb_tx_csum(struct ixgb_adapter *adapter, struct sk_buff *skb)
1243 {
1244         struct ixgb_context_desc *context_desc;
1245         unsigned int i;
1246         uint8_t css, cso;
1247
1248         if(likely(skb->ip_summed == CHECKSUM_PARTIAL)) {
1249                 struct ixgb_buffer *buffer_info;
1250                 css = skb->h.raw - skb->data;
1251                 cso = css + skb->csum_offset;
1252
1253                 i = adapter->tx_ring.next_to_use;
1254                 context_desc = IXGB_CONTEXT_DESC(adapter->tx_ring, i);
1255                 buffer_info = &adapter->tx_ring.buffer_info[i];
1256                 WARN_ON(buffer_info->dma != 0);
1257
1258                 context_desc->tucss = css;
1259                 context_desc->tucso = cso;
1260                 context_desc->tucse = 0;
1261                 /* zero out any previously existing data in one instruction */
1262                 *(uint32_t *)&(context_desc->ipcss) = 0;
1263                 context_desc->status = 0;
1264                 context_desc->hdr_len = 0;
1265                 context_desc->mss = 0;
1266                 context_desc->cmd_type_len =
1267                         cpu_to_le32(IXGB_CONTEXT_DESC_TYPE
1268                                     | IXGB_TX_DESC_CMD_IDE);
1269
1270                 if(++i == adapter->tx_ring.count) i = 0;
1271                 adapter->tx_ring.next_to_use = i;
1272
1273                 return TRUE;
1274         }
1275
1276         return FALSE;
1277 }
1278
1279 #define IXGB_MAX_TXD_PWR        14
1280 #define IXGB_MAX_DATA_PER_TXD   (1<<IXGB_MAX_TXD_PWR)
1281
1282 static int
1283 ixgb_tx_map(struct ixgb_adapter *adapter, struct sk_buff *skb,
1284             unsigned int first)
1285 {
1286         struct ixgb_desc_ring *tx_ring = &adapter->tx_ring;
1287         struct ixgb_buffer *buffer_info;
1288         int len = skb->len;
1289         unsigned int offset = 0, size, count = 0, i;
1290         unsigned int mss = skb_shinfo(skb)->gso_size;
1291
1292         unsigned int nr_frags = skb_shinfo(skb)->nr_frags;
1293         unsigned int f;
1294
1295         len -= skb->data_len;
1296
1297         i = tx_ring->next_to_use;
1298
1299         while(len) {
1300                 buffer_info = &tx_ring->buffer_info[i];
1301                 size = min(len, IXGB_MAX_DATA_PER_TXD);
1302                 /* Workaround for premature desc write-backs
1303                  * in TSO mode.  Append 4-byte sentinel desc */
1304                 if (unlikely(mss && !nr_frags && size == len && size > 8))
1305                         size -= 4;
1306
1307                 buffer_info->length = size;
1308                 WARN_ON(buffer_info->dma != 0);
1309                 buffer_info->dma =
1310                         pci_map_single(adapter->pdev,
1311                                 skb->data + offset,
1312                                 size,
1313                                 PCI_DMA_TODEVICE);
1314                 buffer_info->time_stamp = jiffies;
1315                 buffer_info->next_to_watch = 0;
1316
1317                 len -= size;
1318                 offset += size;
1319                 count++;
1320                 if(++i == tx_ring->count) i = 0;
1321         }
1322
1323         for(f = 0; f < nr_frags; f++) {
1324                 struct skb_frag_struct *frag;
1325
1326                 frag = &skb_shinfo(skb)->frags[f];
1327                 len = frag->size;
1328                 offset = 0;
1329
1330                 while(len) {
1331                         buffer_info = &tx_ring->buffer_info[i];
1332                         size = min(len, IXGB_MAX_DATA_PER_TXD);
1333
1334                         /* Workaround for premature desc write-backs
1335                          * in TSO mode.  Append 4-byte sentinel desc */
1336                         if (unlikely(mss && !nr_frags && size == len
1337                                      && size > 8))
1338                                 size -= 4;
1339
1340                         buffer_info->length = size;
1341                         buffer_info->dma =
1342                                 pci_map_page(adapter->pdev,
1343                                         frag->page,
1344                                         frag->page_offset + offset,
1345                                         size,
1346                                         PCI_DMA_TODEVICE);
1347                         buffer_info->time_stamp = jiffies;
1348                         buffer_info->next_to_watch = 0;
1349
1350                         len -= size;
1351                         offset += size;
1352                         count++;
1353                         if(++i == tx_ring->count) i = 0;
1354                 }
1355         }
1356         i = (i == 0) ? tx_ring->count - 1 : i - 1;
1357         tx_ring->buffer_info[i].skb = skb;
1358         tx_ring->buffer_info[first].next_to_watch = i;
1359
1360         return count;
1361 }
1362
1363 static void
1364 ixgb_tx_queue(struct ixgb_adapter *adapter, int count, int vlan_id,int tx_flags)
1365 {
1366         struct ixgb_desc_ring *tx_ring = &adapter->tx_ring;
1367         struct ixgb_tx_desc *tx_desc = NULL;
1368         struct ixgb_buffer *buffer_info;
1369         uint32_t cmd_type_len = adapter->tx_cmd_type;
1370         uint8_t status = 0;
1371         uint8_t popts = 0;
1372         unsigned int i;
1373
1374         if(tx_flags & IXGB_TX_FLAGS_TSO) {
1375                 cmd_type_len |= IXGB_TX_DESC_CMD_TSE;
1376                 popts |= (IXGB_TX_DESC_POPTS_IXSM | IXGB_TX_DESC_POPTS_TXSM);
1377         }
1378
1379         if(tx_flags & IXGB_TX_FLAGS_CSUM)
1380                 popts |= IXGB_TX_DESC_POPTS_TXSM;
1381
1382         if(tx_flags & IXGB_TX_FLAGS_VLAN) {
1383                 cmd_type_len |= IXGB_TX_DESC_CMD_VLE;
1384         }
1385
1386         i = tx_ring->next_to_use;
1387
1388         while(count--) {
1389                 buffer_info = &tx_ring->buffer_info[i];
1390                 tx_desc = IXGB_TX_DESC(*tx_ring, i);
1391                 tx_desc->buff_addr = cpu_to_le64(buffer_info->dma);
1392                 tx_desc->cmd_type_len =
1393                         cpu_to_le32(cmd_type_len | buffer_info->length);
1394                 tx_desc->status = status;
1395                 tx_desc->popts = popts;
1396                 tx_desc->vlan = cpu_to_le16(vlan_id);
1397
1398                 if(++i == tx_ring->count) i = 0;
1399         }
1400
1401         tx_desc->cmd_type_len |= cpu_to_le32(IXGB_TX_DESC_CMD_EOP 
1402                                 | IXGB_TX_DESC_CMD_RS );
1403
1404         /* Force memory writes to complete before letting h/w
1405          * know there are new descriptors to fetch.  (Only
1406          * applicable for weak-ordered memory model archs,
1407          * such as IA-64). */
1408         wmb();
1409
1410         tx_ring->next_to_use = i;
1411         IXGB_WRITE_REG(&adapter->hw, TDT, i);
1412 }
1413
1414 static int __ixgb_maybe_stop_tx(struct net_device *netdev, int size)
1415 {
1416         struct ixgb_adapter *adapter = netdev_priv(netdev);
1417         struct ixgb_desc_ring *tx_ring = &adapter->tx_ring;
1418
1419         netif_stop_queue(netdev);
1420         /* Herbert's original patch had:
1421          *  smp_mb__after_netif_stop_queue();
1422          * but since that doesn't exist yet, just open code it. */
1423         smp_mb();
1424
1425         /* We need to check again in a case another CPU has just
1426          * made room available. */
1427         if (likely(IXGB_DESC_UNUSED(tx_ring) < size))
1428                 return -EBUSY;
1429
1430         /* A reprieve! */
1431         netif_start_queue(netdev);
1432         ++adapter->restart_queue;
1433         return 0;
1434 }
1435
1436 static int ixgb_maybe_stop_tx(struct net_device *netdev,
1437                               struct ixgb_desc_ring *tx_ring, int size)
1438 {
1439         if (likely(IXGB_DESC_UNUSED(tx_ring) >= size))
1440                 return 0;
1441         return __ixgb_maybe_stop_tx(netdev, size);
1442 }
1443
1444
1445 /* Tx Descriptors needed, worst case */
1446 #define TXD_USE_COUNT(S) (((S) >> IXGB_MAX_TXD_PWR) + \
1447                          (((S) & (IXGB_MAX_DATA_PER_TXD - 1)) ? 1 : 0))
1448 #define DESC_NEEDED TXD_USE_COUNT(IXGB_MAX_DATA_PER_TXD) /* skb->date */ + \
1449         MAX_SKB_FRAGS * TXD_USE_COUNT(PAGE_SIZE) + 1 /* for context */ \
1450         + 1 /* one more needed for sentinel TSO workaround */
1451
1452 static int
1453 ixgb_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
1454 {
1455         struct ixgb_adapter *adapter = netdev_priv(netdev);
1456         unsigned int first;
1457         unsigned int tx_flags = 0;
1458         unsigned long flags;
1459         int vlan_id = 0;
1460         int tso;
1461
1462         if(skb->len <= 0) {
1463                 dev_kfree_skb_any(skb);
1464                 return 0;
1465         }
1466
1467 #ifdef NETIF_F_LLTX
1468         local_irq_save(flags);
1469         if (!spin_trylock(&adapter->tx_lock)) {
1470                 /* Collision - tell upper layer to requeue */
1471                 local_irq_restore(flags);
1472                 return NETDEV_TX_LOCKED;
1473         }
1474 #else
1475         spin_lock_irqsave(&adapter->tx_lock, flags);
1476 #endif
1477
1478         if (unlikely(ixgb_maybe_stop_tx(netdev, &adapter->tx_ring,
1479                      DESC_NEEDED))) {
1480                 netif_stop_queue(netdev);
1481                 spin_unlock_irqrestore(&adapter->tx_lock, flags);
1482                 return NETDEV_TX_BUSY;
1483         }
1484
1485 #ifndef NETIF_F_LLTX
1486         spin_unlock_irqrestore(&adapter->tx_lock, flags);
1487 #endif
1488
1489         if(adapter->vlgrp && vlan_tx_tag_present(skb)) {
1490                 tx_flags |= IXGB_TX_FLAGS_VLAN;
1491                 vlan_id = vlan_tx_tag_get(skb);
1492         }
1493
1494         first = adapter->tx_ring.next_to_use;
1495         
1496         tso = ixgb_tso(adapter, skb);
1497         if (tso < 0) {
1498                 dev_kfree_skb_any(skb);
1499 #ifdef NETIF_F_LLTX
1500                 spin_unlock_irqrestore(&adapter->tx_lock, flags);
1501 #endif
1502                 return NETDEV_TX_OK;
1503         }
1504
1505         if (likely(tso))
1506                 tx_flags |= IXGB_TX_FLAGS_TSO;
1507         else if(ixgb_tx_csum(adapter, skb))
1508                 tx_flags |= IXGB_TX_FLAGS_CSUM;
1509
1510         ixgb_tx_queue(adapter, ixgb_tx_map(adapter, skb, first), vlan_id,
1511                         tx_flags);
1512
1513         netdev->trans_start = jiffies;
1514
1515 #ifdef NETIF_F_LLTX
1516         /* Make sure there is space in the ring for the next send. */
1517         ixgb_maybe_stop_tx(netdev, &adapter->tx_ring, DESC_NEEDED);
1518
1519         spin_unlock_irqrestore(&adapter->tx_lock, flags);
1520
1521 #endif
1522         return NETDEV_TX_OK;
1523 }
1524
1525 /**
1526  * ixgb_tx_timeout - Respond to a Tx Hang
1527  * @netdev: network interface device structure
1528  **/
1529
1530 static void
1531 ixgb_tx_timeout(struct net_device *netdev)
1532 {
1533         struct ixgb_adapter *adapter = netdev_priv(netdev);
1534
1535         /* Do the reset outside of interrupt context */
1536         schedule_work(&adapter->tx_timeout_task);
1537 }
1538
1539 static void
1540 ixgb_tx_timeout_task(struct work_struct *work)
1541 {
1542         struct ixgb_adapter *adapter =
1543                 container_of(work, struct ixgb_adapter, tx_timeout_task);
1544
1545         adapter->tx_timeout_count++;
1546         ixgb_down(adapter, TRUE);
1547         ixgb_up(adapter);
1548 }
1549
1550 /**
1551  * ixgb_get_stats - Get System Network Statistics
1552  * @netdev: network interface device structure
1553  *
1554  * Returns the address of the device statistics structure.
1555  * The statistics are actually updated from the timer callback.
1556  **/
1557
1558 static struct net_device_stats *
1559 ixgb_get_stats(struct net_device *netdev)
1560 {
1561         struct ixgb_adapter *adapter = netdev_priv(netdev);
1562
1563         return &adapter->net_stats;
1564 }
1565
1566 /**
1567  * ixgb_change_mtu - Change the Maximum Transfer Unit
1568  * @netdev: network interface device structure
1569  * @new_mtu: new value for maximum frame size
1570  *
1571  * Returns 0 on success, negative on failure
1572  **/
1573
1574 static int
1575 ixgb_change_mtu(struct net_device *netdev, int new_mtu)
1576 {
1577         struct ixgb_adapter *adapter = netdev_priv(netdev);
1578         int max_frame = new_mtu + ENET_HEADER_SIZE + ENET_FCS_LENGTH;
1579         int old_max_frame = netdev->mtu + ENET_HEADER_SIZE + ENET_FCS_LENGTH;
1580
1581
1582         if((max_frame < IXGB_MIN_ENET_FRAME_SIZE_WITHOUT_FCS + ENET_FCS_LENGTH)
1583            || (max_frame > IXGB_MAX_JUMBO_FRAME_SIZE + ENET_FCS_LENGTH)) {
1584                 DPRINTK(PROBE, ERR, "Invalid MTU setting %d\n", new_mtu);
1585                 return -EINVAL;
1586         }
1587
1588         adapter->rx_buffer_len = max_frame;
1589
1590         netdev->mtu = new_mtu;
1591
1592         if ((old_max_frame != max_frame) && netif_running(netdev)) {
1593                 ixgb_down(adapter, TRUE);
1594                 ixgb_up(adapter);
1595         }
1596
1597         return 0;
1598 }
1599
1600 /**
1601  * ixgb_update_stats - Update the board statistics counters.
1602  * @adapter: board private structure
1603  **/
1604
1605 void
1606 ixgb_update_stats(struct ixgb_adapter *adapter)
1607 {
1608         struct net_device *netdev = adapter->netdev;
1609         struct pci_dev *pdev = adapter->pdev;
1610
1611         /* Prevent stats update while adapter is being reset */
1612         if (pdev->error_state && pdev->error_state != pci_channel_io_normal)
1613                 return;
1614
1615         if((netdev->flags & IFF_PROMISC) || (netdev->flags & IFF_ALLMULTI) ||
1616            (netdev->mc_count > IXGB_MAX_NUM_MULTICAST_ADDRESSES)) {
1617                 u64 multi = IXGB_READ_REG(&adapter->hw, MPRCL);
1618                 u32 bcast_l = IXGB_READ_REG(&adapter->hw, BPRCL);
1619                 u32 bcast_h = IXGB_READ_REG(&adapter->hw, BPRCH);
1620                 u64 bcast = ((u64)bcast_h << 32) | bcast_l; 
1621
1622                 multi |= ((u64)IXGB_READ_REG(&adapter->hw, MPRCH) << 32);
1623                 /* fix up multicast stats by removing broadcasts */
1624                 if(multi >= bcast)
1625                         multi -= bcast;
1626                 
1627                 adapter->stats.mprcl += (multi & 0xFFFFFFFF);
1628                 adapter->stats.mprch += (multi >> 32);
1629                 adapter->stats.bprcl += bcast_l; 
1630                 adapter->stats.bprch += bcast_h;
1631         } else {
1632                 adapter->stats.mprcl += IXGB_READ_REG(&adapter->hw, MPRCL);
1633                 adapter->stats.mprch += IXGB_READ_REG(&adapter->hw, MPRCH);
1634                 adapter->stats.bprcl += IXGB_READ_REG(&adapter->hw, BPRCL);
1635                 adapter->stats.bprch += IXGB_READ_REG(&adapter->hw, BPRCH);
1636         }
1637         adapter->stats.tprl += IXGB_READ_REG(&adapter->hw, TPRL);
1638         adapter->stats.tprh += IXGB_READ_REG(&adapter->hw, TPRH);
1639         adapter->stats.gprcl += IXGB_READ_REG(&adapter->hw, GPRCL);
1640         adapter->stats.gprch += IXGB_READ_REG(&adapter->hw, GPRCH);
1641         adapter->stats.uprcl += IXGB_READ_REG(&adapter->hw, UPRCL);
1642         adapter->stats.uprch += IXGB_READ_REG(&adapter->hw, UPRCH);
1643         adapter->stats.vprcl += IXGB_READ_REG(&adapter->hw, VPRCL);
1644         adapter->stats.vprch += IXGB_READ_REG(&adapter->hw, VPRCH);
1645         adapter->stats.jprcl += IXGB_READ_REG(&adapter->hw, JPRCL);
1646         adapter->stats.jprch += IXGB_READ_REG(&adapter->hw, JPRCH);
1647         adapter->stats.gorcl += IXGB_READ_REG(&adapter->hw, GORCL);
1648         adapter->stats.gorch += IXGB_READ_REG(&adapter->hw, GORCH);
1649         adapter->stats.torl += IXGB_READ_REG(&adapter->hw, TORL);
1650         adapter->stats.torh += IXGB_READ_REG(&adapter->hw, TORH);
1651         adapter->stats.rnbc += IXGB_READ_REG(&adapter->hw, RNBC);
1652         adapter->stats.ruc += IXGB_READ_REG(&adapter->hw, RUC);
1653         adapter->stats.roc += IXGB_READ_REG(&adapter->hw, ROC);
1654         adapter->stats.rlec += IXGB_READ_REG(&adapter->hw, RLEC);
1655         adapter->stats.crcerrs += IXGB_READ_REG(&adapter->hw, CRCERRS);
1656         adapter->stats.icbc += IXGB_READ_REG(&adapter->hw, ICBC);
1657         adapter->stats.ecbc += IXGB_READ_REG(&adapter->hw, ECBC);
1658         adapter->stats.mpc += IXGB_READ_REG(&adapter->hw, MPC);
1659         adapter->stats.tptl += IXGB_READ_REG(&adapter->hw, TPTL);
1660         adapter->stats.tpth += IXGB_READ_REG(&adapter->hw, TPTH);
1661         adapter->stats.gptcl += IXGB_READ_REG(&adapter->hw, GPTCL);
1662         adapter->stats.gptch += IXGB_READ_REG(&adapter->hw, GPTCH);
1663         adapter->stats.bptcl += IXGB_READ_REG(&adapter->hw, BPTCL);
1664         adapter->stats.bptch += IXGB_READ_REG(&adapter->hw, BPTCH);
1665         adapter->stats.mptcl += IXGB_READ_REG(&adapter->hw, MPTCL);
1666         adapter->stats.mptch += IXGB_READ_REG(&adapter->hw, MPTCH);
1667         adapter->stats.uptcl += IXGB_READ_REG(&adapter->hw, UPTCL);
1668         adapter->stats.uptch += IXGB_READ_REG(&adapter->hw, UPTCH);
1669         adapter->stats.vptcl += IXGB_READ_REG(&adapter->hw, VPTCL);
1670         adapter->stats.vptch += IXGB_READ_REG(&adapter->hw, VPTCH);
1671         adapter->stats.jptcl += IXGB_READ_REG(&adapter->hw, JPTCL);
1672         adapter->stats.jptch += IXGB_READ_REG(&adapter->hw, JPTCH);
1673         adapter->stats.gotcl += IXGB_READ_REG(&adapter->hw, GOTCL);
1674         adapter->stats.gotch += IXGB_READ_REG(&adapter->hw, GOTCH);
1675         adapter->stats.totl += IXGB_READ_REG(&adapter->hw, TOTL);
1676         adapter->stats.toth += IXGB_READ_REG(&adapter->hw, TOTH);
1677         adapter->stats.dc += IXGB_READ_REG(&adapter->hw, DC);
1678         adapter->stats.plt64c += IXGB_READ_REG(&adapter->hw, PLT64C);
1679         adapter->stats.tsctc += IXGB_READ_REG(&adapter->hw, TSCTC);
1680         adapter->stats.tsctfc += IXGB_READ_REG(&adapter->hw, TSCTFC);
1681         adapter->stats.ibic += IXGB_READ_REG(&adapter->hw, IBIC);
1682         adapter->stats.rfc += IXGB_READ_REG(&adapter->hw, RFC);
1683         adapter->stats.lfc += IXGB_READ_REG(&adapter->hw, LFC);
1684         adapter->stats.pfrc += IXGB_READ_REG(&adapter->hw, PFRC);
1685         adapter->stats.pftc += IXGB_READ_REG(&adapter->hw, PFTC);
1686         adapter->stats.mcfrc += IXGB_READ_REG(&adapter->hw, MCFRC);
1687         adapter->stats.mcftc += IXGB_READ_REG(&adapter->hw, MCFTC);
1688         adapter->stats.xonrxc += IXGB_READ_REG(&adapter->hw, XONRXC);
1689         adapter->stats.xontxc += IXGB_READ_REG(&adapter->hw, XONTXC);
1690         adapter->stats.xoffrxc += IXGB_READ_REG(&adapter->hw, XOFFRXC);
1691         adapter->stats.xofftxc += IXGB_READ_REG(&adapter->hw, XOFFTXC);
1692         adapter->stats.rjc += IXGB_READ_REG(&adapter->hw, RJC);
1693
1694         /* Fill out the OS statistics structure */
1695
1696         adapter->net_stats.rx_packets = adapter->stats.gprcl;
1697         adapter->net_stats.tx_packets = adapter->stats.gptcl;
1698         adapter->net_stats.rx_bytes = adapter->stats.gorcl;
1699         adapter->net_stats.tx_bytes = adapter->stats.gotcl;
1700         adapter->net_stats.multicast = adapter->stats.mprcl;
1701         adapter->net_stats.collisions = 0;
1702
1703         /* ignore RLEC as it reports errors for padded (<64bytes) frames
1704          * with a length in the type/len field */
1705         adapter->net_stats.rx_errors =
1706             /* adapter->stats.rnbc + */ adapter->stats.crcerrs +
1707             adapter->stats.ruc +
1708             adapter->stats.roc /*+ adapter->stats.rlec */  +
1709             adapter->stats.icbc +
1710             adapter->stats.ecbc + adapter->stats.mpc;
1711
1712         /* see above
1713          * adapter->net_stats.rx_length_errors = adapter->stats.rlec;
1714          */
1715
1716         adapter->net_stats.rx_crc_errors = adapter->stats.crcerrs;
1717         adapter->net_stats.rx_fifo_errors = adapter->stats.mpc;
1718         adapter->net_stats.rx_missed_errors = adapter->stats.mpc;
1719         adapter->net_stats.rx_over_errors = adapter->stats.mpc;
1720
1721         adapter->net_stats.tx_errors = 0;
1722         adapter->net_stats.rx_frame_errors = 0;
1723         adapter->net_stats.tx_aborted_errors = 0;
1724         adapter->net_stats.tx_carrier_errors = 0;
1725         adapter->net_stats.tx_fifo_errors = 0;
1726         adapter->net_stats.tx_heartbeat_errors = 0;
1727         adapter->net_stats.tx_window_errors = 0;
1728 }
1729
1730 #define IXGB_MAX_INTR 10
1731 /**
1732  * ixgb_intr - Interrupt Handler
1733  * @irq: interrupt number
1734  * @data: pointer to a network interface device structure
1735  **/
1736
1737 static irqreturn_t
1738 ixgb_intr(int irq, void *data)
1739 {
1740         struct net_device *netdev = data;
1741         struct ixgb_adapter *adapter = netdev_priv(netdev);
1742         struct ixgb_hw *hw = &adapter->hw;
1743         uint32_t icr = IXGB_READ_REG(hw, ICR);
1744 #ifndef CONFIG_IXGB_NAPI
1745         unsigned int i;
1746 #endif
1747
1748         if(unlikely(!icr))
1749                 return IRQ_NONE;  /* Not our interrupt */
1750
1751         if(unlikely(icr & (IXGB_INT_RXSEQ | IXGB_INT_LSC))) {
1752                 mod_timer(&adapter->watchdog_timer, jiffies);
1753         }
1754
1755 #ifdef CONFIG_IXGB_NAPI
1756         if(netif_rx_schedule_prep(netdev)) {
1757
1758                 /* Disable interrupts and register for poll. The flush 
1759                   of the posted write is intentionally left out.
1760                 */
1761
1762                 atomic_inc(&adapter->irq_sem);
1763                 IXGB_WRITE_REG(&adapter->hw, IMC, ~0);
1764                 __netif_rx_schedule(netdev);
1765         }
1766 #else
1767         /* yes, that is actually a & and it is meant to make sure that
1768          * every pass through this for loop checks both receive and
1769          * transmit queues for completed descriptors, intended to
1770          * avoid starvation issues and assist tx/rx fairness. */
1771         for(i = 0; i < IXGB_MAX_INTR; i++)
1772                 if(!ixgb_clean_rx_irq(adapter) &
1773                    !ixgb_clean_tx_irq(adapter))
1774                         break;
1775 #endif 
1776         return IRQ_HANDLED;
1777 }
1778
1779 #ifdef CONFIG_IXGB_NAPI
1780 /**
1781  * ixgb_clean - NAPI Rx polling callback
1782  * @adapter: board private structure
1783  **/
1784
1785 static int
1786 ixgb_clean(struct net_device *netdev, int *budget)
1787 {
1788         struct ixgb_adapter *adapter = netdev_priv(netdev);
1789         int work_to_do = min(*budget, netdev->quota);
1790         int tx_cleaned;
1791         int work_done = 0;
1792
1793         tx_cleaned = ixgb_clean_tx_irq(adapter);
1794         ixgb_clean_rx_irq(adapter, &work_done, work_to_do);
1795
1796         *budget -= work_done;
1797         netdev->quota -= work_done;
1798
1799         /* if no Tx and not enough Rx work done, exit the polling mode */
1800         if((!tx_cleaned && (work_done == 0)) || !netif_running(netdev)) {
1801                 netif_rx_complete(netdev);
1802                 ixgb_irq_enable(adapter);
1803                 return 0;
1804         }
1805
1806         return 1;
1807 }
1808 #endif
1809
1810 /**
1811  * ixgb_clean_tx_irq - Reclaim resources after transmit completes
1812  * @adapter: board private structure
1813  **/
1814
1815 static boolean_t
1816 ixgb_clean_tx_irq(struct ixgb_adapter *adapter)
1817 {
1818         struct ixgb_desc_ring *tx_ring = &adapter->tx_ring;
1819         struct net_device *netdev = adapter->netdev;
1820         struct ixgb_tx_desc *tx_desc, *eop_desc;
1821         struct ixgb_buffer *buffer_info;
1822         unsigned int i, eop;
1823         boolean_t cleaned = FALSE;
1824
1825         i = tx_ring->next_to_clean;
1826         eop = tx_ring->buffer_info[i].next_to_watch;
1827         eop_desc = IXGB_TX_DESC(*tx_ring, eop);
1828
1829         while(eop_desc->status & IXGB_TX_DESC_STATUS_DD) {
1830
1831                 for(cleaned = FALSE; !cleaned; ) {
1832                         tx_desc = IXGB_TX_DESC(*tx_ring, i);
1833                         buffer_info = &tx_ring->buffer_info[i];
1834
1835                         if (tx_desc->popts
1836                             & (IXGB_TX_DESC_POPTS_TXSM |
1837                                IXGB_TX_DESC_POPTS_IXSM))
1838                                 adapter->hw_csum_tx_good++;
1839
1840                         ixgb_unmap_and_free_tx_resource(adapter, buffer_info);
1841
1842                         *(uint32_t *)&(tx_desc->status) = 0;
1843
1844                         cleaned = (i == eop);
1845                         if(++i == tx_ring->count) i = 0;
1846                 }
1847
1848                 eop = tx_ring->buffer_info[i].next_to_watch;
1849                 eop_desc = IXGB_TX_DESC(*tx_ring, eop);
1850         }
1851
1852         tx_ring->next_to_clean = i;
1853
1854         if (unlikely(netif_queue_stopped(netdev))) {
1855                 spin_lock(&adapter->tx_lock);
1856                 if (netif_queue_stopped(netdev) && netif_carrier_ok(netdev) &&
1857                     (IXGB_DESC_UNUSED(tx_ring) >= DESC_NEEDED))
1858                         netif_wake_queue(netdev);
1859                 spin_unlock(&adapter->tx_lock);
1860         }
1861
1862         if(adapter->detect_tx_hung) {
1863                 /* detect a transmit hang in hardware, this serializes the
1864                  * check with the clearing of time_stamp and movement of i */
1865                 adapter->detect_tx_hung = FALSE;
1866                 if (tx_ring->buffer_info[eop].dma &&
1867                    time_after(jiffies, tx_ring->buffer_info[eop].time_stamp + HZ)
1868                    && !(IXGB_READ_REG(&adapter->hw, STATUS) &
1869                         IXGB_STATUS_TXOFF)) {
1870                         /* detected Tx unit hang */
1871                         DPRINTK(DRV, ERR, "Detected Tx Unit Hang\n"
1872                                         "  TDH                  <%x>\n"
1873                                         "  TDT                  <%x>\n"
1874                                         "  next_to_use          <%x>\n"
1875                                         "  next_to_clean        <%x>\n"
1876                                         "buffer_info[next_to_clean]\n"
1877                                         "  time_stamp           <%lx>\n"
1878                                         "  next_to_watch        <%x>\n"
1879                                         "  jiffies              <%lx>\n"
1880                                         "  next_to_watch.status <%x>\n",
1881                                 IXGB_READ_REG(&adapter->hw, TDH),
1882                                 IXGB_READ_REG(&adapter->hw, TDT),
1883                                 tx_ring->next_to_use,
1884                                 tx_ring->next_to_clean,
1885                                 tx_ring->buffer_info[eop].time_stamp,
1886                                 eop,
1887                                 jiffies,
1888                                 eop_desc->status);
1889                         netif_stop_queue(netdev);
1890                 }
1891         }
1892
1893         return cleaned;
1894 }
1895
1896 /**
1897  * ixgb_rx_checksum - Receive Checksum Offload for 82597.
1898  * @adapter: board private structure
1899  * @rx_desc: receive descriptor
1900  * @sk_buff: socket buffer with received data
1901  **/
1902
1903 static void
1904 ixgb_rx_checksum(struct ixgb_adapter *adapter,
1905                  struct ixgb_rx_desc *rx_desc,
1906                  struct sk_buff *skb)
1907 {
1908         /* Ignore Checksum bit is set OR
1909          * TCP Checksum has not been calculated
1910          */
1911         if((rx_desc->status & IXGB_RX_DESC_STATUS_IXSM) ||
1912            (!(rx_desc->status & IXGB_RX_DESC_STATUS_TCPCS))) {
1913                 skb->ip_summed = CHECKSUM_NONE;
1914                 return;
1915         }
1916
1917         /* At this point we know the hardware did the TCP checksum */
1918         /* now look at the TCP checksum error bit */
1919         if(rx_desc->errors & IXGB_RX_DESC_ERRORS_TCPE) {
1920                 /* let the stack verify checksum errors */
1921                 skb->ip_summed = CHECKSUM_NONE;
1922                 adapter->hw_csum_rx_error++;
1923         } else {
1924                 /* TCP checksum is good */
1925                 skb->ip_summed = CHECKSUM_UNNECESSARY;
1926                 adapter->hw_csum_rx_good++;
1927         }
1928 }
1929
1930 /**
1931  * ixgb_clean_rx_irq - Send received data up the network stack,
1932  * @adapter: board private structure
1933  **/
1934
1935 static boolean_t
1936 #ifdef CONFIG_IXGB_NAPI
1937 ixgb_clean_rx_irq(struct ixgb_adapter *adapter, int *work_done, int work_to_do)
1938 #else
1939 ixgb_clean_rx_irq(struct ixgb_adapter *adapter)
1940 #endif
1941 {
1942         struct ixgb_desc_ring *rx_ring = &adapter->rx_ring;
1943         struct net_device *netdev = adapter->netdev;
1944         struct pci_dev *pdev = adapter->pdev;
1945         struct ixgb_rx_desc *rx_desc, *next_rxd;
1946         struct ixgb_buffer *buffer_info, *next_buffer, *next2_buffer;
1947         uint32_t length;
1948         unsigned int i, j;
1949         boolean_t cleaned = FALSE;
1950
1951         i = rx_ring->next_to_clean;
1952         rx_desc = IXGB_RX_DESC(*rx_ring, i);
1953         buffer_info = &rx_ring->buffer_info[i];
1954
1955         while(rx_desc->status & IXGB_RX_DESC_STATUS_DD) {
1956                 struct sk_buff *skb, *next_skb;
1957                 u8 status;
1958
1959 #ifdef CONFIG_IXGB_NAPI
1960                 if(*work_done >= work_to_do)
1961                         break;
1962
1963                 (*work_done)++;
1964 #endif
1965                 status = rx_desc->status;
1966                 skb = buffer_info->skb;
1967                 buffer_info->skb = NULL;
1968
1969                 prefetch(skb->data);
1970
1971                 if(++i == rx_ring->count) i = 0;
1972                 next_rxd = IXGB_RX_DESC(*rx_ring, i);
1973                 prefetch(next_rxd);
1974
1975                 if((j = i + 1) == rx_ring->count) j = 0;
1976                 next2_buffer = &rx_ring->buffer_info[j];
1977                 prefetch(next2_buffer);
1978
1979                 next_buffer = &rx_ring->buffer_info[i];
1980                 next_skb = next_buffer->skb;
1981                 prefetch(next_skb);
1982
1983                 cleaned = TRUE;
1984
1985                 pci_unmap_single(pdev,
1986                                  buffer_info->dma,
1987                                  buffer_info->length,
1988                                  PCI_DMA_FROMDEVICE);
1989
1990                 length = le16_to_cpu(rx_desc->length);
1991
1992                 if(unlikely(!(status & IXGB_RX_DESC_STATUS_EOP))) {
1993
1994                         /* All receives must fit into a single buffer */
1995
1996                         IXGB_DBG("Receive packet consumed multiple buffers "
1997                                          "length<%x>\n", length);
1998
1999                         dev_kfree_skb_irq(skb);
2000                         goto rxdesc_done;
2001                 }
2002
2003                 if (unlikely(rx_desc->errors
2004                              & (IXGB_RX_DESC_ERRORS_CE | IXGB_RX_DESC_ERRORS_SE
2005                                 | IXGB_RX_DESC_ERRORS_P |
2006                                 IXGB_RX_DESC_ERRORS_RXE))) {
2007
2008                         dev_kfree_skb_irq(skb);
2009                         goto rxdesc_done;
2010                 }
2011
2012                 /* code added for copybreak, this should improve
2013                  * performance for small packets with large amounts
2014                  * of reassembly being done in the stack */
2015 #define IXGB_CB_LENGTH 256
2016                 if (length < IXGB_CB_LENGTH) {
2017                         struct sk_buff *new_skb =
2018                             netdev_alloc_skb(netdev, length + NET_IP_ALIGN);
2019                         if (new_skb) {
2020                                 skb_reserve(new_skb, NET_IP_ALIGN);
2021                                 memcpy(new_skb->data - NET_IP_ALIGN,
2022                                        skb->data - NET_IP_ALIGN,
2023                                        length + NET_IP_ALIGN);
2024                                 /* save the skb in buffer_info as good */
2025                                 buffer_info->skb = skb;
2026                                 skb = new_skb;
2027                         }
2028                 }
2029                 /* end copybreak code */
2030
2031                 /* Good Receive */
2032                 skb_put(skb, length);
2033
2034                 /* Receive Checksum Offload */
2035                 ixgb_rx_checksum(adapter, rx_desc, skb);
2036
2037                 skb->protocol = eth_type_trans(skb, netdev);
2038 #ifdef CONFIG_IXGB_NAPI
2039                 if(adapter->vlgrp && (status & IXGB_RX_DESC_STATUS_VP)) {
2040                         vlan_hwaccel_receive_skb(skb, adapter->vlgrp,
2041                                 le16_to_cpu(rx_desc->special) &
2042                                         IXGB_RX_DESC_SPECIAL_VLAN_MASK);
2043                 } else {
2044                         netif_receive_skb(skb);
2045                 }
2046 #else /* CONFIG_IXGB_NAPI */
2047                 if(adapter->vlgrp && (status & IXGB_RX_DESC_STATUS_VP)) {
2048                         vlan_hwaccel_rx(skb, adapter->vlgrp,
2049                                 le16_to_cpu(rx_desc->special) &
2050                                         IXGB_RX_DESC_SPECIAL_VLAN_MASK);
2051                 } else {
2052                         netif_rx(skb);
2053                 }
2054 #endif /* CONFIG_IXGB_NAPI */
2055                 netdev->last_rx = jiffies;
2056
2057 rxdesc_done:
2058                 /* clean up descriptor, might be written over by hw */
2059                 rx_desc->status = 0;
2060
2061                 /* use prefetched values */
2062                 rx_desc = next_rxd;
2063                 buffer_info = next_buffer;
2064         }
2065
2066         rx_ring->next_to_clean = i;
2067
2068         ixgb_alloc_rx_buffers(adapter);
2069
2070         return cleaned;
2071 }
2072
2073 /**
2074  * ixgb_alloc_rx_buffers - Replace used receive buffers
2075  * @adapter: address of board private structure
2076  **/
2077
2078 static void
2079 ixgb_alloc_rx_buffers(struct ixgb_adapter *adapter)
2080 {
2081         struct ixgb_desc_ring *rx_ring = &adapter->rx_ring;
2082         struct net_device *netdev = adapter->netdev;
2083         struct pci_dev *pdev = adapter->pdev;
2084         struct ixgb_rx_desc *rx_desc;
2085         struct ixgb_buffer *buffer_info;
2086         struct sk_buff *skb;
2087         unsigned int i;
2088         int num_group_tail_writes;
2089         long cleancount;
2090
2091         i = rx_ring->next_to_use;
2092         buffer_info = &rx_ring->buffer_info[i];
2093         cleancount = IXGB_DESC_UNUSED(rx_ring);
2094
2095         num_group_tail_writes = IXGB_RX_BUFFER_WRITE;
2096
2097         /* leave three descriptors unused */
2098         while(--cleancount > 2) {
2099                 /* recycle! its good for you */
2100                 skb = buffer_info->skb;
2101                 if (skb) {
2102                         skb_trim(skb, 0);
2103                         goto map_skb;
2104                 }
2105
2106                 skb = netdev_alloc_skb(netdev, adapter->rx_buffer_len
2107                                        + NET_IP_ALIGN);
2108                 if (unlikely(!skb)) {
2109                         /* Better luck next round */
2110                         adapter->alloc_rx_buff_failed++;
2111                         break;
2112                 }
2113
2114                 /* Make buffer alignment 2 beyond a 16 byte boundary
2115                  * this will result in a 16 byte aligned IP header after
2116                  * the 14 byte MAC header is removed
2117                  */
2118                 skb_reserve(skb, NET_IP_ALIGN);
2119
2120                 buffer_info->skb = skb;
2121                 buffer_info->length = adapter->rx_buffer_len;
2122 map_skb:
2123                 buffer_info->dma = pci_map_single(pdev,
2124                                                   skb->data,
2125                                                   adapter->rx_buffer_len,
2126                                                   PCI_DMA_FROMDEVICE);
2127
2128                 rx_desc = IXGB_RX_DESC(*rx_ring, i);
2129                 rx_desc->buff_addr = cpu_to_le64(buffer_info->dma);
2130                 /* guarantee DD bit not set now before h/w gets descriptor
2131                  * this is the rest of the workaround for h/w double 
2132                  * writeback. */
2133                 rx_desc->status = 0;
2134
2135
2136                 if(++i == rx_ring->count) i = 0;
2137                 buffer_info = &rx_ring->buffer_info[i];
2138         }
2139
2140         if (likely(rx_ring->next_to_use != i)) {
2141                 rx_ring->next_to_use = i;
2142                 if (unlikely(i-- == 0))
2143                         i = (rx_ring->count - 1);
2144
2145                 /* Force memory writes to complete before letting h/w
2146                  * know there are new descriptors to fetch.  (Only
2147                  * applicable for weak-ordered memory model archs, such
2148                  * as IA-64). */
2149                 wmb();
2150                 IXGB_WRITE_REG(&adapter->hw, RDT, i);
2151         }
2152 }
2153
2154 /**
2155  * ixgb_vlan_rx_register - enables or disables vlan tagging/stripping.
2156  * 
2157  * @param netdev network interface device structure
2158  * @param grp indicates to enable or disable tagging/stripping
2159  **/
2160 static void
2161 ixgb_vlan_rx_register(struct net_device *netdev, struct vlan_group *grp)
2162 {
2163         struct ixgb_adapter *adapter = netdev_priv(netdev);
2164         uint32_t ctrl, rctl;
2165
2166         ixgb_irq_disable(adapter);
2167         adapter->vlgrp = grp;
2168
2169         if(grp) {
2170                 /* enable VLAN tag insert/strip */
2171                 ctrl = IXGB_READ_REG(&adapter->hw, CTRL0);
2172                 ctrl |= IXGB_CTRL0_VME;
2173                 IXGB_WRITE_REG(&adapter->hw, CTRL0, ctrl);
2174
2175                 /* enable VLAN receive filtering */
2176
2177                 rctl = IXGB_READ_REG(&adapter->hw, RCTL);
2178                 rctl |= IXGB_RCTL_VFE;
2179                 rctl &= ~IXGB_RCTL_CFIEN;
2180                 IXGB_WRITE_REG(&adapter->hw, RCTL, rctl);
2181         } else {
2182                 /* disable VLAN tag insert/strip */
2183
2184                 ctrl = IXGB_READ_REG(&adapter->hw, CTRL0);
2185                 ctrl &= ~IXGB_CTRL0_VME;
2186                 IXGB_WRITE_REG(&adapter->hw, CTRL0, ctrl);
2187
2188                 /* disable VLAN filtering */
2189
2190                 rctl = IXGB_READ_REG(&adapter->hw, RCTL);
2191                 rctl &= ~IXGB_RCTL_VFE;
2192                 IXGB_WRITE_REG(&adapter->hw, RCTL, rctl);
2193         }
2194
2195         ixgb_irq_enable(adapter);
2196 }
2197
2198 static void
2199 ixgb_vlan_rx_add_vid(struct net_device *netdev, uint16_t vid)
2200 {
2201         struct ixgb_adapter *adapter = netdev_priv(netdev);
2202         uint32_t vfta, index;
2203
2204         /* add VID to filter table */
2205
2206         index = (vid >> 5) & 0x7F;
2207         vfta = IXGB_READ_REG_ARRAY(&adapter->hw, VFTA, index);
2208         vfta |= (1 << (vid & 0x1F));
2209         ixgb_write_vfta(&adapter->hw, index, vfta);
2210 }
2211
2212 static void
2213 ixgb_vlan_rx_kill_vid(struct net_device *netdev, uint16_t vid)
2214 {
2215         struct ixgb_adapter *adapter = netdev_priv(netdev);
2216         uint32_t vfta, index;
2217
2218         ixgb_irq_disable(adapter);
2219
2220         if(adapter->vlgrp)
2221                 adapter->vlgrp->vlan_devices[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(!adapter->vlgrp->vlan_devices[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 */