Add the rt linux 4.1.3-rt3 as base
[kvmfornfv.git] / kernel / drivers / net / ethernet / intel / e1000 / e1000_main.c
1 /*******************************************************************************
2
3   Intel PRO/1000 Linux driver
4   Copyright(c) 1999 - 2006 Intel Corporation.
5
6   This program is free software; you can redistribute it and/or modify it
7   under the terms and conditions of the GNU General Public License,
8   version 2, as published by the Free Software Foundation.
9
10   This program is distributed in the hope it will be useful, but WITHOUT
11   ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12   FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
13   more details.
14
15   You should have received a copy of the GNU General Public License along with
16   this program; if not, write to the Free Software Foundation, Inc.,
17   51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
18
19   The full GNU General Public License is included in this distribution in
20   the file called "COPYING".
21
22   Contact Information:
23   Linux NICS <linux.nics@intel.com>
24   e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
25   Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
26
27 *******************************************************************************/
28
29 #include "e1000.h"
30 #include <net/ip6_checksum.h>
31 #include <linux/io.h>
32 #include <linux/prefetch.h>
33 #include <linux/bitops.h>
34 #include <linux/if_vlan.h>
35
36 char e1000_driver_name[] = "e1000";
37 static char e1000_driver_string[] = "Intel(R) PRO/1000 Network Driver";
38 #define DRV_VERSION "7.3.21-k8-NAPI"
39 const char e1000_driver_version[] = DRV_VERSION;
40 static const char e1000_copyright[] = "Copyright (c) 1999-2006 Intel Corporation.";
41
42 /* e1000_pci_tbl - PCI Device ID Table
43  *
44  * Last entry must be all 0s
45  *
46  * Macro expands to...
47  *   {PCI_DEVICE(PCI_VENDOR_ID_INTEL, device_id)}
48  */
49 static const struct pci_device_id e1000_pci_tbl[] = {
50         INTEL_E1000_ETHERNET_DEVICE(0x1000),
51         INTEL_E1000_ETHERNET_DEVICE(0x1001),
52         INTEL_E1000_ETHERNET_DEVICE(0x1004),
53         INTEL_E1000_ETHERNET_DEVICE(0x1008),
54         INTEL_E1000_ETHERNET_DEVICE(0x1009),
55         INTEL_E1000_ETHERNET_DEVICE(0x100C),
56         INTEL_E1000_ETHERNET_DEVICE(0x100D),
57         INTEL_E1000_ETHERNET_DEVICE(0x100E),
58         INTEL_E1000_ETHERNET_DEVICE(0x100F),
59         INTEL_E1000_ETHERNET_DEVICE(0x1010),
60         INTEL_E1000_ETHERNET_DEVICE(0x1011),
61         INTEL_E1000_ETHERNET_DEVICE(0x1012),
62         INTEL_E1000_ETHERNET_DEVICE(0x1013),
63         INTEL_E1000_ETHERNET_DEVICE(0x1014),
64         INTEL_E1000_ETHERNET_DEVICE(0x1015),
65         INTEL_E1000_ETHERNET_DEVICE(0x1016),
66         INTEL_E1000_ETHERNET_DEVICE(0x1017),
67         INTEL_E1000_ETHERNET_DEVICE(0x1018),
68         INTEL_E1000_ETHERNET_DEVICE(0x1019),
69         INTEL_E1000_ETHERNET_DEVICE(0x101A),
70         INTEL_E1000_ETHERNET_DEVICE(0x101D),
71         INTEL_E1000_ETHERNET_DEVICE(0x101E),
72         INTEL_E1000_ETHERNET_DEVICE(0x1026),
73         INTEL_E1000_ETHERNET_DEVICE(0x1027),
74         INTEL_E1000_ETHERNET_DEVICE(0x1028),
75         INTEL_E1000_ETHERNET_DEVICE(0x1075),
76         INTEL_E1000_ETHERNET_DEVICE(0x1076),
77         INTEL_E1000_ETHERNET_DEVICE(0x1077),
78         INTEL_E1000_ETHERNET_DEVICE(0x1078),
79         INTEL_E1000_ETHERNET_DEVICE(0x1079),
80         INTEL_E1000_ETHERNET_DEVICE(0x107A),
81         INTEL_E1000_ETHERNET_DEVICE(0x107B),
82         INTEL_E1000_ETHERNET_DEVICE(0x107C),
83         INTEL_E1000_ETHERNET_DEVICE(0x108A),
84         INTEL_E1000_ETHERNET_DEVICE(0x1099),
85         INTEL_E1000_ETHERNET_DEVICE(0x10B5),
86         INTEL_E1000_ETHERNET_DEVICE(0x2E6E),
87         /* required last entry */
88         {0,}
89 };
90
91 MODULE_DEVICE_TABLE(pci, e1000_pci_tbl);
92
93 int e1000_up(struct e1000_adapter *adapter);
94 void e1000_down(struct e1000_adapter *adapter);
95 void e1000_reinit_locked(struct e1000_adapter *adapter);
96 void e1000_reset(struct e1000_adapter *adapter);
97 int e1000_setup_all_tx_resources(struct e1000_adapter *adapter);
98 int e1000_setup_all_rx_resources(struct e1000_adapter *adapter);
99 void e1000_free_all_tx_resources(struct e1000_adapter *adapter);
100 void e1000_free_all_rx_resources(struct e1000_adapter *adapter);
101 static int e1000_setup_tx_resources(struct e1000_adapter *adapter,
102                              struct e1000_tx_ring *txdr);
103 static int e1000_setup_rx_resources(struct e1000_adapter *adapter,
104                              struct e1000_rx_ring *rxdr);
105 static void e1000_free_tx_resources(struct e1000_adapter *adapter,
106                              struct e1000_tx_ring *tx_ring);
107 static void e1000_free_rx_resources(struct e1000_adapter *adapter,
108                              struct e1000_rx_ring *rx_ring);
109 void e1000_update_stats(struct e1000_adapter *adapter);
110
111 static int e1000_init_module(void);
112 static void e1000_exit_module(void);
113 static int e1000_probe(struct pci_dev *pdev, const struct pci_device_id *ent);
114 static void e1000_remove(struct pci_dev *pdev);
115 static int e1000_alloc_queues(struct e1000_adapter *adapter);
116 static int e1000_sw_init(struct e1000_adapter *adapter);
117 static int e1000_open(struct net_device *netdev);
118 static int e1000_close(struct net_device *netdev);
119 static void e1000_configure_tx(struct e1000_adapter *adapter);
120 static void e1000_configure_rx(struct e1000_adapter *adapter);
121 static void e1000_setup_rctl(struct e1000_adapter *adapter);
122 static void e1000_clean_all_tx_rings(struct e1000_adapter *adapter);
123 static void e1000_clean_all_rx_rings(struct e1000_adapter *adapter);
124 static void e1000_clean_tx_ring(struct e1000_adapter *adapter,
125                                 struct e1000_tx_ring *tx_ring);
126 static void e1000_clean_rx_ring(struct e1000_adapter *adapter,
127                                 struct e1000_rx_ring *rx_ring);
128 static void e1000_set_rx_mode(struct net_device *netdev);
129 static void e1000_update_phy_info_task(struct work_struct *work);
130 static void e1000_watchdog(struct work_struct *work);
131 static void e1000_82547_tx_fifo_stall_task(struct work_struct *work);
132 static netdev_tx_t e1000_xmit_frame(struct sk_buff *skb,
133                                     struct net_device *netdev);
134 static struct net_device_stats * e1000_get_stats(struct net_device *netdev);
135 static int e1000_change_mtu(struct net_device *netdev, int new_mtu);
136 static int e1000_set_mac(struct net_device *netdev, void *p);
137 static irqreturn_t e1000_intr(int irq, void *data);
138 static bool e1000_clean_tx_irq(struct e1000_adapter *adapter,
139                                struct e1000_tx_ring *tx_ring);
140 static int e1000_clean(struct napi_struct *napi, int budget);
141 static bool e1000_clean_rx_irq(struct e1000_adapter *adapter,
142                                struct e1000_rx_ring *rx_ring,
143                                int *work_done, int work_to_do);
144 static bool e1000_clean_jumbo_rx_irq(struct e1000_adapter *adapter,
145                                      struct e1000_rx_ring *rx_ring,
146                                      int *work_done, int work_to_do);
147 static void e1000_alloc_dummy_rx_buffers(struct e1000_adapter *adapter,
148                                          struct e1000_rx_ring *rx_ring,
149                                          int cleaned_count)
150 {
151 }
152 static void e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
153                                    struct e1000_rx_ring *rx_ring,
154                                    int cleaned_count);
155 static void e1000_alloc_jumbo_rx_buffers(struct e1000_adapter *adapter,
156                                          struct e1000_rx_ring *rx_ring,
157                                          int cleaned_count);
158 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd);
159 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
160                            int cmd);
161 static void e1000_enter_82542_rst(struct e1000_adapter *adapter);
162 static void e1000_leave_82542_rst(struct e1000_adapter *adapter);
163 static void e1000_tx_timeout(struct net_device *dev);
164 static void e1000_reset_task(struct work_struct *work);
165 static void e1000_smartspeed(struct e1000_adapter *adapter);
166 static int e1000_82547_fifo_workaround(struct e1000_adapter *adapter,
167                                        struct sk_buff *skb);
168
169 static bool e1000_vlan_used(struct e1000_adapter *adapter);
170 static void e1000_vlan_mode(struct net_device *netdev,
171                             netdev_features_t features);
172 static void e1000_vlan_filter_on_off(struct e1000_adapter *adapter,
173                                      bool filter_on);
174 static int e1000_vlan_rx_add_vid(struct net_device *netdev,
175                                  __be16 proto, u16 vid);
176 static int e1000_vlan_rx_kill_vid(struct net_device *netdev,
177                                   __be16 proto, u16 vid);
178 static void e1000_restore_vlan(struct e1000_adapter *adapter);
179
180 #ifdef CONFIG_PM
181 static int e1000_suspend(struct pci_dev *pdev, pm_message_t state);
182 static int e1000_resume(struct pci_dev *pdev);
183 #endif
184 static void e1000_shutdown(struct pci_dev *pdev);
185
186 #ifdef CONFIG_NET_POLL_CONTROLLER
187 /* for netdump / net console */
188 static void e1000_netpoll (struct net_device *netdev);
189 #endif
190
191 #define COPYBREAK_DEFAULT 256
192 static unsigned int copybreak __read_mostly = COPYBREAK_DEFAULT;
193 module_param(copybreak, uint, 0644);
194 MODULE_PARM_DESC(copybreak,
195         "Maximum size of packet that is copied to a new buffer on receive");
196
197 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
198                      pci_channel_state_t state);
199 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev);
200 static void e1000_io_resume(struct pci_dev *pdev);
201
202 static const struct pci_error_handlers e1000_err_handler = {
203         .error_detected = e1000_io_error_detected,
204         .slot_reset = e1000_io_slot_reset,
205         .resume = e1000_io_resume,
206 };
207
208 static struct pci_driver e1000_driver = {
209         .name     = e1000_driver_name,
210         .id_table = e1000_pci_tbl,
211         .probe    = e1000_probe,
212         .remove   = e1000_remove,
213 #ifdef CONFIG_PM
214         /* Power Management Hooks */
215         .suspend  = e1000_suspend,
216         .resume   = e1000_resume,
217 #endif
218         .shutdown = e1000_shutdown,
219         .err_handler = &e1000_err_handler
220 };
221
222 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
223 MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
224 MODULE_LICENSE("GPL");
225 MODULE_VERSION(DRV_VERSION);
226
227 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
228 static int debug = -1;
229 module_param(debug, int, 0);
230 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
231
232 /**
233  * e1000_get_hw_dev - return device
234  * used by hardware layer to print debugging information
235  *
236  **/
237 struct net_device *e1000_get_hw_dev(struct e1000_hw *hw)
238 {
239         struct e1000_adapter *adapter = hw->back;
240         return adapter->netdev;
241 }
242
243 /**
244  * e1000_init_module - Driver Registration Routine
245  *
246  * e1000_init_module is the first routine called when the driver is
247  * loaded. All it does is register with the PCI subsystem.
248  **/
249 static int __init e1000_init_module(void)
250 {
251         int ret;
252         pr_info("%s - version %s\n", e1000_driver_string, e1000_driver_version);
253
254         pr_info("%s\n", e1000_copyright);
255
256         ret = pci_register_driver(&e1000_driver);
257         if (copybreak != COPYBREAK_DEFAULT) {
258                 if (copybreak == 0)
259                         pr_info("copybreak disabled\n");
260                 else
261                         pr_info("copybreak enabled for "
262                                    "packets <= %u bytes\n", copybreak);
263         }
264         return ret;
265 }
266
267 module_init(e1000_init_module);
268
269 /**
270  * e1000_exit_module - Driver Exit Cleanup Routine
271  *
272  * e1000_exit_module is called just before the driver is removed
273  * from memory.
274  **/
275 static void __exit e1000_exit_module(void)
276 {
277         pci_unregister_driver(&e1000_driver);
278 }
279
280 module_exit(e1000_exit_module);
281
282 static int e1000_request_irq(struct e1000_adapter *adapter)
283 {
284         struct net_device *netdev = adapter->netdev;
285         irq_handler_t handler = e1000_intr;
286         int irq_flags = IRQF_SHARED;
287         int err;
288
289         err = request_irq(adapter->pdev->irq, handler, irq_flags, netdev->name,
290                           netdev);
291         if (err) {
292                 e_err(probe, "Unable to allocate interrupt Error: %d\n", err);
293         }
294
295         return err;
296 }
297
298 static void e1000_free_irq(struct e1000_adapter *adapter)
299 {
300         struct net_device *netdev = adapter->netdev;
301
302         free_irq(adapter->pdev->irq, netdev);
303 }
304
305 /**
306  * e1000_irq_disable - Mask off interrupt generation on the NIC
307  * @adapter: board private structure
308  **/
309 static void e1000_irq_disable(struct e1000_adapter *adapter)
310 {
311         struct e1000_hw *hw = &adapter->hw;
312
313         ew32(IMC, ~0);
314         E1000_WRITE_FLUSH();
315         synchronize_irq(adapter->pdev->irq);
316 }
317
318 /**
319  * e1000_irq_enable - Enable default interrupt generation settings
320  * @adapter: board private structure
321  **/
322 static void e1000_irq_enable(struct e1000_adapter *adapter)
323 {
324         struct e1000_hw *hw = &adapter->hw;
325
326         ew32(IMS, IMS_ENABLE_MASK);
327         E1000_WRITE_FLUSH();
328 }
329
330 static void e1000_update_mng_vlan(struct e1000_adapter *adapter)
331 {
332         struct e1000_hw *hw = &adapter->hw;
333         struct net_device *netdev = adapter->netdev;
334         u16 vid = hw->mng_cookie.vlan_id;
335         u16 old_vid = adapter->mng_vlan_id;
336
337         if (!e1000_vlan_used(adapter))
338                 return;
339
340         if (!test_bit(vid, adapter->active_vlans)) {
341                 if (hw->mng_cookie.status &
342                     E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT) {
343                         e1000_vlan_rx_add_vid(netdev, htons(ETH_P_8021Q), vid);
344                         adapter->mng_vlan_id = vid;
345                 } else {
346                         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
347                 }
348                 if ((old_vid != (u16)E1000_MNG_VLAN_NONE) &&
349                     (vid != old_vid) &&
350                     !test_bit(old_vid, adapter->active_vlans))
351                         e1000_vlan_rx_kill_vid(netdev, htons(ETH_P_8021Q),
352                                                old_vid);
353         } else {
354                 adapter->mng_vlan_id = vid;
355         }
356 }
357
358 static void e1000_init_manageability(struct e1000_adapter *adapter)
359 {
360         struct e1000_hw *hw = &adapter->hw;
361
362         if (adapter->en_mng_pt) {
363                 u32 manc = er32(MANC);
364
365                 /* disable hardware interception of ARP */
366                 manc &= ~(E1000_MANC_ARP_EN);
367
368                 ew32(MANC, manc);
369         }
370 }
371
372 static void e1000_release_manageability(struct e1000_adapter *adapter)
373 {
374         struct e1000_hw *hw = &adapter->hw;
375
376         if (adapter->en_mng_pt) {
377                 u32 manc = er32(MANC);
378
379                 /* re-enable hardware interception of ARP */
380                 manc |= E1000_MANC_ARP_EN;
381
382                 ew32(MANC, manc);
383         }
384 }
385
386 /**
387  * e1000_configure - configure the hardware for RX and TX
388  * @adapter = private board structure
389  **/
390 static void e1000_configure(struct e1000_adapter *adapter)
391 {
392         struct net_device *netdev = adapter->netdev;
393         int i;
394
395         e1000_set_rx_mode(netdev);
396
397         e1000_restore_vlan(adapter);
398         e1000_init_manageability(adapter);
399
400         e1000_configure_tx(adapter);
401         e1000_setup_rctl(adapter);
402         e1000_configure_rx(adapter);
403         /* call E1000_DESC_UNUSED which always leaves
404          * at least 1 descriptor unused to make sure
405          * next_to_use != next_to_clean
406          */
407         for (i = 0; i < adapter->num_rx_queues; i++) {
408                 struct e1000_rx_ring *ring = &adapter->rx_ring[i];
409                 adapter->alloc_rx_buf(adapter, ring,
410                                       E1000_DESC_UNUSED(ring));
411         }
412 }
413
414 int e1000_up(struct e1000_adapter *adapter)
415 {
416         struct e1000_hw *hw = &adapter->hw;
417
418         /* hardware has been reset, we need to reload some things */
419         e1000_configure(adapter);
420
421         clear_bit(__E1000_DOWN, &adapter->flags);
422
423         napi_enable(&adapter->napi);
424
425         e1000_irq_enable(adapter);
426
427         netif_wake_queue(adapter->netdev);
428
429         /* fire a link change interrupt to start the watchdog */
430         ew32(ICS, E1000_ICS_LSC);
431         return 0;
432 }
433
434 /**
435  * e1000_power_up_phy - restore link in case the phy was powered down
436  * @adapter: address of board private structure
437  *
438  * The phy may be powered down to save power and turn off link when the
439  * driver is unloaded and wake on lan is not enabled (among others)
440  * *** this routine MUST be followed by a call to e1000_reset ***
441  **/
442 void e1000_power_up_phy(struct e1000_adapter *adapter)
443 {
444         struct e1000_hw *hw = &adapter->hw;
445         u16 mii_reg = 0;
446
447         /* Just clear the power down bit to wake the phy back up */
448         if (hw->media_type == e1000_media_type_copper) {
449                 /* according to the manual, the phy will retain its
450                  * settings across a power-down/up cycle
451                  */
452                 e1000_read_phy_reg(hw, PHY_CTRL, &mii_reg);
453                 mii_reg &= ~MII_CR_POWER_DOWN;
454                 e1000_write_phy_reg(hw, PHY_CTRL, mii_reg);
455         }
456 }
457
458 static void e1000_power_down_phy(struct e1000_adapter *adapter)
459 {
460         struct e1000_hw *hw = &adapter->hw;
461
462         /* Power down the PHY so no link is implied when interface is down *
463          * The PHY cannot be powered down if any of the following is true *
464          * (a) WoL is enabled
465          * (b) AMT is active
466          * (c) SoL/IDER session is active
467          */
468         if (!adapter->wol && hw->mac_type >= e1000_82540 &&
469            hw->media_type == e1000_media_type_copper) {
470                 u16 mii_reg = 0;
471
472                 switch (hw->mac_type) {
473                 case e1000_82540:
474                 case e1000_82545:
475                 case e1000_82545_rev_3:
476                 case e1000_82546:
477                 case e1000_ce4100:
478                 case e1000_82546_rev_3:
479                 case e1000_82541:
480                 case e1000_82541_rev_2:
481                 case e1000_82547:
482                 case e1000_82547_rev_2:
483                         if (er32(MANC) & E1000_MANC_SMBUS_EN)
484                                 goto out;
485                         break;
486                 default:
487                         goto out;
488                 }
489                 e1000_read_phy_reg(hw, PHY_CTRL, &mii_reg);
490                 mii_reg |= MII_CR_POWER_DOWN;
491                 e1000_write_phy_reg(hw, PHY_CTRL, mii_reg);
492                 msleep(1);
493         }
494 out:
495         return;
496 }
497
498 static void e1000_down_and_stop(struct e1000_adapter *adapter)
499 {
500         set_bit(__E1000_DOWN, &adapter->flags);
501
502         cancel_delayed_work_sync(&adapter->watchdog_task);
503
504         /*
505          * Since the watchdog task can reschedule other tasks, we should cancel
506          * it first, otherwise we can run into the situation when a work is
507          * still running after the adapter has been turned down.
508          */
509
510         cancel_delayed_work_sync(&adapter->phy_info_task);
511         cancel_delayed_work_sync(&adapter->fifo_stall_task);
512
513         /* Only kill reset task if adapter is not resetting */
514         if (!test_bit(__E1000_RESETTING, &adapter->flags))
515                 cancel_work_sync(&adapter->reset_task);
516 }
517
518 void e1000_down(struct e1000_adapter *adapter)
519 {
520         struct e1000_hw *hw = &adapter->hw;
521         struct net_device *netdev = adapter->netdev;
522         u32 rctl, tctl;
523
524         netif_carrier_off(netdev);
525
526         /* disable receives in the hardware */
527         rctl = er32(RCTL);
528         ew32(RCTL, rctl & ~E1000_RCTL_EN);
529         /* flush and sleep below */
530
531         netif_tx_disable(netdev);
532
533         /* disable transmits in the hardware */
534         tctl = er32(TCTL);
535         tctl &= ~E1000_TCTL_EN;
536         ew32(TCTL, tctl);
537         /* flush both disables and wait for them to finish */
538         E1000_WRITE_FLUSH();
539         msleep(10);
540
541         napi_disable(&adapter->napi);
542
543         e1000_irq_disable(adapter);
544
545         /* Setting DOWN must be after irq_disable to prevent
546          * a screaming interrupt.  Setting DOWN also prevents
547          * tasks from rescheduling.
548          */
549         e1000_down_and_stop(adapter);
550
551         adapter->link_speed = 0;
552         adapter->link_duplex = 0;
553
554         e1000_reset(adapter);
555         e1000_clean_all_tx_rings(adapter);
556         e1000_clean_all_rx_rings(adapter);
557 }
558
559 void e1000_reinit_locked(struct e1000_adapter *adapter)
560 {
561         WARN_ON(in_interrupt());
562         while (test_and_set_bit(__E1000_RESETTING, &adapter->flags))
563                 msleep(1);
564         e1000_down(adapter);
565         e1000_up(adapter);
566         clear_bit(__E1000_RESETTING, &adapter->flags);
567 }
568
569 void e1000_reset(struct e1000_adapter *adapter)
570 {
571         struct e1000_hw *hw = &adapter->hw;
572         u32 pba = 0, tx_space, min_tx_space, min_rx_space;
573         bool legacy_pba_adjust = false;
574         u16 hwm;
575
576         /* Repartition Pba for greater than 9k mtu
577          * To take effect CTRL.RST is required.
578          */
579
580         switch (hw->mac_type) {
581         case e1000_82542_rev2_0:
582         case e1000_82542_rev2_1:
583         case e1000_82543:
584         case e1000_82544:
585         case e1000_82540:
586         case e1000_82541:
587         case e1000_82541_rev_2:
588                 legacy_pba_adjust = true;
589                 pba = E1000_PBA_48K;
590                 break;
591         case e1000_82545:
592         case e1000_82545_rev_3:
593         case e1000_82546:
594         case e1000_ce4100:
595         case e1000_82546_rev_3:
596                 pba = E1000_PBA_48K;
597                 break;
598         case e1000_82547:
599         case e1000_82547_rev_2:
600                 legacy_pba_adjust = true;
601                 pba = E1000_PBA_30K;
602                 break;
603         case e1000_undefined:
604         case e1000_num_macs:
605                 break;
606         }
607
608         if (legacy_pba_adjust) {
609                 if (hw->max_frame_size > E1000_RXBUFFER_8192)
610                         pba -= 8; /* allocate more FIFO for Tx */
611
612                 if (hw->mac_type == e1000_82547) {
613                         adapter->tx_fifo_head = 0;
614                         adapter->tx_head_addr = pba << E1000_TX_HEAD_ADDR_SHIFT;
615                         adapter->tx_fifo_size =
616                                 (E1000_PBA_40K - pba) << E1000_PBA_BYTES_SHIFT;
617                         atomic_set(&adapter->tx_fifo_stall, 0);
618                 }
619         } else if (hw->max_frame_size >  ETH_FRAME_LEN + ETH_FCS_LEN) {
620                 /* adjust PBA for jumbo frames */
621                 ew32(PBA, pba);
622
623                 /* To maintain wire speed transmits, the Tx FIFO should be
624                  * large enough to accommodate two full transmit packets,
625                  * rounded up to the next 1KB and expressed in KB.  Likewise,
626                  * the Rx FIFO should be large enough to accommodate at least
627                  * one full receive packet and is similarly rounded up and
628                  * expressed in KB.
629                  */
630                 pba = er32(PBA);
631                 /* upper 16 bits has Tx packet buffer allocation size in KB */
632                 tx_space = pba >> 16;
633                 /* lower 16 bits has Rx packet buffer allocation size in KB */
634                 pba &= 0xffff;
635                 /* the Tx fifo also stores 16 bytes of information about the Tx
636                  * but don't include ethernet FCS because hardware appends it
637                  */
638                 min_tx_space = (hw->max_frame_size +
639                                 sizeof(struct e1000_tx_desc) -
640                                 ETH_FCS_LEN) * 2;
641                 min_tx_space = ALIGN(min_tx_space, 1024);
642                 min_tx_space >>= 10;
643                 /* software strips receive CRC, so leave room for it */
644                 min_rx_space = hw->max_frame_size;
645                 min_rx_space = ALIGN(min_rx_space, 1024);
646                 min_rx_space >>= 10;
647
648                 /* If current Tx allocation is less than the min Tx FIFO size,
649                  * and the min Tx FIFO size is less than the current Rx FIFO
650                  * allocation, take space away from current Rx allocation
651                  */
652                 if (tx_space < min_tx_space &&
653                     ((min_tx_space - tx_space) < pba)) {
654                         pba = pba - (min_tx_space - tx_space);
655
656                         /* PCI/PCIx hardware has PBA alignment constraints */
657                         switch (hw->mac_type) {
658                         case e1000_82545 ... e1000_82546_rev_3:
659                                 pba &= ~(E1000_PBA_8K - 1);
660                                 break;
661                         default:
662                                 break;
663                         }
664
665                         /* if short on Rx space, Rx wins and must trump Tx
666                          * adjustment or use Early Receive if available
667                          */
668                         if (pba < min_rx_space)
669                                 pba = min_rx_space;
670                 }
671         }
672
673         ew32(PBA, pba);
674
675         /* flow control settings:
676          * The high water mark must be low enough to fit one full frame
677          * (or the size used for early receive) above it in the Rx FIFO.
678          * Set it to the lower of:
679          * - 90% of the Rx FIFO size, and
680          * - the full Rx FIFO size minus the early receive size (for parts
681          *   with ERT support assuming ERT set to E1000_ERT_2048), or
682          * - the full Rx FIFO size minus one full frame
683          */
684         hwm = min(((pba << 10) * 9 / 10),
685                   ((pba << 10) - hw->max_frame_size));
686
687         hw->fc_high_water = hwm & 0xFFF8;       /* 8-byte granularity */
688         hw->fc_low_water = hw->fc_high_water - 8;
689         hw->fc_pause_time = E1000_FC_PAUSE_TIME;
690         hw->fc_send_xon = 1;
691         hw->fc = hw->original_fc;
692
693         /* Allow time for pending master requests to run */
694         e1000_reset_hw(hw);
695         if (hw->mac_type >= e1000_82544)
696                 ew32(WUC, 0);
697
698         if (e1000_init_hw(hw))
699                 e_dev_err("Hardware Error\n");
700         e1000_update_mng_vlan(adapter);
701
702         /* if (adapter->hwflags & HWFLAGS_PHY_PWR_BIT) { */
703         if (hw->mac_type >= e1000_82544 &&
704             hw->autoneg == 1 &&
705             hw->autoneg_advertised == ADVERTISE_1000_FULL) {
706                 u32 ctrl = er32(CTRL);
707                 /* clear phy power management bit if we are in gig only mode,
708                  * which if enabled will attempt negotiation to 100Mb, which
709                  * can cause a loss of link at power off or driver unload
710                  */
711                 ctrl &= ~E1000_CTRL_SWDPIN3;
712                 ew32(CTRL, ctrl);
713         }
714
715         /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
716         ew32(VET, ETHERNET_IEEE_VLAN_TYPE);
717
718         e1000_reset_adaptive(hw);
719         e1000_phy_get_info(hw, &adapter->phy_info);
720
721         e1000_release_manageability(adapter);
722 }
723
724 /* Dump the eeprom for users having checksum issues */
725 static void e1000_dump_eeprom(struct e1000_adapter *adapter)
726 {
727         struct net_device *netdev = adapter->netdev;
728         struct ethtool_eeprom eeprom;
729         const struct ethtool_ops *ops = netdev->ethtool_ops;
730         u8 *data;
731         int i;
732         u16 csum_old, csum_new = 0;
733
734         eeprom.len = ops->get_eeprom_len(netdev);
735         eeprom.offset = 0;
736
737         data = kmalloc(eeprom.len, GFP_KERNEL);
738         if (!data)
739                 return;
740
741         ops->get_eeprom(netdev, &eeprom, data);
742
743         csum_old = (data[EEPROM_CHECKSUM_REG * 2]) +
744                    (data[EEPROM_CHECKSUM_REG * 2 + 1] << 8);
745         for (i = 0; i < EEPROM_CHECKSUM_REG * 2; i += 2)
746                 csum_new += data[i] + (data[i + 1] << 8);
747         csum_new = EEPROM_SUM - csum_new;
748
749         pr_err("/*********************/\n");
750         pr_err("Current EEPROM Checksum : 0x%04x\n", csum_old);
751         pr_err("Calculated              : 0x%04x\n", csum_new);
752
753         pr_err("Offset    Values\n");
754         pr_err("========  ======\n");
755         print_hex_dump(KERN_ERR, "", DUMP_PREFIX_OFFSET, 16, 1, data, 128, 0);
756
757         pr_err("Include this output when contacting your support provider.\n");
758         pr_err("This is not a software error! Something bad happened to\n");
759         pr_err("your hardware or EEPROM image. Ignoring this problem could\n");
760         pr_err("result in further problems, possibly loss of data,\n");
761         pr_err("corruption or system hangs!\n");
762         pr_err("The MAC Address will be reset to 00:00:00:00:00:00,\n");
763         pr_err("which is invalid and requires you to set the proper MAC\n");
764         pr_err("address manually before continuing to enable this network\n");
765         pr_err("device. Please inspect the EEPROM dump and report the\n");
766         pr_err("issue to your hardware vendor or Intel Customer Support.\n");
767         pr_err("/*********************/\n");
768
769         kfree(data);
770 }
771
772 /**
773  * e1000_is_need_ioport - determine if an adapter needs ioport resources or not
774  * @pdev: PCI device information struct
775  *
776  * Return true if an adapter needs ioport resources
777  **/
778 static int e1000_is_need_ioport(struct pci_dev *pdev)
779 {
780         switch (pdev->device) {
781         case E1000_DEV_ID_82540EM:
782         case E1000_DEV_ID_82540EM_LOM:
783         case E1000_DEV_ID_82540EP:
784         case E1000_DEV_ID_82540EP_LOM:
785         case E1000_DEV_ID_82540EP_LP:
786         case E1000_DEV_ID_82541EI:
787         case E1000_DEV_ID_82541EI_MOBILE:
788         case E1000_DEV_ID_82541ER:
789         case E1000_DEV_ID_82541ER_LOM:
790         case E1000_DEV_ID_82541GI:
791         case E1000_DEV_ID_82541GI_LF:
792         case E1000_DEV_ID_82541GI_MOBILE:
793         case E1000_DEV_ID_82544EI_COPPER:
794         case E1000_DEV_ID_82544EI_FIBER:
795         case E1000_DEV_ID_82544GC_COPPER:
796         case E1000_DEV_ID_82544GC_LOM:
797         case E1000_DEV_ID_82545EM_COPPER:
798         case E1000_DEV_ID_82545EM_FIBER:
799         case E1000_DEV_ID_82546EB_COPPER:
800         case E1000_DEV_ID_82546EB_FIBER:
801         case E1000_DEV_ID_82546EB_QUAD_COPPER:
802                 return true;
803         default:
804                 return false;
805         }
806 }
807
808 static netdev_features_t e1000_fix_features(struct net_device *netdev,
809         netdev_features_t features)
810 {
811         /* Since there is no support for separate Rx/Tx vlan accel
812          * enable/disable make sure Tx flag is always in same state as Rx.
813          */
814         if (features & NETIF_F_HW_VLAN_CTAG_RX)
815                 features |= NETIF_F_HW_VLAN_CTAG_TX;
816         else
817                 features &= ~NETIF_F_HW_VLAN_CTAG_TX;
818
819         return features;
820 }
821
822 static int e1000_set_features(struct net_device *netdev,
823         netdev_features_t features)
824 {
825         struct e1000_adapter *adapter = netdev_priv(netdev);
826         netdev_features_t changed = features ^ netdev->features;
827
828         if (changed & NETIF_F_HW_VLAN_CTAG_RX)
829                 e1000_vlan_mode(netdev, features);
830
831         if (!(changed & (NETIF_F_RXCSUM | NETIF_F_RXALL)))
832                 return 0;
833
834         netdev->features = features;
835         adapter->rx_csum = !!(features & NETIF_F_RXCSUM);
836
837         if (netif_running(netdev))
838                 e1000_reinit_locked(adapter);
839         else
840                 e1000_reset(adapter);
841
842         return 0;
843 }
844
845 static const struct net_device_ops e1000_netdev_ops = {
846         .ndo_open               = e1000_open,
847         .ndo_stop               = e1000_close,
848         .ndo_start_xmit         = e1000_xmit_frame,
849         .ndo_get_stats          = e1000_get_stats,
850         .ndo_set_rx_mode        = e1000_set_rx_mode,
851         .ndo_set_mac_address    = e1000_set_mac,
852         .ndo_tx_timeout         = e1000_tx_timeout,
853         .ndo_change_mtu         = e1000_change_mtu,
854         .ndo_do_ioctl           = e1000_ioctl,
855         .ndo_validate_addr      = eth_validate_addr,
856         .ndo_vlan_rx_add_vid    = e1000_vlan_rx_add_vid,
857         .ndo_vlan_rx_kill_vid   = e1000_vlan_rx_kill_vid,
858 #ifdef CONFIG_NET_POLL_CONTROLLER
859         .ndo_poll_controller    = e1000_netpoll,
860 #endif
861         .ndo_fix_features       = e1000_fix_features,
862         .ndo_set_features       = e1000_set_features,
863 };
864
865 /**
866  * e1000_init_hw_struct - initialize members of hw struct
867  * @adapter: board private struct
868  * @hw: structure used by e1000_hw.c
869  *
870  * Factors out initialization of the e1000_hw struct to its own function
871  * that can be called very early at init (just after struct allocation).
872  * Fields are initialized based on PCI device information and
873  * OS network device settings (MTU size).
874  * Returns negative error codes if MAC type setup fails.
875  */
876 static int e1000_init_hw_struct(struct e1000_adapter *adapter,
877                                 struct e1000_hw *hw)
878 {
879         struct pci_dev *pdev = adapter->pdev;
880
881         /* PCI config space info */
882         hw->vendor_id = pdev->vendor;
883         hw->device_id = pdev->device;
884         hw->subsystem_vendor_id = pdev->subsystem_vendor;
885         hw->subsystem_id = pdev->subsystem_device;
886         hw->revision_id = pdev->revision;
887
888         pci_read_config_word(pdev, PCI_COMMAND, &hw->pci_cmd_word);
889
890         hw->max_frame_size = adapter->netdev->mtu +
891                              ENET_HEADER_SIZE + ETHERNET_FCS_SIZE;
892         hw->min_frame_size = MINIMUM_ETHERNET_FRAME_SIZE;
893
894         /* identify the MAC */
895         if (e1000_set_mac_type(hw)) {
896                 e_err(probe, "Unknown MAC Type\n");
897                 return -EIO;
898         }
899
900         switch (hw->mac_type) {
901         default:
902                 break;
903         case e1000_82541:
904         case e1000_82547:
905         case e1000_82541_rev_2:
906         case e1000_82547_rev_2:
907                 hw->phy_init_script = 1;
908                 break;
909         }
910
911         e1000_set_media_type(hw);
912         e1000_get_bus_info(hw);
913
914         hw->wait_autoneg_complete = false;
915         hw->tbi_compatibility_en = true;
916         hw->adaptive_ifs = true;
917
918         /* Copper options */
919
920         if (hw->media_type == e1000_media_type_copper) {
921                 hw->mdix = AUTO_ALL_MODES;
922                 hw->disable_polarity_correction = false;
923                 hw->master_slave = E1000_MASTER_SLAVE;
924         }
925
926         return 0;
927 }
928
929 /**
930  * e1000_probe - Device Initialization Routine
931  * @pdev: PCI device information struct
932  * @ent: entry in e1000_pci_tbl
933  *
934  * Returns 0 on success, negative on failure
935  *
936  * e1000_probe initializes an adapter identified by a pci_dev structure.
937  * The OS initialization, configuring of the adapter private structure,
938  * and a hardware reset occur.
939  **/
940 static int e1000_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
941 {
942         struct net_device *netdev;
943         struct e1000_adapter *adapter;
944         struct e1000_hw *hw;
945
946         static int cards_found = 0;
947         static int global_quad_port_a = 0; /* global ksp3 port a indication */
948         int i, err, pci_using_dac;
949         u16 eeprom_data = 0;
950         u16 tmp = 0;
951         u16 eeprom_apme_mask = E1000_EEPROM_APME;
952         int bars, need_ioport;
953
954         /* do not allocate ioport bars when not needed */
955         need_ioport = e1000_is_need_ioport(pdev);
956         if (need_ioport) {
957                 bars = pci_select_bars(pdev, IORESOURCE_MEM | IORESOURCE_IO);
958                 err = pci_enable_device(pdev);
959         } else {
960                 bars = pci_select_bars(pdev, IORESOURCE_MEM);
961                 err = pci_enable_device_mem(pdev);
962         }
963         if (err)
964                 return err;
965
966         err = pci_request_selected_regions(pdev, bars, e1000_driver_name);
967         if (err)
968                 goto err_pci_reg;
969
970         pci_set_master(pdev);
971         err = pci_save_state(pdev);
972         if (err)
973                 goto err_alloc_etherdev;
974
975         err = -ENOMEM;
976         netdev = alloc_etherdev(sizeof(struct e1000_adapter));
977         if (!netdev)
978                 goto err_alloc_etherdev;
979
980         SET_NETDEV_DEV(netdev, &pdev->dev);
981
982         pci_set_drvdata(pdev, netdev);
983         adapter = netdev_priv(netdev);
984         adapter->netdev = netdev;
985         adapter->pdev = pdev;
986         adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
987         adapter->bars = bars;
988         adapter->need_ioport = need_ioport;
989
990         hw = &adapter->hw;
991         hw->back = adapter;
992
993         err = -EIO;
994         hw->hw_addr = pci_ioremap_bar(pdev, BAR_0);
995         if (!hw->hw_addr)
996                 goto err_ioremap;
997
998         if (adapter->need_ioport) {
999                 for (i = BAR_1; i <= BAR_5; i++) {
1000                         if (pci_resource_len(pdev, i) == 0)
1001                                 continue;
1002                         if (pci_resource_flags(pdev, i) & IORESOURCE_IO) {
1003                                 hw->io_base = pci_resource_start(pdev, i);
1004                                 break;
1005                         }
1006                 }
1007         }
1008
1009         /* make ready for any if (hw->...) below */
1010         err = e1000_init_hw_struct(adapter, hw);
1011         if (err)
1012                 goto err_sw_init;
1013
1014         /* there is a workaround being applied below that limits
1015          * 64-bit DMA addresses to 64-bit hardware.  There are some
1016          * 32-bit adapters that Tx hang when given 64-bit DMA addresses
1017          */
1018         pci_using_dac = 0;
1019         if ((hw->bus_type == e1000_bus_type_pcix) &&
1020             !dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64))) {
1021                 pci_using_dac = 1;
1022         } else {
1023                 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
1024                 if (err) {
1025                         pr_err("No usable DMA config, aborting\n");
1026                         goto err_dma;
1027                 }
1028         }
1029
1030         netdev->netdev_ops = &e1000_netdev_ops;
1031         e1000_set_ethtool_ops(netdev);
1032         netdev->watchdog_timeo = 5 * HZ;
1033         netif_napi_add(netdev, &adapter->napi, e1000_clean, 64);
1034
1035         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
1036
1037         adapter->bd_number = cards_found;
1038
1039         /* setup the private structure */
1040
1041         err = e1000_sw_init(adapter);
1042         if (err)
1043                 goto err_sw_init;
1044
1045         err = -EIO;
1046         if (hw->mac_type == e1000_ce4100) {
1047                 hw->ce4100_gbe_mdio_base_virt =
1048                                         ioremap(pci_resource_start(pdev, BAR_1),
1049                                                 pci_resource_len(pdev, BAR_1));
1050
1051                 if (!hw->ce4100_gbe_mdio_base_virt)
1052                         goto err_mdio_ioremap;
1053         }
1054
1055         if (hw->mac_type >= e1000_82543) {
1056                 netdev->hw_features = NETIF_F_SG |
1057                                    NETIF_F_HW_CSUM |
1058                                    NETIF_F_HW_VLAN_CTAG_RX;
1059                 netdev->features = NETIF_F_HW_VLAN_CTAG_TX |
1060                                    NETIF_F_HW_VLAN_CTAG_FILTER;
1061         }
1062
1063         if ((hw->mac_type >= e1000_82544) &&
1064            (hw->mac_type != e1000_82547))
1065                 netdev->hw_features |= NETIF_F_TSO;
1066
1067         netdev->priv_flags |= IFF_SUPP_NOFCS;
1068
1069         netdev->features |= netdev->hw_features;
1070         netdev->hw_features |= (NETIF_F_RXCSUM |
1071                                 NETIF_F_RXALL |
1072                                 NETIF_F_RXFCS);
1073
1074         if (pci_using_dac) {
1075                 netdev->features |= NETIF_F_HIGHDMA;
1076                 netdev->vlan_features |= NETIF_F_HIGHDMA;
1077         }
1078
1079         netdev->vlan_features |= (NETIF_F_TSO |
1080                                   NETIF_F_HW_CSUM |
1081                                   NETIF_F_SG);
1082
1083         /* Do not set IFF_UNICAST_FLT for VMWare's 82545EM */
1084         if (hw->device_id != E1000_DEV_ID_82545EM_COPPER ||
1085             hw->subsystem_vendor_id != PCI_VENDOR_ID_VMWARE)
1086                 netdev->priv_flags |= IFF_UNICAST_FLT;
1087
1088         adapter->en_mng_pt = e1000_enable_mng_pass_thru(hw);
1089
1090         /* initialize eeprom parameters */
1091         if (e1000_init_eeprom_params(hw)) {
1092                 e_err(probe, "EEPROM initialization failed\n");
1093                 goto err_eeprom;
1094         }
1095
1096         /* before reading the EEPROM, reset the controller to
1097          * put the device in a known good starting state
1098          */
1099
1100         e1000_reset_hw(hw);
1101
1102         /* make sure the EEPROM is good */
1103         if (e1000_validate_eeprom_checksum(hw) < 0) {
1104                 e_err(probe, "The EEPROM Checksum Is Not Valid\n");
1105                 e1000_dump_eeprom(adapter);
1106                 /* set MAC address to all zeroes to invalidate and temporary
1107                  * disable this device for the user. This blocks regular
1108                  * traffic while still permitting ethtool ioctls from reaching
1109                  * the hardware as well as allowing the user to run the
1110                  * interface after manually setting a hw addr using
1111                  * `ip set address`
1112                  */
1113                 memset(hw->mac_addr, 0, netdev->addr_len);
1114         } else {
1115                 /* copy the MAC address out of the EEPROM */
1116                 if (e1000_read_mac_addr(hw))
1117                         e_err(probe, "EEPROM Read Error\n");
1118         }
1119         /* don't block initialization here due to bad MAC address */
1120         memcpy(netdev->dev_addr, hw->mac_addr, netdev->addr_len);
1121
1122         if (!is_valid_ether_addr(netdev->dev_addr))
1123                 e_err(probe, "Invalid MAC Address\n");
1124
1125
1126         INIT_DELAYED_WORK(&adapter->watchdog_task, e1000_watchdog);
1127         INIT_DELAYED_WORK(&adapter->fifo_stall_task,
1128                           e1000_82547_tx_fifo_stall_task);
1129         INIT_DELAYED_WORK(&adapter->phy_info_task, e1000_update_phy_info_task);
1130         INIT_WORK(&adapter->reset_task, e1000_reset_task);
1131
1132         e1000_check_options(adapter);
1133
1134         /* Initial Wake on LAN setting
1135          * If APM wake is enabled in the EEPROM,
1136          * enable the ACPI Magic Packet filter
1137          */
1138
1139         switch (hw->mac_type) {
1140         case e1000_82542_rev2_0:
1141         case e1000_82542_rev2_1:
1142         case e1000_82543:
1143                 break;
1144         case e1000_82544:
1145                 e1000_read_eeprom(hw,
1146                         EEPROM_INIT_CONTROL2_REG, 1, &eeprom_data);
1147                 eeprom_apme_mask = E1000_EEPROM_82544_APM;
1148                 break;
1149         case e1000_82546:
1150         case e1000_82546_rev_3:
1151                 if (er32(STATUS) & E1000_STATUS_FUNC_1){
1152                         e1000_read_eeprom(hw,
1153                                 EEPROM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
1154                         break;
1155                 }
1156                 /* Fall Through */
1157         default:
1158                 e1000_read_eeprom(hw,
1159                         EEPROM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
1160                 break;
1161         }
1162         if (eeprom_data & eeprom_apme_mask)
1163                 adapter->eeprom_wol |= E1000_WUFC_MAG;
1164
1165         /* now that we have the eeprom settings, apply the special cases
1166          * where the eeprom may be wrong or the board simply won't support
1167          * wake on lan on a particular port
1168          */
1169         switch (pdev->device) {
1170         case E1000_DEV_ID_82546GB_PCIE:
1171                 adapter->eeprom_wol = 0;
1172                 break;
1173         case E1000_DEV_ID_82546EB_FIBER:
1174         case E1000_DEV_ID_82546GB_FIBER:
1175                 /* Wake events only supported on port A for dual fiber
1176                  * regardless of eeprom setting
1177                  */
1178                 if (er32(STATUS) & E1000_STATUS_FUNC_1)
1179                         adapter->eeprom_wol = 0;
1180                 break;
1181         case E1000_DEV_ID_82546GB_QUAD_COPPER_KSP3:
1182                 /* if quad port adapter, disable WoL on all but port A */
1183                 if (global_quad_port_a != 0)
1184                         adapter->eeprom_wol = 0;
1185                 else
1186                         adapter->quad_port_a = true;
1187                 /* Reset for multiple quad port adapters */
1188                 if (++global_quad_port_a == 4)
1189                         global_quad_port_a = 0;
1190                 break;
1191         }
1192
1193         /* initialize the wol settings based on the eeprom settings */
1194         adapter->wol = adapter->eeprom_wol;
1195         device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
1196
1197         /* Auto detect PHY address */
1198         if (hw->mac_type == e1000_ce4100) {
1199                 for (i = 0; i < 32; i++) {
1200                         hw->phy_addr = i;
1201                         e1000_read_phy_reg(hw, PHY_ID2, &tmp);
1202                         if (tmp == 0 || tmp == 0xFF) {
1203                                 if (i == 31)
1204                                         goto err_eeprom;
1205                                 continue;
1206                         } else
1207                                 break;
1208                 }
1209         }
1210
1211         /* reset the hardware with the new settings */
1212         e1000_reset(adapter);
1213
1214         strcpy(netdev->name, "eth%d");
1215         err = register_netdev(netdev);
1216         if (err)
1217                 goto err_register;
1218
1219         e1000_vlan_filter_on_off(adapter, false);
1220
1221         /* print bus type/speed/width info */
1222         e_info(probe, "(PCI%s:%dMHz:%d-bit) %pM\n",
1223                ((hw->bus_type == e1000_bus_type_pcix) ? "-X" : ""),
1224                ((hw->bus_speed == e1000_bus_speed_133) ? 133 :
1225                 (hw->bus_speed == e1000_bus_speed_120) ? 120 :
1226                 (hw->bus_speed == e1000_bus_speed_100) ? 100 :
1227                 (hw->bus_speed == e1000_bus_speed_66) ? 66 : 33),
1228                ((hw->bus_width == e1000_bus_width_64) ? 64 : 32),
1229                netdev->dev_addr);
1230
1231         /* carrier off reporting is important to ethtool even BEFORE open */
1232         netif_carrier_off(netdev);
1233
1234         e_info(probe, "Intel(R) PRO/1000 Network Connection\n");
1235
1236         cards_found++;
1237         return 0;
1238
1239 err_register:
1240 err_eeprom:
1241         e1000_phy_hw_reset(hw);
1242
1243         if (hw->flash_address)
1244                 iounmap(hw->flash_address);
1245         kfree(adapter->tx_ring);
1246         kfree(adapter->rx_ring);
1247 err_dma:
1248 err_sw_init:
1249 err_mdio_ioremap:
1250         iounmap(hw->ce4100_gbe_mdio_base_virt);
1251         iounmap(hw->hw_addr);
1252 err_ioremap:
1253         free_netdev(netdev);
1254 err_alloc_etherdev:
1255         pci_release_selected_regions(pdev, bars);
1256 err_pci_reg:
1257         pci_disable_device(pdev);
1258         return err;
1259 }
1260
1261 /**
1262  * e1000_remove - Device Removal Routine
1263  * @pdev: PCI device information struct
1264  *
1265  * e1000_remove is called by the PCI subsystem to alert the driver
1266  * that it should release a PCI device.  The could be caused by a
1267  * Hot-Plug event, or because the driver is going to be removed from
1268  * memory.
1269  **/
1270 static void e1000_remove(struct pci_dev *pdev)
1271 {
1272         struct net_device *netdev = pci_get_drvdata(pdev);
1273         struct e1000_adapter *adapter = netdev_priv(netdev);
1274         struct e1000_hw *hw = &adapter->hw;
1275
1276         e1000_down_and_stop(adapter);
1277         e1000_release_manageability(adapter);
1278
1279         unregister_netdev(netdev);
1280
1281         e1000_phy_hw_reset(hw);
1282
1283         kfree(adapter->tx_ring);
1284         kfree(adapter->rx_ring);
1285
1286         if (hw->mac_type == e1000_ce4100)
1287                 iounmap(hw->ce4100_gbe_mdio_base_virt);
1288         iounmap(hw->hw_addr);
1289         if (hw->flash_address)
1290                 iounmap(hw->flash_address);
1291         pci_release_selected_regions(pdev, adapter->bars);
1292
1293         free_netdev(netdev);
1294
1295         pci_disable_device(pdev);
1296 }
1297
1298 /**
1299  * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
1300  * @adapter: board private structure to initialize
1301  *
1302  * e1000_sw_init initializes the Adapter private data structure.
1303  * e1000_init_hw_struct MUST be called before this function
1304  **/
1305 static int e1000_sw_init(struct e1000_adapter *adapter)
1306 {
1307         adapter->rx_buffer_len = MAXIMUM_ETHERNET_VLAN_SIZE;
1308
1309         adapter->num_tx_queues = 1;
1310         adapter->num_rx_queues = 1;
1311
1312         if (e1000_alloc_queues(adapter)) {
1313                 e_err(probe, "Unable to allocate memory for queues\n");
1314                 return -ENOMEM;
1315         }
1316
1317         /* Explicitly disable IRQ since the NIC can be in any state. */
1318         e1000_irq_disable(adapter);
1319
1320         spin_lock_init(&adapter->stats_lock);
1321
1322         set_bit(__E1000_DOWN, &adapter->flags);
1323
1324         return 0;
1325 }
1326
1327 /**
1328  * e1000_alloc_queues - Allocate memory for all rings
1329  * @adapter: board private structure to initialize
1330  *
1331  * We allocate one ring per queue at run-time since we don't know the
1332  * number of queues at compile-time.
1333  **/
1334 static int e1000_alloc_queues(struct e1000_adapter *adapter)
1335 {
1336         adapter->tx_ring = kcalloc(adapter->num_tx_queues,
1337                                    sizeof(struct e1000_tx_ring), GFP_KERNEL);
1338         if (!adapter->tx_ring)
1339                 return -ENOMEM;
1340
1341         adapter->rx_ring = kcalloc(adapter->num_rx_queues,
1342                                    sizeof(struct e1000_rx_ring), GFP_KERNEL);
1343         if (!adapter->rx_ring) {
1344                 kfree(adapter->tx_ring);
1345                 return -ENOMEM;
1346         }
1347
1348         return E1000_SUCCESS;
1349 }
1350
1351 /**
1352  * e1000_open - Called when a network interface is made active
1353  * @netdev: network interface device structure
1354  *
1355  * Returns 0 on success, negative value on failure
1356  *
1357  * The open entry point is called when a network interface is made
1358  * active by the system (IFF_UP).  At this point all resources needed
1359  * for transmit and receive operations are allocated, the interrupt
1360  * handler is registered with the OS, the watchdog task is started,
1361  * and the stack is notified that the interface is ready.
1362  **/
1363 static int e1000_open(struct net_device *netdev)
1364 {
1365         struct e1000_adapter *adapter = netdev_priv(netdev);
1366         struct e1000_hw *hw = &adapter->hw;
1367         int err;
1368
1369         /* disallow open during test */
1370         if (test_bit(__E1000_TESTING, &adapter->flags))
1371                 return -EBUSY;
1372
1373         netif_carrier_off(netdev);
1374
1375         /* allocate transmit descriptors */
1376         err = e1000_setup_all_tx_resources(adapter);
1377         if (err)
1378                 goto err_setup_tx;
1379
1380         /* allocate receive descriptors */
1381         err = e1000_setup_all_rx_resources(adapter);
1382         if (err)
1383                 goto err_setup_rx;
1384
1385         e1000_power_up_phy(adapter);
1386
1387         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
1388         if ((hw->mng_cookie.status &
1389                           E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT)) {
1390                 e1000_update_mng_vlan(adapter);
1391         }
1392
1393         /* before we allocate an interrupt, we must be ready to handle it.
1394          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
1395          * as soon as we call pci_request_irq, so we have to setup our
1396          * clean_rx handler before we do so.
1397          */
1398         e1000_configure(adapter);
1399
1400         err = e1000_request_irq(adapter);
1401         if (err)
1402                 goto err_req_irq;
1403
1404         /* From here on the code is the same as e1000_up() */
1405         clear_bit(__E1000_DOWN, &adapter->flags);
1406
1407         napi_enable(&adapter->napi);
1408
1409         e1000_irq_enable(adapter);
1410
1411         netif_start_queue(netdev);
1412
1413         /* fire a link status change interrupt to start the watchdog */
1414         ew32(ICS, E1000_ICS_LSC);
1415
1416         return E1000_SUCCESS;
1417
1418 err_req_irq:
1419         e1000_power_down_phy(adapter);
1420         e1000_free_all_rx_resources(adapter);
1421 err_setup_rx:
1422         e1000_free_all_tx_resources(adapter);
1423 err_setup_tx:
1424         e1000_reset(adapter);
1425
1426         return err;
1427 }
1428
1429 /**
1430  * e1000_close - Disables a network interface
1431  * @netdev: network interface device structure
1432  *
1433  * Returns 0, this is not allowed to fail
1434  *
1435  * The close entry point is called when an interface is de-activated
1436  * by the OS.  The hardware is still under the drivers control, but
1437  * needs to be disabled.  A global MAC reset is issued to stop the
1438  * hardware, and all transmit and receive resources are freed.
1439  **/
1440 static int e1000_close(struct net_device *netdev)
1441 {
1442         struct e1000_adapter *adapter = netdev_priv(netdev);
1443         struct e1000_hw *hw = &adapter->hw;
1444         int count = E1000_CHECK_RESET_COUNT;
1445
1446         while (test_bit(__E1000_RESETTING, &adapter->flags) && count--)
1447                 usleep_range(10000, 20000);
1448
1449         WARN_ON(test_bit(__E1000_RESETTING, &adapter->flags));
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 static void e1000_free_frag(const void *data)
2083 {
2084         put_page(virt_to_head_page(data));
2085 }
2086
2087 /**
2088  * e1000_clean_rx_ring - Free Rx Buffers per Queue
2089  * @adapter: board private structure
2090  * @rx_ring: ring to free buffers from
2091  **/
2092 static void e1000_clean_rx_ring(struct e1000_adapter *adapter,
2093                                 struct e1000_rx_ring *rx_ring)
2094 {
2095         struct e1000_hw *hw = &adapter->hw;
2096         struct e1000_rx_buffer *buffer_info;
2097         struct pci_dev *pdev = adapter->pdev;
2098         unsigned long size;
2099         unsigned int i;
2100
2101         /* Free all the Rx netfrags */
2102         for (i = 0; i < rx_ring->count; i++) {
2103                 buffer_info = &rx_ring->buffer_info[i];
2104                 if (adapter->clean_rx == e1000_clean_rx_irq) {
2105                         if (buffer_info->dma)
2106                                 dma_unmap_single(&pdev->dev, buffer_info->dma,
2107                                                  adapter->rx_buffer_len,
2108                                                  DMA_FROM_DEVICE);
2109                         if (buffer_info->rxbuf.data) {
2110                                 e1000_free_frag(buffer_info->rxbuf.data);
2111                                 buffer_info->rxbuf.data = NULL;
2112                         }
2113                 } else if (adapter->clean_rx == e1000_clean_jumbo_rx_irq) {
2114                         if (buffer_info->dma)
2115                                 dma_unmap_page(&pdev->dev, buffer_info->dma,
2116                                                adapter->rx_buffer_len,
2117                                                DMA_FROM_DEVICE);
2118                         if (buffer_info->rxbuf.page) {
2119                                 put_page(buffer_info->rxbuf.page);
2120                                 buffer_info->rxbuf.page = NULL;
2121                         }
2122                 }
2123
2124                 buffer_info->dma = 0;
2125         }
2126
2127         /* there also may be some cached data from a chained receive */
2128         napi_free_frags(&adapter->napi);
2129         rx_ring->rx_skb_top = NULL;
2130
2131         size = sizeof(struct e1000_rx_buffer) * rx_ring->count;
2132         memset(rx_ring->buffer_info, 0, size);
2133
2134         /* Zero out the descriptor ring */
2135         memset(rx_ring->desc, 0, rx_ring->size);
2136
2137         rx_ring->next_to_clean = 0;
2138         rx_ring->next_to_use = 0;
2139
2140         writel(0, hw->hw_addr + rx_ring->rdh);
2141         writel(0, hw->hw_addr + rx_ring->rdt);
2142 }
2143
2144 /**
2145  * e1000_clean_all_rx_rings - Free Rx Buffers for all queues
2146  * @adapter: board private structure
2147  **/
2148 static void e1000_clean_all_rx_rings(struct e1000_adapter *adapter)
2149 {
2150         int i;
2151
2152         for (i = 0; i < adapter->num_rx_queues; i++)
2153                 e1000_clean_rx_ring(adapter, &adapter->rx_ring[i]);
2154 }
2155
2156 /* The 82542 2.0 (revision 2) needs to have the receive unit in reset
2157  * and memory write and invalidate disabled for certain operations
2158  */
2159 static void e1000_enter_82542_rst(struct e1000_adapter *adapter)
2160 {
2161         struct e1000_hw *hw = &adapter->hw;
2162         struct net_device *netdev = adapter->netdev;
2163         u32 rctl;
2164
2165         e1000_pci_clear_mwi(hw);
2166
2167         rctl = er32(RCTL);
2168         rctl |= E1000_RCTL_RST;
2169         ew32(RCTL, rctl);
2170         E1000_WRITE_FLUSH();
2171         mdelay(5);
2172
2173         if (netif_running(netdev))
2174                 e1000_clean_all_rx_rings(adapter);
2175 }
2176
2177 static void e1000_leave_82542_rst(struct e1000_adapter *adapter)
2178 {
2179         struct e1000_hw *hw = &adapter->hw;
2180         struct net_device *netdev = adapter->netdev;
2181         u32 rctl;
2182
2183         rctl = er32(RCTL);
2184         rctl &= ~E1000_RCTL_RST;
2185         ew32(RCTL, rctl);
2186         E1000_WRITE_FLUSH();
2187         mdelay(5);
2188
2189         if (hw->pci_cmd_word & PCI_COMMAND_INVALIDATE)
2190                 e1000_pci_set_mwi(hw);
2191
2192         if (netif_running(netdev)) {
2193                 /* No need to loop, because 82542 supports only 1 queue */
2194                 struct e1000_rx_ring *ring = &adapter->rx_ring[0];
2195                 e1000_configure_rx(adapter);
2196                 adapter->alloc_rx_buf(adapter, ring, E1000_DESC_UNUSED(ring));
2197         }
2198 }
2199
2200 /**
2201  * e1000_set_mac - Change the Ethernet Address of the NIC
2202  * @netdev: network interface device structure
2203  * @p: pointer to an address structure
2204  *
2205  * Returns 0 on success, negative on failure
2206  **/
2207 static int e1000_set_mac(struct net_device *netdev, void *p)
2208 {
2209         struct e1000_adapter *adapter = netdev_priv(netdev);
2210         struct e1000_hw *hw = &adapter->hw;
2211         struct sockaddr *addr = p;
2212
2213         if (!is_valid_ether_addr(addr->sa_data))
2214                 return -EADDRNOTAVAIL;
2215
2216         /* 82542 2.0 needs to be in reset to write receive address registers */
2217
2218         if (hw->mac_type == e1000_82542_rev2_0)
2219                 e1000_enter_82542_rst(adapter);
2220
2221         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
2222         memcpy(hw->mac_addr, addr->sa_data, netdev->addr_len);
2223
2224         e1000_rar_set(hw, hw->mac_addr, 0);
2225
2226         if (hw->mac_type == e1000_82542_rev2_0)
2227                 e1000_leave_82542_rst(adapter);
2228
2229         return 0;
2230 }
2231
2232 /**
2233  * e1000_set_rx_mode - Secondary Unicast, Multicast and Promiscuous mode set
2234  * @netdev: network interface device structure
2235  *
2236  * The set_rx_mode entry point is called whenever the unicast or multicast
2237  * address lists or the network interface flags are updated. This routine is
2238  * responsible for configuring the hardware for proper unicast, multicast,
2239  * promiscuous mode, and all-multi behavior.
2240  **/
2241 static void e1000_set_rx_mode(struct net_device *netdev)
2242 {
2243         struct e1000_adapter *adapter = netdev_priv(netdev);
2244         struct e1000_hw *hw = &adapter->hw;
2245         struct netdev_hw_addr *ha;
2246         bool use_uc = false;
2247         u32 rctl;
2248         u32 hash_value;
2249         int i, rar_entries = E1000_RAR_ENTRIES;
2250         int mta_reg_count = E1000_NUM_MTA_REGISTERS;
2251         u32 *mcarray = kcalloc(mta_reg_count, sizeof(u32), GFP_ATOMIC);
2252
2253         if (!mcarray)
2254                 return;
2255
2256         /* Check for Promiscuous and All Multicast modes */
2257
2258         rctl = er32(RCTL);
2259
2260         if (netdev->flags & IFF_PROMISC) {
2261                 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
2262                 rctl &= ~E1000_RCTL_VFE;
2263         } else {
2264                 if (netdev->flags & IFF_ALLMULTI)
2265                         rctl |= E1000_RCTL_MPE;
2266                 else
2267                         rctl &= ~E1000_RCTL_MPE;
2268                 /* Enable VLAN filter if there is a VLAN */
2269                 if (e1000_vlan_used(adapter))
2270                         rctl |= E1000_RCTL_VFE;
2271         }
2272
2273         if (netdev_uc_count(netdev) > rar_entries - 1) {
2274                 rctl |= E1000_RCTL_UPE;
2275         } else if (!(netdev->flags & IFF_PROMISC)) {
2276                 rctl &= ~E1000_RCTL_UPE;
2277                 use_uc = true;
2278         }
2279
2280         ew32(RCTL, rctl);
2281
2282         /* 82542 2.0 needs to be in reset to write receive address registers */
2283
2284         if (hw->mac_type == e1000_82542_rev2_0)
2285                 e1000_enter_82542_rst(adapter);
2286
2287         /* load the first 14 addresses into the exact filters 1-14. Unicast
2288          * addresses take precedence to avoid disabling unicast filtering
2289          * when possible.
2290          *
2291          * RAR 0 is used for the station MAC address
2292          * if there are not 14 addresses, go ahead and clear the filters
2293          */
2294         i = 1;
2295         if (use_uc)
2296                 netdev_for_each_uc_addr(ha, netdev) {
2297                         if (i == rar_entries)
2298                                 break;
2299                         e1000_rar_set(hw, ha->addr, i++);
2300                 }
2301
2302         netdev_for_each_mc_addr(ha, netdev) {
2303                 if (i == rar_entries) {
2304                         /* load any remaining addresses into the hash table */
2305                         u32 hash_reg, hash_bit, mta;
2306                         hash_value = e1000_hash_mc_addr(hw, ha->addr);
2307                         hash_reg = (hash_value >> 5) & 0x7F;
2308                         hash_bit = hash_value & 0x1F;
2309                         mta = (1 << hash_bit);
2310                         mcarray[hash_reg] |= mta;
2311                 } else {
2312                         e1000_rar_set(hw, ha->addr, i++);
2313                 }
2314         }
2315
2316         for (; i < rar_entries; i++) {
2317                 E1000_WRITE_REG_ARRAY(hw, RA, i << 1, 0);
2318                 E1000_WRITE_FLUSH();
2319                 E1000_WRITE_REG_ARRAY(hw, RA, (i << 1) + 1, 0);
2320                 E1000_WRITE_FLUSH();
2321         }
2322
2323         /* write the hash table completely, write from bottom to avoid
2324          * both stupid write combining chipsets, and flushing each write
2325          */
2326         for (i = mta_reg_count - 1; i >= 0 ; i--) {
2327                 /* If we are on an 82544 has an errata where writing odd
2328                  * offsets overwrites the previous even offset, but writing
2329                  * backwards over the range solves the issue by always
2330                  * writing the odd offset first
2331                  */
2332                 E1000_WRITE_REG_ARRAY(hw, MTA, i, mcarray[i]);
2333         }
2334         E1000_WRITE_FLUSH();
2335
2336         if (hw->mac_type == e1000_82542_rev2_0)
2337                 e1000_leave_82542_rst(adapter);
2338
2339         kfree(mcarray);
2340 }
2341
2342 /**
2343  * e1000_update_phy_info_task - get phy info
2344  * @work: work struct contained inside adapter struct
2345  *
2346  * Need to wait a few seconds after link up to get diagnostic information from
2347  * the phy
2348  */
2349 static void e1000_update_phy_info_task(struct work_struct *work)
2350 {
2351         struct e1000_adapter *adapter = container_of(work,
2352                                                      struct e1000_adapter,
2353                                                      phy_info_task.work);
2354
2355         e1000_phy_get_info(&adapter->hw, &adapter->phy_info);
2356 }
2357
2358 /**
2359  * e1000_82547_tx_fifo_stall_task - task to complete work
2360  * @work: work struct contained inside adapter struct
2361  **/
2362 static void e1000_82547_tx_fifo_stall_task(struct work_struct *work)
2363 {
2364         struct e1000_adapter *adapter = container_of(work,
2365                                                      struct e1000_adapter,
2366                                                      fifo_stall_task.work);
2367         struct e1000_hw *hw = &adapter->hw;
2368         struct net_device *netdev = adapter->netdev;
2369         u32 tctl;
2370
2371         if (atomic_read(&adapter->tx_fifo_stall)) {
2372                 if ((er32(TDT) == er32(TDH)) &&
2373                    (er32(TDFT) == er32(TDFH)) &&
2374                    (er32(TDFTS) == er32(TDFHS))) {
2375                         tctl = er32(TCTL);
2376                         ew32(TCTL, tctl & ~E1000_TCTL_EN);
2377                         ew32(TDFT, adapter->tx_head_addr);
2378                         ew32(TDFH, adapter->tx_head_addr);
2379                         ew32(TDFTS, adapter->tx_head_addr);
2380                         ew32(TDFHS, adapter->tx_head_addr);
2381                         ew32(TCTL, tctl);
2382                         E1000_WRITE_FLUSH();
2383
2384                         adapter->tx_fifo_head = 0;
2385                         atomic_set(&adapter->tx_fifo_stall, 0);
2386                         netif_wake_queue(netdev);
2387                 } else if (!test_bit(__E1000_DOWN, &adapter->flags)) {
2388                         schedule_delayed_work(&adapter->fifo_stall_task, 1);
2389                 }
2390         }
2391 }
2392
2393 bool e1000_has_link(struct e1000_adapter *adapter)
2394 {
2395         struct e1000_hw *hw = &adapter->hw;
2396         bool link_active = false;
2397
2398         /* get_link_status is set on LSC (link status) interrupt or rx
2399          * sequence error interrupt (except on intel ce4100).
2400          * get_link_status will stay false until the
2401          * e1000_check_for_link establishes link for copper adapters
2402          * ONLY
2403          */
2404         switch (hw->media_type) {
2405         case e1000_media_type_copper:
2406                 if (hw->mac_type == e1000_ce4100)
2407                         hw->get_link_status = 1;
2408                 if (hw->get_link_status) {
2409                         e1000_check_for_link(hw);
2410                         link_active = !hw->get_link_status;
2411                 } else {
2412                         link_active = true;
2413                 }
2414                 break;
2415         case e1000_media_type_fiber:
2416                 e1000_check_for_link(hw);
2417                 link_active = !!(er32(STATUS) & E1000_STATUS_LU);
2418                 break;
2419         case e1000_media_type_internal_serdes:
2420                 e1000_check_for_link(hw);
2421                 link_active = hw->serdes_has_link;
2422                 break;
2423         default:
2424                 break;
2425         }
2426
2427         return link_active;
2428 }
2429
2430 /**
2431  * e1000_watchdog - work function
2432  * @work: work struct contained inside adapter struct
2433  **/
2434 static void e1000_watchdog(struct work_struct *work)
2435 {
2436         struct e1000_adapter *adapter = container_of(work,
2437                                                      struct e1000_adapter,
2438                                                      watchdog_task.work);
2439         struct e1000_hw *hw = &adapter->hw;
2440         struct net_device *netdev = adapter->netdev;
2441         struct e1000_tx_ring *txdr = adapter->tx_ring;
2442         u32 link, tctl;
2443
2444         link = e1000_has_link(adapter);
2445         if ((netif_carrier_ok(netdev)) && link)
2446                 goto link_up;
2447
2448         if (link) {
2449                 if (!netif_carrier_ok(netdev)) {
2450                         u32 ctrl;
2451                         bool txb2b = true;
2452                         /* update snapshot of PHY registers on LSC */
2453                         e1000_get_speed_and_duplex(hw,
2454                                                    &adapter->link_speed,
2455                                                    &adapter->link_duplex);
2456
2457                         ctrl = er32(CTRL);
2458                         pr_info("%s NIC Link is Up %d Mbps %s, "
2459                                 "Flow Control: %s\n",
2460                                 netdev->name,
2461                                 adapter->link_speed,
2462                                 adapter->link_duplex == FULL_DUPLEX ?
2463                                 "Full Duplex" : "Half Duplex",
2464                                 ((ctrl & E1000_CTRL_TFCE) && (ctrl &
2465                                 E1000_CTRL_RFCE)) ? "RX/TX" : ((ctrl &
2466                                 E1000_CTRL_RFCE) ? "RX" : ((ctrl &
2467                                 E1000_CTRL_TFCE) ? "TX" : "None")));
2468
2469                         /* adjust timeout factor according to speed/duplex */
2470                         adapter->tx_timeout_factor = 1;
2471                         switch (adapter->link_speed) {
2472                         case SPEED_10:
2473                                 txb2b = false;
2474                                 adapter->tx_timeout_factor = 16;
2475                                 break;
2476                         case SPEED_100:
2477                                 txb2b = false;
2478                                 /* maybe add some timeout factor ? */
2479                                 break;
2480                         }
2481
2482                         /* enable transmits in the hardware */
2483                         tctl = er32(TCTL);
2484                         tctl |= E1000_TCTL_EN;
2485                         ew32(TCTL, tctl);
2486
2487                         netif_carrier_on(netdev);
2488                         if (!test_bit(__E1000_DOWN, &adapter->flags))
2489                                 schedule_delayed_work(&adapter->phy_info_task,
2490                                                       2 * HZ);
2491                         adapter->smartspeed = 0;
2492                 }
2493         } else {
2494                 if (netif_carrier_ok(netdev)) {
2495                         adapter->link_speed = 0;
2496                         adapter->link_duplex = 0;
2497                         pr_info("%s NIC Link is Down\n",
2498                                 netdev->name);
2499                         netif_carrier_off(netdev);
2500
2501                         if (!test_bit(__E1000_DOWN, &adapter->flags))
2502                                 schedule_delayed_work(&adapter->phy_info_task,
2503                                                       2 * HZ);
2504                 }
2505
2506                 e1000_smartspeed(adapter);
2507         }
2508
2509 link_up:
2510         e1000_update_stats(adapter);
2511
2512         hw->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
2513         adapter->tpt_old = adapter->stats.tpt;
2514         hw->collision_delta = adapter->stats.colc - adapter->colc_old;
2515         adapter->colc_old = adapter->stats.colc;
2516
2517         adapter->gorcl = adapter->stats.gorcl - adapter->gorcl_old;
2518         adapter->gorcl_old = adapter->stats.gorcl;
2519         adapter->gotcl = adapter->stats.gotcl - adapter->gotcl_old;
2520         adapter->gotcl_old = adapter->stats.gotcl;
2521
2522         e1000_update_adaptive(hw);
2523
2524         if (!netif_carrier_ok(netdev)) {
2525                 if (E1000_DESC_UNUSED(txdr) + 1 < txdr->count) {
2526                         /* We've lost link, so the controller stops DMA,
2527                          * but we've got queued Tx work that's never going
2528                          * to get done, so reset controller to flush Tx.
2529                          * (Do the reset outside of interrupt context).
2530                          */
2531                         adapter->tx_timeout_count++;
2532                         schedule_work(&adapter->reset_task);
2533                         /* exit immediately since reset is imminent */
2534                         return;
2535                 }
2536         }
2537
2538         /* Simple mode for Interrupt Throttle Rate (ITR) */
2539         if (hw->mac_type >= e1000_82540 && adapter->itr_setting == 4) {
2540                 /* Symmetric Tx/Rx gets a reduced ITR=2000;
2541                  * Total asymmetrical Tx or Rx gets ITR=8000;
2542                  * everyone else is between 2000-8000.
2543                  */
2544                 u32 goc = (adapter->gotcl + adapter->gorcl) / 10000;
2545                 u32 dif = (adapter->gotcl > adapter->gorcl ?
2546                             adapter->gotcl - adapter->gorcl :
2547                             adapter->gorcl - adapter->gotcl) / 10000;
2548                 u32 itr = goc > 0 ? (dif * 6000 / goc + 2000) : 8000;
2549
2550                 ew32(ITR, 1000000000 / (itr * 256));
2551         }
2552
2553         /* Cause software interrupt to ensure rx ring is cleaned */
2554         ew32(ICS, E1000_ICS_RXDMT0);
2555
2556         /* Force detection of hung controller every watchdog period */
2557         adapter->detect_tx_hung = true;
2558
2559         /* Reschedule the task */
2560         if (!test_bit(__E1000_DOWN, &adapter->flags))
2561                 schedule_delayed_work(&adapter->watchdog_task, 2 * HZ);
2562 }
2563
2564 enum latency_range {
2565         lowest_latency = 0,
2566         low_latency = 1,
2567         bulk_latency = 2,
2568         latency_invalid = 255
2569 };
2570
2571 /**
2572  * e1000_update_itr - update the dynamic ITR value based on statistics
2573  * @adapter: pointer to adapter
2574  * @itr_setting: current adapter->itr
2575  * @packets: the number of packets during this measurement interval
2576  * @bytes: the number of bytes during this measurement interval
2577  *
2578  *      Stores a new ITR value based on packets and byte
2579  *      counts during the last interrupt.  The advantage of per interrupt
2580  *      computation is faster updates and more accurate ITR for the current
2581  *      traffic pattern.  Constants in this function were computed
2582  *      based on theoretical maximum wire speed and thresholds were set based
2583  *      on testing data as well as attempting to minimize response time
2584  *      while increasing bulk throughput.
2585  *      this functionality is controlled by the InterruptThrottleRate module
2586  *      parameter (see e1000_param.c)
2587  **/
2588 static unsigned int e1000_update_itr(struct e1000_adapter *adapter,
2589                                      u16 itr_setting, int packets, int bytes)
2590 {
2591         unsigned int retval = itr_setting;
2592         struct e1000_hw *hw = &adapter->hw;
2593
2594         if (unlikely(hw->mac_type < e1000_82540))
2595                 goto update_itr_done;
2596
2597         if (packets == 0)
2598                 goto update_itr_done;
2599
2600         switch (itr_setting) {
2601         case lowest_latency:
2602                 /* jumbo frames get bulk treatment*/
2603                 if (bytes/packets > 8000)
2604                         retval = bulk_latency;
2605                 else if ((packets < 5) && (bytes > 512))
2606                         retval = low_latency;
2607                 break;
2608         case low_latency:  /* 50 usec aka 20000 ints/s */
2609                 if (bytes > 10000) {
2610                         /* jumbo frames need bulk latency setting */
2611                         if (bytes/packets > 8000)
2612                                 retval = bulk_latency;
2613                         else if ((packets < 10) || ((bytes/packets) > 1200))
2614                                 retval = bulk_latency;
2615                         else if ((packets > 35))
2616                                 retval = lowest_latency;
2617                 } else if (bytes/packets > 2000)
2618                         retval = bulk_latency;
2619                 else if (packets <= 2 && bytes < 512)
2620                         retval = lowest_latency;
2621                 break;
2622         case bulk_latency: /* 250 usec aka 4000 ints/s */
2623                 if (bytes > 25000) {
2624                         if (packets > 35)
2625                                 retval = low_latency;
2626                 } else if (bytes < 6000) {
2627                         retval = low_latency;
2628                 }
2629                 break;
2630         }
2631
2632 update_itr_done:
2633         return retval;
2634 }
2635
2636 static void e1000_set_itr(struct e1000_adapter *adapter)
2637 {
2638         struct e1000_hw *hw = &adapter->hw;
2639         u16 current_itr;
2640         u32 new_itr = adapter->itr;
2641
2642         if (unlikely(hw->mac_type < e1000_82540))
2643                 return;
2644
2645         /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
2646         if (unlikely(adapter->link_speed != SPEED_1000)) {
2647                 current_itr = 0;
2648                 new_itr = 4000;
2649                 goto set_itr_now;
2650         }
2651
2652         adapter->tx_itr = e1000_update_itr(adapter, adapter->tx_itr,
2653                                            adapter->total_tx_packets,
2654                                            adapter->total_tx_bytes);
2655         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2656         if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
2657                 adapter->tx_itr = low_latency;
2658
2659         adapter->rx_itr = e1000_update_itr(adapter, adapter->rx_itr,
2660                                            adapter->total_rx_packets,
2661                                            adapter->total_rx_bytes);
2662         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2663         if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
2664                 adapter->rx_itr = low_latency;
2665
2666         current_itr = max(adapter->rx_itr, adapter->tx_itr);
2667
2668         switch (current_itr) {
2669         /* counts and packets in update_itr are dependent on these numbers */
2670         case lowest_latency:
2671                 new_itr = 70000;
2672                 break;
2673         case low_latency:
2674                 new_itr = 20000; /* aka hwitr = ~200 */
2675                 break;
2676         case bulk_latency:
2677                 new_itr = 4000;
2678                 break;
2679         default:
2680                 break;
2681         }
2682
2683 set_itr_now:
2684         if (new_itr != adapter->itr) {
2685                 /* this attempts to bias the interrupt rate towards Bulk
2686                  * by adding intermediate steps when interrupt rate is
2687                  * increasing
2688                  */
2689                 new_itr = new_itr > adapter->itr ?
2690                           min(adapter->itr + (new_itr >> 2), new_itr) :
2691                           new_itr;
2692                 adapter->itr = new_itr;
2693                 ew32(ITR, 1000000000 / (new_itr * 256));
2694         }
2695 }
2696
2697 #define E1000_TX_FLAGS_CSUM             0x00000001
2698 #define E1000_TX_FLAGS_VLAN             0x00000002
2699 #define E1000_TX_FLAGS_TSO              0x00000004
2700 #define E1000_TX_FLAGS_IPV4             0x00000008
2701 #define E1000_TX_FLAGS_NO_FCS           0x00000010
2702 #define E1000_TX_FLAGS_VLAN_MASK        0xffff0000
2703 #define E1000_TX_FLAGS_VLAN_SHIFT       16
2704
2705 static int e1000_tso(struct e1000_adapter *adapter,
2706                      struct e1000_tx_ring *tx_ring, struct sk_buff *skb,
2707                      __be16 protocol)
2708 {
2709         struct e1000_context_desc *context_desc;
2710         struct e1000_tx_buffer *buffer_info;
2711         unsigned int i;
2712         u32 cmd_length = 0;
2713         u16 ipcse = 0, tucse, mss;
2714         u8 ipcss, ipcso, tucss, tucso, hdr_len;
2715
2716         if (skb_is_gso(skb)) {
2717                 int err;
2718
2719                 err = skb_cow_head(skb, 0);
2720                 if (err < 0)
2721                         return err;
2722
2723                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
2724                 mss = skb_shinfo(skb)->gso_size;
2725                 if (protocol == htons(ETH_P_IP)) {
2726                         struct iphdr *iph = ip_hdr(skb);
2727                         iph->tot_len = 0;
2728                         iph->check = 0;
2729                         tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
2730                                                                  iph->daddr, 0,
2731                                                                  IPPROTO_TCP,
2732                                                                  0);
2733                         cmd_length = E1000_TXD_CMD_IP;
2734                         ipcse = skb_transport_offset(skb) - 1;
2735                 } else if (skb_is_gso_v6(skb)) {
2736                         ipv6_hdr(skb)->payload_len = 0;
2737                         tcp_hdr(skb)->check =
2738                                 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
2739                                                  &ipv6_hdr(skb)->daddr,
2740                                                  0, IPPROTO_TCP, 0);
2741                         ipcse = 0;
2742                 }
2743                 ipcss = skb_network_offset(skb);
2744                 ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
2745                 tucss = skb_transport_offset(skb);
2746                 tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
2747                 tucse = 0;
2748
2749                 cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
2750                                E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
2751
2752                 i = tx_ring->next_to_use;
2753                 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
2754                 buffer_info = &tx_ring->buffer_info[i];
2755
2756                 context_desc->lower_setup.ip_fields.ipcss  = ipcss;
2757                 context_desc->lower_setup.ip_fields.ipcso  = ipcso;
2758                 context_desc->lower_setup.ip_fields.ipcse  = cpu_to_le16(ipcse);
2759                 context_desc->upper_setup.tcp_fields.tucss = tucss;
2760                 context_desc->upper_setup.tcp_fields.tucso = tucso;
2761                 context_desc->upper_setup.tcp_fields.tucse = cpu_to_le16(tucse);
2762                 context_desc->tcp_seg_setup.fields.mss     = cpu_to_le16(mss);
2763                 context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
2764                 context_desc->cmd_and_length = cpu_to_le32(cmd_length);
2765
2766                 buffer_info->time_stamp = jiffies;
2767                 buffer_info->next_to_watch = i;
2768
2769                 if (++i == tx_ring->count) i = 0;
2770                 tx_ring->next_to_use = i;
2771
2772                 return true;
2773         }
2774         return false;
2775 }
2776
2777 static bool e1000_tx_csum(struct e1000_adapter *adapter,
2778                           struct e1000_tx_ring *tx_ring, struct sk_buff *skb,
2779                           __be16 protocol)
2780 {
2781         struct e1000_context_desc *context_desc;
2782         struct e1000_tx_buffer *buffer_info;
2783         unsigned int i;
2784         u8 css;
2785         u32 cmd_len = E1000_TXD_CMD_DEXT;
2786
2787         if (skb->ip_summed != CHECKSUM_PARTIAL)
2788                 return false;
2789
2790         switch (protocol) {
2791         case cpu_to_be16(ETH_P_IP):
2792                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
2793                         cmd_len |= E1000_TXD_CMD_TCP;
2794                 break;
2795         case cpu_to_be16(ETH_P_IPV6):
2796                 /* XXX not handling all IPV6 headers */
2797                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
2798                         cmd_len |= E1000_TXD_CMD_TCP;
2799                 break;
2800         default:
2801                 if (unlikely(net_ratelimit()))
2802                         e_warn(drv, "checksum_partial proto=%x!\n",
2803                                skb->protocol);
2804                 break;
2805         }
2806
2807         css = skb_checksum_start_offset(skb);
2808
2809         i = tx_ring->next_to_use;
2810         buffer_info = &tx_ring->buffer_info[i];
2811         context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
2812
2813         context_desc->lower_setup.ip_config = 0;
2814         context_desc->upper_setup.tcp_fields.tucss = css;
2815         context_desc->upper_setup.tcp_fields.tucso =
2816                 css + skb->csum_offset;
2817         context_desc->upper_setup.tcp_fields.tucse = 0;
2818         context_desc->tcp_seg_setup.data = 0;
2819         context_desc->cmd_and_length = cpu_to_le32(cmd_len);
2820
2821         buffer_info->time_stamp = jiffies;
2822         buffer_info->next_to_watch = i;
2823
2824         if (unlikely(++i == tx_ring->count)) i = 0;
2825         tx_ring->next_to_use = i;
2826
2827         return true;
2828 }
2829
2830 #define E1000_MAX_TXD_PWR       12
2831 #define E1000_MAX_DATA_PER_TXD  (1<<E1000_MAX_TXD_PWR)
2832
2833 static int e1000_tx_map(struct e1000_adapter *adapter,
2834                         struct e1000_tx_ring *tx_ring,
2835                         struct sk_buff *skb, unsigned int first,
2836                         unsigned int max_per_txd, unsigned int nr_frags,
2837                         unsigned int mss)
2838 {
2839         struct e1000_hw *hw = &adapter->hw;
2840         struct pci_dev *pdev = adapter->pdev;
2841         struct e1000_tx_buffer *buffer_info;
2842         unsigned int len = skb_headlen(skb);
2843         unsigned int offset = 0, size, count = 0, i;
2844         unsigned int f, bytecount, segs;
2845
2846         i = tx_ring->next_to_use;
2847
2848         while (len) {
2849                 buffer_info = &tx_ring->buffer_info[i];
2850                 size = min(len, max_per_txd);
2851                 /* Workaround for Controller erratum --
2852                  * descriptor for non-tso packet in a linear SKB that follows a
2853                  * tso gets written back prematurely before the data is fully
2854                  * DMA'd to the controller
2855                  */
2856                 if (!skb->data_len && tx_ring->last_tx_tso &&
2857                     !skb_is_gso(skb)) {
2858                         tx_ring->last_tx_tso = false;
2859                         size -= 4;
2860                 }
2861
2862                 /* Workaround for premature desc write-backs
2863                  * in TSO mode.  Append 4-byte sentinel desc
2864                  */
2865                 if (unlikely(mss && !nr_frags && size == len && size > 8))
2866                         size -= 4;
2867                 /* work-around for errata 10 and it applies
2868                  * to all controllers in PCI-X mode
2869                  * The fix is to make sure that the first descriptor of a
2870                  * packet is smaller than 2048 - 16 - 16 (or 2016) bytes
2871                  */
2872                 if (unlikely((hw->bus_type == e1000_bus_type_pcix) &&
2873                                 (size > 2015) && count == 0))
2874                         size = 2015;
2875
2876                 /* Workaround for potential 82544 hang in PCI-X.  Avoid
2877                  * terminating buffers within evenly-aligned dwords.
2878                  */
2879                 if (unlikely(adapter->pcix_82544 &&
2880                    !((unsigned long)(skb->data + offset + size - 1) & 4) &&
2881                    size > 4))
2882                         size -= 4;
2883
2884                 buffer_info->length = size;
2885                 /* set time_stamp *before* dma to help avoid a possible race */
2886                 buffer_info->time_stamp = jiffies;
2887                 buffer_info->mapped_as_page = false;
2888                 buffer_info->dma = dma_map_single(&pdev->dev,
2889                                                   skb->data + offset,
2890                                                   size, DMA_TO_DEVICE);
2891                 if (dma_mapping_error(&pdev->dev, buffer_info->dma))
2892                         goto dma_error;
2893                 buffer_info->next_to_watch = i;
2894
2895                 len -= size;
2896                 offset += size;
2897                 count++;
2898                 if (len) {
2899                         i++;
2900                         if (unlikely(i == tx_ring->count))
2901                                 i = 0;
2902                 }
2903         }
2904
2905         for (f = 0; f < nr_frags; f++) {
2906                 const struct skb_frag_struct *frag;
2907
2908                 frag = &skb_shinfo(skb)->frags[f];
2909                 len = skb_frag_size(frag);
2910                 offset = 0;
2911
2912                 while (len) {
2913                         unsigned long bufend;
2914                         i++;
2915                         if (unlikely(i == tx_ring->count))
2916                                 i = 0;
2917
2918                         buffer_info = &tx_ring->buffer_info[i];
2919                         size = min(len, max_per_txd);
2920                         /* Workaround for premature desc write-backs
2921                          * in TSO mode.  Append 4-byte sentinel desc
2922                          */
2923                         if (unlikely(mss && f == (nr_frags-1) &&
2924                             size == len && size > 8))
2925                                 size -= 4;
2926                         /* Workaround for potential 82544 hang in PCI-X.
2927                          * Avoid terminating buffers within evenly-aligned
2928                          * dwords.
2929                          */
2930                         bufend = (unsigned long)
2931                                 page_to_phys(skb_frag_page(frag));
2932                         bufend += offset + size - 1;
2933                         if (unlikely(adapter->pcix_82544 &&
2934                                      !(bufend & 4) &&
2935                                      size > 4))
2936                                 size -= 4;
2937
2938                         buffer_info->length = size;
2939                         buffer_info->time_stamp = jiffies;
2940                         buffer_info->mapped_as_page = true;
2941                         buffer_info->dma = skb_frag_dma_map(&pdev->dev, frag,
2942                                                 offset, size, DMA_TO_DEVICE);
2943                         if (dma_mapping_error(&pdev->dev, buffer_info->dma))
2944                                 goto dma_error;
2945                         buffer_info->next_to_watch = i;
2946
2947                         len -= size;
2948                         offset += size;
2949                         count++;
2950                 }
2951         }
2952
2953         segs = skb_shinfo(skb)->gso_segs ?: 1;
2954         /* multiply data chunks by size of headers */
2955         bytecount = ((segs - 1) * skb_headlen(skb)) + skb->len;
2956
2957         tx_ring->buffer_info[i].skb = skb;
2958         tx_ring->buffer_info[i].segs = segs;
2959         tx_ring->buffer_info[i].bytecount = bytecount;
2960         tx_ring->buffer_info[first].next_to_watch = i;
2961
2962         return count;
2963
2964 dma_error:
2965         dev_err(&pdev->dev, "TX DMA map failed\n");
2966         buffer_info->dma = 0;
2967         if (count)
2968                 count--;
2969
2970         while (count--) {
2971                 if (i==0)
2972                         i += tx_ring->count;
2973                 i--;
2974                 buffer_info = &tx_ring->buffer_info[i];
2975                 e1000_unmap_and_free_tx_resource(adapter, buffer_info);
2976         }
2977
2978         return 0;
2979 }
2980
2981 static void e1000_tx_queue(struct e1000_adapter *adapter,
2982                            struct e1000_tx_ring *tx_ring, int tx_flags,
2983                            int count)
2984 {
2985         struct e1000_tx_desc *tx_desc = NULL;
2986         struct e1000_tx_buffer *buffer_info;
2987         u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
2988         unsigned int i;
2989
2990         if (likely(tx_flags & E1000_TX_FLAGS_TSO)) {
2991                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
2992                              E1000_TXD_CMD_TSE;
2993                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
2994
2995                 if (likely(tx_flags & E1000_TX_FLAGS_IPV4))
2996                         txd_upper |= E1000_TXD_POPTS_IXSM << 8;
2997         }
2998
2999         if (likely(tx_flags & E1000_TX_FLAGS_CSUM)) {
3000                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
3001                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
3002         }
3003
3004         if (unlikely(tx_flags & E1000_TX_FLAGS_VLAN)) {
3005                 txd_lower |= E1000_TXD_CMD_VLE;
3006                 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
3007         }
3008
3009         if (unlikely(tx_flags & E1000_TX_FLAGS_NO_FCS))
3010                 txd_lower &= ~(E1000_TXD_CMD_IFCS);
3011
3012         i = tx_ring->next_to_use;
3013
3014         while (count--) {
3015                 buffer_info = &tx_ring->buffer_info[i];
3016                 tx_desc = E1000_TX_DESC(*tx_ring, i);
3017                 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
3018                 tx_desc->lower.data =
3019                         cpu_to_le32(txd_lower | buffer_info->length);
3020                 tx_desc->upper.data = cpu_to_le32(txd_upper);
3021                 if (unlikely(++i == tx_ring->count)) 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) >> (X)) + 1 )
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                                 unsigned int pull_size;
3158                         case e1000_82544:
3159                                 /* Make sure we have room to chop off 4 bytes,
3160                                  * and that the end alignment will work out to
3161                                  * this hardware's requirements
3162                                  * NOTE: this is a TSO only workaround
3163                                  * if end byte alignment not correct move us
3164                                  * into the next dword
3165                                  */
3166                                 if ((unsigned long)(skb_tail_pointer(skb) - 1)
3167                                     & 4)
3168                                         break;
3169                                 /* fall through */
3170                                 pull_size = min((unsigned int)4, skb->data_len);
3171                                 if (!__pskb_pull_tail(skb, pull_size)) {
3172                                         e_err(drv, "__pskb_pull_tail "
3173                                               "failed.\n");
3174                                         dev_kfree_skb_any(skb);
3175                                         return NETDEV_TX_OK;
3176                                 }
3177                                 len = skb_headlen(skb);
3178                                 break;
3179                         default:
3180                                 /* do nothing */
3181                                 break;
3182                         }
3183                 }
3184         }
3185
3186         /* reserve a descriptor for the offload context */
3187         if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
3188                 count++;
3189         count++;
3190
3191         /* Controller Erratum workaround */
3192         if (!skb->data_len && tx_ring->last_tx_tso && !skb_is_gso(skb))
3193                 count++;
3194
3195         count += TXD_USE_COUNT(len, max_txd_pwr);
3196
3197         if (adapter->pcix_82544)
3198                 count++;
3199
3200         /* work-around for errata 10 and it applies to all controllers
3201          * in PCI-X mode, so add one more descriptor to the count
3202          */
3203         if (unlikely((hw->bus_type == e1000_bus_type_pcix) &&
3204                         (len > 2015)))
3205                 count++;
3206
3207         nr_frags = skb_shinfo(skb)->nr_frags;
3208         for (f = 0; f < nr_frags; f++)
3209                 count += TXD_USE_COUNT(skb_frag_size(&skb_shinfo(skb)->frags[f]),
3210                                        max_txd_pwr);
3211         if (adapter->pcix_82544)
3212                 count += nr_frags;
3213
3214         /* need: count + 2 desc gap to keep tail from touching
3215          * head, otherwise try next time
3216          */
3217         if (unlikely(e1000_maybe_stop_tx(netdev, tx_ring, count + 2)))
3218                 return NETDEV_TX_BUSY;
3219
3220         if (unlikely((hw->mac_type == e1000_82547) &&
3221                      (e1000_82547_fifo_workaround(adapter, skb)))) {
3222                 netif_stop_queue(netdev);
3223                 if (!test_bit(__E1000_DOWN, &adapter->flags))
3224                         schedule_delayed_work(&adapter->fifo_stall_task, 1);
3225                 return NETDEV_TX_BUSY;
3226         }
3227
3228         if (skb_vlan_tag_present(skb)) {
3229                 tx_flags |= E1000_TX_FLAGS_VLAN;
3230                 tx_flags |= (skb_vlan_tag_get(skb) <<
3231                              E1000_TX_FLAGS_VLAN_SHIFT);
3232         }
3233
3234         first = tx_ring->next_to_use;
3235
3236         tso = e1000_tso(adapter, tx_ring, skb, protocol);
3237         if (tso < 0) {
3238                 dev_kfree_skb_any(skb);
3239                 return NETDEV_TX_OK;
3240         }
3241
3242         if (likely(tso)) {
3243                 if (likely(hw->mac_type != e1000_82544))
3244                         tx_ring->last_tx_tso = true;
3245                 tx_flags |= E1000_TX_FLAGS_TSO;
3246         } else if (likely(e1000_tx_csum(adapter, tx_ring, skb, protocol)))
3247                 tx_flags |= E1000_TX_FLAGS_CSUM;
3248
3249         if (protocol == htons(ETH_P_IP))
3250                 tx_flags |= E1000_TX_FLAGS_IPV4;
3251
3252         if (unlikely(skb->no_fcs))
3253                 tx_flags |= E1000_TX_FLAGS_NO_FCS;
3254
3255         count = e1000_tx_map(adapter, tx_ring, skb, first, max_per_txd,
3256                              nr_frags, mss);
3257
3258         if (count) {
3259                 netdev_sent_queue(netdev, skb->len);
3260                 skb_tx_timestamp(skb);
3261
3262                 e1000_tx_queue(adapter, tx_ring, tx_flags, count);
3263                 /* Make sure there is space in the ring for the next send. */
3264                 e1000_maybe_stop_tx(netdev, tx_ring, MAX_SKB_FRAGS + 2);
3265
3266                 if (!skb->xmit_more ||
3267                     netif_xmit_stopped(netdev_get_tx_queue(netdev, 0))) {
3268                         writel(tx_ring->next_to_use, hw->hw_addr + tx_ring->tdt);
3269                         /* we need this if more than one processor can write to
3270                          * our tail at a time, it synchronizes IO on IA64/Altix
3271                          * systems
3272                          */
3273                         mmiowb();
3274                 }
3275         } else {
3276                 dev_kfree_skb_any(skb);
3277                 tx_ring->buffer_info[first].time_stamp = 0;
3278                 tx_ring->next_to_use = first;
3279         }
3280
3281         return NETDEV_TX_OK;
3282 }
3283
3284 #define NUM_REGS 38 /* 1 based count */
3285 static void e1000_regdump(struct e1000_adapter *adapter)
3286 {
3287         struct e1000_hw *hw = &adapter->hw;
3288         u32 regs[NUM_REGS];
3289         u32 *regs_buff = regs;
3290         int i = 0;
3291
3292         static const char * const reg_name[] = {
3293                 "CTRL",  "STATUS",
3294                 "RCTL", "RDLEN", "RDH", "RDT", "RDTR",
3295                 "TCTL", "TDBAL", "TDBAH", "TDLEN", "TDH", "TDT",
3296                 "TIDV", "TXDCTL", "TADV", "TARC0",
3297                 "TDBAL1", "TDBAH1", "TDLEN1", "TDH1", "TDT1",
3298                 "TXDCTL1", "TARC1",
3299                 "CTRL_EXT", "ERT", "RDBAL", "RDBAH",
3300                 "TDFH", "TDFT", "TDFHS", "TDFTS", "TDFPC",
3301                 "RDFH", "RDFT", "RDFHS", "RDFTS", "RDFPC"
3302         };
3303
3304         regs_buff[0]  = er32(CTRL);
3305         regs_buff[1]  = er32(STATUS);
3306
3307         regs_buff[2]  = er32(RCTL);
3308         regs_buff[3]  = er32(RDLEN);
3309         regs_buff[4]  = er32(RDH);
3310         regs_buff[5]  = er32(RDT);
3311         regs_buff[6]  = er32(RDTR);
3312
3313         regs_buff[7]  = er32(TCTL);
3314         regs_buff[8]  = er32(TDBAL);
3315         regs_buff[9]  = er32(TDBAH);
3316         regs_buff[10] = er32(TDLEN);
3317         regs_buff[11] = er32(TDH);
3318         regs_buff[12] = er32(TDT);
3319         regs_buff[13] = er32(TIDV);
3320         regs_buff[14] = er32(TXDCTL);
3321         regs_buff[15] = er32(TADV);
3322         regs_buff[16] = er32(TARC0);
3323
3324         regs_buff[17] = er32(TDBAL1);
3325         regs_buff[18] = er32(TDBAH1);
3326         regs_buff[19] = er32(TDLEN1);
3327         regs_buff[20] = er32(TDH1);
3328         regs_buff[21] = er32(TDT1);
3329         regs_buff[22] = er32(TXDCTL1);
3330         regs_buff[23] = er32(TARC1);
3331         regs_buff[24] = er32(CTRL_EXT);
3332         regs_buff[25] = er32(ERT);
3333         regs_buff[26] = er32(RDBAL0);
3334         regs_buff[27] = er32(RDBAH0);
3335         regs_buff[28] = er32(TDFH);
3336         regs_buff[29] = er32(TDFT);
3337         regs_buff[30] = er32(TDFHS);
3338         regs_buff[31] = er32(TDFTS);
3339         regs_buff[32] = er32(TDFPC);
3340         regs_buff[33] = er32(RDFH);
3341         regs_buff[34] = er32(RDFT);
3342         regs_buff[35] = er32(RDFHS);
3343         regs_buff[36] = er32(RDFTS);
3344         regs_buff[37] = er32(RDFPC);
3345
3346         pr_info("Register dump\n");
3347         for (i = 0; i < NUM_REGS; i++)
3348                 pr_info("%-15s  %08x\n", reg_name[i], regs_buff[i]);
3349 }
3350
3351 /*
3352  * e1000_dump: Print registers, tx ring and rx ring
3353  */
3354 static void e1000_dump(struct e1000_adapter *adapter)
3355 {
3356         /* this code doesn't handle multiple rings */
3357         struct e1000_tx_ring *tx_ring = adapter->tx_ring;
3358         struct e1000_rx_ring *rx_ring = adapter->rx_ring;
3359         int i;
3360
3361         if (!netif_msg_hw(adapter))
3362                 return;
3363
3364         /* Print Registers */
3365         e1000_regdump(adapter);
3366
3367         /* transmit dump */
3368         pr_info("TX Desc ring0 dump\n");
3369
3370         /* Transmit Descriptor Formats - DEXT[29] is 0 (Legacy) or 1 (Extended)
3371          *
3372          * Legacy Transmit Descriptor
3373          *   +--------------------------------------------------------------+
3374          * 0 |         Buffer Address [63:0] (Reserved on Write Back)       |
3375          *   +--------------------------------------------------------------+
3376          * 8 | Special  |    CSS     | Status |  CMD    |  CSO   |  Length  |
3377          *   +--------------------------------------------------------------+
3378          *   63       48 47        36 35    32 31     24 23    16 15        0
3379          *
3380          * Extended Context Descriptor (DTYP=0x0) for TSO or checksum offload
3381          *   63      48 47    40 39       32 31             16 15    8 7      0
3382          *   +----------------------------------------------------------------+
3383          * 0 |  TUCSE  | TUCS0  |   TUCSS   |     IPCSE       | IPCS0 | IPCSS |
3384          *   +----------------------------------------------------------------+
3385          * 8 |   MSS   | HDRLEN | RSV | STA | TUCMD | DTYP |      PAYLEN      |
3386          *   +----------------------------------------------------------------+
3387          *   63      48 47    40 39 36 35 32 31   24 23  20 19                0
3388          *
3389          * Extended Data Descriptor (DTYP=0x1)
3390          *   +----------------------------------------------------------------+
3391          * 0 |                     Buffer Address [63:0]                      |
3392          *   +----------------------------------------------------------------+
3393          * 8 | VLAN tag |  POPTS  | Rsvd | Status | Command | DTYP |  DTALEN  |
3394          *   +----------------------------------------------------------------+
3395          *   63       48 47     40 39  36 35    32 31     24 23  20 19        0
3396          */
3397         pr_info("Tc[desc]     [Ce CoCsIpceCoS] [MssHlRSCm0Plen] [bi->dma       ] leng  ntw timestmp         bi->skb\n");
3398         pr_info("Td[desc]     [address 63:0  ] [VlaPoRSCm1Dlen] [bi->dma       ] leng  ntw timestmp         bi->skb\n");
3399
3400         if (!netif_msg_tx_done(adapter))
3401                 goto rx_ring_summary;
3402
3403         for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) {
3404                 struct e1000_tx_desc *tx_desc = E1000_TX_DESC(*tx_ring, i);
3405                 struct e1000_tx_buffer *buffer_info = &tx_ring->buffer_info[i];
3406                 struct my_u { __le64 a; __le64 b; };
3407                 struct my_u *u = (struct my_u *)tx_desc;
3408                 const char *type;
3409
3410                 if (i == tx_ring->next_to_use && i == tx_ring->next_to_clean)
3411                         type = "NTC/U";
3412                 else if (i == tx_ring->next_to_use)
3413                         type = "NTU";
3414                 else if (i == tx_ring->next_to_clean)
3415                         type = "NTC";
3416                 else
3417                         type = "";
3418
3419                 pr_info("T%c[0x%03X]    %016llX %016llX %016llX %04X  %3X %016llX %p %s\n",
3420                         ((le64_to_cpu(u->b) & (1<<20)) ? 'd' : 'c'), i,
3421                         le64_to_cpu(u->a), le64_to_cpu(u->b),
3422                         (u64)buffer_info->dma, buffer_info->length,
3423                         buffer_info->next_to_watch,
3424                         (u64)buffer_info->time_stamp, buffer_info->skb, type);
3425         }
3426
3427 rx_ring_summary:
3428         /* receive dump */
3429         pr_info("\nRX Desc ring dump\n");
3430
3431         /* Legacy Receive Descriptor Format
3432          *
3433          * +-----------------------------------------------------+
3434          * |                Buffer Address [63:0]                |
3435          * +-----------------------------------------------------+
3436          * | VLAN Tag | Errors | Status 0 | Packet csum | Length |
3437          * +-----------------------------------------------------+
3438          * 63       48 47    40 39      32 31         16 15      0
3439          */
3440         pr_info("R[desc]      [address 63:0  ] [vl er S cks ln] [bi->dma       ] [bi->skb]\n");
3441
3442         if (!netif_msg_rx_status(adapter))
3443                 goto exit;
3444
3445         for (i = 0; rx_ring->desc && (i < rx_ring->count); i++) {
3446                 struct e1000_rx_desc *rx_desc = E1000_RX_DESC(*rx_ring, i);
3447                 struct e1000_rx_buffer *buffer_info = &rx_ring->buffer_info[i];
3448                 struct my_u { __le64 a; __le64 b; };
3449                 struct my_u *u = (struct my_u *)rx_desc;
3450                 const char *type;
3451
3452                 if (i == rx_ring->next_to_use)
3453                         type = "NTU";
3454                 else if (i == rx_ring->next_to_clean)
3455                         type = "NTC";
3456                 else
3457                         type = "";
3458
3459                 pr_info("R[0x%03X]     %016llX %016llX %016llX %p %s\n",
3460                         i, le64_to_cpu(u->a), le64_to_cpu(u->b),
3461                         (u64)buffer_info->dma, buffer_info->rxbuf.data, type);
3462         } /* for */
3463
3464         /* dump the descriptor caches */
3465         /* rx */
3466         pr_info("Rx descriptor cache in 64bit format\n");
3467         for (i = 0x6000; i <= 0x63FF ; i += 0x10) {
3468                 pr_info("R%04X: %08X|%08X %08X|%08X\n",
3469                         i,
3470                         readl(adapter->hw.hw_addr + i+4),
3471                         readl(adapter->hw.hw_addr + i),
3472                         readl(adapter->hw.hw_addr + i+12),
3473                         readl(adapter->hw.hw_addr + i+8));
3474         }
3475         /* tx */
3476         pr_info("Tx descriptor cache in 64bit format\n");
3477         for (i = 0x7000; i <= 0x73FF ; i += 0x10) {
3478                 pr_info("T%04X: %08X|%08X %08X|%08X\n",
3479                         i,
3480                         readl(adapter->hw.hw_addr + i+4),
3481                         readl(adapter->hw.hw_addr + i),
3482                         readl(adapter->hw.hw_addr + i+12),
3483                         readl(adapter->hw.hw_addr + i+8));
3484         }
3485 exit:
3486         return;
3487 }
3488
3489 /**
3490  * e1000_tx_timeout - Respond to a Tx Hang
3491  * @netdev: network interface device structure
3492  **/
3493 static void e1000_tx_timeout(struct net_device *netdev)
3494 {
3495         struct e1000_adapter *adapter = netdev_priv(netdev);
3496
3497         /* Do the reset outside of interrupt context */
3498         adapter->tx_timeout_count++;
3499         schedule_work(&adapter->reset_task);
3500 }
3501
3502 static void e1000_reset_task(struct work_struct *work)
3503 {
3504         struct e1000_adapter *adapter =
3505                 container_of(work, struct e1000_adapter, reset_task);
3506
3507         e_err(drv, "Reset adapter\n");
3508         e1000_reinit_locked(adapter);
3509 }
3510
3511 /**
3512  * e1000_get_stats - Get System Network Statistics
3513  * @netdev: network interface device structure
3514  *
3515  * Returns the address of the device statistics structure.
3516  * The statistics are actually updated from the watchdog.
3517  **/
3518 static struct net_device_stats *e1000_get_stats(struct net_device *netdev)
3519 {
3520         /* only return the current stats */
3521         return &netdev->stats;
3522 }
3523
3524 /**
3525  * e1000_change_mtu - Change the Maximum Transfer Unit
3526  * @netdev: network interface device structure
3527  * @new_mtu: new value for maximum frame size
3528  *
3529  * Returns 0 on success, negative on failure
3530  **/
3531 static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
3532 {
3533         struct e1000_adapter *adapter = netdev_priv(netdev);
3534         struct e1000_hw *hw = &adapter->hw;
3535         int max_frame = new_mtu + ENET_HEADER_SIZE + ETHERNET_FCS_SIZE;
3536
3537         if ((max_frame < MINIMUM_ETHERNET_FRAME_SIZE) ||
3538             (max_frame > MAX_JUMBO_FRAME_SIZE)) {
3539                 e_err(probe, "Invalid MTU setting\n");
3540                 return -EINVAL;
3541         }
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(napi);
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)) i = 0;
3878                 }
3879
3880                 eop = tx_ring->buffer_info[i].next_to_watch;
3881                 eop_desc = E1000_TX_DESC(*tx_ring, eop);
3882         }
3883
3884         tx_ring->next_to_clean = i;
3885
3886         netdev_completed_queue(netdev, pkts_compl, bytes_compl);
3887
3888 #define TX_WAKE_THRESHOLD 32
3889         if (unlikely(count && netif_carrier_ok(netdev) &&
3890                      E1000_DESC_UNUSED(tx_ring) >= TX_WAKE_THRESHOLD)) {
3891                 /* Make sure that anybody stopping the queue after this
3892                  * sees the new next_to_clean.
3893                  */
3894                 smp_mb();
3895
3896                 if (netif_queue_stopped(netdev) &&
3897                     !(test_bit(__E1000_DOWN, &adapter->flags))) {
3898                         netif_wake_queue(netdev);
3899                         ++adapter->restart_queue;
3900                 }
3901         }
3902
3903         if (adapter->detect_tx_hung) {
3904                 /* Detect a transmit hang in hardware, this serializes the
3905                  * check with the clearing of time_stamp and movement of i
3906                  */
3907                 adapter->detect_tx_hung = false;
3908                 if (tx_ring->buffer_info[eop].time_stamp &&
3909                     time_after(jiffies, tx_ring->buffer_info[eop].time_stamp +
3910                                (adapter->tx_timeout_factor * HZ)) &&
3911                     !(er32(STATUS) & E1000_STATUS_TXOFF)) {
3912
3913                         /* detected Tx unit hang */
3914                         e_err(drv, "Detected Tx Unit Hang\n"
3915                               "  Tx Queue             <%lu>\n"
3916                               "  TDH                  <%x>\n"
3917                               "  TDT                  <%x>\n"
3918                               "  next_to_use          <%x>\n"
3919                               "  next_to_clean        <%x>\n"
3920                               "buffer_info[next_to_clean]\n"
3921                               "  time_stamp           <%lx>\n"
3922                               "  next_to_watch        <%x>\n"
3923                               "  jiffies              <%lx>\n"
3924                               "  next_to_watch.status <%x>\n",
3925                                 (unsigned long)(tx_ring - adapter->tx_ring),
3926                                 readl(hw->hw_addr + tx_ring->tdh),
3927                                 readl(hw->hw_addr + tx_ring->tdt),
3928                                 tx_ring->next_to_use,
3929                                 tx_ring->next_to_clean,
3930                                 tx_ring->buffer_info[eop].time_stamp,
3931                                 eop,
3932                                 jiffies,
3933                                 eop_desc->upper.fields.status);
3934                         e1000_dump(adapter);
3935                         netif_stop_queue(netdev);
3936                 }
3937         }
3938         adapter->total_tx_bytes += total_tx_bytes;
3939         adapter->total_tx_packets += total_tx_packets;
3940         netdev->stats.tx_bytes += total_tx_bytes;
3941         netdev->stats.tx_packets += total_tx_packets;
3942         return count < tx_ring->count;
3943 }
3944
3945 /**
3946  * e1000_rx_checksum - Receive Checksum Offload for 82543
3947  * @adapter:     board private structure
3948  * @status_err:  receive descriptor status and error fields
3949  * @csum:        receive descriptor csum field
3950  * @sk_buff:     socket buffer with received data
3951  **/
3952 static void e1000_rx_checksum(struct e1000_adapter *adapter, u32 status_err,
3953                               u32 csum, struct sk_buff *skb)
3954 {
3955         struct e1000_hw *hw = &adapter->hw;
3956         u16 status = (u16)status_err;
3957         u8 errors = (u8)(status_err >> 24);
3958
3959         skb_checksum_none_assert(skb);
3960
3961         /* 82543 or newer only */
3962         if (unlikely(hw->mac_type < e1000_82543)) return;
3963         /* Ignore Checksum bit is set */
3964         if (unlikely(status & E1000_RXD_STAT_IXSM)) return;
3965         /* TCP/UDP checksum error bit is set */
3966         if (unlikely(errors & E1000_RXD_ERR_TCPE)) {
3967                 /* let the stack verify checksum errors */
3968                 adapter->hw_csum_err++;
3969                 return;
3970         }
3971         /* TCP/UDP Checksum has not been calculated */
3972         if (!(status & E1000_RXD_STAT_TCPCS))
3973                 return;
3974
3975         /* It must be a TCP or UDP packet with a valid checksum */
3976         if (likely(status & E1000_RXD_STAT_TCPCS)) {
3977                 /* TCP checksum is good */
3978                 skb->ip_summed = CHECKSUM_UNNECESSARY;
3979         }
3980         adapter->hw_csum_good++;
3981 }
3982
3983 /**
3984  * e1000_consume_page - helper function for jumbo Rx path
3985  **/
3986 static void e1000_consume_page(struct e1000_rx_buffer *bi, struct sk_buff *skb,
3987                                u16 length)
3988 {
3989         bi->rxbuf.page = NULL;
3990         skb->len += length;
3991         skb->data_len += length;
3992         skb->truesize += PAGE_SIZE;
3993 }
3994
3995 /**
3996  * e1000_receive_skb - helper function to handle rx indications
3997  * @adapter: board private structure
3998  * @status: descriptor status field as written by hardware
3999  * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
4000  * @skb: pointer to sk_buff to be indicated to stack
4001  */
4002 static void e1000_receive_skb(struct e1000_adapter *adapter, u8 status,
4003                               __le16 vlan, struct sk_buff *skb)
4004 {
4005         skb->protocol = eth_type_trans(skb, adapter->netdev);
4006
4007         if (status & E1000_RXD_STAT_VP) {
4008                 u16 vid = le16_to_cpu(vlan) & E1000_RXD_SPC_VLAN_MASK;
4009
4010                 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vid);
4011         }
4012         napi_gro_receive(&adapter->napi, skb);
4013 }
4014
4015 /**
4016  * e1000_tbi_adjust_stats
4017  * @hw: Struct containing variables accessed by shared code
4018  * @frame_len: The length of the frame in question
4019  * @mac_addr: The Ethernet destination address of the frame in question
4020  *
4021  * Adjusts the statistic counters when a frame is accepted by TBI_ACCEPT
4022  */
4023 static void e1000_tbi_adjust_stats(struct e1000_hw *hw,
4024                                    struct e1000_hw_stats *stats,
4025                                    u32 frame_len, const u8 *mac_addr)
4026 {
4027         u64 carry_bit;
4028
4029         /* First adjust the frame length. */
4030         frame_len--;
4031         /* We need to adjust the statistics counters, since the hardware
4032          * counters overcount this packet as a CRC error and undercount
4033          * the packet as a good packet
4034          */
4035         /* This packet should not be counted as a CRC error. */
4036         stats->crcerrs--;
4037         /* This packet does count as a Good Packet Received. */
4038         stats->gprc++;
4039
4040         /* Adjust the Good Octets received counters */
4041         carry_bit = 0x80000000 & stats->gorcl;
4042         stats->gorcl += frame_len;
4043         /* If the high bit of Gorcl (the low 32 bits of the Good Octets
4044          * Received Count) was one before the addition,
4045          * AND it is zero after, then we lost the carry out,
4046          * need to add one to Gorch (Good Octets Received Count High).
4047          * This could be simplified if all environments supported
4048          * 64-bit integers.
4049          */
4050         if (carry_bit && ((stats->gorcl & 0x80000000) == 0))
4051                 stats->gorch++;
4052         /* Is this a broadcast or multicast?  Check broadcast first,
4053          * since the test for a multicast frame will test positive on
4054          * a broadcast frame.
4055          */
4056         if (is_broadcast_ether_addr(mac_addr))
4057                 stats->bprc++;
4058         else if (is_multicast_ether_addr(mac_addr))
4059                 stats->mprc++;
4060
4061         if (frame_len == hw->max_frame_size) {
4062                 /* In this case, the hardware has overcounted the number of
4063                  * oversize frames.
4064                  */
4065                 if (stats->roc > 0)
4066                         stats->roc--;
4067         }
4068
4069         /* Adjust the bin counters when the extra byte put the frame in the
4070          * wrong bin. Remember that the frame_len was adjusted above.
4071          */
4072         if (frame_len == 64) {
4073                 stats->prc64++;
4074                 stats->prc127--;
4075         } else if (frame_len == 127) {
4076                 stats->prc127++;
4077                 stats->prc255--;
4078         } else if (frame_len == 255) {
4079                 stats->prc255++;
4080                 stats->prc511--;
4081         } else if (frame_len == 511) {
4082                 stats->prc511++;
4083                 stats->prc1023--;
4084         } else if (frame_len == 1023) {
4085                 stats->prc1023++;
4086                 stats->prc1522--;
4087         } else if (frame_len == 1522) {
4088                 stats->prc1522++;
4089         }
4090 }
4091
4092 static bool e1000_tbi_should_accept(struct e1000_adapter *adapter,
4093                                     u8 status, u8 errors,
4094                                     u32 length, const u8 *data)
4095 {
4096         struct e1000_hw *hw = &adapter->hw;
4097         u8 last_byte = *(data + length - 1);
4098
4099         if (TBI_ACCEPT(hw, status, errors, length, last_byte)) {
4100                 unsigned long irq_flags;
4101
4102                 spin_lock_irqsave(&adapter->stats_lock, irq_flags);
4103                 e1000_tbi_adjust_stats(hw, &adapter->stats, length, data);
4104                 spin_unlock_irqrestore(&adapter->stats_lock, irq_flags);
4105
4106                 return true;
4107         }
4108
4109         return false;
4110 }
4111
4112 static struct sk_buff *e1000_alloc_rx_skb(struct e1000_adapter *adapter,
4113                                           unsigned int bufsz)
4114 {
4115         struct sk_buff *skb = napi_alloc_skb(&adapter->napi, bufsz);
4116
4117         if (unlikely(!skb))
4118                 adapter->alloc_rx_buff_failed++;
4119         return skb;
4120 }
4121
4122 /**
4123  * e1000_clean_jumbo_rx_irq - Send received data up the network stack; legacy
4124  * @adapter: board private structure
4125  * @rx_ring: ring to clean
4126  * @work_done: amount of napi work completed this call
4127  * @work_to_do: max amount of work allowed for this call to do
4128  *
4129  * the return value indicates whether actual cleaning was done, there
4130  * is no guarantee that everything was cleaned
4131  */
4132 static bool e1000_clean_jumbo_rx_irq(struct e1000_adapter *adapter,
4133                                      struct e1000_rx_ring *rx_ring,
4134                                      int *work_done, int work_to_do)
4135 {
4136         struct net_device *netdev = adapter->netdev;
4137         struct pci_dev *pdev = adapter->pdev;
4138         struct e1000_rx_desc *rx_desc, *next_rxd;
4139         struct e1000_rx_buffer *buffer_info, *next_buffer;
4140         u32 length;
4141         unsigned int i;
4142         int cleaned_count = 0;
4143         bool cleaned = false;
4144         unsigned int total_rx_bytes=0, total_rx_packets=0;
4145
4146         i = rx_ring->next_to_clean;
4147         rx_desc = E1000_RX_DESC(*rx_ring, i);
4148         buffer_info = &rx_ring->buffer_info[i];
4149
4150         while (rx_desc->status & E1000_RXD_STAT_DD) {
4151                 struct sk_buff *skb;
4152                 u8 status;
4153
4154                 if (*work_done >= work_to_do)
4155                         break;
4156                 (*work_done)++;
4157                 dma_rmb(); /* read descriptor and rx_buffer_info after status DD */
4158
4159                 status = rx_desc->status;
4160
4161                 if (++i == rx_ring->count) i = 0;
4162                 next_rxd = E1000_RX_DESC(*rx_ring, i);
4163                 prefetch(next_rxd);
4164
4165                 next_buffer = &rx_ring->buffer_info[i];
4166
4167                 cleaned = true;
4168                 cleaned_count++;
4169                 dma_unmap_page(&pdev->dev, buffer_info->dma,
4170                                adapter->rx_buffer_len, DMA_FROM_DEVICE);
4171                 buffer_info->dma = 0;
4172
4173                 length = le16_to_cpu(rx_desc->length);
4174
4175                 /* errors is only valid for DD + EOP descriptors */
4176                 if (unlikely((status & E1000_RXD_STAT_EOP) &&
4177                     (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK))) {
4178                         u8 *mapped = page_address(buffer_info->rxbuf.page);
4179
4180                         if (e1000_tbi_should_accept(adapter, status,
4181                                                     rx_desc->errors,
4182                                                     length, mapped)) {
4183                                 length--;
4184                         } else if (netdev->features & NETIF_F_RXALL) {
4185                                 goto process_skb;
4186                         } else {
4187                                 /* an error means any chain goes out the window
4188                                  * too
4189                                  */
4190                                 if (rx_ring->rx_skb_top)
4191                                         dev_kfree_skb(rx_ring->rx_skb_top);
4192                                 rx_ring->rx_skb_top = NULL;
4193                                 goto next_desc;
4194                         }
4195                 }
4196
4197 #define rxtop rx_ring->rx_skb_top
4198 process_skb:
4199                 if (!(status & E1000_RXD_STAT_EOP)) {
4200                         /* this descriptor is only the beginning (or middle) */
4201                         if (!rxtop) {
4202                                 /* this is the beginning of a chain */
4203                                 rxtop = napi_get_frags(&adapter->napi);
4204                                 if (!rxtop)
4205                                         break;
4206
4207                                 skb_fill_page_desc(rxtop, 0,
4208                                                    buffer_info->rxbuf.page,
4209                                                    0, length);
4210                         } else {
4211                                 /* this is the middle of a chain */
4212                                 skb_fill_page_desc(rxtop,
4213                                     skb_shinfo(rxtop)->nr_frags,
4214                                     buffer_info->rxbuf.page, 0, length);
4215                         }
4216                         e1000_consume_page(buffer_info, rxtop, length);
4217                         goto next_desc;
4218                 } else {
4219                         if (rxtop) {
4220                                 /* end of the chain */
4221                                 skb_fill_page_desc(rxtop,
4222                                     skb_shinfo(rxtop)->nr_frags,
4223                                     buffer_info->rxbuf.page, 0, length);
4224                                 skb = rxtop;
4225                                 rxtop = NULL;
4226                                 e1000_consume_page(buffer_info, skb, length);
4227                         } else {
4228                                 struct page *p;
4229                                 /* no chain, got EOP, this buf is the packet
4230                                  * copybreak to save the put_page/alloc_page
4231                                  */
4232                                 p = buffer_info->rxbuf.page;
4233                                 if (length <= copybreak) {
4234                                         u8 *vaddr;
4235
4236                                         if (likely(!(netdev->features & NETIF_F_RXFCS)))
4237                                                 length -= 4;
4238                                         skb = e1000_alloc_rx_skb(adapter,
4239                                                                  length);
4240                                         if (!skb)
4241                                                 break;
4242
4243                                         vaddr = kmap_atomic(p);
4244                                         memcpy(skb_tail_pointer(skb), vaddr,
4245                                                length);
4246                                         kunmap_atomic(vaddr);
4247                                         /* re-use the page, so don't erase
4248                                          * buffer_info->rxbuf.page
4249                                          */
4250                                         skb_put(skb, length);
4251                                         e1000_rx_checksum(adapter,
4252                                                           status | rx_desc->errors << 24,
4253                                                           le16_to_cpu(rx_desc->csum), skb);
4254
4255                                         total_rx_bytes += skb->len;
4256                                         total_rx_packets++;
4257
4258                                         e1000_receive_skb(adapter, status,
4259                                                           rx_desc->special, skb);
4260                                         goto next_desc;
4261                                 } else {
4262                                         skb = napi_get_frags(&adapter->napi);
4263                                         if (!skb) {
4264                                                 adapter->alloc_rx_buff_failed++;
4265                                                 break;
4266                                         }
4267                                         skb_fill_page_desc(skb, 0, p, 0,
4268                                                            length);
4269                                         e1000_consume_page(buffer_info, skb,
4270                                                            length);
4271                                 }
4272                         }
4273                 }
4274
4275                 /* Receive Checksum Offload XXX recompute due to CRC strip? */
4276                 e1000_rx_checksum(adapter,
4277                                   (u32)(status) |
4278                                   ((u32)(rx_desc->errors) << 24),
4279                                   le16_to_cpu(rx_desc->csum), skb);
4280
4281                 total_rx_bytes += (skb->len - 4); /* don't count FCS */
4282                 if (likely(!(netdev->features & NETIF_F_RXFCS)))
4283                         pskb_trim(skb, skb->len - 4);
4284                 total_rx_packets++;
4285
4286                 if (status & E1000_RXD_STAT_VP) {
4287                         __le16 vlan = rx_desc->special;
4288                         u16 vid = le16_to_cpu(vlan) & E1000_RXD_SPC_VLAN_MASK;
4289
4290                         __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vid);
4291                 }
4292
4293                 napi_gro_frags(&adapter->napi);
4294
4295 next_desc:
4296                 rx_desc->status = 0;
4297
4298                 /* return some buffers to hardware, one at a time is too slow */
4299                 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
4300                         adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
4301                         cleaned_count = 0;
4302                 }
4303
4304                 /* use prefetched values */
4305                 rx_desc = next_rxd;
4306                 buffer_info = next_buffer;
4307         }
4308         rx_ring->next_to_clean = i;
4309
4310         cleaned_count = E1000_DESC_UNUSED(rx_ring);
4311         if (cleaned_count)
4312                 adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
4313
4314         adapter->total_rx_packets += total_rx_packets;
4315         adapter->total_rx_bytes += total_rx_bytes;
4316         netdev->stats.rx_bytes += total_rx_bytes;
4317         netdev->stats.rx_packets += total_rx_packets;
4318         return cleaned;
4319 }
4320
4321 /* this should improve performance for small packets with large amounts
4322  * of reassembly being done in the stack
4323  */
4324 static struct sk_buff *e1000_copybreak(struct e1000_adapter *adapter,
4325                                        struct e1000_rx_buffer *buffer_info,
4326                                        u32 length, const void *data)
4327 {
4328         struct sk_buff *skb;
4329
4330         if (length > copybreak)
4331                 return NULL;
4332
4333         skb = e1000_alloc_rx_skb(adapter, length);
4334         if (!skb)
4335                 return NULL;
4336
4337         dma_sync_single_for_cpu(&adapter->pdev->dev, buffer_info->dma,
4338                                 length, DMA_FROM_DEVICE);
4339
4340         memcpy(skb_put(skb, length), data, length);
4341
4342         return skb;
4343 }
4344
4345 /**
4346  * e1000_clean_rx_irq - Send received data up the network stack; legacy
4347  * @adapter: board private structure
4348  * @rx_ring: ring to clean
4349  * @work_done: amount of napi work completed this call
4350  * @work_to_do: max amount of work allowed for this call to do
4351  */
4352 static bool e1000_clean_rx_irq(struct e1000_adapter *adapter,
4353                                struct e1000_rx_ring *rx_ring,
4354                                int *work_done, int work_to_do)
4355 {
4356         struct net_device *netdev = adapter->netdev;
4357         struct pci_dev *pdev = adapter->pdev;
4358         struct e1000_rx_desc *rx_desc, *next_rxd;
4359         struct e1000_rx_buffer *buffer_info, *next_buffer;
4360         u32 length;
4361         unsigned int i;
4362         int cleaned_count = 0;
4363         bool cleaned = false;
4364         unsigned int total_rx_bytes=0, total_rx_packets=0;
4365
4366         i = rx_ring->next_to_clean;
4367         rx_desc = E1000_RX_DESC(*rx_ring, i);
4368         buffer_info = &rx_ring->buffer_info[i];
4369
4370         while (rx_desc->status & E1000_RXD_STAT_DD) {
4371                 struct sk_buff *skb;
4372                 u8 *data;
4373                 u8 status;
4374
4375                 if (*work_done >= work_to_do)
4376                         break;
4377                 (*work_done)++;
4378                 dma_rmb(); /* read descriptor and rx_buffer_info after status DD */
4379
4380                 status = rx_desc->status;
4381                 length = le16_to_cpu(rx_desc->length);
4382
4383                 data = buffer_info->rxbuf.data;
4384                 prefetch(data);
4385                 skb = e1000_copybreak(adapter, buffer_info, length, data);
4386                 if (!skb) {
4387                         unsigned int frag_len = e1000_frag_len(adapter);
4388
4389                         skb = build_skb(data - E1000_HEADROOM, frag_len);
4390                         if (!skb) {
4391                                 adapter->alloc_rx_buff_failed++;
4392                                 break;
4393                         }
4394
4395                         skb_reserve(skb, E1000_HEADROOM);
4396                         dma_unmap_single(&pdev->dev, buffer_info->dma,
4397                                          adapter->rx_buffer_len,
4398                                          DMA_FROM_DEVICE);
4399                         buffer_info->dma = 0;
4400                         buffer_info->rxbuf.data = NULL;
4401                 }
4402
4403                 if (++i == rx_ring->count) i = 0;
4404                 next_rxd = E1000_RX_DESC(*rx_ring, i);
4405                 prefetch(next_rxd);
4406
4407                 next_buffer = &rx_ring->buffer_info[i];
4408
4409                 cleaned = true;
4410                 cleaned_count++;
4411
4412                 /* !EOP means multiple descriptors were used to store a single
4413                  * packet, if thats the case we need to toss it.  In fact, we
4414                  * to toss every packet with the EOP bit clear and the next
4415                  * frame that _does_ have the EOP bit set, as it is by
4416                  * definition only a frame fragment
4417                  */
4418                 if (unlikely(!(status & E1000_RXD_STAT_EOP)))
4419                         adapter->discarding = true;
4420
4421                 if (adapter->discarding) {
4422                         /* All receives must fit into a single buffer */
4423                         netdev_dbg(netdev, "Receive packet consumed multiple buffers\n");
4424                         dev_kfree_skb(skb);
4425                         if (status & E1000_RXD_STAT_EOP)
4426                                 adapter->discarding = false;
4427                         goto next_desc;
4428                 }
4429
4430                 if (unlikely(rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK)) {
4431                         if (e1000_tbi_should_accept(adapter, status,
4432                                                     rx_desc->errors,
4433                                                     length, data)) {
4434                                 length--;
4435                         } else if (netdev->features & NETIF_F_RXALL) {
4436                                 goto process_skb;
4437                         } else {
4438                                 dev_kfree_skb(skb);
4439                                 goto next_desc;
4440                         }
4441                 }
4442
4443 process_skb:
4444                 total_rx_bytes += (length - 4); /* don't count FCS */
4445                 total_rx_packets++;
4446
4447                 if (likely(!(netdev->features & NETIF_F_RXFCS)))
4448                         /* adjust length to remove Ethernet CRC, this must be
4449                          * done after the TBI_ACCEPT workaround above
4450                          */
4451                         length -= 4;
4452
4453                 if (buffer_info->rxbuf.data == NULL)
4454                         skb_put(skb, length);
4455                 else /* copybreak skb */
4456                         skb_trim(skb, length);
4457
4458                 /* Receive Checksum Offload */
4459                 e1000_rx_checksum(adapter,
4460                                   (u32)(status) |
4461                                   ((u32)(rx_desc->errors) << 24),
4462                                   le16_to_cpu(rx_desc->csum), skb);
4463
4464                 e1000_receive_skb(adapter, status, rx_desc->special, skb);
4465
4466 next_desc:
4467                 rx_desc->status = 0;
4468
4469                 /* return some buffers to hardware, one at a time is too slow */
4470                 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
4471                         adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
4472                         cleaned_count = 0;
4473                 }
4474
4475                 /* use prefetched values */
4476                 rx_desc = next_rxd;
4477                 buffer_info = next_buffer;
4478         }
4479         rx_ring->next_to_clean = i;
4480
4481         cleaned_count = E1000_DESC_UNUSED(rx_ring);
4482         if (cleaned_count)
4483                 adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
4484
4485         adapter->total_rx_packets += total_rx_packets;
4486         adapter->total_rx_bytes += total_rx_bytes;
4487         netdev->stats.rx_bytes += total_rx_bytes;
4488         netdev->stats.rx_packets += total_rx_packets;
4489         return cleaned;
4490 }
4491
4492 /**
4493  * e1000_alloc_jumbo_rx_buffers - Replace used jumbo receive buffers
4494  * @adapter: address of board private structure
4495  * @rx_ring: pointer to receive ring structure
4496  * @cleaned_count: number of buffers to allocate this pass
4497  **/
4498 static void
4499 e1000_alloc_jumbo_rx_buffers(struct e1000_adapter *adapter,
4500                              struct e1000_rx_ring *rx_ring, int cleaned_count)
4501 {
4502         struct pci_dev *pdev = adapter->pdev;
4503         struct e1000_rx_desc *rx_desc;
4504         struct e1000_rx_buffer *buffer_info;
4505         unsigned int i;
4506
4507         i = rx_ring->next_to_use;
4508         buffer_info = &rx_ring->buffer_info[i];
4509
4510         while (cleaned_count--) {
4511                 /* allocate a new page if necessary */
4512                 if (!buffer_info->rxbuf.page) {
4513                         buffer_info->rxbuf.page = alloc_page(GFP_ATOMIC);
4514                         if (unlikely(!buffer_info->rxbuf.page)) {
4515                                 adapter->alloc_rx_buff_failed++;
4516                                 break;
4517                         }
4518                 }
4519
4520                 if (!buffer_info->dma) {
4521                         buffer_info->dma = dma_map_page(&pdev->dev,
4522                                                         buffer_info->rxbuf.page, 0,
4523                                                         adapter->rx_buffer_len,
4524                                                         DMA_FROM_DEVICE);
4525                         if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
4526                                 put_page(buffer_info->rxbuf.page);
4527                                 buffer_info->rxbuf.page = NULL;
4528                                 buffer_info->dma = 0;
4529                                 adapter->alloc_rx_buff_failed++;
4530                                 break;
4531                         }
4532                 }
4533
4534                 rx_desc = E1000_RX_DESC(*rx_ring, i);
4535                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
4536
4537                 if (unlikely(++i == rx_ring->count))
4538                         i = 0;
4539                 buffer_info = &rx_ring->buffer_info[i];
4540         }
4541
4542         if (likely(rx_ring->next_to_use != i)) {
4543                 rx_ring->next_to_use = i;
4544                 if (unlikely(i-- == 0))
4545                         i = (rx_ring->count - 1);
4546
4547                 /* Force memory writes to complete before letting h/w
4548                  * know there are new descriptors to fetch.  (Only
4549                  * applicable for weak-ordered memory model archs,
4550                  * such as IA-64).
4551                  */
4552                 wmb();
4553                 writel(i, adapter->hw.hw_addr + rx_ring->rdt);
4554         }
4555 }
4556
4557 /**
4558  * e1000_alloc_rx_buffers - Replace used receive buffers; legacy & extended
4559  * @adapter: address of board private structure
4560  **/
4561 static void e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
4562                                    struct e1000_rx_ring *rx_ring,
4563                                    int cleaned_count)
4564 {
4565         struct e1000_hw *hw = &adapter->hw;
4566         struct pci_dev *pdev = adapter->pdev;
4567         struct e1000_rx_desc *rx_desc;
4568         struct e1000_rx_buffer *buffer_info;
4569         unsigned int i;
4570         unsigned int bufsz = adapter->rx_buffer_len;
4571
4572         i = rx_ring->next_to_use;
4573         buffer_info = &rx_ring->buffer_info[i];
4574
4575         while (cleaned_count--) {
4576                 void *data;
4577
4578                 if (buffer_info->rxbuf.data)
4579                         goto skip;
4580
4581                 data = e1000_alloc_frag(adapter);
4582                 if (!data) {
4583                         /* Better luck next round */
4584                         adapter->alloc_rx_buff_failed++;
4585                         break;
4586                 }
4587
4588                 /* Fix for errata 23, can't cross 64kB boundary */
4589                 if (!e1000_check_64k_bound(adapter, data, bufsz)) {
4590                         void *olddata = data;
4591                         e_err(rx_err, "skb align check failed: %u bytes at "
4592                               "%p\n", bufsz, data);
4593                         /* Try again, without freeing the previous */
4594                         data = e1000_alloc_frag(adapter);
4595                         /* Failed allocation, critical failure */
4596                         if (!data) {
4597                                 e1000_free_frag(olddata);
4598                                 adapter->alloc_rx_buff_failed++;
4599                                 break;
4600                         }
4601
4602                         if (!e1000_check_64k_bound(adapter, data, bufsz)) {
4603                                 /* give up */
4604                                 e1000_free_frag(data);
4605                                 e1000_free_frag(olddata);
4606                                 adapter->alloc_rx_buff_failed++;
4607                                 break;
4608                         }
4609
4610                         /* Use new allocation */
4611                         e1000_free_frag(olddata);
4612                 }
4613                 buffer_info->dma = dma_map_single(&pdev->dev,
4614                                                   data,
4615                                                   adapter->rx_buffer_len,
4616                                                   DMA_FROM_DEVICE);
4617                 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
4618                         e1000_free_frag(data);
4619                         buffer_info->dma = 0;
4620                         adapter->alloc_rx_buff_failed++;
4621                         break;
4622                 }
4623
4624                 /* XXX if it was allocated cleanly it will never map to a
4625                  * boundary crossing
4626                  */
4627
4628                 /* Fix for errata 23, can't cross 64kB boundary */
4629                 if (!e1000_check_64k_bound(adapter,
4630                                         (void *)(unsigned long)buffer_info->dma,
4631                                         adapter->rx_buffer_len)) {
4632                         e_err(rx_err, "dma align check failed: %u bytes at "
4633                               "%p\n", adapter->rx_buffer_len,
4634                               (void *)(unsigned long)buffer_info->dma);
4635
4636                         dma_unmap_single(&pdev->dev, buffer_info->dma,
4637                                          adapter->rx_buffer_len,
4638                                          DMA_FROM_DEVICE);
4639
4640                         e1000_free_frag(data);
4641                         buffer_info->rxbuf.data = NULL;
4642                         buffer_info->dma = 0;
4643
4644                         adapter->alloc_rx_buff_failed++;
4645                         break;
4646                 }
4647                 buffer_info->rxbuf.data = data;
4648  skip:
4649                 rx_desc = E1000_RX_DESC(*rx_ring, i);
4650                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
4651
4652                 if (unlikely(++i == rx_ring->count))
4653                         i = 0;
4654                 buffer_info = &rx_ring->buffer_info[i];
4655         }
4656
4657         if (likely(rx_ring->next_to_use != i)) {
4658                 rx_ring->next_to_use = i;
4659                 if (unlikely(i-- == 0))
4660                         i = (rx_ring->count - 1);
4661
4662                 /* Force memory writes to complete before letting h/w
4663                  * know there are new descriptors to fetch.  (Only
4664                  * applicable for weak-ordered memory model archs,
4665                  * such as IA-64).
4666                  */
4667                 wmb();
4668                 writel(i, hw->hw_addr + rx_ring->rdt);
4669         }
4670 }
4671
4672 /**
4673  * e1000_smartspeed - Workaround for SmartSpeed on 82541 and 82547 controllers.
4674  * @adapter:
4675  **/
4676 static void e1000_smartspeed(struct e1000_adapter *adapter)
4677 {
4678         struct e1000_hw *hw = &adapter->hw;
4679         u16 phy_status;
4680         u16 phy_ctrl;
4681
4682         if ((hw->phy_type != e1000_phy_igp) || !hw->autoneg ||
4683            !(hw->autoneg_advertised & ADVERTISE_1000_FULL))
4684                 return;
4685
4686         if (adapter->smartspeed == 0) {
4687                 /* If Master/Slave config fault is asserted twice,
4688                  * we assume back-to-back
4689                  */
4690                 e1000_read_phy_reg(hw, PHY_1000T_STATUS, &phy_status);
4691                 if (!(phy_status & SR_1000T_MS_CONFIG_FAULT)) return;
4692                 e1000_read_phy_reg(hw, PHY_1000T_STATUS, &phy_status);
4693                 if (!(phy_status & SR_1000T_MS_CONFIG_FAULT)) return;
4694                 e1000_read_phy_reg(hw, PHY_1000T_CTRL, &phy_ctrl);
4695                 if (phy_ctrl & CR_1000T_MS_ENABLE) {
4696                         phy_ctrl &= ~CR_1000T_MS_ENABLE;
4697                         e1000_write_phy_reg(hw, PHY_1000T_CTRL,
4698                                             phy_ctrl);
4699                         adapter->smartspeed++;
4700                         if (!e1000_phy_setup_autoneg(hw) &&
4701                            !e1000_read_phy_reg(hw, PHY_CTRL,
4702                                                &phy_ctrl)) {
4703                                 phy_ctrl |= (MII_CR_AUTO_NEG_EN |
4704                                              MII_CR_RESTART_AUTO_NEG);
4705                                 e1000_write_phy_reg(hw, PHY_CTRL,
4706                                                     phy_ctrl);
4707                         }
4708                 }
4709                 return;
4710         } else if (adapter->smartspeed == E1000_SMARTSPEED_DOWNSHIFT) {
4711                 /* If still no link, perhaps using 2/3 pair cable */
4712                 e1000_read_phy_reg(hw, PHY_1000T_CTRL, &phy_ctrl);
4713                 phy_ctrl |= CR_1000T_MS_ENABLE;
4714                 e1000_write_phy_reg(hw, PHY_1000T_CTRL, phy_ctrl);
4715                 if (!e1000_phy_setup_autoneg(hw) &&
4716                    !e1000_read_phy_reg(hw, PHY_CTRL, &phy_ctrl)) {
4717                         phy_ctrl |= (MII_CR_AUTO_NEG_EN |
4718                                      MII_CR_RESTART_AUTO_NEG);
4719                         e1000_write_phy_reg(hw, PHY_CTRL, phy_ctrl);
4720                 }
4721         }
4722         /* Restart process after E1000_SMARTSPEED_MAX iterations */
4723         if (adapter->smartspeed++ == E1000_SMARTSPEED_MAX)
4724                 adapter->smartspeed = 0;
4725 }
4726
4727 /**
4728  * e1000_ioctl -
4729  * @netdev:
4730  * @ifreq:
4731  * @cmd:
4732  **/
4733 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
4734 {
4735         switch (cmd) {
4736         case SIOCGMIIPHY:
4737         case SIOCGMIIREG:
4738         case SIOCSMIIREG:
4739                 return e1000_mii_ioctl(netdev, ifr, cmd);
4740         default:
4741                 return -EOPNOTSUPP;
4742         }
4743 }
4744
4745 /**
4746  * e1000_mii_ioctl -
4747  * @netdev:
4748  * @ifreq:
4749  * @cmd:
4750  **/
4751 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
4752                            int cmd)
4753 {
4754         struct e1000_adapter *adapter = netdev_priv(netdev);
4755         struct e1000_hw *hw = &adapter->hw;
4756         struct mii_ioctl_data *data = if_mii(ifr);
4757         int retval;
4758         u16 mii_reg;
4759         unsigned long flags;
4760
4761         if (hw->media_type != e1000_media_type_copper)
4762                 return -EOPNOTSUPP;
4763
4764         switch (cmd) {
4765         case SIOCGMIIPHY:
4766                 data->phy_id = hw->phy_addr;
4767                 break;
4768         case SIOCGMIIREG:
4769                 spin_lock_irqsave(&adapter->stats_lock, flags);
4770                 if (e1000_read_phy_reg(hw, data->reg_num & 0x1F,
4771                                    &data->val_out)) {
4772                         spin_unlock_irqrestore(&adapter->stats_lock, flags);
4773                         return -EIO;
4774                 }
4775                 spin_unlock_irqrestore(&adapter->stats_lock, flags);
4776                 break;
4777         case SIOCSMIIREG:
4778                 if (data->reg_num & ~(0x1F))
4779                         return -EFAULT;
4780                 mii_reg = data->val_in;
4781                 spin_lock_irqsave(&adapter->stats_lock, flags);
4782                 if (e1000_write_phy_reg(hw, data->reg_num,
4783                                         mii_reg)) {
4784                         spin_unlock_irqrestore(&adapter->stats_lock, flags);
4785                         return -EIO;
4786                 }
4787                 spin_unlock_irqrestore(&adapter->stats_lock, flags);
4788                 if (hw->media_type == e1000_media_type_copper) {
4789                         switch (data->reg_num) {
4790                         case PHY_CTRL:
4791                                 if (mii_reg & MII_CR_POWER_DOWN)
4792                                         break;
4793                                 if (mii_reg & MII_CR_AUTO_NEG_EN) {
4794                                         hw->autoneg = 1;
4795                                         hw->autoneg_advertised = 0x2F;
4796                                 } else {
4797                                         u32 speed;
4798                                         if (mii_reg & 0x40)
4799                                                 speed = SPEED_1000;
4800                                         else if (mii_reg & 0x2000)
4801                                                 speed = SPEED_100;
4802                                         else
4803                                                 speed = SPEED_10;
4804                                         retval = e1000_set_spd_dplx(
4805                                                 adapter, speed,
4806                                                 ((mii_reg & 0x100)
4807                                                  ? DUPLEX_FULL :
4808                                                  DUPLEX_HALF));
4809                                         if (retval)
4810                                                 return retval;
4811                                 }
4812                                 if (netif_running(adapter->netdev))
4813                                         e1000_reinit_locked(adapter);
4814                                 else
4815                                         e1000_reset(adapter);
4816                                 break;
4817                         case M88E1000_PHY_SPEC_CTRL:
4818                         case M88E1000_EXT_PHY_SPEC_CTRL:
4819                                 if (e1000_phy_reset(hw))
4820                                         return -EIO;
4821                                 break;
4822                         }
4823                 } else {
4824                         switch (data->reg_num) {
4825                         case PHY_CTRL:
4826                                 if (mii_reg & MII_CR_POWER_DOWN)
4827                                         break;
4828                                 if (netif_running(adapter->netdev))
4829                                         e1000_reinit_locked(adapter);
4830                                 else
4831                                         e1000_reset(adapter);
4832                                 break;
4833                         }
4834                 }
4835                 break;
4836         default:
4837                 return -EOPNOTSUPP;
4838         }
4839         return E1000_SUCCESS;
4840 }
4841
4842 void e1000_pci_set_mwi(struct e1000_hw *hw)
4843 {
4844         struct e1000_adapter *adapter = hw->back;
4845         int ret_val = pci_set_mwi(adapter->pdev);
4846
4847         if (ret_val)
4848                 e_err(probe, "Error in setting MWI\n");
4849 }
4850
4851 void e1000_pci_clear_mwi(struct e1000_hw *hw)
4852 {
4853         struct e1000_adapter *adapter = hw->back;
4854
4855         pci_clear_mwi(adapter->pdev);
4856 }
4857
4858 int e1000_pcix_get_mmrbc(struct e1000_hw *hw)
4859 {
4860         struct e1000_adapter *adapter = hw->back;
4861         return pcix_get_mmrbc(adapter->pdev);
4862 }
4863
4864 void e1000_pcix_set_mmrbc(struct e1000_hw *hw, int mmrbc)
4865 {
4866         struct e1000_adapter *adapter = hw->back;
4867         pcix_set_mmrbc(adapter->pdev, mmrbc);
4868 }
4869
4870 void e1000_io_write(struct e1000_hw *hw, unsigned long port, u32 value)
4871 {
4872         outl(value, port);
4873 }
4874
4875 static bool e1000_vlan_used(struct e1000_adapter *adapter)
4876 {
4877         u16 vid;
4878
4879         for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
4880                 return true;
4881         return false;
4882 }
4883
4884 static void __e1000_vlan_mode(struct e1000_adapter *adapter,
4885                               netdev_features_t features)
4886 {
4887         struct e1000_hw *hw = &adapter->hw;
4888         u32 ctrl;
4889
4890         ctrl = er32(CTRL);
4891         if (features & NETIF_F_HW_VLAN_CTAG_RX) {
4892                 /* enable VLAN tag insert/strip */
4893                 ctrl |= E1000_CTRL_VME;
4894         } else {
4895                 /* disable VLAN tag insert/strip */
4896                 ctrl &= ~E1000_CTRL_VME;
4897         }
4898         ew32(CTRL, ctrl);
4899 }
4900 static void e1000_vlan_filter_on_off(struct e1000_adapter *adapter,
4901                                      bool filter_on)
4902 {
4903         struct e1000_hw *hw = &adapter->hw;
4904         u32 rctl;
4905
4906         if (!test_bit(__E1000_DOWN, &adapter->flags))
4907                 e1000_irq_disable(adapter);
4908
4909         __e1000_vlan_mode(adapter, adapter->netdev->features);
4910         if (filter_on) {
4911                 /* enable VLAN receive filtering */
4912                 rctl = er32(RCTL);
4913                 rctl &= ~E1000_RCTL_CFIEN;
4914                 if (!(adapter->netdev->flags & IFF_PROMISC))
4915                         rctl |= E1000_RCTL_VFE;
4916                 ew32(RCTL, rctl);
4917                 e1000_update_mng_vlan(adapter);
4918         } else {
4919                 /* disable VLAN receive filtering */
4920                 rctl = er32(RCTL);
4921                 rctl &= ~E1000_RCTL_VFE;
4922                 ew32(RCTL, rctl);
4923         }
4924
4925         if (!test_bit(__E1000_DOWN, &adapter->flags))
4926                 e1000_irq_enable(adapter);
4927 }
4928
4929 static void e1000_vlan_mode(struct net_device *netdev,
4930                             netdev_features_t features)
4931 {
4932         struct e1000_adapter *adapter = netdev_priv(netdev);
4933
4934         if (!test_bit(__E1000_DOWN, &adapter->flags))
4935                 e1000_irq_disable(adapter);
4936
4937         __e1000_vlan_mode(adapter, features);
4938
4939         if (!test_bit(__E1000_DOWN, &adapter->flags))
4940                 e1000_irq_enable(adapter);
4941 }
4942
4943 static int e1000_vlan_rx_add_vid(struct net_device *netdev,
4944                                  __be16 proto, u16 vid)
4945 {
4946         struct e1000_adapter *adapter = netdev_priv(netdev);
4947         struct e1000_hw *hw = &adapter->hw;
4948         u32 vfta, index;
4949
4950         if ((hw->mng_cookie.status &
4951              E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT) &&
4952             (vid == adapter->mng_vlan_id))
4953                 return 0;
4954
4955         if (!e1000_vlan_used(adapter))
4956                 e1000_vlan_filter_on_off(adapter, true);
4957
4958         /* add VID to filter table */
4959         index = (vid >> 5) & 0x7F;
4960         vfta = E1000_READ_REG_ARRAY(hw, VFTA, index);
4961         vfta |= (1 << (vid & 0x1F));
4962         e1000_write_vfta(hw, index, vfta);
4963
4964         set_bit(vid, adapter->active_vlans);
4965
4966         return 0;
4967 }
4968
4969 static int e1000_vlan_rx_kill_vid(struct net_device *netdev,
4970                                   __be16 proto, u16 vid)
4971 {
4972         struct e1000_adapter *adapter = netdev_priv(netdev);
4973         struct e1000_hw *hw = &adapter->hw;
4974         u32 vfta, index;
4975
4976         if (!test_bit(__E1000_DOWN, &adapter->flags))
4977                 e1000_irq_disable(adapter);
4978         if (!test_bit(__E1000_DOWN, &adapter->flags))
4979                 e1000_irq_enable(adapter);
4980
4981         /* remove VID from filter table */
4982         index = (vid >> 5) & 0x7F;
4983         vfta = E1000_READ_REG_ARRAY(hw, VFTA, index);
4984         vfta &= ~(1 << (vid & 0x1F));
4985         e1000_write_vfta(hw, index, vfta);
4986
4987         clear_bit(vid, adapter->active_vlans);
4988
4989         if (!e1000_vlan_used(adapter))
4990                 e1000_vlan_filter_on_off(adapter, false);
4991
4992         return 0;
4993 }
4994
4995 static void e1000_restore_vlan(struct e1000_adapter *adapter)
4996 {
4997         u16 vid;
4998
4999         if (!e1000_vlan_used(adapter))
5000                 return;
5001
5002         e1000_vlan_filter_on_off(adapter, true);
5003         for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
5004                 e1000_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), vid);
5005 }
5006
5007 int e1000_set_spd_dplx(struct e1000_adapter *adapter, u32 spd, u8 dplx)
5008 {
5009         struct e1000_hw *hw = &adapter->hw;
5010
5011         hw->autoneg = 0;
5012
5013         /* Make sure dplx is at most 1 bit and lsb of speed is not set
5014          * for the switch() below to work
5015          */
5016         if ((spd & 1) || (dplx & ~1))
5017                 goto err_inval;
5018
5019         /* Fiber NICs only allow 1000 gbps Full duplex */
5020         if ((hw->media_type == e1000_media_type_fiber) &&
5021             spd != SPEED_1000 &&
5022             dplx != DUPLEX_FULL)
5023                 goto err_inval;
5024
5025         switch (spd + dplx) {
5026         case SPEED_10 + DUPLEX_HALF:
5027                 hw->forced_speed_duplex = e1000_10_half;
5028                 break;
5029         case SPEED_10 + DUPLEX_FULL:
5030                 hw->forced_speed_duplex = e1000_10_full;
5031                 break;
5032         case SPEED_100 + DUPLEX_HALF:
5033                 hw->forced_speed_duplex = e1000_100_half;
5034                 break;
5035         case SPEED_100 + DUPLEX_FULL:
5036                 hw->forced_speed_duplex = e1000_100_full;
5037                 break;
5038         case SPEED_1000 + DUPLEX_FULL:
5039                 hw->autoneg = 1;
5040                 hw->autoneg_advertised = ADVERTISE_1000_FULL;
5041                 break;
5042         case SPEED_1000 + DUPLEX_HALF: /* not supported */
5043         default:
5044                 goto err_inval;
5045         }
5046
5047         /* clear MDI, MDI(-X) override is only allowed when autoneg enabled */
5048         hw->mdix = AUTO_ALL_MODES;
5049
5050         return 0;
5051
5052 err_inval:
5053         e_err(probe, "Unsupported Speed/Duplex configuration\n");
5054         return -EINVAL;
5055 }
5056
5057 static int __e1000_shutdown(struct pci_dev *pdev, bool *enable_wake)
5058 {
5059         struct net_device *netdev = pci_get_drvdata(pdev);
5060         struct e1000_adapter *adapter = netdev_priv(netdev);
5061         struct e1000_hw *hw = &adapter->hw;
5062         u32 ctrl, ctrl_ext, rctl, status;
5063         u32 wufc = adapter->wol;
5064 #ifdef CONFIG_PM
5065         int retval = 0;
5066 #endif
5067
5068         netif_device_detach(netdev);
5069
5070         if (netif_running(netdev)) {
5071                 int count = E1000_CHECK_RESET_COUNT;
5072
5073                 while (test_bit(__E1000_RESETTING, &adapter->flags) && count--)
5074                         usleep_range(10000, 20000);
5075
5076                 WARN_ON(test_bit(__E1000_RESETTING, &adapter->flags));
5077                 e1000_down(adapter);
5078         }
5079
5080 #ifdef CONFIG_PM
5081         retval = pci_save_state(pdev);
5082         if (retval)
5083                 return retval;
5084 #endif
5085
5086         status = er32(STATUS);
5087         if (status & E1000_STATUS_LU)
5088                 wufc &= ~E1000_WUFC_LNKC;
5089
5090         if (wufc) {
5091                 e1000_setup_rctl(adapter);
5092                 e1000_set_rx_mode(netdev);
5093
5094                 rctl = er32(RCTL);
5095
5096                 /* turn on all-multi mode if wake on multicast is enabled */
5097                 if (wufc & E1000_WUFC_MC)
5098                         rctl |= E1000_RCTL_MPE;
5099
5100                 /* enable receives in the hardware */
5101                 ew32(RCTL, rctl | E1000_RCTL_EN);
5102
5103                 if (hw->mac_type >= e1000_82540) {
5104                         ctrl = er32(CTRL);
5105                         /* advertise wake from D3Cold */
5106                         #define E1000_CTRL_ADVD3WUC 0x00100000
5107                         /* phy power management enable */
5108                         #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
5109                         ctrl |= E1000_CTRL_ADVD3WUC |
5110                                 E1000_CTRL_EN_PHY_PWR_MGMT;
5111                         ew32(CTRL, ctrl);
5112                 }
5113
5114                 if (hw->media_type == e1000_media_type_fiber ||
5115                     hw->media_type == e1000_media_type_internal_serdes) {
5116                         /* keep the laser running in D3 */
5117                         ctrl_ext = er32(CTRL_EXT);
5118                         ctrl_ext |= E1000_CTRL_EXT_SDP7_DATA;
5119                         ew32(CTRL_EXT, ctrl_ext);
5120                 }
5121
5122                 ew32(WUC, E1000_WUC_PME_EN);
5123                 ew32(WUFC, wufc);
5124         } else {
5125                 ew32(WUC, 0);
5126                 ew32(WUFC, 0);
5127         }
5128
5129         e1000_release_manageability(adapter);
5130
5131         *enable_wake = !!wufc;
5132
5133         /* make sure adapter isn't asleep if manageability is enabled */
5134         if (adapter->en_mng_pt)
5135                 *enable_wake = true;
5136
5137         if (netif_running(netdev))
5138                 e1000_free_irq(adapter);
5139
5140         pci_disable_device(pdev);
5141
5142         return 0;
5143 }
5144
5145 #ifdef CONFIG_PM
5146 static int e1000_suspend(struct pci_dev *pdev, pm_message_t state)
5147 {
5148         int retval;
5149         bool wake;
5150
5151         retval = __e1000_shutdown(pdev, &wake);
5152         if (retval)
5153                 return retval;
5154
5155         if (wake) {
5156                 pci_prepare_to_sleep(pdev);
5157         } else {
5158                 pci_wake_from_d3(pdev, false);
5159                 pci_set_power_state(pdev, PCI_D3hot);
5160         }
5161
5162         return 0;
5163 }
5164
5165 static int e1000_resume(struct pci_dev *pdev)
5166 {
5167         struct net_device *netdev = pci_get_drvdata(pdev);
5168         struct e1000_adapter *adapter = netdev_priv(netdev);
5169         struct e1000_hw *hw = &adapter->hw;
5170         u32 err;
5171
5172         pci_set_power_state(pdev, PCI_D0);
5173         pci_restore_state(pdev);
5174         pci_save_state(pdev);
5175
5176         if (adapter->need_ioport)
5177                 err = pci_enable_device(pdev);
5178         else
5179                 err = pci_enable_device_mem(pdev);
5180         if (err) {
5181                 pr_err("Cannot enable PCI device from suspend\n");
5182                 return err;
5183         }
5184         pci_set_master(pdev);
5185
5186         pci_enable_wake(pdev, PCI_D3hot, 0);
5187         pci_enable_wake(pdev, PCI_D3cold, 0);
5188
5189         if (netif_running(netdev)) {
5190                 err = e1000_request_irq(adapter);
5191                 if (err)
5192                         return err;
5193         }
5194
5195         e1000_power_up_phy(adapter);
5196         e1000_reset(adapter);
5197         ew32(WUS, ~0);
5198
5199         e1000_init_manageability(adapter);
5200
5201         if (netif_running(netdev))
5202                 e1000_up(adapter);
5203
5204         netif_device_attach(netdev);
5205
5206         return 0;
5207 }
5208 #endif
5209
5210 static void e1000_shutdown(struct pci_dev *pdev)
5211 {
5212         bool wake;
5213
5214         __e1000_shutdown(pdev, &wake);
5215
5216         if (system_state == SYSTEM_POWER_OFF) {
5217                 pci_wake_from_d3(pdev, wake);
5218                 pci_set_power_state(pdev, PCI_D3hot);
5219         }
5220 }
5221
5222 #ifdef CONFIG_NET_POLL_CONTROLLER
5223 /* Polling 'interrupt' - used by things like netconsole to send skbs
5224  * without having to re-enable interrupts. It's not called while
5225  * the interrupt routine is executing.
5226  */
5227 static void e1000_netpoll(struct net_device *netdev)
5228 {
5229         struct e1000_adapter *adapter = netdev_priv(netdev);
5230
5231         disable_irq(adapter->pdev->irq);
5232         e1000_intr(adapter->pdev->irq, netdev);
5233         enable_irq(adapter->pdev->irq);
5234 }
5235 #endif
5236
5237 /**
5238  * e1000_io_error_detected - called when PCI error is detected
5239  * @pdev: Pointer to PCI device
5240  * @state: The current pci connection state
5241  *
5242  * This function is called after a PCI bus error affecting
5243  * this device has been detected.
5244  */
5245 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
5246                                                 pci_channel_state_t state)
5247 {
5248         struct net_device *netdev = pci_get_drvdata(pdev);
5249         struct e1000_adapter *adapter = netdev_priv(netdev);
5250
5251         netif_device_detach(netdev);
5252
5253         if (state == pci_channel_io_perm_failure)
5254                 return PCI_ERS_RESULT_DISCONNECT;
5255
5256         if (netif_running(netdev))
5257                 e1000_down(adapter);
5258         pci_disable_device(pdev);
5259
5260         /* Request a slot slot reset. */
5261         return PCI_ERS_RESULT_NEED_RESET;
5262 }
5263
5264 /**
5265  * e1000_io_slot_reset - called after the pci bus has been reset.
5266  * @pdev: Pointer to PCI device
5267  *
5268  * Restart the card from scratch, as if from a cold-boot. Implementation
5269  * resembles the first-half of the e1000_resume routine.
5270  */
5271 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
5272 {
5273         struct net_device *netdev = pci_get_drvdata(pdev);
5274         struct e1000_adapter *adapter = netdev_priv(netdev);
5275         struct e1000_hw *hw = &adapter->hw;
5276         int err;
5277
5278         if (adapter->need_ioport)
5279                 err = pci_enable_device(pdev);
5280         else
5281                 err = pci_enable_device_mem(pdev);
5282         if (err) {
5283                 pr_err("Cannot re-enable PCI device after reset.\n");
5284                 return PCI_ERS_RESULT_DISCONNECT;
5285         }
5286         pci_set_master(pdev);
5287
5288         pci_enable_wake(pdev, PCI_D3hot, 0);
5289         pci_enable_wake(pdev, PCI_D3cold, 0);
5290
5291         e1000_reset(adapter);
5292         ew32(WUS, ~0);
5293
5294         return PCI_ERS_RESULT_RECOVERED;
5295 }
5296
5297 /**
5298  * e1000_io_resume - called when traffic can start flowing again.
5299  * @pdev: Pointer to PCI device
5300  *
5301  * This callback is called when the error recovery driver tells us that
5302  * its OK to resume normal operation. Implementation resembles the
5303  * second-half of the e1000_resume routine.
5304  */
5305 static void e1000_io_resume(struct pci_dev *pdev)
5306 {
5307         struct net_device *netdev = pci_get_drvdata(pdev);
5308         struct e1000_adapter *adapter = netdev_priv(netdev);
5309
5310         e1000_init_manageability(adapter);
5311
5312         if (netif_running(netdev)) {
5313                 if (e1000_up(adapter)) {
5314                         pr_info("can't bring device back up after reset\n");
5315                         return;
5316                 }
5317         }
5318
5319         netif_device_attach(netdev);
5320 }
5321
5322 /* e1000_main.c */