]> git.karo-electronics.de Git - mv-sheeva.git/blob - drivers/net/e1000/e1000_main.c
Merge branch 'for-patrick' of git://git.kernel.org/pub/scm/linux/kernel/git/horms...
[mv-sheeva.git] / 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-k6-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_get_hw_dev - return device
218  * used by hardware layer to print debugging information
219  *
220  **/
221 struct net_device *e1000_get_hw_dev(struct e1000_hw *hw)
222 {
223         struct e1000_adapter *adapter = hw->back;
224         return adapter->netdev;
225 }
226
227 /**
228  * e1000_init_module - Driver Registration Routine
229  *
230  * e1000_init_module is the first routine called when the driver is
231  * loaded. All it does is register with the PCI subsystem.
232  **/
233
234 static int __init e1000_init_module(void)
235 {
236         int ret;
237         pr_info("%s - version %s\n", e1000_driver_string, e1000_driver_version);
238
239         pr_info("%s\n", e1000_copyright);
240
241         ret = pci_register_driver(&e1000_driver);
242         if (copybreak != COPYBREAK_DEFAULT) {
243                 if (copybreak == 0)
244                         pr_info("copybreak disabled\n");
245                 else
246                         pr_info("copybreak enabled for "
247                                    "packets <= %u bytes\n", copybreak);
248         }
249         return ret;
250 }
251
252 module_init(e1000_init_module);
253
254 /**
255  * e1000_exit_module - Driver Exit Cleanup Routine
256  *
257  * e1000_exit_module is called just before the driver is removed
258  * from memory.
259  **/
260
261 static void __exit e1000_exit_module(void)
262 {
263         pci_unregister_driver(&e1000_driver);
264 }
265
266 module_exit(e1000_exit_module);
267
268 static int e1000_request_irq(struct e1000_adapter *adapter)
269 {
270         struct net_device *netdev = adapter->netdev;
271         irq_handler_t handler = e1000_intr;
272         int irq_flags = IRQF_SHARED;
273         int err;
274
275         err = request_irq(adapter->pdev->irq, handler, irq_flags, netdev->name,
276                           netdev);
277         if (err) {
278                 e_err(probe, "Unable to allocate interrupt Error: %d\n", err);
279         }
280
281         return err;
282 }
283
284 static void e1000_free_irq(struct e1000_adapter *adapter)
285 {
286         struct net_device *netdev = adapter->netdev;
287
288         free_irq(adapter->pdev->irq, netdev);
289 }
290
291 /**
292  * e1000_irq_disable - Mask off interrupt generation on the NIC
293  * @adapter: board private structure
294  **/
295
296 static void e1000_irq_disable(struct e1000_adapter *adapter)
297 {
298         struct e1000_hw *hw = &adapter->hw;
299
300         ew32(IMC, ~0);
301         E1000_WRITE_FLUSH();
302         synchronize_irq(adapter->pdev->irq);
303 }
304
305 /**
306  * e1000_irq_enable - Enable default interrupt generation settings
307  * @adapter: board private structure
308  **/
309
310 static void e1000_irq_enable(struct e1000_adapter *adapter)
311 {
312         struct e1000_hw *hw = &adapter->hw;
313
314         ew32(IMS, IMS_ENABLE_MASK);
315         E1000_WRITE_FLUSH();
316 }
317
318 static void e1000_update_mng_vlan(struct e1000_adapter *adapter)
319 {
320         struct e1000_hw *hw = &adapter->hw;
321         struct net_device *netdev = adapter->netdev;
322         u16 vid = hw->mng_cookie.vlan_id;
323         u16 old_vid = adapter->mng_vlan_id;
324         if (adapter->vlgrp) {
325                 if (!vlan_group_get_device(adapter->vlgrp, vid)) {
326                         if (hw->mng_cookie.status &
327                                 E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT) {
328                                 e1000_vlan_rx_add_vid(netdev, vid);
329                                 adapter->mng_vlan_id = vid;
330                         } else
331                                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
332
333                         if ((old_vid != (u16)E1000_MNG_VLAN_NONE) &&
334                                         (vid != old_vid) &&
335                             !vlan_group_get_device(adapter->vlgrp, old_vid))
336                                 e1000_vlan_rx_kill_vid(netdev, old_vid);
337                 } else
338                         adapter->mng_vlan_id = vid;
339         }
340 }
341
342 static void e1000_init_manageability(struct e1000_adapter *adapter)
343 {
344         struct e1000_hw *hw = &adapter->hw;
345
346         if (adapter->en_mng_pt) {
347                 u32 manc = er32(MANC);
348
349                 /* disable hardware interception of ARP */
350                 manc &= ~(E1000_MANC_ARP_EN);
351
352                 ew32(MANC, manc);
353         }
354 }
355
356 static void e1000_release_manageability(struct e1000_adapter *adapter)
357 {
358         struct e1000_hw *hw = &adapter->hw;
359
360         if (adapter->en_mng_pt) {
361                 u32 manc = er32(MANC);
362
363                 /* re-enable hardware interception of ARP */
364                 manc |= E1000_MANC_ARP_EN;
365
366                 ew32(MANC, manc);
367         }
368 }
369
370 /**
371  * e1000_configure - configure the hardware for RX and TX
372  * @adapter = private board structure
373  **/
374 static void e1000_configure(struct e1000_adapter *adapter)
375 {
376         struct net_device *netdev = adapter->netdev;
377         int i;
378
379         e1000_set_rx_mode(netdev);
380
381         e1000_restore_vlan(adapter);
382         e1000_init_manageability(adapter);
383
384         e1000_configure_tx(adapter);
385         e1000_setup_rctl(adapter);
386         e1000_configure_rx(adapter);
387         /* call E1000_DESC_UNUSED which always leaves
388          * at least 1 descriptor unused to make sure
389          * next_to_use != next_to_clean */
390         for (i = 0; i < adapter->num_rx_queues; i++) {
391                 struct e1000_rx_ring *ring = &adapter->rx_ring[i];
392                 adapter->alloc_rx_buf(adapter, ring,
393                                       E1000_DESC_UNUSED(ring));
394         }
395 }
396
397 int e1000_up(struct e1000_adapter *adapter)
398 {
399         struct e1000_hw *hw = &adapter->hw;
400
401         /* hardware has been reset, we need to reload some things */
402         e1000_configure(adapter);
403
404         clear_bit(__E1000_DOWN, &adapter->flags);
405
406         napi_enable(&adapter->napi);
407
408         e1000_irq_enable(adapter);
409
410         netif_wake_queue(adapter->netdev);
411
412         /* fire a link change interrupt to start the watchdog */
413         ew32(ICS, E1000_ICS_LSC);
414         return 0;
415 }
416
417 /**
418  * e1000_power_up_phy - restore link in case the phy was powered down
419  * @adapter: address of board private structure
420  *
421  * The phy may be powered down to save power and turn off link when the
422  * driver is unloaded and wake on lan is not enabled (among others)
423  * *** this routine MUST be followed by a call to e1000_reset ***
424  *
425  **/
426
427 void e1000_power_up_phy(struct e1000_adapter *adapter)
428 {
429         struct e1000_hw *hw = &adapter->hw;
430         u16 mii_reg = 0;
431
432         /* Just clear the power down bit to wake the phy back up */
433         if (hw->media_type == e1000_media_type_copper) {
434                 /* according to the manual, the phy will retain its
435                  * settings across a power-down/up cycle */
436                 e1000_read_phy_reg(hw, PHY_CTRL, &mii_reg);
437                 mii_reg &= ~MII_CR_POWER_DOWN;
438                 e1000_write_phy_reg(hw, PHY_CTRL, mii_reg);
439         }
440 }
441
442 static void e1000_power_down_phy(struct e1000_adapter *adapter)
443 {
444         struct e1000_hw *hw = &adapter->hw;
445
446         /* Power down the PHY so no link is implied when interface is down *
447          * The PHY cannot be powered down if any of the following is true *
448          * (a) WoL is enabled
449          * (b) AMT is active
450          * (c) SoL/IDER session is active */
451         if (!adapter->wol && hw->mac_type >= e1000_82540 &&
452            hw->media_type == e1000_media_type_copper) {
453                 u16 mii_reg = 0;
454
455                 switch (hw->mac_type) {
456                 case e1000_82540:
457                 case e1000_82545:
458                 case e1000_82545_rev_3:
459                 case e1000_82546:
460                 case e1000_82546_rev_3:
461                 case e1000_82541:
462                 case e1000_82541_rev_2:
463                 case e1000_82547:
464                 case e1000_82547_rev_2:
465                         if (er32(MANC) & E1000_MANC_SMBUS_EN)
466                                 goto out;
467                         break;
468                 default:
469                         goto out;
470                 }
471                 e1000_read_phy_reg(hw, PHY_CTRL, &mii_reg);
472                 mii_reg |= MII_CR_POWER_DOWN;
473                 e1000_write_phy_reg(hw, PHY_CTRL, mii_reg);
474                 mdelay(1);
475         }
476 out:
477         return;
478 }
479
480 void e1000_down(struct e1000_adapter *adapter)
481 {
482         struct e1000_hw *hw = &adapter->hw;
483         struct net_device *netdev = adapter->netdev;
484         u32 rctl, tctl;
485
486         /* signal that we're down so the interrupt handler does not
487          * reschedule our watchdog timer */
488         set_bit(__E1000_DOWN, &adapter->flags);
489
490         /* disable receives in the hardware */
491         rctl = er32(RCTL);
492         ew32(RCTL, rctl & ~E1000_RCTL_EN);
493         /* flush and sleep below */
494
495         netif_tx_disable(netdev);
496
497         /* disable transmits in the hardware */
498         tctl = er32(TCTL);
499         tctl &= ~E1000_TCTL_EN;
500         ew32(TCTL, tctl);
501         /* flush both disables and wait for them to finish */
502         E1000_WRITE_FLUSH();
503         msleep(10);
504
505         napi_disable(&adapter->napi);
506
507         e1000_irq_disable(adapter);
508
509         del_timer_sync(&adapter->tx_fifo_stall_timer);
510         del_timer_sync(&adapter->watchdog_timer);
511         del_timer_sync(&adapter->phy_info_timer);
512
513         adapter->link_speed = 0;
514         adapter->link_duplex = 0;
515         netif_carrier_off(netdev);
516
517         e1000_reset(adapter);
518         e1000_clean_all_tx_rings(adapter);
519         e1000_clean_all_rx_rings(adapter);
520 }
521
522 void e1000_reinit_locked(struct e1000_adapter *adapter)
523 {
524         WARN_ON(in_interrupt());
525         while (test_and_set_bit(__E1000_RESETTING, &adapter->flags))
526                 msleep(1);
527         e1000_down(adapter);
528         e1000_up(adapter);
529         clear_bit(__E1000_RESETTING, &adapter->flags);
530 }
531
532 void e1000_reset(struct e1000_adapter *adapter)
533 {
534         struct e1000_hw *hw = &adapter->hw;
535         u32 pba = 0, tx_space, min_tx_space, min_rx_space;
536         bool legacy_pba_adjust = false;
537         u16 hwm;
538
539         /* Repartition Pba for greater than 9k mtu
540          * To take effect CTRL.RST is required.
541          */
542
543         switch (hw->mac_type) {
544         case e1000_82542_rev2_0:
545         case e1000_82542_rev2_1:
546         case e1000_82543:
547         case e1000_82544:
548         case e1000_82540:
549         case e1000_82541:
550         case e1000_82541_rev_2:
551                 legacy_pba_adjust = true;
552                 pba = E1000_PBA_48K;
553                 break;
554         case e1000_82545:
555         case e1000_82545_rev_3:
556         case e1000_82546:
557         case e1000_82546_rev_3:
558                 pba = E1000_PBA_48K;
559                 break;
560         case e1000_82547:
561         case e1000_82547_rev_2:
562                 legacy_pba_adjust = true;
563                 pba = E1000_PBA_30K;
564                 break;
565         case e1000_undefined:
566         case e1000_num_macs:
567                 break;
568         }
569
570         if (legacy_pba_adjust) {
571                 if (hw->max_frame_size > E1000_RXBUFFER_8192)
572                         pba -= 8; /* allocate more FIFO for Tx */
573
574                 if (hw->mac_type == e1000_82547) {
575                         adapter->tx_fifo_head = 0;
576                         adapter->tx_head_addr = pba << E1000_TX_HEAD_ADDR_SHIFT;
577                         adapter->tx_fifo_size =
578                                 (E1000_PBA_40K - pba) << E1000_PBA_BYTES_SHIFT;
579                         atomic_set(&adapter->tx_fifo_stall, 0);
580                 }
581         } else if (hw->max_frame_size >  ETH_FRAME_LEN + ETH_FCS_LEN) {
582                 /* adjust PBA for jumbo frames */
583                 ew32(PBA, pba);
584
585                 /* To maintain wire speed transmits, the Tx FIFO should be
586                  * large enough to accommodate two full transmit packets,
587                  * rounded up to the next 1KB and expressed in KB.  Likewise,
588                  * the Rx FIFO should be large enough to accommodate at least
589                  * one full receive packet and is similarly rounded up and
590                  * expressed in KB. */
591                 pba = er32(PBA);
592                 /* upper 16 bits has Tx packet buffer allocation size in KB */
593                 tx_space = pba >> 16;
594                 /* lower 16 bits has Rx packet buffer allocation size in KB */
595                 pba &= 0xffff;
596                 /*
597                  * the tx fifo also stores 16 bytes of information about the tx
598                  * but don't include ethernet FCS because hardware appends it
599                  */
600                 min_tx_space = (hw->max_frame_size +
601                                 sizeof(struct e1000_tx_desc) -
602                                 ETH_FCS_LEN) * 2;
603                 min_tx_space = ALIGN(min_tx_space, 1024);
604                 min_tx_space >>= 10;
605                 /* software strips receive CRC, so leave room for it */
606                 min_rx_space = hw->max_frame_size;
607                 min_rx_space = ALIGN(min_rx_space, 1024);
608                 min_rx_space >>= 10;
609
610                 /* If current Tx allocation is less than the min Tx FIFO size,
611                  * and the min Tx FIFO size is less than the current Rx FIFO
612                  * allocation, take space away from current Rx allocation */
613                 if (tx_space < min_tx_space &&
614                     ((min_tx_space - tx_space) < pba)) {
615                         pba = pba - (min_tx_space - tx_space);
616
617                         /* PCI/PCIx hardware has PBA alignment constraints */
618                         switch (hw->mac_type) {
619                         case e1000_82545 ... e1000_82546_rev_3:
620                                 pba &= ~(E1000_PBA_8K - 1);
621                                 break;
622                         default:
623                                 break;
624                         }
625
626                         /* if short on rx space, rx wins and must trump tx
627                          * adjustment or use Early Receive if available */
628                         if (pba < min_rx_space)
629                                 pba = min_rx_space;
630                 }
631         }
632
633         ew32(PBA, pba);
634
635         /*
636          * flow control settings:
637          * The high water mark must be low enough to fit one full frame
638          * (or the size used for early receive) above it in the Rx FIFO.
639          * Set it to the lower of:
640          * - 90% of the Rx FIFO size, and
641          * - the full Rx FIFO size minus the early receive size (for parts
642          *   with ERT support assuming ERT set to E1000_ERT_2048), or
643          * - the full Rx FIFO size minus one full frame
644          */
645         hwm = min(((pba << 10) * 9 / 10),
646                   ((pba << 10) - hw->max_frame_size));
647
648         hw->fc_high_water = hwm & 0xFFF8;       /* 8-byte granularity */
649         hw->fc_low_water = hw->fc_high_water - 8;
650         hw->fc_pause_time = E1000_FC_PAUSE_TIME;
651         hw->fc_send_xon = 1;
652         hw->fc = hw->original_fc;
653
654         /* Allow time for pending master requests to run */
655         e1000_reset_hw(hw);
656         if (hw->mac_type >= e1000_82544)
657                 ew32(WUC, 0);
658
659         if (e1000_init_hw(hw))
660                 e_dev_err("Hardware Error\n");
661         e1000_update_mng_vlan(adapter);
662
663         /* if (adapter->hwflags & HWFLAGS_PHY_PWR_BIT) { */
664         if (hw->mac_type >= e1000_82544 &&
665             hw->autoneg == 1 &&
666             hw->autoneg_advertised == ADVERTISE_1000_FULL) {
667                 u32 ctrl = er32(CTRL);
668                 /* clear phy power management bit if we are in gig only mode,
669                  * which if enabled will attempt negotiation to 100Mb, which
670                  * can cause a loss of link at power off or driver unload */
671                 ctrl &= ~E1000_CTRL_SWDPIN3;
672                 ew32(CTRL, ctrl);
673         }
674
675         /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
676         ew32(VET, ETHERNET_IEEE_VLAN_TYPE);
677
678         e1000_reset_adaptive(hw);
679         e1000_phy_get_info(hw, &adapter->phy_info);
680
681         e1000_release_manageability(adapter);
682 }
683
684 /**
685  *  Dump the eeprom for users having checksum issues
686  **/
687 static void e1000_dump_eeprom(struct e1000_adapter *adapter)
688 {
689         struct net_device *netdev = adapter->netdev;
690         struct ethtool_eeprom eeprom;
691         const struct ethtool_ops *ops = netdev->ethtool_ops;
692         u8 *data;
693         int i;
694         u16 csum_old, csum_new = 0;
695
696         eeprom.len = ops->get_eeprom_len(netdev);
697         eeprom.offset = 0;
698
699         data = kmalloc(eeprom.len, GFP_KERNEL);
700         if (!data) {
701                 pr_err("Unable to allocate memory to dump EEPROM data\n");
702                 return;
703         }
704
705         ops->get_eeprom(netdev, &eeprom, data);
706
707         csum_old = (data[EEPROM_CHECKSUM_REG * 2]) +
708                    (data[EEPROM_CHECKSUM_REG * 2 + 1] << 8);
709         for (i = 0; i < EEPROM_CHECKSUM_REG * 2; i += 2)
710                 csum_new += data[i] + (data[i + 1] << 8);
711         csum_new = EEPROM_SUM - csum_new;
712
713         pr_err("/*********************/\n");
714         pr_err("Current EEPROM Checksum : 0x%04x\n", csum_old);
715         pr_err("Calculated              : 0x%04x\n", csum_new);
716
717         pr_err("Offset    Values\n");
718         pr_err("========  ======\n");
719         print_hex_dump(KERN_ERR, "", DUMP_PREFIX_OFFSET, 16, 1, data, 128, 0);
720
721         pr_err("Include this output when contacting your support provider.\n");
722         pr_err("This is not a software error! Something bad happened to\n");
723         pr_err("your hardware or EEPROM image. Ignoring this problem could\n");
724         pr_err("result in further problems, possibly loss of data,\n");
725         pr_err("corruption or system hangs!\n");
726         pr_err("The MAC Address will be reset to 00:00:00:00:00:00,\n");
727         pr_err("which is invalid and requires you to set the proper MAC\n");
728         pr_err("address manually before continuing to enable this network\n");
729         pr_err("device. Please inspect the EEPROM dump and report the\n");
730         pr_err("issue to your hardware vendor or Intel Customer Support.\n");
731         pr_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_init_hw_struct - initialize members of hw struct
794  * @adapter: board private struct
795  * @hw: structure used by e1000_hw.c
796  *
797  * Factors out initialization of the e1000_hw struct to its own function
798  * that can be called very early at init (just after struct allocation).
799  * Fields are initialized based on PCI device information and
800  * OS network device settings (MTU size).
801  * Returns negative error codes if MAC type setup fails.
802  */
803 static int e1000_init_hw_struct(struct e1000_adapter *adapter,
804                                 struct e1000_hw *hw)
805 {
806         struct pci_dev *pdev = adapter->pdev;
807
808         /* PCI config space info */
809         hw->vendor_id = pdev->vendor;
810         hw->device_id = pdev->device;
811         hw->subsystem_vendor_id = pdev->subsystem_vendor;
812         hw->subsystem_id = pdev->subsystem_device;
813         hw->revision_id = pdev->revision;
814
815         pci_read_config_word(pdev, PCI_COMMAND, &hw->pci_cmd_word);
816
817         hw->max_frame_size = adapter->netdev->mtu +
818                              ENET_HEADER_SIZE + ETHERNET_FCS_SIZE;
819         hw->min_frame_size = MINIMUM_ETHERNET_FRAME_SIZE;
820
821         /* identify the MAC */
822         if (e1000_set_mac_type(hw)) {
823                 e_err(probe, "Unknown MAC Type\n");
824                 return -EIO;
825         }
826
827         switch (hw->mac_type) {
828         default:
829                 break;
830         case e1000_82541:
831         case e1000_82547:
832         case e1000_82541_rev_2:
833         case e1000_82547_rev_2:
834                 hw->phy_init_script = 1;
835                 break;
836         }
837
838         e1000_set_media_type(hw);
839         e1000_get_bus_info(hw);
840
841         hw->wait_autoneg_complete = false;
842         hw->tbi_compatibility_en = true;
843         hw->adaptive_ifs = true;
844
845         /* Copper options */
846
847         if (hw->media_type == e1000_media_type_copper) {
848                 hw->mdix = AUTO_ALL_MODES;
849                 hw->disable_polarity_correction = false;
850                 hw->master_slave = E1000_MASTER_SLAVE;
851         }
852
853         return 0;
854 }
855
856 /**
857  * e1000_probe - Device Initialization Routine
858  * @pdev: PCI device information struct
859  * @ent: entry in e1000_pci_tbl
860  *
861  * Returns 0 on success, negative on failure
862  *
863  * e1000_probe initializes an adapter identified by a pci_dev structure.
864  * The OS initialization, configuring of the adapter private structure,
865  * and a hardware reset occur.
866  **/
867 static int __devinit e1000_probe(struct pci_dev *pdev,
868                                  const struct pci_device_id *ent)
869 {
870         struct net_device *netdev;
871         struct e1000_adapter *adapter;
872         struct e1000_hw *hw;
873
874         static int cards_found = 0;
875         static int global_quad_port_a = 0; /* global ksp3 port a indication */
876         int i, err, pci_using_dac;
877         u16 eeprom_data = 0;
878         u16 eeprom_apme_mask = E1000_EEPROM_APME;
879         int bars, need_ioport;
880
881         /* do not allocate ioport bars when not needed */
882         need_ioport = e1000_is_need_ioport(pdev);
883         if (need_ioport) {
884                 bars = pci_select_bars(pdev, IORESOURCE_MEM | IORESOURCE_IO);
885                 err = pci_enable_device(pdev);
886         } else {
887                 bars = pci_select_bars(pdev, IORESOURCE_MEM);
888                 err = pci_enable_device_mem(pdev);
889         }
890         if (err)
891                 return err;
892
893         err = pci_request_selected_regions(pdev, bars, e1000_driver_name);
894         if (err)
895                 goto err_pci_reg;
896
897         pci_set_master(pdev);
898         err = pci_save_state(pdev);
899         if (err)
900                 goto err_alloc_etherdev;
901
902         err = -ENOMEM;
903         netdev = alloc_etherdev(sizeof(struct e1000_adapter));
904         if (!netdev)
905                 goto err_alloc_etherdev;
906
907         SET_NETDEV_DEV(netdev, &pdev->dev);
908
909         pci_set_drvdata(pdev, netdev);
910         adapter = netdev_priv(netdev);
911         adapter->netdev = netdev;
912         adapter->pdev = pdev;
913         adapter->msg_enable = (1 << debug) - 1;
914         adapter->bars = bars;
915         adapter->need_ioport = need_ioport;
916
917         hw = &adapter->hw;
918         hw->back = adapter;
919
920         err = -EIO;
921         hw->hw_addr = pci_ioremap_bar(pdev, BAR_0);
922         if (!hw->hw_addr)
923                 goto err_ioremap;
924
925         if (adapter->need_ioport) {
926                 for (i = BAR_1; i <= BAR_5; i++) {
927                         if (pci_resource_len(pdev, i) == 0)
928                                 continue;
929                         if (pci_resource_flags(pdev, i) & IORESOURCE_IO) {
930                                 hw->io_base = pci_resource_start(pdev, i);
931                                 break;
932                         }
933                 }
934         }
935
936         /* make ready for any if (hw->...) below */
937         err = e1000_init_hw_struct(adapter, hw);
938         if (err)
939                 goto err_sw_init;
940
941         /*
942          * there is a workaround being applied below that limits
943          * 64-bit DMA addresses to 64-bit hardware.  There are some
944          * 32-bit adapters that Tx hang when given 64-bit DMA addresses
945          */
946         pci_using_dac = 0;
947         if ((hw->bus_type == e1000_bus_type_pcix) &&
948             !dma_set_mask(&pdev->dev, DMA_BIT_MASK(64))) {
949                 /*
950                  * according to DMA-API-HOWTO, coherent calls will always
951                  * succeed if the set call did
952                  */
953                 dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64));
954                 pci_using_dac = 1;
955         } else if (!dma_set_mask(&pdev->dev, DMA_BIT_MASK(32))) {
956                 dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(32));
957         } else {
958                 pr_err("No usable DMA config, aborting\n");
959                 goto err_dma;
960         }
961
962         netdev->netdev_ops = &e1000_netdev_ops;
963         e1000_set_ethtool_ops(netdev);
964         netdev->watchdog_timeo = 5 * HZ;
965         netif_napi_add(netdev, &adapter->napi, e1000_clean, 64);
966
967         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
968
969         adapter->bd_number = cards_found;
970
971         /* setup the private structure */
972
973         err = e1000_sw_init(adapter);
974         if (err)
975                 goto err_sw_init;
976
977         err = -EIO;
978
979         if (hw->mac_type >= e1000_82543) {
980                 netdev->features = NETIF_F_SG |
981                                    NETIF_F_HW_CSUM |
982                                    NETIF_F_HW_VLAN_TX |
983                                    NETIF_F_HW_VLAN_RX |
984                                    NETIF_F_HW_VLAN_FILTER;
985         }
986
987         if ((hw->mac_type >= e1000_82544) &&
988            (hw->mac_type != e1000_82547))
989                 netdev->features |= NETIF_F_TSO;
990
991         if (pci_using_dac)
992                 netdev->features |= NETIF_F_HIGHDMA;
993
994         netdev->vlan_features |= NETIF_F_TSO;
995         netdev->vlan_features |= NETIF_F_HW_CSUM;
996         netdev->vlan_features |= NETIF_F_SG;
997
998         adapter->en_mng_pt = e1000_enable_mng_pass_thru(hw);
999
1000         /* initialize eeprom parameters */
1001         if (e1000_init_eeprom_params(hw)) {
1002                 e_err(probe, "EEPROM initialization failed\n");
1003                 goto err_eeprom;
1004         }
1005
1006         /* before reading the EEPROM, reset the controller to
1007          * put the device in a known good starting state */
1008
1009         e1000_reset_hw(hw);
1010
1011         /* make sure the EEPROM is good */
1012         if (e1000_validate_eeprom_checksum(hw) < 0) {
1013                 e_err(probe, "The EEPROM Checksum Is Not Valid\n");
1014                 e1000_dump_eeprom(adapter);
1015                 /*
1016                  * set MAC address to all zeroes to invalidate and temporary
1017                  * disable this device for the user. This blocks regular
1018                  * traffic while still permitting ethtool ioctls from reaching
1019                  * the hardware as well as allowing the user to run the
1020                  * interface after manually setting a hw addr using
1021                  * `ip set address`
1022                  */
1023                 memset(hw->mac_addr, 0, netdev->addr_len);
1024         } else {
1025                 /* copy the MAC address out of the EEPROM */
1026                 if (e1000_read_mac_addr(hw))
1027                         e_err(probe, "EEPROM Read Error\n");
1028         }
1029         /* don't block initalization here due to bad MAC address */
1030         memcpy(netdev->dev_addr, hw->mac_addr, netdev->addr_len);
1031         memcpy(netdev->perm_addr, hw->mac_addr, netdev->addr_len);
1032
1033         if (!is_valid_ether_addr(netdev->perm_addr))
1034                 e_err(probe, "Invalid MAC Address\n");
1035
1036         init_timer(&adapter->tx_fifo_stall_timer);
1037         adapter->tx_fifo_stall_timer.function = e1000_82547_tx_fifo_stall;
1038         adapter->tx_fifo_stall_timer.data = (unsigned long)adapter;
1039
1040         init_timer(&adapter->watchdog_timer);
1041         adapter->watchdog_timer.function = e1000_watchdog;
1042         adapter->watchdog_timer.data = (unsigned long) adapter;
1043
1044         init_timer(&adapter->phy_info_timer);
1045         adapter->phy_info_timer.function = e1000_update_phy_info;
1046         adapter->phy_info_timer.data = (unsigned long)adapter;
1047
1048         INIT_WORK(&adapter->reset_task, e1000_reset_task);
1049
1050         e1000_check_options(adapter);
1051
1052         /* Initial Wake on LAN setting
1053          * If APM wake is enabled in the EEPROM,
1054          * enable the ACPI Magic Packet filter
1055          */
1056
1057         switch (hw->mac_type) {
1058         case e1000_82542_rev2_0:
1059         case e1000_82542_rev2_1:
1060         case e1000_82543:
1061                 break;
1062         case e1000_82544:
1063                 e1000_read_eeprom(hw,
1064                         EEPROM_INIT_CONTROL2_REG, 1, &eeprom_data);
1065                 eeprom_apme_mask = E1000_EEPROM_82544_APM;
1066                 break;
1067         case e1000_82546:
1068         case e1000_82546_rev_3:
1069                 if (er32(STATUS) & E1000_STATUS_FUNC_1){
1070                         e1000_read_eeprom(hw,
1071                                 EEPROM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
1072                         break;
1073                 }
1074                 /* Fall Through */
1075         default:
1076                 e1000_read_eeprom(hw,
1077                         EEPROM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
1078                 break;
1079         }
1080         if (eeprom_data & eeprom_apme_mask)
1081                 adapter->eeprom_wol |= E1000_WUFC_MAG;
1082
1083         /* now that we have the eeprom settings, apply the special cases
1084          * where the eeprom may be wrong or the board simply won't support
1085          * wake on lan on a particular port */
1086         switch (pdev->device) {
1087         case E1000_DEV_ID_82546GB_PCIE:
1088                 adapter->eeprom_wol = 0;
1089                 break;
1090         case E1000_DEV_ID_82546EB_FIBER:
1091         case E1000_DEV_ID_82546GB_FIBER:
1092                 /* Wake events only supported on port A for dual fiber
1093                  * regardless of eeprom setting */
1094                 if (er32(STATUS) & E1000_STATUS_FUNC_1)
1095                         adapter->eeprom_wol = 0;
1096                 break;
1097         case E1000_DEV_ID_82546GB_QUAD_COPPER_KSP3:
1098                 /* if quad port adapter, disable WoL on all but port A */
1099                 if (global_quad_port_a != 0)
1100                         adapter->eeprom_wol = 0;
1101                 else
1102                         adapter->quad_port_a = 1;
1103                 /* Reset for multiple quad port adapters */
1104                 if (++global_quad_port_a == 4)
1105                         global_quad_port_a = 0;
1106                 break;
1107         }
1108
1109         /* initialize the wol settings based on the eeprom settings */
1110         adapter->wol = adapter->eeprom_wol;
1111         device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
1112
1113         /* reset the hardware with the new settings */
1114         e1000_reset(adapter);
1115
1116         strcpy(netdev->name, "eth%d");
1117         err = register_netdev(netdev);
1118         if (err)
1119                 goto err_register;
1120
1121         /* print bus type/speed/width info */
1122         e_info(probe, "(PCI%s:%dMHz:%d-bit) %pM\n",
1123                ((hw->bus_type == e1000_bus_type_pcix) ? "-X" : ""),
1124                ((hw->bus_speed == e1000_bus_speed_133) ? 133 :
1125                 (hw->bus_speed == e1000_bus_speed_120) ? 120 :
1126                 (hw->bus_speed == e1000_bus_speed_100) ? 100 :
1127                 (hw->bus_speed == e1000_bus_speed_66) ? 66 : 33),
1128                ((hw->bus_width == e1000_bus_width_64) ? 64 : 32),
1129                netdev->dev_addr);
1130
1131         /* carrier off reporting is important to ethtool even BEFORE open */
1132         netif_carrier_off(netdev);
1133
1134         e_info(probe, "Intel(R) PRO/1000 Network Connection\n");
1135
1136         cards_found++;
1137         return 0;
1138
1139 err_register:
1140 err_eeprom:
1141         e1000_phy_hw_reset(hw);
1142
1143         if (hw->flash_address)
1144                 iounmap(hw->flash_address);
1145         kfree(adapter->tx_ring);
1146         kfree(adapter->rx_ring);
1147 err_dma:
1148 err_sw_init:
1149         iounmap(hw->hw_addr);
1150 err_ioremap:
1151         free_netdev(netdev);
1152 err_alloc_etherdev:
1153         pci_release_selected_regions(pdev, bars);
1154 err_pci_reg:
1155         pci_disable_device(pdev);
1156         return err;
1157 }
1158
1159 /**
1160  * e1000_remove - Device Removal Routine
1161  * @pdev: PCI device information struct
1162  *
1163  * e1000_remove is called by the PCI subsystem to alert the driver
1164  * that it should release a PCI device.  The could be caused by a
1165  * Hot-Plug event, or because the driver is going to be removed from
1166  * memory.
1167  **/
1168
1169 static void __devexit e1000_remove(struct pci_dev *pdev)
1170 {
1171         struct net_device *netdev = pci_get_drvdata(pdev);
1172         struct e1000_adapter *adapter = netdev_priv(netdev);
1173         struct e1000_hw *hw = &adapter->hw;
1174
1175         set_bit(__E1000_DOWN, &adapter->flags);
1176         del_timer_sync(&adapter->tx_fifo_stall_timer);
1177         del_timer_sync(&adapter->watchdog_timer);
1178         del_timer_sync(&adapter->phy_info_timer);
1179
1180         cancel_work_sync(&adapter->reset_task);
1181
1182         e1000_release_manageability(adapter);
1183
1184         unregister_netdev(netdev);
1185
1186         e1000_phy_hw_reset(hw);
1187
1188         kfree(adapter->tx_ring);
1189         kfree(adapter->rx_ring);
1190
1191         iounmap(hw->hw_addr);
1192         if (hw->flash_address)
1193                 iounmap(hw->flash_address);
1194         pci_release_selected_regions(pdev, adapter->bars);
1195
1196         free_netdev(netdev);
1197
1198         pci_disable_device(pdev);
1199 }
1200
1201 /**
1202  * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
1203  * @adapter: board private structure to initialize
1204  *
1205  * e1000_sw_init initializes the Adapter private data structure.
1206  * e1000_init_hw_struct MUST be called before this function
1207  **/
1208
1209 static int __devinit e1000_sw_init(struct e1000_adapter *adapter)
1210 {
1211         adapter->rx_buffer_len = MAXIMUM_ETHERNET_VLAN_SIZE;
1212
1213         adapter->num_tx_queues = 1;
1214         adapter->num_rx_queues = 1;
1215
1216         if (e1000_alloc_queues(adapter)) {
1217                 e_err(probe, "Unable to allocate memory for queues\n");
1218                 return -ENOMEM;
1219         }
1220
1221         /* Explicitly disable IRQ since the NIC can be in any state. */
1222         e1000_irq_disable(adapter);
1223
1224         spin_lock_init(&adapter->stats_lock);
1225
1226         set_bit(__E1000_DOWN, &adapter->flags);
1227
1228         return 0;
1229 }
1230
1231 /**
1232  * e1000_alloc_queues - Allocate memory for all rings
1233  * @adapter: board private structure to initialize
1234  *
1235  * We allocate one ring per queue at run-time since we don't know the
1236  * number of queues at compile-time.
1237  **/
1238
1239 static int __devinit e1000_alloc_queues(struct e1000_adapter *adapter)
1240 {
1241         adapter->tx_ring = kcalloc(adapter->num_tx_queues,
1242                                    sizeof(struct e1000_tx_ring), GFP_KERNEL);
1243         if (!adapter->tx_ring)
1244                 return -ENOMEM;
1245
1246         adapter->rx_ring = kcalloc(adapter->num_rx_queues,
1247                                    sizeof(struct e1000_rx_ring), GFP_KERNEL);
1248         if (!adapter->rx_ring) {
1249                 kfree(adapter->tx_ring);
1250                 return -ENOMEM;
1251         }
1252
1253         return E1000_SUCCESS;
1254 }
1255
1256 /**
1257  * e1000_open - Called when a network interface is made active
1258  * @netdev: network interface device structure
1259  *
1260  * Returns 0 on success, negative value on failure
1261  *
1262  * The open entry point is called when a network interface is made
1263  * active by the system (IFF_UP).  At this point all resources needed
1264  * for transmit and receive operations are allocated, the interrupt
1265  * handler is registered with the OS, the watchdog timer is started,
1266  * and the stack is notified that the interface is ready.
1267  **/
1268
1269 static int e1000_open(struct net_device *netdev)
1270 {
1271         struct e1000_adapter *adapter = netdev_priv(netdev);
1272         struct e1000_hw *hw = &adapter->hw;
1273         int err;
1274
1275         /* disallow open during test */
1276         if (test_bit(__E1000_TESTING, &adapter->flags))
1277                 return -EBUSY;
1278
1279         netif_carrier_off(netdev);
1280
1281         /* allocate transmit descriptors */
1282         err = e1000_setup_all_tx_resources(adapter);
1283         if (err)
1284                 goto err_setup_tx;
1285
1286         /* allocate receive descriptors */
1287         err = e1000_setup_all_rx_resources(adapter);
1288         if (err)
1289                 goto err_setup_rx;
1290
1291         e1000_power_up_phy(adapter);
1292
1293         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
1294         if ((hw->mng_cookie.status &
1295                           E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT)) {
1296                 e1000_update_mng_vlan(adapter);
1297         }
1298
1299         /* before we allocate an interrupt, we must be ready to handle it.
1300          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
1301          * as soon as we call pci_request_irq, so we have to setup our
1302          * clean_rx handler before we do so.  */
1303         e1000_configure(adapter);
1304
1305         err = e1000_request_irq(adapter);
1306         if (err)
1307                 goto err_req_irq;
1308
1309         /* From here on the code is the same as e1000_up() */
1310         clear_bit(__E1000_DOWN, &adapter->flags);
1311
1312         napi_enable(&adapter->napi);
1313
1314         e1000_irq_enable(adapter);
1315
1316         netif_start_queue(netdev);
1317
1318         /* fire a link status change interrupt to start the watchdog */
1319         ew32(ICS, E1000_ICS_LSC);
1320
1321         return E1000_SUCCESS;
1322
1323 err_req_irq:
1324         e1000_power_down_phy(adapter);
1325         e1000_free_all_rx_resources(adapter);
1326 err_setup_rx:
1327         e1000_free_all_tx_resources(adapter);
1328 err_setup_tx:
1329         e1000_reset(adapter);
1330
1331         return err;
1332 }
1333
1334 /**
1335  * e1000_close - Disables a network interface
1336  * @netdev: network interface device structure
1337  *
1338  * Returns 0, this is not allowed to fail
1339  *
1340  * The close entry point is called when an interface is de-activated
1341  * by the OS.  The hardware is still under the drivers control, but
1342  * needs to be disabled.  A global MAC reset is issued to stop the
1343  * hardware, and all transmit and receive resources are freed.
1344  **/
1345
1346 static int e1000_close(struct net_device *netdev)
1347 {
1348         struct e1000_adapter *adapter = netdev_priv(netdev);
1349         struct e1000_hw *hw = &adapter->hw;
1350
1351         WARN_ON(test_bit(__E1000_RESETTING, &adapter->flags));
1352         e1000_down(adapter);
1353         e1000_power_down_phy(adapter);
1354         e1000_free_irq(adapter);
1355
1356         e1000_free_all_tx_resources(adapter);
1357         e1000_free_all_rx_resources(adapter);
1358
1359         /* kill manageability vlan ID if supported, but not if a vlan with
1360          * the same ID is registered on the host OS (let 8021q kill it) */
1361         if ((hw->mng_cookie.status &
1362                           E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT) &&
1363              !(adapter->vlgrp &&
1364                vlan_group_get_device(adapter->vlgrp, adapter->mng_vlan_id))) {
1365                 e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
1366         }
1367
1368         return 0;
1369 }
1370
1371 /**
1372  * e1000_check_64k_bound - check that memory doesn't cross 64kB boundary
1373  * @adapter: address of board private structure
1374  * @start: address of beginning of memory
1375  * @len: length of memory
1376  **/
1377 static bool e1000_check_64k_bound(struct e1000_adapter *adapter, void *start,
1378                                   unsigned long len)
1379 {
1380         struct e1000_hw *hw = &adapter->hw;
1381         unsigned long begin = (unsigned long)start;
1382         unsigned long end = begin + len;
1383
1384         /* First rev 82545 and 82546 need to not allow any memory
1385          * write location to cross 64k boundary due to errata 23 */
1386         if (hw->mac_type == e1000_82545 ||
1387             hw->mac_type == e1000_82546) {
1388                 return ((begin ^ (end - 1)) >> 16) != 0 ? false : true;
1389         }
1390
1391         return true;
1392 }
1393
1394 /**
1395  * e1000_setup_tx_resources - allocate Tx resources (Descriptors)
1396  * @adapter: board private structure
1397  * @txdr:    tx descriptor ring (for a specific queue) to setup
1398  *
1399  * Return 0 on success, negative on failure
1400  **/
1401
1402 static int e1000_setup_tx_resources(struct e1000_adapter *adapter,
1403                                     struct e1000_tx_ring *txdr)
1404 {
1405         struct pci_dev *pdev = adapter->pdev;
1406         int size;
1407
1408         size = sizeof(struct e1000_buffer) * txdr->count;
1409         txdr->buffer_info = vmalloc(size);
1410         if (!txdr->buffer_info) {
1411                 e_err(probe, "Unable to allocate memory for the Tx descriptor "
1412                       "ring\n");
1413                 return -ENOMEM;
1414         }
1415         memset(txdr->buffer_info, 0, size);
1416
1417         /* round up to nearest 4K */
1418
1419         txdr->size = txdr->count * sizeof(struct e1000_tx_desc);
1420         txdr->size = ALIGN(txdr->size, 4096);
1421
1422         txdr->desc = dma_alloc_coherent(&pdev->dev, txdr->size, &txdr->dma,
1423                                         GFP_KERNEL);
1424         if (!txdr->desc) {
1425 setup_tx_desc_die:
1426                 vfree(txdr->buffer_info);
1427                 e_err(probe, "Unable to allocate memory for the Tx descriptor "
1428                       "ring\n");
1429                 return -ENOMEM;
1430         }
1431
1432         /* Fix for errata 23, can't cross 64kB boundary */
1433         if (!e1000_check_64k_bound(adapter, txdr->desc, txdr->size)) {
1434                 void *olddesc = txdr->desc;
1435                 dma_addr_t olddma = txdr->dma;
1436                 e_err(tx_err, "txdr align check failed: %u bytes at %p\n",
1437                       txdr->size, txdr->desc);
1438                 /* Try again, without freeing the previous */
1439                 txdr->desc = dma_alloc_coherent(&pdev->dev, txdr->size,
1440                                                 &txdr->dma, GFP_KERNEL);
1441                 /* Failed allocation, critical failure */
1442                 if (!txdr->desc) {
1443                         dma_free_coherent(&pdev->dev, txdr->size, olddesc,
1444                                           olddma);
1445                         goto setup_tx_desc_die;
1446                 }
1447
1448                 if (!e1000_check_64k_bound(adapter, txdr->desc, txdr->size)) {
1449                         /* give up */
1450                         dma_free_coherent(&pdev->dev, txdr->size, txdr->desc,
1451                                           txdr->dma);
1452                         dma_free_coherent(&pdev->dev, txdr->size, olddesc,
1453                                           olddma);
1454                         e_err(probe, "Unable to allocate aligned memory "
1455                               "for the transmit descriptor ring\n");
1456                         vfree(txdr->buffer_info);
1457                         return -ENOMEM;
1458                 } else {
1459                         /* Free old allocation, new allocation was successful */
1460                         dma_free_coherent(&pdev->dev, txdr->size, olddesc,
1461                                           olddma);
1462                 }
1463         }
1464         memset(txdr->desc, 0, txdr->size);
1465
1466         txdr->next_to_use = 0;
1467         txdr->next_to_clean = 0;
1468
1469         return 0;
1470 }
1471
1472 /**
1473  * e1000_setup_all_tx_resources - wrapper to allocate Tx resources
1474  *                                (Descriptors) for all queues
1475  * @adapter: board private structure
1476  *
1477  * Return 0 on success, negative on failure
1478  **/
1479
1480 int e1000_setup_all_tx_resources(struct e1000_adapter *adapter)
1481 {
1482         int i, err = 0;
1483
1484         for (i = 0; i < adapter->num_tx_queues; i++) {
1485                 err = e1000_setup_tx_resources(adapter, &adapter->tx_ring[i]);
1486                 if (err) {
1487                         e_err(probe, "Allocation for Tx Queue %u failed\n", i);
1488                         for (i-- ; i >= 0; i--)
1489                                 e1000_free_tx_resources(adapter,
1490                                                         &adapter->tx_ring[i]);
1491                         break;
1492                 }
1493         }
1494
1495         return err;
1496 }
1497
1498 /**
1499  * e1000_configure_tx - Configure 8254x Transmit Unit after Reset
1500  * @adapter: board private structure
1501  *
1502  * Configure the Tx unit of the MAC after a reset.
1503  **/
1504
1505 static void e1000_configure_tx(struct e1000_adapter *adapter)
1506 {
1507         u64 tdba;
1508         struct e1000_hw *hw = &adapter->hw;
1509         u32 tdlen, tctl, tipg;
1510         u32 ipgr1, ipgr2;
1511
1512         /* Setup the HW Tx Head and Tail descriptor pointers */
1513
1514         switch (adapter->num_tx_queues) {
1515         case 1:
1516         default:
1517                 tdba = adapter->tx_ring[0].dma;
1518                 tdlen = adapter->tx_ring[0].count *
1519                         sizeof(struct e1000_tx_desc);
1520                 ew32(TDLEN, tdlen);
1521                 ew32(TDBAH, (tdba >> 32));
1522                 ew32(TDBAL, (tdba & 0x00000000ffffffffULL));
1523                 ew32(TDT, 0);
1524                 ew32(TDH, 0);
1525                 adapter->tx_ring[0].tdh = ((hw->mac_type >= e1000_82543) ? E1000_TDH : E1000_82542_TDH);
1526                 adapter->tx_ring[0].tdt = ((hw->mac_type >= e1000_82543) ? E1000_TDT : E1000_82542_TDT);
1527                 break;
1528         }
1529
1530         /* Set the default values for the Tx Inter Packet Gap timer */
1531         if ((hw->media_type == e1000_media_type_fiber ||
1532              hw->media_type == e1000_media_type_internal_serdes))
1533                 tipg = DEFAULT_82543_TIPG_IPGT_FIBER;
1534         else
1535                 tipg = DEFAULT_82543_TIPG_IPGT_COPPER;
1536
1537         switch (hw->mac_type) {
1538         case e1000_82542_rev2_0:
1539         case e1000_82542_rev2_1:
1540                 tipg = DEFAULT_82542_TIPG_IPGT;
1541                 ipgr1 = DEFAULT_82542_TIPG_IPGR1;
1542                 ipgr2 = DEFAULT_82542_TIPG_IPGR2;
1543                 break;
1544         default:
1545                 ipgr1 = DEFAULT_82543_TIPG_IPGR1;
1546                 ipgr2 = DEFAULT_82543_TIPG_IPGR2;
1547                 break;
1548         }
1549         tipg |= ipgr1 << E1000_TIPG_IPGR1_SHIFT;
1550         tipg |= ipgr2 << E1000_TIPG_IPGR2_SHIFT;
1551         ew32(TIPG, tipg);
1552
1553         /* Set the Tx Interrupt Delay register */
1554
1555         ew32(TIDV, adapter->tx_int_delay);
1556         if (hw->mac_type >= e1000_82540)
1557                 ew32(TADV, adapter->tx_abs_int_delay);
1558
1559         /* Program the Transmit Control Register */
1560
1561         tctl = er32(TCTL);
1562         tctl &= ~E1000_TCTL_CT;
1563         tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
1564                 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
1565
1566         e1000_config_collision_dist(hw);
1567
1568         /* Setup Transmit Descriptor Settings for eop descriptor */
1569         adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
1570
1571         /* only set IDE if we are delaying interrupts using the timers */
1572         if (adapter->tx_int_delay)
1573                 adapter->txd_cmd |= E1000_TXD_CMD_IDE;
1574
1575         if (hw->mac_type < e1000_82543)
1576                 adapter->txd_cmd |= E1000_TXD_CMD_RPS;
1577         else
1578                 adapter->txd_cmd |= E1000_TXD_CMD_RS;
1579
1580         /* Cache if we're 82544 running in PCI-X because we'll
1581          * need this to apply a workaround later in the send path. */
1582         if (hw->mac_type == e1000_82544 &&
1583             hw->bus_type == e1000_bus_type_pcix)
1584                 adapter->pcix_82544 = 1;
1585
1586         ew32(TCTL, tctl);
1587
1588 }
1589
1590 /**
1591  * e1000_setup_rx_resources - allocate Rx resources (Descriptors)
1592  * @adapter: board private structure
1593  * @rxdr:    rx descriptor ring (for a specific queue) to setup
1594  *
1595  * Returns 0 on success, negative on failure
1596  **/
1597
1598 static int e1000_setup_rx_resources(struct e1000_adapter *adapter,
1599                                     struct e1000_rx_ring *rxdr)
1600 {
1601         struct pci_dev *pdev = adapter->pdev;
1602         int size, desc_len;
1603
1604         size = sizeof(struct e1000_buffer) * rxdr->count;
1605         rxdr->buffer_info = vmalloc(size);
1606         if (!rxdr->buffer_info) {
1607                 e_err(probe, "Unable to allocate memory for the Rx descriptor "
1608                       "ring\n");
1609                 return -ENOMEM;
1610         }
1611         memset(rxdr->buffer_info, 0, size);
1612
1613         desc_len = sizeof(struct e1000_rx_desc);
1614
1615         /* Round up to nearest 4K */
1616
1617         rxdr->size = rxdr->count * desc_len;
1618         rxdr->size = ALIGN(rxdr->size, 4096);
1619
1620         rxdr->desc = dma_alloc_coherent(&pdev->dev, rxdr->size, &rxdr->dma,
1621                                         GFP_KERNEL);
1622
1623         if (!rxdr->desc) {
1624                 e_err(probe, "Unable to allocate memory for the Rx descriptor "
1625                       "ring\n");
1626 setup_rx_desc_die:
1627                 vfree(rxdr->buffer_info);
1628                 return -ENOMEM;
1629         }
1630
1631         /* Fix for errata 23, can't cross 64kB boundary */
1632         if (!e1000_check_64k_bound(adapter, rxdr->desc, rxdr->size)) {
1633                 void *olddesc = rxdr->desc;
1634                 dma_addr_t olddma = rxdr->dma;
1635                 e_err(rx_err, "rxdr align check failed: %u bytes at %p\n",
1636                       rxdr->size, rxdr->desc);
1637                 /* Try again, without freeing the previous */
1638                 rxdr->desc = dma_alloc_coherent(&pdev->dev, rxdr->size,
1639                                                 &rxdr->dma, GFP_KERNEL);
1640                 /* Failed allocation, critical failure */
1641                 if (!rxdr->desc) {
1642                         dma_free_coherent(&pdev->dev, rxdr->size, olddesc,
1643                                           olddma);
1644                         e_err(probe, "Unable to allocate memory for the Rx "
1645                               "descriptor ring\n");
1646                         goto setup_rx_desc_die;
1647                 }
1648
1649                 if (!e1000_check_64k_bound(adapter, rxdr->desc, rxdr->size)) {
1650                         /* give up */
1651                         dma_free_coherent(&pdev->dev, rxdr->size, rxdr->desc,
1652                                           rxdr->dma);
1653                         dma_free_coherent(&pdev->dev, rxdr->size, olddesc,
1654                                           olddma);
1655                         e_err(probe, "Unable to allocate aligned memory for "
1656                               "the Rx descriptor ring\n");
1657                         goto setup_rx_desc_die;
1658                 } else {
1659                         /* Free old allocation, new allocation was successful */
1660                         dma_free_coherent(&pdev->dev, rxdr->size, olddesc,
1661                                           olddma);
1662                 }
1663         }
1664         memset(rxdr->desc, 0, rxdr->size);
1665
1666         rxdr->next_to_clean = 0;
1667         rxdr->next_to_use = 0;
1668         rxdr->rx_skb_top = NULL;
1669
1670         return 0;
1671 }
1672
1673 /**
1674  * e1000_setup_all_rx_resources - wrapper to allocate Rx resources
1675  *                                (Descriptors) for all queues
1676  * @adapter: board private structure
1677  *
1678  * Return 0 on success, negative on failure
1679  **/
1680
1681 int e1000_setup_all_rx_resources(struct e1000_adapter *adapter)
1682 {
1683         int i, err = 0;
1684
1685         for (i = 0; i < adapter->num_rx_queues; i++) {
1686                 err = e1000_setup_rx_resources(adapter, &adapter->rx_ring[i]);
1687                 if (err) {
1688                         e_err(probe, "Allocation for Rx Queue %u failed\n", i);
1689                         for (i-- ; i >= 0; i--)
1690                                 e1000_free_rx_resources(adapter,
1691                                                         &adapter->rx_ring[i]);
1692                         break;
1693                 }
1694         }
1695
1696         return err;
1697 }
1698
1699 /**
1700  * e1000_setup_rctl - configure the receive control registers
1701  * @adapter: Board private structure
1702  **/
1703 static void e1000_setup_rctl(struct e1000_adapter *adapter)
1704 {
1705         struct e1000_hw *hw = &adapter->hw;
1706         u32 rctl;
1707
1708         rctl = er32(RCTL);
1709
1710         rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
1711
1712         rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
1713                 E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
1714                 (hw->mc_filter_type << E1000_RCTL_MO_SHIFT);
1715
1716         if (hw->tbi_compatibility_on == 1)
1717                 rctl |= E1000_RCTL_SBP;
1718         else
1719                 rctl &= ~E1000_RCTL_SBP;
1720
1721         if (adapter->netdev->mtu <= ETH_DATA_LEN)
1722                 rctl &= ~E1000_RCTL_LPE;
1723         else
1724                 rctl |= E1000_RCTL_LPE;
1725
1726         /* Setup buffer sizes */
1727         rctl &= ~E1000_RCTL_SZ_4096;
1728         rctl |= E1000_RCTL_BSEX;
1729         switch (adapter->rx_buffer_len) {
1730                 case E1000_RXBUFFER_2048:
1731                 default:
1732                         rctl |= E1000_RCTL_SZ_2048;
1733                         rctl &= ~E1000_RCTL_BSEX;
1734                         break;
1735                 case E1000_RXBUFFER_4096:
1736                         rctl |= E1000_RCTL_SZ_4096;
1737                         break;
1738                 case E1000_RXBUFFER_8192:
1739                         rctl |= E1000_RCTL_SZ_8192;
1740                         break;
1741                 case E1000_RXBUFFER_16384:
1742                         rctl |= E1000_RCTL_SZ_16384;
1743                         break;
1744         }
1745
1746         ew32(RCTL, rctl);
1747 }
1748
1749 /**
1750  * e1000_configure_rx - Configure 8254x Receive Unit after Reset
1751  * @adapter: board private structure
1752  *
1753  * Configure the Rx unit of the MAC after a reset.
1754  **/
1755
1756 static void e1000_configure_rx(struct e1000_adapter *adapter)
1757 {
1758         u64 rdba;
1759         struct e1000_hw *hw = &adapter->hw;
1760         u32 rdlen, rctl, rxcsum;
1761
1762         if (adapter->netdev->mtu > ETH_DATA_LEN) {
1763                 rdlen = adapter->rx_ring[0].count *
1764                         sizeof(struct e1000_rx_desc);
1765                 adapter->clean_rx = e1000_clean_jumbo_rx_irq;
1766                 adapter->alloc_rx_buf = e1000_alloc_jumbo_rx_buffers;
1767         } else {
1768                 rdlen = adapter->rx_ring[0].count *
1769                         sizeof(struct e1000_rx_desc);
1770                 adapter->clean_rx = e1000_clean_rx_irq;
1771                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
1772         }
1773
1774         /* disable receives while setting up the descriptors */
1775         rctl = er32(RCTL);
1776         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1777
1778         /* set the Receive Delay Timer Register */
1779         ew32(RDTR, adapter->rx_int_delay);
1780
1781         if (hw->mac_type >= e1000_82540) {
1782                 ew32(RADV, adapter->rx_abs_int_delay);
1783                 if (adapter->itr_setting != 0)
1784                         ew32(ITR, 1000000000 / (adapter->itr * 256));
1785         }
1786
1787         /* Setup the HW Rx Head and Tail Descriptor Pointers and
1788          * the Base and Length of the Rx Descriptor Ring */
1789         switch (adapter->num_rx_queues) {
1790         case 1:
1791         default:
1792                 rdba = adapter->rx_ring[0].dma;
1793                 ew32(RDLEN, rdlen);
1794                 ew32(RDBAH, (rdba >> 32));
1795                 ew32(RDBAL, (rdba & 0x00000000ffffffffULL));
1796                 ew32(RDT, 0);
1797                 ew32(RDH, 0);
1798                 adapter->rx_ring[0].rdh = ((hw->mac_type >= e1000_82543) ? E1000_RDH : E1000_82542_RDH);
1799                 adapter->rx_ring[0].rdt = ((hw->mac_type >= e1000_82543) ? E1000_RDT : E1000_82542_RDT);
1800                 break;
1801         }
1802
1803         /* Enable 82543 Receive Checksum Offload for TCP and UDP */
1804         if (hw->mac_type >= e1000_82543) {
1805                 rxcsum = er32(RXCSUM);
1806                 if (adapter->rx_csum)
1807                         rxcsum |= E1000_RXCSUM_TUOFL;
1808                 else
1809                         /* don't need to clear IPPCSE as it defaults to 0 */
1810                         rxcsum &= ~E1000_RXCSUM_TUOFL;
1811                 ew32(RXCSUM, rxcsum);
1812         }
1813
1814         /* Enable Receives */
1815         ew32(RCTL, rctl);
1816 }
1817
1818 /**
1819  * e1000_free_tx_resources - Free Tx Resources per Queue
1820  * @adapter: board private structure
1821  * @tx_ring: Tx descriptor ring for a specific queue
1822  *
1823  * Free all transmit software resources
1824  **/
1825
1826 static void e1000_free_tx_resources(struct e1000_adapter *adapter,
1827                                     struct e1000_tx_ring *tx_ring)
1828 {
1829         struct pci_dev *pdev = adapter->pdev;
1830
1831         e1000_clean_tx_ring(adapter, tx_ring);
1832
1833         vfree(tx_ring->buffer_info);
1834         tx_ring->buffer_info = NULL;
1835
1836         dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
1837                           tx_ring->dma);
1838
1839         tx_ring->desc = NULL;
1840 }
1841
1842 /**
1843  * e1000_free_all_tx_resources - Free Tx Resources for All Queues
1844  * @adapter: board private structure
1845  *
1846  * Free all transmit software resources
1847  **/
1848
1849 void e1000_free_all_tx_resources(struct e1000_adapter *adapter)
1850 {
1851         int i;
1852
1853         for (i = 0; i < adapter->num_tx_queues; i++)
1854                 e1000_free_tx_resources(adapter, &adapter->tx_ring[i]);
1855 }
1856
1857 static void e1000_unmap_and_free_tx_resource(struct e1000_adapter *adapter,
1858                                              struct e1000_buffer *buffer_info)
1859 {
1860         if (buffer_info->dma) {
1861                 if (buffer_info->mapped_as_page)
1862                         dma_unmap_page(&adapter->pdev->dev, buffer_info->dma,
1863                                        buffer_info->length, DMA_TO_DEVICE);
1864                 else
1865                         dma_unmap_single(&adapter->pdev->dev, buffer_info->dma,
1866                                          buffer_info->length,
1867                                          DMA_TO_DEVICE);
1868                 buffer_info->dma = 0;
1869         }
1870         if (buffer_info->skb) {
1871                 dev_kfree_skb_any(buffer_info->skb);
1872                 buffer_info->skb = NULL;
1873         }
1874         buffer_info->time_stamp = 0;
1875         /* buffer_info must be completely set up in the transmit path */
1876 }
1877
1878 /**
1879  * e1000_clean_tx_ring - Free Tx Buffers
1880  * @adapter: board private structure
1881  * @tx_ring: ring to be cleaned
1882  **/
1883
1884 static void e1000_clean_tx_ring(struct e1000_adapter *adapter,
1885                                 struct e1000_tx_ring *tx_ring)
1886 {
1887         struct e1000_hw *hw = &adapter->hw;
1888         struct e1000_buffer *buffer_info;
1889         unsigned long size;
1890         unsigned int i;
1891
1892         /* Free all the Tx ring sk_buffs */
1893
1894         for (i = 0; i < tx_ring->count; i++) {
1895                 buffer_info = &tx_ring->buffer_info[i];
1896                 e1000_unmap_and_free_tx_resource(adapter, buffer_info);
1897         }
1898
1899         size = sizeof(struct e1000_buffer) * tx_ring->count;
1900         memset(tx_ring->buffer_info, 0, size);
1901
1902         /* Zero out the descriptor ring */
1903
1904         memset(tx_ring->desc, 0, tx_ring->size);
1905
1906         tx_ring->next_to_use = 0;
1907         tx_ring->next_to_clean = 0;
1908         tx_ring->last_tx_tso = 0;
1909
1910         writel(0, hw->hw_addr + tx_ring->tdh);
1911         writel(0, hw->hw_addr + tx_ring->tdt);
1912 }
1913
1914 /**
1915  * e1000_clean_all_tx_rings - Free Tx Buffers for all queues
1916  * @adapter: board private structure
1917  **/
1918
1919 static void e1000_clean_all_tx_rings(struct e1000_adapter *adapter)
1920 {
1921         int i;
1922
1923         for (i = 0; i < adapter->num_tx_queues; i++)
1924                 e1000_clean_tx_ring(adapter, &adapter->tx_ring[i]);
1925 }
1926
1927 /**
1928  * e1000_free_rx_resources - Free Rx Resources
1929  * @adapter: board private structure
1930  * @rx_ring: ring to clean the resources from
1931  *
1932  * Free all receive software resources
1933  **/
1934
1935 static void e1000_free_rx_resources(struct e1000_adapter *adapter,
1936                                     struct e1000_rx_ring *rx_ring)
1937 {
1938         struct pci_dev *pdev = adapter->pdev;
1939
1940         e1000_clean_rx_ring(adapter, rx_ring);
1941
1942         vfree(rx_ring->buffer_info);
1943         rx_ring->buffer_info = NULL;
1944
1945         dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
1946                           rx_ring->dma);
1947
1948         rx_ring->desc = NULL;
1949 }
1950
1951 /**
1952  * e1000_free_all_rx_resources - Free Rx Resources for All Queues
1953  * @adapter: board private structure
1954  *
1955  * Free all receive software resources
1956  **/
1957
1958 void e1000_free_all_rx_resources(struct e1000_adapter *adapter)
1959 {
1960         int i;
1961
1962         for (i = 0; i < adapter->num_rx_queues; i++)
1963                 e1000_free_rx_resources(adapter, &adapter->rx_ring[i]);
1964 }
1965
1966 /**
1967  * e1000_clean_rx_ring - Free Rx Buffers per Queue
1968  * @adapter: board private structure
1969  * @rx_ring: ring to free buffers from
1970  **/
1971
1972 static void e1000_clean_rx_ring(struct e1000_adapter *adapter,
1973                                 struct e1000_rx_ring *rx_ring)
1974 {
1975         struct e1000_hw *hw = &adapter->hw;
1976         struct e1000_buffer *buffer_info;
1977         struct pci_dev *pdev = adapter->pdev;
1978         unsigned long size;
1979         unsigned int i;
1980
1981         /* Free all the Rx ring sk_buffs */
1982         for (i = 0; i < rx_ring->count; i++) {
1983                 buffer_info = &rx_ring->buffer_info[i];
1984                 if (buffer_info->dma &&
1985                     adapter->clean_rx == e1000_clean_rx_irq) {
1986                         dma_unmap_single(&pdev->dev, buffer_info->dma,
1987                                          buffer_info->length,
1988                                          DMA_FROM_DEVICE);
1989                 } else if (buffer_info->dma &&
1990                            adapter->clean_rx == e1000_clean_jumbo_rx_irq) {
1991                         dma_unmap_page(&pdev->dev, buffer_info->dma,
1992                                        buffer_info->length,
1993                                        DMA_FROM_DEVICE);
1994                 }
1995
1996                 buffer_info->dma = 0;
1997                 if (buffer_info->page) {
1998                         put_page(buffer_info->page);
1999                         buffer_info->page = NULL;
2000                 }
2001                 if (buffer_info->skb) {
2002                         dev_kfree_skb(buffer_info->skb);
2003                         buffer_info->skb = NULL;
2004                 }
2005         }
2006
2007         /* there also may be some cached data from a chained receive */
2008         if (rx_ring->rx_skb_top) {
2009                 dev_kfree_skb(rx_ring->rx_skb_top);
2010                 rx_ring->rx_skb_top = NULL;
2011         }
2012
2013         size = sizeof(struct e1000_buffer) * rx_ring->count;
2014         memset(rx_ring->buffer_info, 0, size);
2015
2016         /* Zero out the descriptor ring */
2017         memset(rx_ring->desc, 0, rx_ring->size);
2018
2019         rx_ring->next_to_clean = 0;
2020         rx_ring->next_to_use = 0;
2021
2022         writel(0, hw->hw_addr + rx_ring->rdh);
2023         writel(0, hw->hw_addr + rx_ring->rdt);
2024 }
2025
2026 /**
2027  * e1000_clean_all_rx_rings - Free Rx Buffers for all queues
2028  * @adapter: board private structure
2029  **/
2030
2031 static void e1000_clean_all_rx_rings(struct e1000_adapter *adapter)
2032 {
2033         int i;
2034
2035         for (i = 0; i < adapter->num_rx_queues; i++)
2036                 e1000_clean_rx_ring(adapter, &adapter->rx_ring[i]);
2037 }
2038
2039 /* The 82542 2.0 (revision 2) needs to have the receive unit in reset
2040  * and memory write and invalidate disabled for certain operations
2041  */
2042 static void e1000_enter_82542_rst(struct e1000_adapter *adapter)
2043 {
2044         struct e1000_hw *hw = &adapter->hw;
2045         struct net_device *netdev = adapter->netdev;
2046         u32 rctl;
2047
2048         e1000_pci_clear_mwi(hw);
2049
2050         rctl = er32(RCTL);
2051         rctl |= E1000_RCTL_RST;
2052         ew32(RCTL, rctl);
2053         E1000_WRITE_FLUSH();
2054         mdelay(5);
2055
2056         if (netif_running(netdev))
2057                 e1000_clean_all_rx_rings(adapter);
2058 }
2059
2060 static void e1000_leave_82542_rst(struct e1000_adapter *adapter)
2061 {
2062         struct e1000_hw *hw = &adapter->hw;
2063         struct net_device *netdev = adapter->netdev;
2064         u32 rctl;
2065
2066         rctl = er32(RCTL);
2067         rctl &= ~E1000_RCTL_RST;
2068         ew32(RCTL, rctl);
2069         E1000_WRITE_FLUSH();
2070         mdelay(5);
2071
2072         if (hw->pci_cmd_word & PCI_COMMAND_INVALIDATE)
2073                 e1000_pci_set_mwi(hw);
2074
2075         if (netif_running(netdev)) {
2076                 /* No need to loop, because 82542 supports only 1 queue */
2077                 struct e1000_rx_ring *ring = &adapter->rx_ring[0];
2078                 e1000_configure_rx(adapter);
2079                 adapter->alloc_rx_buf(adapter, ring, E1000_DESC_UNUSED(ring));
2080         }
2081 }
2082
2083 /**
2084  * e1000_set_mac - Change the Ethernet Address of the NIC
2085  * @netdev: network interface device structure
2086  * @p: pointer to an address structure
2087  *
2088  * Returns 0 on success, negative on failure
2089  **/
2090
2091 static int e1000_set_mac(struct net_device *netdev, void *p)
2092 {
2093         struct e1000_adapter *adapter = netdev_priv(netdev);
2094         struct e1000_hw *hw = &adapter->hw;
2095         struct sockaddr *addr = p;
2096
2097         if (!is_valid_ether_addr(addr->sa_data))
2098                 return -EADDRNOTAVAIL;
2099
2100         /* 82542 2.0 needs to be in reset to write receive address registers */
2101
2102         if (hw->mac_type == e1000_82542_rev2_0)
2103                 e1000_enter_82542_rst(adapter);
2104
2105         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
2106         memcpy(hw->mac_addr, addr->sa_data, netdev->addr_len);
2107
2108         e1000_rar_set(hw, hw->mac_addr, 0);
2109
2110         if (hw->mac_type == e1000_82542_rev2_0)
2111                 e1000_leave_82542_rst(adapter);
2112
2113         return 0;
2114 }
2115
2116 /**
2117  * e1000_set_rx_mode - Secondary Unicast, Multicast and Promiscuous mode set
2118  * @netdev: network interface device structure
2119  *
2120  * The set_rx_mode entry point is called whenever the unicast or multicast
2121  * address lists or the network interface flags are updated. This routine is
2122  * responsible for configuring the hardware for proper unicast, multicast,
2123  * promiscuous mode, and all-multi behavior.
2124  **/
2125
2126 static void e1000_set_rx_mode(struct net_device *netdev)
2127 {
2128         struct e1000_adapter *adapter = netdev_priv(netdev);
2129         struct e1000_hw *hw = &adapter->hw;
2130         struct netdev_hw_addr *ha;
2131         bool use_uc = false;
2132         u32 rctl;
2133         u32 hash_value;
2134         int i, rar_entries = E1000_RAR_ENTRIES;
2135         int mta_reg_count = E1000_NUM_MTA_REGISTERS;
2136         u32 *mcarray = kcalloc(mta_reg_count, sizeof(u32), GFP_ATOMIC);
2137
2138         if (!mcarray) {
2139                 e_err(probe, "memory allocation failed\n");
2140                 return;
2141         }
2142
2143         /* Check for Promiscuous and All Multicast modes */
2144
2145         rctl = er32(RCTL);
2146
2147         if (netdev->flags & IFF_PROMISC) {
2148                 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
2149                 rctl &= ~E1000_RCTL_VFE;
2150         } else {
2151                 if (netdev->flags & IFF_ALLMULTI)
2152                         rctl |= E1000_RCTL_MPE;
2153                 else
2154                         rctl &= ~E1000_RCTL_MPE;
2155                 /* Enable VLAN filter if there is a VLAN */
2156                 if (adapter->vlgrp)
2157                         rctl |= E1000_RCTL_VFE;
2158         }
2159
2160         if (netdev_uc_count(netdev) > rar_entries - 1) {
2161                 rctl |= E1000_RCTL_UPE;
2162         } else if (!(netdev->flags & IFF_PROMISC)) {
2163                 rctl &= ~E1000_RCTL_UPE;
2164                 use_uc = true;
2165         }
2166
2167         ew32(RCTL, rctl);
2168
2169         /* 82542 2.0 needs to be in reset to write receive address registers */
2170
2171         if (hw->mac_type == e1000_82542_rev2_0)
2172                 e1000_enter_82542_rst(adapter);
2173
2174         /* load the first 14 addresses into the exact filters 1-14. Unicast
2175          * addresses take precedence to avoid disabling unicast filtering
2176          * when possible.
2177          *
2178          * RAR 0 is used for the station MAC adddress
2179          * if there are not 14 addresses, go ahead and clear the filters
2180          */
2181         i = 1;
2182         if (use_uc)
2183                 netdev_for_each_uc_addr(ha, netdev) {
2184                         if (i == rar_entries)
2185                                 break;
2186                         e1000_rar_set(hw, ha->addr, i++);
2187                 }
2188
2189         netdev_for_each_mc_addr(ha, netdev) {
2190                 if (i == rar_entries) {
2191                         /* load any remaining addresses into the hash table */
2192                         u32 hash_reg, hash_bit, mta;
2193                         hash_value = e1000_hash_mc_addr(hw, ha->addr);
2194                         hash_reg = (hash_value >> 5) & 0x7F;
2195                         hash_bit = hash_value & 0x1F;
2196                         mta = (1 << hash_bit);
2197                         mcarray[hash_reg] |= mta;
2198                 } else {
2199                         e1000_rar_set(hw, ha->addr, i++);
2200                 }
2201         }
2202
2203         for (; i < rar_entries; i++) {
2204                 E1000_WRITE_REG_ARRAY(hw, RA, i << 1, 0);
2205                 E1000_WRITE_FLUSH();
2206                 E1000_WRITE_REG_ARRAY(hw, RA, (i << 1) + 1, 0);
2207                 E1000_WRITE_FLUSH();
2208         }
2209
2210         /* write the hash table completely, write from bottom to avoid
2211          * both stupid write combining chipsets, and flushing each write */
2212         for (i = mta_reg_count - 1; i >= 0 ; i--) {
2213                 /*
2214                  * If we are on an 82544 has an errata where writing odd
2215                  * offsets overwrites the previous even offset, but writing
2216                  * backwards over the range solves the issue by always
2217                  * writing the odd offset first
2218                  */
2219                 E1000_WRITE_REG_ARRAY(hw, MTA, i, mcarray[i]);
2220         }
2221         E1000_WRITE_FLUSH();
2222
2223         if (hw->mac_type == e1000_82542_rev2_0)
2224                 e1000_leave_82542_rst(adapter);
2225
2226         kfree(mcarray);
2227 }
2228
2229 /* Need to wait a few seconds after link up to get diagnostic information from
2230  * the phy */
2231
2232 static void e1000_update_phy_info(unsigned long data)
2233 {
2234         struct e1000_adapter *adapter = (struct e1000_adapter *)data;
2235         struct e1000_hw *hw = &adapter->hw;
2236         e1000_phy_get_info(hw, &adapter->phy_info);
2237 }
2238
2239 /**
2240  * e1000_82547_tx_fifo_stall - Timer Call-back
2241  * @data: pointer to adapter cast into an unsigned long
2242  **/
2243
2244 static void e1000_82547_tx_fifo_stall(unsigned long data)
2245 {
2246         struct e1000_adapter *adapter = (struct e1000_adapter *)data;
2247         struct e1000_hw *hw = &adapter->hw;
2248         struct net_device *netdev = adapter->netdev;
2249         u32 tctl;
2250
2251         if (atomic_read(&adapter->tx_fifo_stall)) {
2252                 if ((er32(TDT) == er32(TDH)) &&
2253                    (er32(TDFT) == er32(TDFH)) &&
2254                    (er32(TDFTS) == er32(TDFHS))) {
2255                         tctl = er32(TCTL);
2256                         ew32(TCTL, tctl & ~E1000_TCTL_EN);
2257                         ew32(TDFT, adapter->tx_head_addr);
2258                         ew32(TDFH, adapter->tx_head_addr);
2259                         ew32(TDFTS, adapter->tx_head_addr);
2260                         ew32(TDFHS, adapter->tx_head_addr);
2261                         ew32(TCTL, tctl);
2262                         E1000_WRITE_FLUSH();
2263
2264                         adapter->tx_fifo_head = 0;
2265                         atomic_set(&adapter->tx_fifo_stall, 0);
2266                         netif_wake_queue(netdev);
2267                 } else if (!test_bit(__E1000_DOWN, &adapter->flags)) {
2268                         mod_timer(&adapter->tx_fifo_stall_timer, jiffies + 1);
2269                 }
2270         }
2271 }
2272
2273 bool e1000_has_link(struct e1000_adapter *adapter)
2274 {
2275         struct e1000_hw *hw = &adapter->hw;
2276         bool link_active = false;
2277
2278         /* get_link_status is set on LSC (link status) interrupt or
2279          * rx sequence error interrupt.  get_link_status will stay
2280          * false until the e1000_check_for_link establishes link
2281          * for copper adapters ONLY
2282          */
2283         switch (hw->media_type) {
2284         case e1000_media_type_copper:
2285                 if (hw->get_link_status) {
2286                         e1000_check_for_link(hw);
2287                         link_active = !hw->get_link_status;
2288                 } else {
2289                         link_active = true;
2290                 }
2291                 break;
2292         case e1000_media_type_fiber:
2293                 e1000_check_for_link(hw);
2294                 link_active = !!(er32(STATUS) & E1000_STATUS_LU);
2295                 break;
2296         case e1000_media_type_internal_serdes:
2297                 e1000_check_for_link(hw);
2298                 link_active = hw->serdes_has_link;
2299                 break;
2300         default:
2301                 break;
2302         }
2303
2304         return link_active;
2305 }
2306
2307 /**
2308  * e1000_watchdog - Timer Call-back
2309  * @data: pointer to adapter cast into an unsigned long
2310  **/
2311 static void e1000_watchdog(unsigned long data)
2312 {
2313         struct e1000_adapter *adapter = (struct e1000_adapter *)data;
2314         struct e1000_hw *hw = &adapter->hw;
2315         struct net_device *netdev = adapter->netdev;
2316         struct e1000_tx_ring *txdr = adapter->tx_ring;
2317         u32 link, tctl;
2318
2319         link = e1000_has_link(adapter);
2320         if ((netif_carrier_ok(netdev)) && link)
2321                 goto link_up;
2322
2323         if (link) {
2324                 if (!netif_carrier_ok(netdev)) {
2325                         u32 ctrl;
2326                         bool txb2b = true;
2327                         /* update snapshot of PHY registers on LSC */
2328                         e1000_get_speed_and_duplex(hw,
2329                                                    &adapter->link_speed,
2330                                                    &adapter->link_duplex);
2331
2332                         ctrl = er32(CTRL);
2333                         pr_info("%s NIC Link is Up %d Mbps %s, "
2334                                 "Flow Control: %s\n",
2335                                 netdev->name,
2336                                 adapter->link_speed,
2337                                 adapter->link_duplex == FULL_DUPLEX ?
2338                                 "Full Duplex" : "Half Duplex",
2339                                 ((ctrl & E1000_CTRL_TFCE) && (ctrl &
2340                                 E1000_CTRL_RFCE)) ? "RX/TX" : ((ctrl &
2341                                 E1000_CTRL_RFCE) ? "RX" : ((ctrl &
2342                                 E1000_CTRL_TFCE) ? "TX" : "None")));
2343
2344                         /* adjust timeout factor according to speed/duplex */
2345                         adapter->tx_timeout_factor = 1;
2346                         switch (adapter->link_speed) {
2347                         case SPEED_10:
2348                                 txb2b = false;
2349                                 adapter->tx_timeout_factor = 16;
2350                                 break;
2351                         case SPEED_100:
2352                                 txb2b = false;
2353                                 /* maybe add some timeout factor ? */
2354                                 break;
2355                         }
2356
2357                         /* enable transmits in the hardware */
2358                         tctl = er32(TCTL);
2359                         tctl |= E1000_TCTL_EN;
2360                         ew32(TCTL, tctl);
2361
2362                         netif_carrier_on(netdev);
2363                         if (!test_bit(__E1000_DOWN, &adapter->flags))
2364                                 mod_timer(&adapter->phy_info_timer,
2365                                           round_jiffies(jiffies + 2 * HZ));
2366                         adapter->smartspeed = 0;
2367                 }
2368         } else {
2369                 if (netif_carrier_ok(netdev)) {
2370                         adapter->link_speed = 0;
2371                         adapter->link_duplex = 0;
2372                         pr_info("%s NIC Link is Down\n",
2373                                 netdev->name);
2374                         netif_carrier_off(netdev);
2375
2376                         if (!test_bit(__E1000_DOWN, &adapter->flags))
2377                                 mod_timer(&adapter->phy_info_timer,
2378                                           round_jiffies(jiffies + 2 * HZ));
2379                 }
2380
2381                 e1000_smartspeed(adapter);
2382         }
2383
2384 link_up:
2385         e1000_update_stats(adapter);
2386
2387         hw->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
2388         adapter->tpt_old = adapter->stats.tpt;
2389         hw->collision_delta = adapter->stats.colc - adapter->colc_old;
2390         adapter->colc_old = adapter->stats.colc;
2391
2392         adapter->gorcl = adapter->stats.gorcl - adapter->gorcl_old;
2393         adapter->gorcl_old = adapter->stats.gorcl;
2394         adapter->gotcl = adapter->stats.gotcl - adapter->gotcl_old;
2395         adapter->gotcl_old = adapter->stats.gotcl;
2396
2397         e1000_update_adaptive(hw);
2398
2399         if (!netif_carrier_ok(netdev)) {
2400                 if (E1000_DESC_UNUSED(txdr) + 1 < txdr->count) {
2401                         /* We've lost link, so the controller stops DMA,
2402                          * but we've got queued Tx work that's never going
2403                          * to get done, so reset controller to flush Tx.
2404                          * (Do the reset outside of interrupt context). */
2405                         adapter->tx_timeout_count++;
2406                         schedule_work(&adapter->reset_task);
2407                         /* return immediately since reset is imminent */
2408                         return;
2409                 }
2410         }
2411
2412         /* Simple mode for Interrupt Throttle Rate (ITR) */
2413         if (hw->mac_type >= e1000_82540 && adapter->itr_setting == 4) {
2414                 /*
2415                  * Symmetric Tx/Rx gets a reduced ITR=2000;
2416                  * Total asymmetrical Tx or Rx gets ITR=8000;
2417                  * everyone else is between 2000-8000.
2418                  */
2419                 u32 goc = (adapter->gotcl + adapter->gorcl) / 10000;
2420                 u32 dif = (adapter->gotcl > adapter->gorcl ?
2421                             adapter->gotcl - adapter->gorcl :
2422                             adapter->gorcl - adapter->gotcl) / 10000;
2423                 u32 itr = goc > 0 ? (dif * 6000 / goc + 2000) : 8000;
2424
2425                 ew32(ITR, 1000000000 / (itr * 256));
2426         }
2427
2428         /* Cause software interrupt to ensure rx ring is cleaned */
2429         ew32(ICS, E1000_ICS_RXDMT0);
2430
2431         /* Force detection of hung controller every watchdog period */
2432         adapter->detect_tx_hung = true;
2433
2434         /* Reset the timer */
2435         if (!test_bit(__E1000_DOWN, &adapter->flags))
2436                 mod_timer(&adapter->watchdog_timer,
2437                           round_jiffies(jiffies + 2 * HZ));
2438 }
2439
2440 enum latency_range {
2441         lowest_latency = 0,
2442         low_latency = 1,
2443         bulk_latency = 2,
2444         latency_invalid = 255
2445 };
2446
2447 /**
2448  * e1000_update_itr - update the dynamic ITR value based on statistics
2449  * @adapter: pointer to adapter
2450  * @itr_setting: current adapter->itr
2451  * @packets: the number of packets during this measurement interval
2452  * @bytes: the number of bytes during this measurement interval
2453  *
2454  *      Stores a new ITR value based on packets and byte
2455  *      counts during the last interrupt.  The advantage of per interrupt
2456  *      computation is faster updates and more accurate ITR for the current
2457  *      traffic pattern.  Constants in this function were computed
2458  *      based on theoretical maximum wire speed and thresholds were set based
2459  *      on testing data as well as attempting to minimize response time
2460  *      while increasing bulk throughput.
2461  *      this functionality is controlled by the InterruptThrottleRate module
2462  *      parameter (see e1000_param.c)
2463  **/
2464 static unsigned int e1000_update_itr(struct e1000_adapter *adapter,
2465                                      u16 itr_setting, int packets, int bytes)
2466 {
2467         unsigned int retval = itr_setting;
2468         struct e1000_hw *hw = &adapter->hw;
2469
2470         if (unlikely(hw->mac_type < e1000_82540))
2471                 goto update_itr_done;
2472
2473         if (packets == 0)
2474                 goto update_itr_done;
2475
2476         switch (itr_setting) {
2477         case lowest_latency:
2478                 /* jumbo frames get bulk treatment*/
2479                 if (bytes/packets > 8000)
2480                         retval = bulk_latency;
2481                 else if ((packets < 5) && (bytes > 512))
2482                         retval = low_latency;
2483                 break;
2484         case low_latency:  /* 50 usec aka 20000 ints/s */
2485                 if (bytes > 10000) {
2486                         /* jumbo frames need bulk latency setting */
2487                         if (bytes/packets > 8000)
2488                                 retval = bulk_latency;
2489                         else if ((packets < 10) || ((bytes/packets) > 1200))
2490                                 retval = bulk_latency;
2491                         else if ((packets > 35))
2492                                 retval = lowest_latency;
2493                 } else if (bytes/packets > 2000)
2494                         retval = bulk_latency;
2495                 else if (packets <= 2 && bytes < 512)
2496                         retval = lowest_latency;
2497                 break;
2498         case bulk_latency: /* 250 usec aka 4000 ints/s */
2499                 if (bytes > 25000) {
2500                         if (packets > 35)
2501                                 retval = low_latency;
2502                 } else if (bytes < 6000) {
2503                         retval = low_latency;
2504                 }
2505                 break;
2506         }
2507
2508 update_itr_done:
2509         return retval;
2510 }
2511
2512 static void e1000_set_itr(struct e1000_adapter *adapter)
2513 {
2514         struct e1000_hw *hw = &adapter->hw;
2515         u16 current_itr;
2516         u32 new_itr = adapter->itr;
2517
2518         if (unlikely(hw->mac_type < e1000_82540))
2519                 return;
2520
2521         /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
2522         if (unlikely(adapter->link_speed != SPEED_1000)) {
2523                 current_itr = 0;
2524                 new_itr = 4000;
2525                 goto set_itr_now;
2526         }
2527
2528         adapter->tx_itr = e1000_update_itr(adapter,
2529                                     adapter->tx_itr,
2530                                     adapter->total_tx_packets,
2531                                     adapter->total_tx_bytes);
2532         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2533         if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
2534                 adapter->tx_itr = low_latency;
2535
2536         adapter->rx_itr = e1000_update_itr(adapter,
2537                                     adapter->rx_itr,
2538                                     adapter->total_rx_packets,
2539                                     adapter->total_rx_bytes);
2540         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2541         if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
2542                 adapter->rx_itr = low_latency;
2543
2544         current_itr = max(adapter->rx_itr, adapter->tx_itr);
2545
2546         switch (current_itr) {
2547         /* counts and packets in update_itr are dependent on these numbers */
2548         case lowest_latency:
2549                 new_itr = 70000;
2550                 break;
2551         case low_latency:
2552                 new_itr = 20000; /* aka hwitr = ~200 */
2553                 break;
2554         case bulk_latency:
2555                 new_itr = 4000;
2556                 break;
2557         default:
2558                 break;
2559         }
2560
2561 set_itr_now:
2562         if (new_itr != adapter->itr) {
2563                 /* this attempts to bias the interrupt rate towards Bulk
2564                  * by adding intermediate steps when interrupt rate is
2565                  * increasing */
2566                 new_itr = new_itr > adapter->itr ?
2567                              min(adapter->itr + (new_itr >> 2), new_itr) :
2568                              new_itr;
2569                 adapter->itr = new_itr;
2570                 ew32(ITR, 1000000000 / (new_itr * 256));
2571         }
2572 }
2573
2574 #define E1000_TX_FLAGS_CSUM             0x00000001
2575 #define E1000_TX_FLAGS_VLAN             0x00000002
2576 #define E1000_TX_FLAGS_TSO              0x00000004
2577 #define E1000_TX_FLAGS_IPV4             0x00000008
2578 #define E1000_TX_FLAGS_VLAN_MASK        0xffff0000
2579 #define E1000_TX_FLAGS_VLAN_SHIFT       16
2580
2581 static int e1000_tso(struct e1000_adapter *adapter,
2582                      struct e1000_tx_ring *tx_ring, struct sk_buff *skb)
2583 {
2584         struct e1000_context_desc *context_desc;
2585         struct e1000_buffer *buffer_info;
2586         unsigned int i;
2587         u32 cmd_length = 0;
2588         u16 ipcse = 0, tucse, mss;
2589         u8 ipcss, ipcso, tucss, tucso, hdr_len;
2590         int err;
2591
2592         if (skb_is_gso(skb)) {
2593                 if (skb_header_cloned(skb)) {
2594                         err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2595                         if (err)
2596                                 return err;
2597                 }
2598
2599                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
2600                 mss = skb_shinfo(skb)->gso_size;
2601                 if (skb->protocol == htons(ETH_P_IP)) {
2602                         struct iphdr *iph = ip_hdr(skb);
2603                         iph->tot_len = 0;
2604                         iph->check = 0;
2605                         tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
2606                                                                  iph->daddr, 0,
2607                                                                  IPPROTO_TCP,
2608                                                                  0);
2609                         cmd_length = E1000_TXD_CMD_IP;
2610                         ipcse = skb_transport_offset(skb) - 1;
2611                 } else if (skb->protocol == htons(ETH_P_IPV6)) {
2612                         ipv6_hdr(skb)->payload_len = 0;
2613                         tcp_hdr(skb)->check =
2614                                 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
2615                                                  &ipv6_hdr(skb)->daddr,
2616                                                  0, IPPROTO_TCP, 0);
2617                         ipcse = 0;
2618                 }
2619                 ipcss = skb_network_offset(skb);
2620                 ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
2621                 tucss = skb_transport_offset(skb);
2622                 tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
2623                 tucse = 0;
2624
2625                 cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
2626                                E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
2627
2628                 i = tx_ring->next_to_use;
2629                 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
2630                 buffer_info = &tx_ring->buffer_info[i];
2631
2632                 context_desc->lower_setup.ip_fields.ipcss  = ipcss;
2633                 context_desc->lower_setup.ip_fields.ipcso  = ipcso;
2634                 context_desc->lower_setup.ip_fields.ipcse  = cpu_to_le16(ipcse);
2635                 context_desc->upper_setup.tcp_fields.tucss = tucss;
2636                 context_desc->upper_setup.tcp_fields.tucso = tucso;
2637                 context_desc->upper_setup.tcp_fields.tucse = cpu_to_le16(tucse);
2638                 context_desc->tcp_seg_setup.fields.mss     = cpu_to_le16(mss);
2639                 context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
2640                 context_desc->cmd_and_length = cpu_to_le32(cmd_length);
2641
2642                 buffer_info->time_stamp = jiffies;
2643                 buffer_info->next_to_watch = i;
2644
2645                 if (++i == tx_ring->count) i = 0;
2646                 tx_ring->next_to_use = i;
2647
2648                 return true;
2649         }
2650         return false;
2651 }
2652
2653 static bool e1000_tx_csum(struct e1000_adapter *adapter,
2654                           struct e1000_tx_ring *tx_ring, struct sk_buff *skb)
2655 {
2656         struct e1000_context_desc *context_desc;
2657         struct e1000_buffer *buffer_info;
2658         unsigned int i;
2659         u8 css;
2660         u32 cmd_len = E1000_TXD_CMD_DEXT;
2661
2662         if (skb->ip_summed != CHECKSUM_PARTIAL)
2663                 return false;
2664
2665         switch (skb->protocol) {
2666         case cpu_to_be16(ETH_P_IP):
2667                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
2668                         cmd_len |= E1000_TXD_CMD_TCP;
2669                 break;
2670         case cpu_to_be16(ETH_P_IPV6):
2671                 /* XXX not handling all IPV6 headers */
2672                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
2673                         cmd_len |= E1000_TXD_CMD_TCP;
2674                 break;
2675         default:
2676                 if (unlikely(net_ratelimit()))
2677                         e_warn(drv, "checksum_partial proto=%x!\n",
2678                                skb->protocol);
2679                 break;
2680         }
2681
2682         css = skb_transport_offset(skb);
2683
2684         i = tx_ring->next_to_use;
2685         buffer_info = &tx_ring->buffer_info[i];
2686         context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
2687
2688         context_desc->lower_setup.ip_config = 0;
2689         context_desc->upper_setup.tcp_fields.tucss = css;
2690         context_desc->upper_setup.tcp_fields.tucso =
2691                 css + skb->csum_offset;
2692         context_desc->upper_setup.tcp_fields.tucse = 0;
2693         context_desc->tcp_seg_setup.data = 0;
2694         context_desc->cmd_and_length = cpu_to_le32(cmd_len);
2695
2696         buffer_info->time_stamp = jiffies;
2697         buffer_info->next_to_watch = i;
2698
2699         if (unlikely(++i == tx_ring->count)) i = 0;
2700         tx_ring->next_to_use = i;
2701
2702         return true;
2703 }
2704
2705 #define E1000_MAX_TXD_PWR       12
2706 #define E1000_MAX_DATA_PER_TXD  (1<<E1000_MAX_TXD_PWR)
2707
2708 static int e1000_tx_map(struct e1000_adapter *adapter,
2709                         struct e1000_tx_ring *tx_ring,
2710                         struct sk_buff *skb, unsigned int first,
2711                         unsigned int max_per_txd, unsigned int nr_frags,
2712                         unsigned int mss)
2713 {
2714         struct e1000_hw *hw = &adapter->hw;
2715         struct pci_dev *pdev = adapter->pdev;
2716         struct e1000_buffer *buffer_info;
2717         unsigned int len = skb_headlen(skb);
2718         unsigned int offset = 0, size, count = 0, i;
2719         unsigned int f;
2720
2721         i = tx_ring->next_to_use;
2722
2723         while (len) {
2724                 buffer_info = &tx_ring->buffer_info[i];
2725                 size = min(len, max_per_txd);
2726                 /* Workaround for Controller erratum --
2727                  * descriptor for non-tso packet in a linear SKB that follows a
2728                  * tso gets written back prematurely before the data is fully
2729                  * DMA'd to the controller */
2730                 if (!skb->data_len && tx_ring->last_tx_tso &&
2731                     !skb_is_gso(skb)) {
2732                         tx_ring->last_tx_tso = 0;
2733                         size -= 4;
2734                 }
2735
2736                 /* Workaround for premature desc write-backs
2737                  * in TSO mode.  Append 4-byte sentinel desc */
2738                 if (unlikely(mss && !nr_frags && size == len && size > 8))
2739                         size -= 4;
2740                 /* work-around for errata 10 and it applies
2741                  * to all controllers in PCI-X mode
2742                  * The fix is to make sure that the first descriptor of a
2743                  * packet is smaller than 2048 - 16 - 16 (or 2016) bytes
2744                  */
2745                 if (unlikely((hw->bus_type == e1000_bus_type_pcix) &&
2746                                 (size > 2015) && count == 0))
2747                         size = 2015;
2748
2749                 /* Workaround for potential 82544 hang in PCI-X.  Avoid
2750                  * terminating buffers within evenly-aligned dwords. */
2751                 if (unlikely(adapter->pcix_82544 &&
2752                    !((unsigned long)(skb->data + offset + size - 1) & 4) &&
2753                    size > 4))
2754                         size -= 4;
2755
2756                 buffer_info->length = size;
2757                 /* set time_stamp *before* dma to help avoid a possible race */
2758                 buffer_info->time_stamp = jiffies;
2759                 buffer_info->mapped_as_page = false;
2760                 buffer_info->dma = dma_map_single(&pdev->dev,
2761                                                   skb->data + offset,
2762                                                   size, DMA_TO_DEVICE);
2763                 if (dma_mapping_error(&pdev->dev, buffer_info->dma))
2764                         goto dma_error;
2765                 buffer_info->next_to_watch = i;
2766
2767                 len -= size;
2768                 offset += size;
2769                 count++;
2770                 if (len) {
2771                         i++;
2772                         if (unlikely(i == tx_ring->count))
2773                                 i = 0;
2774                 }
2775         }
2776
2777         for (f = 0; f < nr_frags; f++) {
2778                 struct skb_frag_struct *frag;
2779
2780                 frag = &skb_shinfo(skb)->frags[f];
2781                 len = frag->size;
2782                 offset = frag->page_offset;
2783
2784                 while (len) {
2785                         i++;
2786                         if (unlikely(i == tx_ring->count))
2787                                 i = 0;
2788
2789                         buffer_info = &tx_ring->buffer_info[i];
2790                         size = min(len, max_per_txd);
2791                         /* Workaround for premature desc write-backs
2792                          * in TSO mode.  Append 4-byte sentinel desc */
2793                         if (unlikely(mss && f == (nr_frags-1) && size == len && size > 8))
2794                                 size -= 4;
2795                         /* Workaround for potential 82544 hang in PCI-X.
2796                          * Avoid terminating buffers within evenly-aligned
2797                          * dwords. */
2798                         if (unlikely(adapter->pcix_82544 &&
2799                             !((unsigned long)(page_to_phys(frag->page) + offset
2800                                               + size - 1) & 4) &&
2801                             size > 4))
2802                                 size -= 4;
2803
2804                         buffer_info->length = size;
2805                         buffer_info->time_stamp = jiffies;
2806                         buffer_info->mapped_as_page = true;
2807                         buffer_info->dma = dma_map_page(&pdev->dev, frag->page,
2808                                                         offset, size,
2809                                                         DMA_TO_DEVICE);
2810                         if (dma_mapping_error(&pdev->dev, buffer_info->dma))
2811                                 goto dma_error;
2812                         buffer_info->next_to_watch = i;
2813
2814                         len -= size;
2815                         offset += size;
2816                         count++;
2817                 }
2818         }
2819
2820         tx_ring->buffer_info[i].skb = skb;
2821         tx_ring->buffer_info[first].next_to_watch = i;
2822
2823         return count;
2824
2825 dma_error:
2826         dev_err(&pdev->dev, "TX DMA map failed\n");
2827         buffer_info->dma = 0;
2828         if (count)
2829                 count--;
2830
2831         while (count--) {
2832                 if (i==0)
2833                         i += tx_ring->count;
2834                 i--;
2835                 buffer_info = &tx_ring->buffer_info[i];
2836                 e1000_unmap_and_free_tx_resource(adapter, buffer_info);
2837         }
2838
2839         return 0;
2840 }
2841
2842 static void e1000_tx_queue(struct e1000_adapter *adapter,
2843                            struct e1000_tx_ring *tx_ring, int tx_flags,
2844                            int count)
2845 {
2846         struct e1000_hw *hw = &adapter->hw;
2847         struct e1000_tx_desc *tx_desc = NULL;
2848         struct e1000_buffer *buffer_info;
2849         u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
2850         unsigned int i;
2851
2852         if (likely(tx_flags & E1000_TX_FLAGS_TSO)) {
2853                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
2854                              E1000_TXD_CMD_TSE;
2855                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
2856
2857                 if (likely(tx_flags & E1000_TX_FLAGS_IPV4))
2858                         txd_upper |= E1000_TXD_POPTS_IXSM << 8;
2859         }
2860
2861         if (likely(tx_flags & E1000_TX_FLAGS_CSUM)) {
2862                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
2863                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
2864         }
2865
2866         if (unlikely(tx_flags & E1000_TX_FLAGS_VLAN)) {
2867                 txd_lower |= E1000_TXD_CMD_VLE;
2868                 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
2869         }
2870
2871         i = tx_ring->next_to_use;
2872
2873         while (count--) {
2874                 buffer_info = &tx_ring->buffer_info[i];
2875                 tx_desc = E1000_TX_DESC(*tx_ring, i);
2876                 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
2877                 tx_desc->lower.data =
2878                         cpu_to_le32(txd_lower | buffer_info->length);
2879                 tx_desc->upper.data = cpu_to_le32(txd_upper);
2880                 if (unlikely(++i == tx_ring->count)) i = 0;
2881         }
2882
2883         tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
2884
2885         /* Force memory writes to complete before letting h/w
2886          * know there are new descriptors to fetch.  (Only
2887          * applicable for weak-ordered memory model archs,
2888          * such as IA-64). */
2889         wmb();
2890
2891         tx_ring->next_to_use = i;
2892         writel(i, hw->hw_addr + tx_ring->tdt);
2893         /* we need this if more than one processor can write to our tail
2894          * at a time, it syncronizes IO on IA64/Altix systems */
2895         mmiowb();
2896 }
2897
2898 /**
2899  * 82547 workaround to avoid controller hang in half-duplex environment.
2900  * The workaround is to avoid queuing a large packet that would span
2901  * the internal Tx FIFO ring boundary by notifying the stack to resend
2902  * the packet at a later time.  This gives the Tx FIFO an opportunity to
2903  * flush all packets.  When that occurs, we reset the Tx FIFO pointers
2904  * to the beginning of the Tx FIFO.
2905  **/
2906
2907 #define E1000_FIFO_HDR                  0x10
2908 #define E1000_82547_PAD_LEN             0x3E0
2909
2910 static int e1000_82547_fifo_workaround(struct e1000_adapter *adapter,
2911                                        struct sk_buff *skb)
2912 {
2913         u32 fifo_space = adapter->tx_fifo_size - adapter->tx_fifo_head;
2914         u32 skb_fifo_len = skb->len + E1000_FIFO_HDR;
2915
2916         skb_fifo_len = ALIGN(skb_fifo_len, E1000_FIFO_HDR);
2917
2918         if (adapter->link_duplex != HALF_DUPLEX)
2919                 goto no_fifo_stall_required;
2920
2921         if (atomic_read(&adapter->tx_fifo_stall))
2922                 return 1;
2923
2924         if (skb_fifo_len >= (E1000_82547_PAD_LEN + fifo_space)) {
2925                 atomic_set(&adapter->tx_fifo_stall, 1);
2926                 return 1;
2927         }
2928
2929 no_fifo_stall_required:
2930         adapter->tx_fifo_head += skb_fifo_len;
2931         if (adapter->tx_fifo_head >= adapter->tx_fifo_size)
2932                 adapter->tx_fifo_head -= adapter->tx_fifo_size;
2933         return 0;
2934 }
2935
2936 static int __e1000_maybe_stop_tx(struct net_device *netdev, int size)
2937 {
2938         struct e1000_adapter *adapter = netdev_priv(netdev);
2939         struct e1000_tx_ring *tx_ring = adapter->tx_ring;
2940
2941         netif_stop_queue(netdev);
2942         /* Herbert's original patch had:
2943          *  smp_mb__after_netif_stop_queue();
2944          * but since that doesn't exist yet, just open code it. */
2945         smp_mb();
2946
2947         /* We need to check again in a case another CPU has just
2948          * made room available. */
2949         if (likely(E1000_DESC_UNUSED(tx_ring) < size))
2950                 return -EBUSY;
2951
2952         /* A reprieve! */
2953         netif_start_queue(netdev);
2954         ++adapter->restart_queue;
2955         return 0;
2956 }
2957
2958 static int e1000_maybe_stop_tx(struct net_device *netdev,
2959                                struct e1000_tx_ring *tx_ring, int size)
2960 {
2961         if (likely(E1000_DESC_UNUSED(tx_ring) >= size))
2962                 return 0;
2963         return __e1000_maybe_stop_tx(netdev, size);
2964 }
2965
2966 #define TXD_USE_COUNT(S, X) (((S) >> (X)) + 1 )
2967 static netdev_tx_t e1000_xmit_frame(struct sk_buff *skb,
2968                                     struct net_device *netdev)
2969 {
2970         struct e1000_adapter *adapter = netdev_priv(netdev);
2971         struct e1000_hw *hw = &adapter->hw;
2972         struct e1000_tx_ring *tx_ring;
2973         unsigned int first, max_per_txd = E1000_MAX_DATA_PER_TXD;
2974         unsigned int max_txd_pwr = E1000_MAX_TXD_PWR;
2975         unsigned int tx_flags = 0;
2976         unsigned int len = skb_headlen(skb);
2977         unsigned int nr_frags;
2978         unsigned int mss;
2979         int count = 0;
2980         int tso;
2981         unsigned int f;
2982
2983         /* This goes back to the question of how to logically map a tx queue
2984          * to a flow.  Right now, performance is impacted slightly negatively
2985          * if using multiple tx queues.  If the stack breaks away from a
2986          * single qdisc implementation, we can look at this again. */
2987         tx_ring = adapter->tx_ring;
2988
2989         if (unlikely(skb->len <= 0)) {
2990                 dev_kfree_skb_any(skb);
2991                 return NETDEV_TX_OK;
2992         }
2993
2994         mss = skb_shinfo(skb)->gso_size;
2995         /* The controller does a simple calculation to
2996          * make sure there is enough room in the FIFO before
2997          * initiating the DMA for each buffer.  The calc is:
2998          * 4 = ceil(buffer len/mss).  To make sure we don't
2999          * overrun the FIFO, adjust the max buffer len if mss
3000          * drops. */
3001         if (mss) {
3002                 u8 hdr_len;
3003                 max_per_txd = min(mss << 2, max_per_txd);
3004                 max_txd_pwr = fls(max_per_txd) - 1;
3005
3006                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
3007                 if (skb->data_len && hdr_len == len) {
3008                         switch (hw->mac_type) {
3009                                 unsigned int pull_size;
3010                         case e1000_82544:
3011                                 /* Make sure we have room to chop off 4 bytes,
3012                                  * and that the end alignment will work out to
3013                                  * this hardware's requirements
3014                                  * NOTE: this is a TSO only workaround
3015                                  * if end byte alignment not correct move us
3016                                  * into the next dword */
3017                                 if ((unsigned long)(skb_tail_pointer(skb) - 1) & 4)
3018                                         break;
3019                                 /* fall through */
3020                                 pull_size = min((unsigned int)4, skb->data_len);
3021                                 if (!__pskb_pull_tail(skb, pull_size)) {
3022                                         e_err(drv, "__pskb_pull_tail "
3023                                               "failed.\n");
3024                                         dev_kfree_skb_any(skb);
3025                                         return NETDEV_TX_OK;
3026                                 }
3027                                 len = skb_headlen(skb);
3028                                 break;
3029                         default:
3030                                 /* do nothing */
3031                                 break;
3032                         }
3033                 }
3034         }
3035
3036         /* reserve a descriptor for the offload context */
3037         if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
3038                 count++;
3039         count++;
3040
3041         /* Controller Erratum workaround */
3042         if (!skb->data_len && tx_ring->last_tx_tso && !skb_is_gso(skb))
3043                 count++;
3044
3045         count += TXD_USE_COUNT(len, max_txd_pwr);
3046
3047         if (adapter->pcix_82544)
3048                 count++;
3049
3050         /* work-around for errata 10 and it applies to all controllers
3051          * in PCI-X mode, so add one more descriptor to the count
3052          */
3053         if (unlikely((hw->bus_type == e1000_bus_type_pcix) &&
3054                         (len > 2015)))
3055                 count++;
3056
3057         nr_frags = skb_shinfo(skb)->nr_frags;
3058         for (f = 0; f < nr_frags; f++)
3059                 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size,
3060                                        max_txd_pwr);
3061         if (adapter->pcix_82544)
3062                 count += nr_frags;
3063
3064         /* need: count + 2 desc gap to keep tail from touching
3065          * head, otherwise try next time */
3066         if (unlikely(e1000_maybe_stop_tx(netdev, tx_ring, count + 2)))
3067                 return NETDEV_TX_BUSY;
3068
3069         if (unlikely(hw->mac_type == e1000_82547)) {
3070                 if (unlikely(e1000_82547_fifo_workaround(adapter, skb))) {
3071                         netif_stop_queue(netdev);
3072                         if (!test_bit(__E1000_DOWN, &adapter->flags))
3073                                 mod_timer(&adapter->tx_fifo_stall_timer,
3074                                           jiffies + 1);
3075                         return NETDEV_TX_BUSY;
3076                 }
3077         }
3078
3079         if (unlikely(adapter->vlgrp && vlan_tx_tag_present(skb))) {
3080                 tx_flags |= E1000_TX_FLAGS_VLAN;
3081                 tx_flags |= (vlan_tx_tag_get(skb) << E1000_TX_FLAGS_VLAN_SHIFT);
3082         }
3083
3084         first = tx_ring->next_to_use;
3085
3086         tso = e1000_tso(adapter, tx_ring, skb);
3087         if (tso < 0) {
3088                 dev_kfree_skb_any(skb);
3089                 return NETDEV_TX_OK;
3090         }
3091
3092         if (likely(tso)) {
3093                 if (likely(hw->mac_type != e1000_82544))
3094                         tx_ring->last_tx_tso = 1;
3095                 tx_flags |= E1000_TX_FLAGS_TSO;
3096         } else if (likely(e1000_tx_csum(adapter, tx_ring, skb)))
3097                 tx_flags |= E1000_TX_FLAGS_CSUM;
3098
3099         if (likely(skb->protocol == htons(ETH_P_IP)))
3100                 tx_flags |= E1000_TX_FLAGS_IPV4;
3101
3102         count = e1000_tx_map(adapter, tx_ring, skb, first, max_per_txd,
3103                              nr_frags, mss);
3104
3105         if (count) {
3106                 e1000_tx_queue(adapter, tx_ring, tx_flags, count);
3107                 /* Make sure there is space in the ring for the next send. */
3108                 e1000_maybe_stop_tx(netdev, tx_ring, MAX_SKB_FRAGS + 2);
3109
3110         } else {
3111                 dev_kfree_skb_any(skb);
3112                 tx_ring->buffer_info[first].time_stamp = 0;
3113                 tx_ring->next_to_use = first;
3114         }
3115
3116         return NETDEV_TX_OK;
3117 }
3118
3119 /**
3120  * e1000_tx_timeout - Respond to a Tx Hang
3121  * @netdev: network interface device structure
3122  **/
3123
3124 static void e1000_tx_timeout(struct net_device *netdev)
3125 {
3126         struct e1000_adapter *adapter = netdev_priv(netdev);
3127
3128         /* Do the reset outside of interrupt context */
3129         adapter->tx_timeout_count++;
3130         schedule_work(&adapter->reset_task);
3131 }
3132
3133 static void e1000_reset_task(struct work_struct *work)
3134 {
3135         struct e1000_adapter *adapter =
3136                 container_of(work, struct e1000_adapter, reset_task);
3137
3138         e1000_reinit_locked(adapter);
3139 }
3140
3141 /**
3142  * e1000_get_stats - Get System Network Statistics
3143  * @netdev: network interface device structure
3144  *
3145  * Returns the address of the device statistics structure.
3146  * The statistics are actually updated from the timer callback.
3147  **/
3148
3149 static struct net_device_stats *e1000_get_stats(struct net_device *netdev)
3150 {
3151         /* only return the current stats */
3152         return &netdev->stats;
3153 }
3154
3155 /**
3156  * e1000_change_mtu - Change the Maximum Transfer Unit
3157  * @netdev: network interface device structure
3158  * @new_mtu: new value for maximum frame size
3159  *
3160  * Returns 0 on success, negative on failure
3161  **/
3162
3163 static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
3164 {
3165         struct e1000_adapter *adapter = netdev_priv(netdev);
3166         struct e1000_hw *hw = &adapter->hw;
3167         int max_frame = new_mtu + ENET_HEADER_SIZE + ETHERNET_FCS_SIZE;
3168
3169         if ((max_frame < MINIMUM_ETHERNET_FRAME_SIZE) ||
3170             (max_frame > MAX_JUMBO_FRAME_SIZE)) {
3171                 e_err(probe, "Invalid MTU setting\n");
3172                 return -EINVAL;
3173         }
3174
3175         /* Adapter-specific max frame size limits. */
3176         switch (hw->mac_type) {
3177         case e1000_undefined ... e1000_82542_rev2_1:
3178                 if (max_frame > (ETH_FRAME_LEN + ETH_FCS_LEN)) {
3179                         e_err(probe, "Jumbo Frames not supported.\n");
3180                         return -EINVAL;
3181                 }
3182                 break;
3183         default:
3184                 /* Capable of supporting up to MAX_JUMBO_FRAME_SIZE limit. */
3185                 break;
3186         }
3187
3188         while (test_and_set_bit(__E1000_RESETTING, &adapter->flags))
3189                 msleep(1);
3190         /* e1000_down has a dependency on max_frame_size */
3191         hw->max_frame_size = max_frame;
3192         if (netif_running(netdev))
3193                 e1000_down(adapter);
3194
3195         /* NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
3196          * means we reserve 2 more, this pushes us to allocate from the next
3197          * larger slab size.
3198          * i.e. RXBUFFER_2048 --> size-4096 slab
3199          *  however with the new *_jumbo_rx* routines, jumbo receives will use
3200          *  fragmented skbs */
3201
3202         if (max_frame <= E1000_RXBUFFER_2048)
3203                 adapter->rx_buffer_len = E1000_RXBUFFER_2048;
3204         else
3205 #if (PAGE_SIZE >= E1000_RXBUFFER_16384)
3206                 adapter->rx_buffer_len = E1000_RXBUFFER_16384;
3207 #elif (PAGE_SIZE >= E1000_RXBUFFER_4096)
3208                 adapter->rx_buffer_len = PAGE_SIZE;
3209 #endif
3210
3211         /* adjust allocation if LPE protects us, and we aren't using SBP */
3212         if (!hw->tbi_compatibility_on &&
3213             ((max_frame == (ETH_FRAME_LEN + ETH_FCS_LEN)) ||
3214              (max_frame == MAXIMUM_ETHERNET_VLAN_SIZE)))
3215                 adapter->rx_buffer_len = MAXIMUM_ETHERNET_VLAN_SIZE;
3216
3217         pr_info("%s changing MTU from %d to %d\n",
3218                 netdev->name, netdev->mtu, new_mtu);
3219         netdev->mtu = new_mtu;
3220
3221         if (netif_running(netdev))
3222                 e1000_up(adapter);
3223         else
3224                 e1000_reset(adapter);
3225
3226         clear_bit(__E1000_RESETTING, &adapter->flags);
3227
3228         return 0;
3229 }
3230
3231 /**
3232  * e1000_update_stats - Update the board statistics counters
3233  * @adapter: board private structure
3234  **/
3235
3236 void e1000_update_stats(struct e1000_adapter *adapter)
3237 {
3238         struct net_device *netdev = adapter->netdev;
3239         struct e1000_hw *hw = &adapter->hw;
3240         struct pci_dev *pdev = adapter->pdev;
3241         unsigned long flags;
3242         u16 phy_tmp;
3243
3244 #define PHY_IDLE_ERROR_COUNT_MASK 0x00FF
3245
3246         /*
3247          * Prevent stats update while adapter is being reset, or if the pci
3248          * connection is down.
3249          */
3250         if (adapter->link_speed == 0)
3251                 return;
3252         if (pci_channel_offline(pdev))
3253                 return;
3254
3255         spin_lock_irqsave(&adapter->stats_lock, flags);
3256
3257         /* these counters are modified from e1000_tbi_adjust_stats,
3258          * called from the interrupt context, so they must only
3259          * be written while holding adapter->stats_lock
3260          */
3261
3262         adapter->stats.crcerrs += er32(CRCERRS);
3263         adapter->stats.gprc += er32(GPRC);
3264         adapter->stats.gorcl += er32(GORCL);
3265         adapter->stats.gorch += er32(GORCH);
3266         adapter->stats.bprc += er32(BPRC);
3267         adapter->stats.mprc += er32(MPRC);
3268         adapter->stats.roc += er32(ROC);
3269
3270         adapter->stats.prc64 += er32(PRC64);
3271         adapter->stats.prc127 += er32(PRC127);
3272         adapter->stats.prc255 += er32(PRC255);
3273         adapter->stats.prc511 += er32(PRC511);
3274         adapter->stats.prc1023 += er32(PRC1023);
3275         adapter->stats.prc1522 += er32(PRC1522);
3276
3277         adapter->stats.symerrs += er32(SYMERRS);
3278         adapter->stats.mpc += er32(MPC);
3279         adapter->stats.scc += er32(SCC);
3280         adapter->stats.ecol += er32(ECOL);
3281         adapter->stats.mcc += er32(MCC);
3282         adapter->stats.latecol += er32(LATECOL);
3283         adapter->stats.dc += er32(DC);
3284         adapter->stats.sec += er32(SEC);
3285         adapter->stats.rlec += er32(RLEC);
3286         adapter->stats.xonrxc += er32(XONRXC);
3287         adapter->stats.xontxc += er32(XONTXC);
3288         adapter->stats.xoffrxc += er32(XOFFRXC);
3289         adapter->stats.xofftxc += er32(XOFFTXC);
3290         adapter->stats.fcruc += er32(FCRUC);
3291         adapter->stats.gptc += er32(GPTC);
3292         adapter->stats.gotcl += er32(GOTCL);
3293         adapter->stats.gotch += er32(GOTCH);
3294         adapter->stats.rnbc += er32(RNBC);
3295         adapter->stats.ruc += er32(RUC);
3296         adapter->stats.rfc += er32(RFC);
3297         adapter->stats.rjc += er32(RJC);
3298         adapter->stats.torl += er32(TORL);
3299         adapter->stats.torh += er32(TORH);
3300         adapter->stats.totl += er32(TOTL);
3301         adapter->stats.toth += er32(TOTH);
3302         adapter->stats.tpr += er32(TPR);
3303
3304         adapter->stats.ptc64 += er32(PTC64);
3305         adapter->stats.ptc127 += er32(PTC127);
3306         adapter->stats.ptc255 += er32(PTC255);
3307         adapter->stats.ptc511 += er32(PTC511);
3308         adapter->stats.ptc1023 += er32(PTC1023);
3309         adapter->stats.ptc1522 += er32(PTC1522);
3310
3311         adapter->stats.mptc += er32(MPTC);
3312         adapter->stats.bptc += er32(BPTC);
3313
3314         /* used for adaptive IFS */
3315
3316         hw->tx_packet_delta = er32(TPT);
3317         adapter->stats.tpt += hw->tx_packet_delta;
3318         hw->collision_delta = er32(COLC);
3319         adapter->stats.colc += hw->collision_delta;
3320
3321         if (hw->mac_type >= e1000_82543) {
3322                 adapter->stats.algnerrc += er32(ALGNERRC);
3323                 adapter->stats.rxerrc += er32(RXERRC);
3324                 adapter->stats.tncrs += er32(TNCRS);
3325                 adapter->stats.cexterr += er32(CEXTERR);
3326                 adapter->stats.tsctc += er32(TSCTC);
3327                 adapter->stats.tsctfc += er32(TSCTFC);
3328         }
3329
3330         /* Fill out the OS statistics structure */
3331         netdev->stats.multicast = adapter->stats.mprc;
3332         netdev->stats.collisions = adapter->stats.colc;
3333
3334         /* Rx Errors */
3335
3336         /* RLEC on some newer hardware can be incorrect so build
3337         * our own version based on RUC and ROC */
3338         netdev->stats.rx_errors = adapter->stats.rxerrc +
3339                 adapter->stats.crcerrs + adapter->stats.algnerrc +
3340                 adapter->stats.ruc + adapter->stats.roc +
3341                 adapter->stats.cexterr;
3342         adapter->stats.rlerrc = adapter->stats.ruc + adapter->stats.roc;
3343         netdev->stats.rx_length_errors = adapter->stats.rlerrc;
3344         netdev->stats.rx_crc_errors = adapter->stats.crcerrs;
3345         netdev->stats.rx_frame_errors = adapter->stats.algnerrc;
3346         netdev->stats.rx_missed_errors = adapter->stats.mpc;
3347
3348         /* Tx Errors */
3349         adapter->stats.txerrc = adapter->stats.ecol + adapter->stats.latecol;
3350         netdev->stats.tx_errors = adapter->stats.txerrc;
3351         netdev->stats.tx_aborted_errors = adapter->stats.ecol;
3352         netdev->stats.tx_window_errors = adapter->stats.latecol;
3353         netdev->stats.tx_carrier_errors = adapter->stats.tncrs;
3354         if (hw->bad_tx_carr_stats_fd &&
3355             adapter->link_duplex == FULL_DUPLEX) {
3356                 netdev->stats.tx_carrier_errors = 0;
3357                 adapter->stats.tncrs = 0;
3358         }
3359
3360         /* Tx Dropped needs to be maintained elsewhere */
3361
3362         /* Phy Stats */
3363         if (hw->media_type == e1000_media_type_copper) {
3364                 if ((adapter->link_speed == SPEED_1000) &&
3365                    (!e1000_read_phy_reg(hw, PHY_1000T_STATUS, &phy_tmp))) {
3366                         phy_tmp &= PHY_IDLE_ERROR_COUNT_MASK;
3367                         adapter->phy_stats.idle_errors += phy_tmp;
3368                 }
3369
3370                 if ((hw->mac_type <= e1000_82546) &&
3371                    (hw->phy_type == e1000_phy_m88) &&
3372                    !e1000_read_phy_reg(hw, M88E1000_RX_ERR_CNTR, &phy_tmp))
3373                         adapter->phy_stats.receive_errors += phy_tmp;
3374         }
3375
3376         /* Management Stats */
3377         if (hw->has_smbus) {
3378                 adapter->stats.mgptc += er32(MGTPTC);
3379                 adapter->stats.mgprc += er32(MGTPRC);
3380                 adapter->stats.mgpdc += er32(MGTPDC);
3381         }
3382
3383         spin_unlock_irqrestore(&adapter->stats_lock, flags);
3384 }
3385
3386 /**
3387  * e1000_intr - Interrupt Handler
3388  * @irq: interrupt number
3389  * @data: pointer to a network interface device structure
3390  **/
3391
3392 static irqreturn_t e1000_intr(int irq, void *data)
3393 {
3394         struct net_device *netdev = data;
3395         struct e1000_adapter *adapter = netdev_priv(netdev);
3396         struct e1000_hw *hw = &adapter->hw;
3397         u32 icr = er32(ICR);
3398
3399         if (unlikely((!icr) || test_bit(__E1000_DOWN, &adapter->flags)))
3400                 return IRQ_NONE;  /* Not our interrupt */
3401
3402         if (unlikely(icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC))) {
3403                 hw->get_link_status = 1;
3404                 /* guard against interrupt when we're going down */
3405                 if (!test_bit(__E1000_DOWN, &adapter->flags))
3406                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
3407         }
3408
3409         /* disable interrupts, without the synchronize_irq bit */
3410         ew32(IMC, ~0);
3411         E1000_WRITE_FLUSH();
3412
3413         if (likely(napi_schedule_prep(&adapter->napi))) {
3414                 adapter->total_tx_bytes = 0;
3415                 adapter->total_tx_packets = 0;
3416                 adapter->total_rx_bytes = 0;
3417                 adapter->total_rx_packets = 0;
3418                 __napi_schedule(&adapter->napi);
3419         } else {
3420                 /* this really should not happen! if it does it is basically a
3421                  * bug, but not a hard error, so enable ints and continue */
3422                 if (!test_bit(__E1000_DOWN, &adapter->flags))
3423                         e1000_irq_enable(adapter);
3424         }
3425
3426         return IRQ_HANDLED;
3427 }
3428
3429 /**
3430  * e1000_clean - NAPI Rx polling callback
3431  * @adapter: board private structure
3432  **/
3433 static int e1000_clean(struct napi_struct *napi, int budget)
3434 {
3435         struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter, napi);
3436         int tx_clean_complete = 0, work_done = 0;
3437
3438         tx_clean_complete = e1000_clean_tx_irq(adapter, &adapter->tx_ring[0]);
3439
3440         adapter->clean_rx(adapter, &adapter->rx_ring[0], &work_done, budget);
3441
3442         if (!tx_clean_complete)
3443                 work_done = budget;
3444
3445         /* If budget not fully consumed, exit the polling mode */
3446         if (work_done < budget) {
3447                 if (likely(adapter->itr_setting & 3))
3448                         e1000_set_itr(adapter);
3449                 napi_complete(napi);
3450                 if (!test_bit(__E1000_DOWN, &adapter->flags))
3451                         e1000_irq_enable(adapter);
3452         }
3453
3454         return work_done;
3455 }
3456
3457 /**
3458  * e1000_clean_tx_irq - Reclaim resources after transmit completes
3459  * @adapter: board private structure
3460  **/
3461 static bool e1000_clean_tx_irq(struct e1000_adapter *adapter,
3462                                struct e1000_tx_ring *tx_ring)
3463 {
3464         struct e1000_hw *hw = &adapter->hw;
3465         struct net_device *netdev = adapter->netdev;
3466         struct e1000_tx_desc *tx_desc, *eop_desc;
3467         struct e1000_buffer *buffer_info;
3468         unsigned int i, eop;
3469         unsigned int count = 0;
3470         unsigned int total_tx_bytes=0, total_tx_packets=0;
3471
3472         i = tx_ring->next_to_clean;
3473         eop = tx_ring->buffer_info[i].next_to_watch;
3474         eop_desc = E1000_TX_DESC(*tx_ring, eop);
3475
3476         while ((eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) &&
3477                (count < tx_ring->count)) {
3478                 bool cleaned = false;
3479                 rmb();  /* read buffer_info after eop_desc */
3480                 for ( ; !cleaned; count++) {
3481                         tx_desc = E1000_TX_DESC(*tx_ring, i);
3482                         buffer_info = &tx_ring->buffer_info[i];
3483                         cleaned = (i == eop);
3484
3485                         if (cleaned) {
3486                                 struct sk_buff *skb = buffer_info->skb;
3487                                 unsigned int segs, bytecount;
3488                                 segs = skb_shinfo(skb)->gso_segs ?: 1;
3489                                 /* multiply data chunks by size of headers */
3490                                 bytecount = ((segs - 1) * skb_headlen(skb)) +
3491                                             skb->len;
3492                                 total_tx_packets += segs;
3493                                 total_tx_bytes += bytecount;
3494                         }
3495                         e1000_unmap_and_free_tx_resource(adapter, buffer_info);
3496                         tx_desc->upper.data = 0;
3497
3498                         if (unlikely(++i == tx_ring->count)) i = 0;
3499                 }
3500
3501                 eop = tx_ring->buffer_info[i].next_to_watch;
3502                 eop_desc = E1000_TX_DESC(*tx_ring, eop);
3503         }
3504
3505         tx_ring->next_to_clean = i;
3506
3507 #define TX_WAKE_THRESHOLD 32
3508         if (unlikely(count && netif_carrier_ok(netdev) &&
3509                      E1000_DESC_UNUSED(tx_ring) >= TX_WAKE_THRESHOLD)) {
3510                 /* Make sure that anybody stopping the queue after this
3511                  * sees the new next_to_clean.
3512                  */
3513                 smp_mb();
3514
3515                 if (netif_queue_stopped(netdev) &&
3516                     !(test_bit(__E1000_DOWN, &adapter->flags))) {
3517                         netif_wake_queue(netdev);
3518                         ++adapter->restart_queue;
3519                 }
3520         }
3521
3522         if (adapter->detect_tx_hung) {
3523                 /* Detect a transmit hang in hardware, this serializes the
3524                  * check with the clearing of time_stamp and movement of i */
3525                 adapter->detect_tx_hung = false;
3526                 if (tx_ring->buffer_info[eop].time_stamp &&
3527                     time_after(jiffies, tx_ring->buffer_info[eop].time_stamp +
3528                                (adapter->tx_timeout_factor * HZ)) &&
3529                     !(er32(STATUS) & E1000_STATUS_TXOFF)) {
3530
3531                         /* detected Tx unit hang */
3532                         e_err(drv, "Detected Tx Unit Hang\n"
3533                               "  Tx Queue             <%lu>\n"
3534                               "  TDH                  <%x>\n"
3535                               "  TDT                  <%x>\n"
3536                               "  next_to_use          <%x>\n"
3537                               "  next_to_clean        <%x>\n"
3538                               "buffer_info[next_to_clean]\n"
3539                               "  time_stamp           <%lx>\n"
3540                               "  next_to_watch        <%x>\n"
3541                               "  jiffies              <%lx>\n"
3542                               "  next_to_watch.status <%x>\n",
3543                                 (unsigned long)((tx_ring - adapter->tx_ring) /
3544                                         sizeof(struct e1000_tx_ring)),
3545                                 readl(hw->hw_addr + tx_ring->tdh),
3546                                 readl(hw->hw_addr + tx_ring->tdt),
3547                                 tx_ring->next_to_use,
3548                                 tx_ring->next_to_clean,
3549                                 tx_ring->buffer_info[eop].time_stamp,
3550                                 eop,
3551                                 jiffies,
3552                                 eop_desc->upper.fields.status);
3553                         netif_stop_queue(netdev);
3554                 }
3555         }
3556         adapter->total_tx_bytes += total_tx_bytes;
3557         adapter->total_tx_packets += total_tx_packets;
3558         netdev->stats.tx_bytes += total_tx_bytes;
3559         netdev->stats.tx_packets += total_tx_packets;
3560         return (count < tx_ring->count);
3561 }
3562
3563 /**
3564  * e1000_rx_checksum - Receive Checksum Offload for 82543
3565  * @adapter:     board private structure
3566  * @status_err:  receive descriptor status and error fields
3567  * @csum:        receive descriptor csum field
3568  * @sk_buff:     socket buffer with received data
3569  **/
3570
3571 static void e1000_rx_checksum(struct e1000_adapter *adapter, u32 status_err,
3572                               u32 csum, struct sk_buff *skb)
3573 {
3574         struct e1000_hw *hw = &adapter->hw;
3575         u16 status = (u16)status_err;
3576         u8 errors = (u8)(status_err >> 24);
3577
3578         skb_checksum_none_assert(skb);
3579
3580         /* 82543 or newer only */
3581         if (unlikely(hw->mac_type < e1000_82543)) return;
3582         /* Ignore Checksum bit is set */
3583         if (unlikely(status & E1000_RXD_STAT_IXSM)) return;
3584         /* TCP/UDP checksum error bit is set */
3585         if (unlikely(errors & E1000_RXD_ERR_TCPE)) {
3586                 /* let the stack verify checksum errors */
3587                 adapter->hw_csum_err++;
3588                 return;
3589         }
3590         /* TCP/UDP Checksum has not been calculated */
3591         if (!(status & E1000_RXD_STAT_TCPCS))
3592                 return;
3593
3594         /* It must be a TCP or UDP packet with a valid checksum */
3595         if (likely(status & E1000_RXD_STAT_TCPCS)) {
3596                 /* TCP checksum is good */
3597                 skb->ip_summed = CHECKSUM_UNNECESSARY;
3598         }
3599         adapter->hw_csum_good++;
3600 }
3601
3602 /**
3603  * e1000_consume_page - helper function
3604  **/
3605 static void e1000_consume_page(struct e1000_buffer *bi, struct sk_buff *skb,
3606                                u16 length)
3607 {
3608         bi->page = NULL;
3609         skb->len += length;
3610         skb->data_len += length;
3611         skb->truesize += length;
3612 }
3613
3614 /**
3615  * e1000_receive_skb - helper function to handle rx indications
3616  * @adapter: board private structure
3617  * @status: descriptor status field as written by hardware
3618  * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
3619  * @skb: pointer to sk_buff to be indicated to stack
3620  */
3621 static void e1000_receive_skb(struct e1000_adapter *adapter, u8 status,
3622                               __le16 vlan, struct sk_buff *skb)
3623 {
3624         if (unlikely(adapter->vlgrp && (status & E1000_RXD_STAT_VP))) {
3625                 vlan_hwaccel_receive_skb(skb, adapter->vlgrp,
3626                                          le16_to_cpu(vlan) &
3627                                          E1000_RXD_SPC_VLAN_MASK);
3628         } else {
3629                 netif_receive_skb(skb);
3630         }
3631 }
3632
3633 /**
3634  * e1000_clean_jumbo_rx_irq - Send received data up the network stack; legacy
3635  * @adapter: board private structure
3636  * @rx_ring: ring to clean
3637  * @work_done: amount of napi work completed this call
3638  * @work_to_do: max amount of work allowed for this call to do
3639  *
3640  * the return value indicates whether actual cleaning was done, there
3641  * is no guarantee that everything was cleaned
3642  */
3643 static bool e1000_clean_jumbo_rx_irq(struct e1000_adapter *adapter,
3644                                      struct e1000_rx_ring *rx_ring,
3645                                      int *work_done, int work_to_do)
3646 {
3647         struct e1000_hw *hw = &adapter->hw;
3648         struct net_device *netdev = adapter->netdev;
3649         struct pci_dev *pdev = adapter->pdev;
3650         struct e1000_rx_desc *rx_desc, *next_rxd;
3651         struct e1000_buffer *buffer_info, *next_buffer;
3652         unsigned long irq_flags;
3653         u32 length;
3654         unsigned int i;
3655         int cleaned_count = 0;
3656         bool cleaned = false;
3657         unsigned int total_rx_bytes=0, total_rx_packets=0;
3658
3659         i = rx_ring->next_to_clean;
3660         rx_desc = E1000_RX_DESC(*rx_ring, i);
3661         buffer_info = &rx_ring->buffer_info[i];
3662
3663         while (rx_desc->status & E1000_RXD_STAT_DD) {
3664                 struct sk_buff *skb;
3665                 u8 status;
3666
3667                 if (*work_done >= work_to_do)
3668                         break;
3669                 (*work_done)++;
3670                 rmb(); /* read descriptor and rx_buffer_info after status DD */
3671
3672                 status = rx_desc->status;
3673                 skb = buffer_info->skb;
3674                 buffer_info->skb = NULL;
3675
3676                 if (++i == rx_ring->count) i = 0;
3677                 next_rxd = E1000_RX_DESC(*rx_ring, i);
3678                 prefetch(next_rxd);
3679
3680                 next_buffer = &rx_ring->buffer_info[i];
3681
3682                 cleaned = true;
3683                 cleaned_count++;
3684                 dma_unmap_page(&pdev->dev, buffer_info->dma,
3685                                buffer_info->length, DMA_FROM_DEVICE);
3686                 buffer_info->dma = 0;
3687
3688                 length = le16_to_cpu(rx_desc->length);
3689
3690                 /* errors is only valid for DD + EOP descriptors */
3691                 if (unlikely((status & E1000_RXD_STAT_EOP) &&
3692                     (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK))) {
3693                         u8 last_byte = *(skb->data + length - 1);
3694                         if (TBI_ACCEPT(hw, status, rx_desc->errors, length,
3695                                        last_byte)) {
3696                                 spin_lock_irqsave(&adapter->stats_lock,
3697                                                   irq_flags);
3698                                 e1000_tbi_adjust_stats(hw, &adapter->stats,
3699                                                        length, skb->data);
3700                                 spin_unlock_irqrestore(&adapter->stats_lock,
3701                                                        irq_flags);
3702                                 length--;
3703                         } else {
3704                                 /* recycle both page and skb */
3705                                 buffer_info->skb = skb;
3706                                 /* an error means any chain goes out the window
3707                                  * too */
3708                                 if (rx_ring->rx_skb_top)
3709                                         dev_kfree_skb(rx_ring->rx_skb_top);
3710                                 rx_ring->rx_skb_top = NULL;
3711                                 goto next_desc;
3712                         }
3713                 }
3714
3715 #define rxtop rx_ring->rx_skb_top
3716                 if (!(status & E1000_RXD_STAT_EOP)) {
3717                         /* this descriptor is only the beginning (or middle) */
3718                         if (!rxtop) {
3719                                 /* this is the beginning of a chain */
3720                                 rxtop = skb;
3721                                 skb_fill_page_desc(rxtop, 0, buffer_info->page,
3722                                                    0, length);
3723                         } else {
3724                                 /* this is the middle of a chain */
3725                                 skb_fill_page_desc(rxtop,
3726                                     skb_shinfo(rxtop)->nr_frags,
3727                                     buffer_info->page, 0, length);
3728                                 /* re-use the skb, only consumed the page */
3729                                 buffer_info->skb = skb;
3730                         }
3731                         e1000_consume_page(buffer_info, rxtop, length);
3732                         goto next_desc;
3733                 } else {
3734                         if (rxtop) {
3735                                 /* end of the chain */
3736                                 skb_fill_page_desc(rxtop,
3737                                     skb_shinfo(rxtop)->nr_frags,
3738                                     buffer_info->page, 0, length);
3739                                 /* re-use the current skb, we only consumed the
3740                                  * page */
3741                                 buffer_info->skb = skb;
3742                                 skb = rxtop;
3743                                 rxtop = NULL;
3744                                 e1000_consume_page(buffer_info, skb, length);
3745                         } else {
3746                                 /* no chain, got EOP, this buf is the packet
3747                                  * copybreak to save the put_page/alloc_page */
3748                                 if (length <= copybreak &&
3749                                     skb_tailroom(skb) >= length) {
3750                                         u8 *vaddr;
3751                                         vaddr = kmap_atomic(buffer_info->page,
3752                                                             KM_SKB_DATA_SOFTIRQ);
3753                                         memcpy(skb_tail_pointer(skb), vaddr, length);
3754                                         kunmap_atomic(vaddr,
3755                                                       KM_SKB_DATA_SOFTIRQ);
3756                                         /* re-use the page, so don't erase
3757                                          * buffer_info->page */
3758                                         skb_put(skb, length);
3759                                 } else {
3760                                         skb_fill_page_desc(skb, 0,
3761                                                            buffer_info->page, 0,
3762                                                            length);
3763                                         e1000_consume_page(buffer_info, skb,
3764                                                            length);
3765                                 }
3766                         }
3767                 }
3768
3769                 /* Receive Checksum Offload XXX recompute due to CRC strip? */
3770                 e1000_rx_checksum(adapter,
3771                                   (u32)(status) |
3772                                   ((u32)(rx_desc->errors) << 24),
3773                                   le16_to_cpu(rx_desc->csum), skb);
3774
3775                 pskb_trim(skb, skb->len - 4);
3776
3777                 /* probably a little skewed due to removing CRC */
3778                 total_rx_bytes += skb->len;
3779                 total_rx_packets++;
3780
3781                 /* eth type trans needs skb->data to point to something */
3782                 if (!pskb_may_pull(skb, ETH_HLEN)) {
3783                         e_err(drv, "pskb_may_pull failed.\n");
3784                         dev_kfree_skb(skb);
3785                         goto next_desc;
3786                 }
3787
3788                 skb->protocol = eth_type_trans(skb, netdev);
3789
3790                 e1000_receive_skb(adapter, status, rx_desc->special, skb);
3791
3792 next_desc:
3793                 rx_desc->status = 0;
3794
3795                 /* return some buffers to hardware, one at a time is too slow */
3796                 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
3797                         adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
3798                         cleaned_count = 0;
3799                 }
3800
3801                 /* use prefetched values */
3802                 rx_desc = next_rxd;
3803                 buffer_info = next_buffer;
3804         }
3805         rx_ring->next_to_clean = i;
3806
3807         cleaned_count = E1000_DESC_UNUSED(rx_ring);
3808         if (cleaned_count)
3809                 adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
3810
3811         adapter->total_rx_packets += total_rx_packets;
3812         adapter->total_rx_bytes += total_rx_bytes;
3813         netdev->stats.rx_bytes += total_rx_bytes;
3814         netdev->stats.rx_packets += total_rx_packets;
3815         return cleaned;
3816 }
3817
3818 /*
3819  * this should improve performance for small packets with large amounts
3820  * of reassembly being done in the stack
3821  */
3822 static void e1000_check_copybreak(struct net_device *netdev,
3823                                  struct e1000_buffer *buffer_info,
3824                                  u32 length, struct sk_buff **skb)
3825 {
3826         struct sk_buff *new_skb;
3827
3828         if (length > copybreak)
3829                 return;
3830
3831         new_skb = netdev_alloc_skb_ip_align(netdev, length);
3832         if (!new_skb)
3833                 return;
3834
3835         skb_copy_to_linear_data_offset(new_skb, -NET_IP_ALIGN,
3836                                        (*skb)->data - NET_IP_ALIGN,
3837                                        length + NET_IP_ALIGN);
3838         /* save the skb in buffer_info as good */
3839         buffer_info->skb = *skb;
3840         *skb = new_skb;
3841 }
3842
3843 /**
3844  * e1000_clean_rx_irq - Send received data up the network stack; legacy
3845  * @adapter: board private structure
3846  * @rx_ring: ring to clean
3847  * @work_done: amount of napi work completed this call
3848  * @work_to_do: max amount of work allowed for this call to do
3849  */
3850 static bool e1000_clean_rx_irq(struct e1000_adapter *adapter,
3851                                struct e1000_rx_ring *rx_ring,
3852                                int *work_done, int work_to_do)
3853 {
3854         struct e1000_hw *hw = &adapter->hw;
3855         struct net_device *netdev = adapter->netdev;
3856         struct pci_dev *pdev = adapter->pdev;
3857         struct e1000_rx_desc *rx_desc, *next_rxd;
3858         struct e1000_buffer *buffer_info, *next_buffer;
3859         unsigned long flags;
3860         u32 length;
3861         unsigned int i;
3862         int cleaned_count = 0;
3863         bool cleaned = false;
3864         unsigned int total_rx_bytes=0, total_rx_packets=0;
3865
3866         i = rx_ring->next_to_clean;
3867         rx_desc = E1000_RX_DESC(*rx_ring, i);
3868         buffer_info = &rx_ring->buffer_info[i];
3869
3870         while (rx_desc->status & E1000_RXD_STAT_DD) {
3871                 struct sk_buff *skb;
3872                 u8 status;
3873
3874                 if (*work_done >= work_to_do)
3875                         break;
3876                 (*work_done)++;
3877                 rmb(); /* read descriptor and rx_buffer_info after status DD */
3878
3879                 status = rx_desc->status;
3880                 skb = buffer_info->skb;
3881                 buffer_info->skb = NULL;
3882
3883                 prefetch(skb->data - NET_IP_ALIGN);
3884
3885                 if (++i == rx_ring->count) i = 0;
3886                 next_rxd = E1000_RX_DESC(*rx_ring, i);
3887                 prefetch(next_rxd);
3888
3889                 next_buffer = &rx_ring->buffer_info[i];
3890
3891                 cleaned = true;
3892                 cleaned_count++;
3893                 dma_unmap_single(&pdev->dev, buffer_info->dma,
3894                                  buffer_info->length, DMA_FROM_DEVICE);
3895                 buffer_info->dma = 0;
3896
3897                 length = le16_to_cpu(rx_desc->length);
3898                 /* !EOP means multiple descriptors were used to store a single
3899                  * packet, if thats the case we need to toss it.  In fact, we
3900                  * to toss every packet with the EOP bit clear and the next
3901                  * frame that _does_ have the EOP bit set, as it is by
3902                  * definition only a frame fragment
3903                  */
3904                 if (unlikely(!(status & E1000_RXD_STAT_EOP)))
3905                         adapter->discarding = true;
3906
3907                 if (adapter->discarding) {
3908                         /* All receives must fit into a single buffer */
3909                         e_dbg("Receive packet consumed multiple buffers\n");
3910                         /* recycle */
3911                         buffer_info->skb = skb;
3912                         if (status & E1000_RXD_STAT_EOP)
3913                                 adapter->discarding = false;
3914                         goto next_desc;
3915                 }
3916
3917                 if (unlikely(rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK)) {
3918                         u8 last_byte = *(skb->data + length - 1);
3919                         if (TBI_ACCEPT(hw, status, rx_desc->errors, length,
3920                                        last_byte)) {
3921                                 spin_lock_irqsave(&adapter->stats_lock, flags);
3922                                 e1000_tbi_adjust_stats(hw, &adapter->stats,
3923                                                        length, skb->data);
3924                                 spin_unlock_irqrestore(&adapter->stats_lock,
3925                                                        flags);
3926                                 length--;
3927                         } else {
3928                                 /* recycle */
3929                                 buffer_info->skb = skb;
3930                                 goto next_desc;
3931                         }
3932                 }
3933
3934                 /* adjust length to remove Ethernet CRC, this must be
3935                  * done after the TBI_ACCEPT workaround above */
3936                 length -= 4;
3937
3938                 /* probably a little skewed due to removing CRC */
3939                 total_rx_bytes += length;
3940                 total_rx_packets++;
3941
3942                 e1000_check_copybreak(netdev, buffer_info, length, &skb);
3943
3944                 skb_put(skb, length);
3945
3946                 /* Receive Checksum Offload */
3947                 e1000_rx_checksum(adapter,
3948                                   (u32)(status) |
3949                                   ((u32)(rx_desc->errors) << 24),
3950                                   le16_to_cpu(rx_desc->csum), skb);
3951
3952                 skb->protocol = eth_type_trans(skb, netdev);
3953
3954                 e1000_receive_skb(adapter, status, rx_desc->special, skb);
3955
3956 next_desc:
3957                 rx_desc->status = 0;
3958
3959                 /* return some buffers to hardware, one at a time is too slow */
3960                 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
3961                         adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
3962                         cleaned_count = 0;
3963                 }
3964
3965                 /* use prefetched values */
3966                 rx_desc = next_rxd;
3967                 buffer_info = next_buffer;
3968         }
3969         rx_ring->next_to_clean = i;
3970
3971         cleaned_count = E1000_DESC_UNUSED(rx_ring);
3972         if (cleaned_count)
3973                 adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
3974
3975         adapter->total_rx_packets += total_rx_packets;
3976         adapter->total_rx_bytes += total_rx_bytes;
3977         netdev->stats.rx_bytes += total_rx_bytes;
3978         netdev->stats.rx_packets += total_rx_packets;
3979         return cleaned;
3980 }
3981
3982 /**
3983  * e1000_alloc_jumbo_rx_buffers - Replace used jumbo receive buffers
3984  * @adapter: address of board private structure
3985  * @rx_ring: pointer to receive ring structure
3986  * @cleaned_count: number of buffers to allocate this pass
3987  **/
3988
3989 static void
3990 e1000_alloc_jumbo_rx_buffers(struct e1000_adapter *adapter,
3991                              struct e1000_rx_ring *rx_ring, int cleaned_count)
3992 {
3993         struct net_device *netdev = adapter->netdev;
3994         struct pci_dev *pdev = adapter->pdev;
3995         struct e1000_rx_desc *rx_desc;
3996         struct e1000_buffer *buffer_info;
3997         struct sk_buff *skb;
3998         unsigned int i;
3999         unsigned int bufsz = 256 - 16 /*for skb_reserve */ ;
4000
4001         i = rx_ring->next_to_use;
4002         buffer_info = &rx_ring->buffer_info[i];
4003
4004         while (cleaned_count--) {
4005                 skb = buffer_info->skb;
4006                 if (skb) {
4007                         skb_trim(skb, 0);
4008                         goto check_page;
4009                 }
4010
4011                 skb = netdev_alloc_skb_ip_align(netdev, bufsz);
4012                 if (unlikely(!skb)) {
4013                         /* Better luck next round */
4014                         adapter->alloc_rx_buff_failed++;
4015                         break;
4016                 }
4017
4018                 /* Fix for errata 23, can't cross 64kB boundary */
4019                 if (!e1000_check_64k_bound(adapter, skb->data, bufsz)) {
4020                         struct sk_buff *oldskb = skb;
4021                         e_err(rx_err, "skb align check failed: %u bytes at "
4022                               "%p\n", bufsz, skb->data);
4023                         /* Try again, without freeing the previous */
4024                         skb = netdev_alloc_skb_ip_align(netdev, bufsz);
4025                         /* Failed allocation, critical failure */
4026                         if (!skb) {
4027                                 dev_kfree_skb(oldskb);
4028                                 adapter->alloc_rx_buff_failed++;
4029                                 break;
4030                         }
4031
4032                         if (!e1000_check_64k_bound(adapter, skb->data, bufsz)) {
4033                                 /* give up */
4034                                 dev_kfree_skb(skb);
4035                                 dev_kfree_skb(oldskb);
4036                                 break; /* while (cleaned_count--) */
4037                         }
4038
4039                         /* Use new allocation */
4040                         dev_kfree_skb(oldskb);
4041                 }
4042                 buffer_info->skb = skb;
4043                 buffer_info->length = adapter->rx_buffer_len;
4044 check_page:
4045                 /* allocate a new page if necessary */
4046                 if (!buffer_info->page) {
4047                         buffer_info->page = alloc_page(GFP_ATOMIC);
4048                         if (unlikely(!buffer_info->page)) {
4049                                 adapter->alloc_rx_buff_failed++;
4050                                 break;
4051                         }
4052                 }
4053
4054                 if (!buffer_info->dma) {
4055                         buffer_info->dma = dma_map_page(&pdev->dev,
4056                                                         buffer_info->page, 0,
4057                                                         buffer_info->length,
4058                                                         DMA_FROM_DEVICE);
4059                         if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
4060                                 put_page(buffer_info->page);
4061                                 dev_kfree_skb(skb);
4062                                 buffer_info->page = NULL;
4063                                 buffer_info->skb = NULL;
4064                                 buffer_info->dma = 0;
4065                                 adapter->alloc_rx_buff_failed++;
4066                                 break; /* while !buffer_info->skb */
4067                         }
4068                 }
4069
4070                 rx_desc = E1000_RX_DESC(*rx_ring, i);
4071                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
4072
4073                 if (unlikely(++i == rx_ring->count))
4074                         i = 0;
4075                 buffer_info = &rx_ring->buffer_info[i];
4076         }
4077
4078         if (likely(rx_ring->next_to_use != i)) {
4079                 rx_ring->next_to_use = i;
4080                 if (unlikely(i-- == 0))
4081                         i = (rx_ring->count - 1);
4082
4083                 /* Force memory writes to complete before letting h/w
4084                  * know there are new descriptors to fetch.  (Only
4085                  * applicable for weak-ordered memory model archs,
4086                  * such as IA-64). */
4087                 wmb();
4088                 writel(i, adapter->hw.hw_addr + rx_ring->rdt);
4089         }
4090 }
4091
4092 /**
4093  * e1000_alloc_rx_buffers - Replace used receive buffers; legacy & extended
4094  * @adapter: address of board private structure
4095  **/
4096
4097 static void e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
4098                                    struct e1000_rx_ring *rx_ring,
4099                                    int cleaned_count)
4100 {
4101         struct e1000_hw *hw = &adapter->hw;
4102         struct net_device *netdev = adapter->netdev;
4103         struct pci_dev *pdev = adapter->pdev;
4104         struct e1000_rx_desc *rx_desc;
4105         struct e1000_buffer *buffer_info;
4106         struct sk_buff *skb;
4107         unsigned int i;
4108         unsigned int bufsz = adapter->rx_buffer_len;
4109
4110         i = rx_ring->next_to_use;
4111         buffer_info = &rx_ring->buffer_info[i];
4112
4113         while (cleaned_count--) {
4114                 skb = buffer_info->skb;
4115                 if (skb) {
4116                         skb_trim(skb, 0);
4117                         goto map_skb;
4118                 }
4119
4120                 skb = netdev_alloc_skb_ip_align(netdev, bufsz);
4121                 if (unlikely(!skb)) {
4122                         /* Better luck next round */
4123                         adapter->alloc_rx_buff_failed++;
4124                         break;
4125                 }
4126
4127                 /* Fix for errata 23, can't cross 64kB boundary */
4128                 if (!e1000_check_64k_bound(adapter, skb->data, bufsz)) {
4129                         struct sk_buff *oldskb = skb;
4130                         e_err(rx_err, "skb align check failed: %u bytes at "
4131                               "%p\n", bufsz, skb->data);
4132                         /* Try again, without freeing the previous */
4133                         skb = netdev_alloc_skb_ip_align(netdev, bufsz);
4134                         /* Failed allocation, critical failure */
4135                         if (!skb) {
4136                                 dev_kfree_skb(oldskb);
4137                                 adapter->alloc_rx_buff_failed++;
4138                                 break;
4139                         }
4140
4141                         if (!e1000_check_64k_bound(adapter, skb->data, bufsz)) {
4142                                 /* give up */
4143                                 dev_kfree_skb(skb);
4144                                 dev_kfree_skb(oldskb);
4145                                 adapter->alloc_rx_buff_failed++;
4146                                 break; /* while !buffer_info->skb */
4147                         }
4148
4149                         /* Use new allocation */
4150                         dev_kfree_skb(oldskb);
4151                 }
4152                 buffer_info->skb = skb;
4153                 buffer_info->length = adapter->rx_buffer_len;
4154 map_skb:
4155                 buffer_info->dma = dma_map_single(&pdev->dev,
4156                                                   skb->data,
4157                                                   buffer_info->length,
4158                                                   DMA_FROM_DEVICE);
4159                 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
4160                         dev_kfree_skb(skb);
4161                         buffer_info->skb = NULL;
4162                         buffer_info->dma = 0;
4163                         adapter->alloc_rx_buff_failed++;
4164                         break; /* while !buffer_info->skb */
4165                 }
4166
4167                 /*
4168                  * XXX if it was allocated cleanly it will never map to a
4169                  * boundary crossing
4170                  */
4171
4172                 /* Fix for errata 23, can't cross 64kB boundary */
4173                 if (!e1000_check_64k_bound(adapter,
4174                                         (void *)(unsigned long)buffer_info->dma,
4175                                         adapter->rx_buffer_len)) {
4176                         e_err(rx_err, "dma align check failed: %u bytes at "
4177                               "%p\n", adapter->rx_buffer_len,
4178                               (void *)(unsigned long)buffer_info->dma);
4179                         dev_kfree_skb(skb);
4180                         buffer_info->skb = NULL;
4181
4182                         dma_unmap_single(&pdev->dev, buffer_info->dma,
4183                                          adapter->rx_buffer_len,
4184                                          DMA_FROM_DEVICE);
4185                         buffer_info->dma = 0;
4186
4187                         adapter->alloc_rx_buff_failed++;
4188                         break; /* while !buffer_info->skb */
4189                 }
4190                 rx_desc = E1000_RX_DESC(*rx_ring, i);
4191                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
4192
4193                 if (unlikely(++i == rx_ring->count))
4194                         i = 0;
4195                 buffer_info = &rx_ring->buffer_info[i];
4196         }
4197
4198         if (likely(rx_ring->next_to_use != i)) {
4199                 rx_ring->next_to_use = i;
4200                 if (unlikely(i-- == 0))
4201                         i = (rx_ring->count - 1);
4202
4203                 /* Force memory writes to complete before letting h/w
4204                  * know there are new descriptors to fetch.  (Only
4205                  * applicable for weak-ordered memory model archs,
4206                  * such as IA-64). */
4207                 wmb();
4208                 writel(i, hw->hw_addr + rx_ring->rdt);
4209         }
4210 }
4211
4212 /**
4213  * e1000_smartspeed - Workaround for SmartSpeed on 82541 and 82547 controllers.
4214  * @adapter:
4215  **/
4216
4217 static void e1000_smartspeed(struct e1000_adapter *adapter)
4218 {
4219         struct e1000_hw *hw = &adapter->hw;
4220         u16 phy_status;
4221         u16 phy_ctrl;
4222
4223         if ((hw->phy_type != e1000_phy_igp) || !hw->autoneg ||
4224            !(hw->autoneg_advertised & ADVERTISE_1000_FULL))
4225                 return;
4226
4227         if (adapter->smartspeed == 0) {
4228                 /* If Master/Slave config fault is asserted twice,
4229                  * we assume back-to-back */
4230                 e1000_read_phy_reg(hw, PHY_1000T_STATUS, &phy_status);
4231                 if (!(phy_status & SR_1000T_MS_CONFIG_FAULT)) return;
4232                 e1000_read_phy_reg(hw, PHY_1000T_STATUS, &phy_status);
4233                 if (!(phy_status & SR_1000T_MS_CONFIG_FAULT)) return;
4234                 e1000_read_phy_reg(hw, PHY_1000T_CTRL, &phy_ctrl);
4235                 if (phy_ctrl & CR_1000T_MS_ENABLE) {
4236                         phy_ctrl &= ~CR_1000T_MS_ENABLE;
4237                         e1000_write_phy_reg(hw, PHY_1000T_CTRL,
4238                                             phy_ctrl);
4239                         adapter->smartspeed++;
4240                         if (!e1000_phy_setup_autoneg(hw) &&
4241                            !e1000_read_phy_reg(hw, PHY_CTRL,
4242                                                &phy_ctrl)) {
4243                                 phy_ctrl |= (MII_CR_AUTO_NEG_EN |
4244                                              MII_CR_RESTART_AUTO_NEG);
4245                                 e1000_write_phy_reg(hw, PHY_CTRL,
4246                                                     phy_ctrl);
4247                         }
4248                 }
4249                 return;
4250         } else if (adapter->smartspeed == E1000_SMARTSPEED_DOWNSHIFT) {
4251                 /* If still no link, perhaps using 2/3 pair cable */
4252                 e1000_read_phy_reg(hw, PHY_1000T_CTRL, &phy_ctrl);
4253                 phy_ctrl |= CR_1000T_MS_ENABLE;
4254                 e1000_write_phy_reg(hw, PHY_1000T_CTRL, phy_ctrl);
4255                 if (!e1000_phy_setup_autoneg(hw) &&
4256                    !e1000_read_phy_reg(hw, PHY_CTRL, &phy_ctrl)) {
4257                         phy_ctrl |= (MII_CR_AUTO_NEG_EN |
4258                                      MII_CR_RESTART_AUTO_NEG);
4259                         e1000_write_phy_reg(hw, PHY_CTRL, phy_ctrl);
4260                 }
4261         }
4262         /* Restart process after E1000_SMARTSPEED_MAX iterations */
4263         if (adapter->smartspeed++ == E1000_SMARTSPEED_MAX)
4264                 adapter->smartspeed = 0;
4265 }
4266
4267 /**
4268  * e1000_ioctl -
4269  * @netdev:
4270  * @ifreq:
4271  * @cmd:
4272  **/
4273
4274 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
4275 {
4276         switch (cmd) {
4277         case SIOCGMIIPHY:
4278         case SIOCGMIIREG:
4279         case SIOCSMIIREG:
4280                 return e1000_mii_ioctl(netdev, ifr, cmd);
4281         default:
4282                 return -EOPNOTSUPP;
4283         }
4284 }
4285
4286 /**
4287  * e1000_mii_ioctl -
4288  * @netdev:
4289  * @ifreq:
4290  * @cmd:
4291  **/
4292
4293 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
4294                            int cmd)
4295 {
4296         struct e1000_adapter *adapter = netdev_priv(netdev);
4297         struct e1000_hw *hw = &adapter->hw;
4298         struct mii_ioctl_data *data = if_mii(ifr);
4299         int retval;
4300         u16 mii_reg;
4301         u16 spddplx;
4302         unsigned long flags;
4303
4304         if (hw->media_type != e1000_media_type_copper)
4305                 return -EOPNOTSUPP;
4306
4307         switch (cmd) {
4308         case SIOCGMIIPHY:
4309                 data->phy_id = hw->phy_addr;
4310                 break;
4311         case SIOCGMIIREG:
4312                 spin_lock_irqsave(&adapter->stats_lock, flags);
4313                 if (e1000_read_phy_reg(hw, data->reg_num & 0x1F,
4314                                    &data->val_out)) {
4315                         spin_unlock_irqrestore(&adapter->stats_lock, flags);
4316                         return -EIO;
4317                 }
4318                 spin_unlock_irqrestore(&adapter->stats_lock, flags);
4319                 break;
4320         case SIOCSMIIREG:
4321                 if (data->reg_num & ~(0x1F))
4322                         return -EFAULT;
4323                 mii_reg = data->val_in;
4324                 spin_lock_irqsave(&adapter->stats_lock, flags);
4325                 if (e1000_write_phy_reg(hw, data->reg_num,
4326                                         mii_reg)) {
4327                         spin_unlock_irqrestore(&adapter->stats_lock, flags);
4328                         return -EIO;
4329                 }
4330                 spin_unlock_irqrestore(&adapter->stats_lock, flags);
4331                 if (hw->media_type == e1000_media_type_copper) {
4332                         switch (data->reg_num) {
4333                         case PHY_CTRL:
4334                                 if (mii_reg & MII_CR_POWER_DOWN)
4335                                         break;
4336                                 if (mii_reg & MII_CR_AUTO_NEG_EN) {
4337                                         hw->autoneg = 1;
4338                                         hw->autoneg_advertised = 0x2F;
4339                                 } else {
4340                                         if (mii_reg & 0x40)
4341                                                 spddplx = SPEED_1000;
4342                                         else if (mii_reg & 0x2000)
4343                                                 spddplx = SPEED_100;
4344                                         else
4345                                                 spddplx = SPEED_10;
4346                                         spddplx += (mii_reg & 0x100)
4347                                                    ? DUPLEX_FULL :
4348                                                    DUPLEX_HALF;
4349                                         retval = e1000_set_spd_dplx(adapter,
4350                                                                     spddplx);
4351                                         if (retval)
4352                                                 return retval;
4353                                 }
4354                                 if (netif_running(adapter->netdev))
4355                                         e1000_reinit_locked(adapter);
4356                                 else
4357                                         e1000_reset(adapter);
4358                                 break;
4359                         case M88E1000_PHY_SPEC_CTRL:
4360                         case M88E1000_EXT_PHY_SPEC_CTRL:
4361                                 if (e1000_phy_reset(hw))
4362                                         return -EIO;
4363                                 break;
4364                         }
4365                 } else {
4366                         switch (data->reg_num) {
4367                         case PHY_CTRL:
4368                                 if (mii_reg & MII_CR_POWER_DOWN)
4369                                         break;
4370                                 if (netif_running(adapter->netdev))
4371                                         e1000_reinit_locked(adapter);
4372                                 else
4373                                         e1000_reset(adapter);
4374                                 break;
4375                         }
4376                 }
4377                 break;
4378         default:
4379                 return -EOPNOTSUPP;
4380         }
4381         return E1000_SUCCESS;
4382 }
4383
4384 void e1000_pci_set_mwi(struct e1000_hw *hw)
4385 {
4386         struct e1000_adapter *adapter = hw->back;
4387         int ret_val = pci_set_mwi(adapter->pdev);
4388
4389         if (ret_val)
4390                 e_err(probe, "Error in setting MWI\n");
4391 }
4392
4393 void e1000_pci_clear_mwi(struct e1000_hw *hw)
4394 {
4395         struct e1000_adapter *adapter = hw->back;
4396
4397         pci_clear_mwi(adapter->pdev);
4398 }
4399
4400 int e1000_pcix_get_mmrbc(struct e1000_hw *hw)
4401 {
4402         struct e1000_adapter *adapter = hw->back;
4403         return pcix_get_mmrbc(adapter->pdev);
4404 }
4405
4406 void e1000_pcix_set_mmrbc(struct e1000_hw *hw, int mmrbc)
4407 {
4408         struct e1000_adapter *adapter = hw->back;
4409         pcix_set_mmrbc(adapter->pdev, mmrbc);
4410 }
4411
4412 void e1000_io_write(struct e1000_hw *hw, unsigned long port, u32 value)
4413 {
4414         outl(value, port);
4415 }
4416
4417 static void e1000_vlan_rx_register(struct net_device *netdev,
4418                                    struct vlan_group *grp)
4419 {
4420         struct e1000_adapter *adapter = netdev_priv(netdev);
4421         struct e1000_hw *hw = &adapter->hw;
4422         u32 ctrl, rctl;
4423
4424         if (!test_bit(__E1000_DOWN, &adapter->flags))
4425                 e1000_irq_disable(adapter);
4426         adapter->vlgrp = grp;
4427
4428         if (grp) {
4429                 /* enable VLAN tag insert/strip */
4430                 ctrl = er32(CTRL);
4431                 ctrl |= E1000_CTRL_VME;
4432                 ew32(CTRL, ctrl);
4433
4434                 /* enable VLAN receive filtering */
4435                 rctl = er32(RCTL);
4436                 rctl &= ~E1000_RCTL_CFIEN;
4437                 if (!(netdev->flags & IFF_PROMISC))
4438                         rctl |= E1000_RCTL_VFE;
4439                 ew32(RCTL, rctl);
4440                 e1000_update_mng_vlan(adapter);
4441         } else {
4442                 /* disable VLAN tag insert/strip */
4443                 ctrl = er32(CTRL);
4444                 ctrl &= ~E1000_CTRL_VME;
4445                 ew32(CTRL, ctrl);
4446
4447                 /* disable VLAN receive filtering */
4448                 rctl = er32(RCTL);
4449                 rctl &= ~E1000_RCTL_VFE;
4450                 ew32(RCTL, rctl);
4451
4452                 if (adapter->mng_vlan_id != (u16)E1000_MNG_VLAN_NONE) {
4453                         e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
4454                         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
4455                 }
4456         }
4457
4458         if (!test_bit(__E1000_DOWN, &adapter->flags))
4459                 e1000_irq_enable(adapter);
4460 }
4461
4462 static void e1000_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
4463 {
4464         struct e1000_adapter *adapter = netdev_priv(netdev);
4465         struct e1000_hw *hw = &adapter->hw;
4466         u32 vfta, index;
4467
4468         if ((hw->mng_cookie.status &
4469              E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT) &&
4470             (vid == adapter->mng_vlan_id))
4471                 return;
4472         /* add VID to filter table */
4473         index = (vid >> 5) & 0x7F;
4474         vfta = E1000_READ_REG_ARRAY(hw, VFTA, index);
4475         vfta |= (1 << (vid & 0x1F));
4476         e1000_write_vfta(hw, index, vfta);
4477 }
4478
4479 static void e1000_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
4480 {
4481         struct e1000_adapter *adapter = netdev_priv(netdev);
4482         struct e1000_hw *hw = &adapter->hw;
4483         u32 vfta, index;
4484
4485         if (!test_bit(__E1000_DOWN, &adapter->flags))
4486                 e1000_irq_disable(adapter);
4487         vlan_group_set_device(adapter->vlgrp, vid, NULL);
4488         if (!test_bit(__E1000_DOWN, &adapter->flags))
4489                 e1000_irq_enable(adapter);
4490
4491         /* remove VID from filter table */
4492         index = (vid >> 5) & 0x7F;
4493         vfta = E1000_READ_REG_ARRAY(hw, VFTA, index);
4494         vfta &= ~(1 << (vid & 0x1F));
4495         e1000_write_vfta(hw, index, vfta);
4496 }
4497
4498 static void e1000_restore_vlan(struct e1000_adapter *adapter)
4499 {
4500         e1000_vlan_rx_register(adapter->netdev, adapter->vlgrp);
4501
4502         if (adapter->vlgrp) {
4503                 u16 vid;
4504                 for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
4505                         if (!vlan_group_get_device(adapter->vlgrp, vid))
4506                                 continue;
4507                         e1000_vlan_rx_add_vid(adapter->netdev, vid);
4508                 }
4509         }
4510 }
4511
4512 int e1000_set_spd_dplx(struct e1000_adapter *adapter, u16 spddplx)
4513 {
4514         struct e1000_hw *hw = &adapter->hw;
4515
4516         hw->autoneg = 0;
4517
4518         /* Fiber NICs only allow 1000 gbps Full duplex */
4519         if ((hw->media_type == e1000_media_type_fiber) &&
4520                 spddplx != (SPEED_1000 + DUPLEX_FULL)) {
4521                 e_err(probe, "Unsupported Speed/Duplex configuration\n");
4522                 return -EINVAL;
4523         }
4524
4525         switch (spddplx) {
4526         case SPEED_10 + DUPLEX_HALF:
4527                 hw->forced_speed_duplex = e1000_10_half;
4528                 break;
4529         case SPEED_10 + DUPLEX_FULL:
4530                 hw->forced_speed_duplex = e1000_10_full;
4531                 break;
4532         case SPEED_100 + DUPLEX_HALF:
4533                 hw->forced_speed_duplex = e1000_100_half;
4534                 break;
4535         case SPEED_100 + DUPLEX_FULL:
4536                 hw->forced_speed_duplex = e1000_100_full;
4537                 break;
4538         case SPEED_1000 + DUPLEX_FULL:
4539                 hw->autoneg = 1;
4540                 hw->autoneg_advertised = ADVERTISE_1000_FULL;
4541                 break;
4542         case SPEED_1000 + DUPLEX_HALF: /* not supported */
4543         default:
4544                 e_err(probe, "Unsupported Speed/Duplex configuration\n");
4545                 return -EINVAL;
4546         }
4547         return 0;
4548 }
4549
4550 static int __e1000_shutdown(struct pci_dev *pdev, bool *enable_wake)
4551 {
4552         struct net_device *netdev = pci_get_drvdata(pdev);
4553         struct e1000_adapter *adapter = netdev_priv(netdev);
4554         struct e1000_hw *hw = &adapter->hw;
4555         u32 ctrl, ctrl_ext, rctl, status;
4556         u32 wufc = adapter->wol;
4557 #ifdef CONFIG_PM
4558         int retval = 0;
4559 #endif
4560
4561         netif_device_detach(netdev);
4562
4563         if (netif_running(netdev)) {
4564                 WARN_ON(test_bit(__E1000_RESETTING, &adapter->flags));
4565                 e1000_down(adapter);
4566         }
4567
4568 #ifdef CONFIG_PM
4569         retval = pci_save_state(pdev);
4570         if (retval)
4571                 return retval;
4572 #endif
4573
4574         status = er32(STATUS);
4575         if (status & E1000_STATUS_LU)
4576                 wufc &= ~E1000_WUFC_LNKC;
4577
4578         if (wufc) {
4579                 e1000_setup_rctl(adapter);
4580                 e1000_set_rx_mode(netdev);
4581
4582                 /* turn on all-multi mode if wake on multicast is enabled */
4583                 if (wufc & E1000_WUFC_MC) {
4584                         rctl = er32(RCTL);
4585                         rctl |= E1000_RCTL_MPE;
4586                         ew32(RCTL, rctl);
4587                 }
4588
4589                 if (hw->mac_type >= e1000_82540) {
4590                         ctrl = er32(CTRL);
4591                         /* advertise wake from D3Cold */
4592                         #define E1000_CTRL_ADVD3WUC 0x00100000
4593                         /* phy power management enable */
4594                         #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
4595                         ctrl |= E1000_CTRL_ADVD3WUC |
4596                                 E1000_CTRL_EN_PHY_PWR_MGMT;
4597                         ew32(CTRL, ctrl);
4598                 }
4599
4600                 if (hw->media_type == e1000_media_type_fiber ||
4601                     hw->media_type == e1000_media_type_internal_serdes) {
4602                         /* keep the laser running in D3 */
4603                         ctrl_ext = er32(CTRL_EXT);
4604                         ctrl_ext |= E1000_CTRL_EXT_SDP7_DATA;
4605                         ew32(CTRL_EXT, ctrl_ext);
4606                 }
4607
4608                 ew32(WUC, E1000_WUC_PME_EN);
4609                 ew32(WUFC, wufc);
4610         } else {
4611                 ew32(WUC, 0);
4612                 ew32(WUFC, 0);
4613         }
4614
4615         e1000_release_manageability(adapter);
4616
4617         *enable_wake = !!wufc;
4618
4619         /* make sure adapter isn't asleep if manageability is enabled */
4620         if (adapter->en_mng_pt)
4621                 *enable_wake = true;
4622
4623         if (netif_running(netdev))
4624                 e1000_free_irq(adapter);
4625
4626         pci_disable_device(pdev);
4627
4628         return 0;
4629 }
4630
4631 #ifdef CONFIG_PM
4632 static int e1000_suspend(struct pci_dev *pdev, pm_message_t state)
4633 {
4634         int retval;
4635         bool wake;
4636
4637         retval = __e1000_shutdown(pdev, &wake);
4638         if (retval)
4639                 return retval;
4640
4641         if (wake) {
4642                 pci_prepare_to_sleep(pdev);
4643         } else {
4644                 pci_wake_from_d3(pdev, false);
4645                 pci_set_power_state(pdev, PCI_D3hot);
4646         }
4647
4648         return 0;
4649 }
4650
4651 static int e1000_resume(struct pci_dev *pdev)
4652 {
4653         struct net_device *netdev = pci_get_drvdata(pdev);
4654         struct e1000_adapter *adapter = netdev_priv(netdev);
4655         struct e1000_hw *hw = &adapter->hw;
4656         u32 err;
4657
4658         pci_set_power_state(pdev, PCI_D0);
4659         pci_restore_state(pdev);
4660         pci_save_state(pdev);
4661
4662         if (adapter->need_ioport)
4663                 err = pci_enable_device(pdev);
4664         else
4665                 err = pci_enable_device_mem(pdev);
4666         if (err) {
4667                 pr_err("Cannot enable PCI device from suspend\n");
4668                 return err;
4669         }
4670         pci_set_master(pdev);
4671
4672         pci_enable_wake(pdev, PCI_D3hot, 0);
4673         pci_enable_wake(pdev, PCI_D3cold, 0);
4674
4675         if (netif_running(netdev)) {
4676                 err = e1000_request_irq(adapter);
4677                 if (err)
4678                         return err;
4679         }
4680
4681         e1000_power_up_phy(adapter);
4682         e1000_reset(adapter);
4683         ew32(WUS, ~0);
4684
4685         e1000_init_manageability(adapter);
4686
4687         if (netif_running(netdev))
4688                 e1000_up(adapter);
4689
4690         netif_device_attach(netdev);
4691
4692         return 0;
4693 }
4694 #endif
4695
4696 static void e1000_shutdown(struct pci_dev *pdev)
4697 {
4698         bool wake;
4699
4700         __e1000_shutdown(pdev, &wake);
4701
4702         if (system_state == SYSTEM_POWER_OFF) {
4703                 pci_wake_from_d3(pdev, wake);
4704                 pci_set_power_state(pdev, PCI_D3hot);
4705         }
4706 }
4707
4708 #ifdef CONFIG_NET_POLL_CONTROLLER
4709 /*
4710  * Polling 'interrupt' - used by things like netconsole to send skbs
4711  * without having to re-enable interrupts. It's not called while
4712  * the interrupt routine is executing.
4713  */
4714 static void e1000_netpoll(struct net_device *netdev)
4715 {
4716         struct e1000_adapter *adapter = netdev_priv(netdev);
4717
4718         disable_irq(adapter->pdev->irq);
4719         e1000_intr(adapter->pdev->irq, netdev);
4720         enable_irq(adapter->pdev->irq);
4721 }
4722 #endif
4723
4724 /**
4725  * e1000_io_error_detected - called when PCI error is detected
4726  * @pdev: Pointer to PCI device
4727  * @state: The current pci connection state
4728  *
4729  * This function is called after a PCI bus error affecting
4730  * this device has been detected.
4731  */
4732 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
4733                                                 pci_channel_state_t state)
4734 {
4735         struct net_device *netdev = pci_get_drvdata(pdev);
4736         struct e1000_adapter *adapter = netdev_priv(netdev);
4737
4738         netif_device_detach(netdev);
4739
4740         if (state == pci_channel_io_perm_failure)
4741                 return PCI_ERS_RESULT_DISCONNECT;
4742
4743         if (netif_running(netdev))
4744                 e1000_down(adapter);
4745         pci_disable_device(pdev);
4746
4747         /* Request a slot slot reset. */
4748         return PCI_ERS_RESULT_NEED_RESET;
4749 }
4750
4751 /**
4752  * e1000_io_slot_reset - called after the pci bus has been reset.
4753  * @pdev: Pointer to PCI device
4754  *
4755  * Restart the card from scratch, as if from a cold-boot. Implementation
4756  * resembles the first-half of the e1000_resume routine.
4757  */
4758 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
4759 {
4760         struct net_device *netdev = pci_get_drvdata(pdev);
4761         struct e1000_adapter *adapter = netdev_priv(netdev);
4762         struct e1000_hw *hw = &adapter->hw;
4763         int err;
4764
4765         if (adapter->need_ioport)
4766                 err = pci_enable_device(pdev);
4767         else
4768                 err = pci_enable_device_mem(pdev);
4769         if (err) {
4770                 pr_err("Cannot re-enable PCI device after reset.\n");
4771                 return PCI_ERS_RESULT_DISCONNECT;
4772         }
4773         pci_set_master(pdev);
4774
4775         pci_enable_wake(pdev, PCI_D3hot, 0);
4776         pci_enable_wake(pdev, PCI_D3cold, 0);
4777
4778         e1000_reset(adapter);
4779         ew32(WUS, ~0);
4780
4781         return PCI_ERS_RESULT_RECOVERED;
4782 }
4783
4784 /**
4785  * e1000_io_resume - called when traffic can start flowing again.
4786  * @pdev: Pointer to PCI device
4787  *
4788  * This callback is called when the error recovery driver tells us that
4789  * its OK to resume normal operation. Implementation resembles the
4790  * second-half of the e1000_resume routine.
4791  */
4792 static void e1000_io_resume(struct pci_dev *pdev)
4793 {
4794         struct net_device *netdev = pci_get_drvdata(pdev);
4795         struct e1000_adapter *adapter = netdev_priv(netdev);
4796
4797         e1000_init_manageability(adapter);
4798
4799         if (netif_running(netdev)) {
4800                 if (e1000_up(adapter)) {
4801                         pr_info("can't bring device back up after reset\n");
4802                         return;
4803                 }
4804         }
4805
4806         netif_device_attach(netdev);
4807 }
4808
4809 /* e1000_main.c */