cad6f65fc1e92c1a63ee9dd912da297371bd6779
[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         /* can be netif_tx_disable when NETIF_F_LLTX is removed */
489         netif_stop_queue(netdev);
490
491         /* disable transmits in the hardware */
492         tctl = er32(TCTL);
493         tctl &= ~E1000_TCTL_EN;
494         ew32(TCTL, tctl);
495         /* flush both disables and wait for them to finish */
496         E1000_WRITE_FLUSH();
497         msleep(10);
498
499         napi_disable(&adapter->napi);
500
501         e1000_irq_disable(adapter);
502
503         del_timer_sync(&adapter->tx_fifo_stall_timer);
504         del_timer_sync(&adapter->watchdog_timer);
505         del_timer_sync(&adapter->phy_info_timer);
506
507         netdev->tx_queue_len = adapter->tx_queue_len;
508         adapter->link_speed = 0;
509         adapter->link_duplex = 0;
510         netif_carrier_off(netdev);
511
512         e1000_reset(adapter);
513         e1000_clean_all_tx_rings(adapter);
514         e1000_clean_all_rx_rings(adapter);
515 }
516
517 void e1000_reinit_locked(struct e1000_adapter *adapter)
518 {
519         WARN_ON(in_interrupt());
520         while (test_and_set_bit(__E1000_RESETTING, &adapter->flags))
521                 msleep(1);
522         e1000_down(adapter);
523         e1000_up(adapter);
524         clear_bit(__E1000_RESETTING, &adapter->flags);
525 }
526
527 void e1000_reset(struct e1000_adapter *adapter)
528 {
529         struct e1000_hw *hw = &adapter->hw;
530         u32 pba = 0, tx_space, min_tx_space, min_rx_space;
531         bool legacy_pba_adjust = false;
532         u16 hwm;
533
534         /* Repartition Pba for greater than 9k mtu
535          * To take effect CTRL.RST is required.
536          */
537
538         switch (hw->mac_type) {
539         case e1000_82542_rev2_0:
540         case e1000_82542_rev2_1:
541         case e1000_82543:
542         case e1000_82544:
543         case e1000_82540:
544         case e1000_82541:
545         case e1000_82541_rev_2:
546                 legacy_pba_adjust = true;
547                 pba = E1000_PBA_48K;
548                 break;
549         case e1000_82545:
550         case e1000_82545_rev_3:
551         case e1000_82546:
552         case e1000_82546_rev_3:
553                 pba = E1000_PBA_48K;
554                 break;
555         case e1000_82547:
556         case e1000_82547_rev_2:
557                 legacy_pba_adjust = true;
558                 pba = E1000_PBA_30K;
559                 break;
560         case e1000_undefined:
561         case e1000_num_macs:
562                 break;
563         }
564
565         if (legacy_pba_adjust) {
566                 if (hw->max_frame_size > E1000_RXBUFFER_8192)
567                         pba -= 8; /* allocate more FIFO for Tx */
568
569                 if (hw->mac_type == e1000_82547) {
570                         adapter->tx_fifo_head = 0;
571                         adapter->tx_head_addr = pba << E1000_TX_HEAD_ADDR_SHIFT;
572                         adapter->tx_fifo_size =
573                                 (E1000_PBA_40K - pba) << E1000_PBA_BYTES_SHIFT;
574                         atomic_set(&adapter->tx_fifo_stall, 0);
575                 }
576         } else if (hw->max_frame_size >  ETH_FRAME_LEN + ETH_FCS_LEN) {
577                 /* adjust PBA for jumbo frames */
578                 ew32(PBA, pba);
579
580                 /* To maintain wire speed transmits, the Tx FIFO should be
581                  * large enough to accommodate two full transmit packets,
582                  * rounded up to the next 1KB and expressed in KB.  Likewise,
583                  * the Rx FIFO should be large enough to accommodate at least
584                  * one full receive packet and is similarly rounded up and
585                  * expressed in KB. */
586                 pba = er32(PBA);
587                 /* upper 16 bits has Tx packet buffer allocation size in KB */
588                 tx_space = pba >> 16;
589                 /* lower 16 bits has Rx packet buffer allocation size in KB */
590                 pba &= 0xffff;
591                 /*
592                  * the tx fifo also stores 16 bytes of information about the tx
593                  * but don't include ethernet FCS because hardware appends it
594                  */
595                 min_tx_space = (hw->max_frame_size +
596                                 sizeof(struct e1000_tx_desc) -
597                                 ETH_FCS_LEN) * 2;
598                 min_tx_space = ALIGN(min_tx_space, 1024);
599                 min_tx_space >>= 10;
600                 /* software strips receive CRC, so leave room for it */
601                 min_rx_space = hw->max_frame_size;
602                 min_rx_space = ALIGN(min_rx_space, 1024);
603                 min_rx_space >>= 10;
604
605                 /* If current Tx allocation is less than the min Tx FIFO size,
606                  * and the min Tx FIFO size is less than the current Rx FIFO
607                  * allocation, take space away from current Rx allocation */
608                 if (tx_space < min_tx_space &&
609                     ((min_tx_space - tx_space) < pba)) {
610                         pba = pba - (min_tx_space - tx_space);
611
612                         /* PCI/PCIx hardware has PBA alignment constraints */
613                         switch (hw->mac_type) {
614                         case e1000_82545 ... e1000_82546_rev_3:
615                                 pba &= ~(E1000_PBA_8K - 1);
616                                 break;
617                         default:
618                                 break;
619                         }
620
621                         /* if short on rx space, rx wins and must trump tx
622                          * adjustment or use Early Receive if available */
623                         if (pba < min_rx_space)
624                                 pba = min_rx_space;
625                 }
626         }
627
628         ew32(PBA, pba);
629
630         /*
631          * flow control settings:
632          * The high water mark must be low enough to fit one full frame
633          * (or the size used for early receive) above it in the Rx FIFO.
634          * Set it to the lower of:
635          * - 90% of the Rx FIFO size, and
636          * - the full Rx FIFO size minus the early receive size (for parts
637          *   with ERT support assuming ERT set to E1000_ERT_2048), or
638          * - the full Rx FIFO size minus one full frame
639          */
640         hwm = min(((pba << 10) * 9 / 10),
641                   ((pba << 10) - hw->max_frame_size));
642
643         hw->fc_high_water = hwm & 0xFFF8;       /* 8-byte granularity */
644         hw->fc_low_water = hw->fc_high_water - 8;
645         hw->fc_pause_time = E1000_FC_PAUSE_TIME;
646         hw->fc_send_xon = 1;
647         hw->fc = hw->original_fc;
648
649         /* Allow time for pending master requests to run */
650         e1000_reset_hw(hw);
651         if (hw->mac_type >= e1000_82544)
652                 ew32(WUC, 0);
653
654         if (e1000_init_hw(hw))
655                 DPRINTK(PROBE, ERR, "Hardware Error\n");
656         e1000_update_mng_vlan(adapter);
657
658         /* if (adapter->hwflags & HWFLAGS_PHY_PWR_BIT) { */
659         if (hw->mac_type >= e1000_82544 &&
660             hw->autoneg == 1 &&
661             hw->autoneg_advertised == ADVERTISE_1000_FULL) {
662                 u32 ctrl = er32(CTRL);
663                 /* clear phy power management bit if we are in gig only mode,
664                  * which if enabled will attempt negotiation to 100Mb, which
665                  * can cause a loss of link at power off or driver unload */
666                 ctrl &= ~E1000_CTRL_SWDPIN3;
667                 ew32(CTRL, ctrl);
668         }
669
670         /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
671         ew32(VET, ETHERNET_IEEE_VLAN_TYPE);
672
673         e1000_reset_adaptive(hw);
674         e1000_phy_get_info(hw, &adapter->phy_info);
675
676         e1000_release_manageability(adapter);
677 }
678
679 /**
680  *  Dump the eeprom for users having checksum issues
681  **/
682 static void e1000_dump_eeprom(struct e1000_adapter *adapter)
683 {
684         struct net_device *netdev = adapter->netdev;
685         struct ethtool_eeprom eeprom;
686         const struct ethtool_ops *ops = netdev->ethtool_ops;
687         u8 *data;
688         int i;
689         u16 csum_old, csum_new = 0;
690
691         eeprom.len = ops->get_eeprom_len(netdev);
692         eeprom.offset = 0;
693
694         data = kmalloc(eeprom.len, GFP_KERNEL);
695         if (!data) {
696                 printk(KERN_ERR "Unable to allocate memory to dump EEPROM"
697                        " data\n");
698                 return;
699         }
700
701         ops->get_eeprom(netdev, &eeprom, data);
702
703         csum_old = (data[EEPROM_CHECKSUM_REG * 2]) +
704                    (data[EEPROM_CHECKSUM_REG * 2 + 1] << 8);
705         for (i = 0; i < EEPROM_CHECKSUM_REG * 2; i += 2)
706                 csum_new += data[i] + (data[i + 1] << 8);
707         csum_new = EEPROM_SUM - csum_new;
708
709         printk(KERN_ERR "/*********************/\n");
710         printk(KERN_ERR "Current EEPROM Checksum : 0x%04x\n", csum_old);
711         printk(KERN_ERR "Calculated              : 0x%04x\n", csum_new);
712
713         printk(KERN_ERR "Offset    Values\n");
714         printk(KERN_ERR "========  ======\n");
715         print_hex_dump(KERN_ERR, "", DUMP_PREFIX_OFFSET, 16, 1, data, 128, 0);
716
717         printk(KERN_ERR "Include this output when contacting your support "
718                "provider.\n");
719         printk(KERN_ERR "This is not a software error! Something bad "
720                "happened to your hardware or\n");
721         printk(KERN_ERR "EEPROM image. Ignoring this "
722                "problem could result in further problems,\n");
723         printk(KERN_ERR "possibly loss of data, corruption or system hangs!\n");
724         printk(KERN_ERR "The MAC Address will be reset to 00:00:00:00:00:00, "
725                "which is invalid\n");
726         printk(KERN_ERR "and requires you to set the proper MAC "
727                "address manually before continuing\n");
728         printk(KERN_ERR "to enable this network device.\n");
729         printk(KERN_ERR "Please inspect the EEPROM dump and report the issue "
730                "to your hardware vendor\n");
731         printk(KERN_ERR "or Intel Customer Support.\n");
732         printk(KERN_ERR "/*********************/\n");
733
734         kfree(data);
735 }
736
737 /**
738  * e1000_is_need_ioport - determine if an adapter needs ioport resources or not
739  * @pdev: PCI device information struct
740  *
741  * Return true if an adapter needs ioport resources
742  **/
743 static int e1000_is_need_ioport(struct pci_dev *pdev)
744 {
745         switch (pdev->device) {
746         case E1000_DEV_ID_82540EM:
747         case E1000_DEV_ID_82540EM_LOM:
748         case E1000_DEV_ID_82540EP:
749         case E1000_DEV_ID_82540EP_LOM:
750         case E1000_DEV_ID_82540EP_LP:
751         case E1000_DEV_ID_82541EI:
752         case E1000_DEV_ID_82541EI_MOBILE:
753         case E1000_DEV_ID_82541ER:
754         case E1000_DEV_ID_82541ER_LOM:
755         case E1000_DEV_ID_82541GI:
756         case E1000_DEV_ID_82541GI_LF:
757         case E1000_DEV_ID_82541GI_MOBILE:
758         case E1000_DEV_ID_82544EI_COPPER:
759         case E1000_DEV_ID_82544EI_FIBER:
760         case E1000_DEV_ID_82544GC_COPPER:
761         case E1000_DEV_ID_82544GC_LOM:
762         case E1000_DEV_ID_82545EM_COPPER:
763         case E1000_DEV_ID_82545EM_FIBER:
764         case E1000_DEV_ID_82546EB_COPPER:
765         case E1000_DEV_ID_82546EB_FIBER:
766         case E1000_DEV_ID_82546EB_QUAD_COPPER:
767                 return true;
768         default:
769                 return false;
770         }
771 }
772
773 static const struct net_device_ops e1000_netdev_ops = {
774         .ndo_open               = e1000_open,
775         .ndo_stop               = e1000_close,
776         .ndo_start_xmit         = e1000_xmit_frame,
777         .ndo_get_stats          = e1000_get_stats,
778         .ndo_set_rx_mode        = e1000_set_rx_mode,
779         .ndo_set_mac_address    = e1000_set_mac,
780         .ndo_tx_timeout         = e1000_tx_timeout,
781         .ndo_change_mtu         = e1000_change_mtu,
782         .ndo_do_ioctl           = e1000_ioctl,
783         .ndo_validate_addr      = eth_validate_addr,
784
785         .ndo_vlan_rx_register   = e1000_vlan_rx_register,
786         .ndo_vlan_rx_add_vid    = e1000_vlan_rx_add_vid,
787         .ndo_vlan_rx_kill_vid   = e1000_vlan_rx_kill_vid,
788 #ifdef CONFIG_NET_POLL_CONTROLLER
789         .ndo_poll_controller    = e1000_netpoll,
790 #endif
791 };
792
793 /**
794  * e1000_probe - Device Initialization Routine
795  * @pdev: PCI device information struct
796  * @ent: entry in e1000_pci_tbl
797  *
798  * Returns 0 on success, negative on failure
799  *
800  * e1000_probe initializes an adapter identified by a pci_dev structure.
801  * The OS initialization, configuring of the adapter private structure,
802  * and a hardware reset occur.
803  **/
804 static int __devinit e1000_probe(struct pci_dev *pdev,
805                                  const struct pci_device_id *ent)
806 {
807         struct net_device *netdev;
808         struct e1000_adapter *adapter;
809         struct e1000_hw *hw;
810
811         static int cards_found = 0;
812         static int global_quad_port_a = 0; /* global ksp3 port a indication */
813         int i, err, pci_using_dac;
814         u16 eeprom_data = 0;
815         u16 eeprom_apme_mask = E1000_EEPROM_APME;
816         int bars, need_ioport;
817
818         /* do not allocate ioport bars when not needed */
819         need_ioport = e1000_is_need_ioport(pdev);
820         if (need_ioport) {
821                 bars = pci_select_bars(pdev, IORESOURCE_MEM | IORESOURCE_IO);
822                 err = pci_enable_device(pdev);
823         } else {
824                 bars = pci_select_bars(pdev, IORESOURCE_MEM);
825                 err = pci_enable_device_mem(pdev);
826         }
827         if (err)
828                 return err;
829
830         if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(64)) &&
831             !pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64))) {
832                 pci_using_dac = 1;
833         } else {
834                 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
835                 if (err) {
836                         err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
837                         if (err) {
838                                 E1000_ERR("No usable DMA configuration, "
839                                           "aborting\n");
840                                 goto err_dma;
841                         }
842                 }
843                 pci_using_dac = 0;
844         }
845
846         err = pci_request_selected_regions(pdev, bars, e1000_driver_name);
847         if (err)
848                 goto err_pci_reg;
849
850         pci_set_master(pdev);
851
852         err = -ENOMEM;
853         netdev = alloc_etherdev(sizeof(struct e1000_adapter));
854         if (!netdev)
855                 goto err_alloc_etherdev;
856
857         SET_NETDEV_DEV(netdev, &pdev->dev);
858
859         pci_set_drvdata(pdev, netdev);
860         adapter = netdev_priv(netdev);
861         adapter->netdev = netdev;
862         adapter->pdev = pdev;
863         adapter->msg_enable = (1 << debug) - 1;
864         adapter->bars = bars;
865         adapter->need_ioport = need_ioport;
866
867         hw = &adapter->hw;
868         hw->back = adapter;
869
870         err = -EIO;
871         hw->hw_addr = pci_ioremap_bar(pdev, BAR_0);
872         if (!hw->hw_addr)
873                 goto err_ioremap;
874
875         if (adapter->need_ioport) {
876                 for (i = BAR_1; i <= BAR_5; i++) {
877                         if (pci_resource_len(pdev, i) == 0)
878                                 continue;
879                         if (pci_resource_flags(pdev, i) & IORESOURCE_IO) {
880                                 hw->io_base = pci_resource_start(pdev, i);
881                                 break;
882                         }
883                 }
884         }
885
886         netdev->netdev_ops = &e1000_netdev_ops;
887         e1000_set_ethtool_ops(netdev);
888         netdev->watchdog_timeo = 5 * HZ;
889         netif_napi_add(netdev, &adapter->napi, e1000_clean, 64);
890
891         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
892
893         adapter->bd_number = cards_found;
894
895         /* setup the private structure */
896
897         err = e1000_sw_init(adapter);
898         if (err)
899                 goto err_sw_init;
900
901         err = -EIO;
902
903         if (hw->mac_type >= e1000_82543) {
904                 netdev->features = NETIF_F_SG |
905                                    NETIF_F_HW_CSUM |
906                                    NETIF_F_HW_VLAN_TX |
907                                    NETIF_F_HW_VLAN_RX |
908                                    NETIF_F_HW_VLAN_FILTER;
909         }
910
911         if ((hw->mac_type >= e1000_82544) &&
912            (hw->mac_type != e1000_82547))
913                 netdev->features |= NETIF_F_TSO;
914
915         if (pci_using_dac)
916                 netdev->features |= NETIF_F_HIGHDMA;
917
918         netdev->vlan_features |= NETIF_F_TSO;
919         netdev->vlan_features |= NETIF_F_HW_CSUM;
920         netdev->vlan_features |= NETIF_F_SG;
921
922         adapter->en_mng_pt = e1000_enable_mng_pass_thru(hw);
923
924         /* initialize eeprom parameters */
925         if (e1000_init_eeprom_params(hw)) {
926                 E1000_ERR("EEPROM initialization failed\n");
927                 goto err_eeprom;
928         }
929
930         /* before reading the EEPROM, reset the controller to
931          * put the device in a known good starting state */
932
933         e1000_reset_hw(hw);
934
935         /* make sure the EEPROM is good */
936         if (e1000_validate_eeprom_checksum(hw) < 0) {
937                 DPRINTK(PROBE, ERR, "The EEPROM Checksum Is Not Valid\n");
938                 e1000_dump_eeprom(adapter);
939                 /*
940                  * set MAC address to all zeroes to invalidate and temporary
941                  * disable this device for the user. This blocks regular
942                  * traffic while still permitting ethtool ioctls from reaching
943                  * the hardware as well as allowing the user to run the
944                  * interface after manually setting a hw addr using
945                  * `ip set address`
946                  */
947                 memset(hw->mac_addr, 0, netdev->addr_len);
948         } else {
949                 /* copy the MAC address out of the EEPROM */
950                 if (e1000_read_mac_addr(hw))
951                         DPRINTK(PROBE, ERR, "EEPROM Read Error\n");
952         }
953         /* don't block initalization here due to bad MAC address */
954         memcpy(netdev->dev_addr, hw->mac_addr, netdev->addr_len);
955         memcpy(netdev->perm_addr, hw->mac_addr, netdev->addr_len);
956
957         if (!is_valid_ether_addr(netdev->perm_addr))
958                 DPRINTK(PROBE, ERR, "Invalid MAC Address\n");
959
960         e1000_get_bus_info(hw);
961
962         init_timer(&adapter->tx_fifo_stall_timer);
963         adapter->tx_fifo_stall_timer.function = &e1000_82547_tx_fifo_stall;
964         adapter->tx_fifo_stall_timer.data = (unsigned long)adapter;
965
966         init_timer(&adapter->watchdog_timer);
967         adapter->watchdog_timer.function = &e1000_watchdog;
968         adapter->watchdog_timer.data = (unsigned long) adapter;
969
970         init_timer(&adapter->phy_info_timer);
971         adapter->phy_info_timer.function = &e1000_update_phy_info;
972         adapter->phy_info_timer.data = (unsigned long)adapter;
973
974         INIT_WORK(&adapter->reset_task, e1000_reset_task);
975
976         e1000_check_options(adapter);
977
978         /* Initial Wake on LAN setting
979          * If APM wake is enabled in the EEPROM,
980          * enable the ACPI Magic Packet filter
981          */
982
983         switch (hw->mac_type) {
984         case e1000_82542_rev2_0:
985         case e1000_82542_rev2_1:
986         case e1000_82543:
987                 break;
988         case e1000_82544:
989                 e1000_read_eeprom(hw,
990                         EEPROM_INIT_CONTROL2_REG, 1, &eeprom_data);
991                 eeprom_apme_mask = E1000_EEPROM_82544_APM;
992                 break;
993         case e1000_82546:
994         case e1000_82546_rev_3:
995                 if (er32(STATUS) & E1000_STATUS_FUNC_1){
996                         e1000_read_eeprom(hw,
997                                 EEPROM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
998                         break;
999                 }
1000                 /* Fall Through */
1001         default:
1002                 e1000_read_eeprom(hw,
1003                         EEPROM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
1004                 break;
1005         }
1006         if (eeprom_data & eeprom_apme_mask)
1007                 adapter->eeprom_wol |= E1000_WUFC_MAG;
1008
1009         /* now that we have the eeprom settings, apply the special cases
1010          * where the eeprom may be wrong or the board simply won't support
1011          * wake on lan on a particular port */
1012         switch (pdev->device) {
1013         case E1000_DEV_ID_82546GB_PCIE:
1014                 adapter->eeprom_wol = 0;
1015                 break;
1016         case E1000_DEV_ID_82546EB_FIBER:
1017         case E1000_DEV_ID_82546GB_FIBER:
1018                 /* Wake events only supported on port A for dual fiber
1019                  * regardless of eeprom setting */
1020                 if (er32(STATUS) & E1000_STATUS_FUNC_1)
1021                         adapter->eeprom_wol = 0;
1022                 break;
1023         case E1000_DEV_ID_82546GB_QUAD_COPPER_KSP3:
1024                 /* if quad port adapter, disable WoL on all but port A */
1025                 if (global_quad_port_a != 0)
1026                         adapter->eeprom_wol = 0;
1027                 else
1028                         adapter->quad_port_a = 1;
1029                 /* Reset for multiple quad port adapters */
1030                 if (++global_quad_port_a == 4)
1031                         global_quad_port_a = 0;
1032                 break;
1033         }
1034
1035         /* initialize the wol settings based on the eeprom settings */
1036         adapter->wol = adapter->eeprom_wol;
1037         device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
1038
1039         /* print bus type/speed/width info */
1040         DPRINTK(PROBE, INFO, "(PCI%s:%s:%s) ",
1041                 ((hw->bus_type == e1000_bus_type_pcix) ? "-X" : ""),
1042                 ((hw->bus_speed == e1000_bus_speed_133) ? "133MHz" :
1043                  (hw->bus_speed == e1000_bus_speed_120) ? "120MHz" :
1044                  (hw->bus_speed == e1000_bus_speed_100) ? "100MHz" :
1045                  (hw->bus_speed == e1000_bus_speed_66) ? "66MHz" : "33MHz"),
1046                 ((hw->bus_width == e1000_bus_width_64) ? "64-bit" : "32-bit"));
1047
1048         printk("%pM\n", netdev->dev_addr);
1049
1050         /* reset the hardware with the new settings */
1051         e1000_reset(adapter);
1052
1053         strcpy(netdev->name, "eth%d");
1054         err = register_netdev(netdev);
1055         if (err)
1056                 goto err_register;
1057
1058         /* carrier off reporting is important to ethtool even BEFORE open */
1059         netif_carrier_off(netdev);
1060
1061         DPRINTK(PROBE, INFO, "Intel(R) PRO/1000 Network Connection\n");
1062
1063         cards_found++;
1064         return 0;
1065
1066 err_register:
1067 err_eeprom:
1068         e1000_phy_hw_reset(hw);
1069
1070         if (hw->flash_address)
1071                 iounmap(hw->flash_address);
1072         kfree(adapter->tx_ring);
1073         kfree(adapter->rx_ring);
1074 err_sw_init:
1075         iounmap(hw->hw_addr);
1076 err_ioremap:
1077         free_netdev(netdev);
1078 err_alloc_etherdev:
1079         pci_release_selected_regions(pdev, bars);
1080 err_pci_reg:
1081 err_dma:
1082         pci_disable_device(pdev);
1083         return err;
1084 }
1085
1086 /**
1087  * e1000_remove - Device Removal Routine
1088  * @pdev: PCI device information struct
1089  *
1090  * e1000_remove is called by the PCI subsystem to alert the driver
1091  * that it should release a PCI device.  The could be caused by a
1092  * Hot-Plug event, or because the driver is going to be removed from
1093  * memory.
1094  **/
1095
1096 static void __devexit e1000_remove(struct pci_dev *pdev)
1097 {
1098         struct net_device *netdev = pci_get_drvdata(pdev);
1099         struct e1000_adapter *adapter = netdev_priv(netdev);
1100         struct e1000_hw *hw = &adapter->hw;
1101
1102         cancel_work_sync(&adapter->reset_task);
1103
1104         e1000_release_manageability(adapter);
1105
1106         unregister_netdev(netdev);
1107
1108         e1000_phy_hw_reset(hw);
1109
1110         kfree(adapter->tx_ring);
1111         kfree(adapter->rx_ring);
1112
1113         iounmap(hw->hw_addr);
1114         if (hw->flash_address)
1115                 iounmap(hw->flash_address);
1116         pci_release_selected_regions(pdev, adapter->bars);
1117
1118         free_netdev(netdev);
1119
1120         pci_disable_device(pdev);
1121 }
1122
1123 /**
1124  * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
1125  * @adapter: board private structure to initialize
1126  *
1127  * e1000_sw_init initializes the Adapter private data structure.
1128  * Fields are initialized based on PCI device information and
1129  * OS network device settings (MTU size).
1130  **/
1131
1132 static int __devinit e1000_sw_init(struct e1000_adapter *adapter)
1133 {
1134         struct e1000_hw *hw = &adapter->hw;
1135         struct net_device *netdev = adapter->netdev;
1136         struct pci_dev *pdev = adapter->pdev;
1137
1138         /* PCI config space info */
1139
1140         hw->vendor_id = pdev->vendor;
1141         hw->device_id = pdev->device;
1142         hw->subsystem_vendor_id = pdev->subsystem_vendor;
1143         hw->subsystem_id = pdev->subsystem_device;
1144         hw->revision_id = pdev->revision;
1145
1146         pci_read_config_word(pdev, PCI_COMMAND, &hw->pci_cmd_word);
1147
1148         adapter->rx_buffer_len = MAXIMUM_ETHERNET_VLAN_SIZE;
1149         hw->max_frame_size = netdev->mtu +
1150                              ENET_HEADER_SIZE + ETHERNET_FCS_SIZE;
1151         hw->min_frame_size = MINIMUM_ETHERNET_FRAME_SIZE;
1152
1153         /* identify the MAC */
1154
1155         if (e1000_set_mac_type(hw)) {
1156                 DPRINTK(PROBE, ERR, "Unknown MAC Type\n");
1157                 return -EIO;
1158         }
1159
1160         switch (hw->mac_type) {
1161         default:
1162                 break;
1163         case e1000_82541:
1164         case e1000_82547:
1165         case e1000_82541_rev_2:
1166         case e1000_82547_rev_2:
1167                 hw->phy_init_script = 1;
1168                 break;
1169         }
1170
1171         e1000_set_media_type(hw);
1172
1173         hw->wait_autoneg_complete = false;
1174         hw->tbi_compatibility_en = true;
1175         hw->adaptive_ifs = true;
1176
1177         /* Copper options */
1178
1179         if (hw->media_type == e1000_media_type_copper) {
1180                 hw->mdix = AUTO_ALL_MODES;
1181                 hw->disable_polarity_correction = false;
1182                 hw->master_slave = E1000_MASTER_SLAVE;
1183         }
1184
1185         adapter->num_tx_queues = 1;
1186         adapter->num_rx_queues = 1;
1187
1188         if (e1000_alloc_queues(adapter)) {
1189                 DPRINTK(PROBE, ERR, "Unable to allocate memory for queues\n");
1190                 return -ENOMEM;
1191         }
1192
1193         /* Explicitly disable IRQ since the NIC can be in any state. */
1194         e1000_irq_disable(adapter);
1195
1196         spin_lock_init(&adapter->stats_lock);
1197
1198         set_bit(__E1000_DOWN, &adapter->flags);
1199
1200         return 0;
1201 }
1202
1203 /**
1204  * e1000_alloc_queues - Allocate memory for all rings
1205  * @adapter: board private structure to initialize
1206  *
1207  * We allocate one ring per queue at run-time since we don't know the
1208  * number of queues at compile-time.
1209  **/
1210
1211 static int __devinit e1000_alloc_queues(struct e1000_adapter *adapter)
1212 {
1213         adapter->tx_ring = kcalloc(adapter->num_tx_queues,
1214                                    sizeof(struct e1000_tx_ring), GFP_KERNEL);
1215         if (!adapter->tx_ring)
1216                 return -ENOMEM;
1217
1218         adapter->rx_ring = kcalloc(adapter->num_rx_queues,
1219                                    sizeof(struct e1000_rx_ring), GFP_KERNEL);
1220         if (!adapter->rx_ring) {
1221                 kfree(adapter->tx_ring);
1222                 return -ENOMEM;
1223         }
1224
1225         return E1000_SUCCESS;
1226 }
1227
1228 /**
1229  * e1000_open - Called when a network interface is made active
1230  * @netdev: network interface device structure
1231  *
1232  * Returns 0 on success, negative value on failure
1233  *
1234  * The open entry point is called when a network interface is made
1235  * active by the system (IFF_UP).  At this point all resources needed
1236  * for transmit and receive operations are allocated, the interrupt
1237  * handler is registered with the OS, the watchdog timer is started,
1238  * and the stack is notified that the interface is ready.
1239  **/
1240
1241 static int e1000_open(struct net_device *netdev)
1242 {
1243         struct e1000_adapter *adapter = netdev_priv(netdev);
1244         struct e1000_hw *hw = &adapter->hw;
1245         int err;
1246
1247         /* disallow open during test */
1248         if (test_bit(__E1000_TESTING, &adapter->flags))
1249                 return -EBUSY;
1250
1251         netif_carrier_off(netdev);
1252
1253         /* allocate transmit descriptors */
1254         err = e1000_setup_all_tx_resources(adapter);
1255         if (err)
1256                 goto err_setup_tx;
1257
1258         /* allocate receive descriptors */
1259         err = e1000_setup_all_rx_resources(adapter);
1260         if (err)
1261                 goto err_setup_rx;
1262
1263         e1000_power_up_phy(adapter);
1264
1265         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
1266         if ((hw->mng_cookie.status &
1267                           E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT)) {
1268                 e1000_update_mng_vlan(adapter);
1269         }
1270
1271         /* before we allocate an interrupt, we must be ready to handle it.
1272          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
1273          * as soon as we call pci_request_irq, so we have to setup our
1274          * clean_rx handler before we do so.  */
1275         e1000_configure(adapter);
1276
1277         err = e1000_request_irq(adapter);
1278         if (err)
1279                 goto err_req_irq;
1280
1281         /* From here on the code is the same as e1000_up() */
1282         clear_bit(__E1000_DOWN, &adapter->flags);
1283
1284         napi_enable(&adapter->napi);
1285
1286         e1000_irq_enable(adapter);
1287
1288         netif_start_queue(netdev);
1289
1290         /* fire a link status change interrupt to start the watchdog */
1291         ew32(ICS, E1000_ICS_LSC);
1292
1293         return E1000_SUCCESS;
1294
1295 err_req_irq:
1296         e1000_power_down_phy(adapter);
1297         e1000_free_all_rx_resources(adapter);
1298 err_setup_rx:
1299         e1000_free_all_tx_resources(adapter);
1300 err_setup_tx:
1301         e1000_reset(adapter);
1302
1303         return err;
1304 }
1305
1306 /**
1307  * e1000_close - Disables a network interface
1308  * @netdev: network interface device structure
1309  *
1310  * Returns 0, this is not allowed to fail
1311  *
1312  * The close entry point is called when an interface is de-activated
1313  * by the OS.  The hardware is still under the drivers control, but
1314  * needs to be disabled.  A global MAC reset is issued to stop the
1315  * hardware, and all transmit and receive resources are freed.
1316  **/
1317
1318 static int e1000_close(struct net_device *netdev)
1319 {
1320         struct e1000_adapter *adapter = netdev_priv(netdev);
1321         struct e1000_hw *hw = &adapter->hw;
1322
1323         WARN_ON(test_bit(__E1000_RESETTING, &adapter->flags));
1324         e1000_down(adapter);
1325         e1000_power_down_phy(adapter);
1326         e1000_free_irq(adapter);
1327
1328         e1000_free_all_tx_resources(adapter);
1329         e1000_free_all_rx_resources(adapter);
1330
1331         /* kill manageability vlan ID if supported, but not if a vlan with
1332          * the same ID is registered on the host OS (let 8021q kill it) */
1333         if ((hw->mng_cookie.status &
1334                           E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT) &&
1335              !(adapter->vlgrp &&
1336                vlan_group_get_device(adapter->vlgrp, adapter->mng_vlan_id))) {
1337                 e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
1338         }
1339
1340         return 0;
1341 }
1342
1343 /**
1344  * e1000_check_64k_bound - check that memory doesn't cross 64kB boundary
1345  * @adapter: address of board private structure
1346  * @start: address of beginning of memory
1347  * @len: length of memory
1348  **/
1349 static bool e1000_check_64k_bound(struct e1000_adapter *adapter, void *start,
1350                                   unsigned long len)
1351 {
1352         struct e1000_hw *hw = &adapter->hw;
1353         unsigned long begin = (unsigned long)start;
1354         unsigned long end = begin + len;
1355
1356         /* First rev 82545 and 82546 need to not allow any memory
1357          * write location to cross 64k boundary due to errata 23 */
1358         if (hw->mac_type == e1000_82545 ||
1359             hw->mac_type == e1000_82546) {
1360                 return ((begin ^ (end - 1)) >> 16) != 0 ? false : true;
1361         }
1362
1363         return true;
1364 }
1365
1366 /**
1367  * e1000_setup_tx_resources - allocate Tx resources (Descriptors)
1368  * @adapter: board private structure
1369  * @txdr:    tx descriptor ring (for a specific queue) to setup
1370  *
1371  * Return 0 on success, negative on failure
1372  **/
1373
1374 static int e1000_setup_tx_resources(struct e1000_adapter *adapter,
1375                                     struct e1000_tx_ring *txdr)
1376 {
1377         struct pci_dev *pdev = adapter->pdev;
1378         int size;
1379
1380         size = sizeof(struct e1000_buffer) * txdr->count;
1381         txdr->buffer_info = vmalloc(size);
1382         if (!txdr->buffer_info) {
1383                 DPRINTK(PROBE, ERR,
1384                 "Unable to allocate memory for the transmit descriptor ring\n");
1385                 return -ENOMEM;
1386         }
1387         memset(txdr->buffer_info, 0, size);
1388
1389         /* round up to nearest 4K */
1390
1391         txdr->size = txdr->count * sizeof(struct e1000_tx_desc);
1392         txdr->size = ALIGN(txdr->size, 4096);
1393
1394         txdr->desc = pci_alloc_consistent(pdev, txdr->size, &txdr->dma);
1395         if (!txdr->desc) {
1396 setup_tx_desc_die:
1397                 vfree(txdr->buffer_info);
1398                 DPRINTK(PROBE, ERR,
1399                 "Unable to allocate memory for the transmit descriptor ring\n");
1400                 return -ENOMEM;
1401         }
1402
1403         /* Fix for errata 23, can't cross 64kB boundary */
1404         if (!e1000_check_64k_bound(adapter, txdr->desc, txdr->size)) {
1405                 void *olddesc = txdr->desc;
1406                 dma_addr_t olddma = txdr->dma;
1407                 DPRINTK(TX_ERR, ERR, "txdr align check failed: %u bytes "
1408                                      "at %p\n", txdr->size, txdr->desc);
1409                 /* Try again, without freeing the previous */
1410                 txdr->desc = pci_alloc_consistent(pdev, txdr->size, &txdr->dma);
1411                 /* Failed allocation, critical failure */
1412                 if (!txdr->desc) {
1413                         pci_free_consistent(pdev, txdr->size, olddesc, olddma);
1414                         goto setup_tx_desc_die;
1415                 }
1416
1417                 if (!e1000_check_64k_bound(adapter, txdr->desc, txdr->size)) {
1418                         /* give up */
1419                         pci_free_consistent(pdev, txdr->size, txdr->desc,
1420                                             txdr->dma);
1421                         pci_free_consistent(pdev, txdr->size, olddesc, olddma);
1422                         DPRINTK(PROBE, ERR,
1423                                 "Unable to allocate aligned memory "
1424                                 "for the transmit descriptor ring\n");
1425                         vfree(txdr->buffer_info);
1426                         return -ENOMEM;
1427                 } else {
1428                         /* Free old allocation, new allocation was successful */
1429                         pci_free_consistent(pdev, txdr->size, olddesc, olddma);
1430                 }
1431         }
1432         memset(txdr->desc, 0, txdr->size);
1433
1434         txdr->next_to_use = 0;
1435         txdr->next_to_clean = 0;
1436
1437         return 0;
1438 }
1439
1440 /**
1441  * e1000_setup_all_tx_resources - wrapper to allocate Tx resources
1442  *                                (Descriptors) for all queues
1443  * @adapter: board private structure
1444  *
1445  * Return 0 on success, negative on failure
1446  **/
1447
1448 int e1000_setup_all_tx_resources(struct e1000_adapter *adapter)
1449 {
1450         int i, err = 0;
1451
1452         for (i = 0; i < adapter->num_tx_queues; i++) {
1453                 err = e1000_setup_tx_resources(adapter, &adapter->tx_ring[i]);
1454                 if (err) {
1455                         DPRINTK(PROBE, ERR,
1456                                 "Allocation for Tx Queue %u failed\n", i);
1457                         for (i-- ; i >= 0; i--)
1458                                 e1000_free_tx_resources(adapter,
1459                                                         &adapter->tx_ring[i]);
1460                         break;
1461                 }
1462         }
1463
1464         return err;
1465 }
1466
1467 /**
1468  * e1000_configure_tx - Configure 8254x Transmit Unit after Reset
1469  * @adapter: board private structure
1470  *
1471  * Configure the Tx unit of the MAC after a reset.
1472  **/
1473
1474 static void e1000_configure_tx(struct e1000_adapter *adapter)
1475 {
1476         u64 tdba;
1477         struct e1000_hw *hw = &adapter->hw;
1478         u32 tdlen, tctl, tipg;
1479         u32 ipgr1, ipgr2;
1480
1481         /* Setup the HW Tx Head and Tail descriptor pointers */
1482
1483         switch (adapter->num_tx_queues) {
1484         case 1:
1485         default:
1486                 tdba = adapter->tx_ring[0].dma;
1487                 tdlen = adapter->tx_ring[0].count *
1488                         sizeof(struct e1000_tx_desc);
1489                 ew32(TDLEN, tdlen);
1490                 ew32(TDBAH, (tdba >> 32));
1491                 ew32(TDBAL, (tdba & 0x00000000ffffffffULL));
1492                 ew32(TDT, 0);
1493                 ew32(TDH, 0);
1494                 adapter->tx_ring[0].tdh = ((hw->mac_type >= e1000_82543) ? E1000_TDH : E1000_82542_TDH);
1495                 adapter->tx_ring[0].tdt = ((hw->mac_type >= e1000_82543) ? E1000_TDT : E1000_82542_TDT);
1496                 break;
1497         }
1498
1499         /* Set the default values for the Tx Inter Packet Gap timer */
1500         if ((hw->media_type == e1000_media_type_fiber ||
1501              hw->media_type == e1000_media_type_internal_serdes))
1502                 tipg = DEFAULT_82543_TIPG_IPGT_FIBER;
1503         else
1504                 tipg = DEFAULT_82543_TIPG_IPGT_COPPER;
1505
1506         switch (hw->mac_type) {
1507         case e1000_82542_rev2_0:
1508         case e1000_82542_rev2_1:
1509                 tipg = DEFAULT_82542_TIPG_IPGT;
1510                 ipgr1 = DEFAULT_82542_TIPG_IPGR1;
1511                 ipgr2 = DEFAULT_82542_TIPG_IPGR2;
1512                 break;
1513         default:
1514                 ipgr1 = DEFAULT_82543_TIPG_IPGR1;
1515                 ipgr2 = DEFAULT_82543_TIPG_IPGR2;
1516                 break;
1517         }
1518         tipg |= ipgr1 << E1000_TIPG_IPGR1_SHIFT;
1519         tipg |= ipgr2 << E1000_TIPG_IPGR2_SHIFT;
1520         ew32(TIPG, tipg);
1521
1522         /* Set the Tx Interrupt Delay register */
1523
1524         ew32(TIDV, adapter->tx_int_delay);
1525         if (hw->mac_type >= e1000_82540)
1526                 ew32(TADV, adapter->tx_abs_int_delay);
1527
1528         /* Program the Transmit Control Register */
1529
1530         tctl = er32(TCTL);
1531         tctl &= ~E1000_TCTL_CT;
1532         tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
1533                 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
1534
1535         e1000_config_collision_dist(hw);
1536
1537         /* Setup Transmit Descriptor Settings for eop descriptor */
1538         adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
1539
1540         /* only set IDE if we are delaying interrupts using the timers */
1541         if (adapter->tx_int_delay)
1542                 adapter->txd_cmd |= E1000_TXD_CMD_IDE;
1543
1544         if (hw->mac_type < e1000_82543)
1545                 adapter->txd_cmd |= E1000_TXD_CMD_RPS;
1546         else
1547                 adapter->txd_cmd |= E1000_TXD_CMD_RS;
1548
1549         /* Cache if we're 82544 running in PCI-X because we'll
1550          * need this to apply a workaround later in the send path. */
1551         if (hw->mac_type == e1000_82544 &&
1552             hw->bus_type == e1000_bus_type_pcix)
1553                 adapter->pcix_82544 = 1;
1554
1555         ew32(TCTL, tctl);
1556
1557 }
1558
1559 /**
1560  * e1000_setup_rx_resources - allocate Rx resources (Descriptors)
1561  * @adapter: board private structure
1562  * @rxdr:    rx descriptor ring (for a specific queue) to setup
1563  *
1564  * Returns 0 on success, negative on failure
1565  **/
1566
1567 static int e1000_setup_rx_resources(struct e1000_adapter *adapter,
1568                                     struct e1000_rx_ring *rxdr)
1569 {
1570         struct pci_dev *pdev = adapter->pdev;
1571         int size, desc_len;
1572
1573         size = sizeof(struct e1000_buffer) * rxdr->count;
1574         rxdr->buffer_info = vmalloc(size);
1575         if (!rxdr->buffer_info) {
1576                 DPRINTK(PROBE, ERR,
1577                 "Unable to allocate memory for the receive descriptor ring\n");
1578                 return -ENOMEM;
1579         }
1580         memset(rxdr->buffer_info, 0, size);
1581
1582         desc_len = sizeof(struct e1000_rx_desc);
1583
1584         /* Round up to nearest 4K */
1585
1586         rxdr->size = rxdr->count * desc_len;
1587         rxdr->size = ALIGN(rxdr->size, 4096);
1588
1589         rxdr->desc = pci_alloc_consistent(pdev, rxdr->size, &rxdr->dma);
1590
1591         if (!rxdr->desc) {
1592                 DPRINTK(PROBE, ERR,
1593                 "Unable to allocate memory for the receive descriptor ring\n");
1594 setup_rx_desc_die:
1595                 vfree(rxdr->buffer_info);
1596                 return -ENOMEM;
1597         }
1598
1599         /* Fix for errata 23, can't cross 64kB boundary */
1600         if (!e1000_check_64k_bound(adapter, rxdr->desc, rxdr->size)) {
1601                 void *olddesc = rxdr->desc;
1602                 dma_addr_t olddma = rxdr->dma;
1603                 DPRINTK(RX_ERR, ERR, "rxdr align check failed: %u bytes "
1604                                      "at %p\n", rxdr->size, rxdr->desc);
1605                 /* Try again, without freeing the previous */
1606                 rxdr->desc = pci_alloc_consistent(pdev, rxdr->size, &rxdr->dma);
1607                 /* Failed allocation, critical failure */
1608                 if (!rxdr->desc) {
1609                         pci_free_consistent(pdev, rxdr->size, olddesc, olddma);
1610                         DPRINTK(PROBE, ERR,
1611                                 "Unable to allocate memory "
1612                                 "for the receive descriptor ring\n");
1613                         goto setup_rx_desc_die;
1614                 }
1615
1616                 if (!e1000_check_64k_bound(adapter, rxdr->desc, rxdr->size)) {
1617                         /* give up */
1618                         pci_free_consistent(pdev, rxdr->size, rxdr->desc,
1619                                             rxdr->dma);
1620                         pci_free_consistent(pdev, rxdr->size, olddesc, olddma);
1621                         DPRINTK(PROBE, ERR,
1622                                 "Unable to allocate aligned memory "
1623                                 "for the receive descriptor ring\n");
1624                         goto setup_rx_desc_die;
1625                 } else {
1626                         /* Free old allocation, new allocation was successful */
1627                         pci_free_consistent(pdev, rxdr->size, olddesc, olddma);
1628                 }
1629         }
1630         memset(rxdr->desc, 0, rxdr->size);
1631
1632         rxdr->next_to_clean = 0;
1633         rxdr->next_to_use = 0;
1634         rxdr->rx_skb_top = NULL;
1635
1636         return 0;
1637 }
1638
1639 /**
1640  * e1000_setup_all_rx_resources - wrapper to allocate Rx resources
1641  *                                (Descriptors) for all queues
1642  * @adapter: board private structure
1643  *
1644  * Return 0 on success, negative on failure
1645  **/
1646
1647 int e1000_setup_all_rx_resources(struct e1000_adapter *adapter)
1648 {
1649         int i, err = 0;
1650
1651         for (i = 0; i < adapter->num_rx_queues; i++) {
1652                 err = e1000_setup_rx_resources(adapter, &adapter->rx_ring[i]);
1653                 if (err) {
1654                         DPRINTK(PROBE, ERR,
1655                                 "Allocation for Rx Queue %u failed\n", i);
1656                         for (i-- ; i >= 0; i--)
1657                                 e1000_free_rx_resources(adapter,
1658                                                         &adapter->rx_ring[i]);
1659                         break;
1660                 }
1661         }
1662
1663         return err;
1664 }
1665
1666 /**
1667  * e1000_setup_rctl - configure the receive control registers
1668  * @adapter: Board private structure
1669  **/
1670 static void e1000_setup_rctl(struct e1000_adapter *adapter)
1671 {
1672         struct e1000_hw *hw = &adapter->hw;
1673         u32 rctl;
1674
1675         rctl = er32(RCTL);
1676
1677         rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
1678
1679         rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
1680                 E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
1681                 (hw->mc_filter_type << E1000_RCTL_MO_SHIFT);
1682
1683         if (hw->tbi_compatibility_on == 1)
1684                 rctl |= E1000_RCTL_SBP;
1685         else
1686                 rctl &= ~E1000_RCTL_SBP;
1687
1688         if (adapter->netdev->mtu <= ETH_DATA_LEN)
1689                 rctl &= ~E1000_RCTL_LPE;
1690         else
1691                 rctl |= E1000_RCTL_LPE;
1692
1693         /* Setup buffer sizes */
1694         rctl &= ~E1000_RCTL_SZ_4096;
1695         rctl |= E1000_RCTL_BSEX;
1696         switch (adapter->rx_buffer_len) {
1697                 case E1000_RXBUFFER_256:
1698                         rctl |= E1000_RCTL_SZ_256;
1699                         rctl &= ~E1000_RCTL_BSEX;
1700                         break;
1701                 case E1000_RXBUFFER_512:
1702                         rctl |= E1000_RCTL_SZ_512;
1703                         rctl &= ~E1000_RCTL_BSEX;
1704                         break;
1705                 case E1000_RXBUFFER_1024:
1706                         rctl |= E1000_RCTL_SZ_1024;
1707                         rctl &= ~E1000_RCTL_BSEX;
1708                         break;
1709                 case E1000_RXBUFFER_2048:
1710                 default:
1711                         rctl |= E1000_RCTL_SZ_2048;
1712                         rctl &= ~E1000_RCTL_BSEX;
1713                         break;
1714                 case E1000_RXBUFFER_4096:
1715                         rctl |= E1000_RCTL_SZ_4096;
1716                         break;
1717                 case E1000_RXBUFFER_8192:
1718                         rctl |= E1000_RCTL_SZ_8192;
1719                         break;
1720                 case E1000_RXBUFFER_16384:
1721                         rctl |= E1000_RCTL_SZ_16384;
1722                         break;
1723         }
1724
1725         ew32(RCTL, rctl);
1726 }
1727
1728 /**
1729  * e1000_configure_rx - Configure 8254x Receive Unit after Reset
1730  * @adapter: board private structure
1731  *
1732  * Configure the Rx unit of the MAC after a reset.
1733  **/
1734
1735 static void e1000_configure_rx(struct e1000_adapter *adapter)
1736 {
1737         u64 rdba;
1738         struct e1000_hw *hw = &adapter->hw;
1739         u32 rdlen, rctl, rxcsum;
1740
1741         if (adapter->netdev->mtu > ETH_DATA_LEN) {
1742                 rdlen = adapter->rx_ring[0].count *
1743                         sizeof(struct e1000_rx_desc);
1744                 adapter->clean_rx = e1000_clean_jumbo_rx_irq;
1745                 adapter->alloc_rx_buf = e1000_alloc_jumbo_rx_buffers;
1746         } else {
1747                 rdlen = adapter->rx_ring[0].count *
1748                         sizeof(struct e1000_rx_desc);
1749                 adapter->clean_rx = e1000_clean_rx_irq;
1750                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
1751         }
1752
1753         /* disable receives while setting up the descriptors */
1754         rctl = er32(RCTL);
1755         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1756
1757         /* set the Receive Delay Timer Register */
1758         ew32(RDTR, adapter->rx_int_delay);
1759
1760         if (hw->mac_type >= e1000_82540) {
1761                 ew32(RADV, adapter->rx_abs_int_delay);
1762                 if (adapter->itr_setting != 0)
1763                         ew32(ITR, 1000000000 / (adapter->itr * 256));
1764         }
1765
1766         /* Setup the HW Rx Head and Tail Descriptor Pointers and
1767          * the Base and Length of the Rx Descriptor Ring */
1768         switch (adapter->num_rx_queues) {
1769         case 1:
1770         default:
1771                 rdba = adapter->rx_ring[0].dma;
1772                 ew32(RDLEN, rdlen);
1773                 ew32(RDBAH, (rdba >> 32));
1774                 ew32(RDBAL, (rdba & 0x00000000ffffffffULL));
1775                 ew32(RDT, 0);
1776                 ew32(RDH, 0);
1777                 adapter->rx_ring[0].rdh = ((hw->mac_type >= e1000_82543) ? E1000_RDH : E1000_82542_RDH);
1778                 adapter->rx_ring[0].rdt = ((hw->mac_type >= e1000_82543) ? E1000_RDT : E1000_82542_RDT);
1779                 break;
1780         }
1781
1782         /* Enable 82543 Receive Checksum Offload for TCP and UDP */
1783         if (hw->mac_type >= e1000_82543) {
1784                 rxcsum = er32(RXCSUM);
1785                 if (adapter->rx_csum)
1786                         rxcsum |= E1000_RXCSUM_TUOFL;
1787                 else
1788                         /* don't need to clear IPPCSE as it defaults to 0 */
1789                         rxcsum &= ~E1000_RXCSUM_TUOFL;
1790                 ew32(RXCSUM, rxcsum);
1791         }
1792
1793         /* Enable Receives */
1794         ew32(RCTL, rctl);
1795 }
1796
1797 /**
1798  * e1000_free_tx_resources - Free Tx Resources per Queue
1799  * @adapter: board private structure
1800  * @tx_ring: Tx descriptor ring for a specific queue
1801  *
1802  * Free all transmit software resources
1803  **/
1804
1805 static void e1000_free_tx_resources(struct e1000_adapter *adapter,
1806                                     struct e1000_tx_ring *tx_ring)
1807 {
1808         struct pci_dev *pdev = adapter->pdev;
1809
1810         e1000_clean_tx_ring(adapter, tx_ring);
1811
1812         vfree(tx_ring->buffer_info);
1813         tx_ring->buffer_info = NULL;
1814
1815         pci_free_consistent(pdev, tx_ring->size, tx_ring->desc, tx_ring->dma);
1816
1817         tx_ring->desc = NULL;
1818 }
1819
1820 /**
1821  * e1000_free_all_tx_resources - Free Tx Resources for All Queues
1822  * @adapter: board private structure
1823  *
1824  * Free all transmit software resources
1825  **/
1826
1827 void e1000_free_all_tx_resources(struct e1000_adapter *adapter)
1828 {
1829         int i;
1830
1831         for (i = 0; i < adapter->num_tx_queues; i++)
1832                 e1000_free_tx_resources(adapter, &adapter->tx_ring[i]);
1833 }
1834
1835 static void e1000_unmap_and_free_tx_resource(struct e1000_adapter *adapter,
1836                                              struct e1000_buffer *buffer_info)
1837 {
1838         buffer_info->dma = 0;
1839         if (buffer_info->skb) {
1840                 skb_dma_unmap(&adapter->pdev->dev, buffer_info->skb,
1841                               DMA_TO_DEVICE);
1842                 dev_kfree_skb_any(buffer_info->skb);
1843                 buffer_info->skb = NULL;
1844         }
1845         buffer_info->time_stamp = 0;
1846         /* buffer_info must be completely set up in the transmit path */
1847 }
1848
1849 /**
1850  * e1000_clean_tx_ring - Free Tx Buffers
1851  * @adapter: board private structure
1852  * @tx_ring: ring to be cleaned
1853  **/
1854
1855 static void e1000_clean_tx_ring(struct e1000_adapter *adapter,
1856                                 struct e1000_tx_ring *tx_ring)
1857 {
1858         struct e1000_hw *hw = &adapter->hw;
1859         struct e1000_buffer *buffer_info;
1860         unsigned long size;
1861         unsigned int i;
1862
1863         /* Free all the Tx ring sk_buffs */
1864
1865         for (i = 0; i < tx_ring->count; i++) {
1866                 buffer_info = &tx_ring->buffer_info[i];
1867                 e1000_unmap_and_free_tx_resource(adapter, buffer_info);
1868         }
1869
1870         size = sizeof(struct e1000_buffer) * tx_ring->count;
1871         memset(tx_ring->buffer_info, 0, size);
1872
1873         /* Zero out the descriptor ring */
1874
1875         memset(tx_ring->desc, 0, tx_ring->size);
1876
1877         tx_ring->next_to_use = 0;
1878         tx_ring->next_to_clean = 0;
1879         tx_ring->last_tx_tso = 0;
1880
1881         writel(0, hw->hw_addr + tx_ring->tdh);
1882         writel(0, hw->hw_addr + tx_ring->tdt);
1883 }
1884
1885 /**
1886  * e1000_clean_all_tx_rings - Free Tx Buffers for all queues
1887  * @adapter: board private structure
1888  **/
1889
1890 static void e1000_clean_all_tx_rings(struct e1000_adapter *adapter)
1891 {
1892         int i;
1893
1894         for (i = 0; i < adapter->num_tx_queues; i++)
1895                 e1000_clean_tx_ring(adapter, &adapter->tx_ring[i]);
1896 }
1897
1898 /**
1899  * e1000_free_rx_resources - Free Rx Resources
1900  * @adapter: board private structure
1901  * @rx_ring: ring to clean the resources from
1902  *
1903  * Free all receive software resources
1904  **/
1905
1906 static void e1000_free_rx_resources(struct e1000_adapter *adapter,
1907                                     struct e1000_rx_ring *rx_ring)
1908 {
1909         struct pci_dev *pdev = adapter->pdev;
1910
1911         e1000_clean_rx_ring(adapter, rx_ring);
1912
1913         vfree(rx_ring->buffer_info);
1914         rx_ring->buffer_info = NULL;
1915
1916         pci_free_consistent(pdev, rx_ring->size, rx_ring->desc, rx_ring->dma);
1917
1918         rx_ring->desc = NULL;
1919 }
1920
1921 /**
1922  * e1000_free_all_rx_resources - Free Rx Resources for All Queues
1923  * @adapter: board private structure
1924  *
1925  * Free all receive software resources
1926  **/
1927
1928 void e1000_free_all_rx_resources(struct e1000_adapter *adapter)
1929 {
1930         int i;
1931
1932         for (i = 0; i < adapter->num_rx_queues; i++)
1933                 e1000_free_rx_resources(adapter, &adapter->rx_ring[i]);
1934 }
1935
1936 /**
1937  * e1000_clean_rx_ring - Free Rx Buffers per Queue
1938  * @adapter: board private structure
1939  * @rx_ring: ring to free buffers from
1940  **/
1941
1942 static void e1000_clean_rx_ring(struct e1000_adapter *adapter,
1943                                 struct e1000_rx_ring *rx_ring)
1944 {
1945         struct e1000_hw *hw = &adapter->hw;
1946         struct e1000_buffer *buffer_info;
1947         struct pci_dev *pdev = adapter->pdev;
1948         unsigned long size;
1949         unsigned int i;
1950
1951         /* Free all the Rx ring sk_buffs */
1952         for (i = 0; i < rx_ring->count; i++) {
1953                 buffer_info = &rx_ring->buffer_info[i];
1954                 if (buffer_info->dma &&
1955                     adapter->clean_rx == e1000_clean_rx_irq) {
1956                         pci_unmap_single(pdev, buffer_info->dma,
1957                                          buffer_info->length,
1958                                          PCI_DMA_FROMDEVICE);
1959                 } else if (buffer_info->dma &&
1960                            adapter->clean_rx == e1000_clean_jumbo_rx_irq) {
1961                         pci_unmap_page(pdev, buffer_info->dma,
1962                                        buffer_info->length,
1963                                        PCI_DMA_FROMDEVICE);
1964                 }
1965
1966                 buffer_info->dma = 0;
1967                 if (buffer_info->page) {
1968                         put_page(buffer_info->page);
1969                         buffer_info->page = NULL;
1970                 }
1971                 if (buffer_info->skb) {
1972                         dev_kfree_skb(buffer_info->skb);
1973                         buffer_info->skb = NULL;
1974                 }
1975         }
1976
1977         /* there also may be some cached data from a chained receive */
1978         if (rx_ring->rx_skb_top) {
1979                 dev_kfree_skb(rx_ring->rx_skb_top);
1980                 rx_ring->rx_skb_top = NULL;
1981         }
1982
1983         size = sizeof(struct e1000_buffer) * rx_ring->count;
1984         memset(rx_ring->buffer_info, 0, size);
1985
1986         /* Zero out the descriptor ring */
1987         memset(rx_ring->desc, 0, rx_ring->size);
1988
1989         rx_ring->next_to_clean = 0;
1990         rx_ring->next_to_use = 0;
1991
1992         writel(0, hw->hw_addr + rx_ring->rdh);
1993         writel(0, hw->hw_addr + rx_ring->rdt);
1994 }
1995
1996 /**
1997  * e1000_clean_all_rx_rings - Free Rx Buffers for all queues
1998  * @adapter: board private structure
1999  **/
2000
2001 static void e1000_clean_all_rx_rings(struct e1000_adapter *adapter)
2002 {
2003         int i;
2004
2005         for (i = 0; i < adapter->num_rx_queues; i++)
2006                 e1000_clean_rx_ring(adapter, &adapter->rx_ring[i]);
2007 }
2008
2009 /* The 82542 2.0 (revision 2) needs to have the receive unit in reset
2010  * and memory write and invalidate disabled for certain operations
2011  */
2012 static void e1000_enter_82542_rst(struct e1000_adapter *adapter)
2013 {
2014         struct e1000_hw *hw = &adapter->hw;
2015         struct net_device *netdev = adapter->netdev;
2016         u32 rctl;
2017
2018         e1000_pci_clear_mwi(hw);
2019
2020         rctl = er32(RCTL);
2021         rctl |= E1000_RCTL_RST;
2022         ew32(RCTL, rctl);
2023         E1000_WRITE_FLUSH();
2024         mdelay(5);
2025
2026         if (netif_running(netdev))
2027                 e1000_clean_all_rx_rings(adapter);
2028 }
2029
2030 static void e1000_leave_82542_rst(struct e1000_adapter *adapter)
2031 {
2032         struct e1000_hw *hw = &adapter->hw;
2033         struct net_device *netdev = adapter->netdev;
2034         u32 rctl;
2035
2036         rctl = er32(RCTL);
2037         rctl &= ~E1000_RCTL_RST;
2038         ew32(RCTL, rctl);
2039         E1000_WRITE_FLUSH();
2040         mdelay(5);
2041
2042         if (hw->pci_cmd_word & PCI_COMMAND_INVALIDATE)
2043                 e1000_pci_set_mwi(hw);
2044
2045         if (netif_running(netdev)) {
2046                 /* No need to loop, because 82542 supports only 1 queue */
2047                 struct e1000_rx_ring *ring = &adapter->rx_ring[0];
2048                 e1000_configure_rx(adapter);
2049                 adapter->alloc_rx_buf(adapter, ring, E1000_DESC_UNUSED(ring));
2050         }
2051 }
2052
2053 /**
2054  * e1000_set_mac - Change the Ethernet Address of the NIC
2055  * @netdev: network interface device structure
2056  * @p: pointer to an address structure
2057  *
2058  * Returns 0 on success, negative on failure
2059  **/
2060
2061 static int e1000_set_mac(struct net_device *netdev, void *p)
2062 {
2063         struct e1000_adapter *adapter = netdev_priv(netdev);
2064         struct e1000_hw *hw = &adapter->hw;
2065         struct sockaddr *addr = p;
2066
2067         if (!is_valid_ether_addr(addr->sa_data))
2068                 return -EADDRNOTAVAIL;
2069
2070         /* 82542 2.0 needs to be in reset to write receive address registers */
2071
2072         if (hw->mac_type == e1000_82542_rev2_0)
2073                 e1000_enter_82542_rst(adapter);
2074
2075         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
2076         memcpy(hw->mac_addr, addr->sa_data, netdev->addr_len);
2077
2078         e1000_rar_set(hw, hw->mac_addr, 0);
2079
2080         if (hw->mac_type == e1000_82542_rev2_0)
2081                 e1000_leave_82542_rst(adapter);
2082
2083         return 0;
2084 }
2085
2086 /**
2087  * e1000_set_rx_mode - Secondary Unicast, Multicast and Promiscuous mode set
2088  * @netdev: network interface device structure
2089  *
2090  * The set_rx_mode entry point is called whenever the unicast or multicast
2091  * address lists or the network interface flags are updated. This routine is
2092  * responsible for configuring the hardware for proper unicast, multicast,
2093  * promiscuous mode, and all-multi behavior.
2094  **/
2095
2096 static void e1000_set_rx_mode(struct net_device *netdev)
2097 {
2098         struct e1000_adapter *adapter = netdev_priv(netdev);
2099         struct e1000_hw *hw = &adapter->hw;
2100         struct netdev_hw_addr *ha;
2101         bool use_uc = false;
2102         struct dev_addr_list *mc_ptr;
2103         u32 rctl;
2104         u32 hash_value;
2105         int i, rar_entries = E1000_RAR_ENTRIES;
2106         int mta_reg_count = E1000_NUM_MTA_REGISTERS;
2107         u32 *mcarray = kcalloc(mta_reg_count, sizeof(u32), GFP_ATOMIC);
2108
2109         if (!mcarray) {
2110                 DPRINTK(PROBE, ERR, "memory allocation failed\n");
2111                 return;
2112         }
2113
2114         /* Check for Promiscuous and All Multicast modes */
2115
2116         rctl = er32(RCTL);
2117
2118         if (netdev->flags & IFF_PROMISC) {
2119                 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
2120                 rctl &= ~E1000_RCTL_VFE;
2121         } else {
2122                 if (netdev->flags & IFF_ALLMULTI)
2123                         rctl |= E1000_RCTL_MPE;
2124                 else
2125                         rctl &= ~E1000_RCTL_MPE;
2126                 /* Enable VLAN filter if there is a VLAN */
2127                 if (adapter->vlgrp)
2128                         rctl |= E1000_RCTL_VFE;
2129         }
2130
2131         if (netdev->uc.count > rar_entries - 1) {
2132                 rctl |= E1000_RCTL_UPE;
2133         } else if (!(netdev->flags & IFF_PROMISC)) {
2134                 rctl &= ~E1000_RCTL_UPE;
2135                 use_uc = true;
2136         }
2137
2138         ew32(RCTL, rctl);
2139
2140         /* 82542 2.0 needs to be in reset to write receive address registers */
2141
2142         if (hw->mac_type == e1000_82542_rev2_0)
2143                 e1000_enter_82542_rst(adapter);
2144
2145         /* load the first 14 addresses into the exact filters 1-14. Unicast
2146          * addresses take precedence to avoid disabling unicast filtering
2147          * when possible.
2148          *
2149          * RAR 0 is used for the station MAC adddress
2150          * if there are not 14 addresses, go ahead and clear the filters
2151          */
2152         i = 1;
2153         if (use_uc)
2154                 list_for_each_entry(ha, &netdev->uc.list, list) {
2155                         if (i == rar_entries)
2156                                 break;
2157                         e1000_rar_set(hw, ha->addr, i++);
2158                 }
2159
2160         WARN_ON(i == rar_entries);
2161
2162         mc_ptr = netdev->mc_list;
2163
2164         for (; i < rar_entries; i++) {
2165                 if (mc_ptr) {
2166                         e1000_rar_set(hw, mc_ptr->da_addr, i);
2167                         mc_ptr = mc_ptr->next;
2168                 } else {
2169                         E1000_WRITE_REG_ARRAY(hw, RA, i << 1, 0);
2170                         E1000_WRITE_FLUSH();
2171                         E1000_WRITE_REG_ARRAY(hw, RA, (i << 1) + 1, 0);
2172                         E1000_WRITE_FLUSH();
2173                 }
2174         }
2175
2176         /* load any remaining addresses into the hash table */
2177
2178         for (; mc_ptr; mc_ptr = mc_ptr->next) {
2179                 u32 hash_reg, hash_bit, mta;
2180                 hash_value = e1000_hash_mc_addr(hw, mc_ptr->da_addr);
2181                 hash_reg = (hash_value >> 5) & 0x7F;
2182                 hash_bit = hash_value & 0x1F;
2183                 mta = (1 << hash_bit);
2184                 mcarray[hash_reg] |= mta;
2185         }
2186
2187         /* write the hash table completely, write from bottom to avoid
2188          * both stupid write combining chipsets, and flushing each write */
2189         for (i = mta_reg_count - 1; i >= 0 ; i--) {
2190                 /*
2191                  * If we are on an 82544 has an errata where writing odd
2192                  * offsets overwrites the previous even offset, but writing
2193                  * backwards over the range solves the issue by always
2194                  * writing the odd offset first
2195                  */
2196                 E1000_WRITE_REG_ARRAY(hw, MTA, i, mcarray[i]);
2197         }
2198         E1000_WRITE_FLUSH();
2199
2200         if (hw->mac_type == e1000_82542_rev2_0)
2201                 e1000_leave_82542_rst(adapter);
2202
2203         kfree(mcarray);
2204 }
2205
2206 /* Need to wait a few seconds after link up to get diagnostic information from
2207  * the phy */
2208
2209 static void e1000_update_phy_info(unsigned long data)
2210 {
2211         struct e1000_adapter *adapter = (struct e1000_adapter *)data;
2212         struct e1000_hw *hw = &adapter->hw;
2213         e1000_phy_get_info(hw, &adapter->phy_info);
2214 }
2215
2216 /**
2217  * e1000_82547_tx_fifo_stall - Timer Call-back
2218  * @data: pointer to adapter cast into an unsigned long
2219  **/
2220
2221 static void e1000_82547_tx_fifo_stall(unsigned long data)
2222 {
2223         struct e1000_adapter *adapter = (struct e1000_adapter *)data;
2224         struct e1000_hw *hw = &adapter->hw;
2225         struct net_device *netdev = adapter->netdev;
2226         u32 tctl;
2227
2228         if (atomic_read(&adapter->tx_fifo_stall)) {
2229                 if ((er32(TDT) == er32(TDH)) &&
2230                    (er32(TDFT) == er32(TDFH)) &&
2231                    (er32(TDFTS) == er32(TDFHS))) {
2232                         tctl = er32(TCTL);
2233                         ew32(TCTL, tctl & ~E1000_TCTL_EN);
2234                         ew32(TDFT, adapter->tx_head_addr);
2235                         ew32(TDFH, adapter->tx_head_addr);
2236                         ew32(TDFTS, adapter->tx_head_addr);
2237                         ew32(TDFHS, adapter->tx_head_addr);
2238                         ew32(TCTL, tctl);
2239                         E1000_WRITE_FLUSH();
2240
2241                         adapter->tx_fifo_head = 0;
2242                         atomic_set(&adapter->tx_fifo_stall, 0);
2243                         netif_wake_queue(netdev);
2244                 } else {
2245                         mod_timer(&adapter->tx_fifo_stall_timer, jiffies + 1);
2246                 }
2247         }
2248 }
2249
2250 /**
2251  * e1000_watchdog - Timer Call-back
2252  * @data: pointer to adapter cast into an unsigned long
2253  **/
2254 static void e1000_watchdog(unsigned long data)
2255 {
2256         struct e1000_adapter *adapter = (struct e1000_adapter *)data;
2257         struct e1000_hw *hw = &adapter->hw;
2258         struct net_device *netdev = adapter->netdev;
2259         struct e1000_tx_ring *txdr = adapter->tx_ring;
2260         u32 link, tctl;
2261
2262         e1000_check_for_link(hw);
2263
2264         if ((hw->media_type == e1000_media_type_internal_serdes) &&
2265            !(er32(TXCW) & E1000_TXCW_ANE))
2266                 link = !hw->serdes_link_down;
2267         else
2268                 link = er32(STATUS) & E1000_STATUS_LU;
2269
2270         if (link) {
2271                 if (!netif_carrier_ok(netdev)) {
2272                         u32 ctrl;
2273                         bool txb2b = true;
2274                         e1000_get_speed_and_duplex(hw,
2275                                                    &adapter->link_speed,
2276                                                    &adapter->link_duplex);
2277
2278                         ctrl = er32(CTRL);
2279                         printk(KERN_INFO "e1000: %s NIC Link is Up %d Mbps %s, "
2280                                "Flow Control: %s\n",
2281                                netdev->name,
2282                                adapter->link_speed,
2283                                adapter->link_duplex == FULL_DUPLEX ?
2284                                 "Full Duplex" : "Half Duplex",
2285                                 ((ctrl & E1000_CTRL_TFCE) && (ctrl &
2286                                 E1000_CTRL_RFCE)) ? "RX/TX" : ((ctrl &
2287                                 E1000_CTRL_RFCE) ? "RX" : ((ctrl &
2288                                 E1000_CTRL_TFCE) ? "TX" : "None" )));
2289
2290                         /* tweak tx_queue_len according to speed/duplex
2291                          * and adjust the timeout factor */
2292                         netdev->tx_queue_len = adapter->tx_queue_len;
2293                         adapter->tx_timeout_factor = 1;
2294                         switch (adapter->link_speed) {
2295                         case SPEED_10:
2296                                 txb2b = false;
2297                                 netdev->tx_queue_len = 10;
2298                                 adapter->tx_timeout_factor = 8;
2299                                 break;
2300                         case SPEED_100:
2301                                 txb2b = false;
2302                                 netdev->tx_queue_len = 100;
2303                                 /* maybe add some timeout factor ? */
2304                                 break;
2305                         }
2306
2307                         /* enable transmits in the hardware */
2308                         tctl = er32(TCTL);
2309                         tctl |= E1000_TCTL_EN;
2310                         ew32(TCTL, tctl);
2311
2312                         netif_carrier_on(netdev);
2313                         mod_timer(&adapter->phy_info_timer,
2314                                   round_jiffies(jiffies + 2 * HZ));
2315                         adapter->smartspeed = 0;
2316                 }
2317         } else {
2318                 if (netif_carrier_ok(netdev)) {
2319                         adapter->link_speed = 0;
2320                         adapter->link_duplex = 0;
2321                         printk(KERN_INFO "e1000: %s NIC Link is Down\n",
2322                                netdev->name);
2323                         netif_carrier_off(netdev);
2324                         mod_timer(&adapter->phy_info_timer,
2325                                   round_jiffies(jiffies + 2 * HZ));
2326                 }
2327
2328                 e1000_smartspeed(adapter);
2329         }
2330
2331         e1000_update_stats(adapter);
2332
2333         hw->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
2334         adapter->tpt_old = adapter->stats.tpt;
2335         hw->collision_delta = adapter->stats.colc - adapter->colc_old;
2336         adapter->colc_old = adapter->stats.colc;
2337
2338         adapter->gorcl = adapter->stats.gorcl - adapter->gorcl_old;
2339         adapter->gorcl_old = adapter->stats.gorcl;
2340         adapter->gotcl = adapter->stats.gotcl - adapter->gotcl_old;
2341         adapter->gotcl_old = adapter->stats.gotcl;
2342
2343         e1000_update_adaptive(hw);
2344
2345         if (!netif_carrier_ok(netdev)) {
2346                 if (E1000_DESC_UNUSED(txdr) + 1 < txdr->count) {
2347                         /* We've lost link, so the controller stops DMA,
2348                          * but we've got queued Tx work that's never going
2349                          * to get done, so reset controller to flush Tx.
2350                          * (Do the reset outside of interrupt context). */
2351                         adapter->tx_timeout_count++;
2352                         schedule_work(&adapter->reset_task);
2353                         /* return immediately since reset is imminent */
2354                         return;
2355                 }
2356         }
2357
2358         /* Cause software interrupt to ensure rx ring is cleaned */
2359         ew32(ICS, E1000_ICS_RXDMT0);
2360
2361         /* Force detection of hung controller every watchdog period */
2362         adapter->detect_tx_hung = true;
2363
2364         /* Reset the timer */
2365         mod_timer(&adapter->watchdog_timer, round_jiffies(jiffies + 2 * HZ));
2366 }
2367
2368 enum latency_range {
2369         lowest_latency = 0,
2370         low_latency = 1,
2371         bulk_latency = 2,
2372         latency_invalid = 255
2373 };
2374
2375 /**
2376  * e1000_update_itr - update the dynamic ITR value based on statistics
2377  *      Stores a new ITR value based on packets and byte
2378  *      counts during the last interrupt.  The advantage of per interrupt
2379  *      computation is faster updates and more accurate ITR for the current
2380  *      traffic pattern.  Constants in this function were computed
2381  *      based on theoretical maximum wire speed and thresholds were set based
2382  *      on testing data as well as attempting to minimize response time
2383  *      while increasing bulk throughput.
2384  *      this functionality is controlled by the InterruptThrottleRate module
2385  *      parameter (see e1000_param.c)
2386  * @adapter: pointer to adapter
2387  * @itr_setting: current adapter->itr
2388  * @packets: the number of packets during this measurement interval
2389  * @bytes: the number of bytes during this measurement interval
2390  **/
2391 static unsigned int e1000_update_itr(struct e1000_adapter *adapter,
2392                                      u16 itr_setting, int packets, int bytes)
2393 {
2394         unsigned int retval = itr_setting;
2395         struct e1000_hw *hw = &adapter->hw;
2396
2397         if (unlikely(hw->mac_type < e1000_82540))
2398                 goto update_itr_done;
2399
2400         if (packets == 0)
2401                 goto update_itr_done;
2402
2403         switch (itr_setting) {
2404         case lowest_latency:
2405                 /* jumbo frames get bulk treatment*/
2406                 if (bytes/packets > 8000)
2407                         retval = bulk_latency;
2408                 else if ((packets < 5) && (bytes > 512))
2409                         retval = low_latency;
2410                 break;
2411         case low_latency:  /* 50 usec aka 20000 ints/s */
2412                 if (bytes > 10000) {
2413                         /* jumbo frames need bulk latency setting */
2414                         if (bytes/packets > 8000)
2415                                 retval = bulk_latency;
2416                         else if ((packets < 10) || ((bytes/packets) > 1200))
2417                                 retval = bulk_latency;
2418                         else if ((packets > 35))
2419                                 retval = lowest_latency;
2420                 } else if (bytes/packets > 2000)
2421                         retval = bulk_latency;
2422                 else if (packets <= 2 && bytes < 512)
2423                         retval = lowest_latency;
2424                 break;
2425         case bulk_latency: /* 250 usec aka 4000 ints/s */
2426                 if (bytes > 25000) {
2427                         if (packets > 35)
2428                                 retval = low_latency;
2429                 } else if (bytes < 6000) {
2430                         retval = low_latency;
2431                 }
2432                 break;
2433         }
2434
2435 update_itr_done:
2436         return retval;
2437 }
2438
2439 static void e1000_set_itr(struct e1000_adapter *adapter)
2440 {
2441         struct e1000_hw *hw = &adapter->hw;
2442         u16 current_itr;
2443         u32 new_itr = adapter->itr;
2444
2445         if (unlikely(hw->mac_type < e1000_82540))
2446                 return;
2447
2448         /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
2449         if (unlikely(adapter->link_speed != SPEED_1000)) {
2450                 current_itr = 0;
2451                 new_itr = 4000;
2452                 goto set_itr_now;
2453         }
2454
2455         adapter->tx_itr = e1000_update_itr(adapter,
2456                                     adapter->tx_itr,
2457                                     adapter->total_tx_packets,
2458                                     adapter->total_tx_bytes);
2459         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2460         if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
2461                 adapter->tx_itr = low_latency;
2462
2463         adapter->rx_itr = e1000_update_itr(adapter,
2464                                     adapter->rx_itr,
2465                                     adapter->total_rx_packets,
2466                                     adapter->total_rx_bytes);
2467         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2468         if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
2469                 adapter->rx_itr = low_latency;
2470
2471         current_itr = max(adapter->rx_itr, adapter->tx_itr);
2472
2473         switch (current_itr) {
2474         /* counts and packets in update_itr are dependent on these numbers */
2475         case lowest_latency:
2476                 new_itr = 70000;
2477                 break;
2478         case low_latency:
2479                 new_itr = 20000; /* aka hwitr = ~200 */
2480                 break;
2481         case bulk_latency:
2482                 new_itr = 4000;
2483                 break;
2484         default:
2485                 break;
2486         }
2487
2488 set_itr_now:
2489         if (new_itr != adapter->itr) {
2490                 /* this attempts to bias the interrupt rate towards Bulk
2491                  * by adding intermediate steps when interrupt rate is
2492                  * increasing */
2493                 new_itr = new_itr > adapter->itr ?
2494                              min(adapter->itr + (new_itr >> 2), new_itr) :
2495                              new_itr;
2496                 adapter->itr = new_itr;
2497                 ew32(ITR, 1000000000 / (new_itr * 256));
2498         }
2499
2500         return;
2501 }
2502
2503 #define E1000_TX_FLAGS_CSUM             0x00000001
2504 #define E1000_TX_FLAGS_VLAN             0x00000002
2505 #define E1000_TX_FLAGS_TSO              0x00000004
2506 #define E1000_TX_FLAGS_IPV4             0x00000008
2507 #define E1000_TX_FLAGS_VLAN_MASK        0xffff0000
2508 #define E1000_TX_FLAGS_VLAN_SHIFT       16
2509
2510 static int e1000_tso(struct e1000_adapter *adapter,
2511                      struct e1000_tx_ring *tx_ring, struct sk_buff *skb)
2512 {
2513         struct e1000_context_desc *context_desc;
2514         struct e1000_buffer *buffer_info;
2515         unsigned int i;
2516         u32 cmd_length = 0;
2517         u16 ipcse = 0, tucse, mss;
2518         u8 ipcss, ipcso, tucss, tucso, hdr_len;
2519         int err;
2520
2521         if (skb_is_gso(skb)) {
2522                 if (skb_header_cloned(skb)) {
2523                         err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2524                         if (err)
2525                                 return err;
2526                 }
2527
2528                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
2529                 mss = skb_shinfo(skb)->gso_size;
2530                 if (skb->protocol == htons(ETH_P_IP)) {
2531                         struct iphdr *iph = ip_hdr(skb);
2532                         iph->tot_len = 0;
2533                         iph->check = 0;
2534                         tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
2535                                                                  iph->daddr, 0,
2536                                                                  IPPROTO_TCP,
2537                                                                  0);
2538                         cmd_length = E1000_TXD_CMD_IP;
2539                         ipcse = skb_transport_offset(skb) - 1;
2540                 } else if (skb->protocol == htons(ETH_P_IPV6)) {
2541                         ipv6_hdr(skb)->payload_len = 0;
2542                         tcp_hdr(skb)->check =
2543                                 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
2544                                                  &ipv6_hdr(skb)->daddr,
2545                                                  0, IPPROTO_TCP, 0);
2546                         ipcse = 0;
2547                 }
2548                 ipcss = skb_network_offset(skb);
2549                 ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
2550                 tucss = skb_transport_offset(skb);
2551                 tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
2552                 tucse = 0;
2553
2554                 cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
2555                                E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
2556
2557                 i = tx_ring->next_to_use;
2558                 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
2559                 buffer_info = &tx_ring->buffer_info[i];
2560
2561                 context_desc->lower_setup.ip_fields.ipcss  = ipcss;
2562                 context_desc->lower_setup.ip_fields.ipcso  = ipcso;
2563                 context_desc->lower_setup.ip_fields.ipcse  = cpu_to_le16(ipcse);
2564                 context_desc->upper_setup.tcp_fields.tucss = tucss;
2565                 context_desc->upper_setup.tcp_fields.tucso = tucso;
2566                 context_desc->upper_setup.tcp_fields.tucse = cpu_to_le16(tucse);
2567                 context_desc->tcp_seg_setup.fields.mss     = cpu_to_le16(mss);
2568                 context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
2569                 context_desc->cmd_and_length = cpu_to_le32(cmd_length);
2570
2571                 buffer_info->time_stamp = jiffies;
2572                 buffer_info->next_to_watch = i;
2573
2574                 if (++i == tx_ring->count) i = 0;
2575                 tx_ring->next_to_use = i;
2576
2577                 return true;
2578         }
2579         return false;
2580 }
2581
2582 static bool e1000_tx_csum(struct e1000_adapter *adapter,
2583                           struct e1000_tx_ring *tx_ring, struct sk_buff *skb)
2584 {
2585         struct e1000_context_desc *context_desc;
2586         struct e1000_buffer *buffer_info;
2587         unsigned int i;
2588         u8 css;
2589         u32 cmd_len = E1000_TXD_CMD_DEXT;
2590
2591         if (skb->ip_summed != CHECKSUM_PARTIAL)
2592                 return false;
2593
2594         switch (skb->protocol) {
2595         case cpu_to_be16(ETH_P_IP):
2596                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
2597                         cmd_len |= E1000_TXD_CMD_TCP;
2598                 break;
2599         case cpu_to_be16(ETH_P_IPV6):
2600                 /* XXX not handling all IPV6 headers */
2601                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
2602                         cmd_len |= E1000_TXD_CMD_TCP;
2603                 break;
2604         default:
2605                 if (unlikely(net_ratelimit()))
2606                         DPRINTK(DRV, WARNING,
2607                                 "checksum_partial proto=%x!\n", skb->protocol);
2608                 break;
2609         }
2610
2611         css = skb_transport_offset(skb);
2612
2613         i = tx_ring->next_to_use;
2614         buffer_info = &tx_ring->buffer_info[i];
2615         context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
2616
2617         context_desc->lower_setup.ip_config = 0;
2618         context_desc->upper_setup.tcp_fields.tucss = css;
2619         context_desc->upper_setup.tcp_fields.tucso =
2620                 css + skb->csum_offset;
2621         context_desc->upper_setup.tcp_fields.tucse = 0;
2622         context_desc->tcp_seg_setup.data = 0;
2623         context_desc->cmd_and_length = cpu_to_le32(cmd_len);
2624
2625         buffer_info->time_stamp = jiffies;
2626         buffer_info->next_to_watch = i;
2627
2628         if (unlikely(++i == tx_ring->count)) i = 0;
2629         tx_ring->next_to_use = i;
2630
2631         return true;
2632 }
2633
2634 #define E1000_MAX_TXD_PWR       12
2635 #define E1000_MAX_DATA_PER_TXD  (1<<E1000_MAX_TXD_PWR)
2636
2637 static int e1000_tx_map(struct e1000_adapter *adapter,
2638                         struct e1000_tx_ring *tx_ring,
2639                         struct sk_buff *skb, unsigned int first,
2640                         unsigned int max_per_txd, unsigned int nr_frags,
2641                         unsigned int mss)
2642 {
2643         struct e1000_hw *hw = &adapter->hw;
2644         struct e1000_buffer *buffer_info;
2645         unsigned int len = skb_headlen(skb);
2646         unsigned int offset, size, count = 0, i;
2647         unsigned int f;
2648         dma_addr_t *map;
2649
2650         i = tx_ring->next_to_use;
2651
2652         if (skb_dma_map(&adapter->pdev->dev, skb, DMA_TO_DEVICE)) {
2653                 dev_err(&adapter->pdev->dev, "TX DMA map failed\n");
2654                 return 0;
2655         }
2656
2657         map = skb_shinfo(skb)->dma_maps;
2658         offset = 0;
2659
2660         while (len) {
2661                 buffer_info = &tx_ring->buffer_info[i];
2662                 size = min(len, max_per_txd);
2663                 /* Workaround for Controller erratum --
2664                  * descriptor for non-tso packet in a linear SKB that follows a
2665                  * tso gets written back prematurely before the data is fully
2666                  * DMA'd to the controller */
2667                 if (!skb->data_len && tx_ring->last_tx_tso &&
2668                     !skb_is_gso(skb)) {
2669                         tx_ring->last_tx_tso = 0;
2670                         size -= 4;
2671                 }
2672
2673                 /* Workaround for premature desc write-backs
2674                  * in TSO mode.  Append 4-byte sentinel desc */
2675                 if (unlikely(mss && !nr_frags && size == len && size > 8))
2676                         size -= 4;
2677                 /* work-around for errata 10 and it applies
2678                  * to all controllers in PCI-X mode
2679                  * The fix is to make sure that the first descriptor of a
2680                  * packet is smaller than 2048 - 16 - 16 (or 2016) bytes
2681                  */
2682                 if (unlikely((hw->bus_type == e1000_bus_type_pcix) &&
2683                                 (size > 2015) && count == 0))
2684                         size = 2015;
2685
2686                 /* Workaround for potential 82544 hang in PCI-X.  Avoid
2687                  * terminating buffers within evenly-aligned dwords. */
2688                 if (unlikely(adapter->pcix_82544 &&
2689                    !((unsigned long)(skb->data + offset + size - 1) & 4) &&
2690                    size > 4))
2691                         size -= 4;
2692
2693                 buffer_info->length = size;
2694                 buffer_info->dma = skb_shinfo(skb)->dma_head + offset;
2695                 buffer_info->time_stamp = jiffies;
2696                 buffer_info->next_to_watch = i;
2697
2698                 len -= size;
2699                 offset += size;
2700                 count++;
2701                 if (len) {
2702                         i++;
2703                         if (unlikely(i == tx_ring->count))
2704                                 i = 0;
2705                 }
2706         }
2707
2708         for (f = 0; f < nr_frags; f++) {
2709                 struct skb_frag_struct *frag;
2710
2711                 frag = &skb_shinfo(skb)->frags[f];
2712                 len = frag->size;
2713                 offset = 0;
2714
2715                 while (len) {
2716                         i++;
2717                         if (unlikely(i == tx_ring->count))
2718                                 i = 0;
2719
2720                         buffer_info = &tx_ring->buffer_info[i];
2721                         size = min(len, max_per_txd);
2722                         /* Workaround for premature desc write-backs
2723                          * in TSO mode.  Append 4-byte sentinel desc */
2724                         if (unlikely(mss && f == (nr_frags-1) && size == len && size > 8))
2725                                 size -= 4;
2726                         /* Workaround for potential 82544 hang in PCI-X.
2727                          * Avoid terminating buffers within evenly-aligned
2728                          * dwords. */
2729                         if (unlikely(adapter->pcix_82544 &&
2730                            !((unsigned long)(frag->page+offset+size-1) & 4) &&
2731                            size > 4))
2732                                 size -= 4;
2733
2734                         buffer_info->length = size;
2735                         buffer_info->dma = map[f] + offset;
2736                         buffer_info->time_stamp = jiffies;
2737                         buffer_info->next_to_watch = i;
2738
2739                         len -= size;
2740                         offset += size;
2741                         count++;
2742                 }
2743         }
2744
2745         tx_ring->buffer_info[i].skb = skb;
2746         tx_ring->buffer_info[first].next_to_watch = i;
2747
2748         return count;
2749 }
2750
2751 static void e1000_tx_queue(struct e1000_adapter *adapter,
2752                            struct e1000_tx_ring *tx_ring, int tx_flags,
2753                            int count)
2754 {
2755         struct e1000_hw *hw = &adapter->hw;
2756         struct e1000_tx_desc *tx_desc = NULL;
2757         struct e1000_buffer *buffer_info;
2758         u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
2759         unsigned int i;
2760
2761         if (likely(tx_flags & E1000_TX_FLAGS_TSO)) {
2762                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
2763                              E1000_TXD_CMD_TSE;
2764                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
2765
2766                 if (likely(tx_flags & E1000_TX_FLAGS_IPV4))
2767                         txd_upper |= E1000_TXD_POPTS_IXSM << 8;
2768         }
2769
2770         if (likely(tx_flags & E1000_TX_FLAGS_CSUM)) {
2771                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
2772                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
2773         }
2774
2775         if (unlikely(tx_flags & E1000_TX_FLAGS_VLAN)) {
2776                 txd_lower |= E1000_TXD_CMD_VLE;
2777                 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
2778         }
2779
2780         i = tx_ring->next_to_use;
2781
2782         while (count--) {
2783                 buffer_info = &tx_ring->buffer_info[i];
2784                 tx_desc = E1000_TX_DESC(*tx_ring, i);
2785                 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
2786                 tx_desc->lower.data =
2787                         cpu_to_le32(txd_lower | buffer_info->length);
2788                 tx_desc->upper.data = cpu_to_le32(txd_upper);
2789                 if (unlikely(++i == tx_ring->count)) i = 0;
2790         }
2791
2792         tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
2793
2794         /* Force memory writes to complete before letting h/w
2795          * know there are new descriptors to fetch.  (Only
2796          * applicable for weak-ordered memory model archs,
2797          * such as IA-64). */
2798         wmb();
2799
2800         tx_ring->next_to_use = i;
2801         writel(i, hw->hw_addr + tx_ring->tdt);
2802         /* we need this if more than one processor can write to our tail
2803          * at a time, it syncronizes IO on IA64/Altix systems */
2804         mmiowb();
2805 }
2806
2807 /**
2808  * 82547 workaround to avoid controller hang in half-duplex environment.
2809  * The workaround is to avoid queuing a large packet that would span
2810  * the internal Tx FIFO ring boundary by notifying the stack to resend
2811  * the packet at a later time.  This gives the Tx FIFO an opportunity to
2812  * flush all packets.  When that occurs, we reset the Tx FIFO pointers
2813  * to the beginning of the Tx FIFO.
2814  **/
2815
2816 #define E1000_FIFO_HDR                  0x10
2817 #define E1000_82547_PAD_LEN             0x3E0
2818
2819 static int e1000_82547_fifo_workaround(struct e1000_adapter *adapter,
2820                                        struct sk_buff *skb)
2821 {
2822         u32 fifo_space = adapter->tx_fifo_size - adapter->tx_fifo_head;
2823         u32 skb_fifo_len = skb->len + E1000_FIFO_HDR;
2824
2825         skb_fifo_len = ALIGN(skb_fifo_len, E1000_FIFO_HDR);
2826
2827         if (adapter->link_duplex != HALF_DUPLEX)
2828                 goto no_fifo_stall_required;
2829
2830         if (atomic_read(&adapter->tx_fifo_stall))
2831                 return 1;
2832
2833         if (skb_fifo_len >= (E1000_82547_PAD_LEN + fifo_space)) {
2834                 atomic_set(&adapter->tx_fifo_stall, 1);
2835                 return 1;
2836         }
2837
2838 no_fifo_stall_required:
2839         adapter->tx_fifo_head += skb_fifo_len;
2840         if (adapter->tx_fifo_head >= adapter->tx_fifo_size)
2841                 adapter->tx_fifo_head -= adapter->tx_fifo_size;
2842         return 0;
2843 }
2844
2845 static int __e1000_maybe_stop_tx(struct net_device *netdev, int size)
2846 {
2847         struct e1000_adapter *adapter = netdev_priv(netdev);
2848         struct e1000_tx_ring *tx_ring = adapter->tx_ring;
2849
2850         netif_stop_queue(netdev);
2851         /* Herbert's original patch had:
2852          *  smp_mb__after_netif_stop_queue();
2853          * but since that doesn't exist yet, just open code it. */
2854         smp_mb();
2855
2856         /* We need to check again in a case another CPU has just
2857          * made room available. */
2858         if (likely(E1000_DESC_UNUSED(tx_ring) < size))
2859                 return -EBUSY;
2860
2861         /* A reprieve! */
2862         netif_start_queue(netdev);
2863         ++adapter->restart_queue;
2864         return 0;
2865 }
2866
2867 static int e1000_maybe_stop_tx(struct net_device *netdev,
2868                                struct e1000_tx_ring *tx_ring, int size)
2869 {
2870         if (likely(E1000_DESC_UNUSED(tx_ring) >= size))
2871                 return 0;
2872         return __e1000_maybe_stop_tx(netdev, size);
2873 }
2874
2875 #define TXD_USE_COUNT(S, X) (((S) >> (X)) + 1 )
2876 static netdev_tx_t e1000_xmit_frame(struct sk_buff *skb,
2877                                     struct net_device *netdev)
2878 {
2879         struct e1000_adapter *adapter = netdev_priv(netdev);
2880         struct e1000_hw *hw = &adapter->hw;
2881         struct e1000_tx_ring *tx_ring;
2882         unsigned int first, max_per_txd = E1000_MAX_DATA_PER_TXD;
2883         unsigned int max_txd_pwr = E1000_MAX_TXD_PWR;
2884         unsigned int tx_flags = 0;
2885         unsigned int len = skb->len - skb->data_len;
2886         unsigned int nr_frags;
2887         unsigned int mss;
2888         int count = 0;
2889         int tso;
2890         unsigned int f;
2891
2892         /* This goes back to the question of how to logically map a tx queue
2893          * to a flow.  Right now, performance is impacted slightly negatively
2894          * if using multiple tx queues.  If the stack breaks away from a
2895          * single qdisc implementation, we can look at this again. */
2896         tx_ring = adapter->tx_ring;
2897
2898         if (unlikely(skb->len <= 0)) {
2899                 dev_kfree_skb_any(skb);
2900                 return NETDEV_TX_OK;
2901         }
2902
2903         mss = skb_shinfo(skb)->gso_size;
2904         /* The controller does a simple calculation to
2905          * make sure there is enough room in the FIFO before
2906          * initiating the DMA for each buffer.  The calc is:
2907          * 4 = ceil(buffer len/mss).  To make sure we don't
2908          * overrun the FIFO, adjust the max buffer len if mss
2909          * drops. */
2910         if (mss) {
2911                 u8 hdr_len;
2912                 max_per_txd = min(mss << 2, max_per_txd);
2913                 max_txd_pwr = fls(max_per_txd) - 1;
2914
2915                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
2916                 if (skb->data_len && hdr_len == len) {
2917                         switch (hw->mac_type) {
2918                                 unsigned int pull_size;
2919                         case e1000_82544:
2920                                 /* Make sure we have room to chop off 4 bytes,
2921                                  * and that the end alignment will work out to
2922                                  * this hardware's requirements
2923                                  * NOTE: this is a TSO only workaround
2924                                  * if end byte alignment not correct move us
2925                                  * into the next dword */
2926                                 if ((unsigned long)(skb_tail_pointer(skb) - 1) & 4)
2927                                         break;
2928                                 /* fall through */
2929                                 pull_size = min((unsigned int)4, skb->data_len);
2930                                 if (!__pskb_pull_tail(skb, pull_size)) {
2931                                         DPRINTK(DRV, ERR,
2932                                                 "__pskb_pull_tail failed.\n");
2933                                         dev_kfree_skb_any(skb);
2934                                         return NETDEV_TX_OK;
2935                                 }
2936                                 len = skb->len - skb->data_len;
2937                                 break;
2938                         default:
2939                                 /* do nothing */
2940                                 break;
2941                         }
2942                 }
2943         }
2944
2945         /* reserve a descriptor for the offload context */
2946         if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
2947                 count++;
2948         count++;
2949
2950         /* Controller Erratum workaround */
2951         if (!skb->data_len && tx_ring->last_tx_tso && !skb_is_gso(skb))
2952                 count++;
2953
2954         count += TXD_USE_COUNT(len, max_txd_pwr);
2955
2956         if (adapter->pcix_82544)
2957                 count++;
2958
2959         /* work-around for errata 10 and it applies to all controllers
2960          * in PCI-X mode, so add one more descriptor to the count
2961          */
2962         if (unlikely((hw->bus_type == e1000_bus_type_pcix) &&
2963                         (len > 2015)))
2964                 count++;
2965
2966         nr_frags = skb_shinfo(skb)->nr_frags;
2967         for (f = 0; f < nr_frags; f++)
2968                 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size,
2969                                        max_txd_pwr);
2970         if (adapter->pcix_82544)
2971                 count += nr_frags;
2972
2973         /* need: count + 2 desc gap to keep tail from touching
2974          * head, otherwise try next time */
2975         if (unlikely(e1000_maybe_stop_tx(netdev, tx_ring, count + 2)))
2976                 return NETDEV_TX_BUSY;
2977
2978         if (unlikely(hw->mac_type == e1000_82547)) {
2979                 if (unlikely(e1000_82547_fifo_workaround(adapter, skb))) {
2980                         netif_stop_queue(netdev);
2981                         mod_timer(&adapter->tx_fifo_stall_timer, jiffies + 1);
2982                         return NETDEV_TX_BUSY;
2983                 }
2984         }
2985
2986         if (unlikely(adapter->vlgrp && vlan_tx_tag_present(skb))) {
2987                 tx_flags |= E1000_TX_FLAGS_VLAN;
2988                 tx_flags |= (vlan_tx_tag_get(skb) << E1000_TX_FLAGS_VLAN_SHIFT);
2989         }
2990
2991         first = tx_ring->next_to_use;
2992
2993         tso = e1000_tso(adapter, tx_ring, skb);
2994         if (tso < 0) {
2995                 dev_kfree_skb_any(skb);
2996                 return NETDEV_TX_OK;
2997         }
2998
2999         if (likely(tso)) {
3000                 tx_ring->last_tx_tso = 1;
3001                 tx_flags |= E1000_TX_FLAGS_TSO;
3002         } else if (likely(e1000_tx_csum(adapter, tx_ring, skb)))
3003                 tx_flags |= E1000_TX_FLAGS_CSUM;
3004
3005         if (likely(skb->protocol == htons(ETH_P_IP)))
3006                 tx_flags |= E1000_TX_FLAGS_IPV4;
3007
3008         count = e1000_tx_map(adapter, tx_ring, skb, first, max_per_txd,
3009                              nr_frags, mss);
3010
3011         if (count) {
3012                 e1000_tx_queue(adapter, tx_ring, tx_flags, count);
3013                 /* Make sure there is space in the ring for the next send. */
3014                 e1000_maybe_stop_tx(netdev, tx_ring, MAX_SKB_FRAGS + 2);
3015
3016         } else {
3017                 dev_kfree_skb_any(skb);
3018                 tx_ring->buffer_info[first].time_stamp = 0;
3019                 tx_ring->next_to_use = first;
3020         }
3021
3022         return NETDEV_TX_OK;
3023 }
3024
3025 /**
3026  * e1000_tx_timeout - Respond to a Tx Hang
3027  * @netdev: network interface device structure
3028  **/
3029
3030 static void e1000_tx_timeout(struct net_device *netdev)
3031 {
3032         struct e1000_adapter *adapter = netdev_priv(netdev);
3033
3034         /* Do the reset outside of interrupt context */
3035         adapter->tx_timeout_count++;
3036         schedule_work(&adapter->reset_task);
3037 }
3038
3039 static void e1000_reset_task(struct work_struct *work)
3040 {
3041         struct e1000_adapter *adapter =
3042                 container_of(work, struct e1000_adapter, reset_task);
3043
3044         e1000_reinit_locked(adapter);
3045 }
3046
3047 /**
3048  * e1000_get_stats - Get System Network Statistics
3049  * @netdev: network interface device structure
3050  *
3051  * Returns the address of the device statistics structure.
3052  * The statistics are actually updated from the timer callback.
3053  **/
3054
3055 static struct net_device_stats *e1000_get_stats(struct net_device *netdev)
3056 {
3057         struct e1000_adapter *adapter = netdev_priv(netdev);
3058
3059         /* only return the current stats */
3060         return &adapter->net_stats;
3061 }
3062
3063 /**
3064  * e1000_change_mtu - Change the Maximum Transfer Unit
3065  * @netdev: network interface device structure
3066  * @new_mtu: new value for maximum frame size
3067  *
3068  * Returns 0 on success, negative on failure
3069  **/
3070
3071 static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
3072 {
3073         struct e1000_adapter *adapter = netdev_priv(netdev);
3074         struct e1000_hw *hw = &adapter->hw;
3075         int max_frame = new_mtu + ENET_HEADER_SIZE + ETHERNET_FCS_SIZE;
3076
3077         if ((max_frame < MINIMUM_ETHERNET_FRAME_SIZE) ||
3078             (max_frame > MAX_JUMBO_FRAME_SIZE)) {
3079                 DPRINTK(PROBE, ERR, "Invalid MTU setting\n");
3080                 return -EINVAL;
3081         }
3082
3083         /* Adapter-specific max frame size limits. */
3084         switch (hw->mac_type) {
3085         case e1000_undefined ... e1000_82542_rev2_1:
3086                 if (max_frame > (ETH_FRAME_LEN + ETH_FCS_LEN)) {
3087                         DPRINTK(PROBE, ERR, "Jumbo Frames not supported.\n");
3088                         return -EINVAL;
3089                 }
3090                 break;
3091         default:
3092                 /* Capable of supporting up to MAX_JUMBO_FRAME_SIZE limit. */
3093                 break;
3094         }
3095
3096         /* NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
3097          * means we reserve 2 more, this pushes us to allocate from the next
3098          * larger slab size.
3099          * i.e. RXBUFFER_2048 --> size-4096 slab
3100          *  however with the new *_jumbo_rx* routines, jumbo receives will use
3101          *  fragmented skbs */
3102
3103         if (max_frame <= E1000_RXBUFFER_256)
3104                 adapter->rx_buffer_len = E1000_RXBUFFER_256;
3105         else if (max_frame <= E1000_RXBUFFER_512)
3106                 adapter->rx_buffer_len = E1000_RXBUFFER_512;
3107         else if (max_frame <= E1000_RXBUFFER_1024)
3108                 adapter->rx_buffer_len = E1000_RXBUFFER_1024;
3109         else if (max_frame <= E1000_RXBUFFER_2048)
3110                 adapter->rx_buffer_len = E1000_RXBUFFER_2048;
3111         else
3112 #if (PAGE_SIZE >= E1000_RXBUFFER_16384)
3113                 adapter->rx_buffer_len = E1000_RXBUFFER_16384;
3114 #elif (PAGE_SIZE >= E1000_RXBUFFER_4096)
3115                 adapter->rx_buffer_len = PAGE_SIZE;
3116 #endif
3117
3118         /* adjust allocation if LPE protects us, and we aren't using SBP */
3119         if (!hw->tbi_compatibility_on &&
3120             ((max_frame == (ETH_FRAME_LEN + ETH_FCS_LEN)) ||
3121              (max_frame == MAXIMUM_ETHERNET_VLAN_SIZE)))
3122                 adapter->rx_buffer_len = MAXIMUM_ETHERNET_VLAN_SIZE;
3123
3124         netdev->mtu = new_mtu;
3125         hw->max_frame_size = max_frame;
3126
3127         if (netif_running(netdev))
3128                 e1000_reinit_locked(adapter);
3129
3130         return 0;
3131 }
3132
3133 /**
3134  * e1000_update_stats - Update the board statistics counters
3135  * @adapter: board private structure
3136  **/
3137
3138 void e1000_update_stats(struct e1000_adapter *adapter)
3139 {
3140         struct e1000_hw *hw = &adapter->hw;
3141         struct pci_dev *pdev = adapter->pdev;
3142         unsigned long flags;
3143         u16 phy_tmp;
3144
3145 #define PHY_IDLE_ERROR_COUNT_MASK 0x00FF
3146
3147         /*
3148          * Prevent stats update while adapter is being reset, or if the pci
3149          * connection is down.
3150          */
3151         if (adapter->link_speed == 0)
3152                 return;
3153         if (pci_channel_offline(pdev))
3154                 return;
3155
3156         spin_lock_irqsave(&adapter->stats_lock, flags);
3157
3158         /* these counters are modified from e1000_tbi_adjust_stats,
3159          * called from the interrupt context, so they must only
3160          * be written while holding adapter->stats_lock
3161          */
3162
3163         adapter->stats.crcerrs += er32(CRCERRS);
3164         adapter->stats.gprc += er32(GPRC);
3165         adapter->stats.gorcl += er32(GORCL);
3166         adapter->stats.gorch += er32(GORCH);
3167         adapter->stats.bprc += er32(BPRC);
3168         adapter->stats.mprc += er32(MPRC);
3169         adapter->stats.roc += er32(ROC);
3170
3171         adapter->stats.prc64 += er32(PRC64);
3172         adapter->stats.prc127 += er32(PRC127);
3173         adapter->stats.prc255 += er32(PRC255);
3174         adapter->stats.prc511 += er32(PRC511);
3175         adapter->stats.prc1023 += er32(PRC1023);
3176         adapter->stats.prc1522 += er32(PRC1522);
3177
3178         adapter->stats.symerrs += er32(SYMERRS);
3179         adapter->stats.mpc += er32(MPC);
3180         adapter->stats.scc += er32(SCC);
3181         adapter->stats.ecol += er32(ECOL);
3182         adapter->stats.mcc += er32(MCC);
3183         adapter->stats.latecol += er32(LATECOL);
3184         adapter->stats.dc += er32(DC);
3185         adapter->stats.sec += er32(SEC);
3186         adapter->stats.rlec += er32(RLEC);
3187         adapter->stats.xonrxc += er32(XONRXC);
3188         adapter->stats.xontxc += er32(XONTXC);
3189         adapter->stats.xoffrxc += er32(XOFFRXC);
3190         adapter->stats.xofftxc += er32(XOFFTXC);
3191         adapter->stats.fcruc += er32(FCRUC);
3192         adapter->stats.gptc += er32(GPTC);
3193         adapter->stats.gotcl += er32(GOTCL);
3194         adapter->stats.gotch += er32(GOTCH);
3195         adapter->stats.rnbc += er32(RNBC);
3196         adapter->stats.ruc += er32(RUC);
3197         adapter->stats.rfc += er32(RFC);
3198         adapter->stats.rjc += er32(RJC);
3199         adapter->stats.torl += er32(TORL);
3200         adapter->stats.torh += er32(TORH);
3201         adapter->stats.totl += er32(TOTL);
3202         adapter->stats.toth += er32(TOTH);
3203         adapter->stats.tpr += er32(TPR);
3204
3205         adapter->stats.ptc64 += er32(PTC64);
3206         adapter->stats.ptc127 += er32(PTC127);
3207         adapter->stats.ptc255 += er32(PTC255);
3208         adapter->stats.ptc511 += er32(PTC511);
3209         adapter->stats.ptc1023 += er32(PTC1023);
3210         adapter->stats.ptc1522 += er32(PTC1522);
3211
3212         adapter->stats.mptc += er32(MPTC);
3213         adapter->stats.bptc += er32(BPTC);
3214
3215         /* used for adaptive IFS */
3216
3217         hw->tx_packet_delta = er32(TPT);
3218         adapter->stats.tpt += hw->tx_packet_delta;
3219         hw->collision_delta = er32(COLC);
3220         adapter->stats.colc += hw->collision_delta;
3221
3222         if (hw->mac_type >= e1000_82543) {
3223                 adapter->stats.algnerrc += er32(ALGNERRC);
3224                 adapter->stats.rxerrc += er32(RXERRC);
3225                 adapter->stats.tncrs += er32(TNCRS);
3226                 adapter->stats.cexterr += er32(CEXTERR);
3227                 adapter->stats.tsctc += er32(TSCTC);
3228                 adapter->stats.tsctfc += er32(TSCTFC);
3229         }
3230
3231         /* Fill out the OS statistics structure */
3232         adapter->net_stats.multicast = adapter->stats.mprc;
3233         adapter->net_stats.collisions = adapter->stats.colc;
3234
3235         /* Rx Errors */
3236
3237         /* RLEC on some newer hardware can be incorrect so build
3238         * our own version based on RUC and ROC */
3239         adapter->net_stats.rx_errors = adapter->stats.rxerrc +
3240                 adapter->stats.crcerrs + adapter->stats.algnerrc +
3241                 adapter->stats.ruc + adapter->stats.roc +
3242                 adapter->stats.cexterr;
3243         adapter->stats.rlerrc = adapter->stats.ruc + adapter->stats.roc;
3244         adapter->net_stats.rx_length_errors = adapter->stats.rlerrc;
3245         adapter->net_stats.rx_crc_errors = adapter->stats.crcerrs;
3246         adapter->net_stats.rx_frame_errors = adapter->stats.algnerrc;
3247         adapter->net_stats.rx_missed_errors = adapter->stats.mpc;
3248
3249         /* Tx Errors */
3250         adapter->stats.txerrc = adapter->stats.ecol + adapter->stats.latecol;
3251         adapter->net_stats.tx_errors = adapter->stats.txerrc;
3252         adapter->net_stats.tx_aborted_errors = adapter->stats.ecol;
3253         adapter->net_stats.tx_window_errors = adapter->stats.latecol;
3254         adapter->net_stats.tx_carrier_errors = adapter->stats.tncrs;
3255         if (hw->bad_tx_carr_stats_fd &&
3256             adapter->link_duplex == FULL_DUPLEX) {
3257                 adapter->net_stats.tx_carrier_errors = 0;
3258                 adapter->stats.tncrs = 0;
3259         }
3260
3261         /* Tx Dropped needs to be maintained elsewhere */
3262
3263         /* Phy Stats */
3264         if (hw->media_type == e1000_media_type_copper) {
3265                 if ((adapter->link_speed == SPEED_1000) &&
3266                    (!e1000_read_phy_reg(hw, PHY_1000T_STATUS, &phy_tmp))) {
3267                         phy_tmp &= PHY_IDLE_ERROR_COUNT_MASK;
3268                         adapter->phy_stats.idle_errors += phy_tmp;
3269                 }
3270
3271                 if ((hw->mac_type <= e1000_82546) &&
3272                    (hw->phy_type == e1000_phy_m88) &&
3273                    !e1000_read_phy_reg(hw, M88E1000_RX_ERR_CNTR, &phy_tmp))
3274                         adapter->phy_stats.receive_errors += phy_tmp;
3275         }
3276
3277         /* Management Stats */
3278         if (hw->has_smbus) {
3279                 adapter->stats.mgptc += er32(MGTPTC);
3280                 adapter->stats.mgprc += er32(MGTPRC);
3281                 adapter->stats.mgpdc += er32(MGTPDC);
3282         }
3283
3284         spin_unlock_irqrestore(&adapter->stats_lock, flags);
3285 }
3286
3287 /**
3288  * e1000_intr - Interrupt Handler
3289  * @irq: interrupt number
3290  * @data: pointer to a network interface device structure
3291  **/
3292
3293 static irqreturn_t e1000_intr(int irq, void *data)
3294 {
3295         struct net_device *netdev = data;
3296         struct e1000_adapter *adapter = netdev_priv(netdev);
3297         struct e1000_hw *hw = &adapter->hw;
3298         u32 icr = er32(ICR);
3299
3300         if (unlikely((!icr) || test_bit(__E1000_DOWN, &adapter->flags)))
3301                 return IRQ_NONE;  /* Not our interrupt */
3302
3303         if (unlikely(icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC))) {
3304                 hw->get_link_status = 1;
3305                 /* guard against interrupt when we're going down */
3306                 if (!test_bit(__E1000_DOWN, &adapter->flags))
3307                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
3308         }
3309
3310         /* disable interrupts, without the synchronize_irq bit */
3311         ew32(IMC, ~0);
3312         E1000_WRITE_FLUSH();
3313
3314         if (likely(napi_schedule_prep(&adapter->napi))) {
3315                 adapter->total_tx_bytes = 0;
3316                 adapter->total_tx_packets = 0;
3317                 adapter->total_rx_bytes = 0;
3318                 adapter->total_rx_packets = 0;
3319                 __napi_schedule(&adapter->napi);
3320         } else {
3321                 /* this really should not happen! if it does it is basically a
3322                  * bug, but not a hard error, so enable ints and continue */
3323                 if (!test_bit(__E1000_DOWN, &adapter->flags))
3324                         e1000_irq_enable(adapter);
3325         }
3326
3327         return IRQ_HANDLED;
3328 }
3329
3330 /**
3331  * e1000_clean - NAPI Rx polling callback
3332  * @adapter: board private structure
3333  **/
3334 static int e1000_clean(struct napi_struct *napi, int budget)
3335 {
3336         struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter, napi);
3337         struct net_device *poll_dev = adapter->netdev;
3338         int tx_cleaned = 0, work_done = 0;
3339
3340         adapter = netdev_priv(poll_dev);
3341
3342         tx_cleaned = e1000_clean_tx_irq(adapter, &adapter->tx_ring[0]);
3343
3344         adapter->clean_rx(adapter, &adapter->rx_ring[0],
3345                           &work_done, budget);
3346
3347         if (!tx_cleaned)
3348                 work_done = budget;
3349
3350         /* If budget not fully consumed, exit the polling mode */
3351         if (work_done < budget) {
3352                 if (likely(adapter->itr_setting & 3))
3353                         e1000_set_itr(adapter);
3354                 napi_complete(napi);
3355                 if (!test_bit(__E1000_DOWN, &adapter->flags))
3356                         e1000_irq_enable(adapter);
3357         }
3358
3359         return work_done;
3360 }
3361
3362 /**
3363  * e1000_clean_tx_irq - Reclaim resources after transmit completes
3364  * @adapter: board private structure
3365  **/
3366 static bool e1000_clean_tx_irq(struct e1000_adapter *adapter,
3367                                struct e1000_tx_ring *tx_ring)
3368 {
3369         struct e1000_hw *hw = &adapter->hw;
3370         struct net_device *netdev = adapter->netdev;
3371         struct e1000_tx_desc *tx_desc, *eop_desc;
3372         struct e1000_buffer *buffer_info;
3373         unsigned int i, eop;
3374         unsigned int count = 0;
3375         unsigned int total_tx_bytes=0, total_tx_packets=0;
3376
3377         i = tx_ring->next_to_clean;
3378         eop = tx_ring->buffer_info[i].next_to_watch;
3379         eop_desc = E1000_TX_DESC(*tx_ring, eop);
3380
3381         while ((eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) &&
3382                (count < tx_ring->count)) {
3383                 bool cleaned = false;
3384                 for ( ; !cleaned; count++) {
3385                         tx_desc = E1000_TX_DESC(*tx_ring, i);
3386                         buffer_info = &tx_ring->buffer_info[i];
3387                         cleaned = (i == eop);
3388
3389                         if (cleaned) {
3390                                 struct sk_buff *skb = buffer_info->skb;
3391                                 unsigned int segs, bytecount;
3392                                 segs = skb_shinfo(skb)->gso_segs ?: 1;
3393                                 /* multiply data chunks by size of headers */
3394                                 bytecount = ((segs - 1) * skb_headlen(skb)) +
3395                                             skb->len;
3396                                 total_tx_packets += segs;
3397                                 total_tx_bytes += bytecount;
3398                         }
3399                         e1000_unmap_and_free_tx_resource(adapter, buffer_info);
3400                         tx_desc->upper.data = 0;
3401
3402                         if (unlikely(++i == tx_ring->count)) i = 0;
3403                 }
3404
3405                 eop = tx_ring->buffer_info[i].next_to_watch;
3406                 eop_desc = E1000_TX_DESC(*tx_ring, eop);
3407         }
3408
3409         tx_ring->next_to_clean = i;
3410
3411 #define TX_WAKE_THRESHOLD 32
3412         if (unlikely(count && netif_carrier_ok(netdev) &&
3413                      E1000_DESC_UNUSED(tx_ring) >= TX_WAKE_THRESHOLD)) {
3414                 /* Make sure that anybody stopping the queue after this
3415                  * sees the new next_to_clean.
3416                  */
3417                 smp_mb();
3418                 if (netif_queue_stopped(netdev)) {
3419                         netif_wake_queue(netdev);
3420                         ++adapter->restart_queue;
3421                 }
3422         }
3423
3424         if (adapter->detect_tx_hung) {
3425                 /* Detect a transmit hang in hardware, this serializes the
3426                  * check with the clearing of time_stamp and movement of i */
3427                 adapter->detect_tx_hung = false;
3428                 if (tx_ring->buffer_info[i].time_stamp &&
3429                     time_after(jiffies, tx_ring->buffer_info[i].time_stamp +
3430                                (adapter->tx_timeout_factor * HZ))
3431                     && !(er32(STATUS) & E1000_STATUS_TXOFF)) {
3432
3433                         /* detected Tx unit hang */
3434                         DPRINTK(DRV, ERR, "Detected Tx Unit Hang\n"
3435                                         "  Tx Queue             <%lu>\n"
3436                                         "  TDH                  <%x>\n"
3437                                         "  TDT                  <%x>\n"
3438                                         "  next_to_use          <%x>\n"
3439                                         "  next_to_clean        <%x>\n"
3440                                         "buffer_info[next_to_clean]\n"
3441                                         "  time_stamp           <%lx>\n"
3442                                         "  next_to_watch        <%x>\n"
3443                                         "  jiffies              <%lx>\n"
3444                                         "  next_to_watch.status <%x>\n",
3445                                 (unsigned long)((tx_ring - adapter->tx_ring) /
3446                                         sizeof(struct e1000_tx_ring)),
3447                                 readl(hw->hw_addr + tx_ring->tdh),
3448                                 readl(hw->hw_addr + tx_ring->tdt),
3449                                 tx_ring->next_to_use,
3450                                 tx_ring->next_to_clean,
3451                                 tx_ring->buffer_info[i].time_stamp,
3452                                 eop,
3453                                 jiffies,
3454                                 eop_desc->upper.fields.status);
3455                         netif_stop_queue(netdev);
3456                 }
3457         }
3458         adapter->total_tx_bytes += total_tx_bytes;
3459         adapter->total_tx_packets += total_tx_packets;
3460         adapter->net_stats.tx_bytes += total_tx_bytes;
3461         adapter->net_stats.tx_packets += total_tx_packets;
3462         return (count < tx_ring->count);
3463 }
3464
3465 /**
3466  * e1000_rx_checksum - Receive Checksum Offload for 82543
3467  * @adapter:     board private structure
3468  * @status_err:  receive descriptor status and error fields
3469  * @csum:        receive descriptor csum field
3470  * @sk_buff:     socket buffer with received data
3471  **/
3472
3473 static void e1000_rx_checksum(struct e1000_adapter *adapter, u32 status_err,
3474                               u32 csum, struct sk_buff *skb)
3475 {
3476         struct e1000_hw *hw = &adapter->hw;
3477         u16 status = (u16)status_err;
3478         u8 errors = (u8)(status_err >> 24);
3479         skb->ip_summed = CHECKSUM_NONE;
3480
3481         /* 82543 or newer only */
3482         if (unlikely(hw->mac_type < e1000_82543)) return;
3483         /* Ignore Checksum bit is set */
3484         if (unlikely(status & E1000_RXD_STAT_IXSM)) return;
3485         /* TCP/UDP checksum error bit is set */
3486         if (unlikely(errors & E1000_RXD_ERR_TCPE)) {
3487                 /* let the stack verify checksum errors */
3488                 adapter->hw_csum_err++;
3489                 return;
3490         }
3491         /* TCP/UDP Checksum has not been calculated */
3492         if (!(status & E1000_RXD_STAT_TCPCS))
3493                 return;
3494
3495         /* It must be a TCP or UDP packet with a valid checksum */
3496         if (likely(status & E1000_RXD_STAT_TCPCS)) {
3497                 /* TCP checksum is good */
3498                 skb->ip_summed = CHECKSUM_UNNECESSARY;
3499         }
3500         adapter->hw_csum_good++;
3501 }
3502
3503 /**
3504  * e1000_consume_page - helper function
3505  **/
3506 static void e1000_consume_page(struct e1000_buffer *bi, struct sk_buff *skb,
3507                                u16 length)
3508 {
3509         bi->page = NULL;
3510         skb->len += length;
3511         skb->data_len += length;
3512         skb->truesize += length;
3513 }
3514
3515 /**
3516  * e1000_receive_skb - helper function to handle rx indications
3517  * @adapter: board private structure
3518  * @status: descriptor status field as written by hardware
3519  * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
3520  * @skb: pointer to sk_buff to be indicated to stack
3521  */
3522 static void e1000_receive_skb(struct e1000_adapter *adapter, u8 status,
3523                               __le16 vlan, struct sk_buff *skb)
3524 {
3525         if (unlikely(adapter->vlgrp && (status & E1000_RXD_STAT_VP))) {
3526                 vlan_hwaccel_receive_skb(skb, adapter->vlgrp,
3527                                          le16_to_cpu(vlan) &
3528                                          E1000_RXD_SPC_VLAN_MASK);
3529         } else {
3530                 netif_receive_skb(skb);
3531         }
3532 }
3533
3534 /**
3535  * e1000_clean_jumbo_rx_irq - Send received data up the network stack; legacy
3536  * @adapter: board private structure
3537  * @rx_ring: ring to clean
3538  * @work_done: amount of napi work completed this call
3539  * @work_to_do: max amount of work allowed for this call to do
3540  *
3541  * the return value indicates whether actual cleaning was done, there
3542  * is no guarantee that everything was cleaned
3543  */
3544 static bool e1000_clean_jumbo_rx_irq(struct e1000_adapter *adapter,
3545                                      struct e1000_rx_ring *rx_ring,
3546                                      int *work_done, int work_to_do)
3547 {
3548         struct e1000_hw *hw = &adapter->hw;
3549         struct net_device *netdev = adapter->netdev;
3550         struct pci_dev *pdev = adapter->pdev;
3551         struct e1000_rx_desc *rx_desc, *next_rxd;
3552         struct e1000_buffer *buffer_info, *next_buffer;
3553         unsigned long irq_flags;
3554         u32 length;
3555         unsigned int i;
3556         int cleaned_count = 0;
3557         bool cleaned = false;
3558         unsigned int total_rx_bytes=0, total_rx_packets=0;
3559
3560         i = rx_ring->next_to_clean;
3561         rx_desc = E1000_RX_DESC(*rx_ring, i);
3562         buffer_info = &rx_ring->buffer_info[i];
3563
3564         while (rx_desc->status & E1000_RXD_STAT_DD) {
3565                 struct sk_buff *skb;
3566                 u8 status;
3567
3568                 if (*work_done >= work_to_do)
3569                         break;
3570                 (*work_done)++;
3571
3572                 status = rx_desc->status;
3573                 skb = buffer_info->skb;
3574                 buffer_info->skb = NULL;
3575
3576                 if (++i == rx_ring->count) i = 0;
3577                 next_rxd = E1000_RX_DESC(*rx_ring, i);
3578                 prefetch(next_rxd);
3579
3580                 next_buffer = &rx_ring->buffer_info[i];
3581
3582                 cleaned = true;
3583                 cleaned_count++;
3584                 pci_unmap_page(pdev, buffer_info->dma, buffer_info->length,
3585                                PCI_DMA_FROMDEVICE);
3586                 buffer_info->dma = 0;
3587
3588                 length = le16_to_cpu(rx_desc->length);
3589
3590                 /* errors is only valid for DD + EOP descriptors */
3591                 if (unlikely((status & E1000_RXD_STAT_EOP) &&
3592                     (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK))) {
3593                         u8 last_byte = *(skb->data + length - 1);
3594                         if (TBI_ACCEPT(hw, status, rx_desc->errors, length,
3595                                        last_byte)) {
3596                                 spin_lock_irqsave(&adapter->stats_lock,
3597                                                   irq_flags);
3598                                 e1000_tbi_adjust_stats(hw, &adapter->stats,
3599                                                        length, skb->data);
3600                                 spin_unlock_irqrestore(&adapter->stats_lock,
3601                                                        irq_flags);
3602                                 length--;
3603                         } else {
3604                                 /* recycle both page and skb */
3605                                 buffer_info->skb = skb;
3606                                 /* an error means any chain goes out the window
3607                                  * too */
3608                                 if (rx_ring->rx_skb_top)
3609                                         dev_kfree_skb(rx_ring->rx_skb_top);
3610                                 rx_ring->rx_skb_top = NULL;
3611                                 goto next_desc;
3612                         }
3613                 }
3614
3615 #define rxtop rx_ring->rx_skb_top
3616                 if (!(status & E1000_RXD_STAT_EOP)) {
3617                         /* this descriptor is only the beginning (or middle) */
3618                         if (!rxtop) {
3619                                 /* this is the beginning of a chain */
3620                                 rxtop = skb;
3621                                 skb_fill_page_desc(rxtop, 0, buffer_info->page,
3622                                                    0, length);
3623                         } else {
3624                                 /* this is the middle of a chain */
3625                                 skb_fill_page_desc(rxtop,
3626                                     skb_shinfo(rxtop)->nr_frags,
3627                                     buffer_info->page, 0, length);
3628                                 /* re-use the skb, only consumed the page */
3629                                 buffer_info->skb = skb;
3630                         }
3631                         e1000_consume_page(buffer_info, rxtop, length);
3632                         goto next_desc;
3633                 } else {
3634                         if (rxtop) {
3635                                 /* end of the chain */
3636                                 skb_fill_page_desc(rxtop,
3637                                     skb_shinfo(rxtop)->nr_frags,
3638                                     buffer_info->page, 0, length);
3639                                 /* re-use the current skb, we only consumed the
3640                                  * page */
3641                                 buffer_info->skb = skb;
3642                                 skb = rxtop;
3643                                 rxtop = NULL;
3644                                 e1000_consume_page(buffer_info, skb, length);
3645                         } else {
3646                                 /* no chain, got EOP, this buf is the packet
3647                                  * copybreak to save the put_page/alloc_page */
3648                                 if (length <= copybreak &&
3649                                     skb_tailroom(skb) >= length) {
3650                                         u8 *vaddr;
3651                                         vaddr = kmap_atomic(buffer_info->page,
3652                                                             KM_SKB_DATA_SOFTIRQ);
3653                                         memcpy(skb_tail_pointer(skb), vaddr, length);
3654                                         kunmap_atomic(vaddr,
3655                                                       KM_SKB_DATA_SOFTIRQ);
3656                                         /* re-use the page, so don't erase
3657                                          * buffer_info->page */
3658                                         skb_put(skb, length);
3659                                 } else {
3660                                         skb_fill_page_desc(skb, 0,
3661                                                            buffer_info->page, 0,
3662                                                            length);
3663                                         e1000_consume_page(buffer_info, skb,
3664                                                            length);
3665                                 }
3666                         }
3667                 }
3668
3669                 /* Receive Checksum Offload XXX recompute due to CRC strip? */
3670                 e1000_rx_checksum(adapter,
3671                                   (u32)(status) |
3672                                   ((u32)(rx_desc->errors) << 24),
3673                                   le16_to_cpu(rx_desc->csum), skb);
3674
3675                 pskb_trim(skb, skb->len - 4);
3676
3677                 /* probably a little skewed due to removing CRC */
3678                 total_rx_bytes += skb->len;
3679                 total_rx_packets++;
3680
3681                 /* eth type trans needs skb->data to point to something */
3682                 if (!pskb_may_pull(skb, ETH_HLEN)) {
3683                         DPRINTK(DRV, ERR, "pskb_may_pull failed.\n");
3684                         dev_kfree_skb(skb);
3685                         goto next_desc;
3686                 }
3687
3688                 skb->protocol = eth_type_trans(skb, netdev);
3689
3690                 e1000_receive_skb(adapter, status, rx_desc->special, skb);
3691
3692 next_desc:
3693                 rx_desc->status = 0;
3694
3695                 /* return some buffers to hardware, one at a time is too slow */
3696                 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
3697                         adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
3698                         cleaned_count = 0;
3699                 }
3700
3701                 /* use prefetched values */
3702                 rx_desc = next_rxd;
3703                 buffer_info = next_buffer;
3704         }
3705         rx_ring->next_to_clean = i;
3706
3707         cleaned_count = E1000_DESC_UNUSED(rx_ring);
3708         if (cleaned_count)
3709                 adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
3710
3711         adapter->total_rx_packets += total_rx_packets;
3712         adapter->total_rx_bytes += total_rx_bytes;
3713         adapter->net_stats.rx_bytes += total_rx_bytes;
3714         adapter->net_stats.rx_packets += total_rx_packets;
3715         return cleaned;
3716 }
3717
3718 /**
3719  * e1000_clean_rx_irq - Send received data up the network stack; legacy
3720  * @adapter: board private structure
3721  * @rx_ring: ring to clean
3722  * @work_done: amount of napi work completed this call
3723  * @work_to_do: max amount of work allowed for this call to do
3724  */
3725 static bool e1000_clean_rx_irq(struct e1000_adapter *adapter,
3726                                struct e1000_rx_ring *rx_ring,
3727                                int *work_done, int work_to_do)
3728 {
3729         struct e1000_hw *hw = &adapter->hw;
3730         struct net_device *netdev = adapter->netdev;
3731         struct pci_dev *pdev = adapter->pdev;
3732         struct e1000_rx_desc *rx_desc, *next_rxd;
3733         struct e1000_buffer *buffer_info, *next_buffer;
3734         unsigned long flags;
3735         u32 length;
3736         unsigned int i;
3737         int cleaned_count = 0;
3738         bool cleaned = false;
3739         unsigned int total_rx_bytes=0, total_rx_packets=0;
3740
3741         i = rx_ring->next_to_clean;
3742         rx_desc = E1000_RX_DESC(*rx_ring, i);
3743         buffer_info = &rx_ring->buffer_info[i];
3744
3745         while (rx_desc->status & E1000_RXD_STAT_DD) {
3746                 struct sk_buff *skb;
3747                 u8 status;
3748
3749                 if (*work_done >= work_to_do)
3750                         break;
3751                 (*work_done)++;
3752
3753                 status = rx_desc->status;
3754                 skb = buffer_info->skb;
3755                 buffer_info->skb = NULL;
3756
3757                 prefetch(skb->data - NET_IP_ALIGN);
3758
3759                 if (++i == rx_ring->count) i = 0;
3760                 next_rxd = E1000_RX_DESC(*rx_ring, i);
3761                 prefetch(next_rxd);
3762
3763                 next_buffer = &rx_ring->buffer_info[i];
3764
3765                 cleaned = true;
3766                 cleaned_count++;
3767                 pci_unmap_single(pdev, buffer_info->dma, buffer_info->length,
3768                                  PCI_DMA_FROMDEVICE);
3769                 buffer_info->dma = 0;
3770
3771                 length = le16_to_cpu(rx_desc->length);
3772                 /* !EOP means multiple descriptors were used to store a single
3773                  * packet, also make sure the frame isn't just CRC only */
3774                 if (unlikely(!(status & E1000_RXD_STAT_EOP) || (length <= 4))) {
3775                         /* All receives must fit into a single buffer */
3776                         E1000_DBG("%s: Receive packet consumed multiple"
3777                                   " buffers\n", netdev->name);
3778                         /* recycle */
3779                         buffer_info->skb = skb;
3780                         goto next_desc;
3781                 }
3782
3783                 if (unlikely(rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK)) {
3784                         u8 last_byte = *(skb->data + length - 1);
3785                         if (TBI_ACCEPT(hw, status, rx_desc->errors, length,
3786                                        last_byte)) {
3787                                 spin_lock_irqsave(&adapter->stats_lock, flags);
3788                                 e1000_tbi_adjust_stats(hw, &adapter->stats,
3789                                                        length, skb->data);
3790                                 spin_unlock_irqrestore(&adapter->stats_lock,
3791                                                        flags);
3792                                 length--;
3793                         } else {
3794                                 /* recycle */
3795                                 buffer_info->skb = skb;
3796                                 goto next_desc;
3797                         }
3798                 }
3799
3800                 /* adjust length to remove Ethernet CRC, this must be
3801                  * done after the TBI_ACCEPT workaround above */
3802                 length -= 4;
3803
3804                 /* probably a little skewed due to removing CRC */
3805                 total_rx_bytes += length;
3806                 total_rx_packets++;
3807
3808                 /* code added for copybreak, this should improve
3809                  * performance for small packets with large amounts
3810                  * of reassembly being done in the stack */
3811                 if (length < copybreak) {
3812                         struct sk_buff *new_skb =
3813                             netdev_alloc_skb(netdev, length + NET_IP_ALIGN);
3814                         if (new_skb) {
3815                                 skb_reserve(new_skb, NET_IP_ALIGN);
3816                                 skb_copy_to_linear_data_offset(new_skb,
3817                                                                -NET_IP_ALIGN,
3818                                                                (skb->data -
3819                                                                 NET_IP_ALIGN),
3820                                                                (length +
3821                                                                 NET_IP_ALIGN));
3822                                 /* save the skb in buffer_info as good */
3823                                 buffer_info->skb = skb;
3824                                 skb = new_skb;
3825                         }
3826                         /* else just continue with the old one */
3827                 }
3828                 /* end copybreak code */
3829                 skb_put(skb, length);
3830
3831                 /* Receive Checksum Offload */
3832                 e1000_rx_checksum(adapter,
3833                                   (u32)(status) |
3834                                   ((u32)(rx_desc->errors) << 24),
3835                                   le16_to_cpu(rx_desc->csum), skb);
3836
3837                 skb->protocol = eth_type_trans(skb, netdev);
3838
3839                 e1000_receive_skb(adapter, status, rx_desc->special, skb);
3840
3841 next_desc:
3842                 rx_desc->status = 0;
3843
3844                 /* return some buffers to hardware, one at a time is too slow */
3845                 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
3846                         adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
3847                         cleaned_count = 0;
3848                 }
3849
3850                 /* use prefetched values */
3851                 rx_desc = next_rxd;
3852                 buffer_info = next_buffer;
3853         }
3854         rx_ring->next_to_clean = i;
3855
3856         cleaned_count = E1000_DESC_UNUSED(rx_ring);
3857         if (cleaned_count)
3858                 adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
3859
3860         adapter->total_rx_packets += total_rx_packets;
3861         adapter->total_rx_bytes += total_rx_bytes;
3862         adapter->net_stats.rx_bytes += total_rx_bytes;
3863         adapter->net_stats.rx_packets += total_rx_packets;
3864         return cleaned;
3865 }
3866
3867 /**
3868  * e1000_alloc_jumbo_rx_buffers - Replace used jumbo receive buffers
3869  * @adapter: address of board private structure
3870  * @rx_ring: pointer to receive ring structure
3871  * @cleaned_count: number of buffers to allocate this pass
3872  **/
3873
3874 static void
3875 e1000_alloc_jumbo_rx_buffers(struct e1000_adapter *adapter,
3876                              struct e1000_rx_ring *rx_ring, int cleaned_count)
3877 {
3878         struct net_device *netdev = adapter->netdev;
3879         struct pci_dev *pdev = adapter->pdev;
3880         struct e1000_rx_desc *rx_desc;
3881         struct e1000_buffer *buffer_info;
3882         struct sk_buff *skb;
3883         unsigned int i;
3884         unsigned int bufsz = 256 -
3885                              16 /*for skb_reserve */ -
3886                              NET_IP_ALIGN;
3887
3888         i = rx_ring->next_to_use;
3889         buffer_info = &rx_ring->buffer_info[i];
3890
3891         while (cleaned_count--) {
3892                 skb = buffer_info->skb;
3893                 if (skb) {
3894                         skb_trim(skb, 0);
3895                         goto check_page;
3896                 }
3897
3898                 skb = netdev_alloc_skb(netdev, bufsz);
3899                 if (unlikely(!skb)) {
3900                         /* Better luck next round */
3901                         adapter->alloc_rx_buff_failed++;
3902                         break;
3903                 }
3904
3905                 /* Fix for errata 23, can't cross 64kB boundary */
3906                 if (!e1000_check_64k_bound(adapter, skb->data, bufsz)) {
3907                         struct sk_buff *oldskb = skb;
3908                         DPRINTK(PROBE, ERR, "skb align check failed: %u bytes "
3909                                              "at %p\n", bufsz, skb->data);
3910                         /* Try again, without freeing the previous */
3911                         skb = netdev_alloc_skb(netdev, bufsz);
3912                         /* Failed allocation, critical failure */
3913                         if (!skb) {
3914                                 dev_kfree_skb(oldskb);
3915                                 adapter->alloc_rx_buff_failed++;
3916                                 break;
3917                         }
3918
3919                         if (!e1000_check_64k_bound(adapter, skb->data, bufsz)) {
3920                                 /* give up */
3921                                 dev_kfree_skb(skb);
3922                                 dev_kfree_skb(oldskb);
3923                                 break; /* while (cleaned_count--) */
3924                         }
3925
3926                         /* Use new allocation */
3927                         dev_kfree_skb(oldskb);
3928                 }
3929                 /* Make buffer alignment 2 beyond a 16 byte boundary
3930                  * this will result in a 16 byte aligned IP header after
3931                  * the 14 byte MAC header is removed
3932                  */
3933                 skb_reserve(skb, NET_IP_ALIGN);
3934
3935                 buffer_info->skb = skb;
3936                 buffer_info->length = adapter->rx_buffer_len;
3937 check_page:
3938                 /* allocate a new page if necessary */
3939                 if (!buffer_info->page) {
3940                         buffer_info->page = alloc_page(GFP_ATOMIC);
3941                         if (unlikely(!buffer_info->page)) {
3942                                 adapter->alloc_rx_buff_failed++;
3943                                 break;
3944                         }
3945                 }
3946
3947                 if (!buffer_info->dma)
3948                         buffer_info->dma = pci_map_page(pdev,
3949                                                         buffer_info->page, 0,
3950                                                         buffer_info->length,
3951                                                         PCI_DMA_FROMDEVICE);
3952
3953                 rx_desc = E1000_RX_DESC(*rx_ring, i);
3954                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
3955
3956                 if (unlikely(++i == rx_ring->count))
3957                         i = 0;
3958                 buffer_info = &rx_ring->buffer_info[i];
3959         }
3960
3961         if (likely(rx_ring->next_to_use != i)) {
3962                 rx_ring->next_to_use = i;
3963                 if (unlikely(i-- == 0))
3964                         i = (rx_ring->count - 1);
3965
3966                 /* Force memory writes to complete before letting h/w
3967                  * know there are new descriptors to fetch.  (Only
3968                  * applicable for weak-ordered memory model archs,
3969                  * such as IA-64). */
3970                 wmb();
3971                 writel(i, adapter->hw.hw_addr + rx_ring->rdt);
3972         }
3973 }
3974
3975 /**
3976  * e1000_alloc_rx_buffers - Replace used receive buffers; legacy & extended
3977  * @adapter: address of board private structure
3978  **/
3979
3980 static void e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
3981                                    struct e1000_rx_ring *rx_ring,
3982                                    int cleaned_count)
3983 {
3984         struct e1000_hw *hw = &adapter->hw;
3985         struct net_device *netdev = adapter->netdev;
3986         struct pci_dev *pdev = adapter->pdev;
3987         struct e1000_rx_desc *rx_desc;
3988         struct e1000_buffer *buffer_info;
3989         struct sk_buff *skb;
3990         unsigned int i;
3991         unsigned int bufsz = adapter->rx_buffer_len + NET_IP_ALIGN;
3992
3993         i = rx_ring->next_to_use;
3994         buffer_info = &rx_ring->buffer_info[i];
3995
3996         while (cleaned_count--) {
3997                 skb = buffer_info->skb;
3998                 if (skb) {
3999                         skb_trim(skb, 0);
4000                         goto map_skb;
4001                 }
4002
4003                 skb = netdev_alloc_skb(netdev, bufsz);
4004                 if (unlikely(!skb)) {
4005                         /* Better luck next round */
4006                         adapter->alloc_rx_buff_failed++;
4007                         break;
4008                 }
4009
4010                 /* Fix for errata 23, can't cross 64kB boundary */
4011                 if (!e1000_check_64k_bound(adapter, skb->data, bufsz)) {
4012                         struct sk_buff *oldskb = skb;
4013                         DPRINTK(RX_ERR, ERR, "skb align check failed: %u bytes "
4014                                              "at %p\n", bufsz, skb->data);
4015                         /* Try again, without freeing the previous */
4016                         skb = netdev_alloc_skb(netdev, bufsz);
4017                         /* Failed allocation, critical failure */
4018                         if (!skb) {
4019                                 dev_kfree_skb(oldskb);
4020                                 adapter->alloc_rx_buff_failed++;
4021                                 break;
4022                         }
4023
4024                         if (!e1000_check_64k_bound(adapter, skb->data, bufsz)) {
4025                                 /* give up */
4026                                 dev_kfree_skb(skb);
4027                                 dev_kfree_skb(oldskb);
4028                                 adapter->alloc_rx_buff_failed++;
4029                                 break; /* while !buffer_info->skb */
4030                         }
4031
4032                         /* Use new allocation */
4033                         dev_kfree_skb(oldskb);
4034                 }
4035                 /* Make buffer alignment 2 beyond a 16 byte boundary
4036                  * this will result in a 16 byte aligned IP header after
4037                  * the 14 byte MAC header is removed
4038                  */
4039                 skb_reserve(skb, NET_IP_ALIGN);
4040
4041                 buffer_info->skb = skb;
4042                 buffer_info->length = adapter->rx_buffer_len;
4043 map_skb:
4044                 buffer_info->dma = pci_map_single(pdev,
4045                                                   skb->data,
4046                                                   buffer_info->length,
4047                                                   PCI_DMA_FROMDEVICE);
4048
4049                 /*
4050                  * XXX if it was allocated cleanly it will never map to a
4051                  * boundary crossing
4052                  */
4053
4054                 /* Fix for errata 23, can't cross 64kB boundary */
4055                 if (!e1000_check_64k_bound(adapter,
4056                                         (void *)(unsigned long)buffer_info->dma,
4057                                         adapter->rx_buffer_len)) {
4058                         DPRINTK(RX_ERR, ERR,
4059                                 "dma align check failed: %u bytes at %p\n",
4060                                 adapter->rx_buffer_len,
4061                                 (void *)(unsigned long)buffer_info->dma);
4062                         dev_kfree_skb(skb);
4063                         buffer_info->skb = NULL;
4064
4065                         pci_unmap_single(pdev, buffer_info->dma,
4066                                          adapter->rx_buffer_len,
4067                                          PCI_DMA_FROMDEVICE);
4068                         buffer_info->dma = 0;
4069
4070                         adapter->alloc_rx_buff_failed++;
4071                         break; /* while !buffer_info->skb */
4072                 }
4073                 rx_desc = E1000_RX_DESC(*rx_ring, i);
4074                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
4075
4076                 if (unlikely(++i == rx_ring->count))
4077                         i = 0;
4078                 buffer_info = &rx_ring->buffer_info[i];
4079         }
4080
4081         if (likely(rx_ring->next_to_use != i)) {
4082                 rx_ring->next_to_use = i;
4083                 if (unlikely(i-- == 0))
4084                         i = (rx_ring->count - 1);
4085
4086                 /* Force memory writes to complete before letting h/w
4087                  * know there are new descriptors to fetch.  (Only
4088                  * applicable for weak-ordered memory model archs,
4089                  * such as IA-64). */
4090                 wmb();
4091                 writel(i, hw->hw_addr + rx_ring->rdt);
4092         }
4093 }
4094
4095 /**
4096  * e1000_smartspeed - Workaround for SmartSpeed on 82541 and 82547 controllers.
4097  * @adapter:
4098  **/
4099
4100 static void e1000_smartspeed(struct e1000_adapter *adapter)
4101 {
4102         struct e1000_hw *hw = &adapter->hw;
4103         u16 phy_status;
4104         u16 phy_ctrl;
4105
4106         if ((hw->phy_type != e1000_phy_igp) || !hw->autoneg ||
4107            !(hw->autoneg_advertised & ADVERTISE_1000_FULL))
4108                 return;
4109
4110         if (adapter->smartspeed == 0) {
4111                 /* If Master/Slave config fault is asserted twice,
4112                  * we assume back-to-back */
4113                 e1000_read_phy_reg(hw, PHY_1000T_STATUS, &phy_status);
4114                 if (!(phy_status & SR_1000T_MS_CONFIG_FAULT)) return;
4115                 e1000_read_phy_reg(hw, PHY_1000T_STATUS, &phy_status);
4116                 if (!(phy_status & SR_1000T_MS_CONFIG_FAULT)) return;
4117                 e1000_read_phy_reg(hw, PHY_1000T_CTRL, &phy_ctrl);
4118                 if (phy_ctrl & CR_1000T_MS_ENABLE) {
4119                         phy_ctrl &= ~CR_1000T_MS_ENABLE;
4120                         e1000_write_phy_reg(hw, PHY_1000T_CTRL,
4121                                             phy_ctrl);
4122                         adapter->smartspeed++;
4123                         if (!e1000_phy_setup_autoneg(hw) &&
4124                            !e1000_read_phy_reg(hw, PHY_CTRL,
4125                                                &phy_ctrl)) {
4126                                 phy_ctrl |= (MII_CR_AUTO_NEG_EN |
4127                                              MII_CR_RESTART_AUTO_NEG);
4128                                 e1000_write_phy_reg(hw, PHY_CTRL,
4129                                                     phy_ctrl);
4130                         }
4131                 }
4132                 return;
4133         } else if (adapter->smartspeed == E1000_SMARTSPEED_DOWNSHIFT) {
4134                 /* If still no link, perhaps using 2/3 pair cable */
4135                 e1000_read_phy_reg(hw, PHY_1000T_CTRL, &phy_ctrl);
4136                 phy_ctrl |= CR_1000T_MS_ENABLE;
4137                 e1000_write_phy_reg(hw, PHY_1000T_CTRL, phy_ctrl);
4138                 if (!e1000_phy_setup_autoneg(hw) &&
4139                    !e1000_read_phy_reg(hw, PHY_CTRL, &phy_ctrl)) {
4140                         phy_ctrl |= (MII_CR_AUTO_NEG_EN |
4141                                      MII_CR_RESTART_AUTO_NEG);
4142                         e1000_write_phy_reg(hw, PHY_CTRL, phy_ctrl);
4143                 }
4144         }
4145         /* Restart process after E1000_SMARTSPEED_MAX iterations */
4146         if (adapter->smartspeed++ == E1000_SMARTSPEED_MAX)
4147                 adapter->smartspeed = 0;
4148 }
4149
4150 /**
4151  * e1000_ioctl -
4152  * @netdev:
4153  * @ifreq:
4154  * @cmd:
4155  **/
4156
4157 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
4158 {
4159         switch (cmd) {
4160         case SIOCGMIIPHY:
4161         case SIOCGMIIREG:
4162         case SIOCSMIIREG:
4163                 return e1000_mii_ioctl(netdev, ifr, cmd);
4164         default:
4165                 return -EOPNOTSUPP;
4166         }
4167 }
4168
4169 /**
4170  * e1000_mii_ioctl -
4171  * @netdev:
4172  * @ifreq:
4173  * @cmd:
4174  **/
4175
4176 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
4177                            int cmd)
4178 {
4179         struct e1000_adapter *adapter = netdev_priv(netdev);
4180         struct e1000_hw *hw = &adapter->hw;
4181         struct mii_ioctl_data *data = if_mii(ifr);
4182         int retval;
4183         u16 mii_reg;
4184         u16 spddplx;
4185         unsigned long flags;
4186
4187         if (hw->media_type != e1000_media_type_copper)
4188                 return -EOPNOTSUPP;
4189
4190         switch (cmd) {
4191         case SIOCGMIIPHY:
4192                 data->phy_id = hw->phy_addr;
4193                 break;
4194         case SIOCGMIIREG:
4195                 spin_lock_irqsave(&adapter->stats_lock, flags);
4196                 if (e1000_read_phy_reg(hw, data->reg_num & 0x1F,
4197                                    &data->val_out)) {
4198                         spin_unlock_irqrestore(&adapter->stats_lock, flags);
4199                         return -EIO;
4200                 }
4201                 spin_unlock_irqrestore(&adapter->stats_lock, flags);
4202                 break;
4203         case SIOCSMIIREG:
4204                 if (data->reg_num & ~(0x1F))
4205                         return -EFAULT;
4206                 mii_reg = data->val_in;
4207                 spin_lock_irqsave(&adapter->stats_lock, flags);
4208                 if (e1000_write_phy_reg(hw, data->reg_num,
4209                                         mii_reg)) {
4210                         spin_unlock_irqrestore(&adapter->stats_lock, flags);
4211                         return -EIO;
4212                 }
4213                 spin_unlock_irqrestore(&adapter->stats_lock, flags);
4214                 if (hw->media_type == e1000_media_type_copper) {
4215                         switch (data->reg_num) {
4216                         case PHY_CTRL:
4217                                 if (mii_reg & MII_CR_POWER_DOWN)
4218                                         break;
4219                                 if (mii_reg & MII_CR_AUTO_NEG_EN) {
4220                                         hw->autoneg = 1;
4221                                         hw->autoneg_advertised = 0x2F;
4222                                 } else {
4223                                         if (mii_reg & 0x40)
4224                                                 spddplx = SPEED_1000;
4225                                         else if (mii_reg & 0x2000)
4226                                                 spddplx = SPEED_100;
4227                                         else
4228                                                 spddplx = SPEED_10;
4229                                         spddplx += (mii_reg & 0x100)
4230                                                    ? DUPLEX_FULL :
4231                                                    DUPLEX_HALF;
4232                                         retval = e1000_set_spd_dplx(adapter,
4233                                                                     spddplx);
4234                                         if (retval)
4235                                                 return retval;
4236                                 }
4237                                 if (netif_running(adapter->netdev))
4238                                         e1000_reinit_locked(adapter);
4239                                 else
4240                                         e1000_reset(adapter);
4241                                 break;
4242                         case M88E1000_PHY_SPEC_CTRL:
4243                         case M88E1000_EXT_PHY_SPEC_CTRL:
4244                                 if (e1000_phy_reset(hw))
4245                                         return -EIO;
4246                                 break;
4247                         }
4248                 } else {
4249                         switch (data->reg_num) {
4250                         case PHY_CTRL:
4251                                 if (mii_reg & MII_CR_POWER_DOWN)
4252                                         break;
4253                                 if (netif_running(adapter->netdev))
4254                                         e1000_reinit_locked(adapter);
4255                                 else
4256                                         e1000_reset(adapter);
4257                                 break;
4258                         }
4259                 }
4260                 break;
4261         default:
4262                 return -EOPNOTSUPP;
4263         }
4264         return E1000_SUCCESS;
4265 }
4266
4267 void e1000_pci_set_mwi(struct e1000_hw *hw)
4268 {
4269         struct e1000_adapter *adapter = hw->back;
4270         int ret_val = pci_set_mwi(adapter->pdev);
4271
4272         if (ret_val)
4273                 DPRINTK(PROBE, ERR, "Error in setting MWI\n");
4274 }
4275
4276 void e1000_pci_clear_mwi(struct e1000_hw *hw)
4277 {
4278         struct e1000_adapter *adapter = hw->back;
4279
4280         pci_clear_mwi(adapter->pdev);
4281 }
4282
4283 int e1000_pcix_get_mmrbc(struct e1000_hw *hw)
4284 {
4285         struct e1000_adapter *adapter = hw->back;
4286         return pcix_get_mmrbc(adapter->pdev);
4287 }
4288
4289 void e1000_pcix_set_mmrbc(struct e1000_hw *hw, int mmrbc)
4290 {
4291         struct e1000_adapter *adapter = hw->back;
4292         pcix_set_mmrbc(adapter->pdev, mmrbc);
4293 }
4294
4295 s32 e1000_read_pcie_cap_reg(struct e1000_hw *hw, u32 reg, u16 *value)
4296 {
4297     struct e1000_adapter *adapter = hw->back;
4298     u16 cap_offset;
4299
4300     cap_offset = pci_find_capability(adapter->pdev, PCI_CAP_ID_EXP);
4301     if (!cap_offset)
4302         return -E1000_ERR_CONFIG;
4303
4304     pci_read_config_word(adapter->pdev, cap_offset + reg, value);
4305
4306     return E1000_SUCCESS;
4307 }
4308
4309 void e1000_io_write(struct e1000_hw *hw, unsigned long port, u32 value)
4310 {
4311         outl(value, port);
4312 }
4313
4314 static void e1000_vlan_rx_register(struct net_device *netdev,
4315                                    struct vlan_group *grp)
4316 {
4317         struct e1000_adapter *adapter = netdev_priv(netdev);
4318         struct e1000_hw *hw = &adapter->hw;
4319         u32 ctrl, rctl;
4320
4321         if (!test_bit(__E1000_DOWN, &adapter->flags))
4322                 e1000_irq_disable(adapter);
4323         adapter->vlgrp = grp;
4324
4325         if (grp) {
4326                 /* enable VLAN tag insert/strip */
4327                 ctrl = er32(CTRL);
4328                 ctrl |= E1000_CTRL_VME;
4329                 ew32(CTRL, ctrl);
4330
4331                 /* enable VLAN receive filtering */
4332                 rctl = er32(RCTL);
4333                 rctl &= ~E1000_RCTL_CFIEN;
4334                 if (!(netdev->flags & IFF_PROMISC))
4335                         rctl |= E1000_RCTL_VFE;
4336                 ew32(RCTL, rctl);
4337                 e1000_update_mng_vlan(adapter);
4338         } else {
4339                 /* disable VLAN tag insert/strip */
4340                 ctrl = er32(CTRL);
4341                 ctrl &= ~E1000_CTRL_VME;
4342                 ew32(CTRL, ctrl);
4343
4344                 /* disable VLAN receive filtering */
4345                 rctl = er32(RCTL);
4346                 rctl &= ~E1000_RCTL_VFE;
4347                 ew32(RCTL, rctl);
4348
4349                 if (adapter->mng_vlan_id != (u16)E1000_MNG_VLAN_NONE) {
4350                         e1000_vlan_rx_kill_vid(netdev,
4351                                                adapter->mng_vlan_id);
4352                         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
4353                 }
4354         }
4355
4356         if (!test_bit(__E1000_DOWN, &adapter->flags))
4357                 e1000_irq_enable(adapter);
4358 }
4359
4360 static void e1000_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
4361 {
4362         struct e1000_adapter *adapter = netdev_priv(netdev);
4363         struct e1000_hw *hw = &adapter->hw;
4364         u32 vfta, index;
4365
4366         if ((hw->mng_cookie.status &
4367              E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT) &&
4368             (vid == adapter->mng_vlan_id))
4369                 return;
4370         /* add VID to filter table */
4371         index = (vid >> 5) & 0x7F;
4372         vfta = E1000_READ_REG_ARRAY(hw, VFTA, index);
4373         vfta |= (1 << (vid & 0x1F));
4374         e1000_write_vfta(hw, index, vfta);
4375 }
4376
4377 static void e1000_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
4378 {
4379         struct e1000_adapter *adapter = netdev_priv(netdev);
4380         struct e1000_hw *hw = &adapter->hw;
4381         u32 vfta, index;
4382
4383         if (!test_bit(__E1000_DOWN, &adapter->flags))
4384                 e1000_irq_disable(adapter);
4385         vlan_group_set_device(adapter->vlgrp, vid, NULL);
4386         if (!test_bit(__E1000_DOWN, &adapter->flags))
4387                 e1000_irq_enable(adapter);
4388
4389         /* remove VID from filter table */
4390         index = (vid >> 5) & 0x7F;
4391         vfta = E1000_READ_REG_ARRAY(hw, VFTA, index);
4392         vfta &= ~(1 << (vid & 0x1F));
4393         e1000_write_vfta(hw, index, vfta);
4394 }
4395
4396 static void e1000_restore_vlan(struct e1000_adapter *adapter)
4397 {
4398         e1000_vlan_rx_register(adapter->netdev, adapter->vlgrp);
4399
4400         if (adapter->vlgrp) {
4401                 u16 vid;
4402                 for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
4403                         if (!vlan_group_get_device(adapter->vlgrp, vid))
4404                                 continue;
4405                         e1000_vlan_rx_add_vid(adapter->netdev, vid);
4406                 }
4407         }
4408 }
4409
4410 int e1000_set_spd_dplx(struct e1000_adapter *adapter, u16 spddplx)
4411 {
4412         struct e1000_hw *hw = &adapter->hw;
4413
4414         hw->autoneg = 0;
4415
4416         /* Fiber NICs only allow 1000 gbps Full duplex */
4417         if ((hw->media_type == e1000_media_type_fiber) &&
4418                 spddplx != (SPEED_1000 + DUPLEX_FULL)) {
4419                 DPRINTK(PROBE, ERR, "Unsupported Speed/Duplex configuration\n");
4420                 return -EINVAL;
4421         }
4422
4423         switch (spddplx) {
4424         case SPEED_10 + DUPLEX_HALF:
4425                 hw->forced_speed_duplex = e1000_10_half;
4426                 break;
4427         case SPEED_10 + DUPLEX_FULL:
4428                 hw->forced_speed_duplex = e1000_10_full;
4429                 break;
4430         case SPEED_100 + DUPLEX_HALF:
4431                 hw->forced_speed_duplex = e1000_100_half;
4432                 break;
4433         case SPEED_100 + DUPLEX_FULL:
4434                 hw->forced_speed_duplex = e1000_100_full;
4435                 break;
4436         case SPEED_1000 + DUPLEX_FULL:
4437                 hw->autoneg = 1;
4438                 hw->autoneg_advertised = ADVERTISE_1000_FULL;
4439                 break;
4440         case SPEED_1000 + DUPLEX_HALF: /* not supported */
4441         default:
4442                 DPRINTK(PROBE, ERR, "Unsupported Speed/Duplex configuration\n");
4443                 return -EINVAL;
4444         }
4445         return 0;
4446 }
4447
4448 static int __e1000_shutdown(struct pci_dev *pdev, bool *enable_wake)
4449 {
4450         struct net_device *netdev = pci_get_drvdata(pdev);
4451         struct e1000_adapter *adapter = netdev_priv(netdev);
4452         struct e1000_hw *hw = &adapter->hw;
4453         u32 ctrl, ctrl_ext, rctl, status;
4454         u32 wufc = adapter->wol;
4455 #ifdef CONFIG_PM
4456         int retval = 0;
4457 #endif
4458
4459         netif_device_detach(netdev);
4460
4461         if (netif_running(netdev)) {
4462                 WARN_ON(test_bit(__E1000_RESETTING, &adapter->flags));
4463                 e1000_down(adapter);
4464         }
4465
4466 #ifdef CONFIG_PM
4467         retval = pci_save_state(pdev);
4468         if (retval)
4469                 return retval;
4470 #endif
4471
4472         status = er32(STATUS);
4473         if (status & E1000_STATUS_LU)
4474                 wufc &= ~E1000_WUFC_LNKC;
4475
4476         if (wufc) {
4477                 e1000_setup_rctl(adapter);
4478                 e1000_set_rx_mode(netdev);
4479
4480                 /* turn on all-multi mode if wake on multicast is enabled */
4481                 if (wufc & E1000_WUFC_MC) {
4482                         rctl = er32(RCTL);
4483                         rctl |= E1000_RCTL_MPE;
4484                         ew32(RCTL, rctl);
4485                 }
4486
4487                 if (hw->mac_type >= e1000_82540) {
4488                         ctrl = er32(CTRL);
4489                         /* advertise wake from D3Cold */
4490                         #define E1000_CTRL_ADVD3WUC 0x00100000
4491                         /* phy power management enable */
4492                         #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
4493                         ctrl |= E1000_CTRL_ADVD3WUC |
4494                                 E1000_CTRL_EN_PHY_PWR_MGMT;
4495                         ew32(CTRL, ctrl);
4496                 }
4497
4498                 if (hw->media_type == e1000_media_type_fiber ||
4499                     hw->media_type == e1000_media_type_internal_serdes) {
4500                         /* keep the laser running in D3 */
4501                         ctrl_ext = er32(CTRL_EXT);
4502                         ctrl_ext |= E1000_CTRL_EXT_SDP7_DATA;
4503                         ew32(CTRL_EXT, ctrl_ext);
4504                 }
4505
4506                 ew32(WUC, E1000_WUC_PME_EN);
4507                 ew32(WUFC, wufc);
4508         } else {
4509                 ew32(WUC, 0);
4510                 ew32(WUFC, 0);
4511         }
4512
4513         e1000_release_manageability(adapter);
4514
4515         *enable_wake = !!wufc;
4516
4517         /* make sure adapter isn't asleep if manageability is enabled */
4518         if (adapter->en_mng_pt)
4519                 *enable_wake = true;
4520
4521         if (netif_running(netdev))
4522                 e1000_free_irq(adapter);
4523
4524         pci_disable_device(pdev);
4525
4526         return 0;
4527 }
4528
4529 #ifdef CONFIG_PM
4530 static int e1000_suspend(struct pci_dev *pdev, pm_message_t state)
4531 {
4532         int retval;
4533         bool wake;
4534
4535         retval = __e1000_shutdown(pdev, &wake);
4536         if (retval)
4537                 return retval;
4538
4539         if (wake) {
4540                 pci_prepare_to_sleep(pdev);
4541         } else {
4542                 pci_wake_from_d3(pdev, false);
4543                 pci_set_power_state(pdev, PCI_D3hot);
4544         }
4545
4546         return 0;
4547 }
4548
4549 static int e1000_resume(struct pci_dev *pdev)
4550 {
4551         struct net_device *netdev = pci_get_drvdata(pdev);
4552         struct e1000_adapter *adapter = netdev_priv(netdev);
4553         struct e1000_hw *hw = &adapter->hw;
4554         u32 err;
4555
4556         pci_set_power_state(pdev, PCI_D0);
4557         pci_restore_state(pdev);
4558
4559         if (adapter->need_ioport)
4560                 err = pci_enable_device(pdev);
4561         else
4562                 err = pci_enable_device_mem(pdev);
4563         if (err) {
4564                 printk(KERN_ERR "e1000: Cannot enable PCI device from suspend\n");
4565                 return err;
4566         }
4567         pci_set_master(pdev);
4568
4569         pci_enable_wake(pdev, PCI_D3hot, 0);
4570         pci_enable_wake(pdev, PCI_D3cold, 0);
4571
4572         if (netif_running(netdev)) {
4573                 err = e1000_request_irq(adapter);
4574                 if (err)
4575                         return err;
4576         }
4577
4578         e1000_power_up_phy(adapter);
4579         e1000_reset(adapter);
4580         ew32(WUS, ~0);
4581
4582         e1000_init_manageability(adapter);
4583
4584         if (netif_running(netdev))
4585                 e1000_up(adapter);
4586
4587         netif_device_attach(netdev);
4588
4589         return 0;
4590 }
4591 #endif
4592
4593 static void e1000_shutdown(struct pci_dev *pdev)
4594 {
4595         bool wake;
4596
4597         __e1000_shutdown(pdev, &wake);
4598
4599         if (system_state == SYSTEM_POWER_OFF) {
4600                 pci_wake_from_d3(pdev, wake);
4601                 pci_set_power_state(pdev, PCI_D3hot);
4602         }
4603 }
4604
4605 #ifdef CONFIG_NET_POLL_CONTROLLER
4606 /*
4607  * Polling 'interrupt' - used by things like netconsole to send skbs
4608  * without having to re-enable interrupts. It's not called while
4609  * the interrupt routine is executing.
4610  */
4611 static void e1000_netpoll(struct net_device *netdev)
4612 {
4613         struct e1000_adapter *adapter = netdev_priv(netdev);
4614
4615         disable_irq(adapter->pdev->irq);
4616         e1000_intr(adapter->pdev->irq, netdev);
4617         enable_irq(adapter->pdev->irq);
4618 }
4619 #endif
4620
4621 /**
4622  * e1000_io_error_detected - called when PCI error is detected
4623  * @pdev: Pointer to PCI device
4624  * @state: The current pci conneection state
4625  *
4626  * This function is called after a PCI bus error affecting
4627  * this device has been detected.
4628  */
4629 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
4630                                                 pci_channel_state_t state)
4631 {
4632         struct net_device *netdev = pci_get_drvdata(pdev);
4633         struct e1000_adapter *adapter = netdev_priv(netdev);
4634
4635         netif_device_detach(netdev);
4636
4637         if (state == pci_channel_io_perm_failure)
4638                 return PCI_ERS_RESULT_DISCONNECT;
4639
4640         if (netif_running(netdev))
4641                 e1000_down(adapter);
4642         pci_disable_device(pdev);
4643
4644         /* Request a slot slot reset. */
4645         return PCI_ERS_RESULT_NEED_RESET;
4646 }
4647
4648 /**
4649  * e1000_io_slot_reset - called after the pci bus has been reset.
4650  * @pdev: Pointer to PCI device
4651  *
4652  * Restart the card from scratch, as if from a cold-boot. Implementation
4653  * resembles the first-half of the e1000_resume routine.
4654  */
4655 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
4656 {
4657         struct net_device *netdev = pci_get_drvdata(pdev);
4658         struct e1000_adapter *adapter = netdev_priv(netdev);
4659         struct e1000_hw *hw = &adapter->hw;
4660         int err;
4661
4662         if (adapter->need_ioport)
4663                 err = pci_enable_device(pdev);
4664         else
4665                 err = pci_enable_device_mem(pdev);
4666         if (err) {
4667                 printk(KERN_ERR "e1000: Cannot re-enable PCI device after reset.\n");
4668                 return PCI_ERS_RESULT_DISCONNECT;
4669         }
4670         pci_set_master(pdev);
4671
4672         pci_enable_wake(pdev, PCI_D3hot, 0);
4673         pci_enable_wake(pdev, PCI_D3cold, 0);
4674
4675         e1000_reset(adapter);
4676         ew32(WUS, ~0);
4677
4678         return PCI_ERS_RESULT_RECOVERED;
4679 }
4680
4681 /**
4682  * e1000_io_resume - called when traffic can start flowing again.
4683  * @pdev: Pointer to PCI device
4684  *
4685  * This callback is called when the error recovery driver tells us that
4686  * its OK to resume normal operation. Implementation resembles the
4687  * second-half of the e1000_resume routine.
4688  */
4689 static void e1000_io_resume(struct pci_dev *pdev)
4690 {
4691         struct net_device *netdev = pci_get_drvdata(pdev);
4692         struct e1000_adapter *adapter = netdev_priv(netdev);
4693
4694         e1000_init_manageability(adapter);
4695
4696         if (netif_running(netdev)) {
4697                 if (e1000_up(adapter)) {
4698                         printk("e1000: can't bring device back up after reset\n");
4699                         return;
4700                 }
4701         }
4702
4703         netif_device_attach(netdev);
4704 }
4705
4706 /* e1000_main.c */