Merge branch 'upstream-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/jgarzi...
[linux-2.6] / drivers / net / ixgb / ixgb_main.c
1 /*******************************************************************************
2
3   Intel PRO/10GbE Linux driver
4   Copyright(c) 1999 - 2006 Intel Corporation.
5
6   This program is free software; you can redistribute it and/or modify it
7   under the terms and conditions of the GNU General Public License,
8   version 2, as published by the Free Software Foundation.
9
10   This program is distributed in the hope it will be useful, but WITHOUT
11   ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12   FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
13   more details.
14
15   You should have received a copy of the GNU General Public License along with
16   this program; if not, write to the Free Software Foundation, Inc.,
17   51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
18
19   The full GNU General Public License is included in this distribution in
20   the file called "COPYING".
21
22   Contact Information:
23   Linux NICS <linux.nics@intel.com>
24   e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
25   Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
26
27 *******************************************************************************/
28
29 #include "ixgb.h"
30
31 char ixgb_driver_name[] = "ixgb";
32 static char ixgb_driver_string[] = "Intel(R) PRO/10GbE Network Driver";
33
34 #ifndef CONFIG_IXGB_NAPI
35 #define DRIVERNAPI
36 #else
37 #define DRIVERNAPI "-NAPI"
38 #endif
39 #define DRV_VERSION             "1.0.126-k4"DRIVERNAPI
40 const char ixgb_driver_version[] = DRV_VERSION;
41 static const char ixgb_copyright[] = "Copyright (c) 1999-2006 Intel Corporation.";
42
43 /* 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, bool 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 bool 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 bool ixgb_clean_rx_irq(struct ixgb_adapter *adapter,
102                               int *work_done, int work_to_do);
103 #else
104 static bool 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, u16 vid);
112 static void ixgb_vlan_rx_kill_vid(struct net_device *netdev, u16 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         IXGB_WRITE_REG(&adapter->hw, IMC, ~0);
201         IXGB_WRITE_FLUSH(&adapter->hw);
202         synchronize_irq(adapter->pdev->irq);
203 }
204
205 /**
206  * ixgb_irq_enable - Enable default interrupt generation settings
207  * @adapter: board private structure
208  **/
209
210 static void
211 ixgb_irq_enable(struct ixgb_adapter *adapter)
212 {
213         u32 val = IXGB_INT_RXT0 | IXGB_INT_RXDMT0 |
214                   IXGB_INT_TXDW | IXGB_INT_LSC;
215         if (adapter->hw.subsystem_vendor_id == SUN_SUBVENDOR_ID)
216                 val |= IXGB_INT_GPI0;
217         IXGB_WRITE_REG(&adapter->hw, IMS, val);
218         IXGB_WRITE_FLUSH(&adapter->hw);
219 }
220
221 int
222 ixgb_up(struct ixgb_adapter *adapter)
223 {
224         struct net_device *netdev = adapter->netdev;
225         int err, irq_flags = IRQF_SHARED;
226         int max_frame = netdev->mtu + ENET_HEADER_SIZE + ENET_FCS_LENGTH;
227         struct ixgb_hw *hw = &adapter->hw;
228
229         /* hardware has been reset, we need to reload some things */
230
231         ixgb_rar_set(hw, netdev->dev_addr, 0);
232         ixgb_set_multi(netdev);
233
234         ixgb_restore_vlan(adapter);
235
236         ixgb_configure_tx(adapter);
237         ixgb_setup_rctl(adapter);
238         ixgb_configure_rx(adapter);
239         ixgb_alloc_rx_buffers(adapter);
240
241         /* disable interrupts and get the hardware into a known state */
242         IXGB_WRITE_REG(&adapter->hw, IMC, 0xffffffff);
243
244         /* only enable MSI if bus is in PCI-X mode */
245         if (IXGB_READ_REG(&adapter->hw, STATUS) & IXGB_STATUS_PCIX_MODE) {
246                 err = pci_enable_msi(adapter->pdev);
247                 if (!err) {
248                         adapter->have_msi = 1;
249                         irq_flags = 0;
250                 }
251                 /* proceed to try to request regular interrupt */
252         }
253
254         err = request_irq(adapter->pdev->irq, &ixgb_intr, irq_flags,
255                           netdev->name, netdev);
256         if (err) {
257                 if (adapter->have_msi)
258                         pci_disable_msi(adapter->pdev);
259                 DPRINTK(PROBE, ERR,
260                  "Unable to allocate interrupt Error: %d\n", err);
261                 return err;
262         }
263
264         if((hw->max_frame_size != max_frame) ||
265                 (hw->max_frame_size !=
266                 (IXGB_READ_REG(hw, MFS) >> IXGB_MFS_SHIFT))) {
267
268                 hw->max_frame_size = max_frame;
269
270                 IXGB_WRITE_REG(hw, MFS, hw->max_frame_size << IXGB_MFS_SHIFT);
271
272                 if(hw->max_frame_size >
273                    IXGB_MAX_ENET_FRAME_SIZE_WITHOUT_FCS + ENET_FCS_LENGTH) {
274                         u32 ctrl0 = IXGB_READ_REG(hw, CTRL0);
275
276                         if(!(ctrl0 & IXGB_CTRL0_JFE)) {
277                                 ctrl0 |= IXGB_CTRL0_JFE;
278                                 IXGB_WRITE_REG(hw, CTRL0, ctrl0);
279                         }
280                 }
281         }
282
283         clear_bit(__IXGB_DOWN, &adapter->flags);
284
285 #ifdef CONFIG_IXGB_NAPI
286         napi_enable(&adapter->napi);
287 #endif
288         ixgb_irq_enable(adapter);
289
290         mod_timer(&adapter->watchdog_timer, jiffies);
291
292         return 0;
293 }
294
295 void
296 ixgb_down(struct ixgb_adapter *adapter, bool kill_watchdog)
297 {
298         struct net_device *netdev = adapter->netdev;
299
300         /* prevent the interrupt handler from restarting watchdog */
301         set_bit(__IXGB_DOWN, &adapter->flags);
302
303 #ifdef CONFIG_IXGB_NAPI
304         napi_disable(&adapter->napi);
305 #endif
306         /* waiting for NAPI to complete can re-enable interrupts */
307         ixgb_irq_disable(adapter);
308         free_irq(adapter->pdev->irq, netdev);
309
310         if (adapter->have_msi)
311                 pci_disable_msi(adapter->pdev);
312
313         if(kill_watchdog)
314                 del_timer_sync(&adapter->watchdog_timer);
315
316         adapter->link_speed = 0;
317         adapter->link_duplex = 0;
318         netif_carrier_off(netdev);
319         netif_stop_queue(netdev);
320
321         ixgb_reset(adapter);
322         ixgb_clean_tx_ring(adapter);
323         ixgb_clean_rx_ring(adapter);
324 }
325
326 void
327 ixgb_reset(struct ixgb_adapter *adapter)
328 {
329         struct ixgb_hw *hw = &adapter->hw;
330
331         ixgb_adapter_stop(hw);
332         if (!ixgb_init_hw(hw))
333                 DPRINTK(PROBE, ERR, "ixgb_init_hw failed.\n");
334
335         /* restore frame size information */
336         IXGB_WRITE_REG(hw, MFS, hw->max_frame_size << IXGB_MFS_SHIFT);
337         if (hw->max_frame_size >
338             IXGB_MAX_ENET_FRAME_SIZE_WITHOUT_FCS + ENET_FCS_LENGTH) {
339                 u32 ctrl0 = IXGB_READ_REG(hw, CTRL0);
340                 if (!(ctrl0 & IXGB_CTRL0_JFE)) {
341                         ctrl0 |= IXGB_CTRL0_JFE;
342                         IXGB_WRITE_REG(hw, CTRL0, ctrl0);
343                 }
344         }
345 }
346
347 /**
348  * ixgb_probe - Device Initialization Routine
349  * @pdev: PCI device information struct
350  * @ent: entry in ixgb_pci_tbl
351  *
352  * Returns 0 on success, negative on failure
353  *
354  * ixgb_probe initializes an adapter identified by a pci_dev structure.
355  * The OS initialization, configuring of the adapter private structure,
356  * and a hardware reset occur.
357  **/
358
359 static int __devinit
360 ixgb_probe(struct pci_dev *pdev,
361                 const struct pci_device_id *ent)
362 {
363         struct net_device *netdev = NULL;
364         struct ixgb_adapter *adapter;
365         static int cards_found = 0;
366         unsigned long mmio_start;
367         int mmio_len;
368         int pci_using_dac;
369         int i;
370         int err;
371
372         if((err = pci_enable_device(pdev)))
373                 return err;
374
375         if(!(err = pci_set_dma_mask(pdev, DMA_64BIT_MASK)) &&
376            !(err = pci_set_consistent_dma_mask(pdev, DMA_64BIT_MASK))) {
377                 pci_using_dac = 1;
378         } else {
379                 if((err = pci_set_dma_mask(pdev, DMA_32BIT_MASK)) ||
380                    (err = pci_set_consistent_dma_mask(pdev, DMA_32BIT_MASK))) {
381                         printk(KERN_ERR
382                          "ixgb: No usable DMA configuration, aborting\n");
383                         goto err_dma_mask;
384                 }
385                 pci_using_dac = 0;
386         }
387
388         if((err = pci_request_regions(pdev, ixgb_driver_name)))
389                 goto err_request_regions;
390
391         pci_set_master(pdev);
392
393         netdev = alloc_etherdev(sizeof(struct ixgb_adapter));
394         if(!netdev) {
395                 err = -ENOMEM;
396                 goto err_alloc_etherdev;
397         }
398
399         SET_NETDEV_DEV(netdev, &pdev->dev);
400
401         pci_set_drvdata(pdev, netdev);
402         adapter = netdev_priv(netdev);
403         adapter->netdev = netdev;
404         adapter->pdev = pdev;
405         adapter->hw.back = adapter;
406         adapter->msg_enable = netif_msg_init(debug, DEFAULT_DEBUG_LEVEL_SHIFT);
407
408         mmio_start = pci_resource_start(pdev, BAR_0);
409         mmio_len = pci_resource_len(pdev, BAR_0);
410
411         adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
412         if(!adapter->hw.hw_addr) {
413                 err = -EIO;
414                 goto err_ioremap;
415         }
416
417         for(i = BAR_1; i <= BAR_5; i++) {
418                 if(pci_resource_len(pdev, i) == 0)
419                         continue;
420                 if(pci_resource_flags(pdev, i) & IORESOURCE_IO) {
421                         adapter->hw.io_base = pci_resource_start(pdev, i);
422                         break;
423                 }
424         }
425
426         netdev->open = &ixgb_open;
427         netdev->stop = &ixgb_close;
428         netdev->hard_start_xmit = &ixgb_xmit_frame;
429         netdev->get_stats = &ixgb_get_stats;
430         netdev->set_multicast_list = &ixgb_set_multi;
431         netdev->set_mac_address = &ixgb_set_mac;
432         netdev->change_mtu = &ixgb_change_mtu;
433         ixgb_set_ethtool_ops(netdev);
434         netdev->tx_timeout = &ixgb_tx_timeout;
435         netdev->watchdog_timeo = 5 * HZ;
436 #ifdef CONFIG_IXGB_NAPI
437         netif_napi_add(netdev, &adapter->napi, ixgb_clean, 64);
438 #endif
439         netdev->vlan_rx_register = ixgb_vlan_rx_register;
440         netdev->vlan_rx_add_vid = ixgb_vlan_rx_add_vid;
441         netdev->vlan_rx_kill_vid = ixgb_vlan_rx_kill_vid;
442 #ifdef CONFIG_NET_POLL_CONTROLLER
443         netdev->poll_controller = ixgb_netpoll;
444 #endif
445
446         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
447         netdev->mem_start = mmio_start;
448         netdev->mem_end = mmio_start + mmio_len;
449         netdev->base_addr = adapter->hw.io_base;
450
451         adapter->bd_number = cards_found;
452         adapter->link_speed = 0;
453         adapter->link_duplex = 0;
454
455         /* setup the private structure */
456
457         if((err = ixgb_sw_init(adapter)))
458                 goto err_sw_init;
459
460         netdev->features = NETIF_F_SG |
461                            NETIF_F_HW_CSUM |
462                            NETIF_F_HW_VLAN_TX |
463                            NETIF_F_HW_VLAN_RX |
464                            NETIF_F_HW_VLAN_FILTER;
465         netdev->features |= NETIF_F_TSO;
466 #ifdef NETIF_F_LLTX
467         netdev->features |= NETIF_F_LLTX;
468 #endif
469
470         if(pci_using_dac)
471                 netdev->features |= NETIF_F_HIGHDMA;
472
473         /* make sure the EEPROM is good */
474
475         if(!ixgb_validate_eeprom_checksum(&adapter->hw)) {
476                 DPRINTK(PROBE, ERR, "The EEPROM Checksum Is Not Valid\n");
477                 err = -EIO;
478                 goto err_eeprom;
479         }
480
481         ixgb_get_ee_mac_addr(&adapter->hw, netdev->dev_addr);
482         memcpy(netdev->perm_addr, netdev->dev_addr, netdev->addr_len);
483
484         if(!is_valid_ether_addr(netdev->perm_addr)) {
485                 DPRINTK(PROBE, ERR, "Invalid MAC Address\n");
486                 err = -EIO;
487                 goto err_eeprom;
488         }
489
490         adapter->part_num = ixgb_get_ee_pba_number(&adapter->hw);
491
492         init_timer(&adapter->watchdog_timer);
493         adapter->watchdog_timer.function = &ixgb_watchdog;
494         adapter->watchdog_timer.data = (unsigned long)adapter;
495
496         INIT_WORK(&adapter->tx_timeout_task, ixgb_tx_timeout_task);
497
498         strcpy(netdev->name, "eth%d");
499         if((err = register_netdev(netdev)))
500                 goto err_register;
501
502         /* we're going to reset, so assume we have no link for now */
503
504         netif_carrier_off(netdev);
505         netif_stop_queue(netdev);
506
507         DPRINTK(PROBE, INFO, "Intel(R) PRO/10GbE Network Connection\n");
508         ixgb_check_options(adapter);
509         /* reset the hardware with the new settings */
510
511         ixgb_reset(adapter);
512
513         cards_found++;
514         return 0;
515
516 err_register:
517 err_sw_init:
518 err_eeprom:
519         iounmap(adapter->hw.hw_addr);
520 err_ioremap:
521         free_netdev(netdev);
522 err_alloc_etherdev:
523         pci_release_regions(pdev);
524 err_request_regions:
525 err_dma_mask:
526         pci_disable_device(pdev);
527         return err;
528 }
529
530 /**
531  * ixgb_remove - Device Removal Routine
532  * @pdev: PCI device information struct
533  *
534  * ixgb_remove is called by the PCI subsystem to alert the driver
535  * that it should release a PCI device.  The could be caused by a
536  * Hot-Plug event, or because the driver is going to be removed from
537  * memory.
538  **/
539
540 static void __devexit
541 ixgb_remove(struct pci_dev *pdev)
542 {
543         struct net_device *netdev = pci_get_drvdata(pdev);
544         struct ixgb_adapter *adapter = netdev_priv(netdev);
545
546         unregister_netdev(netdev);
547
548         iounmap(adapter->hw.hw_addr);
549         pci_release_regions(pdev);
550
551         free_netdev(netdev);
552 }
553
554 /**
555  * ixgb_sw_init - Initialize general software structures (struct ixgb_adapter)
556  * @adapter: board private structure to initialize
557  *
558  * ixgb_sw_init initializes the Adapter private data structure.
559  * Fields are initialized based on PCI device information and
560  * OS network device settings (MTU size).
561  **/
562
563 static int __devinit
564 ixgb_sw_init(struct ixgb_adapter *adapter)
565 {
566         struct ixgb_hw *hw = &adapter->hw;
567         struct net_device *netdev = adapter->netdev;
568         struct pci_dev *pdev = adapter->pdev;
569
570         /* PCI config space info */
571
572         hw->vendor_id = pdev->vendor;
573         hw->device_id = pdev->device;
574         hw->subsystem_vendor_id = pdev->subsystem_vendor;
575         hw->subsystem_id = pdev->subsystem_device;
576
577         hw->max_frame_size = netdev->mtu + ENET_HEADER_SIZE + ENET_FCS_LENGTH;
578         adapter->rx_buffer_len = hw->max_frame_size;
579
580         if((hw->device_id == IXGB_DEVICE_ID_82597EX)
581            || (hw->device_id == IXGB_DEVICE_ID_82597EX_CX4)
582            || (hw->device_id == IXGB_DEVICE_ID_82597EX_LR)
583            || (hw->device_id == IXGB_DEVICE_ID_82597EX_SR))
584                         hw->mac_type = ixgb_82597;
585         else {
586                 /* should never have loaded on this device */
587                 DPRINTK(PROBE, ERR, "unsupported device id\n");
588         }
589
590         /* enable flow control to be programmed */
591         hw->fc.send_xon = 1;
592
593         spin_lock_init(&adapter->tx_lock);
594
595         set_bit(__IXGB_DOWN, &adapter->flags);
596         return 0;
597 }
598
599 /**
600  * ixgb_open - Called when a network interface is made active
601  * @netdev: network interface device structure
602  *
603  * Returns 0 on success, negative value on failure
604  *
605  * The open entry point is called when a network interface is made
606  * active by the system (IFF_UP).  At this point all resources needed
607  * for transmit and receive operations are allocated, the interrupt
608  * handler is registered with the OS, the watchdog timer is started,
609  * and the stack is notified that the interface is ready.
610  **/
611
612 static int
613 ixgb_open(struct net_device *netdev)
614 {
615         struct ixgb_adapter *adapter = netdev_priv(netdev);
616         int err;
617
618         /* allocate transmit descriptors */
619
620         if((err = ixgb_setup_tx_resources(adapter)))
621                 goto err_setup_tx;
622
623         /* allocate receive descriptors */
624
625         if((err = ixgb_setup_rx_resources(adapter)))
626                 goto err_setup_rx;
627
628         if((err = ixgb_up(adapter)))
629                 goto err_up;
630
631         return 0;
632
633 err_up:
634         ixgb_free_rx_resources(adapter);
635 err_setup_rx:
636         ixgb_free_tx_resources(adapter);
637 err_setup_tx:
638         ixgb_reset(adapter);
639
640         return err;
641 }
642
643 /**
644  * ixgb_close - Disables a network interface
645  * @netdev: network interface device structure
646  *
647  * Returns 0, this is not allowed to fail
648  *
649  * The close entry point is called when an interface is de-activated
650  * by the OS.  The hardware is still under the drivers control, but
651  * needs to be disabled.  A global MAC reset is issued to stop the
652  * hardware, and all transmit and receive resources are freed.
653  **/
654
655 static int
656 ixgb_close(struct net_device *netdev)
657 {
658         struct ixgb_adapter *adapter = netdev_priv(netdev);
659
660         ixgb_down(adapter, true);
661
662         ixgb_free_tx_resources(adapter);
663         ixgb_free_rx_resources(adapter);
664
665         return 0;
666 }
667
668 /**
669  * ixgb_setup_tx_resources - allocate Tx resources (Descriptors)
670  * @adapter: board private structure
671  *
672  * Return 0 on success, negative on failure
673  **/
674
675 int
676 ixgb_setup_tx_resources(struct ixgb_adapter *adapter)
677 {
678         struct ixgb_desc_ring *txdr = &adapter->tx_ring;
679         struct pci_dev *pdev = adapter->pdev;
680         int size;
681
682         size = sizeof(struct ixgb_buffer) * txdr->count;
683         txdr->buffer_info = vmalloc(size);
684         if(!txdr->buffer_info) {
685                 DPRINTK(PROBE, ERR,
686                  "Unable to allocate transmit descriptor ring memory\n");
687                 return -ENOMEM;
688         }
689         memset(txdr->buffer_info, 0, size);
690
691         /* round up to nearest 4K */
692
693         txdr->size = txdr->count * sizeof(struct ixgb_tx_desc);
694         txdr->size = ALIGN(txdr->size, 4096);
695
696         txdr->desc = pci_alloc_consistent(pdev, txdr->size, &txdr->dma);
697         if(!txdr->desc) {
698                 vfree(txdr->buffer_info);
699                 DPRINTK(PROBE, ERR,
700                  "Unable to allocate transmit descriptor memory\n");
701                 return -ENOMEM;
702         }
703         memset(txdr->desc, 0, txdr->size);
704
705         txdr->next_to_use = 0;
706         txdr->next_to_clean = 0;
707
708         return 0;
709 }
710
711 /**
712  * ixgb_configure_tx - Configure 82597 Transmit Unit after Reset.
713  * @adapter: board private structure
714  *
715  * Configure the Tx unit of the MAC after a reset.
716  **/
717
718 static void
719 ixgb_configure_tx(struct ixgb_adapter *adapter)
720 {
721         u64 tdba = adapter->tx_ring.dma;
722         u32 tdlen = adapter->tx_ring.count * sizeof(struct ixgb_tx_desc);
723         u32 tctl;
724         struct ixgb_hw *hw = &adapter->hw;
725
726         /* Setup the Base and Length of the Tx Descriptor Ring 
727          * tx_ring.dma can be either a 32 or 64 bit value 
728          */
729
730         IXGB_WRITE_REG(hw, TDBAL, (tdba & 0x00000000ffffffffULL));
731         IXGB_WRITE_REG(hw, TDBAH, (tdba >> 32));
732
733         IXGB_WRITE_REG(hw, TDLEN, tdlen);
734
735         /* Setup the HW Tx Head and Tail descriptor pointers */
736
737         IXGB_WRITE_REG(hw, TDH, 0);
738         IXGB_WRITE_REG(hw, TDT, 0);
739
740         /* don't set up txdctl, it induces performance problems if configured
741          * incorrectly */
742         /* Set the Tx Interrupt Delay register */
743
744         IXGB_WRITE_REG(hw, TIDV, adapter->tx_int_delay);
745
746         /* Program the Transmit Control Register */
747
748         tctl = IXGB_TCTL_TCE | IXGB_TCTL_TXEN | IXGB_TCTL_TPDE;
749         IXGB_WRITE_REG(hw, TCTL, tctl);
750
751         /* Setup Transmit Descriptor Settings for this adapter */
752         adapter->tx_cmd_type =
753                 IXGB_TX_DESC_TYPE 
754                 | (adapter->tx_int_delay_enable ? IXGB_TX_DESC_CMD_IDE : 0);
755 }
756
757 /**
758  * ixgb_setup_rx_resources - allocate Rx resources (Descriptors)
759  * @adapter: board private structure
760  *
761  * Returns 0 on success, negative on failure
762  **/
763
764 int
765 ixgb_setup_rx_resources(struct ixgb_adapter *adapter)
766 {
767         struct ixgb_desc_ring *rxdr = &adapter->rx_ring;
768         struct pci_dev *pdev = adapter->pdev;
769         int size;
770
771         size = sizeof(struct ixgb_buffer) * rxdr->count;
772         rxdr->buffer_info = vmalloc(size);
773         if(!rxdr->buffer_info) {
774                 DPRINTK(PROBE, ERR,
775                  "Unable to allocate receive descriptor ring\n");
776                 return -ENOMEM;
777         }
778         memset(rxdr->buffer_info, 0, size);
779
780         /* Round up to nearest 4K */
781
782         rxdr->size = rxdr->count * sizeof(struct ixgb_rx_desc);
783         rxdr->size = ALIGN(rxdr->size, 4096);
784
785         rxdr->desc = pci_alloc_consistent(pdev, rxdr->size, &rxdr->dma);
786
787         if(!rxdr->desc) {
788                 vfree(rxdr->buffer_info);
789                 DPRINTK(PROBE, ERR,
790                  "Unable to allocate receive descriptors\n");
791                 return -ENOMEM;
792         }
793         memset(rxdr->desc, 0, rxdr->size);
794
795         rxdr->next_to_clean = 0;
796         rxdr->next_to_use = 0;
797
798         return 0;
799 }
800
801 /**
802  * ixgb_setup_rctl - configure the receive control register
803  * @adapter: Board private structure
804  **/
805
806 static void
807 ixgb_setup_rctl(struct ixgb_adapter *adapter)
808 {
809         u32 rctl;
810
811         rctl = IXGB_READ_REG(&adapter->hw, RCTL);
812
813         rctl &= ~(3 << IXGB_RCTL_MO_SHIFT);
814
815         rctl |=
816                 IXGB_RCTL_BAM | IXGB_RCTL_RDMTS_1_2 | 
817                 IXGB_RCTL_RXEN | IXGB_RCTL_CFF | 
818                 (adapter->hw.mc_filter_type << IXGB_RCTL_MO_SHIFT);
819
820         rctl |= IXGB_RCTL_SECRC;
821
822         if (adapter->rx_buffer_len <= IXGB_RXBUFFER_2048)
823                 rctl |= IXGB_RCTL_BSIZE_2048;
824         else if (adapter->rx_buffer_len <= IXGB_RXBUFFER_4096)
825                 rctl |= IXGB_RCTL_BSIZE_4096;
826         else if (adapter->rx_buffer_len <= IXGB_RXBUFFER_8192)
827                 rctl |= IXGB_RCTL_BSIZE_8192;
828         else if (adapter->rx_buffer_len <= IXGB_RXBUFFER_16384)
829                 rctl |= IXGB_RCTL_BSIZE_16384;
830
831         IXGB_WRITE_REG(&adapter->hw, RCTL, rctl);
832 }
833
834 /**
835  * ixgb_configure_rx - Configure 82597 Receive Unit after Reset.
836  * @adapter: board private structure
837  *
838  * Configure the Rx unit of the MAC after a reset.
839  **/
840
841 static void
842 ixgb_configure_rx(struct ixgb_adapter *adapter)
843 {
844         u64 rdba = adapter->rx_ring.dma;
845         u32 rdlen = adapter->rx_ring.count * sizeof(struct ixgb_rx_desc);
846         struct ixgb_hw *hw = &adapter->hw;
847         u32 rctl;
848         u32 rxcsum;
849         u32 rxdctl;
850
851         /* make sure receives are disabled while setting up the descriptors */
852
853         rctl = IXGB_READ_REG(hw, RCTL);
854         IXGB_WRITE_REG(hw, RCTL, rctl & ~IXGB_RCTL_RXEN);
855
856         /* set the Receive Delay Timer Register */
857
858         IXGB_WRITE_REG(hw, RDTR, adapter->rx_int_delay);
859
860         /* Setup the Base and Length of the Rx Descriptor Ring */
861
862         IXGB_WRITE_REG(hw, RDBAL, (rdba & 0x00000000ffffffffULL));
863         IXGB_WRITE_REG(hw, RDBAH, (rdba >> 32));
864
865         IXGB_WRITE_REG(hw, RDLEN, rdlen);
866
867         /* Setup the HW Rx Head and Tail Descriptor Pointers */
868         IXGB_WRITE_REG(hw, RDH, 0);
869         IXGB_WRITE_REG(hw, RDT, 0);
870
871         /* set up pre-fetching of receive buffers so we get some before we
872          * run out (default hardware behavior is to run out before fetching
873          * more).  This sets up to fetch if HTHRESH rx descriptors are avail
874          * and the descriptors in hw cache are below PTHRESH.  This avoids
875          * the hardware behavior of fetching <=512 descriptors in a single
876          * burst that pre-empts all other activity, usually causing fifo
877          * overflows. */
878         /* use WTHRESH to burst write 16 descriptors or burst when RXT0 */
879         rxdctl = RXDCTL_WTHRESH_DEFAULT << IXGB_RXDCTL_WTHRESH_SHIFT |
880                  RXDCTL_HTHRESH_DEFAULT << IXGB_RXDCTL_HTHRESH_SHIFT |
881                  RXDCTL_PTHRESH_DEFAULT << IXGB_RXDCTL_PTHRESH_SHIFT;
882         IXGB_WRITE_REG(hw, RXDCTL, rxdctl);
883
884         /* Enable Receive Checksum Offload for TCP and UDP */
885         if (adapter->rx_csum) {
886                 rxcsum = IXGB_READ_REG(hw, RXCSUM);
887                 rxcsum |= IXGB_RXCSUM_TUOFL;
888                 IXGB_WRITE_REG(hw, RXCSUM, rxcsum);
889         }
890
891         /* Enable Receives */
892
893         IXGB_WRITE_REG(hw, RCTL, rctl);
894 }
895
896 /**
897  * ixgb_free_tx_resources - Free Tx Resources
898  * @adapter: board private structure
899  *
900  * Free all transmit software resources
901  **/
902
903 void
904 ixgb_free_tx_resources(struct ixgb_adapter *adapter)
905 {
906         struct pci_dev *pdev = adapter->pdev;
907
908         ixgb_clean_tx_ring(adapter);
909
910         vfree(adapter->tx_ring.buffer_info);
911         adapter->tx_ring.buffer_info = NULL;
912
913         pci_free_consistent(pdev, adapter->tx_ring.size,
914                             adapter->tx_ring.desc, adapter->tx_ring.dma);
915
916         adapter->tx_ring.desc = NULL;
917 }
918
919 static void
920 ixgb_unmap_and_free_tx_resource(struct ixgb_adapter *adapter,
921                                         struct ixgb_buffer *buffer_info)
922 {
923         struct pci_dev *pdev = adapter->pdev;
924
925         if (buffer_info->dma)
926                 pci_unmap_page(pdev, buffer_info->dma, buffer_info->length,
927                                PCI_DMA_TODEVICE);
928
929         if (buffer_info->skb)
930                 dev_kfree_skb_any(buffer_info->skb);
931
932         buffer_info->skb = NULL;
933         buffer_info->dma = 0;
934         buffer_info->time_stamp = 0;
935         /* these fields must always be initialized in tx
936          * buffer_info->length = 0;
937          * buffer_info->next_to_watch = 0; */
938 }
939
940 /**
941  * ixgb_clean_tx_ring - Free Tx Buffers
942  * @adapter: board private structure
943  **/
944
945 static void
946 ixgb_clean_tx_ring(struct ixgb_adapter *adapter)
947 {
948         struct ixgb_desc_ring *tx_ring = &adapter->tx_ring;
949         struct ixgb_buffer *buffer_info;
950         unsigned long size;
951         unsigned int i;
952
953         /* Free all the Tx ring sk_buffs */
954
955         for(i = 0; i < tx_ring->count; i++) {
956                 buffer_info = &tx_ring->buffer_info[i];
957                 ixgb_unmap_and_free_tx_resource(adapter, buffer_info);
958         }
959
960         size = sizeof(struct ixgb_buffer) * tx_ring->count;
961         memset(tx_ring->buffer_info, 0, size);
962
963         /* Zero out the descriptor ring */
964
965         memset(tx_ring->desc, 0, tx_ring->size);
966
967         tx_ring->next_to_use = 0;
968         tx_ring->next_to_clean = 0;
969
970         IXGB_WRITE_REG(&adapter->hw, TDH, 0);
971         IXGB_WRITE_REG(&adapter->hw, TDT, 0);
972 }
973
974 /**
975  * ixgb_free_rx_resources - Free Rx Resources
976  * @adapter: board private structure
977  *
978  * Free all receive software resources
979  **/
980
981 void
982 ixgb_free_rx_resources(struct ixgb_adapter *adapter)
983 {
984         struct ixgb_desc_ring *rx_ring = &adapter->rx_ring;
985         struct pci_dev *pdev = adapter->pdev;
986
987         ixgb_clean_rx_ring(adapter);
988
989         vfree(rx_ring->buffer_info);
990         rx_ring->buffer_info = NULL;
991
992         pci_free_consistent(pdev, rx_ring->size, rx_ring->desc, rx_ring->dma);
993
994         rx_ring->desc = NULL;
995 }
996
997 /**
998  * ixgb_clean_rx_ring - Free Rx Buffers
999  * @adapter: board private structure
1000  **/
1001
1002 static void
1003 ixgb_clean_rx_ring(struct ixgb_adapter *adapter)
1004 {
1005         struct ixgb_desc_ring *rx_ring = &adapter->rx_ring;
1006         struct ixgb_buffer *buffer_info;
1007         struct pci_dev *pdev = adapter->pdev;
1008         unsigned long size;
1009         unsigned int i;
1010
1011         /* Free all the Rx ring sk_buffs */
1012
1013         for(i = 0; i < rx_ring->count; i++) {
1014                 buffer_info = &rx_ring->buffer_info[i];
1015                 if(buffer_info->skb) {
1016
1017                         pci_unmap_single(pdev,
1018                                          buffer_info->dma,
1019                                          buffer_info->length,
1020                                          PCI_DMA_FROMDEVICE);
1021
1022                         dev_kfree_skb(buffer_info->skb);
1023
1024                         buffer_info->skb = NULL;
1025                 }
1026         }
1027
1028         size = sizeof(struct ixgb_buffer) * rx_ring->count;
1029         memset(rx_ring->buffer_info, 0, size);
1030
1031         /* Zero out the descriptor ring */
1032
1033         memset(rx_ring->desc, 0, rx_ring->size);
1034
1035         rx_ring->next_to_clean = 0;
1036         rx_ring->next_to_use = 0;
1037
1038         IXGB_WRITE_REG(&adapter->hw, RDH, 0);
1039         IXGB_WRITE_REG(&adapter->hw, RDT, 0);
1040 }
1041
1042 /**
1043  * ixgb_set_mac - Change the Ethernet Address of the NIC
1044  * @netdev: network interface device structure
1045  * @p: pointer to an address structure
1046  *
1047  * Returns 0 on success, negative on failure
1048  **/
1049
1050 static int
1051 ixgb_set_mac(struct net_device *netdev, void *p)
1052 {
1053         struct ixgb_adapter *adapter = netdev_priv(netdev);
1054         struct sockaddr *addr = p;
1055
1056         if(!is_valid_ether_addr(addr->sa_data))
1057                 return -EADDRNOTAVAIL;
1058
1059         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
1060
1061         ixgb_rar_set(&adapter->hw, addr->sa_data, 0);
1062
1063         return 0;
1064 }
1065
1066 /**
1067  * ixgb_set_multi - Multicast and Promiscuous mode set
1068  * @netdev: network interface device structure
1069  *
1070  * The set_multi entry point is called whenever the multicast address
1071  * list or the network interface flags are updated.  This routine is
1072  * responsible for configuring the hardware for proper multicast,
1073  * promiscuous mode, and all-multi behavior.
1074  **/
1075
1076 static void
1077 ixgb_set_multi(struct net_device *netdev)
1078 {
1079         struct ixgb_adapter *adapter = netdev_priv(netdev);
1080         struct ixgb_hw *hw = &adapter->hw;
1081         struct dev_mc_list *mc_ptr;
1082         u32 rctl;
1083         int i;
1084
1085         /* Check for Promiscuous and All Multicast modes */
1086
1087         rctl = IXGB_READ_REG(hw, RCTL);
1088
1089         if(netdev->flags & IFF_PROMISC) {
1090                 rctl |= (IXGB_RCTL_UPE | IXGB_RCTL_MPE);
1091         } else if(netdev->flags & IFF_ALLMULTI) {
1092                 rctl |= IXGB_RCTL_MPE;
1093                 rctl &= ~IXGB_RCTL_UPE;
1094         } else {
1095                 rctl &= ~(IXGB_RCTL_UPE | IXGB_RCTL_MPE);
1096         }
1097
1098         if(netdev->mc_count > IXGB_MAX_NUM_MULTICAST_ADDRESSES) {
1099                 rctl |= IXGB_RCTL_MPE;
1100                 IXGB_WRITE_REG(hw, RCTL, rctl);
1101         } else {
1102                 u8 mta[IXGB_MAX_NUM_MULTICAST_ADDRESSES *
1103                             IXGB_ETH_LENGTH_OF_ADDRESS];
1104
1105                 IXGB_WRITE_REG(hw, RCTL, rctl);
1106
1107                 for(i = 0, mc_ptr = netdev->mc_list; mc_ptr;
1108                         i++, mc_ptr = mc_ptr->next)
1109                         memcpy(&mta[i * IXGB_ETH_LENGTH_OF_ADDRESS],
1110                                    mc_ptr->dmi_addr, IXGB_ETH_LENGTH_OF_ADDRESS);
1111
1112                 ixgb_mc_addr_list_update(hw, mta, netdev->mc_count, 0);
1113         }
1114 }
1115
1116 /**
1117  * ixgb_watchdog - Timer Call-back
1118  * @data: pointer to netdev cast into an unsigned long
1119  **/
1120
1121 static void
1122 ixgb_watchdog(unsigned long data)
1123 {
1124         struct ixgb_adapter *adapter = (struct ixgb_adapter *)data;
1125         struct net_device *netdev = adapter->netdev;
1126         struct ixgb_desc_ring *txdr = &adapter->tx_ring;
1127
1128         ixgb_check_for_link(&adapter->hw);
1129
1130         if (ixgb_check_for_bad_link(&adapter->hw)) {
1131                 /* force the reset path */
1132                 netif_stop_queue(netdev);
1133         }
1134
1135         if(adapter->hw.link_up) {
1136                 if(!netif_carrier_ok(netdev)) {
1137                         DPRINTK(LINK, INFO,
1138                                 "NIC Link is Up 10000 Mbps Full Duplex\n");
1139                         adapter->link_speed = 10000;
1140                         adapter->link_duplex = FULL_DUPLEX;
1141                         netif_carrier_on(netdev);
1142                         netif_wake_queue(netdev);
1143                 }
1144         } else {
1145                 if(netif_carrier_ok(netdev)) {
1146                         adapter->link_speed = 0;
1147                         adapter->link_duplex = 0;
1148                         DPRINTK(LINK, INFO, "NIC Link is Down\n");
1149                         netif_carrier_off(netdev);
1150                         netif_stop_queue(netdev);
1151
1152                 }
1153         }
1154
1155         ixgb_update_stats(adapter);
1156
1157         if(!netif_carrier_ok(netdev)) {
1158                 if(IXGB_DESC_UNUSED(txdr) + 1 < txdr->count) {
1159                         /* We've lost link, so the controller stops DMA,
1160                          * but we've got queued Tx work that's never going
1161                          * to get done, so reset controller to flush Tx.
1162                          * (Do the reset outside of interrupt context). */
1163                         schedule_work(&adapter->tx_timeout_task);
1164                 }
1165         }
1166
1167         /* Force detection of hung controller every watchdog period */
1168         adapter->detect_tx_hung = true;
1169
1170         /* generate an interrupt to force clean up of any stragglers */
1171         IXGB_WRITE_REG(&adapter->hw, ICS, IXGB_INT_TXDW);
1172
1173         /* Reset the timer */
1174         mod_timer(&adapter->watchdog_timer, jiffies + 2 * HZ);
1175 }
1176
1177 #define IXGB_TX_FLAGS_CSUM              0x00000001
1178 #define IXGB_TX_FLAGS_VLAN              0x00000002
1179 #define IXGB_TX_FLAGS_TSO               0x00000004
1180
1181 static int
1182 ixgb_tso(struct ixgb_adapter *adapter, struct sk_buff *skb)
1183 {
1184         struct ixgb_context_desc *context_desc;
1185         unsigned int i;
1186         u8 ipcss, ipcso, tucss, tucso, hdr_len;
1187         u16 ipcse, tucse, mss;
1188         int err;
1189
1190         if (likely(skb_is_gso(skb))) {
1191                 struct ixgb_buffer *buffer_info;
1192                 struct iphdr *iph;
1193
1194                 if (skb_header_cloned(skb)) {
1195                         err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1196                         if (err)
1197                                 return err;
1198                 }
1199
1200                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
1201                 mss = skb_shinfo(skb)->gso_size;
1202                 iph = ip_hdr(skb);
1203                 iph->tot_len = 0;
1204                 iph->check = 0;
1205                 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
1206                                                          iph->daddr, 0,
1207                                                          IPPROTO_TCP, 0);
1208                 ipcss = skb_network_offset(skb);
1209                 ipcso = (void *)&(iph->check) - (void *)skb->data;
1210                 ipcse = skb_transport_offset(skb) - 1;
1211                 tucss = skb_transport_offset(skb);
1212                 tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
1213                 tucse = 0;
1214
1215                 i = adapter->tx_ring.next_to_use;
1216                 context_desc = IXGB_CONTEXT_DESC(adapter->tx_ring, i);
1217                 buffer_info = &adapter->tx_ring.buffer_info[i];
1218                 WARN_ON(buffer_info->dma != 0);
1219
1220                 context_desc->ipcss = ipcss;
1221                 context_desc->ipcso = ipcso;
1222                 context_desc->ipcse = cpu_to_le16(ipcse);
1223                 context_desc->tucss = tucss;
1224                 context_desc->tucso = tucso;
1225                 context_desc->tucse = cpu_to_le16(tucse);
1226                 context_desc->mss = cpu_to_le16(mss);
1227                 context_desc->hdr_len = hdr_len;
1228                 context_desc->status = 0;
1229                 context_desc->cmd_type_len = cpu_to_le32(
1230                                                   IXGB_CONTEXT_DESC_TYPE 
1231                                                 | IXGB_CONTEXT_DESC_CMD_TSE
1232                                                 | IXGB_CONTEXT_DESC_CMD_IP
1233                                                 | IXGB_CONTEXT_DESC_CMD_TCP
1234                                                 | IXGB_CONTEXT_DESC_CMD_IDE
1235                                                 | (skb->len - (hdr_len)));
1236
1237
1238                 if(++i == adapter->tx_ring.count) i = 0;
1239                 adapter->tx_ring.next_to_use = i;
1240
1241                 return 1;
1242         }
1243
1244         return 0;
1245 }
1246
1247 static bool
1248 ixgb_tx_csum(struct ixgb_adapter *adapter, struct sk_buff *skb)
1249 {
1250         struct ixgb_context_desc *context_desc;
1251         unsigned int i;
1252         u8 css, cso;
1253
1254         if(likely(skb->ip_summed == CHECKSUM_PARTIAL)) {
1255                 struct ixgb_buffer *buffer_info;
1256                 css = skb_transport_offset(skb);
1257                 cso = css + skb->csum_offset;
1258
1259                 i = adapter->tx_ring.next_to_use;
1260                 context_desc = IXGB_CONTEXT_DESC(adapter->tx_ring, i);
1261                 buffer_info = &adapter->tx_ring.buffer_info[i];
1262                 WARN_ON(buffer_info->dma != 0);
1263
1264                 context_desc->tucss = css;
1265                 context_desc->tucso = cso;
1266                 context_desc->tucse = 0;
1267                 /* zero out any previously existing data in one instruction */
1268                 *(u32 *)&(context_desc->ipcss) = 0;
1269                 context_desc->status = 0;
1270                 context_desc->hdr_len = 0;
1271                 context_desc->mss = 0;
1272                 context_desc->cmd_type_len =
1273                         cpu_to_le32(IXGB_CONTEXT_DESC_TYPE
1274                                     | IXGB_TX_DESC_CMD_IDE);
1275
1276                 if(++i == adapter->tx_ring.count) i = 0;
1277                 adapter->tx_ring.next_to_use = i;
1278
1279                 return true;
1280         }
1281
1282         return false;
1283 }
1284
1285 #define IXGB_MAX_TXD_PWR        14
1286 #define IXGB_MAX_DATA_PER_TXD   (1<<IXGB_MAX_TXD_PWR)
1287
1288 static int
1289 ixgb_tx_map(struct ixgb_adapter *adapter, struct sk_buff *skb,
1290             unsigned int first)
1291 {
1292         struct ixgb_desc_ring *tx_ring = &adapter->tx_ring;
1293         struct ixgb_buffer *buffer_info;
1294         int len = skb->len;
1295         unsigned int offset = 0, size, count = 0, i;
1296         unsigned int mss = skb_shinfo(skb)->gso_size;
1297
1298         unsigned int nr_frags = skb_shinfo(skb)->nr_frags;
1299         unsigned int f;
1300
1301         len -= skb->data_len;
1302
1303         i = tx_ring->next_to_use;
1304
1305         while(len) {
1306                 buffer_info = &tx_ring->buffer_info[i];
1307                 size = min(len, IXGB_MAX_DATA_PER_TXD);
1308                 /* Workaround for premature desc write-backs
1309                  * in TSO mode.  Append 4-byte sentinel desc */
1310                 if (unlikely(mss && !nr_frags && size == len && size > 8))
1311                         size -= 4;
1312
1313                 buffer_info->length = size;
1314                 WARN_ON(buffer_info->dma != 0);
1315                 buffer_info->dma =
1316                         pci_map_single(adapter->pdev,
1317                                 skb->data + offset,
1318                                 size,
1319                                 PCI_DMA_TODEVICE);
1320                 buffer_info->time_stamp = jiffies;
1321                 buffer_info->next_to_watch = 0;
1322
1323                 len -= size;
1324                 offset += size;
1325                 count++;
1326                 if(++i == tx_ring->count) i = 0;
1327         }
1328
1329         for(f = 0; f < nr_frags; f++) {
1330                 struct skb_frag_struct *frag;
1331
1332                 frag = &skb_shinfo(skb)->frags[f];
1333                 len = frag->size;
1334                 offset = 0;
1335
1336                 while(len) {
1337                         buffer_info = &tx_ring->buffer_info[i];
1338                         size = min(len, IXGB_MAX_DATA_PER_TXD);
1339
1340                         /* Workaround for premature desc write-backs
1341                          * in TSO mode.  Append 4-byte sentinel desc */
1342                         if (unlikely(mss && (f == (nr_frags - 1))
1343                                      && size == len && size > 8))
1344                                 size -= 4;
1345
1346                         buffer_info->length = size;
1347                         buffer_info->dma =
1348                                 pci_map_page(adapter->pdev,
1349                                         frag->page,
1350                                         frag->page_offset + offset,
1351                                         size,
1352                                         PCI_DMA_TODEVICE);
1353                         buffer_info->time_stamp = jiffies;
1354                         buffer_info->next_to_watch = 0;
1355
1356                         len -= size;
1357                         offset += size;
1358                         count++;
1359                         if(++i == tx_ring->count) i = 0;
1360                 }
1361         }
1362         i = (i == 0) ? tx_ring->count - 1 : i - 1;
1363         tx_ring->buffer_info[i].skb = skb;
1364         tx_ring->buffer_info[first].next_to_watch = i;
1365
1366         return count;
1367 }
1368
1369 static void
1370 ixgb_tx_queue(struct ixgb_adapter *adapter, int count, int vlan_id,int tx_flags)
1371 {
1372         struct ixgb_desc_ring *tx_ring = &adapter->tx_ring;
1373         struct ixgb_tx_desc *tx_desc = NULL;
1374         struct ixgb_buffer *buffer_info;
1375         u32 cmd_type_len = adapter->tx_cmd_type;
1376         u8 status = 0;
1377         u8 popts = 0;
1378         unsigned int i;
1379
1380         if(tx_flags & IXGB_TX_FLAGS_TSO) {
1381                 cmd_type_len |= IXGB_TX_DESC_CMD_TSE;
1382                 popts |= (IXGB_TX_DESC_POPTS_IXSM | IXGB_TX_DESC_POPTS_TXSM);
1383         }
1384
1385         if(tx_flags & IXGB_TX_FLAGS_CSUM)
1386                 popts |= IXGB_TX_DESC_POPTS_TXSM;
1387
1388         if(tx_flags & IXGB_TX_FLAGS_VLAN) {
1389                 cmd_type_len |= IXGB_TX_DESC_CMD_VLE;
1390         }
1391
1392         i = tx_ring->next_to_use;
1393
1394         while(count--) {
1395                 buffer_info = &tx_ring->buffer_info[i];
1396                 tx_desc = IXGB_TX_DESC(*tx_ring, i);
1397                 tx_desc->buff_addr = cpu_to_le64(buffer_info->dma);
1398                 tx_desc->cmd_type_len =
1399                         cpu_to_le32(cmd_type_len | buffer_info->length);
1400                 tx_desc->status = status;
1401                 tx_desc->popts = popts;
1402                 tx_desc->vlan = cpu_to_le16(vlan_id);
1403
1404                 if(++i == tx_ring->count) i = 0;
1405         }
1406
1407         tx_desc->cmd_type_len |= cpu_to_le32(IXGB_TX_DESC_CMD_EOP 
1408                                 | IXGB_TX_DESC_CMD_RS );
1409
1410         /* Force memory writes to complete before letting h/w
1411          * know there are new descriptors to fetch.  (Only
1412          * applicable for weak-ordered memory model archs,
1413          * such as IA-64). */
1414         wmb();
1415
1416         tx_ring->next_to_use = i;
1417         IXGB_WRITE_REG(&adapter->hw, TDT, i);
1418 }
1419
1420 static int __ixgb_maybe_stop_tx(struct net_device *netdev, int size)
1421 {
1422         struct ixgb_adapter *adapter = netdev_priv(netdev);
1423         struct ixgb_desc_ring *tx_ring = &adapter->tx_ring;
1424
1425         netif_stop_queue(netdev);
1426         /* Herbert's original patch had:
1427          *  smp_mb__after_netif_stop_queue();
1428          * but since that doesn't exist yet, just open code it. */
1429         smp_mb();
1430
1431         /* We need to check again in a case another CPU has just
1432          * made room available. */
1433         if (likely(IXGB_DESC_UNUSED(tx_ring) < size))
1434                 return -EBUSY;
1435
1436         /* A reprieve! */
1437         netif_start_queue(netdev);
1438         ++adapter->restart_queue;
1439         return 0;
1440 }
1441
1442 static int ixgb_maybe_stop_tx(struct net_device *netdev,
1443                               struct ixgb_desc_ring *tx_ring, int size)
1444 {
1445         if (likely(IXGB_DESC_UNUSED(tx_ring) >= size))
1446                 return 0;
1447         return __ixgb_maybe_stop_tx(netdev, size);
1448 }
1449
1450
1451 /* Tx Descriptors needed, worst case */
1452 #define TXD_USE_COUNT(S) (((S) >> IXGB_MAX_TXD_PWR) + \
1453                          (((S) & (IXGB_MAX_DATA_PER_TXD - 1)) ? 1 : 0))
1454 #define DESC_NEEDED TXD_USE_COUNT(IXGB_MAX_DATA_PER_TXD) /* skb->date */ + \
1455         MAX_SKB_FRAGS * TXD_USE_COUNT(PAGE_SIZE) + 1 /* for context */ \
1456         + 1 /* one more needed for sentinel TSO workaround */
1457
1458 static int
1459 ixgb_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
1460 {
1461         struct ixgb_adapter *adapter = netdev_priv(netdev);
1462         unsigned int first;
1463         unsigned int tx_flags = 0;
1464         unsigned long flags;
1465         int vlan_id = 0;
1466         int tso;
1467
1468         if (test_bit(__IXGB_DOWN, &adapter->flags)) {
1469                 dev_kfree_skb(skb);
1470                 return NETDEV_TX_OK;
1471         }
1472
1473         if(skb->len <= 0) {
1474                 dev_kfree_skb_any(skb);
1475                 return 0;
1476         }
1477
1478 #ifdef NETIF_F_LLTX
1479         if (!spin_trylock_irqsave(&adapter->tx_lock, flags)) {
1480                 /* Collision - tell upper layer to requeue */
1481                 local_irq_restore(flags);
1482                 return NETDEV_TX_LOCKED;
1483         }
1484 #else
1485         spin_lock_irqsave(&adapter->tx_lock, flags);
1486 #endif
1487
1488         if (unlikely(ixgb_maybe_stop_tx(netdev, &adapter->tx_ring,
1489                      DESC_NEEDED))) {
1490                 netif_stop_queue(netdev);
1491                 spin_unlock_irqrestore(&adapter->tx_lock, flags);
1492                 return NETDEV_TX_BUSY;
1493         }
1494
1495 #ifndef NETIF_F_LLTX
1496         spin_unlock_irqrestore(&adapter->tx_lock, flags);
1497 #endif
1498
1499         if(adapter->vlgrp && vlan_tx_tag_present(skb)) {
1500                 tx_flags |= IXGB_TX_FLAGS_VLAN;
1501                 vlan_id = vlan_tx_tag_get(skb);
1502         }
1503
1504         first = adapter->tx_ring.next_to_use;
1505         
1506         tso = ixgb_tso(adapter, skb);
1507         if (tso < 0) {
1508                 dev_kfree_skb_any(skb);
1509 #ifdef NETIF_F_LLTX
1510                 spin_unlock_irqrestore(&adapter->tx_lock, flags);
1511 #endif
1512                 return NETDEV_TX_OK;
1513         }
1514
1515         if (likely(tso))
1516                 tx_flags |= IXGB_TX_FLAGS_TSO;
1517         else if(ixgb_tx_csum(adapter, skb))
1518                 tx_flags |= IXGB_TX_FLAGS_CSUM;
1519
1520         ixgb_tx_queue(adapter, ixgb_tx_map(adapter, skb, first), vlan_id,
1521                         tx_flags);
1522
1523         netdev->trans_start = jiffies;
1524
1525 #ifdef NETIF_F_LLTX
1526         /* Make sure there is space in the ring for the next send. */
1527         ixgb_maybe_stop_tx(netdev, &adapter->tx_ring, DESC_NEEDED);
1528
1529         spin_unlock_irqrestore(&adapter->tx_lock, flags);
1530
1531 #endif
1532         return NETDEV_TX_OK;
1533 }
1534
1535 /**
1536  * ixgb_tx_timeout - Respond to a Tx Hang
1537  * @netdev: network interface device structure
1538  **/
1539
1540 static void
1541 ixgb_tx_timeout(struct net_device *netdev)
1542 {
1543         struct ixgb_adapter *adapter = netdev_priv(netdev);
1544
1545         /* Do the reset outside of interrupt context */
1546         schedule_work(&adapter->tx_timeout_task);
1547 }
1548
1549 static void
1550 ixgb_tx_timeout_task(struct work_struct *work)
1551 {
1552         struct ixgb_adapter *adapter =
1553                 container_of(work, struct ixgb_adapter, tx_timeout_task);
1554
1555         adapter->tx_timeout_count++;
1556         ixgb_down(adapter, true);
1557         ixgb_up(adapter);
1558 }
1559
1560 /**
1561  * ixgb_get_stats - Get System Network Statistics
1562  * @netdev: network interface device structure
1563  *
1564  * Returns the address of the device statistics structure.
1565  * The statistics are actually updated from the timer callback.
1566  **/
1567
1568 static struct net_device_stats *
1569 ixgb_get_stats(struct net_device *netdev)
1570 {
1571         struct ixgb_adapter *adapter = netdev_priv(netdev);
1572
1573         return &adapter->net_stats;
1574 }
1575
1576 /**
1577  * ixgb_change_mtu - Change the Maximum Transfer Unit
1578  * @netdev: network interface device structure
1579  * @new_mtu: new value for maximum frame size
1580  *
1581  * Returns 0 on success, negative on failure
1582  **/
1583
1584 static int
1585 ixgb_change_mtu(struct net_device *netdev, int new_mtu)
1586 {
1587         struct ixgb_adapter *adapter = netdev_priv(netdev);
1588         int max_frame = new_mtu + ENET_HEADER_SIZE + ENET_FCS_LENGTH;
1589         int old_max_frame = netdev->mtu + ENET_HEADER_SIZE + ENET_FCS_LENGTH;
1590
1591
1592         if((max_frame < IXGB_MIN_ENET_FRAME_SIZE_WITHOUT_FCS + ENET_FCS_LENGTH)
1593            || (max_frame > IXGB_MAX_JUMBO_FRAME_SIZE + ENET_FCS_LENGTH)) {
1594                 DPRINTK(PROBE, ERR, "Invalid MTU setting %d\n", new_mtu);
1595                 return -EINVAL;
1596         }
1597
1598         adapter->rx_buffer_len = max_frame;
1599
1600         netdev->mtu = new_mtu;
1601
1602         if ((old_max_frame != max_frame) && netif_running(netdev)) {
1603                 ixgb_down(adapter, true);
1604                 ixgb_up(adapter);
1605         }
1606
1607         return 0;
1608 }
1609
1610 /**
1611  * ixgb_update_stats - Update the board statistics counters.
1612  * @adapter: board private structure
1613  **/
1614
1615 void
1616 ixgb_update_stats(struct ixgb_adapter *adapter)
1617 {
1618         struct net_device *netdev = adapter->netdev;
1619         struct pci_dev *pdev = adapter->pdev;
1620
1621         /* Prevent stats update while adapter is being reset */
1622         if (pci_channel_offline(pdev))
1623                 return;
1624
1625         if((netdev->flags & IFF_PROMISC) || (netdev->flags & IFF_ALLMULTI) ||
1626            (netdev->mc_count > IXGB_MAX_NUM_MULTICAST_ADDRESSES)) {
1627                 u64 multi = IXGB_READ_REG(&adapter->hw, MPRCL);
1628                 u32 bcast_l = IXGB_READ_REG(&adapter->hw, BPRCL);
1629                 u32 bcast_h = IXGB_READ_REG(&adapter->hw, BPRCH);
1630                 u64 bcast = ((u64)bcast_h << 32) | bcast_l; 
1631
1632                 multi |= ((u64)IXGB_READ_REG(&adapter->hw, MPRCH) << 32);
1633                 /* fix up multicast stats by removing broadcasts */
1634                 if(multi >= bcast)
1635                         multi -= bcast;
1636                 
1637                 adapter->stats.mprcl += (multi & 0xFFFFFFFF);
1638                 adapter->stats.mprch += (multi >> 32);
1639                 adapter->stats.bprcl += bcast_l; 
1640                 adapter->stats.bprch += bcast_h;
1641         } else {
1642                 adapter->stats.mprcl += IXGB_READ_REG(&adapter->hw, MPRCL);
1643                 adapter->stats.mprch += IXGB_READ_REG(&adapter->hw, MPRCH);
1644                 adapter->stats.bprcl += IXGB_READ_REG(&adapter->hw, BPRCL);
1645                 adapter->stats.bprch += IXGB_READ_REG(&adapter->hw, BPRCH);
1646         }
1647         adapter->stats.tprl += IXGB_READ_REG(&adapter->hw, TPRL);
1648         adapter->stats.tprh += IXGB_READ_REG(&adapter->hw, TPRH);
1649         adapter->stats.gprcl += IXGB_READ_REG(&adapter->hw, GPRCL);
1650         adapter->stats.gprch += IXGB_READ_REG(&adapter->hw, GPRCH);
1651         adapter->stats.uprcl += IXGB_READ_REG(&adapter->hw, UPRCL);
1652         adapter->stats.uprch += IXGB_READ_REG(&adapter->hw, UPRCH);
1653         adapter->stats.vprcl += IXGB_READ_REG(&adapter->hw, VPRCL);
1654         adapter->stats.vprch += IXGB_READ_REG(&adapter->hw, VPRCH);
1655         adapter->stats.jprcl += IXGB_READ_REG(&adapter->hw, JPRCL);
1656         adapter->stats.jprch += IXGB_READ_REG(&adapter->hw, JPRCH);
1657         adapter->stats.gorcl += IXGB_READ_REG(&adapter->hw, GORCL);
1658         adapter->stats.gorch += IXGB_READ_REG(&adapter->hw, GORCH);
1659         adapter->stats.torl += IXGB_READ_REG(&adapter->hw, TORL);
1660         adapter->stats.torh += IXGB_READ_REG(&adapter->hw, TORH);
1661         adapter->stats.rnbc += IXGB_READ_REG(&adapter->hw, RNBC);
1662         adapter->stats.ruc += IXGB_READ_REG(&adapter->hw, RUC);
1663         adapter->stats.roc += IXGB_READ_REG(&adapter->hw, ROC);
1664         adapter->stats.rlec += IXGB_READ_REG(&adapter->hw, RLEC);
1665         adapter->stats.crcerrs += IXGB_READ_REG(&adapter->hw, CRCERRS);
1666         adapter->stats.icbc += IXGB_READ_REG(&adapter->hw, ICBC);
1667         adapter->stats.ecbc += IXGB_READ_REG(&adapter->hw, ECBC);
1668         adapter->stats.mpc += IXGB_READ_REG(&adapter->hw, MPC);
1669         adapter->stats.tptl += IXGB_READ_REG(&adapter->hw, TPTL);
1670         adapter->stats.tpth += IXGB_READ_REG(&adapter->hw, TPTH);
1671         adapter->stats.gptcl += IXGB_READ_REG(&adapter->hw, GPTCL);
1672         adapter->stats.gptch += IXGB_READ_REG(&adapter->hw, GPTCH);
1673         adapter->stats.bptcl += IXGB_READ_REG(&adapter->hw, BPTCL);
1674         adapter->stats.bptch += IXGB_READ_REG(&adapter->hw, BPTCH);
1675         adapter->stats.mptcl += IXGB_READ_REG(&adapter->hw, MPTCL);
1676         adapter->stats.mptch += IXGB_READ_REG(&adapter->hw, MPTCH);
1677         adapter->stats.uptcl += IXGB_READ_REG(&adapter->hw, UPTCL);
1678         adapter->stats.uptch += IXGB_READ_REG(&adapter->hw, UPTCH);
1679         adapter->stats.vptcl += IXGB_READ_REG(&adapter->hw, VPTCL);
1680         adapter->stats.vptch += IXGB_READ_REG(&adapter->hw, VPTCH);
1681         adapter->stats.jptcl += IXGB_READ_REG(&adapter->hw, JPTCL);
1682         adapter->stats.jptch += IXGB_READ_REG(&adapter->hw, JPTCH);
1683         adapter->stats.gotcl += IXGB_READ_REG(&adapter->hw, GOTCL);
1684         adapter->stats.gotch += IXGB_READ_REG(&adapter->hw, GOTCH);
1685         adapter->stats.totl += IXGB_READ_REG(&adapter->hw, TOTL);
1686         adapter->stats.toth += IXGB_READ_REG(&adapter->hw, TOTH);
1687         adapter->stats.dc += IXGB_READ_REG(&adapter->hw, DC);
1688         adapter->stats.plt64c += IXGB_READ_REG(&adapter->hw, PLT64C);
1689         adapter->stats.tsctc += IXGB_READ_REG(&adapter->hw, TSCTC);
1690         adapter->stats.tsctfc += IXGB_READ_REG(&adapter->hw, TSCTFC);
1691         adapter->stats.ibic += IXGB_READ_REG(&adapter->hw, IBIC);
1692         adapter->stats.rfc += IXGB_READ_REG(&adapter->hw, RFC);
1693         adapter->stats.lfc += IXGB_READ_REG(&adapter->hw, LFC);
1694         adapter->stats.pfrc += IXGB_READ_REG(&adapter->hw, PFRC);
1695         adapter->stats.pftc += IXGB_READ_REG(&adapter->hw, PFTC);
1696         adapter->stats.mcfrc += IXGB_READ_REG(&adapter->hw, MCFRC);
1697         adapter->stats.mcftc += IXGB_READ_REG(&adapter->hw, MCFTC);
1698         adapter->stats.xonrxc += IXGB_READ_REG(&adapter->hw, XONRXC);
1699         adapter->stats.xontxc += IXGB_READ_REG(&adapter->hw, XONTXC);
1700         adapter->stats.xoffrxc += IXGB_READ_REG(&adapter->hw, XOFFRXC);
1701         adapter->stats.xofftxc += IXGB_READ_REG(&adapter->hw, XOFFTXC);
1702         adapter->stats.rjc += IXGB_READ_REG(&adapter->hw, RJC);
1703
1704         /* Fill out the OS statistics structure */
1705
1706         adapter->net_stats.rx_packets = adapter->stats.gprcl;
1707         adapter->net_stats.tx_packets = adapter->stats.gptcl;
1708         adapter->net_stats.rx_bytes = adapter->stats.gorcl;
1709         adapter->net_stats.tx_bytes = adapter->stats.gotcl;
1710         adapter->net_stats.multicast = adapter->stats.mprcl;
1711         adapter->net_stats.collisions = 0;
1712
1713         /* ignore RLEC as it reports errors for padded (<64bytes) frames
1714          * with a length in the type/len field */
1715         adapter->net_stats.rx_errors =
1716             /* adapter->stats.rnbc + */ adapter->stats.crcerrs +
1717             adapter->stats.ruc +
1718             adapter->stats.roc /*+ adapter->stats.rlec */  +
1719             adapter->stats.icbc +
1720             adapter->stats.ecbc + adapter->stats.mpc;
1721
1722         /* see above
1723          * adapter->net_stats.rx_length_errors = adapter->stats.rlec;
1724          */
1725
1726         adapter->net_stats.rx_crc_errors = adapter->stats.crcerrs;
1727         adapter->net_stats.rx_fifo_errors = adapter->stats.mpc;
1728         adapter->net_stats.rx_missed_errors = adapter->stats.mpc;
1729         adapter->net_stats.rx_over_errors = adapter->stats.mpc;
1730
1731         adapter->net_stats.tx_errors = 0;
1732         adapter->net_stats.rx_frame_errors = 0;
1733         adapter->net_stats.tx_aborted_errors = 0;
1734         adapter->net_stats.tx_carrier_errors = 0;
1735         adapter->net_stats.tx_fifo_errors = 0;
1736         adapter->net_stats.tx_heartbeat_errors = 0;
1737         adapter->net_stats.tx_window_errors = 0;
1738 }
1739
1740 #define IXGB_MAX_INTR 10
1741 /**
1742  * ixgb_intr - Interrupt Handler
1743  * @irq: interrupt number
1744  * @data: pointer to a network interface device structure
1745  **/
1746
1747 static irqreturn_t
1748 ixgb_intr(int irq, void *data)
1749 {
1750         struct net_device *netdev = data;
1751         struct ixgb_adapter *adapter = netdev_priv(netdev);
1752         struct ixgb_hw *hw = &adapter->hw;
1753         u32 icr = IXGB_READ_REG(hw, ICR);
1754 #ifndef CONFIG_IXGB_NAPI
1755         unsigned int i;
1756 #endif
1757
1758         if(unlikely(!icr))
1759                 return IRQ_NONE;  /* Not our interrupt */
1760
1761         if (unlikely(icr & (IXGB_INT_RXSEQ | IXGB_INT_LSC)))
1762                 if (!test_bit(__IXGB_DOWN, &adapter->flags))
1763                         mod_timer(&adapter->watchdog_timer, jiffies);
1764
1765 #ifdef CONFIG_IXGB_NAPI
1766         if (netif_rx_schedule_prep(netdev, &adapter->napi)) {
1767
1768                 /* Disable interrupts and register for poll. The flush 
1769                   of the posted write is intentionally left out.
1770                 */
1771
1772                 IXGB_WRITE_REG(&adapter->hw, IMC, ~0);
1773                 __netif_rx_schedule(netdev, &adapter->napi);
1774         }
1775 #else
1776         /* yes, that is actually a & and it is meant to make sure that
1777          * every pass through this for loop checks both receive and
1778          * transmit queues for completed descriptors, intended to
1779          * avoid starvation issues and assist tx/rx fairness. */
1780         for(i = 0; i < IXGB_MAX_INTR; i++)
1781                 if(!ixgb_clean_rx_irq(adapter) &
1782                    !ixgb_clean_tx_irq(adapter))
1783                         break;
1784 #endif 
1785         return IRQ_HANDLED;
1786 }
1787
1788 #ifdef CONFIG_IXGB_NAPI
1789 /**
1790  * ixgb_clean - NAPI Rx polling callback
1791  * @adapter: board private structure
1792  **/
1793
1794 static int
1795 ixgb_clean(struct napi_struct *napi, int budget)
1796 {
1797         struct ixgb_adapter *adapter = container_of(napi, struct ixgb_adapter, napi);
1798         struct net_device *netdev = adapter->netdev;
1799         int work_done = 0;
1800
1801         ixgb_clean_tx_irq(adapter);
1802         ixgb_clean_rx_irq(adapter, &work_done, budget);
1803
1804         /* If budget not fully consumed, exit the polling mode */
1805         if (work_done < budget) {
1806                 netif_rx_complete(netdev, napi);
1807                 ixgb_irq_enable(adapter);
1808         }
1809
1810         return work_done;
1811 }
1812 #endif
1813
1814 /**
1815  * ixgb_clean_tx_irq - Reclaim resources after transmit completes
1816  * @adapter: board private structure
1817  **/
1818
1819 static bool
1820 ixgb_clean_tx_irq(struct ixgb_adapter *adapter)
1821 {
1822         struct ixgb_desc_ring *tx_ring = &adapter->tx_ring;
1823         struct net_device *netdev = adapter->netdev;
1824         struct ixgb_tx_desc *tx_desc, *eop_desc;
1825         struct ixgb_buffer *buffer_info;
1826         unsigned int i, eop;
1827         bool cleaned = false;
1828
1829         i = tx_ring->next_to_clean;
1830         eop = tx_ring->buffer_info[i].next_to_watch;
1831         eop_desc = IXGB_TX_DESC(*tx_ring, eop);
1832
1833         while(eop_desc->status & IXGB_TX_DESC_STATUS_DD) {
1834
1835                 for (cleaned = false; !cleaned; ) {
1836                         tx_desc = IXGB_TX_DESC(*tx_ring, i);
1837                         buffer_info = &tx_ring->buffer_info[i];
1838
1839                         if (tx_desc->popts
1840                             & (IXGB_TX_DESC_POPTS_TXSM |
1841                                IXGB_TX_DESC_POPTS_IXSM))
1842                                 adapter->hw_csum_tx_good++;
1843
1844                         ixgb_unmap_and_free_tx_resource(adapter, buffer_info);
1845
1846                         *(u32 *)&(tx_desc->status) = 0;
1847
1848                         cleaned = (i == eop);
1849                         if(++i == tx_ring->count) i = 0;
1850                 }
1851
1852                 eop = tx_ring->buffer_info[i].next_to_watch;
1853                 eop_desc = IXGB_TX_DESC(*tx_ring, eop);
1854         }
1855
1856         tx_ring->next_to_clean = i;
1857
1858         if (unlikely(netif_queue_stopped(netdev))) {
1859                 spin_lock(&adapter->tx_lock);
1860                 if (netif_queue_stopped(netdev) && netif_carrier_ok(netdev) &&
1861                     (IXGB_DESC_UNUSED(tx_ring) >= DESC_NEEDED))
1862                         netif_wake_queue(netdev);
1863                 spin_unlock(&adapter->tx_lock);
1864         }
1865
1866         if(adapter->detect_tx_hung) {
1867                 /* detect a transmit hang in hardware, this serializes the
1868                  * check with the clearing of time_stamp and movement of i */
1869                 adapter->detect_tx_hung = false;
1870                 if (tx_ring->buffer_info[eop].dma &&
1871                    time_after(jiffies, tx_ring->buffer_info[eop].time_stamp + HZ)
1872                    && !(IXGB_READ_REG(&adapter->hw, STATUS) &
1873                         IXGB_STATUS_TXOFF)) {
1874                         /* detected Tx unit hang */
1875                         DPRINTK(DRV, ERR, "Detected Tx Unit Hang\n"
1876                                         "  TDH                  <%x>\n"
1877                                         "  TDT                  <%x>\n"
1878                                         "  next_to_use          <%x>\n"
1879                                         "  next_to_clean        <%x>\n"
1880                                         "buffer_info[next_to_clean]\n"
1881                                         "  time_stamp           <%lx>\n"
1882                                         "  next_to_watch        <%x>\n"
1883                                         "  jiffies              <%lx>\n"
1884                                         "  next_to_watch.status <%x>\n",
1885                                 IXGB_READ_REG(&adapter->hw, TDH),
1886                                 IXGB_READ_REG(&adapter->hw, TDT),
1887                                 tx_ring->next_to_use,
1888                                 tx_ring->next_to_clean,
1889                                 tx_ring->buffer_info[eop].time_stamp,
1890                                 eop,
1891                                 jiffies,
1892                                 eop_desc->status);
1893                         netif_stop_queue(netdev);
1894                 }
1895         }
1896
1897         return cleaned;
1898 }
1899
1900 /**
1901  * ixgb_rx_checksum - Receive Checksum Offload for 82597.
1902  * @adapter: board private structure
1903  * @rx_desc: receive descriptor
1904  * @sk_buff: socket buffer with received data
1905  **/
1906
1907 static void
1908 ixgb_rx_checksum(struct ixgb_adapter *adapter,
1909                  struct ixgb_rx_desc *rx_desc,
1910                  struct sk_buff *skb)
1911 {
1912         /* Ignore Checksum bit is set OR
1913          * TCP Checksum has not been calculated
1914          */
1915         if((rx_desc->status & IXGB_RX_DESC_STATUS_IXSM) ||
1916            (!(rx_desc->status & IXGB_RX_DESC_STATUS_TCPCS))) {
1917                 skb->ip_summed = CHECKSUM_NONE;
1918                 return;
1919         }
1920
1921         /* At this point we know the hardware did the TCP checksum */
1922         /* now look at the TCP checksum error bit */
1923         if(rx_desc->errors & IXGB_RX_DESC_ERRORS_TCPE) {
1924                 /* let the stack verify checksum errors */
1925                 skb->ip_summed = CHECKSUM_NONE;
1926                 adapter->hw_csum_rx_error++;
1927         } else {
1928                 /* TCP checksum is good */
1929                 skb->ip_summed = CHECKSUM_UNNECESSARY;
1930                 adapter->hw_csum_rx_good++;
1931         }
1932 }
1933
1934 /**
1935  * ixgb_clean_rx_irq - Send received data up the network stack,
1936  * @adapter: board private structure
1937  **/
1938
1939 static bool
1940 #ifdef CONFIG_IXGB_NAPI
1941 ixgb_clean_rx_irq(struct ixgb_adapter *adapter, int *work_done, int work_to_do)
1942 #else
1943 ixgb_clean_rx_irq(struct ixgb_adapter *adapter)
1944 #endif
1945 {
1946         struct ixgb_desc_ring *rx_ring = &adapter->rx_ring;
1947         struct net_device *netdev = adapter->netdev;
1948         struct pci_dev *pdev = adapter->pdev;
1949         struct ixgb_rx_desc *rx_desc, *next_rxd;
1950         struct ixgb_buffer *buffer_info, *next_buffer, *next2_buffer;
1951         u32 length;
1952         unsigned int i, j;
1953         bool cleaned = false;
1954
1955         i = rx_ring->next_to_clean;
1956         rx_desc = IXGB_RX_DESC(*rx_ring, i);
1957         buffer_info = &rx_ring->buffer_info[i];
1958
1959         while(rx_desc->status & IXGB_RX_DESC_STATUS_DD) {
1960                 struct sk_buff *skb, *next_skb;
1961                 u8 status;
1962
1963 #ifdef CONFIG_IXGB_NAPI
1964                 if(*work_done >= work_to_do)
1965                         break;
1966
1967                 (*work_done)++;
1968 #endif
1969                 status = rx_desc->status;
1970                 skb = buffer_info->skb;
1971                 buffer_info->skb = NULL;
1972
1973                 prefetch(skb->data);
1974
1975                 if(++i == rx_ring->count) i = 0;
1976                 next_rxd = IXGB_RX_DESC(*rx_ring, i);
1977                 prefetch(next_rxd);
1978
1979                 if((j = i + 1) == rx_ring->count) j = 0;
1980                 next2_buffer = &rx_ring->buffer_info[j];
1981                 prefetch(next2_buffer);
1982
1983                 next_buffer = &rx_ring->buffer_info[i];
1984                 next_skb = next_buffer->skb;
1985                 prefetch(next_skb);
1986
1987                 cleaned = true;
1988
1989                 pci_unmap_single(pdev,
1990                                  buffer_info->dma,
1991                                  buffer_info->length,
1992                                  PCI_DMA_FROMDEVICE);
1993
1994                 length = le16_to_cpu(rx_desc->length);
1995
1996                 if(unlikely(!(status & IXGB_RX_DESC_STATUS_EOP))) {
1997
1998                         /* All receives must fit into a single buffer */
1999
2000                         IXGB_DBG("Receive packet consumed multiple buffers "
2001                                          "length<%x>\n", length);
2002
2003                         dev_kfree_skb_irq(skb);
2004                         goto rxdesc_done;
2005                 }
2006
2007                 if (unlikely(rx_desc->errors
2008                              & (IXGB_RX_DESC_ERRORS_CE | IXGB_RX_DESC_ERRORS_SE
2009                                 | IXGB_RX_DESC_ERRORS_P |
2010                                 IXGB_RX_DESC_ERRORS_RXE))) {
2011
2012                         dev_kfree_skb_irq(skb);
2013                         goto rxdesc_done;
2014                 }
2015
2016                 /* code added for copybreak, this should improve
2017                  * performance for small packets with large amounts
2018                  * of reassembly being done in the stack */
2019 #define IXGB_CB_LENGTH 256
2020                 if (length < IXGB_CB_LENGTH) {
2021                         struct sk_buff *new_skb =
2022                             netdev_alloc_skb(netdev, length + NET_IP_ALIGN);
2023                         if (new_skb) {
2024                                 skb_reserve(new_skb, NET_IP_ALIGN);
2025                                 skb_copy_to_linear_data_offset(new_skb,
2026                                                                -NET_IP_ALIGN,
2027                                                                (skb->data -
2028                                                                 NET_IP_ALIGN),
2029                                                                (length +
2030                                                                 NET_IP_ALIGN));
2031                                 /* save the skb in buffer_info as good */
2032                                 buffer_info->skb = skb;
2033                                 skb = new_skb;
2034                         }
2035                 }
2036                 /* end copybreak code */
2037
2038                 /* Good Receive */
2039                 skb_put(skb, length);
2040
2041                 /* Receive Checksum Offload */
2042                 ixgb_rx_checksum(adapter, rx_desc, skb);
2043
2044                 skb->protocol = eth_type_trans(skb, netdev);
2045 #ifdef CONFIG_IXGB_NAPI
2046                 if(adapter->vlgrp && (status & IXGB_RX_DESC_STATUS_VP)) {
2047                         vlan_hwaccel_receive_skb(skb, adapter->vlgrp,
2048                                 le16_to_cpu(rx_desc->special) &
2049                                         IXGB_RX_DESC_SPECIAL_VLAN_MASK);
2050                 } else {
2051                         netif_receive_skb(skb);
2052                 }
2053 #else /* CONFIG_IXGB_NAPI */
2054                 if(adapter->vlgrp && (status & IXGB_RX_DESC_STATUS_VP)) {
2055                         vlan_hwaccel_rx(skb, adapter->vlgrp,
2056                                 le16_to_cpu(rx_desc->special) &
2057                                         IXGB_RX_DESC_SPECIAL_VLAN_MASK);
2058                 } else {
2059                         netif_rx(skb);
2060                 }
2061 #endif /* CONFIG_IXGB_NAPI */
2062                 netdev->last_rx = jiffies;
2063
2064 rxdesc_done:
2065                 /* clean up descriptor, might be written over by hw */
2066                 rx_desc->status = 0;
2067
2068                 /* use prefetched values */
2069                 rx_desc = next_rxd;
2070                 buffer_info = next_buffer;
2071         }
2072
2073         rx_ring->next_to_clean = i;
2074
2075         ixgb_alloc_rx_buffers(adapter);
2076
2077         return cleaned;
2078 }
2079
2080 /**
2081  * ixgb_alloc_rx_buffers - Replace used receive buffers
2082  * @adapter: address of board private structure
2083  **/
2084
2085 static void
2086 ixgb_alloc_rx_buffers(struct ixgb_adapter *adapter)
2087 {
2088         struct ixgb_desc_ring *rx_ring = &adapter->rx_ring;
2089         struct net_device *netdev = adapter->netdev;
2090         struct pci_dev *pdev = adapter->pdev;
2091         struct ixgb_rx_desc *rx_desc;
2092         struct ixgb_buffer *buffer_info;
2093         struct sk_buff *skb;
2094         unsigned int i;
2095         long cleancount;
2096
2097         i = rx_ring->next_to_use;
2098         buffer_info = &rx_ring->buffer_info[i];
2099         cleancount = IXGB_DESC_UNUSED(rx_ring);
2100
2101
2102         /* leave three descriptors unused */
2103         while(--cleancount > 2) {
2104                 /* recycle! its good for you */
2105                 skb = buffer_info->skb;
2106                 if (skb) {
2107                         skb_trim(skb, 0);
2108                         goto map_skb;
2109                 }
2110
2111                 skb = netdev_alloc_skb(netdev, adapter->rx_buffer_len
2112                                        + NET_IP_ALIGN);
2113                 if (unlikely(!skb)) {
2114                         /* Better luck next round */
2115                         adapter->alloc_rx_buff_failed++;
2116                         break;
2117                 }
2118
2119                 /* Make buffer alignment 2 beyond a 16 byte boundary
2120                  * this will result in a 16 byte aligned IP header after
2121                  * the 14 byte MAC header is removed
2122                  */
2123                 skb_reserve(skb, NET_IP_ALIGN);
2124
2125                 buffer_info->skb = skb;
2126                 buffer_info->length = adapter->rx_buffer_len;
2127 map_skb:
2128                 buffer_info->dma = pci_map_single(pdev,
2129                                                   skb->data,
2130                                                   adapter->rx_buffer_len,
2131                                                   PCI_DMA_FROMDEVICE);
2132
2133                 rx_desc = IXGB_RX_DESC(*rx_ring, i);
2134                 rx_desc->buff_addr = cpu_to_le64(buffer_info->dma);
2135                 /* guarantee DD bit not set now before h/w gets descriptor
2136                  * this is the rest of the workaround for h/w double 
2137                  * writeback. */
2138                 rx_desc->status = 0;
2139
2140
2141                 if(++i == rx_ring->count) i = 0;
2142                 buffer_info = &rx_ring->buffer_info[i];
2143         }
2144
2145         if (likely(rx_ring->next_to_use != i)) {
2146                 rx_ring->next_to_use = i;
2147                 if (unlikely(i-- == 0))
2148                         i = (rx_ring->count - 1);
2149
2150                 /* Force memory writes to complete before letting h/w
2151                  * know there are new descriptors to fetch.  (Only
2152                  * applicable for weak-ordered memory model archs, such
2153                  * as IA-64). */
2154                 wmb();
2155                 IXGB_WRITE_REG(&adapter->hw, RDT, i);
2156         }
2157 }
2158
2159 /**
2160  * ixgb_vlan_rx_register - enables or disables vlan tagging/stripping.
2161  * 
2162  * @param netdev network interface device structure
2163  * @param grp indicates to enable or disable tagging/stripping
2164  **/
2165 static void
2166 ixgb_vlan_rx_register(struct net_device *netdev, struct vlan_group *grp)
2167 {
2168         struct ixgb_adapter *adapter = netdev_priv(netdev);
2169         u32 ctrl, rctl;
2170
2171         ixgb_irq_disable(adapter);
2172         adapter->vlgrp = grp;
2173
2174         if(grp) {
2175                 /* enable VLAN tag insert/strip */
2176                 ctrl = IXGB_READ_REG(&adapter->hw, CTRL0);
2177                 ctrl |= IXGB_CTRL0_VME;
2178                 IXGB_WRITE_REG(&adapter->hw, CTRL0, ctrl);
2179
2180                 /* enable VLAN receive filtering */
2181
2182                 rctl = IXGB_READ_REG(&adapter->hw, RCTL);
2183                 rctl |= IXGB_RCTL_VFE;
2184                 rctl &= ~IXGB_RCTL_CFIEN;
2185                 IXGB_WRITE_REG(&adapter->hw, RCTL, rctl);
2186         } else {
2187                 /* disable VLAN tag insert/strip */
2188
2189                 ctrl = IXGB_READ_REG(&adapter->hw, CTRL0);
2190                 ctrl &= ~IXGB_CTRL0_VME;
2191                 IXGB_WRITE_REG(&adapter->hw, CTRL0, ctrl);
2192
2193                 /* disable VLAN filtering */
2194
2195                 rctl = IXGB_READ_REG(&adapter->hw, RCTL);
2196                 rctl &= ~IXGB_RCTL_VFE;
2197                 IXGB_WRITE_REG(&adapter->hw, RCTL, rctl);
2198         }
2199
2200         /* don't enable interrupts unless we are UP */
2201         if (adapter->netdev->flags & IFF_UP)
2202                 ixgb_irq_enable(adapter);
2203 }
2204
2205 static void
2206 ixgb_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
2207 {
2208         struct ixgb_adapter *adapter = netdev_priv(netdev);
2209         u32 vfta, index;
2210
2211         /* add VID to filter table */
2212
2213         index = (vid >> 5) & 0x7F;
2214         vfta = IXGB_READ_REG_ARRAY(&adapter->hw, VFTA, index);
2215         vfta |= (1 << (vid & 0x1F));
2216         ixgb_write_vfta(&adapter->hw, index, vfta);
2217 }
2218
2219 static void
2220 ixgb_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
2221 {
2222         struct ixgb_adapter *adapter = netdev_priv(netdev);
2223         u32 vfta, index;
2224
2225         ixgb_irq_disable(adapter);
2226
2227         vlan_group_set_device(adapter->vlgrp, vid, NULL);
2228
2229         /* don't enable interrupts unless we are UP */
2230         if (adapter->netdev->flags & IFF_UP)
2231                 ixgb_irq_enable(adapter);
2232
2233         /* remove VID from filter table */
2234
2235         index = (vid >> 5) & 0x7F;
2236         vfta = IXGB_READ_REG_ARRAY(&adapter->hw, VFTA, index);
2237         vfta &= ~(1 << (vid & 0x1F));
2238         ixgb_write_vfta(&adapter->hw, index, vfta);
2239 }
2240
2241 static void
2242 ixgb_restore_vlan(struct ixgb_adapter *adapter)
2243 {
2244         ixgb_vlan_rx_register(adapter->netdev, adapter->vlgrp);
2245
2246         if(adapter->vlgrp) {
2247                 u16 vid;
2248                 for(vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
2249                         if(!vlan_group_get_device(adapter->vlgrp, vid))
2250                                 continue;
2251                         ixgb_vlan_rx_add_vid(adapter->netdev, vid);
2252                 }
2253         }
2254 }
2255
2256 #ifdef CONFIG_NET_POLL_CONTROLLER
2257 /*
2258  * Polling 'interrupt' - used by things like netconsole to send skbs
2259  * without having to re-enable interrupts. It's not called while
2260  * the interrupt routine is executing.
2261  */
2262
2263 static void ixgb_netpoll(struct net_device *dev)
2264 {
2265         struct ixgb_adapter *adapter = netdev_priv(dev);
2266
2267         disable_irq(adapter->pdev->irq);
2268         ixgb_intr(adapter->pdev->irq, dev);
2269         enable_irq(adapter->pdev->irq);
2270 }
2271 #endif
2272
2273 /**
2274  * ixgb_io_error_detected() - called when PCI error is detected
2275  * @pdev    pointer to pci device with error
2276  * @state   pci channel state after error
2277  *
2278  * This callback is called by the PCI subsystem whenever
2279  * a PCI bus error is detected.
2280  */
2281 static pci_ers_result_t ixgb_io_error_detected (struct pci_dev *pdev,
2282                                      enum pci_channel_state state)
2283 {
2284         struct net_device *netdev = pci_get_drvdata(pdev);
2285         struct ixgb_adapter *adapter = netdev_priv(netdev);
2286
2287         if(netif_running(netdev))
2288                 ixgb_down(adapter, true);
2289
2290         pci_disable_device(pdev);
2291
2292         /* Request a slot reset. */
2293         return PCI_ERS_RESULT_NEED_RESET;
2294 }
2295
2296 /**
2297  * ixgb_io_slot_reset - called after the pci bus has been reset.
2298  * @pdev    pointer to pci device with error
2299  *
2300  * This callback is called after the PCI buss has been reset.
2301  * Basically, this tries to restart the card from scratch.
2302  * This is a shortened version of the device probe/discovery code,
2303  * it resembles the first-half of the ixgb_probe() routine.
2304  */
2305 static pci_ers_result_t ixgb_io_slot_reset (struct pci_dev *pdev)
2306 {
2307         struct net_device *netdev = pci_get_drvdata(pdev);
2308         struct ixgb_adapter *adapter = netdev_priv(netdev);
2309
2310         if(pci_enable_device(pdev)) {
2311                 DPRINTK(PROBE, ERR, "Cannot re-enable PCI device after reset.\n");
2312                 return PCI_ERS_RESULT_DISCONNECT;
2313         }
2314
2315         /* Perform card reset only on one instance of the card */
2316         if (0 != PCI_FUNC (pdev->devfn))
2317                 return PCI_ERS_RESULT_RECOVERED;
2318
2319         pci_set_master(pdev);
2320
2321         netif_carrier_off(netdev);
2322         netif_stop_queue(netdev);
2323         ixgb_reset(adapter);
2324
2325         /* Make sure the EEPROM is good */
2326         if(!ixgb_validate_eeprom_checksum(&adapter->hw)) {
2327                 DPRINTK(PROBE, ERR, "After reset, the EEPROM checksum is not valid.\n");
2328                 return PCI_ERS_RESULT_DISCONNECT;
2329         }
2330         ixgb_get_ee_mac_addr(&adapter->hw, netdev->dev_addr);
2331         memcpy(netdev->perm_addr, netdev->dev_addr, netdev->addr_len);
2332
2333         if(!is_valid_ether_addr(netdev->perm_addr)) {
2334                 DPRINTK(PROBE, ERR, "After reset, invalid MAC address.\n");
2335                 return PCI_ERS_RESULT_DISCONNECT;
2336         }
2337
2338         return PCI_ERS_RESULT_RECOVERED;
2339 }
2340
2341 /**
2342  * ixgb_io_resume - called when its OK to resume normal operations
2343  * @pdev    pointer to pci device with error
2344  *
2345  * The error recovery driver tells us that its OK to resume
2346  * normal operation. Implementation resembles the second-half
2347  * of the ixgb_probe() routine.
2348  */
2349 static void ixgb_io_resume (struct pci_dev *pdev)
2350 {
2351         struct net_device *netdev = pci_get_drvdata(pdev);
2352         struct ixgb_adapter *adapter = netdev_priv(netdev);
2353
2354         pci_set_master(pdev);
2355
2356         if(netif_running(netdev)) {
2357                 if(ixgb_up(adapter)) {
2358                         printk ("ixgb: can't bring device back up after reset\n");
2359                         return;
2360                 }
2361         }
2362
2363         netif_device_attach(netdev);
2364         mod_timer(&adapter->watchdog_timer, jiffies);
2365 }
2366
2367 /* ixgb_main.c */