Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/roland...
[linux-2.6] / drivers / net / ixgb / ixgb_main.c
1 /*******************************************************************************
2
3   Intel PRO/10GbE Linux driver
4   Copyright(c) 1999 - 2006 Intel Corporation.
5
6   This program is free software; you can redistribute it and/or modify it
7   under the terms and conditions of the GNU General Public License,
8   version 2, as published by the Free Software Foundation.
9
10   This program is distributed in the hope it will be useful, but WITHOUT
11   ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12   FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
13   more details.
14
15   You should have received a copy of the GNU General Public License along with
16   this program; if not, write to the Free Software Foundation, Inc.,
17   51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
18
19   The full GNU General Public License is included in this distribution in
20   the file called "COPYING".
21
22   Contact Information:
23   Linux NICS <linux.nics@intel.com>
24   e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
25   Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
26
27 *******************************************************************************/
28
29 #include "ixgb.h"
30
31 char ixgb_driver_name[] = "ixgb";
32 static char ixgb_driver_string[] = "Intel(R) PRO/10GbE Network Driver";
33
34 #ifndef CONFIG_IXGB_NAPI
35 #define DRIVERNAPI
36 #else
37 #define DRIVERNAPI "-NAPI"
38 #endif
39 #define DRV_VERSION             "1.0.126-k2"DRIVERNAPI
40 const char ixgb_driver_version[] = DRV_VERSION;
41 static const char ixgb_copyright[] = "Copyright (c) 1999-2006 Intel Corporation.";
42
43 /* ixgb_pci_tbl - PCI Device ID Table
44  *
45  * Wildcard entries (PCI_ANY_ID) should come last
46  * Last entry must be all 0s
47  *
48  * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
49  *   Class, Class Mask, private data (not used) }
50  */
51 static struct pci_device_id ixgb_pci_tbl[] = {
52         {INTEL_VENDOR_ID, IXGB_DEVICE_ID_82597EX,
53          PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0},
54         {INTEL_VENDOR_ID, IXGB_DEVICE_ID_82597EX_CX4,
55          PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0},
56         {INTEL_VENDOR_ID, IXGB_DEVICE_ID_82597EX_SR,
57          PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0},
58         {INTEL_VENDOR_ID, IXGB_DEVICE_ID_82597EX_LR,  
59          PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0},
60
61         /* required last entry */
62         {0,}
63 };
64
65 MODULE_DEVICE_TABLE(pci, ixgb_pci_tbl);
66
67 /* Local Function Prototypes */
68
69 int ixgb_up(struct ixgb_adapter *adapter);
70 void ixgb_down(struct ixgb_adapter *adapter, boolean_t kill_watchdog);
71 void ixgb_reset(struct ixgb_adapter *adapter);
72 int ixgb_setup_tx_resources(struct ixgb_adapter *adapter);
73 int ixgb_setup_rx_resources(struct ixgb_adapter *adapter);
74 void ixgb_free_tx_resources(struct ixgb_adapter *adapter);
75 void ixgb_free_rx_resources(struct ixgb_adapter *adapter);
76 void ixgb_update_stats(struct ixgb_adapter *adapter);
77
78 static int ixgb_init_module(void);
79 static void ixgb_exit_module(void);
80 static int ixgb_probe(struct pci_dev *pdev, const struct pci_device_id *ent);
81 static void __devexit ixgb_remove(struct pci_dev *pdev);
82 static int ixgb_sw_init(struct ixgb_adapter *adapter);
83 static int ixgb_open(struct net_device *netdev);
84 static int ixgb_close(struct net_device *netdev);
85 static void ixgb_configure_tx(struct ixgb_adapter *adapter);
86 static void ixgb_configure_rx(struct ixgb_adapter *adapter);
87 static void ixgb_setup_rctl(struct ixgb_adapter *adapter);
88 static void ixgb_clean_tx_ring(struct ixgb_adapter *adapter);
89 static void ixgb_clean_rx_ring(struct ixgb_adapter *adapter);
90 static void ixgb_set_multi(struct net_device *netdev);
91 static void ixgb_watchdog(unsigned long data);
92 static int ixgb_xmit_frame(struct sk_buff *skb, struct net_device *netdev);
93 static struct net_device_stats *ixgb_get_stats(struct net_device *netdev);
94 static int ixgb_change_mtu(struct net_device *netdev, int new_mtu);
95 static int ixgb_set_mac(struct net_device *netdev, void *p);
96 static irqreturn_t ixgb_intr(int irq, void *data);
97 static boolean_t ixgb_clean_tx_irq(struct ixgb_adapter *adapter);
98
99 #ifdef CONFIG_IXGB_NAPI
100 static int ixgb_clean(struct napi_struct *napi, int budget);
101 static boolean_t ixgb_clean_rx_irq(struct ixgb_adapter *adapter,
102                                    int *work_done, int work_to_do);
103 #else
104 static boolean_t ixgb_clean_rx_irq(struct ixgb_adapter *adapter);
105 #endif
106 static void ixgb_alloc_rx_buffers(struct ixgb_adapter *adapter);
107 static void ixgb_tx_timeout(struct net_device *dev);
108 static void ixgb_tx_timeout_task(struct work_struct *work);
109 static void ixgb_vlan_rx_register(struct net_device *netdev,
110                                   struct vlan_group *grp);
111 static void ixgb_vlan_rx_add_vid(struct net_device *netdev, uint16_t vid);
112 static void ixgb_vlan_rx_kill_vid(struct net_device *netdev, uint16_t vid);
113 static void ixgb_restore_vlan(struct ixgb_adapter *adapter);
114
115 #ifdef CONFIG_NET_POLL_CONTROLLER
116 /* for netdump / net console */
117 static void ixgb_netpoll(struct net_device *dev);
118 #endif
119
120 static pci_ers_result_t ixgb_io_error_detected (struct pci_dev *pdev,
121                              enum pci_channel_state state);
122 static pci_ers_result_t ixgb_io_slot_reset (struct pci_dev *pdev);
123 static void ixgb_io_resume (struct pci_dev *pdev);
124
125 static struct pci_error_handlers ixgb_err_handler = {
126         .error_detected = ixgb_io_error_detected,
127         .slot_reset = ixgb_io_slot_reset,
128         .resume = ixgb_io_resume,
129 };
130
131 static struct pci_driver ixgb_driver = {
132         .name     = ixgb_driver_name,
133         .id_table = ixgb_pci_tbl,
134         .probe    = ixgb_probe,
135         .remove   = __devexit_p(ixgb_remove),
136         .err_handler = &ixgb_err_handler
137 };
138
139 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
140 MODULE_DESCRIPTION("Intel(R) PRO/10GbE Network Driver");
141 MODULE_LICENSE("GPL");
142 MODULE_VERSION(DRV_VERSION);
143
144 #define DEFAULT_DEBUG_LEVEL_SHIFT 3
145 static int debug = DEFAULT_DEBUG_LEVEL_SHIFT;
146 module_param(debug, int, 0);
147 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
148
149 /* some defines for controlling descriptor fetches in h/w */
150 #define RXDCTL_WTHRESH_DEFAULT 15  /* chip writes back at this many or RXT0 */
151 #define RXDCTL_PTHRESH_DEFAULT 0   /* chip considers prefech below
152                                     * this */
153 #define RXDCTL_HTHRESH_DEFAULT 0   /* chip will only prefetch if tail
154                                     * is pushed this many descriptors
155                                     * from head */
156
157 /**
158  * ixgb_init_module - Driver Registration Routine
159  *
160  * ixgb_init_module is the first routine called when the driver is
161  * loaded. All it does is register with the PCI subsystem.
162  **/
163
164 static int __init
165 ixgb_init_module(void)
166 {
167         printk(KERN_INFO "%s - version %s\n",
168                ixgb_driver_string, ixgb_driver_version);
169
170         printk(KERN_INFO "%s\n", ixgb_copyright);
171
172         return pci_register_driver(&ixgb_driver);
173 }
174
175 module_init(ixgb_init_module);
176
177 /**
178  * ixgb_exit_module - Driver Exit Cleanup Routine
179  *
180  * ixgb_exit_module is called just before the driver is removed
181  * from memory.
182  **/
183
184 static void __exit
185 ixgb_exit_module(void)
186 {
187         pci_unregister_driver(&ixgb_driver);
188 }
189
190 module_exit(ixgb_exit_module);
191
192 /**
193  * ixgb_irq_disable - Mask off interrupt generation on the NIC
194  * @adapter: board private structure
195  **/
196
197 static void
198 ixgb_irq_disable(struct ixgb_adapter *adapter)
199 {
200         atomic_inc(&adapter->irq_sem);
201         IXGB_WRITE_REG(&adapter->hw, IMC, ~0);
202         IXGB_WRITE_FLUSH(&adapter->hw);
203         synchronize_irq(adapter->pdev->irq);
204 }
205
206 /**
207  * ixgb_irq_enable - Enable default interrupt generation settings
208  * @adapter: board private structure
209  **/
210
211 static void
212 ixgb_irq_enable(struct ixgb_adapter *adapter)
213 {
214         if(atomic_dec_and_test(&adapter->irq_sem)) {
215                 IXGB_WRITE_REG(&adapter->hw, IMS,
216                                IXGB_INT_RXT0 | IXGB_INT_RXDMT0 | IXGB_INT_TXDW |
217                                IXGB_INT_LSC);
218                 IXGB_WRITE_FLUSH(&adapter->hw);
219         }
220 }
221
222 int
223 ixgb_up(struct ixgb_adapter *adapter)
224 {
225         struct net_device *netdev = adapter->netdev;
226         int err, irq_flags = IRQF_SHARED;
227         int max_frame = netdev->mtu + ENET_HEADER_SIZE + ENET_FCS_LENGTH;
228         struct ixgb_hw *hw = &adapter->hw;
229
230         /* hardware has been reset, we need to reload some things */
231
232         ixgb_rar_set(hw, netdev->dev_addr, 0);
233         ixgb_set_multi(netdev);
234
235         ixgb_restore_vlan(adapter);
236
237         ixgb_configure_tx(adapter);
238         ixgb_setup_rctl(adapter);
239         ixgb_configure_rx(adapter);
240         ixgb_alloc_rx_buffers(adapter);
241
242         /* disable interrupts and get the hardware into a known state */
243         IXGB_WRITE_REG(&adapter->hw, IMC, 0xffffffff);
244
245         /* only enable MSI if bus is in PCI-X mode */
246         if (IXGB_READ_REG(&adapter->hw, STATUS) & IXGB_STATUS_PCIX_MODE) {
247                 err = pci_enable_msi(adapter->pdev);
248                 if (!err) {
249                         adapter->have_msi = 1;
250                         irq_flags = 0;
251                 }
252                 /* proceed to try to request regular interrupt */
253         }
254
255         err = request_irq(adapter->pdev->irq, &ixgb_intr, irq_flags,
256                           netdev->name, netdev);
257         if (err) {
258                 if (adapter->have_msi)
259                         pci_disable_msi(adapter->pdev);
260                 DPRINTK(PROBE, ERR,
261                  "Unable to allocate interrupt Error: %d\n", err);
262                 return err;
263         }
264
265         if((hw->max_frame_size != max_frame) ||
266                 (hw->max_frame_size !=
267                 (IXGB_READ_REG(hw, MFS) >> IXGB_MFS_SHIFT))) {
268
269                 hw->max_frame_size = max_frame;
270
271                 IXGB_WRITE_REG(hw, MFS, hw->max_frame_size << IXGB_MFS_SHIFT);
272
273                 if(hw->max_frame_size >
274                    IXGB_MAX_ENET_FRAME_SIZE_WITHOUT_FCS + ENET_FCS_LENGTH) {
275                         uint32_t ctrl0 = IXGB_READ_REG(hw, CTRL0);
276
277                         if(!(ctrl0 & IXGB_CTRL0_JFE)) {
278                                 ctrl0 |= IXGB_CTRL0_JFE;
279                                 IXGB_WRITE_REG(hw, CTRL0, ctrl0);
280                         }
281                 }
282         }
283
284         mod_timer(&adapter->watchdog_timer, jiffies);
285
286 #ifdef CONFIG_IXGB_NAPI
287         napi_enable(&adapter->napi);
288 #endif
289         ixgb_irq_enable(adapter);
290
291         return 0;
292 }
293
294 void
295 ixgb_down(struct ixgb_adapter *adapter, boolean_t kill_watchdog)
296 {
297         struct net_device *netdev = adapter->netdev;
298
299         ixgb_irq_disable(adapter);
300         free_irq(adapter->pdev->irq, netdev);
301
302         if (adapter->have_msi)
303                 pci_disable_msi(adapter->pdev);
304
305         if(kill_watchdog)
306                 del_timer_sync(&adapter->watchdog_timer);
307 #ifdef CONFIG_IXGB_NAPI
308         napi_disable(&adapter->napi);
309 #endif
310         adapter->link_speed = 0;
311         adapter->link_duplex = 0;
312         netif_carrier_off(netdev);
313         netif_stop_queue(netdev);
314
315         ixgb_reset(adapter);
316         ixgb_clean_tx_ring(adapter);
317         ixgb_clean_rx_ring(adapter);
318 }
319
320 void
321 ixgb_reset(struct ixgb_adapter *adapter)
322 {
323         struct ixgb_hw *hw = &adapter->hw;
324
325         ixgb_adapter_stop(hw);
326         if (!ixgb_init_hw(hw))
327                 DPRINTK(PROBE, ERR, "ixgb_init_hw failed.\n");
328
329         /* restore frame size information */
330         IXGB_WRITE_REG(hw, MFS, hw->max_frame_size << IXGB_MFS_SHIFT);
331         if (hw->max_frame_size >
332             IXGB_MAX_ENET_FRAME_SIZE_WITHOUT_FCS + ENET_FCS_LENGTH) {
333                 u32 ctrl0 = IXGB_READ_REG(hw, CTRL0);
334                 if (!(ctrl0 & IXGB_CTRL0_JFE)) {
335                         ctrl0 |= IXGB_CTRL0_JFE;
336                         IXGB_WRITE_REG(hw, CTRL0, ctrl0);
337                 }
338         }
339 }
340
341 /**
342  * ixgb_probe - Device Initialization Routine
343  * @pdev: PCI device information struct
344  * @ent: entry in ixgb_pci_tbl
345  *
346  * Returns 0 on success, negative on failure
347  *
348  * ixgb_probe initializes an adapter identified by a pci_dev structure.
349  * The OS initialization, configuring of the adapter private structure,
350  * and a hardware reset occur.
351  **/
352
353 static int __devinit
354 ixgb_probe(struct pci_dev *pdev,
355                 const struct pci_device_id *ent)
356 {
357         struct net_device *netdev = NULL;
358         struct ixgb_adapter *adapter;
359         static int cards_found = 0;
360         unsigned long mmio_start;
361         int mmio_len;
362         int pci_using_dac;
363         int i;
364         int err;
365
366         if((err = pci_enable_device(pdev)))
367                 return err;
368
369         if(!(err = pci_set_dma_mask(pdev, DMA_64BIT_MASK)) &&
370            !(err = pci_set_consistent_dma_mask(pdev, DMA_64BIT_MASK))) {
371                 pci_using_dac = 1;
372         } else {
373                 if((err = pci_set_dma_mask(pdev, DMA_32BIT_MASK)) ||
374                    (err = pci_set_consistent_dma_mask(pdev, DMA_32BIT_MASK))) {
375                         printk(KERN_ERR
376                          "ixgb: No usable DMA configuration, aborting\n");
377                         goto err_dma_mask;
378                 }
379                 pci_using_dac = 0;
380         }
381
382         if((err = pci_request_regions(pdev, ixgb_driver_name)))
383                 goto err_request_regions;
384
385         pci_set_master(pdev);
386
387         netdev = alloc_etherdev(sizeof(struct ixgb_adapter));
388         if(!netdev) {
389                 err = -ENOMEM;
390                 goto err_alloc_etherdev;
391         }
392
393         SET_NETDEV_DEV(netdev, &pdev->dev);
394
395         pci_set_drvdata(pdev, netdev);
396         adapter = netdev_priv(netdev);
397         adapter->netdev = netdev;
398         adapter->pdev = pdev;
399         adapter->hw.back = adapter;
400         adapter->msg_enable = netif_msg_init(debug, DEFAULT_DEBUG_LEVEL_SHIFT);
401
402         mmio_start = pci_resource_start(pdev, BAR_0);
403         mmio_len = pci_resource_len(pdev, BAR_0);
404
405         adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
406         if(!adapter->hw.hw_addr) {
407                 err = -EIO;
408                 goto err_ioremap;
409         }
410
411         for(i = BAR_1; i <= BAR_5; i++) {
412                 if(pci_resource_len(pdev, i) == 0)
413                         continue;
414                 if(pci_resource_flags(pdev, i) & IORESOURCE_IO) {
415                         adapter->hw.io_base = pci_resource_start(pdev, i);
416                         break;
417                 }
418         }
419
420         netdev->open = &ixgb_open;
421         netdev->stop = &ixgb_close;
422         netdev->hard_start_xmit = &ixgb_xmit_frame;
423         netdev->get_stats = &ixgb_get_stats;
424         netdev->set_multicast_list = &ixgb_set_multi;
425         netdev->set_mac_address = &ixgb_set_mac;
426         netdev->change_mtu = &ixgb_change_mtu;
427         ixgb_set_ethtool_ops(netdev);
428         netdev->tx_timeout = &ixgb_tx_timeout;
429         netdev->watchdog_timeo = 5 * HZ;
430 #ifdef CONFIG_IXGB_NAPI
431         netif_napi_add(netdev, &adapter->napi, ixgb_clean, 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         txdr->size = ALIGN(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         rxdr->size = ALIGN(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[IXGB_MAX_NUM_MULTICAST_ADDRESSES *
1097                             IXGB_ETH_LENGTH_OF_ADDRESS];
1098
1099                 IXGB_WRITE_REG(hw, RCTL, rctl);
1100
1101                 for(i = 0, mc_ptr = netdev->mc_list; mc_ptr;
1102                         i++, mc_ptr = mc_ptr->next)
1103                         memcpy(&mta[i * IXGB_ETH_LENGTH_OF_ADDRESS],
1104                                    mc_ptr->dmi_addr, IXGB_ETH_LENGTH_OF_ADDRESS);
1105
1106                 ixgb_mc_addr_list_update(hw, mta, netdev->mc_count, 0);
1107         }
1108 }
1109
1110 /**
1111  * ixgb_watchdog - Timer Call-back
1112  * @data: pointer to netdev cast into an unsigned long
1113  **/
1114
1115 static void
1116 ixgb_watchdog(unsigned long data)
1117 {
1118         struct ixgb_adapter *adapter = (struct ixgb_adapter *)data;
1119         struct net_device *netdev = adapter->netdev;
1120         struct ixgb_desc_ring *txdr = &adapter->tx_ring;
1121
1122         ixgb_check_for_link(&adapter->hw);
1123
1124         if (ixgb_check_for_bad_link(&adapter->hw)) {
1125                 /* force the reset path */
1126                 netif_stop_queue(netdev);
1127         }
1128
1129         if(adapter->hw.link_up) {
1130                 if(!netif_carrier_ok(netdev)) {
1131                         DPRINTK(LINK, INFO,
1132                                 "NIC Link is Up 10000 Mbps Full Duplex\n");
1133                         adapter->link_speed = 10000;
1134                         adapter->link_duplex = FULL_DUPLEX;
1135                         netif_carrier_on(netdev);
1136                         netif_wake_queue(netdev);
1137                 }
1138         } else {
1139                 if(netif_carrier_ok(netdev)) {
1140                         adapter->link_speed = 0;
1141                         adapter->link_duplex = 0;
1142                         DPRINTK(LINK, INFO, "NIC Link is Down\n");
1143                         netif_carrier_off(netdev);
1144                         netif_stop_queue(netdev);
1145
1146                 }
1147         }
1148
1149         ixgb_update_stats(adapter);
1150
1151         if(!netif_carrier_ok(netdev)) {
1152                 if(IXGB_DESC_UNUSED(txdr) + 1 < txdr->count) {
1153                         /* We've lost link, so the controller stops DMA,
1154                          * but we've got queued Tx work that's never going
1155                          * to get done, so reset controller to flush Tx.
1156                          * (Do the reset outside of interrupt context). */
1157                         schedule_work(&adapter->tx_timeout_task);
1158                 }
1159         }
1160
1161         /* Force detection of hung controller every watchdog period */
1162         adapter->detect_tx_hung = TRUE;
1163
1164         /* generate an interrupt to force clean up of any stragglers */
1165         IXGB_WRITE_REG(&adapter->hw, ICS, IXGB_INT_TXDW);
1166
1167         /* Reset the timer */
1168         mod_timer(&adapter->watchdog_timer, jiffies + 2 * HZ);
1169 }
1170
1171 #define IXGB_TX_FLAGS_CSUM              0x00000001
1172 #define IXGB_TX_FLAGS_VLAN              0x00000002
1173 #define IXGB_TX_FLAGS_TSO               0x00000004
1174
1175 static int
1176 ixgb_tso(struct ixgb_adapter *adapter, struct sk_buff *skb)
1177 {
1178         struct ixgb_context_desc *context_desc;
1179         unsigned int i;
1180         uint8_t ipcss, ipcso, tucss, tucso, hdr_len;
1181         uint16_t ipcse, tucse, mss;
1182         int err;
1183
1184         if (likely(skb_is_gso(skb))) {
1185                 struct ixgb_buffer *buffer_info;
1186                 struct iphdr *iph;
1187
1188                 if (skb_header_cloned(skb)) {
1189                         err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1190                         if (err)
1191                                 return err;
1192                 }
1193
1194                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
1195                 mss = skb_shinfo(skb)->gso_size;
1196                 iph = ip_hdr(skb);
1197                 iph->tot_len = 0;
1198                 iph->check = 0;
1199                 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
1200                                                          iph->daddr, 0,
1201                                                          IPPROTO_TCP, 0);
1202                 ipcss = skb_network_offset(skb);
1203                 ipcso = (void *)&(iph->check) - (void *)skb->data;
1204                 ipcse = skb_transport_offset(skb) - 1;
1205                 tucss = skb_transport_offset(skb);
1206                 tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
1207                 tucse = 0;
1208
1209                 i = adapter->tx_ring.next_to_use;
1210                 context_desc = IXGB_CONTEXT_DESC(adapter->tx_ring, i);
1211                 buffer_info = &adapter->tx_ring.buffer_info[i];
1212                 WARN_ON(buffer_info->dma != 0);
1213
1214                 context_desc->ipcss = ipcss;
1215                 context_desc->ipcso = ipcso;
1216                 context_desc->ipcse = cpu_to_le16(ipcse);
1217                 context_desc->tucss = tucss;
1218                 context_desc->tucso = tucso;
1219                 context_desc->tucse = cpu_to_le16(tucse);
1220                 context_desc->mss = cpu_to_le16(mss);
1221                 context_desc->hdr_len = hdr_len;
1222                 context_desc->status = 0;
1223                 context_desc->cmd_type_len = cpu_to_le32(
1224                                                   IXGB_CONTEXT_DESC_TYPE 
1225                                                 | IXGB_CONTEXT_DESC_CMD_TSE
1226                                                 | IXGB_CONTEXT_DESC_CMD_IP
1227                                                 | IXGB_CONTEXT_DESC_CMD_TCP
1228                                                 | IXGB_CONTEXT_DESC_CMD_IDE
1229                                                 | (skb->len - (hdr_len)));
1230
1231
1232                 if(++i == adapter->tx_ring.count) i = 0;
1233                 adapter->tx_ring.next_to_use = i;
1234
1235                 return 1;
1236         }
1237
1238         return 0;
1239 }
1240
1241 static boolean_t
1242 ixgb_tx_csum(struct ixgb_adapter *adapter, struct sk_buff *skb)
1243 {
1244         struct ixgb_context_desc *context_desc;
1245         unsigned int i;
1246         uint8_t css, cso;
1247
1248         if(likely(skb->ip_summed == CHECKSUM_PARTIAL)) {
1249                 struct ixgb_buffer *buffer_info;
1250                 css = skb_transport_offset(skb);
1251                 cso = css + skb->csum_offset;
1252
1253                 i = adapter->tx_ring.next_to_use;
1254                 context_desc = IXGB_CONTEXT_DESC(adapter->tx_ring, i);
1255                 buffer_info = &adapter->tx_ring.buffer_info[i];
1256                 WARN_ON(buffer_info->dma != 0);
1257
1258                 context_desc->tucss = css;
1259                 context_desc->tucso = cso;
1260                 context_desc->tucse = 0;
1261                 /* zero out any previously existing data in one instruction */
1262                 *(uint32_t *)&(context_desc->ipcss) = 0;
1263                 context_desc->status = 0;
1264                 context_desc->hdr_len = 0;
1265                 context_desc->mss = 0;
1266                 context_desc->cmd_type_len =
1267                         cpu_to_le32(IXGB_CONTEXT_DESC_TYPE
1268                                     | IXGB_TX_DESC_CMD_IDE);
1269
1270                 if(++i == adapter->tx_ring.count) i = 0;
1271                 adapter->tx_ring.next_to_use = i;
1272
1273                 return TRUE;
1274         }
1275
1276         return FALSE;
1277 }
1278
1279 #define IXGB_MAX_TXD_PWR        14
1280 #define IXGB_MAX_DATA_PER_TXD   (1<<IXGB_MAX_TXD_PWR)
1281
1282 static int
1283 ixgb_tx_map(struct ixgb_adapter *adapter, struct sk_buff *skb,
1284             unsigned int first)
1285 {
1286         struct ixgb_desc_ring *tx_ring = &adapter->tx_ring;
1287         struct ixgb_buffer *buffer_info;
1288         int len = skb->len;
1289         unsigned int offset = 0, size, count = 0, i;
1290         unsigned int mss = skb_shinfo(skb)->gso_size;
1291
1292         unsigned int nr_frags = skb_shinfo(skb)->nr_frags;
1293         unsigned int f;
1294
1295         len -= skb->data_len;
1296
1297         i = tx_ring->next_to_use;
1298
1299         while(len) {
1300                 buffer_info = &tx_ring->buffer_info[i];
1301                 size = min(len, IXGB_MAX_DATA_PER_TXD);
1302                 /* Workaround for premature desc write-backs
1303                  * in TSO mode.  Append 4-byte sentinel desc */
1304                 if (unlikely(mss && !nr_frags && size == len && size > 8))
1305                         size -= 4;
1306
1307                 buffer_info->length = size;
1308                 WARN_ON(buffer_info->dma != 0);
1309                 buffer_info->dma =
1310                         pci_map_single(adapter->pdev,
1311                                 skb->data + offset,
1312                                 size,
1313                                 PCI_DMA_TODEVICE);
1314                 buffer_info->time_stamp = jiffies;
1315                 buffer_info->next_to_watch = 0;
1316
1317                 len -= size;
1318                 offset += size;
1319                 count++;
1320                 if(++i == tx_ring->count) i = 0;
1321         }
1322
1323         for(f = 0; f < nr_frags; f++) {
1324                 struct skb_frag_struct *frag;
1325
1326                 frag = &skb_shinfo(skb)->frags[f];
1327                 len = frag->size;
1328                 offset = 0;
1329
1330                 while(len) {
1331                         buffer_info = &tx_ring->buffer_info[i];
1332                         size = min(len, IXGB_MAX_DATA_PER_TXD);
1333
1334                         /* Workaround for premature desc write-backs
1335                          * in TSO mode.  Append 4-byte sentinel desc */
1336                         if (unlikely(mss && (f == (nr_frags - 1))
1337                                      && size == len && size > 8))
1338                                 size -= 4;
1339
1340                         buffer_info->length = size;
1341                         buffer_info->dma =
1342                                 pci_map_page(adapter->pdev,
1343                                         frag->page,
1344                                         frag->page_offset + offset,
1345                                         size,
1346                                         PCI_DMA_TODEVICE);
1347                         buffer_info->time_stamp = jiffies;
1348                         buffer_info->next_to_watch = 0;
1349
1350                         len -= size;
1351                         offset += size;
1352                         count++;
1353                         if(++i == tx_ring->count) i = 0;
1354                 }
1355         }
1356         i = (i == 0) ? tx_ring->count - 1 : i - 1;
1357         tx_ring->buffer_info[i].skb = skb;
1358         tx_ring->buffer_info[first].next_to_watch = i;
1359
1360         return count;
1361 }
1362
1363 static void
1364 ixgb_tx_queue(struct ixgb_adapter *adapter, int count, int vlan_id,int tx_flags)
1365 {
1366         struct ixgb_desc_ring *tx_ring = &adapter->tx_ring;
1367         struct ixgb_tx_desc *tx_desc = NULL;
1368         struct ixgb_buffer *buffer_info;
1369         uint32_t cmd_type_len = adapter->tx_cmd_type;
1370         uint8_t status = 0;
1371         uint8_t popts = 0;
1372         unsigned int i;
1373
1374         if(tx_flags & IXGB_TX_FLAGS_TSO) {
1375                 cmd_type_len |= IXGB_TX_DESC_CMD_TSE;
1376                 popts |= (IXGB_TX_DESC_POPTS_IXSM | IXGB_TX_DESC_POPTS_TXSM);
1377         }
1378
1379         if(tx_flags & IXGB_TX_FLAGS_CSUM)
1380                 popts |= IXGB_TX_DESC_POPTS_TXSM;
1381
1382         if(tx_flags & IXGB_TX_FLAGS_VLAN) {
1383                 cmd_type_len |= IXGB_TX_DESC_CMD_VLE;
1384         }
1385
1386         i = tx_ring->next_to_use;
1387
1388         while(count--) {
1389                 buffer_info = &tx_ring->buffer_info[i];
1390                 tx_desc = IXGB_TX_DESC(*tx_ring, i);
1391                 tx_desc->buff_addr = cpu_to_le64(buffer_info->dma);
1392                 tx_desc->cmd_type_len =
1393                         cpu_to_le32(cmd_type_len | buffer_info->length);
1394                 tx_desc->status = status;
1395                 tx_desc->popts = popts;
1396                 tx_desc->vlan = cpu_to_le16(vlan_id);
1397
1398                 if(++i == tx_ring->count) i = 0;
1399         }
1400
1401         tx_desc->cmd_type_len |= cpu_to_le32(IXGB_TX_DESC_CMD_EOP 
1402                                 | IXGB_TX_DESC_CMD_RS );
1403
1404         /* Force memory writes to complete before letting h/w
1405          * know there are new descriptors to fetch.  (Only
1406          * applicable for weak-ordered memory model archs,
1407          * such as IA-64). */
1408         wmb();
1409
1410         tx_ring->next_to_use = i;
1411         IXGB_WRITE_REG(&adapter->hw, TDT, i);
1412 }
1413
1414 static int __ixgb_maybe_stop_tx(struct net_device *netdev, int size)
1415 {
1416         struct ixgb_adapter *adapter = netdev_priv(netdev);
1417         struct ixgb_desc_ring *tx_ring = &adapter->tx_ring;
1418
1419         netif_stop_queue(netdev);
1420         /* Herbert's original patch had:
1421          *  smp_mb__after_netif_stop_queue();
1422          * but since that doesn't exist yet, just open code it. */
1423         smp_mb();
1424
1425         /* We need to check again in a case another CPU has just
1426          * made room available. */
1427         if (likely(IXGB_DESC_UNUSED(tx_ring) < size))
1428                 return -EBUSY;
1429
1430         /* A reprieve! */
1431         netif_start_queue(netdev);
1432         ++adapter->restart_queue;
1433         return 0;
1434 }
1435
1436 static int ixgb_maybe_stop_tx(struct net_device *netdev,
1437                               struct ixgb_desc_ring *tx_ring, int size)
1438 {
1439         if (likely(IXGB_DESC_UNUSED(tx_ring) >= size))
1440                 return 0;
1441         return __ixgb_maybe_stop_tx(netdev, size);
1442 }
1443
1444
1445 /* Tx Descriptors needed, worst case */
1446 #define TXD_USE_COUNT(S) (((S) >> IXGB_MAX_TXD_PWR) + \
1447                          (((S) & (IXGB_MAX_DATA_PER_TXD - 1)) ? 1 : 0))
1448 #define DESC_NEEDED TXD_USE_COUNT(IXGB_MAX_DATA_PER_TXD) /* skb->date */ + \
1449         MAX_SKB_FRAGS * TXD_USE_COUNT(PAGE_SIZE) + 1 /* for context */ \
1450         + 1 /* one more needed for sentinel TSO workaround */
1451
1452 static int
1453 ixgb_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
1454 {
1455         struct ixgb_adapter *adapter = netdev_priv(netdev);
1456         unsigned int first;
1457         unsigned int tx_flags = 0;
1458         unsigned long flags;
1459         int vlan_id = 0;
1460         int tso;
1461
1462         if(skb->len <= 0) {
1463                 dev_kfree_skb_any(skb);
1464                 return 0;
1465         }
1466
1467 #ifdef NETIF_F_LLTX
1468         local_irq_save(flags);
1469         if (!spin_trylock(&adapter->tx_lock)) {
1470                 /* Collision - tell upper layer to requeue */
1471                 local_irq_restore(flags);
1472                 return NETDEV_TX_LOCKED;
1473         }
1474 #else
1475         spin_lock_irqsave(&adapter->tx_lock, flags);
1476 #endif
1477
1478         if (unlikely(ixgb_maybe_stop_tx(netdev, &adapter->tx_ring,
1479                      DESC_NEEDED))) {
1480                 netif_stop_queue(netdev);
1481                 spin_unlock_irqrestore(&adapter->tx_lock, flags);
1482                 return NETDEV_TX_BUSY;
1483         }
1484
1485 #ifndef NETIF_F_LLTX
1486         spin_unlock_irqrestore(&adapter->tx_lock, flags);
1487 #endif
1488
1489         if(adapter->vlgrp && vlan_tx_tag_present(skb)) {
1490                 tx_flags |= IXGB_TX_FLAGS_VLAN;
1491                 vlan_id = vlan_tx_tag_get(skb);
1492         }
1493
1494         first = adapter->tx_ring.next_to_use;
1495         
1496         tso = ixgb_tso(adapter, skb);
1497         if (tso < 0) {
1498                 dev_kfree_skb_any(skb);
1499 #ifdef NETIF_F_LLTX
1500                 spin_unlock_irqrestore(&adapter->tx_lock, flags);
1501 #endif
1502                 return NETDEV_TX_OK;
1503         }
1504
1505         if (likely(tso))
1506                 tx_flags |= IXGB_TX_FLAGS_TSO;
1507         else if(ixgb_tx_csum(adapter, skb))
1508                 tx_flags |= IXGB_TX_FLAGS_CSUM;
1509
1510         ixgb_tx_queue(adapter, ixgb_tx_map(adapter, skb, first), vlan_id,
1511                         tx_flags);
1512
1513         netdev->trans_start = jiffies;
1514
1515 #ifdef NETIF_F_LLTX
1516         /* Make sure there is space in the ring for the next send. */
1517         ixgb_maybe_stop_tx(netdev, &adapter->tx_ring, DESC_NEEDED);
1518
1519         spin_unlock_irqrestore(&adapter->tx_lock, flags);
1520
1521 #endif
1522         return NETDEV_TX_OK;
1523 }
1524
1525 /**
1526  * ixgb_tx_timeout - Respond to a Tx Hang
1527  * @netdev: network interface device structure
1528  **/
1529
1530 static void
1531 ixgb_tx_timeout(struct net_device *netdev)
1532 {
1533         struct ixgb_adapter *adapter = netdev_priv(netdev);
1534
1535         /* Do the reset outside of interrupt context */
1536         schedule_work(&adapter->tx_timeout_task);
1537 }
1538
1539 static void
1540 ixgb_tx_timeout_task(struct work_struct *work)
1541 {
1542         struct ixgb_adapter *adapter =
1543                 container_of(work, struct ixgb_adapter, tx_timeout_task);
1544
1545         adapter->tx_timeout_count++;
1546         ixgb_down(adapter, TRUE);
1547         ixgb_up(adapter);
1548 }
1549
1550 /**
1551  * ixgb_get_stats - Get System Network Statistics
1552  * @netdev: network interface device structure
1553  *
1554  * Returns the address of the device statistics structure.
1555  * The statistics are actually updated from the timer callback.
1556  **/
1557
1558 static struct net_device_stats *
1559 ixgb_get_stats(struct net_device *netdev)
1560 {
1561         struct ixgb_adapter *adapter = netdev_priv(netdev);
1562
1563         return &adapter->net_stats;
1564 }
1565
1566 /**
1567  * ixgb_change_mtu - Change the Maximum Transfer Unit
1568  * @netdev: network interface device structure
1569  * @new_mtu: new value for maximum frame size
1570  *
1571  * Returns 0 on success, negative on failure
1572  **/
1573
1574 static int
1575 ixgb_change_mtu(struct net_device *netdev, int new_mtu)
1576 {
1577         struct ixgb_adapter *adapter = netdev_priv(netdev);
1578         int max_frame = new_mtu + ENET_HEADER_SIZE + ENET_FCS_LENGTH;
1579         int old_max_frame = netdev->mtu + ENET_HEADER_SIZE + ENET_FCS_LENGTH;
1580
1581
1582         if((max_frame < IXGB_MIN_ENET_FRAME_SIZE_WITHOUT_FCS + ENET_FCS_LENGTH)
1583            || (max_frame > IXGB_MAX_JUMBO_FRAME_SIZE + ENET_FCS_LENGTH)) {
1584                 DPRINTK(PROBE, ERR, "Invalid MTU setting %d\n", new_mtu);
1585                 return -EINVAL;
1586         }
1587
1588         adapter->rx_buffer_len = max_frame;
1589
1590         netdev->mtu = new_mtu;
1591
1592         if ((old_max_frame != max_frame) && netif_running(netdev)) {
1593                 ixgb_down(adapter, TRUE);
1594                 ixgb_up(adapter);
1595         }
1596
1597         return 0;
1598 }
1599
1600 /**
1601  * ixgb_update_stats - Update the board statistics counters.
1602  * @adapter: board private structure
1603  **/
1604
1605 void
1606 ixgb_update_stats(struct ixgb_adapter *adapter)
1607 {
1608         struct net_device *netdev = adapter->netdev;
1609         struct pci_dev *pdev = adapter->pdev;
1610
1611         /* Prevent stats update while adapter is being reset */
1612         if (pci_channel_offline(pdev))
1613                 return;
1614
1615         if((netdev->flags & IFF_PROMISC) || (netdev->flags & IFF_ALLMULTI) ||
1616            (netdev->mc_count > IXGB_MAX_NUM_MULTICAST_ADDRESSES)) {
1617                 u64 multi = IXGB_READ_REG(&adapter->hw, MPRCL);
1618                 u32 bcast_l = IXGB_READ_REG(&adapter->hw, BPRCL);
1619                 u32 bcast_h = IXGB_READ_REG(&adapter->hw, BPRCH);
1620                 u64 bcast = ((u64)bcast_h << 32) | bcast_l; 
1621
1622                 multi |= ((u64)IXGB_READ_REG(&adapter->hw, MPRCH) << 32);
1623                 /* fix up multicast stats by removing broadcasts */
1624                 if(multi >= bcast)
1625                         multi -= bcast;
1626                 
1627                 adapter->stats.mprcl += (multi & 0xFFFFFFFF);
1628                 adapter->stats.mprch += (multi >> 32);
1629                 adapter->stats.bprcl += bcast_l; 
1630                 adapter->stats.bprch += bcast_h;
1631         } else {
1632                 adapter->stats.mprcl += IXGB_READ_REG(&adapter->hw, MPRCL);
1633                 adapter->stats.mprch += IXGB_READ_REG(&adapter->hw, MPRCH);
1634                 adapter->stats.bprcl += IXGB_READ_REG(&adapter->hw, BPRCL);
1635                 adapter->stats.bprch += IXGB_READ_REG(&adapter->hw, BPRCH);
1636         }
1637         adapter->stats.tprl += IXGB_READ_REG(&adapter->hw, TPRL);
1638         adapter->stats.tprh += IXGB_READ_REG(&adapter->hw, TPRH);
1639         adapter->stats.gprcl += IXGB_READ_REG(&adapter->hw, GPRCL);
1640         adapter->stats.gprch += IXGB_READ_REG(&adapter->hw, GPRCH);
1641         adapter->stats.uprcl += IXGB_READ_REG(&adapter->hw, UPRCL);
1642         adapter->stats.uprch += IXGB_READ_REG(&adapter->hw, UPRCH);
1643         adapter->stats.vprcl += IXGB_READ_REG(&adapter->hw, VPRCL);
1644         adapter->stats.vprch += IXGB_READ_REG(&adapter->hw, VPRCH);
1645         adapter->stats.jprcl += IXGB_READ_REG(&adapter->hw, JPRCL);
1646         adapter->stats.jprch += IXGB_READ_REG(&adapter->hw, JPRCH);
1647         adapter->stats.gorcl += IXGB_READ_REG(&adapter->hw, GORCL);
1648         adapter->stats.gorch += IXGB_READ_REG(&adapter->hw, GORCH);
1649         adapter->stats.torl += IXGB_READ_REG(&adapter->hw, TORL);
1650         adapter->stats.torh += IXGB_READ_REG(&adapter->hw, TORH);
1651         adapter->stats.rnbc += IXGB_READ_REG(&adapter->hw, RNBC);
1652         adapter->stats.ruc += IXGB_READ_REG(&adapter->hw, RUC);
1653         adapter->stats.roc += IXGB_READ_REG(&adapter->hw, ROC);
1654         adapter->stats.rlec += IXGB_READ_REG(&adapter->hw, RLEC);
1655         adapter->stats.crcerrs += IXGB_READ_REG(&adapter->hw, CRCERRS);
1656         adapter->stats.icbc += IXGB_READ_REG(&adapter->hw, ICBC);
1657         adapter->stats.ecbc += IXGB_READ_REG(&adapter->hw, ECBC);
1658         adapter->stats.mpc += IXGB_READ_REG(&adapter->hw, MPC);
1659         adapter->stats.tptl += IXGB_READ_REG(&adapter->hw, TPTL);
1660         adapter->stats.tpth += IXGB_READ_REG(&adapter->hw, TPTH);
1661         adapter->stats.gptcl += IXGB_READ_REG(&adapter->hw, GPTCL);
1662         adapter->stats.gptch += IXGB_READ_REG(&adapter->hw, GPTCH);
1663         adapter->stats.bptcl += IXGB_READ_REG(&adapter->hw, BPTCL);
1664         adapter->stats.bptch += IXGB_READ_REG(&adapter->hw, BPTCH);
1665         adapter->stats.mptcl += IXGB_READ_REG(&adapter->hw, MPTCL);
1666         adapter->stats.mptch += IXGB_READ_REG(&adapter->hw, MPTCH);
1667         adapter->stats.uptcl += IXGB_READ_REG(&adapter->hw, UPTCL);
1668         adapter->stats.uptch += IXGB_READ_REG(&adapter->hw, UPTCH);
1669         adapter->stats.vptcl += IXGB_READ_REG(&adapter->hw, VPTCL);
1670         adapter->stats.vptch += IXGB_READ_REG(&adapter->hw, VPTCH);
1671         adapter->stats.jptcl += IXGB_READ_REG(&adapter->hw, JPTCL);
1672         adapter->stats.jptch += IXGB_READ_REG(&adapter->hw, JPTCH);
1673         adapter->stats.gotcl += IXGB_READ_REG(&adapter->hw, GOTCL);
1674         adapter->stats.gotch += IXGB_READ_REG(&adapter->hw, GOTCH);
1675         adapter->stats.totl += IXGB_READ_REG(&adapter->hw, TOTL);
1676         adapter->stats.toth += IXGB_READ_REG(&adapter->hw, TOTH);
1677         adapter->stats.dc += IXGB_READ_REG(&adapter->hw, DC);
1678         adapter->stats.plt64c += IXGB_READ_REG(&adapter->hw, PLT64C);
1679         adapter->stats.tsctc += IXGB_READ_REG(&adapter->hw, TSCTC);
1680         adapter->stats.tsctfc += IXGB_READ_REG(&adapter->hw, TSCTFC);
1681         adapter->stats.ibic += IXGB_READ_REG(&adapter->hw, IBIC);
1682         adapter->stats.rfc += IXGB_READ_REG(&adapter->hw, RFC);
1683         adapter->stats.lfc += IXGB_READ_REG(&adapter->hw, LFC);
1684         adapter->stats.pfrc += IXGB_READ_REG(&adapter->hw, PFRC);
1685         adapter->stats.pftc += IXGB_READ_REG(&adapter->hw, PFTC);
1686         adapter->stats.mcfrc += IXGB_READ_REG(&adapter->hw, MCFRC);
1687         adapter->stats.mcftc += IXGB_READ_REG(&adapter->hw, MCFTC);
1688         adapter->stats.xonrxc += IXGB_READ_REG(&adapter->hw, XONRXC);
1689         adapter->stats.xontxc += IXGB_READ_REG(&adapter->hw, XONTXC);
1690         adapter->stats.xoffrxc += IXGB_READ_REG(&adapter->hw, XOFFRXC);
1691         adapter->stats.xofftxc += IXGB_READ_REG(&adapter->hw, XOFFTXC);
1692         adapter->stats.rjc += IXGB_READ_REG(&adapter->hw, RJC);
1693
1694         /* Fill out the OS statistics structure */
1695
1696         adapter->net_stats.rx_packets = adapter->stats.gprcl;
1697         adapter->net_stats.tx_packets = adapter->stats.gptcl;
1698         adapter->net_stats.rx_bytes = adapter->stats.gorcl;
1699         adapter->net_stats.tx_bytes = adapter->stats.gotcl;
1700         adapter->net_stats.multicast = adapter->stats.mprcl;
1701         adapter->net_stats.collisions = 0;
1702
1703         /* ignore RLEC as it reports errors for padded (<64bytes) frames
1704          * with a length in the type/len field */
1705         adapter->net_stats.rx_errors =
1706             /* adapter->stats.rnbc + */ adapter->stats.crcerrs +
1707             adapter->stats.ruc +
1708             adapter->stats.roc /*+ adapter->stats.rlec */  +
1709             adapter->stats.icbc +
1710             adapter->stats.ecbc + adapter->stats.mpc;
1711
1712         /* see above
1713          * adapter->net_stats.rx_length_errors = adapter->stats.rlec;
1714          */
1715
1716         adapter->net_stats.rx_crc_errors = adapter->stats.crcerrs;
1717         adapter->net_stats.rx_fifo_errors = adapter->stats.mpc;
1718         adapter->net_stats.rx_missed_errors = adapter->stats.mpc;
1719         adapter->net_stats.rx_over_errors = adapter->stats.mpc;
1720
1721         adapter->net_stats.tx_errors = 0;
1722         adapter->net_stats.rx_frame_errors = 0;
1723         adapter->net_stats.tx_aborted_errors = 0;
1724         adapter->net_stats.tx_carrier_errors = 0;
1725         adapter->net_stats.tx_fifo_errors = 0;
1726         adapter->net_stats.tx_heartbeat_errors = 0;
1727         adapter->net_stats.tx_window_errors = 0;
1728 }
1729
1730 #define IXGB_MAX_INTR 10
1731 /**
1732  * ixgb_intr - Interrupt Handler
1733  * @irq: interrupt number
1734  * @data: pointer to a network interface device structure
1735  **/
1736
1737 static irqreturn_t
1738 ixgb_intr(int irq, void *data)
1739 {
1740         struct net_device *netdev = data;
1741         struct ixgb_adapter *adapter = netdev_priv(netdev);
1742         struct ixgb_hw *hw = &adapter->hw;
1743         uint32_t icr = IXGB_READ_REG(hw, ICR);
1744 #ifndef CONFIG_IXGB_NAPI
1745         unsigned int i;
1746 #endif
1747
1748         if(unlikely(!icr))
1749                 return IRQ_NONE;  /* Not our interrupt */
1750
1751         if(unlikely(icr & (IXGB_INT_RXSEQ | IXGB_INT_LSC))) {
1752                 mod_timer(&adapter->watchdog_timer, jiffies);
1753         }
1754
1755 #ifdef CONFIG_IXGB_NAPI
1756         if (netif_rx_schedule_prep(netdev, &adapter->napi)) {
1757
1758                 /* Disable interrupts and register for poll. The flush 
1759                   of the posted write is intentionally left out.
1760                 */
1761
1762                 atomic_inc(&adapter->irq_sem);
1763                 IXGB_WRITE_REG(&adapter->hw, IMC, ~0);
1764                 __netif_rx_schedule(netdev, &adapter->napi);
1765         }
1766 #else
1767         /* yes, that is actually a & and it is meant to make sure that
1768          * every pass through this for loop checks both receive and
1769          * transmit queues for completed descriptors, intended to
1770          * avoid starvation issues and assist tx/rx fairness. */
1771         for(i = 0; i < IXGB_MAX_INTR; i++)
1772                 if(!ixgb_clean_rx_irq(adapter) &
1773                    !ixgb_clean_tx_irq(adapter))
1774                         break;
1775 #endif 
1776         return IRQ_HANDLED;
1777 }
1778
1779 #ifdef CONFIG_IXGB_NAPI
1780 /**
1781  * ixgb_clean - NAPI Rx polling callback
1782  * @adapter: board private structure
1783  **/
1784
1785 static int
1786 ixgb_clean(struct napi_struct *napi, int budget)
1787 {
1788         struct ixgb_adapter *adapter = container_of(napi, struct ixgb_adapter, napi);
1789         struct net_device *netdev = adapter->netdev;
1790         int work_done = 0;
1791
1792         ixgb_clean_tx_irq(adapter);
1793         ixgb_clean_rx_irq(adapter, &work_done, budget);
1794
1795         /* If budget not fully consumed, exit the polling mode */
1796         if (work_done < budget) {
1797                 netif_rx_complete(netdev, napi);
1798                 ixgb_irq_enable(adapter);
1799         }
1800
1801         return work_done;
1802 }
1803 #endif
1804
1805 /**
1806  * ixgb_clean_tx_irq - Reclaim resources after transmit completes
1807  * @adapter: board private structure
1808  **/
1809
1810 static boolean_t
1811 ixgb_clean_tx_irq(struct ixgb_adapter *adapter)
1812 {
1813         struct ixgb_desc_ring *tx_ring = &adapter->tx_ring;
1814         struct net_device *netdev = adapter->netdev;
1815         struct ixgb_tx_desc *tx_desc, *eop_desc;
1816         struct ixgb_buffer *buffer_info;
1817         unsigned int i, eop;
1818         boolean_t cleaned = FALSE;
1819
1820         i = tx_ring->next_to_clean;
1821         eop = tx_ring->buffer_info[i].next_to_watch;
1822         eop_desc = IXGB_TX_DESC(*tx_ring, eop);
1823
1824         while(eop_desc->status & IXGB_TX_DESC_STATUS_DD) {
1825
1826                 for(cleaned = FALSE; !cleaned; ) {
1827                         tx_desc = IXGB_TX_DESC(*tx_ring, i);
1828                         buffer_info = &tx_ring->buffer_info[i];
1829
1830                         if (tx_desc->popts
1831                             & (IXGB_TX_DESC_POPTS_TXSM |
1832                                IXGB_TX_DESC_POPTS_IXSM))
1833                                 adapter->hw_csum_tx_good++;
1834
1835                         ixgb_unmap_and_free_tx_resource(adapter, buffer_info);
1836
1837                         *(uint32_t *)&(tx_desc->status) = 0;
1838
1839                         cleaned = (i == eop);
1840                         if(++i == tx_ring->count) i = 0;
1841                 }
1842
1843                 eop = tx_ring->buffer_info[i].next_to_watch;
1844                 eop_desc = IXGB_TX_DESC(*tx_ring, eop);
1845         }
1846
1847         tx_ring->next_to_clean = i;
1848
1849         if (unlikely(netif_queue_stopped(netdev))) {
1850                 spin_lock(&adapter->tx_lock);
1851                 if (netif_queue_stopped(netdev) && netif_carrier_ok(netdev) &&
1852                     (IXGB_DESC_UNUSED(tx_ring) >= DESC_NEEDED))
1853                         netif_wake_queue(netdev);
1854                 spin_unlock(&adapter->tx_lock);
1855         }
1856
1857         if(adapter->detect_tx_hung) {
1858                 /* detect a transmit hang in hardware, this serializes the
1859                  * check with the clearing of time_stamp and movement of i */
1860                 adapter->detect_tx_hung = FALSE;
1861                 if (tx_ring->buffer_info[eop].dma &&
1862                    time_after(jiffies, tx_ring->buffer_info[eop].time_stamp + HZ)
1863                    && !(IXGB_READ_REG(&adapter->hw, STATUS) &
1864                         IXGB_STATUS_TXOFF)) {
1865                         /* detected Tx unit hang */
1866                         DPRINTK(DRV, ERR, "Detected Tx Unit Hang\n"
1867                                         "  TDH                  <%x>\n"
1868                                         "  TDT                  <%x>\n"
1869                                         "  next_to_use          <%x>\n"
1870                                         "  next_to_clean        <%x>\n"
1871                                         "buffer_info[next_to_clean]\n"
1872                                         "  time_stamp           <%lx>\n"
1873                                         "  next_to_watch        <%x>\n"
1874                                         "  jiffies              <%lx>\n"
1875                                         "  next_to_watch.status <%x>\n",
1876                                 IXGB_READ_REG(&adapter->hw, TDH),
1877                                 IXGB_READ_REG(&adapter->hw, TDT),
1878                                 tx_ring->next_to_use,
1879                                 tx_ring->next_to_clean,
1880                                 tx_ring->buffer_info[eop].time_stamp,
1881                                 eop,
1882                                 jiffies,
1883                                 eop_desc->status);
1884                         netif_stop_queue(netdev);
1885                 }
1886         }
1887
1888         return cleaned;
1889 }
1890
1891 /**
1892  * ixgb_rx_checksum - Receive Checksum Offload for 82597.
1893  * @adapter: board private structure
1894  * @rx_desc: receive descriptor
1895  * @sk_buff: socket buffer with received data
1896  **/
1897
1898 static void
1899 ixgb_rx_checksum(struct ixgb_adapter *adapter,
1900                  struct ixgb_rx_desc *rx_desc,
1901                  struct sk_buff *skb)
1902 {
1903         /* Ignore Checksum bit is set OR
1904          * TCP Checksum has not been calculated
1905          */
1906         if((rx_desc->status & IXGB_RX_DESC_STATUS_IXSM) ||
1907            (!(rx_desc->status & IXGB_RX_DESC_STATUS_TCPCS))) {
1908                 skb->ip_summed = CHECKSUM_NONE;
1909                 return;
1910         }
1911
1912         /* At this point we know the hardware did the TCP checksum */
1913         /* now look at the TCP checksum error bit */
1914         if(rx_desc->errors & IXGB_RX_DESC_ERRORS_TCPE) {
1915                 /* let the stack verify checksum errors */
1916                 skb->ip_summed = CHECKSUM_NONE;
1917                 adapter->hw_csum_rx_error++;
1918         } else {
1919                 /* TCP checksum is good */
1920                 skb->ip_summed = CHECKSUM_UNNECESSARY;
1921                 adapter->hw_csum_rx_good++;
1922         }
1923 }
1924
1925 /**
1926  * ixgb_clean_rx_irq - Send received data up the network stack,
1927  * @adapter: board private structure
1928  **/
1929
1930 static boolean_t
1931 #ifdef CONFIG_IXGB_NAPI
1932 ixgb_clean_rx_irq(struct ixgb_adapter *adapter, int *work_done, int work_to_do)
1933 #else
1934 ixgb_clean_rx_irq(struct ixgb_adapter *adapter)
1935 #endif
1936 {
1937         struct ixgb_desc_ring *rx_ring = &adapter->rx_ring;
1938         struct net_device *netdev = adapter->netdev;
1939         struct pci_dev *pdev = adapter->pdev;
1940         struct ixgb_rx_desc *rx_desc, *next_rxd;
1941         struct ixgb_buffer *buffer_info, *next_buffer, *next2_buffer;
1942         uint32_t length;
1943         unsigned int i, j;
1944         boolean_t cleaned = FALSE;
1945
1946         i = rx_ring->next_to_clean;
1947         rx_desc = IXGB_RX_DESC(*rx_ring, i);
1948         buffer_info = &rx_ring->buffer_info[i];
1949
1950         while(rx_desc->status & IXGB_RX_DESC_STATUS_DD) {
1951                 struct sk_buff *skb, *next_skb;
1952                 u8 status;
1953
1954 #ifdef CONFIG_IXGB_NAPI
1955                 if(*work_done >= work_to_do)
1956                         break;
1957
1958                 (*work_done)++;
1959 #endif
1960                 status = rx_desc->status;
1961                 skb = buffer_info->skb;
1962                 buffer_info->skb = NULL;
1963
1964                 prefetch(skb->data);
1965
1966                 if(++i == rx_ring->count) i = 0;
1967                 next_rxd = IXGB_RX_DESC(*rx_ring, i);
1968                 prefetch(next_rxd);
1969
1970                 if((j = i + 1) == rx_ring->count) j = 0;
1971                 next2_buffer = &rx_ring->buffer_info[j];
1972                 prefetch(next2_buffer);
1973
1974                 next_buffer = &rx_ring->buffer_info[i];
1975                 next_skb = next_buffer->skb;
1976                 prefetch(next_skb);
1977
1978                 cleaned = TRUE;
1979
1980                 pci_unmap_single(pdev,
1981                                  buffer_info->dma,
1982                                  buffer_info->length,
1983                                  PCI_DMA_FROMDEVICE);
1984
1985                 length = le16_to_cpu(rx_desc->length);
1986
1987                 if(unlikely(!(status & IXGB_RX_DESC_STATUS_EOP))) {
1988
1989                         /* All receives must fit into a single buffer */
1990
1991                         IXGB_DBG("Receive packet consumed multiple buffers "
1992                                          "length<%x>\n", length);
1993
1994                         dev_kfree_skb_irq(skb);
1995                         goto rxdesc_done;
1996                 }
1997
1998                 if (unlikely(rx_desc->errors
1999                              & (IXGB_RX_DESC_ERRORS_CE | IXGB_RX_DESC_ERRORS_SE
2000                                 | IXGB_RX_DESC_ERRORS_P |
2001                                 IXGB_RX_DESC_ERRORS_RXE))) {
2002
2003                         dev_kfree_skb_irq(skb);
2004                         goto rxdesc_done;
2005                 }
2006
2007                 /* code added for copybreak, this should improve
2008                  * performance for small packets with large amounts
2009                  * of reassembly being done in the stack */
2010 #define IXGB_CB_LENGTH 256
2011                 if (length < IXGB_CB_LENGTH) {
2012                         struct sk_buff *new_skb =
2013                             netdev_alloc_skb(netdev, length + NET_IP_ALIGN);
2014                         if (new_skb) {
2015                                 skb_reserve(new_skb, NET_IP_ALIGN);
2016                                 skb_copy_to_linear_data_offset(new_skb,
2017                                                                -NET_IP_ALIGN,
2018                                                                (skb->data -
2019                                                                 NET_IP_ALIGN),
2020                                                                (length +
2021                                                                 NET_IP_ALIGN));
2022                                 /* save the skb in buffer_info as good */
2023                                 buffer_info->skb = skb;
2024                                 skb = new_skb;
2025                         }
2026                 }
2027                 /* end copybreak code */
2028
2029                 /* Good Receive */
2030                 skb_put(skb, length);
2031
2032                 /* Receive Checksum Offload */
2033                 ixgb_rx_checksum(adapter, rx_desc, skb);
2034
2035                 skb->protocol = eth_type_trans(skb, netdev);
2036 #ifdef CONFIG_IXGB_NAPI
2037                 if(adapter->vlgrp && (status & IXGB_RX_DESC_STATUS_VP)) {
2038                         vlan_hwaccel_receive_skb(skb, adapter->vlgrp,
2039                                 le16_to_cpu(rx_desc->special) &
2040                                         IXGB_RX_DESC_SPECIAL_VLAN_MASK);
2041                 } else {
2042                         netif_receive_skb(skb);
2043                 }
2044 #else /* CONFIG_IXGB_NAPI */
2045                 if(adapter->vlgrp && (status & IXGB_RX_DESC_STATUS_VP)) {
2046                         vlan_hwaccel_rx(skb, adapter->vlgrp,
2047                                 le16_to_cpu(rx_desc->special) &
2048                                         IXGB_RX_DESC_SPECIAL_VLAN_MASK);
2049                 } else {
2050                         netif_rx(skb);
2051                 }
2052 #endif /* CONFIG_IXGB_NAPI */
2053                 netdev->last_rx = jiffies;
2054
2055 rxdesc_done:
2056                 /* clean up descriptor, might be written over by hw */
2057                 rx_desc->status = 0;
2058
2059                 /* use prefetched values */
2060                 rx_desc = next_rxd;
2061                 buffer_info = next_buffer;
2062         }
2063
2064         rx_ring->next_to_clean = i;
2065
2066         ixgb_alloc_rx_buffers(adapter);
2067
2068         return cleaned;
2069 }
2070
2071 /**
2072  * ixgb_alloc_rx_buffers - Replace used receive buffers
2073  * @adapter: address of board private structure
2074  **/
2075
2076 static void
2077 ixgb_alloc_rx_buffers(struct ixgb_adapter *adapter)
2078 {
2079         struct ixgb_desc_ring *rx_ring = &adapter->rx_ring;
2080         struct net_device *netdev = adapter->netdev;
2081         struct pci_dev *pdev = adapter->pdev;
2082         struct ixgb_rx_desc *rx_desc;
2083         struct ixgb_buffer *buffer_info;
2084         struct sk_buff *skb;
2085         unsigned int i;
2086         int num_group_tail_writes;
2087         long cleancount;
2088
2089         i = rx_ring->next_to_use;
2090         buffer_info = &rx_ring->buffer_info[i];
2091         cleancount = IXGB_DESC_UNUSED(rx_ring);
2092
2093         num_group_tail_writes = IXGB_RX_BUFFER_WRITE;
2094
2095         /* leave three descriptors unused */
2096         while(--cleancount > 2) {
2097                 /* recycle! its good for you */
2098                 skb = buffer_info->skb;
2099                 if (skb) {
2100                         skb_trim(skb, 0);
2101                         goto map_skb;
2102                 }
2103
2104                 skb = netdev_alloc_skb(netdev, adapter->rx_buffer_len
2105                                        + NET_IP_ALIGN);
2106                 if (unlikely(!skb)) {
2107                         /* Better luck next round */
2108                         adapter->alloc_rx_buff_failed++;
2109                         break;
2110                 }
2111
2112                 /* Make buffer alignment 2 beyond a 16 byte boundary
2113                  * this will result in a 16 byte aligned IP header after
2114                  * the 14 byte MAC header is removed
2115                  */
2116                 skb_reserve(skb, NET_IP_ALIGN);
2117
2118                 buffer_info->skb = skb;
2119                 buffer_info->length = adapter->rx_buffer_len;
2120 map_skb:
2121                 buffer_info->dma = pci_map_single(pdev,
2122                                                   skb->data,
2123                                                   adapter->rx_buffer_len,
2124                                                   PCI_DMA_FROMDEVICE);
2125
2126                 rx_desc = IXGB_RX_DESC(*rx_ring, i);
2127                 rx_desc->buff_addr = cpu_to_le64(buffer_info->dma);
2128                 /* guarantee DD bit not set now before h/w gets descriptor
2129                  * this is the rest of the workaround for h/w double 
2130                  * writeback. */
2131                 rx_desc->status = 0;
2132
2133
2134                 if(++i == rx_ring->count) i = 0;
2135                 buffer_info = &rx_ring->buffer_info[i];
2136         }
2137
2138         if (likely(rx_ring->next_to_use != i)) {
2139                 rx_ring->next_to_use = i;
2140                 if (unlikely(i-- == 0))
2141                         i = (rx_ring->count - 1);
2142
2143                 /* Force memory writes to complete before letting h/w
2144                  * know there are new descriptors to fetch.  (Only
2145                  * applicable for weak-ordered memory model archs, such
2146                  * as IA-64). */
2147                 wmb();
2148                 IXGB_WRITE_REG(&adapter->hw, RDT, i);
2149         }
2150 }
2151
2152 /**
2153  * ixgb_vlan_rx_register - enables or disables vlan tagging/stripping.
2154  * 
2155  * @param netdev network interface device structure
2156  * @param grp indicates to enable or disable tagging/stripping
2157  **/
2158 static void
2159 ixgb_vlan_rx_register(struct net_device *netdev, struct vlan_group *grp)
2160 {
2161         struct ixgb_adapter *adapter = netdev_priv(netdev);
2162         uint32_t ctrl, rctl;
2163
2164         ixgb_irq_disable(adapter);
2165         adapter->vlgrp = grp;
2166
2167         if(grp) {
2168                 /* enable VLAN tag insert/strip */
2169                 ctrl = IXGB_READ_REG(&adapter->hw, CTRL0);
2170                 ctrl |= IXGB_CTRL0_VME;
2171                 IXGB_WRITE_REG(&adapter->hw, CTRL0, ctrl);
2172
2173                 /* enable VLAN receive filtering */
2174
2175                 rctl = IXGB_READ_REG(&adapter->hw, RCTL);
2176                 rctl |= IXGB_RCTL_VFE;
2177                 rctl &= ~IXGB_RCTL_CFIEN;
2178                 IXGB_WRITE_REG(&adapter->hw, RCTL, rctl);
2179         } else {
2180                 /* disable VLAN tag insert/strip */
2181
2182                 ctrl = IXGB_READ_REG(&adapter->hw, CTRL0);
2183                 ctrl &= ~IXGB_CTRL0_VME;
2184                 IXGB_WRITE_REG(&adapter->hw, CTRL0, ctrl);
2185
2186                 /* disable VLAN filtering */
2187
2188                 rctl = IXGB_READ_REG(&adapter->hw, RCTL);
2189                 rctl &= ~IXGB_RCTL_VFE;
2190                 IXGB_WRITE_REG(&adapter->hw, RCTL, rctl);
2191         }
2192
2193         ixgb_irq_enable(adapter);
2194 }
2195
2196 static void
2197 ixgb_vlan_rx_add_vid(struct net_device *netdev, uint16_t vid)
2198 {
2199         struct ixgb_adapter *adapter = netdev_priv(netdev);
2200         uint32_t vfta, index;
2201
2202         /* add VID to filter table */
2203
2204         index = (vid >> 5) & 0x7F;
2205         vfta = IXGB_READ_REG_ARRAY(&adapter->hw, VFTA, index);
2206         vfta |= (1 << (vid & 0x1F));
2207         ixgb_write_vfta(&adapter->hw, index, vfta);
2208 }
2209
2210 static void
2211 ixgb_vlan_rx_kill_vid(struct net_device *netdev, uint16_t vid)
2212 {
2213         struct ixgb_adapter *adapter = netdev_priv(netdev);
2214         uint32_t vfta, index;
2215
2216         ixgb_irq_disable(adapter);
2217
2218         vlan_group_set_device(adapter->vlgrp, vid, NULL);
2219
2220         ixgb_irq_enable(adapter);
2221
2222         /* remove VID from filter table*/
2223
2224         index = (vid >> 5) & 0x7F;
2225         vfta = IXGB_READ_REG_ARRAY(&adapter->hw, VFTA, index);
2226         vfta &= ~(1 << (vid & 0x1F));
2227         ixgb_write_vfta(&adapter->hw, index, vfta);
2228 }
2229
2230 static void
2231 ixgb_restore_vlan(struct ixgb_adapter *adapter)
2232 {
2233         ixgb_vlan_rx_register(adapter->netdev, adapter->vlgrp);
2234
2235         if(adapter->vlgrp) {
2236                 uint16_t vid;
2237                 for(vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
2238                         if(!vlan_group_get_device(adapter->vlgrp, vid))
2239                                 continue;
2240                         ixgb_vlan_rx_add_vid(adapter->netdev, vid);
2241                 }
2242         }
2243 }
2244
2245 #ifdef CONFIG_NET_POLL_CONTROLLER
2246 /*
2247  * Polling 'interrupt' - used by things like netconsole to send skbs
2248  * without having to re-enable interrupts. It's not called while
2249  * the interrupt routine is executing.
2250  */
2251
2252 static void ixgb_netpoll(struct net_device *dev)
2253 {
2254         struct ixgb_adapter *adapter = netdev_priv(dev);
2255
2256         disable_irq(adapter->pdev->irq);
2257         ixgb_intr(adapter->pdev->irq, dev);
2258         enable_irq(adapter->pdev->irq);
2259 }
2260 #endif
2261
2262 /**
2263  * ixgb_io_error_detected() - called when PCI error is detected
2264  * @pdev    pointer to pci device with error
2265  * @state   pci channel state after error
2266  *
2267  * This callback is called by the PCI subsystem whenever
2268  * a PCI bus error is detected.
2269  */
2270 static pci_ers_result_t ixgb_io_error_detected (struct pci_dev *pdev,
2271                                      enum pci_channel_state state)
2272 {
2273         struct net_device *netdev = pci_get_drvdata(pdev);
2274         struct ixgb_adapter *adapter = netdev_priv(netdev);
2275
2276         if(netif_running(netdev))
2277                 ixgb_down(adapter, TRUE);
2278
2279         pci_disable_device(pdev);
2280
2281         /* Request a slot reset. */
2282         return PCI_ERS_RESULT_NEED_RESET;
2283 }
2284
2285 /**
2286  * ixgb_io_slot_reset - called after the pci bus has been reset.
2287  * @pdev    pointer to pci device with error
2288  *
2289  * This callback is called after the PCI buss has been reset.
2290  * Basically, this tries to restart the card from scratch.
2291  * This is a shortened version of the device probe/discovery code,
2292  * it resembles the first-half of the ixgb_probe() routine.
2293  */
2294 static pci_ers_result_t ixgb_io_slot_reset (struct pci_dev *pdev)
2295 {
2296         struct net_device *netdev = pci_get_drvdata(pdev);
2297         struct ixgb_adapter *adapter = netdev_priv(netdev);
2298
2299         if(pci_enable_device(pdev)) {
2300                 DPRINTK(PROBE, ERR, "Cannot re-enable PCI device after reset.\n");
2301                 return PCI_ERS_RESULT_DISCONNECT;
2302         }
2303
2304         /* Perform card reset only on one instance of the card */
2305         if (0 != PCI_FUNC (pdev->devfn))
2306                 return PCI_ERS_RESULT_RECOVERED;
2307
2308         pci_set_master(pdev);
2309
2310         netif_carrier_off(netdev);
2311         netif_stop_queue(netdev);
2312         ixgb_reset(adapter);
2313
2314         /* Make sure the EEPROM is good */
2315         if(!ixgb_validate_eeprom_checksum(&adapter->hw)) {
2316                 DPRINTK(PROBE, ERR, "After reset, the EEPROM checksum is not valid.\n");
2317                 return PCI_ERS_RESULT_DISCONNECT;
2318         }
2319         ixgb_get_ee_mac_addr(&adapter->hw, netdev->dev_addr);
2320         memcpy(netdev->perm_addr, netdev->dev_addr, netdev->addr_len);
2321
2322         if(!is_valid_ether_addr(netdev->perm_addr)) {
2323                 DPRINTK(PROBE, ERR, "After reset, invalid MAC address.\n");
2324                 return PCI_ERS_RESULT_DISCONNECT;
2325         }
2326
2327         return PCI_ERS_RESULT_RECOVERED;
2328 }
2329
2330 /**
2331  * ixgb_io_resume - called when its OK to resume normal operations
2332  * @pdev    pointer to pci device with error
2333  *
2334  * The error recovery driver tells us that its OK to resume
2335  * normal operation. Implementation resembles the second-half
2336  * of the ixgb_probe() routine.
2337  */
2338 static void ixgb_io_resume (struct pci_dev *pdev)
2339 {
2340         struct net_device *netdev = pci_get_drvdata(pdev);
2341         struct ixgb_adapter *adapter = netdev_priv(netdev);
2342
2343         pci_set_master(pdev);
2344
2345         if(netif_running(netdev)) {
2346                 if(ixgb_up(adapter)) {
2347                         printk ("ixgb: can't bring device back up after reset\n");
2348                         return;
2349                 }
2350         }
2351
2352         netif_device_attach(netdev);
2353         mod_timer(&adapter->watchdog_timer, jiffies);
2354 }
2355
2356 /* ixgb_main.c */