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