Merge branch 'e1000-fixes' of master.kernel.org:/pub/scm/linux/kernel/git/jgarzik...
[pandora-kernel.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         netdev->features |= NETIF_F_TSO;
460 #ifdef NETIF_F_LLTX
461         netdev->features |= NETIF_F_LLTX;
462 #endif
463
464         if(pci_using_dac)
465                 netdev->features |= NETIF_F_HIGHDMA;
466
467         /* make sure the EEPROM is good */
468
469         if(!ixgb_validate_eeprom_checksum(&adapter->hw)) {
470                 DPRINTK(PROBE, ERR, "The EEPROM Checksum Is Not Valid\n");
471                 err = -EIO;
472                 goto err_eeprom;
473         }
474
475         ixgb_get_ee_mac_addr(&adapter->hw, netdev->dev_addr);
476         memcpy(netdev->perm_addr, netdev->dev_addr, netdev->addr_len);
477
478         if(!is_valid_ether_addr(netdev->perm_addr)) {
479                 DPRINTK(PROBE, ERR, "Invalid MAC Address\n");
480                 err = -EIO;
481                 goto err_eeprom;
482         }
483
484         adapter->part_num = ixgb_get_ee_pba_number(&adapter->hw);
485
486         init_timer(&adapter->watchdog_timer);
487         adapter->watchdog_timer.function = &ixgb_watchdog;
488         adapter->watchdog_timer.data = (unsigned long)adapter;
489
490         INIT_WORK(&adapter->tx_timeout_task, ixgb_tx_timeout_task);
491
492         strcpy(netdev->name, "eth%d");
493         if((err = register_netdev(netdev)))
494                 goto err_register;
495
496         /* we're going to reset, so assume we have no link for now */
497
498         netif_carrier_off(netdev);
499         netif_stop_queue(netdev);
500
501         DPRINTK(PROBE, INFO, "Intel(R) PRO/10GbE Network Connection\n");
502         ixgb_check_options(adapter);
503         /* reset the hardware with the new settings */
504
505         ixgb_reset(adapter);
506
507         cards_found++;
508         return 0;
509
510 err_register:
511 err_sw_init:
512 err_eeprom:
513         iounmap(adapter->hw.hw_addr);
514 err_ioremap:
515         free_netdev(netdev);
516 err_alloc_etherdev:
517         pci_release_regions(pdev);
518 err_request_regions:
519 err_dma_mask:
520         pci_disable_device(pdev);
521         return err;
522 }
523
524 /**
525  * ixgb_remove - Device Removal Routine
526  * @pdev: PCI device information struct
527  *
528  * ixgb_remove is called by the PCI subsystem to alert the driver
529  * that it should release a PCI device.  The could be caused by a
530  * Hot-Plug event, or because the driver is going to be removed from
531  * memory.
532  **/
533
534 static void __devexit
535 ixgb_remove(struct pci_dev *pdev)
536 {
537         struct net_device *netdev = pci_get_drvdata(pdev);
538         struct ixgb_adapter *adapter = netdev_priv(netdev);
539
540         unregister_netdev(netdev);
541
542         iounmap(adapter->hw.hw_addr);
543         pci_release_regions(pdev);
544
545         free_netdev(netdev);
546 }
547
548 /**
549  * ixgb_sw_init - Initialize general software structures (struct ixgb_adapter)
550  * @adapter: board private structure to initialize
551  *
552  * ixgb_sw_init initializes the Adapter private data structure.
553  * Fields are initialized based on PCI device information and
554  * OS network device settings (MTU size).
555  **/
556
557 static int __devinit
558 ixgb_sw_init(struct ixgb_adapter *adapter)
559 {
560         struct ixgb_hw *hw = &adapter->hw;
561         struct net_device *netdev = adapter->netdev;
562         struct pci_dev *pdev = adapter->pdev;
563
564         /* PCI config space info */
565
566         hw->vendor_id = pdev->vendor;
567         hw->device_id = pdev->device;
568         hw->subsystem_vendor_id = pdev->subsystem_vendor;
569         hw->subsystem_id = pdev->subsystem_device;
570
571         hw->max_frame_size = netdev->mtu + ENET_HEADER_SIZE + ENET_FCS_LENGTH;
572         adapter->rx_buffer_len = hw->max_frame_size;
573
574         if((hw->device_id == IXGB_DEVICE_ID_82597EX)
575            || (hw->device_id == IXGB_DEVICE_ID_82597EX_CX4)
576            || (hw->device_id == IXGB_DEVICE_ID_82597EX_LR)
577            || (hw->device_id == IXGB_DEVICE_ID_82597EX_SR))
578                         hw->mac_type = ixgb_82597;
579         else {
580                 /* should never have loaded on this device */
581                 DPRINTK(PROBE, ERR, "unsupported device id\n");
582         }
583
584         /* enable flow control to be programmed */
585         hw->fc.send_xon = 1;
586
587         atomic_set(&adapter->irq_sem, 1);
588         spin_lock_init(&adapter->tx_lock);
589
590         return 0;
591 }
592
593 /**
594  * ixgb_open - Called when a network interface is made active
595  * @netdev: network interface device structure
596  *
597  * Returns 0 on success, negative value on failure
598  *
599  * The open entry point is called when a network interface is made
600  * active by the system (IFF_UP).  At this point all resources needed
601  * for transmit and receive operations are allocated, the interrupt
602  * handler is registered with the OS, the watchdog timer is started,
603  * and the stack is notified that the interface is ready.
604  **/
605
606 static int
607 ixgb_open(struct net_device *netdev)
608 {
609         struct ixgb_adapter *adapter = netdev_priv(netdev);
610         int err;
611
612         /* allocate transmit descriptors */
613
614         if((err = ixgb_setup_tx_resources(adapter)))
615                 goto err_setup_tx;
616
617         /* allocate receive descriptors */
618
619         if((err = ixgb_setup_rx_resources(adapter)))
620                 goto err_setup_rx;
621
622         if((err = ixgb_up(adapter)))
623                 goto err_up;
624
625         return 0;
626
627 err_up:
628         ixgb_free_rx_resources(adapter);
629 err_setup_rx:
630         ixgb_free_tx_resources(adapter);
631 err_setup_tx:
632         ixgb_reset(adapter);
633
634         return err;
635 }
636
637 /**
638  * ixgb_close - Disables a network interface
639  * @netdev: network interface device structure
640  *
641  * Returns 0, this is not allowed to fail
642  *
643  * The close entry point is called when an interface is de-activated
644  * by the OS.  The hardware is still under the drivers control, but
645  * needs to be disabled.  A global MAC reset is issued to stop the
646  * hardware, and all transmit and receive resources are freed.
647  **/
648
649 static int
650 ixgb_close(struct net_device *netdev)
651 {
652         struct ixgb_adapter *adapter = netdev_priv(netdev);
653
654         ixgb_down(adapter, TRUE);
655
656         ixgb_free_tx_resources(adapter);
657         ixgb_free_rx_resources(adapter);
658
659         return 0;
660 }
661
662 /**
663  * ixgb_setup_tx_resources - allocate Tx resources (Descriptors)
664  * @adapter: board private structure
665  *
666  * Return 0 on success, negative on failure
667  **/
668
669 int
670 ixgb_setup_tx_resources(struct ixgb_adapter *adapter)
671 {
672         struct ixgb_desc_ring *txdr = &adapter->tx_ring;
673         struct pci_dev *pdev = adapter->pdev;
674         int size;
675
676         size = sizeof(struct ixgb_buffer) * txdr->count;
677         txdr->buffer_info = vmalloc(size);
678         if(!txdr->buffer_info) {
679                 DPRINTK(PROBE, ERR,
680                  "Unable to allocate transmit descriptor ring memory\n");
681                 return -ENOMEM;
682         }
683         memset(txdr->buffer_info, 0, size);
684
685         /* round up to nearest 4K */
686
687         txdr->size = txdr->count * sizeof(struct ixgb_tx_desc);
688         IXGB_ROUNDUP(txdr->size, 4096);
689
690         txdr->desc = pci_alloc_consistent(pdev, txdr->size, &txdr->dma);
691         if(!txdr->desc) {
692                 vfree(txdr->buffer_info);
693                 DPRINTK(PROBE, ERR,
694                  "Unable to allocate transmit descriptor memory\n");
695                 return -ENOMEM;
696         }
697         memset(txdr->desc, 0, txdr->size);
698
699         txdr->next_to_use = 0;
700         txdr->next_to_clean = 0;
701
702         return 0;
703 }
704
705 /**
706  * ixgb_configure_tx - Configure 82597 Transmit Unit after Reset.
707  * @adapter: board private structure
708  *
709  * Configure the Tx unit of the MAC after a reset.
710  **/
711
712 static void
713 ixgb_configure_tx(struct ixgb_adapter *adapter)
714 {
715         uint64_t tdba = adapter->tx_ring.dma;
716         uint32_t tdlen = adapter->tx_ring.count * sizeof(struct ixgb_tx_desc);
717         uint32_t tctl;
718         struct ixgb_hw *hw = &adapter->hw;
719
720         /* Setup the Base and Length of the Tx Descriptor Ring 
721          * tx_ring.dma can be either a 32 or 64 bit value 
722          */
723
724         IXGB_WRITE_REG(hw, TDBAL, (tdba & 0x00000000ffffffffULL));
725         IXGB_WRITE_REG(hw, TDBAH, (tdba >> 32));
726
727         IXGB_WRITE_REG(hw, TDLEN, tdlen);
728
729         /* Setup the HW Tx Head and Tail descriptor pointers */
730
731         IXGB_WRITE_REG(hw, TDH, 0);
732         IXGB_WRITE_REG(hw, TDT, 0);
733
734         /* don't set up txdctl, it induces performance problems if configured
735          * incorrectly */
736         /* Set the Tx Interrupt Delay register */
737
738         IXGB_WRITE_REG(hw, TIDV, adapter->tx_int_delay);
739
740         /* Program the Transmit Control Register */
741
742         tctl = IXGB_TCTL_TCE | IXGB_TCTL_TXEN | IXGB_TCTL_TPDE;
743         IXGB_WRITE_REG(hw, TCTL, tctl);
744
745         /* Setup Transmit Descriptor Settings for this adapter */
746         adapter->tx_cmd_type =
747                 IXGB_TX_DESC_TYPE 
748                 | (adapter->tx_int_delay_enable ? IXGB_TX_DESC_CMD_IDE : 0);
749 }
750
751 /**
752  * ixgb_setup_rx_resources - allocate Rx resources (Descriptors)
753  * @adapter: board private structure
754  *
755  * Returns 0 on success, negative on failure
756  **/
757
758 int
759 ixgb_setup_rx_resources(struct ixgb_adapter *adapter)
760 {
761         struct ixgb_desc_ring *rxdr = &adapter->rx_ring;
762         struct pci_dev *pdev = adapter->pdev;
763         int size;
764
765         size = sizeof(struct ixgb_buffer) * rxdr->count;
766         rxdr->buffer_info = vmalloc(size);
767         if(!rxdr->buffer_info) {
768                 DPRINTK(PROBE, ERR,
769                  "Unable to allocate receive descriptor ring\n");
770                 return -ENOMEM;
771         }
772         memset(rxdr->buffer_info, 0, size);
773
774         /* Round up to nearest 4K */
775
776         rxdr->size = rxdr->count * sizeof(struct ixgb_rx_desc);
777         IXGB_ROUNDUP(rxdr->size, 4096);
778
779         rxdr->desc = pci_alloc_consistent(pdev, rxdr->size, &rxdr->dma);
780
781         if(!rxdr->desc) {
782                 vfree(rxdr->buffer_info);
783                 DPRINTK(PROBE, ERR,
784                  "Unable to allocate receive descriptors\n");
785                 return -ENOMEM;
786         }
787         memset(rxdr->desc, 0, rxdr->size);
788
789         rxdr->next_to_clean = 0;
790         rxdr->next_to_use = 0;
791
792         return 0;
793 }
794
795 /**
796  * ixgb_setup_rctl - configure the receive control register
797  * @adapter: Board private structure
798  **/
799
800 static void
801 ixgb_setup_rctl(struct ixgb_adapter *adapter)
802 {
803         uint32_t rctl;
804
805         rctl = IXGB_READ_REG(&adapter->hw, RCTL);
806
807         rctl &= ~(3 << IXGB_RCTL_MO_SHIFT);
808
809         rctl |=
810                 IXGB_RCTL_BAM | IXGB_RCTL_RDMTS_1_2 | 
811                 IXGB_RCTL_RXEN | IXGB_RCTL_CFF | 
812                 (adapter->hw.mc_filter_type << IXGB_RCTL_MO_SHIFT);
813
814         rctl |= IXGB_RCTL_SECRC;
815
816         if (adapter->rx_buffer_len <= IXGB_RXBUFFER_2048)
817                 rctl |= IXGB_RCTL_BSIZE_2048;
818         else if (adapter->rx_buffer_len <= IXGB_RXBUFFER_4096)
819                 rctl |= IXGB_RCTL_BSIZE_4096;
820         else if (adapter->rx_buffer_len <= IXGB_RXBUFFER_8192)
821                 rctl |= IXGB_RCTL_BSIZE_8192;
822         else if (adapter->rx_buffer_len <= IXGB_RXBUFFER_16384)
823                 rctl |= IXGB_RCTL_BSIZE_16384;
824
825         IXGB_WRITE_REG(&adapter->hw, RCTL, rctl);
826 }
827
828 /**
829  * ixgb_configure_rx - Configure 82597 Receive Unit after Reset.
830  * @adapter: board private structure
831  *
832  * Configure the Rx unit of the MAC after a reset.
833  **/
834
835 static void
836 ixgb_configure_rx(struct ixgb_adapter *adapter)
837 {
838         uint64_t rdba = adapter->rx_ring.dma;
839         uint32_t rdlen = adapter->rx_ring.count * sizeof(struct ixgb_rx_desc);
840         struct ixgb_hw *hw = &adapter->hw;
841         uint32_t rctl;
842         uint32_t rxcsum;
843         uint32_t rxdctl;
844
845         /* make sure receives are disabled while setting up the descriptors */
846
847         rctl = IXGB_READ_REG(hw, RCTL);
848         IXGB_WRITE_REG(hw, RCTL, rctl & ~IXGB_RCTL_RXEN);
849
850         /* set the Receive Delay Timer Register */
851
852         IXGB_WRITE_REG(hw, RDTR, adapter->rx_int_delay);
853
854         /* Setup the Base and Length of the Rx Descriptor Ring */
855
856         IXGB_WRITE_REG(hw, RDBAL, (rdba & 0x00000000ffffffffULL));
857         IXGB_WRITE_REG(hw, RDBAH, (rdba >> 32));
858
859         IXGB_WRITE_REG(hw, RDLEN, rdlen);
860
861         /* Setup the HW Rx Head and Tail Descriptor Pointers */
862         IXGB_WRITE_REG(hw, RDH, 0);
863         IXGB_WRITE_REG(hw, RDT, 0);
864
865         /* set up pre-fetching of receive buffers so we get some before we
866          * run out (default hardware behavior is to run out before fetching
867          * more).  This sets up to fetch if HTHRESH rx descriptors are avail
868          * and the descriptors in hw cache are below PTHRESH.  This avoids
869          * the hardware behavior of fetching <=512 descriptors in a single
870          * burst that pre-empts all other activity, usually causing fifo
871          * overflows. */
872         /* use WTHRESH to burst write 16 descriptors or burst when RXT0 */
873         rxdctl = RXDCTL_WTHRESH_DEFAULT << IXGB_RXDCTL_WTHRESH_SHIFT |
874                  RXDCTL_HTHRESH_DEFAULT << IXGB_RXDCTL_HTHRESH_SHIFT |
875                  RXDCTL_PTHRESH_DEFAULT << IXGB_RXDCTL_PTHRESH_SHIFT;
876         IXGB_WRITE_REG(hw, RXDCTL, rxdctl);
877
878         /* Enable Receive Checksum Offload for TCP and UDP */
879         if(adapter->rx_csum == TRUE) {
880                 rxcsum = IXGB_READ_REG(hw, RXCSUM);
881                 rxcsum |= IXGB_RXCSUM_TUOFL;
882                 IXGB_WRITE_REG(hw, RXCSUM, rxcsum);
883         }
884
885         /* Enable Receives */
886
887         IXGB_WRITE_REG(hw, RCTL, rctl);
888 }
889
890 /**
891  * ixgb_free_tx_resources - Free Tx Resources
892  * @adapter: board private structure
893  *
894  * Free all transmit software resources
895  **/
896
897 void
898 ixgb_free_tx_resources(struct ixgb_adapter *adapter)
899 {
900         struct pci_dev *pdev = adapter->pdev;
901
902         ixgb_clean_tx_ring(adapter);
903
904         vfree(adapter->tx_ring.buffer_info);
905         adapter->tx_ring.buffer_info = NULL;
906
907         pci_free_consistent(pdev, adapter->tx_ring.size,
908                             adapter->tx_ring.desc, adapter->tx_ring.dma);
909
910         adapter->tx_ring.desc = NULL;
911 }
912
913 static void
914 ixgb_unmap_and_free_tx_resource(struct ixgb_adapter *adapter,
915                                         struct ixgb_buffer *buffer_info)
916 {
917         struct pci_dev *pdev = adapter->pdev;
918
919         if (buffer_info->dma)
920                 pci_unmap_page(pdev, buffer_info->dma, buffer_info->length,
921                                PCI_DMA_TODEVICE);
922
923         if (buffer_info->skb)
924                 dev_kfree_skb_any(buffer_info->skb);
925
926         buffer_info->skb = NULL;
927         buffer_info->dma = 0;
928         buffer_info->time_stamp = 0;
929         /* these fields must always be initialized in tx
930          * buffer_info->length = 0;
931          * buffer_info->next_to_watch = 0; */
932 }
933
934 /**
935  * ixgb_clean_tx_ring - Free Tx Buffers
936  * @adapter: board private structure
937  **/
938
939 static void
940 ixgb_clean_tx_ring(struct ixgb_adapter *adapter)
941 {
942         struct ixgb_desc_ring *tx_ring = &adapter->tx_ring;
943         struct ixgb_buffer *buffer_info;
944         unsigned long size;
945         unsigned int i;
946
947         /* Free all the Tx ring sk_buffs */
948
949         for(i = 0; i < tx_ring->count; i++) {
950                 buffer_info = &tx_ring->buffer_info[i];
951                 ixgb_unmap_and_free_tx_resource(adapter, buffer_info);
952         }
953
954         size = sizeof(struct ixgb_buffer) * tx_ring->count;
955         memset(tx_ring->buffer_info, 0, size);
956
957         /* Zero out the descriptor ring */
958
959         memset(tx_ring->desc, 0, tx_ring->size);
960
961         tx_ring->next_to_use = 0;
962         tx_ring->next_to_clean = 0;
963
964         IXGB_WRITE_REG(&adapter->hw, TDH, 0);
965         IXGB_WRITE_REG(&adapter->hw, TDT, 0);
966 }
967
968 /**
969  * ixgb_free_rx_resources - Free Rx Resources
970  * @adapter: board private structure
971  *
972  * Free all receive software resources
973  **/
974
975 void
976 ixgb_free_rx_resources(struct ixgb_adapter *adapter)
977 {
978         struct ixgb_desc_ring *rx_ring = &adapter->rx_ring;
979         struct pci_dev *pdev = adapter->pdev;
980
981         ixgb_clean_rx_ring(adapter);
982
983         vfree(rx_ring->buffer_info);
984         rx_ring->buffer_info = NULL;
985
986         pci_free_consistent(pdev, rx_ring->size, rx_ring->desc, rx_ring->dma);
987
988         rx_ring->desc = NULL;
989 }
990
991 /**
992  * ixgb_clean_rx_ring - Free Rx Buffers
993  * @adapter: board private structure
994  **/
995
996 static void
997 ixgb_clean_rx_ring(struct ixgb_adapter *adapter)
998 {
999         struct ixgb_desc_ring *rx_ring = &adapter->rx_ring;
1000         struct ixgb_buffer *buffer_info;
1001         struct pci_dev *pdev = adapter->pdev;
1002         unsigned long size;
1003         unsigned int i;
1004
1005         /* Free all the Rx ring sk_buffs */
1006
1007         for(i = 0; i < rx_ring->count; i++) {
1008                 buffer_info = &rx_ring->buffer_info[i];
1009                 if(buffer_info->skb) {
1010
1011                         pci_unmap_single(pdev,
1012                                          buffer_info->dma,
1013                                          buffer_info->length,
1014                                          PCI_DMA_FROMDEVICE);
1015
1016                         dev_kfree_skb(buffer_info->skb);
1017
1018                         buffer_info->skb = NULL;
1019                 }
1020         }
1021
1022         size = sizeof(struct ixgb_buffer) * rx_ring->count;
1023         memset(rx_ring->buffer_info, 0, size);
1024
1025         /* Zero out the descriptor ring */
1026
1027         memset(rx_ring->desc, 0, rx_ring->size);
1028
1029         rx_ring->next_to_clean = 0;
1030         rx_ring->next_to_use = 0;
1031
1032         IXGB_WRITE_REG(&adapter->hw, RDH, 0);
1033         IXGB_WRITE_REG(&adapter->hw, RDT, 0);
1034 }
1035
1036 /**
1037  * ixgb_set_mac - Change the Ethernet Address of the NIC
1038  * @netdev: network interface device structure
1039  * @p: pointer to an address structure
1040  *
1041  * Returns 0 on success, negative on failure
1042  **/
1043
1044 static int
1045 ixgb_set_mac(struct net_device *netdev, void *p)
1046 {
1047         struct ixgb_adapter *adapter = netdev_priv(netdev);
1048         struct sockaddr *addr = p;
1049
1050         if(!is_valid_ether_addr(addr->sa_data))
1051                 return -EADDRNOTAVAIL;
1052
1053         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
1054
1055         ixgb_rar_set(&adapter->hw, addr->sa_data, 0);
1056
1057         return 0;
1058 }
1059
1060 /**
1061  * ixgb_set_multi - Multicast and Promiscuous mode set
1062  * @netdev: network interface device structure
1063  *
1064  * The set_multi entry point is called whenever the multicast address
1065  * list or the network interface flags are updated.  This routine is
1066  * responsible for configuring the hardware for proper multicast,
1067  * promiscuous mode, and all-multi behavior.
1068  **/
1069
1070 static void
1071 ixgb_set_multi(struct net_device *netdev)
1072 {
1073         struct ixgb_adapter *adapter = netdev_priv(netdev);
1074         struct ixgb_hw *hw = &adapter->hw;
1075         struct dev_mc_list *mc_ptr;
1076         uint32_t rctl;
1077         int i;
1078
1079         /* Check for Promiscuous and All Multicast modes */
1080
1081         rctl = IXGB_READ_REG(hw, RCTL);
1082
1083         if(netdev->flags & IFF_PROMISC) {
1084                 rctl |= (IXGB_RCTL_UPE | IXGB_RCTL_MPE);
1085         } else if(netdev->flags & IFF_ALLMULTI) {
1086                 rctl |= IXGB_RCTL_MPE;
1087                 rctl &= ~IXGB_RCTL_UPE;
1088         } else {
1089                 rctl &= ~(IXGB_RCTL_UPE | IXGB_RCTL_MPE);
1090         }
1091
1092         if(netdev->mc_count > IXGB_MAX_NUM_MULTICAST_ADDRESSES) {
1093                 rctl |= IXGB_RCTL_MPE;
1094                 IXGB_WRITE_REG(hw, RCTL, rctl);
1095         } else {
1096                 uint8_t mta[netdev->mc_count * IXGB_ETH_LENGTH_OF_ADDRESS];
1097
1098                 IXGB_WRITE_REG(hw, RCTL, rctl);
1099
1100                 for(i = 0, mc_ptr = netdev->mc_list; mc_ptr;
1101                         i++, mc_ptr = mc_ptr->next)
1102                         memcpy(&mta[i * IXGB_ETH_LENGTH_OF_ADDRESS],
1103                                    mc_ptr->dmi_addr, IXGB_ETH_LENGTH_OF_ADDRESS);
1104
1105                 ixgb_mc_addr_list_update(hw, mta, netdev->mc_count, 0);
1106         }
1107 }
1108
1109 /**
1110  * ixgb_watchdog - Timer Call-back
1111  * @data: pointer to netdev cast into an unsigned long
1112  **/
1113
1114 static void
1115 ixgb_watchdog(unsigned long data)
1116 {
1117         struct ixgb_adapter *adapter = (struct ixgb_adapter *)data;
1118         struct net_device *netdev = adapter->netdev;
1119         struct ixgb_desc_ring *txdr = &adapter->tx_ring;
1120
1121         ixgb_check_for_link(&adapter->hw);
1122
1123         if (ixgb_check_for_bad_link(&adapter->hw)) {
1124                 /* force the reset path */
1125                 netif_stop_queue(netdev);
1126         }
1127
1128         if(adapter->hw.link_up) {
1129                 if(!netif_carrier_ok(netdev)) {
1130                         DPRINTK(LINK, INFO,
1131                                 "NIC Link is Up 10000 Mbps Full Duplex\n");
1132                         adapter->link_speed = 10000;
1133                         adapter->link_duplex = FULL_DUPLEX;
1134                         netif_carrier_on(netdev);
1135                         netif_wake_queue(netdev);
1136                 }
1137         } else {
1138                 if(netif_carrier_ok(netdev)) {
1139                         adapter->link_speed = 0;
1140                         adapter->link_duplex = 0;
1141                         DPRINTK(LINK, INFO, "NIC Link is Down\n");
1142                         netif_carrier_off(netdev);
1143                         netif_stop_queue(netdev);
1144
1145                 }
1146         }
1147
1148         ixgb_update_stats(adapter);
1149
1150         if(!netif_carrier_ok(netdev)) {
1151                 if(IXGB_DESC_UNUSED(txdr) + 1 < txdr->count) {
1152                         /* We've lost link, so the controller stops DMA,
1153                          * but we've got queued Tx work that's never going
1154                          * to get done, so reset controller to flush Tx.
1155                          * (Do the reset outside of interrupt context). */
1156                         schedule_work(&adapter->tx_timeout_task);
1157                 }
1158         }
1159
1160         /* Force detection of hung controller every watchdog period */
1161         adapter->detect_tx_hung = TRUE;
1162
1163         /* generate an interrupt to force clean up of any stragglers */
1164         IXGB_WRITE_REG(&adapter->hw, ICS, IXGB_INT_TXDW);
1165
1166         /* Reset the timer */
1167         mod_timer(&adapter->watchdog_timer, jiffies + 2 * HZ);
1168 }
1169
1170 #define IXGB_TX_FLAGS_CSUM              0x00000001
1171 #define IXGB_TX_FLAGS_VLAN              0x00000002
1172 #define IXGB_TX_FLAGS_TSO               0x00000004
1173
1174 static int
1175 ixgb_tso(struct ixgb_adapter *adapter, struct sk_buff *skb)
1176 {
1177         struct ixgb_context_desc *context_desc;
1178         unsigned int i;
1179         uint8_t ipcss, ipcso, tucss, tucso, hdr_len;
1180         uint16_t ipcse, tucse, mss;
1181         int err;
1182
1183         if (likely(skb_is_gso(skb))) {
1184                 struct ixgb_buffer *buffer_info;
1185                 struct iphdr *iph;
1186
1187                 if (skb_header_cloned(skb)) {
1188                         err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1189                         if (err)
1190                                 return err;
1191                 }
1192
1193                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
1194                 mss = skb_shinfo(skb)->gso_size;
1195                 iph = ip_hdr(skb);
1196                 iph->tot_len = 0;
1197                 iph->check = 0;
1198                 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
1199                                                          iph->daddr, 0,
1200                                                          IPPROTO_TCP, 0);
1201                 ipcss = skb_network_offset(skb);
1202                 ipcso = (void *)&(iph->check) - (void *)skb->data;
1203                 ipcse = skb_transport_offset(skb) - 1;
1204                 tucss = skb_transport_offset(skb);
1205                 tucso = (void *)&(tcp_hdr(skb)->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
1237         return 0;
1238 }
1239
1240 static boolean_t
1241 ixgb_tx_csum(struct ixgb_adapter *adapter, struct sk_buff *skb)
1242 {
1243         struct ixgb_context_desc *context_desc;
1244         unsigned int i;
1245         uint8_t css, cso;
1246
1247         if(likely(skb->ip_summed == CHECKSUM_PARTIAL)) {
1248                 struct ixgb_buffer *buffer_info;
1249                 css = skb_transport_offset(skb);
1250                 cso = css + skb->csum_offset;
1251
1252                 i = adapter->tx_ring.next_to_use;
1253                 context_desc = IXGB_CONTEXT_DESC(adapter->tx_ring, i);
1254                 buffer_info = &adapter->tx_ring.buffer_info[i];
1255                 WARN_ON(buffer_info->dma != 0);
1256
1257                 context_desc->tucss = css;
1258                 context_desc->tucso = cso;
1259                 context_desc->tucse = 0;
1260                 /* zero out any previously existing data in one instruction */
1261                 *(uint32_t *)&(context_desc->ipcss) = 0;
1262                 context_desc->status = 0;
1263                 context_desc->hdr_len = 0;
1264                 context_desc->mss = 0;
1265                 context_desc->cmd_type_len =
1266                         cpu_to_le32(IXGB_CONTEXT_DESC_TYPE
1267                                     | IXGB_TX_DESC_CMD_IDE);
1268
1269                 if(++i == adapter->tx_ring.count) i = 0;
1270                 adapter->tx_ring.next_to_use = i;
1271
1272                 return TRUE;
1273         }
1274
1275         return FALSE;
1276 }
1277
1278 #define IXGB_MAX_TXD_PWR        14
1279 #define IXGB_MAX_DATA_PER_TXD   (1<<IXGB_MAX_TXD_PWR)
1280
1281 static int
1282 ixgb_tx_map(struct ixgb_adapter *adapter, struct sk_buff *skb,
1283             unsigned int first)
1284 {
1285         struct ixgb_desc_ring *tx_ring = &adapter->tx_ring;
1286         struct ixgb_buffer *buffer_info;
1287         int len = skb->len;
1288         unsigned int offset = 0, size, count = 0, i;
1289         unsigned int mss = skb_shinfo(skb)->gso_size;
1290
1291         unsigned int nr_frags = skb_shinfo(skb)->nr_frags;
1292         unsigned int f;
1293
1294         len -= skb->data_len;
1295
1296         i = tx_ring->next_to_use;
1297
1298         while(len) {
1299                 buffer_info = &tx_ring->buffer_info[i];
1300                 size = min(len, IXGB_MAX_DATA_PER_TXD);
1301                 /* Workaround for premature desc write-backs
1302                  * in TSO mode.  Append 4-byte sentinel desc */
1303                 if (unlikely(mss && !nr_frags && size == len && size > 8))
1304                         size -= 4;
1305
1306                 buffer_info->length = size;
1307                 WARN_ON(buffer_info->dma != 0);
1308                 buffer_info->dma =
1309                         pci_map_single(adapter->pdev,
1310                                 skb->data + offset,
1311                                 size,
1312                                 PCI_DMA_TODEVICE);
1313                 buffer_info->time_stamp = jiffies;
1314                 buffer_info->next_to_watch = 0;
1315
1316                 len -= size;
1317                 offset += size;
1318                 count++;
1319                 if(++i == tx_ring->count) i = 0;
1320         }
1321
1322         for(f = 0; f < nr_frags; f++) {
1323                 struct skb_frag_struct *frag;
1324
1325                 frag = &skb_shinfo(skb)->frags[f];
1326                 len = frag->size;
1327                 offset = 0;
1328
1329                 while(len) {
1330                         buffer_info = &tx_ring->buffer_info[i];
1331                         size = min(len, IXGB_MAX_DATA_PER_TXD);
1332
1333                         /* Workaround for premature desc write-backs
1334                          * in TSO mode.  Append 4-byte sentinel desc */
1335                         if (unlikely(mss && !nr_frags && size == len
1336                                      && size > 8))
1337                                 size -= 4;
1338
1339                         buffer_info->length = size;
1340                         buffer_info->dma =
1341                                 pci_map_page(adapter->pdev,
1342                                         frag->page,
1343                                         frag->page_offset + offset,
1344                                         size,
1345                                         PCI_DMA_TODEVICE);
1346                         buffer_info->time_stamp = jiffies;
1347                         buffer_info->next_to_watch = 0;
1348
1349                         len -= size;
1350                         offset += size;
1351                         count++;
1352                         if(++i == tx_ring->count) i = 0;
1353                 }
1354         }
1355         i = (i == 0) ? tx_ring->count - 1 : i - 1;
1356         tx_ring->buffer_info[i].skb = skb;
1357         tx_ring->buffer_info[first].next_to_watch = i;
1358
1359         return count;
1360 }
1361
1362 static void
1363 ixgb_tx_queue(struct ixgb_adapter *adapter, int count, int vlan_id,int tx_flags)
1364 {
1365         struct ixgb_desc_ring *tx_ring = &adapter->tx_ring;
1366         struct ixgb_tx_desc *tx_desc = NULL;
1367         struct ixgb_buffer *buffer_info;
1368         uint32_t cmd_type_len = adapter->tx_cmd_type;
1369         uint8_t status = 0;
1370         uint8_t popts = 0;
1371         unsigned int i;
1372
1373         if(tx_flags & IXGB_TX_FLAGS_TSO) {
1374                 cmd_type_len |= IXGB_TX_DESC_CMD_TSE;
1375                 popts |= (IXGB_TX_DESC_POPTS_IXSM | IXGB_TX_DESC_POPTS_TXSM);
1376         }
1377
1378         if(tx_flags & IXGB_TX_FLAGS_CSUM)
1379                 popts |= IXGB_TX_DESC_POPTS_TXSM;
1380
1381         if(tx_flags & IXGB_TX_FLAGS_VLAN) {
1382                 cmd_type_len |= IXGB_TX_DESC_CMD_VLE;
1383         }
1384
1385         i = tx_ring->next_to_use;
1386
1387         while(count--) {
1388                 buffer_info = &tx_ring->buffer_info[i];
1389                 tx_desc = IXGB_TX_DESC(*tx_ring, i);
1390                 tx_desc->buff_addr = cpu_to_le64(buffer_info->dma);
1391                 tx_desc->cmd_type_len =
1392                         cpu_to_le32(cmd_type_len | buffer_info->length);
1393                 tx_desc->status = status;
1394                 tx_desc->popts = popts;
1395                 tx_desc->vlan = cpu_to_le16(vlan_id);
1396
1397                 if(++i == tx_ring->count) i = 0;
1398         }
1399
1400         tx_desc->cmd_type_len |= cpu_to_le32(IXGB_TX_DESC_CMD_EOP 
1401                                 | IXGB_TX_DESC_CMD_RS );
1402
1403         /* Force memory writes to complete before letting h/w
1404          * know there are new descriptors to fetch.  (Only
1405          * applicable for weak-ordered memory model archs,
1406          * such as IA-64). */
1407         wmb();
1408
1409         tx_ring->next_to_use = i;
1410         IXGB_WRITE_REG(&adapter->hw, TDT, i);
1411 }
1412
1413 static int __ixgb_maybe_stop_tx(struct net_device *netdev, int size)
1414 {
1415         struct ixgb_adapter *adapter = netdev_priv(netdev);
1416         struct ixgb_desc_ring *tx_ring = &adapter->tx_ring;
1417
1418         netif_stop_queue(netdev);
1419         /* Herbert's original patch had:
1420          *  smp_mb__after_netif_stop_queue();
1421          * but since that doesn't exist yet, just open code it. */
1422         smp_mb();
1423
1424         /* We need to check again in a case another CPU has just
1425          * made room available. */
1426         if (likely(IXGB_DESC_UNUSED(tx_ring) < size))
1427                 return -EBUSY;
1428
1429         /* A reprieve! */
1430         netif_start_queue(netdev);
1431         ++adapter->restart_queue;
1432         return 0;
1433 }
1434
1435 static int ixgb_maybe_stop_tx(struct net_device *netdev,
1436                               struct ixgb_desc_ring *tx_ring, int size)
1437 {
1438         if (likely(IXGB_DESC_UNUSED(tx_ring) >= size))
1439                 return 0;
1440         return __ixgb_maybe_stop_tx(netdev, size);
1441 }
1442
1443
1444 /* Tx Descriptors needed, worst case */
1445 #define TXD_USE_COUNT(S) (((S) >> IXGB_MAX_TXD_PWR) + \
1446                          (((S) & (IXGB_MAX_DATA_PER_TXD - 1)) ? 1 : 0))
1447 #define DESC_NEEDED TXD_USE_COUNT(IXGB_MAX_DATA_PER_TXD) /* skb->date */ + \
1448         MAX_SKB_FRAGS * TXD_USE_COUNT(PAGE_SIZE) + 1 /* for context */ \
1449         + 1 /* one more needed for sentinel TSO workaround */
1450
1451 static int
1452 ixgb_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
1453 {
1454         struct ixgb_adapter *adapter = netdev_priv(netdev);
1455         unsigned int first;
1456         unsigned int tx_flags = 0;
1457         unsigned long flags;
1458         int vlan_id = 0;
1459         int tso;
1460
1461         if(skb->len <= 0) {
1462                 dev_kfree_skb_any(skb);
1463                 return 0;
1464         }
1465
1466 #ifdef NETIF_F_LLTX
1467         local_irq_save(flags);
1468         if (!spin_trylock(&adapter->tx_lock)) {
1469                 /* Collision - tell upper layer to requeue */
1470                 local_irq_restore(flags);
1471                 return NETDEV_TX_LOCKED;
1472         }
1473 #else
1474         spin_lock_irqsave(&adapter->tx_lock, flags);
1475 #endif
1476
1477         if (unlikely(ixgb_maybe_stop_tx(netdev, &adapter->tx_ring,
1478                      DESC_NEEDED))) {
1479                 netif_stop_queue(netdev);
1480                 spin_unlock_irqrestore(&adapter->tx_lock, flags);
1481                 return NETDEV_TX_BUSY;
1482         }
1483
1484 #ifndef NETIF_F_LLTX
1485         spin_unlock_irqrestore(&adapter->tx_lock, flags);
1486 #endif
1487
1488         if(adapter->vlgrp && vlan_tx_tag_present(skb)) {
1489                 tx_flags |= IXGB_TX_FLAGS_VLAN;
1490                 vlan_id = vlan_tx_tag_get(skb);
1491         }
1492
1493         first = adapter->tx_ring.next_to_use;
1494         
1495         tso = ixgb_tso(adapter, skb);
1496         if (tso < 0) {
1497                 dev_kfree_skb_any(skb);
1498 #ifdef NETIF_F_LLTX
1499                 spin_unlock_irqrestore(&adapter->tx_lock, flags);
1500 #endif
1501                 return NETDEV_TX_OK;
1502         }
1503
1504         if (likely(tso))
1505                 tx_flags |= IXGB_TX_FLAGS_TSO;
1506         else if(ixgb_tx_csum(adapter, skb))
1507                 tx_flags |= IXGB_TX_FLAGS_CSUM;
1508
1509         ixgb_tx_queue(adapter, ixgb_tx_map(adapter, skb, first), vlan_id,
1510                         tx_flags);
1511
1512         netdev->trans_start = jiffies;
1513
1514 #ifdef NETIF_F_LLTX
1515         /* Make sure there is space in the ring for the next send. */
1516         ixgb_maybe_stop_tx(netdev, &adapter->tx_ring, DESC_NEEDED);
1517
1518         spin_unlock_irqrestore(&adapter->tx_lock, flags);
1519
1520 #endif
1521         return NETDEV_TX_OK;
1522 }
1523
1524 /**
1525  * ixgb_tx_timeout - Respond to a Tx Hang
1526  * @netdev: network interface device structure
1527  **/
1528
1529 static void
1530 ixgb_tx_timeout(struct net_device *netdev)
1531 {
1532         struct ixgb_adapter *adapter = netdev_priv(netdev);
1533
1534         /* Do the reset outside of interrupt context */
1535         schedule_work(&adapter->tx_timeout_task);
1536 }
1537
1538 static void
1539 ixgb_tx_timeout_task(struct work_struct *work)
1540 {
1541         struct ixgb_adapter *adapter =
1542                 container_of(work, struct ixgb_adapter, tx_timeout_task);
1543
1544         adapter->tx_timeout_count++;
1545         ixgb_down(adapter, TRUE);
1546         ixgb_up(adapter);
1547 }
1548
1549 /**
1550  * ixgb_get_stats - Get System Network Statistics
1551  * @netdev: network interface device structure
1552  *
1553  * Returns the address of the device statistics structure.
1554  * The statistics are actually updated from the timer callback.
1555  **/
1556
1557 static struct net_device_stats *
1558 ixgb_get_stats(struct net_device *netdev)
1559 {
1560         struct ixgb_adapter *adapter = netdev_priv(netdev);
1561
1562         return &adapter->net_stats;
1563 }
1564
1565 /**
1566  * ixgb_change_mtu - Change the Maximum Transfer Unit
1567  * @netdev: network interface device structure
1568  * @new_mtu: new value for maximum frame size
1569  *
1570  * Returns 0 on success, negative on failure
1571  **/
1572
1573 static int
1574 ixgb_change_mtu(struct net_device *netdev, int new_mtu)
1575 {
1576         struct ixgb_adapter *adapter = netdev_priv(netdev);
1577         int max_frame = new_mtu + ENET_HEADER_SIZE + ENET_FCS_LENGTH;
1578         int old_max_frame = netdev->mtu + ENET_HEADER_SIZE + ENET_FCS_LENGTH;
1579
1580
1581         if((max_frame < IXGB_MIN_ENET_FRAME_SIZE_WITHOUT_FCS + ENET_FCS_LENGTH)
1582            || (max_frame > IXGB_MAX_JUMBO_FRAME_SIZE + ENET_FCS_LENGTH)) {
1583                 DPRINTK(PROBE, ERR, "Invalid MTU setting %d\n", new_mtu);
1584                 return -EINVAL;
1585         }
1586
1587         adapter->rx_buffer_len = max_frame;
1588
1589         netdev->mtu = new_mtu;
1590
1591         if ((old_max_frame != max_frame) && netif_running(netdev)) {
1592                 ixgb_down(adapter, TRUE);
1593                 ixgb_up(adapter);
1594         }
1595
1596         return 0;
1597 }
1598
1599 /**
1600  * ixgb_update_stats - Update the board statistics counters.
1601  * @adapter: board private structure
1602  **/
1603
1604 void
1605 ixgb_update_stats(struct ixgb_adapter *adapter)
1606 {
1607         struct net_device *netdev = adapter->netdev;
1608         struct pci_dev *pdev = adapter->pdev;
1609
1610         /* Prevent stats update while adapter is being reset */
1611         if (pci_channel_offline(pdev))
1612                 return;
1613
1614         if((netdev->flags & IFF_PROMISC) || (netdev->flags & IFF_ALLMULTI) ||
1615            (netdev->mc_count > IXGB_MAX_NUM_MULTICAST_ADDRESSES)) {
1616                 u64 multi = IXGB_READ_REG(&adapter->hw, MPRCL);
1617                 u32 bcast_l = IXGB_READ_REG(&adapter->hw, BPRCL);
1618                 u32 bcast_h = IXGB_READ_REG(&adapter->hw, BPRCH);
1619                 u64 bcast = ((u64)bcast_h << 32) | bcast_l; 
1620
1621                 multi |= ((u64)IXGB_READ_REG(&adapter->hw, MPRCH) << 32);
1622                 /* fix up multicast stats by removing broadcasts */
1623                 if(multi >= bcast)
1624                         multi -= bcast;
1625                 
1626                 adapter->stats.mprcl += (multi & 0xFFFFFFFF);
1627                 adapter->stats.mprch += (multi >> 32);
1628                 adapter->stats.bprcl += bcast_l; 
1629                 adapter->stats.bprch += bcast_h;
1630         } else {
1631                 adapter->stats.mprcl += IXGB_READ_REG(&adapter->hw, MPRCL);
1632                 adapter->stats.mprch += IXGB_READ_REG(&adapter->hw, MPRCH);
1633                 adapter->stats.bprcl += IXGB_READ_REG(&adapter->hw, BPRCL);
1634                 adapter->stats.bprch += IXGB_READ_REG(&adapter->hw, BPRCH);
1635         }
1636         adapter->stats.tprl += IXGB_READ_REG(&adapter->hw, TPRL);
1637         adapter->stats.tprh += IXGB_READ_REG(&adapter->hw, TPRH);
1638         adapter->stats.gprcl += IXGB_READ_REG(&adapter->hw, GPRCL);
1639         adapter->stats.gprch += IXGB_READ_REG(&adapter->hw, GPRCH);
1640         adapter->stats.uprcl += IXGB_READ_REG(&adapter->hw, UPRCL);
1641         adapter->stats.uprch += IXGB_READ_REG(&adapter->hw, UPRCH);
1642         adapter->stats.vprcl += IXGB_READ_REG(&adapter->hw, VPRCL);
1643         adapter->stats.vprch += IXGB_READ_REG(&adapter->hw, VPRCH);
1644         adapter->stats.jprcl += IXGB_READ_REG(&adapter->hw, JPRCL);
1645         adapter->stats.jprch += IXGB_READ_REG(&adapter->hw, JPRCH);
1646         adapter->stats.gorcl += IXGB_READ_REG(&adapter->hw, GORCL);
1647         adapter->stats.gorch += IXGB_READ_REG(&adapter->hw, GORCH);
1648         adapter->stats.torl += IXGB_READ_REG(&adapter->hw, TORL);
1649         adapter->stats.torh += IXGB_READ_REG(&adapter->hw, TORH);
1650         adapter->stats.rnbc += IXGB_READ_REG(&adapter->hw, RNBC);
1651         adapter->stats.ruc += IXGB_READ_REG(&adapter->hw, RUC);
1652         adapter->stats.roc += IXGB_READ_REG(&adapter->hw, ROC);
1653         adapter->stats.rlec += IXGB_READ_REG(&adapter->hw, RLEC);
1654         adapter->stats.crcerrs += IXGB_READ_REG(&adapter->hw, CRCERRS);
1655         adapter->stats.icbc += IXGB_READ_REG(&adapter->hw, ICBC);
1656         adapter->stats.ecbc += IXGB_READ_REG(&adapter->hw, ECBC);
1657         adapter->stats.mpc += IXGB_READ_REG(&adapter->hw, MPC);
1658         adapter->stats.tptl += IXGB_READ_REG(&adapter->hw, TPTL);
1659         adapter->stats.tpth += IXGB_READ_REG(&adapter->hw, TPTH);
1660         adapter->stats.gptcl += IXGB_READ_REG(&adapter->hw, GPTCL);
1661         adapter->stats.gptch += IXGB_READ_REG(&adapter->hw, GPTCH);
1662         adapter->stats.bptcl += IXGB_READ_REG(&adapter->hw, BPTCL);
1663         adapter->stats.bptch += IXGB_READ_REG(&adapter->hw, BPTCH);
1664         adapter->stats.mptcl += IXGB_READ_REG(&adapter->hw, MPTCL);
1665         adapter->stats.mptch += IXGB_READ_REG(&adapter->hw, MPTCH);
1666         adapter->stats.uptcl += IXGB_READ_REG(&adapter->hw, UPTCL);
1667         adapter->stats.uptch += IXGB_READ_REG(&adapter->hw, UPTCH);
1668         adapter->stats.vptcl += IXGB_READ_REG(&adapter->hw, VPTCL);
1669         adapter->stats.vptch += IXGB_READ_REG(&adapter->hw, VPTCH);
1670         adapter->stats.jptcl += IXGB_READ_REG(&adapter->hw, JPTCL);
1671         adapter->stats.jptch += IXGB_READ_REG(&adapter->hw, JPTCH);
1672         adapter->stats.gotcl += IXGB_READ_REG(&adapter->hw, GOTCL);
1673         adapter->stats.gotch += IXGB_READ_REG(&adapter->hw, GOTCH);
1674         adapter->stats.totl += IXGB_READ_REG(&adapter->hw, TOTL);
1675         adapter->stats.toth += IXGB_READ_REG(&adapter->hw, TOTH);
1676         adapter->stats.dc += IXGB_READ_REG(&adapter->hw, DC);
1677         adapter->stats.plt64c += IXGB_READ_REG(&adapter->hw, PLT64C);
1678         adapter->stats.tsctc += IXGB_READ_REG(&adapter->hw, TSCTC);
1679         adapter->stats.tsctfc += IXGB_READ_REG(&adapter->hw, TSCTFC);
1680         adapter->stats.ibic += IXGB_READ_REG(&adapter->hw, IBIC);
1681         adapter->stats.rfc += IXGB_READ_REG(&adapter->hw, RFC);
1682         adapter->stats.lfc += IXGB_READ_REG(&adapter->hw, LFC);
1683         adapter->stats.pfrc += IXGB_READ_REG(&adapter->hw, PFRC);
1684         adapter->stats.pftc += IXGB_READ_REG(&adapter->hw, PFTC);
1685         adapter->stats.mcfrc += IXGB_READ_REG(&adapter->hw, MCFRC);
1686         adapter->stats.mcftc += IXGB_READ_REG(&adapter->hw, MCFTC);
1687         adapter->stats.xonrxc += IXGB_READ_REG(&adapter->hw, XONRXC);
1688         adapter->stats.xontxc += IXGB_READ_REG(&adapter->hw, XONTXC);
1689         adapter->stats.xoffrxc += IXGB_READ_REG(&adapter->hw, XOFFRXC);
1690         adapter->stats.xofftxc += IXGB_READ_REG(&adapter->hw, XOFFTXC);
1691         adapter->stats.rjc += IXGB_READ_REG(&adapter->hw, RJC);
1692
1693         /* Fill out the OS statistics structure */
1694
1695         adapter->net_stats.rx_packets = adapter->stats.gprcl;
1696         adapter->net_stats.tx_packets = adapter->stats.gptcl;
1697         adapter->net_stats.rx_bytes = adapter->stats.gorcl;
1698         adapter->net_stats.tx_bytes = adapter->stats.gotcl;
1699         adapter->net_stats.multicast = adapter->stats.mprcl;
1700         adapter->net_stats.collisions = 0;
1701
1702         /* ignore RLEC as it reports errors for padded (<64bytes) frames
1703          * with a length in the type/len field */
1704         adapter->net_stats.rx_errors =
1705             /* adapter->stats.rnbc + */ adapter->stats.crcerrs +
1706             adapter->stats.ruc +
1707             adapter->stats.roc /*+ adapter->stats.rlec */  +
1708             adapter->stats.icbc +
1709             adapter->stats.ecbc + adapter->stats.mpc;
1710
1711         /* see above
1712          * adapter->net_stats.rx_length_errors = adapter->stats.rlec;
1713          */
1714
1715         adapter->net_stats.rx_crc_errors = adapter->stats.crcerrs;
1716         adapter->net_stats.rx_fifo_errors = adapter->stats.mpc;
1717         adapter->net_stats.rx_missed_errors = adapter->stats.mpc;
1718         adapter->net_stats.rx_over_errors = adapter->stats.mpc;
1719
1720         adapter->net_stats.tx_errors = 0;
1721         adapter->net_stats.rx_frame_errors = 0;
1722         adapter->net_stats.tx_aborted_errors = 0;
1723         adapter->net_stats.tx_carrier_errors = 0;
1724         adapter->net_stats.tx_fifo_errors = 0;
1725         adapter->net_stats.tx_heartbeat_errors = 0;
1726         adapter->net_stats.tx_window_errors = 0;
1727 }
1728
1729 #define IXGB_MAX_INTR 10
1730 /**
1731  * ixgb_intr - Interrupt Handler
1732  * @irq: interrupt number
1733  * @data: pointer to a network interface device structure
1734  **/
1735
1736 static irqreturn_t
1737 ixgb_intr(int irq, void *data)
1738 {
1739         struct net_device *netdev = data;
1740         struct ixgb_adapter *adapter = netdev_priv(netdev);
1741         struct ixgb_hw *hw = &adapter->hw;
1742         uint32_t icr = IXGB_READ_REG(hw, ICR);
1743 #ifndef CONFIG_IXGB_NAPI
1744         unsigned int i;
1745 #endif
1746
1747         if(unlikely(!icr))
1748                 return IRQ_NONE;  /* Not our interrupt */
1749
1750         if(unlikely(icr & (IXGB_INT_RXSEQ | IXGB_INT_LSC))) {
1751                 mod_timer(&adapter->watchdog_timer, jiffies);
1752         }
1753
1754 #ifdef CONFIG_IXGB_NAPI
1755         if(netif_rx_schedule_prep(netdev)) {
1756
1757                 /* Disable interrupts and register for poll. The flush 
1758                   of the posted write is intentionally left out.
1759                 */
1760
1761                 atomic_inc(&adapter->irq_sem);
1762                 IXGB_WRITE_REG(&adapter->hw, IMC, ~0);
1763                 __netif_rx_schedule(netdev);
1764         }
1765 #else
1766         /* yes, that is actually a & and it is meant to make sure that
1767          * every pass through this for loop checks both receive and
1768          * transmit queues for completed descriptors, intended to
1769          * avoid starvation issues and assist tx/rx fairness. */
1770         for(i = 0; i < IXGB_MAX_INTR; i++)
1771                 if(!ixgb_clean_rx_irq(adapter) &
1772                    !ixgb_clean_tx_irq(adapter))
1773                         break;
1774 #endif 
1775         return IRQ_HANDLED;
1776 }
1777
1778 #ifdef CONFIG_IXGB_NAPI
1779 /**
1780  * ixgb_clean - NAPI Rx polling callback
1781  * @adapter: board private structure
1782  **/
1783
1784 static int
1785 ixgb_clean(struct net_device *netdev, int *budget)
1786 {
1787         struct ixgb_adapter *adapter = netdev_priv(netdev);
1788         int work_to_do = min(*budget, netdev->quota);
1789         int tx_cleaned;
1790         int work_done = 0;
1791
1792         tx_cleaned = ixgb_clean_tx_irq(adapter);
1793         ixgb_clean_rx_irq(adapter, &work_done, work_to_do);
1794
1795         *budget -= work_done;
1796         netdev->quota -= work_done;
1797
1798         /* if no Tx and not enough Rx work done, exit the polling mode */
1799         if((!tx_cleaned && (work_done == 0)) || !netif_running(netdev)) {
1800                 netif_rx_complete(netdev);
1801                 ixgb_irq_enable(adapter);
1802                 return 0;
1803         }
1804
1805         return 1;
1806 }
1807 #endif
1808
1809 /**
1810  * ixgb_clean_tx_irq - Reclaim resources after transmit completes
1811  * @adapter: board private structure
1812  **/
1813
1814 static boolean_t
1815 ixgb_clean_tx_irq(struct ixgb_adapter *adapter)
1816 {
1817         struct ixgb_desc_ring *tx_ring = &adapter->tx_ring;
1818         struct net_device *netdev = adapter->netdev;
1819         struct ixgb_tx_desc *tx_desc, *eop_desc;
1820         struct ixgb_buffer *buffer_info;
1821         unsigned int i, eop;
1822         boolean_t cleaned = FALSE;
1823
1824         i = tx_ring->next_to_clean;
1825         eop = tx_ring->buffer_info[i].next_to_watch;
1826         eop_desc = IXGB_TX_DESC(*tx_ring, eop);
1827
1828         while(eop_desc->status & IXGB_TX_DESC_STATUS_DD) {
1829
1830                 for(cleaned = FALSE; !cleaned; ) {
1831                         tx_desc = IXGB_TX_DESC(*tx_ring, i);
1832                         buffer_info = &tx_ring->buffer_info[i];
1833
1834                         if (tx_desc->popts
1835                             & (IXGB_TX_DESC_POPTS_TXSM |
1836                                IXGB_TX_DESC_POPTS_IXSM))
1837                                 adapter->hw_csum_tx_good++;
1838
1839                         ixgb_unmap_and_free_tx_resource(adapter, buffer_info);
1840
1841                         *(uint32_t *)&(tx_desc->status) = 0;
1842
1843                         cleaned = (i == eop);
1844                         if(++i == tx_ring->count) i = 0;
1845                 }
1846
1847                 eop = tx_ring->buffer_info[i].next_to_watch;
1848                 eop_desc = IXGB_TX_DESC(*tx_ring, eop);
1849         }
1850
1851         tx_ring->next_to_clean = i;
1852
1853         if (unlikely(netif_queue_stopped(netdev))) {
1854                 spin_lock(&adapter->tx_lock);
1855                 if (netif_queue_stopped(netdev) && netif_carrier_ok(netdev) &&
1856                     (IXGB_DESC_UNUSED(tx_ring) >= DESC_NEEDED))
1857                         netif_wake_queue(netdev);
1858                 spin_unlock(&adapter->tx_lock);
1859         }
1860
1861         if(adapter->detect_tx_hung) {
1862                 /* detect a transmit hang in hardware, this serializes the
1863                  * check with the clearing of time_stamp and movement of i */
1864                 adapter->detect_tx_hung = FALSE;
1865                 if (tx_ring->buffer_info[eop].dma &&
1866                    time_after(jiffies, tx_ring->buffer_info[eop].time_stamp + HZ)
1867                    && !(IXGB_READ_REG(&adapter->hw, STATUS) &
1868                         IXGB_STATUS_TXOFF)) {
1869                         /* detected Tx unit hang */
1870                         DPRINTK(DRV, ERR, "Detected Tx Unit Hang\n"
1871                                         "  TDH                  <%x>\n"
1872                                         "  TDT                  <%x>\n"
1873                                         "  next_to_use          <%x>\n"
1874                                         "  next_to_clean        <%x>\n"
1875                                         "buffer_info[next_to_clean]\n"
1876                                         "  time_stamp           <%lx>\n"
1877                                         "  next_to_watch        <%x>\n"
1878                                         "  jiffies              <%lx>\n"
1879                                         "  next_to_watch.status <%x>\n",
1880                                 IXGB_READ_REG(&adapter->hw, TDH),
1881                                 IXGB_READ_REG(&adapter->hw, TDT),
1882                                 tx_ring->next_to_use,
1883                                 tx_ring->next_to_clean,
1884                                 tx_ring->buffer_info[eop].time_stamp,
1885                                 eop,
1886                                 jiffies,
1887                                 eop_desc->status);
1888                         netif_stop_queue(netdev);
1889                 }
1890         }
1891
1892         return cleaned;
1893 }
1894
1895 /**
1896  * ixgb_rx_checksum - Receive Checksum Offload for 82597.
1897  * @adapter: board private structure
1898  * @rx_desc: receive descriptor
1899  * @sk_buff: socket buffer with received data
1900  **/
1901
1902 static void
1903 ixgb_rx_checksum(struct ixgb_adapter *adapter,
1904                  struct ixgb_rx_desc *rx_desc,
1905                  struct sk_buff *skb)
1906 {
1907         /* Ignore Checksum bit is set OR
1908          * TCP Checksum has not been calculated
1909          */
1910         if((rx_desc->status & IXGB_RX_DESC_STATUS_IXSM) ||
1911            (!(rx_desc->status & IXGB_RX_DESC_STATUS_TCPCS))) {
1912                 skb->ip_summed = CHECKSUM_NONE;
1913                 return;
1914         }
1915
1916         /* At this point we know the hardware did the TCP checksum */
1917         /* now look at the TCP checksum error bit */
1918         if(rx_desc->errors & IXGB_RX_DESC_ERRORS_TCPE) {
1919                 /* let the stack verify checksum errors */
1920                 skb->ip_summed = CHECKSUM_NONE;
1921                 adapter->hw_csum_rx_error++;
1922         } else {
1923                 /* TCP checksum is good */
1924                 skb->ip_summed = CHECKSUM_UNNECESSARY;
1925                 adapter->hw_csum_rx_good++;
1926         }
1927 }
1928
1929 /**
1930  * ixgb_clean_rx_irq - Send received data up the network stack,
1931  * @adapter: board private structure
1932  **/
1933
1934 static boolean_t
1935 #ifdef CONFIG_IXGB_NAPI
1936 ixgb_clean_rx_irq(struct ixgb_adapter *adapter, int *work_done, int work_to_do)
1937 #else
1938 ixgb_clean_rx_irq(struct ixgb_adapter *adapter)
1939 #endif
1940 {
1941         struct ixgb_desc_ring *rx_ring = &adapter->rx_ring;
1942         struct net_device *netdev = adapter->netdev;
1943         struct pci_dev *pdev = adapter->pdev;
1944         struct ixgb_rx_desc *rx_desc, *next_rxd;
1945         struct ixgb_buffer *buffer_info, *next_buffer, *next2_buffer;
1946         uint32_t length;
1947         unsigned int i, j;
1948         boolean_t cleaned = FALSE;
1949
1950         i = rx_ring->next_to_clean;
1951         rx_desc = IXGB_RX_DESC(*rx_ring, i);
1952         buffer_info = &rx_ring->buffer_info[i];
1953
1954         while(rx_desc->status & IXGB_RX_DESC_STATUS_DD) {
1955                 struct sk_buff *skb, *next_skb;
1956                 u8 status;
1957
1958 #ifdef CONFIG_IXGB_NAPI
1959                 if(*work_done >= work_to_do)
1960                         break;
1961
1962                 (*work_done)++;
1963 #endif
1964                 status = rx_desc->status;
1965                 skb = buffer_info->skb;
1966                 buffer_info->skb = NULL;
1967
1968                 prefetch(skb->data);
1969
1970                 if(++i == rx_ring->count) i = 0;
1971                 next_rxd = IXGB_RX_DESC(*rx_ring, i);
1972                 prefetch(next_rxd);
1973
1974                 if((j = i + 1) == rx_ring->count) j = 0;
1975                 next2_buffer = &rx_ring->buffer_info[j];
1976                 prefetch(next2_buffer);
1977
1978                 next_buffer = &rx_ring->buffer_info[i];
1979                 next_skb = next_buffer->skb;
1980                 prefetch(next_skb);
1981
1982                 cleaned = TRUE;
1983
1984                 pci_unmap_single(pdev,
1985                                  buffer_info->dma,
1986                                  buffer_info->length,
1987                                  PCI_DMA_FROMDEVICE);
1988
1989                 length = le16_to_cpu(rx_desc->length);
1990
1991                 if(unlikely(!(status & IXGB_RX_DESC_STATUS_EOP))) {
1992
1993                         /* All receives must fit into a single buffer */
1994
1995                         IXGB_DBG("Receive packet consumed multiple buffers "
1996                                          "length<%x>\n", length);
1997
1998                         dev_kfree_skb_irq(skb);
1999                         goto rxdesc_done;
2000                 }
2001
2002                 if (unlikely(rx_desc->errors
2003                              & (IXGB_RX_DESC_ERRORS_CE | IXGB_RX_DESC_ERRORS_SE
2004                                 | IXGB_RX_DESC_ERRORS_P |
2005                                 IXGB_RX_DESC_ERRORS_RXE))) {
2006
2007                         dev_kfree_skb_irq(skb);
2008                         goto rxdesc_done;
2009                 }
2010
2011                 /* code added for copybreak, this should improve
2012                  * performance for small packets with large amounts
2013                  * of reassembly being done in the stack */
2014 #define IXGB_CB_LENGTH 256
2015                 if (length < IXGB_CB_LENGTH) {
2016                         struct sk_buff *new_skb =
2017                             netdev_alloc_skb(netdev, length + NET_IP_ALIGN);
2018                         if (new_skb) {
2019                                 skb_reserve(new_skb, NET_IP_ALIGN);
2020                                 skb_copy_to_linear_data_offset(new_skb,
2021                                                                -NET_IP_ALIGN,
2022                                                                (skb->data -
2023                                                                 NET_IP_ALIGN),
2024                                                                (length +
2025                                                                 NET_IP_ALIGN));
2026                                 /* save the skb in buffer_info as good */
2027                                 buffer_info->skb = skb;
2028                                 skb = new_skb;
2029                         }
2030                 }
2031                 /* end copybreak code */
2032
2033                 /* Good Receive */
2034                 skb_put(skb, length);
2035
2036                 /* Receive Checksum Offload */
2037                 ixgb_rx_checksum(adapter, rx_desc, skb);
2038
2039                 skb->protocol = eth_type_trans(skb, netdev);
2040 #ifdef CONFIG_IXGB_NAPI
2041                 if(adapter->vlgrp && (status & IXGB_RX_DESC_STATUS_VP)) {
2042                         vlan_hwaccel_receive_skb(skb, adapter->vlgrp,
2043                                 le16_to_cpu(rx_desc->special) &
2044                                         IXGB_RX_DESC_SPECIAL_VLAN_MASK);
2045                 } else {
2046                         netif_receive_skb(skb);
2047                 }
2048 #else /* CONFIG_IXGB_NAPI */
2049                 if(adapter->vlgrp && (status & IXGB_RX_DESC_STATUS_VP)) {
2050                         vlan_hwaccel_rx(skb, adapter->vlgrp,
2051                                 le16_to_cpu(rx_desc->special) &
2052                                         IXGB_RX_DESC_SPECIAL_VLAN_MASK);
2053                 } else {
2054                         netif_rx(skb);
2055                 }
2056 #endif /* CONFIG_IXGB_NAPI */
2057                 netdev->last_rx = jiffies;
2058
2059 rxdesc_done:
2060                 /* clean up descriptor, might be written over by hw */
2061                 rx_desc->status = 0;
2062
2063                 /* use prefetched values */
2064                 rx_desc = next_rxd;
2065                 buffer_info = next_buffer;
2066         }
2067
2068         rx_ring->next_to_clean = i;
2069
2070         ixgb_alloc_rx_buffers(adapter);
2071
2072         return cleaned;
2073 }
2074
2075 /**
2076  * ixgb_alloc_rx_buffers - Replace used receive buffers
2077  * @adapter: address of board private structure
2078  **/
2079
2080 static void
2081 ixgb_alloc_rx_buffers(struct ixgb_adapter *adapter)
2082 {
2083         struct ixgb_desc_ring *rx_ring = &adapter->rx_ring;
2084         struct net_device *netdev = adapter->netdev;
2085         struct pci_dev *pdev = adapter->pdev;
2086         struct ixgb_rx_desc *rx_desc;
2087         struct ixgb_buffer *buffer_info;
2088         struct sk_buff *skb;
2089         unsigned int i;
2090         int num_group_tail_writes;
2091         long cleancount;
2092
2093         i = rx_ring->next_to_use;
2094         buffer_info = &rx_ring->buffer_info[i];
2095         cleancount = IXGB_DESC_UNUSED(rx_ring);
2096
2097         num_group_tail_writes = IXGB_RX_BUFFER_WRITE;
2098
2099         /* leave three descriptors unused */
2100         while(--cleancount > 2) {
2101                 /* recycle! its good for you */
2102                 skb = buffer_info->skb;
2103                 if (skb) {
2104                         skb_trim(skb, 0);
2105                         goto map_skb;
2106                 }
2107
2108                 skb = netdev_alloc_skb(netdev, adapter->rx_buffer_len
2109                                        + NET_IP_ALIGN);
2110                 if (unlikely(!skb)) {
2111                         /* Better luck next round */
2112                         adapter->alloc_rx_buff_failed++;
2113                         break;
2114                 }
2115
2116                 /* Make buffer alignment 2 beyond a 16 byte boundary
2117                  * this will result in a 16 byte aligned IP header after
2118                  * the 14 byte MAC header is removed
2119                  */
2120                 skb_reserve(skb, NET_IP_ALIGN);
2121
2122                 buffer_info->skb = skb;
2123                 buffer_info->length = adapter->rx_buffer_len;
2124 map_skb:
2125                 buffer_info->dma = pci_map_single(pdev,
2126                                                   skb->data,
2127                                                   adapter->rx_buffer_len,
2128                                                   PCI_DMA_FROMDEVICE);
2129
2130                 rx_desc = IXGB_RX_DESC(*rx_ring, i);
2131                 rx_desc->buff_addr = cpu_to_le64(buffer_info->dma);
2132                 /* guarantee DD bit not set now before h/w gets descriptor
2133                  * this is the rest of the workaround for h/w double 
2134                  * writeback. */
2135                 rx_desc->status = 0;
2136
2137
2138                 if(++i == rx_ring->count) i = 0;
2139                 buffer_info = &rx_ring->buffer_info[i];
2140         }
2141
2142         if (likely(rx_ring->next_to_use != i)) {
2143                 rx_ring->next_to_use = i;
2144                 if (unlikely(i-- == 0))
2145                         i = (rx_ring->count - 1);
2146
2147                 /* Force memory writes to complete before letting h/w
2148                  * know there are new descriptors to fetch.  (Only
2149                  * applicable for weak-ordered memory model archs, such
2150                  * as IA-64). */
2151                 wmb();
2152                 IXGB_WRITE_REG(&adapter->hw, RDT, i);
2153         }
2154 }
2155
2156 /**
2157  * ixgb_vlan_rx_register - enables or disables vlan tagging/stripping.
2158  * 
2159  * @param netdev network interface device structure
2160  * @param grp indicates to enable or disable tagging/stripping
2161  **/
2162 static void
2163 ixgb_vlan_rx_register(struct net_device *netdev, struct vlan_group *grp)
2164 {
2165         struct ixgb_adapter *adapter = netdev_priv(netdev);
2166         uint32_t ctrl, rctl;
2167
2168         ixgb_irq_disable(adapter);
2169         adapter->vlgrp = grp;
2170
2171         if(grp) {
2172                 /* enable VLAN tag insert/strip */
2173                 ctrl = IXGB_READ_REG(&adapter->hw, CTRL0);
2174                 ctrl |= IXGB_CTRL0_VME;
2175                 IXGB_WRITE_REG(&adapter->hw, CTRL0, ctrl);
2176
2177                 /* enable VLAN receive filtering */
2178
2179                 rctl = IXGB_READ_REG(&adapter->hw, RCTL);
2180                 rctl |= IXGB_RCTL_VFE;
2181                 rctl &= ~IXGB_RCTL_CFIEN;
2182                 IXGB_WRITE_REG(&adapter->hw, RCTL, rctl);
2183         } else {
2184                 /* disable VLAN tag insert/strip */
2185
2186                 ctrl = IXGB_READ_REG(&adapter->hw, CTRL0);
2187                 ctrl &= ~IXGB_CTRL0_VME;
2188                 IXGB_WRITE_REG(&adapter->hw, CTRL0, ctrl);
2189
2190                 /* disable VLAN filtering */
2191
2192                 rctl = IXGB_READ_REG(&adapter->hw, RCTL);
2193                 rctl &= ~IXGB_RCTL_VFE;
2194                 IXGB_WRITE_REG(&adapter->hw, RCTL, rctl);
2195         }
2196
2197         ixgb_irq_enable(adapter);
2198 }
2199
2200 static void
2201 ixgb_vlan_rx_add_vid(struct net_device *netdev, uint16_t vid)
2202 {
2203         struct ixgb_adapter *adapter = netdev_priv(netdev);
2204         uint32_t vfta, index;
2205
2206         /* add VID to filter table */
2207
2208         index = (vid >> 5) & 0x7F;
2209         vfta = IXGB_READ_REG_ARRAY(&adapter->hw, VFTA, index);
2210         vfta |= (1 << (vid & 0x1F));
2211         ixgb_write_vfta(&adapter->hw, index, vfta);
2212 }
2213
2214 static void
2215 ixgb_vlan_rx_kill_vid(struct net_device *netdev, uint16_t vid)
2216 {
2217         struct ixgb_adapter *adapter = netdev_priv(netdev);
2218         uint32_t vfta, index;
2219
2220         ixgb_irq_disable(adapter);
2221
2222         vlan_group_set_device(adapter->vlgrp, vid, NULL);
2223
2224         ixgb_irq_enable(adapter);
2225
2226         /* remove VID from filter table*/
2227
2228         index = (vid >> 5) & 0x7F;
2229         vfta = IXGB_READ_REG_ARRAY(&adapter->hw, VFTA, index);
2230         vfta &= ~(1 << (vid & 0x1F));
2231         ixgb_write_vfta(&adapter->hw, index, vfta);
2232 }
2233
2234 static void
2235 ixgb_restore_vlan(struct ixgb_adapter *adapter)
2236 {
2237         ixgb_vlan_rx_register(adapter->netdev, adapter->vlgrp);
2238
2239         if(adapter->vlgrp) {
2240                 uint16_t vid;
2241                 for(vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
2242                         if(!vlan_group_get_device(adapter->vlgrp, vid))
2243                                 continue;
2244                         ixgb_vlan_rx_add_vid(adapter->netdev, vid);
2245                 }
2246         }
2247 }
2248
2249 #ifdef CONFIG_NET_POLL_CONTROLLER
2250 /*
2251  * Polling 'interrupt' - used by things like netconsole to send skbs
2252  * without having to re-enable interrupts. It's not called while
2253  * the interrupt routine is executing.
2254  */
2255
2256 static void ixgb_netpoll(struct net_device *dev)
2257 {
2258         struct ixgb_adapter *adapter = netdev_priv(dev);
2259
2260         disable_irq(adapter->pdev->irq);
2261         ixgb_intr(adapter->pdev->irq, dev);
2262         enable_irq(adapter->pdev->irq);
2263 }
2264 #endif
2265
2266 /**
2267  * ixgb_io_error_detected() - called when PCI error is detected
2268  * @pdev    pointer to pci device with error
2269  * @state   pci channel state after error
2270  *
2271  * This callback is called by the PCI subsystem whenever
2272  * a PCI bus error is detected.
2273  */
2274 static pci_ers_result_t ixgb_io_error_detected (struct pci_dev *pdev,
2275                                      enum pci_channel_state state)
2276 {
2277         struct net_device *netdev = pci_get_drvdata(pdev);
2278         struct ixgb_adapter *adapter = netdev_priv(netdev);
2279
2280         if(netif_running(netdev))
2281                 ixgb_down(adapter, TRUE);
2282
2283         pci_disable_device(pdev);
2284
2285         /* Request a slot reset. */
2286         return PCI_ERS_RESULT_NEED_RESET;
2287 }
2288
2289 /**
2290  * ixgb_io_slot_reset - called after the pci bus has been reset.
2291  * @pdev    pointer to pci device with error
2292  *
2293  * This callback is called after the PCI buss has been reset.
2294  * Basically, this tries to restart the card from scratch.
2295  * This is a shortened version of the device probe/discovery code,
2296  * it resembles the first-half of the ixgb_probe() routine.
2297  */
2298 static pci_ers_result_t ixgb_io_slot_reset (struct pci_dev *pdev)
2299 {
2300         struct net_device *netdev = pci_get_drvdata(pdev);
2301         struct ixgb_adapter *adapter = netdev_priv(netdev);
2302
2303         if(pci_enable_device(pdev)) {
2304                 DPRINTK(PROBE, ERR, "Cannot re-enable PCI device after reset.\n");
2305                 return PCI_ERS_RESULT_DISCONNECT;
2306         }
2307
2308         /* Perform card reset only on one instance of the card */
2309         if (0 != PCI_FUNC (pdev->devfn))
2310                 return PCI_ERS_RESULT_RECOVERED;
2311
2312         pci_set_master(pdev);
2313
2314         netif_carrier_off(netdev);
2315         netif_stop_queue(netdev);
2316         ixgb_reset(adapter);
2317
2318         /* Make sure the EEPROM is good */
2319         if(!ixgb_validate_eeprom_checksum(&adapter->hw)) {
2320                 DPRINTK(PROBE, ERR, "After reset, the EEPROM checksum is not valid.\n");
2321                 return PCI_ERS_RESULT_DISCONNECT;
2322         }
2323         ixgb_get_ee_mac_addr(&adapter->hw, netdev->dev_addr);
2324         memcpy(netdev->perm_addr, netdev->dev_addr, netdev->addr_len);
2325
2326         if(!is_valid_ether_addr(netdev->perm_addr)) {
2327                 DPRINTK(PROBE, ERR, "After reset, invalid MAC address.\n");
2328                 return PCI_ERS_RESULT_DISCONNECT;
2329         }
2330
2331         return PCI_ERS_RESULT_RECOVERED;
2332 }
2333
2334 /**
2335  * ixgb_io_resume - called when its OK to resume normal operations
2336  * @pdev    pointer to pci device with error
2337  *
2338  * The error recovery driver tells us that its OK to resume
2339  * normal operation. Implementation resembles the second-half
2340  * of the ixgb_probe() routine.
2341  */
2342 static void ixgb_io_resume (struct pci_dev *pdev)
2343 {
2344         struct net_device *netdev = pci_get_drvdata(pdev);
2345         struct ixgb_adapter *adapter = netdev_priv(netdev);
2346
2347         pci_set_master(pdev);
2348
2349         if(netif_running(netdev)) {
2350                 if(ixgb_up(adapter)) {
2351                         printk ("ixgb: can't bring device back up after reset\n");
2352                         return;
2353                 }
2354         }
2355
2356         netif_device_attach(netdev);
2357         mod_timer(&adapter->watchdog_timer, jiffies);
2358 }
2359
2360 /* ixgb_main.c */