e1000: call pci_save_state after pci_restore_state
[safe/jmp/linux-2.6] / drivers / net / 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
32 char e1000_driver_name[] = "e1000";
33 static char e1000_driver_string[] = "Intel(R) PRO/1000 Network Driver";
34 #define DRV_VERSION "7.3.21-k5-NAPI"
35 const char e1000_driver_version[] = DRV_VERSION;
36 static const char e1000_copyright[] = "Copyright (c) 1999-2006 Intel Corporation.";
37
38 /* e1000_pci_tbl - PCI Device ID Table
39  *
40  * Last entry must be all 0s
41  *
42  * Macro expands to...
43  *   {PCI_DEVICE(PCI_VENDOR_ID_INTEL, device_id)}
44  */
45 static DEFINE_PCI_DEVICE_TABLE(e1000_pci_tbl) = {
46         INTEL_E1000_ETHERNET_DEVICE(0x1000),
47         INTEL_E1000_ETHERNET_DEVICE(0x1001),
48         INTEL_E1000_ETHERNET_DEVICE(0x1004),
49         INTEL_E1000_ETHERNET_DEVICE(0x1008),
50         INTEL_E1000_ETHERNET_DEVICE(0x1009),
51         INTEL_E1000_ETHERNET_DEVICE(0x100C),
52         INTEL_E1000_ETHERNET_DEVICE(0x100D),
53         INTEL_E1000_ETHERNET_DEVICE(0x100E),
54         INTEL_E1000_ETHERNET_DEVICE(0x100F),
55         INTEL_E1000_ETHERNET_DEVICE(0x1010),
56         INTEL_E1000_ETHERNET_DEVICE(0x1011),
57         INTEL_E1000_ETHERNET_DEVICE(0x1012),
58         INTEL_E1000_ETHERNET_DEVICE(0x1013),
59         INTEL_E1000_ETHERNET_DEVICE(0x1014),
60         INTEL_E1000_ETHERNET_DEVICE(0x1015),
61         INTEL_E1000_ETHERNET_DEVICE(0x1016),
62         INTEL_E1000_ETHERNET_DEVICE(0x1017),
63         INTEL_E1000_ETHERNET_DEVICE(0x1018),
64         INTEL_E1000_ETHERNET_DEVICE(0x1019),
65         INTEL_E1000_ETHERNET_DEVICE(0x101A),
66         INTEL_E1000_ETHERNET_DEVICE(0x101D),
67         INTEL_E1000_ETHERNET_DEVICE(0x101E),
68         INTEL_E1000_ETHERNET_DEVICE(0x1026),
69         INTEL_E1000_ETHERNET_DEVICE(0x1027),
70         INTEL_E1000_ETHERNET_DEVICE(0x1028),
71         INTEL_E1000_ETHERNET_DEVICE(0x1075),
72         INTEL_E1000_ETHERNET_DEVICE(0x1076),
73         INTEL_E1000_ETHERNET_DEVICE(0x1077),
74         INTEL_E1000_ETHERNET_DEVICE(0x1078),
75         INTEL_E1000_ETHERNET_DEVICE(0x1079),
76         INTEL_E1000_ETHERNET_DEVICE(0x107A),
77         INTEL_E1000_ETHERNET_DEVICE(0x107B),
78         INTEL_E1000_ETHERNET_DEVICE(0x107C),
79         INTEL_E1000_ETHERNET_DEVICE(0x108A),
80         INTEL_E1000_ETHERNET_DEVICE(0x1099),
81         INTEL_E1000_ETHERNET_DEVICE(0x10B5),
82         /* required last entry */
83         {0,}
84 };
85
86 MODULE_DEVICE_TABLE(pci, e1000_pci_tbl);
87
88 int e1000_up(struct e1000_adapter *adapter);
89 void e1000_down(struct e1000_adapter *adapter);
90 void e1000_reinit_locked(struct e1000_adapter *adapter);
91 void e1000_reset(struct e1000_adapter *adapter);
92 int e1000_set_spd_dplx(struct e1000_adapter *adapter, u16 spddplx);
93 int e1000_setup_all_tx_resources(struct e1000_adapter *adapter);
94 int e1000_setup_all_rx_resources(struct e1000_adapter *adapter);
95 void e1000_free_all_tx_resources(struct e1000_adapter *adapter);
96 void e1000_free_all_rx_resources(struct e1000_adapter *adapter);
97 static int e1000_setup_tx_resources(struct e1000_adapter *adapter,
98                              struct e1000_tx_ring *txdr);
99 static int e1000_setup_rx_resources(struct e1000_adapter *adapter,
100                              struct e1000_rx_ring *rxdr);
101 static void e1000_free_tx_resources(struct e1000_adapter *adapter,
102                              struct e1000_tx_ring *tx_ring);
103 static void e1000_free_rx_resources(struct e1000_adapter *adapter,
104                              struct e1000_rx_ring *rx_ring);
105 void e1000_update_stats(struct e1000_adapter *adapter);
106
107 static int e1000_init_module(void);
108 static void e1000_exit_module(void);
109 static int e1000_probe(struct pci_dev *pdev, const struct pci_device_id *ent);
110 static void __devexit e1000_remove(struct pci_dev *pdev);
111 static int e1000_alloc_queues(struct e1000_adapter *adapter);
112 static int e1000_sw_init(struct e1000_adapter *adapter);
113 static int e1000_open(struct net_device *netdev);
114 static int e1000_close(struct net_device *netdev);
115 static void e1000_configure_tx(struct e1000_adapter *adapter);
116 static void e1000_configure_rx(struct e1000_adapter *adapter);
117 static void e1000_setup_rctl(struct e1000_adapter *adapter);
118 static void e1000_clean_all_tx_rings(struct e1000_adapter *adapter);
119 static void e1000_clean_all_rx_rings(struct e1000_adapter *adapter);
120 static void e1000_clean_tx_ring(struct e1000_adapter *adapter,
121                                 struct e1000_tx_ring *tx_ring);
122 static void e1000_clean_rx_ring(struct e1000_adapter *adapter,
123                                 struct e1000_rx_ring *rx_ring);
124 static void e1000_set_rx_mode(struct net_device *netdev);
125 static void e1000_update_phy_info(unsigned long data);
126 static void e1000_watchdog(unsigned long data);
127 static void e1000_82547_tx_fifo_stall(unsigned long data);
128 static netdev_tx_t e1000_xmit_frame(struct sk_buff *skb,
129                                     struct net_device *netdev);
130 static struct net_device_stats * e1000_get_stats(struct net_device *netdev);
131 static int e1000_change_mtu(struct net_device *netdev, int new_mtu);
132 static int e1000_set_mac(struct net_device *netdev, void *p);
133 static irqreturn_t e1000_intr(int irq, void *data);
134 static bool e1000_clean_tx_irq(struct e1000_adapter *adapter,
135                                struct e1000_tx_ring *tx_ring);
136 static int e1000_clean(struct napi_struct *napi, int budget);
137 static bool e1000_clean_rx_irq(struct e1000_adapter *adapter,
138                                struct e1000_rx_ring *rx_ring,
139                                int *work_done, int work_to_do);
140 static bool e1000_clean_jumbo_rx_irq(struct e1000_adapter *adapter,
141                                      struct e1000_rx_ring *rx_ring,
142                                      int *work_done, int work_to_do);
143 static void e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
144                                    struct e1000_rx_ring *rx_ring,
145                                    int cleaned_count);
146 static void e1000_alloc_jumbo_rx_buffers(struct e1000_adapter *adapter,
147                                          struct e1000_rx_ring *rx_ring,
148                                          int cleaned_count);
149 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd);
150 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
151                            int cmd);
152 static void e1000_enter_82542_rst(struct e1000_adapter *adapter);
153 static void e1000_leave_82542_rst(struct e1000_adapter *adapter);
154 static void e1000_tx_timeout(struct net_device *dev);
155 static void e1000_reset_task(struct work_struct *work);
156 static void e1000_smartspeed(struct e1000_adapter *adapter);
157 static int e1000_82547_fifo_workaround(struct e1000_adapter *adapter,
158                                        struct sk_buff *skb);
159
160 static void e1000_vlan_rx_register(struct net_device *netdev, struct vlan_group *grp);
161 static void e1000_vlan_rx_add_vid(struct net_device *netdev, u16 vid);
162 static void e1000_vlan_rx_kill_vid(struct net_device *netdev, u16 vid);
163 static void e1000_restore_vlan(struct e1000_adapter *adapter);
164
165 #ifdef CONFIG_PM
166 static int e1000_suspend(struct pci_dev *pdev, pm_message_t state);
167 static int e1000_resume(struct pci_dev *pdev);
168 #endif
169 static void e1000_shutdown(struct pci_dev *pdev);
170
171 #ifdef CONFIG_NET_POLL_CONTROLLER
172 /* for netdump / net console */
173 static void e1000_netpoll (struct net_device *netdev);
174 #endif
175
176 #define COPYBREAK_DEFAULT 256
177 static unsigned int copybreak __read_mostly = COPYBREAK_DEFAULT;
178 module_param(copybreak, uint, 0644);
179 MODULE_PARM_DESC(copybreak,
180         "Maximum size of packet that is copied to a new buffer on receive");
181
182 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
183                      pci_channel_state_t state);
184 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev);
185 static void e1000_io_resume(struct pci_dev *pdev);
186
187 static struct pci_error_handlers e1000_err_handler = {
188         .error_detected = e1000_io_error_detected,
189         .slot_reset = e1000_io_slot_reset,
190         .resume = e1000_io_resume,
191 };
192
193 static struct pci_driver e1000_driver = {
194         .name     = e1000_driver_name,
195         .id_table = e1000_pci_tbl,
196         .probe    = e1000_probe,
197         .remove   = __devexit_p(e1000_remove),
198 #ifdef CONFIG_PM
199         /* Power Managment Hooks */
200         .suspend  = e1000_suspend,
201         .resume   = e1000_resume,
202 #endif
203         .shutdown = e1000_shutdown,
204         .err_handler = &e1000_err_handler
205 };
206
207 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
208 MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
209 MODULE_LICENSE("GPL");
210 MODULE_VERSION(DRV_VERSION);
211
212 static int debug = NETIF_MSG_DRV | NETIF_MSG_PROBE;
213 module_param(debug, int, 0);
214 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
215
216 /**
217  * e1000_init_module - Driver Registration Routine
218  *
219  * e1000_init_module is the first routine called when the driver is
220  * loaded. All it does is register with the PCI subsystem.
221  **/
222
223 static int __init e1000_init_module(void)
224 {
225         int ret;
226         printk(KERN_INFO "%s - version %s\n",
227                e1000_driver_string, e1000_driver_version);
228
229         printk(KERN_INFO "%s\n", e1000_copyright);
230
231         ret = pci_register_driver(&e1000_driver);
232         if (copybreak != COPYBREAK_DEFAULT) {
233                 if (copybreak == 0)
234                         printk(KERN_INFO "e1000: copybreak disabled\n");
235                 else
236                         printk(KERN_INFO "e1000: copybreak enabled for "
237                                "packets <= %u bytes\n", copybreak);
238         }
239         return ret;
240 }
241
242 module_init(e1000_init_module);
243
244 /**
245  * e1000_exit_module - Driver Exit Cleanup Routine
246  *
247  * e1000_exit_module is called just before the driver is removed
248  * from memory.
249  **/
250
251 static void __exit e1000_exit_module(void)
252 {
253         pci_unregister_driver(&e1000_driver);
254 }
255
256 module_exit(e1000_exit_module);
257
258 static int e1000_request_irq(struct e1000_adapter *adapter)
259 {
260         struct net_device *netdev = adapter->netdev;
261         irq_handler_t handler = e1000_intr;
262         int irq_flags = IRQF_SHARED;
263         int err;
264
265         err = request_irq(adapter->pdev->irq, handler, irq_flags, netdev->name,
266                           netdev);
267         if (err) {
268                 DPRINTK(PROBE, ERR,
269                         "Unable to allocate interrupt Error: %d\n", err);
270         }
271
272         return err;
273 }
274
275 static void e1000_free_irq(struct e1000_adapter *adapter)
276 {
277         struct net_device *netdev = adapter->netdev;
278
279         free_irq(adapter->pdev->irq, netdev);
280 }
281
282 /**
283  * e1000_irq_disable - Mask off interrupt generation on the NIC
284  * @adapter: board private structure
285  **/
286
287 static void e1000_irq_disable(struct e1000_adapter *adapter)
288 {
289         struct e1000_hw *hw = &adapter->hw;
290
291         ew32(IMC, ~0);
292         E1000_WRITE_FLUSH();
293         synchronize_irq(adapter->pdev->irq);
294 }
295
296 /**
297  * e1000_irq_enable - Enable default interrupt generation settings
298  * @adapter: board private structure
299  **/
300
301 static void e1000_irq_enable(struct e1000_adapter *adapter)
302 {
303         struct e1000_hw *hw = &adapter->hw;
304
305         ew32(IMS, IMS_ENABLE_MASK);
306         E1000_WRITE_FLUSH();
307 }
308
309 static void e1000_update_mng_vlan(struct e1000_adapter *adapter)
310 {
311         struct e1000_hw *hw = &adapter->hw;
312         struct net_device *netdev = adapter->netdev;
313         u16 vid = hw->mng_cookie.vlan_id;
314         u16 old_vid = adapter->mng_vlan_id;
315         if (adapter->vlgrp) {
316                 if (!vlan_group_get_device(adapter->vlgrp, vid)) {
317                         if (hw->mng_cookie.status &
318                                 E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT) {
319                                 e1000_vlan_rx_add_vid(netdev, vid);
320                                 adapter->mng_vlan_id = vid;
321                         } else
322                                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
323
324                         if ((old_vid != (u16)E1000_MNG_VLAN_NONE) &&
325                                         (vid != old_vid) &&
326                             !vlan_group_get_device(adapter->vlgrp, old_vid))
327                                 e1000_vlan_rx_kill_vid(netdev, old_vid);
328                 } else
329                         adapter->mng_vlan_id = vid;
330         }
331 }
332
333 static void e1000_init_manageability(struct e1000_adapter *adapter)
334 {
335         struct e1000_hw *hw = &adapter->hw;
336
337         if (adapter->en_mng_pt) {
338                 u32 manc = er32(MANC);
339
340                 /* disable hardware interception of ARP */
341                 manc &= ~(E1000_MANC_ARP_EN);
342
343                 ew32(MANC, manc);
344         }
345 }
346
347 static void e1000_release_manageability(struct e1000_adapter *adapter)
348 {
349         struct e1000_hw *hw = &adapter->hw;
350
351         if (adapter->en_mng_pt) {
352                 u32 manc = er32(MANC);
353
354                 /* re-enable hardware interception of ARP */
355                 manc |= E1000_MANC_ARP_EN;
356
357                 ew32(MANC, manc);
358         }
359 }
360
361 /**
362  * e1000_configure - configure the hardware for RX and TX
363  * @adapter = private board structure
364  **/
365 static void e1000_configure(struct e1000_adapter *adapter)
366 {
367         struct net_device *netdev = adapter->netdev;
368         int i;
369
370         e1000_set_rx_mode(netdev);
371
372         e1000_restore_vlan(adapter);
373         e1000_init_manageability(adapter);
374
375         e1000_configure_tx(adapter);
376         e1000_setup_rctl(adapter);
377         e1000_configure_rx(adapter);
378         /* call E1000_DESC_UNUSED which always leaves
379          * at least 1 descriptor unused to make sure
380          * next_to_use != next_to_clean */
381         for (i = 0; i < adapter->num_rx_queues; i++) {
382                 struct e1000_rx_ring *ring = &adapter->rx_ring[i];
383                 adapter->alloc_rx_buf(adapter, ring,
384                                       E1000_DESC_UNUSED(ring));
385         }
386
387         adapter->tx_queue_len = netdev->tx_queue_len;
388 }
389
390 int e1000_up(struct e1000_adapter *adapter)
391 {
392         struct e1000_hw *hw = &adapter->hw;
393
394         /* hardware has been reset, we need to reload some things */
395         e1000_configure(adapter);
396
397         clear_bit(__E1000_DOWN, &adapter->flags);
398
399         napi_enable(&adapter->napi);
400
401         e1000_irq_enable(adapter);
402
403         netif_wake_queue(adapter->netdev);
404
405         /* fire a link change interrupt to start the watchdog */
406         ew32(ICS, E1000_ICS_LSC);
407         return 0;
408 }
409
410 /**
411  * e1000_power_up_phy - restore link in case the phy was powered down
412  * @adapter: address of board private structure
413  *
414  * The phy may be powered down to save power and turn off link when the
415  * driver is unloaded and wake on lan is not enabled (among others)
416  * *** this routine MUST be followed by a call to e1000_reset ***
417  *
418  **/
419
420 void e1000_power_up_phy(struct e1000_adapter *adapter)
421 {
422         struct e1000_hw *hw = &adapter->hw;
423         u16 mii_reg = 0;
424
425         /* Just clear the power down bit to wake the phy back up */
426         if (hw->media_type == e1000_media_type_copper) {
427                 /* according to the manual, the phy will retain its
428                  * settings across a power-down/up cycle */
429                 e1000_read_phy_reg(hw, PHY_CTRL, &mii_reg);
430                 mii_reg &= ~MII_CR_POWER_DOWN;
431                 e1000_write_phy_reg(hw, PHY_CTRL, mii_reg);
432         }
433 }
434
435 static void e1000_power_down_phy(struct e1000_adapter *adapter)
436 {
437         struct e1000_hw *hw = &adapter->hw;
438
439         /* Power down the PHY so no link is implied when interface is down *
440          * The PHY cannot be powered down if any of the following is true *
441          * (a) WoL is enabled
442          * (b) AMT is active
443          * (c) SoL/IDER session is active */
444         if (!adapter->wol && hw->mac_type >= e1000_82540 &&
445            hw->media_type == e1000_media_type_copper) {
446                 u16 mii_reg = 0;
447
448                 switch (hw->mac_type) {
449                 case e1000_82540:
450                 case e1000_82545:
451                 case e1000_82545_rev_3:
452                 case e1000_82546:
453                 case e1000_82546_rev_3:
454                 case e1000_82541:
455                 case e1000_82541_rev_2:
456                 case e1000_82547:
457                 case e1000_82547_rev_2:
458                         if (er32(MANC) & E1000_MANC_SMBUS_EN)
459                                 goto out;
460                         break;
461                 default:
462                         goto out;
463                 }
464                 e1000_read_phy_reg(hw, PHY_CTRL, &mii_reg);
465                 mii_reg |= MII_CR_POWER_DOWN;
466                 e1000_write_phy_reg(hw, PHY_CTRL, mii_reg);
467                 mdelay(1);
468         }
469 out:
470         return;
471 }
472
473 void e1000_down(struct e1000_adapter *adapter)
474 {
475         struct e1000_hw *hw = &adapter->hw;
476         struct net_device *netdev = adapter->netdev;
477         u32 rctl, tctl;
478
479         /* signal that we're down so the interrupt handler does not
480          * reschedule our watchdog timer */
481         set_bit(__E1000_DOWN, &adapter->flags);
482
483         /* disable receives in the hardware */
484         rctl = er32(RCTL);
485         ew32(RCTL, rctl & ~E1000_RCTL_EN);
486         /* flush and sleep below */
487
488         netif_tx_disable(netdev);
489
490         /* disable transmits in the hardware */
491         tctl = er32(TCTL);
492         tctl &= ~E1000_TCTL_EN;
493         ew32(TCTL, tctl);
494         /* flush both disables and wait for them to finish */
495         E1000_WRITE_FLUSH();
496         msleep(10);
497
498         napi_disable(&adapter->napi);
499
500         e1000_irq_disable(adapter);
501
502         del_timer_sync(&adapter->tx_fifo_stall_timer);
503         del_timer_sync(&adapter->watchdog_timer);
504         del_timer_sync(&adapter->phy_info_timer);
505
506         netdev->tx_queue_len = adapter->tx_queue_len;
507         adapter->link_speed = 0;
508         adapter->link_duplex = 0;
509         netif_carrier_off(netdev);
510
511         e1000_reset(adapter);
512         e1000_clean_all_tx_rings(adapter);
513         e1000_clean_all_rx_rings(adapter);
514 }
515
516 void e1000_reinit_locked(struct e1000_adapter *adapter)
517 {
518         WARN_ON(in_interrupt());
519         while (test_and_set_bit(__E1000_RESETTING, &adapter->flags))
520                 msleep(1);
521         e1000_down(adapter);
522         e1000_up(adapter);
523         clear_bit(__E1000_RESETTING, &adapter->flags);
524 }
525
526 void e1000_reset(struct e1000_adapter *adapter)
527 {
528         struct e1000_hw *hw = &adapter->hw;
529         u32 pba = 0, tx_space, min_tx_space, min_rx_space;
530         bool legacy_pba_adjust = false;
531         u16 hwm;
532
533         /* Repartition Pba for greater than 9k mtu
534          * To take effect CTRL.RST is required.
535          */
536
537         switch (hw->mac_type) {
538         case e1000_82542_rev2_0:
539         case e1000_82542_rev2_1:
540         case e1000_82543:
541         case e1000_82544:
542         case e1000_82540:
543         case e1000_82541:
544         case e1000_82541_rev_2:
545                 legacy_pba_adjust = true;
546                 pba = E1000_PBA_48K;
547                 break;
548         case e1000_82545:
549         case e1000_82545_rev_3:
550         case e1000_82546:
551         case e1000_82546_rev_3:
552                 pba = E1000_PBA_48K;
553                 break;
554         case e1000_82547:
555         case e1000_82547_rev_2:
556                 legacy_pba_adjust = true;
557                 pba = E1000_PBA_30K;
558                 break;
559         case e1000_undefined:
560         case e1000_num_macs:
561                 break;
562         }
563
564         if (legacy_pba_adjust) {
565                 if (hw->max_frame_size > E1000_RXBUFFER_8192)
566                         pba -= 8; /* allocate more FIFO for Tx */
567
568                 if (hw->mac_type == e1000_82547) {
569                         adapter->tx_fifo_head = 0;
570                         adapter->tx_head_addr = pba << E1000_TX_HEAD_ADDR_SHIFT;
571                         adapter->tx_fifo_size =
572                                 (E1000_PBA_40K - pba) << E1000_PBA_BYTES_SHIFT;
573                         atomic_set(&adapter->tx_fifo_stall, 0);
574                 }
575         } else if (hw->max_frame_size >  ETH_FRAME_LEN + ETH_FCS_LEN) {
576                 /* adjust PBA for jumbo frames */
577                 ew32(PBA, pba);
578
579                 /* To maintain wire speed transmits, the Tx FIFO should be
580                  * large enough to accommodate two full transmit packets,
581                  * rounded up to the next 1KB and expressed in KB.  Likewise,
582                  * the Rx FIFO should be large enough to accommodate at least
583                  * one full receive packet and is similarly rounded up and
584                  * expressed in KB. */
585                 pba = er32(PBA);
586                 /* upper 16 bits has Tx packet buffer allocation size in KB */
587                 tx_space = pba >> 16;
588                 /* lower 16 bits has Rx packet buffer allocation size in KB */
589                 pba &= 0xffff;
590                 /*
591                  * the tx fifo also stores 16 bytes of information about the tx
592                  * but don't include ethernet FCS because hardware appends it
593                  */
594                 min_tx_space = (hw->max_frame_size +
595                                 sizeof(struct e1000_tx_desc) -
596                                 ETH_FCS_LEN) * 2;
597                 min_tx_space = ALIGN(min_tx_space, 1024);
598                 min_tx_space >>= 10;
599                 /* software strips receive CRC, so leave room for it */
600                 min_rx_space = hw->max_frame_size;
601                 min_rx_space = ALIGN(min_rx_space, 1024);
602                 min_rx_space >>= 10;
603
604                 /* If current Tx allocation is less than the min Tx FIFO size,
605                  * and the min Tx FIFO size is less than the current Rx FIFO
606                  * allocation, take space away from current Rx allocation */
607                 if (tx_space < min_tx_space &&
608                     ((min_tx_space - tx_space) < pba)) {
609                         pba = pba - (min_tx_space - tx_space);
610
611                         /* PCI/PCIx hardware has PBA alignment constraints */
612                         switch (hw->mac_type) {
613                         case e1000_82545 ... e1000_82546_rev_3:
614                                 pba &= ~(E1000_PBA_8K - 1);
615                                 break;
616                         default:
617                                 break;
618                         }
619
620                         /* if short on rx space, rx wins and must trump tx
621                          * adjustment or use Early Receive if available */
622                         if (pba < min_rx_space)
623                                 pba = min_rx_space;
624                 }
625         }
626
627         ew32(PBA, pba);
628
629         /*
630          * flow control settings:
631          * The high water mark must be low enough to fit one full frame
632          * (or the size used for early receive) above it in the Rx FIFO.
633          * Set it to the lower of:
634          * - 90% of the Rx FIFO size, and
635          * - the full Rx FIFO size minus the early receive size (for parts
636          *   with ERT support assuming ERT set to E1000_ERT_2048), or
637          * - the full Rx FIFO size minus one full frame
638          */
639         hwm = min(((pba << 10) * 9 / 10),
640                   ((pba << 10) - hw->max_frame_size));
641
642         hw->fc_high_water = hwm & 0xFFF8;       /* 8-byte granularity */
643         hw->fc_low_water = hw->fc_high_water - 8;
644         hw->fc_pause_time = E1000_FC_PAUSE_TIME;
645         hw->fc_send_xon = 1;
646         hw->fc = hw->original_fc;
647
648         /* Allow time for pending master requests to run */
649         e1000_reset_hw(hw);
650         if (hw->mac_type >= e1000_82544)
651                 ew32(WUC, 0);
652
653         if (e1000_init_hw(hw))
654                 DPRINTK(PROBE, ERR, "Hardware Error\n");
655         e1000_update_mng_vlan(adapter);
656
657         /* if (adapter->hwflags & HWFLAGS_PHY_PWR_BIT) { */
658         if (hw->mac_type >= e1000_82544 &&
659             hw->autoneg == 1 &&
660             hw->autoneg_advertised == ADVERTISE_1000_FULL) {
661                 u32 ctrl = er32(CTRL);
662                 /* clear phy power management bit if we are in gig only mode,
663                  * which if enabled will attempt negotiation to 100Mb, which
664                  * can cause a loss of link at power off or driver unload */
665                 ctrl &= ~E1000_CTRL_SWDPIN3;
666                 ew32(CTRL, ctrl);
667         }
668
669         /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
670         ew32(VET, ETHERNET_IEEE_VLAN_TYPE);
671
672         e1000_reset_adaptive(hw);
673         e1000_phy_get_info(hw, &adapter->phy_info);
674
675         e1000_release_manageability(adapter);
676 }
677
678 /**
679  *  Dump the eeprom for users having checksum issues
680  **/
681 static void e1000_dump_eeprom(struct e1000_adapter *adapter)
682 {
683         struct net_device *netdev = adapter->netdev;
684         struct ethtool_eeprom eeprom;
685         const struct ethtool_ops *ops = netdev->ethtool_ops;
686         u8 *data;
687         int i;
688         u16 csum_old, csum_new = 0;
689
690         eeprom.len = ops->get_eeprom_len(netdev);
691         eeprom.offset = 0;
692
693         data = kmalloc(eeprom.len, GFP_KERNEL);
694         if (!data) {
695                 printk(KERN_ERR "Unable to allocate memory to dump EEPROM"
696                        " data\n");
697                 return;
698         }
699
700         ops->get_eeprom(netdev, &eeprom, data);
701
702         csum_old = (data[EEPROM_CHECKSUM_REG * 2]) +
703                    (data[EEPROM_CHECKSUM_REG * 2 + 1] << 8);
704         for (i = 0; i < EEPROM_CHECKSUM_REG * 2; i += 2)
705                 csum_new += data[i] + (data[i + 1] << 8);
706         csum_new = EEPROM_SUM - csum_new;
707
708         printk(KERN_ERR "/*********************/\n");
709         printk(KERN_ERR "Current EEPROM Checksum : 0x%04x\n", csum_old);
710         printk(KERN_ERR "Calculated              : 0x%04x\n", csum_new);
711
712         printk(KERN_ERR "Offset    Values\n");
713         printk(KERN_ERR "========  ======\n");
714         print_hex_dump(KERN_ERR, "", DUMP_PREFIX_OFFSET, 16, 1, data, 128, 0);
715
716         printk(KERN_ERR "Include this output when contacting your support "
717                "provider.\n");
718         printk(KERN_ERR "This is not a software error! Something bad "
719                "happened to your hardware or\n");
720         printk(KERN_ERR "EEPROM image. Ignoring this "
721                "problem could result in further problems,\n");
722         printk(KERN_ERR "possibly loss of data, corruption or system hangs!\n");
723         printk(KERN_ERR "The MAC Address will be reset to 00:00:00:00:00:00, "
724                "which is invalid\n");
725         printk(KERN_ERR "and requires you to set the proper MAC "
726                "address manually before continuing\n");
727         printk(KERN_ERR "to enable this network device.\n");
728         printk(KERN_ERR "Please inspect the EEPROM dump and report the issue "
729                "to your hardware vendor\n");
730         printk(KERN_ERR "or Intel Customer Support.\n");
731         printk(KERN_ERR "/*********************/\n");
732
733         kfree(data);
734 }
735
736 /**
737  * e1000_is_need_ioport - determine if an adapter needs ioport resources or not
738  * @pdev: PCI device information struct
739  *
740  * Return true if an adapter needs ioport resources
741  **/
742 static int e1000_is_need_ioport(struct pci_dev *pdev)
743 {
744         switch (pdev->device) {
745         case E1000_DEV_ID_82540EM:
746         case E1000_DEV_ID_82540EM_LOM:
747         case E1000_DEV_ID_82540EP:
748         case E1000_DEV_ID_82540EP_LOM:
749         case E1000_DEV_ID_82540EP_LP:
750         case E1000_DEV_ID_82541EI:
751         case E1000_DEV_ID_82541EI_MOBILE:
752         case E1000_DEV_ID_82541ER:
753         case E1000_DEV_ID_82541ER_LOM:
754         case E1000_DEV_ID_82541GI:
755         case E1000_DEV_ID_82541GI_LF:
756         case E1000_DEV_ID_82541GI_MOBILE:
757         case E1000_DEV_ID_82544EI_COPPER:
758         case E1000_DEV_ID_82544EI_FIBER:
759         case E1000_DEV_ID_82544GC_COPPER:
760         case E1000_DEV_ID_82544GC_LOM:
761         case E1000_DEV_ID_82545EM_COPPER:
762         case E1000_DEV_ID_82545EM_FIBER:
763         case E1000_DEV_ID_82546EB_COPPER:
764         case E1000_DEV_ID_82546EB_FIBER:
765         case E1000_DEV_ID_82546EB_QUAD_COPPER:
766                 return true;
767         default:
768                 return false;
769         }
770 }
771
772 static const struct net_device_ops e1000_netdev_ops = {
773         .ndo_open               = e1000_open,
774         .ndo_stop               = e1000_close,
775         .ndo_start_xmit         = e1000_xmit_frame,
776         .ndo_get_stats          = e1000_get_stats,
777         .ndo_set_rx_mode        = e1000_set_rx_mode,
778         .ndo_set_mac_address    = e1000_set_mac,
779         .ndo_tx_timeout         = e1000_tx_timeout,
780         .ndo_change_mtu         = e1000_change_mtu,
781         .ndo_do_ioctl           = e1000_ioctl,
782         .ndo_validate_addr      = eth_validate_addr,
783
784         .ndo_vlan_rx_register   = e1000_vlan_rx_register,
785         .ndo_vlan_rx_add_vid    = e1000_vlan_rx_add_vid,
786         .ndo_vlan_rx_kill_vid   = e1000_vlan_rx_kill_vid,
787 #ifdef CONFIG_NET_POLL_CONTROLLER
788         .ndo_poll_controller    = e1000_netpoll,
789 #endif
790 };
791
792 /**
793  * e1000_probe - Device Initialization Routine
794  * @pdev: PCI device information struct
795  * @ent: entry in e1000_pci_tbl
796  *
797  * Returns 0 on success, negative on failure
798  *
799  * e1000_probe initializes an adapter identified by a pci_dev structure.
800  * The OS initialization, configuring of the adapter private structure,
801  * and a hardware reset occur.
802  **/
803 static int __devinit e1000_probe(struct pci_dev *pdev,
804                                  const struct pci_device_id *ent)
805 {
806         struct net_device *netdev;
807         struct e1000_adapter *adapter;
808         struct e1000_hw *hw;
809
810         static int cards_found = 0;
811         static int global_quad_port_a = 0; /* global ksp3 port a indication */
812         int i, err, pci_using_dac;
813         u16 eeprom_data = 0;
814         u16 eeprom_apme_mask = E1000_EEPROM_APME;
815         int bars, need_ioport;
816
817         /* do not allocate ioport bars when not needed */
818         need_ioport = e1000_is_need_ioport(pdev);
819         if (need_ioport) {
820                 bars = pci_select_bars(pdev, IORESOURCE_MEM | IORESOURCE_IO);
821                 err = pci_enable_device(pdev);
822         } else {
823                 bars = pci_select_bars(pdev, IORESOURCE_MEM);
824                 err = pci_enable_device_mem(pdev);
825         }
826         if (err)
827                 return err;
828
829         if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(64)) &&
830             !pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64))) {
831                 pci_using_dac = 1;
832         } else {
833                 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
834                 if (err) {
835                         err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
836                         if (err) {
837                                 E1000_ERR("No usable DMA configuration, "
838                                           "aborting\n");
839                                 goto err_dma;
840                         }
841                 }
842                 pci_using_dac = 0;
843         }
844
845         err = pci_request_selected_regions(pdev, bars, e1000_driver_name);
846         if (err)
847                 goto err_pci_reg;
848
849         pci_set_master(pdev);
850         err = pci_save_state(pdev);
851         if (err)
852                 goto err_alloc_etherdev;
853
854         err = -ENOMEM;
855         netdev = alloc_etherdev(sizeof(struct e1000_adapter));
856         if (!netdev)
857                 goto err_alloc_etherdev;
858
859         SET_NETDEV_DEV(netdev, &pdev->dev);
860
861         pci_set_drvdata(pdev, netdev);
862         adapter = netdev_priv(netdev);
863         adapter->netdev = netdev;
864         adapter->pdev = pdev;
865         adapter->msg_enable = (1 << debug) - 1;
866         adapter->bars = bars;
867         adapter->need_ioport = need_ioport;
868
869         hw = &adapter->hw;
870         hw->back = adapter;
871
872         err = -EIO;
873         hw->hw_addr = pci_ioremap_bar(pdev, BAR_0);
874         if (!hw->hw_addr)
875                 goto err_ioremap;
876
877         if (adapter->need_ioport) {
878                 for (i = BAR_1; i <= BAR_5; i++) {
879                         if (pci_resource_len(pdev, i) == 0)
880                                 continue;
881                         if (pci_resource_flags(pdev, i) & IORESOURCE_IO) {
882                                 hw->io_base = pci_resource_start(pdev, i);
883                                 break;
884                         }
885                 }
886         }
887
888         netdev->netdev_ops = &e1000_netdev_ops;
889         e1000_set_ethtool_ops(netdev);
890         netdev->watchdog_timeo = 5 * HZ;
891         netif_napi_add(netdev, &adapter->napi, e1000_clean, 64);
892
893         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
894
895         adapter->bd_number = cards_found;
896
897         /* setup the private structure */
898
899         err = e1000_sw_init(adapter);
900         if (err)
901                 goto err_sw_init;
902
903         err = -EIO;
904
905         if (hw->mac_type >= e1000_82543) {
906                 netdev->features = NETIF_F_SG |
907                                    NETIF_F_HW_CSUM |
908                                    NETIF_F_HW_VLAN_TX |
909                                    NETIF_F_HW_VLAN_RX |
910                                    NETIF_F_HW_VLAN_FILTER;
911         }
912
913         if ((hw->mac_type >= e1000_82544) &&
914            (hw->mac_type != e1000_82547))
915                 netdev->features |= NETIF_F_TSO;
916
917         if (pci_using_dac)
918                 netdev->features |= NETIF_F_HIGHDMA;
919
920         netdev->vlan_features |= NETIF_F_TSO;
921         netdev->vlan_features |= NETIF_F_HW_CSUM;
922         netdev->vlan_features |= NETIF_F_SG;
923
924         adapter->en_mng_pt = e1000_enable_mng_pass_thru(hw);
925
926         /* initialize eeprom parameters */
927         if (e1000_init_eeprom_params(hw)) {
928                 E1000_ERR("EEPROM initialization failed\n");
929                 goto err_eeprom;
930         }
931
932         /* before reading the EEPROM, reset the controller to
933          * put the device in a known good starting state */
934
935         e1000_reset_hw(hw);
936
937         /* make sure the EEPROM is good */
938         if (e1000_validate_eeprom_checksum(hw) < 0) {
939                 DPRINTK(PROBE, ERR, "The EEPROM Checksum Is Not Valid\n");
940                 e1000_dump_eeprom(adapter);
941                 /*
942                  * set MAC address to all zeroes to invalidate and temporary
943                  * disable this device for the user. This blocks regular
944                  * traffic while still permitting ethtool ioctls from reaching
945                  * the hardware as well as allowing the user to run the
946                  * interface after manually setting a hw addr using
947                  * `ip set address`
948                  */
949                 memset(hw->mac_addr, 0, netdev->addr_len);
950         } else {
951                 /* copy the MAC address out of the EEPROM */
952                 if (e1000_read_mac_addr(hw))
953                         DPRINTK(PROBE, ERR, "EEPROM Read Error\n");
954         }
955         /* don't block initalization here due to bad MAC address */
956         memcpy(netdev->dev_addr, hw->mac_addr, netdev->addr_len);
957         memcpy(netdev->perm_addr, hw->mac_addr, netdev->addr_len);
958
959         if (!is_valid_ether_addr(netdev->perm_addr))
960                 DPRINTK(PROBE, ERR, "Invalid MAC Address\n");
961
962         e1000_get_bus_info(hw);
963
964         init_timer(&adapter->tx_fifo_stall_timer);
965         adapter->tx_fifo_stall_timer.function = &e1000_82547_tx_fifo_stall;
966         adapter->tx_fifo_stall_timer.data = (unsigned long)adapter;
967
968         init_timer(&adapter->watchdog_timer);
969         adapter->watchdog_timer.function = &e1000_watchdog;
970         adapter->watchdog_timer.data = (unsigned long) adapter;
971
972         init_timer(&adapter->phy_info_timer);
973         adapter->phy_info_timer.function = &e1000_update_phy_info;
974         adapter->phy_info_timer.data = (unsigned long)adapter;
975
976         INIT_WORK(&adapter->reset_task, e1000_reset_task);
977
978         e1000_check_options(adapter);
979
980         /* Initial Wake on LAN setting
981          * If APM wake is enabled in the EEPROM,
982          * enable the ACPI Magic Packet filter
983          */
984
985         switch (hw->mac_type) {
986         case e1000_82542_rev2_0:
987         case e1000_82542_rev2_1:
988         case e1000_82543:
989                 break;
990         case e1000_82544:
991                 e1000_read_eeprom(hw,
992                         EEPROM_INIT_CONTROL2_REG, 1, &eeprom_data);
993                 eeprom_apme_mask = E1000_EEPROM_82544_APM;
994                 break;
995         case e1000_82546:
996         case e1000_82546_rev_3:
997                 if (er32(STATUS) & E1000_STATUS_FUNC_1){
998                         e1000_read_eeprom(hw,
999                                 EEPROM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
1000                         break;
1001                 }
1002                 /* Fall Through */
1003         default:
1004                 e1000_read_eeprom(hw,
1005                         EEPROM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
1006                 break;
1007         }
1008         if (eeprom_data & eeprom_apme_mask)
1009                 adapter->eeprom_wol |= E1000_WUFC_MAG;
1010
1011         /* now that we have the eeprom settings, apply the special cases
1012          * where the eeprom may be wrong or the board simply won't support
1013          * wake on lan on a particular port */
1014         switch (pdev->device) {
1015         case E1000_DEV_ID_82546GB_PCIE:
1016                 adapter->eeprom_wol = 0;
1017                 break;
1018         case E1000_DEV_ID_82546EB_FIBER:
1019         case E1000_DEV_ID_82546GB_FIBER:
1020                 /* Wake events only supported on port A for dual fiber
1021                  * regardless of eeprom setting */
1022                 if (er32(STATUS) & E1000_STATUS_FUNC_1)
1023                         adapter->eeprom_wol = 0;
1024                 break;
1025         case E1000_DEV_ID_82546GB_QUAD_COPPER_KSP3:
1026                 /* if quad port adapter, disable WoL on all but port A */
1027                 if (global_quad_port_a != 0)
1028                         adapter->eeprom_wol = 0;
1029                 else
1030                         adapter->quad_port_a = 1;
1031                 /* Reset for multiple quad port adapters */
1032                 if (++global_quad_port_a == 4)
1033                         global_quad_port_a = 0;
1034                 break;
1035         }
1036
1037         /* initialize the wol settings based on the eeprom settings */
1038         adapter->wol = adapter->eeprom_wol;
1039         device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
1040
1041         /* print bus type/speed/width info */
1042         DPRINTK(PROBE, INFO, "(PCI%s:%s:%s) ",
1043                 ((hw->bus_type == e1000_bus_type_pcix) ? "-X" : ""),
1044                 ((hw->bus_speed == e1000_bus_speed_133) ? "133MHz" :
1045                  (hw->bus_speed == e1000_bus_speed_120) ? "120MHz" :
1046                  (hw->bus_speed == e1000_bus_speed_100) ? "100MHz" :
1047                  (hw->bus_speed == e1000_bus_speed_66) ? "66MHz" : "33MHz"),
1048                 ((hw->bus_width == e1000_bus_width_64) ? "64-bit" : "32-bit"));
1049
1050         printk("%pM\n", netdev->dev_addr);
1051
1052         /* reset the hardware with the new settings */
1053         e1000_reset(adapter);
1054
1055         strcpy(netdev->name, "eth%d");
1056         err = register_netdev(netdev);
1057         if (err)
1058                 goto err_register;
1059
1060         /* carrier off reporting is important to ethtool even BEFORE open */
1061         netif_carrier_off(netdev);
1062
1063         DPRINTK(PROBE, INFO, "Intel(R) PRO/1000 Network Connection\n");
1064
1065         cards_found++;
1066         return 0;
1067
1068 err_register:
1069 err_eeprom:
1070         e1000_phy_hw_reset(hw);
1071
1072         if (hw->flash_address)
1073                 iounmap(hw->flash_address);
1074         kfree(adapter->tx_ring);
1075         kfree(adapter->rx_ring);
1076 err_sw_init:
1077         iounmap(hw->hw_addr);
1078 err_ioremap:
1079         free_netdev(netdev);
1080 err_alloc_etherdev:
1081         pci_release_selected_regions(pdev, bars);
1082 err_pci_reg:
1083 err_dma:
1084         pci_disable_device(pdev);
1085         return err;
1086 }
1087
1088 /**
1089  * e1000_remove - Device Removal Routine
1090  * @pdev: PCI device information struct
1091  *
1092  * e1000_remove is called by the PCI subsystem to alert the driver
1093  * that it should release a PCI device.  The could be caused by a
1094  * Hot-Plug event, or because the driver is going to be removed from
1095  * memory.
1096  **/
1097
1098 static void __devexit e1000_remove(struct pci_dev *pdev)
1099 {
1100         struct net_device *netdev = pci_get_drvdata(pdev);
1101         struct e1000_adapter *adapter = netdev_priv(netdev);
1102         struct e1000_hw *hw = &adapter->hw;
1103
1104         set_bit(__E1000_DOWN, &adapter->flags);
1105         del_timer_sync(&adapter->tx_fifo_stall_timer);
1106         del_timer_sync(&adapter->watchdog_timer);
1107         del_timer_sync(&adapter->phy_info_timer);
1108
1109         cancel_work_sync(&adapter->reset_task);
1110
1111         e1000_release_manageability(adapter);
1112
1113         unregister_netdev(netdev);
1114
1115         e1000_phy_hw_reset(hw);
1116
1117         kfree(adapter->tx_ring);
1118         kfree(adapter->rx_ring);
1119
1120         iounmap(hw->hw_addr);
1121         if (hw->flash_address)
1122                 iounmap(hw->flash_address);
1123         pci_release_selected_regions(pdev, adapter->bars);
1124
1125         free_netdev(netdev);
1126
1127         pci_disable_device(pdev);
1128 }
1129
1130 /**
1131  * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
1132  * @adapter: board private structure to initialize
1133  *
1134  * e1000_sw_init initializes the Adapter private data structure.
1135  * Fields are initialized based on PCI device information and
1136  * OS network device settings (MTU size).
1137  **/
1138
1139 static int __devinit e1000_sw_init(struct e1000_adapter *adapter)
1140 {
1141         struct e1000_hw *hw = &adapter->hw;
1142         struct net_device *netdev = adapter->netdev;
1143         struct pci_dev *pdev = adapter->pdev;
1144
1145         /* PCI config space info */
1146
1147         hw->vendor_id = pdev->vendor;
1148         hw->device_id = pdev->device;
1149         hw->subsystem_vendor_id = pdev->subsystem_vendor;
1150         hw->subsystem_id = pdev->subsystem_device;
1151         hw->revision_id = pdev->revision;
1152
1153         pci_read_config_word(pdev, PCI_COMMAND, &hw->pci_cmd_word);
1154
1155         adapter->rx_buffer_len = MAXIMUM_ETHERNET_VLAN_SIZE;
1156         hw->max_frame_size = netdev->mtu +
1157                              ENET_HEADER_SIZE + ETHERNET_FCS_SIZE;
1158         hw->min_frame_size = MINIMUM_ETHERNET_FRAME_SIZE;
1159
1160         /* identify the MAC */
1161
1162         if (e1000_set_mac_type(hw)) {
1163                 DPRINTK(PROBE, ERR, "Unknown MAC Type\n");
1164                 return -EIO;
1165         }
1166
1167         switch (hw->mac_type) {
1168         default:
1169                 break;
1170         case e1000_82541:
1171         case e1000_82547:
1172         case e1000_82541_rev_2:
1173         case e1000_82547_rev_2:
1174                 hw->phy_init_script = 1;
1175                 break;
1176         }
1177
1178         e1000_set_media_type(hw);
1179
1180         hw->wait_autoneg_complete = false;
1181         hw->tbi_compatibility_en = true;
1182         hw->adaptive_ifs = true;
1183
1184         /* Copper options */
1185
1186         if (hw->media_type == e1000_media_type_copper) {
1187                 hw->mdix = AUTO_ALL_MODES;
1188                 hw->disable_polarity_correction = false;
1189                 hw->master_slave = E1000_MASTER_SLAVE;
1190         }
1191
1192         adapter->num_tx_queues = 1;
1193         adapter->num_rx_queues = 1;
1194
1195         if (e1000_alloc_queues(adapter)) {
1196                 DPRINTK(PROBE, ERR, "Unable to allocate memory for queues\n");
1197                 return -ENOMEM;
1198         }
1199
1200         /* Explicitly disable IRQ since the NIC can be in any state. */
1201         e1000_irq_disable(adapter);
1202
1203         spin_lock_init(&adapter->stats_lock);
1204
1205         set_bit(__E1000_DOWN, &adapter->flags);
1206
1207         return 0;
1208 }
1209
1210 /**
1211  * e1000_alloc_queues - Allocate memory for all rings
1212  * @adapter: board private structure to initialize
1213  *
1214  * We allocate one ring per queue at run-time since we don't know the
1215  * number of queues at compile-time.
1216  **/
1217
1218 static int __devinit e1000_alloc_queues(struct e1000_adapter *adapter)
1219 {
1220         adapter->tx_ring = kcalloc(adapter->num_tx_queues,
1221                                    sizeof(struct e1000_tx_ring), GFP_KERNEL);
1222         if (!adapter->tx_ring)
1223                 return -ENOMEM;
1224
1225         adapter->rx_ring = kcalloc(adapter->num_rx_queues,
1226                                    sizeof(struct e1000_rx_ring), GFP_KERNEL);
1227         if (!adapter->rx_ring) {
1228                 kfree(adapter->tx_ring);
1229                 return -ENOMEM;
1230         }
1231
1232         return E1000_SUCCESS;
1233 }
1234
1235 /**
1236  * e1000_open - Called when a network interface is made active
1237  * @netdev: network interface device structure
1238  *
1239  * Returns 0 on success, negative value on failure
1240  *
1241  * The open entry point is called when a network interface is made
1242  * active by the system (IFF_UP).  At this point all resources needed
1243  * for transmit and receive operations are allocated, the interrupt
1244  * handler is registered with the OS, the watchdog timer is started,
1245  * and the stack is notified that the interface is ready.
1246  **/
1247
1248 static int e1000_open(struct net_device *netdev)
1249 {
1250         struct e1000_adapter *adapter = netdev_priv(netdev);
1251         struct e1000_hw *hw = &adapter->hw;
1252         int err;
1253
1254         /* disallow open during test */
1255         if (test_bit(__E1000_TESTING, &adapter->flags))
1256                 return -EBUSY;
1257
1258         netif_carrier_off(netdev);
1259
1260         /* allocate transmit descriptors */
1261         err = e1000_setup_all_tx_resources(adapter);
1262         if (err)
1263                 goto err_setup_tx;
1264
1265         /* allocate receive descriptors */
1266         err = e1000_setup_all_rx_resources(adapter);
1267         if (err)
1268                 goto err_setup_rx;
1269
1270         e1000_power_up_phy(adapter);
1271
1272         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
1273         if ((hw->mng_cookie.status &
1274                           E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT)) {
1275                 e1000_update_mng_vlan(adapter);
1276         }
1277
1278         /* before we allocate an interrupt, we must be ready to handle it.
1279          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
1280          * as soon as we call pci_request_irq, so we have to setup our
1281          * clean_rx handler before we do so.  */
1282         e1000_configure(adapter);
1283
1284         err = e1000_request_irq(adapter);
1285         if (err)
1286                 goto err_req_irq;
1287
1288         /* From here on the code is the same as e1000_up() */
1289         clear_bit(__E1000_DOWN, &adapter->flags);
1290
1291         napi_enable(&adapter->napi);
1292
1293         e1000_irq_enable(adapter);
1294
1295         netif_start_queue(netdev);
1296
1297         /* fire a link status change interrupt to start the watchdog */
1298         ew32(ICS, E1000_ICS_LSC);
1299
1300         return E1000_SUCCESS;
1301
1302 err_req_irq:
1303         e1000_power_down_phy(adapter);
1304         e1000_free_all_rx_resources(adapter);
1305 err_setup_rx:
1306         e1000_free_all_tx_resources(adapter);
1307 err_setup_tx:
1308         e1000_reset(adapter);
1309
1310         return err;
1311 }
1312
1313 /**
1314  * e1000_close - Disables a network interface
1315  * @netdev: network interface device structure
1316  *
1317  * Returns 0, this is not allowed to fail
1318  *
1319  * The close entry point is called when an interface is de-activated
1320  * by the OS.  The hardware is still under the drivers control, but
1321  * needs to be disabled.  A global MAC reset is issued to stop the
1322  * hardware, and all transmit and receive resources are freed.
1323  **/
1324
1325 static int e1000_close(struct net_device *netdev)
1326 {
1327         struct e1000_adapter *adapter = netdev_priv(netdev);
1328         struct e1000_hw *hw = &adapter->hw;
1329
1330         WARN_ON(test_bit(__E1000_RESETTING, &adapter->flags));
1331         e1000_down(adapter);
1332         e1000_power_down_phy(adapter);
1333         e1000_free_irq(adapter);
1334
1335         e1000_free_all_tx_resources(adapter);
1336         e1000_free_all_rx_resources(adapter);
1337
1338         /* kill manageability vlan ID if supported, but not if a vlan with
1339          * the same ID is registered on the host OS (let 8021q kill it) */
1340         if ((hw->mng_cookie.status &
1341                           E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT) &&
1342              !(adapter->vlgrp &&
1343                vlan_group_get_device(adapter->vlgrp, adapter->mng_vlan_id))) {
1344                 e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
1345         }
1346
1347         return 0;
1348 }
1349
1350 /**
1351  * e1000_check_64k_bound - check that memory doesn't cross 64kB boundary
1352  * @adapter: address of board private structure
1353  * @start: address of beginning of memory
1354  * @len: length of memory
1355  **/
1356 static bool e1000_check_64k_bound(struct e1000_adapter *adapter, void *start,
1357                                   unsigned long len)
1358 {
1359         struct e1000_hw *hw = &adapter->hw;
1360         unsigned long begin = (unsigned long)start;
1361         unsigned long end = begin + len;
1362
1363         /* First rev 82545 and 82546 need to not allow any memory
1364          * write location to cross 64k boundary due to errata 23 */
1365         if (hw->mac_type == e1000_82545 ||
1366             hw->mac_type == e1000_82546) {
1367                 return ((begin ^ (end - 1)) >> 16) != 0 ? false : true;
1368         }
1369
1370         return true;
1371 }
1372
1373 /**
1374  * e1000_setup_tx_resources - allocate Tx resources (Descriptors)
1375  * @adapter: board private structure
1376  * @txdr:    tx descriptor ring (for a specific queue) to setup
1377  *
1378  * Return 0 on success, negative on failure
1379  **/
1380
1381 static int e1000_setup_tx_resources(struct e1000_adapter *adapter,
1382                                     struct e1000_tx_ring *txdr)
1383 {
1384         struct pci_dev *pdev = adapter->pdev;
1385         int size;
1386
1387         size = sizeof(struct e1000_buffer) * txdr->count;
1388         txdr->buffer_info = vmalloc(size);
1389         if (!txdr->buffer_info) {
1390                 DPRINTK(PROBE, ERR,
1391                 "Unable to allocate memory for the transmit descriptor ring\n");
1392                 return -ENOMEM;
1393         }
1394         memset(txdr->buffer_info, 0, size);
1395
1396         /* round up to nearest 4K */
1397
1398         txdr->size = txdr->count * sizeof(struct e1000_tx_desc);
1399         txdr->size = ALIGN(txdr->size, 4096);
1400
1401         txdr->desc = pci_alloc_consistent(pdev, txdr->size, &txdr->dma);
1402         if (!txdr->desc) {
1403 setup_tx_desc_die:
1404                 vfree(txdr->buffer_info);
1405                 DPRINTK(PROBE, ERR,
1406                 "Unable to allocate memory for the transmit descriptor ring\n");
1407                 return -ENOMEM;
1408         }
1409
1410         /* Fix for errata 23, can't cross 64kB boundary */
1411         if (!e1000_check_64k_bound(adapter, txdr->desc, txdr->size)) {
1412                 void *olddesc = txdr->desc;
1413                 dma_addr_t olddma = txdr->dma;
1414                 DPRINTK(TX_ERR, ERR, "txdr align check failed: %u bytes "
1415                                      "at %p\n", txdr->size, txdr->desc);
1416                 /* Try again, without freeing the previous */
1417                 txdr->desc = pci_alloc_consistent(pdev, txdr->size, &txdr->dma);
1418                 /* Failed allocation, critical failure */
1419                 if (!txdr->desc) {
1420                         pci_free_consistent(pdev, txdr->size, olddesc, olddma);
1421                         goto setup_tx_desc_die;
1422                 }
1423
1424                 if (!e1000_check_64k_bound(adapter, txdr->desc, txdr->size)) {
1425                         /* give up */
1426                         pci_free_consistent(pdev, txdr->size, txdr->desc,
1427                                             txdr->dma);
1428                         pci_free_consistent(pdev, txdr->size, olddesc, olddma);
1429                         DPRINTK(PROBE, ERR,
1430                                 "Unable to allocate aligned memory "
1431                                 "for the transmit descriptor ring\n");
1432                         vfree(txdr->buffer_info);
1433                         return -ENOMEM;
1434                 } else {
1435                         /* Free old allocation, new allocation was successful */
1436                         pci_free_consistent(pdev, txdr->size, olddesc, olddma);
1437                 }
1438         }
1439         memset(txdr->desc, 0, txdr->size);
1440
1441         txdr->next_to_use = 0;
1442         txdr->next_to_clean = 0;
1443
1444         return 0;
1445 }
1446
1447 /**
1448  * e1000_setup_all_tx_resources - wrapper to allocate Tx resources
1449  *                                (Descriptors) for all queues
1450  * @adapter: board private structure
1451  *
1452  * Return 0 on success, negative on failure
1453  **/
1454
1455 int e1000_setup_all_tx_resources(struct e1000_adapter *adapter)
1456 {
1457         int i, err = 0;
1458
1459         for (i = 0; i < adapter->num_tx_queues; i++) {
1460                 err = e1000_setup_tx_resources(adapter, &adapter->tx_ring[i]);
1461                 if (err) {
1462                         DPRINTK(PROBE, ERR,
1463                                 "Allocation for Tx Queue %u failed\n", i);
1464                         for (i-- ; i >= 0; i--)
1465                                 e1000_free_tx_resources(adapter,
1466                                                         &adapter->tx_ring[i]);
1467                         break;
1468                 }
1469         }
1470
1471         return err;
1472 }
1473
1474 /**
1475  * e1000_configure_tx - Configure 8254x Transmit Unit after Reset
1476  * @adapter: board private structure
1477  *
1478  * Configure the Tx unit of the MAC after a reset.
1479  **/
1480
1481 static void e1000_configure_tx(struct e1000_adapter *adapter)
1482 {
1483         u64 tdba;
1484         struct e1000_hw *hw = &adapter->hw;
1485         u32 tdlen, tctl, tipg;
1486         u32 ipgr1, ipgr2;
1487
1488         /* Setup the HW Tx Head and Tail descriptor pointers */
1489
1490         switch (adapter->num_tx_queues) {
1491         case 1:
1492         default:
1493                 tdba = adapter->tx_ring[0].dma;
1494                 tdlen = adapter->tx_ring[0].count *
1495                         sizeof(struct e1000_tx_desc);
1496                 ew32(TDLEN, tdlen);
1497                 ew32(TDBAH, (tdba >> 32));
1498                 ew32(TDBAL, (tdba & 0x00000000ffffffffULL));
1499                 ew32(TDT, 0);
1500                 ew32(TDH, 0);
1501                 adapter->tx_ring[0].tdh = ((hw->mac_type >= e1000_82543) ? E1000_TDH : E1000_82542_TDH);
1502                 adapter->tx_ring[0].tdt = ((hw->mac_type >= e1000_82543) ? E1000_TDT : E1000_82542_TDT);
1503                 break;
1504         }
1505
1506         /* Set the default values for the Tx Inter Packet Gap timer */
1507         if ((hw->media_type == e1000_media_type_fiber ||
1508              hw->media_type == e1000_media_type_internal_serdes))
1509                 tipg = DEFAULT_82543_TIPG_IPGT_FIBER;
1510         else
1511                 tipg = DEFAULT_82543_TIPG_IPGT_COPPER;
1512
1513         switch (hw->mac_type) {
1514         case e1000_82542_rev2_0:
1515         case e1000_82542_rev2_1:
1516                 tipg = DEFAULT_82542_TIPG_IPGT;
1517                 ipgr1 = DEFAULT_82542_TIPG_IPGR1;
1518                 ipgr2 = DEFAULT_82542_TIPG_IPGR2;
1519                 break;
1520         default:
1521                 ipgr1 = DEFAULT_82543_TIPG_IPGR1;
1522                 ipgr2 = DEFAULT_82543_TIPG_IPGR2;
1523                 break;
1524         }
1525         tipg |= ipgr1 << E1000_TIPG_IPGR1_SHIFT;
1526         tipg |= ipgr2 << E1000_TIPG_IPGR2_SHIFT;
1527         ew32(TIPG, tipg);
1528
1529         /* Set the Tx Interrupt Delay register */
1530
1531         ew32(TIDV, adapter->tx_int_delay);
1532         if (hw->mac_type >= e1000_82540)
1533                 ew32(TADV, adapter->tx_abs_int_delay);
1534
1535         /* Program the Transmit Control Register */
1536
1537         tctl = er32(TCTL);
1538         tctl &= ~E1000_TCTL_CT;
1539         tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
1540                 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
1541
1542         e1000_config_collision_dist(hw);
1543
1544         /* Setup Transmit Descriptor Settings for eop descriptor */
1545         adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
1546
1547         /* only set IDE if we are delaying interrupts using the timers */
1548         if (adapter->tx_int_delay)
1549                 adapter->txd_cmd |= E1000_TXD_CMD_IDE;
1550
1551         if (hw->mac_type < e1000_82543)
1552                 adapter->txd_cmd |= E1000_TXD_CMD_RPS;
1553         else
1554                 adapter->txd_cmd |= E1000_TXD_CMD_RS;
1555
1556         /* Cache if we're 82544 running in PCI-X because we'll
1557          * need this to apply a workaround later in the send path. */
1558         if (hw->mac_type == e1000_82544 &&
1559             hw->bus_type == e1000_bus_type_pcix)
1560                 adapter->pcix_82544 = 1;
1561
1562         ew32(TCTL, tctl);
1563
1564 }
1565
1566 /**
1567  * e1000_setup_rx_resources - allocate Rx resources (Descriptors)
1568  * @adapter: board private structure
1569  * @rxdr:    rx descriptor ring (for a specific queue) to setup
1570  *
1571  * Returns 0 on success, negative on failure
1572  **/
1573
1574 static int e1000_setup_rx_resources(struct e1000_adapter *adapter,
1575                                     struct e1000_rx_ring *rxdr)
1576 {
1577         struct pci_dev *pdev = adapter->pdev;
1578         int size, desc_len;
1579
1580         size = sizeof(struct e1000_buffer) * rxdr->count;
1581         rxdr->buffer_info = vmalloc(size);
1582         if (!rxdr->buffer_info) {
1583                 DPRINTK(PROBE, ERR,
1584                 "Unable to allocate memory for the receive descriptor ring\n");
1585                 return -ENOMEM;
1586         }
1587         memset(rxdr->buffer_info, 0, size);
1588
1589         desc_len = sizeof(struct e1000_rx_desc);
1590
1591         /* Round up to nearest 4K */
1592
1593         rxdr->size = rxdr->count * desc_len;
1594         rxdr->size = ALIGN(rxdr->size, 4096);
1595
1596         rxdr->desc = pci_alloc_consistent(pdev, rxdr->size, &rxdr->dma);
1597
1598         if (!rxdr->desc) {
1599                 DPRINTK(PROBE, ERR,
1600                 "Unable to allocate memory for the receive descriptor ring\n");
1601 setup_rx_desc_die:
1602                 vfree(rxdr->buffer_info);
1603                 return -ENOMEM;
1604         }
1605
1606         /* Fix for errata 23, can't cross 64kB boundary */
1607         if (!e1000_check_64k_bound(adapter, rxdr->desc, rxdr->size)) {
1608                 void *olddesc = rxdr->desc;
1609                 dma_addr_t olddma = rxdr->dma;
1610                 DPRINTK(RX_ERR, ERR, "rxdr align check failed: %u bytes "
1611                                      "at %p\n", rxdr->size, rxdr->desc);
1612                 /* Try again, without freeing the previous */
1613                 rxdr->desc = pci_alloc_consistent(pdev, rxdr->size, &rxdr->dma);
1614                 /* Failed allocation, critical failure */
1615                 if (!rxdr->desc) {
1616                         pci_free_consistent(pdev, rxdr->size, olddesc, olddma);
1617                         DPRINTK(PROBE, ERR,
1618                                 "Unable to allocate memory "
1619                                 "for the receive descriptor ring\n");
1620                         goto setup_rx_desc_die;
1621                 }
1622
1623                 if (!e1000_check_64k_bound(adapter, rxdr->desc, rxdr->size)) {
1624                         /* give up */
1625                         pci_free_consistent(pdev, rxdr->size, rxdr->desc,
1626                                             rxdr->dma);
1627                         pci_free_consistent(pdev, rxdr->size, olddesc, olddma);
1628                         DPRINTK(PROBE, ERR,
1629                                 "Unable to allocate aligned memory "
1630                                 "for the receive descriptor ring\n");
1631                         goto setup_rx_desc_die;
1632                 } else {
1633                         /* Free old allocation, new allocation was successful */
1634                         pci_free_consistent(pdev, rxdr->size, olddesc, olddma);
1635                 }
1636         }
1637         memset(rxdr->desc, 0, rxdr->size);
1638
1639         rxdr->next_to_clean = 0;
1640         rxdr->next_to_use = 0;
1641         rxdr->rx_skb_top = NULL;
1642
1643         return 0;
1644 }
1645
1646 /**
1647  * e1000_setup_all_rx_resources - wrapper to allocate Rx resources
1648  *                                (Descriptors) for all queues
1649  * @adapter: board private structure
1650  *
1651  * Return 0 on success, negative on failure
1652  **/
1653
1654 int e1000_setup_all_rx_resources(struct e1000_adapter *adapter)
1655 {
1656         int i, err = 0;
1657
1658         for (i = 0; i < adapter->num_rx_queues; i++) {
1659                 err = e1000_setup_rx_resources(adapter, &adapter->rx_ring[i]);
1660                 if (err) {
1661                         DPRINTK(PROBE, ERR,
1662                                 "Allocation for Rx Queue %u failed\n", i);
1663                         for (i-- ; i >= 0; i--)
1664                                 e1000_free_rx_resources(adapter,
1665                                                         &adapter->rx_ring[i]);
1666                         break;
1667                 }
1668         }
1669
1670         return err;
1671 }
1672
1673 /**
1674  * e1000_setup_rctl - configure the receive control registers
1675  * @adapter: Board private structure
1676  **/
1677 static void e1000_setup_rctl(struct e1000_adapter *adapter)
1678 {
1679         struct e1000_hw *hw = &adapter->hw;
1680         u32 rctl;
1681
1682         rctl = er32(RCTL);
1683
1684         rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
1685
1686         rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
1687                 E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
1688                 (hw->mc_filter_type << E1000_RCTL_MO_SHIFT);
1689
1690         if (hw->tbi_compatibility_on == 1)
1691                 rctl |= E1000_RCTL_SBP;
1692         else
1693                 rctl &= ~E1000_RCTL_SBP;
1694
1695         if (adapter->netdev->mtu <= ETH_DATA_LEN)
1696                 rctl &= ~E1000_RCTL_LPE;
1697         else
1698                 rctl |= E1000_RCTL_LPE;
1699
1700         /* Setup buffer sizes */
1701         rctl &= ~E1000_RCTL_SZ_4096;
1702         rctl |= E1000_RCTL_BSEX;
1703         switch (adapter->rx_buffer_len) {
1704                 case E1000_RXBUFFER_2048:
1705                 default:
1706                         rctl |= E1000_RCTL_SZ_2048;
1707                         rctl &= ~E1000_RCTL_BSEX;
1708                         break;
1709                 case E1000_RXBUFFER_4096:
1710                         rctl |= E1000_RCTL_SZ_4096;
1711                         break;
1712                 case E1000_RXBUFFER_8192:
1713                         rctl |= E1000_RCTL_SZ_8192;
1714                         break;
1715                 case E1000_RXBUFFER_16384:
1716                         rctl |= E1000_RCTL_SZ_16384;
1717                         break;
1718         }
1719
1720         ew32(RCTL, rctl);
1721 }
1722
1723 /**
1724  * e1000_configure_rx - Configure 8254x Receive Unit after Reset
1725  * @adapter: board private structure
1726  *
1727  * Configure the Rx unit of the MAC after a reset.
1728  **/
1729
1730 static void e1000_configure_rx(struct e1000_adapter *adapter)
1731 {
1732         u64 rdba;
1733         struct e1000_hw *hw = &adapter->hw;
1734         u32 rdlen, rctl, rxcsum;
1735
1736         if (adapter->netdev->mtu > ETH_DATA_LEN) {
1737                 rdlen = adapter->rx_ring[0].count *
1738                         sizeof(struct e1000_rx_desc);
1739                 adapter->clean_rx = e1000_clean_jumbo_rx_irq;
1740                 adapter->alloc_rx_buf = e1000_alloc_jumbo_rx_buffers;
1741         } else {
1742                 rdlen = adapter->rx_ring[0].count *
1743                         sizeof(struct e1000_rx_desc);
1744                 adapter->clean_rx = e1000_clean_rx_irq;
1745                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
1746         }
1747
1748         /* disable receives while setting up the descriptors */
1749         rctl = er32(RCTL);
1750         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1751
1752         /* set the Receive Delay Timer Register */
1753         ew32(RDTR, adapter->rx_int_delay);
1754
1755         if (hw->mac_type >= e1000_82540) {
1756                 ew32(RADV, adapter->rx_abs_int_delay);
1757                 if (adapter->itr_setting != 0)
1758                         ew32(ITR, 1000000000 / (adapter->itr * 256));
1759         }
1760
1761         /* Setup the HW Rx Head and Tail Descriptor Pointers and
1762          * the Base and Length of the Rx Descriptor Ring */
1763         switch (adapter->num_rx_queues) {
1764         case 1:
1765         default:
1766                 rdba = adapter->rx_ring[0].dma;
1767                 ew32(RDLEN, rdlen);
1768                 ew32(RDBAH, (rdba >> 32));
1769                 ew32(RDBAL, (rdba & 0x00000000ffffffffULL));
1770                 ew32(RDT, 0);
1771                 ew32(RDH, 0);
1772                 adapter->rx_ring[0].rdh = ((hw->mac_type >= e1000_82543) ? E1000_RDH : E1000_82542_RDH);
1773                 adapter->rx_ring[0].rdt = ((hw->mac_type >= e1000_82543) ? E1000_RDT : E1000_82542_RDT);
1774                 break;
1775         }
1776
1777         /* Enable 82543 Receive Checksum Offload for TCP and UDP */
1778         if (hw->mac_type >= e1000_82543) {
1779                 rxcsum = er32(RXCSUM);
1780                 if (adapter->rx_csum)
1781                         rxcsum |= E1000_RXCSUM_TUOFL;
1782                 else
1783                         /* don't need to clear IPPCSE as it defaults to 0 */
1784                         rxcsum &= ~E1000_RXCSUM_TUOFL;
1785                 ew32(RXCSUM, rxcsum);
1786         }
1787
1788         /* Enable Receives */
1789         ew32(RCTL, rctl);
1790 }
1791
1792 /**
1793  * e1000_free_tx_resources - Free Tx Resources per Queue
1794  * @adapter: board private structure
1795  * @tx_ring: Tx descriptor ring for a specific queue
1796  *
1797  * Free all transmit software resources
1798  **/
1799
1800 static void e1000_free_tx_resources(struct e1000_adapter *adapter,
1801                                     struct e1000_tx_ring *tx_ring)
1802 {
1803         struct pci_dev *pdev = adapter->pdev;
1804
1805         e1000_clean_tx_ring(adapter, tx_ring);
1806
1807         vfree(tx_ring->buffer_info);
1808         tx_ring->buffer_info = NULL;
1809
1810         pci_free_consistent(pdev, tx_ring->size, tx_ring->desc, tx_ring->dma);
1811
1812         tx_ring->desc = NULL;
1813 }
1814
1815 /**
1816  * e1000_free_all_tx_resources - Free Tx Resources for All Queues
1817  * @adapter: board private structure
1818  *
1819  * Free all transmit software resources
1820  **/
1821
1822 void e1000_free_all_tx_resources(struct e1000_adapter *adapter)
1823 {
1824         int i;
1825
1826         for (i = 0; i < adapter->num_tx_queues; i++)
1827                 e1000_free_tx_resources(adapter, &adapter->tx_ring[i]);
1828 }
1829
1830 static void e1000_unmap_and_free_tx_resource(struct e1000_adapter *adapter,
1831                                              struct e1000_buffer *buffer_info)
1832 {
1833         if (buffer_info->dma) {
1834                 if (buffer_info->mapped_as_page)
1835                         pci_unmap_page(adapter->pdev, buffer_info->dma,
1836                                        buffer_info->length, PCI_DMA_TODEVICE);
1837                 else
1838                         pci_unmap_single(adapter->pdev, buffer_info->dma,
1839                                          buffer_info->length,
1840                                          PCI_DMA_TODEVICE);
1841                 buffer_info->dma = 0;
1842         }
1843         if (buffer_info->skb) {
1844                 dev_kfree_skb_any(buffer_info->skb);
1845                 buffer_info->skb = NULL;
1846         }
1847         buffer_info->time_stamp = 0;
1848         /* buffer_info must be completely set up in the transmit path */
1849 }
1850
1851 /**
1852  * e1000_clean_tx_ring - Free Tx Buffers
1853  * @adapter: board private structure
1854  * @tx_ring: ring to be cleaned
1855  **/
1856
1857 static void e1000_clean_tx_ring(struct e1000_adapter *adapter,
1858                                 struct e1000_tx_ring *tx_ring)
1859 {
1860         struct e1000_hw *hw = &adapter->hw;
1861         struct e1000_buffer *buffer_info;
1862         unsigned long size;
1863         unsigned int i;
1864
1865         /* Free all the Tx ring sk_buffs */
1866
1867         for (i = 0; i < tx_ring->count; i++) {
1868                 buffer_info = &tx_ring->buffer_info[i];
1869                 e1000_unmap_and_free_tx_resource(adapter, buffer_info);
1870         }
1871
1872         size = sizeof(struct e1000_buffer) * tx_ring->count;
1873         memset(tx_ring->buffer_info, 0, size);
1874
1875         /* Zero out the descriptor ring */
1876
1877         memset(tx_ring->desc, 0, tx_ring->size);
1878
1879         tx_ring->next_to_use = 0;
1880         tx_ring->next_to_clean = 0;
1881         tx_ring->last_tx_tso = 0;
1882
1883         writel(0, hw->hw_addr + tx_ring->tdh);
1884         writel(0, hw->hw_addr + tx_ring->tdt);
1885 }
1886
1887 /**
1888  * e1000_clean_all_tx_rings - Free Tx Buffers for all queues
1889  * @adapter: board private structure
1890  **/
1891
1892 static void e1000_clean_all_tx_rings(struct e1000_adapter *adapter)
1893 {
1894         int i;
1895
1896         for (i = 0; i < adapter->num_tx_queues; i++)
1897                 e1000_clean_tx_ring(adapter, &adapter->tx_ring[i]);
1898 }
1899
1900 /**
1901  * e1000_free_rx_resources - Free Rx Resources
1902  * @adapter: board private structure
1903  * @rx_ring: ring to clean the resources from
1904  *
1905  * Free all receive software resources
1906  **/
1907
1908 static void e1000_free_rx_resources(struct e1000_adapter *adapter,
1909                                     struct e1000_rx_ring *rx_ring)
1910 {
1911         struct pci_dev *pdev = adapter->pdev;
1912
1913         e1000_clean_rx_ring(adapter, rx_ring);
1914
1915         vfree(rx_ring->buffer_info);
1916         rx_ring->buffer_info = NULL;
1917
1918         pci_free_consistent(pdev, rx_ring->size, rx_ring->desc, rx_ring->dma);
1919
1920         rx_ring->desc = NULL;
1921 }
1922
1923 /**
1924  * e1000_free_all_rx_resources - Free Rx Resources for All Queues
1925  * @adapter: board private structure
1926  *
1927  * Free all receive software resources
1928  **/
1929
1930 void e1000_free_all_rx_resources(struct e1000_adapter *adapter)
1931 {
1932         int i;
1933
1934         for (i = 0; i < adapter->num_rx_queues; i++)
1935                 e1000_free_rx_resources(adapter, &adapter->rx_ring[i]);
1936 }
1937
1938 /**
1939  * e1000_clean_rx_ring - Free Rx Buffers per Queue
1940  * @adapter: board private structure
1941  * @rx_ring: ring to free buffers from
1942  **/
1943
1944 static void e1000_clean_rx_ring(struct e1000_adapter *adapter,
1945                                 struct e1000_rx_ring *rx_ring)
1946 {
1947         struct e1000_hw *hw = &adapter->hw;
1948         struct e1000_buffer *buffer_info;
1949         struct pci_dev *pdev = adapter->pdev;
1950         unsigned long size;
1951         unsigned int i;
1952
1953         /* Free all the Rx ring sk_buffs */
1954         for (i = 0; i < rx_ring->count; i++) {
1955                 buffer_info = &rx_ring->buffer_info[i];
1956                 if (buffer_info->dma &&
1957                     adapter->clean_rx == e1000_clean_rx_irq) {
1958                         pci_unmap_single(pdev, buffer_info->dma,
1959                                          buffer_info->length,
1960                                          PCI_DMA_FROMDEVICE);
1961                 } else if (buffer_info->dma &&
1962                            adapter->clean_rx == e1000_clean_jumbo_rx_irq) {
1963                         pci_unmap_page(pdev, buffer_info->dma,
1964                                        buffer_info->length,
1965                                        PCI_DMA_FROMDEVICE);
1966                 }
1967
1968                 buffer_info->dma = 0;
1969                 if (buffer_info->page) {
1970                         put_page(buffer_info->page);
1971                         buffer_info->page = NULL;
1972                 }
1973                 if (buffer_info->skb) {
1974                         dev_kfree_skb(buffer_info->skb);
1975                         buffer_info->skb = NULL;
1976                 }
1977         }
1978
1979         /* there also may be some cached data from a chained receive */
1980         if (rx_ring->rx_skb_top) {
1981                 dev_kfree_skb(rx_ring->rx_skb_top);
1982                 rx_ring->rx_skb_top = NULL;
1983         }
1984
1985         size = sizeof(struct e1000_buffer) * rx_ring->count;
1986         memset(rx_ring->buffer_info, 0, size);
1987
1988         /* Zero out the descriptor ring */
1989         memset(rx_ring->desc, 0, rx_ring->size);
1990
1991         rx_ring->next_to_clean = 0;
1992         rx_ring->next_to_use = 0;
1993
1994         writel(0, hw->hw_addr + rx_ring->rdh);
1995         writel(0, hw->hw_addr + rx_ring->rdt);
1996 }
1997
1998 /**
1999  * e1000_clean_all_rx_rings - Free Rx Buffers for all queues
2000  * @adapter: board private structure
2001  **/
2002
2003 static void e1000_clean_all_rx_rings(struct e1000_adapter *adapter)
2004 {
2005         int i;
2006
2007         for (i = 0; i < adapter->num_rx_queues; i++)
2008                 e1000_clean_rx_ring(adapter, &adapter->rx_ring[i]);
2009 }
2010
2011 /* The 82542 2.0 (revision 2) needs to have the receive unit in reset
2012  * and memory write and invalidate disabled for certain operations
2013  */
2014 static void e1000_enter_82542_rst(struct e1000_adapter *adapter)
2015 {
2016         struct e1000_hw *hw = &adapter->hw;
2017         struct net_device *netdev = adapter->netdev;
2018         u32 rctl;
2019
2020         e1000_pci_clear_mwi(hw);
2021
2022         rctl = er32(RCTL);
2023         rctl |= E1000_RCTL_RST;
2024         ew32(RCTL, rctl);
2025         E1000_WRITE_FLUSH();
2026         mdelay(5);
2027
2028         if (netif_running(netdev))
2029                 e1000_clean_all_rx_rings(adapter);
2030 }
2031
2032 static void e1000_leave_82542_rst(struct e1000_adapter *adapter)
2033 {
2034         struct e1000_hw *hw = &adapter->hw;
2035         struct net_device *netdev = adapter->netdev;
2036         u32 rctl;
2037
2038         rctl = er32(RCTL);
2039         rctl &= ~E1000_RCTL_RST;
2040         ew32(RCTL, rctl);
2041         E1000_WRITE_FLUSH();
2042         mdelay(5);
2043
2044         if (hw->pci_cmd_word & PCI_COMMAND_INVALIDATE)
2045                 e1000_pci_set_mwi(hw);
2046
2047         if (netif_running(netdev)) {
2048                 /* No need to loop, because 82542 supports only 1 queue */
2049                 struct e1000_rx_ring *ring = &adapter->rx_ring[0];
2050                 e1000_configure_rx(adapter);
2051                 adapter->alloc_rx_buf(adapter, ring, E1000_DESC_UNUSED(ring));
2052         }
2053 }
2054
2055 /**
2056  * e1000_set_mac - Change the Ethernet Address of the NIC
2057  * @netdev: network interface device structure
2058  * @p: pointer to an address structure
2059  *
2060  * Returns 0 on success, negative on failure
2061  **/
2062
2063 static int e1000_set_mac(struct net_device *netdev, void *p)
2064 {
2065         struct e1000_adapter *adapter = netdev_priv(netdev);
2066         struct e1000_hw *hw = &adapter->hw;
2067         struct sockaddr *addr = p;
2068
2069         if (!is_valid_ether_addr(addr->sa_data))
2070                 return -EADDRNOTAVAIL;
2071
2072         /* 82542 2.0 needs to be in reset to write receive address registers */
2073
2074         if (hw->mac_type == e1000_82542_rev2_0)
2075                 e1000_enter_82542_rst(adapter);
2076
2077         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
2078         memcpy(hw->mac_addr, addr->sa_data, netdev->addr_len);
2079
2080         e1000_rar_set(hw, hw->mac_addr, 0);
2081
2082         if (hw->mac_type == e1000_82542_rev2_0)
2083                 e1000_leave_82542_rst(adapter);
2084
2085         return 0;
2086 }
2087
2088 /**
2089  * e1000_set_rx_mode - Secondary Unicast, Multicast and Promiscuous mode set
2090  * @netdev: network interface device structure
2091  *
2092  * The set_rx_mode entry point is called whenever the unicast or multicast
2093  * address lists or the network interface flags are updated. This routine is
2094  * responsible for configuring the hardware for proper unicast, multicast,
2095  * promiscuous mode, and all-multi behavior.
2096  **/
2097
2098 static void e1000_set_rx_mode(struct net_device *netdev)
2099 {
2100         struct e1000_adapter *adapter = netdev_priv(netdev);
2101         struct e1000_hw *hw = &adapter->hw;
2102         struct netdev_hw_addr *ha;
2103         bool use_uc = false;
2104         struct dev_addr_list *mc_ptr;
2105         u32 rctl;
2106         u32 hash_value;
2107         int i, rar_entries = E1000_RAR_ENTRIES;
2108         int mta_reg_count = E1000_NUM_MTA_REGISTERS;
2109         u32 *mcarray = kcalloc(mta_reg_count, sizeof(u32), GFP_ATOMIC);
2110
2111         if (!mcarray) {
2112                 DPRINTK(PROBE, ERR, "memory allocation failed\n");
2113                 return;
2114         }
2115
2116         /* Check for Promiscuous and All Multicast modes */
2117
2118         rctl = er32(RCTL);
2119
2120         if (netdev->flags & IFF_PROMISC) {
2121                 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
2122                 rctl &= ~E1000_RCTL_VFE;
2123         } else {
2124                 if (netdev->flags & IFF_ALLMULTI)
2125                         rctl |= E1000_RCTL_MPE;
2126                 else
2127                         rctl &= ~E1000_RCTL_MPE;
2128                 /* Enable VLAN filter if there is a VLAN */
2129                 if (adapter->vlgrp)
2130                         rctl |= E1000_RCTL_VFE;
2131         }
2132
2133         if (netdev_uc_count(netdev) > rar_entries - 1) {
2134                 rctl |= E1000_RCTL_UPE;
2135         } else if (!(netdev->flags & IFF_PROMISC)) {
2136                 rctl &= ~E1000_RCTL_UPE;
2137                 use_uc = true;
2138         }
2139
2140         ew32(RCTL, rctl);
2141
2142         /* 82542 2.0 needs to be in reset to write receive address registers */
2143
2144         if (hw->mac_type == e1000_82542_rev2_0)
2145                 e1000_enter_82542_rst(adapter);
2146
2147         /* load the first 14 addresses into the exact filters 1-14. Unicast
2148          * addresses take precedence to avoid disabling unicast filtering
2149          * when possible.
2150          *
2151          * RAR 0 is used for the station MAC adddress
2152          * if there are not 14 addresses, go ahead and clear the filters
2153          */
2154         i = 1;
2155         if (use_uc)
2156                 netdev_for_each_uc_addr(ha, netdev) {
2157                         if (i == rar_entries)
2158                                 break;
2159                         e1000_rar_set(hw, ha->addr, i++);
2160                 }
2161
2162         WARN_ON(i == rar_entries);
2163
2164         mc_ptr = netdev->mc_list;
2165
2166         for (; i < rar_entries; i++) {
2167                 if (mc_ptr) {
2168                         e1000_rar_set(hw, mc_ptr->da_addr, i);
2169                         mc_ptr = mc_ptr->next;
2170                 } else {
2171                         E1000_WRITE_REG_ARRAY(hw, RA, i << 1, 0);
2172                         E1000_WRITE_FLUSH();
2173                         E1000_WRITE_REG_ARRAY(hw, RA, (i << 1) + 1, 0);
2174                         E1000_WRITE_FLUSH();
2175                 }
2176         }
2177
2178         /* load any remaining addresses into the hash table */
2179
2180         for (; mc_ptr; mc_ptr = mc_ptr->next) {
2181                 u32 hash_reg, hash_bit, mta;
2182                 hash_value = e1000_hash_mc_addr(hw, mc_ptr->da_addr);
2183                 hash_reg = (hash_value >> 5) & 0x7F;
2184                 hash_bit = hash_value & 0x1F;
2185                 mta = (1 << hash_bit);
2186                 mcarray[hash_reg] |= mta;
2187         }
2188
2189         /* write the hash table completely, write from bottom to avoid
2190          * both stupid write combining chipsets, and flushing each write */
2191         for (i = mta_reg_count - 1; i >= 0 ; i--) {
2192                 /*
2193                  * If we are on an 82544 has an errata where writing odd
2194                  * offsets overwrites the previous even offset, but writing
2195                  * backwards over the range solves the issue by always
2196                  * writing the odd offset first
2197                  */
2198                 E1000_WRITE_REG_ARRAY(hw, MTA, i, mcarray[i]);
2199         }
2200         E1000_WRITE_FLUSH();
2201
2202         if (hw->mac_type == e1000_82542_rev2_0)
2203                 e1000_leave_82542_rst(adapter);
2204
2205         kfree(mcarray);
2206 }
2207
2208 /* Need to wait a few seconds after link up to get diagnostic information from
2209  * the phy */
2210
2211 static void e1000_update_phy_info(unsigned long data)
2212 {
2213         struct e1000_adapter *adapter = (struct e1000_adapter *)data;
2214         struct e1000_hw *hw = &adapter->hw;
2215         e1000_phy_get_info(hw, &adapter->phy_info);
2216 }
2217
2218 /**
2219  * e1000_82547_tx_fifo_stall - Timer Call-back
2220  * @data: pointer to adapter cast into an unsigned long
2221  **/
2222
2223 static void e1000_82547_tx_fifo_stall(unsigned long data)
2224 {
2225         struct e1000_adapter *adapter = (struct e1000_adapter *)data;
2226         struct e1000_hw *hw = &adapter->hw;
2227         struct net_device *netdev = adapter->netdev;
2228         u32 tctl;
2229
2230         if (atomic_read(&adapter->tx_fifo_stall)) {
2231                 if ((er32(TDT) == er32(TDH)) &&
2232                    (er32(TDFT) == er32(TDFH)) &&
2233                    (er32(TDFTS) == er32(TDFHS))) {
2234                         tctl = er32(TCTL);
2235                         ew32(TCTL, tctl & ~E1000_TCTL_EN);
2236                         ew32(TDFT, adapter->tx_head_addr);
2237                         ew32(TDFH, adapter->tx_head_addr);
2238                         ew32(TDFTS, adapter->tx_head_addr);
2239                         ew32(TDFHS, adapter->tx_head_addr);
2240                         ew32(TCTL, tctl);
2241                         E1000_WRITE_FLUSH();
2242
2243                         adapter->tx_fifo_head = 0;
2244                         atomic_set(&adapter->tx_fifo_stall, 0);
2245                         netif_wake_queue(netdev);
2246                 } else if (!test_bit(__E1000_DOWN, &adapter->flags)) {
2247                         mod_timer(&adapter->tx_fifo_stall_timer, jiffies + 1);
2248                 }
2249         }
2250 }
2251
2252 bool e1000_has_link(struct e1000_adapter *adapter)
2253 {
2254         struct e1000_hw *hw = &adapter->hw;
2255         bool link_active = false;
2256
2257         /* get_link_status is set on LSC (link status) interrupt or
2258          * rx sequence error interrupt.  get_link_status will stay
2259          * false until the e1000_check_for_link establishes link
2260          * for copper adapters ONLY
2261          */
2262         switch (hw->media_type) {
2263         case e1000_media_type_copper:
2264                 if (hw->get_link_status) {
2265                         e1000_check_for_link(hw);
2266                         link_active = !hw->get_link_status;
2267                 } else {
2268                         link_active = true;
2269                 }
2270                 break;
2271         case e1000_media_type_fiber:
2272                 e1000_check_for_link(hw);
2273                 link_active = !!(er32(STATUS) & E1000_STATUS_LU);
2274                 break;
2275         case e1000_media_type_internal_serdes:
2276                 e1000_check_for_link(hw);
2277                 link_active = hw->serdes_has_link;
2278                 break;
2279         default:
2280                 break;
2281         }
2282
2283         return link_active;
2284 }
2285
2286 /**
2287  * e1000_watchdog - Timer Call-back
2288  * @data: pointer to adapter cast into an unsigned long
2289  **/
2290 static void e1000_watchdog(unsigned long data)
2291 {
2292         struct e1000_adapter *adapter = (struct e1000_adapter *)data;
2293         struct e1000_hw *hw = &adapter->hw;
2294         struct net_device *netdev = adapter->netdev;
2295         struct e1000_tx_ring *txdr = adapter->tx_ring;
2296         u32 link, tctl;
2297
2298         link = e1000_has_link(adapter);
2299         if ((netif_carrier_ok(netdev)) && link)
2300                 goto link_up;
2301
2302         if (link) {
2303                 if (!netif_carrier_ok(netdev)) {
2304                         u32 ctrl;
2305                         bool txb2b = true;
2306                         /* update snapshot of PHY registers on LSC */
2307                         e1000_get_speed_and_duplex(hw,
2308                                                    &adapter->link_speed,
2309                                                    &adapter->link_duplex);
2310
2311                         ctrl = er32(CTRL);
2312                         printk(KERN_INFO "e1000: %s NIC Link is Up %d Mbps %s, "
2313                                "Flow Control: %s\n",
2314                                netdev->name,
2315                                adapter->link_speed,
2316                                adapter->link_duplex == FULL_DUPLEX ?
2317                                 "Full Duplex" : "Half Duplex",
2318                                 ((ctrl & E1000_CTRL_TFCE) && (ctrl &
2319                                 E1000_CTRL_RFCE)) ? "RX/TX" : ((ctrl &
2320                                 E1000_CTRL_RFCE) ? "RX" : ((ctrl &
2321                                 E1000_CTRL_TFCE) ? "TX" : "None" )));
2322
2323                         /* tweak tx_queue_len according to speed/duplex
2324                          * and adjust the timeout factor */
2325                         netdev->tx_queue_len = adapter->tx_queue_len;
2326                         adapter->tx_timeout_factor = 1;
2327                         switch (adapter->link_speed) {
2328                         case SPEED_10:
2329                                 txb2b = false;
2330                                 netdev->tx_queue_len = 10;
2331                                 adapter->tx_timeout_factor = 16;
2332                                 break;
2333                         case SPEED_100:
2334                                 txb2b = false;
2335                                 netdev->tx_queue_len = 100;
2336                                 /* maybe add some timeout factor ? */
2337                                 break;
2338                         }
2339
2340                         /* enable transmits in the hardware */
2341                         tctl = er32(TCTL);
2342                         tctl |= E1000_TCTL_EN;
2343                         ew32(TCTL, tctl);
2344
2345                         netif_carrier_on(netdev);
2346                         if (!test_bit(__E1000_DOWN, &adapter->flags))
2347                                 mod_timer(&adapter->phy_info_timer,
2348                                           round_jiffies(jiffies + 2 * HZ));
2349                         adapter->smartspeed = 0;
2350                 }
2351         } else {
2352                 if (netif_carrier_ok(netdev)) {
2353                         adapter->link_speed = 0;
2354                         adapter->link_duplex = 0;
2355                         printk(KERN_INFO "e1000: %s NIC Link is Down\n",
2356                                netdev->name);
2357                         netif_carrier_off(netdev);
2358
2359                         if (!test_bit(__E1000_DOWN, &adapter->flags))
2360                                 mod_timer(&adapter->phy_info_timer,
2361                                           round_jiffies(jiffies + 2 * HZ));
2362                 }
2363
2364                 e1000_smartspeed(adapter);
2365         }
2366
2367 link_up:
2368         e1000_update_stats(adapter);
2369
2370         hw->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
2371         adapter->tpt_old = adapter->stats.tpt;
2372         hw->collision_delta = adapter->stats.colc - adapter->colc_old;
2373         adapter->colc_old = adapter->stats.colc;
2374
2375         adapter->gorcl = adapter->stats.gorcl - adapter->gorcl_old;
2376         adapter->gorcl_old = adapter->stats.gorcl;
2377         adapter->gotcl = adapter->stats.gotcl - adapter->gotcl_old;
2378         adapter->gotcl_old = adapter->stats.gotcl;
2379
2380         e1000_update_adaptive(hw);
2381
2382         if (!netif_carrier_ok(netdev)) {
2383                 if (E1000_DESC_UNUSED(txdr) + 1 < txdr->count) {
2384                         /* We've lost link, so the controller stops DMA,
2385                          * but we've got queued Tx work that's never going
2386                          * to get done, so reset controller to flush Tx.
2387                          * (Do the reset outside of interrupt context). */
2388                         adapter->tx_timeout_count++;
2389                         schedule_work(&adapter->reset_task);
2390                         /* return immediately since reset is imminent */
2391                         return;
2392                 }
2393         }
2394
2395         /* Cause software interrupt to ensure rx ring is cleaned */
2396         ew32(ICS, E1000_ICS_RXDMT0);
2397
2398         /* Force detection of hung controller every watchdog period */
2399         adapter->detect_tx_hung = true;
2400
2401         /* Reset the timer */
2402         if (!test_bit(__E1000_DOWN, &adapter->flags))
2403                 mod_timer(&adapter->watchdog_timer,
2404                           round_jiffies(jiffies + 2 * HZ));
2405 }
2406
2407 enum latency_range {
2408         lowest_latency = 0,
2409         low_latency = 1,
2410         bulk_latency = 2,
2411         latency_invalid = 255
2412 };
2413
2414 /**
2415  * e1000_update_itr - update the dynamic ITR value based on statistics
2416  * @adapter: pointer to adapter
2417  * @itr_setting: current adapter->itr
2418  * @packets: the number of packets during this measurement interval
2419  * @bytes: the number of bytes during this measurement interval
2420  *
2421  *      Stores a new ITR value based on packets and byte
2422  *      counts during the last interrupt.  The advantage of per interrupt
2423  *      computation is faster updates and more accurate ITR for the current
2424  *      traffic pattern.  Constants in this function were computed
2425  *      based on theoretical maximum wire speed and thresholds were set based
2426  *      on testing data as well as attempting to minimize response time
2427  *      while increasing bulk throughput.
2428  *      this functionality is controlled by the InterruptThrottleRate module
2429  *      parameter (see e1000_param.c)
2430  **/
2431 static unsigned int e1000_update_itr(struct e1000_adapter *adapter,
2432                                      u16 itr_setting, int packets, int bytes)
2433 {
2434         unsigned int retval = itr_setting;
2435         struct e1000_hw *hw = &adapter->hw;
2436
2437         if (unlikely(hw->mac_type < e1000_82540))
2438                 goto update_itr_done;
2439
2440         if (packets == 0)
2441                 goto update_itr_done;
2442
2443         switch (itr_setting) {
2444         case lowest_latency:
2445                 /* jumbo frames get bulk treatment*/
2446                 if (bytes/packets > 8000)
2447                         retval = bulk_latency;
2448                 else if ((packets < 5) && (bytes > 512))
2449                         retval = low_latency;
2450                 break;
2451         case low_latency:  /* 50 usec aka 20000 ints/s */
2452                 if (bytes > 10000) {
2453                         /* jumbo frames need bulk latency setting */
2454                         if (bytes/packets > 8000)
2455                                 retval = bulk_latency;
2456                         else if ((packets < 10) || ((bytes/packets) > 1200))
2457                                 retval = bulk_latency;
2458                         else if ((packets > 35))
2459                                 retval = lowest_latency;
2460                 } else if (bytes/packets > 2000)
2461                         retval = bulk_latency;
2462                 else if (packets <= 2 && bytes < 512)
2463                         retval = lowest_latency;
2464                 break;
2465         case bulk_latency: /* 250 usec aka 4000 ints/s */
2466                 if (bytes > 25000) {
2467                         if (packets > 35)
2468                                 retval = low_latency;
2469                 } else if (bytes < 6000) {
2470                         retval = low_latency;
2471                 }
2472                 break;
2473         }
2474
2475 update_itr_done:
2476         return retval;
2477 }
2478
2479 static void e1000_set_itr(struct e1000_adapter *adapter)
2480 {
2481         struct e1000_hw *hw = &adapter->hw;
2482         u16 current_itr;
2483         u32 new_itr = adapter->itr;
2484
2485         if (unlikely(hw->mac_type < e1000_82540))
2486                 return;
2487
2488         /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
2489         if (unlikely(adapter->link_speed != SPEED_1000)) {
2490                 current_itr = 0;
2491                 new_itr = 4000;
2492                 goto set_itr_now;
2493         }
2494
2495         adapter->tx_itr = e1000_update_itr(adapter,
2496                                     adapter->tx_itr,
2497                                     adapter->total_tx_packets,
2498                                     adapter->total_tx_bytes);
2499         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2500         if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
2501                 adapter->tx_itr = low_latency;
2502
2503         adapter->rx_itr = e1000_update_itr(adapter,
2504                                     adapter->rx_itr,
2505                                     adapter->total_rx_packets,
2506                                     adapter->total_rx_bytes);
2507         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2508         if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
2509                 adapter->rx_itr = low_latency;
2510
2511         current_itr = max(adapter->rx_itr, adapter->tx_itr);
2512
2513         switch (current_itr) {
2514         /* counts and packets in update_itr are dependent on these numbers */
2515         case lowest_latency:
2516                 new_itr = 70000;
2517                 break;
2518         case low_latency:
2519                 new_itr = 20000; /* aka hwitr = ~200 */
2520                 break;
2521         case bulk_latency:
2522                 new_itr = 4000;
2523                 break;
2524         default:
2525                 break;
2526         }
2527
2528 set_itr_now:
2529         if (new_itr != adapter->itr) {
2530                 /* this attempts to bias the interrupt rate towards Bulk
2531                  * by adding intermediate steps when interrupt rate is
2532                  * increasing */
2533                 new_itr = new_itr > adapter->itr ?
2534                              min(adapter->itr + (new_itr >> 2), new_itr) :
2535                              new_itr;
2536                 adapter->itr = new_itr;
2537                 ew32(ITR, 1000000000 / (new_itr * 256));
2538         }
2539
2540         return;
2541 }
2542
2543 #define E1000_TX_FLAGS_CSUM             0x00000001
2544 #define E1000_TX_FLAGS_VLAN             0x00000002
2545 #define E1000_TX_FLAGS_TSO              0x00000004
2546 #define E1000_TX_FLAGS_IPV4             0x00000008
2547 #define E1000_TX_FLAGS_VLAN_MASK        0xffff0000
2548 #define E1000_TX_FLAGS_VLAN_SHIFT       16
2549
2550 static int e1000_tso(struct e1000_adapter *adapter,
2551                      struct e1000_tx_ring *tx_ring, struct sk_buff *skb)
2552 {
2553         struct e1000_context_desc *context_desc;
2554         struct e1000_buffer *buffer_info;
2555         unsigned int i;
2556         u32 cmd_length = 0;
2557         u16 ipcse = 0, tucse, mss;
2558         u8 ipcss, ipcso, tucss, tucso, hdr_len;
2559         int err;
2560
2561         if (skb_is_gso(skb)) {
2562                 if (skb_header_cloned(skb)) {
2563                         err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2564                         if (err)
2565                                 return err;
2566                 }
2567
2568                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
2569                 mss = skb_shinfo(skb)->gso_size;
2570                 if (skb->protocol == htons(ETH_P_IP)) {
2571                         struct iphdr *iph = ip_hdr(skb);
2572                         iph->tot_len = 0;
2573                         iph->check = 0;
2574                         tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
2575                                                                  iph->daddr, 0,
2576                                                                  IPPROTO_TCP,
2577                                                                  0);
2578                         cmd_length = E1000_TXD_CMD_IP;
2579                         ipcse = skb_transport_offset(skb) - 1;
2580                 } else if (skb->protocol == htons(ETH_P_IPV6)) {
2581                         ipv6_hdr(skb)->payload_len = 0;
2582                         tcp_hdr(skb)->check =
2583                                 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
2584                                                  &ipv6_hdr(skb)->daddr,
2585                                                  0, IPPROTO_TCP, 0);
2586                         ipcse = 0;
2587                 }
2588                 ipcss = skb_network_offset(skb);
2589                 ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
2590                 tucss = skb_transport_offset(skb);
2591                 tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
2592                 tucse = 0;
2593
2594                 cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
2595                                E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
2596
2597                 i = tx_ring->next_to_use;
2598                 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
2599                 buffer_info = &tx_ring->buffer_info[i];
2600
2601                 context_desc->lower_setup.ip_fields.ipcss  = ipcss;
2602                 context_desc->lower_setup.ip_fields.ipcso  = ipcso;
2603                 context_desc->lower_setup.ip_fields.ipcse  = cpu_to_le16(ipcse);
2604                 context_desc->upper_setup.tcp_fields.tucss = tucss;
2605                 context_desc->upper_setup.tcp_fields.tucso = tucso;
2606                 context_desc->upper_setup.tcp_fields.tucse = cpu_to_le16(tucse);
2607                 context_desc->tcp_seg_setup.fields.mss     = cpu_to_le16(mss);
2608                 context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
2609                 context_desc->cmd_and_length = cpu_to_le32(cmd_length);
2610
2611                 buffer_info->time_stamp = jiffies;
2612                 buffer_info->next_to_watch = i;
2613
2614                 if (++i == tx_ring->count) i = 0;
2615                 tx_ring->next_to_use = i;
2616
2617                 return true;
2618         }
2619         return false;
2620 }
2621
2622 static bool e1000_tx_csum(struct e1000_adapter *adapter,
2623                           struct e1000_tx_ring *tx_ring, struct sk_buff *skb)
2624 {
2625         struct e1000_context_desc *context_desc;
2626         struct e1000_buffer *buffer_info;
2627         unsigned int i;
2628         u8 css;
2629         u32 cmd_len = E1000_TXD_CMD_DEXT;
2630
2631         if (skb->ip_summed != CHECKSUM_PARTIAL)
2632                 return false;
2633
2634         switch (skb->protocol) {
2635         case cpu_to_be16(ETH_P_IP):
2636                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
2637                         cmd_len |= E1000_TXD_CMD_TCP;
2638                 break;
2639         case cpu_to_be16(ETH_P_IPV6):
2640                 /* XXX not handling all IPV6 headers */
2641                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
2642                         cmd_len |= E1000_TXD_CMD_TCP;
2643                 break;
2644         default:
2645                 if (unlikely(net_ratelimit()))
2646                         DPRINTK(DRV, WARNING,
2647                                 "checksum_partial proto=%x!\n", skb->protocol);
2648                 break;
2649         }
2650
2651         css = skb_transport_offset(skb);
2652
2653         i = tx_ring->next_to_use;
2654         buffer_info = &tx_ring->buffer_info[i];
2655         context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
2656
2657         context_desc->lower_setup.ip_config = 0;
2658         context_desc->upper_setup.tcp_fields.tucss = css;
2659         context_desc->upper_setup.tcp_fields.tucso =
2660                 css + skb->csum_offset;
2661         context_desc->upper_setup.tcp_fields.tucse = 0;
2662         context_desc->tcp_seg_setup.data = 0;
2663         context_desc->cmd_and_length = cpu_to_le32(cmd_len);
2664
2665         buffer_info->time_stamp = jiffies;
2666         buffer_info->next_to_watch = i;
2667
2668         if (unlikely(++i == tx_ring->count)) i = 0;
2669         tx_ring->next_to_use = i;
2670
2671         return true;
2672 }
2673
2674 #define E1000_MAX_TXD_PWR       12
2675 #define E1000_MAX_DATA_PER_TXD  (1<<E1000_MAX_TXD_PWR)
2676
2677 static int e1000_tx_map(struct e1000_adapter *adapter,
2678                         struct e1000_tx_ring *tx_ring,
2679                         struct sk_buff *skb, unsigned int first,
2680                         unsigned int max_per_txd, unsigned int nr_frags,
2681                         unsigned int mss)
2682 {
2683         struct e1000_hw *hw = &adapter->hw;
2684         struct pci_dev *pdev = adapter->pdev;
2685         struct e1000_buffer *buffer_info;
2686         unsigned int len = skb_headlen(skb);
2687         unsigned int offset = 0, size, count = 0, i;
2688         unsigned int f;
2689
2690         i = tx_ring->next_to_use;
2691
2692         while (len) {
2693                 buffer_info = &tx_ring->buffer_info[i];
2694                 size = min(len, max_per_txd);
2695                 /* Workaround for Controller erratum --
2696                  * descriptor for non-tso packet in a linear SKB that follows a
2697                  * tso gets written back prematurely before the data is fully
2698                  * DMA'd to the controller */
2699                 if (!skb->data_len && tx_ring->last_tx_tso &&
2700                     !skb_is_gso(skb)) {
2701                         tx_ring->last_tx_tso = 0;
2702                         size -= 4;
2703                 }
2704
2705                 /* Workaround for premature desc write-backs
2706                  * in TSO mode.  Append 4-byte sentinel desc */
2707                 if (unlikely(mss && !nr_frags && size == len && size > 8))
2708                         size -= 4;
2709                 /* work-around for errata 10 and it applies
2710                  * to all controllers in PCI-X mode
2711                  * The fix is to make sure that the first descriptor of a
2712                  * packet is smaller than 2048 - 16 - 16 (or 2016) bytes
2713                  */
2714                 if (unlikely((hw->bus_type == e1000_bus_type_pcix) &&
2715                                 (size > 2015) && count == 0))
2716                         size = 2015;
2717
2718                 /* Workaround for potential 82544 hang in PCI-X.  Avoid
2719                  * terminating buffers within evenly-aligned dwords. */
2720                 if (unlikely(adapter->pcix_82544 &&
2721                    !((unsigned long)(skb->data + offset + size - 1) & 4) &&
2722                    size > 4))
2723                         size -= 4;
2724
2725                 buffer_info->length = size;
2726                 /* set time_stamp *before* dma to help avoid a possible race */
2727                 buffer_info->time_stamp = jiffies;
2728                 buffer_info->mapped_as_page = false;
2729                 buffer_info->dma = pci_map_single(pdev, skb->data + offset,
2730                                                   size, PCI_DMA_TODEVICE);
2731                 if (pci_dma_mapping_error(pdev, buffer_info->dma))
2732                         goto dma_error;
2733                 buffer_info->next_to_watch = i;
2734
2735                 len -= size;
2736                 offset += size;
2737                 count++;
2738                 if (len) {
2739                         i++;
2740                         if (unlikely(i == tx_ring->count))
2741                                 i = 0;
2742                 }
2743         }
2744
2745         for (f = 0; f < nr_frags; f++) {
2746                 struct skb_frag_struct *frag;
2747
2748                 frag = &skb_shinfo(skb)->frags[f];
2749                 len = frag->size;
2750                 offset = frag->page_offset;
2751
2752                 while (len) {
2753                         i++;
2754                         if (unlikely(i == tx_ring->count))
2755                                 i = 0;
2756
2757                         buffer_info = &tx_ring->buffer_info[i];
2758                         size = min(len, max_per_txd);
2759                         /* Workaround for premature desc write-backs
2760                          * in TSO mode.  Append 4-byte sentinel desc */
2761                         if (unlikely(mss && f == (nr_frags-1) && size == len && size > 8))
2762                                 size -= 4;
2763                         /* Workaround for potential 82544 hang in PCI-X.
2764                          * Avoid terminating buffers within evenly-aligned
2765                          * dwords. */
2766                         if (unlikely(adapter->pcix_82544 &&
2767                             !((unsigned long)(page_to_phys(frag->page) + offset
2768                                               + size - 1) & 4) &&
2769                             size > 4))
2770                                 size -= 4;
2771
2772                         buffer_info->length = size;
2773                         buffer_info->time_stamp = jiffies;
2774                         buffer_info->mapped_as_page = true;
2775                         buffer_info->dma = pci_map_page(pdev, frag->page,
2776                                                         offset, size,
2777                                                         PCI_DMA_TODEVICE);
2778                         if (pci_dma_mapping_error(pdev, buffer_info->dma))
2779                                 goto dma_error;
2780                         buffer_info->next_to_watch = i;
2781
2782                         len -= size;
2783                         offset += size;
2784                         count++;
2785                 }
2786         }
2787
2788         tx_ring->buffer_info[i].skb = skb;
2789         tx_ring->buffer_info[first].next_to_watch = i;
2790
2791         return count;
2792
2793 dma_error:
2794         dev_err(&pdev->dev, "TX DMA map failed\n");
2795         buffer_info->dma = 0;
2796         if (count)
2797                 count--;
2798
2799         while (count--) {
2800                 if (i==0)
2801                         i += tx_ring->count;
2802                 i--;
2803                 buffer_info = &tx_ring->buffer_info[i];
2804                 e1000_unmap_and_free_tx_resource(adapter, buffer_info);
2805         }
2806
2807         return 0;
2808 }
2809
2810 static void e1000_tx_queue(struct e1000_adapter *adapter,
2811                            struct e1000_tx_ring *tx_ring, int tx_flags,
2812                            int count)
2813 {
2814         struct e1000_hw *hw = &adapter->hw;
2815         struct e1000_tx_desc *tx_desc = NULL;
2816         struct e1000_buffer *buffer_info;
2817         u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
2818         unsigned int i;
2819
2820         if (likely(tx_flags & E1000_TX_FLAGS_TSO)) {
2821                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
2822                              E1000_TXD_CMD_TSE;
2823                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
2824
2825                 if (likely(tx_flags & E1000_TX_FLAGS_IPV4))
2826                         txd_upper |= E1000_TXD_POPTS_IXSM << 8;
2827         }
2828
2829         if (likely(tx_flags & E1000_TX_FLAGS_CSUM)) {
2830                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
2831                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
2832         }
2833
2834         if (unlikely(tx_flags & E1000_TX_FLAGS_VLAN)) {
2835                 txd_lower |= E1000_TXD_CMD_VLE;
2836                 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
2837         }
2838
2839         i = tx_ring->next_to_use;
2840
2841         while (count--) {
2842                 buffer_info = &tx_ring->buffer_info[i];
2843                 tx_desc = E1000_TX_DESC(*tx_ring, i);
2844                 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
2845                 tx_desc->lower.data =
2846                         cpu_to_le32(txd_lower | buffer_info->length);
2847                 tx_desc->upper.data = cpu_to_le32(txd_upper);
2848                 if (unlikely(++i == tx_ring->count)) i = 0;
2849         }
2850
2851         tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
2852
2853         /* Force memory writes to complete before letting h/w
2854          * know there are new descriptors to fetch.  (Only
2855          * applicable for weak-ordered memory model archs,
2856          * such as IA-64). */
2857         wmb();
2858
2859         tx_ring->next_to_use = i;
2860         writel(i, hw->hw_addr + tx_ring->tdt);
2861         /* we need this if more than one processor can write to our tail
2862          * at a time, it syncronizes IO on IA64/Altix systems */
2863         mmiowb();
2864 }
2865
2866 /**
2867  * 82547 workaround to avoid controller hang in half-duplex environment.
2868  * The workaround is to avoid queuing a large packet that would span
2869  * the internal Tx FIFO ring boundary by notifying the stack to resend
2870  * the packet at a later time.  This gives the Tx FIFO an opportunity to
2871  * flush all packets.  When that occurs, we reset the Tx FIFO pointers
2872  * to the beginning of the Tx FIFO.
2873  **/
2874
2875 #define E1000_FIFO_HDR                  0x10
2876 #define E1000_82547_PAD_LEN             0x3E0
2877
2878 static int e1000_82547_fifo_workaround(struct e1000_adapter *adapter,
2879                                        struct sk_buff *skb)
2880 {
2881         u32 fifo_space = adapter->tx_fifo_size - adapter->tx_fifo_head;
2882         u32 skb_fifo_len = skb->len + E1000_FIFO_HDR;
2883
2884         skb_fifo_len = ALIGN(skb_fifo_len, E1000_FIFO_HDR);
2885
2886         if (adapter->link_duplex != HALF_DUPLEX)
2887                 goto no_fifo_stall_required;
2888
2889         if (atomic_read(&adapter->tx_fifo_stall))
2890                 return 1;
2891
2892         if (skb_fifo_len >= (E1000_82547_PAD_LEN + fifo_space)) {
2893                 atomic_set(&adapter->tx_fifo_stall, 1);
2894                 return 1;
2895         }
2896
2897 no_fifo_stall_required:
2898         adapter->tx_fifo_head += skb_fifo_len;
2899         if (adapter->tx_fifo_head >= adapter->tx_fifo_size)
2900                 adapter->tx_fifo_head -= adapter->tx_fifo_size;
2901         return 0;
2902 }
2903
2904 static int __e1000_maybe_stop_tx(struct net_device *netdev, int size)
2905 {
2906         struct e1000_adapter *adapter = netdev_priv(netdev);
2907         struct e1000_tx_ring *tx_ring = adapter->tx_ring;
2908
2909         netif_stop_queue(netdev);
2910         /* Herbert's original patch had:
2911          *  smp_mb__after_netif_stop_queue();
2912          * but since that doesn't exist yet, just open code it. */
2913         smp_mb();
2914
2915         /* We need to check again in a case another CPU has just
2916          * made room available. */
2917         if (likely(E1000_DESC_UNUSED(tx_ring) < size))
2918                 return -EBUSY;
2919
2920         /* A reprieve! */
2921         netif_start_queue(netdev);
2922         ++adapter->restart_queue;
2923         return 0;
2924 }
2925
2926 static int e1000_maybe_stop_tx(struct net_device *netdev,
2927                                struct e1000_tx_ring *tx_ring, int size)
2928 {
2929         if (likely(E1000_DESC_UNUSED(tx_ring) >= size))
2930                 return 0;
2931         return __e1000_maybe_stop_tx(netdev, size);
2932 }
2933
2934 #define TXD_USE_COUNT(S, X) (((S) >> (X)) + 1 )
2935 static netdev_tx_t e1000_xmit_frame(struct sk_buff *skb,
2936                                     struct net_device *netdev)
2937 {
2938         struct e1000_adapter *adapter = netdev_priv(netdev);
2939         struct e1000_hw *hw = &adapter->hw;
2940         struct e1000_tx_ring *tx_ring;
2941         unsigned int first, max_per_txd = E1000_MAX_DATA_PER_TXD;
2942         unsigned int max_txd_pwr = E1000_MAX_TXD_PWR;
2943         unsigned int tx_flags = 0;
2944         unsigned int len = skb->len - skb->data_len;
2945         unsigned int nr_frags;
2946         unsigned int mss;
2947         int count = 0;
2948         int tso;
2949         unsigned int f;
2950
2951         /* This goes back to the question of how to logically map a tx queue
2952          * to a flow.  Right now, performance is impacted slightly negatively
2953          * if using multiple tx queues.  If the stack breaks away from a
2954          * single qdisc implementation, we can look at this again. */
2955         tx_ring = adapter->tx_ring;
2956
2957         if (unlikely(skb->len <= 0)) {
2958                 dev_kfree_skb_any(skb);
2959                 return NETDEV_TX_OK;
2960         }
2961
2962         mss = skb_shinfo(skb)->gso_size;
2963         /* The controller does a simple calculation to
2964          * make sure there is enough room in the FIFO before
2965          * initiating the DMA for each buffer.  The calc is:
2966          * 4 = ceil(buffer len/mss).  To make sure we don't
2967          * overrun the FIFO, adjust the max buffer len if mss
2968          * drops. */
2969         if (mss) {
2970                 u8 hdr_len;
2971                 max_per_txd = min(mss << 2, max_per_txd);
2972                 max_txd_pwr = fls(max_per_txd) - 1;
2973
2974                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
2975                 if (skb->data_len && hdr_len == len) {
2976                         switch (hw->mac_type) {
2977                                 unsigned int pull_size;
2978                         case e1000_82544:
2979                                 /* Make sure we have room to chop off 4 bytes,
2980                                  * and that the end alignment will work out to
2981                                  * this hardware's requirements
2982                                  * NOTE: this is a TSO only workaround
2983                                  * if end byte alignment not correct move us
2984                                  * into the next dword */
2985                                 if ((unsigned long)(skb_tail_pointer(skb) - 1) & 4)
2986                                         break;
2987                                 /* fall through */
2988                                 pull_size = min((unsigned int)4, skb->data_len);
2989                                 if (!__pskb_pull_tail(skb, pull_size)) {
2990                                         DPRINTK(DRV, ERR,
2991                                                 "__pskb_pull_tail failed.\n");
2992                                         dev_kfree_skb_any(skb);
2993                                         return NETDEV_TX_OK;
2994                                 }
2995                                 len = skb->len - skb->data_len;
2996                                 break;
2997                         default:
2998                                 /* do nothing */
2999                                 break;
3000                         }
3001                 }
3002         }
3003
3004         /* reserve a descriptor for the offload context */
3005         if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
3006                 count++;
3007         count++;
3008
3009         /* Controller Erratum workaround */
3010         if (!skb->data_len && tx_ring->last_tx_tso && !skb_is_gso(skb))
3011                 count++;
3012
3013         count += TXD_USE_COUNT(len, max_txd_pwr);
3014
3015         if (adapter->pcix_82544)
3016                 count++;
3017
3018         /* work-around for errata 10 and it applies to all controllers
3019          * in PCI-X mode, so add one more descriptor to the count
3020          */
3021         if (unlikely((hw->bus_type == e1000_bus_type_pcix) &&
3022                         (len > 2015)))
3023                 count++;
3024
3025         nr_frags = skb_shinfo(skb)->nr_frags;
3026         for (f = 0; f < nr_frags; f++)
3027                 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size,
3028                                        max_txd_pwr);
3029         if (adapter->pcix_82544)
3030                 count += nr_frags;
3031
3032         /* need: count + 2 desc gap to keep tail from touching
3033          * head, otherwise try next time */
3034         if (unlikely(e1000_maybe_stop_tx(netdev, tx_ring, count + 2)))
3035                 return NETDEV_TX_BUSY;
3036
3037         if (unlikely(hw->mac_type == e1000_82547)) {
3038                 if (unlikely(e1000_82547_fifo_workaround(adapter, skb))) {
3039                         netif_stop_queue(netdev);
3040                         if (!test_bit(__E1000_DOWN, &adapter->flags))
3041                                 mod_timer(&adapter->tx_fifo_stall_timer,
3042                                           jiffies + 1);
3043                         return NETDEV_TX_BUSY;
3044                 }
3045         }
3046
3047         if (unlikely(adapter->vlgrp && vlan_tx_tag_present(skb))) {
3048                 tx_flags |= E1000_TX_FLAGS_VLAN;
3049                 tx_flags |= (vlan_tx_tag_get(skb) << E1000_TX_FLAGS_VLAN_SHIFT);
3050         }
3051
3052         first = tx_ring->next_to_use;
3053
3054         tso = e1000_tso(adapter, tx_ring, skb);
3055         if (tso < 0) {
3056                 dev_kfree_skb_any(skb);
3057                 return NETDEV_TX_OK;
3058         }
3059
3060         if (likely(tso)) {
3061                 if (likely(hw->mac_type != e1000_82544))
3062                         tx_ring->last_tx_tso = 1;
3063                 tx_flags |= E1000_TX_FLAGS_TSO;
3064         } else if (likely(e1000_tx_csum(adapter, tx_ring, skb)))
3065                 tx_flags |= E1000_TX_FLAGS_CSUM;
3066
3067         if (likely(skb->protocol == htons(ETH_P_IP)))
3068                 tx_flags |= E1000_TX_FLAGS_IPV4;
3069
3070         count = e1000_tx_map(adapter, tx_ring, skb, first, max_per_txd,
3071                              nr_frags, mss);
3072
3073         if (count) {
3074                 e1000_tx_queue(adapter, tx_ring, tx_flags, count);
3075                 /* Make sure there is space in the ring for the next send. */
3076                 e1000_maybe_stop_tx(netdev, tx_ring, MAX_SKB_FRAGS + 2);
3077
3078         } else {
3079                 dev_kfree_skb_any(skb);
3080                 tx_ring->buffer_info[first].time_stamp = 0;
3081                 tx_ring->next_to_use = first;
3082         }
3083
3084         return NETDEV_TX_OK;
3085 }
3086
3087 /**
3088  * e1000_tx_timeout - Respond to a Tx Hang
3089  * @netdev: network interface device structure
3090  **/
3091
3092 static void e1000_tx_timeout(struct net_device *netdev)
3093 {
3094         struct e1000_adapter *adapter = netdev_priv(netdev);
3095
3096         /* Do the reset outside of interrupt context */
3097         adapter->tx_timeout_count++;
3098         schedule_work(&adapter->reset_task);
3099 }
3100
3101 static void e1000_reset_task(struct work_struct *work)
3102 {
3103         struct e1000_adapter *adapter =
3104                 container_of(work, struct e1000_adapter, reset_task);
3105
3106         e1000_reinit_locked(adapter);
3107 }
3108
3109 /**
3110  * e1000_get_stats - Get System Network Statistics
3111  * @netdev: network interface device structure
3112  *
3113  * Returns the address of the device statistics structure.
3114  * The statistics are actually updated from the timer callback.
3115  **/
3116
3117 static struct net_device_stats *e1000_get_stats(struct net_device *netdev)
3118 {
3119         /* only return the current stats */
3120         return &netdev->stats;
3121 }
3122
3123 /**
3124  * e1000_change_mtu - Change the Maximum Transfer Unit
3125  * @netdev: network interface device structure
3126  * @new_mtu: new value for maximum frame size
3127  *
3128  * Returns 0 on success, negative on failure
3129  **/
3130
3131 static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
3132 {
3133         struct e1000_adapter *adapter = netdev_priv(netdev);
3134         struct e1000_hw *hw = &adapter->hw;
3135         int max_frame = new_mtu + ENET_HEADER_SIZE + ETHERNET_FCS_SIZE;
3136
3137         if ((max_frame < MINIMUM_ETHERNET_FRAME_SIZE) ||
3138             (max_frame > MAX_JUMBO_FRAME_SIZE)) {
3139                 DPRINTK(PROBE, ERR, "Invalid MTU setting\n");
3140                 return -EINVAL;
3141         }
3142
3143         /* Adapter-specific max frame size limits. */
3144         switch (hw->mac_type) {
3145         case e1000_undefined ... e1000_82542_rev2_1:
3146                 if (max_frame > (ETH_FRAME_LEN + ETH_FCS_LEN)) {
3147                         DPRINTK(PROBE, ERR, "Jumbo Frames not supported.\n");
3148                         return -EINVAL;
3149                 }
3150                 break;
3151         default:
3152                 /* Capable of supporting up to MAX_JUMBO_FRAME_SIZE limit. */
3153                 break;
3154         }
3155
3156         while (test_and_set_bit(__E1000_RESETTING, &adapter->flags))
3157                 msleep(1);
3158         /* e1000_down has a dependency on max_frame_size */
3159         hw->max_frame_size = max_frame;
3160         if (netif_running(netdev))
3161                 e1000_down(adapter);
3162
3163         /* NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
3164          * means we reserve 2 more, this pushes us to allocate from the next
3165          * larger slab size.
3166          * i.e. RXBUFFER_2048 --> size-4096 slab
3167          *  however with the new *_jumbo_rx* routines, jumbo receives will use
3168          *  fragmented skbs */
3169
3170         if (max_frame <= E1000_RXBUFFER_2048)
3171                 adapter->rx_buffer_len = E1000_RXBUFFER_2048;
3172         else
3173 #if (PAGE_SIZE >= E1000_RXBUFFER_16384)
3174                 adapter->rx_buffer_len = E1000_RXBUFFER_16384;
3175 #elif (PAGE_SIZE >= E1000_RXBUFFER_4096)
3176                 adapter->rx_buffer_len = PAGE_SIZE;
3177 #endif
3178
3179         /* adjust allocation if LPE protects us, and we aren't using SBP */
3180         if (!hw->tbi_compatibility_on &&
3181             ((max_frame == (ETH_FRAME_LEN + ETH_FCS_LEN)) ||
3182              (max_frame == MAXIMUM_ETHERNET_VLAN_SIZE)))
3183                 adapter->rx_buffer_len = MAXIMUM_ETHERNET_VLAN_SIZE;
3184
3185         printk(KERN_INFO "e1000: %s changing MTU from %d to %d\n",
3186                netdev->name, netdev->mtu, new_mtu);
3187         netdev->mtu = new_mtu;
3188
3189         if (netif_running(netdev))
3190                 e1000_up(adapter);
3191         else
3192                 e1000_reset(adapter);
3193
3194         clear_bit(__E1000_RESETTING, &adapter->flags);
3195
3196         return 0;
3197 }
3198
3199 /**
3200  * e1000_update_stats - Update the board statistics counters
3201  * @adapter: board private structure
3202  **/
3203
3204 void e1000_update_stats(struct e1000_adapter *adapter)
3205 {
3206         struct net_device *netdev = adapter->netdev;
3207         struct e1000_hw *hw = &adapter->hw;
3208         struct pci_dev *pdev = adapter->pdev;
3209         unsigned long flags;
3210         u16 phy_tmp;
3211
3212 #define PHY_IDLE_ERROR_COUNT_MASK 0x00FF
3213
3214         /*
3215          * Prevent stats update while adapter is being reset, or if the pci
3216          * connection is down.
3217          */
3218         if (adapter->link_speed == 0)
3219                 return;
3220         if (pci_channel_offline(pdev))
3221                 return;
3222
3223         spin_lock_irqsave(&adapter->stats_lock, flags);
3224
3225         /* these counters are modified from e1000_tbi_adjust_stats,
3226          * called from the interrupt context, so they must only
3227          * be written while holding adapter->stats_lock
3228          */
3229
3230         adapter->stats.crcerrs += er32(CRCERRS);
3231         adapter->stats.gprc += er32(GPRC);
3232         adapter->stats.gorcl += er32(GORCL);
3233         adapter->stats.gorch += er32(GORCH);
3234         adapter->stats.bprc += er32(BPRC);
3235         adapter->stats.mprc += er32(MPRC);
3236         adapter->stats.roc += er32(ROC);
3237
3238         adapter->stats.prc64 += er32(PRC64);
3239         adapter->stats.prc127 += er32(PRC127);
3240         adapter->stats.prc255 += er32(PRC255);
3241         adapter->stats.prc511 += er32(PRC511);
3242         adapter->stats.prc1023 += er32(PRC1023);
3243         adapter->stats.prc1522 += er32(PRC1522);
3244
3245         adapter->stats.symerrs += er32(SYMERRS);
3246         adapter->stats.mpc += er32(MPC);
3247         adapter->stats.scc += er32(SCC);
3248         adapter->stats.ecol += er32(ECOL);
3249         adapter->stats.mcc += er32(MCC);
3250         adapter->stats.latecol += er32(LATECOL);
3251         adapter->stats.dc += er32(DC);
3252         adapter->stats.sec += er32(SEC);
3253         adapter->stats.rlec += er32(RLEC);
3254         adapter->stats.xonrxc += er32(XONRXC);
3255         adapter->stats.xontxc += er32(XONTXC);
3256         adapter->stats.xoffrxc += er32(XOFFRXC);
3257         adapter->stats.xofftxc += er32(XOFFTXC);
3258         adapter->stats.fcruc += er32(FCRUC);
3259         adapter->stats.gptc += er32(GPTC);
3260         adapter->stats.gotcl += er32(GOTCL);
3261         adapter->stats.gotch += er32(GOTCH);
3262         adapter->stats.rnbc += er32(RNBC);
3263         adapter->stats.ruc += er32(RUC);
3264         adapter->stats.rfc += er32(RFC);
3265         adapter->stats.rjc += er32(RJC);
3266         adapter->stats.torl += er32(TORL);
3267         adapter->stats.torh += er32(TORH);
3268         adapter->stats.totl += er32(TOTL);
3269         adapter->stats.toth += er32(TOTH);
3270         adapter->stats.tpr += er32(TPR);
3271
3272         adapter->stats.ptc64 += er32(PTC64);
3273         adapter->stats.ptc127 += er32(PTC127);
3274         adapter->stats.ptc255 += er32(PTC255);
3275         adapter->stats.ptc511 += er32(PTC511);
3276         adapter->stats.ptc1023 += er32(PTC1023);
3277         adapter->stats.ptc1522 += er32(PTC1522);
3278
3279         adapter->stats.mptc += er32(MPTC);
3280         adapter->stats.bptc += er32(BPTC);
3281
3282         /* used for adaptive IFS */
3283
3284         hw->tx_packet_delta = er32(TPT);
3285         adapter->stats.tpt += hw->tx_packet_delta;
3286         hw->collision_delta = er32(COLC);
3287         adapter->stats.colc += hw->collision_delta;
3288
3289         if (hw->mac_type >= e1000_82543) {
3290                 adapter->stats.algnerrc += er32(ALGNERRC);
3291                 adapter->stats.rxerrc += er32(RXERRC);
3292                 adapter->stats.tncrs += er32(TNCRS);
3293                 adapter->stats.cexterr += er32(CEXTERR);
3294                 adapter->stats.tsctc += er32(TSCTC);
3295                 adapter->stats.tsctfc += er32(TSCTFC);
3296         }
3297
3298         /* Fill out the OS statistics structure */
3299         netdev->stats.multicast = adapter->stats.mprc;
3300         netdev->stats.collisions = adapter->stats.colc;
3301
3302         /* Rx Errors */
3303
3304         /* RLEC on some newer hardware can be incorrect so build
3305         * our own version based on RUC and ROC */
3306         netdev->stats.rx_errors = adapter->stats.rxerrc +
3307                 adapter->stats.crcerrs + adapter->stats.algnerrc +
3308                 adapter->stats.ruc + adapter->stats.roc +
3309                 adapter->stats.cexterr;
3310         adapter->stats.rlerrc = adapter->stats.ruc + adapter->stats.roc;
3311         netdev->stats.rx_length_errors = adapter->stats.rlerrc;
3312         netdev->stats.rx_crc_errors = adapter->stats.crcerrs;
3313         netdev->stats.rx_frame_errors = adapter->stats.algnerrc;
3314         netdev->stats.rx_missed_errors = adapter->stats.mpc;
3315
3316         /* Tx Errors */
3317         adapter->stats.txerrc = adapter->stats.ecol + adapter->stats.latecol;
3318         netdev->stats.tx_errors = adapter->stats.txerrc;
3319         netdev->stats.tx_aborted_errors = adapter->stats.ecol;
3320         netdev->stats.tx_window_errors = adapter->stats.latecol;
3321         netdev->stats.tx_carrier_errors = adapter->stats.tncrs;
3322         if (hw->bad_tx_carr_stats_fd &&
3323             adapter->link_duplex == FULL_DUPLEX) {
3324                 netdev->stats.tx_carrier_errors = 0;
3325                 adapter->stats.tncrs = 0;
3326         }
3327
3328         /* Tx Dropped needs to be maintained elsewhere */
3329
3330         /* Phy Stats */
3331         if (hw->media_type == e1000_media_type_copper) {
3332                 if ((adapter->link_speed == SPEED_1000) &&
3333                    (!e1000_read_phy_reg(hw, PHY_1000T_STATUS, &phy_tmp))) {
3334                         phy_tmp &= PHY_IDLE_ERROR_COUNT_MASK;
3335                         adapter->phy_stats.idle_errors += phy_tmp;
3336                 }
3337
3338                 if ((hw->mac_type <= e1000_82546) &&
3339                    (hw->phy_type == e1000_phy_m88) &&
3340                    !e1000_read_phy_reg(hw, M88E1000_RX_ERR_CNTR, &phy_tmp))
3341                         adapter->phy_stats.receive_errors += phy_tmp;
3342         }
3343
3344         /* Management Stats */
3345         if (hw->has_smbus) {
3346                 adapter->stats.mgptc += er32(MGTPTC);
3347                 adapter->stats.mgprc += er32(MGTPRC);
3348                 adapter->stats.mgpdc += er32(MGTPDC);
3349         }
3350
3351         spin_unlock_irqrestore(&adapter->stats_lock, flags);
3352 }
3353
3354 /**
3355  * e1000_intr - Interrupt Handler
3356  * @irq: interrupt number
3357  * @data: pointer to a network interface device structure
3358  **/
3359
3360 static irqreturn_t e1000_intr(int irq, void *data)
3361 {
3362         struct net_device *netdev = data;
3363         struct e1000_adapter *adapter = netdev_priv(netdev);
3364         struct e1000_hw *hw = &adapter->hw;
3365         u32 icr = er32(ICR);
3366
3367         if (unlikely((!icr) || test_bit(__E1000_DOWN, &adapter->flags)))
3368                 return IRQ_NONE;  /* Not our interrupt */
3369
3370         if (unlikely(icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC))) {
3371                 hw->get_link_status = 1;
3372                 /* guard against interrupt when we're going down */
3373                 if (!test_bit(__E1000_DOWN, &adapter->flags))
3374                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
3375         }
3376
3377         /* disable interrupts, without the synchronize_irq bit */
3378         ew32(IMC, ~0);
3379         E1000_WRITE_FLUSH();
3380
3381         if (likely(napi_schedule_prep(&adapter->napi))) {
3382                 adapter->total_tx_bytes = 0;
3383                 adapter->total_tx_packets = 0;
3384                 adapter->total_rx_bytes = 0;
3385                 adapter->total_rx_packets = 0;
3386                 __napi_schedule(&adapter->napi);
3387         } else {
3388                 /* this really should not happen! if it does it is basically a
3389                  * bug, but not a hard error, so enable ints and continue */
3390                 if (!test_bit(__E1000_DOWN, &adapter->flags))
3391                         e1000_irq_enable(adapter);
3392         }
3393
3394         return IRQ_HANDLED;
3395 }
3396
3397 /**
3398  * e1000_clean - NAPI Rx polling callback
3399  * @adapter: board private structure
3400  **/
3401 static int e1000_clean(struct napi_struct *napi, int budget)
3402 {
3403         struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter, napi);
3404         int tx_clean_complete = 0, work_done = 0;
3405
3406         tx_clean_complete = e1000_clean_tx_irq(adapter, &adapter->tx_ring[0]);
3407
3408         adapter->clean_rx(adapter, &adapter->rx_ring[0], &work_done, budget);
3409
3410         if (!tx_clean_complete)
3411                 work_done = budget;
3412
3413         /* If budget not fully consumed, exit the polling mode */
3414         if (work_done < budget) {
3415                 if (likely(adapter->itr_setting & 3))
3416                         e1000_set_itr(adapter);
3417                 napi_complete(napi);
3418                 if (!test_bit(__E1000_DOWN, &adapter->flags))
3419                         e1000_irq_enable(adapter);
3420         }
3421
3422         return work_done;
3423 }
3424
3425 /**
3426  * e1000_clean_tx_irq - Reclaim resources after transmit completes
3427  * @adapter: board private structure
3428  **/
3429 static bool e1000_clean_tx_irq(struct e1000_adapter *adapter,
3430                                struct e1000_tx_ring *tx_ring)
3431 {
3432         struct e1000_hw *hw = &adapter->hw;
3433         struct net_device *netdev = adapter->netdev;
3434         struct e1000_tx_desc *tx_desc, *eop_desc;
3435         struct e1000_buffer *buffer_info;
3436         unsigned int i, eop;
3437         unsigned int count = 0;
3438         unsigned int total_tx_bytes=0, total_tx_packets=0;
3439
3440         i = tx_ring->next_to_clean;
3441         eop = tx_ring->buffer_info[i].next_to_watch;
3442         eop_desc = E1000_TX_DESC(*tx_ring, eop);
3443
3444         while ((eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) &&
3445                (count < tx_ring->count)) {
3446                 bool cleaned = false;
3447                 for ( ; !cleaned; count++) {
3448                         tx_desc = E1000_TX_DESC(*tx_ring, i);
3449                         buffer_info = &tx_ring->buffer_info[i];
3450                         cleaned = (i == eop);
3451
3452                         if (cleaned) {
3453                                 struct sk_buff *skb = buffer_info->skb;
3454                                 unsigned int segs, bytecount;
3455                                 segs = skb_shinfo(skb)->gso_segs ?: 1;
3456                                 /* multiply data chunks by size of headers */
3457                                 bytecount = ((segs - 1) * skb_headlen(skb)) +
3458                                             skb->len;
3459                                 total_tx_packets += segs;
3460                                 total_tx_bytes += bytecount;
3461                         }
3462                         e1000_unmap_and_free_tx_resource(adapter, buffer_info);
3463                         tx_desc->upper.data = 0;
3464
3465                         if (unlikely(++i == tx_ring->count)) i = 0;
3466                 }
3467
3468                 eop = tx_ring->buffer_info[i].next_to_watch;
3469                 eop_desc = E1000_TX_DESC(*tx_ring, eop);
3470         }
3471
3472         tx_ring->next_to_clean = i;
3473
3474 #define TX_WAKE_THRESHOLD 32
3475         if (unlikely(count && netif_carrier_ok(netdev) &&
3476                      E1000_DESC_UNUSED(tx_ring) >= TX_WAKE_THRESHOLD)) {
3477                 /* Make sure that anybody stopping the queue after this
3478                  * sees the new next_to_clean.
3479                  */
3480                 smp_mb();
3481
3482                 if (netif_queue_stopped(netdev) &&
3483                     !(test_bit(__E1000_DOWN, &adapter->flags))) {
3484                         netif_wake_queue(netdev);
3485                         ++adapter->restart_queue;
3486                 }
3487         }
3488
3489         if (adapter->detect_tx_hung) {
3490                 /* Detect a transmit hang in hardware, this serializes the
3491                  * check with the clearing of time_stamp and movement of i */
3492                 adapter->detect_tx_hung = false;
3493                 if (tx_ring->buffer_info[eop].time_stamp &&
3494                     time_after(jiffies, tx_ring->buffer_info[eop].time_stamp +
3495                                (adapter->tx_timeout_factor * HZ)) &&
3496                     !(er32(STATUS) & E1000_STATUS_TXOFF)) {
3497
3498                         /* detected Tx unit hang */
3499                         DPRINTK(DRV, ERR, "Detected Tx Unit Hang\n"
3500                                         "  Tx Queue             <%lu>\n"
3501                                         "  TDH                  <%x>\n"
3502                                         "  TDT                  <%x>\n"
3503                                         "  next_to_use          <%x>\n"
3504                                         "  next_to_clean        <%x>\n"
3505                                         "buffer_info[next_to_clean]\n"
3506                                         "  time_stamp           <%lx>\n"
3507                                         "  next_to_watch        <%x>\n"
3508                                         "  jiffies              <%lx>\n"
3509                                         "  next_to_watch.status <%x>\n",
3510                                 (unsigned long)((tx_ring - adapter->tx_ring) /
3511                                         sizeof(struct e1000_tx_ring)),
3512                                 readl(hw->hw_addr + tx_ring->tdh),
3513                                 readl(hw->hw_addr + tx_ring->tdt),
3514                                 tx_ring->next_to_use,
3515                                 tx_ring->next_to_clean,
3516                                 tx_ring->buffer_info[eop].time_stamp,
3517                                 eop,
3518                                 jiffies,
3519                                 eop_desc->upper.fields.status);
3520                         netif_stop_queue(netdev);
3521                 }
3522         }
3523         adapter->total_tx_bytes += total_tx_bytes;
3524         adapter->total_tx_packets += total_tx_packets;
3525         netdev->stats.tx_bytes += total_tx_bytes;
3526         netdev->stats.tx_packets += total_tx_packets;
3527         return (count < tx_ring->count);
3528 }
3529
3530 /**
3531  * e1000_rx_checksum - Receive Checksum Offload for 82543
3532  * @adapter:     board private structure
3533  * @status_err:  receive descriptor status and error fields
3534  * @csum:        receive descriptor csum field
3535  * @sk_buff:     socket buffer with received data
3536  **/
3537
3538 static void e1000_rx_checksum(struct e1000_adapter *adapter, u32 status_err,
3539                               u32 csum, struct sk_buff *skb)
3540 {
3541         struct e1000_hw *hw = &adapter->hw;
3542         u16 status = (u16)status_err;
3543         u8 errors = (u8)(status_err >> 24);
3544         skb->ip_summed = CHECKSUM_NONE;
3545
3546         /* 82543 or newer only */
3547         if (unlikely(hw->mac_type < e1000_82543)) return;
3548         /* Ignore Checksum bit is set */
3549         if (unlikely(status & E1000_RXD_STAT_IXSM)) return;
3550         /* TCP/UDP checksum error bit is set */
3551         if (unlikely(errors & E1000_RXD_ERR_TCPE)) {
3552                 /* let the stack verify checksum errors */
3553                 adapter->hw_csum_err++;
3554                 return;
3555         }
3556         /* TCP/UDP Checksum has not been calculated */
3557         if (!(status & E1000_RXD_STAT_TCPCS))
3558                 return;
3559
3560         /* It must be a TCP or UDP packet with a valid checksum */
3561         if (likely(status & E1000_RXD_STAT_TCPCS)) {
3562                 /* TCP checksum is good */
3563                 skb->ip_summed = CHECKSUM_UNNECESSARY;
3564         }
3565         adapter->hw_csum_good++;
3566 }
3567
3568 /**
3569  * e1000_consume_page - helper function
3570  **/
3571 static void e1000_consume_page(struct e1000_buffer *bi, struct sk_buff *skb,
3572                                u16 length)
3573 {
3574         bi->page = NULL;
3575         skb->len += length;
3576         skb->data_len += length;
3577         skb->truesize += length;
3578 }
3579
3580 /**
3581  * e1000_receive_skb - helper function to handle rx indications
3582  * @adapter: board private structure
3583  * @status: descriptor status field as written by hardware
3584  * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
3585  * @skb: pointer to sk_buff to be indicated to stack
3586  */
3587 static void e1000_receive_skb(struct e1000_adapter *adapter, u8 status,
3588                               __le16 vlan, struct sk_buff *skb)
3589 {
3590         if (unlikely(adapter->vlgrp && (status & E1000_RXD_STAT_VP))) {
3591                 vlan_hwaccel_receive_skb(skb, adapter->vlgrp,
3592                                          le16_to_cpu(vlan) &
3593                                          E1000_RXD_SPC_VLAN_MASK);
3594         } else {
3595                 netif_receive_skb(skb);
3596         }
3597 }
3598
3599 /**
3600  * e1000_clean_jumbo_rx_irq - Send received data up the network stack; legacy
3601  * @adapter: board private structure
3602  * @rx_ring: ring to clean
3603  * @work_done: amount of napi work completed this call
3604  * @work_to_do: max amount of work allowed for this call to do
3605  *
3606  * the return value indicates whether actual cleaning was done, there
3607  * is no guarantee that everything was cleaned
3608  */
3609 static bool e1000_clean_jumbo_rx_irq(struct e1000_adapter *adapter,
3610                                      struct e1000_rx_ring *rx_ring,
3611                                      int *work_done, int work_to_do)
3612 {
3613         struct e1000_hw *hw = &adapter->hw;
3614         struct net_device *netdev = adapter->netdev;
3615         struct pci_dev *pdev = adapter->pdev;
3616         struct e1000_rx_desc *rx_desc, *next_rxd;
3617         struct e1000_buffer *buffer_info, *next_buffer;
3618         unsigned long irq_flags;
3619         u32 length;
3620         unsigned int i;
3621         int cleaned_count = 0;
3622         bool cleaned = false;
3623         unsigned int total_rx_bytes=0, total_rx_packets=0;
3624
3625         i = rx_ring->next_to_clean;
3626         rx_desc = E1000_RX_DESC(*rx_ring, i);
3627         buffer_info = &rx_ring->buffer_info[i];
3628
3629         while (rx_desc->status & E1000_RXD_STAT_DD) {
3630                 struct sk_buff *skb;
3631                 u8 status;
3632
3633                 if (*work_done >= work_to_do)
3634                         break;
3635                 (*work_done)++;
3636
3637                 status = rx_desc->status;
3638                 skb = buffer_info->skb;
3639                 buffer_info->skb = NULL;
3640
3641                 if (++i == rx_ring->count) i = 0;
3642                 next_rxd = E1000_RX_DESC(*rx_ring, i);
3643                 prefetch(next_rxd);
3644
3645                 next_buffer = &rx_ring->buffer_info[i];
3646
3647                 cleaned = true;
3648                 cleaned_count++;
3649                 pci_unmap_page(pdev, buffer_info->dma, buffer_info->length,
3650                                PCI_DMA_FROMDEVICE);
3651                 buffer_info->dma = 0;
3652
3653                 length = le16_to_cpu(rx_desc->length);
3654
3655                 /* errors is only valid for DD + EOP descriptors */
3656                 if (unlikely((status & E1000_RXD_STAT_EOP) &&
3657                     (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK))) {
3658                         u8 last_byte = *(skb->data + length - 1);
3659                         if (TBI_ACCEPT(hw, status, rx_desc->errors, length,
3660                                        last_byte)) {
3661                                 spin_lock_irqsave(&adapter->stats_lock,
3662                                                   irq_flags);
3663                                 e1000_tbi_adjust_stats(hw, &adapter->stats,
3664                                                        length, skb->data);
3665                                 spin_unlock_irqrestore(&adapter->stats_lock,
3666                                                        irq_flags);
3667                                 length--;
3668                         } else {
3669                                 /* recycle both page and skb */
3670                                 buffer_info->skb = skb;
3671                                 /* an error means any chain goes out the window
3672                                  * too */
3673                                 if (rx_ring->rx_skb_top)
3674                                         dev_kfree_skb(rx_ring->rx_skb_top);
3675                                 rx_ring->rx_skb_top = NULL;
3676                                 goto next_desc;
3677                         }
3678                 }
3679
3680 #define rxtop rx_ring->rx_skb_top
3681                 if (!(status & E1000_RXD_STAT_EOP)) {
3682                         /* this descriptor is only the beginning (or middle) */
3683                         if (!rxtop) {
3684                                 /* this is the beginning of a chain */
3685                                 rxtop = skb;
3686                                 skb_fill_page_desc(rxtop, 0, buffer_info->page,
3687                                                    0, length);
3688                         } else {
3689                                 /* this is the middle of a chain */
3690                                 skb_fill_page_desc(rxtop,
3691                                     skb_shinfo(rxtop)->nr_frags,
3692                                     buffer_info->page, 0, length);
3693                                 /* re-use the skb, only consumed the page */
3694                                 buffer_info->skb = skb;
3695                         }
3696                         e1000_consume_page(buffer_info, rxtop, length);
3697                         goto next_desc;
3698                 } else {
3699                         if (rxtop) {
3700                                 /* end of the chain */
3701                                 skb_fill_page_desc(rxtop,
3702                                     skb_shinfo(rxtop)->nr_frags,
3703                                     buffer_info->page, 0, length);
3704                                 /* re-use the current skb, we only consumed the
3705                                  * page */
3706                                 buffer_info->skb = skb;
3707                                 skb = rxtop;
3708                                 rxtop = NULL;
3709                                 e1000_consume_page(buffer_info, skb, length);
3710                         } else {
3711                                 /* no chain, got EOP, this buf is the packet
3712                                  * copybreak to save the put_page/alloc_page */
3713                                 if (length <= copybreak &&
3714                                     skb_tailroom(skb) >= length) {
3715                                         u8 *vaddr;
3716                                         vaddr = kmap_atomic(buffer_info->page,
3717                                                             KM_SKB_DATA_SOFTIRQ);
3718                                         memcpy(skb_tail_pointer(skb), vaddr, length);
3719                                         kunmap_atomic(vaddr,
3720                                                       KM_SKB_DATA_SOFTIRQ);
3721                                         /* re-use the page, so don't erase
3722                                          * buffer_info->page */
3723                                         skb_put(skb, length);
3724                                 } else {
3725                                         skb_fill_page_desc(skb, 0,
3726                                                            buffer_info->page, 0,
3727                                                            length);
3728                                         e1000_consume_page(buffer_info, skb,
3729                                                            length);
3730                                 }
3731                         }
3732                 }
3733
3734                 /* Receive Checksum Offload XXX recompute due to CRC strip? */
3735                 e1000_rx_checksum(adapter,
3736                                   (u32)(status) |
3737                                   ((u32)(rx_desc->errors) << 24),
3738                                   le16_to_cpu(rx_desc->csum), skb);
3739
3740                 pskb_trim(skb, skb->len - 4);
3741
3742                 /* probably a little skewed due to removing CRC */
3743                 total_rx_bytes += skb->len;
3744                 total_rx_packets++;
3745
3746                 /* eth type trans needs skb->data to point to something */
3747                 if (!pskb_may_pull(skb, ETH_HLEN)) {
3748                         DPRINTK(DRV, ERR, "pskb_may_pull failed.\n");
3749                         dev_kfree_skb(skb);
3750                         goto next_desc;
3751                 }
3752
3753                 skb->protocol = eth_type_trans(skb, netdev);
3754
3755                 e1000_receive_skb(adapter, status, rx_desc->special, skb);
3756
3757 next_desc:
3758                 rx_desc->status = 0;
3759
3760                 /* return some buffers to hardware, one at a time is too slow */
3761                 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
3762                         adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
3763                         cleaned_count = 0;
3764                 }
3765
3766                 /* use prefetched values */
3767                 rx_desc = next_rxd;
3768                 buffer_info = next_buffer;
3769         }
3770         rx_ring->next_to_clean = i;
3771
3772         cleaned_count = E1000_DESC_UNUSED(rx_ring);
3773         if (cleaned_count)
3774                 adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
3775
3776         adapter->total_rx_packets += total_rx_packets;
3777         adapter->total_rx_bytes += total_rx_bytes;
3778         netdev->stats.rx_bytes += total_rx_bytes;
3779         netdev->stats.rx_packets += total_rx_packets;
3780         return cleaned;
3781 }
3782
3783 /**
3784  * e1000_clean_rx_irq - Send received data up the network stack; legacy
3785  * @adapter: board private structure
3786  * @rx_ring: ring to clean
3787  * @work_done: amount of napi work completed this call
3788  * @work_to_do: max amount of work allowed for this call to do
3789  */
3790 static bool e1000_clean_rx_irq(struct e1000_adapter *adapter,
3791                                struct e1000_rx_ring *rx_ring,
3792                                int *work_done, int work_to_do)
3793 {
3794         struct e1000_hw *hw = &adapter->hw;
3795         struct net_device *netdev = adapter->netdev;
3796         struct pci_dev *pdev = adapter->pdev;
3797         struct e1000_rx_desc *rx_desc, *next_rxd;
3798         struct e1000_buffer *buffer_info, *next_buffer;
3799         unsigned long flags;
3800         u32 length;
3801         unsigned int i;
3802         int cleaned_count = 0;
3803         bool cleaned = false;
3804         unsigned int total_rx_bytes=0, total_rx_packets=0;
3805
3806         i = rx_ring->next_to_clean;
3807         rx_desc = E1000_RX_DESC(*rx_ring, i);
3808         buffer_info = &rx_ring->buffer_info[i];
3809
3810         while (rx_desc->status & E1000_RXD_STAT_DD) {
3811                 struct sk_buff *skb;
3812                 u8 status;
3813
3814                 if (*work_done >= work_to_do)
3815                         break;
3816                 (*work_done)++;
3817
3818                 status = rx_desc->status;
3819                 skb = buffer_info->skb;
3820                 buffer_info->skb = NULL;
3821
3822                 prefetch(skb->data - NET_IP_ALIGN);
3823
3824                 if (++i == rx_ring->count) i = 0;
3825                 next_rxd = E1000_RX_DESC(*rx_ring, i);
3826                 prefetch(next_rxd);
3827
3828                 next_buffer = &rx_ring->buffer_info[i];
3829
3830                 cleaned = true;
3831                 cleaned_count++;
3832                 pci_unmap_single(pdev, buffer_info->dma, buffer_info->length,
3833                                  PCI_DMA_FROMDEVICE);
3834                 buffer_info->dma = 0;
3835
3836                 length = le16_to_cpu(rx_desc->length);
3837                 /* !EOP means multiple descriptors were used to store a single
3838                  * packet, if thats the case we need to toss it.  In fact, we
3839                  * to toss every packet with the EOP bit clear and the next
3840                  * frame that _does_ have the EOP bit set, as it is by
3841                  * definition only a frame fragment
3842                  */
3843                 if (unlikely(!(status & E1000_RXD_STAT_EOP)))
3844                         adapter->discarding = true;
3845
3846                 if (adapter->discarding) {
3847                         /* All receives must fit into a single buffer */
3848                         E1000_DBG("%s: Receive packet consumed multiple"
3849                                   " buffers\n", netdev->name);
3850                         /* recycle */
3851                         buffer_info->skb = skb;
3852                         if (status & E1000_RXD_STAT_EOP)
3853                                 adapter->discarding = false;
3854                         goto next_desc;
3855                 }
3856
3857                 if (unlikely(rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK)) {
3858                         u8 last_byte = *(skb->data + length - 1);
3859                         if (TBI_ACCEPT(hw, status, rx_desc->errors, length,
3860                                        last_byte)) {
3861                                 spin_lock_irqsave(&adapter->stats_lock, flags);
3862                                 e1000_tbi_adjust_stats(hw, &adapter->stats,
3863                                                        length, skb->data);
3864                                 spin_unlock_irqrestore(&adapter->stats_lock,
3865                                                        flags);
3866                                 length--;
3867                         } else {
3868                                 /* recycle */
3869                                 buffer_info->skb = skb;
3870                                 goto next_desc;
3871                         }
3872                 }
3873
3874                 /* adjust length to remove Ethernet CRC, this must be
3875                  * done after the TBI_ACCEPT workaround above */
3876                 length -= 4;
3877
3878                 /* probably a little skewed due to removing CRC */
3879                 total_rx_bytes += length;
3880                 total_rx_packets++;
3881
3882                 /* code added for copybreak, this should improve
3883                  * performance for small packets with large amounts
3884                  * of reassembly being done in the stack */
3885                 if (length < copybreak) {
3886                         struct sk_buff *new_skb =
3887                             netdev_alloc_skb_ip_align(netdev, length);
3888                         if (new_skb) {
3889                                 skb_copy_to_linear_data_offset(new_skb,
3890                                                                -NET_IP_ALIGN,
3891                                                                (skb->data -
3892                                                                 NET_IP_ALIGN),
3893                                                                (length +
3894                                                                 NET_IP_ALIGN));
3895                                 /* save the skb in buffer_info as good */
3896                                 buffer_info->skb = skb;
3897                                 skb = new_skb;
3898                         }
3899                         /* else just continue with the old one */
3900                 }
3901                 /* end copybreak code */
3902                 skb_put(skb, length);
3903
3904                 /* Receive Checksum Offload */
3905                 e1000_rx_checksum(adapter,
3906                                   (u32)(status) |
3907                                   ((u32)(rx_desc->errors) << 24),
3908                                   le16_to_cpu(rx_desc->csum), skb);
3909
3910                 skb->protocol = eth_type_trans(skb, netdev);
3911
3912                 e1000_receive_skb(adapter, status, rx_desc->special, skb);
3913
3914 next_desc:
3915                 rx_desc->status = 0;
3916
3917                 /* return some buffers to hardware, one at a time is too slow */
3918                 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
3919                         adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
3920                         cleaned_count = 0;
3921                 }
3922
3923                 /* use prefetched values */
3924                 rx_desc = next_rxd;
3925                 buffer_info = next_buffer;
3926         }
3927         rx_ring->next_to_clean = i;
3928
3929         cleaned_count = E1000_DESC_UNUSED(rx_ring);
3930         if (cleaned_count)
3931                 adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
3932
3933         adapter->total_rx_packets += total_rx_packets;
3934         adapter->total_rx_bytes += total_rx_bytes;
3935         netdev->stats.rx_bytes += total_rx_bytes;
3936         netdev->stats.rx_packets += total_rx_packets;
3937         return cleaned;
3938 }
3939
3940 /**
3941  * e1000_alloc_jumbo_rx_buffers - Replace used jumbo receive buffers
3942  * @adapter: address of board private structure
3943  * @rx_ring: pointer to receive ring structure
3944  * @cleaned_count: number of buffers to allocate this pass
3945  **/
3946
3947 static void
3948 e1000_alloc_jumbo_rx_buffers(struct e1000_adapter *adapter,
3949                              struct e1000_rx_ring *rx_ring, int cleaned_count)
3950 {
3951         struct net_device *netdev = adapter->netdev;
3952         struct pci_dev *pdev = adapter->pdev;
3953         struct e1000_rx_desc *rx_desc;
3954         struct e1000_buffer *buffer_info;
3955         struct sk_buff *skb;
3956         unsigned int i;
3957         unsigned int bufsz = 256 - 16 /*for skb_reserve */ ;
3958
3959         i = rx_ring->next_to_use;
3960         buffer_info = &rx_ring->buffer_info[i];
3961
3962         while (cleaned_count--) {
3963                 skb = buffer_info->skb;
3964                 if (skb) {
3965                         skb_trim(skb, 0);
3966                         goto check_page;
3967                 }
3968
3969                 skb = netdev_alloc_skb_ip_align(netdev, bufsz);
3970                 if (unlikely(!skb)) {
3971                         /* Better luck next round */
3972                         adapter->alloc_rx_buff_failed++;
3973                         break;
3974                 }
3975
3976                 /* Fix for errata 23, can't cross 64kB boundary */
3977                 if (!e1000_check_64k_bound(adapter, skb->data, bufsz)) {
3978                         struct sk_buff *oldskb = skb;
3979                         DPRINTK(PROBE, ERR, "skb align check failed: %u bytes "
3980                                              "at %p\n", bufsz, skb->data);
3981                         /* Try again, without freeing the previous */
3982                         skb = netdev_alloc_skb_ip_align(netdev, bufsz);
3983                         /* Failed allocation, critical failure */
3984                         if (!skb) {
3985                                 dev_kfree_skb(oldskb);
3986                                 adapter->alloc_rx_buff_failed++;
3987                                 break;
3988                         }
3989
3990                         if (!e1000_check_64k_bound(adapter, skb->data, bufsz)) {
3991                                 /* give up */
3992                                 dev_kfree_skb(skb);
3993                                 dev_kfree_skb(oldskb);
3994                                 break; /* while (cleaned_count--) */
3995                         }
3996
3997                         /* Use new allocation */
3998                         dev_kfree_skb(oldskb);
3999                 }
4000                 buffer_info->skb = skb;
4001                 buffer_info->length = adapter->rx_buffer_len;
4002 check_page:
4003                 /* allocate a new page if necessary */
4004                 if (!buffer_info->page) {
4005                         buffer_info->page = alloc_page(GFP_ATOMIC);
4006                         if (unlikely(!buffer_info->page)) {
4007                                 adapter->alloc_rx_buff_failed++;
4008                                 break;
4009                         }
4010                 }
4011
4012                 if (!buffer_info->dma)
4013                         buffer_info->dma = pci_map_page(pdev,
4014                                                         buffer_info->page, 0,
4015                                                         buffer_info->length,
4016                                                         PCI_DMA_FROMDEVICE);
4017
4018                 rx_desc = E1000_RX_DESC(*rx_ring, i);
4019                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
4020
4021                 if (unlikely(++i == rx_ring->count))
4022                         i = 0;
4023                 buffer_info = &rx_ring->buffer_info[i];
4024         }
4025
4026         if (likely(rx_ring->next_to_use != i)) {
4027                 rx_ring->next_to_use = i;
4028                 if (unlikely(i-- == 0))
4029                         i = (rx_ring->count - 1);
4030
4031                 /* Force memory writes to complete before letting h/w
4032                  * know there are new descriptors to fetch.  (Only
4033                  * applicable for weak-ordered memory model archs,
4034                  * such as IA-64). */
4035                 wmb();
4036                 writel(i, adapter->hw.hw_addr + rx_ring->rdt);
4037         }
4038 }
4039
4040 /**
4041  * e1000_alloc_rx_buffers - Replace used receive buffers; legacy & extended
4042  * @adapter: address of board private structure
4043  **/
4044
4045 static void e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
4046                                    struct e1000_rx_ring *rx_ring,
4047                                    int cleaned_count)
4048 {
4049         struct e1000_hw *hw = &adapter->hw;
4050         struct net_device *netdev = adapter->netdev;
4051         struct pci_dev *pdev = adapter->pdev;
4052         struct e1000_rx_desc *rx_desc;
4053         struct e1000_buffer *buffer_info;
4054         struct sk_buff *skb;
4055         unsigned int i;
4056         unsigned int bufsz = adapter->rx_buffer_len;
4057
4058         i = rx_ring->next_to_use;
4059         buffer_info = &rx_ring->buffer_info[i];
4060
4061         while (cleaned_count--) {
4062                 skb = buffer_info->skb;
4063                 if (skb) {
4064                         skb_trim(skb, 0);
4065                         goto map_skb;
4066                 }
4067
4068                 skb = netdev_alloc_skb_ip_align(netdev, bufsz);
4069                 if (unlikely(!skb)) {
4070                         /* Better luck next round */
4071                         adapter->alloc_rx_buff_failed++;
4072                         break;
4073                 }
4074
4075                 /* Fix for errata 23, can't cross 64kB boundary */
4076                 if (!e1000_check_64k_bound(adapter, skb->data, bufsz)) {
4077                         struct sk_buff *oldskb = skb;
4078                         DPRINTK(RX_ERR, ERR, "skb align check failed: %u bytes "
4079                                              "at %p\n", bufsz, skb->data);
4080                         /* Try again, without freeing the previous */
4081                         skb = netdev_alloc_skb_ip_align(netdev, bufsz);
4082                         /* Failed allocation, critical failure */
4083                         if (!skb) {
4084                                 dev_kfree_skb(oldskb);
4085                                 adapter->alloc_rx_buff_failed++;
4086                                 break;
4087                         }
4088
4089                         if (!e1000_check_64k_bound(adapter, skb->data, bufsz)) {
4090                                 /* give up */
4091                                 dev_kfree_skb(skb);
4092                                 dev_kfree_skb(oldskb);
4093                                 adapter->alloc_rx_buff_failed++;
4094                                 break; /* while !buffer_info->skb */
4095                         }
4096
4097                         /* Use new allocation */
4098                         dev_kfree_skb(oldskb);
4099                 }
4100                 buffer_info->skb = skb;
4101                 buffer_info->length = adapter->rx_buffer_len;
4102 map_skb:
4103                 buffer_info->dma = pci_map_single(pdev,
4104                                                   skb->data,
4105                                                   buffer_info->length,
4106                                                   PCI_DMA_FROMDEVICE);
4107
4108                 /*
4109                  * XXX if it was allocated cleanly it will never map to a
4110                  * boundary crossing
4111                  */
4112
4113                 /* Fix for errata 23, can't cross 64kB boundary */
4114                 if (!e1000_check_64k_bound(adapter,
4115                                         (void *)(unsigned long)buffer_info->dma,
4116                                         adapter->rx_buffer_len)) {
4117                         DPRINTK(RX_ERR, ERR,
4118                                 "dma align check failed: %u bytes at %p\n",
4119                                 adapter->rx_buffer_len,
4120                                 (void *)(unsigned long)buffer_info->dma);
4121                         dev_kfree_skb(skb);
4122                         buffer_info->skb = NULL;
4123
4124                         pci_unmap_single(pdev, buffer_info->dma,
4125                                          adapter->rx_buffer_len,
4126                                          PCI_DMA_FROMDEVICE);
4127                         buffer_info->dma = 0;
4128
4129                         adapter->alloc_rx_buff_failed++;
4130                         break; /* while !buffer_info->skb */
4131                 }
4132                 rx_desc = E1000_RX_DESC(*rx_ring, i);
4133                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
4134
4135                 if (unlikely(++i == rx_ring->count))
4136                         i = 0;
4137                 buffer_info = &rx_ring->buffer_info[i];
4138         }
4139
4140         if (likely(rx_ring->next_to_use != i)) {
4141                 rx_ring->next_to_use = i;
4142                 if (unlikely(i-- == 0))
4143                         i = (rx_ring->count - 1);
4144
4145                 /* Force memory writes to complete before letting h/w
4146                  * know there are new descriptors to fetch.  (Only
4147                  * applicable for weak-ordered memory model archs,
4148                  * such as IA-64). */
4149                 wmb();
4150                 writel(i, hw->hw_addr + rx_ring->rdt);
4151         }
4152 }
4153
4154 /**
4155  * e1000_smartspeed - Workaround for SmartSpeed on 82541 and 82547 controllers.
4156  * @adapter:
4157  **/
4158
4159 static void e1000_smartspeed(struct e1000_adapter *adapter)
4160 {
4161         struct e1000_hw *hw = &adapter->hw;
4162         u16 phy_status;
4163         u16 phy_ctrl;
4164
4165         if ((hw->phy_type != e1000_phy_igp) || !hw->autoneg ||
4166            !(hw->autoneg_advertised & ADVERTISE_1000_FULL))
4167                 return;
4168
4169         if (adapter->smartspeed == 0) {
4170                 /* If Master/Slave config fault is asserted twice,
4171                  * we assume back-to-back */
4172                 e1000_read_phy_reg(hw, PHY_1000T_STATUS, &phy_status);
4173                 if (!(phy_status & SR_1000T_MS_CONFIG_FAULT)) return;
4174                 e1000_read_phy_reg(hw, PHY_1000T_STATUS, &phy_status);
4175                 if (!(phy_status & SR_1000T_MS_CONFIG_FAULT)) return;
4176                 e1000_read_phy_reg(hw, PHY_1000T_CTRL, &phy_ctrl);
4177                 if (phy_ctrl & CR_1000T_MS_ENABLE) {
4178                         phy_ctrl &= ~CR_1000T_MS_ENABLE;
4179                         e1000_write_phy_reg(hw, PHY_1000T_CTRL,
4180                                             phy_ctrl);
4181                         adapter->smartspeed++;
4182                         if (!e1000_phy_setup_autoneg(hw) &&
4183                            !e1000_read_phy_reg(hw, PHY_CTRL,
4184                                                &phy_ctrl)) {
4185                                 phy_ctrl |= (MII_CR_AUTO_NEG_EN |
4186                                              MII_CR_RESTART_AUTO_NEG);
4187                                 e1000_write_phy_reg(hw, PHY_CTRL,
4188                                                     phy_ctrl);
4189                         }
4190                 }
4191                 return;
4192         } else if (adapter->smartspeed == E1000_SMARTSPEED_DOWNSHIFT) {
4193                 /* If still no link, perhaps using 2/3 pair cable */
4194                 e1000_read_phy_reg(hw, PHY_1000T_CTRL, &phy_ctrl);
4195                 phy_ctrl |= CR_1000T_MS_ENABLE;
4196                 e1000_write_phy_reg(hw, PHY_1000T_CTRL, phy_ctrl);
4197                 if (!e1000_phy_setup_autoneg(hw) &&
4198                    !e1000_read_phy_reg(hw, PHY_CTRL, &phy_ctrl)) {
4199                         phy_ctrl |= (MII_CR_AUTO_NEG_EN |
4200                                      MII_CR_RESTART_AUTO_NEG);
4201                         e1000_write_phy_reg(hw, PHY_CTRL, phy_ctrl);
4202                 }
4203         }
4204         /* Restart process after E1000_SMARTSPEED_MAX iterations */
4205         if (adapter->smartspeed++ == E1000_SMARTSPEED_MAX)
4206                 adapter->smartspeed = 0;
4207 }
4208
4209 /**
4210  * e1000_ioctl -
4211  * @netdev:
4212  * @ifreq:
4213  * @cmd:
4214  **/
4215
4216 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
4217 {
4218         switch (cmd) {
4219         case SIOCGMIIPHY:
4220         case SIOCGMIIREG:
4221         case SIOCSMIIREG:
4222                 return e1000_mii_ioctl(netdev, ifr, cmd);
4223         default:
4224                 return -EOPNOTSUPP;
4225         }
4226 }
4227
4228 /**
4229  * e1000_mii_ioctl -
4230  * @netdev:
4231  * @ifreq:
4232  * @cmd:
4233  **/
4234
4235 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
4236                            int cmd)
4237 {
4238         struct e1000_adapter *adapter = netdev_priv(netdev);
4239         struct e1000_hw *hw = &adapter->hw;
4240         struct mii_ioctl_data *data = if_mii(ifr);
4241         int retval;
4242         u16 mii_reg;
4243         u16 spddplx;
4244         unsigned long flags;
4245
4246         if (hw->media_type != e1000_media_type_copper)
4247                 return -EOPNOTSUPP;
4248
4249         switch (cmd) {
4250         case SIOCGMIIPHY:
4251                 data->phy_id = hw->phy_addr;
4252                 break;
4253         case SIOCGMIIREG:
4254                 spin_lock_irqsave(&adapter->stats_lock, flags);
4255                 if (e1000_read_phy_reg(hw, data->reg_num & 0x1F,
4256                                    &data->val_out)) {
4257                         spin_unlock_irqrestore(&adapter->stats_lock, flags);
4258                         return -EIO;
4259                 }
4260                 spin_unlock_irqrestore(&adapter->stats_lock, flags);
4261                 break;
4262         case SIOCSMIIREG:
4263                 if (data->reg_num & ~(0x1F))
4264                         return -EFAULT;
4265                 mii_reg = data->val_in;
4266                 spin_lock_irqsave(&adapter->stats_lock, flags);
4267                 if (e1000_write_phy_reg(hw, data->reg_num,
4268                                         mii_reg)) {
4269                         spin_unlock_irqrestore(&adapter->stats_lock, flags);
4270                         return -EIO;
4271                 }
4272                 spin_unlock_irqrestore(&adapter->stats_lock, flags);
4273                 if (hw->media_type == e1000_media_type_copper) {
4274                         switch (data->reg_num) {
4275                         case PHY_CTRL:
4276                                 if (mii_reg & MII_CR_POWER_DOWN)
4277                                         break;
4278                                 if (mii_reg & MII_CR_AUTO_NEG_EN) {
4279                                         hw->autoneg = 1;
4280                                         hw->autoneg_advertised = 0x2F;
4281                                 } else {
4282                                         if (mii_reg & 0x40)
4283                                                 spddplx = SPEED_1000;
4284                                         else if (mii_reg & 0x2000)
4285                                                 spddplx = SPEED_100;
4286                                         else
4287                                                 spddplx = SPEED_10;
4288                                         spddplx += (mii_reg & 0x100)
4289                                                    ? DUPLEX_FULL :
4290                                                    DUPLEX_HALF;
4291                                         retval = e1000_set_spd_dplx(adapter,
4292                                                                     spddplx);
4293                                         if (retval)
4294                                                 return retval;
4295                                 }
4296                                 if (netif_running(adapter->netdev))
4297                                         e1000_reinit_locked(adapter);
4298                                 else
4299                                         e1000_reset(adapter);
4300                                 break;
4301                         case M88E1000_PHY_SPEC_CTRL:
4302                         case M88E1000_EXT_PHY_SPEC_CTRL:
4303                                 if (e1000_phy_reset(hw))
4304                                         return -EIO;
4305                                 break;
4306                         }
4307                 } else {
4308                         switch (data->reg_num) {
4309                         case PHY_CTRL:
4310                                 if (mii_reg & MII_CR_POWER_DOWN)
4311                                         break;
4312                                 if (netif_running(adapter->netdev))
4313                                         e1000_reinit_locked(adapter);
4314                                 else
4315                                         e1000_reset(adapter);
4316                                 break;
4317                         }
4318                 }
4319                 break;
4320         default:
4321                 return -EOPNOTSUPP;
4322         }
4323         return E1000_SUCCESS;
4324 }
4325
4326 void e1000_pci_set_mwi(struct e1000_hw *hw)
4327 {
4328         struct e1000_adapter *adapter = hw->back;
4329         int ret_val = pci_set_mwi(adapter->pdev);
4330
4331         if (ret_val)
4332                 DPRINTK(PROBE, ERR, "Error in setting MWI\n");
4333 }
4334
4335 void e1000_pci_clear_mwi(struct e1000_hw *hw)
4336 {
4337         struct e1000_adapter *adapter = hw->back;
4338
4339         pci_clear_mwi(adapter->pdev);
4340 }
4341
4342 int e1000_pcix_get_mmrbc(struct e1000_hw *hw)
4343 {
4344         struct e1000_adapter *adapter = hw->back;
4345         return pcix_get_mmrbc(adapter->pdev);
4346 }
4347
4348 void e1000_pcix_set_mmrbc(struct e1000_hw *hw, int mmrbc)
4349 {
4350         struct e1000_adapter *adapter = hw->back;
4351         pcix_set_mmrbc(adapter->pdev, mmrbc);
4352 }
4353
4354 void e1000_io_write(struct e1000_hw *hw, unsigned long port, u32 value)
4355 {
4356         outl(value, port);
4357 }
4358
4359 static void e1000_vlan_rx_register(struct net_device *netdev,
4360                                    struct vlan_group *grp)
4361 {
4362         struct e1000_adapter *adapter = netdev_priv(netdev);
4363         struct e1000_hw *hw = &adapter->hw;
4364         u32 ctrl, rctl;
4365
4366         if (!test_bit(__E1000_DOWN, &adapter->flags))
4367                 e1000_irq_disable(adapter);
4368         adapter->vlgrp = grp;
4369
4370         if (grp) {
4371                 /* enable VLAN tag insert/strip */
4372                 ctrl = er32(CTRL);
4373                 ctrl |= E1000_CTRL_VME;
4374                 ew32(CTRL, ctrl);
4375
4376                 /* enable VLAN receive filtering */
4377                 rctl = er32(RCTL);
4378                 rctl &= ~E1000_RCTL_CFIEN;
4379                 if (!(netdev->flags & IFF_PROMISC))
4380                         rctl |= E1000_RCTL_VFE;
4381                 ew32(RCTL, rctl);
4382                 e1000_update_mng_vlan(adapter);
4383         } else {
4384                 /* disable VLAN tag insert/strip */
4385                 ctrl = er32(CTRL);
4386                 ctrl &= ~E1000_CTRL_VME;
4387                 ew32(CTRL, ctrl);
4388
4389                 /* disable VLAN receive filtering */
4390                 rctl = er32(RCTL);
4391                 rctl &= ~E1000_RCTL_VFE;
4392                 ew32(RCTL, rctl);
4393
4394                 if (adapter->mng_vlan_id != (u16)E1000_MNG_VLAN_NONE) {
4395                         e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
4396                         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
4397                 }
4398         }
4399
4400         if (!test_bit(__E1000_DOWN, &adapter->flags))
4401                 e1000_irq_enable(adapter);
4402 }
4403
4404 static void e1000_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
4405 {
4406         struct e1000_adapter *adapter = netdev_priv(netdev);
4407         struct e1000_hw *hw = &adapter->hw;
4408         u32 vfta, index;
4409
4410         if ((hw->mng_cookie.status &
4411              E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT) &&
4412             (vid == adapter->mng_vlan_id))
4413                 return;
4414         /* add VID to filter table */
4415         index = (vid >> 5) & 0x7F;
4416         vfta = E1000_READ_REG_ARRAY(hw, VFTA, index);
4417         vfta |= (1 << (vid & 0x1F));
4418         e1000_write_vfta(hw, index, vfta);
4419 }
4420
4421 static void e1000_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
4422 {
4423         struct e1000_adapter *adapter = netdev_priv(netdev);
4424         struct e1000_hw *hw = &adapter->hw;
4425         u32 vfta, index;
4426
4427         if (!test_bit(__E1000_DOWN, &adapter->flags))
4428                 e1000_irq_disable(adapter);
4429         vlan_group_set_device(adapter->vlgrp, vid, NULL);
4430         if (!test_bit(__E1000_DOWN, &adapter->flags))
4431                 e1000_irq_enable(adapter);
4432
4433         /* remove VID from filter table */
4434         index = (vid >> 5) & 0x7F;
4435         vfta = E1000_READ_REG_ARRAY(hw, VFTA, index);
4436         vfta &= ~(1 << (vid & 0x1F));
4437         e1000_write_vfta(hw, index, vfta);
4438 }
4439
4440 static void e1000_restore_vlan(struct e1000_adapter *adapter)
4441 {
4442         e1000_vlan_rx_register(adapter->netdev, adapter->vlgrp);
4443
4444         if (adapter->vlgrp) {
4445                 u16 vid;
4446                 for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
4447                         if (!vlan_group_get_device(adapter->vlgrp, vid))
4448                                 continue;
4449                         e1000_vlan_rx_add_vid(adapter->netdev, vid);
4450                 }
4451         }
4452 }
4453
4454 int e1000_set_spd_dplx(struct e1000_adapter *adapter, u16 spddplx)
4455 {
4456         struct e1000_hw *hw = &adapter->hw;
4457
4458         hw->autoneg = 0;
4459
4460         /* Fiber NICs only allow 1000 gbps Full duplex */
4461         if ((hw->media_type == e1000_media_type_fiber) &&
4462                 spddplx != (SPEED_1000 + DUPLEX_FULL)) {
4463                 DPRINTK(PROBE, ERR, "Unsupported Speed/Duplex configuration\n");
4464                 return -EINVAL;
4465         }
4466
4467         switch (spddplx) {
4468         case SPEED_10 + DUPLEX_HALF:
4469                 hw->forced_speed_duplex = e1000_10_half;
4470                 break;
4471         case SPEED_10 + DUPLEX_FULL:
4472                 hw->forced_speed_duplex = e1000_10_full;
4473                 break;
4474         case SPEED_100 + DUPLEX_HALF:
4475                 hw->forced_speed_duplex = e1000_100_half;
4476                 break;
4477         case SPEED_100 + DUPLEX_FULL:
4478                 hw->forced_speed_duplex = e1000_100_full;
4479                 break;
4480         case SPEED_1000 + DUPLEX_FULL:
4481                 hw->autoneg = 1;
4482                 hw->autoneg_advertised = ADVERTISE_1000_FULL;
4483                 break;
4484         case SPEED_1000 + DUPLEX_HALF: /* not supported */
4485         default:
4486                 DPRINTK(PROBE, ERR, "Unsupported Speed/Duplex configuration\n");
4487                 return -EINVAL;
4488         }
4489         return 0;
4490 }
4491
4492 static int __e1000_shutdown(struct pci_dev *pdev, bool *enable_wake)
4493 {
4494         struct net_device *netdev = pci_get_drvdata(pdev);
4495         struct e1000_adapter *adapter = netdev_priv(netdev);
4496         struct e1000_hw *hw = &adapter->hw;
4497         u32 ctrl, ctrl_ext, rctl, status;
4498         u32 wufc = adapter->wol;
4499 #ifdef CONFIG_PM
4500         int retval = 0;
4501 #endif
4502
4503         netif_device_detach(netdev);
4504
4505         if (netif_running(netdev)) {
4506                 WARN_ON(test_bit(__E1000_RESETTING, &adapter->flags));
4507                 e1000_down(adapter);
4508         }
4509
4510 #ifdef CONFIG_PM
4511         retval = pci_save_state(pdev);
4512         if (retval)
4513                 return retval;
4514 #endif
4515
4516         status = er32(STATUS);
4517         if (status & E1000_STATUS_LU)
4518                 wufc &= ~E1000_WUFC_LNKC;
4519
4520         if (wufc) {
4521                 e1000_setup_rctl(adapter);
4522                 e1000_set_rx_mode(netdev);
4523
4524                 /* turn on all-multi mode if wake on multicast is enabled */
4525                 if (wufc & E1000_WUFC_MC) {
4526                         rctl = er32(RCTL);
4527                         rctl |= E1000_RCTL_MPE;
4528                         ew32(RCTL, rctl);
4529                 }
4530
4531                 if (hw->mac_type >= e1000_82540) {
4532                         ctrl = er32(CTRL);
4533                         /* advertise wake from D3Cold */
4534                         #define E1000_CTRL_ADVD3WUC 0x00100000
4535                         /* phy power management enable */
4536                         #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
4537                         ctrl |= E1000_CTRL_ADVD3WUC |
4538                                 E1000_CTRL_EN_PHY_PWR_MGMT;
4539                         ew32(CTRL, ctrl);
4540                 }
4541
4542                 if (hw->media_type == e1000_media_type_fiber ||
4543                     hw->media_type == e1000_media_type_internal_serdes) {
4544                         /* keep the laser running in D3 */
4545                         ctrl_ext = er32(CTRL_EXT);
4546                         ctrl_ext |= E1000_CTRL_EXT_SDP7_DATA;
4547                         ew32(CTRL_EXT, ctrl_ext);
4548                 }
4549
4550                 ew32(WUC, E1000_WUC_PME_EN);
4551                 ew32(WUFC, wufc);
4552         } else {
4553                 ew32(WUC, 0);
4554                 ew32(WUFC, 0);
4555         }
4556
4557         e1000_release_manageability(adapter);
4558
4559         *enable_wake = !!wufc;
4560
4561         /* make sure adapter isn't asleep if manageability is enabled */
4562         if (adapter->en_mng_pt)
4563                 *enable_wake = true;
4564
4565         if (netif_running(netdev))
4566                 e1000_free_irq(adapter);
4567
4568         pci_disable_device(pdev);
4569
4570         return 0;
4571 }
4572
4573 #ifdef CONFIG_PM
4574 static int e1000_suspend(struct pci_dev *pdev, pm_message_t state)
4575 {
4576         int retval;
4577         bool wake;
4578
4579         retval = __e1000_shutdown(pdev, &wake);
4580         if (retval)
4581                 return retval;
4582
4583         if (wake) {
4584                 pci_prepare_to_sleep(pdev);
4585         } else {
4586                 pci_wake_from_d3(pdev, false);
4587                 pci_set_power_state(pdev, PCI_D3hot);
4588         }
4589
4590         return 0;
4591 }
4592
4593 static int e1000_resume(struct pci_dev *pdev)
4594 {
4595         struct net_device *netdev = pci_get_drvdata(pdev);
4596         struct e1000_adapter *adapter = netdev_priv(netdev);
4597         struct e1000_hw *hw = &adapter->hw;
4598         u32 err;
4599
4600         pci_set_power_state(pdev, PCI_D0);
4601         pci_restore_state(pdev);
4602         pci_save_state(pdev);
4603
4604         if (adapter->need_ioport)
4605                 err = pci_enable_device(pdev);
4606         else
4607                 err = pci_enable_device_mem(pdev);
4608         if (err) {
4609                 printk(KERN_ERR "e1000: Cannot enable PCI device from suspend\n");
4610                 return err;
4611         }
4612         pci_set_master(pdev);
4613
4614         pci_enable_wake(pdev, PCI_D3hot, 0);
4615         pci_enable_wake(pdev, PCI_D3cold, 0);
4616
4617         if (netif_running(netdev)) {
4618                 err = e1000_request_irq(adapter);
4619                 if (err)
4620                         return err;
4621         }
4622
4623         e1000_power_up_phy(adapter);
4624         e1000_reset(adapter);
4625         ew32(WUS, ~0);
4626
4627         e1000_init_manageability(adapter);
4628
4629         if (netif_running(netdev))
4630                 e1000_up(adapter);
4631
4632         netif_device_attach(netdev);
4633
4634         return 0;
4635 }
4636 #endif
4637
4638 static void e1000_shutdown(struct pci_dev *pdev)
4639 {
4640         bool wake;
4641
4642         __e1000_shutdown(pdev, &wake);
4643
4644         if (system_state == SYSTEM_POWER_OFF) {
4645                 pci_wake_from_d3(pdev, wake);
4646                 pci_set_power_state(pdev, PCI_D3hot);
4647         }
4648 }
4649
4650 #ifdef CONFIG_NET_POLL_CONTROLLER
4651 /*
4652  * Polling 'interrupt' - used by things like netconsole to send skbs
4653  * without having to re-enable interrupts. It's not called while
4654  * the interrupt routine is executing.
4655  */
4656 static void e1000_netpoll(struct net_device *netdev)
4657 {
4658         struct e1000_adapter *adapter = netdev_priv(netdev);
4659
4660         disable_irq(adapter->pdev->irq);
4661         e1000_intr(adapter->pdev->irq, netdev);
4662         enable_irq(adapter->pdev->irq);
4663 }
4664 #endif
4665
4666 /**
4667  * e1000_io_error_detected - called when PCI error is detected
4668  * @pdev: Pointer to PCI device
4669  * @state: The current pci connection state
4670  *
4671  * This function is called after a PCI bus error affecting
4672  * this device has been detected.
4673  */
4674 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
4675                                                 pci_channel_state_t state)
4676 {
4677         struct net_device *netdev = pci_get_drvdata(pdev);
4678         struct e1000_adapter *adapter = netdev_priv(netdev);
4679
4680         netif_device_detach(netdev);
4681
4682         if (state == pci_channel_io_perm_failure)
4683                 return PCI_ERS_RESULT_DISCONNECT;
4684
4685         if (netif_running(netdev))
4686                 e1000_down(adapter);
4687         pci_disable_device(pdev);
4688
4689         /* Request a slot slot reset. */
4690         return PCI_ERS_RESULT_NEED_RESET;
4691 }
4692
4693 /**
4694  * e1000_io_slot_reset - called after the pci bus has been reset.
4695  * @pdev: Pointer to PCI device
4696  *
4697  * Restart the card from scratch, as if from a cold-boot. Implementation
4698  * resembles the first-half of the e1000_resume routine.
4699  */
4700 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
4701 {
4702         struct net_device *netdev = pci_get_drvdata(pdev);
4703         struct e1000_adapter *adapter = netdev_priv(netdev);
4704         struct e1000_hw *hw = &adapter->hw;
4705         int err;
4706
4707         if (adapter->need_ioport)
4708                 err = pci_enable_device(pdev);
4709         else
4710                 err = pci_enable_device_mem(pdev);
4711         if (err) {
4712                 printk(KERN_ERR "e1000: Cannot re-enable PCI device after reset.\n");
4713                 return PCI_ERS_RESULT_DISCONNECT;
4714         }
4715         pci_set_master(pdev);
4716
4717         pci_enable_wake(pdev, PCI_D3hot, 0);
4718         pci_enable_wake(pdev, PCI_D3cold, 0);
4719
4720         e1000_reset(adapter);
4721         ew32(WUS, ~0);
4722
4723         return PCI_ERS_RESULT_RECOVERED;
4724 }
4725
4726 /**
4727  * e1000_io_resume - called when traffic can start flowing again.
4728  * @pdev: Pointer to PCI device
4729  *
4730  * This callback is called when the error recovery driver tells us that
4731  * its OK to resume normal operation. Implementation resembles the
4732  * second-half of the e1000_resume routine.
4733  */
4734 static void e1000_io_resume(struct pci_dev *pdev)
4735 {
4736         struct net_device *netdev = pci_get_drvdata(pdev);
4737         struct e1000_adapter *adapter = netdev_priv(netdev);
4738
4739         e1000_init_manageability(adapter);
4740
4741         if (netif_running(netdev)) {
4742                 if (e1000_up(adapter)) {
4743                         printk("e1000: can't bring device back up after reset\n");
4744                         return;
4745                 }
4746         }
4747
4748         netif_device_attach(netdev);
4749 }
4750
4751 /* e1000_main.c */