[NET]: Make NAPI polling independent of struct net_device objects.
[linux-2.6.git] / drivers / net / ixgb / ixgb_main.c
1 /*******************************************************************************
2
3   Intel PRO/10GbE Linux driver
4   Copyright(c) 1999 - 2006 Intel Corporation.
5
6   This program is free software; you can redistribute it and/or modify it
7   under the terms and conditions of the GNU General Public License,
8   version 2, as published by the Free Software Foundation.
9
10   This program is distributed in the hope it will be useful, but WITHOUT
11   ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12   FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
13   more details.
14
15   You should have received a copy of the GNU General Public License along with
16   this program; if not, write to the Free Software Foundation, Inc.,
17   51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
18
19   The full GNU General Public License is included in this distribution in
20   the file called "COPYING".
21
22   Contact Information:
23   Linux NICS <linux.nics@intel.com>
24   e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
25   Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
26
27 *******************************************************************************/
28
29 #include "ixgb.h"
30
31 char ixgb_driver_name[] = "ixgb";
32 static char ixgb_driver_string[] = "Intel(R) PRO/10GbE Network Driver";
33
34 #ifndef CONFIG_IXGB_NAPI
35 #define DRIVERNAPI
36 #else
37 #define DRIVERNAPI "-NAPI"
38 #endif
39 #define DRV_VERSION             "1.0.126-k2"DRIVERNAPI
40 char ixgb_driver_version[] = DRV_VERSION;
41 static char ixgb_copyright[] = "Copyright (c) 1999-2006 Intel Corporation.";
42
43 /* ixgb_pci_tbl - PCI Device ID Table
44  *
45  * Wildcard entries (PCI_ANY_ID) should come last
46  * Last entry must be all 0s
47  *
48  * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
49  *   Class, Class Mask, private data (not used) }
50  */
51 static struct pci_device_id ixgb_pci_tbl[] = {
52         {INTEL_VENDOR_ID, IXGB_DEVICE_ID_82597EX,
53          PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0},
54         {INTEL_VENDOR_ID, IXGB_DEVICE_ID_82597EX_CX4,
55          PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0},
56         {INTEL_VENDOR_ID, IXGB_DEVICE_ID_82597EX_SR,
57          PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0},
58         {INTEL_VENDOR_ID, IXGB_DEVICE_ID_82597EX_LR,  
59          PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0},
60
61         /* required last entry */
62         {0,}
63 };
64
65 MODULE_DEVICE_TABLE(pci, ixgb_pci_tbl);
66
67 /* Local Function Prototypes */
68
69 int ixgb_up(struct ixgb_adapter *adapter);
70 void ixgb_down(struct ixgb_adapter *adapter, boolean_t kill_watchdog);
71 void ixgb_reset(struct ixgb_adapter *adapter);
72 int ixgb_setup_tx_resources(struct ixgb_adapter *adapter);
73 int ixgb_setup_rx_resources(struct ixgb_adapter *adapter);
74 void ixgb_free_tx_resources(struct ixgb_adapter *adapter);
75 void ixgb_free_rx_resources(struct ixgb_adapter *adapter);
76 void ixgb_update_stats(struct ixgb_adapter *adapter);
77
78 static int ixgb_init_module(void);
79 static void ixgb_exit_module(void);
80 static int ixgb_probe(struct pci_dev *pdev, const struct pci_device_id *ent);
81 static void __devexit ixgb_remove(struct pci_dev *pdev);
82 static int ixgb_sw_init(struct ixgb_adapter *adapter);
83 static int ixgb_open(struct net_device *netdev);
84 static int ixgb_close(struct net_device *netdev);
85 static void ixgb_configure_tx(struct ixgb_adapter *adapter);
86 static void ixgb_configure_rx(struct ixgb_adapter *adapter);
87 static void ixgb_setup_rctl(struct ixgb_adapter *adapter);
88 static void ixgb_clean_tx_ring(struct ixgb_adapter *adapter);
89 static void ixgb_clean_rx_ring(struct ixgb_adapter *adapter);
90 static void ixgb_set_multi(struct net_device *netdev);
91 static void ixgb_watchdog(unsigned long data);
92 static int ixgb_xmit_frame(struct sk_buff *skb, struct net_device *netdev);
93 static struct net_device_stats *ixgb_get_stats(struct net_device *netdev);
94 static int ixgb_change_mtu(struct net_device *netdev, int new_mtu);
95 static int ixgb_set_mac(struct net_device *netdev, void *p);
96 static irqreturn_t ixgb_intr(int irq, void *data);
97 static boolean_t ixgb_clean_tx_irq(struct ixgb_adapter *adapter);
98
99 #ifdef CONFIG_IXGB_NAPI
100 static int ixgb_clean(struct napi_struct *napi, int budget);
101 static boolean_t ixgb_clean_rx_irq(struct ixgb_adapter *adapter,
102                                    int *work_done, int work_to_do);
103 #else
104 static boolean_t ixgb_clean_rx_irq(struct ixgb_adapter *adapter);
105 #endif
106 static void ixgb_alloc_rx_buffers(struct ixgb_adapter *adapter);
107 void ixgb_set_ethtool_ops(struct net_device *netdev);
108 static void ixgb_tx_timeout(struct net_device *dev);
109 static void ixgb_tx_timeout_task(struct work_struct *work);
110 static void ixgb_vlan_rx_register(struct net_device *netdev,
111                                   struct vlan_group *grp);
112 static void ixgb_vlan_rx_add_vid(struct net_device *netdev, uint16_t vid);
113 static void ixgb_vlan_rx_kill_vid(struct net_device *netdev, uint16_t vid);
114 static void ixgb_restore_vlan(struct ixgb_adapter *adapter);
115
116 #ifdef CONFIG_NET_POLL_CONTROLLER
117 /* for netdump / net console */
118 static void ixgb_netpoll(struct net_device *dev);
119 #endif
120
121 static pci_ers_result_t ixgb_io_error_detected (struct pci_dev *pdev,
122                              enum pci_channel_state state);
123 static pci_ers_result_t ixgb_io_slot_reset (struct pci_dev *pdev);
124 static void ixgb_io_resume (struct pci_dev *pdev);
125
126 /* Exported from other modules */
127 extern void ixgb_check_options(struct ixgb_adapter *adapter);
128
129 static struct pci_error_handlers ixgb_err_handler = {
130         .error_detected = ixgb_io_error_detected,
131         .slot_reset = ixgb_io_slot_reset,
132         .resume = ixgb_io_resume,
133 };
134
135 static struct pci_driver ixgb_driver = {
136         .name     = ixgb_driver_name,
137         .id_table = ixgb_pci_tbl,
138         .probe    = ixgb_probe,
139         .remove   = __devexit_p(ixgb_remove),
140         .err_handler = &ixgb_err_handler
141 };
142
143 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
144 MODULE_DESCRIPTION("Intel(R) PRO/10GbE Network Driver");
145 MODULE_LICENSE("GPL");
146 MODULE_VERSION(DRV_VERSION);
147
148 #define DEFAULT_DEBUG_LEVEL_SHIFT 3
149 static int debug = DEFAULT_DEBUG_LEVEL_SHIFT;
150 module_param(debug, int, 0);
151 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
152
153 /* some defines for controlling descriptor fetches in h/w */
154 #define RXDCTL_WTHRESH_DEFAULT 15  /* chip writes back at this many or RXT0 */
155 #define RXDCTL_PTHRESH_DEFAULT 0   /* chip considers prefech below
156                                     * this */
157 #define RXDCTL_HTHRESH_DEFAULT 0   /* chip will only prefetch if tail
158                                     * is pushed this many descriptors
159                                     * from head */
160
161 /**
162  * ixgb_init_module - Driver Registration Routine
163  *
164  * ixgb_init_module is the first routine called when the driver is
165  * loaded. All it does is register with the PCI subsystem.
166  **/
167
168 static int __init
169 ixgb_init_module(void)
170 {
171         printk(KERN_INFO "%s - version %s\n",
172                ixgb_driver_string, ixgb_driver_version);
173
174         printk(KERN_INFO "%s\n", ixgb_copyright);
175
176         return pci_register_driver(&ixgb_driver);
177 }
178
179 module_init(ixgb_init_module);
180
181 /**
182  * ixgb_exit_module - Driver Exit Cleanup Routine
183  *
184  * ixgb_exit_module is called just before the driver is removed
185  * from memory.
186  **/
187
188 static void __exit
189 ixgb_exit_module(void)
190 {
191         pci_unregister_driver(&ixgb_driver);
192 }
193
194 module_exit(ixgb_exit_module);
195
196 /**
197  * ixgb_irq_disable - Mask off interrupt generation on the NIC
198  * @adapter: board private structure
199  **/
200
201 static void
202 ixgb_irq_disable(struct ixgb_adapter *adapter)
203 {
204         atomic_inc(&adapter->irq_sem);
205         IXGB_WRITE_REG(&adapter->hw, IMC, ~0);
206         IXGB_WRITE_FLUSH(&adapter->hw);
207         synchronize_irq(adapter->pdev->irq);
208 }
209
210 /**
211  * ixgb_irq_enable - Enable default interrupt generation settings
212  * @adapter: board private structure
213  **/
214
215 static void
216 ixgb_irq_enable(struct ixgb_adapter *adapter)
217 {
218         if(atomic_dec_and_test(&adapter->irq_sem)) {
219                 IXGB_WRITE_REG(&adapter->hw, IMS,
220                                IXGB_INT_RXT0 | IXGB_INT_RXDMT0 | IXGB_INT_TXDW |
221                                IXGB_INT_LSC);
222                 IXGB_WRITE_FLUSH(&adapter->hw);
223         }
224 }
225
226 int
227 ixgb_up(struct ixgb_adapter *adapter)
228 {
229         struct net_device *netdev = adapter->netdev;
230         int err, irq_flags = IRQF_SHARED;
231         int max_frame = netdev->mtu + ENET_HEADER_SIZE + ENET_FCS_LENGTH;
232         struct ixgb_hw *hw = &adapter->hw;
233
234         /* hardware has been reset, we need to reload some things */
235
236         ixgb_rar_set(hw, netdev->dev_addr, 0);
237         ixgb_set_multi(netdev);
238
239         ixgb_restore_vlan(adapter);
240
241         ixgb_configure_tx(adapter);
242         ixgb_setup_rctl(adapter);
243         ixgb_configure_rx(adapter);
244         ixgb_alloc_rx_buffers(adapter);
245
246         /* disable interrupts and get the hardware into a known state */
247         IXGB_WRITE_REG(&adapter->hw, IMC, 0xffffffff);
248
249         /* only enable MSI if bus is in PCI-X mode */
250         if (IXGB_READ_REG(&adapter->hw, STATUS) & IXGB_STATUS_PCIX_MODE) {
251                 err = pci_enable_msi(adapter->pdev);
252                 if (!err) {
253                         adapter->have_msi = 1;
254                         irq_flags = 0;
255                 }
256                 /* proceed to try to request regular interrupt */
257         }
258
259         err = request_irq(adapter->pdev->irq, &ixgb_intr, irq_flags,
260                           netdev->name, netdev);
261         if (err) {
262                 if (adapter->have_msi)
263                         pci_disable_msi(adapter->pdev);
264                 DPRINTK(PROBE, ERR,
265                  "Unable to allocate interrupt Error: %d\n", err);
266                 return err;
267         }
268
269         if((hw->max_frame_size != max_frame) ||
270                 (hw->max_frame_size !=
271                 (IXGB_READ_REG(hw, MFS) >> IXGB_MFS_SHIFT))) {
272
273                 hw->max_frame_size = max_frame;
274
275                 IXGB_WRITE_REG(hw, MFS, hw->max_frame_size << IXGB_MFS_SHIFT);
276
277                 if(hw->max_frame_size >
278                    IXGB_MAX_ENET_FRAME_SIZE_WITHOUT_FCS + ENET_FCS_LENGTH) {
279                         uint32_t ctrl0 = IXGB_READ_REG(hw, CTRL0);
280
281                         if(!(ctrl0 & IXGB_CTRL0_JFE)) {
282                                 ctrl0 |= IXGB_CTRL0_JFE;
283                                 IXGB_WRITE_REG(hw, CTRL0, ctrl0);
284                         }
285                 }
286         }
287
288         mod_timer(&adapter->watchdog_timer, jiffies);
289
290 #ifdef CONFIG_IXGB_NAPI
291         napi_enable(&adapter->napi);
292 #endif
293         ixgb_irq_enable(adapter);
294
295         return 0;
296 }
297
298 void
299 ixgb_down(struct ixgb_adapter *adapter, boolean_t kill_watchdog)
300 {
301         struct net_device *netdev = adapter->netdev;
302
303         ixgb_irq_disable(adapter);
304         free_irq(adapter->pdev->irq, netdev);
305
306         if (adapter->have_msi)
307                 pci_disable_msi(adapter->pdev);
308
309         if(kill_watchdog)
310                 del_timer_sync(&adapter->watchdog_timer);
311 #ifdef CONFIG_IXGB_NAPI
312         napi_disable(&adapter->napi);
313 #endif
314         adapter->link_speed = 0;
315         adapter->link_duplex = 0;
316         netif_carrier_off(netdev);
317         netif_stop_queue(netdev);
318
319         ixgb_reset(adapter);
320         ixgb_clean_tx_ring(adapter);
321         ixgb_clean_rx_ring(adapter);
322 }
323
324 void
325 ixgb_reset(struct ixgb_adapter *adapter)
326 {
327
328         ixgb_adapter_stop(&adapter->hw);
329         if(!ixgb_init_hw(&adapter->hw))
330                 DPRINTK(PROBE, ERR, "ixgb_init_hw failed.\n");
331 }
332
333 /**
334  * ixgb_probe - Device Initialization Routine
335  * @pdev: PCI device information struct
336  * @ent: entry in ixgb_pci_tbl
337  *
338  * Returns 0 on success, negative on failure
339  *
340  * ixgb_probe initializes an adapter identified by a pci_dev structure.
341  * The OS initialization, configuring of the adapter private structure,
342  * and a hardware reset occur.
343  **/
344
345 static int __devinit
346 ixgb_probe(struct pci_dev *pdev,
347                 const struct pci_device_id *ent)
348 {
349         struct net_device *netdev = NULL;
350         struct ixgb_adapter *adapter;
351         static int cards_found = 0;
352         unsigned long mmio_start;
353         int mmio_len;
354         int pci_using_dac;
355         int i;
356         int err;
357
358         if((err = pci_enable_device(pdev)))
359                 return err;
360
361         if(!(err = pci_set_dma_mask(pdev, DMA_64BIT_MASK)) &&
362            !(err = pci_set_consistent_dma_mask(pdev, DMA_64BIT_MASK))) {
363                 pci_using_dac = 1;
364         } else {
365                 if((err = pci_set_dma_mask(pdev, DMA_32BIT_MASK)) ||
366                    (err = pci_set_consistent_dma_mask(pdev, DMA_32BIT_MASK))) {
367                         printk(KERN_ERR
368                          "ixgb: No usable DMA configuration, aborting\n");
369                         goto err_dma_mask;
370                 }
371                 pci_using_dac = 0;
372         }
373
374         if((err = pci_request_regions(pdev, ixgb_driver_name)))
375                 goto err_request_regions;
376
377         pci_set_master(pdev);
378
379         netdev = alloc_etherdev(sizeof(struct ixgb_adapter));
380         if(!netdev) {
381                 err = -ENOMEM;
382                 goto err_alloc_etherdev;
383         }
384
385         SET_MODULE_OWNER(netdev);
386         SET_NETDEV_DEV(netdev, &pdev->dev);
387
388         pci_set_drvdata(pdev, netdev);
389         adapter = netdev_priv(netdev);
390         adapter->netdev = netdev;
391         adapter->pdev = pdev;
392         adapter->hw.back = adapter;
393         adapter->msg_enable = netif_msg_init(debug, DEFAULT_DEBUG_LEVEL_SHIFT);
394
395         mmio_start = pci_resource_start(pdev, BAR_0);
396         mmio_len = pci_resource_len(pdev, BAR_0);
397
398         adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
399         if(!adapter->hw.hw_addr) {
400                 err = -EIO;
401                 goto err_ioremap;
402         }
403
404         for(i = BAR_1; i <= BAR_5; i++) {
405                 if(pci_resource_len(pdev, i) == 0)
406                         continue;
407                 if(pci_resource_flags(pdev, i) & IORESOURCE_IO) {
408                         adapter->hw.io_base = pci_resource_start(pdev, i);
409                         break;
410                 }
411         }
412
413         netdev->open = &ixgb_open;
414         netdev->stop = &ixgb_close;
415         netdev->hard_start_xmit = &ixgb_xmit_frame;
416         netdev->get_stats = &ixgb_get_stats;
417         netdev->set_multicast_list = &ixgb_set_multi;
418         netdev->set_mac_address = &ixgb_set_mac;
419         netdev->change_mtu = &ixgb_change_mtu;
420         ixgb_set_ethtool_ops(netdev);
421         netdev->tx_timeout = &ixgb_tx_timeout;
422         netdev->watchdog_timeo = 5 * HZ;
423 #ifdef CONFIG_IXGB_NAPI
424         netif_napi_add(netdev, &adapter->napi, ixgb_clean, 64);
425 #endif
426         netdev->vlan_rx_register = ixgb_vlan_rx_register;
427         netdev->vlan_rx_add_vid = ixgb_vlan_rx_add_vid;
428         netdev->vlan_rx_kill_vid = ixgb_vlan_rx_kill_vid;
429 #ifdef CONFIG_NET_POLL_CONTROLLER
430         netdev->poll_controller = ixgb_netpoll;
431 #endif
432
433         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
434         netdev->mem_start = mmio_start;
435         netdev->mem_end = mmio_start + mmio_len;
436         netdev->base_addr = adapter->hw.io_base;
437
438         adapter->bd_number = cards_found;
439         adapter->link_speed = 0;
440         adapter->link_duplex = 0;
441
442         /* setup the private structure */
443
444         if((err = ixgb_sw_init(adapter)))
445                 goto err_sw_init;
446
447         netdev->features = NETIF_F_SG |
448                            NETIF_F_HW_CSUM |
449                            NETIF_F_HW_VLAN_TX |
450                            NETIF_F_HW_VLAN_RX |
451                            NETIF_F_HW_VLAN_FILTER;
452         netdev->features |= NETIF_F_TSO;
453 #ifdef NETIF_F_LLTX
454         netdev->features |= NETIF_F_LLTX;
455 #endif
456
457         if(pci_using_dac)
458                 netdev->features |= NETIF_F_HIGHDMA;
459
460         /* make sure the EEPROM is good */
461
462         if(!ixgb_validate_eeprom_checksum(&adapter->hw)) {
463                 DPRINTK(PROBE, ERR, "The EEPROM Checksum Is Not Valid\n");
464                 err = -EIO;
465                 goto err_eeprom;
466         }
467
468         ixgb_get_ee_mac_addr(&adapter->hw, netdev->dev_addr);
469         memcpy(netdev->perm_addr, netdev->dev_addr, netdev->addr_len);
470
471         if(!is_valid_ether_addr(netdev->perm_addr)) {
472                 DPRINTK(PROBE, ERR, "Invalid MAC Address\n");
473                 err = -EIO;
474                 goto err_eeprom;
475         }
476
477         adapter->part_num = ixgb_get_ee_pba_number(&adapter->hw);
478
479         init_timer(&adapter->watchdog_timer);
480         adapter->watchdog_timer.function = &ixgb_watchdog;
481         adapter->watchdog_timer.data = (unsigned long)adapter;
482
483         INIT_WORK(&adapter->tx_timeout_task, ixgb_tx_timeout_task);
484
485         strcpy(netdev->name, "eth%d");
486         if((err = register_netdev(netdev)))
487                 goto err_register;
488
489         /* we're going to reset, so assume we have no link for now */
490
491         netif_carrier_off(netdev);
492         netif_stop_queue(netdev);
493
494         DPRINTK(PROBE, INFO, "Intel(R) PRO/10GbE Network Connection\n");
495         ixgb_check_options(adapter);
496         /* reset the hardware with the new settings */
497
498         ixgb_reset(adapter);
499
500         cards_found++;
501         return 0;
502
503 err_register:
504 err_sw_init:
505 err_eeprom:
506         iounmap(adapter->hw.hw_addr);
507 err_ioremap:
508         free_netdev(netdev);
509 err_alloc_etherdev:
510         pci_release_regions(pdev);
511 err_request_regions:
512 err_dma_mask:
513         pci_disable_device(pdev);
514         return err;
515 }
516
517 /**
518  * ixgb_remove - Device Removal Routine
519  * @pdev: PCI device information struct
520  *
521  * ixgb_remove is called by the PCI subsystem to alert the driver
522  * that it should release a PCI device.  The could be caused by a
523  * Hot-Plug event, or because the driver is going to be removed from
524  * memory.
525  **/
526
527 static void __devexit
528 ixgb_remove(struct pci_dev *pdev)
529 {
530         struct net_device *netdev = pci_get_drvdata(pdev);
531         struct ixgb_adapter *adapter = netdev_priv(netdev);
532
533         unregister_netdev(netdev);
534
535         iounmap(adapter->hw.hw_addr);
536         pci_release_regions(pdev);
537
538         free_netdev(netdev);
539 }
540
541 /**
542  * ixgb_sw_init - Initialize general software structures (struct ixgb_adapter)
543  * @adapter: board private structure to initialize
544  *
545  * ixgb_sw_init initializes the Adapter private data structure.
546  * Fields are initialized based on PCI device information and
547  * OS network device settings (MTU size).
548  **/
549
550 static int __devinit
551 ixgb_sw_init(struct ixgb_adapter *adapter)
552 {
553         struct ixgb_hw *hw = &adapter->hw;
554         struct net_device *netdev = adapter->netdev;
555         struct pci_dev *pdev = adapter->pdev;
556
557         /* PCI config space info */
558
559         hw->vendor_id = pdev->vendor;
560         hw->device_id = pdev->device;
561         hw->subsystem_vendor_id = pdev->subsystem_vendor;
562         hw->subsystem_id = pdev->subsystem_device;
563
564         hw->max_frame_size = netdev->mtu + ENET_HEADER_SIZE + ENET_FCS_LENGTH;
565         adapter->rx_buffer_len = hw->max_frame_size;
566
567         if((hw->device_id == IXGB_DEVICE_ID_82597EX)
568            || (hw->device_id == IXGB_DEVICE_ID_82597EX_CX4)
569            || (hw->device_id == IXGB_DEVICE_ID_82597EX_LR)
570            || (hw->device_id == IXGB_DEVICE_ID_82597EX_SR))
571                         hw->mac_type = ixgb_82597;
572         else {
573                 /* should never have loaded on this device */
574                 DPRINTK(PROBE, ERR, "unsupported device id\n");
575         }
576
577         /* enable flow control to be programmed */
578         hw->fc.send_xon = 1;
579
580         atomic_set(&adapter->irq_sem, 1);
581         spin_lock_init(&adapter->tx_lock);
582
583         return 0;
584 }
585
586 /**
587  * ixgb_open - Called when a network interface is made active
588  * @netdev: network interface device structure
589  *
590  * Returns 0 on success, negative value on failure
591  *
592  * The open entry point is called when a network interface is made
593  * active by the system (IFF_UP).  At this point all resources needed
594  * for transmit and receive operations are allocated, the interrupt
595  * handler is registered with the OS, the watchdog timer is started,
596  * and the stack is notified that the interface is ready.
597  **/
598
599 static int
600 ixgb_open(struct net_device *netdev)
601 {
602         struct ixgb_adapter *adapter = netdev_priv(netdev);
603         int err;
604
605         /* allocate transmit descriptors */
606
607         if((err = ixgb_setup_tx_resources(adapter)))
608                 goto err_setup_tx;
609
610         /* allocate receive descriptors */
611
612         if((err = ixgb_setup_rx_resources(adapter)))
613                 goto err_setup_rx;
614
615         if((err = ixgb_up(adapter)))
616                 goto err_up;
617
618         return 0;
619
620 err_up:
621         ixgb_free_rx_resources(adapter);
622 err_setup_rx:
623         ixgb_free_tx_resources(adapter);
624 err_setup_tx:
625         ixgb_reset(adapter);
626
627         return err;
628 }
629
630 /**
631  * ixgb_close - Disables a network interface
632  * @netdev: network interface device structure
633  *
634  * Returns 0, this is not allowed to fail
635  *
636  * The close entry point is called when an interface is de-activated
637  * by the OS.  The hardware is still under the drivers control, but
638  * needs to be disabled.  A global MAC reset is issued to stop the
639  * hardware, and all transmit and receive resources are freed.
640  **/
641
642 static int
643 ixgb_close(struct net_device *netdev)
644 {
645         struct ixgb_adapter *adapter = netdev_priv(netdev);
646
647         ixgb_down(adapter, TRUE);
648
649         ixgb_free_tx_resources(adapter);
650         ixgb_free_rx_resources(adapter);
651
652         return 0;
653 }
654
655 /**
656  * ixgb_setup_tx_resources - allocate Tx resources (Descriptors)
657  * @adapter: board private structure
658  *
659  * Return 0 on success, negative on failure
660  **/
661
662 int
663 ixgb_setup_tx_resources(struct ixgb_adapter *adapter)
664 {
665         struct ixgb_desc_ring *txdr = &adapter->tx_ring;
666         struct pci_dev *pdev = adapter->pdev;
667         int size;
668
669         size = sizeof(struct ixgb_buffer) * txdr->count;
670         txdr->buffer_info = vmalloc(size);
671         if(!txdr->buffer_info) {
672                 DPRINTK(PROBE, ERR,
673                  "Unable to allocate transmit descriptor ring memory\n");
674                 return -ENOMEM;
675         }
676         memset(txdr->buffer_info, 0, size);
677
678         /* round up to nearest 4K */
679
680         txdr->size = txdr->count * sizeof(struct ixgb_tx_desc);
681         txdr->size = ALIGN(txdr->size, 4096);
682
683         txdr->desc = pci_alloc_consistent(pdev, txdr->size, &txdr->dma);
684         if(!txdr->desc) {
685                 vfree(txdr->buffer_info);
686                 DPRINTK(PROBE, ERR,
687                  "Unable to allocate transmit descriptor memory\n");
688                 return -ENOMEM;
689         }
690         memset(txdr->desc, 0, txdr->size);
691
692         txdr->next_to_use = 0;
693         txdr->next_to_clean = 0;
694
695         return 0;
696 }
697
698 /**
699  * ixgb_configure_tx - Configure 82597 Transmit Unit after Reset.
700  * @adapter: board private structure
701  *
702  * Configure the Tx unit of the MAC after a reset.
703  **/
704
705 static void
706 ixgb_configure_tx(struct ixgb_adapter *adapter)
707 {
708         uint64_t tdba = adapter->tx_ring.dma;
709         uint32_t tdlen = adapter->tx_ring.count * sizeof(struct ixgb_tx_desc);
710         uint32_t tctl;
711         struct ixgb_hw *hw = &adapter->hw;
712
713         /* Setup the Base and Length of the Tx Descriptor Ring 
714          * tx_ring.dma can be either a 32 or 64 bit value 
715          */
716
717         IXGB_WRITE_REG(hw, TDBAL, (tdba & 0x00000000ffffffffULL));
718         IXGB_WRITE_REG(hw, TDBAH, (tdba >> 32));
719
720         IXGB_WRITE_REG(hw, TDLEN, tdlen);
721
722         /* Setup the HW Tx Head and Tail descriptor pointers */
723
724         IXGB_WRITE_REG(hw, TDH, 0);
725         IXGB_WRITE_REG(hw, TDT, 0);
726
727         /* don't set up txdctl, it induces performance problems if configured
728          * incorrectly */
729         /* Set the Tx Interrupt Delay register */
730
731         IXGB_WRITE_REG(hw, TIDV, adapter->tx_int_delay);
732
733         /* Program the Transmit Control Register */
734
735         tctl = IXGB_TCTL_TCE | IXGB_TCTL_TXEN | IXGB_TCTL_TPDE;
736         IXGB_WRITE_REG(hw, TCTL, tctl);
737
738         /* Setup Transmit Descriptor Settings for this adapter */
739         adapter->tx_cmd_type =
740                 IXGB_TX_DESC_TYPE 
741                 | (adapter->tx_int_delay_enable ? IXGB_TX_DESC_CMD_IDE : 0);
742 }
743
744 /**
745  * ixgb_setup_rx_resources - allocate Rx resources (Descriptors)
746  * @adapter: board private structure
747  *
748  * Returns 0 on success, negative on failure
749  **/
750
751 int
752 ixgb_setup_rx_resources(struct ixgb_adapter *adapter)
753 {
754         struct ixgb_desc_ring *rxdr = &adapter->rx_ring;
755         struct pci_dev *pdev = adapter->pdev;
756         int size;
757
758         size = sizeof(struct ixgb_buffer) * rxdr->count;
759         rxdr->buffer_info = vmalloc(size);
760         if(!rxdr->buffer_info) {
761                 DPRINTK(PROBE, ERR,
762                  "Unable to allocate receive descriptor ring\n");
763                 return -ENOMEM;
764         }
765         memset(rxdr->buffer_info, 0, size);
766
767         /* Round up to nearest 4K */
768
769         rxdr->size = rxdr->count * sizeof(struct ixgb_rx_desc);
770         rxdr->size = ALIGN(rxdr->size, 4096);
771
772         rxdr->desc = pci_alloc_consistent(pdev, rxdr->size, &rxdr->dma);
773
774         if(!rxdr->desc) {
775                 vfree(rxdr->buffer_info);
776                 DPRINTK(PROBE, ERR,
777                  "Unable to allocate receive descriptors\n");
778                 return -ENOMEM;
779         }
780         memset(rxdr->desc, 0, rxdr->size);
781
782         rxdr->next_to_clean = 0;
783         rxdr->next_to_use = 0;
784
785         return 0;
786 }
787
788 /**
789  * ixgb_setup_rctl - configure the receive control register
790  * @adapter: Board private structure
791  **/
792
793 static void
794 ixgb_setup_rctl(struct ixgb_adapter *adapter)
795 {
796         uint32_t rctl;
797
798         rctl = IXGB_READ_REG(&adapter->hw, RCTL);
799
800         rctl &= ~(3 << IXGB_RCTL_MO_SHIFT);
801
802         rctl |=
803                 IXGB_RCTL_BAM | IXGB_RCTL_RDMTS_1_2 | 
804                 IXGB_RCTL_RXEN | IXGB_RCTL_CFF | 
805                 (adapter->hw.mc_filter_type << IXGB_RCTL_MO_SHIFT);
806
807         rctl |= IXGB_RCTL_SECRC;
808
809         if (adapter->rx_buffer_len <= IXGB_RXBUFFER_2048)
810                 rctl |= IXGB_RCTL_BSIZE_2048;
811         else if (adapter->rx_buffer_len <= IXGB_RXBUFFER_4096)
812                 rctl |= IXGB_RCTL_BSIZE_4096;
813         else if (adapter->rx_buffer_len <= IXGB_RXBUFFER_8192)
814                 rctl |= IXGB_RCTL_BSIZE_8192;
815         else if (adapter->rx_buffer_len <= IXGB_RXBUFFER_16384)
816                 rctl |= IXGB_RCTL_BSIZE_16384;
817
818         IXGB_WRITE_REG(&adapter->hw, RCTL, rctl);
819 }
820
821 /**
822  * ixgb_configure_rx - Configure 82597 Receive Unit after Reset.
823  * @adapter: board private structure
824  *
825  * Configure the Rx unit of the MAC after a reset.
826  **/
827
828 static void
829 ixgb_configure_rx(struct ixgb_adapter *adapter)
830 {
831         uint64_t rdba = adapter->rx_ring.dma;
832         uint32_t rdlen = adapter->rx_ring.count * sizeof(struct ixgb_rx_desc);
833         struct ixgb_hw *hw = &adapter->hw;
834         uint32_t rctl;
835         uint32_t rxcsum;
836         uint32_t rxdctl;
837
838         /* make sure receives are disabled while setting up the descriptors */
839
840         rctl = IXGB_READ_REG(hw, RCTL);
841         IXGB_WRITE_REG(hw, RCTL, rctl & ~IXGB_RCTL_RXEN);
842
843         /* set the Receive Delay Timer Register */
844
845         IXGB_WRITE_REG(hw, RDTR, adapter->rx_int_delay);
846
847         /* Setup the Base and Length of the Rx Descriptor Ring */
848
849         IXGB_WRITE_REG(hw, RDBAL, (rdba & 0x00000000ffffffffULL));
850         IXGB_WRITE_REG(hw, RDBAH, (rdba >> 32));
851
852         IXGB_WRITE_REG(hw, RDLEN, rdlen);
853
854         /* Setup the HW Rx Head and Tail Descriptor Pointers */
855         IXGB_WRITE_REG(hw, RDH, 0);
856         IXGB_WRITE_REG(hw, RDT, 0);
857
858         /* set up pre-fetching of receive buffers so we get some before we
859          * run out (default hardware behavior is to run out before fetching
860          * more).  This sets up to fetch if HTHRESH rx descriptors are avail
861          * and the descriptors in hw cache are below PTHRESH.  This avoids
862          * the hardware behavior of fetching <=512 descriptors in a single
863          * burst that pre-empts all other activity, usually causing fifo
864          * overflows. */
865         /* use WTHRESH to burst write 16 descriptors or burst when RXT0 */
866         rxdctl = RXDCTL_WTHRESH_DEFAULT << IXGB_RXDCTL_WTHRESH_SHIFT |
867                  RXDCTL_HTHRESH_DEFAULT << IXGB_RXDCTL_HTHRESH_SHIFT |
868                  RXDCTL_PTHRESH_DEFAULT << IXGB_RXDCTL_PTHRESH_SHIFT;
869         IXGB_WRITE_REG(hw, RXDCTL, rxdctl);
870
871         /* Enable Receive Checksum Offload for TCP and UDP */
872         if(adapter->rx_csum == TRUE) {
873                 rxcsum = IXGB_READ_REG(hw, RXCSUM);
874                 rxcsum |= IXGB_RXCSUM_TUOFL;
875                 IXGB_WRITE_REG(hw, RXCSUM, rxcsum);
876         }
877
878         /* Enable Receives */
879
880         IXGB_WRITE_REG(hw, RCTL, rctl);
881 }
882
883 /**
884  * ixgb_free_tx_resources - Free Tx Resources
885  * @adapter: board private structure
886  *
887  * Free all transmit software resources
888  **/
889
890 void
891 ixgb_free_tx_resources(struct ixgb_adapter *adapter)
892 {
893         struct pci_dev *pdev = adapter->pdev;
894
895         ixgb_clean_tx_ring(adapter);
896
897         vfree(adapter->tx_ring.buffer_info);
898         adapter->tx_ring.buffer_info = NULL;
899
900         pci_free_consistent(pdev, adapter->tx_ring.size,
901                             adapter->tx_ring.desc, adapter->tx_ring.dma);
902
903         adapter->tx_ring.desc = NULL;
904 }
905
906 static void
907 ixgb_unmap_and_free_tx_resource(struct ixgb_adapter *adapter,
908                                         struct ixgb_buffer *buffer_info)
909 {
910         struct pci_dev *pdev = adapter->pdev;
911
912         if (buffer_info->dma)
913                 pci_unmap_page(pdev, buffer_info->dma, buffer_info->length,
914                                PCI_DMA_TODEVICE);
915
916         if (buffer_info->skb)
917                 dev_kfree_skb_any(buffer_info->skb);
918
919         buffer_info->skb = NULL;
920         buffer_info->dma = 0;
921         buffer_info->time_stamp = 0;
922         /* these fields must always be initialized in tx
923          * buffer_info->length = 0;
924          * buffer_info->next_to_watch = 0; */
925 }
926
927 /**
928  * ixgb_clean_tx_ring - Free Tx Buffers
929  * @adapter: board private structure
930  **/
931
932 static void
933 ixgb_clean_tx_ring(struct ixgb_adapter *adapter)
934 {
935         struct ixgb_desc_ring *tx_ring = &adapter->tx_ring;
936         struct ixgb_buffer *buffer_info;
937         unsigned long size;
938         unsigned int i;
939
940         /* Free all the Tx ring sk_buffs */
941
942         for(i = 0; i < tx_ring->count; i++) {
943                 buffer_info = &tx_ring->buffer_info[i];
944                 ixgb_unmap_and_free_tx_resource(adapter, buffer_info);
945         }
946
947         size = sizeof(struct ixgb_buffer) * tx_ring->count;
948         memset(tx_ring->buffer_info, 0, size);
949
950         /* Zero out the descriptor ring */
951
952         memset(tx_ring->desc, 0, tx_ring->size);
953
954         tx_ring->next_to_use = 0;
955         tx_ring->next_to_clean = 0;
956
957         IXGB_WRITE_REG(&adapter->hw, TDH, 0);
958         IXGB_WRITE_REG(&adapter->hw, TDT, 0);
959 }
960
961 /**
962  * ixgb_free_rx_resources - Free Rx Resources
963  * @adapter: board private structure
964  *
965  * Free all receive software resources
966  **/
967
968 void
969 ixgb_free_rx_resources(struct ixgb_adapter *adapter)
970 {
971         struct ixgb_desc_ring *rx_ring = &adapter->rx_ring;
972         struct pci_dev *pdev = adapter->pdev;
973
974         ixgb_clean_rx_ring(adapter);
975
976         vfree(rx_ring->buffer_info);
977         rx_ring->buffer_info = NULL;
978
979         pci_free_consistent(pdev, rx_ring->size, rx_ring->desc, rx_ring->dma);
980
981         rx_ring->desc = NULL;
982 }
983
984 /**
985  * ixgb_clean_rx_ring - Free Rx Buffers
986  * @adapter: board private structure
987  **/
988
989 static void
990 ixgb_clean_rx_ring(struct ixgb_adapter *adapter)
991 {
992         struct ixgb_desc_ring *rx_ring = &adapter->rx_ring;
993         struct ixgb_buffer *buffer_info;
994         struct pci_dev *pdev = adapter->pdev;
995         unsigned long size;
996         unsigned int i;
997
998         /* Free all the Rx ring sk_buffs */
999
1000         for(i = 0; i < rx_ring->count; i++) {
1001                 buffer_info = &rx_ring->buffer_info[i];
1002                 if(buffer_info->skb) {
1003
1004                         pci_unmap_single(pdev,
1005                                          buffer_info->dma,
1006                                          buffer_info->length,
1007                                          PCI_DMA_FROMDEVICE);
1008
1009                         dev_kfree_skb(buffer_info->skb);
1010
1011                         buffer_info->skb = NULL;
1012                 }
1013         }
1014
1015         size = sizeof(struct ixgb_buffer) * rx_ring->count;
1016         memset(rx_ring->buffer_info, 0, size);
1017
1018         /* Zero out the descriptor ring */
1019
1020         memset(rx_ring->desc, 0, rx_ring->size);
1021
1022         rx_ring->next_to_clean = 0;
1023         rx_ring->next_to_use = 0;
1024
1025         IXGB_WRITE_REG(&adapter->hw, RDH, 0);
1026         IXGB_WRITE_REG(&adapter->hw, RDT, 0);
1027 }
1028
1029 /**
1030  * ixgb_set_mac - Change the Ethernet Address of the NIC
1031  * @netdev: network interface device structure
1032  * @p: pointer to an address structure
1033  *
1034  * Returns 0 on success, negative on failure
1035  **/
1036
1037 static int
1038 ixgb_set_mac(struct net_device *netdev, void *p)
1039 {
1040         struct ixgb_adapter *adapter = netdev_priv(netdev);
1041         struct sockaddr *addr = p;
1042
1043         if(!is_valid_ether_addr(addr->sa_data))
1044                 return -EADDRNOTAVAIL;
1045
1046         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
1047
1048         ixgb_rar_set(&adapter->hw, addr->sa_data, 0);
1049
1050         return 0;
1051 }
1052
1053 /**
1054  * ixgb_set_multi - Multicast and Promiscuous mode set
1055  * @netdev: network interface device structure
1056  *
1057  * The set_multi entry point is called whenever the multicast address
1058  * list or the network interface flags are updated.  This routine is
1059  * responsible for configuring the hardware for proper multicast,
1060  * promiscuous mode, and all-multi behavior.
1061  **/
1062
1063 static void
1064 ixgb_set_multi(struct net_device *netdev)
1065 {
1066         struct ixgb_adapter *adapter = netdev_priv(netdev);
1067         struct ixgb_hw *hw = &adapter->hw;
1068         struct dev_mc_list *mc_ptr;
1069         uint32_t rctl;
1070         int i;
1071
1072         /* Check for Promiscuous and All Multicast modes */
1073
1074         rctl = IXGB_READ_REG(hw, RCTL);
1075
1076         if(netdev->flags & IFF_PROMISC) {
1077                 rctl |= (IXGB_RCTL_UPE | IXGB_RCTL_MPE);
1078         } else if(netdev->flags & IFF_ALLMULTI) {
1079                 rctl |= IXGB_RCTL_MPE;
1080                 rctl &= ~IXGB_RCTL_UPE;
1081         } else {
1082                 rctl &= ~(IXGB_RCTL_UPE | IXGB_RCTL_MPE);
1083         }
1084
1085         if(netdev->mc_count > IXGB_MAX_NUM_MULTICAST_ADDRESSES) {
1086                 rctl |= IXGB_RCTL_MPE;
1087                 IXGB_WRITE_REG(hw, RCTL, rctl);
1088         } else {
1089                 uint8_t mta[netdev->mc_count * IXGB_ETH_LENGTH_OF_ADDRESS];
1090
1091                 IXGB_WRITE_REG(hw, RCTL, rctl);
1092
1093                 for(i = 0, mc_ptr = netdev->mc_list; mc_ptr;
1094                         i++, mc_ptr = mc_ptr->next)
1095                         memcpy(&mta[i * IXGB_ETH_LENGTH_OF_ADDRESS],
1096                                    mc_ptr->dmi_addr, IXGB_ETH_LENGTH_OF_ADDRESS);
1097
1098                 ixgb_mc_addr_list_update(hw, mta, netdev->mc_count, 0);
1099         }
1100 }
1101
1102 /**
1103  * ixgb_watchdog - Timer Call-back
1104  * @data: pointer to netdev cast into an unsigned long
1105  **/
1106
1107 static void
1108 ixgb_watchdog(unsigned long data)
1109 {
1110         struct ixgb_adapter *adapter = (struct ixgb_adapter *)data;
1111         struct net_device *netdev = adapter->netdev;
1112         struct ixgb_desc_ring *txdr = &adapter->tx_ring;
1113
1114         ixgb_check_for_link(&adapter->hw);
1115
1116         if (ixgb_check_for_bad_link(&adapter->hw)) {
1117                 /* force the reset path */
1118                 netif_stop_queue(netdev);
1119         }
1120
1121         if(adapter->hw.link_up) {
1122                 if(!netif_carrier_ok(netdev)) {
1123                         DPRINTK(LINK, INFO,
1124                                 "NIC Link is Up 10000 Mbps Full Duplex\n");
1125                         adapter->link_speed = 10000;
1126                         adapter->link_duplex = FULL_DUPLEX;
1127                         netif_carrier_on(netdev);
1128                         netif_wake_queue(netdev);
1129                 }
1130         } else {
1131                 if(netif_carrier_ok(netdev)) {
1132                         adapter->link_speed = 0;
1133                         adapter->link_duplex = 0;
1134                         DPRINTK(LINK, INFO, "NIC Link is Down\n");
1135                         netif_carrier_off(netdev);
1136                         netif_stop_queue(netdev);
1137
1138                 }
1139         }
1140
1141         ixgb_update_stats(adapter);
1142
1143         if(!netif_carrier_ok(netdev)) {
1144                 if(IXGB_DESC_UNUSED(txdr) + 1 < txdr->count) {
1145                         /* We've lost link, so the controller stops DMA,
1146                          * but we've got queued Tx work that's never going
1147                          * to get done, so reset controller to flush Tx.
1148                          * (Do the reset outside of interrupt context). */
1149                         schedule_work(&adapter->tx_timeout_task);
1150                 }
1151         }
1152
1153         /* Force detection of hung controller every watchdog period */
1154         adapter->detect_tx_hung = TRUE;
1155
1156         /* generate an interrupt to force clean up of any stragglers */
1157         IXGB_WRITE_REG(&adapter->hw, ICS, IXGB_INT_TXDW);
1158
1159         /* Reset the timer */
1160         mod_timer(&adapter->watchdog_timer, jiffies + 2 * HZ);
1161 }
1162
1163 #define IXGB_TX_FLAGS_CSUM              0x00000001
1164 #define IXGB_TX_FLAGS_VLAN              0x00000002
1165 #define IXGB_TX_FLAGS_TSO               0x00000004
1166
1167 static int
1168 ixgb_tso(struct ixgb_adapter *adapter, struct sk_buff *skb)
1169 {
1170         struct ixgb_context_desc *context_desc;
1171         unsigned int i;
1172         uint8_t ipcss, ipcso, tucss, tucso, hdr_len;
1173         uint16_t ipcse, tucse, mss;
1174         int err;
1175
1176         if (likely(skb_is_gso(skb))) {
1177                 struct ixgb_buffer *buffer_info;
1178                 struct iphdr *iph;
1179
1180                 if (skb_header_cloned(skb)) {
1181                         err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1182                         if (err)
1183                                 return err;
1184                 }
1185
1186                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
1187                 mss = skb_shinfo(skb)->gso_size;
1188                 iph = ip_hdr(skb);
1189                 iph->tot_len = 0;
1190                 iph->check = 0;
1191                 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
1192                                                          iph->daddr, 0,
1193                                                          IPPROTO_TCP, 0);
1194                 ipcss = skb_network_offset(skb);
1195                 ipcso = (void *)&(iph->check) - (void *)skb->data;
1196                 ipcse = skb_transport_offset(skb) - 1;
1197                 tucss = skb_transport_offset(skb);
1198                 tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
1199                 tucse = 0;
1200
1201                 i = adapter->tx_ring.next_to_use;
1202                 context_desc = IXGB_CONTEXT_DESC(adapter->tx_ring, i);
1203                 buffer_info = &adapter->tx_ring.buffer_info[i];
1204                 WARN_ON(buffer_info->dma != 0);
1205
1206                 context_desc->ipcss = ipcss;
1207                 context_desc->ipcso = ipcso;
1208                 context_desc->ipcse = cpu_to_le16(ipcse);
1209                 context_desc->tucss = tucss;
1210                 context_desc->tucso = tucso;
1211                 context_desc->tucse = cpu_to_le16(tucse);
1212                 context_desc->mss = cpu_to_le16(mss);
1213                 context_desc->hdr_len = hdr_len;
1214                 context_desc->status = 0;
1215                 context_desc->cmd_type_len = cpu_to_le32(
1216                                                   IXGB_CONTEXT_DESC_TYPE 
1217                                                 | IXGB_CONTEXT_DESC_CMD_TSE
1218                                                 | IXGB_CONTEXT_DESC_CMD_IP
1219                                                 | IXGB_CONTEXT_DESC_CMD_TCP
1220                                                 | IXGB_CONTEXT_DESC_CMD_IDE
1221                                                 | (skb->len - (hdr_len)));
1222
1223
1224                 if(++i == adapter->tx_ring.count) i = 0;
1225                 adapter->tx_ring.next_to_use = i;
1226
1227                 return 1;
1228         }
1229
1230         return 0;
1231 }
1232
1233 static boolean_t
1234 ixgb_tx_csum(struct ixgb_adapter *adapter, struct sk_buff *skb)
1235 {
1236         struct ixgb_context_desc *context_desc;
1237         unsigned int i;
1238         uint8_t css, cso;
1239
1240         if(likely(skb->ip_summed == CHECKSUM_PARTIAL)) {
1241                 struct ixgb_buffer *buffer_info;
1242                 css = skb_transport_offset(skb);
1243                 cso = css + skb->csum_offset;
1244
1245                 i = adapter->tx_ring.next_to_use;
1246                 context_desc = IXGB_CONTEXT_DESC(adapter->tx_ring, i);
1247                 buffer_info = &adapter->tx_ring.buffer_info[i];
1248                 WARN_ON(buffer_info->dma != 0);
1249
1250                 context_desc->tucss = css;
1251                 context_desc->tucso = cso;
1252                 context_desc->tucse = 0;
1253                 /* zero out any previously existing data in one instruction */
1254                 *(uint32_t *)&(context_desc->ipcss) = 0;
1255                 context_desc->status = 0;
1256                 context_desc->hdr_len = 0;
1257                 context_desc->mss = 0;
1258                 context_desc->cmd_type_len =
1259                         cpu_to_le32(IXGB_CONTEXT_DESC_TYPE
1260                                     | IXGB_TX_DESC_CMD_IDE);
1261
1262                 if(++i == adapter->tx_ring.count) i = 0;
1263                 adapter->tx_ring.next_to_use = i;
1264
1265                 return TRUE;
1266         }
1267
1268         return FALSE;
1269 }
1270
1271 #define IXGB_MAX_TXD_PWR        14
1272 #define IXGB_MAX_DATA_PER_TXD   (1<<IXGB_MAX_TXD_PWR)
1273
1274 static int
1275 ixgb_tx_map(struct ixgb_adapter *adapter, struct sk_buff *skb,
1276             unsigned int first)
1277 {
1278         struct ixgb_desc_ring *tx_ring = &adapter->tx_ring;
1279         struct ixgb_buffer *buffer_info;
1280         int len = skb->len;
1281         unsigned int offset = 0, size, count = 0, i;
1282         unsigned int mss = skb_shinfo(skb)->gso_size;
1283
1284         unsigned int nr_frags = skb_shinfo(skb)->nr_frags;
1285         unsigned int f;
1286
1287         len -= skb->data_len;
1288
1289         i = tx_ring->next_to_use;
1290
1291         while(len) {
1292                 buffer_info = &tx_ring->buffer_info[i];
1293                 size = min(len, IXGB_MAX_DATA_PER_TXD);
1294                 /* Workaround for premature desc write-backs
1295                  * in TSO mode.  Append 4-byte sentinel desc */
1296                 if (unlikely(mss && !nr_frags && size == len && size > 8))
1297                         size -= 4;
1298
1299                 buffer_info->length = size;
1300                 WARN_ON(buffer_info->dma != 0);
1301                 buffer_info->dma =
1302                         pci_map_single(adapter->pdev,
1303                                 skb->data + offset,
1304                                 size,
1305                                 PCI_DMA_TODEVICE);
1306                 buffer_info->time_stamp = jiffies;
1307                 buffer_info->next_to_watch = 0;
1308
1309                 len -= size;
1310                 offset += size;
1311                 count++;
1312                 if(++i == tx_ring->count) i = 0;
1313         }
1314
1315         for(f = 0; f < nr_frags; f++) {
1316                 struct skb_frag_struct *frag;
1317
1318                 frag = &skb_shinfo(skb)->frags[f];
1319                 len = frag->size;
1320                 offset = 0;
1321
1322                 while(len) {
1323                         buffer_info = &tx_ring->buffer_info[i];
1324                         size = min(len, IXGB_MAX_DATA_PER_TXD);
1325
1326                         /* Workaround for premature desc write-backs
1327                          * in TSO mode.  Append 4-byte sentinel desc */
1328                         if (unlikely(mss && !nr_frags && size == len
1329                                      && size > 8))
1330                                 size -= 4;
1331
1332                         buffer_info->length = size;
1333                         buffer_info->dma =
1334                                 pci_map_page(adapter->pdev,
1335                                         frag->page,
1336                                         frag->page_offset + offset,
1337                                         size,
1338                                         PCI_DMA_TODEVICE);
1339                         buffer_info->time_stamp = jiffies;
1340                         buffer_info->next_to_watch = 0;
1341
1342                         len -= size;
1343                         offset += size;
1344                         count++;
1345                         if(++i == tx_ring->count) i = 0;
1346                 }
1347         }
1348         i = (i == 0) ? tx_ring->count - 1 : i - 1;
1349         tx_ring->buffer_info[i].skb = skb;
1350         tx_ring->buffer_info[first].next_to_watch = i;
1351
1352         return count;
1353 }
1354
1355 static void
1356 ixgb_tx_queue(struct ixgb_adapter *adapter, int count, int vlan_id,int tx_flags)
1357 {
1358         struct ixgb_desc_ring *tx_ring = &adapter->tx_ring;
1359         struct ixgb_tx_desc *tx_desc = NULL;
1360         struct ixgb_buffer *buffer_info;
1361         uint32_t cmd_type_len = adapter->tx_cmd_type;
1362         uint8_t status = 0;
1363         uint8_t popts = 0;
1364         unsigned int i;
1365
1366         if(tx_flags & IXGB_TX_FLAGS_TSO) {
1367                 cmd_type_len |= IXGB_TX_DESC_CMD_TSE;
1368                 popts |= (IXGB_TX_DESC_POPTS_IXSM | IXGB_TX_DESC_POPTS_TXSM);
1369         }
1370
1371         if(tx_flags & IXGB_TX_FLAGS_CSUM)
1372                 popts |= IXGB_TX_DESC_POPTS_TXSM;
1373
1374         if(tx_flags & IXGB_TX_FLAGS_VLAN) {
1375                 cmd_type_len |= IXGB_TX_DESC_CMD_VLE;
1376         }
1377
1378         i = tx_ring->next_to_use;
1379
1380         while(count--) {
1381                 buffer_info = &tx_ring->buffer_info[i];
1382                 tx_desc = IXGB_TX_DESC(*tx_ring, i);
1383                 tx_desc->buff_addr = cpu_to_le64(buffer_info->dma);
1384                 tx_desc->cmd_type_len =
1385                         cpu_to_le32(cmd_type_len | buffer_info->length);
1386                 tx_desc->status = status;
1387                 tx_desc->popts = popts;
1388                 tx_desc->vlan = cpu_to_le16(vlan_id);
1389
1390                 if(++i == tx_ring->count) i = 0;
1391         }
1392
1393         tx_desc->cmd_type_len |= cpu_to_le32(IXGB_TX_DESC_CMD_EOP 
1394                                 | IXGB_TX_DESC_CMD_RS );
1395
1396         /* Force memory writes to complete before letting h/w
1397          * know there are new descriptors to fetch.  (Only
1398          * applicable for weak-ordered memory model archs,
1399          * such as IA-64). */
1400         wmb();
1401
1402         tx_ring->next_to_use = i;
1403         IXGB_WRITE_REG(&adapter->hw, TDT, i);
1404 }
1405
1406 static int __ixgb_maybe_stop_tx(struct net_device *netdev, int size)
1407 {
1408         struct ixgb_adapter *adapter = netdev_priv(netdev);
1409         struct ixgb_desc_ring *tx_ring = &adapter->tx_ring;
1410
1411         netif_stop_queue(netdev);
1412         /* Herbert's original patch had:
1413          *  smp_mb__after_netif_stop_queue();
1414          * but since that doesn't exist yet, just open code it. */
1415         smp_mb();
1416
1417         /* We need to check again in a case another CPU has just
1418          * made room available. */
1419         if (likely(IXGB_DESC_UNUSED(tx_ring) < size))
1420                 return -EBUSY;
1421
1422         /* A reprieve! */
1423         netif_start_queue(netdev);
1424         ++adapter->restart_queue;
1425         return 0;
1426 }
1427
1428 static int ixgb_maybe_stop_tx(struct net_device *netdev,
1429                               struct ixgb_desc_ring *tx_ring, int size)
1430 {
1431         if (likely(IXGB_DESC_UNUSED(tx_ring) >= size))
1432                 return 0;
1433         return __ixgb_maybe_stop_tx(netdev, size);
1434 }
1435
1436
1437 /* Tx Descriptors needed, worst case */
1438 #define TXD_USE_COUNT(S) (((S) >> IXGB_MAX_TXD_PWR) + \
1439                          (((S) & (IXGB_MAX_DATA_PER_TXD - 1)) ? 1 : 0))
1440 #define DESC_NEEDED TXD_USE_COUNT(IXGB_MAX_DATA_PER_TXD) /* skb->date */ + \
1441         MAX_SKB_FRAGS * TXD_USE_COUNT(PAGE_SIZE) + 1 /* for context */ \
1442         + 1 /* one more needed for sentinel TSO workaround */
1443
1444 static int
1445 ixgb_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
1446 {
1447         struct ixgb_adapter *adapter = netdev_priv(netdev);
1448         unsigned int first;
1449         unsigned int tx_flags = 0;
1450         unsigned long flags;
1451         int vlan_id = 0;
1452         int tso;
1453
1454         if(skb->len <= 0) {
1455                 dev_kfree_skb_any(skb);
1456                 return 0;
1457         }
1458
1459 #ifdef NETIF_F_LLTX
1460         local_irq_save(flags);
1461         if (!spin_trylock(&adapter->tx_lock)) {
1462                 /* Collision - tell upper layer to requeue */
1463                 local_irq_restore(flags);
1464                 return NETDEV_TX_LOCKED;
1465         }
1466 #else
1467         spin_lock_irqsave(&adapter->tx_lock, flags);
1468 #endif
1469
1470         if (unlikely(ixgb_maybe_stop_tx(netdev, &adapter->tx_ring,
1471                      DESC_NEEDED))) {
1472                 netif_stop_queue(netdev);
1473                 spin_unlock_irqrestore(&adapter->tx_lock, flags);
1474                 return NETDEV_TX_BUSY;
1475         }
1476
1477 #ifndef NETIF_F_LLTX
1478         spin_unlock_irqrestore(&adapter->tx_lock, flags);
1479 #endif
1480
1481         if(adapter->vlgrp && vlan_tx_tag_present(skb)) {
1482                 tx_flags |= IXGB_TX_FLAGS_VLAN;
1483                 vlan_id = vlan_tx_tag_get(skb);
1484         }
1485
1486         first = adapter->tx_ring.next_to_use;
1487         
1488         tso = ixgb_tso(adapter, skb);
1489         if (tso < 0) {
1490                 dev_kfree_skb_any(skb);
1491 #ifdef NETIF_F_LLTX
1492                 spin_unlock_irqrestore(&adapter->tx_lock, flags);
1493 #endif
1494                 return NETDEV_TX_OK;
1495         }
1496
1497         if (likely(tso))
1498                 tx_flags |= IXGB_TX_FLAGS_TSO;
1499         else if(ixgb_tx_csum(adapter, skb))
1500                 tx_flags |= IXGB_TX_FLAGS_CSUM;
1501
1502         ixgb_tx_queue(adapter, ixgb_tx_map(adapter, skb, first), vlan_id,
1503                         tx_flags);
1504
1505         netdev->trans_start = jiffies;
1506
1507 #ifdef NETIF_F_LLTX
1508         /* Make sure there is space in the ring for the next send. */
1509         ixgb_maybe_stop_tx(netdev, &adapter->tx_ring, DESC_NEEDED);
1510
1511         spin_unlock_irqrestore(&adapter->tx_lock, flags);
1512
1513 #endif
1514         return NETDEV_TX_OK;
1515 }
1516
1517 /**
1518  * ixgb_tx_timeout - Respond to a Tx Hang
1519  * @netdev: network interface device structure
1520  **/
1521
1522 static void
1523 ixgb_tx_timeout(struct net_device *netdev)
1524 {
1525         struct ixgb_adapter *adapter = netdev_priv(netdev);
1526
1527         /* Do the reset outside of interrupt context */
1528         schedule_work(&adapter->tx_timeout_task);
1529 }
1530
1531 static void
1532 ixgb_tx_timeout_task(struct work_struct *work)
1533 {
1534         struct ixgb_adapter *adapter =
1535                 container_of(work, struct ixgb_adapter, tx_timeout_task);
1536
1537         adapter->tx_timeout_count++;
1538         ixgb_down(adapter, TRUE);
1539         ixgb_up(adapter);
1540 }
1541
1542 /**
1543  * ixgb_get_stats - Get System Network Statistics
1544  * @netdev: network interface device structure
1545  *
1546  * Returns the address of the device statistics structure.
1547  * The statistics are actually updated from the timer callback.
1548  **/
1549
1550 static struct net_device_stats *
1551 ixgb_get_stats(struct net_device *netdev)
1552 {
1553         struct ixgb_adapter *adapter = netdev_priv(netdev);
1554
1555         return &adapter->net_stats;
1556 }
1557
1558 /**
1559  * ixgb_change_mtu - Change the Maximum Transfer Unit
1560  * @netdev: network interface device structure
1561  * @new_mtu: new value for maximum frame size
1562  *
1563  * Returns 0 on success, negative on failure
1564  **/
1565
1566 static int
1567 ixgb_change_mtu(struct net_device *netdev, int new_mtu)
1568 {
1569         struct ixgb_adapter *adapter = netdev_priv(netdev);
1570         int max_frame = new_mtu + ENET_HEADER_SIZE + ENET_FCS_LENGTH;
1571         int old_max_frame = netdev->mtu + ENET_HEADER_SIZE + ENET_FCS_LENGTH;
1572
1573
1574         if((max_frame < IXGB_MIN_ENET_FRAME_SIZE_WITHOUT_FCS + ENET_FCS_LENGTH)
1575            || (max_frame > IXGB_MAX_JUMBO_FRAME_SIZE + ENET_FCS_LENGTH)) {
1576                 DPRINTK(PROBE, ERR, "Invalid MTU setting %d\n", new_mtu);
1577                 return -EINVAL;
1578         }
1579
1580         adapter->rx_buffer_len = max_frame;
1581
1582         netdev->mtu = new_mtu;
1583
1584         if ((old_max_frame != max_frame) && netif_running(netdev)) {
1585                 ixgb_down(adapter, TRUE);
1586                 ixgb_up(adapter);
1587         }
1588
1589         return 0;
1590 }
1591
1592 /**
1593  * ixgb_update_stats - Update the board statistics counters.
1594  * @adapter: board private structure
1595  **/
1596
1597 void
1598 ixgb_update_stats(struct ixgb_adapter *adapter)
1599 {
1600         struct net_device *netdev = adapter->netdev;
1601         struct pci_dev *pdev = adapter->pdev;
1602
1603         /* Prevent stats update while adapter is being reset */
1604         if (pci_channel_offline(pdev))
1605                 return;
1606
1607         if((netdev->flags & IFF_PROMISC) || (netdev->flags & IFF_ALLMULTI) ||
1608            (netdev->mc_count > IXGB_MAX_NUM_MULTICAST_ADDRESSES)) {
1609                 u64 multi = IXGB_READ_REG(&adapter->hw, MPRCL);
1610                 u32 bcast_l = IXGB_READ_REG(&adapter->hw, BPRCL);
1611                 u32 bcast_h = IXGB_READ_REG(&adapter->hw, BPRCH);
1612                 u64 bcast = ((u64)bcast_h << 32) | bcast_l; 
1613
1614                 multi |= ((u64)IXGB_READ_REG(&adapter->hw, MPRCH) << 32);
1615                 /* fix up multicast stats by removing broadcasts */
1616                 if(multi >= bcast)
1617                         multi -= bcast;
1618                 
1619                 adapter->stats.mprcl += (multi & 0xFFFFFFFF);
1620                 adapter->stats.mprch += (multi >> 32);
1621                 adapter->stats.bprcl += bcast_l; 
1622                 adapter->stats.bprch += bcast_h;
1623         } else {
1624                 adapter->stats.mprcl += IXGB_READ_REG(&adapter->hw, MPRCL);
1625                 adapter->stats.mprch += IXGB_READ_REG(&adapter->hw, MPRCH);
1626                 adapter->stats.bprcl += IXGB_READ_REG(&adapter->hw, BPRCL);
1627                 adapter->stats.bprch += IXGB_READ_REG(&adapter->hw, BPRCH);
1628         }
1629         adapter->stats.tprl += IXGB_READ_REG(&adapter->hw, TPRL);
1630         adapter->stats.tprh += IXGB_READ_REG(&adapter->hw, TPRH);
1631         adapter->stats.gprcl += IXGB_READ_REG(&adapter->hw, GPRCL);
1632         adapter->stats.gprch += IXGB_READ_REG(&adapter->hw, GPRCH);
1633         adapter->stats.uprcl += IXGB_READ_REG(&adapter->hw, UPRCL);
1634         adapter->stats.uprch += IXGB_READ_REG(&adapter->hw, UPRCH);
1635         adapter->stats.vprcl += IXGB_READ_REG(&adapter->hw, VPRCL);
1636         adapter->stats.vprch += IXGB_READ_REG(&adapter->hw, VPRCH);
1637         adapter->stats.jprcl += IXGB_READ_REG(&adapter->hw, JPRCL);
1638         adapter->stats.jprch += IXGB_READ_REG(&adapter->hw, JPRCH);
1639         adapter->stats.gorcl += IXGB_READ_REG(&adapter->hw, GORCL);
1640         adapter->stats.gorch += IXGB_READ_REG(&adapter->hw, GORCH);
1641         adapter->stats.torl += IXGB_READ_REG(&adapter->hw, TORL);
1642         adapter->stats.torh += IXGB_READ_REG(&adapter->hw, TORH);
1643         adapter->stats.rnbc += IXGB_READ_REG(&adapter->hw, RNBC);
1644         adapter->stats.ruc += IXGB_READ_REG(&adapter->hw, RUC);
1645         adapter->stats.roc += IXGB_READ_REG(&adapter->hw, ROC);
1646         adapter->stats.rlec += IXGB_READ_REG(&adapter->hw, RLEC);
1647         adapter->stats.crcerrs += IXGB_READ_REG(&adapter->hw, CRCERRS);
1648         adapter->stats.icbc += IXGB_READ_REG(&adapter->hw, ICBC);
1649         adapter->stats.ecbc += IXGB_READ_REG(&adapter->hw, ECBC);
1650         adapter->stats.mpc += IXGB_READ_REG(&adapter->hw, MPC);
1651         adapter->stats.tptl += IXGB_READ_REG(&adapter->hw, TPTL);
1652         adapter->stats.tpth += IXGB_READ_REG(&adapter->hw, TPTH);
1653         adapter->stats.gptcl += IXGB_READ_REG(&adapter->hw, GPTCL);
1654         adapter->stats.gptch += IXGB_READ_REG(&adapter->hw, GPTCH);
1655         adapter->stats.bptcl += IXGB_READ_REG(&adapter->hw, BPTCL);
1656         adapter->stats.bptch += IXGB_READ_REG(&adapter->hw, BPTCH);
1657         adapter->stats.mptcl += IXGB_READ_REG(&adapter->hw, MPTCL);
1658         adapter->stats.mptch += IXGB_READ_REG(&adapter->hw, MPTCH);
1659         adapter->stats.uptcl += IXGB_READ_REG(&adapter->hw, UPTCL);
1660         adapter->stats.uptch += IXGB_READ_REG(&adapter->hw, UPTCH);
1661         adapter->stats.vptcl += IXGB_READ_REG(&adapter->hw, VPTCL);
1662         adapter->stats.vptch += IXGB_READ_REG(&adapter->hw, VPTCH);
1663         adapter->stats.jptcl += IXGB_READ_REG(&adapter->hw, JPTCL);
1664         adapter->stats.jptch += IXGB_READ_REG(&adapter->hw, JPTCH);
1665         adapter->stats.gotcl += IXGB_READ_REG(&adapter->hw, GOTCL);
1666         adapter->stats.gotch += IXGB_READ_REG(&adapter->hw, GOTCH);
1667         adapter->stats.totl += IXGB_READ_REG(&adapter->hw, TOTL);
1668         adapter->stats.toth += IXGB_READ_REG(&adapter->hw, TOTH);
1669         adapter->stats.dc += IXGB_READ_REG(&adapter->hw, DC);
1670         adapter->stats.plt64c += IXGB_READ_REG(&adapter->hw, PLT64C);
1671         adapter->stats.tsctc += IXGB_READ_REG(&adapter->hw, TSCTC);
1672         adapter->stats.tsctfc += IXGB_READ_REG(&adapter->hw, TSCTFC);
1673         adapter->stats.ibic += IXGB_READ_REG(&adapter->hw, IBIC);
1674         adapter->stats.rfc += IXGB_READ_REG(&adapter->hw, RFC);
1675         adapter->stats.lfc += IXGB_READ_REG(&adapter->hw, LFC);
1676         adapter->stats.pfrc += IXGB_READ_REG(&adapter->hw, PFRC);
1677         adapter->stats.pftc += IXGB_READ_REG(&adapter->hw, PFTC);
1678         adapter->stats.mcfrc += IXGB_READ_REG(&adapter->hw, MCFRC);
1679         adapter->stats.mcftc += IXGB_READ_REG(&adapter->hw, MCFTC);
1680         adapter->stats.xonrxc += IXGB_READ_REG(&adapter->hw, XONRXC);
1681         adapter->stats.xontxc += IXGB_READ_REG(&adapter->hw, XONTXC);
1682         adapter->stats.xoffrxc += IXGB_READ_REG(&adapter->hw, XOFFRXC);
1683         adapter->stats.xofftxc += IXGB_READ_REG(&adapter->hw, XOFFTXC);
1684         adapter->stats.rjc += IXGB_READ_REG(&adapter->hw, RJC);
1685
1686         /* Fill out the OS statistics structure */
1687
1688         adapter->net_stats.rx_packets = adapter->stats.gprcl;
1689         adapter->net_stats.tx_packets = adapter->stats.gptcl;
1690         adapter->net_stats.rx_bytes = adapter->stats.gorcl;
1691         adapter->net_stats.tx_bytes = adapter->stats.gotcl;
1692         adapter->net_stats.multicast = adapter->stats.mprcl;
1693         adapter->net_stats.collisions = 0;
1694
1695         /* ignore RLEC as it reports errors for padded (<64bytes) frames
1696          * with a length in the type/len field */
1697         adapter->net_stats.rx_errors =
1698             /* adapter->stats.rnbc + */ adapter->stats.crcerrs +
1699             adapter->stats.ruc +
1700             adapter->stats.roc /*+ adapter->stats.rlec */  +
1701             adapter->stats.icbc +
1702             adapter->stats.ecbc + adapter->stats.mpc;
1703
1704         /* see above
1705          * adapter->net_stats.rx_length_errors = adapter->stats.rlec;
1706          */
1707
1708         adapter->net_stats.rx_crc_errors = adapter->stats.crcerrs;
1709         adapter->net_stats.rx_fifo_errors = adapter->stats.mpc;
1710         adapter->net_stats.rx_missed_errors = adapter->stats.mpc;
1711         adapter->net_stats.rx_over_errors = adapter->stats.mpc;
1712
1713         adapter->net_stats.tx_errors = 0;
1714         adapter->net_stats.rx_frame_errors = 0;
1715         adapter->net_stats.tx_aborted_errors = 0;
1716         adapter->net_stats.tx_carrier_errors = 0;
1717         adapter->net_stats.tx_fifo_errors = 0;
1718         adapter->net_stats.tx_heartbeat_errors = 0;
1719         adapter->net_stats.tx_window_errors = 0;
1720 }
1721
1722 #define IXGB_MAX_INTR 10
1723 /**
1724  * ixgb_intr - Interrupt Handler
1725  * @irq: interrupt number
1726  * @data: pointer to a network interface device structure
1727  **/
1728
1729 static irqreturn_t
1730 ixgb_intr(int irq, void *data)
1731 {
1732         struct net_device *netdev = data;
1733         struct ixgb_adapter *adapter = netdev_priv(netdev);
1734         struct ixgb_hw *hw = &adapter->hw;
1735         uint32_t icr = IXGB_READ_REG(hw, ICR);
1736 #ifndef CONFIG_IXGB_NAPI
1737         unsigned int i;
1738 #endif
1739
1740         if(unlikely(!icr))
1741                 return IRQ_NONE;  /* Not our interrupt */
1742
1743         if(unlikely(icr & (IXGB_INT_RXSEQ | IXGB_INT_LSC))) {
1744                 mod_timer(&adapter->watchdog_timer, jiffies);
1745         }
1746
1747 #ifdef CONFIG_IXGB_NAPI
1748         if (netif_rx_schedule_prep(netdev, &adapter->napi)) {
1749
1750                 /* Disable interrupts and register for poll. The flush 
1751                   of the posted write is intentionally left out.
1752                 */
1753
1754                 atomic_inc(&adapter->irq_sem);
1755                 IXGB_WRITE_REG(&adapter->hw, IMC, ~0);
1756                 __netif_rx_schedule(netdev, &adapter->napi);
1757         }
1758 #else
1759         /* yes, that is actually a & and it is meant to make sure that
1760          * every pass through this for loop checks both receive and
1761          * transmit queues for completed descriptors, intended to
1762          * avoid starvation issues and assist tx/rx fairness. */
1763         for(i = 0; i < IXGB_MAX_INTR; i++)
1764                 if(!ixgb_clean_rx_irq(adapter) &
1765                    !ixgb_clean_tx_irq(adapter))
1766                         break;
1767 #endif 
1768         return IRQ_HANDLED;
1769 }
1770
1771 #ifdef CONFIG_IXGB_NAPI
1772 /**
1773  * ixgb_clean - NAPI Rx polling callback
1774  * @adapter: board private structure
1775  **/
1776
1777 static int
1778 ixgb_clean(struct napi_struct *napi, int budget)
1779 {
1780         struct ixgb_adapter *adapter = container_of(napi, struct ixgb_adapter, napi);
1781         struct net_device *netdev = adapter->netdev;
1782         int tx_cleaned;
1783         int work_done = 0;
1784
1785         tx_cleaned = ixgb_clean_tx_irq(adapter);
1786         ixgb_clean_rx_irq(adapter, &work_done, budget);
1787
1788         /* if no Tx and not enough Rx work done, exit the polling mode */
1789         if((!tx_cleaned && (work_done == 0)) || !netif_running(netdev)) {
1790                 netif_rx_complete(netdev, napi);
1791                 ixgb_irq_enable(adapter);
1792         }
1793
1794         return work_done;
1795 }
1796 #endif
1797
1798 /**
1799  * ixgb_clean_tx_irq - Reclaim resources after transmit completes
1800  * @adapter: board private structure
1801  **/
1802
1803 static boolean_t
1804 ixgb_clean_tx_irq(struct ixgb_adapter *adapter)
1805 {
1806         struct ixgb_desc_ring *tx_ring = &adapter->tx_ring;
1807         struct net_device *netdev = adapter->netdev;
1808         struct ixgb_tx_desc *tx_desc, *eop_desc;
1809         struct ixgb_buffer *buffer_info;
1810         unsigned int i, eop;
1811         boolean_t cleaned = FALSE;
1812
1813         i = tx_ring->next_to_clean;
1814         eop = tx_ring->buffer_info[i].next_to_watch;
1815         eop_desc = IXGB_TX_DESC(*tx_ring, eop);
1816
1817         while(eop_desc->status & IXGB_TX_DESC_STATUS_DD) {
1818
1819                 for(cleaned = FALSE; !cleaned; ) {
1820                         tx_desc = IXGB_TX_DESC(*tx_ring, i);
1821                         buffer_info = &tx_ring->buffer_info[i];
1822
1823                         if (tx_desc->popts
1824                             & (IXGB_TX_DESC_POPTS_TXSM |
1825                                IXGB_TX_DESC_POPTS_IXSM))
1826                                 adapter->hw_csum_tx_good++;
1827
1828                         ixgb_unmap_and_free_tx_resource(adapter, buffer_info);
1829
1830                         *(uint32_t *)&(tx_desc->status) = 0;
1831
1832                         cleaned = (i == eop);
1833                         if(++i == tx_ring->count) i = 0;
1834                 }
1835
1836                 eop = tx_ring->buffer_info[i].next_to_watch;
1837                 eop_desc = IXGB_TX_DESC(*tx_ring, eop);
1838         }
1839
1840         tx_ring->next_to_clean = i;
1841
1842         if (unlikely(netif_queue_stopped(netdev))) {
1843                 spin_lock(&adapter->tx_lock);
1844                 if (netif_queue_stopped(netdev) && netif_carrier_ok(netdev) &&
1845                     (IXGB_DESC_UNUSED(tx_ring) >= DESC_NEEDED))
1846                         netif_wake_queue(netdev);
1847                 spin_unlock(&adapter->tx_lock);
1848         }
1849
1850         if(adapter->detect_tx_hung) {
1851                 /* detect a transmit hang in hardware, this serializes the
1852                  * check with the clearing of time_stamp and movement of i */
1853                 adapter->detect_tx_hung = FALSE;
1854                 if (tx_ring->buffer_info[eop].dma &&
1855                    time_after(jiffies, tx_ring->buffer_info[eop].time_stamp + HZ)
1856                    && !(IXGB_READ_REG(&adapter->hw, STATUS) &
1857                         IXGB_STATUS_TXOFF)) {
1858                         /* detected Tx unit hang */
1859                         DPRINTK(DRV, ERR, "Detected Tx Unit Hang\n"
1860                                         "  TDH                  <%x>\n"
1861                                         "  TDT                  <%x>\n"
1862                                         "  next_to_use          <%x>\n"
1863                                         "  next_to_clean        <%x>\n"
1864                                         "buffer_info[next_to_clean]\n"
1865                                         "  time_stamp           <%lx>\n"
1866                                         "  next_to_watch        <%x>\n"
1867                                         "  jiffies              <%lx>\n"
1868                                         "  next_to_watch.status <%x>\n",
1869                                 IXGB_READ_REG(&adapter->hw, TDH),
1870                                 IXGB_READ_REG(&adapter->hw, TDT),
1871                                 tx_ring->next_to_use,
1872                                 tx_ring->next_to_clean,
1873                                 tx_ring->buffer_info[eop].time_stamp,
1874                                 eop,
1875                                 jiffies,
1876                                 eop_desc->status);
1877                         netif_stop_queue(netdev);
1878                 }
1879         }
1880
1881         return cleaned;
1882 }
1883
1884 /**
1885  * ixgb_rx_checksum - Receive Checksum Offload for 82597.
1886  * @adapter: board private structure
1887  * @rx_desc: receive descriptor
1888  * @sk_buff: socket buffer with received data
1889  **/
1890
1891 static void
1892 ixgb_rx_checksum(struct ixgb_adapter *adapter,
1893                  struct ixgb_rx_desc *rx_desc,
1894                  struct sk_buff *skb)
1895 {
1896         /* Ignore Checksum bit is set OR
1897          * TCP Checksum has not been calculated
1898          */
1899         if((rx_desc->status & IXGB_RX_DESC_STATUS_IXSM) ||
1900            (!(rx_desc->status & IXGB_RX_DESC_STATUS_TCPCS))) {
1901                 skb->ip_summed = CHECKSUM_NONE;
1902                 return;
1903         }
1904
1905         /* At this point we know the hardware did the TCP checksum */
1906         /* now look at the TCP checksum error bit */
1907         if(rx_desc->errors & IXGB_RX_DESC_ERRORS_TCPE) {
1908                 /* let the stack verify checksum errors */
1909                 skb->ip_summed = CHECKSUM_NONE;
1910                 adapter->hw_csum_rx_error++;
1911         } else {
1912                 /* TCP checksum is good */
1913                 skb->ip_summed = CHECKSUM_UNNECESSARY;
1914                 adapter->hw_csum_rx_good++;
1915         }
1916 }
1917
1918 /**
1919  * ixgb_clean_rx_irq - Send received data up the network stack,
1920  * @adapter: board private structure
1921  **/
1922
1923 static boolean_t
1924 #ifdef CONFIG_IXGB_NAPI
1925 ixgb_clean_rx_irq(struct ixgb_adapter *adapter, int *work_done, int work_to_do)
1926 #else
1927 ixgb_clean_rx_irq(struct ixgb_adapter *adapter)
1928 #endif
1929 {
1930         struct ixgb_desc_ring *rx_ring = &adapter->rx_ring;
1931         struct net_device *netdev = adapter->netdev;
1932         struct pci_dev *pdev = adapter->pdev;
1933         struct ixgb_rx_desc *rx_desc, *next_rxd;
1934         struct ixgb_buffer *buffer_info, *next_buffer, *next2_buffer;
1935         uint32_t length;
1936         unsigned int i, j;
1937         boolean_t cleaned = FALSE;
1938
1939         i = rx_ring->next_to_clean;
1940         rx_desc = IXGB_RX_DESC(*rx_ring, i);
1941         buffer_info = &rx_ring->buffer_info[i];
1942
1943         while(rx_desc->status & IXGB_RX_DESC_STATUS_DD) {
1944                 struct sk_buff *skb, *next_skb;
1945                 u8 status;
1946
1947 #ifdef CONFIG_IXGB_NAPI
1948                 if(*work_done >= work_to_do)
1949                         break;
1950
1951                 (*work_done)++;
1952 #endif
1953                 status = rx_desc->status;
1954                 skb = buffer_info->skb;
1955                 buffer_info->skb = NULL;
1956
1957                 prefetch(skb->data);
1958
1959                 if(++i == rx_ring->count) i = 0;
1960                 next_rxd = IXGB_RX_DESC(*rx_ring, i);
1961                 prefetch(next_rxd);
1962
1963                 if((j = i + 1) == rx_ring->count) j = 0;
1964                 next2_buffer = &rx_ring->buffer_info[j];
1965                 prefetch(next2_buffer);
1966
1967                 next_buffer = &rx_ring->buffer_info[i];
1968                 next_skb = next_buffer->skb;
1969                 prefetch(next_skb);
1970
1971                 cleaned = TRUE;
1972
1973                 pci_unmap_single(pdev,
1974                                  buffer_info->dma,
1975                                  buffer_info->length,
1976                                  PCI_DMA_FROMDEVICE);
1977
1978                 length = le16_to_cpu(rx_desc->length);
1979
1980                 if(unlikely(!(status & IXGB_RX_DESC_STATUS_EOP))) {
1981
1982                         /* All receives must fit into a single buffer */
1983
1984                         IXGB_DBG("Receive packet consumed multiple buffers "
1985                                          "length<%x>\n", length);
1986
1987                         dev_kfree_skb_irq(skb);
1988                         goto rxdesc_done;
1989                 }
1990
1991                 if (unlikely(rx_desc->errors
1992                              & (IXGB_RX_DESC_ERRORS_CE | IXGB_RX_DESC_ERRORS_SE
1993                                 | IXGB_RX_DESC_ERRORS_P |
1994                                 IXGB_RX_DESC_ERRORS_RXE))) {
1995
1996                         dev_kfree_skb_irq(skb);
1997                         goto rxdesc_done;
1998                 }
1999
2000                 /* code added for copybreak, this should improve
2001                  * performance for small packets with large amounts
2002                  * of reassembly being done in the stack */
2003 #define IXGB_CB_LENGTH 256
2004                 if (length < IXGB_CB_LENGTH) {
2005                         struct sk_buff *new_skb =
2006                             netdev_alloc_skb(netdev, length + NET_IP_ALIGN);
2007                         if (new_skb) {
2008                                 skb_reserve(new_skb, NET_IP_ALIGN);
2009                                 skb_copy_to_linear_data_offset(new_skb,
2010                                                                -NET_IP_ALIGN,
2011                                                                (skb->data -
2012                                                                 NET_IP_ALIGN),
2013                                                                (length +
2014                                                                 NET_IP_ALIGN));
2015                                 /* save the skb in buffer_info as good */
2016                                 buffer_info->skb = skb;
2017                                 skb = new_skb;
2018                         }
2019                 }
2020                 /* end copybreak code */
2021
2022                 /* Good Receive */
2023                 skb_put(skb, length);
2024
2025                 /* Receive Checksum Offload */
2026                 ixgb_rx_checksum(adapter, rx_desc, skb);
2027
2028                 skb->protocol = eth_type_trans(skb, netdev);
2029 #ifdef CONFIG_IXGB_NAPI
2030                 if(adapter->vlgrp && (status & IXGB_RX_DESC_STATUS_VP)) {
2031                         vlan_hwaccel_receive_skb(skb, adapter->vlgrp,
2032                                 le16_to_cpu(rx_desc->special) &
2033                                         IXGB_RX_DESC_SPECIAL_VLAN_MASK);
2034                 } else {
2035                         netif_receive_skb(skb);
2036                 }
2037 #else /* CONFIG_IXGB_NAPI */
2038                 if(adapter->vlgrp && (status & IXGB_RX_DESC_STATUS_VP)) {
2039                         vlan_hwaccel_rx(skb, adapter->vlgrp,
2040                                 le16_to_cpu(rx_desc->special) &
2041                                         IXGB_RX_DESC_SPECIAL_VLAN_MASK);
2042                 } else {
2043                         netif_rx(skb);
2044                 }
2045 #endif /* CONFIG_IXGB_NAPI */
2046                 netdev->last_rx = jiffies;
2047
2048 rxdesc_done:
2049                 /* clean up descriptor, might be written over by hw */
2050                 rx_desc->status = 0;
2051
2052                 /* use prefetched values */
2053                 rx_desc = next_rxd;
2054                 buffer_info = next_buffer;
2055         }
2056
2057         rx_ring->next_to_clean = i;
2058
2059         ixgb_alloc_rx_buffers(adapter);
2060
2061         return cleaned;
2062 }
2063
2064 /**
2065  * ixgb_alloc_rx_buffers - Replace used receive buffers
2066  * @adapter: address of board private structure
2067  **/
2068
2069 static void
2070 ixgb_alloc_rx_buffers(struct ixgb_adapter *adapter)
2071 {
2072         struct ixgb_desc_ring *rx_ring = &adapter->rx_ring;
2073         struct net_device *netdev = adapter->netdev;
2074         struct pci_dev *pdev = adapter->pdev;
2075         struct ixgb_rx_desc *rx_desc;
2076         struct ixgb_buffer *buffer_info;
2077         struct sk_buff *skb;
2078         unsigned int i;
2079         int num_group_tail_writes;
2080         long cleancount;
2081
2082         i = rx_ring->next_to_use;
2083         buffer_info = &rx_ring->buffer_info[i];
2084         cleancount = IXGB_DESC_UNUSED(rx_ring);
2085
2086         num_group_tail_writes = IXGB_RX_BUFFER_WRITE;
2087
2088         /* leave three descriptors unused */
2089         while(--cleancount > 2) {
2090                 /* recycle! its good for you */
2091                 skb = buffer_info->skb;
2092                 if (skb) {
2093                         skb_trim(skb, 0);
2094                         goto map_skb;
2095                 }
2096
2097                 skb = netdev_alloc_skb(netdev, adapter->rx_buffer_len
2098                                        + NET_IP_ALIGN);
2099                 if (unlikely(!skb)) {
2100                         /* Better luck next round */
2101                         adapter->alloc_rx_buff_failed++;
2102                         break;
2103                 }
2104
2105                 /* Make buffer alignment 2 beyond a 16 byte boundary
2106                  * this will result in a 16 byte aligned IP header after
2107                  * the 14 byte MAC header is removed
2108                  */
2109                 skb_reserve(skb, NET_IP_ALIGN);
2110
2111                 buffer_info->skb = skb;
2112                 buffer_info->length = adapter->rx_buffer_len;
2113 map_skb:
2114                 buffer_info->dma = pci_map_single(pdev,
2115                                                   skb->data,
2116                                                   adapter->rx_buffer_len,
2117                                                   PCI_DMA_FROMDEVICE);
2118
2119                 rx_desc = IXGB_RX_DESC(*rx_ring, i);
2120                 rx_desc->buff_addr = cpu_to_le64(buffer_info->dma);
2121                 /* guarantee DD bit not set now before h/w gets descriptor
2122                  * this is the rest of the workaround for h/w double 
2123                  * writeback. */
2124                 rx_desc->status = 0;
2125
2126
2127                 if(++i == rx_ring->count) i = 0;
2128                 buffer_info = &rx_ring->buffer_info[i];
2129         }
2130
2131         if (likely(rx_ring->next_to_use != i)) {
2132                 rx_ring->next_to_use = i;
2133                 if (unlikely(i-- == 0))
2134                         i = (rx_ring->count - 1);
2135
2136                 /* Force memory writes to complete before letting h/w
2137                  * know there are new descriptors to fetch.  (Only
2138                  * applicable for weak-ordered memory model archs, such
2139                  * as IA-64). */
2140                 wmb();
2141                 IXGB_WRITE_REG(&adapter->hw, RDT, i);
2142         }
2143 }
2144
2145 /**
2146  * ixgb_vlan_rx_register - enables or disables vlan tagging/stripping.
2147  * 
2148  * @param netdev network interface device structure
2149  * @param grp indicates to enable or disable tagging/stripping
2150  **/
2151 static void
2152 ixgb_vlan_rx_register(struct net_device *netdev, struct vlan_group *grp)
2153 {
2154         struct ixgb_adapter *adapter = netdev_priv(netdev);
2155         uint32_t ctrl, rctl;
2156
2157         ixgb_irq_disable(adapter);
2158         adapter->vlgrp = grp;
2159
2160         if(grp) {
2161                 /* enable VLAN tag insert/strip */
2162                 ctrl = IXGB_READ_REG(&adapter->hw, CTRL0);
2163                 ctrl |= IXGB_CTRL0_VME;
2164                 IXGB_WRITE_REG(&adapter->hw, CTRL0, ctrl);
2165
2166                 /* enable VLAN receive filtering */
2167
2168                 rctl = IXGB_READ_REG(&adapter->hw, RCTL);
2169                 rctl |= IXGB_RCTL_VFE;
2170                 rctl &= ~IXGB_RCTL_CFIEN;
2171                 IXGB_WRITE_REG(&adapter->hw, RCTL, rctl);
2172         } else {
2173                 /* disable VLAN tag insert/strip */
2174
2175                 ctrl = IXGB_READ_REG(&adapter->hw, CTRL0);
2176                 ctrl &= ~IXGB_CTRL0_VME;
2177                 IXGB_WRITE_REG(&adapter->hw, CTRL0, ctrl);
2178
2179                 /* disable VLAN filtering */
2180
2181                 rctl = IXGB_READ_REG(&adapter->hw, RCTL);
2182                 rctl &= ~IXGB_RCTL_VFE;
2183                 IXGB_WRITE_REG(&adapter->hw, RCTL, rctl);
2184         }
2185
2186         ixgb_irq_enable(adapter);
2187 }
2188
2189 static void
2190 ixgb_vlan_rx_add_vid(struct net_device *netdev, uint16_t vid)
2191 {
2192         struct ixgb_adapter *adapter = netdev_priv(netdev);
2193         uint32_t vfta, index;
2194
2195         /* add VID to filter table */
2196
2197         index = (vid >> 5) & 0x7F;
2198         vfta = IXGB_READ_REG_ARRAY(&adapter->hw, VFTA, index);
2199         vfta |= (1 << (vid & 0x1F));
2200         ixgb_write_vfta(&adapter->hw, index, vfta);
2201 }
2202
2203 static void
2204 ixgb_vlan_rx_kill_vid(struct net_device *netdev, uint16_t vid)
2205 {
2206         struct ixgb_adapter *adapter = netdev_priv(netdev);
2207         uint32_t vfta, index;
2208
2209         ixgb_irq_disable(adapter);
2210
2211         vlan_group_set_device(adapter->vlgrp, vid, NULL);
2212
2213         ixgb_irq_enable(adapter);
2214
2215         /* remove VID from filter table*/
2216
2217         index = (vid >> 5) & 0x7F;
2218         vfta = IXGB_READ_REG_ARRAY(&adapter->hw, VFTA, index);
2219         vfta &= ~(1 << (vid & 0x1F));
2220         ixgb_write_vfta(&adapter->hw, index, vfta);
2221 }
2222
2223 static void
2224 ixgb_restore_vlan(struct ixgb_adapter *adapter)
2225 {
2226         ixgb_vlan_rx_register(adapter->netdev, adapter->vlgrp);
2227
2228         if(adapter->vlgrp) {
2229                 uint16_t vid;
2230                 for(vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
2231                         if(!vlan_group_get_device(adapter->vlgrp, vid))
2232                                 continue;
2233                         ixgb_vlan_rx_add_vid(adapter->netdev, vid);
2234                 }
2235         }
2236 }
2237
2238 #ifdef CONFIG_NET_POLL_CONTROLLER
2239 /*
2240  * Polling 'interrupt' - used by things like netconsole to send skbs
2241  * without having to re-enable interrupts. It's not called while
2242  * the interrupt routine is executing.
2243  */
2244
2245 static void ixgb_netpoll(struct net_device *dev)
2246 {
2247         struct ixgb_adapter *adapter = netdev_priv(dev);
2248
2249         disable_irq(adapter->pdev->irq);
2250         ixgb_intr(adapter->pdev->irq, dev);
2251         enable_irq(adapter->pdev->irq);
2252 }
2253 #endif
2254
2255 /**
2256  * ixgb_io_error_detected() - called when PCI error is detected
2257  * @pdev    pointer to pci device with error
2258  * @state   pci channel state after error
2259  *
2260  * This callback is called by the PCI subsystem whenever
2261  * a PCI bus error is detected.
2262  */
2263 static pci_ers_result_t ixgb_io_error_detected (struct pci_dev *pdev,
2264                                      enum pci_channel_state state)
2265 {
2266         struct net_device *netdev = pci_get_drvdata(pdev);
2267         struct ixgb_adapter *adapter = netdev_priv(netdev);
2268
2269         if(netif_running(netdev))
2270                 ixgb_down(adapter, TRUE);
2271
2272         pci_disable_device(pdev);
2273
2274         /* Request a slot reset. */
2275         return PCI_ERS_RESULT_NEED_RESET;
2276 }
2277
2278 /**
2279  * ixgb_io_slot_reset - called after the pci bus has been reset.
2280  * @pdev    pointer to pci device with error
2281  *
2282  * This callback is called after the PCI buss has been reset.
2283  * Basically, this tries to restart the card from scratch.
2284  * This is a shortened version of the device probe/discovery code,
2285  * it resembles the first-half of the ixgb_probe() routine.
2286  */
2287 static pci_ers_result_t ixgb_io_slot_reset (struct pci_dev *pdev)
2288 {
2289         struct net_device *netdev = pci_get_drvdata(pdev);
2290         struct ixgb_adapter *adapter = netdev_priv(netdev);
2291
2292         if(pci_enable_device(pdev)) {
2293                 DPRINTK(PROBE, ERR, "Cannot re-enable PCI device after reset.\n");
2294                 return PCI_ERS_RESULT_DISCONNECT;
2295         }
2296
2297         /* Perform card reset only on one instance of the card */
2298         if (0 != PCI_FUNC (pdev->devfn))
2299                 return PCI_ERS_RESULT_RECOVERED;
2300
2301         pci_set_master(pdev);
2302
2303         netif_carrier_off(netdev);
2304         netif_stop_queue(netdev);
2305         ixgb_reset(adapter);
2306
2307         /* Make sure the EEPROM is good */
2308         if(!ixgb_validate_eeprom_checksum(&adapter->hw)) {
2309                 DPRINTK(PROBE, ERR, "After reset, the EEPROM checksum is not valid.\n");
2310                 return PCI_ERS_RESULT_DISCONNECT;
2311         }
2312         ixgb_get_ee_mac_addr(&adapter->hw, netdev->dev_addr);
2313         memcpy(netdev->perm_addr, netdev->dev_addr, netdev->addr_len);
2314
2315         if(!is_valid_ether_addr(netdev->perm_addr)) {
2316                 DPRINTK(PROBE, ERR, "After reset, invalid MAC address.\n");
2317                 return PCI_ERS_RESULT_DISCONNECT;
2318         }
2319
2320         return PCI_ERS_RESULT_RECOVERED;
2321 }
2322
2323 /**
2324  * ixgb_io_resume - called when its OK to resume normal operations
2325  * @pdev    pointer to pci device with error
2326  *
2327  * The error recovery driver tells us that its OK to resume
2328  * normal operation. Implementation resembles the second-half
2329  * of the ixgb_probe() routine.
2330  */
2331 static void ixgb_io_resume (struct pci_dev *pdev)
2332 {
2333         struct net_device *netdev = pci_get_drvdata(pdev);
2334         struct ixgb_adapter *adapter = netdev_priv(netdev);
2335
2336         pci_set_master(pdev);
2337
2338         if(netif_running(netdev)) {
2339                 if(ixgb_up(adapter)) {
2340                         printk ("ixgb: can't bring device back up after reset\n");
2341                         return;
2342                 }
2343         }
2344
2345         netif_device_attach(netdev);
2346         mod_timer(&adapter->watchdog_timer, jiffies);
2347 }
2348
2349 /* ixgb_main.c */