e1000: enhance frame fragment detection
[safe/jmp/linux-2.6] / drivers / net / e1000 / e1000_main.c
1 /*******************************************************************************
2
3   Intel PRO/1000 Linux driver
4   Copyright(c) 1999 - 2006 Intel Corporation.
5
6   This program is free software; you can redistribute it and/or modify it
7   under the terms and conditions of the GNU General Public License,
8   version 2, as published by the Free Software Foundation.
9
10   This program is distributed in the hope it will be useful, but WITHOUT
11   ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12   FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
13   more details.
14
15   You should have received a copy of the GNU General Public License along with
16   this program; if not, write to the Free Software Foundation, Inc.,
17   51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
18
19   The full GNU General Public License is included in this distribution in
20   the file called "COPYING".
21
22   Contact Information:
23   Linux NICS <linux.nics@intel.com>
24   e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
25   Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
26
27 *******************************************************************************/
28
29 #include "e1000.h"
30 #include <net/ip6_checksum.h>
31
32 char e1000_driver_name[] = "e1000";
33 static char e1000_driver_string[] = "Intel(R) PRO/1000 Network Driver";
34 #define DRV_VERSION "7.3.21-k5-NAPI"
35 const char e1000_driver_version[] = DRV_VERSION;
36 static const char e1000_copyright[] = "Copyright (c) 1999-2006 Intel Corporation.";
37
38 /* e1000_pci_tbl - PCI Device ID Table
39  *
40  * Last entry must be all 0s
41  *
42  * Macro expands to...
43  *   {PCI_DEVICE(PCI_VENDOR_ID_INTEL, device_id)}
44  */
45 static struct pci_device_id e1000_pci_tbl[] = {
46         INTEL_E1000_ETHERNET_DEVICE(0x1000),
47         INTEL_E1000_ETHERNET_DEVICE(0x1001),
48         INTEL_E1000_ETHERNET_DEVICE(0x1004),
49         INTEL_E1000_ETHERNET_DEVICE(0x1008),
50         INTEL_E1000_ETHERNET_DEVICE(0x1009),
51         INTEL_E1000_ETHERNET_DEVICE(0x100C),
52         INTEL_E1000_ETHERNET_DEVICE(0x100D),
53         INTEL_E1000_ETHERNET_DEVICE(0x100E),
54         INTEL_E1000_ETHERNET_DEVICE(0x100F),
55         INTEL_E1000_ETHERNET_DEVICE(0x1010),
56         INTEL_E1000_ETHERNET_DEVICE(0x1011),
57         INTEL_E1000_ETHERNET_DEVICE(0x1012),
58         INTEL_E1000_ETHERNET_DEVICE(0x1013),
59         INTEL_E1000_ETHERNET_DEVICE(0x1014),
60         INTEL_E1000_ETHERNET_DEVICE(0x1015),
61         INTEL_E1000_ETHERNET_DEVICE(0x1016),
62         INTEL_E1000_ETHERNET_DEVICE(0x1017),
63         INTEL_E1000_ETHERNET_DEVICE(0x1018),
64         INTEL_E1000_ETHERNET_DEVICE(0x1019),
65         INTEL_E1000_ETHERNET_DEVICE(0x101A),
66         INTEL_E1000_ETHERNET_DEVICE(0x101D),
67         INTEL_E1000_ETHERNET_DEVICE(0x101E),
68         INTEL_E1000_ETHERNET_DEVICE(0x1026),
69         INTEL_E1000_ETHERNET_DEVICE(0x1027),
70         INTEL_E1000_ETHERNET_DEVICE(0x1028),
71         INTEL_E1000_ETHERNET_DEVICE(0x1075),
72         INTEL_E1000_ETHERNET_DEVICE(0x1076),
73         INTEL_E1000_ETHERNET_DEVICE(0x1077),
74         INTEL_E1000_ETHERNET_DEVICE(0x1078),
75         INTEL_E1000_ETHERNET_DEVICE(0x1079),
76         INTEL_E1000_ETHERNET_DEVICE(0x107A),
77         INTEL_E1000_ETHERNET_DEVICE(0x107B),
78         INTEL_E1000_ETHERNET_DEVICE(0x107C),
79         INTEL_E1000_ETHERNET_DEVICE(0x108A),
80         INTEL_E1000_ETHERNET_DEVICE(0x1099),
81         INTEL_E1000_ETHERNET_DEVICE(0x10B5),
82         /* required last entry */
83         {0,}
84 };
85
86 MODULE_DEVICE_TABLE(pci, e1000_pci_tbl);
87
88 int e1000_up(struct e1000_adapter *adapter);
89 void e1000_down(struct e1000_adapter *adapter);
90 void e1000_reinit_locked(struct e1000_adapter *adapter);
91 void e1000_reset(struct e1000_adapter *adapter);
92 int e1000_set_spd_dplx(struct e1000_adapter *adapter, u16 spddplx);
93 int e1000_setup_all_tx_resources(struct e1000_adapter *adapter);
94 int e1000_setup_all_rx_resources(struct e1000_adapter *adapter);
95 void e1000_free_all_tx_resources(struct e1000_adapter *adapter);
96 void e1000_free_all_rx_resources(struct e1000_adapter *adapter);
97 static int e1000_setup_tx_resources(struct e1000_adapter *adapter,
98                              struct e1000_tx_ring *txdr);
99 static int e1000_setup_rx_resources(struct e1000_adapter *adapter,
100                              struct e1000_rx_ring *rxdr);
101 static void e1000_free_tx_resources(struct e1000_adapter *adapter,
102                              struct e1000_tx_ring *tx_ring);
103 static void e1000_free_rx_resources(struct e1000_adapter *adapter,
104                              struct e1000_rx_ring *rx_ring);
105 void e1000_update_stats(struct e1000_adapter *adapter);
106
107 static int e1000_init_module(void);
108 static void e1000_exit_module(void);
109 static int e1000_probe(struct pci_dev *pdev, const struct pci_device_id *ent);
110 static void __devexit e1000_remove(struct pci_dev *pdev);
111 static int e1000_alloc_queues(struct e1000_adapter *adapter);
112 static int e1000_sw_init(struct e1000_adapter *adapter);
113 static int e1000_open(struct net_device *netdev);
114 static int e1000_close(struct net_device *netdev);
115 static void e1000_configure_tx(struct e1000_adapter *adapter);
116 static void e1000_configure_rx(struct e1000_adapter *adapter);
117 static void e1000_setup_rctl(struct e1000_adapter *adapter);
118 static void e1000_clean_all_tx_rings(struct e1000_adapter *adapter);
119 static void e1000_clean_all_rx_rings(struct e1000_adapter *adapter);
120 static void e1000_clean_tx_ring(struct e1000_adapter *adapter,
121                                 struct e1000_tx_ring *tx_ring);
122 static void e1000_clean_rx_ring(struct e1000_adapter *adapter,
123                                 struct e1000_rx_ring *rx_ring);
124 static void e1000_set_rx_mode(struct net_device *netdev);
125 static void e1000_update_phy_info(unsigned long data);
126 static void e1000_watchdog(unsigned long data);
127 static void e1000_82547_tx_fifo_stall(unsigned long data);
128 static netdev_tx_t e1000_xmit_frame(struct sk_buff *skb,
129                                     struct net_device *netdev);
130 static struct net_device_stats * e1000_get_stats(struct net_device *netdev);
131 static int e1000_change_mtu(struct net_device *netdev, int new_mtu);
132 static int e1000_set_mac(struct net_device *netdev, void *p);
133 static irqreturn_t e1000_intr(int irq, void *data);
134 static bool e1000_clean_tx_irq(struct e1000_adapter *adapter,
135                                struct e1000_tx_ring *tx_ring);
136 static int e1000_clean(struct napi_struct *napi, int budget);
137 static bool e1000_clean_rx_irq(struct e1000_adapter *adapter,
138                                struct e1000_rx_ring *rx_ring,
139                                int *work_done, int work_to_do);
140 static bool e1000_clean_jumbo_rx_irq(struct e1000_adapter *adapter,
141                                      struct e1000_rx_ring *rx_ring,
142                                      int *work_done, int work_to_do);
143 static void e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
144                                    struct e1000_rx_ring *rx_ring,
145                                    int cleaned_count);
146 static void e1000_alloc_jumbo_rx_buffers(struct e1000_adapter *adapter,
147                                          struct e1000_rx_ring *rx_ring,
148                                          int cleaned_count);
149 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd);
150 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
151                            int cmd);
152 static void e1000_enter_82542_rst(struct e1000_adapter *adapter);
153 static void e1000_leave_82542_rst(struct e1000_adapter *adapter);
154 static void e1000_tx_timeout(struct net_device *dev);
155 static void e1000_reset_task(struct work_struct *work);
156 static void e1000_smartspeed(struct e1000_adapter *adapter);
157 static int e1000_82547_fifo_workaround(struct e1000_adapter *adapter,
158                                        struct sk_buff *skb);
159
160 static void e1000_vlan_rx_register(struct net_device *netdev, struct vlan_group *grp);
161 static void e1000_vlan_rx_add_vid(struct net_device *netdev, u16 vid);
162 static void e1000_vlan_rx_kill_vid(struct net_device *netdev, u16 vid);
163 static void e1000_restore_vlan(struct e1000_adapter *adapter);
164
165 #ifdef CONFIG_PM
166 static int e1000_suspend(struct pci_dev *pdev, pm_message_t state);
167 static int e1000_resume(struct pci_dev *pdev);
168 #endif
169 static void e1000_shutdown(struct pci_dev *pdev);
170
171 #ifdef CONFIG_NET_POLL_CONTROLLER
172 /* for netdump / net console */
173 static void e1000_netpoll (struct net_device *netdev);
174 #endif
175
176 #define COPYBREAK_DEFAULT 256
177 static unsigned int copybreak __read_mostly = COPYBREAK_DEFAULT;
178 module_param(copybreak, uint, 0644);
179 MODULE_PARM_DESC(copybreak,
180         "Maximum size of packet that is copied to a new buffer on receive");
181
182 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
183                      pci_channel_state_t state);
184 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev);
185 static void e1000_io_resume(struct pci_dev *pdev);
186
187 static struct pci_error_handlers e1000_err_handler = {
188         .error_detected = e1000_io_error_detected,
189         .slot_reset = e1000_io_slot_reset,
190         .resume = e1000_io_resume,
191 };
192
193 static struct pci_driver e1000_driver = {
194         .name     = e1000_driver_name,
195         .id_table = e1000_pci_tbl,
196         .probe    = e1000_probe,
197         .remove   = __devexit_p(e1000_remove),
198 #ifdef CONFIG_PM
199         /* Power Managment Hooks */
200         .suspend  = e1000_suspend,
201         .resume   = e1000_resume,
202 #endif
203         .shutdown = e1000_shutdown,
204         .err_handler = &e1000_err_handler
205 };
206
207 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
208 MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
209 MODULE_LICENSE("GPL");
210 MODULE_VERSION(DRV_VERSION);
211
212 static int debug = NETIF_MSG_DRV | NETIF_MSG_PROBE;
213 module_param(debug, int, 0);
214 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
215
216 /**
217  * e1000_init_module - Driver Registration Routine
218  *
219  * e1000_init_module is the first routine called when the driver is
220  * loaded. All it does is register with the PCI subsystem.
221  **/
222
223 static int __init e1000_init_module(void)
224 {
225         int ret;
226         printk(KERN_INFO "%s - version %s\n",
227                e1000_driver_string, e1000_driver_version);
228
229         printk(KERN_INFO "%s\n", e1000_copyright);
230
231         ret = pci_register_driver(&e1000_driver);
232         if (copybreak != COPYBREAK_DEFAULT) {
233                 if (copybreak == 0)
234                         printk(KERN_INFO "e1000: copybreak disabled\n");
235                 else
236                         printk(KERN_INFO "e1000: copybreak enabled for "
237                                "packets <= %u bytes\n", copybreak);
238         }
239         return ret;
240 }
241
242 module_init(e1000_init_module);
243
244 /**
245  * e1000_exit_module - Driver Exit Cleanup Routine
246  *
247  * e1000_exit_module is called just before the driver is removed
248  * from memory.
249  **/
250
251 static void __exit e1000_exit_module(void)
252 {
253         pci_unregister_driver(&e1000_driver);
254 }
255
256 module_exit(e1000_exit_module);
257
258 static int e1000_request_irq(struct e1000_adapter *adapter)
259 {
260         struct net_device *netdev = adapter->netdev;
261         irq_handler_t handler = e1000_intr;
262         int irq_flags = IRQF_SHARED;
263         int err;
264
265         err = request_irq(adapter->pdev->irq, handler, irq_flags, netdev->name,
266                           netdev);
267         if (err) {
268                 DPRINTK(PROBE, ERR,
269                         "Unable to allocate interrupt Error: %d\n", err);
270         }
271
272         return err;
273 }
274
275 static void e1000_free_irq(struct e1000_adapter *adapter)
276 {
277         struct net_device *netdev = adapter->netdev;
278
279         free_irq(adapter->pdev->irq, netdev);
280 }
281
282 /**
283  * e1000_irq_disable - Mask off interrupt generation on the NIC
284  * @adapter: board private structure
285  **/
286
287 static void e1000_irq_disable(struct e1000_adapter *adapter)
288 {
289         struct e1000_hw *hw = &adapter->hw;
290
291         ew32(IMC, ~0);
292         E1000_WRITE_FLUSH();
293         synchronize_irq(adapter->pdev->irq);
294 }
295
296 /**
297  * e1000_irq_enable - Enable default interrupt generation settings
298  * @adapter: board private structure
299  **/
300
301 static void e1000_irq_enable(struct e1000_adapter *adapter)
302 {
303         struct e1000_hw *hw = &adapter->hw;
304
305         ew32(IMS, IMS_ENABLE_MASK);
306         E1000_WRITE_FLUSH();
307 }
308
309 static void e1000_update_mng_vlan(struct e1000_adapter *adapter)
310 {
311         struct e1000_hw *hw = &adapter->hw;
312         struct net_device *netdev = adapter->netdev;
313         u16 vid = hw->mng_cookie.vlan_id;
314         u16 old_vid = adapter->mng_vlan_id;
315         if (adapter->vlgrp) {
316                 if (!vlan_group_get_device(adapter->vlgrp, vid)) {
317                         if (hw->mng_cookie.status &
318                                 E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT) {
319                                 e1000_vlan_rx_add_vid(netdev, vid);
320                                 adapter->mng_vlan_id = vid;
321                         } else
322                                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
323
324                         if ((old_vid != (u16)E1000_MNG_VLAN_NONE) &&
325                                         (vid != old_vid) &&
326                             !vlan_group_get_device(adapter->vlgrp, old_vid))
327                                 e1000_vlan_rx_kill_vid(netdev, old_vid);
328                 } else
329                         adapter->mng_vlan_id = vid;
330         }
331 }
332
333 static void e1000_init_manageability(struct e1000_adapter *adapter)
334 {
335         struct e1000_hw *hw = &adapter->hw;
336
337         if (adapter->en_mng_pt) {
338                 u32 manc = er32(MANC);
339
340                 /* disable hardware interception of ARP */
341                 manc &= ~(E1000_MANC_ARP_EN);
342
343                 ew32(MANC, manc);
344         }
345 }
346
347 static void e1000_release_manageability(struct e1000_adapter *adapter)
348 {
349         struct e1000_hw *hw = &adapter->hw;
350
351         if (adapter->en_mng_pt) {
352                 u32 manc = er32(MANC);
353
354                 /* re-enable hardware interception of ARP */
355                 manc |= E1000_MANC_ARP_EN;
356
357                 ew32(MANC, manc);
358         }
359 }
360
361 /**
362  * e1000_configure - configure the hardware for RX and TX
363  * @adapter = private board structure
364  **/
365 static void e1000_configure(struct e1000_adapter *adapter)
366 {
367         struct net_device *netdev = adapter->netdev;
368         int i;
369
370         e1000_set_rx_mode(netdev);
371
372         e1000_restore_vlan(adapter);
373         e1000_init_manageability(adapter);
374
375         e1000_configure_tx(adapter);
376         e1000_setup_rctl(adapter);
377         e1000_configure_rx(adapter);
378         /* call E1000_DESC_UNUSED which always leaves
379          * at least 1 descriptor unused to make sure
380          * next_to_use != next_to_clean */
381         for (i = 0; i < adapter->num_rx_queues; i++) {
382                 struct e1000_rx_ring *ring = &adapter->rx_ring[i];
383                 adapter->alloc_rx_buf(adapter, ring,
384                                       E1000_DESC_UNUSED(ring));
385         }
386
387         adapter->tx_queue_len = netdev->tx_queue_len;
388 }
389
390 int e1000_up(struct e1000_adapter *adapter)
391 {
392         struct e1000_hw *hw = &adapter->hw;
393
394         /* hardware has been reset, we need to reload some things */
395         e1000_configure(adapter);
396
397         clear_bit(__E1000_DOWN, &adapter->flags);
398
399         napi_enable(&adapter->napi);
400
401         e1000_irq_enable(adapter);
402
403         netif_wake_queue(adapter->netdev);
404
405         /* fire a link change interrupt to start the watchdog */
406         ew32(ICS, E1000_ICS_LSC);
407         return 0;
408 }
409
410 /**
411  * e1000_power_up_phy - restore link in case the phy was powered down
412  * @adapter: address of board private structure
413  *
414  * The phy may be powered down to save power and turn off link when the
415  * driver is unloaded and wake on lan is not enabled (among others)
416  * *** this routine MUST be followed by a call to e1000_reset ***
417  *
418  **/
419
420 void e1000_power_up_phy(struct e1000_adapter *adapter)
421 {
422         struct e1000_hw *hw = &adapter->hw;
423         u16 mii_reg = 0;
424
425         /* Just clear the power down bit to wake the phy back up */
426         if (hw->media_type == e1000_media_type_copper) {
427                 /* according to the manual, the phy will retain its
428                  * settings across a power-down/up cycle */
429                 e1000_read_phy_reg(hw, PHY_CTRL, &mii_reg);
430                 mii_reg &= ~MII_CR_POWER_DOWN;
431                 e1000_write_phy_reg(hw, PHY_CTRL, mii_reg);
432         }
433 }
434
435 static void e1000_power_down_phy(struct e1000_adapter *adapter)
436 {
437         struct e1000_hw *hw = &adapter->hw;
438
439         /* Power down the PHY so no link is implied when interface is down *
440          * The PHY cannot be powered down if any of the following is true *
441          * (a) WoL is enabled
442          * (b) AMT is active
443          * (c) SoL/IDER session is active */
444         if (!adapter->wol && hw->mac_type >= e1000_82540 &&
445            hw->media_type == e1000_media_type_copper) {
446                 u16 mii_reg = 0;
447
448                 switch (hw->mac_type) {
449                 case e1000_82540:
450                 case e1000_82545:
451                 case e1000_82545_rev_3:
452                 case e1000_82546:
453                 case e1000_82546_rev_3:
454                 case e1000_82541:
455                 case e1000_82541_rev_2:
456                 case e1000_82547:
457                 case e1000_82547_rev_2:
458                         if (er32(MANC) & E1000_MANC_SMBUS_EN)
459                                 goto out;
460                         break;
461                 default:
462                         goto out;
463                 }
464                 e1000_read_phy_reg(hw, PHY_CTRL, &mii_reg);
465                 mii_reg |= MII_CR_POWER_DOWN;
466                 e1000_write_phy_reg(hw, PHY_CTRL, mii_reg);
467                 mdelay(1);
468         }
469 out:
470         return;
471 }
472
473 void e1000_down(struct e1000_adapter *adapter)
474 {
475         struct e1000_hw *hw = &adapter->hw;
476         struct net_device *netdev = adapter->netdev;
477         u32 rctl, tctl;
478
479         /* signal that we're down so the interrupt handler does not
480          * reschedule our watchdog timer */
481         set_bit(__E1000_DOWN, &adapter->flags);
482
483         /* disable receives in the hardware */
484         rctl = er32(RCTL);
485         ew32(RCTL, rctl & ~E1000_RCTL_EN);
486         /* flush and sleep below */
487
488         netif_tx_disable(netdev);
489
490         /* disable transmits in the hardware */
491         tctl = er32(TCTL);
492         tctl &= ~E1000_TCTL_EN;
493         ew32(TCTL, tctl);
494         /* flush both disables and wait for them to finish */
495         E1000_WRITE_FLUSH();
496         msleep(10);
497
498         napi_disable(&adapter->napi);
499
500         e1000_irq_disable(adapter);
501
502         del_timer_sync(&adapter->tx_fifo_stall_timer);
503         del_timer_sync(&adapter->watchdog_timer);
504         del_timer_sync(&adapter->phy_info_timer);
505
506         netdev->tx_queue_len = adapter->tx_queue_len;
507         adapter->link_speed = 0;
508         adapter->link_duplex = 0;
509         netif_carrier_off(netdev);
510
511         e1000_reset(adapter);
512         e1000_clean_all_tx_rings(adapter);
513         e1000_clean_all_rx_rings(adapter);
514 }
515
516 void e1000_reinit_locked(struct e1000_adapter *adapter)
517 {
518         WARN_ON(in_interrupt());
519         while (test_and_set_bit(__E1000_RESETTING, &adapter->flags))
520                 msleep(1);
521         e1000_down(adapter);
522         e1000_up(adapter);
523         clear_bit(__E1000_RESETTING, &adapter->flags);
524 }
525
526 void e1000_reset(struct e1000_adapter *adapter)
527 {
528         struct e1000_hw *hw = &adapter->hw;
529         u32 pba = 0, tx_space, min_tx_space, min_rx_space;
530         bool legacy_pba_adjust = false;
531         u16 hwm;
532
533         /* Repartition Pba for greater than 9k mtu
534          * To take effect CTRL.RST is required.
535          */
536
537         switch (hw->mac_type) {
538         case e1000_82542_rev2_0:
539         case e1000_82542_rev2_1:
540         case e1000_82543:
541         case e1000_82544:
542         case e1000_82540:
543         case e1000_82541:
544         case e1000_82541_rev_2:
545                 legacy_pba_adjust = true;
546                 pba = E1000_PBA_48K;
547                 break;
548         case e1000_82545:
549         case e1000_82545_rev_3:
550         case e1000_82546:
551         case e1000_82546_rev_3:
552                 pba = E1000_PBA_48K;
553                 break;
554         case e1000_82547:
555         case e1000_82547_rev_2:
556                 legacy_pba_adjust = true;
557                 pba = E1000_PBA_30K;
558                 break;
559         case e1000_undefined:
560         case e1000_num_macs:
561                 break;
562         }
563
564         if (legacy_pba_adjust) {
565                 if (hw->max_frame_size > E1000_RXBUFFER_8192)
566                         pba -= 8; /* allocate more FIFO for Tx */
567
568                 if (hw->mac_type == e1000_82547) {
569                         adapter->tx_fifo_head = 0;
570                         adapter->tx_head_addr = pba << E1000_TX_HEAD_ADDR_SHIFT;
571                         adapter->tx_fifo_size =
572                                 (E1000_PBA_40K - pba) << E1000_PBA_BYTES_SHIFT;
573                         atomic_set(&adapter->tx_fifo_stall, 0);
574                 }
575         } else if (hw->max_frame_size >  ETH_FRAME_LEN + ETH_FCS_LEN) {
576                 /* adjust PBA for jumbo frames */
577                 ew32(PBA, pba);
578
579                 /* To maintain wire speed transmits, the Tx FIFO should be
580                  * large enough to accommodate two full transmit packets,
581                  * rounded up to the next 1KB and expressed in KB.  Likewise,
582                  * the Rx FIFO should be large enough to accommodate at least
583                  * one full receive packet and is similarly rounded up and
584                  * expressed in KB. */
585                 pba = er32(PBA);
586                 /* upper 16 bits has Tx packet buffer allocation size in KB */
587                 tx_space = pba >> 16;
588                 /* lower 16 bits has Rx packet buffer allocation size in KB */
589                 pba &= 0xffff;
590                 /*
591                  * the tx fifo also stores 16 bytes of information about the tx
592                  * but don't include ethernet FCS because hardware appends it
593                  */
594                 min_tx_space = (hw->max_frame_size +
595                                 sizeof(struct e1000_tx_desc) -
596                                 ETH_FCS_LEN) * 2;
597                 min_tx_space = ALIGN(min_tx_space, 1024);
598                 min_tx_space >>= 10;
599                 /* software strips receive CRC, so leave room for it */
600                 min_rx_space = hw->max_frame_size;
601                 min_rx_space = ALIGN(min_rx_space, 1024);
602                 min_rx_space >>= 10;
603
604                 /* If current Tx allocation is less than the min Tx FIFO size,
605                  * and the min Tx FIFO size is less than the current Rx FIFO
606                  * allocation, take space away from current Rx allocation */
607                 if (tx_space < min_tx_space &&
608                     ((min_tx_space - tx_space) < pba)) {
609                         pba = pba - (min_tx_space - tx_space);
610
611                         /* PCI/PCIx hardware has PBA alignment constraints */
612                         switch (hw->mac_type) {
613                         case e1000_82545 ... e1000_82546_rev_3:
614                                 pba &= ~(E1000_PBA_8K - 1);
615                                 break;
616                         default:
617                                 break;
618                         }
619
620                         /* if short on rx space, rx wins and must trump tx
621                          * adjustment or use Early Receive if available */
622                         if (pba < min_rx_space)
623                                 pba = min_rx_space;
624                 }
625         }
626
627         ew32(PBA, pba);
628
629         /*
630          * flow control settings:
631          * The high water mark must be low enough to fit one full frame
632          * (or the size used for early receive) above it in the Rx FIFO.
633          * Set it to the lower of:
634          * - 90% of the Rx FIFO size, and
635          * - the full Rx FIFO size minus the early receive size (for parts
636          *   with ERT support assuming ERT set to E1000_ERT_2048), or
637          * - the full Rx FIFO size minus one full frame
638          */
639         hwm = min(((pba << 10) * 9 / 10),
640                   ((pba << 10) - hw->max_frame_size));
641
642         hw->fc_high_water = hwm & 0xFFF8;       /* 8-byte granularity */
643         hw->fc_low_water = hw->fc_high_water - 8;
644         hw->fc_pause_time = E1000_FC_PAUSE_TIME;
645         hw->fc_send_xon = 1;
646         hw->fc = hw->original_fc;
647
648         /* Allow time for pending master requests to run */
649         e1000_reset_hw(hw);
650         if (hw->mac_type >= e1000_82544)
651                 ew32(WUC, 0);
652
653         if (e1000_init_hw(hw))
654                 DPRINTK(PROBE, ERR, "Hardware Error\n");
655         e1000_update_mng_vlan(adapter);
656
657         /* if (adapter->hwflags & HWFLAGS_PHY_PWR_BIT) { */
658         if (hw->mac_type >= e1000_82544 &&
659             hw->autoneg == 1 &&
660             hw->autoneg_advertised == ADVERTISE_1000_FULL) {
661                 u32 ctrl = er32(CTRL);
662                 /* clear phy power management bit if we are in gig only mode,
663                  * which if enabled will attempt negotiation to 100Mb, which
664                  * can cause a loss of link at power off or driver unload */
665                 ctrl &= ~E1000_CTRL_SWDPIN3;
666                 ew32(CTRL, ctrl);
667         }
668
669         /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
670         ew32(VET, ETHERNET_IEEE_VLAN_TYPE);
671
672         e1000_reset_adaptive(hw);
673         e1000_phy_get_info(hw, &adapter->phy_info);
674
675         e1000_release_manageability(adapter);
676 }
677
678 /**
679  *  Dump the eeprom for users having checksum issues
680  **/
681 static void e1000_dump_eeprom(struct e1000_adapter *adapter)
682 {
683         struct net_device *netdev = adapter->netdev;
684         struct ethtool_eeprom eeprom;
685         const struct ethtool_ops *ops = netdev->ethtool_ops;
686         u8 *data;
687         int i;
688         u16 csum_old, csum_new = 0;
689
690         eeprom.len = ops->get_eeprom_len(netdev);
691         eeprom.offset = 0;
692
693         data = kmalloc(eeprom.len, GFP_KERNEL);
694         if (!data) {
695                 printk(KERN_ERR "Unable to allocate memory to dump EEPROM"
696                        " data\n");
697                 return;
698         }
699
700         ops->get_eeprom(netdev, &eeprom, data);
701
702         csum_old = (data[EEPROM_CHECKSUM_REG * 2]) +
703                    (data[EEPROM_CHECKSUM_REG * 2 + 1] << 8);
704         for (i = 0; i < EEPROM_CHECKSUM_REG * 2; i += 2)
705                 csum_new += data[i] + (data[i + 1] << 8);
706         csum_new = EEPROM_SUM - csum_new;
707
708         printk(KERN_ERR "/*********************/\n");
709         printk(KERN_ERR "Current EEPROM Checksum : 0x%04x\n", csum_old);
710         printk(KERN_ERR "Calculated              : 0x%04x\n", csum_new);
711
712         printk(KERN_ERR "Offset    Values\n");
713         printk(KERN_ERR "========  ======\n");
714         print_hex_dump(KERN_ERR, "", DUMP_PREFIX_OFFSET, 16, 1, data, 128, 0);
715
716         printk(KERN_ERR "Include this output when contacting your support "
717                "provider.\n");
718         printk(KERN_ERR "This is not a software error! Something bad "
719                "happened to your hardware or\n");
720         printk(KERN_ERR "EEPROM image. Ignoring this "
721                "problem could result in further problems,\n");
722         printk(KERN_ERR "possibly loss of data, corruption or system hangs!\n");
723         printk(KERN_ERR "The MAC Address will be reset to 00:00:00:00:00:00, "
724                "which is invalid\n");
725         printk(KERN_ERR "and requires you to set the proper MAC "
726                "address manually before continuing\n");
727         printk(KERN_ERR "to enable this network device.\n");
728         printk(KERN_ERR "Please inspect the EEPROM dump and report the issue "
729                "to your hardware vendor\n");
730         printk(KERN_ERR "or Intel Customer Support.\n");
731         printk(KERN_ERR "/*********************/\n");
732
733         kfree(data);
734 }
735
736 /**
737  * e1000_is_need_ioport - determine if an adapter needs ioport resources or not
738  * @pdev: PCI device information struct
739  *
740  * Return true if an adapter needs ioport resources
741  **/
742 static int e1000_is_need_ioport(struct pci_dev *pdev)
743 {
744         switch (pdev->device) {
745         case E1000_DEV_ID_82540EM:
746         case E1000_DEV_ID_82540EM_LOM:
747         case E1000_DEV_ID_82540EP:
748         case E1000_DEV_ID_82540EP_LOM:
749         case E1000_DEV_ID_82540EP_LP:
750         case E1000_DEV_ID_82541EI:
751         case E1000_DEV_ID_82541EI_MOBILE:
752         case E1000_DEV_ID_82541ER:
753         case E1000_DEV_ID_82541ER_LOM:
754         case E1000_DEV_ID_82541GI:
755         case E1000_DEV_ID_82541GI_LF:
756         case E1000_DEV_ID_82541GI_MOBILE:
757         case E1000_DEV_ID_82544EI_COPPER:
758         case E1000_DEV_ID_82544EI_FIBER:
759         case E1000_DEV_ID_82544GC_COPPER:
760         case E1000_DEV_ID_82544GC_LOM:
761         case E1000_DEV_ID_82545EM_COPPER:
762         case E1000_DEV_ID_82545EM_FIBER:
763         case E1000_DEV_ID_82546EB_COPPER:
764         case E1000_DEV_ID_82546EB_FIBER:
765         case E1000_DEV_ID_82546EB_QUAD_COPPER:
766                 return true;
767         default:
768                 return false;
769         }
770 }
771
772 static const struct net_device_ops e1000_netdev_ops = {
773         .ndo_open               = e1000_open,
774         .ndo_stop               = e1000_close,
775         .ndo_start_xmit         = e1000_xmit_frame,
776         .ndo_get_stats          = e1000_get_stats,
777         .ndo_set_rx_mode        = e1000_set_rx_mode,
778         .ndo_set_mac_address    = e1000_set_mac,
779         .ndo_tx_timeout         = e1000_tx_timeout,
780         .ndo_change_mtu         = e1000_change_mtu,
781         .ndo_do_ioctl           = e1000_ioctl,
782         .ndo_validate_addr      = eth_validate_addr,
783
784         .ndo_vlan_rx_register   = e1000_vlan_rx_register,
785         .ndo_vlan_rx_add_vid    = e1000_vlan_rx_add_vid,
786         .ndo_vlan_rx_kill_vid   = e1000_vlan_rx_kill_vid,
787 #ifdef CONFIG_NET_POLL_CONTROLLER
788         .ndo_poll_controller    = e1000_netpoll,
789 #endif
790 };
791
792 /**
793  * e1000_probe - Device Initialization Routine
794  * @pdev: PCI device information struct
795  * @ent: entry in e1000_pci_tbl
796  *
797  * Returns 0 on success, negative on failure
798  *
799  * e1000_probe initializes an adapter identified by a pci_dev structure.
800  * The OS initialization, configuring of the adapter private structure,
801  * and a hardware reset occur.
802  **/
803 static int __devinit e1000_probe(struct pci_dev *pdev,
804                                  const struct pci_device_id *ent)
805 {
806         struct net_device *netdev;
807         struct e1000_adapter *adapter;
808         struct e1000_hw *hw;
809
810         static int cards_found = 0;
811         static int global_quad_port_a = 0; /* global ksp3 port a indication */
812         int i, err, pci_using_dac;
813         u16 eeprom_data = 0;
814         u16 eeprom_apme_mask = E1000_EEPROM_APME;
815         int bars, need_ioport;
816
817         /* do not allocate ioport bars when not needed */
818         need_ioport = e1000_is_need_ioport(pdev);
819         if (need_ioport) {
820                 bars = pci_select_bars(pdev, IORESOURCE_MEM | IORESOURCE_IO);
821                 err = pci_enable_device(pdev);
822         } else {
823                 bars = pci_select_bars(pdev, IORESOURCE_MEM);
824                 err = pci_enable_device_mem(pdev);
825         }
826         if (err)
827                 return err;
828
829         if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(64)) &&
830             !pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64))) {
831                 pci_using_dac = 1;
832         } else {
833                 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
834                 if (err) {
835                         err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
836                         if (err) {
837                                 E1000_ERR("No usable DMA configuration, "
838                                           "aborting\n");
839                                 goto err_dma;
840                         }
841                 }
842                 pci_using_dac = 0;
843         }
844
845         err = pci_request_selected_regions(pdev, bars, e1000_driver_name);
846         if (err)
847                 goto err_pci_reg;
848
849         pci_set_master(pdev);
850
851         err = -ENOMEM;
852         netdev = alloc_etherdev(sizeof(struct e1000_adapter));
853         if (!netdev)
854                 goto err_alloc_etherdev;
855
856         SET_NETDEV_DEV(netdev, &pdev->dev);
857
858         pci_set_drvdata(pdev, netdev);
859         adapter = netdev_priv(netdev);
860         adapter->netdev = netdev;
861         adapter->pdev = pdev;
862         adapter->msg_enable = (1 << debug) - 1;
863         adapter->bars = bars;
864         adapter->need_ioport = need_ioport;
865
866         hw = &adapter->hw;
867         hw->back = adapter;
868
869         err = -EIO;
870         hw->hw_addr = pci_ioremap_bar(pdev, BAR_0);
871         if (!hw->hw_addr)
872                 goto err_ioremap;
873
874         if (adapter->need_ioport) {
875                 for (i = BAR_1; i <= BAR_5; i++) {
876                         if (pci_resource_len(pdev, i) == 0)
877                                 continue;
878                         if (pci_resource_flags(pdev, i) & IORESOURCE_IO) {
879                                 hw->io_base = pci_resource_start(pdev, i);
880                                 break;
881                         }
882                 }
883         }
884
885         netdev->netdev_ops = &e1000_netdev_ops;
886         e1000_set_ethtool_ops(netdev);
887         netdev->watchdog_timeo = 5 * HZ;
888         netif_napi_add(netdev, &adapter->napi, e1000_clean, 64);
889
890         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
891
892         adapter->bd_number = cards_found;
893
894         /* setup the private structure */
895
896         err = e1000_sw_init(adapter);
897         if (err)
898                 goto err_sw_init;
899
900         err = -EIO;
901
902         if (hw->mac_type >= e1000_82543) {
903                 netdev->features = NETIF_F_SG |
904                                    NETIF_F_HW_CSUM |
905                                    NETIF_F_HW_VLAN_TX |
906                                    NETIF_F_HW_VLAN_RX |
907                                    NETIF_F_HW_VLAN_FILTER;
908         }
909
910         if ((hw->mac_type >= e1000_82544) &&
911            (hw->mac_type != e1000_82547))
912                 netdev->features |= NETIF_F_TSO;
913
914         if (pci_using_dac)
915                 netdev->features |= NETIF_F_HIGHDMA;
916
917         netdev->vlan_features |= NETIF_F_TSO;
918         netdev->vlan_features |= NETIF_F_HW_CSUM;
919         netdev->vlan_features |= NETIF_F_SG;
920
921         adapter->en_mng_pt = e1000_enable_mng_pass_thru(hw);
922
923         /* initialize eeprom parameters */
924         if (e1000_init_eeprom_params(hw)) {
925                 E1000_ERR("EEPROM initialization failed\n");
926                 goto err_eeprom;
927         }
928
929         /* before reading the EEPROM, reset the controller to
930          * put the device in a known good starting state */
931
932         e1000_reset_hw(hw);
933
934         /* make sure the EEPROM is good */
935         if (e1000_validate_eeprom_checksum(hw) < 0) {
936                 DPRINTK(PROBE, ERR, "The EEPROM Checksum Is Not Valid\n");
937                 e1000_dump_eeprom(adapter);
938                 /*
939                  * set MAC address to all zeroes to invalidate and temporary
940                  * disable this device for the user. This blocks regular
941                  * traffic while still permitting ethtool ioctls from reaching
942                  * the hardware as well as allowing the user to run the
943                  * interface after manually setting a hw addr using
944                  * `ip set address`
945                  */
946                 memset(hw->mac_addr, 0, netdev->addr_len);
947         } else {
948                 /* copy the MAC address out of the EEPROM */
949                 if (e1000_read_mac_addr(hw))
950                         DPRINTK(PROBE, ERR, "EEPROM Read Error\n");
951         }
952         /* don't block initalization here due to bad MAC address */
953         memcpy(netdev->dev_addr, hw->mac_addr, netdev->addr_len);
954         memcpy(netdev->perm_addr, hw->mac_addr, netdev->addr_len);
955
956         if (!is_valid_ether_addr(netdev->perm_addr))
957                 DPRINTK(PROBE, ERR, "Invalid MAC Address\n");
958
959         e1000_get_bus_info(hw);
960
961         init_timer(&adapter->tx_fifo_stall_timer);
962         adapter->tx_fifo_stall_timer.function = &e1000_82547_tx_fifo_stall;
963         adapter->tx_fifo_stall_timer.data = (unsigned long)adapter;
964
965         init_timer(&adapter->watchdog_timer);
966         adapter->watchdog_timer.function = &e1000_watchdog;
967         adapter->watchdog_timer.data = (unsigned long) adapter;
968
969         init_timer(&adapter->phy_info_timer);
970         adapter->phy_info_timer.function = &e1000_update_phy_info;
971         adapter->phy_info_timer.data = (unsigned long)adapter;
972
973         INIT_WORK(&adapter->reset_task, e1000_reset_task);
974
975         e1000_check_options(adapter);
976
977         /* Initial Wake on LAN setting
978          * If APM wake is enabled in the EEPROM,
979          * enable the ACPI Magic Packet filter
980          */
981
982         switch (hw->mac_type) {
983         case e1000_82542_rev2_0:
984         case e1000_82542_rev2_1:
985         case e1000_82543:
986                 break;
987         case e1000_82544:
988                 e1000_read_eeprom(hw,
989                         EEPROM_INIT_CONTROL2_REG, 1, &eeprom_data);
990                 eeprom_apme_mask = E1000_EEPROM_82544_APM;
991                 break;
992         case e1000_82546:
993         case e1000_82546_rev_3:
994                 if (er32(STATUS) & E1000_STATUS_FUNC_1){
995                         e1000_read_eeprom(hw,
996                                 EEPROM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
997                         break;
998                 }
999                 /* Fall Through */
1000         default:
1001                 e1000_read_eeprom(hw,
1002                         EEPROM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
1003                 break;
1004         }
1005         if (eeprom_data & eeprom_apme_mask)
1006                 adapter->eeprom_wol |= E1000_WUFC_MAG;
1007
1008         /* now that we have the eeprom settings, apply the special cases
1009          * where the eeprom may be wrong or the board simply won't support
1010          * wake on lan on a particular port */
1011         switch (pdev->device) {
1012         case E1000_DEV_ID_82546GB_PCIE:
1013                 adapter->eeprom_wol = 0;
1014                 break;
1015         case E1000_DEV_ID_82546EB_FIBER:
1016         case E1000_DEV_ID_82546GB_FIBER:
1017                 /* Wake events only supported on port A for dual fiber
1018                  * regardless of eeprom setting */
1019                 if (er32(STATUS) & E1000_STATUS_FUNC_1)
1020                         adapter->eeprom_wol = 0;
1021                 break;
1022         case E1000_DEV_ID_82546GB_QUAD_COPPER_KSP3:
1023                 /* if quad port adapter, disable WoL on all but port A */
1024                 if (global_quad_port_a != 0)
1025                         adapter->eeprom_wol = 0;
1026                 else
1027                         adapter->quad_port_a = 1;
1028                 /* Reset for multiple quad port adapters */
1029                 if (++global_quad_port_a == 4)
1030                         global_quad_port_a = 0;
1031                 break;
1032         }
1033
1034         /* initialize the wol settings based on the eeprom settings */
1035         adapter->wol = adapter->eeprom_wol;
1036         device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
1037
1038         /* print bus type/speed/width info */
1039         DPRINTK(PROBE, INFO, "(PCI%s:%s:%s) ",
1040                 ((hw->bus_type == e1000_bus_type_pcix) ? "-X" : ""),
1041                 ((hw->bus_speed == e1000_bus_speed_133) ? "133MHz" :
1042                  (hw->bus_speed == e1000_bus_speed_120) ? "120MHz" :
1043                  (hw->bus_speed == e1000_bus_speed_100) ? "100MHz" :
1044                  (hw->bus_speed == e1000_bus_speed_66) ? "66MHz" : "33MHz"),
1045                 ((hw->bus_width == e1000_bus_width_64) ? "64-bit" : "32-bit"));
1046
1047         printk("%pM\n", netdev->dev_addr);
1048
1049         /* reset the hardware with the new settings */
1050         e1000_reset(adapter);
1051
1052         strcpy(netdev->name, "eth%d");
1053         err = register_netdev(netdev);
1054         if (err)
1055                 goto err_register;
1056
1057         /* carrier off reporting is important to ethtool even BEFORE open */
1058         netif_carrier_off(netdev);
1059
1060         DPRINTK(PROBE, INFO, "Intel(R) PRO/1000 Network Connection\n");
1061
1062         cards_found++;
1063         return 0;
1064
1065 err_register:
1066 err_eeprom:
1067         e1000_phy_hw_reset(hw);
1068
1069         if (hw->flash_address)
1070                 iounmap(hw->flash_address);
1071         kfree(adapter->tx_ring);
1072         kfree(adapter->rx_ring);
1073 err_sw_init:
1074         iounmap(hw->hw_addr);
1075 err_ioremap:
1076         free_netdev(netdev);
1077 err_alloc_etherdev:
1078         pci_release_selected_regions(pdev, bars);
1079 err_pci_reg:
1080 err_dma:
1081         pci_disable_device(pdev);
1082         return err;
1083 }
1084
1085 /**
1086  * e1000_remove - Device Removal Routine
1087  * @pdev: PCI device information struct
1088  *
1089  * e1000_remove is called by the PCI subsystem to alert the driver
1090  * that it should release a PCI device.  The could be caused by a
1091  * Hot-Plug event, or because the driver is going to be removed from
1092  * memory.
1093  **/
1094
1095 static void __devexit e1000_remove(struct pci_dev *pdev)
1096 {
1097         struct net_device *netdev = pci_get_drvdata(pdev);
1098         struct e1000_adapter *adapter = netdev_priv(netdev);
1099         struct e1000_hw *hw = &adapter->hw;
1100
1101         set_bit(__E1000_DOWN, &adapter->flags);
1102         del_timer_sync(&adapter->tx_fifo_stall_timer);
1103         del_timer_sync(&adapter->watchdog_timer);
1104         del_timer_sync(&adapter->phy_info_timer);
1105
1106         cancel_work_sync(&adapter->reset_task);
1107
1108         e1000_release_manageability(adapter);
1109
1110         unregister_netdev(netdev);
1111
1112         e1000_phy_hw_reset(hw);
1113
1114         kfree(adapter->tx_ring);
1115         kfree(adapter->rx_ring);
1116
1117         iounmap(hw->hw_addr);
1118         if (hw->flash_address)
1119                 iounmap(hw->flash_address);
1120         pci_release_selected_regions(pdev, adapter->bars);
1121
1122         free_netdev(netdev);
1123
1124         pci_disable_device(pdev);
1125 }
1126
1127 /**
1128  * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
1129  * @adapter: board private structure to initialize
1130  *
1131  * e1000_sw_init initializes the Adapter private data structure.
1132  * Fields are initialized based on PCI device information and
1133  * OS network device settings (MTU size).
1134  **/
1135
1136 static int __devinit e1000_sw_init(struct e1000_adapter *adapter)
1137 {
1138         struct e1000_hw *hw = &adapter->hw;
1139         struct net_device *netdev = adapter->netdev;
1140         struct pci_dev *pdev = adapter->pdev;
1141
1142         /* PCI config space info */
1143
1144         hw->vendor_id = pdev->vendor;
1145         hw->device_id = pdev->device;
1146         hw->subsystem_vendor_id = pdev->subsystem_vendor;
1147         hw->subsystem_id = pdev->subsystem_device;
1148         hw->revision_id = pdev->revision;
1149
1150         pci_read_config_word(pdev, PCI_COMMAND, &hw->pci_cmd_word);
1151
1152         adapter->rx_buffer_len = MAXIMUM_ETHERNET_VLAN_SIZE;
1153         hw->max_frame_size = netdev->mtu +
1154                              ENET_HEADER_SIZE + ETHERNET_FCS_SIZE;
1155         hw->min_frame_size = MINIMUM_ETHERNET_FRAME_SIZE;
1156
1157         /* identify the MAC */
1158
1159         if (e1000_set_mac_type(hw)) {
1160                 DPRINTK(PROBE, ERR, "Unknown MAC Type\n");
1161                 return -EIO;
1162         }
1163
1164         switch (hw->mac_type) {
1165         default:
1166                 break;
1167         case e1000_82541:
1168         case e1000_82547:
1169         case e1000_82541_rev_2:
1170         case e1000_82547_rev_2:
1171                 hw->phy_init_script = 1;
1172                 break;
1173         }
1174
1175         e1000_set_media_type(hw);
1176
1177         hw->wait_autoneg_complete = false;
1178         hw->tbi_compatibility_en = true;
1179         hw->adaptive_ifs = true;
1180
1181         /* Copper options */
1182
1183         if (hw->media_type == e1000_media_type_copper) {
1184                 hw->mdix = AUTO_ALL_MODES;
1185                 hw->disable_polarity_correction = false;
1186                 hw->master_slave = E1000_MASTER_SLAVE;
1187         }
1188
1189         adapter->num_tx_queues = 1;
1190         adapter->num_rx_queues = 1;
1191
1192         if (e1000_alloc_queues(adapter)) {
1193                 DPRINTK(PROBE, ERR, "Unable to allocate memory for queues\n");
1194                 return -ENOMEM;
1195         }
1196
1197         /* Explicitly disable IRQ since the NIC can be in any state. */
1198         e1000_irq_disable(adapter);
1199
1200         spin_lock_init(&adapter->stats_lock);
1201
1202         set_bit(__E1000_DOWN, &adapter->flags);
1203
1204         return 0;
1205 }
1206
1207 /**
1208  * e1000_alloc_queues - Allocate memory for all rings
1209  * @adapter: board private structure to initialize
1210  *
1211  * We allocate one ring per queue at run-time since we don't know the
1212  * number of queues at compile-time.
1213  **/
1214
1215 static int __devinit e1000_alloc_queues(struct e1000_adapter *adapter)
1216 {
1217         adapter->tx_ring = kcalloc(adapter->num_tx_queues,
1218                                    sizeof(struct e1000_tx_ring), GFP_KERNEL);
1219         if (!adapter->tx_ring)
1220                 return -ENOMEM;
1221
1222         adapter->rx_ring = kcalloc(adapter->num_rx_queues,
1223                                    sizeof(struct e1000_rx_ring), GFP_KERNEL);
1224         if (!adapter->rx_ring) {
1225                 kfree(adapter->tx_ring);
1226                 return -ENOMEM;
1227         }
1228
1229         return E1000_SUCCESS;
1230 }
1231
1232 /**
1233  * e1000_open - Called when a network interface is made active
1234  * @netdev: network interface device structure
1235  *
1236  * Returns 0 on success, negative value on failure
1237  *
1238  * The open entry point is called when a network interface is made
1239  * active by the system (IFF_UP).  At this point all resources needed
1240  * for transmit and receive operations are allocated, the interrupt
1241  * handler is registered with the OS, the watchdog timer is started,
1242  * and the stack is notified that the interface is ready.
1243  **/
1244
1245 static int e1000_open(struct net_device *netdev)
1246 {
1247         struct e1000_adapter *adapter = netdev_priv(netdev);
1248         struct e1000_hw *hw = &adapter->hw;
1249         int err;
1250
1251         /* disallow open during test */
1252         if (test_bit(__E1000_TESTING, &adapter->flags))
1253                 return -EBUSY;
1254
1255         netif_carrier_off(netdev);
1256
1257         /* allocate transmit descriptors */
1258         err = e1000_setup_all_tx_resources(adapter);
1259         if (err)
1260                 goto err_setup_tx;
1261
1262         /* allocate receive descriptors */
1263         err = e1000_setup_all_rx_resources(adapter);
1264         if (err)
1265                 goto err_setup_rx;
1266
1267         e1000_power_up_phy(adapter);
1268
1269         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
1270         if ((hw->mng_cookie.status &
1271                           E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT)) {
1272                 e1000_update_mng_vlan(adapter);
1273         }
1274
1275         /* before we allocate an interrupt, we must be ready to handle it.
1276          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
1277          * as soon as we call pci_request_irq, so we have to setup our
1278          * clean_rx handler before we do so.  */
1279         e1000_configure(adapter);
1280
1281         err = e1000_request_irq(adapter);
1282         if (err)
1283                 goto err_req_irq;
1284
1285         /* From here on the code is the same as e1000_up() */
1286         clear_bit(__E1000_DOWN, &adapter->flags);
1287
1288         napi_enable(&adapter->napi);
1289
1290         e1000_irq_enable(adapter);
1291
1292         netif_start_queue(netdev);
1293
1294         /* fire a link status change interrupt to start the watchdog */
1295         ew32(ICS, E1000_ICS_LSC);
1296
1297         return E1000_SUCCESS;
1298
1299 err_req_irq:
1300         e1000_power_down_phy(adapter);
1301         e1000_free_all_rx_resources(adapter);
1302 err_setup_rx:
1303         e1000_free_all_tx_resources(adapter);
1304 err_setup_tx:
1305         e1000_reset(adapter);
1306
1307         return err;
1308 }
1309
1310 /**
1311  * e1000_close - Disables a network interface
1312  * @netdev: network interface device structure
1313  *
1314  * Returns 0, this is not allowed to fail
1315  *
1316  * The close entry point is called when an interface is de-activated
1317  * by the OS.  The hardware is still under the drivers control, but
1318  * needs to be disabled.  A global MAC reset is issued to stop the
1319  * hardware, and all transmit and receive resources are freed.
1320  **/
1321
1322 static int e1000_close(struct net_device *netdev)
1323 {
1324         struct e1000_adapter *adapter = netdev_priv(netdev);
1325         struct e1000_hw *hw = &adapter->hw;
1326
1327         WARN_ON(test_bit(__E1000_RESETTING, &adapter->flags));
1328         e1000_down(adapter);
1329         e1000_power_down_phy(adapter);
1330         e1000_free_irq(adapter);
1331
1332         e1000_free_all_tx_resources(adapter);
1333         e1000_free_all_rx_resources(adapter);
1334
1335         /* kill manageability vlan ID if supported, but not if a vlan with
1336          * the same ID is registered on the host OS (let 8021q kill it) */
1337         if ((hw->mng_cookie.status &
1338                           E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT) &&
1339              !(adapter->vlgrp &&
1340                vlan_group_get_device(adapter->vlgrp, adapter->mng_vlan_id))) {
1341                 e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
1342         }
1343
1344         return 0;
1345 }
1346
1347 /**
1348  * e1000_check_64k_bound - check that memory doesn't cross 64kB boundary
1349  * @adapter: address of board private structure
1350  * @start: address of beginning of memory
1351  * @len: length of memory
1352  **/
1353 static bool e1000_check_64k_bound(struct e1000_adapter *adapter, void *start,
1354                                   unsigned long len)
1355 {
1356         struct e1000_hw *hw = &adapter->hw;
1357         unsigned long begin = (unsigned long)start;
1358         unsigned long end = begin + len;
1359
1360         /* First rev 82545 and 82546 need to not allow any memory
1361          * write location to cross 64k boundary due to errata 23 */
1362         if (hw->mac_type == e1000_82545 ||
1363             hw->mac_type == e1000_82546) {
1364                 return ((begin ^ (end - 1)) >> 16) != 0 ? false : true;
1365         }
1366
1367         return true;
1368 }
1369
1370 /**
1371  * e1000_setup_tx_resources - allocate Tx resources (Descriptors)
1372  * @adapter: board private structure
1373  * @txdr:    tx descriptor ring (for a specific queue) to setup
1374  *
1375  * Return 0 on success, negative on failure
1376  **/
1377
1378 static int e1000_setup_tx_resources(struct e1000_adapter *adapter,
1379                                     struct e1000_tx_ring *txdr)
1380 {
1381         struct pci_dev *pdev = adapter->pdev;
1382         int size;
1383
1384         size = sizeof(struct e1000_buffer) * txdr->count;
1385         txdr->buffer_info = vmalloc(size);
1386         if (!txdr->buffer_info) {
1387                 DPRINTK(PROBE, ERR,
1388                 "Unable to allocate memory for the transmit descriptor ring\n");
1389                 return -ENOMEM;
1390         }
1391         memset(txdr->buffer_info, 0, size);
1392
1393         /* round up to nearest 4K */
1394
1395         txdr->size = txdr->count * sizeof(struct e1000_tx_desc);
1396         txdr->size = ALIGN(txdr->size, 4096);
1397
1398         txdr->desc = pci_alloc_consistent(pdev, txdr->size, &txdr->dma);
1399         if (!txdr->desc) {
1400 setup_tx_desc_die:
1401                 vfree(txdr->buffer_info);
1402                 DPRINTK(PROBE, ERR,
1403                 "Unable to allocate memory for the transmit descriptor ring\n");
1404                 return -ENOMEM;
1405         }
1406
1407         /* Fix for errata 23, can't cross 64kB boundary */
1408         if (!e1000_check_64k_bound(adapter, txdr->desc, txdr->size)) {
1409                 void *olddesc = txdr->desc;
1410                 dma_addr_t olddma = txdr->dma;
1411                 DPRINTK(TX_ERR, ERR, "txdr align check failed: %u bytes "
1412                                      "at %p\n", txdr->size, txdr->desc);
1413                 /* Try again, without freeing the previous */
1414                 txdr->desc = pci_alloc_consistent(pdev, txdr->size, &txdr->dma);
1415                 /* Failed allocation, critical failure */
1416                 if (!txdr->desc) {
1417                         pci_free_consistent(pdev, txdr->size, olddesc, olddma);
1418                         goto setup_tx_desc_die;
1419                 }
1420
1421                 if (!e1000_check_64k_bound(adapter, txdr->desc, txdr->size)) {
1422                         /* give up */
1423                         pci_free_consistent(pdev, txdr->size, txdr->desc,
1424                                             txdr->dma);
1425                         pci_free_consistent(pdev, txdr->size, olddesc, olddma);
1426                         DPRINTK(PROBE, ERR,
1427                                 "Unable to allocate aligned memory "
1428                                 "for the transmit descriptor ring\n");
1429                         vfree(txdr->buffer_info);
1430                         return -ENOMEM;
1431                 } else {
1432                         /* Free old allocation, new allocation was successful */
1433                         pci_free_consistent(pdev, txdr->size, olddesc, olddma);
1434                 }
1435         }
1436         memset(txdr->desc, 0, txdr->size);
1437
1438         txdr->next_to_use = 0;
1439         txdr->next_to_clean = 0;
1440
1441         return 0;
1442 }
1443
1444 /**
1445  * e1000_setup_all_tx_resources - wrapper to allocate Tx resources
1446  *                                (Descriptors) for all queues
1447  * @adapter: board private structure
1448  *
1449  * Return 0 on success, negative on failure
1450  **/
1451
1452 int e1000_setup_all_tx_resources(struct e1000_adapter *adapter)
1453 {
1454         int i, err = 0;
1455
1456         for (i = 0; i < adapter->num_tx_queues; i++) {
1457                 err = e1000_setup_tx_resources(adapter, &adapter->tx_ring[i]);
1458                 if (err) {
1459                         DPRINTK(PROBE, ERR,
1460                                 "Allocation for Tx Queue %u failed\n", i);
1461                         for (i-- ; i >= 0; i--)
1462                                 e1000_free_tx_resources(adapter,
1463                                                         &adapter->tx_ring[i]);
1464                         break;
1465                 }
1466         }
1467
1468         return err;
1469 }
1470
1471 /**
1472  * e1000_configure_tx - Configure 8254x Transmit Unit after Reset
1473  * @adapter: board private structure
1474  *
1475  * Configure the Tx unit of the MAC after a reset.
1476  **/
1477
1478 static void e1000_configure_tx(struct e1000_adapter *adapter)
1479 {
1480         u64 tdba;
1481         struct e1000_hw *hw = &adapter->hw;
1482         u32 tdlen, tctl, tipg;
1483         u32 ipgr1, ipgr2;
1484
1485         /* Setup the HW Tx Head and Tail descriptor pointers */
1486
1487         switch (adapter->num_tx_queues) {
1488         case 1:
1489         default:
1490                 tdba = adapter->tx_ring[0].dma;
1491                 tdlen = adapter->tx_ring[0].count *
1492                         sizeof(struct e1000_tx_desc);
1493                 ew32(TDLEN, tdlen);
1494                 ew32(TDBAH, (tdba >> 32));
1495                 ew32(TDBAL, (tdba & 0x00000000ffffffffULL));
1496                 ew32(TDT, 0);
1497                 ew32(TDH, 0);
1498                 adapter->tx_ring[0].tdh = ((hw->mac_type >= e1000_82543) ? E1000_TDH : E1000_82542_TDH);
1499                 adapter->tx_ring[0].tdt = ((hw->mac_type >= e1000_82543) ? E1000_TDT : E1000_82542_TDT);
1500                 break;
1501         }
1502
1503         /* Set the default values for the Tx Inter Packet Gap timer */
1504         if ((hw->media_type == e1000_media_type_fiber ||
1505              hw->media_type == e1000_media_type_internal_serdes))
1506                 tipg = DEFAULT_82543_TIPG_IPGT_FIBER;
1507         else
1508                 tipg = DEFAULT_82543_TIPG_IPGT_COPPER;
1509
1510         switch (hw->mac_type) {
1511         case e1000_82542_rev2_0:
1512         case e1000_82542_rev2_1:
1513                 tipg = DEFAULT_82542_TIPG_IPGT;
1514                 ipgr1 = DEFAULT_82542_TIPG_IPGR1;
1515                 ipgr2 = DEFAULT_82542_TIPG_IPGR2;
1516                 break;
1517         default:
1518                 ipgr1 = DEFAULT_82543_TIPG_IPGR1;
1519                 ipgr2 = DEFAULT_82543_TIPG_IPGR2;
1520                 break;
1521         }
1522         tipg |= ipgr1 << E1000_TIPG_IPGR1_SHIFT;
1523         tipg |= ipgr2 << E1000_TIPG_IPGR2_SHIFT;
1524         ew32(TIPG, tipg);
1525
1526         /* Set the Tx Interrupt Delay register */
1527
1528         ew32(TIDV, adapter->tx_int_delay);
1529         if (hw->mac_type >= e1000_82540)
1530                 ew32(TADV, adapter->tx_abs_int_delay);
1531
1532         /* Program the Transmit Control Register */
1533
1534         tctl = er32(TCTL);
1535         tctl &= ~E1000_TCTL_CT;
1536         tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
1537                 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
1538
1539         e1000_config_collision_dist(hw);
1540
1541         /* Setup Transmit Descriptor Settings for eop descriptor */
1542         adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
1543
1544         /* only set IDE if we are delaying interrupts using the timers */
1545         if (adapter->tx_int_delay)
1546                 adapter->txd_cmd |= E1000_TXD_CMD_IDE;
1547
1548         if (hw->mac_type < e1000_82543)
1549                 adapter->txd_cmd |= E1000_TXD_CMD_RPS;
1550         else
1551                 adapter->txd_cmd |= E1000_TXD_CMD_RS;
1552
1553         /* Cache if we're 82544 running in PCI-X because we'll
1554          * need this to apply a workaround later in the send path. */
1555         if (hw->mac_type == e1000_82544 &&
1556             hw->bus_type == e1000_bus_type_pcix)
1557                 adapter->pcix_82544 = 1;
1558
1559         ew32(TCTL, tctl);
1560
1561 }
1562
1563 /**
1564  * e1000_setup_rx_resources - allocate Rx resources (Descriptors)
1565  * @adapter: board private structure
1566  * @rxdr:    rx descriptor ring (for a specific queue) to setup
1567  *
1568  * Returns 0 on success, negative on failure
1569  **/
1570
1571 static int e1000_setup_rx_resources(struct e1000_adapter *adapter,
1572                                     struct e1000_rx_ring *rxdr)
1573 {
1574         struct pci_dev *pdev = adapter->pdev;
1575         int size, desc_len;
1576
1577         size = sizeof(struct e1000_buffer) * rxdr->count;
1578         rxdr->buffer_info = vmalloc(size);
1579         if (!rxdr->buffer_info) {
1580                 DPRINTK(PROBE, ERR,
1581                 "Unable to allocate memory for the receive descriptor ring\n");
1582                 return -ENOMEM;
1583         }
1584         memset(rxdr->buffer_info, 0, size);
1585
1586         desc_len = sizeof(struct e1000_rx_desc);
1587
1588         /* Round up to nearest 4K */
1589
1590         rxdr->size = rxdr->count * desc_len;
1591         rxdr->size = ALIGN(rxdr->size, 4096);
1592
1593         rxdr->desc = pci_alloc_consistent(pdev, rxdr->size, &rxdr->dma);
1594
1595         if (!rxdr->desc) {
1596                 DPRINTK(PROBE, ERR,
1597                 "Unable to allocate memory for the receive descriptor ring\n");
1598 setup_rx_desc_die:
1599                 vfree(rxdr->buffer_info);
1600                 return -ENOMEM;
1601         }
1602
1603         /* Fix for errata 23, can't cross 64kB boundary */
1604         if (!e1000_check_64k_bound(adapter, rxdr->desc, rxdr->size)) {
1605                 void *olddesc = rxdr->desc;
1606                 dma_addr_t olddma = rxdr->dma;
1607                 DPRINTK(RX_ERR, ERR, "rxdr align check failed: %u bytes "
1608                                      "at %p\n", rxdr->size, rxdr->desc);
1609                 /* Try again, without freeing the previous */
1610                 rxdr->desc = pci_alloc_consistent(pdev, rxdr->size, &rxdr->dma);
1611                 /* Failed allocation, critical failure */
1612                 if (!rxdr->desc) {
1613                         pci_free_consistent(pdev, rxdr->size, olddesc, olddma);
1614                         DPRINTK(PROBE, ERR,
1615                                 "Unable to allocate memory "
1616                                 "for the receive descriptor ring\n");
1617                         goto setup_rx_desc_die;
1618                 }
1619
1620                 if (!e1000_check_64k_bound(adapter, rxdr->desc, rxdr->size)) {
1621                         /* give up */
1622                         pci_free_consistent(pdev, rxdr->size, rxdr->desc,
1623                                             rxdr->dma);
1624                         pci_free_consistent(pdev, rxdr->size, olddesc, olddma);
1625                         DPRINTK(PROBE, ERR,
1626                                 "Unable to allocate aligned memory "
1627                                 "for the receive descriptor ring\n");
1628                         goto setup_rx_desc_die;
1629                 } else {
1630                         /* Free old allocation, new allocation was successful */
1631                         pci_free_consistent(pdev, rxdr->size, olddesc, olddma);
1632                 }
1633         }
1634         memset(rxdr->desc, 0, rxdr->size);
1635
1636         rxdr->next_to_clean = 0;
1637         rxdr->next_to_use = 0;
1638         rxdr->rx_skb_top = NULL;
1639
1640         return 0;
1641 }
1642
1643 /**
1644  * e1000_setup_all_rx_resources - wrapper to allocate Rx resources
1645  *                                (Descriptors) for all queues
1646  * @adapter: board private structure
1647  *
1648  * Return 0 on success, negative on failure
1649  **/
1650
1651 int e1000_setup_all_rx_resources(struct e1000_adapter *adapter)
1652 {
1653         int i, err = 0;
1654
1655         for (i = 0; i < adapter->num_rx_queues; i++) {
1656                 err = e1000_setup_rx_resources(adapter, &adapter->rx_ring[i]);
1657                 if (err) {
1658                         DPRINTK(PROBE, ERR,
1659                                 "Allocation for Rx Queue %u failed\n", i);
1660                         for (i-- ; i >= 0; i--)
1661                                 e1000_free_rx_resources(adapter,
1662                                                         &adapter->rx_ring[i]);
1663                         break;
1664                 }
1665         }
1666
1667         return err;
1668 }
1669
1670 /**
1671  * e1000_setup_rctl - configure the receive control registers
1672  * @adapter: Board private structure
1673  **/
1674 static void e1000_setup_rctl(struct e1000_adapter *adapter)
1675 {
1676         struct e1000_hw *hw = &adapter->hw;
1677         u32 rctl;
1678
1679         rctl = er32(RCTL);
1680
1681         rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
1682
1683         rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
1684                 E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
1685                 (hw->mc_filter_type << E1000_RCTL_MO_SHIFT);
1686
1687         if (hw->tbi_compatibility_on == 1)
1688                 rctl |= E1000_RCTL_SBP;
1689         else
1690                 rctl &= ~E1000_RCTL_SBP;
1691
1692         if (adapter->netdev->mtu <= ETH_DATA_LEN)
1693                 rctl &= ~E1000_RCTL_LPE;
1694         else
1695                 rctl |= E1000_RCTL_LPE;
1696
1697         /* Setup buffer sizes */
1698         rctl &= ~E1000_RCTL_SZ_4096;
1699         rctl |= E1000_RCTL_BSEX;
1700         switch (adapter->rx_buffer_len) {
1701                 case E1000_RXBUFFER_256:
1702                         rctl |= E1000_RCTL_SZ_256;
1703                         rctl &= ~E1000_RCTL_BSEX;
1704                         break;
1705                 case E1000_RXBUFFER_512:
1706                         rctl |= E1000_RCTL_SZ_512;
1707                         rctl &= ~E1000_RCTL_BSEX;
1708                         break;
1709                 case E1000_RXBUFFER_1024:
1710                         rctl |= E1000_RCTL_SZ_1024;
1711                         rctl &= ~E1000_RCTL_BSEX;
1712                         break;
1713                 case E1000_RXBUFFER_2048:
1714                 default:
1715                         rctl |= E1000_RCTL_SZ_2048;
1716                         rctl &= ~E1000_RCTL_BSEX;
1717                         break;
1718                 case E1000_RXBUFFER_4096:
1719                         rctl |= E1000_RCTL_SZ_4096;
1720                         break;
1721                 case E1000_RXBUFFER_8192:
1722                         rctl |= E1000_RCTL_SZ_8192;
1723                         break;
1724                 case E1000_RXBUFFER_16384:
1725                         rctl |= E1000_RCTL_SZ_16384;
1726                         break;
1727         }
1728
1729         ew32(RCTL, rctl);
1730 }
1731
1732 /**
1733  * e1000_configure_rx - Configure 8254x Receive Unit after Reset
1734  * @adapter: board private structure
1735  *
1736  * Configure the Rx unit of the MAC after a reset.
1737  **/
1738
1739 static void e1000_configure_rx(struct e1000_adapter *adapter)
1740 {
1741         u64 rdba;
1742         struct e1000_hw *hw = &adapter->hw;
1743         u32 rdlen, rctl, rxcsum;
1744
1745         if (adapter->netdev->mtu > ETH_DATA_LEN) {
1746                 rdlen = adapter->rx_ring[0].count *
1747                         sizeof(struct e1000_rx_desc);
1748                 adapter->clean_rx = e1000_clean_jumbo_rx_irq;
1749                 adapter->alloc_rx_buf = e1000_alloc_jumbo_rx_buffers;
1750         } else {
1751                 rdlen = adapter->rx_ring[0].count *
1752                         sizeof(struct e1000_rx_desc);
1753                 adapter->clean_rx = e1000_clean_rx_irq;
1754                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
1755         }
1756
1757         /* disable receives while setting up the descriptors */
1758         rctl = er32(RCTL);
1759         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1760
1761         /* set the Receive Delay Timer Register */
1762         ew32(RDTR, adapter->rx_int_delay);
1763
1764         if (hw->mac_type >= e1000_82540) {
1765                 ew32(RADV, adapter->rx_abs_int_delay);
1766                 if (adapter->itr_setting != 0)
1767                         ew32(ITR, 1000000000 / (adapter->itr * 256));
1768         }
1769
1770         /* Setup the HW Rx Head and Tail Descriptor Pointers and
1771          * the Base and Length of the Rx Descriptor Ring */
1772         switch (adapter->num_rx_queues) {
1773         case 1:
1774         default:
1775                 rdba = adapter->rx_ring[0].dma;
1776                 ew32(RDLEN, rdlen);
1777                 ew32(RDBAH, (rdba >> 32));
1778                 ew32(RDBAL, (rdba & 0x00000000ffffffffULL));
1779                 ew32(RDT, 0);
1780                 ew32(RDH, 0);
1781                 adapter->rx_ring[0].rdh = ((hw->mac_type >= e1000_82543) ? E1000_RDH : E1000_82542_RDH);
1782                 adapter->rx_ring[0].rdt = ((hw->mac_type >= e1000_82543) ? E1000_RDT : E1000_82542_RDT);
1783                 break;
1784         }
1785
1786         /* Enable 82543 Receive Checksum Offload for TCP and UDP */
1787         if (hw->mac_type >= e1000_82543) {
1788                 rxcsum = er32(RXCSUM);
1789                 if (adapter->rx_csum)
1790                         rxcsum |= E1000_RXCSUM_TUOFL;
1791                 else
1792                         /* don't need to clear IPPCSE as it defaults to 0 */
1793                         rxcsum &= ~E1000_RXCSUM_TUOFL;
1794                 ew32(RXCSUM, rxcsum);
1795         }
1796
1797         /* Enable Receives */
1798         ew32(RCTL, rctl);
1799 }
1800
1801 /**
1802  * e1000_free_tx_resources - Free Tx Resources per Queue
1803  * @adapter: board private structure
1804  * @tx_ring: Tx descriptor ring for a specific queue
1805  *
1806  * Free all transmit software resources
1807  **/
1808
1809 static void e1000_free_tx_resources(struct e1000_adapter *adapter,
1810                                     struct e1000_tx_ring *tx_ring)
1811 {
1812         struct pci_dev *pdev = adapter->pdev;
1813
1814         e1000_clean_tx_ring(adapter, tx_ring);
1815
1816         vfree(tx_ring->buffer_info);
1817         tx_ring->buffer_info = NULL;
1818
1819         pci_free_consistent(pdev, tx_ring->size, tx_ring->desc, tx_ring->dma);
1820
1821         tx_ring->desc = NULL;
1822 }
1823
1824 /**
1825  * e1000_free_all_tx_resources - Free Tx Resources for All Queues
1826  * @adapter: board private structure
1827  *
1828  * Free all transmit software resources
1829  **/
1830
1831 void e1000_free_all_tx_resources(struct e1000_adapter *adapter)
1832 {
1833         int i;
1834
1835         for (i = 0; i < adapter->num_tx_queues; i++)
1836                 e1000_free_tx_resources(adapter, &adapter->tx_ring[i]);
1837 }
1838
1839 static void e1000_unmap_and_free_tx_resource(struct e1000_adapter *adapter,
1840                                              struct e1000_buffer *buffer_info)
1841 {
1842         if (buffer_info->dma) {
1843                 if (buffer_info->mapped_as_page)
1844                         pci_unmap_page(adapter->pdev, buffer_info->dma,
1845                                        buffer_info->length, PCI_DMA_TODEVICE);
1846                 else
1847                         pci_unmap_single(adapter->pdev, buffer_info->dma,
1848                                          buffer_info->length,
1849                                          PCI_DMA_TODEVICE);
1850                 buffer_info->dma = 0;
1851         }
1852         if (buffer_info->skb) {
1853                 dev_kfree_skb_any(buffer_info->skb);
1854                 buffer_info->skb = NULL;
1855         }
1856         buffer_info->time_stamp = 0;
1857         /* buffer_info must be completely set up in the transmit path */
1858 }
1859
1860 /**
1861  * e1000_clean_tx_ring - Free Tx Buffers
1862  * @adapter: board private structure
1863  * @tx_ring: ring to be cleaned
1864  **/
1865
1866 static void e1000_clean_tx_ring(struct e1000_adapter *adapter,
1867                                 struct e1000_tx_ring *tx_ring)
1868 {
1869         struct e1000_hw *hw = &adapter->hw;
1870         struct e1000_buffer *buffer_info;
1871         unsigned long size;
1872         unsigned int i;
1873
1874         /* Free all the Tx ring sk_buffs */
1875
1876         for (i = 0; i < tx_ring->count; i++) {
1877                 buffer_info = &tx_ring->buffer_info[i];
1878                 e1000_unmap_and_free_tx_resource(adapter, buffer_info);
1879         }
1880
1881         size = sizeof(struct e1000_buffer) * tx_ring->count;
1882         memset(tx_ring->buffer_info, 0, size);
1883
1884         /* Zero out the descriptor ring */
1885
1886         memset(tx_ring->desc, 0, tx_ring->size);
1887
1888         tx_ring->next_to_use = 0;
1889         tx_ring->next_to_clean = 0;
1890         tx_ring->last_tx_tso = 0;
1891
1892         writel(0, hw->hw_addr + tx_ring->tdh);
1893         writel(0, hw->hw_addr + tx_ring->tdt);
1894 }
1895
1896 /**
1897  * e1000_clean_all_tx_rings - Free Tx Buffers for all queues
1898  * @adapter: board private structure
1899  **/
1900
1901 static void e1000_clean_all_tx_rings(struct e1000_adapter *adapter)
1902 {
1903         int i;
1904
1905         for (i = 0; i < adapter->num_tx_queues; i++)
1906                 e1000_clean_tx_ring(adapter, &adapter->tx_ring[i]);
1907 }
1908
1909 /**
1910  * e1000_free_rx_resources - Free Rx Resources
1911  * @adapter: board private structure
1912  * @rx_ring: ring to clean the resources from
1913  *
1914  * Free all receive software resources
1915  **/
1916
1917 static void e1000_free_rx_resources(struct e1000_adapter *adapter,
1918                                     struct e1000_rx_ring *rx_ring)
1919 {
1920         struct pci_dev *pdev = adapter->pdev;
1921
1922         e1000_clean_rx_ring(adapter, rx_ring);
1923
1924         vfree(rx_ring->buffer_info);
1925         rx_ring->buffer_info = NULL;
1926
1927         pci_free_consistent(pdev, rx_ring->size, rx_ring->desc, rx_ring->dma);
1928
1929         rx_ring->desc = NULL;
1930 }
1931
1932 /**
1933  * e1000_free_all_rx_resources - Free Rx Resources for All Queues
1934  * @adapter: board private structure
1935  *
1936  * Free all receive software resources
1937  **/
1938
1939 void e1000_free_all_rx_resources(struct e1000_adapter *adapter)
1940 {
1941         int i;
1942
1943         for (i = 0; i < adapter->num_rx_queues; i++)
1944                 e1000_free_rx_resources(adapter, &adapter->rx_ring[i]);
1945 }
1946
1947 /**
1948  * e1000_clean_rx_ring - Free Rx Buffers per Queue
1949  * @adapter: board private structure
1950  * @rx_ring: ring to free buffers from
1951  **/
1952
1953 static void e1000_clean_rx_ring(struct e1000_adapter *adapter,
1954                                 struct e1000_rx_ring *rx_ring)
1955 {
1956         struct e1000_hw *hw = &adapter->hw;
1957         struct e1000_buffer *buffer_info;
1958         struct pci_dev *pdev = adapter->pdev;
1959         unsigned long size;
1960         unsigned int i;
1961
1962         /* Free all the Rx ring sk_buffs */
1963         for (i = 0; i < rx_ring->count; i++) {
1964                 buffer_info = &rx_ring->buffer_info[i];
1965                 if (buffer_info->dma &&
1966                     adapter->clean_rx == e1000_clean_rx_irq) {
1967                         pci_unmap_single(pdev, buffer_info->dma,
1968                                          buffer_info->length,
1969                                          PCI_DMA_FROMDEVICE);
1970                 } else if (buffer_info->dma &&
1971                            adapter->clean_rx == e1000_clean_jumbo_rx_irq) {
1972                         pci_unmap_page(pdev, buffer_info->dma,
1973                                        buffer_info->length,
1974                                        PCI_DMA_FROMDEVICE);
1975                 }
1976
1977                 buffer_info->dma = 0;
1978                 if (buffer_info->page) {
1979                         put_page(buffer_info->page);
1980                         buffer_info->page = NULL;
1981                 }
1982                 if (buffer_info->skb) {
1983                         dev_kfree_skb(buffer_info->skb);
1984                         buffer_info->skb = NULL;
1985                 }
1986         }
1987
1988         /* there also may be some cached data from a chained receive */
1989         if (rx_ring->rx_skb_top) {
1990                 dev_kfree_skb(rx_ring->rx_skb_top);
1991                 rx_ring->rx_skb_top = NULL;
1992         }
1993
1994         size = sizeof(struct e1000_buffer) * rx_ring->count;
1995         memset(rx_ring->buffer_info, 0, size);
1996
1997         /* Zero out the descriptor ring */
1998         memset(rx_ring->desc, 0, rx_ring->size);
1999
2000         rx_ring->next_to_clean = 0;
2001         rx_ring->next_to_use = 0;
2002
2003         writel(0, hw->hw_addr + rx_ring->rdh);
2004         writel(0, hw->hw_addr + rx_ring->rdt);
2005 }
2006
2007 /**
2008  * e1000_clean_all_rx_rings - Free Rx Buffers for all queues
2009  * @adapter: board private structure
2010  **/
2011
2012 static void e1000_clean_all_rx_rings(struct e1000_adapter *adapter)
2013 {
2014         int i;
2015
2016         for (i = 0; i < adapter->num_rx_queues; i++)
2017                 e1000_clean_rx_ring(adapter, &adapter->rx_ring[i]);
2018 }
2019
2020 /* The 82542 2.0 (revision 2) needs to have the receive unit in reset
2021  * and memory write and invalidate disabled for certain operations
2022  */
2023 static void e1000_enter_82542_rst(struct e1000_adapter *adapter)
2024 {
2025         struct e1000_hw *hw = &adapter->hw;
2026         struct net_device *netdev = adapter->netdev;
2027         u32 rctl;
2028
2029         e1000_pci_clear_mwi(hw);
2030
2031         rctl = er32(RCTL);
2032         rctl |= E1000_RCTL_RST;
2033         ew32(RCTL, rctl);
2034         E1000_WRITE_FLUSH();
2035         mdelay(5);
2036
2037         if (netif_running(netdev))
2038                 e1000_clean_all_rx_rings(adapter);
2039 }
2040
2041 static void e1000_leave_82542_rst(struct e1000_adapter *adapter)
2042 {
2043         struct e1000_hw *hw = &adapter->hw;
2044         struct net_device *netdev = adapter->netdev;
2045         u32 rctl;
2046
2047         rctl = er32(RCTL);
2048         rctl &= ~E1000_RCTL_RST;
2049         ew32(RCTL, rctl);
2050         E1000_WRITE_FLUSH();
2051         mdelay(5);
2052
2053         if (hw->pci_cmd_word & PCI_COMMAND_INVALIDATE)
2054                 e1000_pci_set_mwi(hw);
2055
2056         if (netif_running(netdev)) {
2057                 /* No need to loop, because 82542 supports only 1 queue */
2058                 struct e1000_rx_ring *ring = &adapter->rx_ring[0];
2059                 e1000_configure_rx(adapter);
2060                 adapter->alloc_rx_buf(adapter, ring, E1000_DESC_UNUSED(ring));
2061         }
2062 }
2063
2064 /**
2065  * e1000_set_mac - Change the Ethernet Address of the NIC
2066  * @netdev: network interface device structure
2067  * @p: pointer to an address structure
2068  *
2069  * Returns 0 on success, negative on failure
2070  **/
2071
2072 static int e1000_set_mac(struct net_device *netdev, void *p)
2073 {
2074         struct e1000_adapter *adapter = netdev_priv(netdev);
2075         struct e1000_hw *hw = &adapter->hw;
2076         struct sockaddr *addr = p;
2077
2078         if (!is_valid_ether_addr(addr->sa_data))
2079                 return -EADDRNOTAVAIL;
2080
2081         /* 82542 2.0 needs to be in reset to write receive address registers */
2082
2083         if (hw->mac_type == e1000_82542_rev2_0)
2084                 e1000_enter_82542_rst(adapter);
2085
2086         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
2087         memcpy(hw->mac_addr, addr->sa_data, netdev->addr_len);
2088
2089         e1000_rar_set(hw, hw->mac_addr, 0);
2090
2091         if (hw->mac_type == e1000_82542_rev2_0)
2092                 e1000_leave_82542_rst(adapter);
2093
2094         return 0;
2095 }
2096
2097 /**
2098  * e1000_set_rx_mode - Secondary Unicast, Multicast and Promiscuous mode set
2099  * @netdev: network interface device structure
2100  *
2101  * The set_rx_mode entry point is called whenever the unicast or multicast
2102  * address lists or the network interface flags are updated. This routine is
2103  * responsible for configuring the hardware for proper unicast, multicast,
2104  * promiscuous mode, and all-multi behavior.
2105  **/
2106
2107 static void e1000_set_rx_mode(struct net_device *netdev)
2108 {
2109         struct e1000_adapter *adapter = netdev_priv(netdev);
2110         struct e1000_hw *hw = &adapter->hw;
2111         struct netdev_hw_addr *ha;
2112         bool use_uc = false;
2113         struct dev_addr_list *mc_ptr;
2114         u32 rctl;
2115         u32 hash_value;
2116         int i, rar_entries = E1000_RAR_ENTRIES;
2117         int mta_reg_count = E1000_NUM_MTA_REGISTERS;
2118         u32 *mcarray = kcalloc(mta_reg_count, sizeof(u32), GFP_ATOMIC);
2119
2120         if (!mcarray) {
2121                 DPRINTK(PROBE, ERR, "memory allocation failed\n");
2122                 return;
2123         }
2124
2125         /* Check for Promiscuous and All Multicast modes */
2126
2127         rctl = er32(RCTL);
2128
2129         if (netdev->flags & IFF_PROMISC) {
2130                 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
2131                 rctl &= ~E1000_RCTL_VFE;
2132         } else {
2133                 if (netdev->flags & IFF_ALLMULTI)
2134                         rctl |= E1000_RCTL_MPE;
2135                 else
2136                         rctl &= ~E1000_RCTL_MPE;
2137                 /* Enable VLAN filter if there is a VLAN */
2138                 if (adapter->vlgrp)
2139                         rctl |= E1000_RCTL_VFE;
2140         }
2141
2142         if (netdev->uc.count > rar_entries - 1) {
2143                 rctl |= E1000_RCTL_UPE;
2144         } else if (!(netdev->flags & IFF_PROMISC)) {
2145                 rctl &= ~E1000_RCTL_UPE;
2146                 use_uc = true;
2147         }
2148
2149         ew32(RCTL, rctl);
2150
2151         /* 82542 2.0 needs to be in reset to write receive address registers */
2152
2153         if (hw->mac_type == e1000_82542_rev2_0)
2154                 e1000_enter_82542_rst(adapter);
2155
2156         /* load the first 14 addresses into the exact filters 1-14. Unicast
2157          * addresses take precedence to avoid disabling unicast filtering
2158          * when possible.
2159          *
2160          * RAR 0 is used for the station MAC adddress
2161          * if there are not 14 addresses, go ahead and clear the filters
2162          */
2163         i = 1;
2164         if (use_uc)
2165                 list_for_each_entry(ha, &netdev->uc.list, list) {
2166                         if (i == rar_entries)
2167                                 break;
2168                         e1000_rar_set(hw, ha->addr, i++);
2169                 }
2170
2171         WARN_ON(i == rar_entries);
2172
2173         mc_ptr = netdev->mc_list;
2174
2175         for (; i < rar_entries; i++) {
2176                 if (mc_ptr) {
2177                         e1000_rar_set(hw, mc_ptr->da_addr, i);
2178                         mc_ptr = mc_ptr->next;
2179                 } else {
2180                         E1000_WRITE_REG_ARRAY(hw, RA, i << 1, 0);
2181                         E1000_WRITE_FLUSH();
2182                         E1000_WRITE_REG_ARRAY(hw, RA, (i << 1) + 1, 0);
2183                         E1000_WRITE_FLUSH();
2184                 }
2185         }
2186
2187         /* load any remaining addresses into the hash table */
2188
2189         for (; mc_ptr; mc_ptr = mc_ptr->next) {
2190                 u32 hash_reg, hash_bit, mta;
2191                 hash_value = e1000_hash_mc_addr(hw, mc_ptr->da_addr);
2192                 hash_reg = (hash_value >> 5) & 0x7F;
2193                 hash_bit = hash_value & 0x1F;
2194                 mta = (1 << hash_bit);
2195                 mcarray[hash_reg] |= mta;
2196         }
2197
2198         /* write the hash table completely, write from bottom to avoid
2199          * both stupid write combining chipsets, and flushing each write */
2200         for (i = mta_reg_count - 1; i >= 0 ; i--) {
2201                 /*
2202                  * If we are on an 82544 has an errata where writing odd
2203                  * offsets overwrites the previous even offset, but writing
2204                  * backwards over the range solves the issue by always
2205                  * writing the odd offset first
2206                  */
2207                 E1000_WRITE_REG_ARRAY(hw, MTA, i, mcarray[i]);
2208         }
2209         E1000_WRITE_FLUSH();
2210
2211         if (hw->mac_type == e1000_82542_rev2_0)
2212                 e1000_leave_82542_rst(adapter);
2213
2214         kfree(mcarray);
2215 }
2216
2217 /* Need to wait a few seconds after link up to get diagnostic information from
2218  * the phy */
2219
2220 static void e1000_update_phy_info(unsigned long data)
2221 {
2222         struct e1000_adapter *adapter = (struct e1000_adapter *)data;
2223         struct e1000_hw *hw = &adapter->hw;
2224         e1000_phy_get_info(hw, &adapter->phy_info);
2225 }
2226
2227 /**
2228  * e1000_82547_tx_fifo_stall - Timer Call-back
2229  * @data: pointer to adapter cast into an unsigned long
2230  **/
2231
2232 static void e1000_82547_tx_fifo_stall(unsigned long data)
2233 {
2234         struct e1000_adapter *adapter = (struct e1000_adapter *)data;
2235         struct e1000_hw *hw = &adapter->hw;
2236         struct net_device *netdev = adapter->netdev;
2237         u32 tctl;
2238
2239         if (atomic_read(&adapter->tx_fifo_stall)) {
2240                 if ((er32(TDT) == er32(TDH)) &&
2241                    (er32(TDFT) == er32(TDFH)) &&
2242                    (er32(TDFTS) == er32(TDFHS))) {
2243                         tctl = er32(TCTL);
2244                         ew32(TCTL, tctl & ~E1000_TCTL_EN);
2245                         ew32(TDFT, adapter->tx_head_addr);
2246                         ew32(TDFH, adapter->tx_head_addr);
2247                         ew32(TDFTS, adapter->tx_head_addr);
2248                         ew32(TDFHS, adapter->tx_head_addr);
2249                         ew32(TCTL, tctl);
2250                         E1000_WRITE_FLUSH();
2251
2252                         adapter->tx_fifo_head = 0;
2253                         atomic_set(&adapter->tx_fifo_stall, 0);
2254                         netif_wake_queue(netdev);
2255                 } else if (!test_bit(__E1000_DOWN, &adapter->flags)) {
2256                         mod_timer(&adapter->tx_fifo_stall_timer, jiffies + 1);
2257                 }
2258         }
2259 }
2260
2261 static bool e1000_has_link(struct e1000_adapter *adapter)
2262 {
2263         struct e1000_hw *hw = &adapter->hw;
2264         bool link_active = false;
2265
2266         /* get_link_status is set on LSC (link status) interrupt or
2267          * rx sequence error interrupt.  get_link_status will stay
2268          * false until the e1000_check_for_link establishes link
2269          * for copper adapters ONLY
2270          */
2271         switch (hw->media_type) {
2272         case e1000_media_type_copper:
2273                 if (hw->get_link_status) {
2274                         e1000_check_for_link(hw);
2275                         link_active = !hw->get_link_status;
2276                 } else {
2277                         link_active = true;
2278                 }
2279                 break;
2280         case e1000_media_type_fiber:
2281                 e1000_check_for_link(hw);
2282                 link_active = !!(er32(STATUS) & E1000_STATUS_LU);
2283                 break;
2284         case e1000_media_type_internal_serdes:
2285                 e1000_check_for_link(hw);
2286                 link_active = hw->serdes_has_link;
2287                 break;
2288         default:
2289                 break;
2290         }
2291
2292         return link_active;
2293 }
2294
2295 /**
2296  * e1000_watchdog - Timer Call-back
2297  * @data: pointer to adapter cast into an unsigned long
2298  **/
2299 static void e1000_watchdog(unsigned long data)
2300 {
2301         struct e1000_adapter *adapter = (struct e1000_adapter *)data;
2302         struct e1000_hw *hw = &adapter->hw;
2303         struct net_device *netdev = adapter->netdev;
2304         struct e1000_tx_ring *txdr = adapter->tx_ring;
2305         u32 link, tctl;
2306
2307         link = e1000_has_link(adapter);
2308         if ((netif_carrier_ok(netdev)) && link)
2309                 goto link_up;
2310
2311         if (link) {
2312                 if (!netif_carrier_ok(netdev)) {
2313                         u32 ctrl;
2314                         bool txb2b = true;
2315                         /* update snapshot of PHY registers on LSC */
2316                         e1000_get_speed_and_duplex(hw,
2317                                                    &adapter->link_speed,
2318                                                    &adapter->link_duplex);
2319
2320                         ctrl = er32(CTRL);
2321                         printk(KERN_INFO "e1000: %s NIC Link is Up %d Mbps %s, "
2322                                "Flow Control: %s\n",
2323                                netdev->name,
2324                                adapter->link_speed,
2325                                adapter->link_duplex == FULL_DUPLEX ?
2326                                 "Full Duplex" : "Half Duplex",
2327                                 ((ctrl & E1000_CTRL_TFCE) && (ctrl &
2328                                 E1000_CTRL_RFCE)) ? "RX/TX" : ((ctrl &
2329                                 E1000_CTRL_RFCE) ? "RX" : ((ctrl &
2330                                 E1000_CTRL_TFCE) ? "TX" : "None" )));
2331
2332                         /* tweak tx_queue_len according to speed/duplex
2333                          * and adjust the timeout factor */
2334                         netdev->tx_queue_len = adapter->tx_queue_len;
2335                         adapter->tx_timeout_factor = 1;
2336                         switch (adapter->link_speed) {
2337                         case SPEED_10:
2338                                 txb2b = false;
2339                                 netdev->tx_queue_len = 10;
2340                                 adapter->tx_timeout_factor = 16;
2341                                 break;
2342                         case SPEED_100:
2343                                 txb2b = false;
2344                                 netdev->tx_queue_len = 100;
2345                                 /* maybe add some timeout factor ? */
2346                                 break;
2347                         }
2348
2349                         /* enable transmits in the hardware */
2350                         tctl = er32(TCTL);
2351                         tctl |= E1000_TCTL_EN;
2352                         ew32(TCTL, tctl);
2353
2354                         netif_carrier_on(netdev);
2355                         if (!test_bit(__E1000_DOWN, &adapter->flags))
2356                                 mod_timer(&adapter->phy_info_timer,
2357                                           round_jiffies(jiffies + 2 * HZ));
2358                         adapter->smartspeed = 0;
2359                 }
2360         } else {
2361                 if (netif_carrier_ok(netdev)) {
2362                         adapter->link_speed = 0;
2363                         adapter->link_duplex = 0;
2364                         printk(KERN_INFO "e1000: %s NIC Link is Down\n",
2365                                netdev->name);
2366                         netif_carrier_off(netdev);
2367
2368                         if (!test_bit(__E1000_DOWN, &adapter->flags))
2369                                 mod_timer(&adapter->phy_info_timer,
2370                                           round_jiffies(jiffies + 2 * HZ));
2371                 }
2372
2373                 e1000_smartspeed(adapter);
2374         }
2375
2376 link_up:
2377         e1000_update_stats(adapter);
2378
2379         hw->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
2380         adapter->tpt_old = adapter->stats.tpt;
2381         hw->collision_delta = adapter->stats.colc - adapter->colc_old;
2382         adapter->colc_old = adapter->stats.colc;
2383
2384         adapter->gorcl = adapter->stats.gorcl - adapter->gorcl_old;
2385         adapter->gorcl_old = adapter->stats.gorcl;
2386         adapter->gotcl = adapter->stats.gotcl - adapter->gotcl_old;
2387         adapter->gotcl_old = adapter->stats.gotcl;
2388
2389         e1000_update_adaptive(hw);
2390
2391         if (!netif_carrier_ok(netdev)) {
2392                 if (E1000_DESC_UNUSED(txdr) + 1 < txdr->count) {
2393                         /* We've lost link, so the controller stops DMA,
2394                          * but we've got queued Tx work that's never going
2395                          * to get done, so reset controller to flush Tx.
2396                          * (Do the reset outside of interrupt context). */
2397                         adapter->tx_timeout_count++;
2398                         schedule_work(&adapter->reset_task);
2399                         /* return immediately since reset is imminent */
2400                         return;
2401                 }
2402         }
2403
2404         /* Cause software interrupt to ensure rx ring is cleaned */
2405         ew32(ICS, E1000_ICS_RXDMT0);
2406
2407         /* Force detection of hung controller every watchdog period */
2408         adapter->detect_tx_hung = true;
2409
2410         /* Reset the timer */
2411         if (!test_bit(__E1000_DOWN, &adapter->flags))
2412                 mod_timer(&adapter->watchdog_timer,
2413                           round_jiffies(jiffies + 2 * HZ));
2414 }
2415
2416 enum latency_range {
2417         lowest_latency = 0,
2418         low_latency = 1,
2419         bulk_latency = 2,
2420         latency_invalid = 255
2421 };
2422
2423 /**
2424  * e1000_update_itr - update the dynamic ITR value based on statistics
2425  * @adapter: pointer to adapter
2426  * @itr_setting: current adapter->itr
2427  * @packets: the number of packets during this measurement interval
2428  * @bytes: the number of bytes during this measurement interval
2429  *
2430  *      Stores a new ITR value based on packets and byte
2431  *      counts during the last interrupt.  The advantage of per interrupt
2432  *      computation is faster updates and more accurate ITR for the current
2433  *      traffic pattern.  Constants in this function were computed
2434  *      based on theoretical maximum wire speed and thresholds were set based
2435  *      on testing data as well as attempting to minimize response time
2436  *      while increasing bulk throughput.
2437  *      this functionality is controlled by the InterruptThrottleRate module
2438  *      parameter (see e1000_param.c)
2439  **/
2440 static unsigned int e1000_update_itr(struct e1000_adapter *adapter,
2441                                      u16 itr_setting, int packets, int bytes)
2442 {
2443         unsigned int retval = itr_setting;
2444         struct e1000_hw *hw = &adapter->hw;
2445
2446         if (unlikely(hw->mac_type < e1000_82540))
2447                 goto update_itr_done;
2448
2449         if (packets == 0)
2450                 goto update_itr_done;
2451
2452         switch (itr_setting) {
2453         case lowest_latency:
2454                 /* jumbo frames get bulk treatment*/
2455                 if (bytes/packets > 8000)
2456                         retval = bulk_latency;
2457                 else if ((packets < 5) && (bytes > 512))
2458                         retval = low_latency;
2459                 break;
2460         case low_latency:  /* 50 usec aka 20000 ints/s */
2461                 if (bytes > 10000) {
2462                         /* jumbo frames need bulk latency setting */
2463                         if (bytes/packets > 8000)
2464                                 retval = bulk_latency;
2465                         else if ((packets < 10) || ((bytes/packets) > 1200))
2466                                 retval = bulk_latency;
2467                         else if ((packets > 35))
2468                                 retval = lowest_latency;
2469                 } else if (bytes/packets > 2000)
2470                         retval = bulk_latency;
2471                 else if (packets <= 2 && bytes < 512)
2472                         retval = lowest_latency;
2473                 break;
2474         case bulk_latency: /* 250 usec aka 4000 ints/s */
2475                 if (bytes > 25000) {
2476                         if (packets > 35)
2477                                 retval = low_latency;
2478                 } else if (bytes < 6000) {
2479                         retval = low_latency;
2480                 }
2481                 break;
2482         }
2483
2484 update_itr_done:
2485         return retval;
2486 }
2487
2488 static void e1000_set_itr(struct e1000_adapter *adapter)
2489 {
2490         struct e1000_hw *hw = &adapter->hw;
2491         u16 current_itr;
2492         u32 new_itr = adapter->itr;
2493
2494         if (unlikely(hw->mac_type < e1000_82540))
2495                 return;
2496
2497         /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
2498         if (unlikely(adapter->link_speed != SPEED_1000)) {
2499                 current_itr = 0;
2500                 new_itr = 4000;
2501                 goto set_itr_now;
2502         }
2503
2504         adapter->tx_itr = e1000_update_itr(adapter,
2505                                     adapter->tx_itr,
2506                                     adapter->total_tx_packets,
2507                                     adapter->total_tx_bytes);
2508         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2509         if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
2510                 adapter->tx_itr = low_latency;
2511
2512         adapter->rx_itr = e1000_update_itr(adapter,
2513                                     adapter->rx_itr,
2514                                     adapter->total_rx_packets,
2515                                     adapter->total_rx_bytes);
2516         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2517         if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
2518                 adapter->rx_itr = low_latency;
2519
2520         current_itr = max(adapter->rx_itr, adapter->tx_itr);
2521
2522         switch (current_itr) {
2523         /* counts and packets in update_itr are dependent on these numbers */
2524         case lowest_latency:
2525                 new_itr = 70000;
2526                 break;
2527         case low_latency:
2528                 new_itr = 20000; /* aka hwitr = ~200 */
2529                 break;
2530         case bulk_latency:
2531                 new_itr = 4000;
2532                 break;
2533         default:
2534                 break;
2535         }
2536
2537 set_itr_now:
2538         if (new_itr != adapter->itr) {
2539                 /* this attempts to bias the interrupt rate towards Bulk
2540                  * by adding intermediate steps when interrupt rate is
2541                  * increasing */
2542                 new_itr = new_itr > adapter->itr ?
2543                              min(adapter->itr + (new_itr >> 2), new_itr) :
2544                              new_itr;
2545                 adapter->itr = new_itr;
2546                 ew32(ITR, 1000000000 / (new_itr * 256));
2547         }
2548
2549         return;
2550 }
2551
2552 #define E1000_TX_FLAGS_CSUM             0x00000001
2553 #define E1000_TX_FLAGS_VLAN             0x00000002
2554 #define E1000_TX_FLAGS_TSO              0x00000004
2555 #define E1000_TX_FLAGS_IPV4             0x00000008
2556 #define E1000_TX_FLAGS_VLAN_MASK        0xffff0000
2557 #define E1000_TX_FLAGS_VLAN_SHIFT       16
2558
2559 static int e1000_tso(struct e1000_adapter *adapter,
2560                      struct e1000_tx_ring *tx_ring, struct sk_buff *skb)
2561 {
2562         struct e1000_context_desc *context_desc;
2563         struct e1000_buffer *buffer_info;
2564         unsigned int i;
2565         u32 cmd_length = 0;
2566         u16 ipcse = 0, tucse, mss;
2567         u8 ipcss, ipcso, tucss, tucso, hdr_len;
2568         int err;
2569
2570         if (skb_is_gso(skb)) {
2571                 if (skb_header_cloned(skb)) {
2572                         err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2573                         if (err)
2574                                 return err;
2575                 }
2576
2577                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
2578                 mss = skb_shinfo(skb)->gso_size;
2579                 if (skb->protocol == htons(ETH_P_IP)) {
2580                         struct iphdr *iph = ip_hdr(skb);
2581                         iph->tot_len = 0;
2582                         iph->check = 0;
2583                         tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
2584                                                                  iph->daddr, 0,
2585                                                                  IPPROTO_TCP,
2586                                                                  0);
2587                         cmd_length = E1000_TXD_CMD_IP;
2588                         ipcse = skb_transport_offset(skb) - 1;
2589                 } else if (skb->protocol == htons(ETH_P_IPV6)) {
2590                         ipv6_hdr(skb)->payload_len = 0;
2591                         tcp_hdr(skb)->check =
2592                                 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
2593                                                  &ipv6_hdr(skb)->daddr,
2594                                                  0, IPPROTO_TCP, 0);
2595                         ipcse = 0;
2596                 }
2597                 ipcss = skb_network_offset(skb);
2598                 ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
2599                 tucss = skb_transport_offset(skb);
2600                 tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
2601                 tucse = 0;
2602
2603                 cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
2604                                E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
2605
2606                 i = tx_ring->next_to_use;
2607                 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
2608                 buffer_info = &tx_ring->buffer_info[i];
2609
2610                 context_desc->lower_setup.ip_fields.ipcss  = ipcss;
2611                 context_desc->lower_setup.ip_fields.ipcso  = ipcso;
2612                 context_desc->lower_setup.ip_fields.ipcse  = cpu_to_le16(ipcse);
2613                 context_desc->upper_setup.tcp_fields.tucss = tucss;
2614                 context_desc->upper_setup.tcp_fields.tucso = tucso;
2615                 context_desc->upper_setup.tcp_fields.tucse = cpu_to_le16(tucse);
2616                 context_desc->tcp_seg_setup.fields.mss     = cpu_to_le16(mss);
2617                 context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
2618                 context_desc->cmd_and_length = cpu_to_le32(cmd_length);
2619
2620                 buffer_info->time_stamp = jiffies;
2621                 buffer_info->next_to_watch = i;
2622
2623                 if (++i == tx_ring->count) i = 0;
2624                 tx_ring->next_to_use = i;
2625
2626                 return true;
2627         }
2628         return false;
2629 }
2630
2631 static bool e1000_tx_csum(struct e1000_adapter *adapter,
2632                           struct e1000_tx_ring *tx_ring, struct sk_buff *skb)
2633 {
2634         struct e1000_context_desc *context_desc;
2635         struct e1000_buffer *buffer_info;
2636         unsigned int i;
2637         u8 css;
2638         u32 cmd_len = E1000_TXD_CMD_DEXT;
2639
2640         if (skb->ip_summed != CHECKSUM_PARTIAL)
2641                 return false;
2642
2643         switch (skb->protocol) {
2644         case cpu_to_be16(ETH_P_IP):
2645                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
2646                         cmd_len |= E1000_TXD_CMD_TCP;
2647                 break;
2648         case cpu_to_be16(ETH_P_IPV6):
2649                 /* XXX not handling all IPV6 headers */
2650                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
2651                         cmd_len |= E1000_TXD_CMD_TCP;
2652                 break;
2653         default:
2654                 if (unlikely(net_ratelimit()))
2655                         DPRINTK(DRV, WARNING,
2656                                 "checksum_partial proto=%x!\n", skb->protocol);
2657                 break;
2658         }
2659
2660         css = skb_transport_offset(skb);
2661
2662         i = tx_ring->next_to_use;
2663         buffer_info = &tx_ring->buffer_info[i];
2664         context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
2665
2666         context_desc->lower_setup.ip_config = 0;
2667         context_desc->upper_setup.tcp_fields.tucss = css;
2668         context_desc->upper_setup.tcp_fields.tucso =
2669                 css + skb->csum_offset;
2670         context_desc->upper_setup.tcp_fields.tucse = 0;
2671         context_desc->tcp_seg_setup.data = 0;
2672         context_desc->cmd_and_length = cpu_to_le32(cmd_len);
2673
2674         buffer_info->time_stamp = jiffies;
2675         buffer_info->next_to_watch = i;
2676
2677         if (unlikely(++i == tx_ring->count)) i = 0;
2678         tx_ring->next_to_use = i;
2679
2680         return true;
2681 }
2682
2683 #define E1000_MAX_TXD_PWR       12
2684 #define E1000_MAX_DATA_PER_TXD  (1<<E1000_MAX_TXD_PWR)
2685
2686 static int e1000_tx_map(struct e1000_adapter *adapter,
2687                         struct e1000_tx_ring *tx_ring,
2688                         struct sk_buff *skb, unsigned int first,
2689                         unsigned int max_per_txd, unsigned int nr_frags,
2690                         unsigned int mss)
2691 {
2692         struct e1000_hw *hw = &adapter->hw;
2693         struct pci_dev *pdev = adapter->pdev;
2694         struct e1000_buffer *buffer_info;
2695         unsigned int len = skb_headlen(skb);
2696         unsigned int offset = 0, size, count = 0, i;
2697         unsigned int f;
2698
2699         i = tx_ring->next_to_use;
2700
2701         while (len) {
2702                 buffer_info = &tx_ring->buffer_info[i];
2703                 size = min(len, max_per_txd);
2704                 /* Workaround for Controller erratum --
2705                  * descriptor for non-tso packet in a linear SKB that follows a
2706                  * tso gets written back prematurely before the data is fully
2707                  * DMA'd to the controller */
2708                 if (!skb->data_len && tx_ring->last_tx_tso &&
2709                     !skb_is_gso(skb)) {
2710                         tx_ring->last_tx_tso = 0;
2711                         size -= 4;
2712                 }
2713
2714                 /* Workaround for premature desc write-backs
2715                  * in TSO mode.  Append 4-byte sentinel desc */
2716                 if (unlikely(mss && !nr_frags && size == len && size > 8))
2717                         size -= 4;
2718                 /* work-around for errata 10 and it applies
2719                  * to all controllers in PCI-X mode
2720                  * The fix is to make sure that the first descriptor of a
2721                  * packet is smaller than 2048 - 16 - 16 (or 2016) bytes
2722                  */
2723                 if (unlikely((hw->bus_type == e1000_bus_type_pcix) &&
2724                                 (size > 2015) && count == 0))
2725                         size = 2015;
2726
2727                 /* Workaround for potential 82544 hang in PCI-X.  Avoid
2728                  * terminating buffers within evenly-aligned dwords. */
2729                 if (unlikely(adapter->pcix_82544 &&
2730                    !((unsigned long)(skb->data + offset + size - 1) & 4) &&
2731                    size > 4))
2732                         size -= 4;
2733
2734                 buffer_info->length = size;
2735                 /* set time_stamp *before* dma to help avoid a possible race */
2736                 buffer_info->time_stamp = jiffies;
2737                 buffer_info->mapped_as_page = false;
2738                 buffer_info->dma = pci_map_single(pdev, skb->data + offset,
2739                                                   size, PCI_DMA_TODEVICE);
2740                 if (pci_dma_mapping_error(pdev, buffer_info->dma))
2741                         goto dma_error;
2742                 buffer_info->next_to_watch = i;
2743
2744                 len -= size;
2745                 offset += size;
2746                 count++;
2747                 if (len) {
2748                         i++;
2749                         if (unlikely(i == tx_ring->count))
2750                                 i = 0;
2751                 }
2752         }
2753
2754         for (f = 0; f < nr_frags; f++) {
2755                 struct skb_frag_struct *frag;
2756
2757                 frag = &skb_shinfo(skb)->frags[f];
2758                 len = frag->size;
2759                 offset = frag->page_offset;
2760
2761                 while (len) {
2762                         i++;
2763                         if (unlikely(i == tx_ring->count))
2764                                 i = 0;
2765
2766                         buffer_info = &tx_ring->buffer_info[i];
2767                         size = min(len, max_per_txd);
2768                         /* Workaround for premature desc write-backs
2769                          * in TSO mode.  Append 4-byte sentinel desc */
2770                         if (unlikely(mss && f == (nr_frags-1) && size == len && size > 8))
2771                                 size -= 4;
2772                         /* Workaround for potential 82544 hang in PCI-X.
2773                          * Avoid terminating buffers within evenly-aligned
2774                          * dwords. */
2775                         if (unlikely(adapter->pcix_82544 &&
2776                             !((unsigned long)(page_to_phys(frag->page) + offset
2777                                               + size - 1) & 4) &&
2778                             size > 4))
2779                                 size -= 4;
2780
2781                         buffer_info->length = size;
2782                         buffer_info->time_stamp = jiffies;
2783                         buffer_info->mapped_as_page = true;
2784                         buffer_info->dma = pci_map_page(pdev, frag->page,
2785                                                         offset, size,
2786                                                         PCI_DMA_TODEVICE);
2787                         if (pci_dma_mapping_error(pdev, buffer_info->dma))
2788                                 goto dma_error;
2789                         buffer_info->next_to_watch = i;
2790
2791                         len -= size;
2792                         offset += size;
2793                         count++;
2794                 }
2795         }
2796
2797         tx_ring->buffer_info[i].skb = skb;
2798         tx_ring->buffer_info[first].next_to_watch = i;
2799
2800         return count;
2801
2802 dma_error:
2803         dev_err(&pdev->dev, "TX DMA map failed\n");
2804         buffer_info->dma = 0;
2805         if (count)
2806                 count--;
2807
2808         while (count--) {
2809                 if (i==0)
2810                         i += tx_ring->count;
2811                 i--;
2812                 buffer_info = &tx_ring->buffer_info[i];
2813                 e1000_unmap_and_free_tx_resource(adapter, buffer_info);
2814         }
2815
2816         return 0;
2817 }
2818
2819 static void e1000_tx_queue(struct e1000_adapter *adapter,
2820                            struct e1000_tx_ring *tx_ring, int tx_flags,
2821                            int count)
2822 {
2823         struct e1000_hw *hw = &adapter->hw;
2824         struct e1000_tx_desc *tx_desc = NULL;
2825         struct e1000_buffer *buffer_info;
2826         u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
2827         unsigned int i;
2828
2829         if (likely(tx_flags & E1000_TX_FLAGS_TSO)) {
2830                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
2831                              E1000_TXD_CMD_TSE;
2832                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
2833
2834                 if (likely(tx_flags & E1000_TX_FLAGS_IPV4))
2835                         txd_upper |= E1000_TXD_POPTS_IXSM << 8;
2836         }
2837
2838         if (likely(tx_flags & E1000_TX_FLAGS_CSUM)) {
2839                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
2840                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
2841         }
2842
2843         if (unlikely(tx_flags & E1000_TX_FLAGS_VLAN)) {
2844                 txd_lower |= E1000_TXD_CMD_VLE;
2845                 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
2846         }
2847
2848         i = tx_ring->next_to_use;
2849
2850         while (count--) {
2851                 buffer_info = &tx_ring->buffer_info[i];
2852                 tx_desc = E1000_TX_DESC(*tx_ring, i);
2853                 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
2854                 tx_desc->lower.data =
2855                         cpu_to_le32(txd_lower | buffer_info->length);
2856                 tx_desc->upper.data = cpu_to_le32(txd_upper);
2857                 if (unlikely(++i == tx_ring->count)) i = 0;
2858         }
2859
2860         tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
2861
2862         /* Force memory writes to complete before letting h/w
2863          * know there are new descriptors to fetch.  (Only
2864          * applicable for weak-ordered memory model archs,
2865          * such as IA-64). */
2866         wmb();
2867
2868         tx_ring->next_to_use = i;
2869         writel(i, hw->hw_addr + tx_ring->tdt);
2870         /* we need this if more than one processor can write to our tail
2871          * at a time, it syncronizes IO on IA64/Altix systems */
2872         mmiowb();
2873 }
2874
2875 /**
2876  * 82547 workaround to avoid controller hang in half-duplex environment.
2877  * The workaround is to avoid queuing a large packet that would span
2878  * the internal Tx FIFO ring boundary by notifying the stack to resend
2879  * the packet at a later time.  This gives the Tx FIFO an opportunity to
2880  * flush all packets.  When that occurs, we reset the Tx FIFO pointers
2881  * to the beginning of the Tx FIFO.
2882  **/
2883
2884 #define E1000_FIFO_HDR                  0x10
2885 #define E1000_82547_PAD_LEN             0x3E0
2886
2887 static int e1000_82547_fifo_workaround(struct e1000_adapter *adapter,
2888                                        struct sk_buff *skb)
2889 {
2890         u32 fifo_space = adapter->tx_fifo_size - adapter->tx_fifo_head;
2891         u32 skb_fifo_len = skb->len + E1000_FIFO_HDR;
2892
2893         skb_fifo_len = ALIGN(skb_fifo_len, E1000_FIFO_HDR);
2894
2895         if (adapter->link_duplex != HALF_DUPLEX)
2896                 goto no_fifo_stall_required;
2897
2898         if (atomic_read(&adapter->tx_fifo_stall))
2899                 return 1;
2900
2901         if (skb_fifo_len >= (E1000_82547_PAD_LEN + fifo_space)) {
2902                 atomic_set(&adapter->tx_fifo_stall, 1);
2903                 return 1;
2904         }
2905
2906 no_fifo_stall_required:
2907         adapter->tx_fifo_head += skb_fifo_len;
2908         if (adapter->tx_fifo_head >= adapter->tx_fifo_size)
2909                 adapter->tx_fifo_head -= adapter->tx_fifo_size;
2910         return 0;
2911 }
2912
2913 static int __e1000_maybe_stop_tx(struct net_device *netdev, int size)
2914 {
2915         struct e1000_adapter *adapter = netdev_priv(netdev);
2916         struct e1000_tx_ring *tx_ring = adapter->tx_ring;
2917
2918         netif_stop_queue(netdev);
2919         /* Herbert's original patch had:
2920          *  smp_mb__after_netif_stop_queue();
2921          * but since that doesn't exist yet, just open code it. */
2922         smp_mb();
2923
2924         /* We need to check again in a case another CPU has just
2925          * made room available. */
2926         if (likely(E1000_DESC_UNUSED(tx_ring) < size))
2927                 return -EBUSY;
2928
2929         /* A reprieve! */
2930         netif_start_queue(netdev);
2931         ++adapter->restart_queue;
2932         return 0;
2933 }
2934
2935 static int e1000_maybe_stop_tx(struct net_device *netdev,
2936                                struct e1000_tx_ring *tx_ring, int size)
2937 {
2938         if (likely(E1000_DESC_UNUSED(tx_ring) >= size))
2939                 return 0;
2940         return __e1000_maybe_stop_tx(netdev, size);
2941 }
2942
2943 #define TXD_USE_COUNT(S, X) (((S) >> (X)) + 1 )
2944 static netdev_tx_t e1000_xmit_frame(struct sk_buff *skb,
2945                                     struct net_device *netdev)
2946 {
2947         struct e1000_adapter *adapter = netdev_priv(netdev);
2948         struct e1000_hw *hw = &adapter->hw;
2949         struct e1000_tx_ring *tx_ring;
2950         unsigned int first, max_per_txd = E1000_MAX_DATA_PER_TXD;
2951         unsigned int max_txd_pwr = E1000_MAX_TXD_PWR;
2952         unsigned int tx_flags = 0;
2953         unsigned int len = skb->len - skb->data_len;
2954         unsigned int nr_frags;
2955         unsigned int mss;
2956         int count = 0;
2957         int tso;
2958         unsigned int f;
2959
2960         /* This goes back to the question of how to logically map a tx queue
2961          * to a flow.  Right now, performance is impacted slightly negatively
2962          * if using multiple tx queues.  If the stack breaks away from a
2963          * single qdisc implementation, we can look at this again. */
2964         tx_ring = adapter->tx_ring;
2965
2966         if (unlikely(skb->len <= 0)) {
2967                 dev_kfree_skb_any(skb);
2968                 return NETDEV_TX_OK;
2969         }
2970
2971         mss = skb_shinfo(skb)->gso_size;
2972         /* The controller does a simple calculation to
2973          * make sure there is enough room in the FIFO before
2974          * initiating the DMA for each buffer.  The calc is:
2975          * 4 = ceil(buffer len/mss).  To make sure we don't
2976          * overrun the FIFO, adjust the max buffer len if mss
2977          * drops. */
2978         if (mss) {
2979                 u8 hdr_len;
2980                 max_per_txd = min(mss << 2, max_per_txd);
2981                 max_txd_pwr = fls(max_per_txd) - 1;
2982
2983                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
2984                 if (skb->data_len && hdr_len == len) {
2985                         switch (hw->mac_type) {
2986                                 unsigned int pull_size;
2987                         case e1000_82544:
2988                                 /* Make sure we have room to chop off 4 bytes,
2989                                  * and that the end alignment will work out to
2990                                  * this hardware's requirements
2991                                  * NOTE: this is a TSO only workaround
2992                                  * if end byte alignment not correct move us
2993                                  * into the next dword */
2994                                 if ((unsigned long)(skb_tail_pointer(skb) - 1) & 4)
2995                                         break;
2996                                 /* fall through */
2997                                 pull_size = min((unsigned int)4, skb->data_len);
2998                                 if (!__pskb_pull_tail(skb, pull_size)) {
2999                                         DPRINTK(DRV, ERR,
3000                                                 "__pskb_pull_tail failed.\n");
3001                                         dev_kfree_skb_any(skb);
3002                                         return NETDEV_TX_OK;
3003                                 }
3004                                 len = skb->len - skb->data_len;
3005                                 break;
3006                         default:
3007                                 /* do nothing */
3008                                 break;
3009                         }
3010                 }
3011         }
3012
3013         /* reserve a descriptor for the offload context */
3014         if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
3015                 count++;
3016         count++;
3017
3018         /* Controller Erratum workaround */
3019         if (!skb->data_len && tx_ring->last_tx_tso && !skb_is_gso(skb))
3020                 count++;
3021
3022         count += TXD_USE_COUNT(len, max_txd_pwr);
3023
3024         if (adapter->pcix_82544)
3025                 count++;
3026
3027         /* work-around for errata 10 and it applies to all controllers
3028          * in PCI-X mode, so add one more descriptor to the count
3029          */
3030         if (unlikely((hw->bus_type == e1000_bus_type_pcix) &&
3031                         (len > 2015)))
3032                 count++;
3033
3034         nr_frags = skb_shinfo(skb)->nr_frags;
3035         for (f = 0; f < nr_frags; f++)
3036                 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size,
3037                                        max_txd_pwr);
3038         if (adapter->pcix_82544)
3039                 count += nr_frags;
3040
3041         /* need: count + 2 desc gap to keep tail from touching
3042          * head, otherwise try next time */
3043         if (unlikely(e1000_maybe_stop_tx(netdev, tx_ring, count + 2)))
3044                 return NETDEV_TX_BUSY;
3045
3046         if (unlikely(hw->mac_type == e1000_82547)) {
3047                 if (unlikely(e1000_82547_fifo_workaround(adapter, skb))) {
3048                         netif_stop_queue(netdev);
3049                         if (!test_bit(__E1000_DOWN, &adapter->flags))
3050                                 mod_timer(&adapter->tx_fifo_stall_timer,
3051                                           jiffies + 1);
3052                         return NETDEV_TX_BUSY;
3053                 }
3054         }
3055
3056         if (unlikely(adapter->vlgrp && vlan_tx_tag_present(skb))) {
3057                 tx_flags |= E1000_TX_FLAGS_VLAN;
3058                 tx_flags |= (vlan_tx_tag_get(skb) << E1000_TX_FLAGS_VLAN_SHIFT);
3059         }
3060
3061         first = tx_ring->next_to_use;
3062
3063         tso = e1000_tso(adapter, tx_ring, skb);
3064         if (tso < 0) {
3065                 dev_kfree_skb_any(skb);
3066                 return NETDEV_TX_OK;
3067         }
3068
3069         if (likely(tso)) {
3070                 if (likely(hw->mac_type != e1000_82544))
3071                         tx_ring->last_tx_tso = 1;
3072                 tx_flags |= E1000_TX_FLAGS_TSO;
3073         } else if (likely(e1000_tx_csum(adapter, tx_ring, skb)))
3074                 tx_flags |= E1000_TX_FLAGS_CSUM;
3075
3076         if (likely(skb->protocol == htons(ETH_P_IP)))
3077                 tx_flags |= E1000_TX_FLAGS_IPV4;
3078
3079         count = e1000_tx_map(adapter, tx_ring, skb, first, max_per_txd,
3080                              nr_frags, mss);
3081
3082         if (count) {
3083                 e1000_tx_queue(adapter, tx_ring, tx_flags, count);
3084                 /* Make sure there is space in the ring for the next send. */
3085                 e1000_maybe_stop_tx(netdev, tx_ring, MAX_SKB_FRAGS + 2);
3086
3087         } else {
3088                 dev_kfree_skb_any(skb);
3089                 tx_ring->buffer_info[first].time_stamp = 0;
3090                 tx_ring->next_to_use = first;
3091         }
3092
3093         return NETDEV_TX_OK;
3094 }
3095
3096 /**
3097  * e1000_tx_timeout - Respond to a Tx Hang
3098  * @netdev: network interface device structure
3099  **/
3100
3101 static void e1000_tx_timeout(struct net_device *netdev)
3102 {
3103         struct e1000_adapter *adapter = netdev_priv(netdev);
3104
3105         /* Do the reset outside of interrupt context */
3106         adapter->tx_timeout_count++;
3107         schedule_work(&adapter->reset_task);
3108 }
3109
3110 static void e1000_reset_task(struct work_struct *work)
3111 {
3112         struct e1000_adapter *adapter =
3113                 container_of(work, struct e1000_adapter, reset_task);
3114
3115         e1000_reinit_locked(adapter);
3116 }
3117
3118 /**
3119  * e1000_get_stats - Get System Network Statistics
3120  * @netdev: network interface device structure
3121  *
3122  * Returns the address of the device statistics structure.
3123  * The statistics are actually updated from the timer callback.
3124  **/
3125
3126 static struct net_device_stats *e1000_get_stats(struct net_device *netdev)
3127 {
3128         /* only return the current stats */
3129         return &netdev->stats;
3130 }
3131
3132 /**
3133  * e1000_change_mtu - Change the Maximum Transfer Unit
3134  * @netdev: network interface device structure
3135  * @new_mtu: new value for maximum frame size
3136  *
3137  * Returns 0 on success, negative on failure
3138  **/
3139
3140 static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
3141 {
3142         struct e1000_adapter *adapter = netdev_priv(netdev);
3143         struct e1000_hw *hw = &adapter->hw;
3144         int max_frame = new_mtu + ENET_HEADER_SIZE + ETHERNET_FCS_SIZE;
3145
3146         if ((max_frame < MINIMUM_ETHERNET_FRAME_SIZE) ||
3147             (max_frame > MAX_JUMBO_FRAME_SIZE)) {
3148                 DPRINTK(PROBE, ERR, "Invalid MTU setting\n");
3149                 return -EINVAL;
3150         }
3151
3152         /* Adapter-specific max frame size limits. */
3153         switch (hw->mac_type) {
3154         case e1000_undefined ... e1000_82542_rev2_1:
3155                 if (max_frame > (ETH_FRAME_LEN + ETH_FCS_LEN)) {
3156                         DPRINTK(PROBE, ERR, "Jumbo Frames not supported.\n");
3157                         return -EINVAL;
3158                 }
3159                 break;
3160         default:
3161                 /* Capable of supporting up to MAX_JUMBO_FRAME_SIZE limit. */
3162                 break;
3163         }
3164
3165         while (test_and_set_bit(__E1000_RESETTING, &adapter->flags))
3166                 msleep(1);
3167         /* e1000_down has a dependency on max_frame_size */
3168         hw->max_frame_size = max_frame;
3169         if (netif_running(netdev))
3170                 e1000_down(adapter);
3171
3172         /* NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
3173          * means we reserve 2 more, this pushes us to allocate from the next
3174          * larger slab size.
3175          * i.e. RXBUFFER_2048 --> size-4096 slab
3176          *  however with the new *_jumbo_rx* routines, jumbo receives will use
3177          *  fragmented skbs */
3178
3179         if (max_frame <= E1000_RXBUFFER_256)
3180                 adapter->rx_buffer_len = E1000_RXBUFFER_256;
3181         else if (max_frame <= E1000_RXBUFFER_512)
3182                 adapter->rx_buffer_len = E1000_RXBUFFER_512;
3183         else if (max_frame <= E1000_RXBUFFER_1024)
3184                 adapter->rx_buffer_len = E1000_RXBUFFER_1024;
3185         else if (max_frame <= E1000_RXBUFFER_2048)
3186                 adapter->rx_buffer_len = E1000_RXBUFFER_2048;
3187         else
3188 #if (PAGE_SIZE >= E1000_RXBUFFER_16384)
3189                 adapter->rx_buffer_len = E1000_RXBUFFER_16384;
3190 #elif (PAGE_SIZE >= E1000_RXBUFFER_4096)
3191                 adapter->rx_buffer_len = PAGE_SIZE;
3192 #endif
3193
3194         /* adjust allocation if LPE protects us, and we aren't using SBP */
3195         if (!hw->tbi_compatibility_on &&
3196             ((max_frame == (ETH_FRAME_LEN + ETH_FCS_LEN)) ||
3197              (max_frame == MAXIMUM_ETHERNET_VLAN_SIZE)))
3198                 adapter->rx_buffer_len = MAXIMUM_ETHERNET_VLAN_SIZE;
3199
3200         printk(KERN_INFO "e1000: %s changing MTU from %d to %d\n",
3201                netdev->name, netdev->mtu, new_mtu);
3202         netdev->mtu = new_mtu;
3203
3204         if (netif_running(netdev))
3205                 e1000_up(adapter);
3206         else
3207                 e1000_reset(adapter);
3208
3209         clear_bit(__E1000_RESETTING, &adapter->flags);
3210
3211         return 0;
3212 }
3213
3214 /**
3215  * e1000_update_stats - Update the board statistics counters
3216  * @adapter: board private structure
3217  **/
3218
3219 void e1000_update_stats(struct e1000_adapter *adapter)
3220 {
3221         struct net_device *netdev = adapter->netdev;
3222         struct e1000_hw *hw = &adapter->hw;
3223         struct pci_dev *pdev = adapter->pdev;
3224         unsigned long flags;
3225         u16 phy_tmp;
3226
3227 #define PHY_IDLE_ERROR_COUNT_MASK 0x00FF
3228
3229         /*
3230          * Prevent stats update while adapter is being reset, or if the pci
3231          * connection is down.
3232          */
3233         if (adapter->link_speed == 0)
3234                 return;
3235         if (pci_channel_offline(pdev))
3236                 return;
3237
3238         spin_lock_irqsave(&adapter->stats_lock, flags);
3239
3240         /* these counters are modified from e1000_tbi_adjust_stats,
3241          * called from the interrupt context, so they must only
3242          * be written while holding adapter->stats_lock
3243          */
3244
3245         adapter->stats.crcerrs += er32(CRCERRS);
3246         adapter->stats.gprc += er32(GPRC);
3247         adapter->stats.gorcl += er32(GORCL);
3248         adapter->stats.gorch += er32(GORCH);
3249         adapter->stats.bprc += er32(BPRC);
3250         adapter->stats.mprc += er32(MPRC);
3251         adapter->stats.roc += er32(ROC);
3252
3253         adapter->stats.prc64 += er32(PRC64);
3254         adapter->stats.prc127 += er32(PRC127);
3255         adapter->stats.prc255 += er32(PRC255);
3256         adapter->stats.prc511 += er32(PRC511);
3257         adapter->stats.prc1023 += er32(PRC1023);
3258         adapter->stats.prc1522 += er32(PRC1522);
3259
3260         adapter->stats.symerrs += er32(SYMERRS);
3261         adapter->stats.mpc += er32(MPC);
3262         adapter->stats.scc += er32(SCC);
3263         adapter->stats.ecol += er32(ECOL);
3264         adapter->stats.mcc += er32(MCC);
3265         adapter->stats.latecol += er32(LATECOL);
3266         adapter->stats.dc += er32(DC);
3267         adapter->stats.sec += er32(SEC);
3268         adapter->stats.rlec += er32(RLEC);
3269         adapter->stats.xonrxc += er32(XONRXC);
3270         adapter->stats.xontxc += er32(XONTXC);
3271         adapter->stats.xoffrxc += er32(XOFFRXC);
3272         adapter->stats.xofftxc += er32(XOFFTXC);
3273         adapter->stats.fcruc += er32(FCRUC);
3274         adapter->stats.gptc += er32(GPTC);
3275         adapter->stats.gotcl += er32(GOTCL);
3276         adapter->stats.gotch += er32(GOTCH);
3277         adapter->stats.rnbc += er32(RNBC);
3278         adapter->stats.ruc += er32(RUC);
3279         adapter->stats.rfc += er32(RFC);
3280         adapter->stats.rjc += er32(RJC);
3281         adapter->stats.torl += er32(TORL);
3282         adapter->stats.torh += er32(TORH);
3283         adapter->stats.totl += er32(TOTL);
3284         adapter->stats.toth += er32(TOTH);
3285         adapter->stats.tpr += er32(TPR);
3286
3287         adapter->stats.ptc64 += er32(PTC64);
3288         adapter->stats.ptc127 += er32(PTC127);
3289         adapter->stats.ptc255 += er32(PTC255);
3290         adapter->stats.ptc511 += er32(PTC511);
3291         adapter->stats.ptc1023 += er32(PTC1023);
3292         adapter->stats.ptc1522 += er32(PTC1522);
3293
3294         adapter->stats.mptc += er32(MPTC);
3295         adapter->stats.bptc += er32(BPTC);
3296
3297         /* used for adaptive IFS */
3298
3299         hw->tx_packet_delta = er32(TPT);
3300         adapter->stats.tpt += hw->tx_packet_delta;
3301         hw->collision_delta = er32(COLC);
3302         adapter->stats.colc += hw->collision_delta;
3303
3304         if (hw->mac_type >= e1000_82543) {
3305                 adapter->stats.algnerrc += er32(ALGNERRC);
3306                 adapter->stats.rxerrc += er32(RXERRC);
3307                 adapter->stats.tncrs += er32(TNCRS);
3308                 adapter->stats.cexterr += er32(CEXTERR);
3309                 adapter->stats.tsctc += er32(TSCTC);
3310                 adapter->stats.tsctfc += er32(TSCTFC);
3311         }
3312
3313         /* Fill out the OS statistics structure */
3314         netdev->stats.multicast = adapter->stats.mprc;
3315         netdev->stats.collisions = adapter->stats.colc;
3316
3317         /* Rx Errors */
3318
3319         /* RLEC on some newer hardware can be incorrect so build
3320         * our own version based on RUC and ROC */
3321         netdev->stats.rx_errors = adapter->stats.rxerrc +
3322                 adapter->stats.crcerrs + adapter->stats.algnerrc +
3323                 adapter->stats.ruc + adapter->stats.roc +
3324                 adapter->stats.cexterr;
3325         adapter->stats.rlerrc = adapter->stats.ruc + adapter->stats.roc;
3326         netdev->stats.rx_length_errors = adapter->stats.rlerrc;
3327         netdev->stats.rx_crc_errors = adapter->stats.crcerrs;
3328         netdev->stats.rx_frame_errors = adapter->stats.algnerrc;
3329         netdev->stats.rx_missed_errors = adapter->stats.mpc;
3330
3331         /* Tx Errors */
3332         adapter->stats.txerrc = adapter->stats.ecol + adapter->stats.latecol;
3333         netdev->stats.tx_errors = adapter->stats.txerrc;
3334         netdev->stats.tx_aborted_errors = adapter->stats.ecol;
3335         netdev->stats.tx_window_errors = adapter->stats.latecol;
3336         netdev->stats.tx_carrier_errors = adapter->stats.tncrs;
3337         if (hw->bad_tx_carr_stats_fd &&
3338             adapter->link_duplex == FULL_DUPLEX) {
3339                 netdev->stats.tx_carrier_errors = 0;
3340                 adapter->stats.tncrs = 0;
3341         }
3342
3343         /* Tx Dropped needs to be maintained elsewhere */
3344
3345         /* Phy Stats */
3346         if (hw->media_type == e1000_media_type_copper) {
3347                 if ((adapter->link_speed == SPEED_1000) &&
3348                    (!e1000_read_phy_reg(hw, PHY_1000T_STATUS, &phy_tmp))) {
3349                         phy_tmp &= PHY_IDLE_ERROR_COUNT_MASK;
3350                         adapter->phy_stats.idle_errors += phy_tmp;
3351                 }
3352
3353                 if ((hw->mac_type <= e1000_82546) &&
3354                    (hw->phy_type == e1000_phy_m88) &&
3355                    !e1000_read_phy_reg(hw, M88E1000_RX_ERR_CNTR, &phy_tmp))
3356                         adapter->phy_stats.receive_errors += phy_tmp;
3357         }
3358
3359         /* Management Stats */
3360         if (hw->has_smbus) {
3361                 adapter->stats.mgptc += er32(MGTPTC);
3362                 adapter->stats.mgprc += er32(MGTPRC);
3363                 adapter->stats.mgpdc += er32(MGTPDC);
3364         }
3365
3366         spin_unlock_irqrestore(&adapter->stats_lock, flags);
3367 }
3368
3369 /**
3370  * e1000_intr - Interrupt Handler
3371  * @irq: interrupt number
3372  * @data: pointer to a network interface device structure
3373  **/
3374
3375 static irqreturn_t e1000_intr(int irq, void *data)
3376 {
3377         struct net_device *netdev = data;
3378         struct e1000_adapter *adapter = netdev_priv(netdev);
3379         struct e1000_hw *hw = &adapter->hw;
3380         u32 icr = er32(ICR);
3381
3382         if (unlikely((!icr) || test_bit(__E1000_DOWN, &adapter->flags)))
3383                 return IRQ_NONE;  /* Not our interrupt */
3384
3385         if (unlikely(icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC))) {
3386                 hw->get_link_status = 1;
3387                 /* guard against interrupt when we're going down */
3388                 if (!test_bit(__E1000_DOWN, &adapter->flags))
3389                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
3390         }
3391
3392         /* disable interrupts, without the synchronize_irq bit */
3393         ew32(IMC, ~0);
3394         E1000_WRITE_FLUSH();
3395
3396         if (likely(napi_schedule_prep(&adapter->napi))) {
3397                 adapter->total_tx_bytes = 0;
3398                 adapter->total_tx_packets = 0;
3399                 adapter->total_rx_bytes = 0;
3400                 adapter->total_rx_packets = 0;
3401                 __napi_schedule(&adapter->napi);
3402         } else {
3403                 /* this really should not happen! if it does it is basically a
3404                  * bug, but not a hard error, so enable ints and continue */
3405                 if (!test_bit(__E1000_DOWN, &adapter->flags))
3406                         e1000_irq_enable(adapter);
3407         }
3408
3409         return IRQ_HANDLED;
3410 }
3411
3412 /**
3413  * e1000_clean - NAPI Rx polling callback
3414  * @adapter: board private structure
3415  **/
3416 static int e1000_clean(struct napi_struct *napi, int budget)
3417 {
3418         struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter, napi);
3419         int tx_clean_complete = 0, work_done = 0;
3420
3421         tx_clean_complete = e1000_clean_tx_irq(adapter, &adapter->tx_ring[0]);
3422
3423         adapter->clean_rx(adapter, &adapter->rx_ring[0], &work_done, budget);
3424
3425         if (!tx_clean_complete)
3426                 work_done = budget;
3427
3428         /* If budget not fully consumed, exit the polling mode */
3429         if (work_done < budget) {
3430                 if (likely(adapter->itr_setting & 3))
3431                         e1000_set_itr(adapter);
3432                 napi_complete(napi);
3433                 if (!test_bit(__E1000_DOWN, &adapter->flags))
3434                         e1000_irq_enable(adapter);
3435         }
3436
3437         return work_done;
3438 }
3439
3440 /**
3441  * e1000_clean_tx_irq - Reclaim resources after transmit completes
3442  * @adapter: board private structure
3443  **/
3444 static bool e1000_clean_tx_irq(struct e1000_adapter *adapter,
3445                                struct e1000_tx_ring *tx_ring)
3446 {
3447         struct e1000_hw *hw = &adapter->hw;
3448         struct net_device *netdev = adapter->netdev;
3449         struct e1000_tx_desc *tx_desc, *eop_desc;
3450         struct e1000_buffer *buffer_info;
3451         unsigned int i, eop;
3452         unsigned int count = 0;
3453         unsigned int total_tx_bytes=0, total_tx_packets=0;
3454
3455         i = tx_ring->next_to_clean;
3456         eop = tx_ring->buffer_info[i].next_to_watch;
3457         eop_desc = E1000_TX_DESC(*tx_ring, eop);
3458
3459         while ((eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) &&
3460                (count < tx_ring->count)) {
3461                 bool cleaned = false;
3462                 for ( ; !cleaned; count++) {
3463                         tx_desc = E1000_TX_DESC(*tx_ring, i);
3464                         buffer_info = &tx_ring->buffer_info[i];
3465                         cleaned = (i == eop);
3466
3467                         if (cleaned) {
3468                                 struct sk_buff *skb = buffer_info->skb;
3469                                 unsigned int segs, bytecount;
3470                                 segs = skb_shinfo(skb)->gso_segs ?: 1;
3471                                 /* multiply data chunks by size of headers */
3472                                 bytecount = ((segs - 1) * skb_headlen(skb)) +
3473                                             skb->len;
3474                                 total_tx_packets += segs;
3475                                 total_tx_bytes += bytecount;
3476                         }
3477                         e1000_unmap_and_free_tx_resource(adapter, buffer_info);
3478                         tx_desc->upper.data = 0;
3479
3480                         if (unlikely(++i == tx_ring->count)) i = 0;
3481                 }
3482
3483                 eop = tx_ring->buffer_info[i].next_to_watch;
3484                 eop_desc = E1000_TX_DESC(*tx_ring, eop);
3485         }
3486
3487         tx_ring->next_to_clean = i;
3488
3489 #define TX_WAKE_THRESHOLD 32
3490         if (unlikely(count && netif_carrier_ok(netdev) &&
3491                      E1000_DESC_UNUSED(tx_ring) >= TX_WAKE_THRESHOLD)) {
3492                 /* Make sure that anybody stopping the queue after this
3493                  * sees the new next_to_clean.
3494                  */
3495                 smp_mb();
3496
3497                 if (netif_queue_stopped(netdev) &&
3498                     !(test_bit(__E1000_DOWN, &adapter->flags))) {
3499                         netif_wake_queue(netdev);
3500                         ++adapter->restart_queue;
3501                 }
3502         }
3503
3504         if (adapter->detect_tx_hung) {
3505                 /* Detect a transmit hang in hardware, this serializes the
3506                  * check with the clearing of time_stamp and movement of i */
3507                 adapter->detect_tx_hung = false;
3508                 if (tx_ring->buffer_info[eop].time_stamp &&
3509                     time_after(jiffies, tx_ring->buffer_info[eop].time_stamp +
3510                                (adapter->tx_timeout_factor * HZ)) &&
3511                     !(er32(STATUS) & E1000_STATUS_TXOFF)) {
3512
3513                         /* detected Tx unit hang */
3514                         DPRINTK(DRV, ERR, "Detected Tx Unit Hang\n"
3515                                         "  Tx Queue             <%lu>\n"
3516                                         "  TDH                  <%x>\n"
3517                                         "  TDT                  <%x>\n"
3518                                         "  next_to_use          <%x>\n"
3519                                         "  next_to_clean        <%x>\n"
3520                                         "buffer_info[next_to_clean]\n"
3521                                         "  time_stamp           <%lx>\n"
3522                                         "  next_to_watch        <%x>\n"
3523                                         "  jiffies              <%lx>\n"
3524                                         "  next_to_watch.status <%x>\n",
3525                                 (unsigned long)((tx_ring - adapter->tx_ring) /
3526                                         sizeof(struct e1000_tx_ring)),
3527                                 readl(hw->hw_addr + tx_ring->tdh),
3528                                 readl(hw->hw_addr + tx_ring->tdt),
3529                                 tx_ring->next_to_use,
3530                                 tx_ring->next_to_clean,
3531                                 tx_ring->buffer_info[eop].time_stamp,
3532                                 eop,
3533                                 jiffies,
3534                                 eop_desc->upper.fields.status);
3535                         netif_stop_queue(netdev);
3536                 }
3537         }
3538         adapter->total_tx_bytes += total_tx_bytes;
3539         adapter->total_tx_packets += total_tx_packets;
3540         netdev->stats.tx_bytes += total_tx_bytes;
3541         netdev->stats.tx_packets += total_tx_packets;
3542         return (count < tx_ring->count);
3543 }
3544
3545 /**
3546  * e1000_rx_checksum - Receive Checksum Offload for 82543
3547  * @adapter:     board private structure
3548  * @status_err:  receive descriptor status and error fields
3549  * @csum:        receive descriptor csum field
3550  * @sk_buff:     socket buffer with received data
3551  **/
3552
3553 static void e1000_rx_checksum(struct e1000_adapter *adapter, u32 status_err,
3554                               u32 csum, struct sk_buff *skb)
3555 {
3556         struct e1000_hw *hw = &adapter->hw;
3557         u16 status = (u16)status_err;
3558         u8 errors = (u8)(status_err >> 24);
3559         skb->ip_summed = CHECKSUM_NONE;
3560
3561         /* 82543 or newer only */
3562         if (unlikely(hw->mac_type < e1000_82543)) return;
3563         /* Ignore Checksum bit is set */
3564         if (unlikely(status & E1000_RXD_STAT_IXSM)) return;
3565         /* TCP/UDP checksum error bit is set */
3566         if (unlikely(errors & E1000_RXD_ERR_TCPE)) {
3567                 /* let the stack verify checksum errors */
3568                 adapter->hw_csum_err++;
3569                 return;
3570         }
3571         /* TCP/UDP Checksum has not been calculated */
3572         if (!(status & E1000_RXD_STAT_TCPCS))
3573                 return;
3574
3575         /* It must be a TCP or UDP packet with a valid checksum */
3576         if (likely(status & E1000_RXD_STAT_TCPCS)) {
3577                 /* TCP checksum is good */
3578                 skb->ip_summed = CHECKSUM_UNNECESSARY;
3579         }
3580         adapter->hw_csum_good++;
3581 }
3582
3583 /**
3584  * e1000_consume_page - helper function
3585  **/
3586 static void e1000_consume_page(struct e1000_buffer *bi, struct sk_buff *skb,
3587                                u16 length)
3588 {
3589         bi->page = NULL;
3590         skb->len += length;
3591         skb->data_len += length;
3592         skb->truesize += length;
3593 }
3594
3595 /**
3596  * e1000_receive_skb - helper function to handle rx indications
3597  * @adapter: board private structure
3598  * @status: descriptor status field as written by hardware
3599  * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
3600  * @skb: pointer to sk_buff to be indicated to stack
3601  */
3602 static void e1000_receive_skb(struct e1000_adapter *adapter, u8 status,
3603                               __le16 vlan, struct sk_buff *skb)
3604 {
3605         if (unlikely(adapter->vlgrp && (status & E1000_RXD_STAT_VP))) {
3606                 vlan_hwaccel_receive_skb(skb, adapter->vlgrp,
3607                                          le16_to_cpu(vlan) &
3608                                          E1000_RXD_SPC_VLAN_MASK);
3609         } else {
3610                 netif_receive_skb(skb);
3611         }
3612 }
3613
3614 /**
3615  * e1000_clean_jumbo_rx_irq - Send received data up the network stack; legacy
3616  * @adapter: board private structure
3617  * @rx_ring: ring to clean
3618  * @work_done: amount of napi work completed this call
3619  * @work_to_do: max amount of work allowed for this call to do
3620  *
3621  * the return value indicates whether actual cleaning was done, there
3622  * is no guarantee that everything was cleaned
3623  */
3624 static bool e1000_clean_jumbo_rx_irq(struct e1000_adapter *adapter,
3625                                      struct e1000_rx_ring *rx_ring,
3626                                      int *work_done, int work_to_do)
3627 {
3628         struct e1000_hw *hw = &adapter->hw;
3629         struct net_device *netdev = adapter->netdev;
3630         struct pci_dev *pdev = adapter->pdev;
3631         struct e1000_rx_desc *rx_desc, *next_rxd;
3632         struct e1000_buffer *buffer_info, *next_buffer;
3633         unsigned long irq_flags;
3634         u32 length;
3635         unsigned int i;
3636         int cleaned_count = 0;
3637         bool cleaned = false;
3638         unsigned int total_rx_bytes=0, total_rx_packets=0;
3639
3640         i = rx_ring->next_to_clean;
3641         rx_desc = E1000_RX_DESC(*rx_ring, i);
3642         buffer_info = &rx_ring->buffer_info[i];
3643
3644         while (rx_desc->status & E1000_RXD_STAT_DD) {
3645                 struct sk_buff *skb;
3646                 u8 status;
3647
3648                 if (*work_done >= work_to_do)
3649                         break;
3650                 (*work_done)++;
3651
3652                 status = rx_desc->status;
3653                 skb = buffer_info->skb;
3654                 buffer_info->skb = NULL;
3655
3656                 if (++i == rx_ring->count) i = 0;
3657                 next_rxd = E1000_RX_DESC(*rx_ring, i);
3658                 prefetch(next_rxd);
3659
3660                 next_buffer = &rx_ring->buffer_info[i];
3661
3662                 cleaned = true;
3663                 cleaned_count++;
3664                 pci_unmap_page(pdev, buffer_info->dma, buffer_info->length,
3665                                PCI_DMA_FROMDEVICE);
3666                 buffer_info->dma = 0;
3667
3668                 length = le16_to_cpu(rx_desc->length);
3669
3670                 /* errors is only valid for DD + EOP descriptors */
3671                 if (unlikely((status & E1000_RXD_STAT_EOP) &&
3672                     (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK))) {
3673                         u8 last_byte = *(skb->data + length - 1);
3674                         if (TBI_ACCEPT(hw, status, rx_desc->errors, length,
3675                                        last_byte)) {
3676                                 spin_lock_irqsave(&adapter->stats_lock,
3677                                                   irq_flags);
3678                                 e1000_tbi_adjust_stats(hw, &adapter->stats,
3679                                                        length, skb->data);
3680                                 spin_unlock_irqrestore(&adapter->stats_lock,
3681                                                        irq_flags);
3682                                 length--;
3683                         } else {
3684                                 /* recycle both page and skb */
3685                                 buffer_info->skb = skb;
3686                                 /* an error means any chain goes out the window
3687                                  * too */
3688                                 if (rx_ring->rx_skb_top)
3689                                         dev_kfree_skb(rx_ring->rx_skb_top);
3690                                 rx_ring->rx_skb_top = NULL;
3691                                 goto next_desc;
3692                         }
3693                 }
3694
3695 #define rxtop rx_ring->rx_skb_top
3696                 if (!(status & E1000_RXD_STAT_EOP)) {
3697                         /* this descriptor is only the beginning (or middle) */
3698                         if (!rxtop) {
3699                                 /* this is the beginning of a chain */
3700                                 rxtop = skb;
3701                                 skb_fill_page_desc(rxtop, 0, buffer_info->page,
3702                                                    0, length);
3703                         } else {
3704                                 /* this is the middle of a chain */
3705                                 skb_fill_page_desc(rxtop,
3706                                     skb_shinfo(rxtop)->nr_frags,
3707                                     buffer_info->page, 0, length);
3708                                 /* re-use the skb, only consumed the page */
3709                                 buffer_info->skb = skb;
3710                         }
3711                         e1000_consume_page(buffer_info, rxtop, length);
3712                         goto next_desc;
3713                 } else {
3714                         if (rxtop) {
3715                                 /* end of the chain */
3716                                 skb_fill_page_desc(rxtop,
3717                                     skb_shinfo(rxtop)->nr_frags,
3718                                     buffer_info->page, 0, length);
3719                                 /* re-use the current skb, we only consumed the
3720                                  * page */
3721                                 buffer_info->skb = skb;
3722                                 skb = rxtop;
3723                                 rxtop = NULL;
3724                                 e1000_consume_page(buffer_info, skb, length);
3725                         } else {
3726                                 /* no chain, got EOP, this buf is the packet
3727                                  * copybreak to save the put_page/alloc_page */
3728                                 if (length <= copybreak &&
3729                                     skb_tailroom(skb) >= length) {
3730                                         u8 *vaddr;
3731                                         vaddr = kmap_atomic(buffer_info->page,
3732                                                             KM_SKB_DATA_SOFTIRQ);
3733                                         memcpy(skb_tail_pointer(skb), vaddr, length);
3734                                         kunmap_atomic(vaddr,
3735                                                       KM_SKB_DATA_SOFTIRQ);
3736                                         /* re-use the page, so don't erase
3737                                          * buffer_info->page */
3738                                         skb_put(skb, length);
3739                                 } else {
3740                                         skb_fill_page_desc(skb, 0,
3741                                                            buffer_info->page, 0,
3742                                                            length);
3743                                         e1000_consume_page(buffer_info, skb,
3744                                                            length);
3745                                 }
3746                         }
3747                 }
3748
3749                 /* Receive Checksum Offload XXX recompute due to CRC strip? */
3750                 e1000_rx_checksum(adapter,
3751                                   (u32)(status) |
3752                                   ((u32)(rx_desc->errors) << 24),
3753                                   le16_to_cpu(rx_desc->csum), skb);
3754
3755                 pskb_trim(skb, skb->len - 4);
3756
3757                 /* probably a little skewed due to removing CRC */
3758                 total_rx_bytes += skb->len;
3759                 total_rx_packets++;
3760
3761                 /* eth type trans needs skb->data to point to something */
3762                 if (!pskb_may_pull(skb, ETH_HLEN)) {
3763                         DPRINTK(DRV, ERR, "pskb_may_pull failed.\n");
3764                         dev_kfree_skb(skb);
3765                         goto next_desc;
3766                 }
3767
3768                 skb->protocol = eth_type_trans(skb, netdev);
3769
3770                 e1000_receive_skb(adapter, status, rx_desc->special, skb);
3771
3772 next_desc:
3773                 rx_desc->status = 0;
3774
3775                 /* return some buffers to hardware, one at a time is too slow */
3776                 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
3777                         adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
3778                         cleaned_count = 0;
3779                 }
3780
3781                 /* use prefetched values */
3782                 rx_desc = next_rxd;
3783                 buffer_info = next_buffer;
3784         }
3785         rx_ring->next_to_clean = i;
3786
3787         cleaned_count = E1000_DESC_UNUSED(rx_ring);
3788         if (cleaned_count)
3789                 adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
3790
3791         adapter->total_rx_packets += total_rx_packets;
3792         adapter->total_rx_bytes += total_rx_bytes;
3793         netdev->stats.rx_bytes += total_rx_bytes;
3794         netdev->stats.rx_packets += total_rx_packets;
3795         return cleaned;
3796 }
3797
3798 /**
3799  * e1000_clean_rx_irq - Send received data up the network stack; legacy
3800  * @adapter: board private structure
3801  * @rx_ring: ring to clean
3802  * @work_done: amount of napi work completed this call
3803  * @work_to_do: max amount of work allowed for this call to do
3804  */
3805 static bool e1000_clean_rx_irq(struct e1000_adapter *adapter,
3806                                struct e1000_rx_ring *rx_ring,
3807                                int *work_done, int work_to_do)
3808 {
3809         struct e1000_hw *hw = &adapter->hw;
3810         struct net_device *netdev = adapter->netdev;
3811         struct pci_dev *pdev = adapter->pdev;
3812         struct e1000_rx_desc *rx_desc, *next_rxd;
3813         struct e1000_buffer *buffer_info, *next_buffer;
3814         unsigned long flags;
3815         u32 length;
3816         unsigned int i;
3817         int cleaned_count = 0;
3818         bool cleaned = false;
3819         unsigned int total_rx_bytes=0, total_rx_packets=0;
3820
3821         i = rx_ring->next_to_clean;
3822         rx_desc = E1000_RX_DESC(*rx_ring, i);
3823         buffer_info = &rx_ring->buffer_info[i];
3824
3825         while (rx_desc->status & E1000_RXD_STAT_DD) {
3826                 struct sk_buff *skb;
3827                 u8 status;
3828
3829                 if (*work_done >= work_to_do)
3830                         break;
3831                 (*work_done)++;
3832
3833                 status = rx_desc->status;
3834                 skb = buffer_info->skb;
3835                 buffer_info->skb = NULL;
3836
3837                 prefetch(skb->data - NET_IP_ALIGN);
3838
3839                 if (++i == rx_ring->count) i = 0;
3840                 next_rxd = E1000_RX_DESC(*rx_ring, i);
3841                 prefetch(next_rxd);
3842
3843                 next_buffer = &rx_ring->buffer_info[i];
3844
3845                 cleaned = true;
3846                 cleaned_count++;
3847                 pci_unmap_single(pdev, buffer_info->dma, buffer_info->length,
3848                                  PCI_DMA_FROMDEVICE);
3849                 buffer_info->dma = 0;
3850
3851                 length = le16_to_cpu(rx_desc->length);
3852                 /* !EOP means multiple descriptors were used to store a single
3853                  * packet, if thats the case we need to toss it.  In fact, we
3854                  * to toss every packet with the EOP bit clear and the next
3855                  * frame that _does_ have the EOP bit set, as it is by
3856                  * definition only a frame fragment
3857                  */
3858                 if (unlikely(!(status & E1000_RXD_STAT_EOP)))
3859                         adapter->discarding = true;
3860
3861                 if (adapter->discarding) {
3862                         /* All receives must fit into a single buffer */
3863                         E1000_DBG("%s: Receive packet consumed multiple"
3864                                   " buffers\n", netdev->name);
3865                         /* recycle */
3866                         buffer_info->skb = skb;
3867                         if (status & E1000_RXD_STAT_EOP)
3868                                 adapter->discarding = false;
3869                         goto next_desc;
3870                 }
3871
3872                 if (unlikely(rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK)) {
3873                         u8 last_byte = *(skb->data + length - 1);
3874                         if (TBI_ACCEPT(hw, status, rx_desc->errors, length,
3875                                        last_byte)) {
3876                                 spin_lock_irqsave(&adapter->stats_lock, flags);
3877                                 e1000_tbi_adjust_stats(hw, &adapter->stats,
3878                                                        length, skb->data);
3879                                 spin_unlock_irqrestore(&adapter->stats_lock,
3880                                                        flags);
3881                                 length--;
3882                         } else {
3883                                 /* recycle */
3884                                 buffer_info->skb = skb;
3885                                 goto next_desc;
3886                         }
3887                 }
3888
3889                 /* adjust length to remove Ethernet CRC, this must be
3890                  * done after the TBI_ACCEPT workaround above */
3891                 length -= 4;
3892
3893                 /* probably a little skewed due to removing CRC */
3894                 total_rx_bytes += length;
3895                 total_rx_packets++;
3896
3897                 /* code added for copybreak, this should improve
3898                  * performance for small packets with large amounts
3899                  * of reassembly being done in the stack */
3900                 if (length < copybreak) {
3901                         struct sk_buff *new_skb =
3902                             netdev_alloc_skb_ip_align(netdev, length);
3903                         if (new_skb) {
3904                                 skb_copy_to_linear_data_offset(new_skb,
3905                                                                -NET_IP_ALIGN,
3906                                                                (skb->data -
3907                                                                 NET_IP_ALIGN),
3908                                                                (length +
3909                                                                 NET_IP_ALIGN));
3910                                 /* save the skb in buffer_info as good */
3911                                 buffer_info->skb = skb;
3912                                 skb = new_skb;
3913                         }
3914                         /* else just continue with the old one */
3915                 }
3916                 /* end copybreak code */
3917                 skb_put(skb, length);
3918
3919                 /* Receive Checksum Offload */
3920                 e1000_rx_checksum(adapter,
3921                                   (u32)(status) |
3922                                   ((u32)(rx_desc->errors) << 24),
3923                                   le16_to_cpu(rx_desc->csum), skb);
3924
3925                 skb->protocol = eth_type_trans(skb, netdev);
3926
3927                 e1000_receive_skb(adapter, status, rx_desc->special, skb);
3928
3929 next_desc:
3930                 rx_desc->status = 0;
3931
3932                 /* return some buffers to hardware, one at a time is too slow */
3933                 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
3934                         adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
3935                         cleaned_count = 0;
3936                 }
3937
3938                 /* use prefetched values */
3939                 rx_desc = next_rxd;
3940                 buffer_info = next_buffer;
3941         }
3942         rx_ring->next_to_clean = i;
3943
3944         cleaned_count = E1000_DESC_UNUSED(rx_ring);
3945         if (cleaned_count)
3946                 adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
3947
3948         adapter->total_rx_packets += total_rx_packets;
3949         adapter->total_rx_bytes += total_rx_bytes;
3950         netdev->stats.rx_bytes += total_rx_bytes;
3951         netdev->stats.rx_packets += total_rx_packets;
3952         return cleaned;
3953 }
3954
3955 /**
3956  * e1000_alloc_jumbo_rx_buffers - Replace used jumbo receive buffers
3957  * @adapter: address of board private structure
3958  * @rx_ring: pointer to receive ring structure
3959  * @cleaned_count: number of buffers to allocate this pass
3960  **/
3961
3962 static void
3963 e1000_alloc_jumbo_rx_buffers(struct e1000_adapter *adapter,
3964                              struct e1000_rx_ring *rx_ring, int cleaned_count)
3965 {
3966         struct net_device *netdev = adapter->netdev;
3967         struct pci_dev *pdev = adapter->pdev;
3968         struct e1000_rx_desc *rx_desc;
3969         struct e1000_buffer *buffer_info;
3970         struct sk_buff *skb;
3971         unsigned int i;
3972         unsigned int bufsz = 256 - 16 /*for skb_reserve */ ;
3973
3974         i = rx_ring->next_to_use;
3975         buffer_info = &rx_ring->buffer_info[i];
3976
3977         while (cleaned_count--) {
3978                 skb = buffer_info->skb;
3979                 if (skb) {
3980                         skb_trim(skb, 0);
3981                         goto check_page;
3982                 }
3983
3984                 skb = netdev_alloc_skb_ip_align(netdev, bufsz);
3985                 if (unlikely(!skb)) {
3986                         /* Better luck next round */
3987                         adapter->alloc_rx_buff_failed++;
3988                         break;
3989                 }
3990
3991                 /* Fix for errata 23, can't cross 64kB boundary */
3992                 if (!e1000_check_64k_bound(adapter, skb->data, bufsz)) {
3993                         struct sk_buff *oldskb = skb;
3994                         DPRINTK(PROBE, ERR, "skb align check failed: %u bytes "
3995                                              "at %p\n", bufsz, skb->data);
3996                         /* Try again, without freeing the previous */
3997                         skb = netdev_alloc_skb_ip_align(netdev, bufsz);
3998                         /* Failed allocation, critical failure */
3999                         if (!skb) {
4000                                 dev_kfree_skb(oldskb);
4001                                 adapter->alloc_rx_buff_failed++;
4002                                 break;
4003                         }
4004
4005                         if (!e1000_check_64k_bound(adapter, skb->data, bufsz)) {
4006                                 /* give up */
4007                                 dev_kfree_skb(skb);
4008                                 dev_kfree_skb(oldskb);
4009                                 break; /* while (cleaned_count--) */
4010                         }
4011
4012                         /* Use new allocation */
4013                         dev_kfree_skb(oldskb);
4014                 }
4015                 buffer_info->skb = skb;
4016                 buffer_info->length = adapter->rx_buffer_len;
4017 check_page:
4018                 /* allocate a new page if necessary */
4019                 if (!buffer_info->page) {
4020                         buffer_info->page = alloc_page(GFP_ATOMIC);
4021                         if (unlikely(!buffer_info->page)) {
4022                                 adapter->alloc_rx_buff_failed++;
4023                                 break;
4024                         }
4025                 }
4026
4027                 if (!buffer_info->dma)
4028                         buffer_info->dma = pci_map_page(pdev,
4029                                                         buffer_info->page, 0,
4030                                                         buffer_info->length,
4031                                                         PCI_DMA_FROMDEVICE);
4032
4033                 rx_desc = E1000_RX_DESC(*rx_ring, i);
4034                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
4035
4036                 if (unlikely(++i == rx_ring->count))
4037                         i = 0;
4038                 buffer_info = &rx_ring->buffer_info[i];
4039         }
4040
4041         if (likely(rx_ring->next_to_use != i)) {
4042                 rx_ring->next_to_use = i;
4043                 if (unlikely(i-- == 0))
4044                         i = (rx_ring->count - 1);
4045
4046                 /* Force memory writes to complete before letting h/w
4047                  * know there are new descriptors to fetch.  (Only
4048                  * applicable for weak-ordered memory model archs,
4049                  * such as IA-64). */
4050                 wmb();
4051                 writel(i, adapter->hw.hw_addr + rx_ring->rdt);
4052         }
4053 }
4054
4055 /**
4056  * e1000_alloc_rx_buffers - Replace used receive buffers; legacy & extended
4057  * @adapter: address of board private structure
4058  **/
4059
4060 static void e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
4061                                    struct e1000_rx_ring *rx_ring,
4062                                    int cleaned_count)
4063 {
4064         struct e1000_hw *hw = &adapter->hw;
4065         struct net_device *netdev = adapter->netdev;
4066         struct pci_dev *pdev = adapter->pdev;
4067         struct e1000_rx_desc *rx_desc;
4068         struct e1000_buffer *buffer_info;
4069         struct sk_buff *skb;
4070         unsigned int i;
4071         unsigned int bufsz = adapter->rx_buffer_len;
4072
4073         i = rx_ring->next_to_use;
4074         buffer_info = &rx_ring->buffer_info[i];
4075
4076         while (cleaned_count--) {
4077                 skb = buffer_info->skb;
4078                 if (skb) {
4079                         skb_trim(skb, 0);
4080                         goto map_skb;
4081                 }
4082
4083                 skb = netdev_alloc_skb_ip_align(netdev, bufsz);
4084                 if (unlikely(!skb)) {
4085                         /* Better luck next round */
4086                         adapter->alloc_rx_buff_failed++;
4087                         break;
4088                 }
4089
4090                 /* Fix for errata 23, can't cross 64kB boundary */
4091                 if (!e1000_check_64k_bound(adapter, skb->data, bufsz)) {
4092                         struct sk_buff *oldskb = skb;
4093                         DPRINTK(RX_ERR, ERR, "skb align check failed: %u bytes "
4094                                              "at %p\n", bufsz, skb->data);
4095                         /* Try again, without freeing the previous */
4096                         skb = netdev_alloc_skb_ip_align(netdev, bufsz);
4097                         /* Failed allocation, critical failure */
4098                         if (!skb) {
4099                                 dev_kfree_skb(oldskb);
4100                                 adapter->alloc_rx_buff_failed++;
4101                                 break;
4102                         }
4103
4104                         if (!e1000_check_64k_bound(adapter, skb->data, bufsz)) {
4105                                 /* give up */
4106                                 dev_kfree_skb(skb);
4107                                 dev_kfree_skb(oldskb);
4108                                 adapter->alloc_rx_buff_failed++;
4109                                 break; /* while !buffer_info->skb */
4110                         }
4111
4112                         /* Use new allocation */
4113                         dev_kfree_skb(oldskb);
4114                 }
4115                 buffer_info->skb = skb;
4116                 buffer_info->length = adapter->rx_buffer_len;
4117 map_skb:
4118                 buffer_info->dma = pci_map_single(pdev,
4119                                                   skb->data,
4120                                                   buffer_info->length,
4121                                                   PCI_DMA_FROMDEVICE);
4122
4123                 /*
4124                  * XXX if it was allocated cleanly it will never map to a
4125                  * boundary crossing
4126                  */
4127
4128                 /* Fix for errata 23, can't cross 64kB boundary */
4129                 if (!e1000_check_64k_bound(adapter,
4130                                         (void *)(unsigned long)buffer_info->dma,
4131                                         adapter->rx_buffer_len)) {
4132                         DPRINTK(RX_ERR, ERR,
4133                                 "dma align check failed: %u bytes at %p\n",
4134                                 adapter->rx_buffer_len,
4135                                 (void *)(unsigned long)buffer_info->dma);
4136                         dev_kfree_skb(skb);
4137                         buffer_info->skb = NULL;
4138
4139                         pci_unmap_single(pdev, buffer_info->dma,
4140                                          adapter->rx_buffer_len,
4141                                          PCI_DMA_FROMDEVICE);
4142                         buffer_info->dma = 0;
4143
4144                         adapter->alloc_rx_buff_failed++;
4145                         break; /* while !buffer_info->skb */
4146                 }
4147                 rx_desc = E1000_RX_DESC(*rx_ring, i);
4148                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
4149
4150                 if (unlikely(++i == rx_ring->count))
4151                         i = 0;
4152                 buffer_info = &rx_ring->buffer_info[i];
4153         }
4154
4155         if (likely(rx_ring->next_to_use != i)) {
4156                 rx_ring->next_to_use = i;
4157                 if (unlikely(i-- == 0))
4158                         i = (rx_ring->count - 1);
4159
4160                 /* Force memory writes to complete before letting h/w
4161                  * know there are new descriptors to fetch.  (Only
4162                  * applicable for weak-ordered memory model archs,
4163                  * such as IA-64). */
4164                 wmb();
4165                 writel(i, hw->hw_addr + rx_ring->rdt);
4166         }
4167 }
4168
4169 /**
4170  * e1000_smartspeed - Workaround for SmartSpeed on 82541 and 82547 controllers.
4171  * @adapter:
4172  **/
4173
4174 static void e1000_smartspeed(struct e1000_adapter *adapter)
4175 {
4176         struct e1000_hw *hw = &adapter->hw;
4177         u16 phy_status;
4178         u16 phy_ctrl;
4179
4180         if ((hw->phy_type != e1000_phy_igp) || !hw->autoneg ||
4181            !(hw->autoneg_advertised & ADVERTISE_1000_FULL))
4182                 return;
4183
4184         if (adapter->smartspeed == 0) {
4185                 /* If Master/Slave config fault is asserted twice,
4186                  * we assume back-to-back */
4187                 e1000_read_phy_reg(hw, PHY_1000T_STATUS, &phy_status);
4188                 if (!(phy_status & SR_1000T_MS_CONFIG_FAULT)) return;
4189                 e1000_read_phy_reg(hw, PHY_1000T_STATUS, &phy_status);
4190                 if (!(phy_status & SR_1000T_MS_CONFIG_FAULT)) return;
4191                 e1000_read_phy_reg(hw, PHY_1000T_CTRL, &phy_ctrl);
4192                 if (phy_ctrl & CR_1000T_MS_ENABLE) {
4193                         phy_ctrl &= ~CR_1000T_MS_ENABLE;
4194                         e1000_write_phy_reg(hw, PHY_1000T_CTRL,
4195                                             phy_ctrl);
4196                         adapter->smartspeed++;
4197                         if (!e1000_phy_setup_autoneg(hw) &&
4198                            !e1000_read_phy_reg(hw, PHY_CTRL,
4199                                                &phy_ctrl)) {
4200                                 phy_ctrl |= (MII_CR_AUTO_NEG_EN |
4201                                              MII_CR_RESTART_AUTO_NEG);
4202                                 e1000_write_phy_reg(hw, PHY_CTRL,
4203                                                     phy_ctrl);
4204                         }
4205                 }
4206                 return;
4207         } else if (adapter->smartspeed == E1000_SMARTSPEED_DOWNSHIFT) {
4208                 /* If still no link, perhaps using 2/3 pair cable */
4209                 e1000_read_phy_reg(hw, PHY_1000T_CTRL, &phy_ctrl);
4210                 phy_ctrl |= CR_1000T_MS_ENABLE;
4211                 e1000_write_phy_reg(hw, PHY_1000T_CTRL, phy_ctrl);
4212                 if (!e1000_phy_setup_autoneg(hw) &&
4213                    !e1000_read_phy_reg(hw, PHY_CTRL, &phy_ctrl)) {
4214                         phy_ctrl |= (MII_CR_AUTO_NEG_EN |
4215                                      MII_CR_RESTART_AUTO_NEG);
4216                         e1000_write_phy_reg(hw, PHY_CTRL, phy_ctrl);
4217                 }
4218         }
4219         /* Restart process after E1000_SMARTSPEED_MAX iterations */
4220         if (adapter->smartspeed++ == E1000_SMARTSPEED_MAX)
4221                 adapter->smartspeed = 0;
4222 }
4223
4224 /**
4225  * e1000_ioctl -
4226  * @netdev:
4227  * @ifreq:
4228  * @cmd:
4229  **/
4230
4231 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
4232 {
4233         switch (cmd) {
4234         case SIOCGMIIPHY:
4235         case SIOCGMIIREG:
4236         case SIOCSMIIREG:
4237                 return e1000_mii_ioctl(netdev, ifr, cmd);
4238         default:
4239                 return -EOPNOTSUPP;
4240         }
4241 }
4242
4243 /**
4244  * e1000_mii_ioctl -
4245  * @netdev:
4246  * @ifreq:
4247  * @cmd:
4248  **/
4249
4250 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
4251                            int cmd)
4252 {
4253         struct e1000_adapter *adapter = netdev_priv(netdev);
4254         struct e1000_hw *hw = &adapter->hw;
4255         struct mii_ioctl_data *data = if_mii(ifr);
4256         int retval;
4257         u16 mii_reg;
4258         u16 spddplx;
4259         unsigned long flags;
4260
4261         if (hw->media_type != e1000_media_type_copper)
4262                 return -EOPNOTSUPP;
4263
4264         switch (cmd) {
4265         case SIOCGMIIPHY:
4266                 data->phy_id = hw->phy_addr;
4267                 break;
4268         case SIOCGMIIREG:
4269                 spin_lock_irqsave(&adapter->stats_lock, flags);
4270                 if (e1000_read_phy_reg(hw, data->reg_num & 0x1F,
4271                                    &data->val_out)) {
4272                         spin_unlock_irqrestore(&adapter->stats_lock, flags);
4273                         return -EIO;
4274                 }
4275                 spin_unlock_irqrestore(&adapter->stats_lock, flags);
4276                 break;
4277         case SIOCSMIIREG:
4278                 if (data->reg_num & ~(0x1F))
4279                         return -EFAULT;
4280                 mii_reg = data->val_in;
4281                 spin_lock_irqsave(&adapter->stats_lock, flags);
4282                 if (e1000_write_phy_reg(hw, data->reg_num,
4283                                         mii_reg)) {
4284                         spin_unlock_irqrestore(&adapter->stats_lock, flags);
4285                         return -EIO;
4286                 }
4287                 spin_unlock_irqrestore(&adapter->stats_lock, flags);
4288                 if (hw->media_type == e1000_media_type_copper) {
4289                         switch (data->reg_num) {
4290                         case PHY_CTRL:
4291                                 if (mii_reg & MII_CR_POWER_DOWN)
4292                                         break;
4293                                 if (mii_reg & MII_CR_AUTO_NEG_EN) {
4294                                         hw->autoneg = 1;
4295                                         hw->autoneg_advertised = 0x2F;
4296                                 } else {
4297                                         if (mii_reg & 0x40)
4298                                                 spddplx = SPEED_1000;
4299                                         else if (mii_reg & 0x2000)
4300                                                 spddplx = SPEED_100;
4301                                         else
4302                                                 spddplx = SPEED_10;
4303                                         spddplx += (mii_reg & 0x100)
4304                                                    ? DUPLEX_FULL :
4305                                                    DUPLEX_HALF;
4306                                         retval = e1000_set_spd_dplx(adapter,
4307                                                                     spddplx);
4308                                         if (retval)
4309                                                 return retval;
4310                                 }
4311                                 if (netif_running(adapter->netdev))
4312                                         e1000_reinit_locked(adapter);
4313                                 else
4314                                         e1000_reset(adapter);
4315                                 break;
4316                         case M88E1000_PHY_SPEC_CTRL:
4317                         case M88E1000_EXT_PHY_SPEC_CTRL:
4318                                 if (e1000_phy_reset(hw))
4319                                         return -EIO;
4320                                 break;
4321                         }
4322                 } else {
4323                         switch (data->reg_num) {
4324                         case PHY_CTRL:
4325                                 if (mii_reg & MII_CR_POWER_DOWN)
4326                                         break;
4327                                 if (netif_running(adapter->netdev))
4328                                         e1000_reinit_locked(adapter);
4329                                 else
4330                                         e1000_reset(adapter);
4331                                 break;
4332                         }
4333                 }
4334                 break;
4335         default:
4336                 return -EOPNOTSUPP;
4337         }
4338         return E1000_SUCCESS;
4339 }
4340
4341 void e1000_pci_set_mwi(struct e1000_hw *hw)
4342 {
4343         struct e1000_adapter *adapter = hw->back;
4344         int ret_val = pci_set_mwi(adapter->pdev);
4345
4346         if (ret_val)
4347                 DPRINTK(PROBE, ERR, "Error in setting MWI\n");
4348 }
4349
4350 void e1000_pci_clear_mwi(struct e1000_hw *hw)
4351 {
4352         struct e1000_adapter *adapter = hw->back;
4353
4354         pci_clear_mwi(adapter->pdev);
4355 }
4356
4357 int e1000_pcix_get_mmrbc(struct e1000_hw *hw)
4358 {
4359         struct e1000_adapter *adapter = hw->back;
4360         return pcix_get_mmrbc(adapter->pdev);
4361 }
4362
4363 void e1000_pcix_set_mmrbc(struct e1000_hw *hw, int mmrbc)
4364 {
4365         struct e1000_adapter *adapter = hw->back;
4366         pcix_set_mmrbc(adapter->pdev, mmrbc);
4367 }
4368
4369 void e1000_io_write(struct e1000_hw *hw, unsigned long port, u32 value)
4370 {
4371         outl(value, port);
4372 }
4373
4374 static void e1000_vlan_rx_register(struct net_device *netdev,
4375                                    struct vlan_group *grp)
4376 {
4377         struct e1000_adapter *adapter = netdev_priv(netdev);
4378         struct e1000_hw *hw = &adapter->hw;
4379         u32 ctrl, rctl;
4380
4381         if (!test_bit(__E1000_DOWN, &adapter->flags))
4382                 e1000_irq_disable(adapter);
4383         adapter->vlgrp = grp;
4384
4385         if (grp) {
4386                 /* enable VLAN tag insert/strip */
4387                 ctrl = er32(CTRL);
4388                 ctrl |= E1000_CTRL_VME;
4389                 ew32(CTRL, ctrl);
4390
4391                 /* enable VLAN receive filtering */
4392                 rctl = er32(RCTL);
4393                 rctl &= ~E1000_RCTL_CFIEN;
4394                 if (!(netdev->flags & IFF_PROMISC))
4395                         rctl |= E1000_RCTL_VFE;
4396                 ew32(RCTL, rctl);
4397                 e1000_update_mng_vlan(adapter);
4398         } else {
4399                 /* disable VLAN tag insert/strip */
4400                 ctrl = er32(CTRL);
4401                 ctrl &= ~E1000_CTRL_VME;
4402                 ew32(CTRL, ctrl);
4403
4404                 /* disable VLAN receive filtering */
4405                 rctl = er32(RCTL);
4406                 rctl &= ~E1000_RCTL_VFE;
4407                 ew32(RCTL, rctl);
4408
4409                 if (adapter->mng_vlan_id != (u16)E1000_MNG_VLAN_NONE) {
4410                         e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
4411                         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
4412                 }
4413         }
4414
4415         if (!test_bit(__E1000_DOWN, &adapter->flags))
4416                 e1000_irq_enable(adapter);
4417 }
4418
4419 static void e1000_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
4420 {
4421         struct e1000_adapter *adapter = netdev_priv(netdev);
4422         struct e1000_hw *hw = &adapter->hw;
4423         u32 vfta, index;
4424
4425         if ((hw->mng_cookie.status &
4426              E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT) &&
4427             (vid == adapter->mng_vlan_id))
4428                 return;
4429         /* add VID to filter table */
4430         index = (vid >> 5) & 0x7F;
4431         vfta = E1000_READ_REG_ARRAY(hw, VFTA, index);
4432         vfta |= (1 << (vid & 0x1F));
4433         e1000_write_vfta(hw, index, vfta);
4434 }
4435
4436 static void e1000_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
4437 {
4438         struct e1000_adapter *adapter = netdev_priv(netdev);
4439         struct e1000_hw *hw = &adapter->hw;
4440         u32 vfta, index;
4441
4442         if (!test_bit(__E1000_DOWN, &adapter->flags))
4443                 e1000_irq_disable(adapter);
4444         vlan_group_set_device(adapter->vlgrp, vid, NULL);
4445         if (!test_bit(__E1000_DOWN, &adapter->flags))
4446                 e1000_irq_enable(adapter);
4447
4448         /* remove VID from filter table */
4449         index = (vid >> 5) & 0x7F;
4450         vfta = E1000_READ_REG_ARRAY(hw, VFTA, index);
4451         vfta &= ~(1 << (vid & 0x1F));
4452         e1000_write_vfta(hw, index, vfta);
4453 }
4454
4455 static void e1000_restore_vlan(struct e1000_adapter *adapter)
4456 {
4457         e1000_vlan_rx_register(adapter->netdev, adapter->vlgrp);
4458
4459         if (adapter->vlgrp) {
4460                 u16 vid;
4461                 for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
4462                         if (!vlan_group_get_device(adapter->vlgrp, vid))
4463                                 continue;
4464                         e1000_vlan_rx_add_vid(adapter->netdev, vid);
4465                 }
4466         }
4467 }
4468
4469 int e1000_set_spd_dplx(struct e1000_adapter *adapter, u16 spddplx)
4470 {
4471         struct e1000_hw *hw = &adapter->hw;
4472
4473         hw->autoneg = 0;
4474
4475         /* Fiber NICs only allow 1000 gbps Full duplex */
4476         if ((hw->media_type == e1000_media_type_fiber) &&
4477                 spddplx != (SPEED_1000 + DUPLEX_FULL)) {
4478                 DPRINTK(PROBE, ERR, "Unsupported Speed/Duplex configuration\n");
4479                 return -EINVAL;
4480         }
4481
4482         switch (spddplx) {
4483         case SPEED_10 + DUPLEX_HALF:
4484                 hw->forced_speed_duplex = e1000_10_half;
4485                 break;
4486         case SPEED_10 + DUPLEX_FULL:
4487                 hw->forced_speed_duplex = e1000_10_full;
4488                 break;
4489         case SPEED_100 + DUPLEX_HALF:
4490                 hw->forced_speed_duplex = e1000_100_half;
4491                 break;
4492         case SPEED_100 + DUPLEX_FULL:
4493                 hw->forced_speed_duplex = e1000_100_full;
4494                 break;
4495         case SPEED_1000 + DUPLEX_FULL:
4496                 hw->autoneg = 1;
4497                 hw->autoneg_advertised = ADVERTISE_1000_FULL;
4498                 break;
4499         case SPEED_1000 + DUPLEX_HALF: /* not supported */
4500         default:
4501                 DPRINTK(PROBE, ERR, "Unsupported Speed/Duplex configuration\n");
4502                 return -EINVAL;
4503         }
4504         return 0;
4505 }
4506
4507 static int __e1000_shutdown(struct pci_dev *pdev, bool *enable_wake)
4508 {
4509         struct net_device *netdev = pci_get_drvdata(pdev);
4510         struct e1000_adapter *adapter = netdev_priv(netdev);
4511         struct e1000_hw *hw = &adapter->hw;
4512         u32 ctrl, ctrl_ext, rctl, status;
4513         u32 wufc = adapter->wol;
4514 #ifdef CONFIG_PM
4515         int retval = 0;
4516 #endif
4517
4518         netif_device_detach(netdev);
4519
4520         if (netif_running(netdev)) {
4521                 WARN_ON(test_bit(__E1000_RESETTING, &adapter->flags));
4522                 e1000_down(adapter);
4523         }
4524
4525 #ifdef CONFIG_PM
4526         retval = pci_save_state(pdev);
4527         if (retval)
4528                 return retval;
4529 #endif
4530
4531         status = er32(STATUS);
4532         if (status & E1000_STATUS_LU)
4533                 wufc &= ~E1000_WUFC_LNKC;
4534
4535         if (wufc) {
4536                 e1000_setup_rctl(adapter);
4537                 e1000_set_rx_mode(netdev);
4538
4539                 /* turn on all-multi mode if wake on multicast is enabled */
4540                 if (wufc & E1000_WUFC_MC) {
4541                         rctl = er32(RCTL);
4542                         rctl |= E1000_RCTL_MPE;
4543                         ew32(RCTL, rctl);
4544                 }
4545
4546                 if (hw->mac_type >= e1000_82540) {
4547                         ctrl = er32(CTRL);
4548                         /* advertise wake from D3Cold */
4549                         #define E1000_CTRL_ADVD3WUC 0x00100000
4550                         /* phy power management enable */
4551                         #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
4552                         ctrl |= E1000_CTRL_ADVD3WUC |
4553                                 E1000_CTRL_EN_PHY_PWR_MGMT;
4554                         ew32(CTRL, ctrl);
4555                 }
4556
4557                 if (hw->media_type == e1000_media_type_fiber ||
4558                     hw->media_type == e1000_media_type_internal_serdes) {
4559                         /* keep the laser running in D3 */
4560                         ctrl_ext = er32(CTRL_EXT);
4561                         ctrl_ext |= E1000_CTRL_EXT_SDP7_DATA;
4562                         ew32(CTRL_EXT, ctrl_ext);
4563                 }
4564
4565                 ew32(WUC, E1000_WUC_PME_EN);
4566                 ew32(WUFC, wufc);
4567         } else {
4568                 ew32(WUC, 0);
4569                 ew32(WUFC, 0);
4570         }
4571
4572         e1000_release_manageability(adapter);
4573
4574         *enable_wake = !!wufc;
4575
4576         /* make sure adapter isn't asleep if manageability is enabled */
4577         if (adapter->en_mng_pt)
4578                 *enable_wake = true;
4579
4580         if (netif_running(netdev))
4581                 e1000_free_irq(adapter);
4582
4583         pci_disable_device(pdev);
4584
4585         return 0;
4586 }
4587
4588 #ifdef CONFIG_PM
4589 static int e1000_suspend(struct pci_dev *pdev, pm_message_t state)
4590 {
4591         int retval;
4592         bool wake;
4593
4594         retval = __e1000_shutdown(pdev, &wake);
4595         if (retval)
4596                 return retval;
4597
4598         if (wake) {
4599                 pci_prepare_to_sleep(pdev);
4600         } else {
4601                 pci_wake_from_d3(pdev, false);
4602                 pci_set_power_state(pdev, PCI_D3hot);
4603         }
4604
4605         return 0;
4606 }
4607
4608 static int e1000_resume(struct pci_dev *pdev)
4609 {
4610         struct net_device *netdev = pci_get_drvdata(pdev);
4611         struct e1000_adapter *adapter = netdev_priv(netdev);
4612         struct e1000_hw *hw = &adapter->hw;
4613         u32 err;
4614
4615         pci_set_power_state(pdev, PCI_D0);
4616         pci_restore_state(pdev);
4617
4618         if (adapter->need_ioport)
4619                 err = pci_enable_device(pdev);
4620         else
4621                 err = pci_enable_device_mem(pdev);
4622         if (err) {
4623                 printk(KERN_ERR "e1000: Cannot enable PCI device from suspend\n");
4624                 return err;
4625         }
4626         pci_set_master(pdev);
4627
4628         pci_enable_wake(pdev, PCI_D3hot, 0);
4629         pci_enable_wake(pdev, PCI_D3cold, 0);
4630
4631         if (netif_running(netdev)) {
4632                 err = e1000_request_irq(adapter);
4633                 if (err)
4634                         return err;
4635         }
4636
4637         e1000_power_up_phy(adapter);
4638         e1000_reset(adapter);
4639         ew32(WUS, ~0);
4640
4641         e1000_init_manageability(adapter);
4642
4643         if (netif_running(netdev))
4644                 e1000_up(adapter);
4645
4646         netif_device_attach(netdev);
4647
4648         return 0;
4649 }
4650 #endif
4651
4652 static void e1000_shutdown(struct pci_dev *pdev)
4653 {
4654         bool wake;
4655
4656         __e1000_shutdown(pdev, &wake);
4657
4658         if (system_state == SYSTEM_POWER_OFF) {
4659                 pci_wake_from_d3(pdev, wake);
4660                 pci_set_power_state(pdev, PCI_D3hot);
4661         }
4662 }
4663
4664 #ifdef CONFIG_NET_POLL_CONTROLLER
4665 /*
4666  * Polling 'interrupt' - used by things like netconsole to send skbs
4667  * without having to re-enable interrupts. It's not called while
4668  * the interrupt routine is executing.
4669  */
4670 static void e1000_netpoll(struct net_device *netdev)
4671 {
4672         struct e1000_adapter *adapter = netdev_priv(netdev);
4673
4674         disable_irq(adapter->pdev->irq);
4675         e1000_intr(adapter->pdev->irq, netdev);
4676         enable_irq(adapter->pdev->irq);
4677 }
4678 #endif
4679
4680 /**
4681  * e1000_io_error_detected - called when PCI error is detected
4682  * @pdev: Pointer to PCI device
4683  * @state: The current pci connection state
4684  *
4685  * This function is called after a PCI bus error affecting
4686  * this device has been detected.
4687  */
4688 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
4689                                                 pci_channel_state_t state)
4690 {
4691         struct net_device *netdev = pci_get_drvdata(pdev);
4692         struct e1000_adapter *adapter = netdev_priv(netdev);
4693
4694         netif_device_detach(netdev);
4695
4696         if (state == pci_channel_io_perm_failure)
4697                 return PCI_ERS_RESULT_DISCONNECT;
4698
4699         if (netif_running(netdev))
4700                 e1000_down(adapter);
4701         pci_disable_device(pdev);
4702
4703         /* Request a slot slot reset. */
4704         return PCI_ERS_RESULT_NEED_RESET;
4705 }
4706
4707 /**
4708  * e1000_io_slot_reset - called after the pci bus has been reset.
4709  * @pdev: Pointer to PCI device
4710  *
4711  * Restart the card from scratch, as if from a cold-boot. Implementation
4712  * resembles the first-half of the e1000_resume routine.
4713  */
4714 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
4715 {
4716         struct net_device *netdev = pci_get_drvdata(pdev);
4717         struct e1000_adapter *adapter = netdev_priv(netdev);
4718         struct e1000_hw *hw = &adapter->hw;
4719         int err;
4720
4721         if (adapter->need_ioport)
4722                 err = pci_enable_device(pdev);
4723         else
4724                 err = pci_enable_device_mem(pdev);
4725         if (err) {
4726                 printk(KERN_ERR "e1000: Cannot re-enable PCI device after reset.\n");
4727                 return PCI_ERS_RESULT_DISCONNECT;
4728         }
4729         pci_set_master(pdev);
4730
4731         pci_enable_wake(pdev, PCI_D3hot, 0);
4732         pci_enable_wake(pdev, PCI_D3cold, 0);
4733
4734         e1000_reset(adapter);
4735         ew32(WUS, ~0);
4736
4737         return PCI_ERS_RESULT_RECOVERED;
4738 }
4739
4740 /**
4741  * e1000_io_resume - called when traffic can start flowing again.
4742  * @pdev: Pointer to PCI device
4743  *
4744  * This callback is called when the error recovery driver tells us that
4745  * its OK to resume normal operation. Implementation resembles the
4746  * second-half of the e1000_resume routine.
4747  */
4748 static void e1000_io_resume(struct pci_dev *pdev)
4749 {
4750         struct net_device *netdev = pci_get_drvdata(pdev);
4751         struct e1000_adapter *adapter = netdev_priv(netdev);
4752
4753         e1000_init_manageability(adapter);
4754
4755         if (netif_running(netdev)) {
4756                 if (e1000_up(adapter)) {
4757                         printk("e1000: can't bring device back up after reset\n");
4758                         return;
4759                 }
4760         }
4761
4762         netif_device_attach(netdev);
4763 }
4764
4765 /* e1000_main.c */