pci: use pci_ioremap_bar() in drivers/net
[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 struct pci_device_id ixgb_pci_tbl[] = {
54         {INTEL_VENDOR_ID, IXGB_DEVICE_ID_82597EX,
55          PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0},
56         {INTEL_VENDOR_ID, IXGB_DEVICE_ID_82597EX_CX4,
57          PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0},
58         {INTEL_VENDOR_ID, IXGB_DEVICE_ID_82597EX_SR,
59          PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0},
60         {INTEL_VENDOR_ID, IXGB_DEVICE_ID_82597EX_LR,
61          PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0},
62
63         /* required last entry */
64         {0,}
65 };
66
67 MODULE_DEVICE_TABLE(pci, ixgb_pci_tbl);
68
69 /* Local Function Prototypes */
70 static int ixgb_init_module(void);
71 static void ixgb_exit_module(void);
72 static int ixgb_probe(struct pci_dev *pdev, const struct pci_device_id *ent);
73 static void __devexit ixgb_remove(struct pci_dev *pdev);
74 static int ixgb_sw_init(struct ixgb_adapter *adapter);
75 static int ixgb_open(struct net_device *netdev);
76 static int ixgb_close(struct net_device *netdev);
77 static void ixgb_configure_tx(struct ixgb_adapter *adapter);
78 static void ixgb_configure_rx(struct ixgb_adapter *adapter);
79 static void ixgb_setup_rctl(struct ixgb_adapter *adapter);
80 static void ixgb_clean_tx_ring(struct ixgb_adapter *adapter);
81 static void ixgb_clean_rx_ring(struct ixgb_adapter *adapter);
82 static void ixgb_set_multi(struct net_device *netdev);
83 static void ixgb_watchdog(unsigned long data);
84 static int ixgb_xmit_frame(struct sk_buff *skb, struct net_device *netdev);
85 static struct net_device_stats *ixgb_get_stats(struct net_device *netdev);
86 static int ixgb_change_mtu(struct net_device *netdev, int new_mtu);
87 static int ixgb_set_mac(struct net_device *netdev, void *p);
88 static irqreturn_t ixgb_intr(int irq, void *data);
89 static bool ixgb_clean_tx_irq(struct ixgb_adapter *adapter);
90
91 static int ixgb_clean(struct napi_struct *, int);
92 static bool ixgb_clean_rx_irq(struct ixgb_adapter *, int *, int);
93 static void ixgb_alloc_rx_buffers(struct ixgb_adapter *, int);
94
95 static void ixgb_tx_timeout(struct net_device *dev);
96 static void ixgb_tx_timeout_task(struct work_struct *work);
97
98 static void ixgb_vlan_rx_register(struct net_device *netdev,
99                                   struct vlan_group *grp);
100 static void ixgb_vlan_rx_add_vid(struct net_device *netdev, u16 vid);
101 static void ixgb_vlan_rx_kill_vid(struct net_device *netdev, u16 vid);
102 static void ixgb_restore_vlan(struct ixgb_adapter *adapter);
103
104 #ifdef CONFIG_NET_POLL_CONTROLLER
105 /* for netdump / net console */
106 static void ixgb_netpoll(struct net_device *dev);
107 #endif
108
109 static pci_ers_result_t ixgb_io_error_detected (struct pci_dev *pdev,
110                              enum pci_channel_state state);
111 static pci_ers_result_t ixgb_io_slot_reset (struct pci_dev *pdev);
112 static void ixgb_io_resume (struct pci_dev *pdev);
113
114 static struct pci_error_handlers ixgb_err_handler = {
115         .error_detected = ixgb_io_error_detected,
116         .slot_reset = ixgb_io_slot_reset,
117         .resume = ixgb_io_resume,
118 };
119
120 static struct pci_driver ixgb_driver = {
121         .name     = ixgb_driver_name,
122         .id_table = ixgb_pci_tbl,
123         .probe    = ixgb_probe,
124         .remove   = __devexit_p(ixgb_remove),
125         .err_handler = &ixgb_err_handler
126 };
127
128 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
129 MODULE_DESCRIPTION("Intel(R) PRO/10GbE Network Driver");
130 MODULE_LICENSE("GPL");
131 MODULE_VERSION(DRV_VERSION);
132
133 #define DEFAULT_DEBUG_LEVEL_SHIFT 3
134 static int debug = DEFAULT_DEBUG_LEVEL_SHIFT;
135 module_param(debug, int, 0);
136 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
137
138 /**
139  * ixgb_init_module - Driver Registration Routine
140  *
141  * ixgb_init_module is the first routine called when the driver is
142  * loaded. All it does is register with the PCI subsystem.
143  **/
144
145 static int __init
146 ixgb_init_module(void)
147 {
148         printk(KERN_INFO "%s - version %s\n",
149                ixgb_driver_string, ixgb_driver_version);
150
151         printk(KERN_INFO "%s\n", ixgb_copyright);
152
153         return pci_register_driver(&ixgb_driver);
154 }
155
156 module_init(ixgb_init_module);
157
158 /**
159  * ixgb_exit_module - Driver Exit Cleanup Routine
160  *
161  * ixgb_exit_module is called just before the driver is removed
162  * from memory.
163  **/
164
165 static void __exit
166 ixgb_exit_module(void)
167 {
168         pci_unregister_driver(&ixgb_driver);
169 }
170
171 module_exit(ixgb_exit_module);
172
173 /**
174  * ixgb_irq_disable - Mask off interrupt generation on the NIC
175  * @adapter: board private structure
176  **/
177
178 static void
179 ixgb_irq_disable(struct ixgb_adapter *adapter)
180 {
181         IXGB_WRITE_REG(&adapter->hw, IMC, ~0);
182         IXGB_WRITE_FLUSH(&adapter->hw);
183         synchronize_irq(adapter->pdev->irq);
184 }
185
186 /**
187  * ixgb_irq_enable - Enable default interrupt generation settings
188  * @adapter: board private structure
189  **/
190
191 static void
192 ixgb_irq_enable(struct ixgb_adapter *adapter)
193 {
194         u32 val = IXGB_INT_RXT0 | IXGB_INT_RXDMT0 |
195                   IXGB_INT_TXDW | IXGB_INT_LSC;
196         if (adapter->hw.subsystem_vendor_id == SUN_SUBVENDOR_ID)
197                 val |= IXGB_INT_GPI0;
198         IXGB_WRITE_REG(&adapter->hw, IMS, val);
199         IXGB_WRITE_FLUSH(&adapter->hw);
200 }
201
202 int
203 ixgb_up(struct ixgb_adapter *adapter)
204 {
205         struct net_device *netdev = adapter->netdev;
206         int err, irq_flags = IRQF_SHARED;
207         int max_frame = netdev->mtu + ENET_HEADER_SIZE + ENET_FCS_LENGTH;
208         struct ixgb_hw *hw = &adapter->hw;
209
210         /* hardware has been reset, we need to reload some things */
211
212         ixgb_rar_set(hw, netdev->dev_addr, 0);
213         ixgb_set_multi(netdev);
214
215         ixgb_restore_vlan(adapter);
216
217         ixgb_configure_tx(adapter);
218         ixgb_setup_rctl(adapter);
219         ixgb_configure_rx(adapter);
220         ixgb_alloc_rx_buffers(adapter, IXGB_DESC_UNUSED(&adapter->rx_ring));
221
222         /* disable interrupts and get the hardware into a known state */
223         IXGB_WRITE_REG(&adapter->hw, IMC, 0xffffffff);
224
225         /* only enable MSI if bus is in PCI-X mode */
226         if (IXGB_READ_REG(&adapter->hw, STATUS) & IXGB_STATUS_PCIX_MODE) {
227                 err = pci_enable_msi(adapter->pdev);
228                 if (!err) {
229                         adapter->have_msi = 1;
230                         irq_flags = 0;
231                 }
232                 /* proceed to try to request regular interrupt */
233         }
234
235         err = request_irq(adapter->pdev->irq, &ixgb_intr, irq_flags,
236                           netdev->name, netdev);
237         if (err) {
238                 if (adapter->have_msi)
239                         pci_disable_msi(adapter->pdev);
240                 DPRINTK(PROBE, ERR,
241                  "Unable to allocate interrupt Error: %d\n", err);
242                 return err;
243         }
244
245         if ((hw->max_frame_size != max_frame) ||
246                 (hw->max_frame_size !=
247                 (IXGB_READ_REG(hw, MFS) >> IXGB_MFS_SHIFT))) {
248
249                 hw->max_frame_size = max_frame;
250
251                 IXGB_WRITE_REG(hw, MFS, hw->max_frame_size << IXGB_MFS_SHIFT);
252
253                 if (hw->max_frame_size >
254                    IXGB_MAX_ENET_FRAME_SIZE_WITHOUT_FCS + ENET_FCS_LENGTH) {
255                         u32 ctrl0 = IXGB_READ_REG(hw, CTRL0);
256
257                         if (!(ctrl0 & IXGB_CTRL0_JFE)) {
258                                 ctrl0 |= IXGB_CTRL0_JFE;
259                                 IXGB_WRITE_REG(hw, CTRL0, ctrl0);
260                         }
261                 }
262         }
263
264         clear_bit(__IXGB_DOWN, &adapter->flags);
265
266         napi_enable(&adapter->napi);
267         ixgb_irq_enable(adapter);
268
269         mod_timer(&adapter->watchdog_timer, jiffies);
270
271         return 0;
272 }
273
274 void
275 ixgb_down(struct ixgb_adapter *adapter, bool kill_watchdog)
276 {
277         struct net_device *netdev = adapter->netdev;
278
279         /* prevent the interrupt handler from restarting watchdog */
280         set_bit(__IXGB_DOWN, &adapter->flags);
281
282         napi_disable(&adapter->napi);
283         /* waiting for NAPI to complete can re-enable interrupts */
284         ixgb_irq_disable(adapter);
285         free_irq(adapter->pdev->irq, netdev);
286
287         if (adapter->have_msi)
288                 pci_disable_msi(adapter->pdev);
289
290         if (kill_watchdog)
291                 del_timer_sync(&adapter->watchdog_timer);
292
293         adapter->link_speed = 0;
294         adapter->link_duplex = 0;
295         netif_carrier_off(netdev);
296         netif_stop_queue(netdev);
297
298         ixgb_reset(adapter);
299         ixgb_clean_tx_ring(adapter);
300         ixgb_clean_rx_ring(adapter);
301 }
302
303 void
304 ixgb_reset(struct ixgb_adapter *adapter)
305 {
306         struct ixgb_hw *hw = &adapter->hw;
307
308         ixgb_adapter_stop(hw);
309         if (!ixgb_init_hw(hw))
310                 DPRINTK(PROBE, ERR, "ixgb_init_hw failed.\n");
311
312         /* restore frame size information */
313         IXGB_WRITE_REG(hw, MFS, hw->max_frame_size << IXGB_MFS_SHIFT);
314         if (hw->max_frame_size >
315             IXGB_MAX_ENET_FRAME_SIZE_WITHOUT_FCS + ENET_FCS_LENGTH) {
316                 u32 ctrl0 = IXGB_READ_REG(hw, CTRL0);
317                 if (!(ctrl0 & IXGB_CTRL0_JFE)) {
318                         ctrl0 |= IXGB_CTRL0_JFE;
319                         IXGB_WRITE_REG(hw, CTRL0, ctrl0);
320                 }
321         }
322 }
323
324 /**
325  * ixgb_probe - Device Initialization Routine
326  * @pdev: PCI device information struct
327  * @ent: entry in ixgb_pci_tbl
328  *
329  * Returns 0 on success, negative on failure
330  *
331  * ixgb_probe initializes an adapter identified by a pci_dev structure.
332  * The OS initialization, configuring of the adapter private structure,
333  * and a hardware reset occur.
334  **/
335
336 static int __devinit
337 ixgb_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
338 {
339         struct net_device *netdev = NULL;
340         struct ixgb_adapter *adapter;
341         static int cards_found = 0;
342         int pci_using_dac;
343         int i;
344         int err;
345
346         err = pci_enable_device(pdev);
347         if (err)
348                 return err;
349
350         if (!(err = pci_set_dma_mask(pdev, DMA_64BIT_MASK)) &&
351             !(err = pci_set_consistent_dma_mask(pdev, DMA_64BIT_MASK))) {
352                 pci_using_dac = 1;
353         } else {
354                 if ((err = pci_set_dma_mask(pdev, DMA_32BIT_MASK)) ||
355                     (err = pci_set_consistent_dma_mask(pdev, DMA_32BIT_MASK))) {
356                         printk(KERN_ERR
357                          "ixgb: No usable DMA configuration, aborting\n");
358                         goto err_dma_mask;
359                 }
360                 pci_using_dac = 0;
361         }
362
363         err = pci_request_regions(pdev, ixgb_driver_name);
364         if (err)
365                 goto err_request_regions;
366
367         pci_set_master(pdev);
368
369         netdev = alloc_etherdev(sizeof(struct ixgb_adapter));
370         if (!netdev) {
371                 err = -ENOMEM;
372                 goto err_alloc_etherdev;
373         }
374
375         SET_NETDEV_DEV(netdev, &pdev->dev);
376
377         pci_set_drvdata(pdev, netdev);
378         adapter = netdev_priv(netdev);
379         adapter->netdev = netdev;
380         adapter->pdev = pdev;
381         adapter->hw.back = adapter;
382         adapter->msg_enable = netif_msg_init(debug, DEFAULT_DEBUG_LEVEL_SHIFT);
383
384         adapter->hw.hw_addr = pci_ioremap_bar(pdev, BAR_0);
385         if (!adapter->hw.hw_addr) {
386                 err = -EIO;
387                 goto err_ioremap;
388         }
389
390         for (i = BAR_1; i <= BAR_5; i++) {
391                 if (pci_resource_len(pdev, i) == 0)
392                         continue;
393                 if (pci_resource_flags(pdev, i) & IORESOURCE_IO) {
394                         adapter->hw.io_base = pci_resource_start(pdev, i);
395                         break;
396                 }
397         }
398
399         netdev->open = &ixgb_open;
400         netdev->stop = &ixgb_close;
401         netdev->hard_start_xmit = &ixgb_xmit_frame;
402         netdev->get_stats = &ixgb_get_stats;
403         netdev->set_multicast_list = &ixgb_set_multi;
404         netdev->set_mac_address = &ixgb_set_mac;
405         netdev->change_mtu = &ixgb_change_mtu;
406         ixgb_set_ethtool_ops(netdev);
407         netdev->tx_timeout = &ixgb_tx_timeout;
408         netdev->watchdog_timeo = 5 * HZ;
409         netif_napi_add(netdev, &adapter->napi, ixgb_clean, 64);
410         netdev->vlan_rx_register = ixgb_vlan_rx_register;
411         netdev->vlan_rx_add_vid = ixgb_vlan_rx_add_vid;
412         netdev->vlan_rx_kill_vid = ixgb_vlan_rx_kill_vid;
413 #ifdef CONFIG_NET_POLL_CONTROLLER
414         netdev->poll_controller = ixgb_netpoll;
415 #endif
416
417         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
418
419         adapter->bd_number = cards_found;
420         adapter->link_speed = 0;
421         adapter->link_duplex = 0;
422
423         /* setup the private structure */
424
425         err = ixgb_sw_init(adapter);
426         if (err)
427                 goto err_sw_init;
428
429         netdev->features = NETIF_F_SG |
430                            NETIF_F_HW_CSUM |
431                            NETIF_F_HW_VLAN_TX |
432                            NETIF_F_HW_VLAN_RX |
433                            NETIF_F_HW_VLAN_FILTER;
434         netdev->features |= NETIF_F_TSO;
435
436         if (pci_using_dac)
437                 netdev->features |= NETIF_F_HIGHDMA;
438
439         /* make sure the EEPROM is good */
440
441         if (!ixgb_validate_eeprom_checksum(&adapter->hw)) {
442                 DPRINTK(PROBE, ERR, "The EEPROM Checksum Is Not Valid\n");
443                 err = -EIO;
444                 goto err_eeprom;
445         }
446
447         ixgb_get_ee_mac_addr(&adapter->hw, netdev->dev_addr);
448         memcpy(netdev->perm_addr, netdev->dev_addr, netdev->addr_len);
449
450         if (!is_valid_ether_addr(netdev->perm_addr)) {
451                 DPRINTK(PROBE, ERR, "Invalid MAC Address\n");
452                 err = -EIO;
453                 goto err_eeprom;
454         }
455
456         adapter->part_num = ixgb_get_ee_pba_number(&adapter->hw);
457
458         init_timer(&adapter->watchdog_timer);
459         adapter->watchdog_timer.function = &ixgb_watchdog;
460         adapter->watchdog_timer.data = (unsigned long)adapter;
461
462         INIT_WORK(&adapter->tx_timeout_task, ixgb_tx_timeout_task);
463
464         strcpy(netdev->name, "eth%d");
465         err = register_netdev(netdev);
466         if (err)
467                 goto err_register;
468
469         /* we're going to reset, so assume we have no link for now */
470
471         netif_carrier_off(netdev);
472         netif_stop_queue(netdev);
473
474         DPRINTK(PROBE, INFO, "Intel(R) PRO/10GbE Network Connection\n");
475         ixgb_check_options(adapter);
476         /* reset the hardware with the new settings */
477
478         ixgb_reset(adapter);
479
480         cards_found++;
481         return 0;
482
483 err_register:
484 err_sw_init:
485 err_eeprom:
486         iounmap(adapter->hw.hw_addr);
487 err_ioremap:
488         free_netdev(netdev);
489 err_alloc_etherdev:
490         pci_release_regions(pdev);
491 err_request_regions:
492 err_dma_mask:
493         pci_disable_device(pdev);
494         return err;
495 }
496
497 /**
498  * ixgb_remove - Device Removal Routine
499  * @pdev: PCI device information struct
500  *
501  * ixgb_remove is called by the PCI subsystem to alert the driver
502  * that it should release a PCI device.  The could be caused by a
503  * Hot-Plug event, or because the driver is going to be removed from
504  * memory.
505  **/
506
507 static void __devexit
508 ixgb_remove(struct pci_dev *pdev)
509 {
510         struct net_device *netdev = pci_get_drvdata(pdev);
511         struct ixgb_adapter *adapter = netdev_priv(netdev);
512
513         flush_scheduled_work();
514
515         unregister_netdev(netdev);
516
517         iounmap(adapter->hw.hw_addr);
518         pci_release_regions(pdev);
519
520         free_netdev(netdev);
521 }
522
523 /**
524  * ixgb_sw_init - Initialize general software structures (struct ixgb_adapter)
525  * @adapter: board private structure to initialize
526  *
527  * ixgb_sw_init initializes the Adapter private data structure.
528  * Fields are initialized based on PCI device information and
529  * OS network device settings (MTU size).
530  **/
531
532 static int __devinit
533 ixgb_sw_init(struct ixgb_adapter *adapter)
534 {
535         struct ixgb_hw *hw = &adapter->hw;
536         struct net_device *netdev = adapter->netdev;
537         struct pci_dev *pdev = adapter->pdev;
538
539         /* PCI config space info */
540
541         hw->vendor_id = pdev->vendor;
542         hw->device_id = pdev->device;
543         hw->subsystem_vendor_id = pdev->subsystem_vendor;
544         hw->subsystem_id = pdev->subsystem_device;
545
546         hw->max_frame_size = netdev->mtu + ENET_HEADER_SIZE + ENET_FCS_LENGTH;
547         adapter->rx_buffer_len = hw->max_frame_size + 8; /* + 8 for errata */
548
549         if ((hw->device_id == IXGB_DEVICE_ID_82597EX)
550            || (hw->device_id == IXGB_DEVICE_ID_82597EX_CX4)
551            || (hw->device_id == IXGB_DEVICE_ID_82597EX_LR)
552            || (hw->device_id == IXGB_DEVICE_ID_82597EX_SR))
553                 hw->mac_type = ixgb_82597;
554         else {
555                 /* should never have loaded on this device */
556                 DPRINTK(PROBE, ERR, "unsupported device id\n");
557         }
558
559         /* enable flow control to be programmed */
560         hw->fc.send_xon = 1;
561
562         set_bit(__IXGB_DOWN, &adapter->flags);
563         return 0;
564 }
565
566 /**
567  * ixgb_open - Called when a network interface is made active
568  * @netdev: network interface device structure
569  *
570  * Returns 0 on success, negative value on failure
571  *
572  * The open entry point is called when a network interface is made
573  * active by the system (IFF_UP).  At this point all resources needed
574  * for transmit and receive operations are allocated, the interrupt
575  * handler is registered with the OS, the watchdog timer is started,
576  * and the stack is notified that the interface is ready.
577  **/
578
579 static int
580 ixgb_open(struct net_device *netdev)
581 {
582         struct ixgb_adapter *adapter = netdev_priv(netdev);
583         int err;
584
585         /* allocate transmit descriptors */
586         err = ixgb_setup_tx_resources(adapter);
587         if (err)
588                 goto err_setup_tx;
589
590         /* allocate receive descriptors */
591
592         err = ixgb_setup_rx_resources(adapter);
593         if (err)
594                 goto err_setup_rx;
595
596         err = ixgb_up(adapter);
597         if (err)
598                 goto err_up;
599
600         return 0;
601
602 err_up:
603         ixgb_free_rx_resources(adapter);
604 err_setup_rx:
605         ixgb_free_tx_resources(adapter);
606 err_setup_tx:
607         ixgb_reset(adapter);
608
609         return err;
610 }
611
612 /**
613  * ixgb_close - Disables a network interface
614  * @netdev: network interface device structure
615  *
616  * Returns 0, this is not allowed to fail
617  *
618  * The close entry point is called when an interface is de-activated
619  * by the OS.  The hardware is still under the drivers control, but
620  * needs to be disabled.  A global MAC reset is issued to stop the
621  * hardware, and all transmit and receive resources are freed.
622  **/
623
624 static int
625 ixgb_close(struct net_device *netdev)
626 {
627         struct ixgb_adapter *adapter = netdev_priv(netdev);
628
629         ixgb_down(adapter, true);
630
631         ixgb_free_tx_resources(adapter);
632         ixgb_free_rx_resources(adapter);
633
634         return 0;
635 }
636
637 /**
638  * ixgb_setup_tx_resources - allocate Tx resources (Descriptors)
639  * @adapter: board private structure
640  *
641  * Return 0 on success, negative on failure
642  **/
643
644 int
645 ixgb_setup_tx_resources(struct ixgb_adapter *adapter)
646 {
647         struct ixgb_desc_ring *txdr = &adapter->tx_ring;
648         struct pci_dev *pdev = adapter->pdev;
649         int size;
650
651         size = sizeof(struct ixgb_buffer) * txdr->count;
652         txdr->buffer_info = vmalloc(size);
653         if (!txdr->buffer_info) {
654                 DPRINTK(PROBE, ERR,
655                  "Unable to allocate transmit descriptor ring memory\n");
656                 return -ENOMEM;
657         }
658         memset(txdr->buffer_info, 0, size);
659
660         /* round up to nearest 4K */
661
662         txdr->size = txdr->count * sizeof(struct ixgb_tx_desc);
663         txdr->size = ALIGN(txdr->size, 4096);
664
665         txdr->desc = pci_alloc_consistent(pdev, txdr->size, &txdr->dma);
666         if (!txdr->desc) {
667                 vfree(txdr->buffer_info);
668                 DPRINTK(PROBE, ERR,
669                  "Unable to allocate transmit descriptor memory\n");
670                 return -ENOMEM;
671         }
672         memset(txdr->desc, 0, txdr->size);
673
674         txdr->next_to_use = 0;
675         txdr->next_to_clean = 0;
676
677         return 0;
678 }
679
680 /**
681  * ixgb_configure_tx - Configure 82597 Transmit Unit after Reset.
682  * @adapter: board private structure
683  *
684  * Configure the Tx unit of the MAC after a reset.
685  **/
686
687 static void
688 ixgb_configure_tx(struct ixgb_adapter *adapter)
689 {
690         u64 tdba = adapter->tx_ring.dma;
691         u32 tdlen = adapter->tx_ring.count * sizeof(struct ixgb_tx_desc);
692         u32 tctl;
693         struct ixgb_hw *hw = &adapter->hw;
694
695         /* Setup the Base and Length of the Tx Descriptor Ring
696          * tx_ring.dma can be either a 32 or 64 bit value
697          */
698
699         IXGB_WRITE_REG(hw, TDBAL, (tdba & 0x00000000ffffffffULL));
700         IXGB_WRITE_REG(hw, TDBAH, (tdba >> 32));
701
702         IXGB_WRITE_REG(hw, TDLEN, tdlen);
703
704         /* Setup the HW Tx Head and Tail descriptor pointers */
705
706         IXGB_WRITE_REG(hw, TDH, 0);
707         IXGB_WRITE_REG(hw, TDT, 0);
708
709         /* don't set up txdctl, it induces performance problems if configured
710          * incorrectly */
711         /* Set the Tx Interrupt Delay register */
712
713         IXGB_WRITE_REG(hw, TIDV, adapter->tx_int_delay);
714
715         /* Program the Transmit Control Register */
716
717         tctl = IXGB_TCTL_TCE | IXGB_TCTL_TXEN | IXGB_TCTL_TPDE;
718         IXGB_WRITE_REG(hw, TCTL, tctl);
719
720         /* Setup Transmit Descriptor Settings for this adapter */
721         adapter->tx_cmd_type =
722                 IXGB_TX_DESC_TYPE |
723                 (adapter->tx_int_delay_enable ? IXGB_TX_DESC_CMD_IDE : 0);
724 }
725
726 /**
727  * ixgb_setup_rx_resources - allocate Rx resources (Descriptors)
728  * @adapter: board private structure
729  *
730  * Returns 0 on success, negative on failure
731  **/
732
733 int
734 ixgb_setup_rx_resources(struct ixgb_adapter *adapter)
735 {
736         struct ixgb_desc_ring *rxdr = &adapter->rx_ring;
737         struct pci_dev *pdev = adapter->pdev;
738         int size;
739
740         size = sizeof(struct ixgb_buffer) * rxdr->count;
741         rxdr->buffer_info = vmalloc(size);
742         if (!rxdr->buffer_info) {
743                 DPRINTK(PROBE, ERR,
744                  "Unable to allocate receive descriptor ring\n");
745                 return -ENOMEM;
746         }
747         memset(rxdr->buffer_info, 0, size);
748
749         /* Round up to nearest 4K */
750
751         rxdr->size = rxdr->count * sizeof(struct ixgb_rx_desc);
752         rxdr->size = ALIGN(rxdr->size, 4096);
753
754         rxdr->desc = pci_alloc_consistent(pdev, rxdr->size, &rxdr->dma);
755
756         if (!rxdr->desc) {
757                 vfree(rxdr->buffer_info);
758                 DPRINTK(PROBE, ERR,
759                  "Unable to allocate receive descriptors\n");
760                 return -ENOMEM;
761         }
762         memset(rxdr->desc, 0, rxdr->size);
763
764         rxdr->next_to_clean = 0;
765         rxdr->next_to_use = 0;
766
767         return 0;
768 }
769
770 /**
771  * ixgb_setup_rctl - configure the receive control register
772  * @adapter: Board private structure
773  **/
774
775 static void
776 ixgb_setup_rctl(struct ixgb_adapter *adapter)
777 {
778         u32 rctl;
779
780         rctl = IXGB_READ_REG(&adapter->hw, RCTL);
781
782         rctl &= ~(3 << IXGB_RCTL_MO_SHIFT);
783
784         rctl |=
785                 IXGB_RCTL_BAM | IXGB_RCTL_RDMTS_1_2 |
786                 IXGB_RCTL_RXEN | IXGB_RCTL_CFF |
787                 (adapter->hw.mc_filter_type << IXGB_RCTL_MO_SHIFT);
788
789         rctl |= IXGB_RCTL_SECRC;
790
791         if (adapter->rx_buffer_len <= IXGB_RXBUFFER_2048)
792                 rctl |= IXGB_RCTL_BSIZE_2048;
793         else if (adapter->rx_buffer_len <= IXGB_RXBUFFER_4096)
794                 rctl |= IXGB_RCTL_BSIZE_4096;
795         else if (adapter->rx_buffer_len <= IXGB_RXBUFFER_8192)
796                 rctl |= IXGB_RCTL_BSIZE_8192;
797         else if (adapter->rx_buffer_len <= IXGB_RXBUFFER_16384)
798                 rctl |= IXGB_RCTL_BSIZE_16384;
799
800         IXGB_WRITE_REG(&adapter->hw, RCTL, rctl);
801 }
802
803 /**
804  * ixgb_configure_rx - Configure 82597 Receive Unit after Reset.
805  * @adapter: board private structure
806  *
807  * Configure the Rx unit of the MAC after a reset.
808  **/
809
810 static void
811 ixgb_configure_rx(struct ixgb_adapter *adapter)
812 {
813         u64 rdba = adapter->rx_ring.dma;
814         u32 rdlen = adapter->rx_ring.count * sizeof(struct ixgb_rx_desc);
815         struct ixgb_hw *hw = &adapter->hw;
816         u32 rctl;
817         u32 rxcsum;
818
819         /* make sure receives are disabled while setting up the descriptors */
820
821         rctl = IXGB_READ_REG(hw, RCTL);
822         IXGB_WRITE_REG(hw, RCTL, rctl & ~IXGB_RCTL_RXEN);
823
824         /* set the Receive Delay Timer Register */
825
826         IXGB_WRITE_REG(hw, RDTR, adapter->rx_int_delay);
827
828         /* Setup the Base and Length of the Rx Descriptor Ring */
829
830         IXGB_WRITE_REG(hw, RDBAL, (rdba & 0x00000000ffffffffULL));
831         IXGB_WRITE_REG(hw, RDBAH, (rdba >> 32));
832
833         IXGB_WRITE_REG(hw, RDLEN, rdlen);
834
835         /* Setup the HW Rx Head and Tail Descriptor Pointers */
836         IXGB_WRITE_REG(hw, RDH, 0);
837         IXGB_WRITE_REG(hw, RDT, 0);
838
839         /* due to the hardware errata with RXDCTL, we are unable to use any of
840          * the performance enhancing features of it without causing other
841          * subtle bugs, some of the bugs could include receive length
842          * corruption at high data rates (WTHRESH > 0) and/or receive
843          * descriptor ring irregularites (particularly in hardware cache) */
844         IXGB_WRITE_REG(hw, RXDCTL, 0);
845
846         /* Enable Receive Checksum Offload for TCP and UDP */
847         if (adapter->rx_csum) {
848                 rxcsum = IXGB_READ_REG(hw, RXCSUM);
849                 rxcsum |= IXGB_RXCSUM_TUOFL;
850                 IXGB_WRITE_REG(hw, RXCSUM, rxcsum);
851         }
852
853         /* Enable Receives */
854
855         IXGB_WRITE_REG(hw, RCTL, rctl);
856 }
857
858 /**
859  * ixgb_free_tx_resources - Free Tx Resources
860  * @adapter: board private structure
861  *
862  * Free all transmit software resources
863  **/
864
865 void
866 ixgb_free_tx_resources(struct ixgb_adapter *adapter)
867 {
868         struct pci_dev *pdev = adapter->pdev;
869
870         ixgb_clean_tx_ring(adapter);
871
872         vfree(adapter->tx_ring.buffer_info);
873         adapter->tx_ring.buffer_info = NULL;
874
875         pci_free_consistent(pdev, adapter->tx_ring.size,
876                             adapter->tx_ring.desc, adapter->tx_ring.dma);
877
878         adapter->tx_ring.desc = NULL;
879 }
880
881 static void
882 ixgb_unmap_and_free_tx_resource(struct ixgb_adapter *adapter,
883                                 struct ixgb_buffer *buffer_info)
884 {
885         struct pci_dev *pdev = adapter->pdev;
886
887         if (buffer_info->dma)
888                 pci_unmap_page(pdev, buffer_info->dma, buffer_info->length,
889                                PCI_DMA_TODEVICE);
890
891         /* okay to call kfree_skb here instead of kfree_skb_any because
892          * this is never called in interrupt context */
893         if (buffer_info->skb)
894                 dev_kfree_skb(buffer_info->skb);
895
896         buffer_info->skb = NULL;
897         buffer_info->dma = 0;
898         buffer_info->time_stamp = 0;
899         /* these fields must always be initialized in tx
900          * buffer_info->length = 0;
901          * buffer_info->next_to_watch = 0; */
902 }
903
904 /**
905  * ixgb_clean_tx_ring - Free Tx Buffers
906  * @adapter: board private structure
907  **/
908
909 static void
910 ixgb_clean_tx_ring(struct ixgb_adapter *adapter)
911 {
912         struct ixgb_desc_ring *tx_ring = &adapter->tx_ring;
913         struct ixgb_buffer *buffer_info;
914         unsigned long size;
915         unsigned int i;
916
917         /* Free all the Tx ring sk_buffs */
918
919         for (i = 0; i < tx_ring->count; i++) {
920                 buffer_info = &tx_ring->buffer_info[i];
921                 ixgb_unmap_and_free_tx_resource(adapter, buffer_info);
922         }
923
924         size = sizeof(struct ixgb_buffer) * tx_ring->count;
925         memset(tx_ring->buffer_info, 0, size);
926
927         /* Zero out the descriptor ring */
928
929         memset(tx_ring->desc, 0, tx_ring->size);
930
931         tx_ring->next_to_use = 0;
932         tx_ring->next_to_clean = 0;
933
934         IXGB_WRITE_REG(&adapter->hw, TDH, 0);
935         IXGB_WRITE_REG(&adapter->hw, TDT, 0);
936 }
937
938 /**
939  * ixgb_free_rx_resources - Free Rx Resources
940  * @adapter: board private structure
941  *
942  * Free all receive software resources
943  **/
944
945 void
946 ixgb_free_rx_resources(struct ixgb_adapter *adapter)
947 {
948         struct ixgb_desc_ring *rx_ring = &adapter->rx_ring;
949         struct pci_dev *pdev = adapter->pdev;
950
951         ixgb_clean_rx_ring(adapter);
952
953         vfree(rx_ring->buffer_info);
954         rx_ring->buffer_info = NULL;
955
956         pci_free_consistent(pdev, rx_ring->size, rx_ring->desc, rx_ring->dma);
957
958         rx_ring->desc = NULL;
959 }
960
961 /**
962  * ixgb_clean_rx_ring - Free Rx Buffers
963  * @adapter: board private structure
964  **/
965
966 static void
967 ixgb_clean_rx_ring(struct ixgb_adapter *adapter)
968 {
969         struct ixgb_desc_ring *rx_ring = &adapter->rx_ring;
970         struct ixgb_buffer *buffer_info;
971         struct pci_dev *pdev = adapter->pdev;
972         unsigned long size;
973         unsigned int i;
974
975         /* Free all the Rx ring sk_buffs */
976
977         for (i = 0; i < rx_ring->count; i++) {
978                 buffer_info = &rx_ring->buffer_info[i];
979                 if (buffer_info->dma) {
980                         pci_unmap_single(pdev,
981                                          buffer_info->dma,
982                                          buffer_info->length,
983                                          PCI_DMA_FROMDEVICE);
984                         buffer_info->dma = 0;
985                         buffer_info->length = 0;
986                 }
987
988                 if (buffer_info->skb) {
989                         dev_kfree_skb(buffer_info->skb);
990                         buffer_info->skb = NULL;
991                 }
992         }
993
994         size = sizeof(struct ixgb_buffer) * rx_ring->count;
995         memset(rx_ring->buffer_info, 0, size);
996
997         /* Zero out the descriptor ring */
998
999         memset(rx_ring->desc, 0, rx_ring->size);
1000
1001         rx_ring->next_to_clean = 0;
1002         rx_ring->next_to_use = 0;
1003
1004         IXGB_WRITE_REG(&adapter->hw, RDH, 0);
1005         IXGB_WRITE_REG(&adapter->hw, RDT, 0);
1006 }
1007
1008 /**
1009  * ixgb_set_mac - Change the Ethernet Address of the NIC
1010  * @netdev: network interface device structure
1011  * @p: pointer to an address structure
1012  *
1013  * Returns 0 on success, negative on failure
1014  **/
1015
1016 static int
1017 ixgb_set_mac(struct net_device *netdev, void *p)
1018 {
1019         struct ixgb_adapter *adapter = netdev_priv(netdev);
1020         struct sockaddr *addr = p;
1021
1022         if (!is_valid_ether_addr(addr->sa_data))
1023                 return -EADDRNOTAVAIL;
1024
1025         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
1026
1027         ixgb_rar_set(&adapter->hw, addr->sa_data, 0);
1028
1029         return 0;
1030 }
1031
1032 /**
1033  * ixgb_set_multi - Multicast and Promiscuous mode set
1034  * @netdev: network interface device structure
1035  *
1036  * The set_multi entry point is called whenever the multicast address
1037  * list or the network interface flags are updated.  This routine is
1038  * responsible for configuring the hardware for proper multicast,
1039  * promiscuous mode, and all-multi behavior.
1040  **/
1041
1042 static void
1043 ixgb_set_multi(struct net_device *netdev)
1044 {
1045         struct ixgb_adapter *adapter = netdev_priv(netdev);
1046         struct ixgb_hw *hw = &adapter->hw;
1047         struct dev_mc_list *mc_ptr;
1048         u32 rctl;
1049         int i;
1050
1051         /* Check for Promiscuous and All Multicast modes */
1052
1053         rctl = IXGB_READ_REG(hw, RCTL);
1054
1055         if (netdev->flags & IFF_PROMISC) {
1056                 rctl |= (IXGB_RCTL_UPE | IXGB_RCTL_MPE);
1057                 rctl &= ~IXGB_RCTL_VFE;
1058         } else {
1059                 if (netdev->flags & IFF_ALLMULTI) {
1060                         rctl |= IXGB_RCTL_MPE;
1061                         rctl &= ~IXGB_RCTL_UPE;
1062                 } else {
1063                         rctl &= ~(IXGB_RCTL_UPE | IXGB_RCTL_MPE);
1064                 }
1065                 rctl |= IXGB_RCTL_VFE;
1066         }
1067
1068         if (netdev->mc_count > IXGB_MAX_NUM_MULTICAST_ADDRESSES) {
1069                 rctl |= IXGB_RCTL_MPE;
1070                 IXGB_WRITE_REG(hw, RCTL, rctl);
1071         } else {
1072                 u8 mta[IXGB_MAX_NUM_MULTICAST_ADDRESSES *
1073                             IXGB_ETH_LENGTH_OF_ADDRESS];
1074
1075                 IXGB_WRITE_REG(hw, RCTL, rctl);
1076
1077                 for (i = 0, mc_ptr = netdev->mc_list;
1078                      mc_ptr;
1079                      i++, mc_ptr = mc_ptr->next)
1080                         memcpy(&mta[i * IXGB_ETH_LENGTH_OF_ADDRESS],
1081                                mc_ptr->dmi_addr, IXGB_ETH_LENGTH_OF_ADDRESS);
1082
1083                 ixgb_mc_addr_list_update(hw, mta, netdev->mc_count, 0);
1084         }
1085 }
1086
1087 /**
1088  * ixgb_watchdog - Timer Call-back
1089  * @data: pointer to netdev cast into an unsigned long
1090  **/
1091
1092 static void
1093 ixgb_watchdog(unsigned long data)
1094 {
1095         struct ixgb_adapter *adapter = (struct ixgb_adapter *)data;
1096         struct net_device *netdev = adapter->netdev;
1097         struct ixgb_desc_ring *txdr = &adapter->tx_ring;
1098
1099         ixgb_check_for_link(&adapter->hw);
1100
1101         if (ixgb_check_for_bad_link(&adapter->hw)) {
1102                 /* force the reset path */
1103                 netif_stop_queue(netdev);
1104         }
1105
1106         if (adapter->hw.link_up) {
1107                 if (!netif_carrier_ok(netdev)) {
1108                         DPRINTK(LINK, INFO,
1109                                 "NIC Link is Up 10000 Mbps Full Duplex\n");
1110                         adapter->link_speed = 10000;
1111                         adapter->link_duplex = FULL_DUPLEX;
1112                         netif_carrier_on(netdev);
1113                         netif_wake_queue(netdev);
1114                 }
1115         } else {
1116                 if (netif_carrier_ok(netdev)) {
1117                         adapter->link_speed = 0;
1118                         adapter->link_duplex = 0;
1119                         DPRINTK(LINK, INFO, "NIC Link is Down\n");
1120                         netif_carrier_off(netdev);
1121                         netif_stop_queue(netdev);
1122
1123                 }
1124         }
1125
1126         ixgb_update_stats(adapter);
1127
1128         if (!netif_carrier_ok(netdev)) {
1129                 if (IXGB_DESC_UNUSED(txdr) + 1 < txdr->count) {
1130                         /* We've lost link, so the controller stops DMA,
1131                          * but we've got queued Tx work that's never going
1132                          * to get done, so reset controller to flush Tx.
1133                          * (Do the reset outside of interrupt context). */
1134                         schedule_work(&adapter->tx_timeout_task);
1135                 }
1136         }
1137
1138         /* Force detection of hung controller every watchdog period */
1139         adapter->detect_tx_hung = true;
1140
1141         /* generate an interrupt to force clean up of any stragglers */
1142         IXGB_WRITE_REG(&adapter->hw, ICS, IXGB_INT_TXDW);
1143
1144         /* Reset the timer */
1145         mod_timer(&adapter->watchdog_timer, jiffies + 2 * HZ);
1146 }
1147
1148 #define IXGB_TX_FLAGS_CSUM              0x00000001
1149 #define IXGB_TX_FLAGS_VLAN              0x00000002
1150 #define IXGB_TX_FLAGS_TSO               0x00000004
1151
1152 static int
1153 ixgb_tso(struct ixgb_adapter *adapter, struct sk_buff *skb)
1154 {
1155         struct ixgb_context_desc *context_desc;
1156         unsigned int i;
1157         u8 ipcss, ipcso, tucss, tucso, hdr_len;
1158         u16 ipcse, tucse, mss;
1159         int err;
1160
1161         if (likely(skb_is_gso(skb))) {
1162                 struct ixgb_buffer *buffer_info;
1163                 struct iphdr *iph;
1164
1165                 if (skb_header_cloned(skb)) {
1166                         err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1167                         if (err)
1168                                 return err;
1169                 }
1170
1171                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
1172                 mss = skb_shinfo(skb)->gso_size;
1173                 iph = ip_hdr(skb);
1174                 iph->tot_len = 0;
1175                 iph->check = 0;
1176                 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
1177                                                          iph->daddr, 0,
1178                                                          IPPROTO_TCP, 0);
1179                 ipcss = skb_network_offset(skb);
1180                 ipcso = (void *)&(iph->check) - (void *)skb->data;
1181                 ipcse = skb_transport_offset(skb) - 1;
1182                 tucss = skb_transport_offset(skb);
1183                 tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
1184                 tucse = 0;
1185
1186                 i = adapter->tx_ring.next_to_use;
1187                 context_desc = IXGB_CONTEXT_DESC(adapter->tx_ring, i);
1188                 buffer_info = &adapter->tx_ring.buffer_info[i];
1189                 WARN_ON(buffer_info->dma != 0);
1190
1191                 context_desc->ipcss = ipcss;
1192                 context_desc->ipcso = ipcso;
1193                 context_desc->ipcse = cpu_to_le16(ipcse);
1194                 context_desc->tucss = tucss;
1195                 context_desc->tucso = tucso;
1196                 context_desc->tucse = cpu_to_le16(tucse);
1197                 context_desc->mss = cpu_to_le16(mss);
1198                 context_desc->hdr_len = hdr_len;
1199                 context_desc->status = 0;
1200                 context_desc->cmd_type_len = cpu_to_le32(
1201                                                   IXGB_CONTEXT_DESC_TYPE
1202                                                 | IXGB_CONTEXT_DESC_CMD_TSE
1203                                                 | IXGB_CONTEXT_DESC_CMD_IP
1204                                                 | IXGB_CONTEXT_DESC_CMD_TCP
1205                                                 | IXGB_CONTEXT_DESC_CMD_IDE
1206                                                 | (skb->len - (hdr_len)));
1207
1208
1209                 if (++i == adapter->tx_ring.count) i = 0;
1210                 adapter->tx_ring.next_to_use = i;
1211
1212                 return 1;
1213         }
1214
1215         return 0;
1216 }
1217
1218 static bool
1219 ixgb_tx_csum(struct ixgb_adapter *adapter, struct sk_buff *skb)
1220 {
1221         struct ixgb_context_desc *context_desc;
1222         unsigned int i;
1223         u8 css, cso;
1224
1225         if (likely(skb->ip_summed == CHECKSUM_PARTIAL)) {
1226                 struct ixgb_buffer *buffer_info;
1227                 css = skb_transport_offset(skb);
1228                 cso = css + skb->csum_offset;
1229
1230                 i = adapter->tx_ring.next_to_use;
1231                 context_desc = IXGB_CONTEXT_DESC(adapter->tx_ring, i);
1232                 buffer_info = &adapter->tx_ring.buffer_info[i];
1233                 WARN_ON(buffer_info->dma != 0);
1234
1235                 context_desc->tucss = css;
1236                 context_desc->tucso = cso;
1237                 context_desc->tucse = 0;
1238                 /* zero out any previously existing data in one instruction */
1239                 *(u32 *)&(context_desc->ipcss) = 0;
1240                 context_desc->status = 0;
1241                 context_desc->hdr_len = 0;
1242                 context_desc->mss = 0;
1243                 context_desc->cmd_type_len =
1244                         cpu_to_le32(IXGB_CONTEXT_DESC_TYPE
1245                                     | IXGB_TX_DESC_CMD_IDE);
1246
1247                 if (++i == adapter->tx_ring.count) i = 0;
1248                 adapter->tx_ring.next_to_use = i;
1249
1250                 return true;
1251         }
1252
1253         return false;
1254 }
1255
1256 #define IXGB_MAX_TXD_PWR        14
1257 #define IXGB_MAX_DATA_PER_TXD   (1<<IXGB_MAX_TXD_PWR)
1258
1259 static int
1260 ixgb_tx_map(struct ixgb_adapter *adapter, struct sk_buff *skb,
1261             unsigned int first)
1262 {
1263         struct ixgb_desc_ring *tx_ring = &adapter->tx_ring;
1264         struct ixgb_buffer *buffer_info;
1265         int len = skb->len;
1266         unsigned int offset = 0, size, count = 0, i;
1267         unsigned int mss = skb_shinfo(skb)->gso_size;
1268
1269         unsigned int nr_frags = skb_shinfo(skb)->nr_frags;
1270         unsigned int f;
1271
1272         len -= skb->data_len;
1273
1274         i = tx_ring->next_to_use;
1275
1276         while (len) {
1277                 buffer_info = &tx_ring->buffer_info[i];
1278                 size = min(len, IXGB_MAX_DATA_PER_TXD);
1279                 /* Workaround for premature desc write-backs
1280                  * in TSO mode.  Append 4-byte sentinel desc */
1281                 if (unlikely(mss && !nr_frags && size == len && size > 8))
1282                         size -= 4;
1283
1284                 buffer_info->length = size;
1285                 WARN_ON(buffer_info->dma != 0);
1286                 buffer_info->time_stamp = jiffies;
1287                 buffer_info->dma =
1288                         pci_map_single(adapter->pdev,
1289                                 skb->data + offset,
1290                                 size,
1291                                 PCI_DMA_TODEVICE);
1292                 buffer_info->next_to_watch = 0;
1293
1294                 len -= size;
1295                 offset += size;
1296                 count++;
1297                 if (++i == tx_ring->count) i = 0;
1298         }
1299
1300         for (f = 0; f < nr_frags; f++) {
1301                 struct skb_frag_struct *frag;
1302
1303                 frag = &skb_shinfo(skb)->frags[f];
1304                 len = frag->size;
1305                 offset = 0;
1306
1307                 while (len) {
1308                         buffer_info = &tx_ring->buffer_info[i];
1309                         size = min(len, IXGB_MAX_DATA_PER_TXD);
1310
1311                         /* Workaround for premature desc write-backs
1312                          * in TSO mode.  Append 4-byte sentinel desc */
1313                         if (unlikely(mss && (f == (nr_frags - 1))
1314                                      && size == len && size > 8))
1315                                 size -= 4;
1316
1317                         buffer_info->length = size;
1318                         buffer_info->time_stamp = jiffies;
1319                         buffer_info->dma =
1320                                 pci_map_page(adapter->pdev,
1321                                         frag->page,
1322                                         frag->page_offset + offset,
1323                                         size,
1324                                         PCI_DMA_TODEVICE);
1325                         buffer_info->next_to_watch = 0;
1326
1327                         len -= size;
1328                         offset += size;
1329                         count++;
1330                         if (++i == tx_ring->count) i = 0;
1331                 }
1332         }
1333         i = (i == 0) ? tx_ring->count - 1 : i - 1;
1334         tx_ring->buffer_info[i].skb = skb;
1335         tx_ring->buffer_info[first].next_to_watch = i;
1336
1337         return count;
1338 }
1339
1340 static void
1341 ixgb_tx_queue(struct ixgb_adapter *adapter, int count, int vlan_id,int tx_flags)
1342 {
1343         struct ixgb_desc_ring *tx_ring = &adapter->tx_ring;
1344         struct ixgb_tx_desc *tx_desc = NULL;
1345         struct ixgb_buffer *buffer_info;
1346         u32 cmd_type_len = adapter->tx_cmd_type;
1347         u8 status = 0;
1348         u8 popts = 0;
1349         unsigned int i;
1350
1351         if (tx_flags & IXGB_TX_FLAGS_TSO) {
1352                 cmd_type_len |= IXGB_TX_DESC_CMD_TSE;
1353                 popts |= (IXGB_TX_DESC_POPTS_IXSM | IXGB_TX_DESC_POPTS_TXSM);
1354         }
1355
1356         if (tx_flags & IXGB_TX_FLAGS_CSUM)
1357                 popts |= IXGB_TX_DESC_POPTS_TXSM;
1358
1359         if (tx_flags & IXGB_TX_FLAGS_VLAN)
1360                 cmd_type_len |= IXGB_TX_DESC_CMD_VLE;
1361
1362         i = tx_ring->next_to_use;
1363
1364         while (count--) {
1365                 buffer_info = &tx_ring->buffer_info[i];
1366                 tx_desc = IXGB_TX_DESC(*tx_ring, i);
1367                 tx_desc->buff_addr = cpu_to_le64(buffer_info->dma);
1368                 tx_desc->cmd_type_len =
1369                         cpu_to_le32(cmd_type_len | buffer_info->length);
1370                 tx_desc->status = status;
1371                 tx_desc->popts = popts;
1372                 tx_desc->vlan = cpu_to_le16(vlan_id);
1373
1374                 if (++i == tx_ring->count) i = 0;
1375         }
1376
1377         tx_desc->cmd_type_len |=
1378                 cpu_to_le32(IXGB_TX_DESC_CMD_EOP | IXGB_TX_DESC_CMD_RS);
1379
1380         /* Force memory writes to complete before letting h/w
1381          * know there are new descriptors to fetch.  (Only
1382          * applicable for weak-ordered memory model archs,
1383          * such as IA-64). */
1384         wmb();
1385
1386         tx_ring->next_to_use = i;
1387         IXGB_WRITE_REG(&adapter->hw, TDT, i);
1388 }
1389
1390 static int __ixgb_maybe_stop_tx(struct net_device *netdev, int size)
1391 {
1392         struct ixgb_adapter *adapter = netdev_priv(netdev);
1393         struct ixgb_desc_ring *tx_ring = &adapter->tx_ring;
1394
1395         netif_stop_queue(netdev);
1396         /* Herbert's original patch had:
1397          *  smp_mb__after_netif_stop_queue();
1398          * but since that doesn't exist yet, just open code it. */
1399         smp_mb();
1400
1401         /* We need to check again in a case another CPU has just
1402          * made room available. */
1403         if (likely(IXGB_DESC_UNUSED(tx_ring) < size))
1404                 return -EBUSY;
1405
1406         /* A reprieve! */
1407         netif_start_queue(netdev);
1408         ++adapter->restart_queue;
1409         return 0;
1410 }
1411
1412 static int ixgb_maybe_stop_tx(struct net_device *netdev,
1413                               struct ixgb_desc_ring *tx_ring, int size)
1414 {
1415         if (likely(IXGB_DESC_UNUSED(tx_ring) >= size))
1416                 return 0;
1417         return __ixgb_maybe_stop_tx(netdev, size);
1418 }
1419
1420
1421 /* Tx Descriptors needed, worst case */
1422 #define TXD_USE_COUNT(S) (((S) >> IXGB_MAX_TXD_PWR) + \
1423                          (((S) & (IXGB_MAX_DATA_PER_TXD - 1)) ? 1 : 0))
1424 #define DESC_NEEDED TXD_USE_COUNT(IXGB_MAX_DATA_PER_TXD) /* skb->date */ + \
1425         MAX_SKB_FRAGS * TXD_USE_COUNT(PAGE_SIZE) + 1 /* for context */ \
1426         + 1 /* one more needed for sentinel TSO workaround */
1427
1428 static int
1429 ixgb_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
1430 {
1431         struct ixgb_adapter *adapter = netdev_priv(netdev);
1432         unsigned int first;
1433         unsigned int tx_flags = 0;
1434         int vlan_id = 0;
1435         int tso;
1436
1437         if (test_bit(__IXGB_DOWN, &adapter->flags)) {
1438                 dev_kfree_skb(skb);
1439                 return NETDEV_TX_OK;
1440         }
1441
1442         if (skb->len <= 0) {
1443                 dev_kfree_skb(skb);
1444                 return 0;
1445         }
1446
1447         if (unlikely(ixgb_maybe_stop_tx(netdev, &adapter->tx_ring,
1448                      DESC_NEEDED)))
1449                 return NETDEV_TX_BUSY;
1450
1451         if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
1452                 tx_flags |= IXGB_TX_FLAGS_VLAN;
1453                 vlan_id = vlan_tx_tag_get(skb);
1454         }
1455
1456         first = adapter->tx_ring.next_to_use;
1457
1458         tso = ixgb_tso(adapter, skb);
1459         if (tso < 0) {
1460                 dev_kfree_skb(skb);
1461                 return NETDEV_TX_OK;
1462         }
1463
1464         if (likely(tso))
1465                 tx_flags |= IXGB_TX_FLAGS_TSO;
1466         else if (ixgb_tx_csum(adapter, skb))
1467                 tx_flags |= IXGB_TX_FLAGS_CSUM;
1468
1469         ixgb_tx_queue(adapter, ixgb_tx_map(adapter, skb, first), vlan_id,
1470                         tx_flags);
1471
1472         netdev->trans_start = jiffies;
1473
1474         /* Make sure there is space in the ring for the next send. */
1475         ixgb_maybe_stop_tx(netdev, &adapter->tx_ring, DESC_NEEDED);
1476
1477         return NETDEV_TX_OK;
1478 }
1479
1480 /**
1481  * ixgb_tx_timeout - Respond to a Tx Hang
1482  * @netdev: network interface device structure
1483  **/
1484
1485 static void
1486 ixgb_tx_timeout(struct net_device *netdev)
1487 {
1488         struct ixgb_adapter *adapter = netdev_priv(netdev);
1489
1490         /* Do the reset outside of interrupt context */
1491         schedule_work(&adapter->tx_timeout_task);
1492 }
1493
1494 static void
1495 ixgb_tx_timeout_task(struct work_struct *work)
1496 {
1497         struct ixgb_adapter *adapter =
1498                 container_of(work, struct ixgb_adapter, tx_timeout_task);
1499
1500         adapter->tx_timeout_count++;
1501         ixgb_down(adapter, true);
1502         ixgb_up(adapter);
1503 }
1504
1505 /**
1506  * ixgb_get_stats - Get System Network Statistics
1507  * @netdev: network interface device structure
1508  *
1509  * Returns the address of the device statistics structure.
1510  * The statistics are actually updated from the timer callback.
1511  **/
1512
1513 static struct net_device_stats *
1514 ixgb_get_stats(struct net_device *netdev)
1515 {
1516         struct ixgb_adapter *adapter = netdev_priv(netdev);
1517
1518         return &adapter->net_stats;
1519 }
1520
1521 /**
1522  * ixgb_change_mtu - Change the Maximum Transfer Unit
1523  * @netdev: network interface device structure
1524  * @new_mtu: new value for maximum frame size
1525  *
1526  * Returns 0 on success, negative on failure
1527  **/
1528
1529 static int
1530 ixgb_change_mtu(struct net_device *netdev, int new_mtu)
1531 {
1532         struct ixgb_adapter *adapter = netdev_priv(netdev);
1533         int max_frame = new_mtu + ENET_HEADER_SIZE + ENET_FCS_LENGTH;
1534         int old_max_frame = netdev->mtu + ENET_HEADER_SIZE + ENET_FCS_LENGTH;
1535
1536         /* MTU < 68 is an error for IPv4 traffic, just don't allow it */
1537         if ((new_mtu < 68) ||
1538             (max_frame > IXGB_MAX_JUMBO_FRAME_SIZE + ENET_FCS_LENGTH)) {
1539                 DPRINTK(PROBE, ERR, "Invalid MTU setting %d\n", new_mtu);
1540                 return -EINVAL;
1541         }
1542
1543         if (old_max_frame == max_frame)
1544                 return 0;
1545
1546         if (netif_running(netdev))
1547                 ixgb_down(adapter, true);
1548
1549         adapter->rx_buffer_len = max_frame + 8; /* + 8 for errata */
1550
1551         netdev->mtu = new_mtu;
1552
1553         if (netif_running(netdev))
1554                 ixgb_up(adapter);
1555
1556         return 0;
1557 }
1558
1559 /**
1560  * ixgb_update_stats - Update the board statistics counters.
1561  * @adapter: board private structure
1562  **/
1563
1564 void
1565 ixgb_update_stats(struct ixgb_adapter *adapter)
1566 {
1567         struct net_device *netdev = adapter->netdev;
1568         struct pci_dev *pdev = adapter->pdev;
1569
1570         /* Prevent stats update while adapter is being reset */
1571         if (pci_channel_offline(pdev))
1572                 return;
1573
1574         if ((netdev->flags & IFF_PROMISC) || (netdev->flags & IFF_ALLMULTI) ||
1575            (netdev->mc_count > IXGB_MAX_NUM_MULTICAST_ADDRESSES)) {
1576                 u64 multi = IXGB_READ_REG(&adapter->hw, MPRCL);
1577                 u32 bcast_l = IXGB_READ_REG(&adapter->hw, BPRCL);
1578                 u32 bcast_h = IXGB_READ_REG(&adapter->hw, BPRCH);
1579                 u64 bcast = ((u64)bcast_h << 32) | bcast_l;
1580
1581                 multi |= ((u64)IXGB_READ_REG(&adapter->hw, MPRCH) << 32);
1582                 /* fix up multicast stats by removing broadcasts */
1583                 if (multi >= bcast)
1584                         multi -= bcast;
1585
1586                 adapter->stats.mprcl += (multi & 0xFFFFFFFF);
1587                 adapter->stats.mprch += (multi >> 32);
1588                 adapter->stats.bprcl += bcast_l;
1589                 adapter->stats.bprch += bcast_h;
1590         } else {
1591                 adapter->stats.mprcl += IXGB_READ_REG(&adapter->hw, MPRCL);
1592                 adapter->stats.mprch += IXGB_READ_REG(&adapter->hw, MPRCH);
1593                 adapter->stats.bprcl += IXGB_READ_REG(&adapter->hw, BPRCL);
1594                 adapter->stats.bprch += IXGB_READ_REG(&adapter->hw, BPRCH);
1595         }
1596         adapter->stats.tprl += IXGB_READ_REG(&adapter->hw, TPRL);
1597         adapter->stats.tprh += IXGB_READ_REG(&adapter->hw, TPRH);
1598         adapter->stats.gprcl += IXGB_READ_REG(&adapter->hw, GPRCL);
1599         adapter->stats.gprch += IXGB_READ_REG(&adapter->hw, GPRCH);
1600         adapter->stats.uprcl += IXGB_READ_REG(&adapter->hw, UPRCL);
1601         adapter->stats.uprch += IXGB_READ_REG(&adapter->hw, UPRCH);
1602         adapter->stats.vprcl += IXGB_READ_REG(&adapter->hw, VPRCL);
1603         adapter->stats.vprch += IXGB_READ_REG(&adapter->hw, VPRCH);
1604         adapter->stats.jprcl += IXGB_READ_REG(&adapter->hw, JPRCL);
1605         adapter->stats.jprch += IXGB_READ_REG(&adapter->hw, JPRCH);
1606         adapter->stats.gorcl += IXGB_READ_REG(&adapter->hw, GORCL);
1607         adapter->stats.gorch += IXGB_READ_REG(&adapter->hw, GORCH);
1608         adapter->stats.torl += IXGB_READ_REG(&adapter->hw, TORL);
1609         adapter->stats.torh += IXGB_READ_REG(&adapter->hw, TORH);
1610         adapter->stats.rnbc += IXGB_READ_REG(&adapter->hw, RNBC);
1611         adapter->stats.ruc += IXGB_READ_REG(&adapter->hw, RUC);
1612         adapter->stats.roc += IXGB_READ_REG(&adapter->hw, ROC);
1613         adapter->stats.rlec += IXGB_READ_REG(&adapter->hw, RLEC);
1614         adapter->stats.crcerrs += IXGB_READ_REG(&adapter->hw, CRCERRS);
1615         adapter->stats.icbc += IXGB_READ_REG(&adapter->hw, ICBC);
1616         adapter->stats.ecbc += IXGB_READ_REG(&adapter->hw, ECBC);
1617         adapter->stats.mpc += IXGB_READ_REG(&adapter->hw, MPC);
1618         adapter->stats.tptl += IXGB_READ_REG(&adapter->hw, TPTL);
1619         adapter->stats.tpth += IXGB_READ_REG(&adapter->hw, TPTH);
1620         adapter->stats.gptcl += IXGB_READ_REG(&adapter->hw, GPTCL);
1621         adapter->stats.gptch += IXGB_READ_REG(&adapter->hw, GPTCH);
1622         adapter->stats.bptcl += IXGB_READ_REG(&adapter->hw, BPTCL);
1623         adapter->stats.bptch += IXGB_READ_REG(&adapter->hw, BPTCH);
1624         adapter->stats.mptcl += IXGB_READ_REG(&adapter->hw, MPTCL);
1625         adapter->stats.mptch += IXGB_READ_REG(&adapter->hw, MPTCH);
1626         adapter->stats.uptcl += IXGB_READ_REG(&adapter->hw, UPTCL);
1627         adapter->stats.uptch += IXGB_READ_REG(&adapter->hw, UPTCH);
1628         adapter->stats.vptcl += IXGB_READ_REG(&adapter->hw, VPTCL);
1629         adapter->stats.vptch += IXGB_READ_REG(&adapter->hw, VPTCH);
1630         adapter->stats.jptcl += IXGB_READ_REG(&adapter->hw, JPTCL);
1631         adapter->stats.jptch += IXGB_READ_REG(&adapter->hw, JPTCH);
1632         adapter->stats.gotcl += IXGB_READ_REG(&adapter->hw, GOTCL);
1633         adapter->stats.gotch += IXGB_READ_REG(&adapter->hw, GOTCH);
1634         adapter->stats.totl += IXGB_READ_REG(&adapter->hw, TOTL);
1635         adapter->stats.toth += IXGB_READ_REG(&adapter->hw, TOTH);
1636         adapter->stats.dc += IXGB_READ_REG(&adapter->hw, DC);
1637         adapter->stats.plt64c += IXGB_READ_REG(&adapter->hw, PLT64C);
1638         adapter->stats.tsctc += IXGB_READ_REG(&adapter->hw, TSCTC);
1639         adapter->stats.tsctfc += IXGB_READ_REG(&adapter->hw, TSCTFC);
1640         adapter->stats.ibic += IXGB_READ_REG(&adapter->hw, IBIC);
1641         adapter->stats.rfc += IXGB_READ_REG(&adapter->hw, RFC);
1642         adapter->stats.lfc += IXGB_READ_REG(&adapter->hw, LFC);
1643         adapter->stats.pfrc += IXGB_READ_REG(&adapter->hw, PFRC);
1644         adapter->stats.pftc += IXGB_READ_REG(&adapter->hw, PFTC);
1645         adapter->stats.mcfrc += IXGB_READ_REG(&adapter->hw, MCFRC);
1646         adapter->stats.mcftc += IXGB_READ_REG(&adapter->hw, MCFTC);
1647         adapter->stats.xonrxc += IXGB_READ_REG(&adapter->hw, XONRXC);
1648         adapter->stats.xontxc += IXGB_READ_REG(&adapter->hw, XONTXC);
1649         adapter->stats.xoffrxc += IXGB_READ_REG(&adapter->hw, XOFFRXC);
1650         adapter->stats.xofftxc += IXGB_READ_REG(&adapter->hw, XOFFTXC);
1651         adapter->stats.rjc += IXGB_READ_REG(&adapter->hw, RJC);
1652
1653         /* Fill out the OS statistics structure */
1654
1655         adapter->net_stats.rx_packets = adapter->stats.gprcl;
1656         adapter->net_stats.tx_packets = adapter->stats.gptcl;
1657         adapter->net_stats.rx_bytes = adapter->stats.gorcl;
1658         adapter->net_stats.tx_bytes = adapter->stats.gotcl;
1659         adapter->net_stats.multicast = adapter->stats.mprcl;
1660         adapter->net_stats.collisions = 0;
1661
1662         /* ignore RLEC as it reports errors for padded (<64bytes) frames
1663          * with a length in the type/len field */
1664         adapter->net_stats.rx_errors =
1665             /* adapter->stats.rnbc + */ adapter->stats.crcerrs +
1666             adapter->stats.ruc +
1667             adapter->stats.roc /*+ adapter->stats.rlec */  +
1668             adapter->stats.icbc +
1669             adapter->stats.ecbc + adapter->stats.mpc;
1670
1671         /* see above
1672          * adapter->net_stats.rx_length_errors = adapter->stats.rlec;
1673          */
1674
1675         adapter->net_stats.rx_crc_errors = adapter->stats.crcerrs;
1676         adapter->net_stats.rx_fifo_errors = adapter->stats.mpc;
1677         adapter->net_stats.rx_missed_errors = adapter->stats.mpc;
1678         adapter->net_stats.rx_over_errors = adapter->stats.mpc;
1679
1680         adapter->net_stats.tx_errors = 0;
1681         adapter->net_stats.rx_frame_errors = 0;
1682         adapter->net_stats.tx_aborted_errors = 0;
1683         adapter->net_stats.tx_carrier_errors = 0;
1684         adapter->net_stats.tx_fifo_errors = 0;
1685         adapter->net_stats.tx_heartbeat_errors = 0;
1686         adapter->net_stats.tx_window_errors = 0;
1687 }
1688
1689 #define IXGB_MAX_INTR 10
1690 /**
1691  * ixgb_intr - Interrupt Handler
1692  * @irq: interrupt number
1693  * @data: pointer to a network interface device structure
1694  **/
1695
1696 static irqreturn_t
1697 ixgb_intr(int irq, void *data)
1698 {
1699         struct net_device *netdev = data;
1700         struct ixgb_adapter *adapter = netdev_priv(netdev);
1701         struct ixgb_hw *hw = &adapter->hw;
1702         u32 icr = IXGB_READ_REG(hw, ICR);
1703
1704         if (unlikely(!icr))
1705                 return IRQ_NONE;  /* Not our interrupt */
1706
1707         if (unlikely(icr & (IXGB_INT_RXSEQ | IXGB_INT_LSC)))
1708                 if (!test_bit(__IXGB_DOWN, &adapter->flags))
1709                         mod_timer(&adapter->watchdog_timer, jiffies);
1710
1711         if (netif_rx_schedule_prep(netdev, &adapter->napi)) {
1712
1713                 /* Disable interrupts and register for poll. The flush
1714                   of the posted write is intentionally left out.
1715                 */
1716
1717                 IXGB_WRITE_REG(&adapter->hw, IMC, ~0);
1718                 __netif_rx_schedule(netdev, &adapter->napi);
1719         }
1720         return IRQ_HANDLED;
1721 }
1722
1723 /**
1724  * ixgb_clean - NAPI Rx polling callback
1725  * @adapter: board private structure
1726  **/
1727
1728 static int
1729 ixgb_clean(struct napi_struct *napi, int budget)
1730 {
1731         struct ixgb_adapter *adapter = container_of(napi, struct ixgb_adapter, napi);
1732         struct net_device *netdev = adapter->netdev;
1733         int work_done = 0;
1734
1735         ixgb_clean_tx_irq(adapter);
1736         ixgb_clean_rx_irq(adapter, &work_done, budget);
1737
1738         /* If budget not fully consumed, exit the polling mode */
1739         if (work_done < budget) {
1740                 netif_rx_complete(netdev, napi);
1741                 if (!test_bit(__IXGB_DOWN, &adapter->flags))
1742                         ixgb_irq_enable(adapter);
1743         }
1744
1745         return work_done;
1746 }
1747
1748 /**
1749  * ixgb_clean_tx_irq - Reclaim resources after transmit completes
1750  * @adapter: board private structure
1751  **/
1752
1753 static bool
1754 ixgb_clean_tx_irq(struct ixgb_adapter *adapter)
1755 {
1756         struct ixgb_desc_ring *tx_ring = &adapter->tx_ring;
1757         struct net_device *netdev = adapter->netdev;
1758         struct ixgb_tx_desc *tx_desc, *eop_desc;
1759         struct ixgb_buffer *buffer_info;
1760         unsigned int i, eop;
1761         bool cleaned = false;
1762
1763         i = tx_ring->next_to_clean;
1764         eop = tx_ring->buffer_info[i].next_to_watch;
1765         eop_desc = IXGB_TX_DESC(*tx_ring, eop);
1766
1767         while (eop_desc->status & IXGB_TX_DESC_STATUS_DD) {
1768
1769                 for (cleaned = false; !cleaned; ) {
1770                         tx_desc = IXGB_TX_DESC(*tx_ring, i);
1771                         buffer_info = &tx_ring->buffer_info[i];
1772
1773                         if (tx_desc->popts &
1774                            (IXGB_TX_DESC_POPTS_TXSM |
1775                             IXGB_TX_DESC_POPTS_IXSM))
1776                                 adapter->hw_csum_tx_good++;
1777
1778                         ixgb_unmap_and_free_tx_resource(adapter, buffer_info);
1779
1780                         *(u32 *)&(tx_desc->status) = 0;
1781
1782                         cleaned = (i == eop);
1783                         if (++i == tx_ring->count) i = 0;
1784                 }
1785
1786                 eop = tx_ring->buffer_info[i].next_to_watch;
1787                 eop_desc = IXGB_TX_DESC(*tx_ring, eop);
1788         }
1789
1790         tx_ring->next_to_clean = i;
1791
1792         if (unlikely(cleaned && netif_carrier_ok(netdev) &&
1793                      IXGB_DESC_UNUSED(tx_ring) >= DESC_NEEDED)) {
1794                 /* Make sure that anybody stopping the queue after this
1795                  * sees the new next_to_clean. */
1796                 smp_mb();
1797
1798                 if (netif_queue_stopped(netdev) &&
1799                     !(test_bit(__IXGB_DOWN, &adapter->flags))) {
1800                         netif_wake_queue(netdev);
1801                         ++adapter->restart_queue;
1802                 }
1803         }
1804
1805         if (adapter->detect_tx_hung) {
1806                 /* detect a transmit hang in hardware, this serializes the
1807                  * check with the clearing of time_stamp and movement of i */
1808                 adapter->detect_tx_hung = false;
1809                 if (tx_ring->buffer_info[eop].dma &&
1810                    time_after(jiffies, tx_ring->buffer_info[eop].time_stamp + HZ)
1811                    && !(IXGB_READ_REG(&adapter->hw, STATUS) &
1812                         IXGB_STATUS_TXOFF)) {
1813                         /* detected Tx unit hang */
1814                         DPRINTK(DRV, ERR, "Detected Tx Unit Hang\n"
1815                                         "  TDH                  <%x>\n"
1816                                         "  TDT                  <%x>\n"
1817                                         "  next_to_use          <%x>\n"
1818                                         "  next_to_clean        <%x>\n"
1819                                         "buffer_info[next_to_clean]\n"
1820                                         "  time_stamp           <%lx>\n"
1821                                         "  next_to_watch        <%x>\n"
1822                                         "  jiffies              <%lx>\n"
1823                                         "  next_to_watch.status <%x>\n",
1824                                 IXGB_READ_REG(&adapter->hw, TDH),
1825                                 IXGB_READ_REG(&adapter->hw, TDT),
1826                                 tx_ring->next_to_use,
1827                                 tx_ring->next_to_clean,
1828                                 tx_ring->buffer_info[eop].time_stamp,
1829                                 eop,
1830                                 jiffies,
1831                                 eop_desc->status);
1832                         netif_stop_queue(netdev);
1833                 }
1834         }
1835
1836         return cleaned;
1837 }
1838
1839 /**
1840  * ixgb_rx_checksum - Receive Checksum Offload for 82597.
1841  * @adapter: board private structure
1842  * @rx_desc: receive descriptor
1843  * @sk_buff: socket buffer with received data
1844  **/
1845
1846 static void
1847 ixgb_rx_checksum(struct ixgb_adapter *adapter,
1848                  struct ixgb_rx_desc *rx_desc,
1849                  struct sk_buff *skb)
1850 {
1851         /* Ignore Checksum bit is set OR
1852          * TCP Checksum has not been calculated
1853          */
1854         if ((rx_desc->status & IXGB_RX_DESC_STATUS_IXSM) ||
1855            (!(rx_desc->status & IXGB_RX_DESC_STATUS_TCPCS))) {
1856                 skb->ip_summed = CHECKSUM_NONE;
1857                 return;
1858         }
1859
1860         /* At this point we know the hardware did the TCP checksum */
1861         /* now look at the TCP checksum error bit */
1862         if (rx_desc->errors & IXGB_RX_DESC_ERRORS_TCPE) {
1863                 /* let the stack verify checksum errors */
1864                 skb->ip_summed = CHECKSUM_NONE;
1865                 adapter->hw_csum_rx_error++;
1866         } else {
1867                 /* TCP checksum is good */
1868                 skb->ip_summed = CHECKSUM_UNNECESSARY;
1869                 adapter->hw_csum_rx_good++;
1870         }
1871 }
1872
1873 /**
1874  * ixgb_clean_rx_irq - Send received data up the network stack,
1875  * @adapter: board private structure
1876  **/
1877
1878 static bool
1879 ixgb_clean_rx_irq(struct ixgb_adapter *adapter, int *work_done, int work_to_do)
1880 {
1881         struct ixgb_desc_ring *rx_ring = &adapter->rx_ring;
1882         struct net_device *netdev = adapter->netdev;
1883         struct pci_dev *pdev = adapter->pdev;
1884         struct ixgb_rx_desc *rx_desc, *next_rxd;
1885         struct ixgb_buffer *buffer_info, *next_buffer, *next2_buffer;
1886         u32 length;
1887         unsigned int i, j;
1888         int cleaned_count = 0;
1889         bool cleaned = false;
1890
1891         i = rx_ring->next_to_clean;
1892         rx_desc = IXGB_RX_DESC(*rx_ring, i);
1893         buffer_info = &rx_ring->buffer_info[i];
1894
1895         while (rx_desc->status & IXGB_RX_DESC_STATUS_DD) {
1896                 struct sk_buff *skb;
1897                 u8 status;
1898
1899                 if (*work_done >= work_to_do)
1900                         break;
1901
1902                 (*work_done)++;
1903                 status = rx_desc->status;
1904                 skb = buffer_info->skb;
1905                 buffer_info->skb = NULL;
1906
1907                 prefetch(skb->data - NET_IP_ALIGN);
1908
1909                 if (++i == rx_ring->count) i = 0;
1910                 next_rxd = IXGB_RX_DESC(*rx_ring, i);
1911                 prefetch(next_rxd);
1912
1913                 if ((j = i + 1) == rx_ring->count) j = 0;
1914                 next2_buffer = &rx_ring->buffer_info[j];
1915                 prefetch(next2_buffer);
1916
1917                 next_buffer = &rx_ring->buffer_info[i];
1918
1919                 cleaned = true;
1920                 cleaned_count++;
1921
1922                 pci_unmap_single(pdev,
1923                                  buffer_info->dma,
1924                                  buffer_info->length,
1925                                  PCI_DMA_FROMDEVICE);
1926                 buffer_info->dma = 0;
1927
1928                 length = le16_to_cpu(rx_desc->length);
1929                 rx_desc->length = 0;
1930
1931                 if (unlikely(!(status & IXGB_RX_DESC_STATUS_EOP))) {
1932
1933                         /* All receives must fit into a single buffer */
1934
1935                         IXGB_DBG("Receive packet consumed multiple buffers "
1936                                          "length<%x>\n", length);
1937
1938                         dev_kfree_skb_irq(skb);
1939                         goto rxdesc_done;
1940                 }
1941
1942                 if (unlikely(rx_desc->errors &
1943                     (IXGB_RX_DESC_ERRORS_CE | IXGB_RX_DESC_ERRORS_SE |
1944                      IXGB_RX_DESC_ERRORS_P | IXGB_RX_DESC_ERRORS_RXE))) {
1945                         dev_kfree_skb_irq(skb);
1946                         goto rxdesc_done;
1947                 }
1948
1949                 /* code added for copybreak, this should improve
1950                  * performance for small packets with large amounts
1951                  * of reassembly being done in the stack */
1952                 if (length < copybreak) {
1953                         struct sk_buff *new_skb =
1954                             netdev_alloc_skb(netdev, length + NET_IP_ALIGN);
1955                         if (new_skb) {
1956                                 skb_reserve(new_skb, NET_IP_ALIGN);
1957                                 skb_copy_to_linear_data_offset(new_skb,
1958                                                                -NET_IP_ALIGN,
1959                                                                (skb->data -
1960                                                                 NET_IP_ALIGN),
1961                                                                (length +
1962                                                                 NET_IP_ALIGN));
1963                                 /* save the skb in buffer_info as good */
1964                                 buffer_info->skb = skb;
1965                                 skb = new_skb;
1966                         }
1967                 }
1968                 /* end copybreak code */
1969
1970                 /* Good Receive */
1971                 skb_put(skb, length);
1972
1973                 /* Receive Checksum Offload */
1974                 ixgb_rx_checksum(adapter, rx_desc, skb);
1975
1976                 skb->protocol = eth_type_trans(skb, netdev);
1977                 if (adapter->vlgrp && (status & IXGB_RX_DESC_STATUS_VP)) {
1978                         vlan_hwaccel_receive_skb(skb, adapter->vlgrp,
1979                                                 le16_to_cpu(rx_desc->special));
1980                 } else {
1981                         netif_receive_skb(skb);
1982                 }
1983                 netdev->last_rx = jiffies;
1984
1985 rxdesc_done:
1986                 /* clean up descriptor, might be written over by hw */
1987                 rx_desc->status = 0;
1988
1989                 /* return some buffers to hardware, one at a time is too slow */
1990                 if (unlikely(cleaned_count >= IXGB_RX_BUFFER_WRITE)) {
1991                         ixgb_alloc_rx_buffers(adapter, cleaned_count);
1992                         cleaned_count = 0;
1993                 }
1994
1995                 /* use prefetched values */
1996                 rx_desc = next_rxd;
1997                 buffer_info = next_buffer;
1998         }
1999
2000         rx_ring->next_to_clean = i;
2001
2002         cleaned_count = IXGB_DESC_UNUSED(rx_ring);
2003         if (cleaned_count)
2004                 ixgb_alloc_rx_buffers(adapter, cleaned_count);
2005
2006         return cleaned;
2007 }
2008
2009 /**
2010  * ixgb_alloc_rx_buffers - Replace used receive buffers
2011  * @adapter: address of board private structure
2012  **/
2013
2014 static void
2015 ixgb_alloc_rx_buffers(struct ixgb_adapter *adapter, int cleaned_count)
2016 {
2017         struct ixgb_desc_ring *rx_ring = &adapter->rx_ring;
2018         struct net_device *netdev = adapter->netdev;
2019         struct pci_dev *pdev = adapter->pdev;
2020         struct ixgb_rx_desc *rx_desc;
2021         struct ixgb_buffer *buffer_info;
2022         struct sk_buff *skb;
2023         unsigned int i;
2024         long cleancount;
2025
2026         i = rx_ring->next_to_use;
2027         buffer_info = &rx_ring->buffer_info[i];
2028         cleancount = IXGB_DESC_UNUSED(rx_ring);
2029
2030
2031         /* leave three descriptors unused */
2032         while (--cleancount > 2 && cleaned_count--) {
2033                 /* recycle! its good for you */
2034                 skb = buffer_info->skb;
2035                 if (skb) {
2036                         skb_trim(skb, 0);
2037                         goto map_skb;
2038                 }
2039
2040                 skb = netdev_alloc_skb(netdev, adapter->rx_buffer_len
2041                                        + NET_IP_ALIGN);
2042                 if (unlikely(!skb)) {
2043                         /* Better luck next round */
2044                         adapter->alloc_rx_buff_failed++;
2045                         break;
2046                 }
2047
2048                 /* Make buffer alignment 2 beyond a 16 byte boundary
2049                  * this will result in a 16 byte aligned IP header after
2050                  * the 14 byte MAC header is removed
2051                  */
2052                 skb_reserve(skb, NET_IP_ALIGN);
2053
2054                 buffer_info->skb = skb;
2055                 buffer_info->length = adapter->rx_buffer_len;
2056 map_skb:
2057                 buffer_info->dma = pci_map_single(pdev,
2058                                                   skb->data,
2059                                                   adapter->rx_buffer_len,
2060                                                   PCI_DMA_FROMDEVICE);
2061
2062                 rx_desc = IXGB_RX_DESC(*rx_ring, i);
2063                 rx_desc->buff_addr = cpu_to_le64(buffer_info->dma);
2064                 /* guarantee DD bit not set now before h/w gets descriptor
2065                  * this is the rest of the workaround for h/w double
2066                  * writeback. */
2067                 rx_desc->status = 0;
2068
2069
2070                 if (++i == rx_ring->count) i = 0;
2071                 buffer_info = &rx_ring->buffer_info[i];
2072         }
2073
2074         if (likely(rx_ring->next_to_use != i)) {
2075                 rx_ring->next_to_use = i;
2076                 if (unlikely(i-- == 0))
2077                         i = (rx_ring->count - 1);
2078
2079                 /* Force memory writes to complete before letting h/w
2080                  * know there are new descriptors to fetch.  (Only
2081                  * applicable for weak-ordered memory model archs, such
2082                  * as IA-64). */
2083                 wmb();
2084                 IXGB_WRITE_REG(&adapter->hw, RDT, i);
2085         }
2086 }
2087
2088 /**
2089  * ixgb_vlan_rx_register - enables or disables vlan tagging/stripping.
2090  *
2091  * @param netdev network interface device structure
2092  * @param grp indicates to enable or disable tagging/stripping
2093  **/
2094 static void
2095 ixgb_vlan_rx_register(struct net_device *netdev, struct vlan_group *grp)
2096 {
2097         struct ixgb_adapter *adapter = netdev_priv(netdev);
2098         u32 ctrl, rctl;
2099
2100         ixgb_irq_disable(adapter);
2101         adapter->vlgrp = grp;
2102
2103         if (grp) {
2104                 /* enable VLAN tag insert/strip */
2105                 ctrl = IXGB_READ_REG(&adapter->hw, CTRL0);
2106                 ctrl |= IXGB_CTRL0_VME;
2107                 IXGB_WRITE_REG(&adapter->hw, CTRL0, ctrl);
2108
2109                 /* enable VLAN receive filtering */
2110
2111                 rctl = IXGB_READ_REG(&adapter->hw, RCTL);
2112                 rctl &= ~IXGB_RCTL_CFIEN;
2113                 IXGB_WRITE_REG(&adapter->hw, RCTL, rctl);
2114         } else {
2115                 /* disable VLAN tag insert/strip */
2116
2117                 ctrl = IXGB_READ_REG(&adapter->hw, CTRL0);
2118                 ctrl &= ~IXGB_CTRL0_VME;
2119                 IXGB_WRITE_REG(&adapter->hw, CTRL0, ctrl);
2120         }
2121
2122         /* don't enable interrupts unless we are UP */
2123         if (adapter->netdev->flags & IFF_UP)
2124                 ixgb_irq_enable(adapter);
2125 }
2126
2127 static void
2128 ixgb_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
2129 {
2130         struct ixgb_adapter *adapter = netdev_priv(netdev);
2131         u32 vfta, index;
2132
2133         /* add VID to filter table */
2134
2135         index = (vid >> 5) & 0x7F;
2136         vfta = IXGB_READ_REG_ARRAY(&adapter->hw, VFTA, index);
2137         vfta |= (1 << (vid & 0x1F));
2138         ixgb_write_vfta(&adapter->hw, index, vfta);
2139 }
2140
2141 static void
2142 ixgb_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
2143 {
2144         struct ixgb_adapter *adapter = netdev_priv(netdev);
2145         u32 vfta, index;
2146
2147         ixgb_irq_disable(adapter);
2148
2149         vlan_group_set_device(adapter->vlgrp, vid, NULL);
2150
2151         /* don't enable interrupts unless we are UP */
2152         if (adapter->netdev->flags & IFF_UP)
2153                 ixgb_irq_enable(adapter);
2154
2155         /* remove VID from filter table */
2156
2157         index = (vid >> 5) & 0x7F;
2158         vfta = IXGB_READ_REG_ARRAY(&adapter->hw, VFTA, index);
2159         vfta &= ~(1 << (vid & 0x1F));
2160         ixgb_write_vfta(&adapter->hw, index, vfta);
2161 }
2162
2163 static void
2164 ixgb_restore_vlan(struct ixgb_adapter *adapter)
2165 {
2166         ixgb_vlan_rx_register(adapter->netdev, adapter->vlgrp);
2167
2168         if (adapter->vlgrp) {
2169                 u16 vid;
2170                 for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
2171                         if (!vlan_group_get_device(adapter->vlgrp, vid))
2172                                 continue;
2173                         ixgb_vlan_rx_add_vid(adapter->netdev, vid);
2174                 }
2175         }
2176 }
2177
2178 #ifdef CONFIG_NET_POLL_CONTROLLER
2179 /*
2180  * Polling 'interrupt' - used by things like netconsole to send skbs
2181  * without having to re-enable interrupts. It's not called while
2182  * the interrupt routine is executing.
2183  */
2184
2185 static void ixgb_netpoll(struct net_device *dev)
2186 {
2187         struct ixgb_adapter *adapter = netdev_priv(dev);
2188
2189         disable_irq(adapter->pdev->irq);
2190         ixgb_intr(adapter->pdev->irq, dev);
2191         enable_irq(adapter->pdev->irq);
2192 }
2193 #endif
2194
2195 /**
2196  * ixgb_io_error_detected() - called when PCI error is detected
2197  * @pdev    pointer to pci device with error
2198  * @state   pci channel state after error
2199  *
2200  * This callback is called by the PCI subsystem whenever
2201  * a PCI bus error is detected.
2202  */
2203 static pci_ers_result_t ixgb_io_error_detected(struct pci_dev *pdev,
2204                                                enum pci_channel_state state)
2205 {
2206         struct net_device *netdev = pci_get_drvdata(pdev);
2207         struct ixgb_adapter *adapter = netdev_priv(netdev);
2208
2209         if (netif_running(netdev))
2210                 ixgb_down(adapter, true);
2211
2212         pci_disable_device(pdev);
2213
2214         /* Request a slot reset. */
2215         return PCI_ERS_RESULT_NEED_RESET;
2216 }
2217
2218 /**
2219  * ixgb_io_slot_reset - called after the pci bus has been reset.
2220  * @pdev    pointer to pci device with error
2221  *
2222  * This callback is called after the PCI bus has been reset.
2223  * Basically, this tries to restart the card from scratch.
2224  * This is a shortened version of the device probe/discovery code,
2225  * it resembles the first-half of the ixgb_probe() routine.
2226  */
2227 static pci_ers_result_t ixgb_io_slot_reset(struct pci_dev *pdev)
2228 {
2229         struct net_device *netdev = pci_get_drvdata(pdev);
2230         struct ixgb_adapter *adapter = netdev_priv(netdev);
2231
2232         if (pci_enable_device(pdev)) {
2233                 DPRINTK(PROBE, ERR, "Cannot re-enable PCI device after reset.\n");
2234                 return PCI_ERS_RESULT_DISCONNECT;
2235         }
2236
2237         /* Perform card reset only on one instance of the card */
2238         if (0 != PCI_FUNC (pdev->devfn))
2239                 return PCI_ERS_RESULT_RECOVERED;
2240
2241         pci_set_master(pdev);
2242
2243         netif_carrier_off(netdev);
2244         netif_stop_queue(netdev);
2245         ixgb_reset(adapter);
2246
2247         /* Make sure the EEPROM is good */
2248         if (!ixgb_validate_eeprom_checksum(&adapter->hw)) {
2249                 DPRINTK(PROBE, ERR, "After reset, the EEPROM checksum is not valid.\n");
2250                 return PCI_ERS_RESULT_DISCONNECT;
2251         }
2252         ixgb_get_ee_mac_addr(&adapter->hw, netdev->dev_addr);
2253         memcpy(netdev->perm_addr, netdev->dev_addr, netdev->addr_len);
2254
2255         if (!is_valid_ether_addr(netdev->perm_addr)) {
2256                 DPRINTK(PROBE, ERR, "After reset, invalid MAC address.\n");
2257                 return PCI_ERS_RESULT_DISCONNECT;
2258         }
2259
2260         return PCI_ERS_RESULT_RECOVERED;
2261 }
2262
2263 /**
2264  * ixgb_io_resume - called when its OK to resume normal operations
2265  * @pdev    pointer to pci device with error
2266  *
2267  * The error recovery driver tells us that its OK to resume
2268  * normal operation. Implementation resembles the second-half
2269  * of the ixgb_probe() routine.
2270  */
2271 static void ixgb_io_resume(struct pci_dev *pdev)
2272 {
2273         struct net_device *netdev = pci_get_drvdata(pdev);
2274         struct ixgb_adapter *adapter = netdev_priv(netdev);
2275
2276         pci_set_master(pdev);
2277
2278         if (netif_running(netdev)) {
2279                 if (ixgb_up(adapter)) {
2280                         printk ("ixgb: can't bring device back up after reset\n");
2281                         return;
2282                 }
2283         }
2284
2285         netif_device_attach(netdev);
2286         mod_timer(&adapter->watchdog_timer, jiffies);
2287 }
2288
2289 /* ixgb_main.c */