GNU Linux-libre 4.19.295-gnu1
[releases.git] / drivers / net / ethernet / intel / e1000 / e1000_main.c
1 // SPDX-License-Identifier: GPL-2.0
2 /* Copyright(c) 1999 - 2006 Intel Corporation. */
3
4 #include "e1000.h"
5 #include <net/ip6_checksum.h>
6 #include <linux/io.h>
7 #include <linux/prefetch.h>
8 #include <linux/bitops.h>
9 #include <linux/if_vlan.h>
10
11 char e1000_driver_name[] = "e1000";
12 static char e1000_driver_string[] = "Intel(R) PRO/1000 Network Driver";
13 #define DRV_VERSION "7.3.21-k8-NAPI"
14 const char e1000_driver_version[] = DRV_VERSION;
15 static const char e1000_copyright[] = "Copyright (c) 1999-2006 Intel Corporation.";
16
17 /* e1000_pci_tbl - PCI Device ID Table
18  *
19  * Last entry must be all 0s
20  *
21  * Macro expands to...
22  *   {PCI_DEVICE(PCI_VENDOR_ID_INTEL, device_id)}
23  */
24 static const struct pci_device_id e1000_pci_tbl[] = {
25         INTEL_E1000_ETHERNET_DEVICE(0x1000),
26         INTEL_E1000_ETHERNET_DEVICE(0x1001),
27         INTEL_E1000_ETHERNET_DEVICE(0x1004),
28         INTEL_E1000_ETHERNET_DEVICE(0x1008),
29         INTEL_E1000_ETHERNET_DEVICE(0x1009),
30         INTEL_E1000_ETHERNET_DEVICE(0x100C),
31         INTEL_E1000_ETHERNET_DEVICE(0x100D),
32         INTEL_E1000_ETHERNET_DEVICE(0x100E),
33         INTEL_E1000_ETHERNET_DEVICE(0x100F),
34         INTEL_E1000_ETHERNET_DEVICE(0x1010),
35         INTEL_E1000_ETHERNET_DEVICE(0x1011),
36         INTEL_E1000_ETHERNET_DEVICE(0x1012),
37         INTEL_E1000_ETHERNET_DEVICE(0x1013),
38         INTEL_E1000_ETHERNET_DEVICE(0x1014),
39         INTEL_E1000_ETHERNET_DEVICE(0x1015),
40         INTEL_E1000_ETHERNET_DEVICE(0x1016),
41         INTEL_E1000_ETHERNET_DEVICE(0x1017),
42         INTEL_E1000_ETHERNET_DEVICE(0x1018),
43         INTEL_E1000_ETHERNET_DEVICE(0x1019),
44         INTEL_E1000_ETHERNET_DEVICE(0x101A),
45         INTEL_E1000_ETHERNET_DEVICE(0x101D),
46         INTEL_E1000_ETHERNET_DEVICE(0x101E),
47         INTEL_E1000_ETHERNET_DEVICE(0x1026),
48         INTEL_E1000_ETHERNET_DEVICE(0x1027),
49         INTEL_E1000_ETHERNET_DEVICE(0x1028),
50         INTEL_E1000_ETHERNET_DEVICE(0x1075),
51         INTEL_E1000_ETHERNET_DEVICE(0x1076),
52         INTEL_E1000_ETHERNET_DEVICE(0x1077),
53         INTEL_E1000_ETHERNET_DEVICE(0x1078),
54         INTEL_E1000_ETHERNET_DEVICE(0x1079),
55         INTEL_E1000_ETHERNET_DEVICE(0x107A),
56         INTEL_E1000_ETHERNET_DEVICE(0x107B),
57         INTEL_E1000_ETHERNET_DEVICE(0x107C),
58         INTEL_E1000_ETHERNET_DEVICE(0x108A),
59         INTEL_E1000_ETHERNET_DEVICE(0x1099),
60         INTEL_E1000_ETHERNET_DEVICE(0x10B5),
61         INTEL_E1000_ETHERNET_DEVICE(0x2E6E),
62         /* required last entry */
63         {0,}
64 };
65
66 MODULE_DEVICE_TABLE(pci, e1000_pci_tbl);
67
68 int e1000_up(struct e1000_adapter *adapter);
69 void e1000_down(struct e1000_adapter *adapter);
70 void e1000_reinit_locked(struct e1000_adapter *adapter);
71 void e1000_reset(struct e1000_adapter *adapter);
72 int e1000_setup_all_tx_resources(struct e1000_adapter *adapter);
73 int e1000_setup_all_rx_resources(struct e1000_adapter *adapter);
74 void e1000_free_all_tx_resources(struct e1000_adapter *adapter);
75 void e1000_free_all_rx_resources(struct e1000_adapter *adapter);
76 static int e1000_setup_tx_resources(struct e1000_adapter *adapter,
77                                     struct e1000_tx_ring *txdr);
78 static int e1000_setup_rx_resources(struct e1000_adapter *adapter,
79                                     struct e1000_rx_ring *rxdr);
80 static void e1000_free_tx_resources(struct e1000_adapter *adapter,
81                                     struct e1000_tx_ring *tx_ring);
82 static void e1000_free_rx_resources(struct e1000_adapter *adapter,
83                                     struct e1000_rx_ring *rx_ring);
84 void e1000_update_stats(struct e1000_adapter *adapter);
85
86 static int e1000_init_module(void);
87 static void e1000_exit_module(void);
88 static int e1000_probe(struct pci_dev *pdev, const struct pci_device_id *ent);
89 static void e1000_remove(struct pci_dev *pdev);
90 static int e1000_alloc_queues(struct e1000_adapter *adapter);
91 static int e1000_sw_init(struct e1000_adapter *adapter);
92 int e1000_open(struct net_device *netdev);
93 int e1000_close(struct net_device *netdev);
94 static void e1000_configure_tx(struct e1000_adapter *adapter);
95 static void e1000_configure_rx(struct e1000_adapter *adapter);
96 static void e1000_setup_rctl(struct e1000_adapter *adapter);
97 static void e1000_clean_all_tx_rings(struct e1000_adapter *adapter);
98 static void e1000_clean_all_rx_rings(struct e1000_adapter *adapter);
99 static void e1000_clean_tx_ring(struct e1000_adapter *adapter,
100                                 struct e1000_tx_ring *tx_ring);
101 static void e1000_clean_rx_ring(struct e1000_adapter *adapter,
102                                 struct e1000_rx_ring *rx_ring);
103 static void e1000_set_rx_mode(struct net_device *netdev);
104 static void e1000_update_phy_info_task(struct work_struct *work);
105 static void e1000_watchdog(struct work_struct *work);
106 static void e1000_82547_tx_fifo_stall_task(struct work_struct *work);
107 static netdev_tx_t e1000_xmit_frame(struct sk_buff *skb,
108                                     struct net_device *netdev);
109 static int e1000_change_mtu(struct net_device *netdev, int new_mtu);
110 static int e1000_set_mac(struct net_device *netdev, void *p);
111 static irqreturn_t e1000_intr(int irq, void *data);
112 static bool e1000_clean_tx_irq(struct e1000_adapter *adapter,
113                                struct e1000_tx_ring *tx_ring);
114 static int e1000_clean(struct napi_struct *napi, int budget);
115 static bool e1000_clean_rx_irq(struct e1000_adapter *adapter,
116                                struct e1000_rx_ring *rx_ring,
117                                int *work_done, int work_to_do);
118 static bool e1000_clean_jumbo_rx_irq(struct e1000_adapter *adapter,
119                                      struct e1000_rx_ring *rx_ring,
120                                      int *work_done, int work_to_do);
121 static void e1000_alloc_dummy_rx_buffers(struct e1000_adapter *adapter,
122                                          struct e1000_rx_ring *rx_ring,
123                                          int cleaned_count)
124 {
125 }
126 static void e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
127                                    struct e1000_rx_ring *rx_ring,
128                                    int cleaned_count);
129 static void e1000_alloc_jumbo_rx_buffers(struct e1000_adapter *adapter,
130                                          struct e1000_rx_ring *rx_ring,
131                                          int cleaned_count);
132 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd);
133 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
134                            int cmd);
135 static void e1000_enter_82542_rst(struct e1000_adapter *adapter);
136 static void e1000_leave_82542_rst(struct e1000_adapter *adapter);
137 static void e1000_tx_timeout(struct net_device *dev);
138 static void e1000_reset_task(struct work_struct *work);
139 static void e1000_smartspeed(struct e1000_adapter *adapter);
140 static int e1000_82547_fifo_workaround(struct e1000_adapter *adapter,
141                                        struct sk_buff *skb);
142
143 static bool e1000_vlan_used(struct e1000_adapter *adapter);
144 static void e1000_vlan_mode(struct net_device *netdev,
145                             netdev_features_t features);
146 static void e1000_vlan_filter_on_off(struct e1000_adapter *adapter,
147                                      bool filter_on);
148 static int e1000_vlan_rx_add_vid(struct net_device *netdev,
149                                  __be16 proto, u16 vid);
150 static int e1000_vlan_rx_kill_vid(struct net_device *netdev,
151                                   __be16 proto, u16 vid);
152 static void e1000_restore_vlan(struct e1000_adapter *adapter);
153
154 #ifdef CONFIG_PM
155 static int e1000_suspend(struct pci_dev *pdev, pm_message_t state);
156 static int e1000_resume(struct pci_dev *pdev);
157 #endif
158 static void e1000_shutdown(struct pci_dev *pdev);
159
160 #ifdef CONFIG_NET_POLL_CONTROLLER
161 /* for netdump / net console */
162 static void e1000_netpoll (struct net_device *netdev);
163 #endif
164
165 #define COPYBREAK_DEFAULT 256
166 static unsigned int copybreak __read_mostly = COPYBREAK_DEFAULT;
167 module_param(copybreak, uint, 0644);
168 MODULE_PARM_DESC(copybreak,
169         "Maximum size of packet that is copied to a new buffer on receive");
170
171 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
172                                                 pci_channel_state_t state);
173 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev);
174 static void e1000_io_resume(struct pci_dev *pdev);
175
176 static const struct pci_error_handlers e1000_err_handler = {
177         .error_detected = e1000_io_error_detected,
178         .slot_reset = e1000_io_slot_reset,
179         .resume = e1000_io_resume,
180 };
181
182 static struct pci_driver e1000_driver = {
183         .name     = e1000_driver_name,
184         .id_table = e1000_pci_tbl,
185         .probe    = e1000_probe,
186         .remove   = e1000_remove,
187 #ifdef CONFIG_PM
188         /* Power Management Hooks */
189         .suspend  = e1000_suspend,
190         .resume   = e1000_resume,
191 #endif
192         .shutdown = e1000_shutdown,
193         .err_handler = &e1000_err_handler
194 };
195
196 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
197 MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
198 MODULE_LICENSE("GPL");
199 MODULE_VERSION(DRV_VERSION);
200
201 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
202 static int debug = -1;
203 module_param(debug, int, 0);
204 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
205
206 /**
207  * e1000_get_hw_dev - return device
208  * used by hardware layer to print debugging information
209  *
210  **/
211 struct net_device *e1000_get_hw_dev(struct e1000_hw *hw)
212 {
213         struct e1000_adapter *adapter = hw->back;
214         return adapter->netdev;
215 }
216
217 /**
218  * e1000_init_module - Driver Registration Routine
219  *
220  * e1000_init_module is the first routine called when the driver is
221  * loaded. All it does is register with the PCI subsystem.
222  **/
223 static int __init e1000_init_module(void)
224 {
225         int ret;
226         pr_info("%s - version %s\n", e1000_driver_string, e1000_driver_version);
227
228         pr_info("%s\n", e1000_copyright);
229
230         ret = pci_register_driver(&e1000_driver);
231         if (copybreak != COPYBREAK_DEFAULT) {
232                 if (copybreak == 0)
233                         pr_info("copybreak disabled\n");
234                 else
235                         pr_info("copybreak enabled for "
236                                    "packets <= %u bytes\n", copybreak);
237         }
238         return ret;
239 }
240
241 module_init(e1000_init_module);
242
243 /**
244  * e1000_exit_module - Driver Exit Cleanup Routine
245  *
246  * e1000_exit_module is called just before the driver is removed
247  * from memory.
248  **/
249 static void __exit e1000_exit_module(void)
250 {
251         pci_unregister_driver(&e1000_driver);
252 }
253
254 module_exit(e1000_exit_module);
255
256 static int e1000_request_irq(struct e1000_adapter *adapter)
257 {
258         struct net_device *netdev = adapter->netdev;
259         irq_handler_t handler = e1000_intr;
260         int irq_flags = IRQF_SHARED;
261         int err;
262
263         err = request_irq(adapter->pdev->irq, handler, irq_flags, netdev->name,
264                           netdev);
265         if (err) {
266                 e_err(probe, "Unable to allocate interrupt Error: %d\n", err);
267         }
268
269         return err;
270 }
271
272 static void e1000_free_irq(struct e1000_adapter *adapter)
273 {
274         struct net_device *netdev = adapter->netdev;
275
276         free_irq(adapter->pdev->irq, netdev);
277 }
278
279 /**
280  * e1000_irq_disable - Mask off interrupt generation on the NIC
281  * @adapter: board private structure
282  **/
283 static void e1000_irq_disable(struct e1000_adapter *adapter)
284 {
285         struct e1000_hw *hw = &adapter->hw;
286
287         ew32(IMC, ~0);
288         E1000_WRITE_FLUSH();
289         synchronize_irq(adapter->pdev->irq);
290 }
291
292 /**
293  * e1000_irq_enable - Enable default interrupt generation settings
294  * @adapter: board private structure
295  **/
296 static void e1000_irq_enable(struct e1000_adapter *adapter)
297 {
298         struct e1000_hw *hw = &adapter->hw;
299
300         ew32(IMS, IMS_ENABLE_MASK);
301         E1000_WRITE_FLUSH();
302 }
303
304 static void e1000_update_mng_vlan(struct e1000_adapter *adapter)
305 {
306         struct e1000_hw *hw = &adapter->hw;
307         struct net_device *netdev = adapter->netdev;
308         u16 vid = hw->mng_cookie.vlan_id;
309         u16 old_vid = adapter->mng_vlan_id;
310
311         if (!e1000_vlan_used(adapter))
312                 return;
313
314         if (!test_bit(vid, adapter->active_vlans)) {
315                 if (hw->mng_cookie.status &
316                     E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT) {
317                         e1000_vlan_rx_add_vid(netdev, htons(ETH_P_8021Q), vid);
318                         adapter->mng_vlan_id = vid;
319                 } else {
320                         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
321                 }
322                 if ((old_vid != (u16)E1000_MNG_VLAN_NONE) &&
323                     (vid != old_vid) &&
324                     !test_bit(old_vid, adapter->active_vlans))
325                         e1000_vlan_rx_kill_vid(netdev, htons(ETH_P_8021Q),
326                                                old_vid);
327         } else {
328                 adapter->mng_vlan_id = vid;
329         }
330 }
331
332 static void e1000_init_manageability(struct e1000_adapter *adapter)
333 {
334         struct e1000_hw *hw = &adapter->hw;
335
336         if (adapter->en_mng_pt) {
337                 u32 manc = er32(MANC);
338
339                 /* disable hardware interception of ARP */
340                 manc &= ~(E1000_MANC_ARP_EN);
341
342                 ew32(MANC, manc);
343         }
344 }
345
346 static void e1000_release_manageability(struct e1000_adapter *adapter)
347 {
348         struct e1000_hw *hw = &adapter->hw;
349
350         if (adapter->en_mng_pt) {
351                 u32 manc = er32(MANC);
352
353                 /* re-enable hardware interception of ARP */
354                 manc |= E1000_MANC_ARP_EN;
355
356                 ew32(MANC, manc);
357         }
358 }
359
360 /**
361  * e1000_configure - configure the hardware for RX and TX
362  * @adapter = private board structure
363  **/
364 static void e1000_configure(struct e1000_adapter *adapter)
365 {
366         struct net_device *netdev = adapter->netdev;
367         int i;
368
369         e1000_set_rx_mode(netdev);
370
371         e1000_restore_vlan(adapter);
372         e1000_init_manageability(adapter);
373
374         e1000_configure_tx(adapter);
375         e1000_setup_rctl(adapter);
376         e1000_configure_rx(adapter);
377         /* call E1000_DESC_UNUSED which always leaves
378          * at least 1 descriptor unused to make sure
379          * next_to_use != next_to_clean
380          */
381         for (i = 0; i < adapter->num_rx_queues; i++) {
382                 struct e1000_rx_ring *ring = &adapter->rx_ring[i];
383                 adapter->alloc_rx_buf(adapter, ring,
384                                       E1000_DESC_UNUSED(ring));
385         }
386 }
387
388 int e1000_up(struct e1000_adapter *adapter)
389 {
390         struct e1000_hw *hw = &adapter->hw;
391
392         /* hardware has been reset, we need to reload some things */
393         e1000_configure(adapter);
394
395         clear_bit(__E1000_DOWN, &adapter->flags);
396
397         napi_enable(&adapter->napi);
398
399         e1000_irq_enable(adapter);
400
401         netif_wake_queue(adapter->netdev);
402
403         /* fire a link change interrupt to start the watchdog */
404         ew32(ICS, E1000_ICS_LSC);
405         return 0;
406 }
407
408 /**
409  * e1000_power_up_phy - restore link in case the phy was powered down
410  * @adapter: address of board private structure
411  *
412  * The phy may be powered down to save power and turn off link when the
413  * driver is unloaded and wake on lan is not enabled (among others)
414  * *** this routine MUST be followed by a call to e1000_reset ***
415  **/
416 void e1000_power_up_phy(struct e1000_adapter *adapter)
417 {
418         struct e1000_hw *hw = &adapter->hw;
419         u16 mii_reg = 0;
420
421         /* Just clear the power down bit to wake the phy back up */
422         if (hw->media_type == e1000_media_type_copper) {
423                 /* according to the manual, the phy will retain its
424                  * settings across a power-down/up cycle
425                  */
426                 e1000_read_phy_reg(hw, PHY_CTRL, &mii_reg);
427                 mii_reg &= ~MII_CR_POWER_DOWN;
428                 e1000_write_phy_reg(hw, PHY_CTRL, mii_reg);
429         }
430 }
431
432 static void e1000_power_down_phy(struct e1000_adapter *adapter)
433 {
434         struct e1000_hw *hw = &adapter->hw;
435
436         /* Power down the PHY so no link is implied when interface is down *
437          * The PHY cannot be powered down if any of the following is true *
438          * (a) WoL is enabled
439          * (b) AMT is active
440          * (c) SoL/IDER session is active
441          */
442         if (!adapter->wol && hw->mac_type >= e1000_82540 &&
443            hw->media_type == e1000_media_type_copper) {
444                 u16 mii_reg = 0;
445
446                 switch (hw->mac_type) {
447                 case e1000_82540:
448                 case e1000_82545:
449                 case e1000_82545_rev_3:
450                 case e1000_82546:
451                 case e1000_ce4100:
452                 case e1000_82546_rev_3:
453                 case e1000_82541:
454                 case e1000_82541_rev_2:
455                 case e1000_82547:
456                 case e1000_82547_rev_2:
457                         if (er32(MANC) & E1000_MANC_SMBUS_EN)
458                                 goto out;
459                         break;
460                 default:
461                         goto out;
462                 }
463                 e1000_read_phy_reg(hw, PHY_CTRL, &mii_reg);
464                 mii_reg |= MII_CR_POWER_DOWN;
465                 e1000_write_phy_reg(hw, PHY_CTRL, mii_reg);
466                 msleep(1);
467         }
468 out:
469         return;
470 }
471
472 static void e1000_down_and_stop(struct e1000_adapter *adapter)
473 {
474         set_bit(__E1000_DOWN, &adapter->flags);
475
476         cancel_delayed_work_sync(&adapter->watchdog_task);
477
478         /*
479          * Since the watchdog task can reschedule other tasks, we should cancel
480          * it first, otherwise we can run into the situation when a work is
481          * still running after the adapter has been turned down.
482          */
483
484         cancel_delayed_work_sync(&adapter->phy_info_task);
485         cancel_delayed_work_sync(&adapter->fifo_stall_task);
486
487         /* Only kill reset task if adapter is not resetting */
488         if (!test_bit(__E1000_RESETTING, &adapter->flags))
489                 cancel_work_sync(&adapter->reset_task);
490 }
491
492 void e1000_down(struct e1000_adapter *adapter)
493 {
494         struct e1000_hw *hw = &adapter->hw;
495         struct net_device *netdev = adapter->netdev;
496         u32 rctl, tctl;
497
498         /* disable receives in the hardware */
499         rctl = er32(RCTL);
500         ew32(RCTL, rctl & ~E1000_RCTL_EN);
501         /* flush and sleep below */
502
503         netif_tx_disable(netdev);
504
505         /* disable transmits in the hardware */
506         tctl = er32(TCTL);
507         tctl &= ~E1000_TCTL_EN;
508         ew32(TCTL, tctl);
509         /* flush both disables and wait for them to finish */
510         E1000_WRITE_FLUSH();
511         msleep(10);
512
513         /* Set the carrier off after transmits have been disabled in the
514          * hardware, to avoid race conditions with e1000_watchdog() (which
515          * may be running concurrently to us, checking for the carrier
516          * bit to decide whether it should enable transmits again). Such
517          * a race condition would result into transmission being disabled
518          * in the hardware until the next IFF_DOWN+IFF_UP cycle.
519          */
520         netif_carrier_off(netdev);
521
522         napi_disable(&adapter->napi);
523
524         e1000_irq_disable(adapter);
525
526         /* Setting DOWN must be after irq_disable to prevent
527          * a screaming interrupt.  Setting DOWN also prevents
528          * tasks from rescheduling.
529          */
530         e1000_down_and_stop(adapter);
531
532         adapter->link_speed = 0;
533         adapter->link_duplex = 0;
534
535         e1000_reset(adapter);
536         e1000_clean_all_tx_rings(adapter);
537         e1000_clean_all_rx_rings(adapter);
538 }
539
540 void e1000_reinit_locked(struct e1000_adapter *adapter)
541 {
542         WARN_ON(in_interrupt());
543         while (test_and_set_bit(__E1000_RESETTING, &adapter->flags))
544                 msleep(1);
545
546         /* only run the task if not already down */
547         if (!test_bit(__E1000_DOWN, &adapter->flags)) {
548                 e1000_down(adapter);
549                 e1000_up(adapter);
550         }
551
552         clear_bit(__E1000_RESETTING, &adapter->flags);
553 }
554
555 void e1000_reset(struct e1000_adapter *adapter)
556 {
557         struct e1000_hw *hw = &adapter->hw;
558         u32 pba = 0, tx_space, min_tx_space, min_rx_space;
559         bool legacy_pba_adjust = false;
560         u16 hwm;
561
562         /* Repartition Pba for greater than 9k mtu
563          * To take effect CTRL.RST is required.
564          */
565
566         switch (hw->mac_type) {
567         case e1000_82542_rev2_0:
568         case e1000_82542_rev2_1:
569         case e1000_82543:
570         case e1000_82544:
571         case e1000_82540:
572         case e1000_82541:
573         case e1000_82541_rev_2:
574                 legacy_pba_adjust = true;
575                 pba = E1000_PBA_48K;
576                 break;
577         case e1000_82545:
578         case e1000_82545_rev_3:
579         case e1000_82546:
580         case e1000_ce4100:
581         case e1000_82546_rev_3:
582                 pba = E1000_PBA_48K;
583                 break;
584         case e1000_82547:
585         case e1000_82547_rev_2:
586                 legacy_pba_adjust = true;
587                 pba = E1000_PBA_30K;
588                 break;
589         case e1000_undefined:
590         case e1000_num_macs:
591                 break;
592         }
593
594         if (legacy_pba_adjust) {
595                 if (hw->max_frame_size > E1000_RXBUFFER_8192)
596                         pba -= 8; /* allocate more FIFO for Tx */
597
598                 if (hw->mac_type == e1000_82547) {
599                         adapter->tx_fifo_head = 0;
600                         adapter->tx_head_addr = pba << E1000_TX_HEAD_ADDR_SHIFT;
601                         adapter->tx_fifo_size =
602                                 (E1000_PBA_40K - pba) << E1000_PBA_BYTES_SHIFT;
603                         atomic_set(&adapter->tx_fifo_stall, 0);
604                 }
605         } else if (hw->max_frame_size >  ETH_FRAME_LEN + ETH_FCS_LEN) {
606                 /* adjust PBA for jumbo frames */
607                 ew32(PBA, pba);
608
609                 /* To maintain wire speed transmits, the Tx FIFO should be
610                  * large enough to accommodate two full transmit packets,
611                  * rounded up to the next 1KB and expressed in KB.  Likewise,
612                  * the Rx FIFO should be large enough to accommodate at least
613                  * one full receive packet and is similarly rounded up and
614                  * expressed in KB.
615                  */
616                 pba = er32(PBA);
617                 /* upper 16 bits has Tx packet buffer allocation size in KB */
618                 tx_space = pba >> 16;
619                 /* lower 16 bits has Rx packet buffer allocation size in KB */
620                 pba &= 0xffff;
621                 /* the Tx fifo also stores 16 bytes of information about the Tx
622                  * but don't include ethernet FCS because hardware appends it
623                  */
624                 min_tx_space = (hw->max_frame_size +
625                                 sizeof(struct e1000_tx_desc) -
626                                 ETH_FCS_LEN) * 2;
627                 min_tx_space = ALIGN(min_tx_space, 1024);
628                 min_tx_space >>= 10;
629                 /* software strips receive CRC, so leave room for it */
630                 min_rx_space = hw->max_frame_size;
631                 min_rx_space = ALIGN(min_rx_space, 1024);
632                 min_rx_space >>= 10;
633
634                 /* If current Tx allocation is less than the min Tx FIFO size,
635                  * and the min Tx FIFO size is less than the current Rx FIFO
636                  * allocation, take space away from current Rx allocation
637                  */
638                 if (tx_space < min_tx_space &&
639                     ((min_tx_space - tx_space) < pba)) {
640                         pba = pba - (min_tx_space - tx_space);
641
642                         /* PCI/PCIx hardware has PBA alignment constraints */
643                         switch (hw->mac_type) {
644                         case e1000_82545 ... e1000_82546_rev_3:
645                                 pba &= ~(E1000_PBA_8K - 1);
646                                 break;
647                         default:
648                                 break;
649                         }
650
651                         /* if short on Rx space, Rx wins and must trump Tx
652                          * adjustment or use Early Receive if available
653                          */
654                         if (pba < min_rx_space)
655                                 pba = min_rx_space;
656                 }
657         }
658
659         ew32(PBA, pba);
660
661         /* flow control settings:
662          * The high water mark must be low enough to fit one full frame
663          * (or the size used for early receive) above it in the Rx FIFO.
664          * Set it to the lower of:
665          * - 90% of the Rx FIFO size, and
666          * - the full Rx FIFO size minus the early receive size (for parts
667          *   with ERT support assuming ERT set to E1000_ERT_2048), or
668          * - the full Rx FIFO size minus one full frame
669          */
670         hwm = min(((pba << 10) * 9 / 10),
671                   ((pba << 10) - hw->max_frame_size));
672
673         hw->fc_high_water = hwm & 0xFFF8;       /* 8-byte granularity */
674         hw->fc_low_water = hw->fc_high_water - 8;
675         hw->fc_pause_time = E1000_FC_PAUSE_TIME;
676         hw->fc_send_xon = 1;
677         hw->fc = hw->original_fc;
678
679         /* Allow time for pending master requests to run */
680         e1000_reset_hw(hw);
681         if (hw->mac_type >= e1000_82544)
682                 ew32(WUC, 0);
683
684         if (e1000_init_hw(hw))
685                 e_dev_err("Hardware Error\n");
686         e1000_update_mng_vlan(adapter);
687
688         /* if (adapter->hwflags & HWFLAGS_PHY_PWR_BIT) { */
689         if (hw->mac_type >= e1000_82544 &&
690             hw->autoneg == 1 &&
691             hw->autoneg_advertised == ADVERTISE_1000_FULL) {
692                 u32 ctrl = er32(CTRL);
693                 /* clear phy power management bit if we are in gig only mode,
694                  * which if enabled will attempt negotiation to 100Mb, which
695                  * can cause a loss of link at power off or driver unload
696                  */
697                 ctrl &= ~E1000_CTRL_SWDPIN3;
698                 ew32(CTRL, ctrl);
699         }
700
701         /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
702         ew32(VET, ETHERNET_IEEE_VLAN_TYPE);
703
704         e1000_reset_adaptive(hw);
705         e1000_phy_get_info(hw, &adapter->phy_info);
706
707         e1000_release_manageability(adapter);
708 }
709
710 /* Dump the eeprom for users having checksum issues */
711 static void e1000_dump_eeprom(struct e1000_adapter *adapter)
712 {
713         struct net_device *netdev = adapter->netdev;
714         struct ethtool_eeprom eeprom;
715         const struct ethtool_ops *ops = netdev->ethtool_ops;
716         u8 *data;
717         int i;
718         u16 csum_old, csum_new = 0;
719
720         eeprom.len = ops->get_eeprom_len(netdev);
721         eeprom.offset = 0;
722
723         data = kmalloc(eeprom.len, GFP_KERNEL);
724         if (!data)
725                 return;
726
727         ops->get_eeprom(netdev, &eeprom, data);
728
729         csum_old = (data[EEPROM_CHECKSUM_REG * 2]) +
730                    (data[EEPROM_CHECKSUM_REG * 2 + 1] << 8);
731         for (i = 0; i < EEPROM_CHECKSUM_REG * 2; i += 2)
732                 csum_new += data[i] + (data[i + 1] << 8);
733         csum_new = EEPROM_SUM - csum_new;
734
735         pr_err("/*********************/\n");
736         pr_err("Current EEPROM Checksum : 0x%04x\n", csum_old);
737         pr_err("Calculated              : 0x%04x\n", csum_new);
738
739         pr_err("Offset    Values\n");
740         pr_err("========  ======\n");
741         print_hex_dump(KERN_ERR, "", DUMP_PREFIX_OFFSET, 16, 1, data, 128, 0);
742
743         pr_err("Include this output when contacting your support provider.\n");
744         pr_err("This is not a software error! Something bad happened to\n");
745         pr_err("your hardware or EEPROM image. Ignoring this problem could\n");
746         pr_err("result in further problems, possibly loss of data,\n");
747         pr_err("corruption or system hangs!\n");
748         pr_err("The MAC Address will be reset to 00:00:00:00:00:00,\n");
749         pr_err("which is invalid and requires you to set the proper MAC\n");
750         pr_err("address manually before continuing to enable this network\n");
751         pr_err("device. Please inspect the EEPROM dump and report the\n");
752         pr_err("issue to your hardware vendor or Intel Customer Support.\n");
753         pr_err("/*********************/\n");
754
755         kfree(data);
756 }
757
758 /**
759  * e1000_is_need_ioport - determine if an adapter needs ioport resources or not
760  * @pdev: PCI device information struct
761  *
762  * Return true if an adapter needs ioport resources
763  **/
764 static int e1000_is_need_ioport(struct pci_dev *pdev)
765 {
766         switch (pdev->device) {
767         case E1000_DEV_ID_82540EM:
768         case E1000_DEV_ID_82540EM_LOM:
769         case E1000_DEV_ID_82540EP:
770         case E1000_DEV_ID_82540EP_LOM:
771         case E1000_DEV_ID_82540EP_LP:
772         case E1000_DEV_ID_82541EI:
773         case E1000_DEV_ID_82541EI_MOBILE:
774         case E1000_DEV_ID_82541ER:
775         case E1000_DEV_ID_82541ER_LOM:
776         case E1000_DEV_ID_82541GI:
777         case E1000_DEV_ID_82541GI_LF:
778         case E1000_DEV_ID_82541GI_MOBILE:
779         case E1000_DEV_ID_82544EI_COPPER:
780         case E1000_DEV_ID_82544EI_FIBER:
781         case E1000_DEV_ID_82544GC_COPPER:
782         case E1000_DEV_ID_82544GC_LOM:
783         case E1000_DEV_ID_82545EM_COPPER:
784         case E1000_DEV_ID_82545EM_FIBER:
785         case E1000_DEV_ID_82546EB_COPPER:
786         case E1000_DEV_ID_82546EB_FIBER:
787         case E1000_DEV_ID_82546EB_QUAD_COPPER:
788                 return true;
789         default:
790                 return false;
791         }
792 }
793
794 static netdev_features_t e1000_fix_features(struct net_device *netdev,
795         netdev_features_t features)
796 {
797         /* Since there is no support for separate Rx/Tx vlan accel
798          * enable/disable make sure Tx flag is always in same state as Rx.
799          */
800         if (features & NETIF_F_HW_VLAN_CTAG_RX)
801                 features |= NETIF_F_HW_VLAN_CTAG_TX;
802         else
803                 features &= ~NETIF_F_HW_VLAN_CTAG_TX;
804
805         return features;
806 }
807
808 static int e1000_set_features(struct net_device *netdev,
809         netdev_features_t features)
810 {
811         struct e1000_adapter *adapter = netdev_priv(netdev);
812         netdev_features_t changed = features ^ netdev->features;
813
814         if (changed & NETIF_F_HW_VLAN_CTAG_RX)
815                 e1000_vlan_mode(netdev, features);
816
817         if (!(changed & (NETIF_F_RXCSUM | NETIF_F_RXALL)))
818                 return 0;
819
820         netdev->features = features;
821         adapter->rx_csum = !!(features & NETIF_F_RXCSUM);
822
823         if (netif_running(netdev))
824                 e1000_reinit_locked(adapter);
825         else
826                 e1000_reset(adapter);
827
828         return 0;
829 }
830
831 static const struct net_device_ops e1000_netdev_ops = {
832         .ndo_open               = e1000_open,
833         .ndo_stop               = e1000_close,
834         .ndo_start_xmit         = e1000_xmit_frame,
835         .ndo_set_rx_mode        = e1000_set_rx_mode,
836         .ndo_set_mac_address    = e1000_set_mac,
837         .ndo_tx_timeout         = e1000_tx_timeout,
838         .ndo_change_mtu         = e1000_change_mtu,
839         .ndo_do_ioctl           = e1000_ioctl,
840         .ndo_validate_addr      = eth_validate_addr,
841         .ndo_vlan_rx_add_vid    = e1000_vlan_rx_add_vid,
842         .ndo_vlan_rx_kill_vid   = e1000_vlan_rx_kill_vid,
843 #ifdef CONFIG_NET_POLL_CONTROLLER
844         .ndo_poll_controller    = e1000_netpoll,
845 #endif
846         .ndo_fix_features       = e1000_fix_features,
847         .ndo_set_features       = e1000_set_features,
848 };
849
850 /**
851  * e1000_init_hw_struct - initialize members of hw struct
852  * @adapter: board private struct
853  * @hw: structure used by e1000_hw.c
854  *
855  * Factors out initialization of the e1000_hw struct to its own function
856  * that can be called very early at init (just after struct allocation).
857  * Fields are initialized based on PCI device information and
858  * OS network device settings (MTU size).
859  * Returns negative error codes if MAC type setup fails.
860  */
861 static int e1000_init_hw_struct(struct e1000_adapter *adapter,
862                                 struct e1000_hw *hw)
863 {
864         struct pci_dev *pdev = adapter->pdev;
865
866         /* PCI config space info */
867         hw->vendor_id = pdev->vendor;
868         hw->device_id = pdev->device;
869         hw->subsystem_vendor_id = pdev->subsystem_vendor;
870         hw->subsystem_id = pdev->subsystem_device;
871         hw->revision_id = pdev->revision;
872
873         pci_read_config_word(pdev, PCI_COMMAND, &hw->pci_cmd_word);
874
875         hw->max_frame_size = adapter->netdev->mtu +
876                              ENET_HEADER_SIZE + ETHERNET_FCS_SIZE;
877         hw->min_frame_size = MINIMUM_ETHERNET_FRAME_SIZE;
878
879         /* identify the MAC */
880         if (e1000_set_mac_type(hw)) {
881                 e_err(probe, "Unknown MAC Type\n");
882                 return -EIO;
883         }
884
885         switch (hw->mac_type) {
886         default:
887                 break;
888         case e1000_82541:
889         case e1000_82547:
890         case e1000_82541_rev_2:
891         case e1000_82547_rev_2:
892                 hw->phy_init_script = 1;
893                 break;
894         }
895
896         e1000_set_media_type(hw);
897         e1000_get_bus_info(hw);
898
899         hw->wait_autoneg_complete = false;
900         hw->tbi_compatibility_en = true;
901         hw->adaptive_ifs = true;
902
903         /* Copper options */
904
905         if (hw->media_type == e1000_media_type_copper) {
906                 hw->mdix = AUTO_ALL_MODES;
907                 hw->disable_polarity_correction = false;
908                 hw->master_slave = E1000_MASTER_SLAVE;
909         }
910
911         return 0;
912 }
913
914 /**
915  * e1000_probe - Device Initialization Routine
916  * @pdev: PCI device information struct
917  * @ent: entry in e1000_pci_tbl
918  *
919  * Returns 0 on success, negative on failure
920  *
921  * e1000_probe initializes an adapter identified by a pci_dev structure.
922  * The OS initialization, configuring of the adapter private structure,
923  * and a hardware reset occur.
924  **/
925 static int e1000_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
926 {
927         struct net_device *netdev;
928         struct e1000_adapter *adapter = NULL;
929         struct e1000_hw *hw;
930
931         static int cards_found;
932         static int global_quad_port_a; /* global ksp3 port a indication */
933         int i, err, pci_using_dac;
934         u16 eeprom_data = 0;
935         u16 tmp = 0;
936         u16 eeprom_apme_mask = E1000_EEPROM_APME;
937         int bars, need_ioport;
938         bool disable_dev = false;
939
940         /* do not allocate ioport bars when not needed */
941         need_ioport = e1000_is_need_ioport(pdev);
942         if (need_ioport) {
943                 bars = pci_select_bars(pdev, IORESOURCE_MEM | IORESOURCE_IO);
944                 err = pci_enable_device(pdev);
945         } else {
946                 bars = pci_select_bars(pdev, IORESOURCE_MEM);
947                 err = pci_enable_device_mem(pdev);
948         }
949         if (err)
950                 return err;
951
952         err = pci_request_selected_regions(pdev, bars, e1000_driver_name);
953         if (err)
954                 goto err_pci_reg;
955
956         pci_set_master(pdev);
957         err = pci_save_state(pdev);
958         if (err)
959                 goto err_alloc_etherdev;
960
961         err = -ENOMEM;
962         netdev = alloc_etherdev(sizeof(struct e1000_adapter));
963         if (!netdev)
964                 goto err_alloc_etherdev;
965
966         SET_NETDEV_DEV(netdev, &pdev->dev);
967
968         pci_set_drvdata(pdev, netdev);
969         adapter = netdev_priv(netdev);
970         adapter->netdev = netdev;
971         adapter->pdev = pdev;
972         adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
973         adapter->bars = bars;
974         adapter->need_ioport = need_ioport;
975
976         hw = &adapter->hw;
977         hw->back = adapter;
978
979         err = -EIO;
980         hw->hw_addr = pci_ioremap_bar(pdev, BAR_0);
981         if (!hw->hw_addr)
982                 goto err_ioremap;
983
984         if (adapter->need_ioport) {
985                 for (i = BAR_1; i <= BAR_5; i++) {
986                         if (pci_resource_len(pdev, i) == 0)
987                                 continue;
988                         if (pci_resource_flags(pdev, i) & IORESOURCE_IO) {
989                                 hw->io_base = pci_resource_start(pdev, i);
990                                 break;
991                         }
992                 }
993         }
994
995         /* make ready for any if (hw->...) below */
996         err = e1000_init_hw_struct(adapter, hw);
997         if (err)
998                 goto err_sw_init;
999
1000         /* there is a workaround being applied below that limits
1001          * 64-bit DMA addresses to 64-bit hardware.  There are some
1002          * 32-bit adapters that Tx hang when given 64-bit DMA addresses
1003          */
1004         pci_using_dac = 0;
1005         if ((hw->bus_type == e1000_bus_type_pcix) &&
1006             !dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64))) {
1007                 pci_using_dac = 1;
1008         } else {
1009                 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
1010                 if (err) {
1011                         pr_err("No usable DMA config, aborting\n");
1012                         goto err_dma;
1013                 }
1014         }
1015
1016         netdev->netdev_ops = &e1000_netdev_ops;
1017         e1000_set_ethtool_ops(netdev);
1018         netdev->watchdog_timeo = 5 * HZ;
1019         netif_napi_add(netdev, &adapter->napi, e1000_clean, 64);
1020
1021         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
1022
1023         adapter->bd_number = cards_found;
1024
1025         /* setup the private structure */
1026
1027         err = e1000_sw_init(adapter);
1028         if (err)
1029                 goto err_sw_init;
1030
1031         err = -EIO;
1032         if (hw->mac_type == e1000_ce4100) {
1033                 hw->ce4100_gbe_mdio_base_virt =
1034                                         ioremap(pci_resource_start(pdev, BAR_1),
1035                                                 pci_resource_len(pdev, BAR_1));
1036
1037                 if (!hw->ce4100_gbe_mdio_base_virt)
1038                         goto err_mdio_ioremap;
1039         }
1040
1041         if (hw->mac_type >= e1000_82543) {
1042                 netdev->hw_features = NETIF_F_SG |
1043                                    NETIF_F_HW_CSUM |
1044                                    NETIF_F_HW_VLAN_CTAG_RX;
1045                 netdev->features = NETIF_F_HW_VLAN_CTAG_TX |
1046                                    NETIF_F_HW_VLAN_CTAG_FILTER;
1047         }
1048
1049         if ((hw->mac_type >= e1000_82544) &&
1050            (hw->mac_type != e1000_82547))
1051                 netdev->hw_features |= NETIF_F_TSO;
1052
1053         netdev->priv_flags |= IFF_SUPP_NOFCS;
1054
1055         netdev->features |= netdev->hw_features;
1056         netdev->hw_features |= (NETIF_F_RXCSUM |
1057                                 NETIF_F_RXALL |
1058                                 NETIF_F_RXFCS);
1059
1060         if (pci_using_dac) {
1061                 netdev->features |= NETIF_F_HIGHDMA;
1062                 netdev->vlan_features |= NETIF_F_HIGHDMA;
1063         }
1064
1065         netdev->vlan_features |= (NETIF_F_TSO |
1066                                   NETIF_F_HW_CSUM |
1067                                   NETIF_F_SG);
1068
1069         /* Do not set IFF_UNICAST_FLT for VMWare's 82545EM */
1070         if (hw->device_id != E1000_DEV_ID_82545EM_COPPER ||
1071             hw->subsystem_vendor_id != PCI_VENDOR_ID_VMWARE)
1072                 netdev->priv_flags |= IFF_UNICAST_FLT;
1073
1074         /* MTU range: 46 - 16110 */
1075         netdev->min_mtu = ETH_ZLEN - ETH_HLEN;
1076         netdev->max_mtu = MAX_JUMBO_FRAME_SIZE - (ETH_HLEN + ETH_FCS_LEN);
1077
1078         adapter->en_mng_pt = e1000_enable_mng_pass_thru(hw);
1079
1080         /* initialize eeprom parameters */
1081         if (e1000_init_eeprom_params(hw)) {
1082                 e_err(probe, "EEPROM initialization failed\n");
1083                 goto err_eeprom;
1084         }
1085
1086         /* before reading the EEPROM, reset the controller to
1087          * put the device in a known good starting state
1088          */
1089
1090         e1000_reset_hw(hw);
1091
1092         /* make sure the EEPROM is good */
1093         if (e1000_validate_eeprom_checksum(hw) < 0) {
1094                 e_err(probe, "The EEPROM Checksum Is Not Valid\n");
1095                 e1000_dump_eeprom(adapter);
1096                 /* set MAC address to all zeroes to invalidate and temporary
1097                  * disable this device for the user. This blocks regular
1098                  * traffic while still permitting ethtool ioctls from reaching
1099                  * the hardware as well as allowing the user to run the
1100                  * interface after manually setting a hw addr using
1101                  * `ip set address`
1102                  */
1103                 memset(hw->mac_addr, 0, netdev->addr_len);
1104         } else {
1105                 /* copy the MAC address out of the EEPROM */
1106                 if (e1000_read_mac_addr(hw))
1107                         e_err(probe, "EEPROM Read Error\n");
1108         }
1109         /* don't block initialization here due to bad MAC address */
1110         memcpy(netdev->dev_addr, hw->mac_addr, netdev->addr_len);
1111
1112         if (!is_valid_ether_addr(netdev->dev_addr))
1113                 e_err(probe, "Invalid MAC Address\n");
1114
1115
1116         INIT_DELAYED_WORK(&adapter->watchdog_task, e1000_watchdog);
1117         INIT_DELAYED_WORK(&adapter->fifo_stall_task,
1118                           e1000_82547_tx_fifo_stall_task);
1119         INIT_DELAYED_WORK(&adapter->phy_info_task, e1000_update_phy_info_task);
1120         INIT_WORK(&adapter->reset_task, e1000_reset_task);
1121
1122         e1000_check_options(adapter);
1123
1124         /* Initial Wake on LAN setting
1125          * If APM wake is enabled in the EEPROM,
1126          * enable the ACPI Magic Packet filter
1127          */
1128
1129         switch (hw->mac_type) {
1130         case e1000_82542_rev2_0:
1131         case e1000_82542_rev2_1:
1132         case e1000_82543:
1133                 break;
1134         case e1000_82544:
1135                 e1000_read_eeprom(hw,
1136                         EEPROM_INIT_CONTROL2_REG, 1, &eeprom_data);
1137                 eeprom_apme_mask = E1000_EEPROM_82544_APM;
1138                 break;
1139         case e1000_82546:
1140         case e1000_82546_rev_3:
1141                 if (er32(STATUS) & E1000_STATUS_FUNC_1) {
1142                         e1000_read_eeprom(hw,
1143                                 EEPROM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
1144                         break;
1145                 }
1146                 /* Fall Through */
1147         default:
1148                 e1000_read_eeprom(hw,
1149                         EEPROM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
1150                 break;
1151         }
1152         if (eeprom_data & eeprom_apme_mask)
1153                 adapter->eeprom_wol |= E1000_WUFC_MAG;
1154
1155         /* now that we have the eeprom settings, apply the special cases
1156          * where the eeprom may be wrong or the board simply won't support
1157          * wake on lan on a particular port
1158          */
1159         switch (pdev->device) {
1160         case E1000_DEV_ID_82546GB_PCIE:
1161                 adapter->eeprom_wol = 0;
1162                 break;
1163         case E1000_DEV_ID_82546EB_FIBER:
1164         case E1000_DEV_ID_82546GB_FIBER:
1165                 /* Wake events only supported on port A for dual fiber
1166                  * regardless of eeprom setting
1167                  */
1168                 if (er32(STATUS) & E1000_STATUS_FUNC_1)
1169                         adapter->eeprom_wol = 0;
1170                 break;
1171         case E1000_DEV_ID_82546GB_QUAD_COPPER_KSP3:
1172                 /* if quad port adapter, disable WoL on all but port A */
1173                 if (global_quad_port_a != 0)
1174                         adapter->eeprom_wol = 0;
1175                 else
1176                         adapter->quad_port_a = true;
1177                 /* Reset for multiple quad port adapters */
1178                 if (++global_quad_port_a == 4)
1179                         global_quad_port_a = 0;
1180                 break;
1181         }
1182
1183         /* initialize the wol settings based on the eeprom settings */
1184         adapter->wol = adapter->eeprom_wol;
1185         device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
1186
1187         /* Auto detect PHY address */
1188         if (hw->mac_type == e1000_ce4100) {
1189                 for (i = 0; i < 32; i++) {
1190                         hw->phy_addr = i;
1191                         e1000_read_phy_reg(hw, PHY_ID2, &tmp);
1192
1193                         if (tmp != 0 && tmp != 0xFF)
1194                                 break;
1195                 }
1196
1197                 if (i >= 32)
1198                         goto err_eeprom;
1199         }
1200
1201         /* reset the hardware with the new settings */
1202         e1000_reset(adapter);
1203
1204         strcpy(netdev->name, "eth%d");
1205         err = register_netdev(netdev);
1206         if (err)
1207                 goto err_register;
1208
1209         e1000_vlan_filter_on_off(adapter, false);
1210
1211         /* print bus type/speed/width info */
1212         e_info(probe, "(PCI%s:%dMHz:%d-bit) %pM\n",
1213                ((hw->bus_type == e1000_bus_type_pcix) ? "-X" : ""),
1214                ((hw->bus_speed == e1000_bus_speed_133) ? 133 :
1215                 (hw->bus_speed == e1000_bus_speed_120) ? 120 :
1216                 (hw->bus_speed == e1000_bus_speed_100) ? 100 :
1217                 (hw->bus_speed == e1000_bus_speed_66) ? 66 : 33),
1218                ((hw->bus_width == e1000_bus_width_64) ? 64 : 32),
1219                netdev->dev_addr);
1220
1221         /* carrier off reporting is important to ethtool even BEFORE open */
1222         netif_carrier_off(netdev);
1223
1224         e_info(probe, "Intel(R) PRO/1000 Network Connection\n");
1225
1226         cards_found++;
1227         return 0;
1228
1229 err_register:
1230 err_eeprom:
1231         e1000_phy_hw_reset(hw);
1232
1233         if (hw->flash_address)
1234                 iounmap(hw->flash_address);
1235         kfree(adapter->tx_ring);
1236         kfree(adapter->rx_ring);
1237 err_dma:
1238 err_sw_init:
1239 err_mdio_ioremap:
1240         iounmap(hw->ce4100_gbe_mdio_base_virt);
1241         iounmap(hw->hw_addr);
1242 err_ioremap:
1243         disable_dev = !test_and_set_bit(__E1000_DISABLED, &adapter->flags);
1244         free_netdev(netdev);
1245 err_alloc_etherdev:
1246         pci_release_selected_regions(pdev, bars);
1247 err_pci_reg:
1248         if (!adapter || disable_dev)
1249                 pci_disable_device(pdev);
1250         return err;
1251 }
1252
1253 /**
1254  * e1000_remove - Device Removal Routine
1255  * @pdev: PCI device information struct
1256  *
1257  * e1000_remove is called by the PCI subsystem to alert the driver
1258  * that it should release a PCI device. That could be caused by a
1259  * Hot-Plug event, or because the driver is going to be removed from
1260  * memory.
1261  **/
1262 static void e1000_remove(struct pci_dev *pdev)
1263 {
1264         struct net_device *netdev = pci_get_drvdata(pdev);
1265         struct e1000_adapter *adapter = netdev_priv(netdev);
1266         struct e1000_hw *hw = &adapter->hw;
1267         bool disable_dev;
1268
1269         e1000_down_and_stop(adapter);
1270         e1000_release_manageability(adapter);
1271
1272         unregister_netdev(netdev);
1273
1274         e1000_phy_hw_reset(hw);
1275
1276         kfree(adapter->tx_ring);
1277         kfree(adapter->rx_ring);
1278
1279         if (hw->mac_type == e1000_ce4100)
1280                 iounmap(hw->ce4100_gbe_mdio_base_virt);
1281         iounmap(hw->hw_addr);
1282         if (hw->flash_address)
1283                 iounmap(hw->flash_address);
1284         pci_release_selected_regions(pdev, adapter->bars);
1285
1286         disable_dev = !test_and_set_bit(__E1000_DISABLED, &adapter->flags);
1287         free_netdev(netdev);
1288
1289         if (disable_dev)
1290                 pci_disable_device(pdev);
1291 }
1292
1293 /**
1294  * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
1295  * @adapter: board private structure to initialize
1296  *
1297  * e1000_sw_init initializes the Adapter private data structure.
1298  * e1000_init_hw_struct MUST be called before this function
1299  **/
1300 static int e1000_sw_init(struct e1000_adapter *adapter)
1301 {
1302         adapter->rx_buffer_len = MAXIMUM_ETHERNET_VLAN_SIZE;
1303
1304         adapter->num_tx_queues = 1;
1305         adapter->num_rx_queues = 1;
1306
1307         if (e1000_alloc_queues(adapter)) {
1308                 e_err(probe, "Unable to allocate memory for queues\n");
1309                 return -ENOMEM;
1310         }
1311
1312         /* Explicitly disable IRQ since the NIC can be in any state. */
1313         e1000_irq_disable(adapter);
1314
1315         spin_lock_init(&adapter->stats_lock);
1316
1317         set_bit(__E1000_DOWN, &adapter->flags);
1318
1319         return 0;
1320 }
1321
1322 /**
1323  * e1000_alloc_queues - Allocate memory for all rings
1324  * @adapter: board private structure to initialize
1325  *
1326  * We allocate one ring per queue at run-time since we don't know the
1327  * number of queues at compile-time.
1328  **/
1329 static int e1000_alloc_queues(struct e1000_adapter *adapter)
1330 {
1331         adapter->tx_ring = kcalloc(adapter->num_tx_queues,
1332                                    sizeof(struct e1000_tx_ring), GFP_KERNEL);
1333         if (!adapter->tx_ring)
1334                 return -ENOMEM;
1335
1336         adapter->rx_ring = kcalloc(adapter->num_rx_queues,
1337                                    sizeof(struct e1000_rx_ring), GFP_KERNEL);
1338         if (!adapter->rx_ring) {
1339                 kfree(adapter->tx_ring);
1340                 return -ENOMEM;
1341         }
1342
1343         return E1000_SUCCESS;
1344 }
1345
1346 /**
1347  * e1000_open - Called when a network interface is made active
1348  * @netdev: network interface device structure
1349  *
1350  * Returns 0 on success, negative value on failure
1351  *
1352  * The open entry point is called when a network interface is made
1353  * active by the system (IFF_UP).  At this point all resources needed
1354  * for transmit and receive operations are allocated, the interrupt
1355  * handler is registered with the OS, the watchdog task is started,
1356  * and the stack is notified that the interface is ready.
1357  **/
1358 int e1000_open(struct net_device *netdev)
1359 {
1360         struct e1000_adapter *adapter = netdev_priv(netdev);
1361         struct e1000_hw *hw = &adapter->hw;
1362         int err;
1363
1364         /* disallow open during test */
1365         if (test_bit(__E1000_TESTING, &adapter->flags))
1366                 return -EBUSY;
1367
1368         netif_carrier_off(netdev);
1369
1370         /* allocate transmit descriptors */
1371         err = e1000_setup_all_tx_resources(adapter);
1372         if (err)
1373                 goto err_setup_tx;
1374
1375         /* allocate receive descriptors */
1376         err = e1000_setup_all_rx_resources(adapter);
1377         if (err)
1378                 goto err_setup_rx;
1379
1380         e1000_power_up_phy(adapter);
1381
1382         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
1383         if ((hw->mng_cookie.status &
1384                           E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT)) {
1385                 e1000_update_mng_vlan(adapter);
1386         }
1387
1388         /* before we allocate an interrupt, we must be ready to handle it.
1389          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
1390          * as soon as we call pci_request_irq, so we have to setup our
1391          * clean_rx handler before we do so.
1392          */
1393         e1000_configure(adapter);
1394
1395         err = e1000_request_irq(adapter);
1396         if (err)
1397                 goto err_req_irq;
1398
1399         /* From here on the code is the same as e1000_up() */
1400         clear_bit(__E1000_DOWN, &adapter->flags);
1401
1402         napi_enable(&adapter->napi);
1403
1404         e1000_irq_enable(adapter);
1405
1406         netif_start_queue(netdev);
1407
1408         /* fire a link status change interrupt to start the watchdog */
1409         ew32(ICS, E1000_ICS_LSC);
1410
1411         return E1000_SUCCESS;
1412
1413 err_req_irq:
1414         e1000_power_down_phy(adapter);
1415         e1000_free_all_rx_resources(adapter);
1416 err_setup_rx:
1417         e1000_free_all_tx_resources(adapter);
1418 err_setup_tx:
1419         e1000_reset(adapter);
1420
1421         return err;
1422 }
1423
1424 /**
1425  * e1000_close - Disables a network interface
1426  * @netdev: network interface device structure
1427  *
1428  * Returns 0, this is not allowed to fail
1429  *
1430  * The close entry point is called when an interface is de-activated
1431  * by the OS.  The hardware is still under the drivers control, but
1432  * needs to be disabled.  A global MAC reset is issued to stop the
1433  * hardware, and all transmit and receive resources are freed.
1434  **/
1435 int e1000_close(struct net_device *netdev)
1436 {
1437         struct e1000_adapter *adapter = netdev_priv(netdev);
1438         struct e1000_hw *hw = &adapter->hw;
1439         int count = E1000_CHECK_RESET_COUNT;
1440
1441         while (test_and_set_bit(__E1000_RESETTING, &adapter->flags) && count--)
1442                 usleep_range(10000, 20000);
1443
1444         WARN_ON(count < 0);
1445
1446         /* signal that we're down so that the reset task will no longer run */
1447         set_bit(__E1000_DOWN, &adapter->flags);
1448         clear_bit(__E1000_RESETTING, &adapter->flags);
1449
1450         e1000_down(adapter);
1451         e1000_power_down_phy(adapter);
1452         e1000_free_irq(adapter);
1453
1454         e1000_free_all_tx_resources(adapter);
1455         e1000_free_all_rx_resources(adapter);
1456
1457         /* kill manageability vlan ID if supported, but not if a vlan with
1458          * the same ID is registered on the host OS (let 8021q kill it)
1459          */
1460         if ((hw->mng_cookie.status &
1461              E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT) &&
1462             !test_bit(adapter->mng_vlan_id, adapter->active_vlans)) {
1463                 e1000_vlan_rx_kill_vid(netdev, htons(ETH_P_8021Q),
1464                                        adapter->mng_vlan_id);
1465         }
1466
1467         return 0;
1468 }
1469
1470 /**
1471  * e1000_check_64k_bound - check that memory doesn't cross 64kB boundary
1472  * @adapter: address of board private structure
1473  * @start: address of beginning of memory
1474  * @len: length of memory
1475  **/
1476 static bool e1000_check_64k_bound(struct e1000_adapter *adapter, void *start,
1477                                   unsigned long len)
1478 {
1479         struct e1000_hw *hw = &adapter->hw;
1480         unsigned long begin = (unsigned long)start;
1481         unsigned long end = begin + len;
1482
1483         /* First rev 82545 and 82546 need to not allow any memory
1484          * write location to cross 64k boundary due to errata 23
1485          */
1486         if (hw->mac_type == e1000_82545 ||
1487             hw->mac_type == e1000_ce4100 ||
1488             hw->mac_type == e1000_82546) {
1489                 return ((begin ^ (end - 1)) >> 16) != 0 ? false : true;
1490         }
1491
1492         return true;
1493 }
1494
1495 /**
1496  * e1000_setup_tx_resources - allocate Tx resources (Descriptors)
1497  * @adapter: board private structure
1498  * @txdr:    tx descriptor ring (for a specific queue) to setup
1499  *
1500  * Return 0 on success, negative on failure
1501  **/
1502 static int e1000_setup_tx_resources(struct e1000_adapter *adapter,
1503                                     struct e1000_tx_ring *txdr)
1504 {
1505         struct pci_dev *pdev = adapter->pdev;
1506         int size;
1507
1508         size = sizeof(struct e1000_tx_buffer) * txdr->count;
1509         txdr->buffer_info = vzalloc(size);
1510         if (!txdr->buffer_info)
1511                 return -ENOMEM;
1512
1513         /* round up to nearest 4K */
1514
1515         txdr->size = txdr->count * sizeof(struct e1000_tx_desc);
1516         txdr->size = ALIGN(txdr->size, 4096);
1517
1518         txdr->desc = dma_alloc_coherent(&pdev->dev, txdr->size, &txdr->dma,
1519                                         GFP_KERNEL);
1520         if (!txdr->desc) {
1521 setup_tx_desc_die:
1522                 vfree(txdr->buffer_info);
1523                 return -ENOMEM;
1524         }
1525
1526         /* Fix for errata 23, can't cross 64kB boundary */
1527         if (!e1000_check_64k_bound(adapter, txdr->desc, txdr->size)) {
1528                 void *olddesc = txdr->desc;
1529                 dma_addr_t olddma = txdr->dma;
1530                 e_err(tx_err, "txdr align check failed: %u bytes at %p\n",
1531                       txdr->size, txdr->desc);
1532                 /* Try again, without freeing the previous */
1533                 txdr->desc = dma_alloc_coherent(&pdev->dev, txdr->size,
1534                                                 &txdr->dma, GFP_KERNEL);
1535                 /* Failed allocation, critical failure */
1536                 if (!txdr->desc) {
1537                         dma_free_coherent(&pdev->dev, txdr->size, olddesc,
1538                                           olddma);
1539                         goto setup_tx_desc_die;
1540                 }
1541
1542                 if (!e1000_check_64k_bound(adapter, txdr->desc, txdr->size)) {
1543                         /* give up */
1544                         dma_free_coherent(&pdev->dev, txdr->size, txdr->desc,
1545                                           txdr->dma);
1546                         dma_free_coherent(&pdev->dev, txdr->size, olddesc,
1547                                           olddma);
1548                         e_err(probe, "Unable to allocate aligned memory "
1549                               "for the transmit descriptor ring\n");
1550                         vfree(txdr->buffer_info);
1551                         return -ENOMEM;
1552                 } else {
1553                         /* Free old allocation, new allocation was successful */
1554                         dma_free_coherent(&pdev->dev, txdr->size, olddesc,
1555                                           olddma);
1556                 }
1557         }
1558         memset(txdr->desc, 0, txdr->size);
1559
1560         txdr->next_to_use = 0;
1561         txdr->next_to_clean = 0;
1562
1563         return 0;
1564 }
1565
1566 /**
1567  * e1000_setup_all_tx_resources - wrapper to allocate Tx resources
1568  *                                (Descriptors) for all queues
1569  * @adapter: board private structure
1570  *
1571  * Return 0 on success, negative on failure
1572  **/
1573 int e1000_setup_all_tx_resources(struct e1000_adapter *adapter)
1574 {
1575         int i, err = 0;
1576
1577         for (i = 0; i < adapter->num_tx_queues; i++) {
1578                 err = e1000_setup_tx_resources(adapter, &adapter->tx_ring[i]);
1579                 if (err) {
1580                         e_err(probe, "Allocation for Tx Queue %u failed\n", i);
1581                         for (i-- ; i >= 0; i--)
1582                                 e1000_free_tx_resources(adapter,
1583                                                         &adapter->tx_ring[i]);
1584                         break;
1585                 }
1586         }
1587
1588         return err;
1589 }
1590
1591 /**
1592  * e1000_configure_tx - Configure 8254x Transmit Unit after Reset
1593  * @adapter: board private structure
1594  *
1595  * Configure the Tx unit of the MAC after a reset.
1596  **/
1597 static void e1000_configure_tx(struct e1000_adapter *adapter)
1598 {
1599         u64 tdba;
1600         struct e1000_hw *hw = &adapter->hw;
1601         u32 tdlen, tctl, tipg;
1602         u32 ipgr1, ipgr2;
1603
1604         /* Setup the HW Tx Head and Tail descriptor pointers */
1605
1606         switch (adapter->num_tx_queues) {
1607         case 1:
1608         default:
1609                 tdba = adapter->tx_ring[0].dma;
1610                 tdlen = adapter->tx_ring[0].count *
1611                         sizeof(struct e1000_tx_desc);
1612                 ew32(TDLEN, tdlen);
1613                 ew32(TDBAH, (tdba >> 32));
1614                 ew32(TDBAL, (tdba & 0x00000000ffffffffULL));
1615                 ew32(TDT, 0);
1616                 ew32(TDH, 0);
1617                 adapter->tx_ring[0].tdh = ((hw->mac_type >= e1000_82543) ?
1618                                            E1000_TDH : E1000_82542_TDH);
1619                 adapter->tx_ring[0].tdt = ((hw->mac_type >= e1000_82543) ?
1620                                            E1000_TDT : E1000_82542_TDT);
1621                 break;
1622         }
1623
1624         /* Set the default values for the Tx Inter Packet Gap timer */
1625         if ((hw->media_type == e1000_media_type_fiber ||
1626              hw->media_type == e1000_media_type_internal_serdes))
1627                 tipg = DEFAULT_82543_TIPG_IPGT_FIBER;
1628         else
1629                 tipg = DEFAULT_82543_TIPG_IPGT_COPPER;
1630
1631         switch (hw->mac_type) {
1632         case e1000_82542_rev2_0:
1633         case e1000_82542_rev2_1:
1634                 tipg = DEFAULT_82542_TIPG_IPGT;
1635                 ipgr1 = DEFAULT_82542_TIPG_IPGR1;
1636                 ipgr2 = DEFAULT_82542_TIPG_IPGR2;
1637                 break;
1638         default:
1639                 ipgr1 = DEFAULT_82543_TIPG_IPGR1;
1640                 ipgr2 = DEFAULT_82543_TIPG_IPGR2;
1641                 break;
1642         }
1643         tipg |= ipgr1 << E1000_TIPG_IPGR1_SHIFT;
1644         tipg |= ipgr2 << E1000_TIPG_IPGR2_SHIFT;
1645         ew32(TIPG, tipg);
1646
1647         /* Set the Tx Interrupt Delay register */
1648
1649         ew32(TIDV, adapter->tx_int_delay);
1650         if (hw->mac_type >= e1000_82540)
1651                 ew32(TADV, adapter->tx_abs_int_delay);
1652
1653         /* Program the Transmit Control Register */
1654
1655         tctl = er32(TCTL);
1656         tctl &= ~E1000_TCTL_CT;
1657         tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
1658                 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
1659
1660         e1000_config_collision_dist(hw);
1661
1662         /* Setup Transmit Descriptor Settings for eop descriptor */
1663         adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
1664
1665         /* only set IDE if we are delaying interrupts using the timers */
1666         if (adapter->tx_int_delay)
1667                 adapter->txd_cmd |= E1000_TXD_CMD_IDE;
1668
1669         if (hw->mac_type < e1000_82543)
1670                 adapter->txd_cmd |= E1000_TXD_CMD_RPS;
1671         else
1672                 adapter->txd_cmd |= E1000_TXD_CMD_RS;
1673
1674         /* Cache if we're 82544 running in PCI-X because we'll
1675          * need this to apply a workaround later in the send path.
1676          */
1677         if (hw->mac_type == e1000_82544 &&
1678             hw->bus_type == e1000_bus_type_pcix)
1679                 adapter->pcix_82544 = true;
1680
1681         ew32(TCTL, tctl);
1682
1683 }
1684
1685 /**
1686  * e1000_setup_rx_resources - allocate Rx resources (Descriptors)
1687  * @adapter: board private structure
1688  * @rxdr:    rx descriptor ring (for a specific queue) to setup
1689  *
1690  * Returns 0 on success, negative on failure
1691  **/
1692 static int e1000_setup_rx_resources(struct e1000_adapter *adapter,
1693                                     struct e1000_rx_ring *rxdr)
1694 {
1695         struct pci_dev *pdev = adapter->pdev;
1696         int size, desc_len;
1697
1698         size = sizeof(struct e1000_rx_buffer) * rxdr->count;
1699         rxdr->buffer_info = vzalloc(size);
1700         if (!rxdr->buffer_info)
1701                 return -ENOMEM;
1702
1703         desc_len = sizeof(struct e1000_rx_desc);
1704
1705         /* Round up to nearest 4K */
1706
1707         rxdr->size = rxdr->count * desc_len;
1708         rxdr->size = ALIGN(rxdr->size, 4096);
1709
1710         rxdr->desc = dma_alloc_coherent(&pdev->dev, rxdr->size, &rxdr->dma,
1711                                         GFP_KERNEL);
1712         if (!rxdr->desc) {
1713 setup_rx_desc_die:
1714                 vfree(rxdr->buffer_info);
1715                 return -ENOMEM;
1716         }
1717
1718         /* Fix for errata 23, can't cross 64kB boundary */
1719         if (!e1000_check_64k_bound(adapter, rxdr->desc, rxdr->size)) {
1720                 void *olddesc = rxdr->desc;
1721                 dma_addr_t olddma = rxdr->dma;
1722                 e_err(rx_err, "rxdr align check failed: %u bytes at %p\n",
1723                       rxdr->size, rxdr->desc);
1724                 /* Try again, without freeing the previous */
1725                 rxdr->desc = dma_alloc_coherent(&pdev->dev, rxdr->size,
1726                                                 &rxdr->dma, GFP_KERNEL);
1727                 /* Failed allocation, critical failure */
1728                 if (!rxdr->desc) {
1729                         dma_free_coherent(&pdev->dev, rxdr->size, olddesc,
1730                                           olddma);
1731                         goto setup_rx_desc_die;
1732                 }
1733
1734                 if (!e1000_check_64k_bound(adapter, rxdr->desc, rxdr->size)) {
1735                         /* give up */
1736                         dma_free_coherent(&pdev->dev, rxdr->size, rxdr->desc,
1737                                           rxdr->dma);
1738                         dma_free_coherent(&pdev->dev, rxdr->size, olddesc,
1739                                           olddma);
1740                         e_err(probe, "Unable to allocate aligned memory for "
1741                               "the Rx descriptor ring\n");
1742                         goto setup_rx_desc_die;
1743                 } else {
1744                         /* Free old allocation, new allocation was successful */
1745                         dma_free_coherent(&pdev->dev, rxdr->size, olddesc,
1746                                           olddma);
1747                 }
1748         }
1749         memset(rxdr->desc, 0, rxdr->size);
1750
1751         rxdr->next_to_clean = 0;
1752         rxdr->next_to_use = 0;
1753         rxdr->rx_skb_top = NULL;
1754
1755         return 0;
1756 }
1757
1758 /**
1759  * e1000_setup_all_rx_resources - wrapper to allocate Rx resources
1760  *                                (Descriptors) for all queues
1761  * @adapter: board private structure
1762  *
1763  * Return 0 on success, negative on failure
1764  **/
1765 int e1000_setup_all_rx_resources(struct e1000_adapter *adapter)
1766 {
1767         int i, err = 0;
1768
1769         for (i = 0; i < adapter->num_rx_queues; i++) {
1770                 err = e1000_setup_rx_resources(adapter, &adapter->rx_ring[i]);
1771                 if (err) {
1772                         e_err(probe, "Allocation for Rx Queue %u failed\n", i);
1773                         for (i-- ; i >= 0; i--)
1774                                 e1000_free_rx_resources(adapter,
1775                                                         &adapter->rx_ring[i]);
1776                         break;
1777                 }
1778         }
1779
1780         return err;
1781 }
1782
1783 /**
1784  * e1000_setup_rctl - configure the receive control registers
1785  * @adapter: Board private structure
1786  **/
1787 static void e1000_setup_rctl(struct e1000_adapter *adapter)
1788 {
1789         struct e1000_hw *hw = &adapter->hw;
1790         u32 rctl;
1791
1792         rctl = er32(RCTL);
1793
1794         rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
1795
1796         rctl |= E1000_RCTL_BAM | E1000_RCTL_LBM_NO |
1797                 E1000_RCTL_RDMTS_HALF |
1798                 (hw->mc_filter_type << E1000_RCTL_MO_SHIFT);
1799
1800         if (hw->tbi_compatibility_on == 1)
1801                 rctl |= E1000_RCTL_SBP;
1802         else
1803                 rctl &= ~E1000_RCTL_SBP;
1804
1805         if (adapter->netdev->mtu <= ETH_DATA_LEN)
1806                 rctl &= ~E1000_RCTL_LPE;
1807         else
1808                 rctl |= E1000_RCTL_LPE;
1809
1810         /* Setup buffer sizes */
1811         rctl &= ~E1000_RCTL_SZ_4096;
1812         rctl |= E1000_RCTL_BSEX;
1813         switch (adapter->rx_buffer_len) {
1814         case E1000_RXBUFFER_2048:
1815         default:
1816                 rctl |= E1000_RCTL_SZ_2048;
1817                 rctl &= ~E1000_RCTL_BSEX;
1818                 break;
1819         case E1000_RXBUFFER_4096:
1820                 rctl |= E1000_RCTL_SZ_4096;
1821                 break;
1822         case E1000_RXBUFFER_8192:
1823                 rctl |= E1000_RCTL_SZ_8192;
1824                 break;
1825         case E1000_RXBUFFER_16384:
1826                 rctl |= E1000_RCTL_SZ_16384;
1827                 break;
1828         }
1829
1830         /* This is useful for sniffing bad packets. */
1831         if (adapter->netdev->features & NETIF_F_RXALL) {
1832                 /* UPE and MPE will be handled by normal PROMISC logic
1833                  * in e1000e_set_rx_mode
1834                  */
1835                 rctl |= (E1000_RCTL_SBP | /* Receive bad packets */
1836                          E1000_RCTL_BAM | /* RX All Bcast Pkts */
1837                          E1000_RCTL_PMCF); /* RX All MAC Ctrl Pkts */
1838
1839                 rctl &= ~(E1000_RCTL_VFE | /* Disable VLAN filter */
1840                           E1000_RCTL_DPF | /* Allow filtered pause */
1841                           E1000_RCTL_CFIEN); /* Dis VLAN CFIEN Filter */
1842                 /* Do not mess with E1000_CTRL_VME, it affects transmit as well,
1843                  * and that breaks VLANs.
1844                  */
1845         }
1846
1847         ew32(RCTL, rctl);
1848 }
1849
1850 /**
1851  * e1000_configure_rx - Configure 8254x Receive Unit after Reset
1852  * @adapter: board private structure
1853  *
1854  * Configure the Rx unit of the MAC after a reset.
1855  **/
1856 static void e1000_configure_rx(struct e1000_adapter *adapter)
1857 {
1858         u64 rdba;
1859         struct e1000_hw *hw = &adapter->hw;
1860         u32 rdlen, rctl, rxcsum;
1861
1862         if (adapter->netdev->mtu > ETH_DATA_LEN) {
1863                 rdlen = adapter->rx_ring[0].count *
1864                         sizeof(struct e1000_rx_desc);
1865                 adapter->clean_rx = e1000_clean_jumbo_rx_irq;
1866                 adapter->alloc_rx_buf = e1000_alloc_jumbo_rx_buffers;
1867         } else {
1868                 rdlen = adapter->rx_ring[0].count *
1869                         sizeof(struct e1000_rx_desc);
1870                 adapter->clean_rx = e1000_clean_rx_irq;
1871                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
1872         }
1873
1874         /* disable receives while setting up the descriptors */
1875         rctl = er32(RCTL);
1876         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1877
1878         /* set the Receive Delay Timer Register */
1879         ew32(RDTR, adapter->rx_int_delay);
1880
1881         if (hw->mac_type >= e1000_82540) {
1882                 ew32(RADV, adapter->rx_abs_int_delay);
1883                 if (adapter->itr_setting != 0)
1884                         ew32(ITR, 1000000000 / (adapter->itr * 256));
1885         }
1886
1887         /* Setup the HW Rx Head and Tail Descriptor Pointers and
1888          * the Base and Length of the Rx Descriptor Ring
1889          */
1890         switch (adapter->num_rx_queues) {
1891         case 1:
1892         default:
1893                 rdba = adapter->rx_ring[0].dma;
1894                 ew32(RDLEN, rdlen);
1895                 ew32(RDBAH, (rdba >> 32));
1896                 ew32(RDBAL, (rdba & 0x00000000ffffffffULL));
1897                 ew32(RDT, 0);
1898                 ew32(RDH, 0);
1899                 adapter->rx_ring[0].rdh = ((hw->mac_type >= e1000_82543) ?
1900                                            E1000_RDH : E1000_82542_RDH);
1901                 adapter->rx_ring[0].rdt = ((hw->mac_type >= e1000_82543) ?
1902                                            E1000_RDT : E1000_82542_RDT);
1903                 break;
1904         }
1905
1906         /* Enable 82543 Receive Checksum Offload for TCP and UDP */
1907         if (hw->mac_type >= e1000_82543) {
1908                 rxcsum = er32(RXCSUM);
1909                 if (adapter->rx_csum)
1910                         rxcsum |= E1000_RXCSUM_TUOFL;
1911                 else
1912                         /* don't need to clear IPPCSE as it defaults to 0 */
1913                         rxcsum &= ~E1000_RXCSUM_TUOFL;
1914                 ew32(RXCSUM, rxcsum);
1915         }
1916
1917         /* Enable Receives */
1918         ew32(RCTL, rctl | E1000_RCTL_EN);
1919 }
1920
1921 /**
1922  * e1000_free_tx_resources - Free Tx Resources per Queue
1923  * @adapter: board private structure
1924  * @tx_ring: Tx descriptor ring for a specific queue
1925  *
1926  * Free all transmit software resources
1927  **/
1928 static void e1000_free_tx_resources(struct e1000_adapter *adapter,
1929                                     struct e1000_tx_ring *tx_ring)
1930 {
1931         struct pci_dev *pdev = adapter->pdev;
1932
1933         e1000_clean_tx_ring(adapter, tx_ring);
1934
1935         vfree(tx_ring->buffer_info);
1936         tx_ring->buffer_info = NULL;
1937
1938         dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
1939                           tx_ring->dma);
1940
1941         tx_ring->desc = NULL;
1942 }
1943
1944 /**
1945  * e1000_free_all_tx_resources - Free Tx Resources for All Queues
1946  * @adapter: board private structure
1947  *
1948  * Free all transmit software resources
1949  **/
1950 void e1000_free_all_tx_resources(struct e1000_adapter *adapter)
1951 {
1952         int i;
1953
1954         for (i = 0; i < adapter->num_tx_queues; i++)
1955                 e1000_free_tx_resources(adapter, &adapter->tx_ring[i]);
1956 }
1957
1958 static void
1959 e1000_unmap_and_free_tx_resource(struct e1000_adapter *adapter,
1960                                  struct e1000_tx_buffer *buffer_info)
1961 {
1962         if (buffer_info->dma) {
1963                 if (buffer_info->mapped_as_page)
1964                         dma_unmap_page(&adapter->pdev->dev, buffer_info->dma,
1965                                        buffer_info->length, DMA_TO_DEVICE);
1966                 else
1967                         dma_unmap_single(&adapter->pdev->dev, buffer_info->dma,
1968                                          buffer_info->length,
1969                                          DMA_TO_DEVICE);
1970                 buffer_info->dma = 0;
1971         }
1972         if (buffer_info->skb) {
1973                 dev_kfree_skb_any(buffer_info->skb);
1974                 buffer_info->skb = NULL;
1975         }
1976         buffer_info->time_stamp = 0;
1977         /* buffer_info must be completely set up in the transmit path */
1978 }
1979
1980 /**
1981  * e1000_clean_tx_ring - Free Tx Buffers
1982  * @adapter: board private structure
1983  * @tx_ring: ring to be cleaned
1984  **/
1985 static void e1000_clean_tx_ring(struct e1000_adapter *adapter,
1986                                 struct e1000_tx_ring *tx_ring)
1987 {
1988         struct e1000_hw *hw = &adapter->hw;
1989         struct e1000_tx_buffer *buffer_info;
1990         unsigned long size;
1991         unsigned int i;
1992
1993         /* Free all the Tx ring sk_buffs */
1994
1995         for (i = 0; i < tx_ring->count; i++) {
1996                 buffer_info = &tx_ring->buffer_info[i];
1997                 e1000_unmap_and_free_tx_resource(adapter, buffer_info);
1998         }
1999
2000         netdev_reset_queue(adapter->netdev);
2001         size = sizeof(struct e1000_tx_buffer) * tx_ring->count;
2002         memset(tx_ring->buffer_info, 0, size);
2003
2004         /* Zero out the descriptor ring */
2005
2006         memset(tx_ring->desc, 0, tx_ring->size);
2007
2008         tx_ring->next_to_use = 0;
2009         tx_ring->next_to_clean = 0;
2010         tx_ring->last_tx_tso = false;
2011
2012         writel(0, hw->hw_addr + tx_ring->tdh);
2013         writel(0, hw->hw_addr + tx_ring->tdt);
2014 }
2015
2016 /**
2017  * e1000_clean_all_tx_rings - Free Tx Buffers for all queues
2018  * @adapter: board private structure
2019  **/
2020 static void e1000_clean_all_tx_rings(struct e1000_adapter *adapter)
2021 {
2022         int i;
2023
2024         for (i = 0; i < adapter->num_tx_queues; i++)
2025                 e1000_clean_tx_ring(adapter, &adapter->tx_ring[i]);
2026 }
2027
2028 /**
2029  * e1000_free_rx_resources - Free Rx Resources
2030  * @adapter: board private structure
2031  * @rx_ring: ring to clean the resources from
2032  *
2033  * Free all receive software resources
2034  **/
2035 static void e1000_free_rx_resources(struct e1000_adapter *adapter,
2036                                     struct e1000_rx_ring *rx_ring)
2037 {
2038         struct pci_dev *pdev = adapter->pdev;
2039
2040         e1000_clean_rx_ring(adapter, rx_ring);
2041
2042         vfree(rx_ring->buffer_info);
2043         rx_ring->buffer_info = NULL;
2044
2045         dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
2046                           rx_ring->dma);
2047
2048         rx_ring->desc = NULL;
2049 }
2050
2051 /**
2052  * e1000_free_all_rx_resources - Free Rx Resources for All Queues
2053  * @adapter: board private structure
2054  *
2055  * Free all receive software resources
2056  **/
2057 void e1000_free_all_rx_resources(struct e1000_adapter *adapter)
2058 {
2059         int i;
2060
2061         for (i = 0; i < adapter->num_rx_queues; i++)
2062                 e1000_free_rx_resources(adapter, &adapter->rx_ring[i]);
2063 }
2064
2065 #define E1000_HEADROOM (NET_SKB_PAD + NET_IP_ALIGN)
2066 static unsigned int e1000_frag_len(const struct e1000_adapter *a)
2067 {
2068         return SKB_DATA_ALIGN(a->rx_buffer_len + E1000_HEADROOM) +
2069                 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
2070 }
2071
2072 static void *e1000_alloc_frag(const struct e1000_adapter *a)
2073 {
2074         unsigned int len = e1000_frag_len(a);
2075         u8 *data = netdev_alloc_frag(len);
2076
2077         if (likely(data))
2078                 data += E1000_HEADROOM;
2079         return data;
2080 }
2081
2082 /**
2083  * e1000_clean_rx_ring - Free Rx Buffers per Queue
2084  * @adapter: board private structure
2085  * @rx_ring: ring to free buffers from
2086  **/
2087 static void e1000_clean_rx_ring(struct e1000_adapter *adapter,
2088                                 struct e1000_rx_ring *rx_ring)
2089 {
2090         struct e1000_hw *hw = &adapter->hw;
2091         struct e1000_rx_buffer *buffer_info;
2092         struct pci_dev *pdev = adapter->pdev;
2093         unsigned long size;
2094         unsigned int i;
2095
2096         /* Free all the Rx netfrags */
2097         for (i = 0; i < rx_ring->count; i++) {
2098                 buffer_info = &rx_ring->buffer_info[i];
2099                 if (adapter->clean_rx == e1000_clean_rx_irq) {
2100                         if (buffer_info->dma)
2101                                 dma_unmap_single(&pdev->dev, buffer_info->dma,
2102                                                  adapter->rx_buffer_len,
2103                                                  DMA_FROM_DEVICE);
2104                         if (buffer_info->rxbuf.data) {
2105                                 skb_free_frag(buffer_info->rxbuf.data);
2106                                 buffer_info->rxbuf.data = NULL;
2107                         }
2108                 } else if (adapter->clean_rx == e1000_clean_jumbo_rx_irq) {
2109                         if (buffer_info->dma)
2110                                 dma_unmap_page(&pdev->dev, buffer_info->dma,
2111                                                adapter->rx_buffer_len,
2112                                                DMA_FROM_DEVICE);
2113                         if (buffer_info->rxbuf.page) {
2114                                 put_page(buffer_info->rxbuf.page);
2115                                 buffer_info->rxbuf.page = NULL;
2116                         }
2117                 }
2118
2119                 buffer_info->dma = 0;
2120         }
2121
2122         /* there also may be some cached data from a chained receive */
2123         napi_free_frags(&adapter->napi);
2124         rx_ring->rx_skb_top = NULL;
2125
2126         size = sizeof(struct e1000_rx_buffer) * rx_ring->count;
2127         memset(rx_ring->buffer_info, 0, size);
2128
2129         /* Zero out the descriptor ring */
2130         memset(rx_ring->desc, 0, rx_ring->size);
2131
2132         rx_ring->next_to_clean = 0;
2133         rx_ring->next_to_use = 0;
2134
2135         writel(0, hw->hw_addr + rx_ring->rdh);
2136         writel(0, hw->hw_addr + rx_ring->rdt);
2137 }
2138
2139 /**
2140  * e1000_clean_all_rx_rings - Free Rx Buffers for all queues
2141  * @adapter: board private structure
2142  **/
2143 static void e1000_clean_all_rx_rings(struct e1000_adapter *adapter)
2144 {
2145         int i;
2146
2147         for (i = 0; i < adapter->num_rx_queues; i++)
2148                 e1000_clean_rx_ring(adapter, &adapter->rx_ring[i]);
2149 }
2150
2151 /* The 82542 2.0 (revision 2) needs to have the receive unit in reset
2152  * and memory write and invalidate disabled for certain operations
2153  */
2154 static void e1000_enter_82542_rst(struct e1000_adapter *adapter)
2155 {
2156         struct e1000_hw *hw = &adapter->hw;
2157         struct net_device *netdev = adapter->netdev;
2158         u32 rctl;
2159
2160         e1000_pci_clear_mwi(hw);
2161
2162         rctl = er32(RCTL);
2163         rctl |= E1000_RCTL_RST;
2164         ew32(RCTL, rctl);
2165         E1000_WRITE_FLUSH();
2166         mdelay(5);
2167
2168         if (netif_running(netdev))
2169                 e1000_clean_all_rx_rings(adapter);
2170 }
2171
2172 static void e1000_leave_82542_rst(struct e1000_adapter *adapter)
2173 {
2174         struct e1000_hw *hw = &adapter->hw;
2175         struct net_device *netdev = adapter->netdev;
2176         u32 rctl;
2177
2178         rctl = er32(RCTL);
2179         rctl &= ~E1000_RCTL_RST;
2180         ew32(RCTL, rctl);
2181         E1000_WRITE_FLUSH();
2182         mdelay(5);
2183
2184         if (hw->pci_cmd_word & PCI_COMMAND_INVALIDATE)
2185                 e1000_pci_set_mwi(hw);
2186
2187         if (netif_running(netdev)) {
2188                 /* No need to loop, because 82542 supports only 1 queue */
2189                 struct e1000_rx_ring *ring = &adapter->rx_ring[0];
2190                 e1000_configure_rx(adapter);
2191                 adapter->alloc_rx_buf(adapter, ring, E1000_DESC_UNUSED(ring));
2192         }
2193 }
2194
2195 /**
2196  * e1000_set_mac - Change the Ethernet Address of the NIC
2197  * @netdev: network interface device structure
2198  * @p: pointer to an address structure
2199  *
2200  * Returns 0 on success, negative on failure
2201  **/
2202 static int e1000_set_mac(struct net_device *netdev, void *p)
2203 {
2204         struct e1000_adapter *adapter = netdev_priv(netdev);
2205         struct e1000_hw *hw = &adapter->hw;
2206         struct sockaddr *addr = p;
2207
2208         if (!is_valid_ether_addr(addr->sa_data))
2209                 return -EADDRNOTAVAIL;
2210
2211         /* 82542 2.0 needs to be in reset to write receive address registers */
2212
2213         if (hw->mac_type == e1000_82542_rev2_0)
2214                 e1000_enter_82542_rst(adapter);
2215
2216         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
2217         memcpy(hw->mac_addr, addr->sa_data, netdev->addr_len);
2218
2219         e1000_rar_set(hw, hw->mac_addr, 0);
2220
2221         if (hw->mac_type == e1000_82542_rev2_0)
2222                 e1000_leave_82542_rst(adapter);
2223
2224         return 0;
2225 }
2226
2227 /**
2228  * e1000_set_rx_mode - Secondary Unicast, Multicast and Promiscuous mode set
2229  * @netdev: network interface device structure
2230  *
2231  * The set_rx_mode entry point is called whenever the unicast or multicast
2232  * address lists or the network interface flags are updated. This routine is
2233  * responsible for configuring the hardware for proper unicast, multicast,
2234  * promiscuous mode, and all-multi behavior.
2235  **/
2236 static void e1000_set_rx_mode(struct net_device *netdev)
2237 {
2238         struct e1000_adapter *adapter = netdev_priv(netdev);
2239         struct e1000_hw *hw = &adapter->hw;
2240         struct netdev_hw_addr *ha;
2241         bool use_uc = false;
2242         u32 rctl;
2243         u32 hash_value;
2244         int i, rar_entries = E1000_RAR_ENTRIES;
2245         int mta_reg_count = E1000_NUM_MTA_REGISTERS;
2246         u32 *mcarray = kcalloc(mta_reg_count, sizeof(u32), GFP_ATOMIC);
2247
2248         if (!mcarray)
2249                 return;
2250
2251         /* Check for Promiscuous and All Multicast modes */
2252
2253         rctl = er32(RCTL);
2254
2255         if (netdev->flags & IFF_PROMISC) {
2256                 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
2257                 rctl &= ~E1000_RCTL_VFE;
2258         } else {
2259                 if (netdev->flags & IFF_ALLMULTI)
2260                         rctl |= E1000_RCTL_MPE;
2261                 else
2262                         rctl &= ~E1000_RCTL_MPE;
2263                 /* Enable VLAN filter if there is a VLAN */
2264                 if (e1000_vlan_used(adapter))
2265                         rctl |= E1000_RCTL_VFE;
2266         }
2267
2268         if (netdev_uc_count(netdev) > rar_entries - 1) {
2269                 rctl |= E1000_RCTL_UPE;
2270         } else if (!(netdev->flags & IFF_PROMISC)) {
2271                 rctl &= ~E1000_RCTL_UPE;
2272                 use_uc = true;
2273         }
2274
2275         ew32(RCTL, rctl);
2276
2277         /* 82542 2.0 needs to be in reset to write receive address registers */
2278
2279         if (hw->mac_type == e1000_82542_rev2_0)
2280                 e1000_enter_82542_rst(adapter);
2281
2282         /* load the first 14 addresses into the exact filters 1-14. Unicast
2283          * addresses take precedence to avoid disabling unicast filtering
2284          * when possible.
2285          *
2286          * RAR 0 is used for the station MAC address
2287          * if there are not 14 addresses, go ahead and clear the filters
2288          */
2289         i = 1;
2290         if (use_uc)
2291                 netdev_for_each_uc_addr(ha, netdev) {
2292                         if (i == rar_entries)
2293                                 break;
2294                         e1000_rar_set(hw, ha->addr, i++);
2295                 }
2296
2297         netdev_for_each_mc_addr(ha, netdev) {
2298                 if (i == rar_entries) {
2299                         /* load any remaining addresses into the hash table */
2300                         u32 hash_reg, hash_bit, mta;
2301                         hash_value = e1000_hash_mc_addr(hw, ha->addr);
2302                         hash_reg = (hash_value >> 5) & 0x7F;
2303                         hash_bit = hash_value & 0x1F;
2304                         mta = (1 << hash_bit);
2305                         mcarray[hash_reg] |= mta;
2306                 } else {
2307                         e1000_rar_set(hw, ha->addr, i++);
2308                 }
2309         }
2310
2311         for (; i < rar_entries; i++) {
2312                 E1000_WRITE_REG_ARRAY(hw, RA, i << 1, 0);
2313                 E1000_WRITE_FLUSH();
2314                 E1000_WRITE_REG_ARRAY(hw, RA, (i << 1) + 1, 0);
2315                 E1000_WRITE_FLUSH();
2316         }
2317
2318         /* write the hash table completely, write from bottom to avoid
2319          * both stupid write combining chipsets, and flushing each write
2320          */
2321         for (i = mta_reg_count - 1; i >= 0 ; i--) {
2322                 /* If we are on an 82544 has an errata where writing odd
2323                  * offsets overwrites the previous even offset, but writing
2324                  * backwards over the range solves the issue by always
2325                  * writing the odd offset first
2326                  */
2327                 E1000_WRITE_REG_ARRAY(hw, MTA, i, mcarray[i]);
2328         }
2329         E1000_WRITE_FLUSH();
2330
2331         if (hw->mac_type == e1000_82542_rev2_0)
2332                 e1000_leave_82542_rst(adapter);
2333
2334         kfree(mcarray);
2335 }
2336
2337 /**
2338  * e1000_update_phy_info_task - get phy info
2339  * @work: work struct contained inside adapter struct
2340  *
2341  * Need to wait a few seconds after link up to get diagnostic information from
2342  * the phy
2343  */
2344 static void e1000_update_phy_info_task(struct work_struct *work)
2345 {
2346         struct e1000_adapter *adapter = container_of(work,
2347                                                      struct e1000_adapter,
2348                                                      phy_info_task.work);
2349
2350         e1000_phy_get_info(&adapter->hw, &adapter->phy_info);
2351 }
2352
2353 /**
2354  * e1000_82547_tx_fifo_stall_task - task to complete work
2355  * @work: work struct contained inside adapter struct
2356  **/
2357 static void e1000_82547_tx_fifo_stall_task(struct work_struct *work)
2358 {
2359         struct e1000_adapter *adapter = container_of(work,
2360                                                      struct e1000_adapter,
2361                                                      fifo_stall_task.work);
2362         struct e1000_hw *hw = &adapter->hw;
2363         struct net_device *netdev = adapter->netdev;
2364         u32 tctl;
2365
2366         if (atomic_read(&adapter->tx_fifo_stall)) {
2367                 if ((er32(TDT) == er32(TDH)) &&
2368                    (er32(TDFT) == er32(TDFH)) &&
2369                    (er32(TDFTS) == er32(TDFHS))) {
2370                         tctl = er32(TCTL);
2371                         ew32(TCTL, tctl & ~E1000_TCTL_EN);
2372                         ew32(TDFT, adapter->tx_head_addr);
2373                         ew32(TDFH, adapter->tx_head_addr);
2374                         ew32(TDFTS, adapter->tx_head_addr);
2375                         ew32(TDFHS, adapter->tx_head_addr);
2376                         ew32(TCTL, tctl);
2377                         E1000_WRITE_FLUSH();
2378
2379                         adapter->tx_fifo_head = 0;
2380                         atomic_set(&adapter->tx_fifo_stall, 0);
2381                         netif_wake_queue(netdev);
2382                 } else if (!test_bit(__E1000_DOWN, &adapter->flags)) {
2383                         schedule_delayed_work(&adapter->fifo_stall_task, 1);
2384                 }
2385         }
2386 }
2387
2388 bool e1000_has_link(struct e1000_adapter *adapter)
2389 {
2390         struct e1000_hw *hw = &adapter->hw;
2391         bool link_active = false;
2392
2393         /* get_link_status is set on LSC (link status) interrupt or rx
2394          * sequence error interrupt (except on intel ce4100).
2395          * get_link_status will stay false until the
2396          * e1000_check_for_link establishes link for copper adapters
2397          * ONLY
2398          */
2399         switch (hw->media_type) {
2400         case e1000_media_type_copper:
2401                 if (hw->mac_type == e1000_ce4100)
2402                         hw->get_link_status = 1;
2403                 if (hw->get_link_status) {
2404                         e1000_check_for_link(hw);
2405                         link_active = !hw->get_link_status;
2406                 } else {
2407                         link_active = true;
2408                 }
2409                 break;
2410         case e1000_media_type_fiber:
2411                 e1000_check_for_link(hw);
2412                 link_active = !!(er32(STATUS) & E1000_STATUS_LU);
2413                 break;
2414         case e1000_media_type_internal_serdes:
2415                 e1000_check_for_link(hw);
2416                 link_active = hw->serdes_has_link;
2417                 break;
2418         default:
2419                 break;
2420         }
2421
2422         return link_active;
2423 }
2424
2425 /**
2426  * e1000_watchdog - work function
2427  * @work: work struct contained inside adapter struct
2428  **/
2429 static void e1000_watchdog(struct work_struct *work)
2430 {
2431         struct e1000_adapter *adapter = container_of(work,
2432                                                      struct e1000_adapter,
2433                                                      watchdog_task.work);
2434         struct e1000_hw *hw = &adapter->hw;
2435         struct net_device *netdev = adapter->netdev;
2436         struct e1000_tx_ring *txdr = adapter->tx_ring;
2437         u32 link, tctl;
2438
2439         link = e1000_has_link(adapter);
2440         if ((netif_carrier_ok(netdev)) && link)
2441                 goto link_up;
2442
2443         if (link) {
2444                 if (!netif_carrier_ok(netdev)) {
2445                         u32 ctrl;
2446                         bool txb2b = true;
2447                         /* update snapshot of PHY registers on LSC */
2448                         e1000_get_speed_and_duplex(hw,
2449                                                    &adapter->link_speed,
2450                                                    &adapter->link_duplex);
2451
2452                         ctrl = er32(CTRL);
2453                         pr_info("%s NIC Link is Up %d Mbps %s, "
2454                                 "Flow Control: %s\n",
2455                                 netdev->name,
2456                                 adapter->link_speed,
2457                                 adapter->link_duplex == FULL_DUPLEX ?
2458                                 "Full Duplex" : "Half Duplex",
2459                                 ((ctrl & E1000_CTRL_TFCE) && (ctrl &
2460                                 E1000_CTRL_RFCE)) ? "RX/TX" : ((ctrl &
2461                                 E1000_CTRL_RFCE) ? "RX" : ((ctrl &
2462                                 E1000_CTRL_TFCE) ? "TX" : "None")));
2463
2464                         /* adjust timeout factor according to speed/duplex */
2465                         adapter->tx_timeout_factor = 1;
2466                         switch (adapter->link_speed) {
2467                         case SPEED_10:
2468                                 txb2b = false;
2469                                 adapter->tx_timeout_factor = 16;
2470                                 break;
2471                         case SPEED_100:
2472                                 txb2b = false;
2473                                 /* maybe add some timeout factor ? */
2474                                 break;
2475                         }
2476
2477                         /* enable transmits in the hardware */
2478                         tctl = er32(TCTL);
2479                         tctl |= E1000_TCTL_EN;
2480                         ew32(TCTL, tctl);
2481
2482                         netif_carrier_on(netdev);
2483                         if (!test_bit(__E1000_DOWN, &adapter->flags))
2484                                 schedule_delayed_work(&adapter->phy_info_task,
2485                                                       2 * HZ);
2486                         adapter->smartspeed = 0;
2487                 }
2488         } else {
2489                 if (netif_carrier_ok(netdev)) {
2490                         adapter->link_speed = 0;
2491                         adapter->link_duplex = 0;
2492                         pr_info("%s NIC Link is Down\n",
2493                                 netdev->name);
2494                         netif_carrier_off(netdev);
2495
2496                         if (!test_bit(__E1000_DOWN, &adapter->flags))
2497                                 schedule_delayed_work(&adapter->phy_info_task,
2498                                                       2 * HZ);
2499                 }
2500
2501                 e1000_smartspeed(adapter);
2502         }
2503
2504 link_up:
2505         e1000_update_stats(adapter);
2506
2507         hw->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
2508         adapter->tpt_old = adapter->stats.tpt;
2509         hw->collision_delta = adapter->stats.colc - adapter->colc_old;
2510         adapter->colc_old = adapter->stats.colc;
2511
2512         adapter->gorcl = adapter->stats.gorcl - adapter->gorcl_old;
2513         adapter->gorcl_old = adapter->stats.gorcl;
2514         adapter->gotcl = adapter->stats.gotcl - adapter->gotcl_old;
2515         adapter->gotcl_old = adapter->stats.gotcl;
2516
2517         e1000_update_adaptive(hw);
2518
2519         if (!netif_carrier_ok(netdev)) {
2520                 if (E1000_DESC_UNUSED(txdr) + 1 < txdr->count) {
2521                         /* We've lost link, so the controller stops DMA,
2522                          * but we've got queued Tx work that's never going
2523                          * to get done, so reset controller to flush Tx.
2524                          * (Do the reset outside of interrupt context).
2525                          */
2526                         adapter->tx_timeout_count++;
2527                         schedule_work(&adapter->reset_task);
2528                         /* exit immediately since reset is imminent */
2529                         return;
2530                 }
2531         }
2532
2533         /* Simple mode for Interrupt Throttle Rate (ITR) */
2534         if (hw->mac_type >= e1000_82540 && adapter->itr_setting == 4) {
2535                 /* Symmetric Tx/Rx gets a reduced ITR=2000;
2536                  * Total asymmetrical Tx or Rx gets ITR=8000;
2537                  * everyone else is between 2000-8000.
2538                  */
2539                 u32 goc = (adapter->gotcl + adapter->gorcl) / 10000;
2540                 u32 dif = (adapter->gotcl > adapter->gorcl ?
2541                             adapter->gotcl - adapter->gorcl :
2542                             adapter->gorcl - adapter->gotcl) / 10000;
2543                 u32 itr = goc > 0 ? (dif * 6000 / goc + 2000) : 8000;
2544
2545                 ew32(ITR, 1000000000 / (itr * 256));
2546         }
2547
2548         /* Cause software interrupt to ensure rx ring is cleaned */
2549         ew32(ICS, E1000_ICS_RXDMT0);
2550
2551         /* Force detection of hung controller every watchdog period */
2552         adapter->detect_tx_hung = true;
2553
2554         /* Reschedule the task */
2555         if (!test_bit(__E1000_DOWN, &adapter->flags))
2556                 schedule_delayed_work(&adapter->watchdog_task, 2 * HZ);
2557 }
2558
2559 enum latency_range {
2560         lowest_latency = 0,
2561         low_latency = 1,
2562         bulk_latency = 2,
2563         latency_invalid = 255
2564 };
2565
2566 /**
2567  * e1000_update_itr - update the dynamic ITR value based on statistics
2568  * @adapter: pointer to adapter
2569  * @itr_setting: current adapter->itr
2570  * @packets: the number of packets during this measurement interval
2571  * @bytes: the number of bytes during this measurement interval
2572  *
2573  *      Stores a new ITR value based on packets and byte
2574  *      counts during the last interrupt.  The advantage of per interrupt
2575  *      computation is faster updates and more accurate ITR for the current
2576  *      traffic pattern.  Constants in this function were computed
2577  *      based on theoretical maximum wire speed and thresholds were set based
2578  *      on testing data as well as attempting to minimize response time
2579  *      while increasing bulk throughput.
2580  *      this functionality is controlled by the InterruptThrottleRate module
2581  *      parameter (see e1000_param.c)
2582  **/
2583 static unsigned int e1000_update_itr(struct e1000_adapter *adapter,
2584                                      u16 itr_setting, int packets, int bytes)
2585 {
2586         unsigned int retval = itr_setting;
2587         struct e1000_hw *hw = &adapter->hw;
2588
2589         if (unlikely(hw->mac_type < e1000_82540))
2590                 goto update_itr_done;
2591
2592         if (packets == 0)
2593                 goto update_itr_done;
2594
2595         switch (itr_setting) {
2596         case lowest_latency:
2597                 /* jumbo frames get bulk treatment*/
2598                 if (bytes/packets > 8000)
2599                         retval = bulk_latency;
2600                 else if ((packets < 5) && (bytes > 512))
2601                         retval = low_latency;
2602                 break;
2603         case low_latency:  /* 50 usec aka 20000 ints/s */
2604                 if (bytes > 10000) {
2605                         /* jumbo frames need bulk latency setting */
2606                         if (bytes/packets > 8000)
2607                                 retval = bulk_latency;
2608                         else if ((packets < 10) || ((bytes/packets) > 1200))
2609                                 retval = bulk_latency;
2610                         else if ((packets > 35))
2611                                 retval = lowest_latency;
2612                 } else if (bytes/packets > 2000)
2613                         retval = bulk_latency;
2614                 else if (packets <= 2 && bytes < 512)
2615                         retval = lowest_latency;
2616                 break;
2617         case bulk_latency: /* 250 usec aka 4000 ints/s */
2618                 if (bytes > 25000) {
2619                         if (packets > 35)
2620                                 retval = low_latency;
2621                 } else if (bytes < 6000) {
2622                         retval = low_latency;
2623                 }
2624                 break;
2625         }
2626
2627 update_itr_done:
2628         return retval;
2629 }
2630
2631 static void e1000_set_itr(struct e1000_adapter *adapter)
2632 {
2633         struct e1000_hw *hw = &adapter->hw;
2634         u16 current_itr;
2635         u32 new_itr = adapter->itr;
2636
2637         if (unlikely(hw->mac_type < e1000_82540))
2638                 return;
2639
2640         /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
2641         if (unlikely(adapter->link_speed != SPEED_1000)) {
2642                 current_itr = 0;
2643                 new_itr = 4000;
2644                 goto set_itr_now;
2645         }
2646
2647         adapter->tx_itr = e1000_update_itr(adapter, adapter->tx_itr,
2648                                            adapter->total_tx_packets,
2649                                            adapter->total_tx_bytes);
2650         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2651         if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
2652                 adapter->tx_itr = low_latency;
2653
2654         adapter->rx_itr = e1000_update_itr(adapter, adapter->rx_itr,
2655                                            adapter->total_rx_packets,
2656                                            adapter->total_rx_bytes);
2657         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2658         if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
2659                 adapter->rx_itr = low_latency;
2660
2661         current_itr = max(adapter->rx_itr, adapter->tx_itr);
2662
2663         switch (current_itr) {
2664         /* counts and packets in update_itr are dependent on these numbers */
2665         case lowest_latency:
2666                 new_itr = 70000;
2667                 break;
2668         case low_latency:
2669                 new_itr = 20000; /* aka hwitr = ~200 */
2670                 break;
2671         case bulk_latency:
2672                 new_itr = 4000;
2673                 break;
2674         default:
2675                 break;
2676         }
2677
2678 set_itr_now:
2679         if (new_itr != adapter->itr) {
2680                 /* this attempts to bias the interrupt rate towards Bulk
2681                  * by adding intermediate steps when interrupt rate is
2682                  * increasing
2683                  */
2684                 new_itr = new_itr > adapter->itr ?
2685                           min(adapter->itr + (new_itr >> 2), new_itr) :
2686                           new_itr;
2687                 adapter->itr = new_itr;
2688                 ew32(ITR, 1000000000 / (new_itr * 256));
2689         }
2690 }
2691
2692 #define E1000_TX_FLAGS_CSUM             0x00000001
2693 #define E1000_TX_FLAGS_VLAN             0x00000002
2694 #define E1000_TX_FLAGS_TSO              0x00000004
2695 #define E1000_TX_FLAGS_IPV4             0x00000008
2696 #define E1000_TX_FLAGS_NO_FCS           0x00000010
2697 #define E1000_TX_FLAGS_VLAN_MASK        0xffff0000
2698 #define E1000_TX_FLAGS_VLAN_SHIFT       16
2699
2700 static int e1000_tso(struct e1000_adapter *adapter,
2701                      struct e1000_tx_ring *tx_ring, struct sk_buff *skb,
2702                      __be16 protocol)
2703 {
2704         struct e1000_context_desc *context_desc;
2705         struct e1000_tx_buffer *buffer_info;
2706         unsigned int i;
2707         u32 cmd_length = 0;
2708         u16 ipcse = 0, tucse, mss;
2709         u8 ipcss, ipcso, tucss, tucso, hdr_len;
2710
2711         if (skb_is_gso(skb)) {
2712                 int err;
2713
2714                 err = skb_cow_head(skb, 0);
2715                 if (err < 0)
2716                         return err;
2717
2718                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
2719                 mss = skb_shinfo(skb)->gso_size;
2720                 if (protocol == htons(ETH_P_IP)) {
2721                         struct iphdr *iph = ip_hdr(skb);
2722                         iph->tot_len = 0;
2723                         iph->check = 0;
2724                         tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
2725                                                                  iph->daddr, 0,
2726                                                                  IPPROTO_TCP,
2727                                                                  0);
2728                         cmd_length = E1000_TXD_CMD_IP;
2729                         ipcse = skb_transport_offset(skb) - 1;
2730                 } else if (skb_is_gso_v6(skb)) {
2731                         ipv6_hdr(skb)->payload_len = 0;
2732                         tcp_hdr(skb)->check =
2733                                 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
2734                                                  &ipv6_hdr(skb)->daddr,
2735                                                  0, IPPROTO_TCP, 0);
2736                         ipcse = 0;
2737                 }
2738                 ipcss = skb_network_offset(skb);
2739                 ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
2740                 tucss = skb_transport_offset(skb);
2741                 tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
2742                 tucse = 0;
2743
2744                 cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
2745                                E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
2746
2747                 i = tx_ring->next_to_use;
2748                 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
2749                 buffer_info = &tx_ring->buffer_info[i];
2750
2751                 context_desc->lower_setup.ip_fields.ipcss  = ipcss;
2752                 context_desc->lower_setup.ip_fields.ipcso  = ipcso;
2753                 context_desc->lower_setup.ip_fields.ipcse  = cpu_to_le16(ipcse);
2754                 context_desc->upper_setup.tcp_fields.tucss = tucss;
2755                 context_desc->upper_setup.tcp_fields.tucso = tucso;
2756                 context_desc->upper_setup.tcp_fields.tucse = cpu_to_le16(tucse);
2757                 context_desc->tcp_seg_setup.fields.mss     = cpu_to_le16(mss);
2758                 context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
2759                 context_desc->cmd_and_length = cpu_to_le32(cmd_length);
2760
2761                 buffer_info->time_stamp = jiffies;
2762                 buffer_info->next_to_watch = i;
2763
2764                 if (++i == tx_ring->count)
2765                         i = 0;
2766
2767                 tx_ring->next_to_use = i;
2768
2769                 return true;
2770         }
2771         return false;
2772 }
2773
2774 static bool e1000_tx_csum(struct e1000_adapter *adapter,
2775                           struct e1000_tx_ring *tx_ring, struct sk_buff *skb,
2776                           __be16 protocol)
2777 {
2778         struct e1000_context_desc *context_desc;
2779         struct e1000_tx_buffer *buffer_info;
2780         unsigned int i;
2781         u8 css;
2782         u32 cmd_len = E1000_TXD_CMD_DEXT;
2783
2784         if (skb->ip_summed != CHECKSUM_PARTIAL)
2785                 return false;
2786
2787         switch (protocol) {
2788         case cpu_to_be16(ETH_P_IP):
2789                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
2790                         cmd_len |= E1000_TXD_CMD_TCP;
2791                 break;
2792         case cpu_to_be16(ETH_P_IPV6):
2793                 /* XXX not handling all IPV6 headers */
2794                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
2795                         cmd_len |= E1000_TXD_CMD_TCP;
2796                 break;
2797         default:
2798                 if (unlikely(net_ratelimit()))
2799                         e_warn(drv, "checksum_partial proto=%x!\n",
2800                                skb->protocol);
2801                 break;
2802         }
2803
2804         css = skb_checksum_start_offset(skb);
2805
2806         i = tx_ring->next_to_use;
2807         buffer_info = &tx_ring->buffer_info[i];
2808         context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
2809
2810         context_desc->lower_setup.ip_config = 0;
2811         context_desc->upper_setup.tcp_fields.tucss = css;
2812         context_desc->upper_setup.tcp_fields.tucso =
2813                 css + skb->csum_offset;
2814         context_desc->upper_setup.tcp_fields.tucse = 0;
2815         context_desc->tcp_seg_setup.data = 0;
2816         context_desc->cmd_and_length = cpu_to_le32(cmd_len);
2817
2818         buffer_info->time_stamp = jiffies;
2819         buffer_info->next_to_watch = i;
2820
2821         if (unlikely(++i == tx_ring->count))
2822                 i = 0;
2823
2824         tx_ring->next_to_use = i;
2825
2826         return true;
2827 }
2828
2829 #define E1000_MAX_TXD_PWR       12
2830 #define E1000_MAX_DATA_PER_TXD  (1<<E1000_MAX_TXD_PWR)
2831
2832 static int e1000_tx_map(struct e1000_adapter *adapter,
2833                         struct e1000_tx_ring *tx_ring,
2834                         struct sk_buff *skb, unsigned int first,
2835                         unsigned int max_per_txd, unsigned int nr_frags,
2836                         unsigned int mss)
2837 {
2838         struct e1000_hw *hw = &adapter->hw;
2839         struct pci_dev *pdev = adapter->pdev;
2840         struct e1000_tx_buffer *buffer_info;
2841         unsigned int len = skb_headlen(skb);
2842         unsigned int offset = 0, size, count = 0, i;
2843         unsigned int f, bytecount, segs;
2844
2845         i = tx_ring->next_to_use;
2846
2847         while (len) {
2848                 buffer_info = &tx_ring->buffer_info[i];
2849                 size = min(len, max_per_txd);
2850                 /* Workaround for Controller erratum --
2851                  * descriptor for non-tso packet in a linear SKB that follows a
2852                  * tso gets written back prematurely before the data is fully
2853                  * DMA'd to the controller
2854                  */
2855                 if (!skb->data_len && tx_ring->last_tx_tso &&
2856                     !skb_is_gso(skb)) {
2857                         tx_ring->last_tx_tso = false;
2858                         size -= 4;
2859                 }
2860
2861                 /* Workaround for premature desc write-backs
2862                  * in TSO mode.  Append 4-byte sentinel desc
2863                  */
2864                 if (unlikely(mss && !nr_frags && size == len && size > 8))
2865                         size -= 4;
2866                 /* work-around for errata 10 and it applies
2867                  * to all controllers in PCI-X mode
2868                  * The fix is to make sure that the first descriptor of a
2869                  * packet is smaller than 2048 - 16 - 16 (or 2016) bytes
2870                  */
2871                 if (unlikely((hw->bus_type == e1000_bus_type_pcix) &&
2872                              (size > 2015) && count == 0))
2873                         size = 2015;
2874
2875                 /* Workaround for potential 82544 hang in PCI-X.  Avoid
2876                  * terminating buffers within evenly-aligned dwords.
2877                  */
2878                 if (unlikely(adapter->pcix_82544 &&
2879                    !((unsigned long)(skb->data + offset + size - 1) & 4) &&
2880                    size > 4))
2881                         size -= 4;
2882
2883                 buffer_info->length = size;
2884                 /* set time_stamp *before* dma to help avoid a possible race */
2885                 buffer_info->time_stamp = jiffies;
2886                 buffer_info->mapped_as_page = false;
2887                 buffer_info->dma = dma_map_single(&pdev->dev,
2888                                                   skb->data + offset,
2889                                                   size, DMA_TO_DEVICE);
2890                 if (dma_mapping_error(&pdev->dev, buffer_info->dma))
2891                         goto dma_error;
2892                 buffer_info->next_to_watch = i;
2893
2894                 len -= size;
2895                 offset += size;
2896                 count++;
2897                 if (len) {
2898                         i++;
2899                         if (unlikely(i == tx_ring->count))
2900                                 i = 0;
2901                 }
2902         }
2903
2904         for (f = 0; f < nr_frags; f++) {
2905                 const struct skb_frag_struct *frag;
2906
2907                 frag = &skb_shinfo(skb)->frags[f];
2908                 len = skb_frag_size(frag);
2909                 offset = 0;
2910
2911                 while (len) {
2912                         unsigned long bufend;
2913                         i++;
2914                         if (unlikely(i == tx_ring->count))
2915                                 i = 0;
2916
2917                         buffer_info = &tx_ring->buffer_info[i];
2918                         size = min(len, max_per_txd);
2919                         /* Workaround for premature desc write-backs
2920                          * in TSO mode.  Append 4-byte sentinel desc
2921                          */
2922                         if (unlikely(mss && f == (nr_frags-1) &&
2923                             size == len && size > 8))
2924                                 size -= 4;
2925                         /* Workaround for potential 82544 hang in PCI-X.
2926                          * Avoid terminating buffers within evenly-aligned
2927                          * dwords.
2928                          */
2929                         bufend = (unsigned long)
2930                                 page_to_phys(skb_frag_page(frag));
2931                         bufend += offset + size - 1;
2932                         if (unlikely(adapter->pcix_82544 &&
2933                                      !(bufend & 4) &&
2934                                      size > 4))
2935                                 size -= 4;
2936
2937                         buffer_info->length = size;
2938                         buffer_info->time_stamp = jiffies;
2939                         buffer_info->mapped_as_page = true;
2940                         buffer_info->dma = skb_frag_dma_map(&pdev->dev, frag,
2941                                                 offset, size, DMA_TO_DEVICE);
2942                         if (dma_mapping_error(&pdev->dev, buffer_info->dma))
2943                                 goto dma_error;
2944                         buffer_info->next_to_watch = i;
2945
2946                         len -= size;
2947                         offset += size;
2948                         count++;
2949                 }
2950         }
2951
2952         segs = skb_shinfo(skb)->gso_segs ?: 1;
2953         /* multiply data chunks by size of headers */
2954         bytecount = ((segs - 1) * skb_headlen(skb)) + skb->len;
2955
2956         tx_ring->buffer_info[i].skb = skb;
2957         tx_ring->buffer_info[i].segs = segs;
2958         tx_ring->buffer_info[i].bytecount = bytecount;
2959         tx_ring->buffer_info[first].next_to_watch = i;
2960
2961         return count;
2962
2963 dma_error:
2964         dev_err(&pdev->dev, "TX DMA map failed\n");
2965         buffer_info->dma = 0;
2966         if (count)
2967                 count--;
2968
2969         while (count--) {
2970                 if (i == 0)
2971                         i += tx_ring->count;
2972                 i--;
2973                 buffer_info = &tx_ring->buffer_info[i];
2974                 e1000_unmap_and_free_tx_resource(adapter, buffer_info);
2975         }
2976
2977         return 0;
2978 }
2979
2980 static void e1000_tx_queue(struct e1000_adapter *adapter,
2981                            struct e1000_tx_ring *tx_ring, int tx_flags,
2982                            int count)
2983 {
2984         struct e1000_tx_desc *tx_desc = NULL;
2985         struct e1000_tx_buffer *buffer_info;
2986         u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
2987         unsigned int i;
2988
2989         if (likely(tx_flags & E1000_TX_FLAGS_TSO)) {
2990                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
2991                              E1000_TXD_CMD_TSE;
2992                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
2993
2994                 if (likely(tx_flags & E1000_TX_FLAGS_IPV4))
2995                         txd_upper |= E1000_TXD_POPTS_IXSM << 8;
2996         }
2997
2998         if (likely(tx_flags & E1000_TX_FLAGS_CSUM)) {
2999                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
3000                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
3001         }
3002
3003         if (unlikely(tx_flags & E1000_TX_FLAGS_VLAN)) {
3004                 txd_lower |= E1000_TXD_CMD_VLE;
3005                 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
3006         }
3007
3008         if (unlikely(tx_flags & E1000_TX_FLAGS_NO_FCS))
3009                 txd_lower &= ~(E1000_TXD_CMD_IFCS);
3010
3011         i = tx_ring->next_to_use;
3012
3013         while (count--) {
3014                 buffer_info = &tx_ring->buffer_info[i];
3015                 tx_desc = E1000_TX_DESC(*tx_ring, i);
3016                 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
3017                 tx_desc->lower.data =
3018                         cpu_to_le32(txd_lower | buffer_info->length);
3019                 tx_desc->upper.data = cpu_to_le32(txd_upper);
3020                 if (unlikely(++i == tx_ring->count))
3021                         i = 0;
3022         }
3023
3024         tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
3025
3026         /* txd_cmd re-enables FCS, so we'll re-disable it here as desired. */
3027         if (unlikely(tx_flags & E1000_TX_FLAGS_NO_FCS))
3028                 tx_desc->lower.data &= ~(cpu_to_le32(E1000_TXD_CMD_IFCS));
3029
3030         /* Force memory writes to complete before letting h/w
3031          * know there are new descriptors to fetch.  (Only
3032          * applicable for weak-ordered memory model archs,
3033          * such as IA-64).
3034          */
3035         wmb();
3036
3037         tx_ring->next_to_use = i;
3038 }
3039
3040 /* 82547 workaround to avoid controller hang in half-duplex environment.
3041  * The workaround is to avoid queuing a large packet that would span
3042  * the internal Tx FIFO ring boundary by notifying the stack to resend
3043  * the packet at a later time.  This gives the Tx FIFO an opportunity to
3044  * flush all packets.  When that occurs, we reset the Tx FIFO pointers
3045  * to the beginning of the Tx FIFO.
3046  */
3047
3048 #define E1000_FIFO_HDR                  0x10
3049 #define E1000_82547_PAD_LEN             0x3E0
3050
3051 static int e1000_82547_fifo_workaround(struct e1000_adapter *adapter,
3052                                        struct sk_buff *skb)
3053 {
3054         u32 fifo_space = adapter->tx_fifo_size - adapter->tx_fifo_head;
3055         u32 skb_fifo_len = skb->len + E1000_FIFO_HDR;
3056
3057         skb_fifo_len = ALIGN(skb_fifo_len, E1000_FIFO_HDR);
3058
3059         if (adapter->link_duplex != HALF_DUPLEX)
3060                 goto no_fifo_stall_required;
3061
3062         if (atomic_read(&adapter->tx_fifo_stall))
3063                 return 1;
3064
3065         if (skb_fifo_len >= (E1000_82547_PAD_LEN + fifo_space)) {
3066                 atomic_set(&adapter->tx_fifo_stall, 1);
3067                 return 1;
3068         }
3069
3070 no_fifo_stall_required:
3071         adapter->tx_fifo_head += skb_fifo_len;
3072         if (adapter->tx_fifo_head >= adapter->tx_fifo_size)
3073                 adapter->tx_fifo_head -= adapter->tx_fifo_size;
3074         return 0;
3075 }
3076
3077 static int __e1000_maybe_stop_tx(struct net_device *netdev, int size)
3078 {
3079         struct e1000_adapter *adapter = netdev_priv(netdev);
3080         struct e1000_tx_ring *tx_ring = adapter->tx_ring;
3081
3082         netif_stop_queue(netdev);
3083         /* Herbert's original patch had:
3084          *  smp_mb__after_netif_stop_queue();
3085          * but since that doesn't exist yet, just open code it.
3086          */
3087         smp_mb();
3088
3089         /* We need to check again in a case another CPU has just
3090          * made room available.
3091          */
3092         if (likely(E1000_DESC_UNUSED(tx_ring) < size))
3093                 return -EBUSY;
3094
3095         /* A reprieve! */
3096         netif_start_queue(netdev);
3097         ++adapter->restart_queue;
3098         return 0;
3099 }
3100
3101 static int e1000_maybe_stop_tx(struct net_device *netdev,
3102                                struct e1000_tx_ring *tx_ring, int size)
3103 {
3104         if (likely(E1000_DESC_UNUSED(tx_ring) >= size))
3105                 return 0;
3106         return __e1000_maybe_stop_tx(netdev, size);
3107 }
3108
3109 #define TXD_USE_COUNT(S, X) (((S) + ((1 << (X)) - 1)) >> (X))
3110 static netdev_tx_t e1000_xmit_frame(struct sk_buff *skb,
3111                                     struct net_device *netdev)
3112 {
3113         struct e1000_adapter *adapter = netdev_priv(netdev);
3114         struct e1000_hw *hw = &adapter->hw;
3115         struct e1000_tx_ring *tx_ring;
3116         unsigned int first, max_per_txd = E1000_MAX_DATA_PER_TXD;
3117         unsigned int max_txd_pwr = E1000_MAX_TXD_PWR;
3118         unsigned int tx_flags = 0;
3119         unsigned int len = skb_headlen(skb);
3120         unsigned int nr_frags;
3121         unsigned int mss;
3122         int count = 0;
3123         int tso;
3124         unsigned int f;
3125         __be16 protocol = vlan_get_protocol(skb);
3126
3127         /* This goes back to the question of how to logically map a Tx queue
3128          * to a flow.  Right now, performance is impacted slightly negatively
3129          * if using multiple Tx queues.  If the stack breaks away from a
3130          * single qdisc implementation, we can look at this again.
3131          */
3132         tx_ring = adapter->tx_ring;
3133
3134         /* On PCI/PCI-X HW, if packet size is less than ETH_ZLEN,
3135          * packets may get corrupted during padding by HW.
3136          * To WA this issue, pad all small packets manually.
3137          */
3138         if (eth_skb_pad(skb))
3139                 return NETDEV_TX_OK;
3140
3141         mss = skb_shinfo(skb)->gso_size;
3142         /* The controller does a simple calculation to
3143          * make sure there is enough room in the FIFO before
3144          * initiating the DMA for each buffer.  The calc is:
3145          * 4 = ceil(buffer len/mss).  To make sure we don't
3146          * overrun the FIFO, adjust the max buffer len if mss
3147          * drops.
3148          */
3149         if (mss) {
3150                 u8 hdr_len;
3151                 max_per_txd = min(mss << 2, max_per_txd);
3152                 max_txd_pwr = fls(max_per_txd) - 1;
3153
3154                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
3155                 if (skb->data_len && hdr_len == len) {
3156                         switch (hw->mac_type) {
3157                         case e1000_82544: {
3158                                 unsigned int pull_size;
3159
3160                                 /* Make sure we have room to chop off 4 bytes,
3161                                  * and that the end alignment will work out to
3162                                  * this hardware's requirements
3163                                  * NOTE: this is a TSO only workaround
3164                                  * if end byte alignment not correct move us
3165                                  * into the next dword
3166                                  */
3167                                 if ((unsigned long)(skb_tail_pointer(skb) - 1)
3168                                     & 4)
3169                                         break;
3170                                 /* fall through */
3171                                 pull_size = min((unsigned int)4, skb->data_len);
3172                                 if (!__pskb_pull_tail(skb, pull_size)) {
3173                                         e_err(drv, "__pskb_pull_tail "
3174                                               "failed.\n");
3175                                         dev_kfree_skb_any(skb);
3176                                         return NETDEV_TX_OK;
3177                                 }
3178                                 len = skb_headlen(skb);
3179                                 break;
3180                         }
3181                         default:
3182                                 /* do nothing */
3183                                 break;
3184                         }
3185                 }
3186         }
3187
3188         /* reserve a descriptor for the offload context */
3189         if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
3190                 count++;
3191         count++;
3192
3193         /* Controller Erratum workaround */
3194         if (!skb->data_len && tx_ring->last_tx_tso && !skb_is_gso(skb))
3195                 count++;
3196
3197         count += TXD_USE_COUNT(len, max_txd_pwr);
3198
3199         if (adapter->pcix_82544)
3200                 count++;
3201
3202         /* work-around for errata 10 and it applies to all controllers
3203          * in PCI-X mode, so add one more descriptor to the count
3204          */
3205         if (unlikely((hw->bus_type == e1000_bus_type_pcix) &&
3206                         (len > 2015)))
3207                 count++;
3208
3209         nr_frags = skb_shinfo(skb)->nr_frags;
3210         for (f = 0; f < nr_frags; f++)
3211                 count += TXD_USE_COUNT(skb_frag_size(&skb_shinfo(skb)->frags[f]),
3212                                        max_txd_pwr);
3213         if (adapter->pcix_82544)
3214                 count += nr_frags;
3215
3216         /* need: count + 2 desc gap to keep tail from touching
3217          * head, otherwise try next time
3218          */
3219         if (unlikely(e1000_maybe_stop_tx(netdev, tx_ring, count + 2)))
3220                 return NETDEV_TX_BUSY;
3221
3222         if (unlikely((hw->mac_type == e1000_82547) &&
3223                      (e1000_82547_fifo_workaround(adapter, skb)))) {
3224                 netif_stop_queue(netdev);
3225                 if (!test_bit(__E1000_DOWN, &adapter->flags))
3226                         schedule_delayed_work(&adapter->fifo_stall_task, 1);
3227                 return NETDEV_TX_BUSY;
3228         }
3229
3230         if (skb_vlan_tag_present(skb)) {
3231                 tx_flags |= E1000_TX_FLAGS_VLAN;
3232                 tx_flags |= (skb_vlan_tag_get(skb) <<
3233                              E1000_TX_FLAGS_VLAN_SHIFT);
3234         }
3235
3236         first = tx_ring->next_to_use;
3237
3238         tso = e1000_tso(adapter, tx_ring, skb, protocol);
3239         if (tso < 0) {
3240                 dev_kfree_skb_any(skb);
3241                 return NETDEV_TX_OK;
3242         }
3243
3244         if (likely(tso)) {
3245                 if (likely(hw->mac_type != e1000_82544))
3246                         tx_ring->last_tx_tso = true;
3247                 tx_flags |= E1000_TX_FLAGS_TSO;
3248         } else if (likely(e1000_tx_csum(adapter, tx_ring, skb, protocol)))
3249                 tx_flags |= E1000_TX_FLAGS_CSUM;
3250
3251         if (protocol == htons(ETH_P_IP))
3252                 tx_flags |= E1000_TX_FLAGS_IPV4;
3253
3254         if (unlikely(skb->no_fcs))
3255                 tx_flags |= E1000_TX_FLAGS_NO_FCS;
3256
3257         count = e1000_tx_map(adapter, tx_ring, skb, first, max_per_txd,
3258                              nr_frags, mss);
3259
3260         if (count) {
3261                 /* The descriptors needed is higher than other Intel drivers
3262                  * due to a number of workarounds.  The breakdown is below:
3263                  * Data descriptors: MAX_SKB_FRAGS + 1
3264                  * Context Descriptor: 1
3265                  * Keep head from touching tail: 2
3266                  * Workarounds: 3
3267                  */
3268                 int desc_needed = MAX_SKB_FRAGS + 7;
3269
3270                 netdev_sent_queue(netdev, skb->len);
3271                 skb_tx_timestamp(skb);
3272
3273                 e1000_tx_queue(adapter, tx_ring, tx_flags, count);
3274
3275                 /* 82544 potentially requires twice as many data descriptors
3276                  * in order to guarantee buffers don't end on evenly-aligned
3277                  * dwords
3278                  */
3279                 if (adapter->pcix_82544)
3280                         desc_needed += MAX_SKB_FRAGS + 1;
3281
3282                 /* Make sure there is space in the ring for the next send. */
3283                 e1000_maybe_stop_tx(netdev, tx_ring, desc_needed);
3284
3285                 if (!skb->xmit_more ||
3286                     netif_xmit_stopped(netdev_get_tx_queue(netdev, 0))) {
3287                         writel(tx_ring->next_to_use, hw->hw_addr + tx_ring->tdt);
3288                         /* we need this if more than one processor can write to
3289                          * our tail at a time, it synchronizes IO on IA64/Altix
3290                          * systems
3291                          */
3292                         mmiowb();
3293                 }
3294         } else {
3295                 dev_kfree_skb_any(skb);
3296                 tx_ring->buffer_info[first].time_stamp = 0;
3297                 tx_ring->next_to_use = first;
3298         }
3299
3300         return NETDEV_TX_OK;
3301 }
3302
3303 #define NUM_REGS 38 /* 1 based count */
3304 static void e1000_regdump(struct e1000_adapter *adapter)
3305 {
3306         struct e1000_hw *hw = &adapter->hw;
3307         u32 regs[NUM_REGS];
3308         u32 *regs_buff = regs;
3309         int i = 0;
3310
3311         static const char * const reg_name[] = {
3312                 "CTRL",  "STATUS",
3313                 "RCTL", "RDLEN", "RDH", "RDT", "RDTR",
3314                 "TCTL", "TDBAL", "TDBAH", "TDLEN", "TDH", "TDT",
3315                 "TIDV", "TXDCTL", "TADV", "TARC0",
3316                 "TDBAL1", "TDBAH1", "TDLEN1", "TDH1", "TDT1",
3317                 "TXDCTL1", "TARC1",
3318                 "CTRL_EXT", "ERT", "RDBAL", "RDBAH",
3319                 "TDFH", "TDFT", "TDFHS", "TDFTS", "TDFPC",
3320                 "RDFH", "RDFT", "RDFHS", "RDFTS", "RDFPC"
3321         };
3322
3323         regs_buff[0]  = er32(CTRL);
3324         regs_buff[1]  = er32(STATUS);
3325
3326         regs_buff[2]  = er32(RCTL);
3327         regs_buff[3]  = er32(RDLEN);
3328         regs_buff[4]  = er32(RDH);
3329         regs_buff[5]  = er32(RDT);
3330         regs_buff[6]  = er32(RDTR);
3331
3332         regs_buff[7]  = er32(TCTL);
3333         regs_buff[8]  = er32(TDBAL);
3334         regs_buff[9]  = er32(TDBAH);
3335         regs_buff[10] = er32(TDLEN);
3336         regs_buff[11] = er32(TDH);
3337         regs_buff[12] = er32(TDT);
3338         regs_buff[13] = er32(TIDV);
3339         regs_buff[14] = er32(TXDCTL);
3340         regs_buff[15] = er32(TADV);
3341         regs_buff[16] = er32(TARC0);
3342
3343         regs_buff[17] = er32(TDBAL1);
3344         regs_buff[18] = er32(TDBAH1);
3345         regs_buff[19] = er32(TDLEN1);
3346         regs_buff[20] = er32(TDH1);
3347         regs_buff[21] = er32(TDT1);
3348         regs_buff[22] = er32(TXDCTL1);
3349         regs_buff[23] = er32(TARC1);
3350         regs_buff[24] = er32(CTRL_EXT);
3351         regs_buff[25] = er32(ERT);
3352         regs_buff[26] = er32(RDBAL0);
3353         regs_buff[27] = er32(RDBAH0);
3354         regs_buff[28] = er32(TDFH);
3355         regs_buff[29] = er32(TDFT);
3356         regs_buff[30] = er32(TDFHS);
3357         regs_buff[31] = er32(TDFTS);
3358         regs_buff[32] = er32(TDFPC);
3359         regs_buff[33] = er32(RDFH);
3360         regs_buff[34] = er32(RDFT);
3361         regs_buff[35] = er32(RDFHS);
3362         regs_buff[36] = er32(RDFTS);
3363         regs_buff[37] = er32(RDFPC);
3364
3365         pr_info("Register dump\n");
3366         for (i = 0; i < NUM_REGS; i++)
3367                 pr_info("%-15s  %08x\n", reg_name[i], regs_buff[i]);
3368 }
3369
3370 /*
3371  * e1000_dump: Print registers, tx ring and rx ring
3372  */
3373 static void e1000_dump(struct e1000_adapter *adapter)
3374 {
3375         /* this code doesn't handle multiple rings */
3376         struct e1000_tx_ring *tx_ring = adapter->tx_ring;
3377         struct e1000_rx_ring *rx_ring = adapter->rx_ring;
3378         int i;
3379
3380         if (!netif_msg_hw(adapter))
3381                 return;
3382
3383         /* Print Registers */
3384         e1000_regdump(adapter);
3385
3386         /* transmit dump */
3387         pr_info("TX Desc ring0 dump\n");
3388
3389         /* Transmit Descriptor Formats - DEXT[29] is 0 (Legacy) or 1 (Extended)
3390          *
3391          * Legacy Transmit Descriptor
3392          *   +--------------------------------------------------------------+
3393          * 0 |         Buffer Address [63:0] (Reserved on Write Back)       |
3394          *   +--------------------------------------------------------------+
3395          * 8 | Special  |    CSS     | Status |  CMD    |  CSO   |  Length  |
3396          *   +--------------------------------------------------------------+
3397          *   63       48 47        36 35    32 31     24 23    16 15        0
3398          *
3399          * Extended Context Descriptor (DTYP=0x0) for TSO or checksum offload
3400          *   63      48 47    40 39       32 31             16 15    8 7      0
3401          *   +----------------------------------------------------------------+
3402          * 0 |  TUCSE  | TUCS0  |   TUCSS   |     IPCSE       | IPCS0 | IPCSS |
3403          *   +----------------------------------------------------------------+
3404          * 8 |   MSS   | HDRLEN | RSV | STA | TUCMD | DTYP |      PAYLEN      |
3405          *   +----------------------------------------------------------------+
3406          *   63      48 47    40 39 36 35 32 31   24 23  20 19                0
3407          *
3408          * Extended Data Descriptor (DTYP=0x1)
3409          *   +----------------------------------------------------------------+
3410          * 0 |                     Buffer Address [63:0]                      |
3411          *   +----------------------------------------------------------------+
3412          * 8 | VLAN tag |  POPTS  | Rsvd | Status | Command | DTYP |  DTALEN  |
3413          *   +----------------------------------------------------------------+
3414          *   63       48 47     40 39  36 35    32 31     24 23  20 19        0
3415          */
3416         pr_info("Tc[desc]     [Ce CoCsIpceCoS] [MssHlRSCm0Plen] [bi->dma       ] leng  ntw timestmp         bi->skb\n");
3417         pr_info("Td[desc]     [address 63:0  ] [VlaPoRSCm1Dlen] [bi->dma       ] leng  ntw timestmp         bi->skb\n");
3418
3419         if (!netif_msg_tx_done(adapter))
3420                 goto rx_ring_summary;
3421
3422         for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) {
3423                 struct e1000_tx_desc *tx_desc = E1000_TX_DESC(*tx_ring, i);
3424                 struct e1000_tx_buffer *buffer_info = &tx_ring->buffer_info[i];
3425                 struct my_u { __le64 a; __le64 b; };
3426                 struct my_u *u = (struct my_u *)tx_desc;
3427                 const char *type;
3428
3429                 if (i == tx_ring->next_to_use && i == tx_ring->next_to_clean)
3430                         type = "NTC/U";
3431                 else if (i == tx_ring->next_to_use)
3432                         type = "NTU";
3433                 else if (i == tx_ring->next_to_clean)
3434                         type = "NTC";
3435                 else
3436                         type = "";
3437
3438                 pr_info("T%c[0x%03X]    %016llX %016llX %016llX %04X  %3X %016llX %p %s\n",
3439                         ((le64_to_cpu(u->b) & (1<<20)) ? 'd' : 'c'), i,
3440                         le64_to_cpu(u->a), le64_to_cpu(u->b),
3441                         (u64)buffer_info->dma, buffer_info->length,
3442                         buffer_info->next_to_watch,
3443                         (u64)buffer_info->time_stamp, buffer_info->skb, type);
3444         }
3445
3446 rx_ring_summary:
3447         /* receive dump */
3448         pr_info("\nRX Desc ring dump\n");
3449
3450         /* Legacy Receive Descriptor Format
3451          *
3452          * +-----------------------------------------------------+
3453          * |                Buffer Address [63:0]                |
3454          * +-----------------------------------------------------+
3455          * | VLAN Tag | Errors | Status 0 | Packet csum | Length |
3456          * +-----------------------------------------------------+
3457          * 63       48 47    40 39      32 31         16 15      0
3458          */
3459         pr_info("R[desc]      [address 63:0  ] [vl er S cks ln] [bi->dma       ] [bi->skb]\n");
3460
3461         if (!netif_msg_rx_status(adapter))
3462                 goto exit;
3463
3464         for (i = 0; rx_ring->desc && (i < rx_ring->count); i++) {
3465                 struct e1000_rx_desc *rx_desc = E1000_RX_DESC(*rx_ring, i);
3466                 struct e1000_rx_buffer *buffer_info = &rx_ring->buffer_info[i];
3467                 struct my_u { __le64 a; __le64 b; };
3468                 struct my_u *u = (struct my_u *)rx_desc;
3469                 const char *type;
3470
3471                 if (i == rx_ring->next_to_use)
3472                         type = "NTU";
3473                 else if (i == rx_ring->next_to_clean)
3474                         type = "NTC";
3475                 else
3476                         type = "";
3477
3478                 pr_info("R[0x%03X]     %016llX %016llX %016llX %p %s\n",
3479                         i, le64_to_cpu(u->a), le64_to_cpu(u->b),
3480                         (u64)buffer_info->dma, buffer_info->rxbuf.data, type);
3481         } /* for */
3482
3483         /* dump the descriptor caches */
3484         /* rx */
3485         pr_info("Rx descriptor cache in 64bit format\n");
3486         for (i = 0x6000; i <= 0x63FF ; i += 0x10) {
3487                 pr_info("R%04X: %08X|%08X %08X|%08X\n",
3488                         i,
3489                         readl(adapter->hw.hw_addr + i+4),
3490                         readl(adapter->hw.hw_addr + i),
3491                         readl(adapter->hw.hw_addr + i+12),
3492                         readl(adapter->hw.hw_addr + i+8));
3493         }
3494         /* tx */
3495         pr_info("Tx descriptor cache in 64bit format\n");
3496         for (i = 0x7000; i <= 0x73FF ; i += 0x10) {
3497                 pr_info("T%04X: %08X|%08X %08X|%08X\n",
3498                         i,
3499                         readl(adapter->hw.hw_addr + i+4),
3500                         readl(adapter->hw.hw_addr + i),
3501                         readl(adapter->hw.hw_addr + i+12),
3502                         readl(adapter->hw.hw_addr + i+8));
3503         }
3504 exit:
3505         return;
3506 }
3507
3508 /**
3509  * e1000_tx_timeout - Respond to a Tx Hang
3510  * @netdev: network interface device structure
3511  **/
3512 static void e1000_tx_timeout(struct net_device *netdev)
3513 {
3514         struct e1000_adapter *adapter = netdev_priv(netdev);
3515
3516         /* Do the reset outside of interrupt context */
3517         adapter->tx_timeout_count++;
3518         schedule_work(&adapter->reset_task);
3519 }
3520
3521 static void e1000_reset_task(struct work_struct *work)
3522 {
3523         struct e1000_adapter *adapter =
3524                 container_of(work, struct e1000_adapter, reset_task);
3525
3526         e_err(drv, "Reset adapter\n");
3527         e1000_reinit_locked(adapter);
3528 }
3529
3530 /**
3531  * e1000_change_mtu - Change the Maximum Transfer Unit
3532  * @netdev: network interface device structure
3533  * @new_mtu: new value for maximum frame size
3534  *
3535  * Returns 0 on success, negative on failure
3536  **/
3537 static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
3538 {
3539         struct e1000_adapter *adapter = netdev_priv(netdev);
3540         struct e1000_hw *hw = &adapter->hw;
3541         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
3542
3543         /* Adapter-specific max frame size limits. */
3544         switch (hw->mac_type) {
3545         case e1000_undefined ... e1000_82542_rev2_1:
3546                 if (max_frame > (ETH_FRAME_LEN + ETH_FCS_LEN)) {
3547                         e_err(probe, "Jumbo Frames not supported.\n");
3548                         return -EINVAL;
3549                 }
3550                 break;
3551         default:
3552                 /* Capable of supporting up to MAX_JUMBO_FRAME_SIZE limit. */
3553                 break;
3554         }
3555
3556         while (test_and_set_bit(__E1000_RESETTING, &adapter->flags))
3557                 msleep(1);
3558         /* e1000_down has a dependency on max_frame_size */
3559         hw->max_frame_size = max_frame;
3560         if (netif_running(netdev)) {
3561                 /* prevent buffers from being reallocated */
3562                 adapter->alloc_rx_buf = e1000_alloc_dummy_rx_buffers;
3563                 e1000_down(adapter);
3564         }
3565
3566         /* NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
3567          * means we reserve 2 more, this pushes us to allocate from the next
3568          * larger slab size.
3569          * i.e. RXBUFFER_2048 --> size-4096 slab
3570          * however with the new *_jumbo_rx* routines, jumbo receives will use
3571          * fragmented skbs
3572          */
3573
3574         if (max_frame <= E1000_RXBUFFER_2048)
3575                 adapter->rx_buffer_len = E1000_RXBUFFER_2048;
3576         else
3577 #if (PAGE_SIZE >= E1000_RXBUFFER_16384)
3578                 adapter->rx_buffer_len = E1000_RXBUFFER_16384;
3579 #elif (PAGE_SIZE >= E1000_RXBUFFER_4096)
3580                 adapter->rx_buffer_len = PAGE_SIZE;
3581 #endif
3582
3583         /* adjust allocation if LPE protects us, and we aren't using SBP */
3584         if (!hw->tbi_compatibility_on &&
3585             ((max_frame == (ETH_FRAME_LEN + ETH_FCS_LEN)) ||
3586              (max_frame == MAXIMUM_ETHERNET_VLAN_SIZE)))
3587                 adapter->rx_buffer_len = MAXIMUM_ETHERNET_VLAN_SIZE;
3588
3589         pr_info("%s changing MTU from %d to %d\n",
3590                 netdev->name, netdev->mtu, new_mtu);
3591         netdev->mtu = new_mtu;
3592
3593         if (netif_running(netdev))
3594                 e1000_up(adapter);
3595         else
3596                 e1000_reset(adapter);
3597
3598         clear_bit(__E1000_RESETTING, &adapter->flags);
3599
3600         return 0;
3601 }
3602
3603 /**
3604  * e1000_update_stats - Update the board statistics counters
3605  * @adapter: board private structure
3606  **/
3607 void e1000_update_stats(struct e1000_adapter *adapter)
3608 {
3609         struct net_device *netdev = adapter->netdev;
3610         struct e1000_hw *hw = &adapter->hw;
3611         struct pci_dev *pdev = adapter->pdev;
3612         unsigned long flags;
3613         u16 phy_tmp;
3614
3615 #define PHY_IDLE_ERROR_COUNT_MASK 0x00FF
3616
3617         /* Prevent stats update while adapter is being reset, or if the pci
3618          * connection is down.
3619          */
3620         if (adapter->link_speed == 0)
3621                 return;
3622         if (pci_channel_offline(pdev))
3623                 return;
3624
3625         spin_lock_irqsave(&adapter->stats_lock, flags);
3626
3627         /* these counters are modified from e1000_tbi_adjust_stats,
3628          * called from the interrupt context, so they must only
3629          * be written while holding adapter->stats_lock
3630          */
3631
3632         adapter->stats.crcerrs += er32(CRCERRS);
3633         adapter->stats.gprc += er32(GPRC);
3634         adapter->stats.gorcl += er32(GORCL);
3635         adapter->stats.gorch += er32(GORCH);
3636         adapter->stats.bprc += er32(BPRC);
3637         adapter->stats.mprc += er32(MPRC);
3638         adapter->stats.roc += er32(ROC);
3639
3640         adapter->stats.prc64 += er32(PRC64);
3641         adapter->stats.prc127 += er32(PRC127);
3642         adapter->stats.prc255 += er32(PRC255);
3643         adapter->stats.prc511 += er32(PRC511);
3644         adapter->stats.prc1023 += er32(PRC1023);
3645         adapter->stats.prc1522 += er32(PRC1522);
3646
3647         adapter->stats.symerrs += er32(SYMERRS);
3648         adapter->stats.mpc += er32(MPC);
3649         adapter->stats.scc += er32(SCC);
3650         adapter->stats.ecol += er32(ECOL);
3651         adapter->stats.mcc += er32(MCC);
3652         adapter->stats.latecol += er32(LATECOL);
3653         adapter->stats.dc += er32(DC);
3654         adapter->stats.sec += er32(SEC);
3655         adapter->stats.rlec += er32(RLEC);
3656         adapter->stats.xonrxc += er32(XONRXC);
3657         adapter->stats.xontxc += er32(XONTXC);
3658         adapter->stats.xoffrxc += er32(XOFFRXC);
3659         adapter->stats.xofftxc += er32(XOFFTXC);
3660         adapter->stats.fcruc += er32(FCRUC);
3661         adapter->stats.gptc += er32(GPTC);
3662         adapter->stats.gotcl += er32(GOTCL);
3663         adapter->stats.gotch += er32(GOTCH);
3664         adapter->stats.rnbc += er32(RNBC);
3665         adapter->stats.ruc += er32(RUC);
3666         adapter->stats.rfc += er32(RFC);
3667         adapter->stats.rjc += er32(RJC);
3668         adapter->stats.torl += er32(TORL);
3669         adapter->stats.torh += er32(TORH);
3670         adapter->stats.totl += er32(TOTL);
3671         adapter->stats.toth += er32(TOTH);
3672         adapter->stats.tpr += er32(TPR);
3673
3674         adapter->stats.ptc64 += er32(PTC64);
3675         adapter->stats.ptc127 += er32(PTC127);
3676         adapter->stats.ptc255 += er32(PTC255);
3677         adapter->stats.ptc511 += er32(PTC511);
3678         adapter->stats.ptc1023 += er32(PTC1023);
3679         adapter->stats.ptc1522 += er32(PTC1522);
3680
3681         adapter->stats.mptc += er32(MPTC);
3682         adapter->stats.bptc += er32(BPTC);
3683
3684         /* used for adaptive IFS */
3685
3686         hw->tx_packet_delta = er32(TPT);
3687         adapter->stats.tpt += hw->tx_packet_delta;
3688         hw->collision_delta = er32(COLC);
3689         adapter->stats.colc += hw->collision_delta;
3690
3691         if (hw->mac_type >= e1000_82543) {
3692                 adapter->stats.algnerrc += er32(ALGNERRC);
3693                 adapter->stats.rxerrc += er32(RXERRC);
3694                 adapter->stats.tncrs += er32(TNCRS);
3695                 adapter->stats.cexterr += er32(CEXTERR);
3696                 adapter->stats.tsctc += er32(TSCTC);
3697                 adapter->stats.tsctfc += er32(TSCTFC);
3698         }
3699
3700         /* Fill out the OS statistics structure */
3701         netdev->stats.multicast = adapter->stats.mprc;
3702         netdev->stats.collisions = adapter->stats.colc;
3703
3704         /* Rx Errors */
3705
3706         /* RLEC on some newer hardware can be incorrect so build
3707          * our own version based on RUC and ROC
3708          */
3709         netdev->stats.rx_errors = adapter->stats.rxerrc +
3710                 adapter->stats.crcerrs + adapter->stats.algnerrc +
3711                 adapter->stats.ruc + adapter->stats.roc +
3712                 adapter->stats.cexterr;
3713         adapter->stats.rlerrc = adapter->stats.ruc + adapter->stats.roc;
3714         netdev->stats.rx_length_errors = adapter->stats.rlerrc;
3715         netdev->stats.rx_crc_errors = adapter->stats.crcerrs;
3716         netdev->stats.rx_frame_errors = adapter->stats.algnerrc;
3717         netdev->stats.rx_missed_errors = adapter->stats.mpc;
3718
3719         /* Tx Errors */
3720         adapter->stats.txerrc = adapter->stats.ecol + adapter->stats.latecol;
3721         netdev->stats.tx_errors = adapter->stats.txerrc;
3722         netdev->stats.tx_aborted_errors = adapter->stats.ecol;
3723         netdev->stats.tx_window_errors = adapter->stats.latecol;
3724         netdev->stats.tx_carrier_errors = adapter->stats.tncrs;
3725         if (hw->bad_tx_carr_stats_fd &&
3726             adapter->link_duplex == FULL_DUPLEX) {
3727                 netdev->stats.tx_carrier_errors = 0;
3728                 adapter->stats.tncrs = 0;
3729         }
3730
3731         /* Tx Dropped needs to be maintained elsewhere */
3732
3733         /* Phy Stats */
3734         if (hw->media_type == e1000_media_type_copper) {
3735                 if ((adapter->link_speed == SPEED_1000) &&
3736                    (!e1000_read_phy_reg(hw, PHY_1000T_STATUS, &phy_tmp))) {
3737                         phy_tmp &= PHY_IDLE_ERROR_COUNT_MASK;
3738                         adapter->phy_stats.idle_errors += phy_tmp;
3739                 }
3740
3741                 if ((hw->mac_type <= e1000_82546) &&
3742                    (hw->phy_type == e1000_phy_m88) &&
3743                    !e1000_read_phy_reg(hw, M88E1000_RX_ERR_CNTR, &phy_tmp))
3744                         adapter->phy_stats.receive_errors += phy_tmp;
3745         }
3746
3747         /* Management Stats */
3748         if (hw->has_smbus) {
3749                 adapter->stats.mgptc += er32(MGTPTC);
3750                 adapter->stats.mgprc += er32(MGTPRC);
3751                 adapter->stats.mgpdc += er32(MGTPDC);
3752         }
3753
3754         spin_unlock_irqrestore(&adapter->stats_lock, flags);
3755 }
3756
3757 /**
3758  * e1000_intr - Interrupt Handler
3759  * @irq: interrupt number
3760  * @data: pointer to a network interface device structure
3761  **/
3762 static irqreturn_t e1000_intr(int irq, void *data)
3763 {
3764         struct net_device *netdev = data;
3765         struct e1000_adapter *adapter = netdev_priv(netdev);
3766         struct e1000_hw *hw = &adapter->hw;
3767         u32 icr = er32(ICR);
3768
3769         if (unlikely((!icr)))
3770                 return IRQ_NONE;  /* Not our interrupt */
3771
3772         /* we might have caused the interrupt, but the above
3773          * read cleared it, and just in case the driver is
3774          * down there is nothing to do so return handled
3775          */
3776         if (unlikely(test_bit(__E1000_DOWN, &adapter->flags)))
3777                 return IRQ_HANDLED;
3778
3779         if (unlikely(icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC))) {
3780                 hw->get_link_status = 1;
3781                 /* guard against interrupt when we're going down */
3782                 if (!test_bit(__E1000_DOWN, &adapter->flags))
3783                         schedule_delayed_work(&adapter->watchdog_task, 1);
3784         }
3785
3786         /* disable interrupts, without the synchronize_irq bit */
3787         ew32(IMC, ~0);
3788         E1000_WRITE_FLUSH();
3789
3790         if (likely(napi_schedule_prep(&adapter->napi))) {
3791                 adapter->total_tx_bytes = 0;
3792                 adapter->total_tx_packets = 0;
3793                 adapter->total_rx_bytes = 0;
3794                 adapter->total_rx_packets = 0;
3795                 __napi_schedule(&adapter->napi);
3796         } else {
3797                 /* this really should not happen! if it does it is basically a
3798                  * bug, but not a hard error, so enable ints and continue
3799                  */
3800                 if (!test_bit(__E1000_DOWN, &adapter->flags))
3801                         e1000_irq_enable(adapter);
3802         }
3803
3804         return IRQ_HANDLED;
3805 }
3806
3807 /**
3808  * e1000_clean - NAPI Rx polling callback
3809  * @adapter: board private structure
3810  **/
3811 static int e1000_clean(struct napi_struct *napi, int budget)
3812 {
3813         struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter,
3814                                                      napi);
3815         int tx_clean_complete = 0, work_done = 0;
3816
3817         tx_clean_complete = e1000_clean_tx_irq(adapter, &adapter->tx_ring[0]);
3818
3819         adapter->clean_rx(adapter, &adapter->rx_ring[0], &work_done, budget);
3820
3821         if (!tx_clean_complete)
3822                 work_done = budget;
3823
3824         /* If budget not fully consumed, exit the polling mode */
3825         if (work_done < budget) {
3826                 if (likely(adapter->itr_setting & 3))
3827                         e1000_set_itr(adapter);
3828                 napi_complete_done(napi, work_done);
3829                 if (!test_bit(__E1000_DOWN, &adapter->flags))
3830                         e1000_irq_enable(adapter);
3831         }
3832
3833         return work_done;
3834 }
3835
3836 /**
3837  * e1000_clean_tx_irq - Reclaim resources after transmit completes
3838  * @adapter: board private structure
3839  **/
3840 static bool e1000_clean_tx_irq(struct e1000_adapter *adapter,
3841                                struct e1000_tx_ring *tx_ring)
3842 {
3843         struct e1000_hw *hw = &adapter->hw;
3844         struct net_device *netdev = adapter->netdev;
3845         struct e1000_tx_desc *tx_desc, *eop_desc;
3846         struct e1000_tx_buffer *buffer_info;
3847         unsigned int i, eop;
3848         unsigned int count = 0;
3849         unsigned int total_tx_bytes = 0, total_tx_packets = 0;
3850         unsigned int bytes_compl = 0, pkts_compl = 0;
3851
3852         i = tx_ring->next_to_clean;
3853         eop = tx_ring->buffer_info[i].next_to_watch;
3854         eop_desc = E1000_TX_DESC(*tx_ring, eop);
3855
3856         while ((eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) &&
3857                (count < tx_ring->count)) {
3858                 bool cleaned = false;
3859                 dma_rmb();      /* read buffer_info after eop_desc */
3860                 for ( ; !cleaned; count++) {
3861                         tx_desc = E1000_TX_DESC(*tx_ring, i);
3862                         buffer_info = &tx_ring->buffer_info[i];
3863                         cleaned = (i == eop);
3864
3865                         if (cleaned) {
3866                                 total_tx_packets += buffer_info->segs;
3867                                 total_tx_bytes += buffer_info->bytecount;
3868                                 if (buffer_info->skb) {
3869                                         bytes_compl += buffer_info->skb->len;
3870                                         pkts_compl++;
3871                                 }
3872
3873                         }
3874                         e1000_unmap_and_free_tx_resource(adapter, buffer_info);
3875                         tx_desc->upper.data = 0;
3876
3877                         if (unlikely(++i == tx_ring->count))
3878                                 i = 0;
3879                 }
3880
3881                 eop = tx_ring->buffer_info[i].next_to_watch;
3882                 eop_desc = E1000_TX_DESC(*tx_ring, eop);
3883         }
3884
3885         /* Synchronize with E1000_DESC_UNUSED called from e1000_xmit_frame,
3886          * which will reuse the cleaned buffers.
3887          */
3888         smp_store_release(&tx_ring->next_to_clean, i);
3889
3890         netdev_completed_queue(netdev, pkts_compl, bytes_compl);
3891
3892 #define TX_WAKE_THRESHOLD 32
3893         if (unlikely(count && netif_carrier_ok(netdev) &&
3894                      E1000_DESC_UNUSED(tx_ring) >= TX_WAKE_THRESHOLD)) {
3895                 /* Make sure that anybody stopping the queue after this
3896                  * sees the new next_to_clean.
3897                  */
3898                 smp_mb();
3899
3900                 if (netif_queue_stopped(netdev) &&
3901                     !(test_bit(__E1000_DOWN, &adapter->flags))) {
3902                         netif_wake_queue(netdev);
3903                         ++adapter->restart_queue;
3904                 }
3905         }
3906
3907         if (adapter->detect_tx_hung) {
3908                 /* Detect a transmit hang in hardware, this serializes the
3909                  * check with the clearing of time_stamp and movement of i
3910                  */
3911                 adapter->detect_tx_hung = false;
3912                 if (tx_ring->buffer_info[eop].time_stamp &&
3913                     time_after(jiffies, tx_ring->buffer_info[eop].time_stamp +
3914                                (adapter->tx_timeout_factor * HZ)) &&
3915                     !(er32(STATUS) & E1000_STATUS_TXOFF)) {
3916
3917                         /* detected Tx unit hang */
3918                         e_err(drv, "Detected Tx Unit Hang\n"
3919                               "  Tx Queue             <%lu>\n"
3920                               "  TDH                  <%x>\n"
3921                               "  TDT                  <%x>\n"
3922                               "  next_to_use          <%x>\n"
3923                               "  next_to_clean        <%x>\n"
3924                               "buffer_info[next_to_clean]\n"
3925                               "  time_stamp           <%lx>\n"
3926                               "  next_to_watch        <%x>\n"
3927                               "  jiffies              <%lx>\n"
3928                               "  next_to_watch.status <%x>\n",
3929                                 (unsigned long)(tx_ring - adapter->tx_ring),
3930                                 readl(hw->hw_addr + tx_ring->tdh),
3931                                 readl(hw->hw_addr + tx_ring->tdt),
3932                                 tx_ring->next_to_use,
3933                                 tx_ring->next_to_clean,
3934                                 tx_ring->buffer_info[eop].time_stamp,
3935                                 eop,
3936                                 jiffies,
3937                                 eop_desc->upper.fields.status);
3938                         e1000_dump(adapter);
3939                         netif_stop_queue(netdev);
3940                 }
3941         }
3942         adapter->total_tx_bytes += total_tx_bytes;
3943         adapter->total_tx_packets += total_tx_packets;
3944         netdev->stats.tx_bytes += total_tx_bytes;
3945         netdev->stats.tx_packets += total_tx_packets;
3946         return count < tx_ring->count;
3947 }
3948
3949 /**
3950  * e1000_rx_checksum - Receive Checksum Offload for 82543
3951  * @adapter:     board private structure
3952  * @status_err:  receive descriptor status and error fields
3953  * @csum:        receive descriptor csum field
3954  * @sk_buff:     socket buffer with received data
3955  **/
3956 static void e1000_rx_checksum(struct e1000_adapter *adapter, u32 status_err,
3957                               u32 csum, struct sk_buff *skb)
3958 {
3959         struct e1000_hw *hw = &adapter->hw;
3960         u16 status = (u16)status_err;
3961         u8 errors = (u8)(status_err >> 24);
3962
3963         skb_checksum_none_assert(skb);
3964
3965         /* 82543 or newer only */
3966         if (unlikely(hw->mac_type < e1000_82543))
3967                 return;
3968         /* Ignore Checksum bit is set */
3969         if (unlikely(status & E1000_RXD_STAT_IXSM))
3970                 return;
3971         /* TCP/UDP checksum error bit is set */
3972         if (unlikely(errors & E1000_RXD_ERR_TCPE)) {
3973                 /* let the stack verify checksum errors */
3974                 adapter->hw_csum_err++;
3975                 return;
3976         }
3977         /* TCP/UDP Checksum has not been calculated */
3978         if (!(status & E1000_RXD_STAT_TCPCS))
3979                 return;
3980
3981         /* It must be a TCP or UDP packet with a valid checksum */
3982         if (likely(status & E1000_RXD_STAT_TCPCS)) {
3983                 /* TCP checksum is good */
3984                 skb->ip_summed = CHECKSUM_UNNECESSARY;
3985         }
3986         adapter->hw_csum_good++;
3987 }
3988
3989 /**
3990  * e1000_consume_page - helper function for jumbo Rx path
3991  **/
3992 static void e1000_consume_page(struct e1000_rx_buffer *bi, struct sk_buff *skb,
3993                                u16 length)
3994 {
3995         bi->rxbuf.page = NULL;
3996         skb->len += length;
3997         skb->data_len += length;
3998         skb->truesize += PAGE_SIZE;
3999 }
4000
4001 /**
4002  * e1000_receive_skb - helper function to handle rx indications
4003  * @adapter: board private structure
4004  * @status: descriptor status field as written by hardware
4005  * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
4006  * @skb: pointer to sk_buff to be indicated to stack
4007  */
4008 static void e1000_receive_skb(struct e1000_adapter *adapter, u8 status,
4009                               __le16 vlan, struct sk_buff *skb)
4010 {
4011         skb->protocol = eth_type_trans(skb, adapter->netdev);
4012
4013         if (status & E1000_RXD_STAT_VP) {
4014                 u16 vid = le16_to_cpu(vlan) & E1000_RXD_SPC_VLAN_MASK;
4015
4016                 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vid);
4017         }
4018         napi_gro_receive(&adapter->napi, skb);
4019 }
4020
4021 /**
4022  * e1000_tbi_adjust_stats
4023  * @hw: Struct containing variables accessed by shared code
4024  * @frame_len: The length of the frame in question
4025  * @mac_addr: The Ethernet destination address of the frame in question
4026  *
4027  * Adjusts the statistic counters when a frame is accepted by TBI_ACCEPT
4028  */
4029 static void e1000_tbi_adjust_stats(struct e1000_hw *hw,
4030                                    struct e1000_hw_stats *stats,
4031                                    u32 frame_len, const u8 *mac_addr)
4032 {
4033         u64 carry_bit;
4034
4035         /* First adjust the frame length. */
4036         frame_len--;
4037         /* We need to adjust the statistics counters, since the hardware
4038          * counters overcount this packet as a CRC error and undercount
4039          * the packet as a good packet
4040          */
4041         /* This packet should not be counted as a CRC error. */
4042         stats->crcerrs--;
4043         /* This packet does count as a Good Packet Received. */
4044         stats->gprc++;
4045
4046         /* Adjust the Good Octets received counters */
4047         carry_bit = 0x80000000 & stats->gorcl;
4048         stats->gorcl += frame_len;
4049         /* If the high bit of Gorcl (the low 32 bits of the Good Octets
4050          * Received Count) was one before the addition,
4051          * AND it is zero after, then we lost the carry out,
4052          * need to add one to Gorch (Good Octets Received Count High).
4053          * This could be simplified if all environments supported
4054          * 64-bit integers.
4055          */
4056         if (carry_bit && ((stats->gorcl & 0x80000000) == 0))
4057                 stats->gorch++;
4058         /* Is this a broadcast or multicast?  Check broadcast first,
4059          * since the test for a multicast frame will test positive on
4060          * a broadcast frame.
4061          */
4062         if (is_broadcast_ether_addr(mac_addr))
4063                 stats->bprc++;
4064         else if (is_multicast_ether_addr(mac_addr))
4065                 stats->mprc++;
4066
4067         if (frame_len == hw->max_frame_size) {
4068                 /* In this case, the hardware has overcounted the number of
4069                  * oversize frames.
4070                  */
4071                 if (stats->roc > 0)
4072                         stats->roc--;
4073         }
4074
4075         /* Adjust the bin counters when the extra byte put the frame in the
4076          * wrong bin. Remember that the frame_len was adjusted above.
4077          */
4078         if (frame_len == 64) {
4079                 stats->prc64++;
4080                 stats->prc127--;
4081         } else if (frame_len == 127) {
4082                 stats->prc127++;
4083                 stats->prc255--;
4084         } else if (frame_len == 255) {
4085                 stats->prc255++;
4086                 stats->prc511--;
4087         } else if (frame_len == 511) {
4088                 stats->prc511++;
4089                 stats->prc1023--;
4090         } else if (frame_len == 1023) {
4091                 stats->prc1023++;
4092                 stats->prc1522--;
4093         } else if (frame_len == 1522) {
4094                 stats->prc1522++;
4095         }
4096 }
4097
4098 static bool e1000_tbi_should_accept(struct e1000_adapter *adapter,
4099                                     u8 status, u8 errors,
4100                                     u32 length, const u8 *data)
4101 {
4102         struct e1000_hw *hw = &adapter->hw;
4103         u8 last_byte = *(data + length - 1);
4104
4105         if (TBI_ACCEPT(hw, status, errors, length, last_byte)) {
4106                 unsigned long irq_flags;
4107
4108                 spin_lock_irqsave(&adapter->stats_lock, irq_flags);
4109                 e1000_tbi_adjust_stats(hw, &adapter->stats, length, data);
4110                 spin_unlock_irqrestore(&adapter->stats_lock, irq_flags);
4111
4112                 return true;
4113         }
4114
4115         return false;
4116 }
4117
4118 static struct sk_buff *e1000_alloc_rx_skb(struct e1000_adapter *adapter,
4119                                           unsigned int bufsz)
4120 {
4121         struct sk_buff *skb = napi_alloc_skb(&adapter->napi, bufsz);
4122
4123         if (unlikely(!skb))
4124                 adapter->alloc_rx_buff_failed++;
4125         return skb;
4126 }
4127
4128 /**
4129  * e1000_clean_jumbo_rx_irq - Send received data up the network stack; legacy
4130  * @adapter: board private structure
4131  * @rx_ring: ring to clean
4132  * @work_done: amount of napi work completed this call
4133  * @work_to_do: max amount of work allowed for this call to do
4134  *
4135  * the return value indicates whether actual cleaning was done, there
4136  * is no guarantee that everything was cleaned
4137  */
4138 static bool e1000_clean_jumbo_rx_irq(struct e1000_adapter *adapter,
4139                                      struct e1000_rx_ring *rx_ring,
4140                                      int *work_done, int work_to_do)
4141 {
4142         struct net_device *netdev = adapter->netdev;
4143         struct pci_dev *pdev = adapter->pdev;
4144         struct e1000_rx_desc *rx_desc, *next_rxd;
4145         struct e1000_rx_buffer *buffer_info, *next_buffer;
4146         u32 length;
4147         unsigned int i;
4148         int cleaned_count = 0;
4149         bool cleaned = false;
4150         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
4151
4152         i = rx_ring->next_to_clean;
4153         rx_desc = E1000_RX_DESC(*rx_ring, i);
4154         buffer_info = &rx_ring->buffer_info[i];
4155
4156         while (rx_desc->status & E1000_RXD_STAT_DD) {
4157                 struct sk_buff *skb;
4158                 u8 status;
4159
4160                 if (*work_done >= work_to_do)
4161                         break;
4162                 (*work_done)++;
4163                 dma_rmb(); /* read descriptor and rx_buffer_info after status DD */
4164
4165                 status = rx_desc->status;
4166
4167                 if (++i == rx_ring->count)
4168                         i = 0;
4169
4170                 next_rxd = E1000_RX_DESC(*rx_ring, i);
4171                 prefetch(next_rxd);
4172
4173                 next_buffer = &rx_ring->buffer_info[i];
4174
4175                 cleaned = true;
4176                 cleaned_count++;
4177                 dma_unmap_page(&pdev->dev, buffer_info->dma,
4178                                adapter->rx_buffer_len, DMA_FROM_DEVICE);
4179                 buffer_info->dma = 0;
4180
4181                 length = le16_to_cpu(rx_desc->length);
4182
4183                 /* errors is only valid for DD + EOP descriptors */
4184                 if (unlikely((status & E1000_RXD_STAT_EOP) &&
4185                     (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK))) {
4186                         u8 *mapped = page_address(buffer_info->rxbuf.page);
4187
4188                         if (e1000_tbi_should_accept(adapter, status,
4189                                                     rx_desc->errors,
4190                                                     length, mapped)) {
4191                                 length--;
4192                         } else if (netdev->features & NETIF_F_RXALL) {
4193                                 goto process_skb;
4194                         } else {
4195                                 /* an error means any chain goes out the window
4196                                  * too
4197                                  */
4198                                 if (rx_ring->rx_skb_top)
4199                                         dev_kfree_skb(rx_ring->rx_skb_top);
4200                                 rx_ring->rx_skb_top = NULL;
4201                                 goto next_desc;
4202                         }
4203                 }
4204
4205 #define rxtop rx_ring->rx_skb_top
4206 process_skb:
4207                 if (!(status & E1000_RXD_STAT_EOP)) {
4208                         /* this descriptor is only the beginning (or middle) */
4209                         if (!rxtop) {
4210                                 /* this is the beginning of a chain */
4211                                 rxtop = napi_get_frags(&adapter->napi);
4212                                 if (!rxtop)
4213                                         break;
4214
4215                                 skb_fill_page_desc(rxtop, 0,
4216                                                    buffer_info->rxbuf.page,
4217                                                    0, length);
4218                         } else {
4219                                 /* this is the middle of a chain */
4220                                 skb_fill_page_desc(rxtop,
4221                                     skb_shinfo(rxtop)->nr_frags,
4222                                     buffer_info->rxbuf.page, 0, length);
4223                         }
4224                         e1000_consume_page(buffer_info, rxtop, length);
4225                         goto next_desc;
4226                 } else {
4227                         if (rxtop) {
4228                                 /* end of the chain */
4229                                 skb_fill_page_desc(rxtop,
4230                                     skb_shinfo(rxtop)->nr_frags,
4231                                     buffer_info->rxbuf.page, 0, length);
4232                                 skb = rxtop;
4233                                 rxtop = NULL;
4234                                 e1000_consume_page(buffer_info, skb, length);
4235                         } else {
4236                                 struct page *p;
4237                                 /* no chain, got EOP, this buf is the packet
4238                                  * copybreak to save the put_page/alloc_page
4239                                  */
4240                                 p = buffer_info->rxbuf.page;
4241                                 if (length <= copybreak) {
4242                                         u8 *vaddr;
4243
4244                                         if (likely(!(netdev->features & NETIF_F_RXFCS)))
4245                                                 length -= 4;
4246                                         skb = e1000_alloc_rx_skb(adapter,
4247                                                                  length);
4248                                         if (!skb)
4249                                                 break;
4250
4251                                         vaddr = kmap_atomic(p);
4252                                         memcpy(skb_tail_pointer(skb), vaddr,
4253                                                length);
4254                                         kunmap_atomic(vaddr);
4255                                         /* re-use the page, so don't erase
4256                                          * buffer_info->rxbuf.page
4257                                          */
4258                                         skb_put(skb, length);
4259                                         e1000_rx_checksum(adapter,
4260                                                           status | rx_desc->errors << 24,
4261                                                           le16_to_cpu(rx_desc->csum), skb);
4262
4263                                         total_rx_bytes += skb->len;
4264                                         total_rx_packets++;
4265
4266                                         e1000_receive_skb(adapter, status,
4267                                                           rx_desc->special, skb);
4268                                         goto next_desc;
4269                                 } else {
4270                                         skb = napi_get_frags(&adapter->napi);
4271                                         if (!skb) {
4272                                                 adapter->alloc_rx_buff_failed++;
4273                                                 break;
4274                                         }
4275                                         skb_fill_page_desc(skb, 0, p, 0,
4276                                                            length);
4277                                         e1000_consume_page(buffer_info, skb,
4278                                                            length);
4279                                 }
4280                         }
4281                 }
4282
4283                 /* Receive Checksum Offload XXX recompute due to CRC strip? */
4284                 e1000_rx_checksum(adapter,
4285                                   (u32)(status) |
4286                                   ((u32)(rx_desc->errors) << 24),
4287                                   le16_to_cpu(rx_desc->csum), skb);
4288
4289                 total_rx_bytes += (skb->len - 4); /* don't count FCS */
4290                 if (likely(!(netdev->features & NETIF_F_RXFCS)))
4291                         pskb_trim(skb, skb->len - 4);
4292                 total_rx_packets++;
4293
4294                 if (status & E1000_RXD_STAT_VP) {
4295                         __le16 vlan = rx_desc->special;
4296                         u16 vid = le16_to_cpu(vlan) & E1000_RXD_SPC_VLAN_MASK;
4297
4298                         __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vid);
4299                 }
4300
4301                 napi_gro_frags(&adapter->napi);
4302
4303 next_desc:
4304                 rx_desc->status = 0;
4305
4306                 /* return some buffers to hardware, one at a time is too slow */
4307                 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
4308                         adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
4309                         cleaned_count = 0;
4310                 }
4311
4312                 /* use prefetched values */
4313                 rx_desc = next_rxd;
4314                 buffer_info = next_buffer;
4315         }
4316         rx_ring->next_to_clean = i;
4317
4318         cleaned_count = E1000_DESC_UNUSED(rx_ring);
4319         if (cleaned_count)
4320                 adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
4321
4322         adapter->total_rx_packets += total_rx_packets;
4323         adapter->total_rx_bytes += total_rx_bytes;
4324         netdev->stats.rx_bytes += total_rx_bytes;
4325         netdev->stats.rx_packets += total_rx_packets;
4326         return cleaned;
4327 }
4328
4329 /* this should improve performance for small packets with large amounts
4330  * of reassembly being done in the stack
4331  */
4332 static struct sk_buff *e1000_copybreak(struct e1000_adapter *adapter,
4333                                        struct e1000_rx_buffer *buffer_info,
4334                                        u32 length, const void *data)
4335 {
4336         struct sk_buff *skb;
4337
4338         if (length > copybreak)
4339                 return NULL;
4340
4341         skb = e1000_alloc_rx_skb(adapter, length);
4342         if (!skb)
4343                 return NULL;
4344
4345         dma_sync_single_for_cpu(&adapter->pdev->dev, buffer_info->dma,
4346                                 length, DMA_FROM_DEVICE);
4347
4348         skb_put_data(skb, data, length);
4349
4350         return skb;
4351 }
4352
4353 /**
4354  * e1000_clean_rx_irq - Send received data up the network stack; legacy
4355  * @adapter: board private structure
4356  * @rx_ring: ring to clean
4357  * @work_done: amount of napi work completed this call
4358  * @work_to_do: max amount of work allowed for this call to do
4359  */
4360 static bool e1000_clean_rx_irq(struct e1000_adapter *adapter,
4361                                struct e1000_rx_ring *rx_ring,
4362                                int *work_done, int work_to_do)
4363 {
4364         struct net_device *netdev = adapter->netdev;
4365         struct pci_dev *pdev = adapter->pdev;
4366         struct e1000_rx_desc *rx_desc, *next_rxd;
4367         struct e1000_rx_buffer *buffer_info, *next_buffer;
4368         u32 length;
4369         unsigned int i;
4370         int cleaned_count = 0;
4371         bool cleaned = false;
4372         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
4373
4374         i = rx_ring->next_to_clean;
4375         rx_desc = E1000_RX_DESC(*rx_ring, i);
4376         buffer_info = &rx_ring->buffer_info[i];
4377
4378         while (rx_desc->status & E1000_RXD_STAT_DD) {
4379                 struct sk_buff *skb;
4380                 u8 *data;
4381                 u8 status;
4382
4383                 if (*work_done >= work_to_do)
4384                         break;
4385                 (*work_done)++;
4386                 dma_rmb(); /* read descriptor and rx_buffer_info after status DD */
4387
4388                 status = rx_desc->status;
4389                 length = le16_to_cpu(rx_desc->length);
4390
4391                 data = buffer_info->rxbuf.data;
4392                 prefetch(data);
4393                 skb = e1000_copybreak(adapter, buffer_info, length, data);
4394                 if (!skb) {
4395                         unsigned int frag_len = e1000_frag_len(adapter);
4396
4397                         skb = build_skb(data - E1000_HEADROOM, frag_len);
4398                         if (!skb) {
4399                                 adapter->alloc_rx_buff_failed++;
4400                                 break;
4401                         }
4402
4403                         skb_reserve(skb, E1000_HEADROOM);
4404                         dma_unmap_single(&pdev->dev, buffer_info->dma,
4405                                          adapter->rx_buffer_len,
4406                                          DMA_FROM_DEVICE);
4407                         buffer_info->dma = 0;
4408                         buffer_info->rxbuf.data = NULL;
4409                 }
4410
4411                 if (++i == rx_ring->count)
4412                         i = 0;
4413
4414                 next_rxd = E1000_RX_DESC(*rx_ring, i);
4415                 prefetch(next_rxd);
4416
4417                 next_buffer = &rx_ring->buffer_info[i];
4418
4419                 cleaned = true;
4420                 cleaned_count++;
4421
4422                 /* !EOP means multiple descriptors were used to store a single
4423                  * packet, if thats the case we need to toss it.  In fact, we
4424                  * to toss every packet with the EOP bit clear and the next
4425                  * frame that _does_ have the EOP bit set, as it is by
4426                  * definition only a frame fragment
4427                  */
4428                 if (unlikely(!(status & E1000_RXD_STAT_EOP)))
4429                         adapter->discarding = true;
4430
4431                 if (adapter->discarding) {
4432                         /* All receives must fit into a single buffer */
4433                         netdev_dbg(netdev, "Receive packet consumed multiple buffers\n");
4434                         dev_kfree_skb(skb);
4435                         if (status & E1000_RXD_STAT_EOP)
4436                                 adapter->discarding = false;
4437                         goto next_desc;
4438                 }
4439
4440                 if (unlikely(rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK)) {
4441                         if (e1000_tbi_should_accept(adapter, status,
4442                                                     rx_desc->errors,
4443                                                     length, data)) {
4444                                 length--;
4445                         } else if (netdev->features & NETIF_F_RXALL) {
4446                                 goto process_skb;
4447                         } else {
4448                                 dev_kfree_skb(skb);
4449                                 goto next_desc;
4450                         }
4451                 }
4452
4453 process_skb:
4454                 total_rx_bytes += (length - 4); /* don't count FCS */
4455                 total_rx_packets++;
4456
4457                 if (likely(!(netdev->features & NETIF_F_RXFCS)))
4458                         /* adjust length to remove Ethernet CRC, this must be
4459                          * done after the TBI_ACCEPT workaround above
4460                          */
4461                         length -= 4;
4462
4463                 if (buffer_info->rxbuf.data == NULL)
4464                         skb_put(skb, length);
4465                 else /* copybreak skb */
4466                         skb_trim(skb, length);
4467
4468                 /* Receive Checksum Offload */
4469                 e1000_rx_checksum(adapter,
4470                                   (u32)(status) |
4471                                   ((u32)(rx_desc->errors) << 24),
4472                                   le16_to_cpu(rx_desc->csum), skb);
4473
4474                 e1000_receive_skb(adapter, status, rx_desc->special, skb);
4475
4476 next_desc:
4477                 rx_desc->status = 0;
4478
4479                 /* return some buffers to hardware, one at a time is too slow */
4480                 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
4481                         adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
4482                         cleaned_count = 0;
4483                 }
4484
4485                 /* use prefetched values */
4486                 rx_desc = next_rxd;
4487                 buffer_info = next_buffer;
4488         }
4489         rx_ring->next_to_clean = i;
4490
4491         cleaned_count = E1000_DESC_UNUSED(rx_ring);
4492         if (cleaned_count)
4493                 adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
4494
4495         adapter->total_rx_packets += total_rx_packets;
4496         adapter->total_rx_bytes += total_rx_bytes;
4497         netdev->stats.rx_bytes += total_rx_bytes;
4498         netdev->stats.rx_packets += total_rx_packets;
4499         return cleaned;
4500 }
4501
4502 /**
4503  * e1000_alloc_jumbo_rx_buffers - Replace used jumbo receive buffers
4504  * @adapter: address of board private structure
4505  * @rx_ring: pointer to receive ring structure
4506  * @cleaned_count: number of buffers to allocate this pass
4507  **/
4508 static void
4509 e1000_alloc_jumbo_rx_buffers(struct e1000_adapter *adapter,
4510                              struct e1000_rx_ring *rx_ring, int cleaned_count)
4511 {
4512         struct pci_dev *pdev = adapter->pdev;
4513         struct e1000_rx_desc *rx_desc;
4514         struct e1000_rx_buffer *buffer_info;
4515         unsigned int i;
4516
4517         i = rx_ring->next_to_use;
4518         buffer_info = &rx_ring->buffer_info[i];
4519
4520         while (cleaned_count--) {
4521                 /* allocate a new page if necessary */
4522                 if (!buffer_info->rxbuf.page) {
4523                         buffer_info->rxbuf.page = alloc_page(GFP_ATOMIC);
4524                         if (unlikely(!buffer_info->rxbuf.page)) {
4525                                 adapter->alloc_rx_buff_failed++;
4526                                 break;
4527                         }
4528                 }
4529
4530                 if (!buffer_info->dma) {
4531                         buffer_info->dma = dma_map_page(&pdev->dev,
4532                                                         buffer_info->rxbuf.page, 0,
4533                                                         adapter->rx_buffer_len,
4534                                                         DMA_FROM_DEVICE);
4535                         if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
4536                                 put_page(buffer_info->rxbuf.page);
4537                                 buffer_info->rxbuf.page = NULL;
4538                                 buffer_info->dma = 0;
4539                                 adapter->alloc_rx_buff_failed++;
4540                                 break;
4541                         }
4542                 }
4543
4544                 rx_desc = E1000_RX_DESC(*rx_ring, i);
4545                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
4546
4547                 if (unlikely(++i == rx_ring->count))
4548                         i = 0;
4549                 buffer_info = &rx_ring->buffer_info[i];
4550         }
4551
4552         if (likely(rx_ring->next_to_use != i)) {
4553                 rx_ring->next_to_use = i;
4554                 if (unlikely(i-- == 0))
4555                         i = (rx_ring->count - 1);
4556
4557                 /* Force memory writes to complete before letting h/w
4558                  * know there are new descriptors to fetch.  (Only
4559                  * applicable for weak-ordered memory model archs,
4560                  * such as IA-64).
4561                  */
4562                 wmb();
4563                 writel(i, adapter->hw.hw_addr + rx_ring->rdt);
4564         }
4565 }
4566
4567 /**
4568  * e1000_alloc_rx_buffers - Replace used receive buffers; legacy & extended
4569  * @adapter: address of board private structure
4570  **/
4571 static void e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
4572                                    struct e1000_rx_ring *rx_ring,
4573                                    int cleaned_count)
4574 {
4575         struct e1000_hw *hw = &adapter->hw;
4576         struct pci_dev *pdev = adapter->pdev;
4577         struct e1000_rx_desc *rx_desc;
4578         struct e1000_rx_buffer *buffer_info;
4579         unsigned int i;
4580         unsigned int bufsz = adapter->rx_buffer_len;
4581
4582         i = rx_ring->next_to_use;
4583         buffer_info = &rx_ring->buffer_info[i];
4584
4585         while (cleaned_count--) {
4586                 void *data;
4587
4588                 if (buffer_info->rxbuf.data)
4589                         goto skip;
4590
4591                 data = e1000_alloc_frag(adapter);
4592                 if (!data) {
4593                         /* Better luck next round */
4594                         adapter->alloc_rx_buff_failed++;
4595                         break;
4596                 }
4597
4598                 /* Fix for errata 23, can't cross 64kB boundary */
4599                 if (!e1000_check_64k_bound(adapter, data, bufsz)) {
4600                         void *olddata = data;
4601                         e_err(rx_err, "skb align check failed: %u bytes at "
4602                               "%p\n", bufsz, data);
4603                         /* Try again, without freeing the previous */
4604                         data = e1000_alloc_frag(adapter);
4605                         /* Failed allocation, critical failure */
4606                         if (!data) {
4607                                 skb_free_frag(olddata);
4608                                 adapter->alloc_rx_buff_failed++;
4609                                 break;
4610                         }
4611
4612                         if (!e1000_check_64k_bound(adapter, data, bufsz)) {
4613                                 /* give up */
4614                                 skb_free_frag(data);
4615                                 skb_free_frag(olddata);
4616                                 adapter->alloc_rx_buff_failed++;
4617                                 break;
4618                         }
4619
4620                         /* Use new allocation */
4621                         skb_free_frag(olddata);
4622                 }
4623                 buffer_info->dma = dma_map_single(&pdev->dev,
4624                                                   data,
4625                                                   adapter->rx_buffer_len,
4626                                                   DMA_FROM_DEVICE);
4627                 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
4628                         skb_free_frag(data);
4629                         buffer_info->dma = 0;
4630                         adapter->alloc_rx_buff_failed++;
4631                         break;
4632                 }
4633
4634                 /* XXX if it was allocated cleanly it will never map to a
4635                  * boundary crossing
4636                  */
4637
4638                 /* Fix for errata 23, can't cross 64kB boundary */
4639                 if (!e1000_check_64k_bound(adapter,
4640                                         (void *)(unsigned long)buffer_info->dma,
4641                                         adapter->rx_buffer_len)) {
4642                         e_err(rx_err, "dma align check failed: %u bytes at "
4643                               "%p\n", adapter->rx_buffer_len,
4644                               (void *)(unsigned long)buffer_info->dma);
4645
4646                         dma_unmap_single(&pdev->dev, buffer_info->dma,
4647                                          adapter->rx_buffer_len,
4648                                          DMA_FROM_DEVICE);
4649
4650                         skb_free_frag(data);
4651                         buffer_info->rxbuf.data = NULL;
4652                         buffer_info->dma = 0;
4653
4654                         adapter->alloc_rx_buff_failed++;
4655                         break;
4656                 }
4657                 buffer_info->rxbuf.data = data;
4658  skip:
4659                 rx_desc = E1000_RX_DESC(*rx_ring, i);
4660                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
4661
4662                 if (unlikely(++i == rx_ring->count))
4663                         i = 0;
4664                 buffer_info = &rx_ring->buffer_info[i];
4665         }
4666
4667         if (likely(rx_ring->next_to_use != i)) {
4668                 rx_ring->next_to_use = i;
4669                 if (unlikely(i-- == 0))
4670                         i = (rx_ring->count - 1);
4671
4672                 /* Force memory writes to complete before letting h/w
4673                  * know there are new descriptors to fetch.  (Only
4674                  * applicable for weak-ordered memory model archs,
4675                  * such as IA-64).
4676                  */
4677                 wmb();
4678                 writel(i, hw->hw_addr + rx_ring->rdt);
4679         }
4680 }
4681
4682 /**
4683  * e1000_smartspeed - Workaround for SmartSpeed on 82541 and 82547 controllers.
4684  * @adapter:
4685  **/
4686 static void e1000_smartspeed(struct e1000_adapter *adapter)
4687 {
4688         struct e1000_hw *hw = &adapter->hw;
4689         u16 phy_status;
4690         u16 phy_ctrl;
4691
4692         if ((hw->phy_type != e1000_phy_igp) || !hw->autoneg ||
4693            !(hw->autoneg_advertised & ADVERTISE_1000_FULL))
4694                 return;
4695
4696         if (adapter->smartspeed == 0) {
4697                 /* If Master/Slave config fault is asserted twice,
4698                  * we assume back-to-back
4699                  */
4700                 e1000_read_phy_reg(hw, PHY_1000T_STATUS, &phy_status);
4701                 if (!(phy_status & SR_1000T_MS_CONFIG_FAULT))
4702                         return;
4703                 e1000_read_phy_reg(hw, PHY_1000T_STATUS, &phy_status);
4704                 if (!(phy_status & SR_1000T_MS_CONFIG_FAULT))
4705                         return;
4706                 e1000_read_phy_reg(hw, PHY_1000T_CTRL, &phy_ctrl);
4707                 if (phy_ctrl & CR_1000T_MS_ENABLE) {
4708                         phy_ctrl &= ~CR_1000T_MS_ENABLE;
4709                         e1000_write_phy_reg(hw, PHY_1000T_CTRL,
4710                                             phy_ctrl);
4711                         adapter->smartspeed++;
4712                         if (!e1000_phy_setup_autoneg(hw) &&
4713                            !e1000_read_phy_reg(hw, PHY_CTRL,
4714                                                &phy_ctrl)) {
4715                                 phy_ctrl |= (MII_CR_AUTO_NEG_EN |
4716                                              MII_CR_RESTART_AUTO_NEG);
4717                                 e1000_write_phy_reg(hw, PHY_CTRL,
4718                                                     phy_ctrl);
4719                         }
4720                 }
4721                 return;
4722         } else if (adapter->smartspeed == E1000_SMARTSPEED_DOWNSHIFT) {
4723                 /* If still no link, perhaps using 2/3 pair cable */
4724                 e1000_read_phy_reg(hw, PHY_1000T_CTRL, &phy_ctrl);
4725                 phy_ctrl |= CR_1000T_MS_ENABLE;
4726                 e1000_write_phy_reg(hw, PHY_1000T_CTRL, phy_ctrl);
4727                 if (!e1000_phy_setup_autoneg(hw) &&
4728                    !e1000_read_phy_reg(hw, PHY_CTRL, &phy_ctrl)) {
4729                         phy_ctrl |= (MII_CR_AUTO_NEG_EN |
4730                                      MII_CR_RESTART_AUTO_NEG);
4731                         e1000_write_phy_reg(hw, PHY_CTRL, phy_ctrl);
4732                 }
4733         }
4734         /* Restart process after E1000_SMARTSPEED_MAX iterations */
4735         if (adapter->smartspeed++ == E1000_SMARTSPEED_MAX)
4736                 adapter->smartspeed = 0;
4737 }
4738
4739 /**
4740  * e1000_ioctl -
4741  * @netdev:
4742  * @ifreq:
4743  * @cmd:
4744  **/
4745 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
4746 {
4747         switch (cmd) {
4748         case SIOCGMIIPHY:
4749         case SIOCGMIIREG:
4750         case SIOCSMIIREG:
4751                 return e1000_mii_ioctl(netdev, ifr, cmd);
4752         default:
4753                 return -EOPNOTSUPP;
4754         }
4755 }
4756
4757 /**
4758  * e1000_mii_ioctl -
4759  * @netdev:
4760  * @ifreq:
4761  * @cmd:
4762  **/
4763 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
4764                            int cmd)
4765 {
4766         struct e1000_adapter *adapter = netdev_priv(netdev);
4767         struct e1000_hw *hw = &adapter->hw;
4768         struct mii_ioctl_data *data = if_mii(ifr);
4769         int retval;
4770         u16 mii_reg;
4771         unsigned long flags;
4772
4773         if (hw->media_type != e1000_media_type_copper)
4774                 return -EOPNOTSUPP;
4775
4776         switch (cmd) {
4777         case SIOCGMIIPHY:
4778                 data->phy_id = hw->phy_addr;
4779                 break;
4780         case SIOCGMIIREG:
4781                 spin_lock_irqsave(&adapter->stats_lock, flags);
4782                 if (e1000_read_phy_reg(hw, data->reg_num & 0x1F,
4783                                    &data->val_out)) {
4784                         spin_unlock_irqrestore(&adapter->stats_lock, flags);
4785                         return -EIO;
4786                 }
4787                 spin_unlock_irqrestore(&adapter->stats_lock, flags);
4788                 break;
4789         case SIOCSMIIREG:
4790                 if (data->reg_num & ~(0x1F))
4791                         return -EFAULT;
4792                 mii_reg = data->val_in;
4793                 spin_lock_irqsave(&adapter->stats_lock, flags);
4794                 if (e1000_write_phy_reg(hw, data->reg_num,
4795                                         mii_reg)) {
4796                         spin_unlock_irqrestore(&adapter->stats_lock, flags);
4797                         return -EIO;
4798                 }
4799                 spin_unlock_irqrestore(&adapter->stats_lock, flags);
4800                 if (hw->media_type == e1000_media_type_copper) {
4801                         switch (data->reg_num) {
4802                         case PHY_CTRL:
4803                                 if (mii_reg & MII_CR_POWER_DOWN)
4804                                         break;
4805                                 if (mii_reg & MII_CR_AUTO_NEG_EN) {
4806                                         hw->autoneg = 1;
4807                                         hw->autoneg_advertised = 0x2F;
4808                                 } else {
4809                                         u32 speed;
4810                                         if (mii_reg & 0x40)
4811                                                 speed = SPEED_1000;
4812                                         else if (mii_reg & 0x2000)
4813                                                 speed = SPEED_100;
4814                                         else
4815                                                 speed = SPEED_10;
4816                                         retval = e1000_set_spd_dplx(
4817                                                 adapter, speed,
4818                                                 ((mii_reg & 0x100)
4819                                                  ? DUPLEX_FULL :
4820                                                  DUPLEX_HALF));
4821                                         if (retval)
4822                                                 return retval;
4823                                 }
4824                                 if (netif_running(adapter->netdev))
4825                                         e1000_reinit_locked(adapter);
4826                                 else
4827                                         e1000_reset(adapter);
4828                                 break;
4829                         case M88E1000_PHY_SPEC_CTRL:
4830                         case M88E1000_EXT_PHY_SPEC_CTRL:
4831                                 if (e1000_phy_reset(hw))
4832                                         return -EIO;
4833                                 break;
4834                         }
4835                 } else {
4836                         switch (data->reg_num) {
4837                         case PHY_CTRL:
4838                                 if (mii_reg & MII_CR_POWER_DOWN)
4839                                         break;
4840                                 if (netif_running(adapter->netdev))
4841                                         e1000_reinit_locked(adapter);
4842                                 else
4843                                         e1000_reset(adapter);
4844                                 break;
4845                         }
4846                 }
4847                 break;
4848         default:
4849                 return -EOPNOTSUPP;
4850         }
4851         return E1000_SUCCESS;
4852 }
4853
4854 void e1000_pci_set_mwi(struct e1000_hw *hw)
4855 {
4856         struct e1000_adapter *adapter = hw->back;
4857         int ret_val = pci_set_mwi(adapter->pdev);
4858
4859         if (ret_val)
4860                 e_err(probe, "Error in setting MWI\n");
4861 }
4862
4863 void e1000_pci_clear_mwi(struct e1000_hw *hw)
4864 {
4865         struct e1000_adapter *adapter = hw->back;
4866
4867         pci_clear_mwi(adapter->pdev);
4868 }
4869
4870 int e1000_pcix_get_mmrbc(struct e1000_hw *hw)
4871 {
4872         struct e1000_adapter *adapter = hw->back;
4873         return pcix_get_mmrbc(adapter->pdev);
4874 }
4875
4876 void e1000_pcix_set_mmrbc(struct e1000_hw *hw, int mmrbc)
4877 {
4878         struct e1000_adapter *adapter = hw->back;
4879         pcix_set_mmrbc(adapter->pdev, mmrbc);
4880 }
4881
4882 void e1000_io_write(struct e1000_hw *hw, unsigned long port, u32 value)
4883 {
4884         outl(value, port);
4885 }
4886
4887 static bool e1000_vlan_used(struct e1000_adapter *adapter)
4888 {
4889         u16 vid;
4890
4891         for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
4892                 return true;
4893         return false;
4894 }
4895
4896 static void __e1000_vlan_mode(struct e1000_adapter *adapter,
4897                               netdev_features_t features)
4898 {
4899         struct e1000_hw *hw = &adapter->hw;
4900         u32 ctrl;
4901
4902         ctrl = er32(CTRL);
4903         if (features & NETIF_F_HW_VLAN_CTAG_RX) {
4904                 /* enable VLAN tag insert/strip */
4905                 ctrl |= E1000_CTRL_VME;
4906         } else {
4907                 /* disable VLAN tag insert/strip */
4908                 ctrl &= ~E1000_CTRL_VME;
4909         }
4910         ew32(CTRL, ctrl);
4911 }
4912 static void e1000_vlan_filter_on_off(struct e1000_adapter *adapter,
4913                                      bool filter_on)
4914 {
4915         struct e1000_hw *hw = &adapter->hw;
4916         u32 rctl;
4917
4918         if (!test_bit(__E1000_DOWN, &adapter->flags))
4919                 e1000_irq_disable(adapter);
4920
4921         __e1000_vlan_mode(adapter, adapter->netdev->features);
4922         if (filter_on) {
4923                 /* enable VLAN receive filtering */
4924                 rctl = er32(RCTL);
4925                 rctl &= ~E1000_RCTL_CFIEN;
4926                 if (!(adapter->netdev->flags & IFF_PROMISC))
4927                         rctl |= E1000_RCTL_VFE;
4928                 ew32(RCTL, rctl);
4929                 e1000_update_mng_vlan(adapter);
4930         } else {
4931                 /* disable VLAN receive filtering */
4932                 rctl = er32(RCTL);
4933                 rctl &= ~E1000_RCTL_VFE;
4934                 ew32(RCTL, rctl);
4935         }
4936
4937         if (!test_bit(__E1000_DOWN, &adapter->flags))
4938                 e1000_irq_enable(adapter);
4939 }
4940
4941 static void e1000_vlan_mode(struct net_device *netdev,
4942                             netdev_features_t features)
4943 {
4944         struct e1000_adapter *adapter = netdev_priv(netdev);
4945
4946         if (!test_bit(__E1000_DOWN, &adapter->flags))
4947                 e1000_irq_disable(adapter);
4948
4949         __e1000_vlan_mode(adapter, features);
4950
4951         if (!test_bit(__E1000_DOWN, &adapter->flags))
4952                 e1000_irq_enable(adapter);
4953 }
4954
4955 static int e1000_vlan_rx_add_vid(struct net_device *netdev,
4956                                  __be16 proto, u16 vid)
4957 {
4958         struct e1000_adapter *adapter = netdev_priv(netdev);
4959         struct e1000_hw *hw = &adapter->hw;
4960         u32 vfta, index;
4961
4962         if ((hw->mng_cookie.status &
4963              E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT) &&
4964             (vid == adapter->mng_vlan_id))
4965                 return 0;
4966
4967         if (!e1000_vlan_used(adapter))
4968                 e1000_vlan_filter_on_off(adapter, true);
4969
4970         /* add VID to filter table */
4971         index = (vid >> 5) & 0x7F;
4972         vfta = E1000_READ_REG_ARRAY(hw, VFTA, index);
4973         vfta |= (1 << (vid & 0x1F));
4974         e1000_write_vfta(hw, index, vfta);
4975
4976         set_bit(vid, adapter->active_vlans);
4977
4978         return 0;
4979 }
4980
4981 static int e1000_vlan_rx_kill_vid(struct net_device *netdev,
4982                                   __be16 proto, u16 vid)
4983 {
4984         struct e1000_adapter *adapter = netdev_priv(netdev);
4985         struct e1000_hw *hw = &adapter->hw;
4986         u32 vfta, index;
4987
4988         if (!test_bit(__E1000_DOWN, &adapter->flags))
4989                 e1000_irq_disable(adapter);
4990         if (!test_bit(__E1000_DOWN, &adapter->flags))
4991                 e1000_irq_enable(adapter);
4992
4993         /* remove VID from filter table */
4994         index = (vid >> 5) & 0x7F;
4995         vfta = E1000_READ_REG_ARRAY(hw, VFTA, index);
4996         vfta &= ~(1 << (vid & 0x1F));
4997         e1000_write_vfta(hw, index, vfta);
4998
4999         clear_bit(vid, adapter->active_vlans);
5000
5001         if (!e1000_vlan_used(adapter))
5002                 e1000_vlan_filter_on_off(adapter, false);
5003
5004         return 0;
5005 }
5006
5007 static void e1000_restore_vlan(struct e1000_adapter *adapter)
5008 {
5009         u16 vid;
5010
5011         if (!e1000_vlan_used(adapter))
5012                 return;
5013
5014         e1000_vlan_filter_on_off(adapter, true);
5015         for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
5016                 e1000_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), vid);
5017 }
5018
5019 int e1000_set_spd_dplx(struct e1000_adapter *adapter, u32 spd, u8 dplx)
5020 {
5021         struct e1000_hw *hw = &adapter->hw;
5022
5023         hw->autoneg = 0;
5024
5025         /* Make sure dplx is at most 1 bit and lsb of speed is not set
5026          * for the switch() below to work
5027          */
5028         if ((spd & 1) || (dplx & ~1))
5029                 goto err_inval;
5030
5031         /* Fiber NICs only allow 1000 gbps Full duplex */
5032         if ((hw->media_type == e1000_media_type_fiber) &&
5033             spd != SPEED_1000 &&
5034             dplx != DUPLEX_FULL)
5035                 goto err_inval;
5036
5037         switch (spd + dplx) {
5038         case SPEED_10 + DUPLEX_HALF:
5039                 hw->forced_speed_duplex = e1000_10_half;
5040                 break;
5041         case SPEED_10 + DUPLEX_FULL:
5042                 hw->forced_speed_duplex = e1000_10_full;
5043                 break;
5044         case SPEED_100 + DUPLEX_HALF:
5045                 hw->forced_speed_duplex = e1000_100_half;
5046                 break;
5047         case SPEED_100 + DUPLEX_FULL:
5048                 hw->forced_speed_duplex = e1000_100_full;
5049                 break;
5050         case SPEED_1000 + DUPLEX_FULL:
5051                 hw->autoneg = 1;
5052                 hw->autoneg_advertised = ADVERTISE_1000_FULL;
5053                 break;
5054         case SPEED_1000 + DUPLEX_HALF: /* not supported */
5055         default:
5056                 goto err_inval;
5057         }
5058
5059         /* clear MDI, MDI(-X) override is only allowed when autoneg enabled */
5060         hw->mdix = AUTO_ALL_MODES;
5061
5062         return 0;
5063
5064 err_inval:
5065         e_err(probe, "Unsupported Speed/Duplex configuration\n");
5066         return -EINVAL;
5067 }
5068
5069 static int __e1000_shutdown(struct pci_dev *pdev, bool *enable_wake)
5070 {
5071         struct net_device *netdev = pci_get_drvdata(pdev);
5072         struct e1000_adapter *adapter = netdev_priv(netdev);
5073         struct e1000_hw *hw = &adapter->hw;
5074         u32 ctrl, ctrl_ext, rctl, status;
5075         u32 wufc = adapter->wol;
5076 #ifdef CONFIG_PM
5077         int retval = 0;
5078 #endif
5079
5080         netif_device_detach(netdev);
5081
5082         if (netif_running(netdev)) {
5083                 int count = E1000_CHECK_RESET_COUNT;
5084
5085                 while (test_bit(__E1000_RESETTING, &adapter->flags) && count--)
5086                         usleep_range(10000, 20000);
5087
5088                 WARN_ON(test_bit(__E1000_RESETTING, &adapter->flags));
5089                 e1000_down(adapter);
5090         }
5091
5092 #ifdef CONFIG_PM
5093         retval = pci_save_state(pdev);
5094         if (retval)
5095                 return retval;
5096 #endif
5097
5098         status = er32(STATUS);
5099         if (status & E1000_STATUS_LU)
5100                 wufc &= ~E1000_WUFC_LNKC;
5101
5102         if (wufc) {
5103                 e1000_setup_rctl(adapter);
5104                 e1000_set_rx_mode(netdev);
5105
5106                 rctl = er32(RCTL);
5107
5108                 /* turn on all-multi mode if wake on multicast is enabled */
5109                 if (wufc & E1000_WUFC_MC)
5110                         rctl |= E1000_RCTL_MPE;
5111
5112                 /* enable receives in the hardware */
5113                 ew32(RCTL, rctl | E1000_RCTL_EN);
5114
5115                 if (hw->mac_type >= e1000_82540) {
5116                         ctrl = er32(CTRL);
5117                         /* advertise wake from D3Cold */
5118                         #define E1000_CTRL_ADVD3WUC 0x00100000
5119                         /* phy power management enable */
5120                         #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
5121                         ctrl |= E1000_CTRL_ADVD3WUC |
5122                                 E1000_CTRL_EN_PHY_PWR_MGMT;
5123                         ew32(CTRL, ctrl);
5124                 }
5125
5126                 if (hw->media_type == e1000_media_type_fiber ||
5127                     hw->media_type == e1000_media_type_internal_serdes) {
5128                         /* keep the laser running in D3 */
5129                         ctrl_ext = er32(CTRL_EXT);
5130                         ctrl_ext |= E1000_CTRL_EXT_SDP7_DATA;
5131                         ew32(CTRL_EXT, ctrl_ext);
5132                 }
5133
5134                 ew32(WUC, E1000_WUC_PME_EN);
5135                 ew32(WUFC, wufc);
5136         } else {
5137                 ew32(WUC, 0);
5138                 ew32(WUFC, 0);
5139         }
5140
5141         e1000_release_manageability(adapter);
5142
5143         *enable_wake = !!wufc;
5144
5145         /* make sure adapter isn't asleep if manageability is enabled */
5146         if (adapter->en_mng_pt)
5147                 *enable_wake = true;
5148
5149         if (netif_running(netdev))
5150                 e1000_free_irq(adapter);
5151
5152         if (!test_and_set_bit(__E1000_DISABLED, &adapter->flags))
5153                 pci_disable_device(pdev);
5154
5155         return 0;
5156 }
5157
5158 #ifdef CONFIG_PM
5159 static int e1000_suspend(struct pci_dev *pdev, pm_message_t state)
5160 {
5161         int retval;
5162         bool wake;
5163
5164         retval = __e1000_shutdown(pdev, &wake);
5165         if (retval)
5166                 return retval;
5167
5168         if (wake) {
5169                 pci_prepare_to_sleep(pdev);
5170         } else {
5171                 pci_wake_from_d3(pdev, false);
5172                 pci_set_power_state(pdev, PCI_D3hot);
5173         }
5174
5175         return 0;
5176 }
5177
5178 static int e1000_resume(struct pci_dev *pdev)
5179 {
5180         struct net_device *netdev = pci_get_drvdata(pdev);
5181         struct e1000_adapter *adapter = netdev_priv(netdev);
5182         struct e1000_hw *hw = &adapter->hw;
5183         u32 err;
5184
5185         pci_set_power_state(pdev, PCI_D0);
5186         pci_restore_state(pdev);
5187         pci_save_state(pdev);
5188
5189         if (adapter->need_ioport)
5190                 err = pci_enable_device(pdev);
5191         else
5192                 err = pci_enable_device_mem(pdev);
5193         if (err) {
5194                 pr_err("Cannot enable PCI device from suspend\n");
5195                 return err;
5196         }
5197
5198         /* flush memory to make sure state is correct */
5199         smp_mb__before_atomic();
5200         clear_bit(__E1000_DISABLED, &adapter->flags);
5201         pci_set_master(pdev);
5202
5203         pci_enable_wake(pdev, PCI_D3hot, 0);
5204         pci_enable_wake(pdev, PCI_D3cold, 0);
5205
5206         if (netif_running(netdev)) {
5207                 err = e1000_request_irq(adapter);
5208                 if (err)
5209                         return err;
5210         }
5211
5212         e1000_power_up_phy(adapter);
5213         e1000_reset(adapter);
5214         ew32(WUS, ~0);
5215
5216         e1000_init_manageability(adapter);
5217
5218         if (netif_running(netdev))
5219                 e1000_up(adapter);
5220
5221         netif_device_attach(netdev);
5222
5223         return 0;
5224 }
5225 #endif
5226
5227 static void e1000_shutdown(struct pci_dev *pdev)
5228 {
5229         bool wake;
5230
5231         __e1000_shutdown(pdev, &wake);
5232
5233         if (system_state == SYSTEM_POWER_OFF) {
5234                 pci_wake_from_d3(pdev, wake);
5235                 pci_set_power_state(pdev, PCI_D3hot);
5236         }
5237 }
5238
5239 #ifdef CONFIG_NET_POLL_CONTROLLER
5240 /* Polling 'interrupt' - used by things like netconsole to send skbs
5241  * without having to re-enable interrupts. It's not called while
5242  * the interrupt routine is executing.
5243  */
5244 static void e1000_netpoll(struct net_device *netdev)
5245 {
5246         struct e1000_adapter *adapter = netdev_priv(netdev);
5247
5248         if (disable_hardirq(adapter->pdev->irq))
5249                 e1000_intr(adapter->pdev->irq, netdev);
5250         enable_irq(adapter->pdev->irq);
5251 }
5252 #endif
5253
5254 /**
5255  * e1000_io_error_detected - called when PCI error is detected
5256  * @pdev: Pointer to PCI device
5257  * @state: The current pci connection state
5258  *
5259  * This function is called after a PCI bus error affecting
5260  * this device has been detected.
5261  */
5262 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
5263                                                 pci_channel_state_t state)
5264 {
5265         struct net_device *netdev = pci_get_drvdata(pdev);
5266         struct e1000_adapter *adapter = netdev_priv(netdev);
5267
5268         netif_device_detach(netdev);
5269
5270         if (state == pci_channel_io_perm_failure)
5271                 return PCI_ERS_RESULT_DISCONNECT;
5272
5273         if (netif_running(netdev))
5274                 e1000_down(adapter);
5275
5276         if (!test_and_set_bit(__E1000_DISABLED, &adapter->flags))
5277                 pci_disable_device(pdev);
5278
5279         /* Request a slot slot reset. */
5280         return PCI_ERS_RESULT_NEED_RESET;
5281 }
5282
5283 /**
5284  * e1000_io_slot_reset - called after the pci bus has been reset.
5285  * @pdev: Pointer to PCI device
5286  *
5287  * Restart the card from scratch, as if from a cold-boot. Implementation
5288  * resembles the first-half of the e1000_resume routine.
5289  */
5290 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
5291 {
5292         struct net_device *netdev = pci_get_drvdata(pdev);
5293         struct e1000_adapter *adapter = netdev_priv(netdev);
5294         struct e1000_hw *hw = &adapter->hw;
5295         int err;
5296
5297         if (adapter->need_ioport)
5298                 err = pci_enable_device(pdev);
5299         else
5300                 err = pci_enable_device_mem(pdev);
5301         if (err) {
5302                 pr_err("Cannot re-enable PCI device after reset.\n");
5303                 return PCI_ERS_RESULT_DISCONNECT;
5304         }
5305
5306         /* flush memory to make sure state is correct */
5307         smp_mb__before_atomic();
5308         clear_bit(__E1000_DISABLED, &adapter->flags);
5309         pci_set_master(pdev);
5310
5311         pci_enable_wake(pdev, PCI_D3hot, 0);
5312         pci_enable_wake(pdev, PCI_D3cold, 0);
5313
5314         e1000_reset(adapter);
5315         ew32(WUS, ~0);
5316
5317         return PCI_ERS_RESULT_RECOVERED;
5318 }
5319
5320 /**
5321  * e1000_io_resume - called when traffic can start flowing again.
5322  * @pdev: Pointer to PCI device
5323  *
5324  * This callback is called when the error recovery driver tells us that
5325  * its OK to resume normal operation. Implementation resembles the
5326  * second-half of the e1000_resume routine.
5327  */
5328 static void e1000_io_resume(struct pci_dev *pdev)
5329 {
5330         struct net_device *netdev = pci_get_drvdata(pdev);
5331         struct e1000_adapter *adapter = netdev_priv(netdev);
5332
5333         e1000_init_manageability(adapter);
5334
5335         if (netif_running(netdev)) {
5336                 if (e1000_up(adapter)) {
5337                         pr_info("can't bring device back up after reset\n");
5338                         return;
5339                 }
5340         }
5341
5342         netif_device_attach(netdev);
5343 }
5344
5345 /* e1000_main.c */