Merge branch 'x86/crashdump' into x86/urgent
[linux-2.6] / drivers / net / ixgb / ixgb_main.c
1 /*******************************************************************************
2
3   Intel PRO/10GbE Linux driver
4   Copyright(c) 1999 - 2008 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 #define DRIVERNAPI "-NAPI"
35 #define DRV_VERSION "1.0.135-k2" DRIVERNAPI
36 const char ixgb_driver_version[] = DRV_VERSION;
37 static const char ixgb_copyright[] = "Copyright (c) 1999-2008 Intel Corporation.";
38
39 #define IXGB_CB_LENGTH 256
40 static unsigned int copybreak __read_mostly = IXGB_CB_LENGTH;
41 module_param(copybreak, uint, 0644);
42 MODULE_PARM_DESC(copybreak,
43         "Maximum size of packet that is copied to a new buffer on receive");
44
45 /* ixgb_pci_tbl - PCI Device ID Table
46  *
47  * Wildcard entries (PCI_ANY_ID) should come last
48  * Last entry must be all 0s
49  *
50  * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
51  *   Class, Class Mask, private data (not used) }
52  */
53 static struct pci_device_id ixgb_pci_tbl[] = {
54         {INTEL_VENDOR_ID, IXGB_DEVICE_ID_82597EX,
55          PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0},
56         {INTEL_VENDOR_ID, IXGB_DEVICE_ID_82597EX_CX4,
57          PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0},
58         {INTEL_VENDOR_ID, IXGB_DEVICE_ID_82597EX_SR,
59          PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0},
60         {INTEL_VENDOR_ID, IXGB_DEVICE_ID_82597EX_LR,
61          PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0},
62
63         /* required last entry */
64         {0,}
65 };
66
67 MODULE_DEVICE_TABLE(pci, ixgb_pci_tbl);
68
69 /* Local Function Prototypes */
70 static int ixgb_init_module(void);
71 static void ixgb_exit_module(void);
72 static int ixgb_probe(struct pci_dev *pdev, const struct pci_device_id *ent);
73 static void __devexit ixgb_remove(struct pci_dev *pdev);
74 static int ixgb_sw_init(struct ixgb_adapter *adapter);
75 static int ixgb_open(struct net_device *netdev);
76 static int ixgb_close(struct net_device *netdev);
77 static void ixgb_configure_tx(struct ixgb_adapter *adapter);
78 static void ixgb_configure_rx(struct ixgb_adapter *adapter);
79 static void ixgb_setup_rctl(struct ixgb_adapter *adapter);
80 static void ixgb_clean_tx_ring(struct ixgb_adapter *adapter);
81 static void ixgb_clean_rx_ring(struct ixgb_adapter *adapter);
82 static void ixgb_set_multi(struct net_device *netdev);
83 static void ixgb_watchdog(unsigned long data);
84 static int ixgb_xmit_frame(struct sk_buff *skb, struct net_device *netdev);
85 static struct net_device_stats *ixgb_get_stats(struct net_device *netdev);
86 static int ixgb_change_mtu(struct net_device *netdev, int new_mtu);
87 static int ixgb_set_mac(struct net_device *netdev, void *p);
88 static irqreturn_t ixgb_intr(int irq, void *data);
89 static bool ixgb_clean_tx_irq(struct ixgb_adapter *adapter);
90
91 static int ixgb_clean(struct napi_struct *, int);
92 static bool ixgb_clean_rx_irq(struct ixgb_adapter *, int *, int);
93 static void ixgb_alloc_rx_buffers(struct ixgb_adapter *, int);
94
95 static void ixgb_tx_timeout(struct net_device *dev);
96 static void ixgb_tx_timeout_task(struct work_struct *work);
97
98 static void ixgb_vlan_rx_register(struct net_device *netdev,
99                                   struct vlan_group *grp);
100 static void ixgb_vlan_rx_add_vid(struct net_device *netdev, u16 vid);
101 static void ixgb_vlan_rx_kill_vid(struct net_device *netdev, u16 vid);
102 static void ixgb_restore_vlan(struct ixgb_adapter *adapter);
103
104 #ifdef CONFIG_NET_POLL_CONTROLLER
105 /* for netdump / net console */
106 static void ixgb_netpoll(struct net_device *dev);
107 #endif
108
109 static pci_ers_result_t ixgb_io_error_detected (struct pci_dev *pdev,
110                              enum pci_channel_state state);
111 static pci_ers_result_t ixgb_io_slot_reset (struct pci_dev *pdev);
112 static void ixgb_io_resume (struct pci_dev *pdev);
113
114 static struct pci_error_handlers ixgb_err_handler = {
115         .error_detected = ixgb_io_error_detected,
116         .slot_reset = ixgb_io_slot_reset,
117         .resume = ixgb_io_resume,
118 };
119
120 static struct pci_driver ixgb_driver = {
121         .name     = ixgb_driver_name,
122         .id_table = ixgb_pci_tbl,
123         .probe    = ixgb_probe,
124         .remove   = __devexit_p(ixgb_remove),
125         .err_handler = &ixgb_err_handler
126 };
127
128 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
129 MODULE_DESCRIPTION("Intel(R) PRO/10GbE Network Driver");
130 MODULE_LICENSE("GPL");
131 MODULE_VERSION(DRV_VERSION);
132
133 #define DEFAULT_DEBUG_LEVEL_SHIFT 3
134 static int debug = DEFAULT_DEBUG_LEVEL_SHIFT;
135 module_param(debug, int, 0);
136 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
137
138 /**
139  * ixgb_init_module - Driver Registration Routine
140  *
141  * ixgb_init_module is the first routine called when the driver is
142  * loaded. All it does is register with the PCI subsystem.
143  **/
144
145 static int __init
146 ixgb_init_module(void)
147 {
148         printk(KERN_INFO "%s - version %s\n",
149                ixgb_driver_string, ixgb_driver_version);
150
151         printk(KERN_INFO "%s\n", ixgb_copyright);
152
153         return pci_register_driver(&ixgb_driver);
154 }
155
156 module_init(ixgb_init_module);
157
158 /**
159  * ixgb_exit_module - Driver Exit Cleanup Routine
160  *
161  * ixgb_exit_module is called just before the driver is removed
162  * from memory.
163  **/
164
165 static void __exit
166 ixgb_exit_module(void)
167 {
168         pci_unregister_driver(&ixgb_driver);
169 }
170
171 module_exit(ixgb_exit_module);
172
173 /**
174  * ixgb_irq_disable - Mask off interrupt generation on the NIC
175  * @adapter: board private structure
176  **/
177
178 static void
179 ixgb_irq_disable(struct ixgb_adapter *adapter)
180 {
181         IXGB_WRITE_REG(&adapter->hw, IMC, ~0);
182         IXGB_WRITE_FLUSH(&adapter->hw);
183         synchronize_irq(adapter->pdev->irq);
184 }
185
186 /**
187  * ixgb_irq_enable - Enable default interrupt generation settings
188  * @adapter: board private structure
189  **/
190
191 static void
192 ixgb_irq_enable(struct ixgb_adapter *adapter)
193 {
194         u32 val = IXGB_INT_RXT0 | IXGB_INT_RXDMT0 |
195                   IXGB_INT_TXDW | IXGB_INT_LSC;
196         if (adapter->hw.subsystem_vendor_id == SUN_SUBVENDOR_ID)
197                 val |= IXGB_INT_GPI0;
198         IXGB_WRITE_REG(&adapter->hw, IMS, val);
199         IXGB_WRITE_FLUSH(&adapter->hw);
200 }
201
202 int
203 ixgb_up(struct ixgb_adapter *adapter)
204 {
205         struct net_device *netdev = adapter->netdev;
206         int err, irq_flags = IRQF_SHARED;
207         int max_frame = netdev->mtu + ENET_HEADER_SIZE + ENET_FCS_LENGTH;
208         struct ixgb_hw *hw = &adapter->hw;
209
210         /* hardware has been reset, we need to reload some things */
211
212         ixgb_rar_set(hw, netdev->dev_addr, 0);
213         ixgb_set_multi(netdev);
214
215         ixgb_restore_vlan(adapter);
216
217         ixgb_configure_tx(adapter);
218         ixgb_setup_rctl(adapter);
219         ixgb_configure_rx(adapter);
220         ixgb_alloc_rx_buffers(adapter, IXGB_DESC_UNUSED(&adapter->rx_ring));
221
222         /* disable interrupts and get the hardware into a known state */
223         IXGB_WRITE_REG(&adapter->hw, IMC, 0xffffffff);
224
225         /* only enable MSI if bus is in PCI-X mode */
226         if (IXGB_READ_REG(&adapter->hw, STATUS) & IXGB_STATUS_PCIX_MODE) {
227                 err = pci_enable_msi(adapter->pdev);
228                 if (!err) {
229                         adapter->have_msi = 1;
230                         irq_flags = 0;
231                 }
232                 /* proceed to try to request regular interrupt */
233         }
234
235         err = request_irq(adapter->pdev->irq, &ixgb_intr, irq_flags,
236                           netdev->name, netdev);
237         if (err) {
238                 if (adapter->have_msi)
239                         pci_disable_msi(adapter->pdev);
240                 DPRINTK(PROBE, ERR,
241                  "Unable to allocate interrupt Error: %d\n", err);
242                 return err;
243         }
244
245         if ((hw->max_frame_size != max_frame) ||
246                 (hw->max_frame_size !=
247                 (IXGB_READ_REG(hw, MFS) >> IXGB_MFS_SHIFT))) {
248
249                 hw->max_frame_size = max_frame;
250
251                 IXGB_WRITE_REG(hw, MFS, hw->max_frame_size << IXGB_MFS_SHIFT);
252
253                 if (hw->max_frame_size >
254                    IXGB_MAX_ENET_FRAME_SIZE_WITHOUT_FCS + ENET_FCS_LENGTH) {
255                         u32 ctrl0 = IXGB_READ_REG(hw, CTRL0);
256
257                         if (!(ctrl0 & IXGB_CTRL0_JFE)) {
258                                 ctrl0 |= IXGB_CTRL0_JFE;
259                                 IXGB_WRITE_REG(hw, CTRL0, ctrl0);
260                         }
261                 }
262         }
263
264         clear_bit(__IXGB_DOWN, &adapter->flags);
265
266         napi_enable(&adapter->napi);
267         ixgb_irq_enable(adapter);
268
269         mod_timer(&adapter->watchdog_timer, jiffies);
270
271         return 0;
272 }
273
274 void
275 ixgb_down(struct ixgb_adapter *adapter, bool kill_watchdog)
276 {
277         struct net_device *netdev = adapter->netdev;
278
279         /* prevent the interrupt handler from restarting watchdog */
280         set_bit(__IXGB_DOWN, &adapter->flags);
281
282         napi_disable(&adapter->napi);
283         /* waiting for NAPI to complete can re-enable interrupts */
284         ixgb_irq_disable(adapter);
285         free_irq(adapter->pdev->irq, netdev);
286
287         if (adapter->have_msi)
288                 pci_disable_msi(adapter->pdev);
289
290         if (kill_watchdog)
291                 del_timer_sync(&adapter->watchdog_timer);
292
293         adapter->link_speed = 0;
294         adapter->link_duplex = 0;
295         netif_carrier_off(netdev);
296         netif_stop_queue(netdev);
297
298         ixgb_reset(adapter);
299         ixgb_clean_tx_ring(adapter);
300         ixgb_clean_rx_ring(adapter);
301 }
302
303 void
304 ixgb_reset(struct ixgb_adapter *adapter)
305 {
306         struct ixgb_hw *hw = &adapter->hw;
307
308         ixgb_adapter_stop(hw);
309         if (!ixgb_init_hw(hw))
310                 DPRINTK(PROBE, ERR, "ixgb_init_hw failed.\n");
311
312         /* restore frame size information */
313         IXGB_WRITE_REG(hw, MFS, hw->max_frame_size << IXGB_MFS_SHIFT);
314         if (hw->max_frame_size >
315             IXGB_MAX_ENET_FRAME_SIZE_WITHOUT_FCS + ENET_FCS_LENGTH) {
316                 u32 ctrl0 = IXGB_READ_REG(hw, CTRL0);
317                 if (!(ctrl0 & IXGB_CTRL0_JFE)) {
318                         ctrl0 |= IXGB_CTRL0_JFE;
319                         IXGB_WRITE_REG(hw, CTRL0, ctrl0);
320                 }
321         }
322 }
323
324 /**
325  * ixgb_probe - Device Initialization Routine
326  * @pdev: PCI device information struct
327  * @ent: entry in ixgb_pci_tbl
328  *
329  * Returns 0 on success, negative on failure
330  *
331  * ixgb_probe initializes an adapter identified by a pci_dev structure.
332  * The OS initialization, configuring of the adapter private structure,
333  * and a hardware reset occur.
334  **/
335
336 static int __devinit
337 ixgb_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
338 {
339         struct net_device *netdev = NULL;
340         struct ixgb_adapter *adapter;
341         static int cards_found = 0;
342         int pci_using_dac;
343         int i;
344         int err;
345
346         err = pci_enable_device(pdev);
347         if (err)
348                 return err;
349
350         if (!(err = pci_set_dma_mask(pdev, DMA_64BIT_MASK)) &&
351             !(err = pci_set_consistent_dma_mask(pdev, DMA_64BIT_MASK))) {
352                 pci_using_dac = 1;
353         } else {
354                 if ((err = pci_set_dma_mask(pdev, DMA_32BIT_MASK)) ||
355                     (err = pci_set_consistent_dma_mask(pdev, DMA_32BIT_MASK))) {
356                         printk(KERN_ERR
357                          "ixgb: No usable DMA configuration, aborting\n");
358                         goto err_dma_mask;
359                 }
360                 pci_using_dac = 0;
361         }
362
363         err = pci_request_regions(pdev, ixgb_driver_name);
364         if (err)
365                 goto err_request_regions;
366
367         pci_set_master(pdev);
368
369         netdev = alloc_etherdev(sizeof(struct ixgb_adapter));
370         if (!netdev) {
371                 err = -ENOMEM;
372                 goto err_alloc_etherdev;
373         }
374
375         SET_NETDEV_DEV(netdev, &pdev->dev);
376
377         pci_set_drvdata(pdev, netdev);
378         adapter = netdev_priv(netdev);
379         adapter->netdev = netdev;
380         adapter->pdev = pdev;
381         adapter->hw.back = adapter;
382         adapter->msg_enable = netif_msg_init(debug, DEFAULT_DEBUG_LEVEL_SHIFT);
383
384         adapter->hw.hw_addr = ioremap(pci_resource_start(pdev, BAR_0),
385                                       pci_resource_len(pdev, BAR_0));
386         if (!adapter->hw.hw_addr) {
387                 err = -EIO;
388                 goto err_ioremap;
389         }
390
391         for (i = BAR_1; i <= BAR_5; i++) {
392                 if (pci_resource_len(pdev, i) == 0)
393                         continue;
394                 if (pci_resource_flags(pdev, i) & IORESOURCE_IO) {
395                         adapter->hw.io_base = pci_resource_start(pdev, i);
396                         break;
397                 }
398         }
399
400         netdev->open = &ixgb_open;
401         netdev->stop = &ixgb_close;
402         netdev->hard_start_xmit = &ixgb_xmit_frame;
403         netdev->get_stats = &ixgb_get_stats;
404         netdev->set_multicast_list = &ixgb_set_multi;
405         netdev->set_mac_address = &ixgb_set_mac;
406         netdev->change_mtu = &ixgb_change_mtu;
407         ixgb_set_ethtool_ops(netdev);
408         netdev->tx_timeout = &ixgb_tx_timeout;
409         netdev->watchdog_timeo = 5 * HZ;
410         netif_napi_add(netdev, &adapter->napi, ixgb_clean, 64);
411         netdev->vlan_rx_register = ixgb_vlan_rx_register;
412         netdev->vlan_rx_add_vid = ixgb_vlan_rx_add_vid;
413         netdev->vlan_rx_kill_vid = ixgb_vlan_rx_kill_vid;
414 #ifdef CONFIG_NET_POLL_CONTROLLER
415         netdev->poll_controller = ixgb_netpoll;
416 #endif
417
418         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
419
420         adapter->bd_number = cards_found;
421         adapter->link_speed = 0;
422         adapter->link_duplex = 0;
423
424         /* setup the private structure */
425
426         err = ixgb_sw_init(adapter);
427         if (err)
428                 goto err_sw_init;
429
430         netdev->features = NETIF_F_SG |
431                            NETIF_F_HW_CSUM |
432                            NETIF_F_HW_VLAN_TX |
433                            NETIF_F_HW_VLAN_RX |
434                            NETIF_F_HW_VLAN_FILTER;
435         netdev->features |= NETIF_F_TSO;
436
437         if (pci_using_dac)
438                 netdev->features |= NETIF_F_HIGHDMA;
439
440         /* make sure the EEPROM is good */
441
442         if (!ixgb_validate_eeprom_checksum(&adapter->hw)) {
443                 DPRINTK(PROBE, ERR, "The EEPROM Checksum Is Not Valid\n");
444                 err = -EIO;
445                 goto err_eeprom;
446         }
447
448         ixgb_get_ee_mac_addr(&adapter->hw, netdev->dev_addr);
449         memcpy(netdev->perm_addr, netdev->dev_addr, netdev->addr_len);
450
451         if (!is_valid_ether_addr(netdev->perm_addr)) {
452                 DPRINTK(PROBE, ERR, "Invalid MAC Address\n");
453                 err = -EIO;
454                 goto err_eeprom;
455         }
456
457         adapter->part_num = ixgb_get_ee_pba_number(&adapter->hw);
458
459         init_timer(&adapter->watchdog_timer);
460         adapter->watchdog_timer.function = &ixgb_watchdog;
461         adapter->watchdog_timer.data = (unsigned long)adapter;
462
463         INIT_WORK(&adapter->tx_timeout_task, ixgb_tx_timeout_task);
464
465         strcpy(netdev->name, "eth%d");
466         err = register_netdev(netdev);
467         if (err)
468                 goto err_register;
469
470         /* we're going to reset, so assume we have no link for now */
471
472         netif_carrier_off(netdev);
473         netif_stop_queue(netdev);
474
475         DPRINTK(PROBE, INFO, "Intel(R) PRO/10GbE Network Connection\n");
476         ixgb_check_options(adapter);
477         /* reset the hardware with the new settings */
478
479         ixgb_reset(adapter);
480
481         cards_found++;
482         return 0;
483
484 err_register:
485 err_sw_init:
486 err_eeprom:
487         iounmap(adapter->hw.hw_addr);
488 err_ioremap:
489         free_netdev(netdev);
490 err_alloc_etherdev:
491         pci_release_regions(pdev);
492 err_request_regions:
493 err_dma_mask:
494         pci_disable_device(pdev);
495         return err;
496 }
497
498 /**
499  * ixgb_remove - Device Removal Routine
500  * @pdev: PCI device information struct
501  *
502  * ixgb_remove is called by the PCI subsystem to alert the driver
503  * that it should release a PCI device.  The could be caused by a
504  * Hot-Plug event, or because the driver is going to be removed from
505  * memory.
506  **/
507
508 static void __devexit
509 ixgb_remove(struct pci_dev *pdev)
510 {
511         struct net_device *netdev = pci_get_drvdata(pdev);
512         struct ixgb_adapter *adapter = netdev_priv(netdev);
513
514         flush_scheduled_work();
515
516         unregister_netdev(netdev);
517
518         iounmap(adapter->hw.hw_addr);
519         pci_release_regions(pdev);
520
521         free_netdev(netdev);
522 }
523
524 /**
525  * ixgb_sw_init - Initialize general software structures (struct ixgb_adapter)
526  * @adapter: board private structure to initialize
527  *
528  * ixgb_sw_init initializes the Adapter private data structure.
529  * Fields are initialized based on PCI device information and
530  * OS network device settings (MTU size).
531  **/
532
533 static int __devinit
534 ixgb_sw_init(struct ixgb_adapter *adapter)
535 {
536         struct ixgb_hw *hw = &adapter->hw;
537         struct net_device *netdev = adapter->netdev;
538         struct pci_dev *pdev = adapter->pdev;
539
540         /* PCI config space info */
541
542         hw->vendor_id = pdev->vendor;
543         hw->device_id = pdev->device;
544         hw->subsystem_vendor_id = pdev->subsystem_vendor;
545         hw->subsystem_id = pdev->subsystem_device;
546
547         hw->max_frame_size = netdev->mtu + ENET_HEADER_SIZE + ENET_FCS_LENGTH;
548         adapter->rx_buffer_len = hw->max_frame_size + 8; /* + 8 for errata */
549
550         if ((hw->device_id == IXGB_DEVICE_ID_82597EX)
551            || (hw->device_id == IXGB_DEVICE_ID_82597EX_CX4)
552            || (hw->device_id == IXGB_DEVICE_ID_82597EX_LR)
553            || (hw->device_id == IXGB_DEVICE_ID_82597EX_SR))
554                 hw->mac_type = ixgb_82597;
555         else {
556                 /* should never have loaded on this device */
557                 DPRINTK(PROBE, ERR, "unsupported device id\n");
558         }
559
560         /* enable flow control to be programmed */
561         hw->fc.send_xon = 1;
562
563         set_bit(__IXGB_DOWN, &adapter->flags);
564         return 0;
565 }
566
567 /**
568  * ixgb_open - Called when a network interface is made active
569  * @netdev: network interface device structure
570  *
571  * Returns 0 on success, negative value on failure
572  *
573  * The open entry point is called when a network interface is made
574  * active by the system (IFF_UP).  At this point all resources needed
575  * for transmit and receive operations are allocated, the interrupt
576  * handler is registered with the OS, the watchdog timer is started,
577  * and the stack is notified that the interface is ready.
578  **/
579
580 static int
581 ixgb_open(struct net_device *netdev)
582 {
583         struct ixgb_adapter *adapter = netdev_priv(netdev);
584         int err;
585
586         /* allocate transmit descriptors */
587         err = ixgb_setup_tx_resources(adapter);
588         if (err)
589                 goto err_setup_tx;
590
591         /* allocate receive descriptors */
592
593         err = ixgb_setup_rx_resources(adapter);
594         if (err)
595                 goto err_setup_rx;
596
597         err = ixgb_up(adapter);
598         if (err)
599                 goto err_up;
600
601         return 0;
602
603 err_up:
604         ixgb_free_rx_resources(adapter);
605 err_setup_rx:
606         ixgb_free_tx_resources(adapter);
607 err_setup_tx:
608         ixgb_reset(adapter);
609
610         return err;
611 }
612
613 /**
614  * ixgb_close - Disables a network interface
615  * @netdev: network interface device structure
616  *
617  * Returns 0, this is not allowed to fail
618  *
619  * The close entry point is called when an interface is de-activated
620  * by the OS.  The hardware is still under the drivers control, but
621  * needs to be disabled.  A global MAC reset is issued to stop the
622  * hardware, and all transmit and receive resources are freed.
623  **/
624
625 static int
626 ixgb_close(struct net_device *netdev)
627 {
628         struct ixgb_adapter *adapter = netdev_priv(netdev);
629
630         ixgb_down(adapter, true);
631
632         ixgb_free_tx_resources(adapter);
633         ixgb_free_rx_resources(adapter);
634
635         return 0;
636 }
637
638 /**
639  * ixgb_setup_tx_resources - allocate Tx resources (Descriptors)
640  * @adapter: board private structure
641  *
642  * Return 0 on success, negative on failure
643  **/
644
645 int
646 ixgb_setup_tx_resources(struct ixgb_adapter *adapter)
647 {
648         struct ixgb_desc_ring *txdr = &adapter->tx_ring;
649         struct pci_dev *pdev = adapter->pdev;
650         int size;
651
652         size = sizeof(struct ixgb_buffer) * txdr->count;
653         txdr->buffer_info = vmalloc(size);
654         if (!txdr->buffer_info) {
655                 DPRINTK(PROBE, ERR,
656                  "Unable to allocate transmit descriptor ring memory\n");
657                 return -ENOMEM;
658         }
659         memset(txdr->buffer_info, 0, size);
660
661         /* round up to nearest 4K */
662
663         txdr->size = txdr->count * sizeof(struct ixgb_tx_desc);
664         txdr->size = ALIGN(txdr->size, 4096);
665
666         txdr->desc = pci_alloc_consistent(pdev, txdr->size, &txdr->dma);
667         if (!txdr->desc) {
668                 vfree(txdr->buffer_info);
669                 DPRINTK(PROBE, ERR,
670                  "Unable to allocate transmit descriptor memory\n");
671                 return -ENOMEM;
672         }
673         memset(txdr->desc, 0, txdr->size);
674
675         txdr->next_to_use = 0;
676         txdr->next_to_clean = 0;
677
678         return 0;
679 }
680
681 /**
682  * ixgb_configure_tx - Configure 82597 Transmit Unit after Reset.
683  * @adapter: board private structure
684  *
685  * Configure the Tx unit of the MAC after a reset.
686  **/
687
688 static void
689 ixgb_configure_tx(struct ixgb_adapter *adapter)
690 {
691         u64 tdba = adapter->tx_ring.dma;
692         u32 tdlen = adapter->tx_ring.count * sizeof(struct ixgb_tx_desc);
693         u32 tctl;
694         struct ixgb_hw *hw = &adapter->hw;
695
696         /* Setup the Base and Length of the Tx Descriptor Ring
697          * tx_ring.dma can be either a 32 or 64 bit value
698          */
699
700         IXGB_WRITE_REG(hw, TDBAL, (tdba & 0x00000000ffffffffULL));
701         IXGB_WRITE_REG(hw, TDBAH, (tdba >> 32));
702
703         IXGB_WRITE_REG(hw, TDLEN, tdlen);
704
705         /* Setup the HW Tx Head and Tail descriptor pointers */
706
707         IXGB_WRITE_REG(hw, TDH, 0);
708         IXGB_WRITE_REG(hw, TDT, 0);
709
710         /* don't set up txdctl, it induces performance problems if configured
711          * incorrectly */
712         /* Set the Tx Interrupt Delay register */
713
714         IXGB_WRITE_REG(hw, TIDV, adapter->tx_int_delay);
715
716         /* Program the Transmit Control Register */
717
718         tctl = IXGB_TCTL_TCE | IXGB_TCTL_TXEN | IXGB_TCTL_TPDE;
719         IXGB_WRITE_REG(hw, TCTL, tctl);
720
721         /* Setup Transmit Descriptor Settings for this adapter */
722         adapter->tx_cmd_type =
723                 IXGB_TX_DESC_TYPE |
724                 (adapter->tx_int_delay_enable ? IXGB_TX_DESC_CMD_IDE : 0);
725 }
726
727 /**
728  * ixgb_setup_rx_resources - allocate Rx resources (Descriptors)
729  * @adapter: board private structure
730  *
731  * Returns 0 on success, negative on failure
732  **/
733
734 int
735 ixgb_setup_rx_resources(struct ixgb_adapter *adapter)
736 {
737         struct ixgb_desc_ring *rxdr = &adapter->rx_ring;
738         struct pci_dev *pdev = adapter->pdev;
739         int size;
740
741         size = sizeof(struct ixgb_buffer) * rxdr->count;
742         rxdr->buffer_info = vmalloc(size);
743         if (!rxdr->buffer_info) {
744                 DPRINTK(PROBE, ERR,
745                  "Unable to allocate receive descriptor ring\n");
746                 return -ENOMEM;
747         }
748         memset(rxdr->buffer_info, 0, size);
749
750         /* Round up to nearest 4K */
751
752         rxdr->size = rxdr->count * sizeof(struct ixgb_rx_desc);
753         rxdr->size = ALIGN(rxdr->size, 4096);
754
755         rxdr->desc = pci_alloc_consistent(pdev, rxdr->size, &rxdr->dma);
756
757         if (!rxdr->desc) {
758                 vfree(rxdr->buffer_info);
759                 DPRINTK(PROBE, ERR,
760                  "Unable to allocate receive descriptors\n");
761                 return -ENOMEM;
762         }
763         memset(rxdr->desc, 0, rxdr->size);
764
765         rxdr->next_to_clean = 0;
766         rxdr->next_to_use = 0;
767
768         return 0;
769 }
770
771 /**
772  * ixgb_setup_rctl - configure the receive control register
773  * @adapter: Board private structure
774  **/
775
776 static void
777 ixgb_setup_rctl(struct ixgb_adapter *adapter)
778 {
779         u32 rctl;
780
781         rctl = IXGB_READ_REG(&adapter->hw, RCTL);
782
783         rctl &= ~(3 << IXGB_RCTL_MO_SHIFT);
784
785         rctl |=
786                 IXGB_RCTL_BAM | IXGB_RCTL_RDMTS_1_2 |
787                 IXGB_RCTL_RXEN | IXGB_RCTL_CFF |
788                 (adapter->hw.mc_filter_type << IXGB_RCTL_MO_SHIFT);
789
790         rctl |= IXGB_RCTL_SECRC;
791
792         if (adapter->rx_buffer_len <= IXGB_RXBUFFER_2048)
793                 rctl |= IXGB_RCTL_BSIZE_2048;
794         else if (adapter->rx_buffer_len <= IXGB_RXBUFFER_4096)
795                 rctl |= IXGB_RCTL_BSIZE_4096;
796         else if (adapter->rx_buffer_len <= IXGB_RXBUFFER_8192)
797                 rctl |= IXGB_RCTL_BSIZE_8192;
798         else if (adapter->rx_buffer_len <= IXGB_RXBUFFER_16384)
799                 rctl |= IXGB_RCTL_BSIZE_16384;
800
801         IXGB_WRITE_REG(&adapter->hw, RCTL, rctl);
802 }
803
804 /**
805  * ixgb_configure_rx - Configure 82597 Receive Unit after Reset.
806  * @adapter: board private structure
807  *
808  * Configure the Rx unit of the MAC after a reset.
809  **/
810
811 static void
812 ixgb_configure_rx(struct ixgb_adapter *adapter)
813 {
814         u64 rdba = adapter->rx_ring.dma;
815         u32 rdlen = adapter->rx_ring.count * sizeof(struct ixgb_rx_desc);
816         struct ixgb_hw *hw = &adapter->hw;
817         u32 rctl;
818         u32 rxcsum;
819
820         /* make sure receives are disabled while setting up the descriptors */
821
822         rctl = IXGB_READ_REG(hw, RCTL);
823         IXGB_WRITE_REG(hw, RCTL, rctl & ~IXGB_RCTL_RXEN);
824
825         /* set the Receive Delay Timer Register */
826
827         IXGB_WRITE_REG(hw, RDTR, adapter->rx_int_delay);
828
829         /* Setup the Base and Length of the Rx Descriptor Ring */
830
831         IXGB_WRITE_REG(hw, RDBAL, (rdba & 0x00000000ffffffffULL));
832         IXGB_WRITE_REG(hw, RDBAH, (rdba >> 32));
833
834         IXGB_WRITE_REG(hw, RDLEN, rdlen);
835
836         /* Setup the HW Rx Head and Tail Descriptor Pointers */
837         IXGB_WRITE_REG(hw, RDH, 0);
838         IXGB_WRITE_REG(hw, RDT, 0);
839
840         /* due to the hardware errata with RXDCTL, we are unable to use any of
841          * the performance enhancing features of it without causing other
842          * subtle bugs, some of the bugs could include receive length
843          * corruption at high data rates (WTHRESH > 0) and/or receive
844          * descriptor ring irregularites (particularly in hardware cache) */
845         IXGB_WRITE_REG(hw, RXDCTL, 0);
846
847         /* Enable Receive Checksum Offload for TCP and UDP */
848         if (adapter->rx_csum) {
849                 rxcsum = IXGB_READ_REG(hw, RXCSUM);
850                 rxcsum |= IXGB_RXCSUM_TUOFL;
851                 IXGB_WRITE_REG(hw, RXCSUM, rxcsum);
852         }
853
854         /* Enable Receives */
855
856         IXGB_WRITE_REG(hw, RCTL, rctl);
857 }
858
859 /**
860  * ixgb_free_tx_resources - Free Tx Resources
861  * @adapter: board private structure
862  *
863  * Free all transmit software resources
864  **/
865
866 void
867 ixgb_free_tx_resources(struct ixgb_adapter *adapter)
868 {
869         struct pci_dev *pdev = adapter->pdev;
870
871         ixgb_clean_tx_ring(adapter);
872
873         vfree(adapter->tx_ring.buffer_info);
874         adapter->tx_ring.buffer_info = NULL;
875
876         pci_free_consistent(pdev, adapter->tx_ring.size,
877                             adapter->tx_ring.desc, adapter->tx_ring.dma);
878
879         adapter->tx_ring.desc = NULL;
880 }
881
882 static void
883 ixgb_unmap_and_free_tx_resource(struct ixgb_adapter *adapter,
884                                 struct ixgb_buffer *buffer_info)
885 {
886         struct pci_dev *pdev = adapter->pdev;
887
888         if (buffer_info->dma)
889                 pci_unmap_page(pdev, buffer_info->dma, buffer_info->length,
890                                PCI_DMA_TODEVICE);
891
892         /* okay to call kfree_skb here instead of kfree_skb_any because
893          * this is never called in interrupt context */
894         if (buffer_info->skb)
895                 dev_kfree_skb(buffer_info->skb);
896
897         buffer_info->skb = NULL;
898         buffer_info->dma = 0;
899         buffer_info->time_stamp = 0;
900         /* these fields must always be initialized in tx
901          * buffer_info->length = 0;
902          * buffer_info->next_to_watch = 0; */
903 }
904
905 /**
906  * ixgb_clean_tx_ring - Free Tx Buffers
907  * @adapter: board private structure
908  **/
909
910 static void
911 ixgb_clean_tx_ring(struct ixgb_adapter *adapter)
912 {
913         struct ixgb_desc_ring *tx_ring = &adapter->tx_ring;
914         struct ixgb_buffer *buffer_info;
915         unsigned long size;
916         unsigned int i;
917
918         /* Free all the Tx ring sk_buffs */
919
920         for (i = 0; i < tx_ring->count; i++) {
921                 buffer_info = &tx_ring->buffer_info[i];
922                 ixgb_unmap_and_free_tx_resource(adapter, buffer_info);
923         }
924
925         size = sizeof(struct ixgb_buffer) * tx_ring->count;
926         memset(tx_ring->buffer_info, 0, size);
927
928         /* Zero out the descriptor ring */
929
930         memset(tx_ring->desc, 0, tx_ring->size);
931
932         tx_ring->next_to_use = 0;
933         tx_ring->next_to_clean = 0;
934
935         IXGB_WRITE_REG(&adapter->hw, TDH, 0);
936         IXGB_WRITE_REG(&adapter->hw, TDT, 0);
937 }
938
939 /**
940  * ixgb_free_rx_resources - Free Rx Resources
941  * @adapter: board private structure
942  *
943  * Free all receive software resources
944  **/
945
946 void
947 ixgb_free_rx_resources(struct ixgb_adapter *adapter)
948 {
949         struct ixgb_desc_ring *rx_ring = &adapter->rx_ring;
950         struct pci_dev *pdev = adapter->pdev;
951
952         ixgb_clean_rx_ring(adapter);
953
954         vfree(rx_ring->buffer_info);
955         rx_ring->buffer_info = NULL;
956
957         pci_free_consistent(pdev, rx_ring->size, rx_ring->desc, rx_ring->dma);
958
959         rx_ring->desc = NULL;
960 }
961
962 /**
963  * ixgb_clean_rx_ring - Free Rx Buffers
964  * @adapter: board private structure
965  **/
966
967 static void
968 ixgb_clean_rx_ring(struct ixgb_adapter *adapter)
969 {
970         struct ixgb_desc_ring *rx_ring = &adapter->rx_ring;
971         struct ixgb_buffer *buffer_info;
972         struct pci_dev *pdev = adapter->pdev;
973         unsigned long size;
974         unsigned int i;
975
976         /* Free all the Rx ring sk_buffs */
977
978         for (i = 0; i < rx_ring->count; i++) {
979                 buffer_info = &rx_ring->buffer_info[i];
980                 if (buffer_info->skb) {
981
982                         pci_unmap_single(pdev,
983                                          buffer_info->dma,
984                                          buffer_info->length,
985                                          PCI_DMA_FROMDEVICE);
986
987                         dev_kfree_skb(buffer_info->skb);
988
989                         buffer_info->skb = NULL;
990                 }
991         }
992
993         size = sizeof(struct ixgb_buffer) * rx_ring->count;
994         memset(rx_ring->buffer_info, 0, size);
995
996         /* Zero out the descriptor ring */
997
998         memset(rx_ring->desc, 0, rx_ring->size);
999
1000         rx_ring->next_to_clean = 0;
1001         rx_ring->next_to_use = 0;
1002
1003         IXGB_WRITE_REG(&adapter->hw, RDH, 0);
1004         IXGB_WRITE_REG(&adapter->hw, RDT, 0);
1005 }
1006
1007 /**
1008  * ixgb_set_mac - Change the Ethernet Address of the NIC
1009  * @netdev: network interface device structure
1010  * @p: pointer to an address structure
1011  *
1012  * Returns 0 on success, negative on failure
1013  **/
1014
1015 static int
1016 ixgb_set_mac(struct net_device *netdev, void *p)
1017 {
1018         struct ixgb_adapter *adapter = netdev_priv(netdev);
1019         struct sockaddr *addr = p;
1020
1021         if (!is_valid_ether_addr(addr->sa_data))
1022                 return -EADDRNOTAVAIL;
1023
1024         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
1025
1026         ixgb_rar_set(&adapter->hw, addr->sa_data, 0);
1027
1028         return 0;
1029 }
1030
1031 /**
1032  * ixgb_set_multi - Multicast and Promiscuous mode set
1033  * @netdev: network interface device structure
1034  *
1035  * The set_multi entry point is called whenever the multicast address
1036  * list or the network interface flags are updated.  This routine is
1037  * responsible for configuring the hardware for proper multicast,
1038  * promiscuous mode, and all-multi behavior.
1039  **/
1040
1041 static void
1042 ixgb_set_multi(struct net_device *netdev)
1043 {
1044         struct ixgb_adapter *adapter = netdev_priv(netdev);
1045         struct ixgb_hw *hw = &adapter->hw;
1046         struct dev_mc_list *mc_ptr;
1047         u32 rctl;
1048         int i;
1049
1050         /* Check for Promiscuous and All Multicast modes */
1051
1052         rctl = IXGB_READ_REG(hw, RCTL);
1053
1054         if (netdev->flags & IFF_PROMISC) {
1055                 rctl |= (IXGB_RCTL_UPE | IXGB_RCTL_MPE);
1056                 rctl &= ~IXGB_RCTL_VFE;
1057         } else {
1058                 if (netdev->flags & IFF_ALLMULTI) {
1059                         rctl |= IXGB_RCTL_MPE;
1060                         rctl &= ~IXGB_RCTL_UPE;
1061                 } else {
1062                         rctl &= ~(IXGB_RCTL_UPE | IXGB_RCTL_MPE);
1063                 }
1064                 rctl |= IXGB_RCTL_VFE;
1065         }
1066
1067         if (netdev->mc_count > IXGB_MAX_NUM_MULTICAST_ADDRESSES) {
1068                 rctl |= IXGB_RCTL_MPE;
1069                 IXGB_WRITE_REG(hw, RCTL, rctl);
1070         } else {
1071                 u8 mta[IXGB_MAX_NUM_MULTICAST_ADDRESSES *
1072                             IXGB_ETH_LENGTH_OF_ADDRESS];
1073
1074                 IXGB_WRITE_REG(hw, RCTL, rctl);
1075
1076                 for (i = 0, mc_ptr = netdev->mc_list;
1077                      mc_ptr;
1078                      i++, mc_ptr = mc_ptr->next)
1079                         memcpy(&mta[i * IXGB_ETH_LENGTH_OF_ADDRESS],
1080                                mc_ptr->dmi_addr, IXGB_ETH_LENGTH_OF_ADDRESS);
1081
1082                 ixgb_mc_addr_list_update(hw, mta, netdev->mc_count, 0);
1083         }
1084 }
1085
1086 /**
1087  * ixgb_watchdog - Timer Call-back
1088  * @data: pointer to netdev cast into an unsigned long
1089  **/
1090
1091 static void
1092 ixgb_watchdog(unsigned long data)
1093 {
1094         struct ixgb_adapter *adapter = (struct ixgb_adapter *)data;
1095         struct net_device *netdev = adapter->netdev;
1096         struct ixgb_desc_ring *txdr = &adapter->tx_ring;
1097
1098         ixgb_check_for_link(&adapter->hw);
1099
1100         if (ixgb_check_for_bad_link(&adapter->hw)) {
1101                 /* force the reset path */
1102                 netif_stop_queue(netdev);
1103         }
1104
1105         if (adapter->hw.link_up) {
1106                 if (!netif_carrier_ok(netdev)) {
1107                         DPRINTK(LINK, INFO,
1108                                 "NIC Link is Up 10000 Mbps Full Duplex\n");
1109                         adapter->link_speed = 10000;
1110                         adapter->link_duplex = FULL_DUPLEX;
1111                         netif_carrier_on(netdev);
1112                         netif_wake_queue(netdev);
1113                 }
1114         } else {
1115                 if (netif_carrier_ok(netdev)) {
1116                         adapter->link_speed = 0;
1117                         adapter->link_duplex = 0;
1118                         DPRINTK(LINK, INFO, "NIC Link is Down\n");
1119                         netif_carrier_off(netdev);
1120                         netif_stop_queue(netdev);
1121
1122                 }
1123         }
1124
1125         ixgb_update_stats(adapter);
1126
1127         if (!netif_carrier_ok(netdev)) {
1128                 if (IXGB_DESC_UNUSED(txdr) + 1 < txdr->count) {
1129                         /* We've lost link, so the controller stops DMA,
1130                          * but we've got queued Tx work that's never going
1131                          * to get done, so reset controller to flush Tx.
1132                          * (Do the reset outside of interrupt context). */
1133                         schedule_work(&adapter->tx_timeout_task);
1134                 }
1135         }
1136
1137         /* Force detection of hung controller every watchdog period */
1138         adapter->detect_tx_hung = true;
1139
1140         /* generate an interrupt to force clean up of any stragglers */
1141         IXGB_WRITE_REG(&adapter->hw, ICS, IXGB_INT_TXDW);
1142
1143         /* Reset the timer */
1144         mod_timer(&adapter->watchdog_timer, jiffies + 2 * HZ);
1145 }
1146
1147 #define IXGB_TX_FLAGS_CSUM              0x00000001
1148 #define IXGB_TX_FLAGS_VLAN              0x00000002
1149 #define IXGB_TX_FLAGS_TSO               0x00000004
1150
1151 static int
1152 ixgb_tso(struct ixgb_adapter *adapter, struct sk_buff *skb)
1153 {
1154         struct ixgb_context_desc *context_desc;
1155         unsigned int i;
1156         u8 ipcss, ipcso, tucss, tucso, hdr_len;
1157         u16 ipcse, tucse, mss;
1158         int err;
1159
1160         if (likely(skb_is_gso(skb))) {
1161                 struct ixgb_buffer *buffer_info;
1162                 struct iphdr *iph;
1163
1164                 if (skb_header_cloned(skb)) {
1165                         err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1166                         if (err)
1167                                 return err;
1168                 }
1169
1170                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
1171                 mss = skb_shinfo(skb)->gso_size;
1172                 iph = ip_hdr(skb);
1173                 iph->tot_len = 0;
1174                 iph->check = 0;
1175                 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
1176                                                          iph->daddr, 0,
1177                                                          IPPROTO_TCP, 0);
1178                 ipcss = skb_network_offset(skb);
1179                 ipcso = (void *)&(iph->check) - (void *)skb->data;
1180                 ipcse = skb_transport_offset(skb) - 1;
1181                 tucss = skb_transport_offset(skb);
1182                 tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
1183                 tucse = 0;
1184
1185                 i = adapter->tx_ring.next_to_use;
1186                 context_desc = IXGB_CONTEXT_DESC(adapter->tx_ring, i);
1187                 buffer_info = &adapter->tx_ring.buffer_info[i];
1188                 WARN_ON(buffer_info->dma != 0);
1189
1190                 context_desc->ipcss = ipcss;
1191                 context_desc->ipcso = ipcso;
1192                 context_desc->ipcse = cpu_to_le16(ipcse);
1193                 context_desc->tucss = tucss;
1194                 context_desc->tucso = tucso;
1195                 context_desc->tucse = cpu_to_le16(tucse);
1196                 context_desc->mss = cpu_to_le16(mss);
1197                 context_desc->hdr_len = hdr_len;
1198                 context_desc->status = 0;
1199                 context_desc->cmd_type_len = cpu_to_le32(
1200                                                   IXGB_CONTEXT_DESC_TYPE
1201                                                 | IXGB_CONTEXT_DESC_CMD_TSE
1202                                                 | IXGB_CONTEXT_DESC_CMD_IP
1203                                                 | IXGB_CONTEXT_DESC_CMD_TCP
1204                                                 | IXGB_CONTEXT_DESC_CMD_IDE
1205                                                 | (skb->len - (hdr_len)));
1206
1207
1208                 if (++i == adapter->tx_ring.count) i = 0;
1209                 adapter->tx_ring.next_to_use = i;
1210
1211                 return 1;
1212         }
1213
1214         return 0;
1215 }
1216
1217 static bool
1218 ixgb_tx_csum(struct ixgb_adapter *adapter, struct sk_buff *skb)
1219 {
1220         struct ixgb_context_desc *context_desc;
1221         unsigned int i;
1222         u8 css, cso;
1223
1224         if (likely(skb->ip_summed == CHECKSUM_PARTIAL)) {
1225                 struct ixgb_buffer *buffer_info;
1226                 css = skb_transport_offset(skb);
1227                 cso = css + skb->csum_offset;
1228
1229                 i = adapter->tx_ring.next_to_use;
1230                 context_desc = IXGB_CONTEXT_DESC(adapter->tx_ring, i);
1231                 buffer_info = &adapter->tx_ring.buffer_info[i];
1232                 WARN_ON(buffer_info->dma != 0);
1233
1234                 context_desc->tucss = css;
1235                 context_desc->tucso = cso;
1236                 context_desc->tucse = 0;
1237                 /* zero out any previously existing data in one instruction */
1238                 *(u32 *)&(context_desc->ipcss) = 0;
1239                 context_desc->status = 0;
1240                 context_desc->hdr_len = 0;
1241                 context_desc->mss = 0;
1242                 context_desc->cmd_type_len =
1243                         cpu_to_le32(IXGB_CONTEXT_DESC_TYPE
1244                                     | IXGB_TX_DESC_CMD_IDE);
1245
1246                 if (++i == adapter->tx_ring.count) i = 0;
1247                 adapter->tx_ring.next_to_use = i;
1248
1249                 return true;
1250         }
1251
1252         return false;
1253 }
1254
1255 #define IXGB_MAX_TXD_PWR        14
1256 #define IXGB_MAX_DATA_PER_TXD   (1<<IXGB_MAX_TXD_PWR)
1257
1258 static int
1259 ixgb_tx_map(struct ixgb_adapter *adapter, struct sk_buff *skb,
1260             unsigned int first)
1261 {
1262         struct ixgb_desc_ring *tx_ring = &adapter->tx_ring;
1263         struct ixgb_buffer *buffer_info;
1264         int len = skb->len;
1265         unsigned int offset = 0, size, count = 0, i;
1266         unsigned int mss = skb_shinfo(skb)->gso_size;
1267
1268         unsigned int nr_frags = skb_shinfo(skb)->nr_frags;
1269         unsigned int f;
1270
1271         len -= skb->data_len;
1272
1273         i = tx_ring->next_to_use;
1274
1275         while (len) {
1276                 buffer_info = &tx_ring->buffer_info[i];
1277                 size = min(len, IXGB_MAX_DATA_PER_TXD);
1278                 /* Workaround for premature desc write-backs
1279                  * in TSO mode.  Append 4-byte sentinel desc */
1280                 if (unlikely(mss && !nr_frags && size == len && size > 8))
1281                         size -= 4;
1282
1283                 buffer_info->length = size;
1284                 WARN_ON(buffer_info->dma != 0);
1285                 buffer_info->time_stamp = jiffies;
1286                 buffer_info->dma =
1287                         pci_map_single(adapter->pdev,
1288                                 skb->data + offset,
1289                                 size,
1290                                 PCI_DMA_TODEVICE);
1291                 buffer_info->next_to_watch = 0;
1292
1293                 len -= size;
1294                 offset += size;
1295                 count++;
1296                 if (++i == tx_ring->count) i = 0;
1297         }
1298
1299         for (f = 0; f < nr_frags; f++) {
1300                 struct skb_frag_struct *frag;
1301
1302                 frag = &skb_shinfo(skb)->frags[f];
1303                 len = frag->size;
1304                 offset = 0;
1305
1306                 while (len) {
1307                         buffer_info = &tx_ring->buffer_info[i];
1308                         size = min(len, IXGB_MAX_DATA_PER_TXD);
1309
1310                         /* Workaround for premature desc write-backs
1311                          * in TSO mode.  Append 4-byte sentinel desc */
1312                         if (unlikely(mss && (f == (nr_frags - 1))
1313                                      && size == len && size > 8))
1314                                 size -= 4;
1315
1316                         buffer_info->length = size;
1317                         buffer_info->time_stamp = jiffies;
1318                         buffer_info->dma =
1319                                 pci_map_page(adapter->pdev,
1320                                         frag->page,
1321                                         frag->page_offset + offset,
1322                                         size,
1323                                         PCI_DMA_TODEVICE);
1324                         buffer_info->next_to_watch = 0;
1325
1326                         len -= size;
1327                         offset += size;
1328                         count++;
1329                         if (++i == tx_ring->count) i = 0;
1330                 }
1331         }
1332         i = (i == 0) ? tx_ring->count - 1 : i - 1;
1333         tx_ring->buffer_info[i].skb = skb;
1334         tx_ring->buffer_info[first].next_to_watch = i;
1335
1336         return count;
1337 }
1338
1339 static void
1340 ixgb_tx_queue(struct ixgb_adapter *adapter, int count, int vlan_id,int tx_flags)
1341 {
1342         struct ixgb_desc_ring *tx_ring = &adapter->tx_ring;
1343         struct ixgb_tx_desc *tx_desc = NULL;
1344         struct ixgb_buffer *buffer_info;
1345         u32 cmd_type_len = adapter->tx_cmd_type;
1346         u8 status = 0;
1347         u8 popts = 0;
1348         unsigned int i;
1349
1350         if (tx_flags & IXGB_TX_FLAGS_TSO) {
1351                 cmd_type_len |= IXGB_TX_DESC_CMD_TSE;
1352                 popts |= (IXGB_TX_DESC_POPTS_IXSM | IXGB_TX_DESC_POPTS_TXSM);
1353         }
1354
1355         if (tx_flags & IXGB_TX_FLAGS_CSUM)
1356                 popts |= IXGB_TX_DESC_POPTS_TXSM;
1357
1358         if (tx_flags & IXGB_TX_FLAGS_VLAN)
1359                 cmd_type_len |= IXGB_TX_DESC_CMD_VLE;
1360
1361         i = tx_ring->next_to_use;
1362
1363         while (count--) {
1364                 buffer_info = &tx_ring->buffer_info[i];
1365                 tx_desc = IXGB_TX_DESC(*tx_ring, i);
1366                 tx_desc->buff_addr = cpu_to_le64(buffer_info->dma);
1367                 tx_desc->cmd_type_len =
1368                         cpu_to_le32(cmd_type_len | buffer_info->length);
1369                 tx_desc->status = status;
1370                 tx_desc->popts = popts;
1371                 tx_desc->vlan = cpu_to_le16(vlan_id);
1372
1373                 if (++i == tx_ring->count) i = 0;
1374         }
1375
1376         tx_desc->cmd_type_len |=
1377                 cpu_to_le32(IXGB_TX_DESC_CMD_EOP | IXGB_TX_DESC_CMD_RS);
1378
1379         /* Force memory writes to complete before letting h/w
1380          * know there are new descriptors to fetch.  (Only
1381          * applicable for weak-ordered memory model archs,
1382          * such as IA-64). */
1383         wmb();
1384
1385         tx_ring->next_to_use = i;
1386         IXGB_WRITE_REG(&adapter->hw, TDT, i);
1387 }
1388
1389 static int __ixgb_maybe_stop_tx(struct net_device *netdev, int size)
1390 {
1391         struct ixgb_adapter *adapter = netdev_priv(netdev);
1392         struct ixgb_desc_ring *tx_ring = &adapter->tx_ring;
1393
1394         netif_stop_queue(netdev);
1395         /* Herbert's original patch had:
1396          *  smp_mb__after_netif_stop_queue();
1397          * but since that doesn't exist yet, just open code it. */
1398         smp_mb();
1399
1400         /* We need to check again in a case another CPU has just
1401          * made room available. */
1402         if (likely(IXGB_DESC_UNUSED(tx_ring) < size))
1403                 return -EBUSY;
1404
1405         /* A reprieve! */
1406         netif_start_queue(netdev);
1407         ++adapter->restart_queue;
1408         return 0;
1409 }
1410
1411 static int ixgb_maybe_stop_tx(struct net_device *netdev,
1412                               struct ixgb_desc_ring *tx_ring, int size)
1413 {
1414         if (likely(IXGB_DESC_UNUSED(tx_ring) >= size))
1415                 return 0;
1416         return __ixgb_maybe_stop_tx(netdev, size);
1417 }
1418
1419
1420 /* Tx Descriptors needed, worst case */
1421 #define TXD_USE_COUNT(S) (((S) >> IXGB_MAX_TXD_PWR) + \
1422                          (((S) & (IXGB_MAX_DATA_PER_TXD - 1)) ? 1 : 0))
1423 #define DESC_NEEDED TXD_USE_COUNT(IXGB_MAX_DATA_PER_TXD) /* skb->date */ + \
1424         MAX_SKB_FRAGS * TXD_USE_COUNT(PAGE_SIZE) + 1 /* for context */ \
1425         + 1 /* one more needed for sentinel TSO workaround */
1426
1427 static int
1428 ixgb_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
1429 {
1430         struct ixgb_adapter *adapter = netdev_priv(netdev);
1431         unsigned int first;
1432         unsigned int tx_flags = 0;
1433         int vlan_id = 0;
1434         int tso;
1435
1436         if (test_bit(__IXGB_DOWN, &adapter->flags)) {
1437                 dev_kfree_skb(skb);
1438                 return NETDEV_TX_OK;
1439         }
1440
1441         if (skb->len <= 0) {
1442                 dev_kfree_skb(skb);
1443                 return 0;
1444         }
1445
1446         if (unlikely(ixgb_maybe_stop_tx(netdev, &adapter->tx_ring,
1447                      DESC_NEEDED)))
1448                 return NETDEV_TX_BUSY;
1449
1450         if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
1451                 tx_flags |= IXGB_TX_FLAGS_VLAN;
1452                 vlan_id = vlan_tx_tag_get(skb);
1453         }
1454
1455         first = adapter->tx_ring.next_to_use;
1456
1457         tso = ixgb_tso(adapter, skb);
1458         if (tso < 0) {
1459                 dev_kfree_skb(skb);
1460                 return NETDEV_TX_OK;
1461         }
1462
1463         if (likely(tso))
1464                 tx_flags |= IXGB_TX_FLAGS_TSO;
1465         else if (ixgb_tx_csum(adapter, skb))
1466                 tx_flags |= IXGB_TX_FLAGS_CSUM;
1467
1468         ixgb_tx_queue(adapter, ixgb_tx_map(adapter, skb, first), vlan_id,
1469                         tx_flags);
1470
1471         netdev->trans_start = jiffies;
1472
1473         /* Make sure there is space in the ring for the next send. */
1474         ixgb_maybe_stop_tx(netdev, &adapter->tx_ring, DESC_NEEDED);
1475
1476         return NETDEV_TX_OK;
1477 }
1478
1479 /**
1480  * ixgb_tx_timeout - Respond to a Tx Hang
1481  * @netdev: network interface device structure
1482  **/
1483
1484 static void
1485 ixgb_tx_timeout(struct net_device *netdev)
1486 {
1487         struct ixgb_adapter *adapter = netdev_priv(netdev);
1488
1489         /* Do the reset outside of interrupt context */
1490         schedule_work(&adapter->tx_timeout_task);
1491 }
1492
1493 static void
1494 ixgb_tx_timeout_task(struct work_struct *work)
1495 {
1496         struct ixgb_adapter *adapter =
1497                 container_of(work, struct ixgb_adapter, tx_timeout_task);
1498
1499         adapter->tx_timeout_count++;
1500         ixgb_down(adapter, true);
1501         ixgb_up(adapter);
1502 }
1503
1504 /**
1505  * ixgb_get_stats - Get System Network Statistics
1506  * @netdev: network interface device structure
1507  *
1508  * Returns the address of the device statistics structure.
1509  * The statistics are actually updated from the timer callback.
1510  **/
1511
1512 static struct net_device_stats *
1513 ixgb_get_stats(struct net_device *netdev)
1514 {
1515         struct ixgb_adapter *adapter = netdev_priv(netdev);
1516
1517         return &adapter->net_stats;
1518 }
1519
1520 /**
1521  * ixgb_change_mtu - Change the Maximum Transfer Unit
1522  * @netdev: network interface device structure
1523  * @new_mtu: new value for maximum frame size
1524  *
1525  * Returns 0 on success, negative on failure
1526  **/
1527
1528 static int
1529 ixgb_change_mtu(struct net_device *netdev, int new_mtu)
1530 {
1531         struct ixgb_adapter *adapter = netdev_priv(netdev);
1532         int max_frame = new_mtu + ENET_HEADER_SIZE + ENET_FCS_LENGTH;
1533         int old_max_frame = netdev->mtu + ENET_HEADER_SIZE + ENET_FCS_LENGTH;
1534
1535         /* MTU < 68 is an error for IPv4 traffic, just don't allow it */
1536         if ((new_mtu < 68) ||
1537             (max_frame > IXGB_MAX_JUMBO_FRAME_SIZE + ENET_FCS_LENGTH)) {
1538                 DPRINTK(PROBE, ERR, "Invalid MTU setting %d\n", new_mtu);
1539                 return -EINVAL;
1540         }
1541
1542         if (old_max_frame == max_frame)
1543                 return 0;
1544
1545         if (netif_running(netdev))
1546                 ixgb_down(adapter, true);
1547
1548         adapter->rx_buffer_len = max_frame + 8; /* + 8 for errata */
1549
1550         netdev->mtu = new_mtu;
1551
1552         if (netif_running(netdev))
1553                 ixgb_up(adapter);
1554
1555         return 0;
1556 }
1557
1558 /**
1559  * ixgb_update_stats - Update the board statistics counters.
1560  * @adapter: board private structure
1561  **/
1562
1563 void
1564 ixgb_update_stats(struct ixgb_adapter *adapter)
1565 {
1566         struct net_device *netdev = adapter->netdev;
1567         struct pci_dev *pdev = adapter->pdev;
1568
1569         /* Prevent stats update while adapter is being reset */
1570         if (pci_channel_offline(pdev))
1571                 return;
1572
1573         if ((netdev->flags & IFF_PROMISC) || (netdev->flags & IFF_ALLMULTI) ||
1574            (netdev->mc_count > IXGB_MAX_NUM_MULTICAST_ADDRESSES)) {
1575                 u64 multi = IXGB_READ_REG(&adapter->hw, MPRCL);
1576                 u32 bcast_l = IXGB_READ_REG(&adapter->hw, BPRCL);
1577                 u32 bcast_h = IXGB_READ_REG(&adapter->hw, BPRCH);
1578                 u64 bcast = ((u64)bcast_h << 32) | bcast_l;
1579
1580                 multi |= ((u64)IXGB_READ_REG(&adapter->hw, MPRCH) << 32);
1581                 /* fix up multicast stats by removing broadcasts */
1582                 if (multi >= bcast)
1583                         multi -= bcast;
1584
1585                 adapter->stats.mprcl += (multi & 0xFFFFFFFF);
1586                 adapter->stats.mprch += (multi >> 32);
1587                 adapter->stats.bprcl += bcast_l;
1588                 adapter->stats.bprch += bcast_h;
1589         } else {
1590                 adapter->stats.mprcl += IXGB_READ_REG(&adapter->hw, MPRCL);
1591                 adapter->stats.mprch += IXGB_READ_REG(&adapter->hw, MPRCH);
1592                 adapter->stats.bprcl += IXGB_READ_REG(&adapter->hw, BPRCL);
1593                 adapter->stats.bprch += IXGB_READ_REG(&adapter->hw, BPRCH);
1594         }
1595         adapter->stats.tprl += IXGB_READ_REG(&adapter->hw, TPRL);
1596         adapter->stats.tprh += IXGB_READ_REG(&adapter->hw, TPRH);
1597         adapter->stats.gprcl += IXGB_READ_REG(&adapter->hw, GPRCL);
1598         adapter->stats.gprch += IXGB_READ_REG(&adapter->hw, GPRCH);
1599         adapter->stats.uprcl += IXGB_READ_REG(&adapter->hw, UPRCL);
1600         adapter->stats.uprch += IXGB_READ_REG(&adapter->hw, UPRCH);
1601         adapter->stats.vprcl += IXGB_READ_REG(&adapter->hw, VPRCL);
1602         adapter->stats.vprch += IXGB_READ_REG(&adapter->hw, VPRCH);
1603         adapter->stats.jprcl += IXGB_READ_REG(&adapter->hw, JPRCL);
1604         adapter->stats.jprch += IXGB_READ_REG(&adapter->hw, JPRCH);
1605         adapter->stats.gorcl += IXGB_READ_REG(&adapter->hw, GORCL);
1606         adapter->stats.gorch += IXGB_READ_REG(&adapter->hw, GORCH);
1607         adapter->stats.torl += IXGB_READ_REG(&adapter->hw, TORL);
1608         adapter->stats.torh += IXGB_READ_REG(&adapter->hw, TORH);
1609         adapter->stats.rnbc += IXGB_READ_REG(&adapter->hw, RNBC);
1610         adapter->stats.ruc += IXGB_READ_REG(&adapter->hw, RUC);
1611         adapter->stats.roc += IXGB_READ_REG(&adapter->hw, ROC);
1612         adapter->stats.rlec += IXGB_READ_REG(&adapter->hw, RLEC);
1613         adapter->stats.crcerrs += IXGB_READ_REG(&adapter->hw, CRCERRS);
1614         adapter->stats.icbc += IXGB_READ_REG(&adapter->hw, ICBC);
1615         adapter->stats.ecbc += IXGB_READ_REG(&adapter->hw, ECBC);
1616         adapter->stats.mpc += IXGB_READ_REG(&adapter->hw, MPC);
1617         adapter->stats.tptl += IXGB_READ_REG(&adapter->hw, TPTL);
1618         adapter->stats.tpth += IXGB_READ_REG(&adapter->hw, TPTH);
1619         adapter->stats.gptcl += IXGB_READ_REG(&adapter->hw, GPTCL);
1620         adapter->stats.gptch += IXGB_READ_REG(&adapter->hw, GPTCH);
1621         adapter->stats.bptcl += IXGB_READ_REG(&adapter->hw, BPTCL);
1622         adapter->stats.bptch += IXGB_READ_REG(&adapter->hw, BPTCH);
1623         adapter->stats.mptcl += IXGB_READ_REG(&adapter->hw, MPTCL);
1624         adapter->stats.mptch += IXGB_READ_REG(&adapter->hw, MPTCH);
1625         adapter->stats.uptcl += IXGB_READ_REG(&adapter->hw, UPTCL);
1626         adapter->stats.uptch += IXGB_READ_REG(&adapter->hw, UPTCH);
1627         adapter->stats.vptcl += IXGB_READ_REG(&adapter->hw, VPTCL);
1628         adapter->stats.vptch += IXGB_READ_REG(&adapter->hw, VPTCH);
1629         adapter->stats.jptcl += IXGB_READ_REG(&adapter->hw, JPTCL);
1630         adapter->stats.jptch += IXGB_READ_REG(&adapter->hw, JPTCH);
1631         adapter->stats.gotcl += IXGB_READ_REG(&adapter->hw, GOTCL);
1632         adapter->stats.gotch += IXGB_READ_REG(&adapter->hw, GOTCH);
1633         adapter->stats.totl += IXGB_READ_REG(&adapter->hw, TOTL);
1634         adapter->stats.toth += IXGB_READ_REG(&adapter->hw, TOTH);
1635         adapter->stats.dc += IXGB_READ_REG(&adapter->hw, DC);
1636         adapter->stats.plt64c += IXGB_READ_REG(&adapter->hw, PLT64C);
1637         adapter->stats.tsctc += IXGB_READ_REG(&adapter->hw, TSCTC);
1638         adapter->stats.tsctfc += IXGB_READ_REG(&adapter->hw, TSCTFC);
1639         adapter->stats.ibic += IXGB_READ_REG(&adapter->hw, IBIC);
1640         adapter->stats.rfc += IXGB_READ_REG(&adapter->hw, RFC);
1641         adapter->stats.lfc += IXGB_READ_REG(&adapter->hw, LFC);
1642         adapter->stats.pfrc += IXGB_READ_REG(&adapter->hw, PFRC);
1643         adapter->stats.pftc += IXGB_READ_REG(&adapter->hw, PFTC);
1644         adapter->stats.mcfrc += IXGB_READ_REG(&adapter->hw, MCFRC);
1645         adapter->stats.mcftc += IXGB_READ_REG(&adapter->hw, MCFTC);
1646         adapter->stats.xonrxc += IXGB_READ_REG(&adapter->hw, XONRXC);
1647         adapter->stats.xontxc += IXGB_READ_REG(&adapter->hw, XONTXC);
1648         adapter->stats.xoffrxc += IXGB_READ_REG(&adapter->hw, XOFFRXC);
1649         adapter->stats.xofftxc += IXGB_READ_REG(&adapter->hw, XOFFTXC);
1650         adapter->stats.rjc += IXGB_READ_REG(&adapter->hw, RJC);
1651
1652         /* Fill out the OS statistics structure */
1653
1654         adapter->net_stats.rx_packets = adapter->stats.gprcl;
1655         adapter->net_stats.tx_packets = adapter->stats.gptcl;
1656         adapter->net_stats.rx_bytes = adapter->stats.gorcl;
1657         adapter->net_stats.tx_bytes = adapter->stats.gotcl;
1658         adapter->net_stats.multicast = adapter->stats.mprcl;
1659         adapter->net_stats.collisions = 0;
1660
1661         /* ignore RLEC as it reports errors for padded (<64bytes) frames
1662          * with a length in the type/len field */
1663         adapter->net_stats.rx_errors =
1664             /* adapter->stats.rnbc + */ adapter->stats.crcerrs +
1665             adapter->stats.ruc +
1666             adapter->stats.roc /*+ adapter->stats.rlec */  +
1667             adapter->stats.icbc +
1668             adapter->stats.ecbc + adapter->stats.mpc;
1669
1670         /* see above
1671          * adapter->net_stats.rx_length_errors = adapter->stats.rlec;
1672          */
1673
1674         adapter->net_stats.rx_crc_errors = adapter->stats.crcerrs;
1675         adapter->net_stats.rx_fifo_errors = adapter->stats.mpc;
1676         adapter->net_stats.rx_missed_errors = adapter->stats.mpc;
1677         adapter->net_stats.rx_over_errors = adapter->stats.mpc;
1678
1679         adapter->net_stats.tx_errors = 0;
1680         adapter->net_stats.rx_frame_errors = 0;
1681         adapter->net_stats.tx_aborted_errors = 0;
1682         adapter->net_stats.tx_carrier_errors = 0;
1683         adapter->net_stats.tx_fifo_errors = 0;
1684         adapter->net_stats.tx_heartbeat_errors = 0;
1685         adapter->net_stats.tx_window_errors = 0;
1686 }
1687
1688 #define IXGB_MAX_INTR 10
1689 /**
1690  * ixgb_intr - Interrupt Handler
1691  * @irq: interrupt number
1692  * @data: pointer to a network interface device structure
1693  **/
1694
1695 static irqreturn_t
1696 ixgb_intr(int irq, void *data)
1697 {
1698         struct net_device *netdev = data;
1699         struct ixgb_adapter *adapter = netdev_priv(netdev);
1700         struct ixgb_hw *hw = &adapter->hw;
1701         u32 icr = IXGB_READ_REG(hw, ICR);
1702
1703         if (unlikely(!icr))
1704                 return IRQ_NONE;  /* Not our interrupt */
1705
1706         if (unlikely(icr & (IXGB_INT_RXSEQ | IXGB_INT_LSC)))
1707                 if (!test_bit(__IXGB_DOWN, &adapter->flags))
1708                         mod_timer(&adapter->watchdog_timer, jiffies);
1709
1710         if (netif_rx_schedule_prep(netdev, &adapter->napi)) {
1711
1712                 /* Disable interrupts and register for poll. The flush
1713                   of the posted write is intentionally left out.
1714                 */
1715
1716                 IXGB_WRITE_REG(&adapter->hw, IMC, ~0);
1717                 __netif_rx_schedule(netdev, &adapter->napi);
1718         }
1719         return IRQ_HANDLED;
1720 }
1721
1722 /**
1723  * ixgb_clean - NAPI Rx polling callback
1724  * @adapter: board private structure
1725  **/
1726
1727 static int
1728 ixgb_clean(struct napi_struct *napi, int budget)
1729 {
1730         struct ixgb_adapter *adapter = container_of(napi, struct ixgb_adapter, napi);
1731         struct net_device *netdev = adapter->netdev;
1732         int work_done = 0;
1733
1734         ixgb_clean_tx_irq(adapter);
1735         ixgb_clean_rx_irq(adapter, &work_done, budget);
1736
1737         /* If budget not fully consumed, exit the polling mode */
1738         if (work_done < budget) {
1739                 netif_rx_complete(netdev, napi);
1740                 if (!test_bit(__IXGB_DOWN, &adapter->flags))
1741                         ixgb_irq_enable(adapter);
1742         }
1743
1744         return work_done;
1745 }
1746
1747 /**
1748  * ixgb_clean_tx_irq - Reclaim resources after transmit completes
1749  * @adapter: board private structure
1750  **/
1751
1752 static bool
1753 ixgb_clean_tx_irq(struct ixgb_adapter *adapter)
1754 {
1755         struct ixgb_desc_ring *tx_ring = &adapter->tx_ring;
1756         struct net_device *netdev = adapter->netdev;
1757         struct ixgb_tx_desc *tx_desc, *eop_desc;
1758         struct ixgb_buffer *buffer_info;
1759         unsigned int i, eop;
1760         bool cleaned = false;
1761
1762         i = tx_ring->next_to_clean;
1763         eop = tx_ring->buffer_info[i].next_to_watch;
1764         eop_desc = IXGB_TX_DESC(*tx_ring, eop);
1765
1766         while (eop_desc->status & IXGB_TX_DESC_STATUS_DD) {
1767
1768                 for (cleaned = false; !cleaned; ) {
1769                         tx_desc = IXGB_TX_DESC(*tx_ring, i);
1770                         buffer_info = &tx_ring->buffer_info[i];
1771
1772                         if (tx_desc->popts &
1773                            (IXGB_TX_DESC_POPTS_TXSM |
1774                             IXGB_TX_DESC_POPTS_IXSM))
1775                                 adapter->hw_csum_tx_good++;
1776
1777                         ixgb_unmap_and_free_tx_resource(adapter, buffer_info);
1778
1779                         *(u32 *)&(tx_desc->status) = 0;
1780
1781                         cleaned = (i == eop);
1782                         if (++i == tx_ring->count) i = 0;
1783                 }
1784
1785                 eop = tx_ring->buffer_info[i].next_to_watch;
1786                 eop_desc = IXGB_TX_DESC(*tx_ring, eop);
1787         }
1788
1789         tx_ring->next_to_clean = i;
1790
1791         if (unlikely(cleaned && netif_carrier_ok(netdev) &&
1792                      IXGB_DESC_UNUSED(tx_ring) >= DESC_NEEDED)) {
1793                 /* Make sure that anybody stopping the queue after this
1794                  * sees the new next_to_clean. */
1795                 smp_mb();
1796
1797                 if (netif_queue_stopped(netdev) &&
1798                     !(test_bit(__IXGB_DOWN, &adapter->flags))) {
1799                         netif_wake_queue(netdev);
1800                         ++adapter->restart_queue;
1801                 }
1802         }
1803
1804         if (adapter->detect_tx_hung) {
1805                 /* detect a transmit hang in hardware, this serializes the
1806                  * check with the clearing of time_stamp and movement of i */
1807                 adapter->detect_tx_hung = false;
1808                 if (tx_ring->buffer_info[eop].dma &&
1809                    time_after(jiffies, tx_ring->buffer_info[eop].time_stamp + HZ)
1810                    && !(IXGB_READ_REG(&adapter->hw, STATUS) &
1811                         IXGB_STATUS_TXOFF)) {
1812                         /* detected Tx unit hang */
1813                         DPRINTK(DRV, ERR, "Detected Tx Unit Hang\n"
1814                                         "  TDH                  <%x>\n"
1815                                         "  TDT                  <%x>\n"
1816                                         "  next_to_use          <%x>\n"
1817                                         "  next_to_clean        <%x>\n"
1818                                         "buffer_info[next_to_clean]\n"
1819                                         "  time_stamp           <%lx>\n"
1820                                         "  next_to_watch        <%x>\n"
1821                                         "  jiffies              <%lx>\n"
1822                                         "  next_to_watch.status <%x>\n",
1823                                 IXGB_READ_REG(&adapter->hw, TDH),
1824                                 IXGB_READ_REG(&adapter->hw, TDT),
1825                                 tx_ring->next_to_use,
1826                                 tx_ring->next_to_clean,
1827                                 tx_ring->buffer_info[eop].time_stamp,
1828                                 eop,
1829                                 jiffies,
1830                                 eop_desc->status);
1831                         netif_stop_queue(netdev);
1832                 }
1833         }
1834
1835         return cleaned;
1836 }
1837
1838 /**
1839  * ixgb_rx_checksum - Receive Checksum Offload for 82597.
1840  * @adapter: board private structure
1841  * @rx_desc: receive descriptor
1842  * @sk_buff: socket buffer with received data
1843  **/
1844
1845 static void
1846 ixgb_rx_checksum(struct ixgb_adapter *adapter,
1847                  struct ixgb_rx_desc *rx_desc,
1848                  struct sk_buff *skb)
1849 {
1850         /* Ignore Checksum bit is set OR
1851          * TCP Checksum has not been calculated
1852          */
1853         if ((rx_desc->status & IXGB_RX_DESC_STATUS_IXSM) ||
1854            (!(rx_desc->status & IXGB_RX_DESC_STATUS_TCPCS))) {
1855                 skb->ip_summed = CHECKSUM_NONE;
1856                 return;
1857         }
1858
1859         /* At this point we know the hardware did the TCP checksum */
1860         /* now look at the TCP checksum error bit */
1861         if (rx_desc->errors & IXGB_RX_DESC_ERRORS_TCPE) {
1862                 /* let the stack verify checksum errors */
1863                 skb->ip_summed = CHECKSUM_NONE;
1864                 adapter->hw_csum_rx_error++;
1865         } else {
1866                 /* TCP checksum is good */
1867                 skb->ip_summed = CHECKSUM_UNNECESSARY;
1868                 adapter->hw_csum_rx_good++;
1869         }
1870 }
1871
1872 /**
1873  * ixgb_clean_rx_irq - Send received data up the network stack,
1874  * @adapter: board private structure
1875  **/
1876
1877 static bool
1878 ixgb_clean_rx_irq(struct ixgb_adapter *adapter, int *work_done, int work_to_do)
1879 {
1880         struct ixgb_desc_ring *rx_ring = &adapter->rx_ring;
1881         struct net_device *netdev = adapter->netdev;
1882         struct pci_dev *pdev = adapter->pdev;
1883         struct ixgb_rx_desc *rx_desc, *next_rxd;
1884         struct ixgb_buffer *buffer_info, *next_buffer, *next2_buffer;
1885         u32 length;
1886         unsigned int i, j;
1887         int cleaned_count = 0;
1888         bool cleaned = false;
1889
1890         i = rx_ring->next_to_clean;
1891         rx_desc = IXGB_RX_DESC(*rx_ring, i);
1892         buffer_info = &rx_ring->buffer_info[i];
1893
1894         while (rx_desc->status & IXGB_RX_DESC_STATUS_DD) {
1895                 struct sk_buff *skb;
1896                 u8 status;
1897
1898                 if (*work_done >= work_to_do)
1899                         break;
1900
1901                 (*work_done)++;
1902                 status = rx_desc->status;
1903                 skb = buffer_info->skb;
1904                 buffer_info->skb = NULL;
1905
1906                 prefetch(skb->data - NET_IP_ALIGN);
1907
1908                 if (++i == rx_ring->count) i = 0;
1909                 next_rxd = IXGB_RX_DESC(*rx_ring, i);
1910                 prefetch(next_rxd);
1911
1912                 if ((j = i + 1) == rx_ring->count) j = 0;
1913                 next2_buffer = &rx_ring->buffer_info[j];
1914                 prefetch(next2_buffer);
1915
1916                 next_buffer = &rx_ring->buffer_info[i];
1917
1918                 cleaned = true;
1919                 cleaned_count++;
1920
1921                 pci_unmap_single(pdev,
1922                                  buffer_info->dma,
1923                                  buffer_info->length,
1924                                  PCI_DMA_FROMDEVICE);
1925                 buffer_info->dma = 0;
1926
1927                 length = le16_to_cpu(rx_desc->length);
1928                 rx_desc->length = 0;
1929
1930                 if (unlikely(!(status & IXGB_RX_DESC_STATUS_EOP))) {
1931
1932                         /* All receives must fit into a single buffer */
1933
1934                         IXGB_DBG("Receive packet consumed multiple buffers "
1935                                          "length<%x>\n", length);
1936
1937                         dev_kfree_skb_irq(skb);
1938                         goto rxdesc_done;
1939                 }
1940
1941                 if (unlikely(rx_desc->errors &
1942                     (IXGB_RX_DESC_ERRORS_CE | IXGB_RX_DESC_ERRORS_SE |
1943                      IXGB_RX_DESC_ERRORS_P | IXGB_RX_DESC_ERRORS_RXE))) {
1944                         dev_kfree_skb_irq(skb);
1945                         goto rxdesc_done;
1946                 }
1947
1948                 /* code added for copybreak, this should improve
1949                  * performance for small packets with large amounts
1950                  * of reassembly being done in the stack */
1951                 if (length < copybreak) {
1952                         struct sk_buff *new_skb =
1953                             netdev_alloc_skb(netdev, length + NET_IP_ALIGN);
1954                         if (new_skb) {
1955                                 skb_reserve(new_skb, NET_IP_ALIGN);
1956                                 skb_copy_to_linear_data_offset(new_skb,
1957                                                                -NET_IP_ALIGN,
1958                                                                (skb->data -
1959                                                                 NET_IP_ALIGN),
1960                                                                (length +
1961                                                                 NET_IP_ALIGN));
1962                                 /* save the skb in buffer_info as good */
1963                                 buffer_info->skb = skb;
1964                                 skb = new_skb;
1965                         }
1966                 }
1967                 /* end copybreak code */
1968
1969                 /* Good Receive */
1970                 skb_put(skb, length);
1971
1972                 /* Receive Checksum Offload */
1973                 ixgb_rx_checksum(adapter, rx_desc, skb);
1974
1975                 skb->protocol = eth_type_trans(skb, netdev);
1976                 if (adapter->vlgrp && (status & IXGB_RX_DESC_STATUS_VP)) {
1977                         vlan_hwaccel_receive_skb(skb, adapter->vlgrp,
1978                                                 le16_to_cpu(rx_desc->special));
1979                 } else {
1980                         netif_receive_skb(skb);
1981                 }
1982                 netdev->last_rx = jiffies;
1983
1984 rxdesc_done:
1985                 /* clean up descriptor, might be written over by hw */
1986                 rx_desc->status = 0;
1987
1988                 /* return some buffers to hardware, one at a time is too slow */
1989                 if (unlikely(cleaned_count >= IXGB_RX_BUFFER_WRITE)) {
1990                         ixgb_alloc_rx_buffers(adapter, cleaned_count);
1991                         cleaned_count = 0;
1992                 }
1993
1994                 /* use prefetched values */
1995                 rx_desc = next_rxd;
1996                 buffer_info = next_buffer;
1997         }
1998
1999         rx_ring->next_to_clean = i;
2000
2001         cleaned_count = IXGB_DESC_UNUSED(rx_ring);
2002         if (cleaned_count)
2003                 ixgb_alloc_rx_buffers(adapter, cleaned_count);
2004
2005         return cleaned;
2006 }
2007
2008 /**
2009  * ixgb_alloc_rx_buffers - Replace used receive buffers
2010  * @adapter: address of board private structure
2011  **/
2012
2013 static void
2014 ixgb_alloc_rx_buffers(struct ixgb_adapter *adapter, int cleaned_count)
2015 {
2016         struct ixgb_desc_ring *rx_ring = &adapter->rx_ring;
2017         struct net_device *netdev = adapter->netdev;
2018         struct pci_dev *pdev = adapter->pdev;
2019         struct ixgb_rx_desc *rx_desc;
2020         struct ixgb_buffer *buffer_info;
2021         struct sk_buff *skb;
2022         unsigned int i;
2023         long cleancount;
2024
2025         i = rx_ring->next_to_use;
2026         buffer_info = &rx_ring->buffer_info[i];
2027         cleancount = IXGB_DESC_UNUSED(rx_ring);
2028
2029
2030         /* leave three descriptors unused */
2031         while (--cleancount > 2 && cleaned_count--) {
2032                 /* recycle! its good for you */
2033                 skb = buffer_info->skb;
2034                 if (skb) {
2035                         skb_trim(skb, 0);
2036                         goto map_skb;
2037                 }
2038
2039                 skb = netdev_alloc_skb(netdev, adapter->rx_buffer_len
2040                                        + NET_IP_ALIGN);
2041                 if (unlikely(!skb)) {
2042                         /* Better luck next round */
2043                         adapter->alloc_rx_buff_failed++;
2044                         break;
2045                 }
2046
2047                 /* Make buffer alignment 2 beyond a 16 byte boundary
2048                  * this will result in a 16 byte aligned IP header after
2049                  * the 14 byte MAC header is removed
2050                  */
2051                 skb_reserve(skb, NET_IP_ALIGN);
2052
2053                 buffer_info->skb = skb;
2054                 buffer_info->length = adapter->rx_buffer_len;
2055 map_skb:
2056                 buffer_info->dma = pci_map_single(pdev,
2057                                                   skb->data,
2058                                                   adapter->rx_buffer_len,
2059                                                   PCI_DMA_FROMDEVICE);
2060
2061                 rx_desc = IXGB_RX_DESC(*rx_ring, i);
2062                 rx_desc->buff_addr = cpu_to_le64(buffer_info->dma);
2063                 /* guarantee DD bit not set now before h/w gets descriptor
2064                  * this is the rest of the workaround for h/w double
2065                  * writeback. */
2066                 rx_desc->status = 0;
2067
2068
2069                 if (++i == rx_ring->count) i = 0;
2070                 buffer_info = &rx_ring->buffer_info[i];
2071         }
2072
2073         if (likely(rx_ring->next_to_use != i)) {
2074                 rx_ring->next_to_use = i;
2075                 if (unlikely(i-- == 0))
2076                         i = (rx_ring->count - 1);
2077
2078                 /* Force memory writes to complete before letting h/w
2079                  * know there are new descriptors to fetch.  (Only
2080                  * applicable for weak-ordered memory model archs, such
2081                  * as IA-64). */
2082                 wmb();
2083                 IXGB_WRITE_REG(&adapter->hw, RDT, i);
2084         }
2085 }
2086
2087 /**
2088  * ixgb_vlan_rx_register - enables or disables vlan tagging/stripping.
2089  *
2090  * @param netdev network interface device structure
2091  * @param grp indicates to enable or disable tagging/stripping
2092  **/
2093 static void
2094 ixgb_vlan_rx_register(struct net_device *netdev, struct vlan_group *grp)
2095 {
2096         struct ixgb_adapter *adapter = netdev_priv(netdev);
2097         u32 ctrl, rctl;
2098
2099         ixgb_irq_disable(adapter);
2100         adapter->vlgrp = grp;
2101
2102         if (grp) {
2103                 /* enable VLAN tag insert/strip */
2104                 ctrl = IXGB_READ_REG(&adapter->hw, CTRL0);
2105                 ctrl |= IXGB_CTRL0_VME;
2106                 IXGB_WRITE_REG(&adapter->hw, CTRL0, ctrl);
2107
2108                 /* enable VLAN receive filtering */
2109
2110                 rctl = IXGB_READ_REG(&adapter->hw, RCTL);
2111                 rctl &= ~IXGB_RCTL_CFIEN;
2112                 IXGB_WRITE_REG(&adapter->hw, RCTL, rctl);
2113         } else {
2114                 /* disable VLAN tag insert/strip */
2115
2116                 ctrl = IXGB_READ_REG(&adapter->hw, CTRL0);
2117                 ctrl &= ~IXGB_CTRL0_VME;
2118                 IXGB_WRITE_REG(&adapter->hw, CTRL0, ctrl);
2119         }
2120
2121         /* don't enable interrupts unless we are UP */
2122         if (adapter->netdev->flags & IFF_UP)
2123                 ixgb_irq_enable(adapter);
2124 }
2125
2126 static void
2127 ixgb_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
2128 {
2129         struct ixgb_adapter *adapter = netdev_priv(netdev);
2130         u32 vfta, index;
2131
2132         /* add VID to filter table */
2133
2134         index = (vid >> 5) & 0x7F;
2135         vfta = IXGB_READ_REG_ARRAY(&adapter->hw, VFTA, index);
2136         vfta |= (1 << (vid & 0x1F));
2137         ixgb_write_vfta(&adapter->hw, index, vfta);
2138 }
2139
2140 static void
2141 ixgb_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
2142 {
2143         struct ixgb_adapter *adapter = netdev_priv(netdev);
2144         u32 vfta, index;
2145
2146         ixgb_irq_disable(adapter);
2147
2148         vlan_group_set_device(adapter->vlgrp, vid, NULL);
2149
2150         /* don't enable interrupts unless we are UP */
2151         if (adapter->netdev->flags & IFF_UP)
2152                 ixgb_irq_enable(adapter);
2153
2154         /* remove VID from filter table */
2155
2156         index = (vid >> 5) & 0x7F;
2157         vfta = IXGB_READ_REG_ARRAY(&adapter->hw, VFTA, index);
2158         vfta &= ~(1 << (vid & 0x1F));
2159         ixgb_write_vfta(&adapter->hw, index, vfta);
2160 }
2161
2162 static void
2163 ixgb_restore_vlan(struct ixgb_adapter *adapter)
2164 {
2165         ixgb_vlan_rx_register(adapter->netdev, adapter->vlgrp);
2166
2167         if (adapter->vlgrp) {
2168                 u16 vid;
2169                 for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
2170                         if (!vlan_group_get_device(adapter->vlgrp, vid))
2171                                 continue;
2172                         ixgb_vlan_rx_add_vid(adapter->netdev, vid);
2173                 }
2174         }
2175 }
2176
2177 #ifdef CONFIG_NET_POLL_CONTROLLER
2178 /*
2179  * Polling 'interrupt' - used by things like netconsole to send skbs
2180  * without having to re-enable interrupts. It's not called while
2181  * the interrupt routine is executing.
2182  */
2183
2184 static void ixgb_netpoll(struct net_device *dev)
2185 {
2186         struct ixgb_adapter *adapter = netdev_priv(dev);
2187
2188         disable_irq(adapter->pdev->irq);
2189         ixgb_intr(adapter->pdev->irq, dev);
2190         enable_irq(adapter->pdev->irq);
2191 }
2192 #endif
2193
2194 /**
2195  * ixgb_io_error_detected() - called when PCI error is detected
2196  * @pdev    pointer to pci device with error
2197  * @state   pci channel state after error
2198  *
2199  * This callback is called by the PCI subsystem whenever
2200  * a PCI bus error is detected.
2201  */
2202 static pci_ers_result_t ixgb_io_error_detected(struct pci_dev *pdev,
2203                                                enum pci_channel_state state)
2204 {
2205         struct net_device *netdev = pci_get_drvdata(pdev);
2206         struct ixgb_adapter *adapter = netdev_priv(netdev);
2207
2208         if (netif_running(netdev))
2209                 ixgb_down(adapter, true);
2210
2211         pci_disable_device(pdev);
2212
2213         /* Request a slot reset. */
2214         return PCI_ERS_RESULT_NEED_RESET;
2215 }
2216
2217 /**
2218  * ixgb_io_slot_reset - called after the pci bus has been reset.
2219  * @pdev    pointer to pci device with error
2220  *
2221  * This callback is called after the PCI bus has been reset.
2222  * Basically, this tries to restart the card from scratch.
2223  * This is a shortened version of the device probe/discovery code,
2224  * it resembles the first-half of the ixgb_probe() routine.
2225  */
2226 static pci_ers_result_t ixgb_io_slot_reset(struct pci_dev *pdev)
2227 {
2228         struct net_device *netdev = pci_get_drvdata(pdev);
2229         struct ixgb_adapter *adapter = netdev_priv(netdev);
2230
2231         if (pci_enable_device(pdev)) {
2232                 DPRINTK(PROBE, ERR, "Cannot re-enable PCI device after reset.\n");
2233                 return PCI_ERS_RESULT_DISCONNECT;
2234         }
2235
2236         /* Perform card reset only on one instance of the card */
2237         if (0 != PCI_FUNC (pdev->devfn))
2238                 return PCI_ERS_RESULT_RECOVERED;
2239
2240         pci_set_master(pdev);
2241
2242         netif_carrier_off(netdev);
2243         netif_stop_queue(netdev);
2244         ixgb_reset(adapter);
2245
2246         /* Make sure the EEPROM is good */
2247         if (!ixgb_validate_eeprom_checksum(&adapter->hw)) {
2248                 DPRINTK(PROBE, ERR, "After reset, the EEPROM checksum is not valid.\n");
2249                 return PCI_ERS_RESULT_DISCONNECT;
2250         }
2251         ixgb_get_ee_mac_addr(&adapter->hw, netdev->dev_addr);
2252         memcpy(netdev->perm_addr, netdev->dev_addr, netdev->addr_len);
2253
2254         if (!is_valid_ether_addr(netdev->perm_addr)) {
2255                 DPRINTK(PROBE, ERR, "After reset, invalid MAC address.\n");
2256                 return PCI_ERS_RESULT_DISCONNECT;
2257         }
2258
2259         return PCI_ERS_RESULT_RECOVERED;
2260 }
2261
2262 /**
2263  * ixgb_io_resume - called when its OK to resume normal operations
2264  * @pdev    pointer to pci device with error
2265  *
2266  * The error recovery driver tells us that its OK to resume
2267  * normal operation. Implementation resembles the second-half
2268  * of the ixgb_probe() routine.
2269  */
2270 static void ixgb_io_resume(struct pci_dev *pdev)
2271 {
2272         struct net_device *netdev = pci_get_drvdata(pdev);
2273         struct ixgb_adapter *adapter = netdev_priv(netdev);
2274
2275         pci_set_master(pdev);
2276
2277         if (netif_running(netdev)) {
2278                 if (ixgb_up(adapter)) {
2279                         printk ("ixgb: can't bring device back up after reset\n");
2280                         return;
2281                 }
2282         }
2283
2284         netif_device_attach(netdev);
2285         mod_timer(&adapter->watchdog_timer, jiffies);
2286 }
2287
2288 /* ixgb_main.c */