drivers/net/: use DEFINE_PCI_DEVICE_TABLE()
[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 #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 DEFINE_PCI_DEVICE_TABLE(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 netdev_tx_t ixgb_xmit_frame(struct sk_buff *skb,
85                                    struct net_device *netdev);
86 static struct net_device_stats *ixgb_get_stats(struct net_device *netdev);
87 static int ixgb_change_mtu(struct net_device *netdev, int new_mtu);
88 static int ixgb_set_mac(struct net_device *netdev, void *p);
89 static irqreturn_t ixgb_intr(int irq, void *data);
90 static bool ixgb_clean_tx_irq(struct ixgb_adapter *adapter);
91
92 static int ixgb_clean(struct napi_struct *, int);
93 static bool ixgb_clean_rx_irq(struct ixgb_adapter *, int *, int);
94 static void ixgb_alloc_rx_buffers(struct ixgb_adapter *, int);
95
96 static void ixgb_tx_timeout(struct net_device *dev);
97 static void ixgb_tx_timeout_task(struct work_struct *work);
98
99 static void ixgb_vlan_rx_register(struct net_device *netdev,
100                                   struct vlan_group *grp);
101 static void ixgb_vlan_rx_add_vid(struct net_device *netdev, u16 vid);
102 static void ixgb_vlan_rx_kill_vid(struct net_device *netdev, u16 vid);
103 static void ixgb_restore_vlan(struct ixgb_adapter *adapter);
104
105 #ifdef CONFIG_NET_POLL_CONTROLLER
106 /* for netdump / net console */
107 static void ixgb_netpoll(struct net_device *dev);
108 #endif
109
110 static pci_ers_result_t ixgb_io_error_detected (struct pci_dev *pdev,
111                              enum pci_channel_state state);
112 static pci_ers_result_t ixgb_io_slot_reset (struct pci_dev *pdev);
113 static void ixgb_io_resume (struct pci_dev *pdev);
114
115 static struct pci_error_handlers ixgb_err_handler = {
116         .error_detected = ixgb_io_error_detected,
117         .slot_reset = ixgb_io_slot_reset,
118         .resume = ixgb_io_resume,
119 };
120
121 static struct pci_driver ixgb_driver = {
122         .name     = ixgb_driver_name,
123         .id_table = ixgb_pci_tbl,
124         .probe    = ixgb_probe,
125         .remove   = __devexit_p(ixgb_remove),
126         .err_handler = &ixgb_err_handler
127 };
128
129 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
130 MODULE_DESCRIPTION("Intel(R) PRO/10GbE Network Driver");
131 MODULE_LICENSE("GPL");
132 MODULE_VERSION(DRV_VERSION);
133
134 #define DEFAULT_DEBUG_LEVEL_SHIFT 3
135 static int debug = DEFAULT_DEBUG_LEVEL_SHIFT;
136 module_param(debug, int, 0);
137 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
138
139 /**
140  * ixgb_init_module - Driver Registration Routine
141  *
142  * ixgb_init_module is the first routine called when the driver is
143  * loaded. All it does is register with the PCI subsystem.
144  **/
145
146 static int __init
147 ixgb_init_module(void)
148 {
149         printk(KERN_INFO "%s - version %s\n",
150                ixgb_driver_string, ixgb_driver_version);
151
152         printk(KERN_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                 DPRINTK(PROBE, ERR,
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                 DPRINTK(PROBE, ERR, "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         if (!(err = pci_set_dma_mask(pdev, DMA_BIT_MASK(64))) &&
372             !(err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64)))) {
373                 pci_using_dac = 1;
374         } else {
375                 if ((err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32))) ||
376                     (err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32)))) {
377                         printk(KERN_ERR
378                          "ixgb: No usable DMA configuration, aborting\n");
379                         goto err_dma_mask;
380                 }
381                 pci_using_dac = 0;
382         }
383
384         err = pci_request_regions(pdev, ixgb_driver_name);
385         if (err)
386                 goto err_request_regions;
387
388         pci_set_master(pdev);
389
390         netdev = alloc_etherdev(sizeof(struct ixgb_adapter));
391         if (!netdev) {
392                 err = -ENOMEM;
393                 goto err_alloc_etherdev;
394         }
395
396         SET_NETDEV_DEV(netdev, &pdev->dev);
397
398         pci_set_drvdata(pdev, netdev);
399         adapter = netdev_priv(netdev);
400         adapter->netdev = netdev;
401         adapter->pdev = pdev;
402         adapter->hw.back = adapter;
403         adapter->msg_enable = netif_msg_init(debug, DEFAULT_DEBUG_LEVEL_SHIFT);
404
405         adapter->hw.hw_addr = pci_ioremap_bar(pdev, BAR_0);
406         if (!adapter->hw.hw_addr) {
407                 err = -EIO;
408                 goto err_ioremap;
409         }
410
411         for (i = BAR_1; i <= BAR_5; i++) {
412                 if (pci_resource_len(pdev, i) == 0)
413                         continue;
414                 if (pci_resource_flags(pdev, i) & IORESOURCE_IO) {
415                         adapter->hw.io_base = pci_resource_start(pdev, i);
416                         break;
417                 }
418         }
419
420         netdev->netdev_ops = &ixgb_netdev_ops;
421         ixgb_set_ethtool_ops(netdev);
422         netdev->watchdog_timeo = 5 * HZ;
423         netif_napi_add(netdev, &adapter->napi, ixgb_clean, 64);
424
425         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
426
427         adapter->bd_number = cards_found;
428         adapter->link_speed = 0;
429         adapter->link_duplex = 0;
430
431         /* setup the private structure */
432
433         err = ixgb_sw_init(adapter);
434         if (err)
435                 goto err_sw_init;
436
437         netdev->features = NETIF_F_SG |
438                            NETIF_F_HW_CSUM |
439                            NETIF_F_HW_VLAN_TX |
440                            NETIF_F_HW_VLAN_RX |
441                            NETIF_F_HW_VLAN_FILTER;
442         netdev->features |= NETIF_F_TSO;
443
444         if (pci_using_dac)
445                 netdev->features |= NETIF_F_HIGHDMA;
446
447         /* make sure the EEPROM is good */
448
449         if (!ixgb_validate_eeprom_checksum(&adapter->hw)) {
450                 DPRINTK(PROBE, ERR, "The EEPROM Checksum Is Not Valid\n");
451                 err = -EIO;
452                 goto err_eeprom;
453         }
454
455         ixgb_get_ee_mac_addr(&adapter->hw, netdev->dev_addr);
456         memcpy(netdev->perm_addr, netdev->dev_addr, netdev->addr_len);
457
458         if (!is_valid_ether_addr(netdev->perm_addr)) {
459                 DPRINTK(PROBE, ERR, "Invalid MAC Address\n");
460                 err = -EIO;
461                 goto err_eeprom;
462         }
463
464         adapter->part_num = ixgb_get_ee_pba_number(&adapter->hw);
465
466         init_timer(&adapter->watchdog_timer);
467         adapter->watchdog_timer.function = &ixgb_watchdog;
468         adapter->watchdog_timer.data = (unsigned long)adapter;
469
470         INIT_WORK(&adapter->tx_timeout_task, ixgb_tx_timeout_task);
471
472         strcpy(netdev->name, "eth%d");
473         err = register_netdev(netdev);
474         if (err)
475                 goto err_register;
476
477         /* carrier off reporting is important to ethtool even BEFORE open */
478         netif_carrier_off(netdev);
479
480         DPRINTK(PROBE, INFO, "Intel(R) PRO/10GbE Network Connection\n");
481         ixgb_check_options(adapter);
482         /* reset the hardware with the new settings */
483
484         ixgb_reset(adapter);
485
486         cards_found++;
487         return 0;
488
489 err_register:
490 err_sw_init:
491 err_eeprom:
492         iounmap(adapter->hw.hw_addr);
493 err_ioremap:
494         free_netdev(netdev);
495 err_alloc_etherdev:
496         pci_release_regions(pdev);
497 err_request_regions:
498 err_dma_mask:
499         pci_disable_device(pdev);
500         return err;
501 }
502
503 /**
504  * ixgb_remove - Device Removal Routine
505  * @pdev: PCI device information struct
506  *
507  * ixgb_remove is called by the PCI subsystem to alert the driver
508  * that it should release a PCI device.  The could be caused by a
509  * Hot-Plug event, or because the driver is going to be removed from
510  * memory.
511  **/
512
513 static void __devexit
514 ixgb_remove(struct pci_dev *pdev)
515 {
516         struct net_device *netdev = pci_get_drvdata(pdev);
517         struct ixgb_adapter *adapter = netdev_priv(netdev);
518
519         flush_scheduled_work();
520
521         unregister_netdev(netdev);
522
523         iounmap(adapter->hw.hw_addr);
524         pci_release_regions(pdev);
525
526         free_netdev(netdev);
527 }
528
529 /**
530  * ixgb_sw_init - Initialize general software structures (struct ixgb_adapter)
531  * @adapter: board private structure to initialize
532  *
533  * ixgb_sw_init initializes the Adapter private data structure.
534  * Fields are initialized based on PCI device information and
535  * OS network device settings (MTU size).
536  **/
537
538 static int __devinit
539 ixgb_sw_init(struct ixgb_adapter *adapter)
540 {
541         struct ixgb_hw *hw = &adapter->hw;
542         struct net_device *netdev = adapter->netdev;
543         struct pci_dev *pdev = adapter->pdev;
544
545         /* PCI config space info */
546
547         hw->vendor_id = pdev->vendor;
548         hw->device_id = pdev->device;
549         hw->subsystem_vendor_id = pdev->subsystem_vendor;
550         hw->subsystem_id = pdev->subsystem_device;
551
552         hw->max_frame_size = netdev->mtu + ENET_HEADER_SIZE + ENET_FCS_LENGTH;
553         adapter->rx_buffer_len = hw->max_frame_size + 8; /* + 8 for errata */
554
555         if ((hw->device_id == IXGB_DEVICE_ID_82597EX)
556            || (hw->device_id == IXGB_DEVICE_ID_82597EX_CX4)
557            || (hw->device_id == IXGB_DEVICE_ID_82597EX_LR)
558            || (hw->device_id == IXGB_DEVICE_ID_82597EX_SR))
559                 hw->mac_type = ixgb_82597;
560         else {
561                 /* should never have loaded on this device */
562                 DPRINTK(PROBE, ERR, "unsupported device id\n");
563         }
564
565         /* enable flow control to be programmed */
566         hw->fc.send_xon = 1;
567
568         set_bit(__IXGB_DOWN, &adapter->flags);
569         return 0;
570 }
571
572 /**
573  * ixgb_open - Called when a network interface is made active
574  * @netdev: network interface device structure
575  *
576  * Returns 0 on success, negative value on failure
577  *
578  * The open entry point is called when a network interface is made
579  * active by the system (IFF_UP).  At this point all resources needed
580  * for transmit and receive operations are allocated, the interrupt
581  * handler is registered with the OS, the watchdog timer is started,
582  * and the stack is notified that the interface is ready.
583  **/
584
585 static int
586 ixgb_open(struct net_device *netdev)
587 {
588         struct ixgb_adapter *adapter = netdev_priv(netdev);
589         int err;
590
591         /* allocate transmit descriptors */
592         err = ixgb_setup_tx_resources(adapter);
593         if (err)
594                 goto err_setup_tx;
595
596         netif_carrier_off(netdev);
597
598         /* allocate receive descriptors */
599
600         err = ixgb_setup_rx_resources(adapter);
601         if (err)
602                 goto err_setup_rx;
603
604         err = ixgb_up(adapter);
605         if (err)
606                 goto err_up;
607
608         netif_start_queue(netdev);
609
610         return 0;
611
612 err_up:
613         ixgb_free_rx_resources(adapter);
614 err_setup_rx:
615         ixgb_free_tx_resources(adapter);
616 err_setup_tx:
617         ixgb_reset(adapter);
618
619         return err;
620 }
621
622 /**
623  * ixgb_close - Disables a network interface
624  * @netdev: network interface device structure
625  *
626  * Returns 0, this is not allowed to fail
627  *
628  * The close entry point is called when an interface is de-activated
629  * by the OS.  The hardware is still under the drivers control, but
630  * needs to be disabled.  A global MAC reset is issued to stop the
631  * hardware, and all transmit and receive resources are freed.
632  **/
633
634 static int
635 ixgb_close(struct net_device *netdev)
636 {
637         struct ixgb_adapter *adapter = netdev_priv(netdev);
638
639         ixgb_down(adapter, true);
640
641         ixgb_free_tx_resources(adapter);
642         ixgb_free_rx_resources(adapter);
643
644         return 0;
645 }
646
647 /**
648  * ixgb_setup_tx_resources - allocate Tx resources (Descriptors)
649  * @adapter: board private structure
650  *
651  * Return 0 on success, negative on failure
652  **/
653
654 int
655 ixgb_setup_tx_resources(struct ixgb_adapter *adapter)
656 {
657         struct ixgb_desc_ring *txdr = &adapter->tx_ring;
658         struct pci_dev *pdev = adapter->pdev;
659         int size;
660
661         size = sizeof(struct ixgb_buffer) * txdr->count;
662         txdr->buffer_info = vmalloc(size);
663         if (!txdr->buffer_info) {
664                 DPRINTK(PROBE, ERR,
665                  "Unable to allocate transmit descriptor ring memory\n");
666                 return -ENOMEM;
667         }
668         memset(txdr->buffer_info, 0, size);
669
670         /* round up to nearest 4K */
671
672         txdr->size = txdr->count * sizeof(struct ixgb_tx_desc);
673         txdr->size = ALIGN(txdr->size, 4096);
674
675         txdr->desc = pci_alloc_consistent(pdev, txdr->size, &txdr->dma);
676         if (!txdr->desc) {
677                 vfree(txdr->buffer_info);
678                 DPRINTK(PROBE, ERR,
679                  "Unable to allocate transmit descriptor memory\n");
680                 return -ENOMEM;
681         }
682         memset(txdr->desc, 0, txdr->size);
683
684         txdr->next_to_use = 0;
685         txdr->next_to_clean = 0;
686
687         return 0;
688 }
689
690 /**
691  * ixgb_configure_tx - Configure 82597 Transmit Unit after Reset.
692  * @adapter: board private structure
693  *
694  * Configure the Tx unit of the MAC after a reset.
695  **/
696
697 static void
698 ixgb_configure_tx(struct ixgb_adapter *adapter)
699 {
700         u64 tdba = adapter->tx_ring.dma;
701         u32 tdlen = adapter->tx_ring.count * sizeof(struct ixgb_tx_desc);
702         u32 tctl;
703         struct ixgb_hw *hw = &adapter->hw;
704
705         /* Setup the Base and Length of the Tx Descriptor Ring
706          * tx_ring.dma can be either a 32 or 64 bit value
707          */
708
709         IXGB_WRITE_REG(hw, TDBAL, (tdba & 0x00000000ffffffffULL));
710         IXGB_WRITE_REG(hw, TDBAH, (tdba >> 32));
711
712         IXGB_WRITE_REG(hw, TDLEN, tdlen);
713
714         /* Setup the HW Tx Head and Tail descriptor pointers */
715
716         IXGB_WRITE_REG(hw, TDH, 0);
717         IXGB_WRITE_REG(hw, TDT, 0);
718
719         /* don't set up txdctl, it induces performance problems if configured
720          * incorrectly */
721         /* Set the Tx Interrupt Delay register */
722
723         IXGB_WRITE_REG(hw, TIDV, adapter->tx_int_delay);
724
725         /* Program the Transmit Control Register */
726
727         tctl = IXGB_TCTL_TCE | IXGB_TCTL_TXEN | IXGB_TCTL_TPDE;
728         IXGB_WRITE_REG(hw, TCTL, tctl);
729
730         /* Setup Transmit Descriptor Settings for this adapter */
731         adapter->tx_cmd_type =
732                 IXGB_TX_DESC_TYPE |
733                 (adapter->tx_int_delay_enable ? IXGB_TX_DESC_CMD_IDE : 0);
734 }
735
736 /**
737  * ixgb_setup_rx_resources - allocate Rx resources (Descriptors)
738  * @adapter: board private structure
739  *
740  * Returns 0 on success, negative on failure
741  **/
742
743 int
744 ixgb_setup_rx_resources(struct ixgb_adapter *adapter)
745 {
746         struct ixgb_desc_ring *rxdr = &adapter->rx_ring;
747         struct pci_dev *pdev = adapter->pdev;
748         int size;
749
750         size = sizeof(struct ixgb_buffer) * rxdr->count;
751         rxdr->buffer_info = vmalloc(size);
752         if (!rxdr->buffer_info) {
753                 DPRINTK(PROBE, ERR,
754                  "Unable to allocate receive descriptor ring\n");
755                 return -ENOMEM;
756         }
757         memset(rxdr->buffer_info, 0, size);
758
759         /* Round up to nearest 4K */
760
761         rxdr->size = rxdr->count * sizeof(struct ixgb_rx_desc);
762         rxdr->size = ALIGN(rxdr->size, 4096);
763
764         rxdr->desc = pci_alloc_consistent(pdev, rxdr->size, &rxdr->dma);
765
766         if (!rxdr->desc) {
767                 vfree(rxdr->buffer_info);
768                 DPRINTK(PROBE, ERR,
769                  "Unable to allocate receive descriptors\n");
770                 return -ENOMEM;
771         }
772         memset(rxdr->desc, 0, rxdr->size);
773
774         rxdr->next_to_clean = 0;
775         rxdr->next_to_use = 0;
776
777         return 0;
778 }
779
780 /**
781  * ixgb_setup_rctl - configure the receive control register
782  * @adapter: Board private structure
783  **/
784
785 static void
786 ixgb_setup_rctl(struct ixgb_adapter *adapter)
787 {
788         u32 rctl;
789
790         rctl = IXGB_READ_REG(&adapter->hw, RCTL);
791
792         rctl &= ~(3 << IXGB_RCTL_MO_SHIFT);
793
794         rctl |=
795                 IXGB_RCTL_BAM | IXGB_RCTL_RDMTS_1_2 |
796                 IXGB_RCTL_RXEN | IXGB_RCTL_CFF |
797                 (adapter->hw.mc_filter_type << IXGB_RCTL_MO_SHIFT);
798
799         rctl |= IXGB_RCTL_SECRC;
800
801         if (adapter->rx_buffer_len <= IXGB_RXBUFFER_2048)
802                 rctl |= IXGB_RCTL_BSIZE_2048;
803         else if (adapter->rx_buffer_len <= IXGB_RXBUFFER_4096)
804                 rctl |= IXGB_RCTL_BSIZE_4096;
805         else if (adapter->rx_buffer_len <= IXGB_RXBUFFER_8192)
806                 rctl |= IXGB_RCTL_BSIZE_8192;
807         else if (adapter->rx_buffer_len <= IXGB_RXBUFFER_16384)
808                 rctl |= IXGB_RCTL_BSIZE_16384;
809
810         IXGB_WRITE_REG(&adapter->hw, RCTL, rctl);
811 }
812
813 /**
814  * ixgb_configure_rx - Configure 82597 Receive Unit after Reset.
815  * @adapter: board private structure
816  *
817  * Configure the Rx unit of the MAC after a reset.
818  **/
819
820 static void
821 ixgb_configure_rx(struct ixgb_adapter *adapter)
822 {
823         u64 rdba = adapter->rx_ring.dma;
824         u32 rdlen = adapter->rx_ring.count * sizeof(struct ixgb_rx_desc);
825         struct ixgb_hw *hw = &adapter->hw;
826         u32 rctl;
827         u32 rxcsum;
828
829         /* make sure receives are disabled while setting up the descriptors */
830
831         rctl = IXGB_READ_REG(hw, RCTL);
832         IXGB_WRITE_REG(hw, RCTL, rctl & ~IXGB_RCTL_RXEN);
833
834         /* set the Receive Delay Timer Register */
835
836         IXGB_WRITE_REG(hw, RDTR, adapter->rx_int_delay);
837
838         /* Setup the Base and Length of the Rx Descriptor Ring */
839
840         IXGB_WRITE_REG(hw, RDBAL, (rdba & 0x00000000ffffffffULL));
841         IXGB_WRITE_REG(hw, RDBAH, (rdba >> 32));
842
843         IXGB_WRITE_REG(hw, RDLEN, rdlen);
844
845         /* Setup the HW Rx Head and Tail Descriptor Pointers */
846         IXGB_WRITE_REG(hw, RDH, 0);
847         IXGB_WRITE_REG(hw, RDT, 0);
848
849         /* due to the hardware errata with RXDCTL, we are unable to use any of
850          * the performance enhancing features of it without causing other
851          * subtle bugs, some of the bugs could include receive length
852          * corruption at high data rates (WTHRESH > 0) and/or receive
853          * descriptor ring irregularites (particularly in hardware cache) */
854         IXGB_WRITE_REG(hw, RXDCTL, 0);
855
856         /* Enable Receive Checksum Offload for TCP and UDP */
857         if (adapter->rx_csum) {
858                 rxcsum = IXGB_READ_REG(hw, RXCSUM);
859                 rxcsum |= IXGB_RXCSUM_TUOFL;
860                 IXGB_WRITE_REG(hw, RXCSUM, rxcsum);
861         }
862
863         /* Enable Receives */
864
865         IXGB_WRITE_REG(hw, RCTL, rctl);
866 }
867
868 /**
869  * ixgb_free_tx_resources - Free Tx Resources
870  * @adapter: board private structure
871  *
872  * Free all transmit software resources
873  **/
874
875 void
876 ixgb_free_tx_resources(struct ixgb_adapter *adapter)
877 {
878         struct pci_dev *pdev = adapter->pdev;
879
880         ixgb_clean_tx_ring(adapter);
881
882         vfree(adapter->tx_ring.buffer_info);
883         adapter->tx_ring.buffer_info = NULL;
884
885         pci_free_consistent(pdev, adapter->tx_ring.size,
886                             adapter->tx_ring.desc, adapter->tx_ring.dma);
887
888         adapter->tx_ring.desc = NULL;
889 }
890
891 static void
892 ixgb_unmap_and_free_tx_resource(struct ixgb_adapter *adapter,
893                                 struct ixgb_buffer *buffer_info)
894 {
895         if (buffer_info->dma) {
896                 if (buffer_info->mapped_as_page)
897                         pci_unmap_page(adapter->pdev, buffer_info->dma,
898                                        buffer_info->length, PCI_DMA_TODEVICE);
899                 else
900                         pci_unmap_single(adapter->pdev, buffer_info->dma,
901                                          buffer_info->length,
902                                          PCI_DMA_TODEVICE);
903                 buffer_info->dma = 0;
904         }
905
906         if (buffer_info->skb) {
907                 dev_kfree_skb_any(buffer_info->skb);
908                 buffer_info->skb = NULL;
909         }
910         buffer_info->time_stamp = 0;
911         /* these fields must always be initialized in tx
912          * buffer_info->length = 0;
913          * buffer_info->next_to_watch = 0; */
914 }
915
916 /**
917  * ixgb_clean_tx_ring - Free Tx Buffers
918  * @adapter: board private structure
919  **/
920
921 static void
922 ixgb_clean_tx_ring(struct ixgb_adapter *adapter)
923 {
924         struct ixgb_desc_ring *tx_ring = &adapter->tx_ring;
925         struct ixgb_buffer *buffer_info;
926         unsigned long size;
927         unsigned int i;
928
929         /* Free all the Tx ring sk_buffs */
930
931         for (i = 0; i < tx_ring->count; i++) {
932                 buffer_info = &tx_ring->buffer_info[i];
933                 ixgb_unmap_and_free_tx_resource(adapter, buffer_info);
934         }
935
936         size = sizeof(struct ixgb_buffer) * tx_ring->count;
937         memset(tx_ring->buffer_info, 0, size);
938
939         /* Zero out the descriptor ring */
940
941         memset(tx_ring->desc, 0, tx_ring->size);
942
943         tx_ring->next_to_use = 0;
944         tx_ring->next_to_clean = 0;
945
946         IXGB_WRITE_REG(&adapter->hw, TDH, 0);
947         IXGB_WRITE_REG(&adapter->hw, TDT, 0);
948 }
949
950 /**
951  * ixgb_free_rx_resources - Free Rx Resources
952  * @adapter: board private structure
953  *
954  * Free all receive software resources
955  **/
956
957 void
958 ixgb_free_rx_resources(struct ixgb_adapter *adapter)
959 {
960         struct ixgb_desc_ring *rx_ring = &adapter->rx_ring;
961         struct pci_dev *pdev = adapter->pdev;
962
963         ixgb_clean_rx_ring(adapter);
964
965         vfree(rx_ring->buffer_info);
966         rx_ring->buffer_info = NULL;
967
968         pci_free_consistent(pdev, rx_ring->size, rx_ring->desc, rx_ring->dma);
969
970         rx_ring->desc = NULL;
971 }
972
973 /**
974  * ixgb_clean_rx_ring - Free Rx Buffers
975  * @adapter: board private structure
976  **/
977
978 static void
979 ixgb_clean_rx_ring(struct ixgb_adapter *adapter)
980 {
981         struct ixgb_desc_ring *rx_ring = &adapter->rx_ring;
982         struct ixgb_buffer *buffer_info;
983         struct pci_dev *pdev = adapter->pdev;
984         unsigned long size;
985         unsigned int i;
986
987         /* Free all the Rx ring sk_buffs */
988
989         for (i = 0; i < rx_ring->count; i++) {
990                 buffer_info = &rx_ring->buffer_info[i];
991                 if (buffer_info->dma) {
992                         pci_unmap_single(pdev,
993                                          buffer_info->dma,
994                                          buffer_info->length,
995                                          PCI_DMA_FROMDEVICE);
996                         buffer_info->dma = 0;
997                         buffer_info->length = 0;
998                 }
999
1000                 if (buffer_info->skb) {
1001                         dev_kfree_skb(buffer_info->skb);
1002                         buffer_info->skb = NULL;
1003                 }
1004         }
1005
1006         size = sizeof(struct ixgb_buffer) * rx_ring->count;
1007         memset(rx_ring->buffer_info, 0, size);
1008
1009         /* Zero out the descriptor ring */
1010
1011         memset(rx_ring->desc, 0, rx_ring->size);
1012
1013         rx_ring->next_to_clean = 0;
1014         rx_ring->next_to_use = 0;
1015
1016         IXGB_WRITE_REG(&adapter->hw, RDH, 0);
1017         IXGB_WRITE_REG(&adapter->hw, RDT, 0);
1018 }
1019
1020 /**
1021  * ixgb_set_mac - Change the Ethernet Address of the NIC
1022  * @netdev: network interface device structure
1023  * @p: pointer to an address structure
1024  *
1025  * Returns 0 on success, negative on failure
1026  **/
1027
1028 static int
1029 ixgb_set_mac(struct net_device *netdev, void *p)
1030 {
1031         struct ixgb_adapter *adapter = netdev_priv(netdev);
1032         struct sockaddr *addr = p;
1033
1034         if (!is_valid_ether_addr(addr->sa_data))
1035                 return -EADDRNOTAVAIL;
1036
1037         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
1038
1039         ixgb_rar_set(&adapter->hw, addr->sa_data, 0);
1040
1041         return 0;
1042 }
1043
1044 /**
1045  * ixgb_set_multi - Multicast and Promiscuous mode set
1046  * @netdev: network interface device structure
1047  *
1048  * The set_multi entry point is called whenever the multicast address
1049  * list or the network interface flags are updated.  This routine is
1050  * responsible for configuring the hardware for proper multicast,
1051  * promiscuous mode, and all-multi behavior.
1052  **/
1053
1054 static void
1055 ixgb_set_multi(struct net_device *netdev)
1056 {
1057         struct ixgb_adapter *adapter = netdev_priv(netdev);
1058         struct ixgb_hw *hw = &adapter->hw;
1059         struct dev_mc_list *mc_ptr;
1060         u32 rctl;
1061         int i;
1062
1063         /* Check for Promiscuous and All Multicast modes */
1064
1065         rctl = IXGB_READ_REG(hw, RCTL);
1066
1067         if (netdev->flags & IFF_PROMISC) {
1068                 rctl |= (IXGB_RCTL_UPE | IXGB_RCTL_MPE);
1069                 rctl &= ~IXGB_RCTL_VFE;
1070         } else {
1071                 if (netdev->flags & IFF_ALLMULTI) {
1072                         rctl |= IXGB_RCTL_MPE;
1073                         rctl &= ~IXGB_RCTL_UPE;
1074                 } else {
1075                         rctl &= ~(IXGB_RCTL_UPE | IXGB_RCTL_MPE);
1076                 }
1077                 rctl |= IXGB_RCTL_VFE;
1078         }
1079
1080         if (netdev->mc_count > IXGB_MAX_NUM_MULTICAST_ADDRESSES) {
1081                 rctl |= IXGB_RCTL_MPE;
1082                 IXGB_WRITE_REG(hw, RCTL, rctl);
1083         } else {
1084                 u8 mta[IXGB_MAX_NUM_MULTICAST_ADDRESSES *
1085                             IXGB_ETH_LENGTH_OF_ADDRESS];
1086
1087                 IXGB_WRITE_REG(hw, RCTL, rctl);
1088
1089                 for (i = 0, mc_ptr = netdev->mc_list;
1090                      mc_ptr;
1091                      i++, mc_ptr = mc_ptr->next)
1092                         memcpy(&mta[i * IXGB_ETH_LENGTH_OF_ADDRESS],
1093                                mc_ptr->dmi_addr, IXGB_ETH_LENGTH_OF_ADDRESS);
1094
1095                 ixgb_mc_addr_list_update(hw, mta, netdev->mc_count, 0);
1096         }
1097 }
1098
1099 /**
1100  * ixgb_watchdog - Timer Call-back
1101  * @data: pointer to netdev cast into an unsigned long
1102  **/
1103
1104 static void
1105 ixgb_watchdog(unsigned long data)
1106 {
1107         struct ixgb_adapter *adapter = (struct ixgb_adapter *)data;
1108         struct net_device *netdev = adapter->netdev;
1109         struct ixgb_desc_ring *txdr = &adapter->tx_ring;
1110
1111         ixgb_check_for_link(&adapter->hw);
1112
1113         if (ixgb_check_for_bad_link(&adapter->hw)) {
1114                 /* force the reset path */
1115                 netif_stop_queue(netdev);
1116         }
1117
1118         if (adapter->hw.link_up) {
1119                 if (!netif_carrier_ok(netdev)) {
1120                         printk(KERN_INFO "ixgb: %s NIC Link is Up 10 Gbps "
1121                                "Full Duplex, Flow Control: %s\n",
1122                                netdev->name,
1123                                (adapter->hw.fc.type == ixgb_fc_full) ?
1124                                 "RX/TX" :
1125                                 ((adapter->hw.fc.type == ixgb_fc_rx_pause) ?
1126                                  "RX" :
1127                                  ((adapter->hw.fc.type == ixgb_fc_tx_pause) ?
1128                                   "TX" : "None")));
1129                         adapter->link_speed = 10000;
1130                         adapter->link_duplex = FULL_DUPLEX;
1131                         netif_carrier_on(netdev);
1132                 }
1133         } else {
1134                 if (netif_carrier_ok(netdev)) {
1135                         adapter->link_speed = 0;
1136                         adapter->link_duplex = 0;
1137                         printk(KERN_INFO "ixgb: %s NIC Link is Down\n",
1138                                netdev->name);
1139                         netif_carrier_off(netdev);
1140                 }
1141         }
1142
1143         ixgb_update_stats(adapter);
1144
1145         if (!netif_carrier_ok(netdev)) {
1146                 if (IXGB_DESC_UNUSED(txdr) + 1 < txdr->count) {
1147                         /* We've lost link, so the controller stops DMA,
1148                          * but we've got queued Tx work that's never going
1149                          * to get done, so reset controller to flush Tx.
1150                          * (Do the reset outside of interrupt context). */
1151                         schedule_work(&adapter->tx_timeout_task);
1152                         /* return immediately since reset is imminent */
1153                         return;
1154                 }
1155         }
1156
1157         /* Force detection of hung controller every watchdog period */
1158         adapter->detect_tx_hung = true;
1159
1160         /* generate an interrupt to force clean up of any stragglers */
1161         IXGB_WRITE_REG(&adapter->hw, ICS, IXGB_INT_TXDW);
1162
1163         /* Reset the timer */
1164         mod_timer(&adapter->watchdog_timer, jiffies + 2 * HZ);
1165 }
1166
1167 #define IXGB_TX_FLAGS_CSUM              0x00000001
1168 #define IXGB_TX_FLAGS_VLAN              0x00000002
1169 #define IXGB_TX_FLAGS_TSO               0x00000004
1170
1171 static int
1172 ixgb_tso(struct ixgb_adapter *adapter, struct sk_buff *skb)
1173 {
1174         struct ixgb_context_desc *context_desc;
1175         unsigned int i;
1176         u8 ipcss, ipcso, tucss, tucso, hdr_len;
1177         u16 ipcse, tucse, mss;
1178         int err;
1179
1180         if (likely(skb_is_gso(skb))) {
1181                 struct ixgb_buffer *buffer_info;
1182                 struct iphdr *iph;
1183
1184                 if (skb_header_cloned(skb)) {
1185                         err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1186                         if (err)
1187                                 return err;
1188                 }
1189
1190                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
1191                 mss = skb_shinfo(skb)->gso_size;
1192                 iph = ip_hdr(skb);
1193                 iph->tot_len = 0;
1194                 iph->check = 0;
1195                 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
1196                                                          iph->daddr, 0,
1197                                                          IPPROTO_TCP, 0);
1198                 ipcss = skb_network_offset(skb);
1199                 ipcso = (void *)&(iph->check) - (void *)skb->data;
1200                 ipcse = skb_transport_offset(skb) - 1;
1201                 tucss = skb_transport_offset(skb);
1202                 tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
1203                 tucse = 0;
1204
1205                 i = adapter->tx_ring.next_to_use;
1206                 context_desc = IXGB_CONTEXT_DESC(adapter->tx_ring, i);
1207                 buffer_info = &adapter->tx_ring.buffer_info[i];
1208                 WARN_ON(buffer_info->dma != 0);
1209
1210                 context_desc->ipcss = ipcss;
1211                 context_desc->ipcso = ipcso;
1212                 context_desc->ipcse = cpu_to_le16(ipcse);
1213                 context_desc->tucss = tucss;
1214                 context_desc->tucso = tucso;
1215                 context_desc->tucse = cpu_to_le16(tucse);
1216                 context_desc->mss = cpu_to_le16(mss);
1217                 context_desc->hdr_len = hdr_len;
1218                 context_desc->status = 0;
1219                 context_desc->cmd_type_len = cpu_to_le32(
1220                                                   IXGB_CONTEXT_DESC_TYPE
1221                                                 | IXGB_CONTEXT_DESC_CMD_TSE
1222                                                 | IXGB_CONTEXT_DESC_CMD_IP
1223                                                 | IXGB_CONTEXT_DESC_CMD_TCP
1224                                                 | IXGB_CONTEXT_DESC_CMD_IDE
1225                                                 | (skb->len - (hdr_len)));
1226
1227
1228                 if (++i == adapter->tx_ring.count) i = 0;
1229                 adapter->tx_ring.next_to_use = i;
1230
1231                 return 1;
1232         }
1233
1234         return 0;
1235 }
1236
1237 static bool
1238 ixgb_tx_csum(struct ixgb_adapter *adapter, struct sk_buff *skb)
1239 {
1240         struct ixgb_context_desc *context_desc;
1241         unsigned int i;
1242         u8 css, cso;
1243
1244         if (likely(skb->ip_summed == CHECKSUM_PARTIAL)) {
1245                 struct ixgb_buffer *buffer_info;
1246                 css = skb_transport_offset(skb);
1247                 cso = css + skb->csum_offset;
1248
1249                 i = adapter->tx_ring.next_to_use;
1250                 context_desc = IXGB_CONTEXT_DESC(adapter->tx_ring, i);
1251                 buffer_info = &adapter->tx_ring.buffer_info[i];
1252                 WARN_ON(buffer_info->dma != 0);
1253
1254                 context_desc->tucss = css;
1255                 context_desc->tucso = cso;
1256                 context_desc->tucse = 0;
1257                 /* zero out any previously existing data in one instruction */
1258                 *(u32 *)&(context_desc->ipcss) = 0;
1259                 context_desc->status = 0;
1260                 context_desc->hdr_len = 0;
1261                 context_desc->mss = 0;
1262                 context_desc->cmd_type_len =
1263                         cpu_to_le32(IXGB_CONTEXT_DESC_TYPE
1264                                     | IXGB_TX_DESC_CMD_IDE);
1265
1266                 if (++i == adapter->tx_ring.count) i = 0;
1267                 adapter->tx_ring.next_to_use = i;
1268
1269                 return true;
1270         }
1271
1272         return false;
1273 }
1274
1275 #define IXGB_MAX_TXD_PWR        14
1276 #define IXGB_MAX_DATA_PER_TXD   (1<<IXGB_MAX_TXD_PWR)
1277
1278 static int
1279 ixgb_tx_map(struct ixgb_adapter *adapter, struct sk_buff *skb,
1280             unsigned int first)
1281 {
1282         struct ixgb_desc_ring *tx_ring = &adapter->tx_ring;
1283         struct pci_dev *pdev = adapter->pdev;
1284         struct ixgb_buffer *buffer_info;
1285         int len = skb_headlen(skb);
1286         unsigned int offset = 0, size, count = 0, i;
1287         unsigned int mss = skb_shinfo(skb)->gso_size;
1288         unsigned int nr_frags = skb_shinfo(skb)->nr_frags;
1289         unsigned int f;
1290
1291         i = tx_ring->next_to_use;
1292
1293         while (len) {
1294                 buffer_info = &tx_ring->buffer_info[i];
1295                 size = min(len, IXGB_MAX_DATA_PER_TXD);
1296                 /* Workaround for premature desc write-backs
1297                  * in TSO mode.  Append 4-byte sentinel desc */
1298                 if (unlikely(mss && !nr_frags && size == len && size > 8))
1299                         size -= 4;
1300
1301                 buffer_info->length = size;
1302                 WARN_ON(buffer_info->dma != 0);
1303                 buffer_info->time_stamp = jiffies;
1304                 buffer_info->mapped_as_page = false;
1305                 buffer_info->dma = pci_map_single(pdev, skb->data + offset,
1306                                                   size, PCI_DMA_TODEVICE);
1307                 if (pci_dma_mapping_error(pdev, buffer_info->dma))
1308                         goto dma_error;
1309                 buffer_info->next_to_watch = 0;
1310
1311                 len -= size;
1312                 offset += size;
1313                 count++;
1314                 if (len) {
1315                         i++;
1316                         if (i == tx_ring->count)
1317                                 i = 0;
1318                 }
1319         }
1320
1321         for (f = 0; f < nr_frags; f++) {
1322                 struct skb_frag_struct *frag;
1323
1324                 frag = &skb_shinfo(skb)->frags[f];
1325                 len = frag->size;
1326                 offset = frag->page_offset;
1327
1328                 while (len) {
1329                         i++;
1330                         if (i == tx_ring->count)
1331                                 i = 0;
1332
1333                         buffer_info = &tx_ring->buffer_info[i];
1334                         size = min(len, IXGB_MAX_DATA_PER_TXD);
1335
1336                         /* Workaround for premature desc write-backs
1337                          * in TSO mode.  Append 4-byte sentinel desc */
1338                         if (unlikely(mss && (f == (nr_frags - 1))
1339                                      && size == len && size > 8))
1340                                 size -= 4;
1341
1342                         buffer_info->length = size;
1343                         buffer_info->time_stamp = jiffies;
1344                         buffer_info->mapped_as_page = true;
1345                         buffer_info->dma =
1346                                 pci_map_page(pdev, frag->page,
1347                                              offset, size,
1348                                              PCI_DMA_TODEVICE);
1349                         if (pci_dma_mapping_error(pdev, buffer_info->dma))
1350                                 goto dma_error;
1351                         buffer_info->next_to_watch = 0;
1352
1353                         len -= size;
1354                         offset += size;
1355                         count++;
1356                 }
1357         }
1358         tx_ring->buffer_info[i].skb = skb;
1359         tx_ring->buffer_info[first].next_to_watch = i;
1360
1361         return count;
1362
1363 dma_error:
1364         dev_err(&pdev->dev, "TX DMA map failed\n");
1365         buffer_info->dma = 0;
1366         count--;
1367
1368         while (count >= 0) {
1369                 count--;
1370                 i--;
1371                 if (i < 0)
1372                         i += tx_ring->count;
1373                 buffer_info = &tx_ring->buffer_info[i];
1374                 ixgb_unmap_and_free_tx_resource(adapter, buffer_info);
1375         }
1376
1377         return 0;
1378 }
1379
1380 static void
1381 ixgb_tx_queue(struct ixgb_adapter *adapter, int count, int vlan_id,int tx_flags)
1382 {
1383         struct ixgb_desc_ring *tx_ring = &adapter->tx_ring;
1384         struct ixgb_tx_desc *tx_desc = NULL;
1385         struct ixgb_buffer *buffer_info;
1386         u32 cmd_type_len = adapter->tx_cmd_type;
1387         u8 status = 0;
1388         u8 popts = 0;
1389         unsigned int i;
1390
1391         if (tx_flags & IXGB_TX_FLAGS_TSO) {
1392                 cmd_type_len |= IXGB_TX_DESC_CMD_TSE;
1393                 popts |= (IXGB_TX_DESC_POPTS_IXSM | IXGB_TX_DESC_POPTS_TXSM);
1394         }
1395
1396         if (tx_flags & IXGB_TX_FLAGS_CSUM)
1397                 popts |= IXGB_TX_DESC_POPTS_TXSM;
1398
1399         if (tx_flags & IXGB_TX_FLAGS_VLAN)
1400                 cmd_type_len |= IXGB_TX_DESC_CMD_VLE;
1401
1402         i = tx_ring->next_to_use;
1403
1404         while (count--) {
1405                 buffer_info = &tx_ring->buffer_info[i];
1406                 tx_desc = IXGB_TX_DESC(*tx_ring, i);
1407                 tx_desc->buff_addr = cpu_to_le64(buffer_info->dma);
1408                 tx_desc->cmd_type_len =
1409                         cpu_to_le32(cmd_type_len | buffer_info->length);
1410                 tx_desc->status = status;
1411                 tx_desc->popts = popts;
1412                 tx_desc->vlan = cpu_to_le16(vlan_id);
1413
1414                 if (++i == tx_ring->count) i = 0;
1415         }
1416
1417         tx_desc->cmd_type_len |=
1418                 cpu_to_le32(IXGB_TX_DESC_CMD_EOP | IXGB_TX_DESC_CMD_RS);
1419
1420         /* Force memory writes to complete before letting h/w
1421          * know there are new descriptors to fetch.  (Only
1422          * applicable for weak-ordered memory model archs,
1423          * such as IA-64). */
1424         wmb();
1425
1426         tx_ring->next_to_use = i;
1427         IXGB_WRITE_REG(&adapter->hw, TDT, i);
1428 }
1429
1430 static int __ixgb_maybe_stop_tx(struct net_device *netdev, int size)
1431 {
1432         struct ixgb_adapter *adapter = netdev_priv(netdev);
1433         struct ixgb_desc_ring *tx_ring = &adapter->tx_ring;
1434
1435         netif_stop_queue(netdev);
1436         /* Herbert's original patch had:
1437          *  smp_mb__after_netif_stop_queue();
1438          * but since that doesn't exist yet, just open code it. */
1439         smp_mb();
1440
1441         /* We need to check again in a case another CPU has just
1442          * made room available. */
1443         if (likely(IXGB_DESC_UNUSED(tx_ring) < size))
1444                 return -EBUSY;
1445
1446         /* A reprieve! */
1447         netif_start_queue(netdev);
1448         ++adapter->restart_queue;
1449         return 0;
1450 }
1451
1452 static int ixgb_maybe_stop_tx(struct net_device *netdev,
1453                               struct ixgb_desc_ring *tx_ring, int size)
1454 {
1455         if (likely(IXGB_DESC_UNUSED(tx_ring) >= size))
1456                 return 0;
1457         return __ixgb_maybe_stop_tx(netdev, size);
1458 }
1459
1460
1461 /* Tx Descriptors needed, worst case */
1462 #define TXD_USE_COUNT(S) (((S) >> IXGB_MAX_TXD_PWR) + \
1463                          (((S) & (IXGB_MAX_DATA_PER_TXD - 1)) ? 1 : 0))
1464 #define DESC_NEEDED TXD_USE_COUNT(IXGB_MAX_DATA_PER_TXD) /* skb->date */ + \
1465         MAX_SKB_FRAGS * TXD_USE_COUNT(PAGE_SIZE) + 1 /* for context */ \
1466         + 1 /* one more needed for sentinel TSO workaround */
1467
1468 static netdev_tx_t
1469 ixgb_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
1470 {
1471         struct ixgb_adapter *adapter = netdev_priv(netdev);
1472         unsigned int first;
1473         unsigned int tx_flags = 0;
1474         int vlan_id = 0;
1475         int count = 0;
1476         int tso;
1477
1478         if (test_bit(__IXGB_DOWN, &adapter->flags)) {
1479                 dev_kfree_skb(skb);
1480                 return NETDEV_TX_OK;
1481         }
1482
1483         if (skb->len <= 0) {
1484                 dev_kfree_skb(skb);
1485                 return NETDEV_TX_OK;
1486         }
1487
1488         if (unlikely(ixgb_maybe_stop_tx(netdev, &adapter->tx_ring,
1489                      DESC_NEEDED)))
1490                 return NETDEV_TX_BUSY;
1491
1492         if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
1493                 tx_flags |= IXGB_TX_FLAGS_VLAN;
1494                 vlan_id = vlan_tx_tag_get(skb);
1495         }
1496
1497         first = adapter->tx_ring.next_to_use;
1498
1499         tso = ixgb_tso(adapter, skb);
1500         if (tso < 0) {
1501                 dev_kfree_skb(skb);
1502                 return NETDEV_TX_OK;
1503         }
1504
1505         if (likely(tso))
1506                 tx_flags |= IXGB_TX_FLAGS_TSO;
1507         else if (ixgb_tx_csum(adapter, skb))
1508                 tx_flags |= IXGB_TX_FLAGS_CSUM;
1509
1510         count = ixgb_tx_map(adapter, skb, first);
1511
1512         if (count) {
1513                 ixgb_tx_queue(adapter, count, vlan_id, tx_flags);
1514                 /* Make sure there is space in the ring for the next send. */
1515                 ixgb_maybe_stop_tx(netdev, &adapter->tx_ring, DESC_NEEDED);
1516
1517         } else {
1518                 dev_kfree_skb_any(skb);
1519                 adapter->tx_ring.buffer_info[first].time_stamp = 0;
1520                 adapter->tx_ring.next_to_use = first;
1521         }
1522
1523         return NETDEV_TX_OK;
1524 }
1525
1526 /**
1527  * ixgb_tx_timeout - Respond to a Tx Hang
1528  * @netdev: network interface device structure
1529  **/
1530
1531 static void
1532 ixgb_tx_timeout(struct net_device *netdev)
1533 {
1534         struct ixgb_adapter *adapter = netdev_priv(netdev);
1535
1536         /* Do the reset outside of interrupt context */
1537         schedule_work(&adapter->tx_timeout_task);
1538 }
1539
1540 static void
1541 ixgb_tx_timeout_task(struct work_struct *work)
1542 {
1543         struct ixgb_adapter *adapter =
1544                 container_of(work, struct ixgb_adapter, tx_timeout_task);
1545
1546         adapter->tx_timeout_count++;
1547         ixgb_down(adapter, true);
1548         ixgb_up(adapter);
1549 }
1550
1551 /**
1552  * ixgb_get_stats - Get System Network Statistics
1553  * @netdev: network interface device structure
1554  *
1555  * Returns the address of the device statistics structure.
1556  * The statistics are actually updated from the timer callback.
1557  **/
1558
1559 static struct net_device_stats *
1560 ixgb_get_stats(struct net_device *netdev)
1561 {
1562         return &netdev->stats;
1563 }
1564
1565 /**
1566  * ixgb_change_mtu - Change the Maximum Transfer Unit
1567  * @netdev: network interface device structure
1568  * @new_mtu: new value for maximum frame size
1569  *
1570  * Returns 0 on success, negative on failure
1571  **/
1572
1573 static int
1574 ixgb_change_mtu(struct net_device *netdev, int new_mtu)
1575 {
1576         struct ixgb_adapter *adapter = netdev_priv(netdev);
1577         int max_frame = new_mtu + ENET_HEADER_SIZE + ENET_FCS_LENGTH;
1578         int old_max_frame = netdev->mtu + ENET_HEADER_SIZE + ENET_FCS_LENGTH;
1579
1580         /* MTU < 68 is an error for IPv4 traffic, just don't allow it */
1581         if ((new_mtu < 68) ||
1582             (max_frame > IXGB_MAX_JUMBO_FRAME_SIZE + ENET_FCS_LENGTH)) {
1583                 DPRINTK(PROBE, ERR, "Invalid MTU setting %d\n", new_mtu);
1584                 return -EINVAL;
1585         }
1586
1587         if (old_max_frame == max_frame)
1588                 return 0;
1589
1590         if (netif_running(netdev))
1591                 ixgb_down(adapter, true);
1592
1593         adapter->rx_buffer_len = max_frame + 8; /* + 8 for errata */
1594
1595         netdev->mtu = new_mtu;
1596
1597         if (netif_running(netdev))
1598                 ixgb_up(adapter);
1599
1600         return 0;
1601 }
1602
1603 /**
1604  * ixgb_update_stats - Update the board statistics counters.
1605  * @adapter: board private structure
1606  **/
1607
1608 void
1609 ixgb_update_stats(struct ixgb_adapter *adapter)
1610 {
1611         struct net_device *netdev = adapter->netdev;
1612         struct pci_dev *pdev = adapter->pdev;
1613
1614         /* Prevent stats update while adapter is being reset */
1615         if (pci_channel_offline(pdev))
1616                 return;
1617
1618         if ((netdev->flags & IFF_PROMISC) || (netdev->flags & IFF_ALLMULTI) ||
1619            (netdev->mc_count > IXGB_MAX_NUM_MULTICAST_ADDRESSES)) {
1620                 u64 multi = IXGB_READ_REG(&adapter->hw, MPRCL);
1621                 u32 bcast_l = IXGB_READ_REG(&adapter->hw, BPRCL);
1622                 u32 bcast_h = IXGB_READ_REG(&adapter->hw, BPRCH);
1623                 u64 bcast = ((u64)bcast_h << 32) | bcast_l;
1624
1625                 multi |= ((u64)IXGB_READ_REG(&adapter->hw, MPRCH) << 32);
1626                 /* fix up multicast stats by removing broadcasts */
1627                 if (multi >= bcast)
1628                         multi -= bcast;
1629
1630                 adapter->stats.mprcl += (multi & 0xFFFFFFFF);
1631                 adapter->stats.mprch += (multi >> 32);
1632                 adapter->stats.bprcl += bcast_l;
1633                 adapter->stats.bprch += bcast_h;
1634         } else {
1635                 adapter->stats.mprcl += IXGB_READ_REG(&adapter->hw, MPRCL);
1636                 adapter->stats.mprch += IXGB_READ_REG(&adapter->hw, MPRCH);
1637                 adapter->stats.bprcl += IXGB_READ_REG(&adapter->hw, BPRCL);
1638                 adapter->stats.bprch += IXGB_READ_REG(&adapter->hw, BPRCH);
1639         }
1640         adapter->stats.tprl += IXGB_READ_REG(&adapter->hw, TPRL);
1641         adapter->stats.tprh += IXGB_READ_REG(&adapter->hw, TPRH);
1642         adapter->stats.gprcl += IXGB_READ_REG(&adapter->hw, GPRCL);
1643         adapter->stats.gprch += IXGB_READ_REG(&adapter->hw, GPRCH);
1644         adapter->stats.uprcl += IXGB_READ_REG(&adapter->hw, UPRCL);
1645         adapter->stats.uprch += IXGB_READ_REG(&adapter->hw, UPRCH);
1646         adapter->stats.vprcl += IXGB_READ_REG(&adapter->hw, VPRCL);
1647         adapter->stats.vprch += IXGB_READ_REG(&adapter->hw, VPRCH);
1648         adapter->stats.jprcl += IXGB_READ_REG(&adapter->hw, JPRCL);
1649         adapter->stats.jprch += IXGB_READ_REG(&adapter->hw, JPRCH);
1650         adapter->stats.gorcl += IXGB_READ_REG(&adapter->hw, GORCL);
1651         adapter->stats.gorch += IXGB_READ_REG(&adapter->hw, GORCH);
1652         adapter->stats.torl += IXGB_READ_REG(&adapter->hw, TORL);
1653         adapter->stats.torh += IXGB_READ_REG(&adapter->hw, TORH);
1654         adapter->stats.rnbc += IXGB_READ_REG(&adapter->hw, RNBC);
1655         adapter->stats.ruc += IXGB_READ_REG(&adapter->hw, RUC);
1656         adapter->stats.roc += IXGB_READ_REG(&adapter->hw, ROC);
1657         adapter->stats.rlec += IXGB_READ_REG(&adapter->hw, RLEC);
1658         adapter->stats.crcerrs += IXGB_READ_REG(&adapter->hw, CRCERRS);
1659         adapter->stats.icbc += IXGB_READ_REG(&adapter->hw, ICBC);
1660         adapter->stats.ecbc += IXGB_READ_REG(&adapter->hw, ECBC);
1661         adapter->stats.mpc += IXGB_READ_REG(&adapter->hw, MPC);
1662         adapter->stats.tptl += IXGB_READ_REG(&adapter->hw, TPTL);
1663         adapter->stats.tpth += IXGB_READ_REG(&adapter->hw, TPTH);
1664         adapter->stats.gptcl += IXGB_READ_REG(&adapter->hw, GPTCL);
1665         adapter->stats.gptch += IXGB_READ_REG(&adapter->hw, GPTCH);
1666         adapter->stats.bptcl += IXGB_READ_REG(&adapter->hw, BPTCL);
1667         adapter->stats.bptch += IXGB_READ_REG(&adapter->hw, BPTCH);
1668         adapter->stats.mptcl += IXGB_READ_REG(&adapter->hw, MPTCL);
1669         adapter->stats.mptch += IXGB_READ_REG(&adapter->hw, MPTCH);
1670         adapter->stats.uptcl += IXGB_READ_REG(&adapter->hw, UPTCL);
1671         adapter->stats.uptch += IXGB_READ_REG(&adapter->hw, UPTCH);
1672         adapter->stats.vptcl += IXGB_READ_REG(&adapter->hw, VPTCL);
1673         adapter->stats.vptch += IXGB_READ_REG(&adapter->hw, VPTCH);
1674         adapter->stats.jptcl += IXGB_READ_REG(&adapter->hw, JPTCL);
1675         adapter->stats.jptch += IXGB_READ_REG(&adapter->hw, JPTCH);
1676         adapter->stats.gotcl += IXGB_READ_REG(&adapter->hw, GOTCL);
1677         adapter->stats.gotch += IXGB_READ_REG(&adapter->hw, GOTCH);
1678         adapter->stats.totl += IXGB_READ_REG(&adapter->hw, TOTL);
1679         adapter->stats.toth += IXGB_READ_REG(&adapter->hw, TOTH);
1680         adapter->stats.dc += IXGB_READ_REG(&adapter->hw, DC);
1681         adapter->stats.plt64c += IXGB_READ_REG(&adapter->hw, PLT64C);
1682         adapter->stats.tsctc += IXGB_READ_REG(&adapter->hw, TSCTC);
1683         adapter->stats.tsctfc += IXGB_READ_REG(&adapter->hw, TSCTFC);
1684         adapter->stats.ibic += IXGB_READ_REG(&adapter->hw, IBIC);
1685         adapter->stats.rfc += IXGB_READ_REG(&adapter->hw, RFC);
1686         adapter->stats.lfc += IXGB_READ_REG(&adapter->hw, LFC);
1687         adapter->stats.pfrc += IXGB_READ_REG(&adapter->hw, PFRC);
1688         adapter->stats.pftc += IXGB_READ_REG(&adapter->hw, PFTC);
1689         adapter->stats.mcfrc += IXGB_READ_REG(&adapter->hw, MCFRC);
1690         adapter->stats.mcftc += IXGB_READ_REG(&adapter->hw, MCFTC);
1691         adapter->stats.xonrxc += IXGB_READ_REG(&adapter->hw, XONRXC);
1692         adapter->stats.xontxc += IXGB_READ_REG(&adapter->hw, XONTXC);
1693         adapter->stats.xoffrxc += IXGB_READ_REG(&adapter->hw, XOFFRXC);
1694         adapter->stats.xofftxc += IXGB_READ_REG(&adapter->hw, XOFFTXC);
1695         adapter->stats.rjc += IXGB_READ_REG(&adapter->hw, RJC);
1696
1697         /* Fill out the OS statistics structure */
1698
1699         netdev->stats.rx_packets = adapter->stats.gprcl;
1700         netdev->stats.tx_packets = adapter->stats.gptcl;
1701         netdev->stats.rx_bytes = adapter->stats.gorcl;
1702         netdev->stats.tx_bytes = adapter->stats.gotcl;
1703         netdev->stats.multicast = adapter->stats.mprcl;
1704         netdev->stats.collisions = 0;
1705
1706         /* ignore RLEC as it reports errors for padded (<64bytes) frames
1707          * with a length in the type/len field */
1708         netdev->stats.rx_errors =
1709             /* adapter->stats.rnbc + */ adapter->stats.crcerrs +
1710             adapter->stats.ruc +
1711             adapter->stats.roc /*+ adapter->stats.rlec */  +
1712             adapter->stats.icbc +
1713             adapter->stats.ecbc + adapter->stats.mpc;
1714
1715         /* see above
1716          * netdev->stats.rx_length_errors = adapter->stats.rlec;
1717          */
1718
1719         netdev->stats.rx_crc_errors = adapter->stats.crcerrs;
1720         netdev->stats.rx_fifo_errors = adapter->stats.mpc;
1721         netdev->stats.rx_missed_errors = adapter->stats.mpc;
1722         netdev->stats.rx_over_errors = adapter->stats.mpc;
1723
1724         netdev->stats.tx_errors = 0;
1725         netdev->stats.rx_frame_errors = 0;
1726         netdev->stats.tx_aborted_errors = 0;
1727         netdev->stats.tx_carrier_errors = 0;
1728         netdev->stats.tx_fifo_errors = 0;
1729         netdev->stats.tx_heartbeat_errors = 0;
1730         netdev->stats.tx_window_errors = 0;
1731 }
1732
1733 #define IXGB_MAX_INTR 10
1734 /**
1735  * ixgb_intr - Interrupt Handler
1736  * @irq: interrupt number
1737  * @data: pointer to a network interface device structure
1738  **/
1739
1740 static irqreturn_t
1741 ixgb_intr(int irq, void *data)
1742 {
1743         struct net_device *netdev = data;
1744         struct ixgb_adapter *adapter = netdev_priv(netdev);
1745         struct ixgb_hw *hw = &adapter->hw;
1746         u32 icr = IXGB_READ_REG(hw, ICR);
1747
1748         if (unlikely(!icr))
1749                 return IRQ_NONE;  /* Not our interrupt */
1750
1751         if (unlikely(icr & (IXGB_INT_RXSEQ | IXGB_INT_LSC)))
1752                 if (!test_bit(__IXGB_DOWN, &adapter->flags))
1753                         mod_timer(&adapter->watchdog_timer, jiffies);
1754
1755         if (napi_schedule_prep(&adapter->napi)) {
1756
1757                 /* Disable interrupts and register for poll. The flush
1758                   of the posted write is intentionally left out.
1759                 */
1760
1761                 IXGB_WRITE_REG(&adapter->hw, IMC, ~0);
1762                 __napi_schedule(&adapter->napi);
1763         }
1764         return IRQ_HANDLED;
1765 }
1766
1767 /**
1768  * ixgb_clean - NAPI Rx polling callback
1769  * @adapter: board private structure
1770  **/
1771
1772 static int
1773 ixgb_clean(struct napi_struct *napi, int budget)
1774 {
1775         struct ixgb_adapter *adapter = container_of(napi, struct ixgb_adapter, napi);
1776         int work_done = 0;
1777
1778         ixgb_clean_tx_irq(adapter);
1779         ixgb_clean_rx_irq(adapter, &work_done, budget);
1780
1781         /* If budget not fully consumed, exit the polling mode */
1782         if (work_done < budget) {
1783                 napi_complete(napi);
1784                 if (!test_bit(__IXGB_DOWN, &adapter->flags))
1785                         ixgb_irq_enable(adapter);
1786         }
1787
1788         return work_done;
1789 }
1790
1791 /**
1792  * ixgb_clean_tx_irq - Reclaim resources after transmit completes
1793  * @adapter: board private structure
1794  **/
1795
1796 static bool
1797 ixgb_clean_tx_irq(struct ixgb_adapter *adapter)
1798 {
1799         struct ixgb_desc_ring *tx_ring = &adapter->tx_ring;
1800         struct net_device *netdev = adapter->netdev;
1801         struct ixgb_tx_desc *tx_desc, *eop_desc;
1802         struct ixgb_buffer *buffer_info;
1803         unsigned int i, eop;
1804         bool cleaned = false;
1805
1806         i = tx_ring->next_to_clean;
1807         eop = tx_ring->buffer_info[i].next_to_watch;
1808         eop_desc = IXGB_TX_DESC(*tx_ring, eop);
1809
1810         while (eop_desc->status & IXGB_TX_DESC_STATUS_DD) {
1811
1812                 for (cleaned = false; !cleaned; ) {
1813                         tx_desc = IXGB_TX_DESC(*tx_ring, i);
1814                         buffer_info = &tx_ring->buffer_info[i];
1815
1816                         if (tx_desc->popts &
1817                            (IXGB_TX_DESC_POPTS_TXSM |
1818                             IXGB_TX_DESC_POPTS_IXSM))
1819                                 adapter->hw_csum_tx_good++;
1820
1821                         ixgb_unmap_and_free_tx_resource(adapter, buffer_info);
1822
1823                         *(u32 *)&(tx_desc->status) = 0;
1824
1825                         cleaned = (i == eop);
1826                         if (++i == tx_ring->count) i = 0;
1827                 }
1828
1829                 eop = tx_ring->buffer_info[i].next_to_watch;
1830                 eop_desc = IXGB_TX_DESC(*tx_ring, eop);
1831         }
1832
1833         tx_ring->next_to_clean = i;
1834
1835         if (unlikely(cleaned && netif_carrier_ok(netdev) &&
1836                      IXGB_DESC_UNUSED(tx_ring) >= DESC_NEEDED)) {
1837                 /* Make sure that anybody stopping the queue after this
1838                  * sees the new next_to_clean. */
1839                 smp_mb();
1840
1841                 if (netif_queue_stopped(netdev) &&
1842                     !(test_bit(__IXGB_DOWN, &adapter->flags))) {
1843                         netif_wake_queue(netdev);
1844                         ++adapter->restart_queue;
1845                 }
1846         }
1847
1848         if (adapter->detect_tx_hung) {
1849                 /* detect a transmit hang in hardware, this serializes the
1850                  * check with the clearing of time_stamp and movement of i */
1851                 adapter->detect_tx_hung = false;
1852                 if (tx_ring->buffer_info[eop].time_stamp &&
1853                    time_after(jiffies, tx_ring->buffer_info[eop].time_stamp + HZ)
1854                    && !(IXGB_READ_REG(&adapter->hw, STATUS) &
1855                         IXGB_STATUS_TXOFF)) {
1856                         /* detected Tx unit hang */
1857                         DPRINTK(DRV, ERR, "Detected Tx Unit Hang\n"
1858                                         "  TDH                  <%x>\n"
1859                                         "  TDT                  <%x>\n"
1860                                         "  next_to_use          <%x>\n"
1861                                         "  next_to_clean        <%x>\n"
1862                                         "buffer_info[next_to_clean]\n"
1863                                         "  time_stamp           <%lx>\n"
1864                                         "  next_to_watch        <%x>\n"
1865                                         "  jiffies              <%lx>\n"
1866                                         "  next_to_watch.status <%x>\n",
1867                                 IXGB_READ_REG(&adapter->hw, TDH),
1868                                 IXGB_READ_REG(&adapter->hw, TDT),
1869                                 tx_ring->next_to_use,
1870                                 tx_ring->next_to_clean,
1871                                 tx_ring->buffer_info[eop].time_stamp,
1872                                 eop,
1873                                 jiffies,
1874                                 eop_desc->status);
1875                         netif_stop_queue(netdev);
1876                 }
1877         }
1878
1879         return cleaned;
1880 }
1881
1882 /**
1883  * ixgb_rx_checksum - Receive Checksum Offload for 82597.
1884  * @adapter: board private structure
1885  * @rx_desc: receive descriptor
1886  * @sk_buff: socket buffer with received data
1887  **/
1888
1889 static void
1890 ixgb_rx_checksum(struct ixgb_adapter *adapter,
1891                  struct ixgb_rx_desc *rx_desc,
1892                  struct sk_buff *skb)
1893 {
1894         /* Ignore Checksum bit is set OR
1895          * TCP Checksum has not been calculated
1896          */
1897         if ((rx_desc->status & IXGB_RX_DESC_STATUS_IXSM) ||
1898            (!(rx_desc->status & IXGB_RX_DESC_STATUS_TCPCS))) {
1899                 skb->ip_summed = CHECKSUM_NONE;
1900                 return;
1901         }
1902
1903         /* At this point we know the hardware did the TCP checksum */
1904         /* now look at the TCP checksum error bit */
1905         if (rx_desc->errors & IXGB_RX_DESC_ERRORS_TCPE) {
1906                 /* let the stack verify checksum errors */
1907                 skb->ip_summed = CHECKSUM_NONE;
1908                 adapter->hw_csum_rx_error++;
1909         } else {
1910                 /* TCP checksum is good */
1911                 skb->ip_summed = CHECKSUM_UNNECESSARY;
1912                 adapter->hw_csum_rx_good++;
1913         }
1914 }
1915
1916 /**
1917  * ixgb_clean_rx_irq - Send received data up the network stack,
1918  * @adapter: board private structure
1919  **/
1920
1921 static bool
1922 ixgb_clean_rx_irq(struct ixgb_adapter *adapter, int *work_done, int work_to_do)
1923 {
1924         struct ixgb_desc_ring *rx_ring = &adapter->rx_ring;
1925         struct net_device *netdev = adapter->netdev;
1926         struct pci_dev *pdev = adapter->pdev;
1927         struct ixgb_rx_desc *rx_desc, *next_rxd;
1928         struct ixgb_buffer *buffer_info, *next_buffer, *next2_buffer;
1929         u32 length;
1930         unsigned int i, j;
1931         int cleaned_count = 0;
1932         bool cleaned = false;
1933
1934         i = rx_ring->next_to_clean;
1935         rx_desc = IXGB_RX_DESC(*rx_ring, i);
1936         buffer_info = &rx_ring->buffer_info[i];
1937
1938         while (rx_desc->status & IXGB_RX_DESC_STATUS_DD) {
1939                 struct sk_buff *skb;
1940                 u8 status;
1941
1942                 if (*work_done >= work_to_do)
1943                         break;
1944
1945                 (*work_done)++;
1946                 status = rx_desc->status;
1947                 skb = buffer_info->skb;
1948                 buffer_info->skb = NULL;
1949
1950                 prefetch(skb->data - NET_IP_ALIGN);
1951
1952                 if (++i == rx_ring->count) i = 0;
1953                 next_rxd = IXGB_RX_DESC(*rx_ring, i);
1954                 prefetch(next_rxd);
1955
1956                 if ((j = i + 1) == rx_ring->count) j = 0;
1957                 next2_buffer = &rx_ring->buffer_info[j];
1958                 prefetch(next2_buffer);
1959
1960                 next_buffer = &rx_ring->buffer_info[i];
1961
1962                 cleaned = true;
1963                 cleaned_count++;
1964
1965                 pci_unmap_single(pdev,
1966                                  buffer_info->dma,
1967                                  buffer_info->length,
1968                                  PCI_DMA_FROMDEVICE);
1969                 buffer_info->dma = 0;
1970
1971                 length = le16_to_cpu(rx_desc->length);
1972                 rx_desc->length = 0;
1973
1974                 if (unlikely(!(status & IXGB_RX_DESC_STATUS_EOP))) {
1975
1976                         /* All receives must fit into a single buffer */
1977
1978                         IXGB_DBG("Receive packet consumed multiple buffers "
1979                                          "length<%x>\n", length);
1980
1981                         dev_kfree_skb_irq(skb);
1982                         goto rxdesc_done;
1983                 }
1984
1985                 if (unlikely(rx_desc->errors &
1986                     (IXGB_RX_DESC_ERRORS_CE | IXGB_RX_DESC_ERRORS_SE |
1987                      IXGB_RX_DESC_ERRORS_P | IXGB_RX_DESC_ERRORS_RXE))) {
1988                         dev_kfree_skb_irq(skb);
1989                         goto rxdesc_done;
1990                 }
1991
1992                 /* code added for copybreak, this should improve
1993                  * performance for small packets with large amounts
1994                  * of reassembly being done in the stack */
1995                 if (length < copybreak) {
1996                         struct sk_buff *new_skb =
1997                             netdev_alloc_skb_ip_align(netdev, length);
1998                         if (new_skb) {
1999                                 skb_copy_to_linear_data_offset(new_skb,
2000                                                                -NET_IP_ALIGN,
2001                                                                (skb->data -
2002                                                                 NET_IP_ALIGN),
2003                                                                (length +
2004                                                                 NET_IP_ALIGN));
2005                                 /* save the skb in buffer_info as good */
2006                                 buffer_info->skb = skb;
2007                                 skb = new_skb;
2008                         }
2009                 }
2010                 /* end copybreak code */
2011
2012                 /* Good Receive */
2013                 skb_put(skb, length);
2014
2015                 /* Receive Checksum Offload */
2016                 ixgb_rx_checksum(adapter, rx_desc, skb);
2017
2018                 skb->protocol = eth_type_trans(skb, netdev);
2019                 if (adapter->vlgrp && (status & IXGB_RX_DESC_STATUS_VP)) {
2020                         vlan_hwaccel_receive_skb(skb, adapter->vlgrp,
2021                                                 le16_to_cpu(rx_desc->special));
2022                 } else {
2023                         netif_receive_skb(skb);
2024                 }
2025
2026 rxdesc_done:
2027                 /* clean up descriptor, might be written over by hw */
2028                 rx_desc->status = 0;
2029
2030                 /* return some buffers to hardware, one at a time is too slow */
2031                 if (unlikely(cleaned_count >= IXGB_RX_BUFFER_WRITE)) {
2032                         ixgb_alloc_rx_buffers(adapter, cleaned_count);
2033                         cleaned_count = 0;
2034                 }
2035
2036                 /* use prefetched values */
2037                 rx_desc = next_rxd;
2038                 buffer_info = next_buffer;
2039         }
2040
2041         rx_ring->next_to_clean = i;
2042
2043         cleaned_count = IXGB_DESC_UNUSED(rx_ring);
2044         if (cleaned_count)
2045                 ixgb_alloc_rx_buffers(adapter, cleaned_count);
2046
2047         return cleaned;
2048 }
2049
2050 /**
2051  * ixgb_alloc_rx_buffers - Replace used receive buffers
2052  * @adapter: address of board private structure
2053  **/
2054
2055 static void
2056 ixgb_alloc_rx_buffers(struct ixgb_adapter *adapter, int cleaned_count)
2057 {
2058         struct ixgb_desc_ring *rx_ring = &adapter->rx_ring;
2059         struct net_device *netdev = adapter->netdev;
2060         struct pci_dev *pdev = adapter->pdev;
2061         struct ixgb_rx_desc *rx_desc;
2062         struct ixgb_buffer *buffer_info;
2063         struct sk_buff *skb;
2064         unsigned int i;
2065         long cleancount;
2066
2067         i = rx_ring->next_to_use;
2068         buffer_info = &rx_ring->buffer_info[i];
2069         cleancount = IXGB_DESC_UNUSED(rx_ring);
2070
2071
2072         /* leave three descriptors unused */
2073         while (--cleancount > 2 && cleaned_count--) {
2074                 /* recycle! its good for you */
2075                 skb = buffer_info->skb;
2076                 if (skb) {
2077                         skb_trim(skb, 0);
2078                         goto map_skb;
2079                 }
2080
2081                 skb = netdev_alloc_skb_ip_align(netdev, adapter->rx_buffer_len);
2082                 if (unlikely(!skb)) {
2083                         /* Better luck next round */
2084                         adapter->alloc_rx_buff_failed++;
2085                         break;
2086                 }
2087
2088                 buffer_info->skb = skb;
2089                 buffer_info->length = adapter->rx_buffer_len;
2090 map_skb:
2091                 buffer_info->dma = pci_map_single(pdev,
2092                                                   skb->data,
2093                                                   adapter->rx_buffer_len,
2094                                                   PCI_DMA_FROMDEVICE);
2095
2096                 rx_desc = IXGB_RX_DESC(*rx_ring, i);
2097                 rx_desc->buff_addr = cpu_to_le64(buffer_info->dma);
2098                 /* guarantee DD bit not set now before h/w gets descriptor
2099                  * this is the rest of the workaround for h/w double
2100                  * writeback. */
2101                 rx_desc->status = 0;
2102
2103
2104                 if (++i == rx_ring->count) i = 0;
2105                 buffer_info = &rx_ring->buffer_info[i];
2106         }
2107
2108         if (likely(rx_ring->next_to_use != i)) {
2109                 rx_ring->next_to_use = i;
2110                 if (unlikely(i-- == 0))
2111                         i = (rx_ring->count - 1);
2112
2113                 /* Force memory writes to complete before letting h/w
2114                  * know there are new descriptors to fetch.  (Only
2115                  * applicable for weak-ordered memory model archs, such
2116                  * as IA-64). */
2117                 wmb();
2118                 IXGB_WRITE_REG(&adapter->hw, RDT, i);
2119         }
2120 }
2121
2122 /**
2123  * ixgb_vlan_rx_register - enables or disables vlan tagging/stripping.
2124  *
2125  * @param netdev network interface device structure
2126  * @param grp indicates to enable or disable tagging/stripping
2127  **/
2128 static void
2129 ixgb_vlan_rx_register(struct net_device *netdev, struct vlan_group *grp)
2130 {
2131         struct ixgb_adapter *adapter = netdev_priv(netdev);
2132         u32 ctrl, rctl;
2133
2134         ixgb_irq_disable(adapter);
2135         adapter->vlgrp = grp;
2136
2137         if (grp) {
2138                 /* enable VLAN tag insert/strip */
2139                 ctrl = IXGB_READ_REG(&adapter->hw, CTRL0);
2140                 ctrl |= IXGB_CTRL0_VME;
2141                 IXGB_WRITE_REG(&adapter->hw, CTRL0, ctrl);
2142
2143                 /* enable VLAN receive filtering */
2144
2145                 rctl = IXGB_READ_REG(&adapter->hw, RCTL);
2146                 rctl &= ~IXGB_RCTL_CFIEN;
2147                 IXGB_WRITE_REG(&adapter->hw, RCTL, rctl);
2148         } else {
2149                 /* disable VLAN tag insert/strip */
2150
2151                 ctrl = IXGB_READ_REG(&adapter->hw, CTRL0);
2152                 ctrl &= ~IXGB_CTRL0_VME;
2153                 IXGB_WRITE_REG(&adapter->hw, CTRL0, ctrl);
2154         }
2155
2156         /* don't enable interrupts unless we are UP */
2157         if (adapter->netdev->flags & IFF_UP)
2158                 ixgb_irq_enable(adapter);
2159 }
2160
2161 static void
2162 ixgb_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
2163 {
2164         struct ixgb_adapter *adapter = netdev_priv(netdev);
2165         u32 vfta, index;
2166
2167         /* add VID to filter table */
2168
2169         index = (vid >> 5) & 0x7F;
2170         vfta = IXGB_READ_REG_ARRAY(&adapter->hw, VFTA, index);
2171         vfta |= (1 << (vid & 0x1F));
2172         ixgb_write_vfta(&adapter->hw, index, vfta);
2173 }
2174
2175 static void
2176 ixgb_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
2177 {
2178         struct ixgb_adapter *adapter = netdev_priv(netdev);
2179         u32 vfta, index;
2180
2181         ixgb_irq_disable(adapter);
2182
2183         vlan_group_set_device(adapter->vlgrp, vid, NULL);
2184
2185         /* don't enable interrupts unless we are UP */
2186         if (adapter->netdev->flags & IFF_UP)
2187                 ixgb_irq_enable(adapter);
2188
2189         /* remove VID from filter table */
2190
2191         index = (vid >> 5) & 0x7F;
2192         vfta = IXGB_READ_REG_ARRAY(&adapter->hw, VFTA, index);
2193         vfta &= ~(1 << (vid & 0x1F));
2194         ixgb_write_vfta(&adapter->hw, index, vfta);
2195 }
2196
2197 static void
2198 ixgb_restore_vlan(struct ixgb_adapter *adapter)
2199 {
2200         ixgb_vlan_rx_register(adapter->netdev, adapter->vlgrp);
2201
2202         if (adapter->vlgrp) {
2203                 u16 vid;
2204                 for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
2205                         if (!vlan_group_get_device(adapter->vlgrp, vid))
2206                                 continue;
2207                         ixgb_vlan_rx_add_vid(adapter->netdev, vid);
2208                 }
2209         }
2210 }
2211
2212 #ifdef CONFIG_NET_POLL_CONTROLLER
2213 /*
2214  * Polling 'interrupt' - used by things like netconsole to send skbs
2215  * without having to re-enable interrupts. It's not called while
2216  * the interrupt routine is executing.
2217  */
2218
2219 static void ixgb_netpoll(struct net_device *dev)
2220 {
2221         struct ixgb_adapter *adapter = netdev_priv(dev);
2222
2223         disable_irq(adapter->pdev->irq);
2224         ixgb_intr(adapter->pdev->irq, dev);
2225         enable_irq(adapter->pdev->irq);
2226 }
2227 #endif
2228
2229 /**
2230  * ixgb_io_error_detected() - called when PCI error is detected
2231  * @pdev    pointer to pci device with error
2232  * @state   pci channel state after error
2233  *
2234  * This callback is called by the PCI subsystem whenever
2235  * a PCI bus error is detected.
2236  */
2237 static pci_ers_result_t ixgb_io_error_detected(struct pci_dev *pdev,
2238                                                enum pci_channel_state state)
2239 {
2240         struct net_device *netdev = pci_get_drvdata(pdev);
2241         struct ixgb_adapter *adapter = netdev_priv(netdev);
2242
2243         netif_device_detach(netdev);
2244
2245         if (state == pci_channel_io_perm_failure)
2246                 return PCI_ERS_RESULT_DISCONNECT;
2247
2248         if (netif_running(netdev))
2249                 ixgb_down(adapter, true);
2250
2251         pci_disable_device(pdev);
2252
2253         /* Request a slot reset. */
2254         return PCI_ERS_RESULT_NEED_RESET;
2255 }
2256
2257 /**
2258  * ixgb_io_slot_reset - called after the pci bus has been reset.
2259  * @pdev    pointer to pci device with error
2260  *
2261  * This callback is called after the PCI bus has been reset.
2262  * Basically, this tries to restart the card from scratch.
2263  * This is a shortened version of the device probe/discovery code,
2264  * it resembles the first-half of the ixgb_probe() routine.
2265  */
2266 static pci_ers_result_t ixgb_io_slot_reset(struct pci_dev *pdev)
2267 {
2268         struct net_device *netdev = pci_get_drvdata(pdev);
2269         struct ixgb_adapter *adapter = netdev_priv(netdev);
2270
2271         if (pci_enable_device(pdev)) {
2272                 DPRINTK(PROBE, ERR, "Cannot re-enable PCI device after reset.\n");
2273                 return PCI_ERS_RESULT_DISCONNECT;
2274         }
2275
2276         /* Perform card reset only on one instance of the card */
2277         if (0 != PCI_FUNC (pdev->devfn))
2278                 return PCI_ERS_RESULT_RECOVERED;
2279
2280         pci_set_master(pdev);
2281
2282         netif_carrier_off(netdev);
2283         netif_stop_queue(netdev);
2284         ixgb_reset(adapter);
2285
2286         /* Make sure the EEPROM is good */
2287         if (!ixgb_validate_eeprom_checksum(&adapter->hw)) {
2288                 DPRINTK(PROBE, ERR, "After reset, the EEPROM checksum is not valid.\n");
2289                 return PCI_ERS_RESULT_DISCONNECT;
2290         }
2291         ixgb_get_ee_mac_addr(&adapter->hw, netdev->dev_addr);
2292         memcpy(netdev->perm_addr, netdev->dev_addr, netdev->addr_len);
2293
2294         if (!is_valid_ether_addr(netdev->perm_addr)) {
2295                 DPRINTK(PROBE, ERR, "After reset, invalid MAC address.\n");
2296                 return PCI_ERS_RESULT_DISCONNECT;
2297         }
2298
2299         return PCI_ERS_RESULT_RECOVERED;
2300 }
2301
2302 /**
2303  * ixgb_io_resume - called when its OK to resume normal operations
2304  * @pdev    pointer to pci device with error
2305  *
2306  * The error recovery driver tells us that its OK to resume
2307  * normal operation. Implementation resembles the second-half
2308  * of the ixgb_probe() routine.
2309  */
2310 static void ixgb_io_resume(struct pci_dev *pdev)
2311 {
2312         struct net_device *netdev = pci_get_drvdata(pdev);
2313         struct ixgb_adapter *adapter = netdev_priv(netdev);
2314
2315         pci_set_master(pdev);
2316
2317         if (netif_running(netdev)) {
2318                 if (ixgb_up(adapter)) {
2319                         printk ("ixgb: can't bring device back up after reset\n");
2320                         return;
2321                 }
2322         }
2323
2324         netif_device_attach(netdev);
2325         mod_timer(&adapter->watchdog_timer, jiffies);
2326 }
2327
2328 /* ixgb_main.c */