211a1694667e41a6d38739bb890013534ce50308
[linux-2.6.git] / 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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
30
31 #include "ixgb.h"
32
33 char ixgb_driver_name[] = "ixgb";
34 static char ixgb_driver_string[] = "Intel(R) PRO/10GbE Network Driver";
35
36 #define DRIVERNAPI "-NAPI"
37 #define DRV_VERSION "1.0.135-k2" DRIVERNAPI
38 const char ixgb_driver_version[] = DRV_VERSION;
39 static const char ixgb_copyright[] = "Copyright (c) 1999-2008 Intel Corporation.";
40
41 #define IXGB_CB_LENGTH 256
42 static unsigned int copybreak __read_mostly = IXGB_CB_LENGTH;
43 module_param(copybreak, uint, 0644);
44 MODULE_PARM_DESC(copybreak,
45         "Maximum size of packet that is copied to a new buffer on receive");
46
47 /* ixgb_pci_tbl - PCI Device ID Table
48  *
49  * Wildcard entries (PCI_ANY_ID) should come last
50  * Last entry must be all 0s
51  *
52  * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
53  *   Class, Class Mask, private data (not used) }
54  */
55 static DEFINE_PCI_DEVICE_TABLE(ixgb_pci_tbl) = {
56         {INTEL_VENDOR_ID, IXGB_DEVICE_ID_82597EX,
57          PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0},
58         {INTEL_VENDOR_ID, IXGB_DEVICE_ID_82597EX_CX4,
59          PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0},
60         {INTEL_VENDOR_ID, IXGB_DEVICE_ID_82597EX_SR,
61          PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0},
62         {INTEL_VENDOR_ID, IXGB_DEVICE_ID_82597EX_LR,
63          PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0},
64
65         /* required last entry */
66         {0,}
67 };
68
69 MODULE_DEVICE_TABLE(pci, ixgb_pci_tbl);
70
71 /* Local Function Prototypes */
72 static int ixgb_init_module(void);
73 static void ixgb_exit_module(void);
74 static int ixgb_probe(struct pci_dev *pdev, const struct pci_device_id *ent);
75 static void __devexit ixgb_remove(struct pci_dev *pdev);
76 static int ixgb_sw_init(struct ixgb_adapter *adapter);
77 static int ixgb_open(struct net_device *netdev);
78 static int ixgb_close(struct net_device *netdev);
79 static void ixgb_configure_tx(struct ixgb_adapter *adapter);
80 static void ixgb_configure_rx(struct ixgb_adapter *adapter);
81 static void ixgb_setup_rctl(struct ixgb_adapter *adapter);
82 static void ixgb_clean_tx_ring(struct ixgb_adapter *adapter);
83 static void ixgb_clean_rx_ring(struct ixgb_adapter *adapter);
84 static void ixgb_set_multi(struct net_device *netdev);
85 static void ixgb_watchdog(unsigned long data);
86 static netdev_tx_t ixgb_xmit_frame(struct sk_buff *skb,
87                                    struct net_device *netdev);
88 static struct net_device_stats *ixgb_get_stats(struct net_device *netdev);
89 static int ixgb_change_mtu(struct net_device *netdev, int new_mtu);
90 static int ixgb_set_mac(struct net_device *netdev, void *p);
91 static irqreturn_t ixgb_intr(int irq, void *data);
92 static bool ixgb_clean_tx_irq(struct ixgb_adapter *adapter);
93
94 static int ixgb_clean(struct napi_struct *, int);
95 static bool ixgb_clean_rx_irq(struct ixgb_adapter *, int *, int);
96 static void ixgb_alloc_rx_buffers(struct ixgb_adapter *, int);
97
98 static void ixgb_tx_timeout(struct net_device *dev);
99 static void ixgb_tx_timeout_task(struct work_struct *work);
100
101 static void ixgb_vlan_rx_register(struct net_device *netdev,
102                                   struct vlan_group *grp);
103 static void ixgb_vlan_rx_add_vid(struct net_device *netdev, u16 vid);
104 static void ixgb_vlan_rx_kill_vid(struct net_device *netdev, u16 vid);
105 static void ixgb_restore_vlan(struct ixgb_adapter *adapter);
106
107 #ifdef CONFIG_NET_POLL_CONTROLLER
108 /* for netdump / net console */
109 static void ixgb_netpoll(struct net_device *dev);
110 #endif
111
112 static pci_ers_result_t ixgb_io_error_detected (struct pci_dev *pdev,
113                              enum pci_channel_state state);
114 static pci_ers_result_t ixgb_io_slot_reset (struct pci_dev *pdev);
115 static void ixgb_io_resume (struct pci_dev *pdev);
116
117 static struct pci_error_handlers ixgb_err_handler = {
118         .error_detected = ixgb_io_error_detected,
119         .slot_reset = ixgb_io_slot_reset,
120         .resume = ixgb_io_resume,
121 };
122
123 static struct pci_driver ixgb_driver = {
124         .name     = ixgb_driver_name,
125         .id_table = ixgb_pci_tbl,
126         .probe    = ixgb_probe,
127         .remove   = __devexit_p(ixgb_remove),
128         .err_handler = &ixgb_err_handler
129 };
130
131 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
132 MODULE_DESCRIPTION("Intel(R) PRO/10GbE Network Driver");
133 MODULE_LICENSE("GPL");
134 MODULE_VERSION(DRV_VERSION);
135
136 #define DEFAULT_DEBUG_LEVEL_SHIFT 3
137 static int debug = DEFAULT_DEBUG_LEVEL_SHIFT;
138 module_param(debug, int, 0);
139 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
140
141 /**
142  * ixgb_init_module - Driver Registration Routine
143  *
144  * ixgb_init_module is the first routine called when the driver is
145  * loaded. All it does is register with the PCI subsystem.
146  **/
147
148 static int __init
149 ixgb_init_module(void)
150 {
151         pr_info("%s - version %s\n", ixgb_driver_string, ixgb_driver_version);
152         pr_info("%s\n", ixgb_copyright);
153
154         return pci_register_driver(&ixgb_driver);
155 }
156
157 module_init(ixgb_init_module);
158
159 /**
160  * ixgb_exit_module - Driver Exit Cleanup Routine
161  *
162  * ixgb_exit_module is called just before the driver is removed
163  * from memory.
164  **/
165
166 static void __exit
167 ixgb_exit_module(void)
168 {
169         pci_unregister_driver(&ixgb_driver);
170 }
171
172 module_exit(ixgb_exit_module);
173
174 /**
175  * ixgb_irq_disable - Mask off interrupt generation on the NIC
176  * @adapter: board private structure
177  **/
178
179 static void
180 ixgb_irq_disable(struct ixgb_adapter *adapter)
181 {
182         IXGB_WRITE_REG(&adapter->hw, IMC, ~0);
183         IXGB_WRITE_FLUSH(&adapter->hw);
184         synchronize_irq(adapter->pdev->irq);
185 }
186
187 /**
188  * ixgb_irq_enable - Enable default interrupt generation settings
189  * @adapter: board private structure
190  **/
191
192 static void
193 ixgb_irq_enable(struct ixgb_adapter *adapter)
194 {
195         u32 val = IXGB_INT_RXT0 | IXGB_INT_RXDMT0 |
196                   IXGB_INT_TXDW | IXGB_INT_LSC;
197         if (adapter->hw.subsystem_vendor_id == SUN_SUBVENDOR_ID)
198                 val |= IXGB_INT_GPI0;
199         IXGB_WRITE_REG(&adapter->hw, IMS, val);
200         IXGB_WRITE_FLUSH(&adapter->hw);
201 }
202
203 int
204 ixgb_up(struct ixgb_adapter *adapter)
205 {
206         struct net_device *netdev = adapter->netdev;
207         int err, irq_flags = IRQF_SHARED;
208         int max_frame = netdev->mtu + ENET_HEADER_SIZE + ENET_FCS_LENGTH;
209         struct ixgb_hw *hw = &adapter->hw;
210
211         /* hardware has been reset, we need to reload some things */
212
213         ixgb_rar_set(hw, netdev->dev_addr, 0);
214         ixgb_set_multi(netdev);
215
216         ixgb_restore_vlan(adapter);
217
218         ixgb_configure_tx(adapter);
219         ixgb_setup_rctl(adapter);
220         ixgb_configure_rx(adapter);
221         ixgb_alloc_rx_buffers(adapter, IXGB_DESC_UNUSED(&adapter->rx_ring));
222
223         /* disable interrupts and get the hardware into a known state */
224         IXGB_WRITE_REG(&adapter->hw, IMC, 0xffffffff);
225
226         /* only enable MSI if bus is in PCI-X mode */
227         if (IXGB_READ_REG(&adapter->hw, STATUS) & IXGB_STATUS_PCIX_MODE) {
228                 err = pci_enable_msi(adapter->pdev);
229                 if (!err) {
230                         adapter->have_msi = 1;
231                         irq_flags = 0;
232                 }
233                 /* proceed to try to request regular interrupt */
234         }
235
236         err = request_irq(adapter->pdev->irq, ixgb_intr, irq_flags,
237                           netdev->name, netdev);
238         if (err) {
239                 if (adapter->have_msi)
240                         pci_disable_msi(adapter->pdev);
241                 netif_err(adapter, probe, adapter->netdev,
242                           "Unable to allocate interrupt Error: %d\n", err);
243                 return err;
244         }
245
246         if ((hw->max_frame_size != max_frame) ||
247                 (hw->max_frame_size !=
248                 (IXGB_READ_REG(hw, MFS) >> IXGB_MFS_SHIFT))) {
249
250                 hw->max_frame_size = max_frame;
251
252                 IXGB_WRITE_REG(hw, MFS, hw->max_frame_size << IXGB_MFS_SHIFT);
253
254                 if (hw->max_frame_size >
255                    IXGB_MAX_ENET_FRAME_SIZE_WITHOUT_FCS + ENET_FCS_LENGTH) {
256                         u32 ctrl0 = IXGB_READ_REG(hw, CTRL0);
257
258                         if (!(ctrl0 & IXGB_CTRL0_JFE)) {
259                                 ctrl0 |= IXGB_CTRL0_JFE;
260                                 IXGB_WRITE_REG(hw, CTRL0, ctrl0);
261                         }
262                 }
263         }
264
265         clear_bit(__IXGB_DOWN, &adapter->flags);
266
267         napi_enable(&adapter->napi);
268         ixgb_irq_enable(adapter);
269
270         netif_wake_queue(netdev);
271
272         mod_timer(&adapter->watchdog_timer, jiffies);
273
274         return 0;
275 }
276
277 void
278 ixgb_down(struct ixgb_adapter *adapter, bool kill_watchdog)
279 {
280         struct net_device *netdev = adapter->netdev;
281
282         /* prevent the interrupt handler from restarting watchdog */
283         set_bit(__IXGB_DOWN, &adapter->flags);
284
285         napi_disable(&adapter->napi);
286         /* waiting for NAPI to complete can re-enable interrupts */
287         ixgb_irq_disable(adapter);
288         free_irq(adapter->pdev->irq, netdev);
289
290         if (adapter->have_msi)
291                 pci_disable_msi(adapter->pdev);
292
293         if (kill_watchdog)
294                 del_timer_sync(&adapter->watchdog_timer);
295
296         adapter->link_speed = 0;
297         adapter->link_duplex = 0;
298         netif_carrier_off(netdev);
299         netif_stop_queue(netdev);
300
301         ixgb_reset(adapter);
302         ixgb_clean_tx_ring(adapter);
303         ixgb_clean_rx_ring(adapter);
304 }
305
306 void
307 ixgb_reset(struct ixgb_adapter *adapter)
308 {
309         struct ixgb_hw *hw = &adapter->hw;
310
311         ixgb_adapter_stop(hw);
312         if (!ixgb_init_hw(hw))
313                 netif_err(adapter, probe, adapter->netdev, "ixgb_init_hw failed\n");
314
315         /* restore frame size information */
316         IXGB_WRITE_REG(hw, MFS, hw->max_frame_size << IXGB_MFS_SHIFT);
317         if (hw->max_frame_size >
318             IXGB_MAX_ENET_FRAME_SIZE_WITHOUT_FCS + ENET_FCS_LENGTH) {
319                 u32 ctrl0 = IXGB_READ_REG(hw, CTRL0);
320                 if (!(ctrl0 & IXGB_CTRL0_JFE)) {
321                         ctrl0 |= IXGB_CTRL0_JFE;
322                         IXGB_WRITE_REG(hw, CTRL0, ctrl0);
323                 }
324         }
325 }
326
327 static const struct net_device_ops ixgb_netdev_ops = {
328         .ndo_open               = ixgb_open,
329         .ndo_stop               = ixgb_close,
330         .ndo_start_xmit         = ixgb_xmit_frame,
331         .ndo_get_stats          = ixgb_get_stats,
332         .ndo_set_multicast_list = ixgb_set_multi,
333         .ndo_validate_addr      = eth_validate_addr,
334         .ndo_set_mac_address    = ixgb_set_mac,
335         .ndo_change_mtu         = ixgb_change_mtu,
336         .ndo_tx_timeout         = ixgb_tx_timeout,
337         .ndo_vlan_rx_register   = ixgb_vlan_rx_register,
338         .ndo_vlan_rx_add_vid    = ixgb_vlan_rx_add_vid,
339         .ndo_vlan_rx_kill_vid   = ixgb_vlan_rx_kill_vid,
340 #ifdef CONFIG_NET_POLL_CONTROLLER
341         .ndo_poll_controller    = ixgb_netpoll,
342 #endif
343 };
344
345 /**
346  * ixgb_probe - Device Initialization Routine
347  * @pdev: PCI device information struct
348  * @ent: entry in ixgb_pci_tbl
349  *
350  * Returns 0 on success, negative on failure
351  *
352  * ixgb_probe initializes an adapter identified by a pci_dev structure.
353  * The OS initialization, configuring of the adapter private structure,
354  * and a hardware reset occur.
355  **/
356
357 static int __devinit
358 ixgb_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
359 {
360         struct net_device *netdev = NULL;
361         struct ixgb_adapter *adapter;
362         static int cards_found = 0;
363         int pci_using_dac;
364         int i;
365         int err;
366
367         err = pci_enable_device(pdev);
368         if (err)
369                 return err;
370
371         pci_using_dac = 0;
372         err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(64));
373         if (!err) {
374                 err = dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64));
375                 if (!err)
376                         pci_using_dac = 1;
377         } else {
378                 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
379                 if (err) {
380                         err = dma_set_coherent_mask(&pdev->dev,
381                                                     DMA_BIT_MASK(32));
382                         if (err) {
383                                 pr_err("No usable DMA configuration, aborting\n");
384                                 goto err_dma_mask;
385                         }
386                 }
387         }
388
389         err = pci_request_regions(pdev, ixgb_driver_name);
390         if (err)
391                 goto err_request_regions;
392
393         pci_set_master(pdev);
394
395         netdev = alloc_etherdev(sizeof(struct ixgb_adapter));
396         if (!netdev) {
397                 err = -ENOMEM;
398                 goto err_alloc_etherdev;
399         }
400
401         SET_NETDEV_DEV(netdev, &pdev->dev);
402
403         pci_set_drvdata(pdev, netdev);
404         adapter = netdev_priv(netdev);
405         adapter->netdev = netdev;
406         adapter->pdev = pdev;
407         adapter->hw.back = adapter;
408         adapter->msg_enable = netif_msg_init(debug, DEFAULT_DEBUG_LEVEL_SHIFT);
409
410         adapter->hw.hw_addr = pci_ioremap_bar(pdev, BAR_0);
411         if (!adapter->hw.hw_addr) {
412                 err = -EIO;
413                 goto err_ioremap;
414         }
415
416         for (i = BAR_1; i <= BAR_5; i++) {
417                 if (pci_resource_len(pdev, i) == 0)
418                         continue;
419                 if (pci_resource_flags(pdev, i) & IORESOURCE_IO) {
420                         adapter->hw.io_base = pci_resource_start(pdev, i);
421                         break;
422                 }
423         }
424
425         netdev->netdev_ops = &ixgb_netdev_ops;
426         ixgb_set_ethtool_ops(netdev);
427         netdev->watchdog_timeo = 5 * HZ;
428         netif_napi_add(netdev, &adapter->napi, ixgb_clean, 64);
429
430         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
431
432         adapter->bd_number = cards_found;
433         adapter->link_speed = 0;
434         adapter->link_duplex = 0;
435
436         /* setup the private structure */
437
438         err = ixgb_sw_init(adapter);
439         if (err)
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
449         if (pci_using_dac) {
450                 netdev->features |= NETIF_F_HIGHDMA;
451                 netdev->vlan_features |= NETIF_F_HIGHDMA;
452         }
453
454         /* make sure the EEPROM is good */
455
456         if (!ixgb_validate_eeprom_checksum(&adapter->hw)) {
457                 netif_err(adapter, probe, adapter->netdev,
458                           "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                 netif_err(adapter, probe, adapter->netdev, "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         err = register_netdev(netdev);
482         if (err)
483                 goto err_register;
484
485         /* carrier off reporting is important to ethtool even BEFORE open */
486         netif_carrier_off(netdev);
487
488         netif_info(adapter, probe, adapter->netdev,
489                    "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         flush_scheduled_work();
529
530         unregister_netdev(netdev);
531
532         iounmap(adapter->hw.hw_addr);
533         pci_release_regions(pdev);
534
535         free_netdev(netdev);
536         pci_disable_device(pdev);
537 }
538
539 /**
540  * ixgb_sw_init - Initialize general software structures (struct ixgb_adapter)
541  * @adapter: board private structure to initialize
542  *
543  * ixgb_sw_init initializes the Adapter private data structure.
544  * Fields are initialized based on PCI device information and
545  * OS network device settings (MTU size).
546  **/
547
548 static int __devinit
549 ixgb_sw_init(struct ixgb_adapter *adapter)
550 {
551         struct ixgb_hw *hw = &adapter->hw;
552         struct net_device *netdev = adapter->netdev;
553         struct pci_dev *pdev = adapter->pdev;
554
555         /* PCI config space info */
556
557         hw->vendor_id = pdev->vendor;
558         hw->device_id = pdev->device;
559         hw->subsystem_vendor_id = pdev->subsystem_vendor;
560         hw->subsystem_id = pdev->subsystem_device;
561
562         hw->max_frame_size = netdev->mtu + ENET_HEADER_SIZE + ENET_FCS_LENGTH;
563         adapter->rx_buffer_len = hw->max_frame_size + 8; /* + 8 for errata */
564
565         if ((hw->device_id == IXGB_DEVICE_ID_82597EX) ||
566             (hw->device_id == IXGB_DEVICE_ID_82597EX_CX4) ||
567             (hw->device_id == IXGB_DEVICE_ID_82597EX_LR) ||
568             (hw->device_id == IXGB_DEVICE_ID_82597EX_SR))
569                 hw->mac_type = ixgb_82597;
570         else {
571                 /* should never have loaded on this device */
572                 netif_err(adapter, probe, adapter->netdev, "unsupported device id\n");
573         }
574
575         /* enable flow control to be programmed */
576         hw->fc.send_xon = 1;
577
578         set_bit(__IXGB_DOWN, &adapter->flags);
579         return 0;
580 }
581
582 /**
583  * ixgb_open - Called when a network interface is made active
584  * @netdev: network interface device structure
585  *
586  * Returns 0 on success, negative value on failure
587  *
588  * The open entry point is called when a network interface is made
589  * active by the system (IFF_UP).  At this point all resources needed
590  * for transmit and receive operations are allocated, the interrupt
591  * handler is registered with the OS, the watchdog timer is started,
592  * and the stack is notified that the interface is ready.
593  **/
594
595 static int
596 ixgb_open(struct net_device *netdev)
597 {
598         struct ixgb_adapter *adapter = netdev_priv(netdev);
599         int err;
600
601         /* allocate transmit descriptors */
602         err = ixgb_setup_tx_resources(adapter);
603         if (err)
604                 goto err_setup_tx;
605
606         netif_carrier_off(netdev);
607
608         /* allocate receive descriptors */
609
610         err = ixgb_setup_rx_resources(adapter);
611         if (err)
612                 goto err_setup_rx;
613
614         err = ixgb_up(adapter);
615         if (err)
616                 goto err_up;
617
618         netif_start_queue(netdev);
619
620         return 0;
621
622 err_up:
623         ixgb_free_rx_resources(adapter);
624 err_setup_rx:
625         ixgb_free_tx_resources(adapter);
626 err_setup_tx:
627         ixgb_reset(adapter);
628
629         return err;
630 }
631
632 /**
633  * ixgb_close - Disables a network interface
634  * @netdev: network interface device structure
635  *
636  * Returns 0, this is not allowed to fail
637  *
638  * The close entry point is called when an interface is de-activated
639  * by the OS.  The hardware is still under the drivers control, but
640  * needs to be disabled.  A global MAC reset is issued to stop the
641  * hardware, and all transmit and receive resources are freed.
642  **/
643
644 static int
645 ixgb_close(struct net_device *netdev)
646 {
647         struct ixgb_adapter *adapter = netdev_priv(netdev);
648
649         ixgb_down(adapter, true);
650
651         ixgb_free_tx_resources(adapter);
652         ixgb_free_rx_resources(adapter);
653
654         return 0;
655 }
656
657 /**
658  * ixgb_setup_tx_resources - allocate Tx resources (Descriptors)
659  * @adapter: board private structure
660  *
661  * Return 0 on success, negative on failure
662  **/
663
664 int
665 ixgb_setup_tx_resources(struct ixgb_adapter *adapter)
666 {
667         struct ixgb_desc_ring *txdr = &adapter->tx_ring;
668         struct pci_dev *pdev = adapter->pdev;
669         int size;
670
671         size = sizeof(struct ixgb_buffer) * txdr->count;
672         txdr->buffer_info = vzalloc(size);
673         if (!txdr->buffer_info) {
674                 netif_err(adapter, probe, adapter->netdev,
675                           "Unable to allocate transmit descriptor ring memory\n");
676                 return -ENOMEM;
677         }
678
679         /* round up to nearest 4K */
680
681         txdr->size = txdr->count * sizeof(struct ixgb_tx_desc);
682         txdr->size = ALIGN(txdr->size, 4096);
683
684         txdr->desc = dma_alloc_coherent(&pdev->dev, txdr->size, &txdr->dma,
685                                         GFP_KERNEL);
686         if (!txdr->desc) {
687                 vfree(txdr->buffer_info);
688                 netif_err(adapter, probe, adapter->netdev,
689                           "Unable to allocate transmit descriptor memory\n");
690                 return -ENOMEM;
691         }
692         memset(txdr->desc, 0, txdr->size);
693
694         txdr->next_to_use = 0;
695         txdr->next_to_clean = 0;
696
697         return 0;
698 }
699
700 /**
701  * ixgb_configure_tx - Configure 82597 Transmit Unit after Reset.
702  * @adapter: board private structure
703  *
704  * Configure the Tx unit of the MAC after a reset.
705  **/
706
707 static void
708 ixgb_configure_tx(struct ixgb_adapter *adapter)
709 {
710         u64 tdba = adapter->tx_ring.dma;
711         u32 tdlen = adapter->tx_ring.count * sizeof(struct ixgb_tx_desc);
712         u32 tctl;
713         struct ixgb_hw *hw = &adapter->hw;
714
715         /* Setup the Base and Length of the Tx Descriptor Ring
716          * tx_ring.dma can be either a 32 or 64 bit value
717          */
718
719         IXGB_WRITE_REG(hw, TDBAL, (tdba & 0x00000000ffffffffULL));
720         IXGB_WRITE_REG(hw, TDBAH, (tdba >> 32));
721
722         IXGB_WRITE_REG(hw, TDLEN, tdlen);
723
724         /* Setup the HW Tx Head and Tail descriptor pointers */
725
726         IXGB_WRITE_REG(hw, TDH, 0);
727         IXGB_WRITE_REG(hw, TDT, 0);
728
729         /* don't set up txdctl, it induces performance problems if configured
730          * incorrectly */
731         /* Set the Tx Interrupt Delay register */
732
733         IXGB_WRITE_REG(hw, TIDV, adapter->tx_int_delay);
734
735         /* Program the Transmit Control Register */
736
737         tctl = IXGB_TCTL_TCE | IXGB_TCTL_TXEN | IXGB_TCTL_TPDE;
738         IXGB_WRITE_REG(hw, TCTL, tctl);
739
740         /* Setup Transmit Descriptor Settings for this adapter */
741         adapter->tx_cmd_type =
742                 IXGB_TX_DESC_TYPE |
743                 (adapter->tx_int_delay_enable ? IXGB_TX_DESC_CMD_IDE : 0);
744 }
745
746 /**
747  * ixgb_setup_rx_resources - allocate Rx resources (Descriptors)
748  * @adapter: board private structure
749  *
750  * Returns 0 on success, negative on failure
751  **/
752
753 int
754 ixgb_setup_rx_resources(struct ixgb_adapter *adapter)
755 {
756         struct ixgb_desc_ring *rxdr = &adapter->rx_ring;
757         struct pci_dev *pdev = adapter->pdev;
758         int size;
759
760         size = sizeof(struct ixgb_buffer) * rxdr->count;
761         rxdr->buffer_info = vzalloc(size);
762         if (!rxdr->buffer_info) {
763                 netif_err(adapter, probe, adapter->netdev,
764                           "Unable to allocate receive descriptor ring\n");
765                 return -ENOMEM;
766         }
767
768         /* Round up to nearest 4K */
769
770         rxdr->size = rxdr->count * sizeof(struct ixgb_rx_desc);
771         rxdr->size = ALIGN(rxdr->size, 4096);
772
773         rxdr->desc = dma_alloc_coherent(&pdev->dev, rxdr->size, &rxdr->dma,
774                                         GFP_KERNEL);
775
776         if (!rxdr->desc) {
777                 vfree(rxdr->buffer_info);
778                 netif_err(adapter, probe, adapter->netdev,
779                           "Unable to allocate receive descriptors\n");
780                 return -ENOMEM;
781         }
782         memset(rxdr->desc, 0, rxdr->size);
783
784         rxdr->next_to_clean = 0;
785         rxdr->next_to_use = 0;
786
787         return 0;
788 }
789
790 /**
791  * ixgb_setup_rctl - configure the receive control register
792  * @adapter: Board private structure
793  **/
794
795 static void
796 ixgb_setup_rctl(struct ixgb_adapter *adapter)
797 {
798         u32 rctl;
799
800         rctl = IXGB_READ_REG(&adapter->hw, RCTL);
801
802         rctl &= ~(3 << IXGB_RCTL_MO_SHIFT);
803
804         rctl |=
805                 IXGB_RCTL_BAM | IXGB_RCTL_RDMTS_1_2 |
806                 IXGB_RCTL_RXEN | IXGB_RCTL_CFF |
807                 (adapter->hw.mc_filter_type << IXGB_RCTL_MO_SHIFT);
808
809         rctl |= IXGB_RCTL_SECRC;
810
811         if (adapter->rx_buffer_len <= IXGB_RXBUFFER_2048)
812                 rctl |= IXGB_RCTL_BSIZE_2048;
813         else if (adapter->rx_buffer_len <= IXGB_RXBUFFER_4096)
814                 rctl |= IXGB_RCTL_BSIZE_4096;
815         else if (adapter->rx_buffer_len <= IXGB_RXBUFFER_8192)
816                 rctl |= IXGB_RCTL_BSIZE_8192;
817         else if (adapter->rx_buffer_len <= IXGB_RXBUFFER_16384)
818                 rctl |= IXGB_RCTL_BSIZE_16384;
819
820         IXGB_WRITE_REG(&adapter->hw, RCTL, rctl);
821 }
822
823 /**
824  * ixgb_configure_rx - Configure 82597 Receive Unit after Reset.
825  * @adapter: board private structure
826  *
827  * Configure the Rx unit of the MAC after a reset.
828  **/
829
830 static void
831 ixgb_configure_rx(struct ixgb_adapter *adapter)
832 {
833         u64 rdba = adapter->rx_ring.dma;
834         u32 rdlen = adapter->rx_ring.count * sizeof(struct ixgb_rx_desc);
835         struct ixgb_hw *hw = &adapter->hw;
836         u32 rctl;
837         u32 rxcsum;
838
839         /* make sure receives are disabled while setting up the descriptors */
840
841         rctl = IXGB_READ_REG(hw, RCTL);
842         IXGB_WRITE_REG(hw, RCTL, rctl & ~IXGB_RCTL_RXEN);
843
844         /* set the Receive Delay Timer Register */
845
846         IXGB_WRITE_REG(hw, RDTR, adapter->rx_int_delay);
847
848         /* Setup the Base and Length of the Rx Descriptor Ring */
849
850         IXGB_WRITE_REG(hw, RDBAL, (rdba & 0x00000000ffffffffULL));
851         IXGB_WRITE_REG(hw, RDBAH, (rdba >> 32));
852
853         IXGB_WRITE_REG(hw, RDLEN, rdlen);
854
855         /* Setup the HW Rx Head and Tail Descriptor Pointers */
856         IXGB_WRITE_REG(hw, RDH, 0);
857         IXGB_WRITE_REG(hw, RDT, 0);
858
859         /* due to the hardware errata with RXDCTL, we are unable to use any of
860          * the performance enhancing features of it without causing other
861          * subtle bugs, some of the bugs could include receive length
862          * corruption at high data rates (WTHRESH > 0) and/or receive
863          * descriptor ring irregularites (particularly in hardware cache) */
864         IXGB_WRITE_REG(hw, RXDCTL, 0);
865
866         /* Enable Receive Checksum Offload for TCP and UDP */
867         if (adapter->rx_csum) {
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         dma_free_coherent(&pdev->dev, 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         if (buffer_info->dma) {
906                 if (buffer_info->mapped_as_page)
907                         dma_unmap_page(&adapter->pdev->dev, buffer_info->dma,
908                                        buffer_info->length, DMA_TO_DEVICE);
909                 else
910                         dma_unmap_single(&adapter->pdev->dev, buffer_info->dma,
911                                          buffer_info->length, DMA_TO_DEVICE);
912                 buffer_info->dma = 0;
913         }
914
915         if (buffer_info->skb) {
916                 dev_kfree_skb_any(buffer_info->skb);
917                 buffer_info->skb = NULL;
918         }
919         buffer_info->time_stamp = 0;
920         /* these fields must always be initialized in tx
921          * buffer_info->length = 0;
922          * buffer_info->next_to_watch = 0; */
923 }
924
925 /**
926  * ixgb_clean_tx_ring - Free Tx Buffers
927  * @adapter: board private structure
928  **/
929
930 static void
931 ixgb_clean_tx_ring(struct ixgb_adapter *adapter)
932 {
933         struct ixgb_desc_ring *tx_ring = &adapter->tx_ring;
934         struct ixgb_buffer *buffer_info;
935         unsigned long size;
936         unsigned int i;
937
938         /* Free all the Tx ring sk_buffs */
939
940         for (i = 0; i < tx_ring->count; i++) {
941                 buffer_info = &tx_ring->buffer_info[i];
942                 ixgb_unmap_and_free_tx_resource(adapter, buffer_info);
943         }
944
945         size = sizeof(struct ixgb_buffer) * tx_ring->count;
946         memset(tx_ring->buffer_info, 0, size);
947
948         /* Zero out the descriptor ring */
949
950         memset(tx_ring->desc, 0, tx_ring->size);
951
952         tx_ring->next_to_use = 0;
953         tx_ring->next_to_clean = 0;
954
955         IXGB_WRITE_REG(&adapter->hw, TDH, 0);
956         IXGB_WRITE_REG(&adapter->hw, TDT, 0);
957 }
958
959 /**
960  * ixgb_free_rx_resources - Free Rx Resources
961  * @adapter: board private structure
962  *
963  * Free all receive software resources
964  **/
965
966 void
967 ixgb_free_rx_resources(struct ixgb_adapter *adapter)
968 {
969         struct ixgb_desc_ring *rx_ring = &adapter->rx_ring;
970         struct pci_dev *pdev = adapter->pdev;
971
972         ixgb_clean_rx_ring(adapter);
973
974         vfree(rx_ring->buffer_info);
975         rx_ring->buffer_info = NULL;
976
977         dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
978                           rx_ring->dma);
979
980         rx_ring->desc = NULL;
981 }
982
983 /**
984  * ixgb_clean_rx_ring - Free Rx Buffers
985  * @adapter: board private structure
986  **/
987
988 static void
989 ixgb_clean_rx_ring(struct ixgb_adapter *adapter)
990 {
991         struct ixgb_desc_ring *rx_ring = &adapter->rx_ring;
992         struct ixgb_buffer *buffer_info;
993         struct pci_dev *pdev = adapter->pdev;
994         unsigned long size;
995         unsigned int i;
996
997         /* Free all the Rx ring sk_buffs */
998
999         for (i = 0; i < rx_ring->count; i++) {
1000                 buffer_info = &rx_ring->buffer_info[i];
1001                 if (buffer_info->dma) {
1002                         dma_unmap_single(&pdev->dev,
1003                                          buffer_info->dma,
1004                                          buffer_info->length,
1005                                          DMA_FROM_DEVICE);
1006                         buffer_info->dma = 0;
1007                         buffer_info->length = 0;
1008                 }
1009
1010                 if (buffer_info->skb) {
1011                         dev_kfree_skb(buffer_info->skb);
1012                         buffer_info->skb = NULL;
1013                 }
1014         }
1015
1016         size = sizeof(struct ixgb_buffer) * rx_ring->count;
1017         memset(rx_ring->buffer_info, 0, size);
1018
1019         /* Zero out the descriptor ring */
1020
1021         memset(rx_ring->desc, 0, rx_ring->size);
1022
1023         rx_ring->next_to_clean = 0;
1024         rx_ring->next_to_use = 0;
1025
1026         IXGB_WRITE_REG(&adapter->hw, RDH, 0);
1027         IXGB_WRITE_REG(&adapter->hw, RDT, 0);
1028 }
1029
1030 /**
1031  * ixgb_set_mac - Change the Ethernet Address of the NIC
1032  * @netdev: network interface device structure
1033  * @p: pointer to an address structure
1034  *
1035  * Returns 0 on success, negative on failure
1036  **/
1037
1038 static int
1039 ixgb_set_mac(struct net_device *netdev, void *p)
1040 {
1041         struct ixgb_adapter *adapter = netdev_priv(netdev);
1042         struct sockaddr *addr = p;
1043
1044         if (!is_valid_ether_addr(addr->sa_data))
1045                 return -EADDRNOTAVAIL;
1046
1047         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
1048
1049         ixgb_rar_set(&adapter->hw, addr->sa_data, 0);
1050
1051         return 0;
1052 }
1053
1054 /**
1055  * ixgb_set_multi - Multicast and Promiscuous mode set
1056  * @netdev: network interface device structure
1057  *
1058  * The set_multi entry point is called whenever the multicast address
1059  * list or the network interface flags are updated.  This routine is
1060  * responsible for configuring the hardware for proper multicast,
1061  * promiscuous mode, and all-multi behavior.
1062  **/
1063
1064 static void
1065 ixgb_set_multi(struct net_device *netdev)
1066 {
1067         struct ixgb_adapter *adapter = netdev_priv(netdev);
1068         struct ixgb_hw *hw = &adapter->hw;
1069         struct netdev_hw_addr *ha;
1070         u32 rctl;
1071         int i;
1072
1073         /* Check for Promiscuous and All Multicast modes */
1074
1075         rctl = IXGB_READ_REG(hw, RCTL);
1076
1077         if (netdev->flags & IFF_PROMISC) {
1078                 rctl |= (IXGB_RCTL_UPE | IXGB_RCTL_MPE);
1079                 rctl &= ~IXGB_RCTL_VFE;
1080         } else {
1081                 if (netdev->flags & IFF_ALLMULTI) {
1082                         rctl |= IXGB_RCTL_MPE;
1083                         rctl &= ~IXGB_RCTL_UPE;
1084                 } else {
1085                         rctl &= ~(IXGB_RCTL_UPE | IXGB_RCTL_MPE);
1086                 }
1087                 rctl |= IXGB_RCTL_VFE;
1088         }
1089
1090         if (netdev_mc_count(netdev) > IXGB_MAX_NUM_MULTICAST_ADDRESSES) {
1091                 rctl |= IXGB_RCTL_MPE;
1092                 IXGB_WRITE_REG(hw, RCTL, rctl);
1093         } else {
1094                 u8 mta[IXGB_MAX_NUM_MULTICAST_ADDRESSES *
1095                             IXGB_ETH_LENGTH_OF_ADDRESS];
1096
1097                 IXGB_WRITE_REG(hw, RCTL, rctl);
1098
1099                 i = 0;
1100                 netdev_for_each_mc_addr(ha, netdev)
1101                         memcpy(&mta[i++ * IXGB_ETH_LENGTH_OF_ADDRESS],
1102                                ha->addr, IXGB_ETH_LENGTH_OF_ADDRESS);
1103
1104                 ixgb_mc_addr_list_update(hw, mta, netdev_mc_count(netdev), 0);
1105         }
1106 }
1107
1108 /**
1109  * ixgb_watchdog - Timer Call-back
1110  * @data: pointer to netdev cast into an unsigned long
1111  **/
1112
1113 static void
1114 ixgb_watchdog(unsigned long data)
1115 {
1116         struct ixgb_adapter *adapter = (struct ixgb_adapter *)data;
1117         struct net_device *netdev = adapter->netdev;
1118         struct ixgb_desc_ring *txdr = &adapter->tx_ring;
1119
1120         ixgb_check_for_link(&adapter->hw);
1121
1122         if (ixgb_check_for_bad_link(&adapter->hw)) {
1123                 /* force the reset path */
1124                 netif_stop_queue(netdev);
1125         }
1126
1127         if (adapter->hw.link_up) {
1128                 if (!netif_carrier_ok(netdev)) {
1129                         netdev_info(netdev,
1130                                     "NIC Link is Up 10 Gbps Full Duplex, Flow Control: %s\n",
1131                                     (adapter->hw.fc.type == ixgb_fc_full) ?
1132                                     "RX/TX" :
1133                                     (adapter->hw.fc.type == ixgb_fc_rx_pause) ?
1134                                      "RX" :
1135                                     (adapter->hw.fc.type == ixgb_fc_tx_pause) ?
1136                                     "TX" : "None");
1137                         adapter->link_speed = 10000;
1138                         adapter->link_duplex = FULL_DUPLEX;
1139                         netif_carrier_on(netdev);
1140                 }
1141         } else {
1142                 if (netif_carrier_ok(netdev)) {
1143                         adapter->link_speed = 0;
1144                         adapter->link_duplex = 0;
1145                         netdev_info(netdev, "NIC Link is Down\n");
1146                         netif_carrier_off(netdev);
1147                 }
1148         }
1149
1150         ixgb_update_stats(adapter);
1151
1152         if (!netif_carrier_ok(netdev)) {
1153                 if (IXGB_DESC_UNUSED(txdr) + 1 < txdr->count) {
1154                         /* We've lost link, so the controller stops DMA,
1155                          * but we've got queued Tx work that's never going
1156                          * to get done, so reset controller to flush Tx.
1157                          * (Do the reset outside of interrupt context). */
1158                         schedule_work(&adapter->tx_timeout_task);
1159                         /* return immediately since reset is imminent */
1160                         return;
1161                 }
1162         }
1163
1164         /* Force detection of hung controller every watchdog period */
1165         adapter->detect_tx_hung = true;
1166
1167         /* generate an interrupt to force clean up of any stragglers */
1168         IXGB_WRITE_REG(&adapter->hw, ICS, IXGB_INT_TXDW);
1169
1170         /* Reset the timer */
1171         mod_timer(&adapter->watchdog_timer, jiffies + 2 * HZ);
1172 }
1173
1174 #define IXGB_TX_FLAGS_CSUM              0x00000001
1175 #define IXGB_TX_FLAGS_VLAN              0x00000002
1176 #define IXGB_TX_FLAGS_TSO               0x00000004
1177
1178 static int
1179 ixgb_tso(struct ixgb_adapter *adapter, struct sk_buff *skb)
1180 {
1181         struct ixgb_context_desc *context_desc;
1182         unsigned int i;
1183         u8 ipcss, ipcso, tucss, tucso, hdr_len;
1184         u16 ipcse, tucse, mss;
1185         int err;
1186
1187         if (likely(skb_is_gso(skb))) {
1188                 struct ixgb_buffer *buffer_info;
1189                 struct iphdr *iph;
1190
1191                 if (skb_header_cloned(skb)) {
1192                         err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1193                         if (err)
1194                                 return err;
1195                 }
1196
1197                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
1198                 mss = skb_shinfo(skb)->gso_size;
1199                 iph = ip_hdr(skb);
1200                 iph->tot_len = 0;
1201                 iph->check = 0;
1202                 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
1203                                                          iph->daddr, 0,
1204                                                          IPPROTO_TCP, 0);
1205                 ipcss = skb_network_offset(skb);
1206                 ipcso = (void *)&(iph->check) - (void *)skb->data;
1207                 ipcse = skb_transport_offset(skb) - 1;
1208                 tucss = skb_transport_offset(skb);
1209                 tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
1210                 tucse = 0;
1211
1212                 i = adapter->tx_ring.next_to_use;
1213                 context_desc = IXGB_CONTEXT_DESC(adapter->tx_ring, i);
1214                 buffer_info = &adapter->tx_ring.buffer_info[i];
1215                 WARN_ON(buffer_info->dma != 0);
1216
1217                 context_desc->ipcss = ipcss;
1218                 context_desc->ipcso = ipcso;
1219                 context_desc->ipcse = cpu_to_le16(ipcse);
1220                 context_desc->tucss = tucss;
1221                 context_desc->tucso = tucso;
1222                 context_desc->tucse = cpu_to_le16(tucse);
1223                 context_desc->mss = cpu_to_le16(mss);
1224                 context_desc->hdr_len = hdr_len;
1225                 context_desc->status = 0;
1226                 context_desc->cmd_type_len = cpu_to_le32(
1227                                                   IXGB_CONTEXT_DESC_TYPE
1228                                                 | IXGB_CONTEXT_DESC_CMD_TSE
1229                                                 | IXGB_CONTEXT_DESC_CMD_IP
1230                                                 | IXGB_CONTEXT_DESC_CMD_TCP
1231                                                 | IXGB_CONTEXT_DESC_CMD_IDE
1232                                                 | (skb->len - (hdr_len)));
1233
1234
1235                 if (++i == adapter->tx_ring.count) i = 0;
1236                 adapter->tx_ring.next_to_use = i;
1237
1238                 return 1;
1239         }
1240
1241         return 0;
1242 }
1243
1244 static bool
1245 ixgb_tx_csum(struct ixgb_adapter *adapter, struct sk_buff *skb)
1246 {
1247         struct ixgb_context_desc *context_desc;
1248         unsigned int i;
1249         u8 css, cso;
1250
1251         if (likely(skb->ip_summed == CHECKSUM_PARTIAL)) {
1252                 struct ixgb_buffer *buffer_info;
1253                 css = skb_transport_offset(skb);
1254                 cso = css + skb->csum_offset;
1255
1256                 i = adapter->tx_ring.next_to_use;
1257                 context_desc = IXGB_CONTEXT_DESC(adapter->tx_ring, i);
1258                 buffer_info = &adapter->tx_ring.buffer_info[i];
1259                 WARN_ON(buffer_info->dma != 0);
1260
1261                 context_desc->tucss = css;
1262                 context_desc->tucso = cso;
1263                 context_desc->tucse = 0;
1264                 /* zero out any previously existing data in one instruction */
1265                 *(u32 *)&(context_desc->ipcss) = 0;
1266                 context_desc->status = 0;
1267                 context_desc->hdr_len = 0;
1268                 context_desc->mss = 0;
1269                 context_desc->cmd_type_len =
1270                         cpu_to_le32(IXGB_CONTEXT_DESC_TYPE
1271                                     | IXGB_TX_DESC_CMD_IDE);
1272
1273                 if (++i == adapter->tx_ring.count) i = 0;
1274                 adapter->tx_ring.next_to_use = i;
1275
1276                 return true;
1277         }
1278
1279         return false;
1280 }
1281
1282 #define IXGB_MAX_TXD_PWR        14
1283 #define IXGB_MAX_DATA_PER_TXD   (1<<IXGB_MAX_TXD_PWR)
1284
1285 static int
1286 ixgb_tx_map(struct ixgb_adapter *adapter, struct sk_buff *skb,
1287             unsigned int first)
1288 {
1289         struct ixgb_desc_ring *tx_ring = &adapter->tx_ring;
1290         struct pci_dev *pdev = adapter->pdev;
1291         struct ixgb_buffer *buffer_info;
1292         int len = skb_headlen(skb);
1293         unsigned int offset = 0, size, count = 0, i;
1294         unsigned int mss = skb_shinfo(skb)->gso_size;
1295         unsigned int nr_frags = skb_shinfo(skb)->nr_frags;
1296         unsigned int f;
1297
1298         i = tx_ring->next_to_use;
1299
1300         while (len) {
1301                 buffer_info = &tx_ring->buffer_info[i];
1302                 size = min(len, IXGB_MAX_DATA_PER_TXD);
1303                 /* Workaround for premature desc write-backs
1304                  * in TSO mode.  Append 4-byte sentinel desc */
1305                 if (unlikely(mss && !nr_frags && size == len && size > 8))
1306                         size -= 4;
1307
1308                 buffer_info->length = size;
1309                 WARN_ON(buffer_info->dma != 0);
1310                 buffer_info->time_stamp = jiffies;
1311                 buffer_info->mapped_as_page = false;
1312                 buffer_info->dma = dma_map_single(&pdev->dev,
1313                                                   skb->data + offset,
1314                                                   size, DMA_TO_DEVICE);
1315                 if (dma_mapping_error(&pdev->dev, buffer_info->dma))
1316                         goto dma_error;
1317                 buffer_info->next_to_watch = 0;
1318
1319                 len -= size;
1320                 offset += size;
1321                 count++;
1322                 if (len) {
1323                         i++;
1324                         if (i == tx_ring->count)
1325                                 i = 0;
1326                 }
1327         }
1328
1329         for (f = 0; f < nr_frags; f++) {
1330                 struct skb_frag_struct *frag;
1331
1332                 frag = &skb_shinfo(skb)->frags[f];
1333                 len = frag->size;
1334                 offset = frag->page_offset;
1335
1336                 while (len) {
1337                         i++;
1338                         if (i == tx_ring->count)
1339                                 i = 0;
1340
1341                         buffer_info = &tx_ring->buffer_info[i];
1342                         size = min(len, IXGB_MAX_DATA_PER_TXD);
1343
1344                         /* Workaround for premature desc write-backs
1345                          * in TSO mode.  Append 4-byte sentinel desc */
1346                         if (unlikely(mss && (f == (nr_frags - 1))
1347                                      && size == len && size > 8))
1348                                 size -= 4;
1349
1350                         buffer_info->length = size;
1351                         buffer_info->time_stamp = jiffies;
1352                         buffer_info->mapped_as_page = true;
1353                         buffer_info->dma =
1354                                 dma_map_page(&pdev->dev, frag->page,
1355                                              offset, size, DMA_TO_DEVICE);
1356                         if (dma_mapping_error(&pdev->dev, buffer_info->dma))
1357                                 goto dma_error;
1358                         buffer_info->next_to_watch = 0;
1359
1360                         len -= size;
1361                         offset += size;
1362                         count++;
1363                 }
1364         }
1365         tx_ring->buffer_info[i].skb = skb;
1366         tx_ring->buffer_info[first].next_to_watch = i;
1367
1368         return count;
1369
1370 dma_error:
1371         dev_err(&pdev->dev, "TX DMA map failed\n");
1372         buffer_info->dma = 0;
1373         if (count)
1374                 count--;
1375
1376         while (count--) {
1377                 if (i==0)
1378                         i += tx_ring->count;
1379                 i--;
1380                 buffer_info = &tx_ring->buffer_info[i];
1381                 ixgb_unmap_and_free_tx_resource(adapter, buffer_info);
1382         }
1383
1384         return 0;
1385 }
1386
1387 static void
1388 ixgb_tx_queue(struct ixgb_adapter *adapter, int count, int vlan_id,int tx_flags)
1389 {
1390         struct ixgb_desc_ring *tx_ring = &adapter->tx_ring;
1391         struct ixgb_tx_desc *tx_desc = NULL;
1392         struct ixgb_buffer *buffer_info;
1393         u32 cmd_type_len = adapter->tx_cmd_type;
1394         u8 status = 0;
1395         u8 popts = 0;
1396         unsigned int i;
1397
1398         if (tx_flags & IXGB_TX_FLAGS_TSO) {
1399                 cmd_type_len |= IXGB_TX_DESC_CMD_TSE;
1400                 popts |= (IXGB_TX_DESC_POPTS_IXSM | IXGB_TX_DESC_POPTS_TXSM);
1401         }
1402
1403         if (tx_flags & IXGB_TX_FLAGS_CSUM)
1404                 popts |= IXGB_TX_DESC_POPTS_TXSM;
1405
1406         if (tx_flags & IXGB_TX_FLAGS_VLAN)
1407                 cmd_type_len |= IXGB_TX_DESC_CMD_VLE;
1408
1409         i = tx_ring->next_to_use;
1410
1411         while (count--) {
1412                 buffer_info = &tx_ring->buffer_info[i];
1413                 tx_desc = IXGB_TX_DESC(*tx_ring, i);
1414                 tx_desc->buff_addr = cpu_to_le64(buffer_info->dma);
1415                 tx_desc->cmd_type_len =
1416                         cpu_to_le32(cmd_type_len | buffer_info->length);
1417                 tx_desc->status = status;
1418                 tx_desc->popts = popts;
1419                 tx_desc->vlan = cpu_to_le16(vlan_id);
1420
1421                 if (++i == tx_ring->count) i = 0;
1422         }
1423
1424         tx_desc->cmd_type_len |=
1425                 cpu_to_le32(IXGB_TX_DESC_CMD_EOP | IXGB_TX_DESC_CMD_RS);
1426
1427         /* Force memory writes to complete before letting h/w
1428          * know there are new descriptors to fetch.  (Only
1429          * applicable for weak-ordered memory model archs,
1430          * such as IA-64). */
1431         wmb();
1432
1433         tx_ring->next_to_use = i;
1434         IXGB_WRITE_REG(&adapter->hw, TDT, i);
1435 }
1436
1437 static int __ixgb_maybe_stop_tx(struct net_device *netdev, int size)
1438 {
1439         struct ixgb_adapter *adapter = netdev_priv(netdev);
1440         struct ixgb_desc_ring *tx_ring = &adapter->tx_ring;
1441
1442         netif_stop_queue(netdev);
1443         /* Herbert's original patch had:
1444          *  smp_mb__after_netif_stop_queue();
1445          * but since that doesn't exist yet, just open code it. */
1446         smp_mb();
1447
1448         /* We need to check again in a case another CPU has just
1449          * made room available. */
1450         if (likely(IXGB_DESC_UNUSED(tx_ring) < size))
1451                 return -EBUSY;
1452
1453         /* A reprieve! */
1454         netif_start_queue(netdev);
1455         ++adapter->restart_queue;
1456         return 0;
1457 }
1458
1459 static int ixgb_maybe_stop_tx(struct net_device *netdev,
1460                               struct ixgb_desc_ring *tx_ring, int size)
1461 {
1462         if (likely(IXGB_DESC_UNUSED(tx_ring) >= size))
1463                 return 0;
1464         return __ixgb_maybe_stop_tx(netdev, size);
1465 }
1466
1467
1468 /* Tx Descriptors needed, worst case */
1469 #define TXD_USE_COUNT(S) (((S) >> IXGB_MAX_TXD_PWR) + \
1470                          (((S) & (IXGB_MAX_DATA_PER_TXD - 1)) ? 1 : 0))
1471 #define DESC_NEEDED TXD_USE_COUNT(IXGB_MAX_DATA_PER_TXD) /* skb->date */ + \
1472         MAX_SKB_FRAGS * TXD_USE_COUNT(PAGE_SIZE) + 1 /* for context */ \
1473         + 1 /* one more needed for sentinel TSO workaround */
1474
1475 static netdev_tx_t
1476 ixgb_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
1477 {
1478         struct ixgb_adapter *adapter = netdev_priv(netdev);
1479         unsigned int first;
1480         unsigned int tx_flags = 0;
1481         int vlan_id = 0;
1482         int count = 0;
1483         int tso;
1484
1485         if (test_bit(__IXGB_DOWN, &adapter->flags)) {
1486                 dev_kfree_skb(skb);
1487                 return NETDEV_TX_OK;
1488         }
1489
1490         if (skb->len <= 0) {
1491                 dev_kfree_skb(skb);
1492                 return NETDEV_TX_OK;
1493         }
1494
1495         if (unlikely(ixgb_maybe_stop_tx(netdev, &adapter->tx_ring,
1496                      DESC_NEEDED)))
1497                 return NETDEV_TX_BUSY;
1498
1499         if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
1500                 tx_flags |= IXGB_TX_FLAGS_VLAN;
1501                 vlan_id = vlan_tx_tag_get(skb);
1502         }
1503
1504         first = adapter->tx_ring.next_to_use;
1505
1506         tso = ixgb_tso(adapter, skb);
1507         if (tso < 0) {
1508                 dev_kfree_skb(skb);
1509                 return NETDEV_TX_OK;
1510         }
1511
1512         if (likely(tso))
1513                 tx_flags |= IXGB_TX_FLAGS_TSO;
1514         else if (ixgb_tx_csum(adapter, skb))
1515                 tx_flags |= IXGB_TX_FLAGS_CSUM;
1516
1517         count = ixgb_tx_map(adapter, skb, first);
1518
1519         if (count) {
1520                 ixgb_tx_queue(adapter, count, vlan_id, tx_flags);
1521                 /* Make sure there is space in the ring for the next send. */
1522                 ixgb_maybe_stop_tx(netdev, &adapter->tx_ring, DESC_NEEDED);
1523
1524         } else {
1525                 dev_kfree_skb_any(skb);
1526                 adapter->tx_ring.buffer_info[first].time_stamp = 0;
1527                 adapter->tx_ring.next_to_use = first;
1528         }
1529
1530         return NETDEV_TX_OK;
1531 }
1532
1533 /**
1534  * ixgb_tx_timeout - Respond to a Tx Hang
1535  * @netdev: network interface device structure
1536  **/
1537
1538 static void
1539 ixgb_tx_timeout(struct net_device *netdev)
1540 {
1541         struct ixgb_adapter *adapter = netdev_priv(netdev);
1542
1543         /* Do the reset outside of interrupt context */
1544         schedule_work(&adapter->tx_timeout_task);
1545 }
1546
1547 static void
1548 ixgb_tx_timeout_task(struct work_struct *work)
1549 {
1550         struct ixgb_adapter *adapter =
1551                 container_of(work, struct ixgb_adapter, tx_timeout_task);
1552
1553         adapter->tx_timeout_count++;
1554         ixgb_down(adapter, true);
1555         ixgb_up(adapter);
1556 }
1557
1558 /**
1559  * ixgb_get_stats - Get System Network Statistics
1560  * @netdev: network interface device structure
1561  *
1562  * Returns the address of the device statistics structure.
1563  * The statistics are actually updated from the timer callback.
1564  **/
1565
1566 static struct net_device_stats *
1567 ixgb_get_stats(struct net_device *netdev)
1568 {
1569         return &netdev->stats;
1570 }
1571
1572 /**
1573  * ixgb_change_mtu - Change the Maximum Transfer Unit
1574  * @netdev: network interface device structure
1575  * @new_mtu: new value for maximum frame size
1576  *
1577  * Returns 0 on success, negative on failure
1578  **/
1579
1580 static int
1581 ixgb_change_mtu(struct net_device *netdev, int new_mtu)
1582 {
1583         struct ixgb_adapter *adapter = netdev_priv(netdev);
1584         int max_frame = new_mtu + ENET_HEADER_SIZE + ENET_FCS_LENGTH;
1585         int old_max_frame = netdev->mtu + ENET_HEADER_SIZE + ENET_FCS_LENGTH;
1586
1587         /* MTU < 68 is an error for IPv4 traffic, just don't allow it */
1588         if ((new_mtu < 68) ||
1589             (max_frame > IXGB_MAX_JUMBO_FRAME_SIZE + ENET_FCS_LENGTH)) {
1590                 netif_err(adapter, probe, adapter->netdev,
1591                           "Invalid MTU setting %d\n", new_mtu);
1592                 return -EINVAL;
1593         }
1594
1595         if (old_max_frame == max_frame)
1596                 return 0;
1597
1598         if (netif_running(netdev))
1599                 ixgb_down(adapter, true);
1600
1601         adapter->rx_buffer_len = max_frame + 8; /* + 8 for errata */
1602
1603         netdev->mtu = new_mtu;
1604
1605         if (netif_running(netdev))
1606                 ixgb_up(adapter);
1607
1608         return 0;
1609 }
1610
1611 /**
1612  * ixgb_update_stats - Update the board statistics counters.
1613  * @adapter: board private structure
1614  **/
1615
1616 void
1617 ixgb_update_stats(struct ixgb_adapter *adapter)
1618 {
1619         struct net_device *netdev = adapter->netdev;
1620         struct pci_dev *pdev = adapter->pdev;
1621
1622         /* Prevent stats update while adapter is being reset */
1623         if (pci_channel_offline(pdev))
1624                 return;
1625
1626         if ((netdev->flags & IFF_PROMISC) || (netdev->flags & IFF_ALLMULTI) ||
1627            (netdev_mc_count(netdev) > IXGB_MAX_NUM_MULTICAST_ADDRESSES)) {
1628                 u64 multi = IXGB_READ_REG(&adapter->hw, MPRCL);
1629                 u32 bcast_l = IXGB_READ_REG(&adapter->hw, BPRCL);
1630                 u32 bcast_h = IXGB_READ_REG(&adapter->hw, BPRCH);
1631                 u64 bcast = ((u64)bcast_h << 32) | bcast_l;
1632
1633                 multi |= ((u64)IXGB_READ_REG(&adapter->hw, MPRCH) << 32);
1634                 /* fix up multicast stats by removing broadcasts */
1635                 if (multi >= bcast)
1636                         multi -= bcast;
1637
1638                 adapter->stats.mprcl += (multi & 0xFFFFFFFF);
1639                 adapter->stats.mprch += (multi >> 32);
1640                 adapter->stats.bprcl += bcast_l;
1641                 adapter->stats.bprch += bcast_h;
1642         } else {
1643                 adapter->stats.mprcl += IXGB_READ_REG(&adapter->hw, MPRCL);
1644                 adapter->stats.mprch += IXGB_READ_REG(&adapter->hw, MPRCH);
1645                 adapter->stats.bprcl += IXGB_READ_REG(&adapter->hw, BPRCL);
1646                 adapter->stats.bprch += IXGB_READ_REG(&adapter->hw, BPRCH);
1647         }
1648         adapter->stats.tprl += IXGB_READ_REG(&adapter->hw, TPRL);
1649         adapter->stats.tprh += IXGB_READ_REG(&adapter->hw, TPRH);
1650         adapter->stats.gprcl += IXGB_READ_REG(&adapter->hw, GPRCL);
1651         adapter->stats.gprch += IXGB_READ_REG(&adapter->hw, GPRCH);
1652         adapter->stats.uprcl += IXGB_READ_REG(&adapter->hw, UPRCL);
1653         adapter->stats.uprch += IXGB_READ_REG(&adapter->hw, UPRCH);
1654         adapter->stats.vprcl += IXGB_READ_REG(&adapter->hw, VPRCL);
1655         adapter->stats.vprch += IXGB_READ_REG(&adapter->hw, VPRCH);
1656         adapter->stats.jprcl += IXGB_READ_REG(&adapter->hw, JPRCL);
1657         adapter->stats.jprch += IXGB_READ_REG(&adapter->hw, JPRCH);
1658         adapter->stats.gorcl += IXGB_READ_REG(&adapter->hw, GORCL);
1659         adapter->stats.gorch += IXGB_READ_REG(&adapter->hw, GORCH);
1660         adapter->stats.torl += IXGB_READ_REG(&adapter->hw, TORL);
1661         adapter->stats.torh += IXGB_READ_REG(&adapter->hw, TORH);
1662         adapter->stats.rnbc += IXGB_READ_REG(&adapter->hw, RNBC);
1663         adapter->stats.ruc += IXGB_READ_REG(&adapter->hw, RUC);
1664         adapter->stats.roc += IXGB_READ_REG(&adapter->hw, ROC);
1665         adapter->stats.rlec += IXGB_READ_REG(&adapter->hw, RLEC);
1666         adapter->stats.crcerrs += IXGB_READ_REG(&adapter->hw, CRCERRS);
1667         adapter->stats.icbc += IXGB_READ_REG(&adapter->hw, ICBC);
1668         adapter->stats.ecbc += IXGB_READ_REG(&adapter->hw, ECBC);
1669         adapter->stats.mpc += IXGB_READ_REG(&adapter->hw, MPC);
1670         adapter->stats.tptl += IXGB_READ_REG(&adapter->hw, TPTL);
1671         adapter->stats.tpth += IXGB_READ_REG(&adapter->hw, TPTH);
1672         adapter->stats.gptcl += IXGB_READ_REG(&adapter->hw, GPTCL);
1673         adapter->stats.gptch += IXGB_READ_REG(&adapter->hw, GPTCH);
1674         adapter->stats.bptcl += IXGB_READ_REG(&adapter->hw, BPTCL);
1675         adapter->stats.bptch += IXGB_READ_REG(&adapter->hw, BPTCH);
1676         adapter->stats.mptcl += IXGB_READ_REG(&adapter->hw, MPTCL);
1677         adapter->stats.mptch += IXGB_READ_REG(&adapter->hw, MPTCH);
1678         adapter->stats.uptcl += IXGB_READ_REG(&adapter->hw, UPTCL);
1679         adapter->stats.uptch += IXGB_READ_REG(&adapter->hw, UPTCH);
1680         adapter->stats.vptcl += IXGB_READ_REG(&adapter->hw, VPTCL);
1681         adapter->stats.vptch += IXGB_READ_REG(&adapter->hw, VPTCH);
1682         adapter->stats.jptcl += IXGB_READ_REG(&adapter->hw, JPTCL);
1683         adapter->stats.jptch += IXGB_READ_REG(&adapter->hw, JPTCH);
1684         adapter->stats.gotcl += IXGB_READ_REG(&adapter->hw, GOTCL);
1685         adapter->stats.gotch += IXGB_READ_REG(&adapter->hw, GOTCH);
1686         adapter->stats.totl += IXGB_READ_REG(&adapter->hw, TOTL);
1687         adapter->stats.toth += IXGB_READ_REG(&adapter->hw, TOTH);
1688         adapter->stats.dc += IXGB_READ_REG(&adapter->hw, DC);
1689         adapter->stats.plt64c += IXGB_READ_REG(&adapter->hw, PLT64C);
1690         adapter->stats.tsctc += IXGB_READ_REG(&adapter->hw, TSCTC);
1691         adapter->stats.tsctfc += IXGB_READ_REG(&adapter->hw, TSCTFC);
1692         adapter->stats.ibic += IXGB_READ_REG(&adapter->hw, IBIC);
1693         adapter->stats.rfc += IXGB_READ_REG(&adapter->hw, RFC);
1694         adapter->stats.lfc += IXGB_READ_REG(&adapter->hw, LFC);
1695         adapter->stats.pfrc += IXGB_READ_REG(&adapter->hw, PFRC);
1696         adapter->stats.pftc += IXGB_READ_REG(&adapter->hw, PFTC);
1697         adapter->stats.mcfrc += IXGB_READ_REG(&adapter->hw, MCFRC);
1698         adapter->stats.mcftc += IXGB_READ_REG(&adapter->hw, MCFTC);
1699         adapter->stats.xonrxc += IXGB_READ_REG(&adapter->hw, XONRXC);
1700         adapter->stats.xontxc += IXGB_READ_REG(&adapter->hw, XONTXC);
1701         adapter->stats.xoffrxc += IXGB_READ_REG(&adapter->hw, XOFFRXC);
1702         adapter->stats.xofftxc += IXGB_READ_REG(&adapter->hw, XOFFTXC);
1703         adapter->stats.rjc += IXGB_READ_REG(&adapter->hw, RJC);
1704
1705         /* Fill out the OS statistics structure */
1706
1707         netdev->stats.rx_packets = adapter->stats.gprcl;
1708         netdev->stats.tx_packets = adapter->stats.gptcl;
1709         netdev->stats.rx_bytes = adapter->stats.gorcl;
1710         netdev->stats.tx_bytes = adapter->stats.gotcl;
1711         netdev->stats.multicast = adapter->stats.mprcl;
1712         netdev->stats.collisions = 0;
1713
1714         /* ignore RLEC as it reports errors for padded (<64bytes) frames
1715          * with a length in the type/len field */
1716         netdev->stats.rx_errors =
1717             /* adapter->stats.rnbc + */ adapter->stats.crcerrs +
1718             adapter->stats.ruc +
1719             adapter->stats.roc /*+ adapter->stats.rlec */  +
1720             adapter->stats.icbc +
1721             adapter->stats.ecbc + adapter->stats.mpc;
1722
1723         /* see above
1724          * netdev->stats.rx_length_errors = adapter->stats.rlec;
1725          */
1726
1727         netdev->stats.rx_crc_errors = adapter->stats.crcerrs;
1728         netdev->stats.rx_fifo_errors = adapter->stats.mpc;
1729         netdev->stats.rx_missed_errors = adapter->stats.mpc;
1730         netdev->stats.rx_over_errors = adapter->stats.mpc;
1731
1732         netdev->stats.tx_errors = 0;
1733         netdev->stats.rx_frame_errors = 0;
1734         netdev->stats.tx_aborted_errors = 0;
1735         netdev->stats.tx_carrier_errors = 0;
1736         netdev->stats.tx_fifo_errors = 0;
1737         netdev->stats.tx_heartbeat_errors = 0;
1738         netdev->stats.tx_window_errors = 0;
1739 }
1740
1741 #define IXGB_MAX_INTR 10
1742 /**
1743  * ixgb_intr - Interrupt Handler
1744  * @irq: interrupt number
1745  * @data: pointer to a network interface device structure
1746  **/
1747
1748 static irqreturn_t
1749 ixgb_intr(int irq, void *data)
1750 {
1751         struct net_device *netdev = data;
1752         struct ixgb_adapter *adapter = netdev_priv(netdev);
1753         struct ixgb_hw *hw = &adapter->hw;
1754         u32 icr = IXGB_READ_REG(hw, ICR);
1755
1756         if (unlikely(!icr))
1757                 return IRQ_NONE;  /* Not our interrupt */
1758
1759         if (unlikely(icr & (IXGB_INT_RXSEQ | IXGB_INT_LSC)))
1760                 if (!test_bit(__IXGB_DOWN, &adapter->flags))
1761                         mod_timer(&adapter->watchdog_timer, jiffies);
1762
1763         if (napi_schedule_prep(&adapter->napi)) {
1764
1765                 /* Disable interrupts and register for poll. The flush
1766                   of the posted write is intentionally left out.
1767                 */
1768
1769                 IXGB_WRITE_REG(&adapter->hw, IMC, ~0);
1770                 __napi_schedule(&adapter->napi);
1771         }
1772         return IRQ_HANDLED;
1773 }
1774
1775 /**
1776  * ixgb_clean - NAPI Rx polling callback
1777  * @adapter: board private structure
1778  **/
1779
1780 static int
1781 ixgb_clean(struct napi_struct *napi, int budget)
1782 {
1783         struct ixgb_adapter *adapter = container_of(napi, struct ixgb_adapter, napi);
1784         int work_done = 0;
1785
1786         ixgb_clean_tx_irq(adapter);
1787         ixgb_clean_rx_irq(adapter, &work_done, budget);
1788
1789         /* If budget not fully consumed, exit the polling mode */
1790         if (work_done < budget) {
1791                 napi_complete(napi);
1792                 if (!test_bit(__IXGB_DOWN, &adapter->flags))
1793                         ixgb_irq_enable(adapter);
1794         }
1795
1796         return work_done;
1797 }
1798
1799 /**
1800  * ixgb_clean_tx_irq - Reclaim resources after transmit completes
1801  * @adapter: board private structure
1802  **/
1803
1804 static bool
1805 ixgb_clean_tx_irq(struct ixgb_adapter *adapter)
1806 {
1807         struct ixgb_desc_ring *tx_ring = &adapter->tx_ring;
1808         struct net_device *netdev = adapter->netdev;
1809         struct ixgb_tx_desc *tx_desc, *eop_desc;
1810         struct ixgb_buffer *buffer_info;
1811         unsigned int i, eop;
1812         bool cleaned = false;
1813
1814         i = tx_ring->next_to_clean;
1815         eop = tx_ring->buffer_info[i].next_to_watch;
1816         eop_desc = IXGB_TX_DESC(*tx_ring, eop);
1817
1818         while (eop_desc->status & IXGB_TX_DESC_STATUS_DD) {
1819
1820                 rmb(); /* read buffer_info after eop_desc */
1821                 for (cleaned = false; !cleaned; ) {
1822                         tx_desc = IXGB_TX_DESC(*tx_ring, i);
1823                         buffer_info = &tx_ring->buffer_info[i];
1824
1825                         if (tx_desc->popts &
1826                            (IXGB_TX_DESC_POPTS_TXSM |
1827                             IXGB_TX_DESC_POPTS_IXSM))
1828                                 adapter->hw_csum_tx_good++;
1829
1830                         ixgb_unmap_and_free_tx_resource(adapter, buffer_info);
1831
1832                         *(u32 *)&(tx_desc->status) = 0;
1833
1834                         cleaned = (i == eop);
1835                         if (++i == tx_ring->count) i = 0;
1836                 }
1837
1838                 eop = tx_ring->buffer_info[i].next_to_watch;
1839                 eop_desc = IXGB_TX_DESC(*tx_ring, eop);
1840         }
1841
1842         tx_ring->next_to_clean = i;
1843
1844         if (unlikely(cleaned && netif_carrier_ok(netdev) &&
1845                      IXGB_DESC_UNUSED(tx_ring) >= DESC_NEEDED)) {
1846                 /* Make sure that anybody stopping the queue after this
1847                  * sees the new next_to_clean. */
1848                 smp_mb();
1849
1850                 if (netif_queue_stopped(netdev) &&
1851                     !(test_bit(__IXGB_DOWN, &adapter->flags))) {
1852                         netif_wake_queue(netdev);
1853                         ++adapter->restart_queue;
1854                 }
1855         }
1856
1857         if (adapter->detect_tx_hung) {
1858                 /* detect a transmit hang in hardware, this serializes the
1859                  * check with the clearing of time_stamp and movement of i */
1860                 adapter->detect_tx_hung = false;
1861                 if (tx_ring->buffer_info[eop].time_stamp &&
1862                    time_after(jiffies, tx_ring->buffer_info[eop].time_stamp + HZ)
1863                    && !(IXGB_READ_REG(&adapter->hw, STATUS) &
1864                         IXGB_STATUS_TXOFF)) {
1865                         /* detected Tx unit hang */
1866                         netif_err(adapter, drv, adapter->netdev,
1867                                   "Detected Tx Unit Hang\n"
1868                                   "  TDH                  <%x>\n"
1869                                   "  TDT                  <%x>\n"
1870                                   "  next_to_use          <%x>\n"
1871                                   "  next_to_clean        <%x>\n"
1872                                   "buffer_info[next_to_clean]\n"
1873                                   "  time_stamp           <%lx>\n"
1874                                   "  next_to_watch        <%x>\n"
1875                                   "  jiffies              <%lx>\n"
1876                                   "  next_to_watch.status <%x>\n",
1877                                   IXGB_READ_REG(&adapter->hw, TDH),
1878                                   IXGB_READ_REG(&adapter->hw, TDT),
1879                                   tx_ring->next_to_use,
1880                                   tx_ring->next_to_clean,
1881                                   tx_ring->buffer_info[eop].time_stamp,
1882                                   eop,
1883                                   jiffies,
1884                                   eop_desc->status);
1885                         netif_stop_queue(netdev);
1886                 }
1887         }
1888
1889         return cleaned;
1890 }
1891
1892 /**
1893  * ixgb_rx_checksum - Receive Checksum Offload for 82597.
1894  * @adapter: board private structure
1895  * @rx_desc: receive descriptor
1896  * @sk_buff: socket buffer with received data
1897  **/
1898
1899 static void
1900 ixgb_rx_checksum(struct ixgb_adapter *adapter,
1901                  struct ixgb_rx_desc *rx_desc,
1902                  struct sk_buff *skb)
1903 {
1904         /* Ignore Checksum bit is set OR
1905          * TCP Checksum has not been calculated
1906          */
1907         if ((rx_desc->status & IXGB_RX_DESC_STATUS_IXSM) ||
1908            (!(rx_desc->status & IXGB_RX_DESC_STATUS_TCPCS))) {
1909                 skb_checksum_none_assert(skb);
1910                 return;
1911         }
1912
1913         /* At this point we know the hardware did the TCP checksum */
1914         /* now look at the TCP checksum error bit */
1915         if (rx_desc->errors & IXGB_RX_DESC_ERRORS_TCPE) {
1916                 /* let the stack verify checksum errors */
1917                 skb_checksum_none_assert(skb);
1918                 adapter->hw_csum_rx_error++;
1919         } else {
1920                 /* TCP checksum is good */
1921                 skb->ip_summed = CHECKSUM_UNNECESSARY;
1922                 adapter->hw_csum_rx_good++;
1923         }
1924 }
1925
1926 /*
1927  * this should improve performance for small packets with large amounts
1928  * of reassembly being done in the stack
1929  */
1930 static void ixgb_check_copybreak(struct net_device *netdev,
1931                                  struct ixgb_buffer *buffer_info,
1932                                  u32 length, struct sk_buff **skb)
1933 {
1934         struct sk_buff *new_skb;
1935
1936         if (length > copybreak)
1937                 return;
1938
1939         new_skb = netdev_alloc_skb_ip_align(netdev, length);
1940         if (!new_skb)
1941                 return;
1942
1943         skb_copy_to_linear_data_offset(new_skb, -NET_IP_ALIGN,
1944                                        (*skb)->data - NET_IP_ALIGN,
1945                                        length + NET_IP_ALIGN);
1946         /* save the skb in buffer_info as good */
1947         buffer_info->skb = *skb;
1948         *skb = new_skb;
1949 }
1950
1951 /**
1952  * ixgb_clean_rx_irq - Send received data up the network stack,
1953  * @adapter: board private structure
1954  **/
1955
1956 static bool
1957 ixgb_clean_rx_irq(struct ixgb_adapter *adapter, int *work_done, int work_to_do)
1958 {
1959         struct ixgb_desc_ring *rx_ring = &adapter->rx_ring;
1960         struct net_device *netdev = adapter->netdev;
1961         struct pci_dev *pdev = adapter->pdev;
1962         struct ixgb_rx_desc *rx_desc, *next_rxd;
1963         struct ixgb_buffer *buffer_info, *next_buffer, *next2_buffer;
1964         u32 length;
1965         unsigned int i, j;
1966         int cleaned_count = 0;
1967         bool cleaned = false;
1968
1969         i = rx_ring->next_to_clean;
1970         rx_desc = IXGB_RX_DESC(*rx_ring, i);
1971         buffer_info = &rx_ring->buffer_info[i];
1972
1973         while (rx_desc->status & IXGB_RX_DESC_STATUS_DD) {
1974                 struct sk_buff *skb;
1975                 u8 status;
1976
1977                 if (*work_done >= work_to_do)
1978                         break;
1979
1980                 (*work_done)++;
1981                 rmb();  /* read descriptor and rx_buffer_info after status DD */
1982                 status = rx_desc->status;
1983                 skb = buffer_info->skb;
1984                 buffer_info->skb = NULL;
1985
1986                 prefetch(skb->data - NET_IP_ALIGN);
1987
1988                 if (++i == rx_ring->count)
1989                         i = 0;
1990                 next_rxd = IXGB_RX_DESC(*rx_ring, i);
1991                 prefetch(next_rxd);
1992
1993                 j = i + 1;
1994                 if (j == rx_ring->count)
1995                         j = 0;
1996                 next2_buffer = &rx_ring->buffer_info[j];
1997                 prefetch(next2_buffer);
1998
1999                 next_buffer = &rx_ring->buffer_info[i];
2000
2001                 cleaned = true;
2002                 cleaned_count++;
2003
2004                 dma_unmap_single(&pdev->dev,
2005                                  buffer_info->dma,
2006                                  buffer_info->length,
2007                                  DMA_FROM_DEVICE);
2008                 buffer_info->dma = 0;
2009
2010                 length = le16_to_cpu(rx_desc->length);
2011                 rx_desc->length = 0;
2012
2013                 if (unlikely(!(status & IXGB_RX_DESC_STATUS_EOP))) {
2014
2015                         /* All receives must fit into a single buffer */
2016
2017                         IXGB_DBG("Receive packet consumed multiple buffers "
2018                                          "length<%x>\n", length);
2019
2020                         dev_kfree_skb_irq(skb);
2021                         goto rxdesc_done;
2022                 }
2023
2024                 if (unlikely(rx_desc->errors &
2025                     (IXGB_RX_DESC_ERRORS_CE | IXGB_RX_DESC_ERRORS_SE |
2026                      IXGB_RX_DESC_ERRORS_P | IXGB_RX_DESC_ERRORS_RXE))) {
2027                         dev_kfree_skb_irq(skb);
2028                         goto rxdesc_done;
2029                 }
2030
2031                 ixgb_check_copybreak(netdev, buffer_info, length, &skb);
2032
2033                 /* Good Receive */
2034                 skb_put(skb, length);
2035
2036                 /* Receive Checksum Offload */
2037                 ixgb_rx_checksum(adapter, rx_desc, skb);
2038
2039                 skb->protocol = eth_type_trans(skb, netdev);
2040                 if (adapter->vlgrp && (status & IXGB_RX_DESC_STATUS_VP)) {
2041                         vlan_hwaccel_receive_skb(skb, adapter->vlgrp,
2042                                                 le16_to_cpu(rx_desc->special));
2043                 } else {
2044                         netif_receive_skb(skb);
2045                 }
2046
2047 rxdesc_done:
2048                 /* clean up descriptor, might be written over by hw */
2049                 rx_desc->status = 0;
2050
2051                 /* return some buffers to hardware, one at a time is too slow */
2052                 if (unlikely(cleaned_count >= IXGB_RX_BUFFER_WRITE)) {
2053                         ixgb_alloc_rx_buffers(adapter, cleaned_count);
2054                         cleaned_count = 0;
2055                 }
2056
2057                 /* use prefetched values */
2058                 rx_desc = next_rxd;
2059                 buffer_info = next_buffer;
2060         }
2061
2062         rx_ring->next_to_clean = i;
2063
2064         cleaned_count = IXGB_DESC_UNUSED(rx_ring);
2065         if (cleaned_count)
2066                 ixgb_alloc_rx_buffers(adapter, cleaned_count);
2067
2068         return cleaned;
2069 }
2070
2071 /**
2072  * ixgb_alloc_rx_buffers - Replace used receive buffers
2073  * @adapter: address of board private structure
2074  **/
2075
2076 static void
2077 ixgb_alloc_rx_buffers(struct ixgb_adapter *adapter, int cleaned_count)
2078 {
2079         struct ixgb_desc_ring *rx_ring = &adapter->rx_ring;
2080         struct net_device *netdev = adapter->netdev;
2081         struct pci_dev *pdev = adapter->pdev;
2082         struct ixgb_rx_desc *rx_desc;
2083         struct ixgb_buffer *buffer_info;
2084         struct sk_buff *skb;
2085         unsigned int i;
2086         long cleancount;
2087
2088         i = rx_ring->next_to_use;
2089         buffer_info = &rx_ring->buffer_info[i];
2090         cleancount = IXGB_DESC_UNUSED(rx_ring);
2091
2092
2093         /* leave three descriptors unused */
2094         while (--cleancount > 2 && cleaned_count--) {
2095                 /* recycle! its good for you */
2096                 skb = buffer_info->skb;
2097                 if (skb) {
2098                         skb_trim(skb, 0);
2099                         goto map_skb;
2100                 }
2101
2102                 skb = netdev_alloc_skb_ip_align(netdev, adapter->rx_buffer_len);
2103                 if (unlikely(!skb)) {
2104                         /* Better luck next round */
2105                         adapter->alloc_rx_buff_failed++;
2106                         break;
2107                 }
2108
2109                 buffer_info->skb = skb;
2110                 buffer_info->length = adapter->rx_buffer_len;
2111 map_skb:
2112                 buffer_info->dma = dma_map_single(&pdev->dev,
2113                                                   skb->data,
2114                                                   adapter->rx_buffer_len,
2115                                                   DMA_FROM_DEVICE);
2116
2117                 rx_desc = IXGB_RX_DESC(*rx_ring, i);
2118                 rx_desc->buff_addr = cpu_to_le64(buffer_info->dma);
2119                 /* guarantee DD bit not set now before h/w gets descriptor
2120                  * this is the rest of the workaround for h/w double
2121                  * writeback. */
2122                 rx_desc->status = 0;
2123
2124
2125                 if (++i == rx_ring->count) i = 0;
2126                 buffer_info = &rx_ring->buffer_info[i];
2127         }
2128
2129         if (likely(rx_ring->next_to_use != i)) {
2130                 rx_ring->next_to_use = i;
2131                 if (unlikely(i-- == 0))
2132                         i = (rx_ring->count - 1);
2133
2134                 /* Force memory writes to complete before letting h/w
2135                  * know there are new descriptors to fetch.  (Only
2136                  * applicable for weak-ordered memory model archs, such
2137                  * as IA-64). */
2138                 wmb();
2139                 IXGB_WRITE_REG(&adapter->hw, RDT, i);
2140         }
2141 }
2142
2143 /**
2144  * ixgb_vlan_rx_register - enables or disables vlan tagging/stripping.
2145  *
2146  * @param netdev network interface device structure
2147  * @param grp indicates to enable or disable tagging/stripping
2148  **/
2149 static void
2150 ixgb_vlan_rx_register(struct net_device *netdev, struct vlan_group *grp)
2151 {
2152         struct ixgb_adapter *adapter = netdev_priv(netdev);
2153         u32 ctrl, rctl;
2154
2155         ixgb_irq_disable(adapter);
2156         adapter->vlgrp = grp;
2157
2158         if (grp) {
2159                 /* enable VLAN tag insert/strip */
2160                 ctrl = IXGB_READ_REG(&adapter->hw, CTRL0);
2161                 ctrl |= IXGB_CTRL0_VME;
2162                 IXGB_WRITE_REG(&adapter->hw, CTRL0, ctrl);
2163
2164                 /* enable VLAN receive filtering */
2165
2166                 rctl = IXGB_READ_REG(&adapter->hw, RCTL);
2167                 rctl &= ~IXGB_RCTL_CFIEN;
2168                 IXGB_WRITE_REG(&adapter->hw, RCTL, rctl);
2169         } else {
2170                 /* disable VLAN tag insert/strip */
2171
2172                 ctrl = IXGB_READ_REG(&adapter->hw, CTRL0);
2173                 ctrl &= ~IXGB_CTRL0_VME;
2174                 IXGB_WRITE_REG(&adapter->hw, CTRL0, ctrl);
2175         }
2176
2177         /* don't enable interrupts unless we are UP */
2178         if (adapter->netdev->flags & IFF_UP)
2179                 ixgb_irq_enable(adapter);
2180 }
2181
2182 static void
2183 ixgb_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
2184 {
2185         struct ixgb_adapter *adapter = netdev_priv(netdev);
2186         u32 vfta, index;
2187
2188         /* add VID to filter table */
2189
2190         index = (vid >> 5) & 0x7F;
2191         vfta = IXGB_READ_REG_ARRAY(&adapter->hw, VFTA, index);
2192         vfta |= (1 << (vid & 0x1F));
2193         ixgb_write_vfta(&adapter->hw, index, vfta);
2194 }
2195
2196 static void
2197 ixgb_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
2198 {
2199         struct ixgb_adapter *adapter = netdev_priv(netdev);
2200         u32 vfta, index;
2201
2202         ixgb_irq_disable(adapter);
2203
2204         vlan_group_set_device(adapter->vlgrp, vid, NULL);
2205
2206         /* don't enable interrupts unless we are UP */
2207         if (adapter->netdev->flags & IFF_UP)
2208                 ixgb_irq_enable(adapter);
2209
2210         /* remove VID from filter table */
2211
2212         index = (vid >> 5) & 0x7F;
2213         vfta = IXGB_READ_REG_ARRAY(&adapter->hw, VFTA, index);
2214         vfta &= ~(1 << (vid & 0x1F));
2215         ixgb_write_vfta(&adapter->hw, index, vfta);
2216 }
2217
2218 static void
2219 ixgb_restore_vlan(struct ixgb_adapter *adapter)
2220 {
2221         ixgb_vlan_rx_register(adapter->netdev, adapter->vlgrp);
2222
2223         if (adapter->vlgrp) {
2224                 u16 vid;
2225                 for (vid = 0; vid < VLAN_N_VID; vid++) {
2226                         if (!vlan_group_get_device(adapter->vlgrp, vid))
2227                                 continue;
2228                         ixgb_vlan_rx_add_vid(adapter->netdev, vid);
2229                 }
2230         }
2231 }
2232
2233 #ifdef CONFIG_NET_POLL_CONTROLLER
2234 /*
2235  * Polling 'interrupt' - used by things like netconsole to send skbs
2236  * without having to re-enable interrupts. It's not called while
2237  * the interrupt routine is executing.
2238  */
2239
2240 static void ixgb_netpoll(struct net_device *dev)
2241 {
2242         struct ixgb_adapter *adapter = netdev_priv(dev);
2243
2244         disable_irq(adapter->pdev->irq);
2245         ixgb_intr(adapter->pdev->irq, dev);
2246         enable_irq(adapter->pdev->irq);
2247 }
2248 #endif
2249
2250 /**
2251  * ixgb_io_error_detected() - called when PCI error is detected
2252  * @pdev    pointer to pci device with error
2253  * @state   pci channel state after error
2254  *
2255  * This callback is called by the PCI subsystem whenever
2256  * a PCI bus error is detected.
2257  */
2258 static pci_ers_result_t ixgb_io_error_detected(struct pci_dev *pdev,
2259                                                enum pci_channel_state state)
2260 {
2261         struct net_device *netdev = pci_get_drvdata(pdev);
2262         struct ixgb_adapter *adapter = netdev_priv(netdev);
2263
2264         netif_device_detach(netdev);
2265
2266         if (state == pci_channel_io_perm_failure)
2267                 return PCI_ERS_RESULT_DISCONNECT;
2268
2269         if (netif_running(netdev))
2270                 ixgb_down(adapter, true);
2271
2272         pci_disable_device(pdev);
2273
2274         /* Request a slot reset. */
2275         return PCI_ERS_RESULT_NEED_RESET;
2276 }
2277
2278 /**
2279  * ixgb_io_slot_reset - called after the pci bus has been reset.
2280  * @pdev    pointer to pci device with error
2281  *
2282  * This callback is called after the PCI bus has been reset.
2283  * Basically, this tries to restart the card from scratch.
2284  * This is a shortened version of the device probe/discovery code,
2285  * it resembles the first-half of the ixgb_probe() routine.
2286  */
2287 static pci_ers_result_t ixgb_io_slot_reset(struct pci_dev *pdev)
2288 {
2289         struct net_device *netdev = pci_get_drvdata(pdev);
2290         struct ixgb_adapter *adapter = netdev_priv(netdev);
2291
2292         if (pci_enable_device(pdev)) {
2293                 netif_err(adapter, probe, adapter->netdev,
2294                           "Cannot re-enable PCI device after reset\n");
2295                 return PCI_ERS_RESULT_DISCONNECT;
2296         }
2297
2298         /* Perform card reset only on one instance of the card */
2299         if (0 != PCI_FUNC (pdev->devfn))
2300                 return PCI_ERS_RESULT_RECOVERED;
2301
2302         pci_set_master(pdev);
2303
2304         netif_carrier_off(netdev);
2305         netif_stop_queue(netdev);
2306         ixgb_reset(adapter);
2307
2308         /* Make sure the EEPROM is good */
2309         if (!ixgb_validate_eeprom_checksum(&adapter->hw)) {
2310                 netif_err(adapter, probe, adapter->netdev,
2311                           "After reset, the EEPROM checksum is not valid\n");
2312                 return PCI_ERS_RESULT_DISCONNECT;
2313         }
2314         ixgb_get_ee_mac_addr(&adapter->hw, netdev->dev_addr);
2315         memcpy(netdev->perm_addr, netdev->dev_addr, netdev->addr_len);
2316
2317         if (!is_valid_ether_addr(netdev->perm_addr)) {
2318                 netif_err(adapter, probe, adapter->netdev,
2319                           "After reset, invalid MAC address\n");
2320                 return PCI_ERS_RESULT_DISCONNECT;
2321         }
2322
2323         return PCI_ERS_RESULT_RECOVERED;
2324 }
2325
2326 /**
2327  * ixgb_io_resume - called when its OK to resume normal operations
2328  * @pdev    pointer to pci device with error
2329  *
2330  * The error recovery driver tells us that its OK to resume
2331  * normal operation. Implementation resembles the second-half
2332  * of the ixgb_probe() routine.
2333  */
2334 static void ixgb_io_resume(struct pci_dev *pdev)
2335 {
2336         struct net_device *netdev = pci_get_drvdata(pdev);
2337         struct ixgb_adapter *adapter = netdev_priv(netdev);
2338
2339         pci_set_master(pdev);
2340
2341         if (netif_running(netdev)) {
2342                 if (ixgb_up(adapter)) {
2343                         pr_err("can't bring device back up after reset\n");
2344                         return;
2345                 }
2346         }
2347
2348         netif_device_attach(netdev);
2349         mod_timer(&adapter->watchdog_timer, jiffies);
2350 }
2351
2352 /* ixgb_main.c */